Anda di halaman 1dari 95

1998 Nature America Inc. http://neurosci.nature.

com

contents

volume 1 no 5

september 1998

http://neurosci.nature.com

1998 Nature America Inc. http://neurosci.nature.com

On page 359, Xiong et al. show that the parietal eye of the sideblotched lizard uses an unusual G-protein signaling pathway. See also News and Views, page 339. Photograph courtesy of John Finn and K.-W. Yau.

editorial
Analyzing the auditory scene . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333

letter to the editor


Has a new color area been discovered? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335

news and views


Of vulcan ears, human ears and earprints. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337 Fred Wightman and Doris Kistler SEE ARTICLE, PAGE 417 Opening the third eye. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339 Fred Rieke SEE ARTICLE, PAGE 359 Intrathalamic connections: a new way to modulate cortical plasticity? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341 Jon Kaas and Ford Ebner SEE ARTICLE, PAGE 389 Discriminating native sounds: language-specific brain responses in infants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343 Kalyani Narasimhan SEE SCIENTIFIC CORRESPONDENCE, PAGE 351 How do our brains analyze temporal structure in sound? . . . . . . . . . . . . . . . . . . 343 Robert Zatorre SEE ARTICLE, PAGE 422

Language-specific responses in the infant brain. Pages 343 and 351

book review
The Minds Past . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347 by M Gazzaniga REVIEWED BY A R DAMASIO

scientific correspondence
Excitotoxic cell death requires mitochondrial calcium uptake. Page 366

Molecular identification of the corticosterone-sensitive extraneuronal catecholamine transporter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349 D Grndemann, B Schechinger, G A Rappold and E Schmig

Nature Neuroscience (ISSN 1097-6256) is published monthly by Nature America Inc., headquartered at 345 Park Avenue South, New York, NY 10010-1707. Editorial Office: 345 Park Avenue South, New York, NY 10010. Telephone 212 726 9200, Fax (212) 696 9635. North American Advertising: Nature Neuroscience, 345 Park Avenue South, New York, NY 100101707. Telephone (212) 726-9200. Fax (212) 696-9006. European Advertising: Nature Neuroscience, Porters South, Crinan Street, London N1 9SQ. Telephone (0171) 833 4000. Fax (0171) 843 4596. New subscriptions, renewals, changes of address, back issues, and all customer service questions in North America should be addressed to Nature Neuroscience Subscription Department, PO Box 5054, Brentwood, TN 37024-5054. Telephone (800) 524-0328, Direct Dial (615) 377 3322, Fax (615) 377 0525. Outside North America: Nature Neuroscience, Macmillan Magazines Ltd, Brunel Road, Basingstoke, Hants RG212XS, U.K. Annual subscription rates: U.S./Canada: U.S. $595, Canada add 7% for GST (institutional/corporate), U.S. $195, Canada add 7% for GST (individual making personal payment BN: 14091 1595 RT); U.K./Europe:395 (institutional/corporate), 175 (individual making personal payment); Rest of world (excluding Japan): 450 (institutional/corporate), 195 (individual making personal payment); Japan: Contact Japan Publications Trading Co. Ltd., 2-1 Sarugaku-cho 1 chome, Chiyoda-ku, Tokyo 101, Japan, phone (03) 292-3755. Back issues: U.S./Canada, $45, Canada add 7% for GST; Rest of world: surface U.S. $43, air mail U.S. $45. Reprints: Nature Neuroscience Reprints Department, 345 Park Avenue South, New York, NY 10010-1707. Subscription information is available at the Nature Neuroscience homepage at http://neurosci.nature.com. POSTMASTER: Send address changes to Nature Neuroscience Subscription Department, P.O. Box 5054, Brentwood, TN 37024-5054. Executive Officers of Nature America Inc: Nicholas Byam Shaw, Chairman of the Board; Mary Waltham, President; Edward Valis, Secretary-Treasurer. Printed by Publishers Press, Shepherdsville, KY, USA. Copyright 1998 Nature America Inc.

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

contents

Development of language-specific phoneme representations in the infant brain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351 M Cheour, R Ceponiene, A Lehtokoski, A Luuk, J Allik, K Alho and R Ntnen SEE NEWS AND VIEWS, PAGE 343

review
New communication between dorsal thalamic nuclei. Pages 341 and 389

Genetic dissection of Alzheimers disease and related dementias: amyloid and its relationship to tau . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355 J Hardy, K Duff, K G Hardy, J Perez-Tur and M Hutton

articles
An unusual cGMP pathway underlying depolarizing light response of the vertebrate parietal-eye photoreceptor . . . . . . . . . . . . . . . . . . . . 359 W-H Xiong, E C Solessio and K-W Yau SEE NEWS AND VIEWS, PAGE 339
1998 Nature America Inc. http://neurosci.nature.com

Glutamate-induced neuron death requires mitochondrial calcium uptake . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366 A K Stout, H M Raphael, B I Kanterewicz, E Klann and I J Reynolds A null mutation in TGF- leads to a reduction in midbrain dopaminergic neurons in the substantia nigra . . . . . . . . . . . . . . . . . . . . . . . . . . . 374 M Blum Presynaptic long-term depression at a central glutamatergic synapse: a role for CaMKII. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378 T W Margrie, J A P Rostas and P Sah
How complex receptive fields are built. Page 395

Mechanoelectrical transduction assisted by Brownian motion: a role for noise in the auditory system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384 F Jaramillo and K Wiesenfeld A new intrathalamic pathway linking modality-related nuclei in the dorsal thalamus. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389 J W Crabtree, G L Collingridge and J T R Isaac
SEE NEWS AND VIEWS, PAGE 341

Functional connectivity between simple cells and complex cells in cat striate cortex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395 J-M Alonso and L M Martinez A neural correlate for vestibulo-ocular reflex suppression during voluntary eyehead gaze shifts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404 J E Roy and K E Cullen Expectation of reward modulates cognitive signals in the basal ganglia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411 R Kawagoe, Y Takikawa and O Hikosaka Relearning sound localization with new ears . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417 P M Hofman, J G A Van Riswick and A J Van Opstal
SEE NEWS AND VIEWS, PAGE 337
Localizing sound with modified outer ears. Pages 337 and 417

Analysis of temporal structure in sound by the human brain . . . . . . . . . . . . . . . . 422 T D Griffiths, C Bchel, R S J Frackowiak and R D Patterson
SEE NEWS AND VIEWS, PAGE 343

classified advertising
see back pages

nature neuroscience volume 1 no 5 september 1998

ii

1998 Nature America Inc. http://neurosci.nature.com

editorial

Analyzing the auditory scene


The auditory system may be less well understood than the visual system, but it is no less remarkable. Vibrations on the eardrums must be analyzed to yield not only pitch, but also speech and music, as well as sound location and its changes through time. Three papers in this issue examine how the human brain analyzes sound, illustrating the complexity of the tasks needed to build up our rich perception of the auditory world. All auditory processing requires the integration of signals over time. For real-world sounds, this means analyzing multiple time scales, from the milliseconds of pitch to the seconds, even minutes, that define speech and music. Griffiths et al. (on page 422) use PET imaging to ask which brain areas are involved in processing pitch and melody. The popular account of pitch perception is that vibration at a given frequency activates hair cells in a restricted region of the cochlea. This cochlear frequency map is projected (via the brainstem, inferior colliculus and thalamus) to the auditory cortex; thus frequency can be represented through the early stages of the auditory system by a simple place code. In reality, however, things are much more complex. Pitch perception can also arise from temporal information in a noisy stimulus, even though there is no energy peak at the corresponding frequency, and the simple place-coding model cannot account for the perception of pitch from such stimuli. Instead, temporal regularities in the stimulus must be conveyed to the brain through the timing of action potentials, in other words by a temporal code rather than a place code. Griffiths et al. show that this signal has been decoded by the time it reaches the primary auditory cortex. They then take advantage of these noisy stimuli to identify brain regions that are specifically activated by melody. The beauty of this design is that by varying the amount of temporal information in the stimulus, they can cause the melody to emerge gradually from the background noise. In this way, they identify cortical regions whose activity correlates with the emergence of melody, but not with notes per se. These areas are distinct from the primary auditory cortex, and the authors speculate that they may also be involved in processing speech. The ability to discriminate speech sounds develops at a very young age. On page 351, Cheour and colleagues describe what is apparently the earliest known neural correlate of this process. Behavioral tests show that early language exposure affects a childs ability to discriminate phonemes; young babies can discriminate a wide range of sounds, but they gradually lose the ability to make discriminations that are not important within their native language. (One example is the inability of many Japanese speakers to distinguish the sounds /l/ and /r/.) This has been described as the perceptual magnet effect; sounds sufficiently similar to a prototypic vowel or consonant are captured and all perceived as examples of that sound, so that differences between them (which might be meaningful in another language) are undetected.
nature neuroscience volume 1 no 5 september 1998

Cheour and colleagues study the emergence of this effect in the first year of life, using scalp electrodes to record the responses of Finnish and Estonian children to sounds that are either common to both languages or unique to Estonian. They take advantage of a phenomenon called mismatch negativity (MMN). When a series of repeated sounds is interrupted by an unexpected oddball, it produces a distinctive electrical signal, believed to originate in or near the primary auditory cortex. The size of the MMN response presumably indicates a mismatch between the neural representations of the two sounds. At six months, Finnish children show MMN signals that correlate with the acoustical differences between the different vowels. At one year, however, they show a stronger response to the oddball sound that is a vowel in their native language than to the acoustically more dissimilar vowel that is unique to Estonian. By contrast, Estonian one-year-olds, who have been exposed to both vowels, show responses that correlate with acoustical differences. Like the visual system, the auditory system analyzes not only what but also where. Humans can localize sounds to within a few degrees, not only in the horizontal but also in the vertical (elevation) dimension. Although horizontal location can be extracted from differences in loudness and timing between the two ears, this cannot explain how we perceive elevation. This complex computation depends on the folded shape of the ears. The outer ear causes an elaborate transformation of the sound spectrum that varies with elevation, as illustrated by Van Opstal and colleagues (page 417) in their Fig. 1. To localize the source of the sound, the brain must compare the actual (transformed) sound against an assumed template. Ears, though, come in many different varieties; how does the brain know the shape of its owners ears? Presumably the transformation function must be learned, and the authors have tested this by altering their subjects ears with plastic implants. Although the implants initially disrupt perception of sound elevation, subjects learn to localize accurately with their new ears within a few weeks. Horizontal localization remains normal throughout, indicating that despite the apparent seamlessness of auditory space, the two axes arise by independent neural mechanisms. Surprisingly, subjects perform normally as soon as the implants are removed. In other words, unlike other forms of sensory adaptation (for instance visual adaptation to prism glasses), the original map is preserved alongside the newly acquired one. Whereas the visual system of humans resembles that of other primates both physiologically and psychophysically, it is uncertain to what extent this is true for the auditory system. Language is a uniquely human ability, but human sound localization and pitch perception seem to share features in common with other animals. The challenge for the field will be to find ways to appropriately relate human and animal studies, which will be essential if we are to fully understand our ability to hear.
333

1998 Nature America Inc. http://neurosci.nature.com

1998 Nature America Inc. http://neurosci.nature.com

letter to the editor

Has a new color area been discovered?


TO THE EDITOR A recent paper in Nature Neuroscience1 claims to show a previously undifferentiated cortical area that we call V8 in the human fusiform gyrus. This claim has given hopes to some2 that the cortical area responsible for the conscious perception of colors in humans has at last been found. However, the Talairach coordinates for this new area V8 (ref. 1) are identical to those that we had published for V4 (ref. 3). The authors have therefore not found a new area; instead they have rediscovered and tried to rename area V4. Furthermore, their report1 states, in reference to our paper3, that a prior study also concluded that this human color-selective region included a representation of upper and lower visual fields. How, then, can they state that colors activate area V8 but not V4 (ref. 1)? The answer is simple: it hinges on the use of the letter v, enabling one to write of V4 or V4v. To understand how a single letter can lead to such confusion, one has to retrace the history of the subject briefly. In 1995, Sereno and his colleagues, including Roger Tootell, co-author of ref. 1, reported the results of their mapping experiments in human visual cortex 4 . Many of the areas described had maps similar to ones found earlier in the macaque. Their map of what they supposed to be human V4 was not so straightforward. They distinguished a ventral V4v, located in the fusiform gyrus, from a dorsal V4d, located dorsolaterally, the two separated from each other by a relatively large expanse of cortex. V4v was clearly shown in the diagrams, but not V4d. This separation was unlike the V4 map in the monkey, where the two subdivisions, representing lower and upper visual fields respectively, are continuous with each other5. This made us suspicious, because the clinical evidence shows that lesions in the fusiform gyrus, where we had located V4 (refs 3, 6) can result in total hemi-achromatopsias 7,8 that include both upper and lower quadrants of the visual hemifield. We therefore undertook a mapping experiment3 and found, unlike the Sereno report 4 , that both quadrants are mapped within the color center (area V4) in the fusiform gyrus of each hemisphere. Human V4, like monkey V4, therefore contains a complete map of the visual hemifield in each hemisphere. It is this crucial finding that Hadjikhani et al.1 have now confirmed. Not surprisingly, the Talairach coordinates of their new area are identical to those of V4 (26, -67, -9 for our V4 and 33, -65, -14 for the new color area) but differ significantly from the coordinates of the more posterior V4v, at 32, -87, -16 (Fig. 1). Now we can see how one can write that color selectivity is located in area V8, rather than in V4 (ref. 1) or pretend that cerebral achromatopsia is produced by lesions of area V8, not the favorite candidate V4 (ref. 2). One can do so by simply dropping the v from the V4v and not stating explicitly that the human color center is distinct from V4v, though identical in position to V4 (refs 3, 6). Both the above quotes would be correct, if the v were reinstated into the V4; both are wrong without it, as is the statement that the human color center is distinct from area V4 (ref. 1). Whether this attempted renaming solves any of the mysteries of conscious color perception remains to be seen. By stating that color-selective activity is located in area V8 rather than in V4 (without adding the v), they have misled readers into believing that they have identified a new color area2. That is the central issue that the authors ought now to address unambiguously, by acknowledging that they have not discovered a previously undescribed area, that color activates V4 selectively but not V4v, and that the coordinates of the human color center coincide with what we have given for V4 (refs 3, 6) but differs from that of their V4v. We leave it to them to describe in another context their area V4v, which is the area that seems to have no monkey equivalent and thus represents the real new discovery.
S. Zeki, D. J. McKeefry, A. Bartels and R. S. J. Frackowiak Wellcome Department of Cognitive Neurology, Institute of Neurology, University College London, London WC1E 6BT, UK

1998 Nature America Inc. http://neurosci.nature.com

Fig. 1. The figure shows the locations of the three areas that are discussed in the text, in a glass brain projection. The areas were located by using the Talairach coordinates of the three areas given in the paper by Hadjikhani et al. (1998): O corresponds to area V4 defined in Lueck et al. Nature 340, 386389 (1989); Zeki et al. J. Neurosci. 11, 641649 (1991); McKeefry and Zeki Brain 120, 22292242 (1997). X corresponds to the new area V8 of Hadjikhani et al. and the + to the area V4v defined by Sereno, M.I., et al. Science 268, 889893 (1995).

T OOTELL AND H ADJIKHANI REPLY Macaque V4 is a cortical visual area that is often subdivided into dorsal and ventral parts (V4d and V4v), comprising the retinotopic representations of the lower and upper visual fields, respectively. This has been reported in numerous articles and is not in dispute. Nevertheless, Zeki and colleagues make the following claims: (1) human area V4v seems to have no monkey equivalent, and thus [it] represents the real new discovery, and (2) our human area V8 is actually equivalent to macaque V4. Their claim is based on three properties that they require of human V4: (3) human V4 should include a representation of both upper and lower visual fields, (4) human V4 should be colorselective and (5) human V4 coincides with the neural region affected in clinical achromatopsia. Our response to each of their five points follows. (1) Human V4v is not a newly discovered area, as claimed by
335

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

letter to the editor

Macaque (after Boussaoud et al, 1991)

Macaque (after Felleman and Van Essen, 1991)

c
1998 Nature America Inc. http://neurosci.nature.com

1982, ref. 13). If one does not presume that human V4 is necessarily color selective, it is easier to evaluate the human retinotopy on its own merits. If one is instead convinced that human V4 must be color-selective, the complicated rearrangement in Fig. 2d may be the best fit one can make to the actual evidence. (5) Hadjikhani et al. 1 confirm the reports of Dr. Zeki and colleagues of color selectivity in a region implicated in achromatopsia (in the collateral sulcus/fusiform gyrus). However, our analysis indicates that this color-selective area is not equivalent to macaque V4, which is why this area required a new name (V8).
R. B. H. Tootell and N. Hadjikhani Nuclear Magnetic Resonance Center, Massachusetts General Hospital, 149 13th Street, Charlestown, Massachusetts 02129, USA
1. Hadjikhani, N., Liu, A. K., Dale, A., Cavanagh, P. & Tootell, R. B. H. Nature Neurosci. 1, 235241 (1998). 2. Heywood, C. & Cowey, A. Nature Neurosci. 1, 171173 (1998). 3. McKeefry, D. & Zeki, S. Brain 120, 22292242 (1997). 4. Sereno, M. I. et al. Science 268, 889893 (1995). 5. Gattass, R., Sousa, A. P. B. & Gross, C. G. J. Neurosci. 8, 18311845 (1988). 6. Zeki, S. et al. J. Neurosci. 11, 641649 (1991).

Human (after Hadjikhani et al, 1998)

Human (after Zeki et al, 1998)

Comparison of visual cortical maps in Old World primates, drawn by different authors. All maps were drawn from right hemispheres in flattened cortical format, to approximately the same scale. (a) and (b) were based on electrophysiological and connection evidence from macaque monkeys. (c) is based on the retintopic fMRI maps from five human hemispheres, digitally averaged (see Hadjikhani et al., 1998). (d) is the revision suggested by Zeti et al.,1998, as we understand it.

Zeki et al. Retinotopically and topographically, human V4v is entirely equivalent to macaque V4v (for example, refs 5, 9, 10; compare Fig. 2c to a and b ). V4v is not even novel in humans, since it was defined equivalently by us4,11 and by others12 prior to Hadjikhani et al.1. (2) Almost all work from macaque V4 (including Zekis reports of color selectivity) has been based on recordings from the dorsal subdivision of V4, V4d. Can the human equivalent of macaque V4d be the area involved in human achromatopsia (in the collateral sulcus/fusiform gyrus)? All aspects of the human cortical map (including the topography of V4v) confirm that any human homologue of V4d would be located superior to V4v, and between V3A and MT, exactly as in macaques (Fig. 2ac). Thus, V4d should be even farther from the human color area than V4v. Retinotopic maps taken from human V4d have not been published
336

yet, but the macaque maps would predict a lower visual field representation there. (3) We agree with Dr. Zeki that any human V4 should represent the complete contralateral visual field, including both upper and lower visual fields. However, all visual areas are expected to have this property, so this argument does not help us find human V4. (4) Unlike Dr. Zeki, we had no preconceptions that retinotopically defined human V4 would also show high color selectivity. After much research and controversy2,13, macaque V4 (or any subdivision thereof) does not seem unusually color selective. For instance, after Zekis initial report claiming that 100% of the cells in V4 were color selective 14 , the percentage of reported color-selective cells shrank over the succeeding years, under increasingly detailed scrutiny (87% in 1977, ref. 15; 68% in 1978, ref. 16; 32% in 1978, ref. 17; 20% in 1981, ref. 18; 10% in 1981, ref. 19; and 18% in

7. Verrey, L. Arch. DOphtalmol. (Paris) 8, 289300 (1888). 8. Damasio, A., Yamada, T., Damasio, H., Corbett, J. & McKee, J. Neurology 30, 10641071 (1980). 9. Felleman, D. J. & Van Essen, D. C. Cereb. Cortex 1, 147 (1991). 10. Boussaoud, D., Desimone, R. & Ungerleider, L. G. J. Comp. Neurol. 306, 554575 (1991). 11. Tootell, R. B. et al. J. Neurosci. 17, 70607078 (1997). 12. DeYoe, E. A. et al. Proc. Natl Acad. Sci. USA 93, 23822386 (1996). 13. Schein, S. J., Marrocco, R. T. & de Monasterio, F. M. J. Neurophysiol. 47, 193213 (1982). 14. Zeki, S. M. Brain Res. 53, 422427 (1973). 15. Zeki, S. M. Proc. R. Soc. Lond. B 197, 195223 (1977). 16. Zeki, S. M. J. Physiol. (Lond.) 277, 273290 (1978). 17. Van Essen, D. C. & Zeki, S. M. J. Physiol. (Lond.) 277, 193226 (1978). 18. Fischer, B., Boch, R. & Bach, M. Exp. Brain Res. 43, 6977 (1981). 19. Van Essen, D. C., Maunsell, J. H. & Bixby, J. L. J. Comp. Neurol. 199, 293326 (1981).

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

news and views

Of vulcan ears, human ears and earprints


Fred Wightman and Doris Kistler
Our outer ears are critical for localizing sound elevation. Van Opstal and colleagues show that humans adapt to new ear shapes. This process resembles learning a second language because after adaptation people can localize equally well with their own or modified ears.

1998 Nature America Inc. http://neurosci.nature.com

As any Star Trek fan can tell you, First Officer Spocks ears are decidedly nonhuman. Whether or not Vulcans hear things differently (or better) than we do is hard to know, but for humans and some other Earthly creatures, ear shape is important to our perception of the auditory world. For example, our ability to localize a sound is determined in

ability, as previous research would suggest1,2. After six weeks of wearing these molds continuously, though, all four listeners seemed to have learned the new ears, so that their localization was normal again. Even more surprising was that when the molds were removed, localization was still normal. In other words, there

Pinna cues... are as individual as fingerprints, thus the term earprints.


large part by what happens when a sound wave reflects off the ridges and folds of our outer ears, the pinnae. Pinna shapes vary tremendously from person to person (Fig. 1), and to localize accurately we must learn the acoustical characteristics of our own pinnae. Mr. Spock would be the first to point out that it is logical to assume that he has learned his own ears, in spite of their very different shape, and that he can localize at least as well as we Earthlings. One wonders how well Leonard Nimoy can localize when he is using Spocks ears rather than his own. In this issue of Nature Neuroscience, Hofman, Van Riswick and Van Opstal (pp. 417421) offer an intriguing new insight into how our brains learn the characteristics of our own ears and how we might learn to use a different set of ears. They measured the localization ability of four people before and after the shapes of their ears were changed by inserting plastic molds into the pinna cavities. Changing ear shape had a dramatic deleterious effect on localization
Fred Wightman and Doris Kistler are at the Waisman Center, University of Wisconsin Madison, 1500 Highland Avenue, Madison, Wisconsin 53705, USA email: wightman@waisman.wisc.edu

was no aftereffect of the kind typically reported in sensory adaptation studies. It was as if the listeners had learned a new language and now had two sets of ears with which they were proficient. Acquisition and maintenance of a second mapping without erasing the first seems more like learning a second language than like learning a sensory perceptual transformation. Auditory researchers understand roughly how people localize sounds, but some puzzles remain. The process is very different from visual localization, in which the location of an object in the environment directly corresponds to the retinal location of the image of that object. Audition has no such direct representation of the spatial arrangement of objects. Instead, apparent sound direction must be derived from the neural representations of incoming sound waves. Until recently, the apparent direction of a sound was thought to be determined mainly by two acoustical cues that have neural representations: the interaural differences in the time of arrival of a sound wave at the two ears and the interaural differences in level or loudness. Fig. 1. The shape of the outer ear or pinna varies The problem with interaural substantially among individuals.
337

differences is that they are ambiguous cues. Consider interaural time difference. This sub-millisecond difference in time of arrival of a sound wave at a persons two ears does depend on sound source direction, and thus is a potential cue for direction. Figure 2 shows interaural time differences actually measured from a persons ears and plotted as a function of source position. Contours of constant ITD are plotted underneath the colored surface. These contours mark off sound-source azimuths and elevations for which the ITD cue is constant, and thus clearly reveal the ambiguity of this cue. Note that the contours are roughly circular, suggesting that the ITD produced by a source at 90 degrees azimuth and 30 degrees ele-

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

news and views

vation (90,30) would be about the same as the ITDs produced by sources at (90,30), (60,0) and (120,0). The other interaural difference cue, in sound level, is similarly ambiguous. To the extent that listeners rely on interaural difference cues to determine sound source direction, they should make localization errors that reflect the ambiguity of the cues. Although such errors do not seem to be a problem in real life, they are not infrequent in laboratory studies of sound localization. Most numerous are those called front-back confusions. An example of a front-back confusion would be a listeners report that the apparent direction of a sound was 120 degrees to the right of straight ahead (thus, in the rear hemifield) when in fact the source was only 60 degrees to the right. Both azimuths would produce roughly the same interaural time difference. Some listeners make large numbers of front-back confusions, and some listeners make very few. If apparent sound position were determined mainly by interaural differences, one would expect many more confusions than are observed. Clearly some other source of information is used to resolve ambiguities. This is where ear shape enters the picture. In recent years, a great deal of research has focused on the acoustical cues to sound direction produced by our pinnae. Diffraction of a sound wave around the shoulders, head and pinnae introduces large changes into the frequency content of the sound wave at each ear. These changes, pinna cues, are exactly analogous to what would happen if the sound were passed through a pair of electronic filters: certain frequencies are amplified and others are attenuated. The frequency response of these filters, commonly called headrelated transfer functions or HRTFs for short, are highly dependent on sound direction and thus offer potential localization cues. Not only are pinna cues used to determine the direction of a sound, they are essential. Among other things, pinna cues are required for accurate perception of sound source elevation 3 , and they seem to be needed to give us the out there character of everyday sounds. When pinna cues are stripped away and sounds are presented to listeners over headphones, the most common report is that the sounds are heard inside the head4. Pinna cues are idiosyncratic. Because there is a wide variety of ear shapes in the population, the laws of acoustics dictate
338

1998 Nature America Inc. http://neurosci.nature.com

Fig. 2. The difference in time of arrival of a sound wave at the two ears (ITD) plotted as a function of the azimuth (leftright) and elevation (updown) of the sound source relative to the listeners head. The ITD is given in microseconds, and azimuth and elevation in degrees. At zero degrees azimuth and zero degrees elevation, the source is directly in front of the listener. Positive angles are to the right and above. The ITDs are computed relative to the left ear, so when the source is on the left (negative azimuths), ITD is positive (blue). Note that for this listener, ITD reaches a maximum of about 500 s at an azimuth of about -90 and an elevation of 0. An ITD of 500 s implies roughly that the sound wave must travel an extra 17 cm to get to the opposite ear. The red contours plotted beneath the colored surface illustrate the ambiguity of ITD as a localization cue. All locations defined by each contour line have the same ITD.

that HRTFs will also vary a great deal from individual to individual. They are as individual as fingerprints, thus the term earprints. The perceptual relevance of the individual differences in pinna cues has been demonstrated in a number of ways, but perhaps most convincingly in experiments using virtual auditory space techniques5,6. With these virtual techniques, pinna cues are synthesized from individually measured HRTFs and then added to sounds presented through headphones. Thus, it is possible to present sounds to listeners that simulate the effect of listening through someone elses ears. The results of such experiments1,2 suggest that the most accurate localization is always achieved when ones own HRTFs are used, and that localization accuracy progressively deteriorates as the HRTFs used to synthesize sounds in virtual auditory space increasingly differ from ones own2. The most frequent kind of error made when listening through other peoples HRTFs is the front-back confusion. This result suggests that pinna cues are indeed used to resolve frontback ambiguities. Most theoretical accounts of how the brain might use pinna cues include the concept of a central repository of direc-

tion templates, derived either from the HRTFs or a transformation of them79. The idea is that the frequency content of an incoming sound is compared to each of the templates, and the one that fits the best then codes the direction of the incoming sound. Presumably the template bank is built up through long experience listening to sounds (with ones own HRTFs) in everyday life, where visual or other cues give the listener feedback about actual sound source direction. Nothing is known for certain about either how the templates are formed or how the brain uses the templates to determine a sounds direction. At this point the template theory is simply an untested hypothesis that seems to be consistent with much of the empirical data. Among the many important, but as yet unexplored, issues is how templates might develop in the early years of life when changes in head and pinna size and shape produce significant changes in the HRTFs. One obvious possibility is that the mechanism is adaptable, at least during development. The study by Hofman and colleagues offers convincing evidence for the adaptability of pinna cue processing. There is no doubt that pinna cues

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

news and views

offer sufficient information for resolving the ambiguities produced by interaural difference cues, and thus for reducing front-back confusions. However, because many listeners make large numbers of front-back confusions even when their own pinna cues are available, more information must be necessary for resolving front-back confusions, at least for these listeners. It is possible that they use dynamic cues. Hans Wallach10 originally articulated the theory that head movements could provide the information needed to resolve the ambiguity in interaural difference cues. The basic idea is that, given a fixed sound source, the way the interaural differences change during a head movement uniquely determines whether the sound is in the front or rear hemisphere. Indeed, frontback confusions nearly disappear when head movements are allowed1113. Our own recent work13 suggests that there are large individual differences in how pinna cues and dynamic cues are used. In the freestyle condition of this experiment, listeners were encouraged to move their heads, but only if they felt it would help them localize the sounds, and in a control (restricted) condition, listeners were told not to move their heads. The 2.5-second wideband noise bursts were presented from one of several hundred possible locations all around the listener, starting when the listener was facing straight ahead. Head position was continuously monitored with a magnetic tracker. The striking result was that only those listeners who made frequent front-back confusions in the restricted condition moved their heads in the freestyle condition, implying that even without feedback some listeners know that they make front-back confusions and that head movements can help them avoid such errors. Subjects in the Hofman study moved freely as they learned the new mapping, so presumably they could have used movement cues to help calibrate their new pinnae. Modern research on sound localization is informing us of the importance of earprints and providing new understanding of how they are processed. The suggestion from this new study that learning earprints is like learning a second language is especially exciting in this regard. This and other research demonstrates that sound localization involves an individual-specific integration of several different kinds of information, including interaural difference

cues, pinna cues and dynamic cues. Fascinating!, Spock would say.
1. Wenzel, E. M., Arruda, M., Kistler, D. J. & Wightman, F. L. J. Acoust. Soc. Am. 94, 111123 (1993). 2. Wightman, F. L. & Kistler, D. J. in Proceedings of the ASSP (IEEE) Workshop on Applications of Signal Processing to Audio and Acoustics. (IEEE Press, New York, 1993). 3. Gardner, M. B. & Gardner, R. S. J. Acoust. Soc. Am. 53, 400408 (1973). 4. Plenge, G. J. Acoust. Soc. Am., 56, 944951 (1974). 5. Wightman, F. L. & Kistler, D. J. J. Acoust. Soc. Am. 85, 858867 (1989).

6. Wightman, F. L. & Kistler, D. J. J. Acoust. Soc. Am. 85, 868878 (1989). 7. Macpherson, E. A. J. Acoust. Soc. Am. 101, 3105 (1997). 8. Middlebrooks, J. C. J. Acoust. Soc. Am. 92, 26072624 (1992). 9. Zakarauskas, P. & Cynader, M. S. J. Acoust. Soc. Am. 94, 13231332 (1993). 10. Wallach, H. J. Exp. Psychol. 27, 339368 (1940). 11. Perrett, S. & Noble, W. J. Acoust. Soc. Am. 102, 23252332 (1997). 12. Perrett, S. & Noble, W. Percept. Psychophys. 59, 10181026 (1997). 13. Wightman, F. L. & Kistler, D. J. J. Acoust. Soc. Am. (in press).

1998 Nature America Inc. http://neurosci.nature.com

Opening the third eye


Fred Rieke
Dark-adapted parietal-eye photoreceptors depolarize in response to light. A new study reveals that an unusual Gprotein signaling pathway mediates this response.

In Hindu mythology, the god of destruction, Shiva, is often depicted with a third eye in the middle of his forehead. Legend has it that when this third eye is opened, it can destroy anything it sees. Humans, perhaps fortunately, have no such third eye, but some lizards do have a parietal or third eye on top of their head (Fig. 1). In this issue of Nature Neuroscience (page 359365), Xiong, Solessio and Yau describe the unusual transduction cascade that underlies the response to light in photoreceptors from the lizard parietal eye. The function of the parietal eye is not entirely clear, although it seems likely to detect day/night changes in light intensity and spectral composition1. The retina of the parietal eye has two types of cells, photoreceptors and ganglion cells, but lacks the bipolar, horizontal and amacrine cells found in the retina of the vertebrate lateral eye. A curious feature of the dark-adapted parietal photoreceptors is that they depolarize when exposed to light1, whereas other vertebrate photoreceptors hyperpolarize in response to a light flash. As with vertebrate and invertebrate lateral-eye photoreceptors, phototransduction in the parietal-eye photoreceptors is
Fred Rieke is at the Department of Physiology and Biophysics, Box 357290, University of Washington, Seattle, Washington 98195, USA email: rieke@u.washington.edu

mediated by a G-protein intracellular signaling cascade. However, Xiong and colleagues show that the parietal-eye phototransduction cascade differs from other known G-protein cascades in two important ways. First, G-protein activation increases the cGMP concentration by inhibiting the cGMP phosphodiesterase (PDE, the enzyme that hydrolyzes cGMP). G-protein inhibition of PDE has not been observed previously2. Second, the PDE activity in parietal-eye photoreceptors is regulated by two G proteins operating

Fig. 1. Photograph of a side-blotched lizard, showing the parietal (third) eye on the top of its head. Photograph provided by John Finn and King-Wai Yau.
339

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

news and views

antagonistically; one G protein is constitutively active and the other G protein is activated by the photopigment. Such antagonistic control of PDE has not been found elsewhere, although similar regulation occurs for adenyl cyclase3. These new results on phototransduction in the parietal-eye photoreceptors add to the diversity of known G-protein signaling mechanisms. Second-messenger cascades use G proteins to communicate and amplify the activity of a receptor to downstream elements of the cascade. Such G-protein signaling cascades mediate a vast array of biological processes, from the transduction of sensory input signals to the regulation of blood flow. These G-protein cascades provide the amplification required to detect a single odorant molecule4 or a single photon 58 , they fine-tune the excitability of neurons in response to a variety of extracellular signals9, and they mediate slow synaptic potentials. Although all these cascades involve G proteins, the diversity of their downstream signaling produces very different neural responses. In vertebrate rod and cone photoreceptors6, a G-protein cascade (Fig. 2a) links the photopigment (P) to the membrane current, acting as a molecular amplifier and preventing the influx of more than 106 sodium ions during the course of the single-photon response in the rod photoreceptor 7,8. The light-activated photopigment (P*) catalyzes the activation of the G protein transducin; in the rod, a single photopigment molecule can activate several thousand transducin molecules while it is active. Transducin activates PDE, and each activated PDE molecule can hydrolyze 50100 cGMP molecules during its active lifetime. The resulting drop in cGMP concentration causes channels in the surface membrane of the outer segment to close, the influx of cations to decrease and the cell to hyperpolarize.

...activity in parietal-eye photoreceptors is regulated by two G proteins operating antagonistically...


Phototransduction in invertebrate photoreceptors differs both mechanistically and functionally from transduction in vertebrate rods and cones. Invertebrate photoreceptors depolarize rather than hyperpolarize in response to light, and the phototransduction cascade is remarkably
340

fast; the single-photon response in invertebrate photoreceptors can be less than 30 ms in duraa Rod and cone photoreceptors tion, which is about 5'GMP 10 times faster than G* + PDE PDE* mammalian rods and 100 times faster than P* cGMP light amphibian rods. current G GC Invertebrate photoreP GTP ceptors seem to use a phosphoinositide signaling cascade 10,11 b Invertebrate photoreceptors (Fig. 2b) rather than ? the cGMP signaling PLC* G* + PLC opens current cascade of vertebrate P* light rods and cones. As in G vertebrates, the activated photopigment P molecule activates a G c Parietal eye photoreceptors protein, but in this 5'GMP G +* + case the G protein PDE G-* activates phospholicGMP pase (PLC). The actiP* light current vated phospholipase GC Gcauses channels in the GTP P cell membrane to open, increasing the influx of cations and depolarizing the cell. Fig. 2. Summary of the key steps in three phototransduction cascades. (a) Vertebrate rods and cones. (b) Invertebrate phoTo date, the identity toreceptors. (c) Parietal-eye photoreceptors. P, photopigment; of the signal permitG, G protein; PDE, phosphodiesterase; GC, guanylate cyclase; ting phospholipase to cGMP, cyclic GMP; PLC, phospholipase C. open membrane channels has remained elusive. opposite response polarities arise? In this The parietal-eye photoreceptors share new study, Xiong and colleagues studied some properties with vertebrate rods and phototransduction in isolated parietal-eye cones and some with invertebrate phophotoreceptors using whole-cell or perfotoreceptors. In morphology they resemrated-patch voltage-clamp recordings. They ble vertebrate cones, but the depolarizing found that inhibition of the PDE activity in light response of the dark-adapted pariisolated photoreceptors rapidly increased etal-eye photoreceptors suggests a transthe current, indicating that in darkness duction cascade similar to that of cGMP is synthesized continuously by invertebrate photoreceptors. Earlier work guanylate cyclase but rapidly hydrolyzed by by Finn, Solessio and PDE. Blocking G-protein activation also Yau12 showed that memrapidly increased the current, suggesting brane patches from the that PDE activity in the dark is produced by outer segment of the paria constitutively active G protein. In addietal-eye photoreceptors tion, the light-induced increase in inward had a high density of current and the resulting depolarization cGMP-gated channels were suppressed by PDE inhibitors. Based with properties similar to on this and other evidence, they propose the those of vertebrate rods following transduction scheme (Fig. 2c): and cones. Based on this light acts by inhibiting the PDE (rather than similarity, they suggested that the depoactivating PDE as in rods and cones), perlarizing light response was produced by mitting the cGMP level to increase and an increase in the internal cGMP channels in the surface membrane to open. concentration. Because the photopigment is a G-proteinThis suggests a transduction cascade coupled receptor, PDE inhibition by light is using much of the same machinery as rod probably mediated by a second G protein. and cone photoreceptors. How then do the
nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

1998 Nature America Inc. http://neurosci.nature.com

news and views

http://neurosci.nature.com

In addition to their implications for the study of G-protein signaling, these studies raise a number of interesting questions. Calcium is critical to adaptation to steady light in photoreceptors from vertebrate and invertebrate lateral eyes10,13. It will be interesting to see if the same is true for parietaleye photoreceptors. Vertebrate rod and cone photoreceptors obey the principle of univariance, which states that absorbed photons produce identical responses independent of wavelength. Parietal-eye photoreceptors do not; when exposed to steady light, the parietal-eye photoreceptors can produce hyperpolarizing responses to blue light and depolarizing responses to green light1. Invertebrate photoreceptors exhibit a similar effect in which long-wavelength light can turn the photopigment off14, the opposite of the response to short-wavelength light. This provides a simpler pathway for the recycling of light-activated photopigment molecules than that found in vertebrate rods and cones. A more general question that continues to bother many of us who study rod

and cone photoreceptors is the significance of the hyperpolarizing light response. This question becomes more puzzling, given that similar biochemical machinery is used by the parietal-eye photoreceptors to produce a depolarizing response. Simple energetic arguments dont seem to strongly favor one response polarity over the other, as each type of photoreceptor is active for roughly half of a 24-hour day. Perhaps a more likely explanation is that the hyperpolarizing responses allow a graceful transition from rod to cone vision; as the light level increases, the rods hyperpolarize and stop releasing neurotransmitter, thus no longer providing input to the visual system. Were the cells to depolarize in light instead, they would continue to be active in light and could provide unwanted visual input. This speculation aside, a better functional and mechanistic understanding of the parietal-eye photoreceptors should shed new light on intracellular signaling cascades and surprise us with the richness and diversity of G-protein signaling.

1. Solessio, E. & Engbretson, G. A. Nature 364, 442445 (1993). 2. Wedel, B. J. & Garbers, D. L. FEBS Lett. 410, 2933 (1997). 3. Gudermann, T., Schoneberg, T. & Schultz, G. Annu. Rev. Neurosci. 20, 399427 (1997). 4. Menini, A., Picco, C. & Firestein, S. Nature 373, 435437 (1995). 5. Baylor, D. A., Lamb, T. D. & Yau, K.-W. J. Physiol. 288, 613634 (1979). 6. Baylor, D. A. Proc. Natl Acad. Sci. USA 93, 560565 (1996). 7. Rieke, F. & Baylor, D. A. Rev. Mod. Physics 70, 10271036 (1998). 8. Pugh, E. N. Jr & Lamb, T. D. Biochem. Biophys. Acta 1141, 111149 (1993). 9. Hille, B. Trends Neurosci. 17, 531536 (1994). 10. Scott, K. & Zuker, C. Trends Biochem. Sci. 22, 350354 (1997) 11. ODay, P. M., Bacigalupo, J., Vergara, C. & Haab, J. E. Mol. Neurobiol. 15, 4163 (1997). 12. Finn, J. T., Solessio, E. C. & Yau, K.-W. Nature 385, 815819 (1997). 13. Koutalos, Y. & Yau, K.-W. Trends Neurosci. 19, 7381 (1996). 14. Richard, E. A. & Lisman, J. Nature 356, 336338 (1992).

1998 Nature America Inc.

Intrathalamic connections: a new way to modulate cortical plasticity?


Jon Kaas and Ford Ebner
Crabtree and colleagues report newly discovered connections between nuclei of the somatosensory thalamus, challenging the traditional view that thalamic nuclei do not communicate.

Imagine sending a letter, only to discover that the post office had changed the contents before delivery. Many neurobiologists tend to think of the thalamus as a sort of post office, whose job is to relay sensory information to the neocortex through distinct and functionally isolated nuclei. Different cortical areas are known to be richly connected with each other and to interact in a complex way, but thalamic neurons do not interconnect, and hence
Jon Kaas and Ford Ebner are in the Department of Psychology, 301 Wilson Hall, Vanderbilt University, Nashville, Tennesee 37240, USA e-mail: jon.kaas@vanderbilt.edu and ford.ebner@vanderbilt.edu

have been thought to do their job rather rigidly. Thus, we are surprised by the finding of Crabtree, Collingridge and Isaac 1 on page 389 of this issue that neurons in separate somatosensory nuclei of the dorsal thalamus do influence one anothers activity. This interaction was demonstrated in a tissue slice, where stimulating neurons in the ventroposterior nucleus (VP, also called the ventrobasal nucleus) by local application of glutamate inhibited the activity of neurons in the medial posterior nucleus (POm).

These two nuclei had to interact through intrathalamic connections, because other parts of the brain, including somatosensory cortex, were eliminated from the slice. The pathway for the intrathalamic interaction seems to be through the thalamic reticular nucleus (TRN). This nucleus 2 , part of the ventral thalamus, consists of a pure population of inhibitory GABAergic neurons. Axons coursing between the thalamus and cortex send side branches to synapse within the TRN as they pass through. Thus, the TRN receives activating inputs from both the dorsal thalamus and the cortex, but it projects exclusively to the dorsal thalamus. These connections have been considered to be highly topographic, in that axons from each restricted region of the thalamus terminate on the same TRN neurons that project back to that region of the thalamus3.

...one function of intranuclear inhibition... could be to regulate the ease with which the receptive field properties of cortical neurons can be changed.
341

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

news and views

another thalamic nucleus, POm. Similar intranuclear interactions may occur in many thalamic nuclei. What might this newly revealed intrathalamic pathway do? A major effect of TRN activity is to reduce the sizes of receptive fields of its target neurons in the dorsal thalamus through inhibition, which suppresses all but the earliest and most potent component of an activating sensory volley 4 . TRN neurons are powerfully activated by ascending sensory inputs via thalamocortical axon collaterals, and by inputs from corticothalamic axons. Both sources of reticular nucleus activation would converge on the Fig. 1. A possible role for intrathalamic inhibition in regusame set of TRN neurons, lating cortical plasticity. VP projects to the barrel domains in layer IV of S1 cortex, and POm axons project to the albeit at slightly different septa around the barrels, where they promote intracorti- times following a sensory cal spread of activity. The TRN normally inhibits POm neu- volley, and suppress rons via an intrathalamic route. Arousal activates neurons delayed, weak and off-focus that inhibit the TRN, thereby reducing its inhibition of sources of thalamic activation. Thus, suppression of POm neurons, which increases cortical plasticity. cortical activating inputs to the TRN and dorsal thalamus can lead to large increases in the sizes of receptive fields However, Crabtree and colleagues for VP neurons6. demonstrate that a given region of the TRN can project to two nuclei within The identification of intranuclear the dorsal thalamus. They injected tracinhibition via the TRN adds another ers into their tissue slices and revealed dimension to this already-complex netthat both VP and POm project to the work (Fig. 1). One possible role for this same sector of TRN. In the brain slice, newly described pathway is in controlthe TRN neurons complete the only ling cortical placticity. Projections from preserved pathway between the two VP activate neurons in primary nuclei, and thus the authors conclude somatosensory cortex (S1), and projecthat VP inputs to TRN activated an tions from S1, in turn, provide excitainhibitory pathway to POm. Of course, tory feedback onto neurons in VP and the well known connections from a POm as well as to the TRN. Although at given nucleus to TRN and back to the least parts of POm receive ascending same nucleus predominate. Thus, one sensory projections, inputs from S1 are important function of TRN neurons necessary for POm neurons to respond could be to act very much like the to sensory stimulation 7 . In addition, intrinsic inhibitory neurons in the dorPOm cells generally have larger recepsal thalamus, especially in rats, which tive fields than VP neurons when meahave no intrinsic inhibitory neurons in sured under anesthesia8. Large receptive VP at all. The TRN neurons would fields are compatible with our proposed inhibit the same set of neurons that actirole for POm projections in plasticity. vate them, and perhaps a fringe of surPOm neurons project to locations in S1 rounding neurons. However, Crabtree that would facilitate the horizontal and colleagues 1 provide evidence for spread of activity in cortex and could potentially promote the integration of another, quite different role for TRN, in inputs during experience-related cortiwhich neurons in one nucleus of the cal plasticity. For example, repeated dorsal thalamus, VP, excite TRN neusimultaneous stimulation of pairs of rons, which then inhibit neurons in
342

weak-effect and strong-effect whiskers on the face increases the responsiveness of cortical neurons after 50100 trials to the weak-effect whisker in awake rats9. The facilitory role of POm neurons in this process would be reduced by VPinitiated inhibition through the TRN circuit. Thus, one function of intranuclear inhibition in the thalamus through the reticular nucleus could be to regulate the ease with which the receptive field properties of cortical neurons can be changed. Under many circumstances, a low threshold for the synaptic modifications that produce cortical plasticity could be detrimental and unwanted. During these conditions, intrathalamic inhibition would suppress the POm relay to cortex, reducing its role in facilitating plasticity. During other circumstances, however, it could be advantageous for neuronal properties to change following a single experience. For example, it would be adaptive for one encounter with a tiger to permanently change an animals behavior. In this case, change could be potentiated by shutting down the reticular nucleus. Inputs to the TRN from novelty-activated, ascending cholinergic brainstem pathways are inhibitory 10 . Therefore, during quiet behavior, plasticity would be reduced, but during peak periods of arousal and attention, the suppression of the TRN would enhance cortical plasticity. These are, of course, rather unsupported speculations. Nevertheless, they usefully indicate a possible way that the VP to TRN to POm pathway might have a powerful influence on cortical function.
1. Crabtree, J. W., Collingridge, G. I. and Issac, J. T. R. Nature Neurosci. 1, 389394 (1998). 2. Jones, E. G. The Thalamus (Plenum, New York, 1985). 3. Cox, C. L., Huguenord, J. R. & Prince, D. A. J. Comp. Neurol. 366, 416430 (1996). 4. Lee, S. M., Friedberg, M. H. & Ebner, F. F. J. Neurophysiol. 71, 17021715 (1994). 5. Pinault, D., Bourassa, J. & Deschenes, M. Eur. J. Neurosci. 7, 3140 (1995). 6. Ergenzinger, E. R., Glasier, M. M., Hahn, J. D. & Pons,T. P. Nature Neurosci. 1, 226229 (1998). 7. Diamond, M. E., Armstrong-James, M., Budway, M. J. & Ebner, F. F. J. Comp.Neurol. 319, 6684 (1992). 8. Diamond, M. E., Armstrong-James, M. & Ebner, F. F. J. Comp. Neurol. 318, 462476 (1992). 9. Delacour, J., Houcine, O. & Talbi, B. J. Neurosci. 23, 6371 (1987). 10. Dingledine, R. & Kelly, J. S. J. Physiol. (Lond.) 271, 135154 (1977).

1998 Nature America Inc. http://neurosci.nature.com

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

news and views

Discriminating native sounds: language-specific brain responses in infants


Most adults find it difficult to hear the differences among sounds that do not occur in their own language. How do children learn to perceive the sounds of their native language more easily than the sounds of a foreign one? A study by Marie Cheour and colleagues (on page 351) reports the first neural correlate of this learning in infants. The authors examined mismatch negativity (measured by scalp electrodes, see photo), which is an electrical signal produced in response to an unexpected auditory stimulus in a series of repeating stimuli (in this case, the vowel /e/). Six-month-old Finnish infants had a slightly smaller mismatch negativity response to the Finnish vowel // (which also occurs in Estonian) than to the uniquely Estonian vowel //. By one year, the same Finnish infants responded much more strongly to the vowel from their native language, whereas Estonian children of the same age perceived both sounds equally well. This study shows that language-specific memory traces in the human brain emerge between six months and one year of age.

. http://neurosci.nature.com

Kalyani Narasimhan

1998 Nature America In

How do our brains analyze temporal structure in sound?


Robert Zatorre
How does the human brain process the temporal structure of a musical sound? A PET imaging study identifies cortical regions involved in pitch computation and melodic pattern recognition.

As a consequence of its physical nature, sound unfolds over time. This presents unique problems for the nervous system, which is faced with the task of extracting and encoding information contained in vibrating air molecules over short time periods, as well as integrating successive sonic events over longer time frames. The auditory system is therefore highly specialized for processing temporal events, which constitute the lowest common denominator of everything from the bark of a dog to a Bach cantata. The brain imaging study
Robert Zatorre is at the Department of Neuropsychology, Montreal Neurological Hospital, 3901 University, Montreal PQ H3A 2B4 Canada email: md37@musica.mcgill.ca

by Griffiths and colleagues in this issue of Nature Neuroscience (pp 422427) examines how the human brain may process the temporal structure contained in a musical sound. They identify different cortical regions involved in the processing of short-term (pitch computation) and longerterm (melodic pattern) temporal information. These investigators took advantage of a phenomenon described some three hundred years ago by Huygens1, who noted that the periodic reflections of the noise made by a fountain from the stone steps of a staircase resulted in an audible pitch (Fig. 1). The same phenomenon can be reproduced and studied in the laboratory by tak-

ing a sample of random noise and passing it through a cascade of delay-and-add networks. That is, the noise is displaced by a brief time delay and added to itself; then the output of that procedure is again added to itself with the same delay, and so on for some number of iterations. This process results in an audible pitch corresponding to the reciprocal of the time delay constant used (ref. 1 and Fourcin, A., Fifth Intl. Congress Acoust., B42, 1965). This pitch sensation is of particular interest because it is perceived despite the absence of any auditory spectral cues that could activate a particular frequency representation in the cochlea, in contrast to a typical complex tone. Instead, the perception of pitch must result from the central nervous systems

...seeking areas whose activity correlated... with increasing salience of the melody.
capacity to encode temporal regularity in the neural firing pattern. Since the time of Helmholtz, it has been controversial whether temporal or spectral information is the basis for the sensation of pitch. This duality arises because temporal regularity in a signal, or periodicity, gives rise both to temporally regular neur343

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

news and views

produces a stronger sense of pitch. This approach, known as a parametric design, is a powerful one that is becoming increasingly common in the brain imaging community. By looking for CBF changes that covary with an input variable Bob Crimi (in this instance, numFig. 1. Illustration of the phenomenon described by Christian ber of iterations), one Huygens in 1693. He noted that the noise produced by a foun- avoids the problems tain at the chteau of Chantilly de la Cour was reflected by a associated with substone staircase in such a way that it produced a musical tone. traction analysis, in He correctly deduced that this was due to the successively which it is necessary to longer time intervals taken for the reflections from each step assume that two experto reach the listeners ear. imental conditions differ only by a discrete number of components. In a parametric al firing patterns and to spatially distinct design, by contrast, all that is required is that neural activity for different frequencies. In the CBF changes be related in a roughly linother words, frequency of vibration is repear fashion to the input variable. Moreover, resented by the nervous system as both a although techniques such as PET and fMRI time code and a place code. Both phesuffer from relatively poor temporal resonomena arise in the inner ear; the time lution, it is nonetheless possible to use them code is made possible because auditory to study temporal processing at a much fibers originating in the cochlea fire action finer time scale by varying the stimulus potentials that are phase-locked to the dimension of interest appropriately. That is, stimulus, whereas the place code, known rather than measuring the millisecond-level as tonotopy, arises because frequency preftiming of responses, the researcher measures erence is mapped along the length of the the differences in response to stimuli that basilar membrane in the cochlea. Because vary on that time scale. of the orderly projection of fibers through As the number of iterations was the auditory pathway, this tonotopic map increased, the authors observed a nearly linis maintained in the primary auditory corear increase in CBF bilaterally in the supetex and adjacent areas2. It has therefore trarior temporal gyrus, near the primary auditory cortex of Heschls gyrus (see Fig. ditionally been difficult to determine 2), as determined from anatomical probawhether a particular effect may arise from bility maps3. These results directly implitemporal or spectral aspects of pitch processing. The stimulus used by Griffiths and cate a region within or adjacent to the colleagues is particularly well suited to primary auditory cortex in processing temstudying temporal processing in isolation, poral fine structure, and indicate that tembecause it results in an essentially flat specporal integration probably occurs at or trum of cochlear activation (as shown by before the primary auditory cortex, perthe simulations illustrated in their Fig. 1), haps in the inferior colliculus or medial with no peak at the frequency correspondgeniculate nucleus of the thalamus. One ing to the perceived pitch. Hence, the pitch question that arises from these data is the that is perceived as a result, and any assoextent to which the CBF pattern directly ciated neural activity that is correlated with correlates with perceived pitch salience, as pitch perception, must result from tempoopposed to the physical structure of the ral processing. stimulus. It would be of interest, for examHow and where does the brain extract ple, to investigate how perceived pitch senpitch information from temporal cues? In sation (as rated by the subjects) relates to their first experiment, Griffiths and colthe observed pattern of cerebral activity. It leagues used positron emission tomography will also be important to compare more (PET) to analyze changes in cerebral blood directly the cortical mapping of temporalflow (CBF, which reflects neural activity) as ly based pitch and of spectrally determined a function of increasing number of iterapitch to see to what extent they correspond. tions of the noise-delay stimulus. Each sucIn other functional imaging studies, cessive iteration of the stimulus sequence specific activation patterns seem to be
344

1998 Nature America Inc. http://neurosci.nature.com

related to the processing of stimuli containing spectral modulation but whose temporal structure is constant (Thivard, L. et al., NeuroImage, 7, 373, 1998). These data complement those of Griffiths and colleagues and seem to point to distinct, functionally specialized cortical fields that are concerned with spectral or temporal aspects of stimulus processing. Such findings are also in keeping with neurophysiological data showing sensitivity of neurons in the superior temporal gyrus of the macaque to stimuli with varying spectral energy peaks4. These peaks, or formants, are interesting because they are relevant for the perception of speech, which depends on both spectral and temporal cues. What remains to be done, however, is to specify which areas in particular are involved, what their boundaries are, and most importantly what their precise functional contribution to different types of auditory processing may be. In their second experiment, Griffiths and colleagues extended their parametric approach to examine the CBF pattern associated with longer-term temporal integration processes. They used the same iterated noise stimuli described above, but this time the pitches formed a brief structured melodic pattern (whereas in the prior condition they only formed a relatively monotonous tonal pattern). This manipulation allowed the authors to investigate the emer-

Bob Crimi

Fig. 2. Lateral view of the human brain illustrating areas of cortex activated by delay-and-add noise. The primary auditory cortex is within the Sylvian fissure, on the internal surface of the temporal lobe. The pitch-sequence detection areas described by Griffiths and colleagues are on the external surface of the temporal lobe. These areas are activated by delayand-add notes arranged into musical phrases.

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

news and views

1998 Nature America Inc. http://neurosci.nature.com

gence of longer-term structure in the stimulus by seeking areas that showed an interaction between number of iterations and the presence or absence of a melodic pattern. In other words, they were seeking areas whose activity was correlated not with the perception of pitch per se, but rather with the increasing salience of the melody (as its component pitches were strengthened by increasing the number of iterations). Four such areas were identified that increased their response as a function of iteration much more for tonal melodic patterns than for non-melodic pitch patterns. These regions were located symmetrically in the two hemispheres: two in the superior temporal gyrus posterior to Heschls gyrus and two in the superior temporal sulcus/middle temporal gyrus (Fig. 2). These four regions were spatially distinct from the peri-primary areas identified in the previous analysis, suggesting that the fine temporal processing related to pitch extraction involves different regions and mechanisms than the longer-term time processing that would be involved in the detection of melodic patterns. A number of previous studies that have

examined pitch processing either by functional imaging5,6 or by studying braindamaged patients7,8 have suggested that tonal processing is lateralized to the right temporal cortex. This conclusion seems to be at odds with the bilateral activation observed by Griffiths and colleagues. However, pitch processing is clearly very complex, and the various studies may have been examining different aspects of this phenomenon. For instance, the right-side bias that has emerged from some earlier studies may reflect a process based largely on spectral cues, whereas the fine-grained temporal analysis of periodicity in the present study may be a more bilaterally distributed function. Furthermore, the lesion data indicate that whereas relatively mild deficits occur after unilateral cortical damage, much more devastating effects are observed with bilateral lesions9, hence indicating that hemispheric differences are not absolute. Nevertheless, the differences raise a general question that arises with any functional imaging experiment. Functional imaging, as exemplified by this carefully designed and executed study, can provide crucially important information about

human perceptual and cognitive functions. Nonetheless, it can only identify neural correlates of these functions, and it cannot determine the degree to which a given site of activity provides essential computations. In the long run, it will be important to combine functional imaging data with the results of lesion studies if we are to gain a clearer understanding of the complexities of the human auditory system.

1. Bilsen, F. & Ritsma, R. Acoustica 22, 6373 (1969/70). 2. Merzenich, M. M. & Brugge, J. F. Brain Res. 60, 315333 (1973). 3. Penhune, V. B., Zatorre, R. J., MacDonald, J. D. & Evans, A. C. Cereb. Cortex 6, 661672 (1996). 4. Rauschecker, J. P., Tian, B. & Hauser, M. Science 268, 111114 (1995). 5. Zatorre, R. J., Evans, A. C., Meyer, E. & Gjedde, A. Science 256, 846849 (1992). 6. Zatorre, R. J., Evans, A. C. & Meyer, E. J. Neurosci. 14, 19081919 (1994). 7. Zatorre, R. J. J. Acoust. Soc. Am. 84, 566572 (1988). 8. Zatorre, R. J. & Samson, S. Brain 114, 24032417 (1991). 9. Peretz, I. et al. Brain 117, 12831301 (1994).

nature neuroscience volume 1 no 5 september 1998

345

1998 Nature America Inc. http://neurosci.nature.com

scientific correspondence Molecular identification of the corticosteronesensitive extraneuronal catecholamine transporter


Dirk Grndemann1, Birgit Schechinger2, Gudrun A. Rappold2 and Edgar Schmig1
1

Department of Pharmacology, Im Neuenheimer Feld 366 and 2Department of Human Genetics, Im Neuenheimer Feld 328, University of Heidelberg, 69120 Heidelberg, Germany Correspondence should be addressed to E.S. (edgar.schoemig@urz.uni.heidelberg.de)

Catecholaminergic signaling regulates various physiological functions, such as blood pressure1 and is implicated in drug dependence, affective disorders and male aggressive behavior2,3. The actions of released catecholamines are terminated by sodium-driven, high-affinity transporters in the plasma membrane of the releasing neurons4,5 and by a corticosteronesensitive, low-affinity, high-capacity extraneuronal transport system6, originally named uptake2, found in sympathetically innervated tissues7 and in central nervous system glia8. Here we report the molecular identification and pharmacological characterization of the extraneuronal catecholamine transporter, which is unrelated to the family of sodium-driven neuronal monoamine transporters5. Extraneuronal uptake is closely related to the non-neuronal metabolism of catecholamines7 by catechol-O-methyltransferase (COMT), which exists almost exclusively in non-neuronal cells9, and monoamine oxidases (MAO-A and MAO-B). Extraneuronal transport is the predominant pathway for terminating the actions of circulating adrenaline and noradrenaline10. Although neuronal and extraneuronal uptake compete for released catecholamines, these transport systems have distinct pharmacological profiles, that is, affinity for substrates and sensitivity to various drugs. Overlap in the antagonist sensitivity between extraneuronal catecholamine uptake and apical renal transport of organic cations by OCT2 (ref. 11) raised the possibility that the extraneuronal transporter might belong to the recently defined family of amphiphilic solute facilitators (ASF)12. Degenerate oligonucleotides were derived from common sequence motifs of the ASF family and used for PCR on cDNA from Caki-1 cells, a human kidney carcinoma cell line known to express extraneuronal noradrenaline transport13. Amplicons of the expected sizes were isolated by ultraviolet-protected gel electrophoresis, cloned and sequenced. A fragment with similarity to members of the ASF family was identified. This fragment, in northern analysis of Caki-1 mRNA, detected a single band with a length of approximately 3.4 kb. The full-length cDNA of the corresponding transporter was assembled from a Caki-1 cDNA library clone and a fragment from inverse PCR. Because of its functional characteristics, we named this new transporter EMT (extraneuronal transporter for monoamine transmitters). Amino-acid sequence analysis identified EMT as a new member of the ASF transporter family12. No proteins highly homologous to EMT are yet known. EMT is similar to the OCT and OAT proteins (renal transporters for organic cations and organic anions), with identity (similarity) scores of about
nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

Fig. 1. Amino-acid sequence of EMT. PCR with degenerate oligonucleotides directed against common sequence motifs of the ASF family of transporters was done on cDNA prepared from human Caki-1 cells (ATCC HTB-46). The sequence of EMT is deposited in GenBank/EBI Data Bank under the accession number AJ001417. A 210-bp fragment with similarity to ASF family members was obtained using the forward primer 5GARTTRTAYCCNAC3 and the reverse primer 5TTNGTYTCNGGYAA3. A primary plasmid library in Escherichia coli DH10B was constructed from size-fractionated (2.83.8 kb) Caki-1 cDNA and screened by colony hybridization. Four independent positive clones were isolated, all with an incomplete coding sequence at the 5-end. To determine the missing sequence information, inverse PCR was done using specific primers for first-strand cDNA synthesis (reverse complementary to positions 15701587 of the EMT cDNA) and subsequent PCR (positions 473496, reverse, and 14951518, forward). The main product of 570 bp was cloned and sequenced. A cDNA with the complete open reading frame of EMT was assembled from the PCR fragment (forward primer, positions 1639; reverse primer, 730753) and one of the incomplete library clones (positions 5662659), using the BstBI site at position 732. Boxes indicate 12 putative transmembrane segments. Consensus sequences for glycosylation are marked by asterisks.

50% (70%) and 32% (55%), respectively. EMT consists of 556 amino acids ( Fig. 1 ) with 12 putative transmembrane segments. To determine whether the functional properties of EMT match the characteristics of extraneuronal transport of catecholamines, the EMT cDNA was inserted into the expression vector pcDNA3 and transfected into 293 cells, a cell line from human embryonic kidney. Cells stably transfected with pcDNA3EMT, but not with pcDNA3, show uptake of known substrates of the extraneuronal catecholamine transporter, such as tyramine, adrenaline, noradrenaline, 5-hydroxytryptamine and the neurotoxin 1-methyl-4-phenylpyridinium (MPP+) (Fig. 2a). In accordance with data from perfused rat heart, adrenaline was taken up three times more efficiently than noradrenaline7. Tetraethylammonium (TEA), a prototypical substrate for the OCT proteins, was not accepted as a substrate by EMT. At the short incubation period of 45 seconds, uptake of dopamine by EMT was minimal. Initial rates of specific noradrenaline uptake were saturable (Fig. 2b), with an apparent Michaelis-Menten constant, K m, of 510 M (95% confidence interval, 360730) and a maximal transport rate, Vmax, of 3.9 0.3 nmol/min per mg protein (which translates into 0.9 fmol/min per cell). This measured Km corresponds well with the 252 M K m of nora349

1998 Nature America Inc. http://neurosci.nature.com

scientific correspondence

a
Specific uptake (pmol/min per mg protein)

b
Specific 3H-noradrenaline uptake (nmol/min per mg protein)

c
3H-MPP+ uptake (relative to control)

e in

PP

lin

lin

in

5-

am

am

TE

re

re

Ty r

Ad

ad

1998 Nature America Inc. http://neurosci.nature.com

e d Fig. 2. Functional characteristics of EMT. The EMT cDNA was inserted into the KpnI and BamHI sites of pcDNA3 (Invitrogen, NV Leek, The Netherlands) to produce pcDNA3EMT. 293 cells (CRL-1573; ATCC) were stably transfected with pcDNA3EMT or pcDNA3 (controls). (a) Specific uptake of radiolabeled substrates, each at 0.1 M. The cells were incubated for 45 seconds at 37C. Specific uptake is that fraction of total uptake that is sensitive to 1 M disprocynium24. (b) Saturation of initial rates of specific 3H-noradrenaline uptake. Specific uptake was calculated by subtraction Ki for noradrenaline uptake into from total uptake of the uptake into control Caki-1cells (M) 3 + cells. (c) Inhibition of specific H-MPP uptake by disprocynium24 (filled circle), corticosterone (open square), O-methylisoprenaline (filled diamond), cimetidine (open circle), procainamide (filled square) and (-)noradrenaline (open diamond). (d) Correlation between Ki values for MPP+ uptake into 293 cells expressing EMT and for extraneuronal noradrenaline uptake into Caki-1 cells. A highly significant correlation was found (r = 0.998, n = 6, and p < 0.001). The slope of the regression line is 1.11 0.03. (e) Effect of specific catecholamine transport inhibitors on EMT. In all experiments involving catecholamines, the solutions contained 1 mM ascorbic acid, 10 M U0521 and 10 M pargyline to prevent non-enzymatic and enzymatic degradation. Shown are means standard error. (n = 34).
Ki for MPP+ uptake into 293 cells expressing EMTh (M)
3H-MPP+ uptake (pmol/min per mg protein)

or

op

3H-Noradrenaline

(mM)

Inhibitor (M)

na

na

drenaline uptake by the extraneuronal system in the isolated perfused rat heart 6. Although the K m for noradrenaline of EMT is considerably higher than the K m for neuronal uptake, the Vmax of noradrenaline uptake by EMT is between two and three orders of magnitude higher than the Vmax of noradren-

aline uptake by the heterologously expressed neuronal noradrenaline transporter in tissue culture 4 and in intact rat heart 6,14. Thus, the high V max of EMT compensates for the relatively low Km, resulting in comparable efficiencies (rate of transport per cell or tissue weight) of extraneuronal and neu-

Fig. 3. Tissue distribution and chromosoa b mal localization of EMT. (a) Detection by RT-PCR of EMT mRNA in total RNA from human tissues with primers specific for EMT (positions 10771100, forward; and positions 17081731, reverse) and, as a control, GAPDH (annealing temperature 55 C, 34 cycles). The mRNA-derived signal is the difference of band intensities between samples generated in the presence (+) or absence (-) of reverse transcriptase (RT). (b) Fluorescence in situ hybridization with an EMT-containing PAC and probes specific for chromosome 6. PAC DNA was nick-translated with biotin-11-dUTP, and the DNA chromosome 6 from sorted human chromosome 6 was nick-translated with digoxigenin-11-dUTP. PAC DNA and DNA specific for chromosome 6 were cohybridized to metaphase chromosomes from stimulated lymphocytes of a caucasian male individual. The biotinylated probe was detected by avidin-conjugated FITC (yellow), and the probe labeled with digoxigenin was detected by mouse-anti-digoxigenin and Cy3-conjugated sheep-anti-mouse antibodies (red). Chromosomes were counterstained with DAPI. Chromosomes 6 are displayed separately, showing the localization of the EMT gene in 6q27.
350 nature neuroscience volume 1 no 5 september 1998

30 rpin nM e ip ra 30 m D 0 in is pr n e oc M yn i 30 um 0 2 nM 4 D es

nt

ro

Co

Re

se

1998 Nature America Inc. http://neurosci.nature.com

scientific correspondence

ronal transport at non-saturating substrate concentrations. We also characterized EMT with 3H-MPP+ as its substrate, which is superior to endogenous amines for in vitro studies of the extraneuronal transporter8. Transport of MPP+ was sensitive to various structurally unrelated inhibitors of the extraneuronal monoamine transporter (Fig. 2c). The isocyanine disprocynium24 (Ki = 15 nM) and the steroid corticosterone (Ki = 120 nM), which inhibits extraneuronal transport by a direct, non-genomic mechanism, were the most potent inhibitors. The values for Ki for the inhibition of EMT-mediated MPP + uptake were almost identical to those of extraneuronal noradrenaline uptake into Caki-1 cells (Fig. 2d) and MPP+ uptake into human astrocytes8. Also, the rank order of antagonists perfectly fits data from the classical preparations for extraneuronal monoamine transport, perfused rat heart and isolated incubated rabbit aorta13. In control experiments, EMT was resistant to reserpine, a specific inhibitor of vesicular monoamine transporters, and to desipramine, a specific inhibitor of the neuronal noradrenaline transporter (Fig. 2e). In a limited survey of human tissue distribution by RT-PCR on total RNA from biopsy samples, the EMT mRNA was detected in liver, heart and brain cortex ( Fig. 3a). This corresponds well with the anticipated broad tissue distribution of the extraneuronal catecholamine transporter7 and supports the idea that uptake of catecholamines into human glia is mediated by the extraneuronal transporter8. Using fluorescence in situ hybridization, we have mapped the EMT gene to 6q27, the end of the long arm of chromosome 6 ( Fig. 3b ). In addition to putative tumor suppressor genes, susceptibility loci for type 1 diabetes15 have been genetically mapped in the human 6q27 region. Decreased hepatic inactivation of released catecholamines could account for enhanced glycogenolysis and thus contribute to elevated blood glucose levels. EMT acts as a widely distributed second line of defense, inactivating the fraction of released monoamine transmitters that escapes neuronal reuptake and thus preventing uncontrolled spreading of the signal. Pharmacological blockade of EMT considerably increases plasma catecholamine levels10. A chronically elevated sympathetic tone resulting from loss of EMT function may induce vasoconstriction and vascular hypertrophy, leading to progressive increases in peripheral resistance and elevated blood pressure. In the central nervous system, most of the aminergic synapses are closely surrounded by glia, which express considerable amounts of COMT and MAO. Thus, EMT may also represent a key for non-neuronal inactivation of catecholamines in the central nervous system and, thus, a new molecular target for the development of drugs that aim at an elevation of free monoamine transmitters. The cloning of EMT provides a new candidate not only for diseases based on autonomic dysfunction but also for neuropsychiatric disorders.

3. Brunner, H., Nelen, M., Breakefield, O., Ropers, H. & vanOost, B. Science 262, 578580 (1993). 4. Pacholczyk, T., Blakely, R. & Amara, S. Nature 350, 350354 (1991). 5. Borowski, B. & Hoffman, B. Int. Rev. Neurobiol. 38, 139-199 (1995). 6. Iversen, L. Br. J. Pharmacol. 25, 18-33 (1965). 7. Trendelenburg, U. in Handbook of Experimental Pharmacology Vol. 90 (eds Trendelenburg, U. & Weiner, N.) 279319 (Springer, New York, 1988). 8. Russ, H., Staudt, K., Martel, F., Gliese, M. & Schmig, E. Eur. J. Neurosci. 8, 12561264 (1996). 9. Karhunen, T., Tilgmann, C., Ulmanen, I., Julkunen, I. & Panula, P. J. Histochem. Cytochem. 42, 10791090 (1994). 10. Eisenhofer, G., McCarty, R., Pacak, K., Russ, H. & Schmig, E. NaunynSchmiedebergs Arch. Pharmacol. 354, 287294 (1996). 11. Grndemann, D. et al. J. Biol. Chem. 272, 1040810413 (1997). 12. Schmig, E. et al. FEBS Lett. 425, 7986 (1998). 13. Schmig, E. & Schnfeld, C.-L. Naunyn-Schmiedebergs Arch. Pharmacol. 341, 404410 (1990). 14. Iversen, L. Brit. J. Pharmacol. 24, 387394 (1965). 15. Luo, D.-F. et al. Am. J. Hum. Genet. 57, 911919 (1995).

1998 Nature America Inc. http://neurosci.nature.com

Development of language-specific phoneme representations in the infant brain


Marie Cheour1, Rita Ceponiene1, Anne Lehtokoski1, Aavo Luuk2, Jri Allik2, Kimmo Alho1,3 and Risto Ntnen1,3
1

Cognitive Brain Research Unit, Department of Psychology, University of Helsinki, FIN-00014, Finland Department of Psychology, University of Tartu, Estonia BioMag laboratory, P.O.Box 508 00029 HYKS, Finland Correspondence should be addressed to M.C. (Marie.Cheour@Helsinki.fi)

2 3

Acknowledgements
Our thanks to D. Keppler, J. Babin-Ebell and M. Gliese for tissue samples and to A. Ripperger and B. Wallenwein for technical assistance. This work was supported by grants from the Deutsche Forschungsgemeinschaft (Un34/19-1/B2 and SFB601/A4).

RECEIVED 18 JUNE: ACCEPTED 13 JULY 1998


1. Mark, A. J. Hypertension 14, 159165 (1996). 2. Schulteis, G. & Koob, G. Nature 371, 108109 (1994).

Studies using behavioral methods, such as head-turning experiments, in which children are conditioned to turn their heads toward the sound source when they detect a change in the sound, have shown that environment has an important effect on how infants perceive language14. Young infants are able to discriminate almost all phonetic contrasts, whereas older infants discriminate better between phonemes that occur in the language that they normally hear, rather than foreign-language phonemes. Here we demonstrate the development of language-specific memory traces in the brains of the same group of infants between six months and one year of age. The Estonian and Finnish languages, which are closely related to each other, have very similar vowel structures 5. For example, the vowels /e/ and //, which differ only in the second-formant (F2) frequency, exist in both languages. However, only Estonian has the vowel //, which is approximately between // and /o/ (Fig. 1). We used these phonemes to investigate the development of language-specific memory traces in infants. This was done by analyzing mismatch negativity (MMN), an electric brain response automatically elicited by infrequent changes (deviant stimulus) in a repetitive (standard) stimulus612. The MMN amplitude increases with an increasing acoustic difference between deviant and standard stimuli 9 . Thus, when the phoneme /e/ is the standard, the MMN amplitude should be larger when the deviant stimulus
351

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

scientific correspondence

2nd formant (Hz)

a
1998 Nature America Inc. http://neurosci.nature.com

Fig. 1. Stimuli used in the experiment. Vowels of 75 dB SPL, lasting 400 ms (with 10 ms rise and fall times), were presented through two loudspeakers about 50 cm from the infants ears. The (onset-to-onset) interstimulus interval was 1100 ms. The upward arrows show the locations of the vowel stimuli in F1F2 space. The standard stimulus was the Finnish and Estonian vowel /e/. The deviant stimulus // was a vowel shared by both languages; the deviant stimulus // is a vowel only in Estonian. The F2 frequency chosen for the frequent stimulus produced the best model of the /e/ prototype of the Finnish and Estonian languages as judged by adult listeners 6. For /e/ it was 1940 Hz; for //, 1533 Hz; and for //, 1311 Hz. The first (F1), third (F3), and fourth (F4) formants for each vowel were constant at 450, 2540, and 3500 Hz, respectively. The average fundamental frequency (F0) was 105 Hz. The vowels were generated by the production of synthetic stimuli from natural glotted excitation in conjunction with a vocal-tract model 6.

is // than when it is //. Nevertheless, the MMN amplitude in Finnish adults 6 was smaller for // than //, possibly because // is not a vowel in Finnish. Consistent with this, the MMN attenuation for // relative to // did not occur in Estonian adults, in whose language // is a vowel. Perhaps as // does not exist Finns 6 mo Finns 12 mo Estonians 12 mo in Finnish, Finns usually have very little experience with it, and therefore a memory trace for it has not been developed in their brain. We recorded the brain responses of nine Finnish (seven male and two female) and nine Estonian (six male and three female) healthy infants without hearing deficits, all from monolingual families, while the infants sat Fig. 2. The MMN amplitude at the central Cz electrode (grand-average, deviant-standard on a safety seat. To distract the chil- difference waveform, averaged across nine infants) reflects the development of languagedren from the stimuli and to minimize specific memory traces in Finnish infants. At six months of age, their MMN amplitude eye and body movements, an assistant reflects only the acoustical difference between the deviant and standard stimuli. In conentertained them with silent toys dur- trast, at one year of age, the MMN amplitude in the same children was considerably ing the experiment. The Finnish enhanced for the Finnish vowel // and considerably attenuated for the Estonian vowel //. infants were studied at six months of In Estonian one-year-old infants, the MMN amplitude reflected only the acoustic difference age (average, six months one day; between deviant and standard stimuli, as both deviant stimuli are vowels in Estonian. range, six months ten days) and standard /e/ - deviant //, a vowel shared by Finnish and Estonian languages, - - - standard again at one year of age (average, one /e/ - deviant //, an Estonian vowel.
352 nature neuroscience volume 1 no 5 september 1998

year three days; range, one year fourteen days). The Estonian infants were studied when they were one year old (average, one year eight days; range, one year fourteen days). Parents gave informed consent for their childs participation. The /e/ phoneme was the standard stimulus, and it was randomly replaced by deviant // or // 5 (Fig. 1), with a probability of occurrence of 0.1 for each. Electroencephalograms (EEGs), amplified by SynAmps at DC-100 Hz and digitized at 250 Hz, were recorded at frontal (Fz) and central (Cz) electrodes. Eye movements were monitored with electro-oculogram (EOG) electrodes attached below and at the lateral corner of the left eye. All electrodes were referred to a left-mastoid electrode. EEG epochs with EEG or EOG artifacts exceeding 250 V at any electrode were discarded, as were epochs for the first two stimuli of each block. For each infant, there were at least 90 acceptable EEG epochs for each deviant stimulus. Frequencies higher than 15 Hz or lower than 0.1 Hz were digitally filtered out off-line. A baseline of 0 V was set at the mean amplitude over the 50millisecond pre-stimulus period. The MMN amplitude was measured from the difference waves (obtained by subtracting the average standard-stimulus response from the average deviant-stimulus response) separately for each subject and electrode, as the mean amplitude, relative to the baseline, of the 80-ms period centered on the largest peak between 250450 ms. The differences in the MMN amplitude in response to the deviant // and // stimuli were tested with three-way ANOVA analysis: group (Estonian, Finnish oneyear-old, Finnish six-month-old) stimulus type (//, //) electrode (Fz, Cz). Considerable MMN responses were elicited by both deviant stimuli in all groups (Figs 2 and 3). In six-month-old Finnish infants, the MMN amplitude was larger, although not significantly, for // than for //, that is, larger for the acoustically more different stimulus. Thus, these six-month-old infants perceived the differences between these stimuli mainly or only acoustically. In contrast, at one year of age, the same Finnish infants had a much smaller MMN amplitude for the Estonian vowel // than for the Finnish vowel //, although the acoustical difference from /e/ was larger to // than to //. A threeway ANOVA showed a highly significant group (six- versus twelve-month-old Finns) stimulus type (// versus //)

1st formant (Hz)

1998 Nature America Inc. http://neurosci.nature.com

scientific correspondence

Fig. 3. The MMN peak MMN amplitude amplitude (at Cz) as a function of the deviant stimulus. Results are shown for three groups: six-month-old (squares, thick solid line) and one-year-old (circles, thin solid line) Finnish infants and one-year-old Estonian (triangles, dotted line) infants. These are arranged in the order of increasing F2 difference from the standard stimulus. , 6-month-old Finns; b, one-year-old Finns; --, one-year-old Estonians.

tude for the Estonian vowel // was significantly larger in the Estonian than in the Finnish one-year-old infants (p < 0.004). Our data indicate that language-dependent memory traces in the human brain emerge before the age of 12 months. To our knowledge, this is the first definite neurophysiological evidence for the development of brain memory traces for speech sounds in infants. Our results show that by the age of one year, the MMN amplitude increases for native phonemes and decreases for non-native phonemes. Thus, during their cognitive development, the ability of infants to discriminate native speech sounds improves, while at the same time they seem to lose some of their ability to discriminate non-native speech sounds14,1315.

Acknowledgements
This work was supported by the Academy of Finland. We thank P. Alku for producing the stimuli and Sanna Kurjenluoma, Nina Penttinen and Marieke Saher for data collection.

1998 Nature America Inc. http://neurosci.nature.com

interaction (F(1,16) = 16.23; p < 0.001). The least significant difference post-hoc test showed that the MMN amplitude for the Finnish vowel // was increased ( p < 0.01) and for the Estonian vowel // was decreased (p < 0.02) for Finnish infants between six and twelve months of age. Therefore, it seems that the memory trace for the native vowel // was formed before the age of one year in these Finnish children. According to these results, the MMN amplitude for // should not be smaller than that for // in Estonian one-yearold infants, as both are vowels in Estonian. Indeed, there was no distinct difference in the MMN amplitude between these two vowels; if anything, the MMN amplitude for // was larger than that for //, apparently because of the greater acoustical difference from // to /e/ than from // to /e/. Moreover, a comparison of one-year-old Finnish and Estonian infants yielded a highly significant group stimulus type interaction (F (1,16) = 14.88; p < 0.002). The least significant difference post-hoc test showed that this resulted from the MMN amplitude for // being larger than that for // in the one-year-old Finnish infants ( p < 0.002). Furthermore, the MMN ampli-

RECEIVED 5 JUNE: ACCEPTED 23 JULY 1998


1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. Kuhl, P. K., Williams, K. A. & Lacerda, F. Science 255, 606608 (1992). Werker, J. F. & Tees, R. C. Infant Behav. Dev. 7, 4963 (1984). Werker, J. F.& Lalonde, C. E. Dev. Psychol. 24, 672683 (1988). Best, T. C., McRoberts, G. W. & Sithole, N. M. J. Exp. Psychol. Hum. Percept. Perform. 14, 345360 (1988). Iivonen, A. Phonetica 52, 221224 (1995). Ntnen, R. et al. Nature 385, 432434 (1997). Ntnen, R., Gaillard, A. W. K. & Mntysalo, S. Acta Psychol. 4, 313329 (1978). Ntnen, R. in Attention and Brain Function (Lawrence Erlbaum, Hillsdale, New Jersey, 1992). Tiitinen, H., May, P., Reinikainen, K. & Ntnen, R. Nature 372, 9092 (1994). Kraus, N. et al. Science 273, 971973 (1996). Cheour, M. et al. Psychophysiology 33, 478481 (1996). Cheour, M. et al. Neuroreport 8, 17851787 (1997). Polka, L. & Werker, J. F. J. Exp. Psychol. Hum. Percept. Perform. 20, 421435 (1994). Strange, W. & Dittman, S., Percept. Psychophys. 36, 131145 (1984). Werker, J. F. in The Development of Speech Perception (Bradford Book, Massachusetts, 1994).

nature neuroscience volume 1 no 5 september 1998

353

1998 Nature America Inc. http://neurosci.nature.com

review

Genetic dissection of Alzheimers disease and related dementias: amyloid and its relationship to tau
John Hardy, Karen Duff, Katrina Gwinn Hardy, Jordi Perez-Tur and Mike Hutton
Neurogenetics and Transgenics Laboratories, Mayo Clinic Jacksonville, Jacksonville, Florida 32224, USA Correspondence should be addressed to J.H. (hardy.john@mayo.edu)

1998 Nature America Inc. http://neurosci.nature.com

Molecular genetic analysis is revealing the etiologies of Alzheimers disease (AD) and related dementias. Here we review genetic and molecular biological evidence suggesting that the peptide A42 is central to the etiology of AD. Recent data also suggests that dysfunction in the cytoskeletal protein tau is on the pathway that leads to neurodegeneration and dementia. Tau is produced either indirectly, by A42, or directly, in some forms of frontotemporal dementia by mutations in tau itself. These data support and refine the amyloid cascade hypothesis for AD and suggest that understanding the causes and consquences of tau dysfunction is an important priority for dementia research.

The pathology of Alzheimers disease (AD), the most common cause of dementia, is characterized by primarily extracellular plaques and intracellular neurofibrillary tangles. Plaques are composed mainly of the amyloid- peptide (A), whereas tangles are composed mainly of the cytoskeletal protein, tau. The relationship between these lesions and the disease process has long been debated. However, the currently dominant theory of AD etiology and pathogenesis is the amyloid cascade hypothesis13, which states that overproduction of A , or failure to clear this peptide, leads to AD primarily through amyloid deposition, which produces neurofibrillary tangles; these lesions then are associated with cell death, which is reflected in memory impairment, the hallmark of this dementia. Over the last eight years, the amyloid cascade hypothesis has gained strength through the observation that AD-causing mutations identified in amyloid- precursor protein (APP)4 and the presenilin5,6 genes alter APP metabolism such that more of the insoluble A peptide 7 (A 42/43) is produced 816 . This A42/43 peptide also forms amyloid fibrils more readily than the more prevalent A40, and as a result, amyloid fibrils are then selectively deposited in the neuritic plaque1719. Although the amyloid cascade hypothesis of AD pathogenesis has not been seriously challenged by alternative theories, it has been vigorously attacked for several potential shortcomings discussed below. In addition, the mechanism by which neuritic plaques relate to neurofibrillary tangles remains elusive, as does our understanding of whether the latter is an important intermediate in cell death. Recently, several important observations have indicated that the amyloid cascade hypothesis needs to be recast (Fig. 1). These findings suggest that tau-containing neurofibrillary tangles are involved in several other neurodegenerative diseases, even when those tangles are induced by factors other than A, in contrast to AD. The purpose of this review is to highlight these observations and to attempt a greater synthesis of the pathogenic processes that lead to neurodegeneration, not only in AD but also in other diseases.
nature neuroscience volume 1 no 5 september 1998

DO NEURITIC PLAQUES CAUSE NEURODEGENERATION?


An implicit assumption of the cascade hypothesis is that the formation of neuritic amyloid plaques initiates neurodegeneration. Historically, there have been two criticisms of this view: first, that the degree of dementia does not correlate with the number of plaques20, and second, that neurofibrillary tangle formation seems to predate plaque formation21. In addition, transgenic mice that overexpress APP develop little if any neurodegeneration22, even with extensive amyloid deposition23, and they also show behavioral changes before the onset of this deposition2426. Finally, an atypical AD family with a presenilin-1 mutation27 has been described in which few neuritic plaques are found in the brains of affected individuals28. This family has a presenilin-1 mutation that results in the loss of exon 9 from the affected mRNA transcripts. The phenotype is unusual in that it begins with spastic paraparesis, a clinical syndrome in which lower extremity spacticity occurs insidiously and progressively, with no associated weakness, pain or other symptoms. This spastic paraparesis is probably secondary to the neuropathologically observed corticospinal tract degeneration, and it occurs clinically before any marked cognitive changes. The amyloid plaques in affected individuals of this family are large and diffuse, with an appearance reminiscent of cotton wool balls. These plaques have neither abnormal neurites nor complement cascade activation to suggest an inflammatory response. The description of this family is particularly significant because it suggests that Alzheimers disease can occur not only in the absence of neuritic plaques but also in the absence of an inflammatory response to those plaques. These arguments that the formation of neuritic plaques is not the crucial pathogenic process in AD have been countered by several possible explanations. First, a lack of correlation between amyloid burden and degree of dementia may not take into account the possibility that plaques are dynamic structures that can be broken down29. Second, it is possible that transgenic rodent brains simply do not replicate the neurofibrillary changes typical of human AD. Finally, it may be unwise to dismiss the importance of neuritic plaques based on their absence in a small number of families with a specific presenilin-1 mutation. Nonetheless, it is widely believed that neu355

1998 Nature America Inc. http://neurosci.nature.com

review

Down Syndrome

1998 Nature America Inc. http://neurosci.nature.com

Fig. 1. The amyloid cascade hypothesis, showing the proposed relationships between A and tau and between Alzheimers disease and FTDP-17. The link between A42 overproduction and tau dysfunction is presently uncertain and is therefore represented by a question mark. In addition, it is unclear whether tau dysfunction leads directly to cell death or if the formation of neurofibrillary tangles are a necessary intermediate.

These clinical data suggest that A and APOE interact, although why APOE does not influence age of onset in families with presenilin mutations40,41 remains an enigma. A clear experimental demonstration of a biochemical connection between A and APOE has come from the pathological analysis of a cross between an APOE knockout mouse and a transgenic, amyloid-depositing mouse42. When this mouse is depleted of APOE, amyloid deposits accumulate much more slowly. Whether the influence of APOE is on APP/A generation, accumulation or clearance remains obscure. However, it is clear that the control of APOE expression is very likely to be important in Alzheimers pathogenesis. Consistent with this, AD risk is modified by genetic variability in promoter elements, which determine APOE expression levels4346. These findings led to the suggestion that the protein isoform of APOE is not the only determinant of the pathogenesis of AD at the APOE locus; the amount of neural expression is also important.
FTDP-17 AND TAU MUTATIONS

ritic plaques may not be central to the disease process, but may rather represent the end point of a pathological process in which they are simply the clearing stations of accumulated peptide30. This hypothesis is supported by observations in the primate brain that A amyloid fibrils can induce neurodegeneration and abnormal tau phosphorylation, a likely first step in the formation of tangles. Notably, this neurodegeneration occurs despite the absence of neuritic plaques. Another conclusion suggested by these studies is that there may be species differences in neuronal sensitivity to amyloid, because similar A injections in rat brains produced neither neurodegeneration nor tau hyperphosphorylation. This species difference could explain the absence of neurodegeneration, despite extensive amyloid deposition, in the brains of transgenic mice in which human APP is overexpressed. It is of crucial importance to determine where and how A is toxic, because this toxicity is an obvious potential target for pharmacotherapy. All the pathogenic mutations in APP and the presenilins documented to date lead to elevated levels of A42, suggesting that this is the toxic moiety. Thus, specific manipulation of APP metabolism is a credible target for drug discovery programs. Considerable A42 is produced intracellularly31,32, and intracellular accumulation, rather than extracellular deposition, of amyloid fibrils composed of A may be critical to the pathogenic process33,12. Unfortunately, this is the area of research where the least data is available, in part because of the lack of cell loss or overt dysfunction in amyloid-depositing transgenic mice, which complicates the use of this otherwise valuable model.

A major current debate in AD research regards the relative importance of tau and amyloid to the pathogenesis of the disease. The normal function of the tau protein is to bind and thus maintain the structure of neuronal microtubules. The tau gene (on chr17q21) contains a total of 15 exons with the major tau protein isoforms being encoded by 11 of them47. The gene undergoes alternative splicing of exons 2, 3 and 10 to generate a total of six different tau mRNAs that encode proteins of 352441 amino acids48 (Fig. 2). Exons 913 encode four microtubule-binding motifs that are imperfect repeats of 31 or 32 residues. The alternative splicing of exon 10 thus generates tau protein with three or four microtubule-binding repeats49. Mutations in the tau gene are now known to cause frontal temporal dementia with Parkinsonism linked to chromosome 17 (FTDP-17)5052. Clinically this syndrome resembles AD, although the frontal-lobe abnormalities that predominate are clearly distinct from the typical AD phenotype. Historically many cases of this autosomal dominant disease would have been diagnosed as having Picks disease. In the majority of cases with FTDP-17, frontal lobe dysfunction is responsible for the earliest detected symptoms, such as behavioral and personality changes (disinhibition, apathy). These are followed by progressive cognitive decline. In contrast, behavioral changes in AD occur relatively late in the course of the disease. Pathologically FTDP-17 is marked by pronounced atrophy of the frontal and temporal lobes with neuronal cell loss, glial cell proliferation and activation (gliosis), and spongiform changes in the superficial cortical layers. Ballooned (swollen) neurons are also present in the brains of affected individuals from several separate families. In addition, affected individuals in families with FTDP-17 have abnormal
Tau protein isoforms with 3 microtubule binding repeats (exon 10-)

DO APOE AND THE A PATHWAY INTERACT?


Another criticism of the amyloid cascade hypothesis has been that there is no known biochemical association between apolipoprotein E (APOE)34 and APP metabolism or A deposition, even though genetic variability in APOE is a major risk factor for AD3436. Furthermore, APOE genotype influences age of AD onset in patients with Down syndrome (i.e. three copies of the chromosome-21-encoded APP gene)37 and in those with APP mutations38,39.
356

Tau protein isoforms with 4 microtubule binding repeats (exon 10+)

Fig. 2. The six major tau protein isoforms that are generated by alternative splicing of exons 2, 3 and 10. Exons 912 encode the four microtubule binding repeats (indicated by black boxes). Alternative splicing of exon 10 leads to the generation of tau protein isoforms with four or three repeats.
nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

review

intraneuronal accumulations of hyperphosphorylated tau protein. However, there is significant variability in the severity, distribution and morphology of the tau accumulations in FTDP-17.

HOW DO TAU MUTATIONS RELATE TO TANGLES?


Hyperphosphorylated tau is the major component of the neurofibrillary tangles found in the brains of AD patients. In some FTDP17 families, the tangles are composed of tau in paired helical filaments similar to those found in AD53,54. Such accumulations contain all six major tau protein isoforms (Table 1) and are restricted to neurons53,54. However, in other FTDP-17 families, the accumulations have a very different morphology and are composed of tau filaments that resemble twisted ribbons with a longer periodicity than that found in the AD-like tangles53. In these latter FTDP17 families, the accumulations consist almost entirely of four-repeat tau isoforms, and they are found in both neurons and glia. This variability in the pathology of FTDP-17 is caused by the variety of mutations in the tau gene that correlate with distinct pathological findings in the different FTDP-17 families. The most common mutations in FTDP-17 occur in the 5 splice site of exon 10 and affect the alternate splicing of this exon (Fig. 3)51,52. In the normal human brain, this exon is skipped in most tau mRNA transcripts, leading to production of tau protein with three microtubule-binding repeats. However, in individuals with splicesite mutations, increased splicing-in of exon 10 increases production of the four-repeat tau protein 51,52. Overproduction of four-repeat tau in adults causes the characteristic tau accumulation that is observed in the brains of affected patients with these splice site mutations, with the longer twisted-ribbon filaments rather than the AD-like paired helical filaments53. Similar four-repeat, predominantly twisted-ribbon accumulations are also observed in families with a missense mutation in exon 10. In families with this mutation, the level of tau RNA containing exon 10 is unaltered51, but the deposited protein is the mutant four-repeat isoform. The location of this mutation suggests that it is likely to disrupt microtubule binding in the four-repeat tau isoforms, although three-repeat tau isoforms are unafffected. The final group of FTDP-17 families have missense mutations outside of exon 10 (refs 50, 51), which therefore affect both three-repeat and four-repeat tau. In this last group of families, the neurofibrillary tangles consist of paired helical filaments and contain all six tau isoforms53,54 (Table 1). In other words, the splice mutations and missense mutations occurring in exon 10 lead to four-repeat tau deposits and twisted ribbons, and mutations in other regions retain normal splicing, leading to the deposition of both three- and four-repeat tau, and to the formation of paired helical filaments resembling those found in AD. Taken together, these mutations suggest that the spectrum of tauopathies referred to as FTDP-17 results from disruption of tau microtubule binding. It is presently not clear if this disruption causes cell death

Fig. 3. Diagram of the 3 end of exon 10 of the tau gene, showing the pathogenic mutations in this region and how they affect (and destabilize) the stem-loop structure, possibly preventing the splicing out of exon 10 (modified from ref. 51).

directly by causing the degeneration of the microtubules or if it simply leads to an increase in unbound tau, which in turn leads to the formation of filaments and eventually tangles that are the actual cause of neurodegeneration. These findings are remarkable for two reasons. First, they show that tau dysfunction can lead to neuronal damage and death, resulting in pathological and clinical phenotype with similarities to AD55. Second, they show that amyloid deposition is not an inevitable consequence of tau dysfunction. These findings are completely consistent with the amyloid cascade hypothesis (Fig. 1). There are of course significant differences between the clinical and the pathological phenotypes of FTDP-17 and AD. Perhaps most interesting is the different distribution of the pathology in the two diseases. That is, in FTDP-17, as is clear from the name, the frontal and temporal regions of the brain are disproportionately affected, whereas in AD the frontoparietal regions are particularly vulnerable to degeneration. This difference in distribution would seem most likely to reflect the difference in the initial cause of the neurodegeneration in each disease (A in AD, tau dysfunction in FTDP-17). Thus in AD we anticipate that the distribution of the pathology will be a function of two factors: local concentration of A in each brain region and sensitivity of the neurons to damage induced by A and dependent on tau dysfunction. In contrast, we expect the pathology in FTDP-17 to reflect the sensitivity of neurons only to the tau dysfunction induced by the specific mutation in each family. It is perhaps not surprising then that FTDP-17 displays a wide range of clinical and pathological phenotypes, given that the different causative mutations can have such different effects on the tau molecule. Neuronal sensitivity to tau dysfunction may also underlie the distribution of the pathology observed in other tauopathies such as Picks disease and progressive supranuclear palsy (PSP). In Picks disease, the tau accumulations known as Pick bodies consist almost entirely of three-repeat tau isoforms56, for reasons that are not presently clear. One brain region that is dramatically affected in Picks disease is the granule cell layer of the hippocampus. Interestingly, Table 1. Tau mutations relative to FTDP-17 pathology these neurons have previously been shown to generate only three-repeat Mutation Soluble Tau Tau inclusions Tau filaments tau, consistent with the presence of Missense mutation Ratio of four-repeat to All six isoforms AD-like paired three-repeat accumulations (Pick outside exon 10 three-repeat unaltered included helical filaments bodies) in these cells. Thus, Missense mutation Ratio of four-repeat to Predominantly Unknown although it is clear that the distrib1 in exon 10 three-repeat unaltered four-repeat ution of the pathology (tangles, cell Exon 10, Increased Predominantly Twisted-ribbon death, gliosis) in AD and FTDP-17 5 splice site mutation four-repeat four-repeat filaments is different, it seems likely that this reflects a difference in the patho1Inferred from unchanged ratio of RNA with and without exon 10. genic agent in each disease and not Proposed relationship between site of mutation in tau gene and type of pathogenic tau mutation, based on refs 50 necessarily a difference in the and 51.
nature neuroscience volume 1 no 5 september 1998 357

1998 Nature America Inc. http://neurosci.nature.com

1998 Nature America Inc. http://neurosci.nature.com

review

downstream pathways that lead to neurodegeneration.

Conclusion In conclusion, these results suggest that A is the primary culprit in AD, but they also suggest that tau dysfunction is directly on the route to neuronal degeneration. However, the relationship between A and tau in AD remains unclear and will doubtless become a major focus for future research. It may be that both neuritic plaques and neurofibrillary tangles are merely markers of the dysfunction of their constituent molecules rather than pathogenic in themselves. From a therapeutic point of view, targeting A production makes sense for AD; however, these results suggest that targeting tau dysfunction would have the possible advantage that treatments based on this latter strategy would be appropriate for the myriad of other tauopathies besides AD. Because clinically the distinction can be difficult to make, especially early in the course of these illnesses, therapeutic strategies that are broad-based will be of great value.
1998 Nature America Inc. http://neurosci.nature.com

23. 24. 25. 26. 27. 28. 29. 30. 31. 32.

Acknowledgements
Supported by the Mayo Foundation and by an NIA Program Project Grant and an NINDS Project Grant (M.H.).

33. 34.

RECEIVED 29 MAY: ACCEPTED 31 JULY 1998


35. 1. Hardy, J. & Allsop, D. Amyloid deposition as the central event in the aetiology of Alzheimers disease. Trends Pharmacol. Sci. 12, 383388 (1991). 2. Selkoe, D. J. Amyloid -protein and the genetics of Alzheimers disease. J. Biol. Chem. 271, 1829518298 (1996). 3. Hardy, J. Amyloid, the presenilins and Alzheimers disease. Trends Neurosci. 20, 154159 (1997). 4. Goate, A. et al. Segregation of a missense mutation in the amyloid precursor protein gene with familial Alzheimers disease. Nature 349, 704706 (1991). 5. Sherrington, R. et al. Cloning of a gene bearing missense mutations in early-onset familial Alzheimers disease. Nature 375, 754760 (1995). 6. Levy-Lahad, E. et al. Candidate gene for the chromosome 1 familial Alzheimers disease locus. Science 269, 973977 (1995). 7. Jarrett, J. T. et al. The carboxy terminus of the amyloid protein is critical for the seeding of amyloid formation: implications for the pathogenesis of Alzheimers disease. Biochemistry 32, 46934697 (1993). 8. Citron, M. et al. Mutation of the -amyloid precursor protein in familial Alzheimers disease increases -protein production. Nature 360, 672674 (1992). 9. Cai, X. D. et al. Release of excess amyloid -protein from a mutant amyloid protein precursor. Science 259, 514516 (1993). 10. Suzuki, N. et al. An increased percentage of long amyloid -protein secreted by familial amyloid -protein precursor (-APP717) mutants. Science 264, 13361340 (1994). 11. Scheuner, D. et al. Secreted amyloid -protein similar to that in the senile plaques of Alzheimers disease is increased in vivo by the presenilin 1 and 2 and APP mutations linked to familial Alzheimers disease. Nat. Med. 2, 864870 (1996). 12. Duff, K. et al. Increased amyloid-42(43) in brains of mice expressing mutant presenilin 1. Nature 383, 710713 (1996). 13. Borchelt, D. R. et al. Familial Alzheimers disease-linked presenilin 1 variants elevate A1-42/1-40 ratio in vitro and in vivo. Neuron 17, 10051013 (1996). 14. Citron, M. et al. Mutant presenilins of Alzheimers disease increase production of 42-residue amyloid -protein in both transfected cells and transgenic mice. Nat. Med. 3, 6772 (1997). 15. Xia, W. et al. Enhanced production and oligomerization of the 42-residue amyloid -protein by Chinese hamster ovary cells stably expressing mutant presenilins. J. Biol. Chem. 272, 79777982 (1997). 16. Mehta, N. D. et al. Increased A42(43) from cell lines expressing presenilin 1 mutations. Ann. Neurol. 43, 256258 (1998). 17. Iwatsubo, T. et al. Visualization of A42(43) and A40 in senile plaques with endspecific A monoclonals: evidence that an initially deposited species is A42(43). Neuron 13, 4553 (1994). 18. Mann D. M. et al. Amyloid- protein (A) deposition in chromosome 14-linked Alzheimers disease: predominance of A42(43). Ann. Neurol. 40, 149156 (1996). 19. Lemere, C. A. et al. The E280A presenilin 1 Alzheimer mutation produces increased A42 deposition and severe cerebellar pathology. Nat. Med. 2, 11461150 (1996). 20. Terry, R. D. The pathogenesis of Alzheimer disease: an alternative to the amyloid hypothesis. J. Neuropathol. Exp. Neurol. 55, 10231025 (1996). 21. Braak, H. et al. Age, neurofibrillary changes, A-amyloid and the onset of Alzheimers disease. Neurosci. Lett. 210, 8790 (1996). 22. Irizarry, M. C. et al. A deposition is associated with neuropil changes, but not 36. 37. 38. 39. 40. 41. 42. 43. 44. 45. 46. 47. 48.

49. 50. 51. 52. 53. 54. 55. 56.

with overt neuronal loss in the human amyloid precursor protein V717F (PDAPP) transgenic mouse. J. Neurosci. 17, 70537059 (1997). Games, D. et al. Alzheimer-type neuropathology in transgenic mice overexpressing V717F -amyloid precursor protein. Nature 373, 523527 (1995). Hsiao, K. et al. Correlative memory deficits, A elevation, and amyloid plaques in transgenic mice. Science 274, 99102 (1996). Borchelt, D. R. et al. Accelerated amyloid deposition in the brains of transgenic mice coexpressing mutant presenilin 1 and amyloid precursor proteins. Neuron 19, 939945 (1997). Holcomb, L. et al. Accelerated Alzheimer-type phenotype in transgenic mice carrying both mutant amyloid precursor protein and presenilin 1 transgenes. Nat. Med. 4, 97100 (1998). Perez-Tur, J. et al. A mutation in Alzheimers disease destroying a splice acceptor site in the presenilin-1 gene. Neuroreport 7, 297301 (1995). Crook, R. et al. A variant of Alzheimers disease with spastic paraparesis and unusual plaques due to deletion of exon 9 of presenilin 1. Nat. Med. 4, 452455 (1998). Cruz, L. et al. Aggregation and disaggregation of senile plaques in Alzheimer disease. Proc. Natl Acad. Sci. USA 94, 76127616 (1997). Kuo, Y. M. et al. Water-soluble A (N-40, N-42) oligomers in normal and Alzheimer disease brains. J. Biol. Chem. 271, 40774081 (1996). Wild-Bode, C. et al. Intracellular generation and accumulation of amyloid betapeptide terminating at amino acid 42. J. Biol. Chem. 272, 1608516088 (1997). Hartmann, T. et al. Distinct sites of intracellular production for Alzheimers disease A40/42 amyloid peptides. Nat. Med. 3, 10161020 (1997). Yang, A. J. et al. Intracellular A1-42 aggregates stimulate the accumulation of stable, insoluble amyloidogenic fragments of the amyloid precursor protein in transfected cells. J. Biol. Chem. 270, 1478614792 (1995). Roses, A. D. Apolipoprotein E affects the rate of Alzheimer disease expression: amyloid burden is a secondary consequence dependent on APOE genotype and duration of disease. J Neuropathol. Exp. Neurol. 53, 429437 (1994). Corder, E. H. et al. Gene dose of apolipoprotein E type 4 allele and the risk of Alzheimers disease in late onset families. Science 261, 921923 (1993). Roses, A. D. Alzheimer diseases: a model of gene mutations and susceptibility polymorphisms for complex psychiatric diseases. Am. J. Med. Genet. 81, 4957 (1998). Royston, M. C. et al. Apolipoprotein E epsilon 2 allele promotes longevity and protects patients with Downs syndrome from dementia. Neuroreport 5, 25832585 (1994). Houlden, H. et al. APOE genotype and Alzheimers disease. Lancet 342, 737738 (1993). Sorbi, S. et al. Epistatic effect of APP717 mutation and apolipoprotein E genotype in familial Alzheimers disease. Ann. Neurol. 38, 124127 (1995). Van Broeckhoven, C. et al. APOE genotype does not modulate age of onset in families with chromosome 14 encoded Alzheimers disease. Neurosci. Lett. 169, 179180 (1994). Lendon, C. L. et al. E280A PS-1 mutation causes Alzheimers disease but age of onset is not modified by ApoE alleles. Hum. Mutat. 10, 186195 (1997). Bales, K. R. et al. Lack of apolipoprotein E dramatically reduces amyloid betapeptide deposition. Nat. Genet. 17, 263264 (1997). Lambert, J. C. et al. A new polymorphism in the APOE promoter associated with risk of developing Alzheimers disease. Hum. Mol. Genet. 7, 533540 (1998). Lambert, J. C. et al. Distortion of allelic expression of apolipoprotein E in Alzheimers disease. Hum. Mol. Genet. 6, 21512154 (1997). Artiga, M. J. et al. Allelic polymorphisms in the transcriptional regulatory region of apolipoprotein E gene. FEBS Lett. 421, 105108 (1998). Bullido, M. J. et al. A polymorphism in the regulatory region of APOE associated with risk for Alzheimers dementia. Nat. Genet. 18, 6971 (1998). Andreadis, A., Brown, W. M. & Kosik, K. S. Structure and novel exons of the human tau gene. Biochemistry 31, 1062610633 (1992). Goedert, M., Spillantini, M. G., Jakes, R., Rutherford, D. & Crowther, R. A. Multiple isoforms of human microtubule-associated protein tau: sequence and localization in neurofibrillary tangles of Alzheimers disease. Neuron 3, 519526 (1989). Goedert, M. et al. Assembly of microtubule-associated protein tau into Alzheimer-like filaments induced by sulphated glycosaminoglycans. Nature 383, 550553 (1996). Poorkaj, P. et al. Tau is a candidate gene for chromosome 17 frontotemporal dementia. Ann. Neurol. 43, 815826 (1998). Hutton, M. et al. Coding and 5 splice site mutations in tau associated with inherited dementia (FTDP-17). Nature 393, 702705 (1998). Spillantini, M. G. et al. Mutation in the tau gene in familial multiple system tauopathy with presenile dementia. Proc. Natl Acad. Sci. USA 95, 77377741 (1998). Spillantini, M. G. et al. Frontotemporal dementia and Parkinsonism linked to chromosome 17: a new group of Tauopathies. Brain Pathol. 8, 387402 (1998). Reed, L. A. et al. Autosomal dominant dementia with widespread neurofibrillary tangles. Ann. Neurol. 42, 564572 (1997). Foster, N. L. et al. Frontotemporal dementia and parkinsonism linked to chromosome 17: a consensus conference. Ann. Neurol. 41, 706715 (1997). Delacourte, A. et al. Vulnerable neuronal subsets in Alzheimers and Picks disease are distinguished by their distribution and phosphorylation. Ann. Neurol. 43, 193204 (1998).

358

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

An unusual cGMP pathway underlying depolarizing light response of the vertebrate parietal-eye photoreceptor
Wei-Hong Xiong1,2, Eduardo C. Solessio3 and King-Wai Yau1,2,4
1

Howard Hughes Medical Institute and the Departments of 2Neuroscience and 4Ophthalmology, Johns Hopkins University School of Medicine, Baltimore, Maryland 21205, USA John A. Moran Eye Center, University of Utah Health Science Center, Salt Lake City, Utah 84132, USA Correspondence should be addressed to K.-W.Y. (kwyau@welchlink.welch.jhu.edu)

1998 Nature America Inc. http://neurosci.nature.com

All cellular signaling pathways currently known to elevate cGMP involve the activation of a guanylyl cyclase to synthesize cGMP. Here we describe an exception to this rule. In the vertebrate parietal eye, the photoreceptors depolarize to light under dark-adapted conditions, unlike rods and cones but like most invertebrate photoreceptors. We report that the signaling pathway for this response involves a rise in intracellular cGMP resulting from an inhibition of the phosphodiesterase that hydrolyzes cGMP. Furthermore, this phosphodiesterase is driven by an active G protein in darkness. These results indicate an antagonistic control of the phosphodiesterase by two G proteins, analogous to the Gs/Gi control of adenylyl cyclase. Our findings demonstrate an unusual phototransduction mechanism and at the same time indicate that signaling involving cyclic nucleotides is more elaborate than previously known.

In vertebrates, the rods and cones of the lateral eyes hyperpolarize in response to illumination. This response is generated by a G-protein-coupled signaling pathway that stimulates a cGMPphosphodiesterase and thus the hydrolysis of cGMP14. In darkness, cytoplasmic cGMP binds to and opens cGMP-activated, nonselective cation channels5 in the outer-segment membrane of the cells, sustaining an inward dark current and keeping the cell depolarized. In light, the decrease in cGMP concentration closes these channels and consequently produces a membrane hyperpolarization as the electrical response. In lizards and some other lower vertebrates, there is a parietal (third) eye6 at the top of the head. This eye, like the lateral eyes, has a cornea, a lens and a retina, but this retina has only photoreceptors and ganglion cells, lacking bipolar, horizontal and amacrine cells. The outer segments of these photoreceptors resemble the cone outer segment in morphology, in that they have many membranous discs in continuity with the plasma membrane6; by analogy to cones, this is presumably where phototransduction takes place. Under dark-adapted conditions, the parietal-eye photoreceptor depolarizes to light7, opposite to the hyperpolarizing light responses of rods and cones. Depolarizing light responses are characteristic of most invertebrate photoreceptors and are thought to involve a phosphoinositide signaling pathway810. Thus the question arises as to whether the parietal-eye photoreceptor uses a cGMP or a phosphoinositide signaling cascade for phototransduction. We have previously found, using excised patches of plasma membrane, that a cGMP-gated cation channel is present selectively on the outer segment of this photoreceptor 11 . However, these experiments provided no direct information about the phototransduction mechanism. Here we have addressed this question by recording from intact, dissociated photoreceptors. Our findings demonstrate that although the
nature neuroscience volume 1 no 5 september 1998

depolarizing response of these cells is generated by a cGMP signaling pathway, the nature of this pathway is very unusual.

Results Whole-cell recording and perforated-patch recording with nystatin12 were made from the cell bodies of dark-adapted, single dissociated parietal-eye photoreceptors of the side-blotched lizard. In the whole-cell recording configuration, dialysis of 50 M cGMP from the pipet into a cell invariably induced little or no membrane current in darkness (Fig. 1a). However, at higher cGMP concentrations, progressively larger inward currents were observed, reaching a maximum of 170270 pA at 45 mV (mean standard deviation, 210 37 pA at 1 mM cGMP; n = 8 cells; Fig. 1a and d, upper panel). The variation in the saturated current among experiments might be due to a variation in outer-segment length produced by the dissociation procedure. The hydrolysis-resistant analog 8-bromo-cGMP induced similar maximal currents but was much more effective at lower concentrations, producing a considerable inward current at concentrations as low as 5 M (Fig. 1b and d, upper panel). Although 1 mM IP3 or 5 mM calcium dialyzed in the same manner was also able to elicit a small current, this current was outward, opposite to that induced by light7 (Fig. 1c and d, upper panel). For photoreceptors that had lost the outer segment during dissociation, dialysis with as much as 1 mM cGMP produced negligible inward current, whereas dialysis with 1 mM IP3 or 5 mM calcium induced similar outward currents as before (Fig. 1d, lower panel). Dialysis of cAMP (2 mM) or 8-bromo-cAMP (5 mM) into intact photoreceptors produced no detectable current. Finally, dialysis of cAMP with cGMP also had little influence on the cGMP-induced current (data not shown). Hence, among the common second messengers that
359

1998 Nature America Inc. http://neurosci.nature.com

articles

a d With O.S. we have tested, only cGMP is able to mimic light by inducing an inward current. To determine whether the opening of cGMP-activated channels11 underlies the light response, we used two chemicals known to block cGMPs activated channels, L-cis b diltiazem13 and dichlorobenzamil14. The inward current induced by exogenous cGMP dialyzed from a pipet in the whole-cell configuration, and that by a light flash in perforated-patch recordings, were Without O.S. both transiently suppressed s c by a puff of 0.1 mM L-cis-diltiazem directed at the recorded cells ( Fig. 2a and b ). Similar results were obtained with dichlorobenzamil (data s not shown). Furthermore, voltage-ramp experiments indicated that both currents had identical currentvoltage s relations, with characteristic 5 outward rectification (Fig. Fig. 1. Whole-cell dialysis experiments on dissociated parietal-eye photoreceptors. (ac) Membrane cur2c). Thus, the depolarizing rents induced by whole-cell dialysis of cGMP (a), 8-bromo-cGMP (b), IP3 or Ca2+(c) in darkness. Wholelight response indeed arises cell recording began at time zero, with rupture of the membrane patch underneath the gigaseal patch from the opening of cGMP- electrode. Each trace represents a separate experiment. Indicated concentrations corresponded to those activated channels, and there- in the whole-cell pipet. (d) Upper panel, collected results from experiments (ac); lower panel, collected fore light must elevate cGMP results from similar experiments on cells lacking the outer segment (O.S.). Means and standard deviations of the currents are shown. Numbers in parentheses indicate number of experiments. Membrane voltage in the outer segment. If a basal metabolic flux was clamped at 45 mV. of cGMP exists in the photoreceptor in darkness, it should be possible to elevate maximal IBMX-induced current (36.3 15.5 pA; n = 4 experintracellular cGMP with a phosphodiesterase inhibitor, such as iments) was considerably smaller than the maximal current 3-isobutyl-1-methyl-xanthine (IBMX). A brief, strong puff of induced by cGMP dialyzed from a pipet (about 200 pA, see this highly membrane-permeant chemical (1 mM) ejected from Fig. 1). To confirm this difference, we measured both currents a pipet at a cell under perforated-patch recording indeed elicited from the same cell with successive perforated-patch and wholea substantial transient inward current (Fig. 3a, upper panel); cell recordings by including both nystatin and cGMP (50 M weaker puffs gave smaller currents (not shown). This IBMX8-bromo-cGMP, which is sufficient to saturate the cGMPinduced current resembled in time course the response of the induced current; Fig. 1b and d) in the patch-pipet solution. The same cell to a bright flash (Fig. 3a, lower panel); the slower rise nystatin allowed initial perforated-patch recording of the maxtime at the foot of the light response presumably reflected the imal IBMX-induced current as described in Fig. 3c; afterwards, delay of the phototransduction cascade. The maximal inward the membrane was ruptured by gentle suction applied to the current induced by a saturating puff of IBMX in Fig. 3a is very patch-electrode to achieve the whole-cell recording mode and to similar to that induced by a bright flash; when averaged over dialyze the 8-bromo-cGMP into the cell (Fig. 4a). The results many experiments, the former was slightly larger, by 10% or less. showed that light failed to trigger any current over and above The two currents also had identical currentvoltage relations the maximal IBMX-induced current even when the latter was (Fig. 3b and legend). The IBMX-induced current is interpreted to only a small fraction (16 6%; n = 4 experiments) of the maxreflect a transient rise in intracellular cGMP due to sustained synimal cGMP-induced current. This observation indicates that thesis and transiently inhibited hydrolysis. light acts by inhibiting the phosphodiesterase activity. If light To determine whether light elevates intracellular cGMP by activated the guanylyl cyclase at all, the light response in Fig. activating a guanylyl cyclase, inhibiting a phosphodiesterase, or 4a would have risen above the maximal IBMX-induced current. both, we applied a saturating puff of IBMX and a light flash in The small fraction of open channels resulting from even a strong rapid succession. We found that light was able to elicit an inward puff of IBMX indicates that cGMP synthesis by the guanylyl current only when the IBMX-induced current had already cyclase was relatively slow, so that the transient inhibition of declined from its peak (Fig. 3c). The lack of summation might the phosphodiesterase by IBMX produced only a small elevahave meant that all cGMP-activated channels were already tion of cGMP. Results similar to those in Figs 3 and 4a were opened by a saturating puff of IBMX. However, on average, the
pA pA pA

1998 Nature America Inc. http://neurosci.nature.com

360

pA

nature neuroscience volume 1 no 5 september 1998

pA

pA

1998 Nature America Inc. http://neurosci.nature.com

articles

1998 Nature America Inc. http://neurosci.nature.com

Fig. 2. Similarity between a b cGMP-induced and lightinduced currents. (a) Inhibition of the cGMP- and light-induced currents by the blocker L-cisdiltiazem. The drug (0.1 mM) in s s medium CO2-independent (CIM) was puffed at the recorded cell from a neighboring pipet. Upper panel, inward current induced by whole-cell dialysis of 0.5 mM 8-bromos s cGMP (at time zero), and L-cisdiltiazem puffed for 500 and 50 c ms, respectively. Lower panel, inward current induced by a flash (1 s, delivering 2.2 x 106 photons per m2 at 520 nm) measured with nystatinperforated-patch recording; the traces are in chronological order: flash only (1), flash with a 100 ms L-cis-diltiazem puff (2), flash only (3) and Lcis-diltiazem puff only (4). mV mV (b) Same experiments as in (a), but with puff of CIM containing no L-cis-diltiazem, showing the small artifact from puffing. Same 8-bromo-cGMP concentration and flash intensity as in (a), but a different cell was used. Traces in lower panel are, in chronological order: flash only (1), flash with CIM puff (2) and flash only (3). CIM puff is 500 ms, upper, and 100 ms, lower. (c) Currentvoltage relations for the currents induced by 100 M cGMP dialysis and by light. The relations were obtained with voltage ramps. For the cGMP experiment, the voltage ramp was run several times immediately after membrane breakthrough (and before development of the cGMPinduced current), and the resulting currentvoltage relations were averaged; the same was done after the current activated by the 100 M cGMP-dialysis had reached steady-state. The difference between the two averages gave the depicted relation. For the experiment with light, the same procedure was done in darkness and during a steady light (2.2 x 10 6 photons per m2 per second, at 520 nm). Left, each relation is from a different cell. Right, comparison of the two relations after normalization at +60 mV, showing their strong similarity. Membrane voltage was clamped at 45 mV in (a) and (b).
pA pA pA pA Norm. Curr. pA

obtained with zaprinast 15,16 , a specific inhibitor of cyclic nucleotide-specific phosphodiesterase (data not shown), supporting the conclusions arrived at with IBMX, which is a less specific inhibitor. To rule out any input of cAMP into cGMP metabolism15, we also bath-applied 100 M forskolin, a membrane-permeant activator of adenylyl cyclase, to a photoreceptor recorded in the perforated-patch configuration, but observed no effect in either darkness or light (data not shown). Another experiment supported the conclusion that light acts by inhibiting the dark phosphodiesterase activity. With whole-cell recording, the inward current was small or undetectable in darkness when the pipet solution contained a low concentration of cGMP (see earlier), but this was substantially enhanced by a light flash (Fig. 4b ), suggesting that light inhibited the dark phosphodiesterase activity to allow more exogenous cGMP to open the channels. The same experiments with cAMP dialysis gave no light response (data not shown), again indicating cAMP is not involved. Light responses were not observed when the pipet solution contained GTP but lacked cGMP, suggesting that the guanylyl cyclase activity was labile under whole-cell recording, perhaps because of washout of a factor important for its function. However, the phosphodiesterase activity and its control by light remained intact. The visual pigment is a G-protein-coupled receptor and is
nature neuroscience volume 1 no 5 september 1998

thus expected to signal through a G protein. Is the phosphodiesterase activity in darkness driven by a G protein? When GTPS (1 mM), which permanently activates G proteins, and 50 M cGMP were dialyzed into a cell from a whole-cell pipet, no inward dark current or light response was observed, unlike the result in Fig. 4b, although a strong puff of IBMX was still able to elicit a transient inward current (Fig. 5a). Our interpretation is that the dark phosphodiesterase activity was substantially increased by the presence of GTPS, so that the light flashes (at the intensity used) were not bright enough, despite enhancement by GTPS, to inhibit the hydrolysis sufficiently to allow the exogenous cGMP to open channels. The effectiveness of IBMX under these conditions presumably resulted from its ability to inhibit the phosphodiesterase noncompetitively by affecting the enzymes catalytic site. With a much lower concentration of GTPS (10 M), flash responses were observed that were longer-lasting than control responses, as would be expected from a G protein mediating the light action (not shown). A similar result was obtained with AlF4 (20 M), which is known to act like GTPS17,18 (Fig. 5b). With the low concentration of AlF4 used here, the progressive disappearance of the light response over time could be followed. Though not shown in Fig. 5b, IBMX was still able to elicit an inward current after the flash responses had disappeared in the pres361

1998 Nature America Inc. http://neurosci.nature.com

articles

1998 Nature America Inc. http://neurosci.nature.com

Fig. 3. Comparison of IBMXa b induced and light-induced currents. (a) Similar inward currents induced by a saturating puff of IBMX and by a strong light flash from a photoreceptor. Nystatin perforated-patch recording. Puff (40 ms) of 1 mM IBMX in CIM; flash (1 s) delivered 2.2 106 photons (520 nm) per m2. mV (b) Currentvoltage relations for s the currents induced by IBMX and light, normalized at +60 mV to c show their similarity. A different cell was used for each relation. The relation for the light-induced current was identical to that in Fig. 2c. For the experiment with IBMX, the voltage ramps were run in control solution and again after the current induced by continuous bath application of 1 mM IBMX had reached steady state; the difference between the averaged curs rentvoltage relations in the two conditions were then calculated. s (c) Lack of summation of the currents induced by IBMX and light at the peak of the current induced by a saturating puff of IBMX. Puff (20 ms) of 1 mM IBMX in CIM; flash (1 s) delivered 2.9 x 105 photons per m2 at 520 nm. At right, traces 1, 2 and 3 replotted from left, scaled to the same height and superimposed. Nystatinperforated-patch recording. Different cell from those used in (a) and (b). Membrane voltage was clamped at 45 mV in (a) and (c).
pA pA pA Norm. Curr. pA

ence of 20 M AlF4, as with GTPS. We have not used higher concentrations of AlF4. When GDPS (1 mM), which inhibits G proteins, was included in the pipet solution with 50 M cGMP, the dark current became very large (Fig. 5c). This current was similar in amplitude to the maximal current induced by IBMX in the absence of GDPS (not shown). Furthermore, in the presence of GDPS, IBMX elicited little or no addition-

al inward current (not shown), indicating that the dark phosphodiesterase activity was already mostly inhibited by the GDPS, so that the exogenous cGMP became very effective in opening channels. The lack of light response in this case was not surprising, because the phosphodiesterase was already mostly suppressed and because the presence of GDPS together with the absence of GTP should also inhibit phototrans-

Fig. 4. Light inhibits the dark 0 M cGMP a Nystatin + 50 M 8-Br-cGMP b cGMP-phosphodiesterase activity in the photoreceptor. (a) Lack of summation of the IBMX- and lightinduced currents (as in Fig. 3c), even though these currents reflected the opening of only a 10 M small percentage of the cGMP-activated channels. Top panel, nystatinperforated-patch recording s until break of trace; thereafter, whole-cell recording (WCR) with Nystatin alone 50 M 8-bromo-cGMP dialysis. F, 50 M flash; P, puff of IBMX; P,F, IBMX and flash in rapid succession. Bottom panel, control experiment in which 8-bromo-cGMP had been omitted from pipet. 30 ms puff of IBMX at 1 mM. (b) Light increased the inward current induced by wholes cell cGMP dialysis. Each trace cors responds to a different cGMP concentration in the whole-cell pipet and is from a different cell. Whole-cell dialysis began at time zero. Flash (1 s) delivered 2.2 x 106 photons per m2 at 520 nm. Membrane voltage was clamped at 45 mV.
pA pA

362

nature neuroscience volume 1 no 5 september 1998

pA

pA

pA

1998 Nature America Inc. http://neurosci.nature.com

articles

duction. In control experiments with dialysis of 1 mM GDPS but no cGMP included, no inward current developed in darkness. Taken together, these results suggest that the phosphodiesterase activity in darkness is largely stimulated by an active G protein. Most of this enzyme activity seems to be suppressible by light, as indicated by the very similar inward currents induced by light and by IBMX (Fig. 3a).

a
pA

1 mM GTPS + 50 M cGMP

Discussion Here we have shown that the depolarizing light response of parietal-eye photoreceptors arises from an increase in cGMP concentration and the consequent activation of cGMP-gated cation channels on the outer segment of the cell. This increase in cGMP results from a decrease in phosphodiesterase activity rather than an increase in guanylyl cyclase activity. What is the underlying signaling mechanism? A cascade essentially identical, but anti-symmetric, to the phototransductioin cascade in rods and cones (Fig. 6a, left panel) might explain all of our findings. In this scheme, all of the photopigment molecules are active in darkness, driving a G protein and stimulating a phosphodiesterase to hydrolyze cGMP and keep the channels closed. Light inactivates the pigment and hence the cascade, causing the cGMP level to rebound and the cGMP-activated channels to open. This cascade is esthetically simple, and it embodies the key features of an active G protein and an active phosphodiesterase in darkness that are inhibited by light. However, the above scheme cannot explain the saturating responseintensity relation observed for the parietal-eye photoreceptor7 (also W.-H. X. and K.-W. Y., unpublished data). Based on enzyme kinetics, the increase in phosphodiesterase activity would be expected to depend on the amount of active pigment according to a function of the Michaelis-Menten type. Consequently, the number of open channels becoming closed should depend on the amount of active pigment in a similar manner (modified by the cooperativity function13 between open channels and cGMP, which, however, does not affect our present argument). Therefore, with increasing amounts of active pigment, the dependence would approach an asymptote, corresponding to all channels being closed (Fig. 6a, curve 1). Thus, if an initially all-active pigment is inactivated by light in direct proportion to light intensity, the number of channels that are reopened, and hence the amplitude of the depolarizing light response, should be initially very insensitive to light because the relation is still riding on the asymptote, but the response will eventually increase steeply when the last few pigment molecules are inactivated. This prediction is considerably different from the form of the responseintensity function observed experimentally (Fig. 6a, curve 2), in which the response initially rises linearly with light intensity and then saturates at high intensities7. Even if the pigment content of the cell is so low that the phosphodiesterase activity rises linearly throughout the range of active pigment fractions, the model still cannot predict an observed saturating intensityresponse relation. Thus, the cascade in Fig. 6a cannot be correct. A more likely phototransduction cascade is one in which two signaling pathways, each associated with a distinct G protein, control the phosphodiesterase activity. One G protein (G1) is active in darkness and stimulates the enzyme, and the other G protein (G2) is stimulated by light and inactivates the enzyme (Fig. 6c). The simpler alternative in which a single G protein mediates both pathways ( Fig 6b) is improbable because this would require the visual pigment to inhibit the G protein directly, a polarity of interaction that has not been described for any receptor coupled to a G protein. It is still not known whether the
nature neuroscience volume 1 no 5 september 1998

b
pA

20 M AlF4 + 50 M cGMP

1998 Nature America Inc. http://neurosci.nature.com

1 mM GDPS + 50 M cGMP

pA

Fig. 5. A G protein is involved in the control of the cGMP-phosphodiesterase activity in darkness. cGMP (50 M) was dialyzed with 1 mM GTPS (a), 20 M AlF4 (b) or 1 mM GDPS (c) by wholecell recording (which began at time zero) into a cell. Note the gradual disappearance of the light response in (b). Different cells were used in (a), (b) and (c). Light flashes (1 s) delivered 2.2 x 106 photons per m2 at 520 nm. IBMX puff in (a) was at 1 mM for 30 ms. Membrane voltage was clamped at 45 mV.

stimulatory G protein is spontaneously active or is driven by an upstream signal in darkness, hence the question mark in Fig. 6c. Based on an observed chromatic antagonism7, the photopigment in this photoreceptor may be bistable, as in invertebrates7,19. A candidate for driving G1 in darkness would then be the putative second stable state of the pigment. It is also not known whether the inhibitory G2 protein acts directly on the phosphodiesterase, or indirectly through inhibiting G1. Possibilities include a direct inhibition of the phosphodiesterase by G2 or G2, and an inhibition of G1 by G217,20. These questions may be answered by the molecular identification of the G proteins and the phosphodiesterase. By analogy to retinal rods and cones1, the phosphodiesterase is probably composed of , and subunits. We have attempted to use cholera and pertussis toxins, which ADP-ribosylate specific -subunits of G proteins and cause their permanent activation or inactivation1,17,2123, to investigate G1 and G2, but the experiments were inconclusive, perhaps because of access problems in the outer segment. Molecular cloning of the genes may be the best approach. Our results demonstrate not only an unusual phototrans363

1998 Nature America Inc. http://neurosci.nature.com

articles

1998 Nature America Inc. http://neurosci.nature.com

duction mechanism, but also newly discovered features about signaling involving cyclic nucleotides. First, all signaling pathways known thus far to elevate cGMP invariably involve the activation of a guanylyl cyclase: either a ligand-activated membranous cyclase or a nitric-oxide-activated soluble cyclase2426. However, in the parietal-eye photoreceptor, the rise in cGMP results from an inhibition of the phosphodiesterase. Because the cGMP level represents a balance between synthesis and hydrolysis, a regulation of either step can achieve the same result. By inhibiting the phosphodiesterase to elevate cGMP, a cell can allow the guanylyl cyclase activity to limit the maximal rate of cGMP increase regardless of stimulus strength in a signaling pathway. Second, the antagonistic control of the cGMP phosphodiesterase by two G proteins described here has no prior example, but it bears a striking parallel to the Gs/Gi antagonistic control of adenylyl cyclase in cAMP signaling17,20,22,23,27. Finally, the inhibition of phosphodiesterase activity may be exploited also by other cell types as a mechanism to elevate not just cGMP, but cAMP as well, instead of the conventional mechanism through activation of the adenylyl cyclase. From our discoveries about the parietal-eye photoreceptor, and those by Gomez and Nasi28 on the scallop hyperpolarizing photoreceptorboth of which are ciliary photoreceptors like retinal rods and conesit is now clear that all ciliary photoreceptors, whether vertebrate or invertebrate, depolarizing or hyperpolarizing, use a cGMP signaling cascade for phototransduction. Nonetheless, it is still unclear from the evolutionary point of view how much variation in details exists within this common motif. For example, although light also elevates cGMP in the scallop hyperpolarizing photoreceptor, it is unknown28 whether this involves an inhibition of the phosphodiesterase or a stimulation of the guanylyl cyclase.

Methods I SOLATION OF PARIETAL - EYE PHOTORECEPTORS . Except for minor modifications in the procedure, the parietal-eye photoreceptors from the side-blotched lizard, Uta stansburiana, were isolated as described7,11. The parietal eye, still attached to the overlying skin, was dissected from the pithed lizard and immersed for 15 min in CO 2-independent medium (CIM; Life Technologies) containing collagenase (CLS1, 5 mg/ml; Worthington, Freehold, New Jersey). After washes in CIM, the eye was removed from the skin using tweezers and was incubated in CIM containing pronase (2 mg/ml; Boehringer) for 13 min so that the retina could be detached from the eye capsule. The retina was placed in a solution free of divalent cations (140 mM NaCl, 5 mM KCl, 5 mM Na-EGTA, 5 mM Na-EDTA, 10 mM Na-HEPES, pH 7.4) containing trypsin (2 mg/ml; Sigma) for 11 min, rinsed in CIM containing 10% newborn calf serum (Life Technologies) for 10 min and then in CIM for 10 min. Finally, the tissue was triturated gently with glass pipets having progressively smaller diameters (500100 m). The cell suspension was plated on a glass coverslip pretreated with concanavalin A (Sigma) and used for as long as six hours. The dissection and solution changes were done in room light, whereas the incubations and subsequent storage were done in complete darkness, all at room temperature (2325C).

Fig. 6. Possible phototransduction schemes for the parietal-eye photoreceptor. (a) Left, the simplest phototransduction scheme that can explain the observations qualitatively, but not quantitatively. Right, curve 1, the responseintensity relation expected from this scheme, ignoring cooperativity in channel activation by cGMP; curve 2, the form of the responseintensity relation observed experimentally7. Rh, inactive pigment; Rh*, active pigment. (b) A more complex phototransduction scheme that involves a single G protein but is unlikely to be correct. (c) The probable phototransduction scheme describing the operation of the parietal-eye photoreceptor, involving an excitatory and an inhibitory G protein. Question marks (b,c) indicate an unknown; that is, whether the stimulatory G1 protein is spontaneously active or is driven by an upstream signal in darkness.

was done under microscope light; immediately after, all procedures were done in the dark, except for stimulation with light flashes. For wholecell recording, dark adaptation for about 10 min was allowed before membrane breakthrough; for perforated-patch recording, patch perforation by nystatin generally took 25 min, again in the dark. The patch pipets were pulled from aluminosilicate glass capillaries (World Precision Instruments, Sarasota, Florida) to a tip lumen diameter of less than 0.5 m. For whole-cell recording, the pipets were filled with a pseudointracellular solution (10 mM KCl, 120 mM K-gluconate, 5 mM MgCl2, 1 mM Na-EGTA, 10 mM K-HEPES, 3 mM Na2ATP, 1 mM Na2GTP, pH 7.4), to which test chemicals were added, as needed, from stock solutions prepared daily; for GTPS and GDPS, they were equimolar substitutions for GTP. For perforated-patch recording, a fresh stock solution of nystatin (50 mg/ml; Sigma) in dimethylsulfoxide (DMSO) was diluted to 100200 g/ml in a base-pipet solution7 (4 mM KCl, 120 mM K-gluconate, 4 mM NaCl, 1 mM CaCl2, 2 mM MgCl2, 11 mM Na-EGTA, 10 mM K-HEPES, pH 7.4). The pipet, including its tip, was filled with this nystatin-containing solution. The progress of perforation
nature neuroscience volume 1 no 5 september 1998

ELECTROPHYSIOLOGICAL RECORDING. Whole-cell recording and perforated-patch recording with nystatin12 were made from cell bodies of the dissociated parietal-eye photoreceptors. The initial gigaseal formation
364

1998 Nature America Inc. http://neurosci.nature.com

articles

was assessed from the decrease in access resistance (to about 30 M). The liquid-junction potential was measured to be about 7 mV for perforated-patch recording, but has not been subtracted from the holding potential indicated. In a few experiments, perforated-patch recording was intentionally changed to whole-cell recording by applying gentle suction inside the pipet to rupture the underlying plasma membrane. In both recording configurations, the membrane potential was clamped at 45 mV unless otherwise indicated. In experiments to measure the currentvoltage relation, the membrane potential was first stepped to 60 mV for 10 milliseconds (ms) and then ramped from 60 mV to +60 mV at 267 mV/s. All nucleotides were obtained from Sigma. During recordings, the chamber containing the cells was continuously perfused with CIM at 0.70.8 ml/min. Puffer pipets were made in the same manner as patch pipets except for their larger tip diameters (about 23 m). They were filled with CIM solution, with or without a test chemical, and controlled by a picospritzer (General Valve, Fairfield, New Jersey) operating at about 10 psi. Light flashes were all at 520 nm. Because of the low light sensitivity of these cells, the flash duration was set at 1 s. The recordings were done at room temperature. Forskolin (Sigma) was dissolved in DMSO before being diluted with CIM to the final concentration. (Final concentration of DMSO was about 0.1%.) The cholera and pertussis toxins (List Biochemicals, Campbell, California) were dialyzed, either as the holoenzyme or as the (catalytic) subunit A alone (10 g/ml for cholera toxin and 1 g/ml for pertussis toxin), with 50 M cGMP, 1 mM nicotinamide adenine dinucleotide and 0.1 mM dithiothreitol from the whole-cell pipet into the cell29,30.

Acknowledgements
We thank Roger C. Hardie and Yiannis Koutalos for suggestions about experiments, and Joe Beavo, Jackie D. Corbin, Peter N. Devreotes, Vincent Manganiello, Enrico Nasi, Carol Parent, and W. Joe Thompson for discussions. Robert D. Barber, Peter G. Gillespie, Maria E. Grunwald, Dietmar Krautwurst and Haining Zhong provided comments on the manuscript. We also thank Eric Lasater for his generosity in making this collaboration possible. The L-cisdiltiazem was a gift from Tanabe Seyaku Co. (Osaka, Japan). This work was supported by a grant from the U.S. National Eye Institute.

RECEIVED 23 JUNE: ACCEPTED 20 JULY 1998


1. Stryer, L. Cyclic GMP cascade of vision. Annu. Rev. Neurosci. 9, 87119 (1986). 2. Lagnado, L. & Baylor, D. A. Signal flow in visual transduction. Neuron 8, 9951002 (1992). 3. Pugh, E. N. Jr & Lamb, T. D. Amplification and kinetics of the activation steps in phototransduction. Biochim. Biophys. Acta 1141, 111149 (1993). 4. Yarfitz, S. & Hurley, J. B. Transduction mechanism of vertebrate and invertebrate photoreceptors. J. Biol. Chem. 269, 1432914332 (1994). 5. Yau, K.-W. & Baylor, D. A. Cyclic GMP-activated conductance of retinal photoreceptor cells. Annu. Rev. Neurosci. 12, 289327 (1989).

6. Eakin, R. M. in The Third Eye (University of California Press, Berkeley, 1973). 7. Solessio, E. & Engbretson, G. A. Antagonistic chromatic mechanisms in photoreceptors of the parietal eye of lizards. Nature 364, 442445 (1993). 8. Hardie, R. C. & Minke, B. Phosphoinositide-mediated phototransduction in Drosophila photoreceptors: the role of Ca2+ and trp. Cell Calcium 18, 256274 (1995). 9. Ranganathan, R., Malicki, D. M. & Zuker, C. S. Signal transduction in Drosophila photoreceptor. Annu. Rev. Neurosci. 18, 283317 (1995). 10. Shin, J., Richard, E. A. & Lisman, J. E. Ca2+ is an obligatory intermediate in the excitation cascade of limulus photoreceptors. Neuron 11, 845855 (1993). 11. Finn, J. T., Solessio, E. C. & Yau, K.-W. A cGMP-gated cation channel in depolarizing photoreceptors of the lizard parietal eye. Nature 385, 815819 (1997). 12. Horn, R. & Marty, A. Muscarinic activation of ionic currents measured by a new whole-cell recording method. J. Gen. Physiol. 92, 145159 (1988). 13. Koch, K.-W. & Kaupp, U. B. Cyclic GMP directly regulates a cation conductance in membranes of bovine rods by a cooperative mechanism. J. Biol. Chem. 260, 67886800 (1985). 14. Nicol, G. D., Schnetkamp, P. P., Saimi, Y., Cragoe, E. J. Jr & Bownds, M. D. A derivative of amiloride blocks both the light-regulated and cyclic GMPregulated conductances in rod photoreceptors. J. Gen. Physiol. 90, 651669 (1987). 15. Beavo, J. A. Cyclic nucleotide phosphodiesterases: functional implications of multiple isoforms. Physiol. Rev. 75, 725748 (1995). 16 Gillespie, P. G. & Beavo, J. A. Inhibition and stimulation of photoreceptor phosphodiesterases by dipyridamole and M&B 22,948. Mol. Pharmacol. 36, 773781 (1989). 17. Gilman, A. G. G proteins: transducers of receptor-generated signals. Annu. Rev. Biochem. 56, 615649 (1987). 18. Bigay, J., Deterre, P., Pfister, C. & Chabre, M. Fluoroaluminates activate transducin-GDP by mimicking the gamma-phosphate of GTP in its binding site. FEBS Lett. 191, 181185 (1985). 19. Pugh, E. N. Jr, Lisman, J. & Payne, R. Strange case of the third eye. Nature 364, 389390 (1993). 20. Sunahara, R. K., Dessauer, C. W. & Gilman, A. G. Complexity and diversity of mammalian adenylyl cyclases. Annu. Rev. Pharmacol. Toxicol. 36, 461480 (1996). 21. Ui, M. & Katada, T. Bacterial toxins as probe for receptor-Gi coupling. Adv. Second Messenger Phosphoprotein Res. 24, 6369 (1990). 22. Birnbaumer, L. G proteins in signal transduction. Annu. Rev. Pharmacol. Toxicol. 30, 675705 (1990). 23. Simon, M. I., Strathmann, M. P. & Gautam, N. Diversity of G proteins in signal transduction. Science 252, 802808 (1991). 24. Drewett, J. G. & Garbers, D. L. The family of guanylyl cyclase receptors and their ligands. Endocr. Rev. 15, 135162 (1994). 25. Wedel, B. J. & Garbers, D. L. New insights on the functions of the guanylyl cyclase receptors. FEBS Lett. 410, 2933 (1997). 26. Zhang, J. & Snyder, S. H. Nitric oxide in the nervous system. Annu. Rev. Pharmacol. Toxicol. 35, 213233 (1995). 27. Gudermann, T., Schoneberg, T. & Schultz, G. Functional and structural complexity of signal transduction via G-protein-coupled receptors. Annu. Rev. Neurosci. 20, 399427 (1997). 28. Gomez, M. & Nasi, E. Activation of light-dependent K+ channels in ciliary invertebrate photoreceptors involves cGMP but not the IP3/Ca2+ cascade. Neuron 15, 607618 (1995). 29. Li, Y., Hanf, R., Otero, A. S., Fischmeister, R. & Szabo, G. Differential effects of pertussis toxin on the muscarinic regulation of Ca2+ and K+ currents in frog cardiac myocytes. J. Gen. Physiol. 104, 941959 (1994). 30. Zong, X. & Lux, H. D. Augmentation of calcium channel currents in response to G protein activation by GTP gamma S in chick sensory neurons. J. Neurosci. 14, 48474853 (1994).

1998 Nature America Inc. http://neurosci.nature.com

nature neuroscience volume 1 no 5 september 1998

365

1998 Nature America Inc. http://neurosci.nature.com

articles

Glutamate-induced neuron death requires mitochondrial calcium uptake


Amy K. Stout1, Heather M. Raphael1, Beatriz I. Kanterewicz2, Eric Klann2 and Ian J. Reynolds1
1 2

Department of Pharmacology, E1354 Biomedical Science Tower, University of Pittsburgh, Pittsburgh, Pennsylvania 15261, USA Department of Neuroscience, 446 Crawford Hall, University of Pittsburgh, Pittsburgh, Pennsylvania 15260, USA Correspondence should be addressed to I.J.R. (ijr@prophet.pharm.pitt.edu)

1998 Nature America Inc. http://neurosci.nature.com

We have investigated the role of mitochondrial calcium buffering in excitotoxic cell death. Glutamate acts at NMDA receptors in cultured rat forebrain neurons to increase the intracellular free calcium concentration. Although concurrent inhibition of mitochondrial calcium uptake substantially enhanced this cytoplasmic calcium increase, it significantly reduced glutamatestimulated neuronal cell death. Mitochondrial inhibition did not affect nitric oxide production or MAP kinase phosphorylation, which have been proposed to mediate excitotoxicity. These results indicate that very high levels of cytoplasmic calcium are not necessarily toxic to forebrain neurons, and that potential-driven uptake of calcium into mitochondria is required to trigger NMDAreceptor-stimulated neuronal death.

Despite being the predominant excitatory neurotransmitter in mammalian brain, glutamate can be toxic to neurons. Some of the tissue damage that occurs with traumatic brain injury or ischemia is thought to result from the release of glutamate and other excitatory amino acids, and so cell-culture models of glutamate toxicity have been used to investigate the mechanisms of neuronal injury associated with these pathologies. The delayed neuronal injury that occurs after intense glutamate stimulation depends on calcium influx through the NMDA subtype of glutamate receptor1. Although it is unclear whether calcium influx through NMDA receptors is uniquely toxic because of the subcellular localization of these receptors2 or because of their high calcium permeability3, it has generally been accepted that there is some correlation between cytoplasmic calcium concentrations and toxicity. Many cytoplasmic calcium-dependent enzymes, such as nitric oxide (NO) synthase and the protease calpain, have been proposed to mediate excitotoxicity4,5. Alternatively, reactive oxygen species production may be the ultimate determinant of cell death610, and it is also possible that the impaired bioenergetic state that occurs because of the calcium load could contribute to loss of cell viability11,12. Mitochondria may be involved in glutamate toxicity1315. Because the mitochondria have a large capacity for calcium uptake (through the potential-driven calcium uniporter)11,12,16, they might be neuroprotective by removing calcium from the cytoplasm17. However, calcium uptake into the mitochondria might have detrimental effects, as glutamate stimulation results in both enhanced reactive oxygen species production by the mitochondria1820 and mitochondrial membrane depolarization (presumably through opening of the permeability transition pore)1315. Also, mitochondrial calcium cycling may contribute to bioenergetic failure11,12. Here we have attempted to discriminate the causes of excitotoxic cell death by blocking mitochondrial function. We dissi366

pated the mitochondrial membrane potential and eliminated calcium uptake into mitochondria using three different inhibitor treatments: the protonophore and mitochondrial uncoupler FCCP (carbonyl cyanide p-(trifluoromethoxy)phenyl-hydrazone), the structurally distinct uncoupler dinitrophenol (DNP), or the electron-transport-chain inhibitor rotenone in combination with the ATP-synthase inhibitor oligomycin. Although glutamate-stimulated increases in [Ca2+]i (intracellular free calcium concentration) were significantly potentiated by these drugs, transient inhibition of mitochondrial function during glutamate stimulation prevented glutamate-induced cell death. These results indicate that mitochondrial calcium uptake is necessary for expression of excitotoxicity.

Results

INHIBITORS DEPOLARIZE MITOCHONDRIAL MEMBRANES


We previously measured mitochondrial membrane potential in our cultured neurons using the fluorescent dye JC-1 and showed that FCCP caused a rapid, robust and reversible mitochondrial membrane depolarization14. Those studies also demonstrated that glutamate exposure can cause mitochondrial membrane depolarization14. Here we measured changes in mitochondrial membrane potential upon glutamate stimulation in the presence and absence of mitochondrial inhibitors in primary cultures of fetal rat forebrain neurons (Fig. 1). Stimulation with 100 M glutamate in the presence of 10 M glycine caused a significant mitochondrial membrane depolarization. The mean normalized JC-1 ratio, a semiquantitative estimate of the mitochondrial membrane potential, obtained after a five-minute exposure to glutamate was 0.79 0.02 ( n = 82; p < 0.05 compared with buffer-treated controls, 0.94 0.02; n = 88). Mitochondrial inhibitors significantly potentiated the glutamate-stimulated decrease in the JC-1 ratio. When neurons were stimulated with glutamate and glycine in the presence of 5 M
nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

Time (min)

1998 Nature America Inc. http://neurosci.nature.com

Fig. 1. Mitochondrial inhibitors enhance glutamate-stimulated mitochondrial membrane depolarization. Mitochondrial membrane potential was measured as the ratio of the 590 nm/530 nm fluorescence of JC-1, and all ratios were normalized to a starting value of one. Neurons were stimulated with 100 M glutamate and 10 M glycine for five minutes (solid bar) at room temperature. FCCP (750 nM) or rotenone (5 M) and oligomycin (10 M) were present during stimulation as indicated; rotenone and oligomycin were also included for five minutes before stimulation (dashed line). After a 20-min recovery period, cells were stimulated with 750 nM FCCP (three minutes, arrow) as a positive control for mitochondrial membrane depolarization. The mean results from four to six experiments (5788 neurons) are shown.

even more (1.53 0.07; n = 314; p < 0.05). Glutamate-stimulated increases in [Ca2+]i were also significantly enhanced by a second, structurally distinct protonophore/uncoupler, DNP (300 M, 0.92 0.03; n = 94; p < 0.05 compared with glutamate alone). Experiments using JC-1 confirmed that this concentration of DNP causes a rapid and reversible mitochondrial membrane depolarization similar to that observed with 750 nM FCCP (not shown). Preliminary experiments using the electron transport chain inhibitor antimycin A also indicate that Magfura 2 responses are augmented in the presence of this drug (not shown). These results indicate that mitochondrial inhibitors with different mechanisms of action can block calcium uptake into the mitochondria and amplify the size of the glutamate-induced cytoplasmic calcium transient.

Normalized JC-1 ratio

MITOCHONDRIAL INHIBITORS REDUCE GLUTAMATE TOXICITY


We assessed cell viability by calcein AM loading approximately 24 hours after glutamate exposure (Fig. 3a). A ten-minute exposure to glutamate and glycine reduced neuronal viability to 54% 2 of buffer-treated controls (p < 0.05). When 750 nM FCCP was included during glutamate stimulation, toxicity was significantly blocked (80% 4 of control; p < 0.05 compared with glutamate alone). In contrast, rotenone and oligomycin significantly potentiated glutamate toxicity (25% 2 of control; p < 0.05 compared with glutamate alone), probably because rotenone and oligomycin irreversibly inhibit mitochondrial function 21 . Substantiating this interpretation, rotenone and oligomycin alone reduced cell viability (17% 10 control) to the same degree as rotenone and oligomycin plus glutamate (p = 0.33). Prolonged exposure to FCCP was also toxic. When neurons were exposed to FCCP for 24 hours, viability was considerably decreased regardless of exposure to glutamate: with glutamate exposure, 36% 17 of control, p = 0.18 compared with glutamate and rotenone and oligomycin; without glutamate exposure, 30% 11 of control, p = 0.23 compared with rotenone and oligomycin. Transient inhibition of mitochondrial function with a ten-minute exposure to FCCP alone had no significant effect on cell viability (94% 5 control; p = 0.23 compared with buffer alone).

rotenone and 10 M oligomycin (after a five-minute pretreatment with the inhibitors), the mitochondrial membrane depolarization was significantly enhanced (JC-1 ratio, 0.50 0.02; n = 84; p < 0.05) compared with glutamate alone. Similarly, coapplication of 750 nM FCCP with glutamate and glycine also potentiated the response (0.41 0.01; n = 57; p < 0.05) compared with glutamate alone. The depolarization observed with glutamate and FCCP stimulation was not different from that observed with FCCP alone. The mean normalized JC-1 ratio obtained at the end of a three-minute exposure to FCCP in the buffer-treated controls was 0.44 0.02 (n = 88; p = 0.37 compared with glutamate and FCCP). Unlike FCCP, rotenone and oligomycin treatment alone did not cause significant mitochondrial membrane depolarization. At the end of the five-minute pretreatment with rotenone and oligomycin, the normalized JC-1 ratio was 0.91 0.02 (n = 84). However, the depolarization observed with glutamate, rotenone and oligomycin was similar to that found with glutamate and FCCP treatment (Fig. 1). These results indicate that the potential-driven mitochondrial calcium uptake normally associated with glutamate stimulation is compromised by mitochondrial inhibitors. They further show that inhibition of mitochondrial function by FCCP is reversible, but inhibition by rotenone and oligomycin is not (Fig. 1). We used the low-affinity fluorescent calcium indicator Magfura 2 to measure glutamate-induced increases in [Ca2+]i (Fig. 2). The mean peak Magfura 2 ratio obtained upon stimulation with glutamate and glycine was 0.65 0.02 (n = 580). When neurons were stimulated with glutamate and glycine in the presence of 5 M rotenone and 10 M oligomycin (after a five-minute inhibitor pretreatment), the glutamate-stimulated increase in [Ca2+]i was significantly enhanced (mean peak Magfura 2 ratio, 1.11 0.03; n = 149; p < 0.05). Similarly, co-application of 750 nM FCCP with glutamate and glycine amplified the response
nature neuroscience volume 1 no 5 september 1998

MITOCHONDRIAL INHIBITORS POTENTIATE INCREASES IN [Ca2+]i

Magfura 2 ratio

Time (min)

Fig. 2. Mitochondrial inhibitors enhance glutamate-stimulated increases in [Ca2+]i measured using Magfura 2. Neurons were stimulated with 100 M glutamate and 10 M glycine for five minutes (as indicated by the bar) at room temperature. FCCP (750 nM) or rotenone (5 M) and oligomycin (10 M) were present during glutamate stimulation as indicated. In addition, rotenone and oligomycin were also included for five minutes before stimulation. The mean responses from three fields of 2628 neurons from three separate coverslips are shown. Similar results were obtained from four to nine additional coverslips.
367

1998 Nature America Inc. http://neurosci.nature.com

articles

a
Calcein fluorescence, % control

b
(100) (96)

(55)

c
(100)

(88) (87)

(49)

(55)

(43)

Buffer

Glu

We also assessed neuronal viability using cell counting after trypan blue staining (Fig. 3bd). Glutamate and glycine reduced cell counts by about 50% in these studies, and FCCP completely blocked the toxicity associated with a five-minute glutamate stimulation without having any effect on cell viability by itself (Fig. 3b). Similarly, DNP also protected cells from glutamate toxicity (Fig. 3c). When cells were stimulated with glutamate in the presence of 300 M DNP, cell counts were not significantly different from those obtained in buffer-treated controls (115 1 versus 133 6; n = 3; p = 0.06, 87% of control). Also, DNP by itself did not have any significant effect on cell viability (cell counts, 134 6; n = 3; p > 0.8, 101% of control). These results indicate that mitochondrial uncoupling is neuroprotective and that the protection found with FCCP is not because of some unique structural feature of this compound compared with other uncouplers.

INTRACELLULAR pH CHANGES DO NOT CAUSE CELL DEATH


FCCP can inhibit glutamate-stimulated decreases in intracellular pH (pHi) in neurons22. Thus, the neuroprotection observed with
368

FCCP may have been related to its effects on pHi rather than its ability to block mitochondrial calcium uptake. To investigate this, we measured changes in pHi in our neurons with the fluorescent indicator 2,7-bis-(2-carboxyethyl)-5-(and-6)-carboxyfluorescein (BCECF). Similar to previous results, both glutamate and FCCP exposures caused a decrease in pHi, and FCCP counteracted the glutamate-induced acidification when cells were exposed to both agents simultaneously ( Fig. 4a and b ). The BCECF ratio measured at the end of a five-minute glutamate stimulation was 2.56 0.03 (n = 248), whereas the ratio measured at the end of a five-minute glutamate stimulation in the presence of 750 nM FCCP was 4.21 0.04 (n = 95, p < 0.05 compared to glutamate alone). FCCP caused an intracellular acidification that was similar to that triggered by glutamate (mean final BCECF ratio, 2.68 0.03; n = 126) but was not toxic to neurons (Fig. 3a and b), which indicates that intracellular acidification is not the primary determinant of neuronal cell death and that pH effects can be dissociated from toxicity. Furthermore, exposure of neurons to glutamate in the presence of rotenone and
nature neuroscience volume 1 no 5 september 1998

Buffer pH 8.5

Glu + DNP

Glu pH 8.5

Buffer

DNP

Glu

Fig. 3. Transient inhibition of mitochondrial calcium uptake protects against glutamate toxicity. (a) Neurons were stimulated with 100 M glutamate and 10 M glycine for 10 min, and viability was measured by calcein AM loading approximately 24 h after glutamate exposure. FCCP (750 nM) or rotenone (5 M) and oligomycin (10 M) were present during glutamate stimulation as indicated. In addition, rotenone and oligomycin were included for five minutes before stimulation. FCCP was also included during the 24-hour post-glutamate incubation for treatments labelled as Long FCCP. Each bar represents the mean results from three to sixteen experiments done in triplicate. Asterisks (*) indicate calcein fluorescence was significantly different from buffer-treated controls (p < 0.05, one-way ANOVA with Bonferroni multiple comparisons test for post-hoc comparisons). Double asterisks (**) indicate results were significantly different from glutamate treatment alone (p < 0.05, one-way ANOVA with Bonferroni multiple comparisons test for post-hoc comparisons). (bd) Neurons were stimulated with 100 M glutamate and 10 M glycine for five minutes, and viability was measured by cell counting after trypan blue staining approximately 24 h after glutamate exposure. Each bar represents the mean results from three experiments in which four or more fields from each of three coverslips were counted for each treatment condition in a blinded manner. Asterisks (*) indicate results were significantly different from buffer-treated controls (p < 0.05, oneway ANOVA with Bonferroni multiple comparisons test for post-hoc comparisons). Cell counts expressed as percent of buffer-treated controls are indicated in parentheses at the top of each bar. FCCP (750 nM; b) or DNP (300 M; c) was present during stimulation, or stimulations were done in buffer adjusted to pH 8.5 (d), as indicated.

Buffer

Glu + FCCP

FCCP

Glu

Glu + Rot Oligo

Rot + Oligo

Glu + Long FCCP

Long FCCP

Trypan blue-excluding cells

Buffer

1998 Nature America Inc. http://neurosci.nature.com

Trypan blue-excluding cells

Trypan blue-excluding cells

Glu + FCCP

FCCP

Glu

1998 Nature America Inc. http://neurosci.nature.com

articles

a
BCECF ratio

Final BCECF ratio

FCCP

Glu

Glu pH 8.5

Time (min)

1998 Nature America Inc. http://neurosci.nature.com

Fig. 4. Mitochondrial inhibitors and elevated extracellular pH inhibit glutamate-stimulated intracellular acidification. (a) Neurons were stimulated with 100 M glutamate and 10 M glycine, with 750 nM FCCP or with glutamate, glycine and FCCP for 5 min (bar). The mean BCECF ratios from three fields of 2125 neurons from three separate coverslips are shown. Similar results were obtained from three to nine additional coverslips. (b) Final BCECF ratios were determined at the end of five-minute stimulations. FCCP (750 nM) or rotenone (5 M) and oligomycin (10 M) were present during stimulation as indicated, and rotenone and oligomycin were also included for 5 min before stimulation. Each bar represents the mean results from three to ten experiments. Asterisks (*) indicate results were significantly different from glutamate treatment alone (p < 0.05, one-way ANOVA with Bonferroni multiple comparisons test for post-hoc comparisons). The mean baseline BCECF ratio from all experiments is shown for comparison.

oligomycin also significantly inhibited glutamate-induced acidification (3.64 0.02; n = 98; p < 0.05). Although co-exposure with rotenone and oligomycin did not block glutamate toxicity (Fig. 3a), it did inhibit glutamate-induced acidification, indicating that inhibition of glutamate-induced acidification is not sufficient for neuroprotection and that FCCP-induced neuroprotection is not due to its ability to counteract glutamateinduced acidification. Changes in extracellular pH can have significant effects on pHi23. Consistent with this, when we stimulated cells in an extracellular buffer adjusted to pH 8.5, glutamate-induced intracellular acidification was again significantly inhibited (Fig. 4b, mean final BCECF ratio, 3.83 0.04; n = 81; p < 0.05). However, glutamate stimulation in these conditions was still neurotoxic (Fig. 3d). Again, these results indicate that blocking glutamate-induced intracellular acidification is insufficient for neuroprotection and that the FCCP neuroprotection is not due to its effects on intracellular pH.
FCCP DOES NOT AFFECT NOS ACTIVITY

compared with buffer-treated controls). However, FCCP had no effect on basal unstimulated NOS activity (3.04 0.30; p = 0.62 compared with buffer-treated controls) or on glutamate-stimulated NOS activity (6.92 0.97; p = 0.30 compared with glutamate alone). These results indicate that the neuroprotection observed when FCCP is applied in combination with glutamate cannot be explained by an inhibition of NO production.
FCCP DOES NOT AFFECT MAPK ACTIVITY

Many reports demonstrate a correlation between activation of calcium-dependent cytoplasmic enzymes and glutamate toxicity4,5,7,24,25, but we have shown here a condition in which glutamate-stimulated increases in [Ca2+]i are extremely high with little apparent toxicity. We therefore determined the effect of FCCP on the glutamate-induced activation of these proposed cytoplasmic mediators of excitotoxicity in an attempt to explain the neuroprotective effects of FCCP. We measured nitric oxide synthase (NOS) activity by determining the conversion of radiolabelled L-arginine to L-citrulline (Fig. 5). As expected, glutamate significantly enhanced NOS activity (percent conversion, 2.66 0.29 for buffer-treated controls and 6.10 0.64 for glutamate stimulation; p < 0.05), which was completely blocked by the competitive NOS inhibitor N-nitro-L-arginine (2.34 0.07; p = 0.74
nature neuroscience volume 1 no 5 september 1998

Glutamate-receptor stimulation causes phosphorylation and activation of mitogen-activated protein kinases (MAPKs) in neurons in a calcium-dependent manner 26, and activation of a member of the MAPK family may mediate glutamate toxicity27. However, activation of MAPKs has also been shown to ameliorate toxicity28. Therefore, we measured glutamate-stimulated p42/p44 MAPK activation in our experimental system (Fig. 6). Using an antibody to the dual-phosphorylated, activated form of p42/p44 MAPK, we observed that glutamate stimulation caused a significant enhancement of both p42 and p44 MAPK activity (311% 57 and 215% 23 control, respectively; p < 0.05). Similar to the results with NOS activity, FCCP had no effect on the basal unstimulated MAPK activity (p42 activity, 124% 21; p44 activity, 90 8% control; p = 0.66 and 0.69 compared with buffer alone) or glutamate-induced MAPK activation (p42 activity, 276% 42; p44 activity, 227% 22 control; p = 0.52 and 0.63 compared with glutamate alone). These results provide an additional example of a cytosolic signalling pathway postulated to participate in glutamate toxicity that is unimpaired by exposure of neurons to FCCP.

Discussion We have shown that transient inhibition of mitochondrial function concurrent with glutamate exposure significantly reduces delayed neurotoxicity. These results indicate that very high cyto369

Glu + Rot + Oligo

Mean basal

Glu + FCCP

Rot + Oligo

1998 Nature America Inc. http://neurosci.nature.com

articles

plasmic calcium concentrations alone are not toxic to neurons, but that potential-driven uptake of calcium into the mitochondria is essential to trigger excitotoxic neuronal death. Consistent with our conclusion that blockade of calcium uptake into the mitochondria is neuroprotective, the onset of neuronal cell death is delayed by inclusion of rotenone and oligomycin in an experimental design that uses prolonged glutamate exposures to induce toxicity29. However, in contrast to the results reported here, calcium influx into the cytoplasm was inhibited with rotenone and oligomycin treatment in that study29. It is therefore possible that a reduction of the glutamateinduced calcium load that occurs after mitochondrial inhibition with rotenone and oligomycin (by an unknown mechanism) may underlie the neuroprotection observed in this previous study. Although we used FCCP to inhibit mitochondrial calcium uptake, FCCP may have several effects in addition to uncoupling mitochondria. FCCP does inhibit glutamate-stimulated intracellular acidification (Fig. 4), but this is probably not the mechanism of neuroprotection, as other experimental conditions that blocked glutamate-induced acidification (oligomycin and rotenone or pH 8.5, Fig. 4b) were not neuroprotective (Fig. 3a and d). Furthermore, FCCP exposure by itself caused an intracellular acidification comparable to that induced by glutamate (Fig. 4) but was not toxic (Fig. 3a and b), which also indicates that intracellular acidification is not the primary determinant of neuronal cell injury. In addition, FCCP treatment reduces the neuronal ATP/ADP ratio within minutes30. However, this alone is evidently not neurotoxic, because five- to ten-minute applications of FCCP did not injure neurons (Fig. 3a and b, and ref. 22). Thus, it seems that the protective effects of FCCP are mostly attributable to its ability to modify mitochondrial calcium transport. Although the effects of FCCP on intracellular pH might be viewed as unfortunate confounding side effects of this drug related to its protonophore nature, results presented here and previously31 indicate that glutamate-induced acidification is a direct result of mitochondrial calcium uptake. Therefore, all mitochondrial inhibitors, not just the protonophore inhibitors, should prevent glutamate-induced acidification by enhancing mitochondrial membrane depolarization and consequently inhibiting potentialdependent mitochondrial calcium uptake. Instead, it seems that the transience of mitochondrial inhibition by the uncouplers used in the present study is the key to neuroprotection. We used Magfura 2 because of its ability to report more accurately the high [Ca2+]i associated with glutamate stimulation in the presence of mitochondrial inhibitors. More conventional dyes such as Fura 2 would underestimate changes of this magnitude32,33. Although we have previously used Magfura 2 as an indicator of changes in intracellular free magnesium concentrations34
%[3H] arginine converted to [3H] citrulline Glu Buffer Buffer + N-Arg

1998 Nature America Inc. http://neurosci.nature.com

([Mg2+]i), Magfura 2 is obviously sensitive to high levels of cytoplasmic calcium in addition to normal [Mg2+]i. We obtained results similar to Fig. 2 in neurons loaded with Calcium Green 5N. Because this dye is insensitive to changes in [Mg2+]i33,35, it is unlikely that the results we obtained here with Magfura 2 are attributable to a large alteration in any magnesium response. There is efflux of calcium from the mitochondria into the cytoplasm during recovery from a glutamate stimulation in these cells14, and the time required for recovery to baseline [Ca2+]i depends on the extent of mitochondrial calcium loading36,37. Moreover, calcium can be rapidly cleared from the cytoplasm, and recovery to baseline [Ca2+]i is enhanced if mitochondrial calcium efflux is blocked pharmacologically14. Therefore, the enhanced recovery after stimulation with glutamate and FCCP compared with that after glutamate stimulation alone is consistent with previous findings. With FCCP present during stimulation, mitochondrial calcium uptake should have been reduced or eliminated. Thus, during recovery there would be very little mitochondrial calcium efflux to delay the return to baseline cytoplasmic calcium levels. The results shown here indicate that blocking mitochondrial calcium uptake substantially reduces glutamate toxicity. The mechanism of mitochondrial mediation of neuronal injury remains uncertain. Glutamate stimulates increases in reactive oxygen species (ROS) production, and the mitochondria are the postulated source of these free radicals1820. FCCP effectively prevents the glutamate-mediated generation of ROS18, which would be consistent with ROS being the critical mediator of injury. We have demonstrated here that excitotoxic cell death can be dissociated from nitric oxide production. Although there is evidence that NO is the mediator of calcium-depennature neuroscience volume 1 no 5 september 1998

370

Glu + FCCP + N-Arg

Fig. 5. FCCP does not block glutamate-stimulated increases in NOS activity. Neurons were preloaded with radiolabelled arginine and treated with buffer, with 100 M glutamate and 10 M glycine, with 100 M glutamate, 10 M glycine and 750 nM FCCP or with 750 nM FCCP alone for 10 min. The percent of L-[3H]arginine converted to L-[3H]citrulline was determined from cell extracts immediately after glutamate exposure. Each bar represents the mean results from three to six experiments done in triplicate. As indicated, neurons were pretreated with the competitive NOS inhibitor N-nitro-L-arginine methyl ester (N-Arg, 300 M) for 10 min before and during L-[3H]arginine loading and stimulation. Asterisks (*) indicate NOS activity was significantly different from buffer-treated controls (p < 0.05, one-way ANOVA with Bonferroni multiple comparisons test for post-hoc comparisons). Double asterisks (**) indicate results were significantly different from the same stimulation performed in the absence of N-Arg (p < 0.05, one-way ANOVA with Bonferroni multiple comparisons test for post-hoc comparisons).

FCCP

FCCP + N-Arg

Glu + N-Arg

Glu + FCCP

1998 Nature America Inc. http://neurosci.nature.com

articles

articles

p44 p42

b
Normalized active MAPK immunoreactivity (%)

Buffer

Glu

Glu + FCCP

FCCP

1998 Nature America Inc. http://neurosci.nature.com

Fig. 6. FCCP does not alter glutamate-stimulated increases in MAPK activity. (a) Representative anti-active MAPK western blot of neurons treated with buffer, with 100 M glutamate and 10 M glycine, with 100 M glutamate, 10 M glycine and 750 nM FCCP or with 750 nM FCCP alone for 10 min. Cell lysates were collected immediately after stimulation. (b) Normalized active MAPK immunoreactivity of p42 MAPK (open bars) and p44 MAPK (hashed bars). Each bar represents the mean results from six experiments. Asterisks (*) indicate MAPK activity was significantly different from buffer-treated controls (p < 0.05, one-way ANOVA with Bonferroni multiple comparisons test for post-hoc comparisons).

Buffer

Glu

Glu + FCCP

FCCP

dent glutamate-stimulated cell death4,38, there are also several conflicting reports7,3941. However, the ultimate determinant of toxicity may be the formation of peroxynitrite rather than NO 42. Thus, the neuroprotection observed with FCCP here would be consistent with this hypothesis, as the inhibition of mitochondrial superoxide production that occurs with FCCP could reduce peroxynitrite production. The finding that mitochondrial calcium uptake is essential for neuronal injury indicates that inhibition of the calcium uniporter may provide a new approach to protection against excitotoxicity. Currently available inhibitors have serious limitations for this purpose. For example, ruthenium red penetrates cells poorly and has very limited specificity for the uniporter compared with the ryanodine receptor in intact cells. However, related compounds may eventually prove useful for neuroprotection. Mitochondria are increasingly thought to contribute to apoptotic cell death. After activation of apoptotic cascades, factors released by mitochondria (including cytochrome c) activate caspases, which execute the apoptotic death pathway4346. In our experiments, cytochrome c might be released, and caspases activated as a result. However, the models in which these mechanisms have been established do not require large increases in [Ca2+]i to trigger cell death, nor are they associated with rapid mitochondrial depolarization or almost immediate superoxide production43,47 as observed in glutamate-treated neurons. Thus, it may be inappropriate to assume that cytochrome c release and caspase activation occur in our experiments, especially as glutamate toxicity seems to be a predominantly necrotic form of injury. The full extent of the similarities and differences between the various forms of injury mediated by mitochondria remains to be established.
Methods CELL CULTURE. Primary cultures were prepared from Sprague-Dawley rat fetuses at embryonic day 17 as described48. Dissociated forebrain neurons were plated on poly-L-lysine-coated 31-mm glass coverslips in Dulnature neuroscience volume 1 no 5 september 1998

beccos Modified Eagles medium (DMEM) supplemented with 10% fetal bovine serum, penicillin (24 U/ml) and streptomycin (24 g/ml). After 24 h, the medium was replaced with one containing 10% horse serum, and the coverslips were inverted to inhibit proliferation of glia. These culture conditions provide the sparse, glia-poor neuronal cultures that are necessary for optimal results in fluorescence microscopy experiments. Neurons were used after 1418 days in culture. All procedures using animals were in accordance with the NIH Guide for the Care and Use of Laboratory Animals and were approved by the University of Pittsburghs Institutional Animal Care and Use Committee. SOLUTIONS. For perfusion of coverslips in the fluorescence microscopy experiments and for stimulation in the toxicity and MAPK assays, we used a HEPES-buffered salt solution (HBSS, adjusted to pH 7.4 with NaOH): 137 mM NaCl, 5 mM KCl, 10 mM NaHCO3, 20 mM HEPES, 5.5 mM glucose, 0.6 mM KH2PO4, 0.6 mM Na2HPO4, 1.4 mM CaCl2 and 0.9 mM MgSO4. A different HEPES-buffered salt solution (NOS buffer, adjusted to pH 7.4 with NaOH) was used for stimulation in the NOS assay: 140 mM NaCl, 5.4 mM KCl, 25 mM HEPES, 5.6 mM glucose, 1.4 mM CaCl2 and 0.9 mM MgCl2. Stimulation was terminated in the NOS assay with a stop buffer: 118 mM NaCl, 4.7 mM KCl, 24.8 mM NaHCO3, 1.2 mM KH2PO4, 4 mM EDTA and 5 mM L-arginine. In the MAPK assay, cells were scraped and sonicated in a homogenization buffer (adjusted to pH 7.5 with HCl): 50 mM Tris, 1 mM EDTA and 1 mM EGTA. Sodium pyrophosphate (2 mM), p-nitrophenyl phosphate (4 mM), sodium orthovanadate (1 mM), aprotinin (100 ng/ml), benzamidine (10 M) and leupeptin (100 ng/ml) were also added to the homogenization solution immediately before use. MEASUREMENT OF MITOCHONDRIAL MEMBRANE POTENTIAL. Semi-quantitative estimates of the mitochondrial membrane potential were made using the dye JC-1 (Molecular Probes, Eugene, Oregon) as described14. Dye loading was achieved by incubation in HBSS containing 3 M JC-1 and 0.5% DMSO at 37C for 20 min. After loading, coverslips were incubated with HBSS for 20 min at room temperature (2025 C) and then mounted on the stage of an ACAS 570c confocal microscope (Meridian Instruments, Okemos, Michigan). Fields were illuminated using the 488-nm line of an argon laser, and emission at 530 and 590 was monitored. Emitted light was passed through a pinhole to limit data acquisition to a horizontal slice that was approximately 1 m in thickness. Ratios were calculated by dividing the signal at 590 nm by the signal at 530 nm after background
371

1998 Nature America Inc. http://neurosci.nature.com

articles

subtraction. Ratios were also normalized to a starting value of 1 for comparison between cells. A decrease in the normalized JC-1 fluorescence ratio indicates mitochondrial membrane depolarization. The coverslips were not continuously perfused, and solution changes were made by aspirating and replacing the contents of the recording chamber. MEASUREMENT OF [Ca2+]i AND pHi. Stock solutions (1 mM) of the acetoxy methyl esters of Magfura 2 and BCECF (Molecular Probes, Eugene, Oregon) were prepared in anhydrous dimethyl sulfoxide (DMSO). Dye loading was achieved by incubation in HBSS containing 5 M Magfura 2 or BCECF, 5 mg/ml bovine serum albumin and 0.5% DMSO at 37C for 20 min. After being loaded, coverslips were rinsed with HBSS, mounted in a recording chamber and perfused with HBSS at a rate of 20 ml/min. There was an exchange of the 1 ml chamber volume about every three seconds with this high flow rate. All recordings were made at room temperature (2025C). Background fluorescence values (determined from cell-free regions of each coverslip) were subtracted from all signals. The imaging system used in these studies consisted of a Nikon Diaphot 300 microscope fitted with a 40 quartz objective, a Dage-MTI cooled-CCD camera with 640 480-pixel resolution in combination with a Dage-MTI Gen II Sys image intensifier, a software package from Compix (Cranberry, Pennsylvania), and a 75 watt Xenon lamp-based monochromator light source (Applied Scientific Instrumentation, Eugene, Oregon). This light source can switch to any wavelength between 260 and 680 nm in about 50 ms under computer control using a bandwidth of 12 nm, and the imaging software package can acquire the full 640 480 camera field at up to 10 frames per second. Cells were alternately illuminated with 335 nm and 375 nm light for Magfura 2 measurements or with 498 nm and 450 nm light for BCECF measurements. Attenuation of incident light was achieved with neutral density filters (ND 2 -/+ ND 1 for 1% or 0.1% transmittance in the Magfura 2 and BCECF experiments, respectively; Omega Optical, Brattleboro, Vermont). For both the Magfura 2 and the BCECF measurements, emitted light passed through a 515-nm dichroic mirror and a 535 12.5-nm band-pass filter (Omega Optical, Brattleboro, Vermont). TOXICITY ASSAYS. Coverslips were inverted to orient the cultured neurons face-up, and the cells were rinsed twice with sterile HBSS. After a 5- or 10-min pre-incubation with HBSS alone, neurons were exposed to the appropriate stimulation solutions for 5 or 10 min. After stimulus exposure, neurons were returned to the incubator in Minimum Essential Medium containing penicillin (24 U/ml) and streptomycin (24 g/ml) for 2024 h. In these experiments, the results represent the mean values obtained from three coverslips for each treatment condition. Neuronal viability was assessed using two methods. For the data shown in Fig. 3a, neuronal viability was assayed by calcein loading. Cells were incubated in 5 M Calcein AM (Molecular Probes, Eugene, Oregon) for 1 h, and the fluorescence at 530 nm was measured using a CytoFluor II platereader (PerSeptive Biosystems, Framingham, Massachusetts). Only viable neurons should be able to cleave and trap the fluorescent calcein product. Viability was expressed as a percentage of the signal obtained from coverslips exposed to buffer changes alone. For the data shown in Fig. 3bd, neuronal viability was assayed by cell counting. Cells were incubated with phosphate-buffered saline (PBS) containing 0.4% trypan blue for five minutes. Coverslips were then washed two or three times with PBS and fixed in 4% paraformaldehyde for 10 min. Cells excluding trypan blue were counted from four or more fields from each coverslip by a researcher blind to their treatment. It was not possible to mark specific fields on the glass coverslips and count the cells from these designated fields before and after stimulation. Also, there was some loss of adherent fields due to coverslip flipping and manipulation that was independent of the exposure condition being tested. For these reasons, we typically counted cells from optimal fields on each coverslip; this practice of counting from the best fields on each coverslip probably underestimates the cell death caused by glutamate stimulation.
NOS ASSAY. NOS activity was determined by measuring the formation of L[3H]citrulline from L-[3H]arginine using a modification of a published method49. Coverslips were rinsed twice with NOS buffer, inverted to orient the cultured neurons face up, and transferred to clean 6-well culture plates containing 1 ml of NOS buffer per well. Cells were pre-incubated

in NOS buffer for 10 min. Neurons were preloaded with L-[2,3,4,53H]arginine monohydrochloride (3 Ci/well, Amersham) for 10 min and then stimulated for 10 min. The competitive NOS inhibitor N-nitroL-arginine methyl ester hydrochloride (N-Arg, 300 M; Research Biochemical International, Natick, Massachusetts) was included during pre-incubation, L-[3H]arginine loading, and stimulation in some experiments. After stimulation, the stimulation solution was removed, the cells were rinsed twice with ice-cold stop buffer, and 1 ml of 0.3 N HClO4 was added to each well. After 10 min, the HClO4 extract was neutralized with K2CO3. L-[3H]Arginine was separated from L-[3H]citrulline by filtration of 500 l cell extract aliquots through columns containing 2 ml of Dowex AG50WX-8 (Na+ form, Bio-Rad, Richmond, California). The ion-exchange columns were rinsed three times with 2 ml of water each time. L-[3H]citrulline content of the eluants was determined by liquid scintillation spectrosocopy. Total uptake of L-[3H]arginine was determined by liquid scintillation spectroscopy of 50 l aliquots of the cell extracts. NOS activity was expressed as percent conversion of L-[3H]arginine to L-[3H]citrulline.
MAPK ASSAY. Coverslips were rinsed twice with HBSS, inverted to orient the

1998 Nature America Inc. http://neurosci.nature.com

cultured neurons face up, and transferred to clean 6-well culture plates containing 1 ml of HBSS per well. Cells were pre-incubated in HBSS for 10 min and then stimulated for 10 min. After stimulation, the stimulation solution was removed, the cells were rinse twice with ice cold phosphatebuffered saline, and 100 l of homogenization solution was added to each well. Cells were scraped, and the cell scrapings were sonicated and stored at 70C. Protein concentration of the samples was measured by the Bradford method using bovine serum albumin as the standard. Equivalent amounts of protein for each sample were resolved by 10% SDSPAGE, blotted electrophoretically to Immobilon membranes, and incubated in Tris-buffered saline with Tween-20 [50 mM Tris-HCl (pH 7.5-8), 150 mM NaCl, and 0.1% Tween-20] containing 3% bovine serum albumin for 30 min. Blots were then incubated with an active MAPK polyclonal antibody (1:20,000 dilution; Promega) followed by incubation with horseradish peroxidase-linked goat anti-rabbit IgG (1:4,000 dilution; Amersham) and developed using enhanced chemiluminiscence (ECL, Amersham). Densitometric analysis of the active MAPK immunoreactivity was done using NIH Image software. STATISTICS. Results were compared by one-way analyses of variance with Bonferroni multiple comparisons test for post-hoc comparisons (InStat 3.0, GraphPad Software, San Diego, CA).

Acknowledgements
This work was supported by NIH grants NS 34138 (I.J.R.), NS 09998 (A.K.S.) and NS 34007 (E.K.). I.J.R. is an Established Investigator of the American Heart Association.

RECEIVED 24 JUNE: ACCEPTED 23 JULY 1998


1. Choi, D. W. Ionic dependence of glutamate neurotoxicity. J. Neurosci. 7, 369379 (1987). 2. Tymianski, M., Charlton, M. P., Carlen, P. L. & Tator, C. H. Source specificity of early calcium neurotoxicity in cultured embryonic spinal neurons. J. Neurosci. 13, 20852104 (1993). 3. Hartley, D. M., Kurth, M. C., Bjerkness, L., Weiss, J. H. & Choi, D. W. Glutamate receptor-induced 45Ca2+ accumulation in cortical culture correlates with subsequent neuronal degeneration. J. Neurosci. 13, 19932000 (1993). 4. Dawson, V. L., Dawson, T. M., London, E. D., Bredt, D. S. & Snyder, S. H. Nitric oxide mediates glutamate neurotoxicity in primary cortical cultures. Proc. Natl Acad. Sci. USA 88, 63686371 (1991). 5. Siman, R., Noszek, J. C. & Kegerise, C. Calpain I activation is specifically related to excitatory amino acid induction of hippocampal damage. J. Neurosci. 9, 15791590 (1989). 6. Monyer, H., Hartley, D. M. & Choi, D. W. 21-aminosteroids attenuate excitotoxic neuronal injury in cortical cell cultures. Neuron 5, 121126 (1990). 7. Lafon-Cazal, M., Pietri, S., Culcasi, M. & Bockaert, J. NMDA-dependent superoxide production and neurotoxicity. Nature 364, 535537 (1993). 8. Coyle, J. T. & Puttfarcken, P. Oxidative stress, glutamate, and neurodegenerative disorders. Science 262, 689695 (1993).

372

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

9. Dykens, J. A. Isolated cerebral and cerebellar mitochondria produce free radicals when exposed to elevated Ca2+ and Na+: implications for neurodegeneration. J. Neurochem. 63, 584591 (1994). 10. Patel, M., Day, B. J., Crapo, J. D., Fridovich, I. & McNamara, J. O. Requirement for superoxide in excitotoxic cell death. Neuron 16, 345355 (1996). 11. Nicholls, D. G. & Akerman, K. E. O. Mitochondrial calcium transport. Biochim. Biophys. Acta 683, 5788 (1982). 12. Gunter, T. E., Gunter, K. K., Sheu, S.-S. & Gavin, C. E. Mitochondrial calcium transport: Physiological and pathological relevance. Am. J. Physiol. Cell Physiol. 267, C313C339 (1994). 13. Ankarcrona, M. et al. Glutamate-induced neuronal death: a succession of necrosis or apoptosis depending on mitochondrial function. Neuron 15, 961973 (1995). 14. White, R. J. & Reynolds, I. J. Mitochondrial depolarization in glutamatestimulated neurons: An early signal specific to excitotoxin exposure. J. Neurosci. 16, 56885697 (1996). 15. Schinder, A. F., Olson, E. C., Spitzer, N. C. & Montal, M. Mitochondrial dysfunction is a primary event in glutamate neurotoxicity. J. Neurosci. 16, 61256133 (1996). 16. Gunter, T. E. & Pfeiffer, D. R. Mechanisms by which mitochondria transport calcium. Am. J. Physiol. Cell Physiol. 258, C755C786 (1990). 17. Nicholls, D. G. in Molecular Mechanisms of Ischemic Brain Damage (eds Kogure, K., Hossman, K. A., Siesj, B. K. & Welsh, F. A.) 97106 (Elsevier, Amsterdam, 1985). 18. Reynolds, I. J. & Hastings, T. G. Glutamate induces the production of reactive oxygen species in cultured forebrain neurons following NMDA receptor activation. J. Neurosci. 15, 33183327 (1995). 19. Dugan, L. L. et al. Mitochondrial production of reactive oxygen species in cortical neurons following exposure to N-methyl-D-aspartate. J. Neurosci. 15, 63776388 (1995). 20. Bindokas, V. P., Jordan, J., Lee, C. C. & Miller, R. J. Superoxide production in rat hippocampal neurons: selective imaging with hydroethidine. J. Neurosci. 16, 13241336 (1996). 21. Arkles, B. & Brinigar, W. S. Respiratory properties of rat liver mitochondria immobilized on an alkylsilyated glass surface. J. Biol. Chem. 250, 88568862 (1975). 22. Wang, G. J., Richardson, S. R. & Thayer, S. A. Intracellular acidification is not a prerequisite for glutamate-triggered death of cultured hippocampal neurons. Neurosci. Lett. 186, 139144 (1995). 23. Hoyt, K. R. & Reynolds, I. J. Alkalinization prolongs recovery from glutamateinduced increases in [Ca2+]i by enhancing Ca2+ efflux through the mitochondrial Na+/Ca2+ exchanger in cultured rat forebrain neurons. J. Neurochem. (in press). 24. Favaron, M. et al. Gangliosides prevent glutamate and kainate neurotoxicity in primary neuronal culture of neonatal rat cerebellum and cortex. Proc. Natl Acad. Sci. USA 85, 73517355 (1988). 25. Dykens, J. A., Stern, A. & Trenkner, E. Mechanism of kainate toxicity to cerebellar neurons in vitro is analogous to reperfusion tissue injury. J. Neurochem. 49, 12221228 (1987). 26. Bading, H. & Greenberg, M. E. Stimulation of protein tyrosine phosphorylation by NMDA receptor activation. Science 253, 912914 (1991). 27. Kawasaki, H. et al. Activation and involvement of p38 mitogen-activated protein kinase in glutamate-induced apoptosis in rat cerebellar granule cells. J. Biol. Chem. 272, 1851818521 (1997). 28. Guyton, K. Z., Liu, Y., Gorospe, M., Xu, Q. & Holbrook, N. J. Activation of mitogen-activated protein kinase by H2O2: role in cell survival following oxidant injury. J. Biol. Chem. 271, 41384142 (1996). 29. Budd, S. L. & Nicholls, D. G. Mitochondria, calcium regulation and acute glutamate excitotoxicity in cultured cerebellar granule cells. J. Neurochem. 67, 22822291 (1996). 30. Budd, S. L. & Nicholls, D. G. A reevaluation of the role of mitochondria in neuronal Ca2+ homeostasis. J. Neurochem. 66, 403411 (1996).

31. Wang, G. J., Randall, R. D. & Thayer, S. A. Glutamate-induced intracellular acidification of cultured hippocampal neurons demonstrates altered energy metabolism resulting from Ca 2+ loads. J. Neurophysiol. 72, 25632569 (1994). 32. Hyrc, K., Handran, S. D., Rothman, S. M. & Goldberg, M. P. Ionized intracellular calcium concentration predicts excitotoxic neuronal death: observations with low affinity fluorescent calcium indicators. J. Neurosci. 17, 66696677 (1997). 33. Stout, A. K. & Reynolds, I. J. High-affinity calcium indicators underestimate increases in intracellular calcium concentrations associated with excitotoxic glutamate stimulations. Neuroscience (in press). 34. Brocard, J. B., Rajdev, S. & Reynolds, I. J. Glutamate induced increases in intracellular free Mg 2+ in cultured cortical neurons. Neuron 11, 751757 (1993). 35. Rajdev, S. & Reynolds, I. J. Calcium green 5N, a novel fluorescent probe for monitoring high intracellular free Ca2+ concentrations associated with glutamate excitotoxicity in cultured rat brain neurons. Neurosci. Lett. 162, 149152 (1993). 36. White, R. J. & Reynolds, I. J. Mitochondria accumulate Ca 2+ following intense glutamate stimulation of cultured rat forebrain neurones. J. Physiol. (Lond.) 498, 3147 (1997). 37. Hoyt, K. R., Stout, A. K., Cardman, J. M. & Reynolds, I. J. An evaluation of intracellular sodium and mitochondria in the buffering of kainateinduced intracellular free calcium changes in rat forebrain neurons. J. Physiol. (Lond.) 509, 103116 (1998). 38. Dawson, V. L., Kizushi, V. M., Huang, P. L., Snyder, S. H. & Dawson, T. L. Resistance to neurotoxicity in cortical cultures from neuronal nitric oxide deficient mice. J. Neurosci. 16, 24792487 (1996). 39. Hewett, S. J., Corbett, J. A., McDaniel, M. L. & Choi, D. W. Inhibition of nitric oxide formation does not protect murine cortical cell cultures from N-methyl-D-aspartate neurotoxicity. Brain Res. 625, 337341 (1993). 40. Pauwels, P. J. & Leysen, J. E. Blockade of nitric oxide formation does not prevent glutamate-induced neurotoxicity in neuronal cultures from rat hippocampus. Neurosci. Lett. 143, 2730 (1992). 41. Demerl-Pallardy, C., Lonchampt, M.-O., Chabrier, P.-E. & Braquet, P. Absence of implication of L-arginine/nitric oxide pathway on neuronal cell injury induced by L-glutamate or hypoxia. Biochem. Biophys. Res. Commun. 181, 456464 (1991). 42. Beckman, J. S., Beckman, T. W., Chen, J., Marshall, P. A. & Freeman, B. A. Apparent hydroxyl radical production by peroxynitrite: implications for endothelial injury from nitric oxide and superoxide. Proc. Natl Acad. Sci. USA 87, 16201624 (1990). 43. Newmeyer, D. D., Farschon, D. M. & Reed, J. C. Cell-free apoptosis in Xenopus egg extracts: inhibition by Bcl-2 and requirement for an organelle fraction enriched in mitochondria. Cell 79, 189192 (1994). 44. Yang, J. et al. Prevention of apoptosis by Bcl-2: release of cytochrome c from mitochondria blocked. Science 275, 11291132 (1997). 45. Susin, S.A. et al. Bcl-2 inhibits the mitochondrial release of an apoptogenic protease. J. Exp. Med. 184, 13311341 (1996). 46. Kluck, R. M., Bossy-Wetzel, E., Green, D. R. & Newmeyer, D. D. The release of cytochrome c from mitochondria: a primary site for Bcl-2 regulation of apoptosis. Science 275, 11321136 (1997). 47. Kim, C. N. et al. Overexpression of Bcl-X(L) inhibits ara-C-induced mitochondrial loss of cytochrome c and other pertubations that activate the molecular cascade of apoptosis. Cancer Res. 57, 31153120 (1997). 48. White, R. J. & Reynolds, I. J. Mitochondria and Na+/Ca2+ exchange buffer glutamate-induced calcium loads in cultured cortical neurons. J. Neurosci. 15, 13181328 (1995). 49. Bredt, D. S., Mourey, R. J. & Snyder, S. H. A simple, sensitive, and specific radioreceptor assay for inositol 1,4,5-trisphosphate in biological tissues. Biochem. Biophys. Res. Comm. 159, 976982 (1989).

1998 Nature America Inc. http://neurosci.nature.com

nature neuroscience volume 1 no 5 september 1998

373

1998 Nature America Inc. http://neurosci.nature.com

articles

A null mutation in TGF- leads to a reduction in midbrain dopaminergic neurons in the substantia nigra
Mariann Blum
Fishberg Research Center for Neurobiology, Mt. Sinai School of Medicine, One Gustave Levy Place, Box 1065, New York, New York 10029, USA Correspondence should be addressed to M.B. (blum@cortex.neuro.mssm.edu)

1998 Nature America Inc. http://neurosci.nature.com

Transforming growth factor (TGF)- is neurotrophic for midbrain dopaminergic neurons in vitro. Here I investigated whether a null mutation in the TGF- gene affects the normal development or survival of dopaminergic neurons in either the substantial nigra (SN) or the ventral tegmental area (VTA). The SN of TGF- knockout mice contained 50% fewer dopaminergic neurons than the control SN, but VTA neuron number was unchanged. In addition, the overall volume of the dorsal striatum was reduced by 20%. Newborn mice showed a similar decrease in the number of SN dopaminergic neurons, suggesting that TGF- is unlikely to regulate developmental neuron death. These studies indicate that TGF- is required for the normal proliferation or differentiation of a select population of dopaminergic neurons within the SN.

The midbrain contains three populations of dopaminergic neurons. One group, in the substantia nigra (SN), projects to the dorsal striatum. These neurons are important for the control of movement and degenerate in Parkinsons disease. A second group of dopaminergic neurons, in the ventral tegmental area (VTA), projects to the ventral striatum and frontal cortex. The VTA neurons are involved in reward and cognitive behaviors and have been implicated in substance abuse and schizophrenia. Much investigation has been focused on identifying factors that may prevent neurodegeneration or stimulate the proliferation of dopaminergic progenitor cells because it may lead to new treatments for Parkinsons disease or to better understanding of the normal development of dopaminergic neurons. Many neurotrophic factors enhance the survival of midbrain dopaminergic neurons in vitro or stimulate the proliferation of cultured mesencephalic neuronal progenitors. Endogenous expression of many of these neurotrophic factors in the basal ganglia suggests that they may have a physiological role in the development or survival of dopaminergic neurons. However, disruption of genes encoding two putative dopaminergic neurotrophic factors, brain-derived neurotrophic factor and glial cell line-derived neurotrophic factor, did not seem to affect the differentiation or survival of dopaminergic neurons15. Transgenic mice have been generated with a null mutation in the gene encoding another putative dopaminergic neurotrophic factor, transforming growth factor- (TGF-)6,7, a member of the epidermal growth factor (EGF) family that binds and transduces its signal via the EGF receptor. TGF- is important for determining skin architecture and patterns of hair growth. Here I report that the lack of TGF- reduces dopaminergic neuron number in the substantia nigra but not in the ventral tegmental area, probably by regulating neuron proliferation or differentiation.

Results TGF-/ mice7 on a mixed C57Bl/6 129 genetic background were purchased from Jackson Laboratories. To minimize genet374

ic differences, F2 littermates from a C57Bl/6 129 cross breeding were used for TGF-+/+ control mice. The midbrain and forebrain of young adult mice (approximately 90 days old) were stained with antibodies recognizing the rate-limiting enzyme for catecholamine biosynthesis, tyrosine hydroxylase (TH). To obtain an unbiased estimate of the number of dopaminergic neurons in the SN and VTA, a stereological probe called an optical fractionator was used8,9. Counts of dopaminergic neurons showed that there were almost half the number of TH-immunoreactive neurons in the SN of TGF-/ mice, compared to control mice, with no difference in neuron number in the VTA (Fig. 1). To control for possible differences in genetic background between TGF-/ and the control B129F2/J mice, I analyzed neuron number in the SN and VTA of two additional strains of mice. Although the number of dopaminergic neurons varied across genetic strains, the number of dopaminergic neurons in the SN and VTA within a strain were always equal in the three genetic backgrounds thus far analyzed (Table 1). A similar observation has been reported10 for unbiased cell counts on another mouse strain (CBA/J). Thus, cell counts from wild-type mice from four different genetic backgrounds suggest that the nearly 50% difference in the number of dopaminergic neurons in the SN compared with the VTA in the TGF- / mice is due to the absence of TGF- rather than to the genetic background. To determine whether the decrease in TH-immunoreactive cells observed in the SN results from an actual loss of nigrostriatal dopaminergic neurons or merely from a decrease in TH expression, fluorescently labeled microspheres were injected into the dorsal striatum. Because all SN neurons projecting to the striatum are dopaminergic, the detection of retrogradely labeled cells in the SN that are not TH-immunoreactive would suggest that the decrease in cell counts results from a loss of TH expression and not a loss of dopaminergic neurons. However, all retrogradely labeled cells expressed TH (data not shown), suggesting that the decrease in TH-immunoreactive cells results
nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

+/+

+/+

-/-

-/-

1998 Nature America Inc. http://neurosci.nature.com

Fig. 1. TH-immunoreactive cell counts. (a) TH immunocytochemistry from coronal section containing the SN in a TGF-+/+ mouse. (b) High-power view of the SN shown in (a). (c) TH immunocytochemistry from a coronal section containing the VTA in a TGF-+/+ mouse. (d) TH immunocytochemistry from a coronal section containing the SN in a TGF-/ mouse. (e) High-power view of the SN shown in (d). (f) TH immunocytochemistry from a coronal section containing the VTA in a TGF-/ mouse. (g) TH-immunoreactive cell counts from the SN. The data (n = 4) are presented as the mean standard error. (h) Cell counts from the VTA. A statistically significant decrease in TH-immunoreactive cells was detected in the SN (p 0.05, t-test) of TGF-/ mice compared to controls, with no difference in the VTA. Scale bar, 1000 m in (a), (c), (d) and (f), and 250 m in (b) and (e).

g
TH-immunoreactive cells

TGF-+/+

h
TH-immunoreactive cells

TGF-+/+

from a loss of dopaminergic neurons. To further support this idea, Nissl counts of neurons in the SN were performed. Similarly to the TH-immunoreactive cell counts, a 50% reduction in the number of Nissl-stained SN neurons was observed in TGF-/ (13,347 1,370) compared to TGF-+/+ mice (27,593 2470, p < 0.01, t-test). To address whether the absence of TGF- expression may lead to the degeneration of SN dopaminergic neurons in the TGF/ mice, the dopaminergic innervation pattern to the striatum was analyzed, for changes can be readily observed following the degeneration of dopaminergic neurons. For example, after the death of the nigrostriatal dopaminergic neurons that occurs as a result of the weaver mutation, dopaminergic innervation density is greatly decreased in the dorsal striatum, whereas the pattern in the ventral striatum remains nearly normal11. However, when the striatum of TGF-/ mice was immunostained with TH, no difference in the density of the THimmunoreactive innervation pattern Table 1. Dopaminergic cell numbers across genetic backgrounds in the SN and VTA. was observed (Fig. 2). This observaSwiss Webster C57Black/6 CBA C57Black/6 S129 tion was further supported by bioSN:VTA Ratio 1.23 0.94 1.02 chemical studies measuring dopamine 4,389 666 6,589 790 11,395 1,614 uptake into striatal synaptosomes via SN neuron number 3,564 454 7,044 1,684 11,184 916 the high-affinity dopamine trans- VTA neuron number
nature neuroscience volume 1 no 5 september 1998

porter, which is a quantitative measure reflecting dopaminergic terminal density. No difference was detected in the concentration of uptake sites in either the dorsal (4.75 0.071 versus 4.73 0.43) or ventral striatum (1.72 0.23 versus 1.86 0.20 fmols/min/g protein; mean -/standard error). However, although there was no change in the density of dopaminergic terminals, the lateral ventricle was greatly enlarged in TGF-/ mice, suggesting that the size of the striatum may be decreased (Fig. 2). Thus, the volume of the dorsal and ventral striatum was measured independently. The border between the dorsal and ventral striatum was established by drawing a line from the bottom of the lateral ventricle diagonally to the lateral border of the olfactory tubercle, and volumes were measured using the estimation method of Cavalieri12. The dorsal striatum was near-/TGF- ly 20% smaller in TGF- / compared with TGF-+/+ mice (2.96 0.13 versus 3.58 0.24 mm3, p 0. 05, t-test). However, there was no difference in the volume of the ventral striatum (1.26 0.09 versus 1.39 0.11 mm3), which is innervated by the unaffected dopaminergic neurons residing in the VTA. This decrease in the volume of the dorsal striatum explains why there is no apparent change in the density of TGF--/the dopaminergic innervation in TGF/ mice; fewer dopaminergic neurons innervating a smaller volume would result in no change in the innervation density. It is not clear, however, whether the reduction in the size of the dorsal striatum is a consequence of fewer SN dopaminergic neurons or a direct effect of the TGF- null mutation on the striatum. The appearance of dopaminergic terminals in the developing striatum during the peak of striatal neurogenesis13,14, in combination with dopamine receptor expression in striatal neural progenitors15, has led to the suggestion that the dopaminergic innervation to the striatum may be involved in defining its final target size. Thus, if fewer SN dopaminergic neurons are born in the TGF-/ mouse, this could subsequently affect the extent of proliferation or differentiation of striatal neuronal progenitors. On the other hand, TGF-/EGF receptor expression is found in the striatal ventricular zone of both rats and mice1618, so lack of TGF-, rather than decreased dopaminergic innervation, may regulate neurogenesis. However, the greatest level of TGF-/EGF

+/+

375

1998 Nature America Inc. http://neurosci.nature.com

articles

Fig. 2. TH immunocytochemistry of the striatum. (a) TGF-+/+ mice. (b) TGF-/ mice. Scale bar, 1 mm.

receptor expression seems to occur after the peak of striatal neurogenesis, so TGF- is more likely to be involved in modulating the proliferation of glial cells18. In accordance with this finding, there are fewer astrocytes in the striatum of another TGF- mutant, waved-1, in which there is a 90% reduction in TGF-16. In addition, infusion of EGF into adult animals stimulates the proliferation of astrocytes19. Therefore, in the TGF-/ mouse, there may be fewer astrocytes in the striatum, which may possibly account for the reduction in size. Alternatively, TGF- may be required for the survival of striatal GABAergic interneurons, which express TGF- /EGF receptors 20 . Further analysis is required to determine which of these possibilities accounts for the reduction in the volume of the dorsal striatum. EGF receptor expression has been detected in the rat midbrain as early as embryonic day 14 in the germinal zone in the region of the developing SN and VTA precursors18. Thus, TGF may be important for modulating the proliferation or differentiation of dopaminergic neuronal progenitors. In rats, TH-immunoreactive neurons in the SN undergo a period of programmed cell death during the first and second postnatal weeks21. Therefore, to address whether the absence of TGF- is affecting the proliferation and/or differentiation of SN dopaminergic neurons or modulating programmed cell death, mice were collected at birth. Counts of TH-immunoreactive cells in the SN of mice at postnatal day 1 showed that the number of dopaminergic neurons in TGF-/ mice was already significantly reduced at this young age (Fig. 3). In addition, the number of TH-immunoreactive cells in TGF-+/+ mice at postnatal day 1 was the same as

1998 Nature America Inc. http://neurosci.nature.com

that found in adult mice, supporting the observation10 that midbrain dopaminergic neurons do not undergo programmed cell death in mice, in contrast to rats. To address whether mice may undergo a period of programmed cell death at an earlier developmental stage, and to determine whether the absence of TGF- leads to cell death, embryos from both TGF-+/+ and TGF-/ mice were collected on embryonic day 14, approximately two days after the majority of dopaminergic neurons have been born22 and examined for the presence of cell death using TUNEL (TdT-mediated dUTP-X nick end labeling, Boehringer Mannheim). Although TUNEL-positive cells could be detected in the dorsal region above the ventricle throughout the rostral caudal extent of the midbrain, no TUNEL-positive cells were found in the ventral midbrain, which contained TH-immunoreactive cells in embryos from either phenotype (data not shown). Thus, these data suggest that dopaminergic neurons in mice do not go through a period of programmed cell death and that TGF- is necessary for the proliferation and/or differentiation of SN dopaminergic neurons.

TH-immunoreactive cells

Discussion A question unanswered in this study is what distinguishes the remaining 50% of SN dopaminergic neurons in the TGF-/ mouse from those that are lost. Possibly, TGF- regulates the number of times dopaminergic progenitors divide before exiting the cell cycle, as the loss of a single cell cycle would decrease the number of dopaminergic neurons by half. However another possibility, which other studies indicate is more likely, is that TGF- only acts on a subpopulation of the dopaminergic progenitor pool. In rats, not all TH-positive cells in the germinal zone express TGF-/EGF receptors, and the peak of expression occurs after a large portion of dopaminergic neurons have been born18. In addition, in vitro studies support the idea that the stimulation of proliferating midbrain progenitors through the TGF-/EGF receptor does not occur during the entire phase of development. Thus, EGF can stimulate the proliferation of neuronal progenitors in cultures from embryonic day 16 (ref. 23) but not from lessdeveloped embryonic day 12 rat midbrains24. Thus, these studies suggest that TGF- may be important for the proliferation of the late-generated dopaminergic neurons. Interestingly, it is these late-generated dopaminergic neurons that are first affected in the weaver mutant mouse22. Thus, these findings suggest that TGF- is required for the development of the most vulnerable population of SN dopaminergic neurons. This is particularly relevant to the intensive effort to generate a readily available source of dopaminergic neurons that can be used to transplant into patients with Parkinsons disease to replace lost nigrostriatal dopaminergic neurons, as the results of this study suggest that TGF- may be a useful factor for stimulating the production of dopaminergic neurons.
Methods IMMUNOCYTOCHEMISTRY. Adult mice were transcardially perfused with 4% paraformaldehyde. Brains were collected and cut into 34 mm blocks containing the midbrain and striatum and postfixed for five hours. After decapitation, the brains were collected from postnatal day (PN) 1 mice and immersion fixed for five hours. Subsequently, the brain blocks were placed in cold 30% sucrose overnight, and 40 m vibratome sections were cut from the adult and PN 1 midbrain blocks. The forebrain blocks were frozen in OCT (Lipshaw), and 30 m cryosections were cut. Using a random start, 1:8 (adult) and 1: 6 (PN1) series of sections were collected for TH immunocytochemistry. Free-floating sections were incubated overnight at room temperature with a rabbit-anti-rat TH polyclonal antibody (Pel-freeze, 1:1000 in PBS, 3% goat serum and 0. nature neuroscience volume 1 no 5 september 1998

TGF-+/+

TGF-/

Fig. 3. TH-immunoreactive cell counts in the SN of postnatal day 1 TGF-+/+ and TGF-/ mice. The data (n = 4) are presented as the mean standard error, and the decrease in the TGF-/ mice is statistically significant (p 0. 01, t-test).
376

1998 Nature America Inc. http://neurosci.nature.com

articles

3% Triton X-100). After removal of the primary antibody and washing with PBS, the sections were incubated for two hours at room temperature with an anti-rabbit IgG conjugated with biotin (Amersham, 1:200 in PBS, 3% goat serum and 0.3% Triton X-100). The secondary was removed, and sections were washed, quenched for five minutes in 0. 3% H2O2 in PBS and incubated with ExtraAvidin-peroxidase (Sigma, 1:200) for one hour before being incubated with 3,3-diaminobenzidine. STEREOLOGY. To count the TH-immunoreactive cells and to estimate the volume of the striatum, a computer-assisted image analysis system was used. The system consisted of a Zeiss Axiophot photomicroscope equipped with a Zeiss MSP65 computer-controlled motorized stage, a ZVS-47E video camera, a Macintosh 840AV workstation, and NeuroZoom morphometry software developed in collaboration between the Neurobiology of Aging Laboratories at Mt. Sinai School of Medicine and the Scripps Research Institute25. The midbrain sections were viewed at low power (10x objective), and the SN and VTA were outlined, using the medial lemniscal fiber bundle to demarcate the border between them. Then at a random start, the number of TH-immunoreactive cells in a counting frame (90 125 m) were counted at 150-m steps in the xy axis, and through a dissector height of 10 m, which was analyzed in 2m steps in the z-axis8,9. A 63x oil-immersion objective (NA 1. 25) and condensor (NA 1. 4) were used for these cell counts. After all TH immunoreactive cells (Q) were counted, the total number (Ntotal) of dopaminergic neurons in the SN or VTA was calculated according to the equation, (Ntotal) = (Q) thickness/height 1/(areal sampling fraction) 1/(section sampling fraction). The volumes, not corrected for tissue shrinkage, of the dorsal and ventral striatum were determined from a 1:8 series of sections immunostained with TH by the method of Cavalieri12.

Acknowledgements
I would like to thank John Morrison and members of his laboratory for support and discussions concerning the stereological methods used in this study, Peter Rapp for comments on the manuscript, and Susan Lasky and Jeremy Kay for technical support. This work was supported by NIH grant AG08538.

RECEIVED 6 MAY: ACCEPTED 29 JULY 1998


1. Ernfors, P., Lee, K.-F. & Jaenisch, R. Mice lacking brain-derived neurotrophic factor develop with sensory deficits. Nature 368, 147150 (1994). 2. Jones, K. R., Farias, I., Backus, C. & Reichardt, L. F. Targeted disruption of the BDNF gene perturbs brain and sensory neuron development but not motor neuron development. Cell 76, 989999 (1994). 3. Moore, M. W. et al. Renal and neuronal abnormalities in mice lacking GDNF. Nature 382, 7679 (1996). 4. Snchez, M. P., Silos-Santiago, I., Frisn, J., Lira, S. A. & Barbacid, M. Renal agenesis and the absence of enteric neurons in mice lacking GDNF. Nature 382, 7073 (1996). 5. Pichel, J. G. et al. Defects in enteric innervation and kidney development in

mice lacking GDNF. Nature 382, 7376 (1996). 6. Luetteke, N. C. et al. TGF- deficiency results in hair follicle and eye abnormalities in targeted and waved-1 mice. Cell 73, 263278 (1993). 7. Mann, G. B. et al. Mice with a null mutation of the TGF- gene have abnormal skin architecture, wavy hair and curly whiskers and often develop corneal inflammation. Cell 73, 249261 (1993). 8. West, M. J. & Gundersen, H. J. G. Unbiased stereological estimation of the number of neurons in the human hippocampus. J. Comp. Neurol. 296, 122 (1990). 9. West, M. J. New stereological methods for counting neurons. Neurobiol. Aging 14, 275285 (1993). 10. Lieb, K. et al. Pre- and postnatal development of dopaminergic neuron numbers in the male and female mouse midbrain. Dev. Brain Res. 94, 3743 (1996). 11. Graybiel, A. M., Ohta, K. & Roffler-Tarlov, S. Patterns of cell and fiber vulnerability in the mesostriatal system of the mutant mouse weaver. I. Gradients and compartments. J. Neurosci. 10, 720733 (1990). 12. Gundersen, H. J. G. et al. The new stereological tools: Disector, Fractionator, Nucleator and point sampled intercepts and their use in pathological research and diagnosis. Acta Pathol. Microbiol. Immunol. Scand. 96, 857881 (1988). 13. Bayer, S. A. Neurogenesis in the rat neostriatum. Int. J. Dev. Neurosci. 2, 163175 (1984). 14. van der Kooy, D. & Fishell, G. Neuronal birthdate underlies the development of striatal compartments. Brain Res. 401, 155161 (1987). 15. Schambra, U. B. et al. Ontogeny of D1A and D2 dopamine receptor subtypes in rat brain using in situ hybridization and receptor binding. Neuroscience 62, 6585 (1994). 16. Weickert, C. S. & Blum, M. Striatal TGF-: Postnatal developmental expression and evidence for a role in the proliferation of subependymal cells. Dev. Brain Res. 86, 203216 (1995). 17. Seroogy, K. B., Gall, C. M., Lee, D. C. & Kornblum, H. I. Proliferative zones of postnatal rat brain express epidermal growth factor receptor mRNA. Brain Res. 670, 157164 (1995). 18. Kornblum, H. I. et al. Prenatal ontogeny of the epidermal growth factor receptor and its ligand, transforming growth factor alpha, in the rat brain. J. Comp. Neurol. 380, 243261 (1997). 19. Kuhn, H. G., Winkler, J., Kempermann, G., Thal, L. J. & Gage, F. H. Epidermal growth factor and fibroblast growth factor-2 have different effects on neural progenitors in the adult rat brain. J. Neurosci. 17, 58205829 (1997). 20. Kornblum, H. I., Gall, C. M., Sergoogy, K. B. & Lauterborn, J. C. A subpopulation of striatal GABAergic neurons expresses the epidermal growth factor receptor. Neuroscience 69, 10251029 (1995). 21. Oo, T. F. & Burke, R. E. The time course of developmental cell death in phenotypically defined dopaminergic neurons of the substantia nigra. Dev. Brain Res. 98, 191196 (1997). 22. Bayer, S. A. et al. Selective vulnerability of late-generated dopaminergic neurons of the substantia nigra in weaver mutant mice. Proc. Natl Acad. Sci. USA 92, 91379140 (1995). 23. Mytilineou, C., Park, T. & Shen, J. Epidermal growth factor-induced survival and proliferation of neuronal precursor cells from embryonic rat mesencephalon. Neurosci. Lett. 135, 6266 (1992). 24. Bouvier, M. M. & Mytilineou, C. Basic fibroblast growth factor increases division and delays differentiation of dopamine precursors in vitro. J. Neurosci. 15, 71417149 (1995). 25. Bloom, F. E., Young, W. G., Nimchinsky, E. A., Hof, P. R. & Morrison, J. H. in Neuroinformatics - An Overview of the Human Brain Project (eds Koslow, S. H. & Huerta, M. F.) 83123 (Lawrence Erlbaum, Mahwah, 1997).

1998 Nature America Inc. http://neurosci.nature.com

nature neuroscience volume 1 no 5 september 1998

377

1998 Nature America Inc. http://neurosci.nature.com

articles

Presynaptic long-term depression at a central glutamatergic synapse: a role for CaMKII


Troy W. Margrie1, John A. P. Rostas1 and Pankaj Sah2,3
1

The Neuroscience Group and 2The Disciplines of Medical Biochemistry and Human Physiology, Faculty of Medicine and Health Sciences, University of Newcastle, Callaghan NSW, 2308, Australia Current address: Division of Neuroscience, John Curtin School of Medical Research, GPO Box 334, ACT 2601, Australia Correspondence should be addressed to P.S. (pankaj.sah@anu.edu.au)

1998 Nature America Inc. http://neurosci.nature.com

CaMKII is a calcium-activated kinase that is abundant in neurons and has been strongly implicated in memory and learning. Here we show that low-frequency stimulation of glutamatergic afferents in hippocampal slices from juvenile domestic chicks results in long-term depression of synaptic transmission. This reduction does not require activation of NMDA or metabotropic glutamate receptors and does not require a rise in postsynaptic calcium. However, buffering presynaptic calcium prevents the reduction of the excitatory postsynaptic potential or current that is induced by low-frequency stimulation. In addition, application of the calmodulin antagonist calmidazolium, or the specific CaMKII antagonist KN-93, completely blocks long-term depression. These findings demonstrate a newly discovered form of long-term synaptic depression in the avian hippocampus.

Long-term potentiation (LTP) and long-term depression (LTD) are two extensively studied forms of synaptic plasticity. Although most studies have focused on the mammalian brain, the avian brain offers a powerful tool to study the cellular mechanisms of learning and memory 1,2. When compared with those of the mammalian forebrain, synapses in the avian brain undergo a prolonged period of maturation during the first ten weeks after hatching3. In the first two weeks of development, during which the formation of most synapses in the forebrain is completed, there is an initial enrichment of cytosolic calcium-calmodulinstimulated protein kinase II (CaMKII). This is followed by the synapse maturation phase, in which CaMKII undergoes a slow translocation from the cytosol to particulate structures, including neuronal membranes and the cytoskeleton3,4. CaMKII is thus enriched in the cytosol during the first two weeks after hatching, after which it is translocated to the plasma membrane. CaMKII is a multifunctional calcium-dependent serinethreonine kinase that is abundant in central nervous tissue. At synapses, it is present in both the presynaptic and postsynaptic compartments, where it can phosphorylate proteins involved in transmitter release and reception. After its activation by calcium/calmodulin, CaMKII can become autophosphorylated on threonine 286/7, causing the enzyme to remain active independent of calcium/calmodulin binding and increase its affinity for calcium/calmodulin 5. These changes in activity may allow CaMKII to act as a molecular switch to trigger longterm synaptic change68. Our understanding of the postsynaptic actions of CaMKII comes mostly from studies of mammalian hippocampal CA1 pyramidal neurons. In these cells, activation of CaMKII or injection of a constitutively active form of CaMKII into the postsynaptic cell9,10 increases the amplitude of glutamatergic EPSCs. Blocking CaMKII by either pharmacological11,12 or genetic13
378

means blocks the induction of LTP. These data indicate that postsynaptic activation of CaMKII produces a long-lasting increase in excitatory synaptic transmission. In the presynaptic terminal, injection of CaMKII in invertebrates14,15, or in rat brain synaptosomes16, increases transmitter release. Therefore, activation of presynaptic CaMKII may produce an increase in transmitter release. However, the effects of physiological activation of presynaptic CaMKII are not known. Tetanic stimulation of glutamatergic afferents in the adult chicken hippocampus results in a presynaptic form of LTP17. Here we show that low-frequency stimulation in the juvenile chick hippocampus results in a long-term reduction in synaptic transmission. This LTD seems to rely on activation of a presynaptic CaMKII.

Results Stimulation of afferents in the avian hippocampus activates glutamatergic synapses that, like their mammalian counterparts, show components mediated by both NMDA and non-NMDA glutamate receptor subtypes. Tetanic stimulation of these afferents in hippocampal slices from adult chicks results in LTP17. In slices from young chickens, brief, low-frequency stimulation (LFS; 0.5 Hz, 150 or 300 stimuli) caused a longlasting reduction in the slope of the EPSP (Fig. 1a). This reduction in the EPSP was maintained for the duration of the recording (up to 80 minutes) and was readily reversed by a high-frequency train (n = 3, Fig. 1c), showing that synapses that had undergone LTD could be repotentiated. LTD was saturable and could also be reversed by bath-application of forskolin, an activator of adenylyl cyclase (n = 2; data not shown). The average magnitude of LTD measured 25 minutes after LFS was 41% 8 of the baseline (n = 9; Fig. 1b). To determine if LTD was limited to inputs that had undergone low-frequency stimulation, we stimulated two
nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

a
% EPSP slope

c
% EPSP slope

Time (min)

Time (min)

b
% EPSP slope

d
% EPSC amplitude

1998 Nature America Inc. http://neurosci.nature.com

Time (min)

Time (min)

Fig. 1. Low-frequency stimulation (LFS) induces long-term depression (LTD) of synaptic transmission. (a) Raw data for a single cell showing the timecourse of LTD. Averages of five consecutive sweeps for the times indicated by the numbers (1 and 2) are shown below the graph. (b) Average LTD for nine cells. The magnitude of LTD, measured 25 minutes after LFS, was 41% 8 of baseline. (c) LTD can be reversed by tetanic stimulation. The initial slope of the EPSP is plotted for one cell against time. LFS leads to a sustained decrease in the EPSP slope, which is readily reversed by tetanic stimulation (100 Hz, 1 second; asterisk). (d) LTD is input-specific. Two independent inputs (S1, open circles, and S2, filled circles) stimulated the same neuron. After a baseline period of 10 minutes, LFS was delivered to S2. Raw sweep data showing superimposed EPSCs before and after LFS for S1 and S2 from one cell are shown above the graph. The short, solid horizontal lines (ad) show the time during which LFS (0.5 Hz) was applied.

independent inputs to the same cell. After obtaining a stable baseline, we delivered LFS to one of the inputs. LTD was confined to inputs that had undergone LFS, showing that like LTP at these synapses17, this form of LTD is input-specific (n = 3; Fig. 1d). Application of the NMDA receptor antagonist D-APV (30 M) completely abolished the component of the excitatory postsynaptic current mediated by NMDA receptors (n = 5; data not shown). However, in the presence of D-APV (Fig. 2a and b), LTD was not significantly different from control (49% 6; n = 5; p > 0.05), indicating that NMDA receptors are not required in this form of LTD. To determine if a rise in postsynaptic calcium was necessary for the induction of LTD, we loaded cells with the calcium chelator BAPTA (n = 5, Fig. 3a and b). After recording stable baselines for at least 15 minutes, we delivered LFS. BAPTA injection had no effect on LTD induction. Although chelation of postsynaptic

Fig. 2. NMDA-receptor activation is not required for LTD. (a) Raw data from an experiment in which LTD was tested in the presence of the NMDA receptor antagonist APV (50 M). Sample traces at the times indicated (1 and 2) are shown below the graph. (b) The time-course of LTD in cells in hippocampal slices that were superfused with APV (n = 5). There was no significant difference in the average LTD response between APVtreated and control cells (49% 6 versus 41% 8; p > 0.05). The short, solid horizontal lines show the time during which LFS (0.5 Hz) was applied. The cross-hatched bars indicate the time during which DAPV was superfused over the hippocampal slices.

a
% EPSP slope

Time (min)

b
% EPSP slope

Time (min)

calcium with sharp intracellular electrodes is effective at preventing calcium-dependent LTP in other systems18, an inadequate amount of the BAPTA may have diffused out of the electrode to the dendritic spines. To improve delivery of BAPTA, we used low-resistance whole-cell patch pipettes (n = 4; Fig. 3c and d) containing BAPTA (10 mM) to chelate postsynaptic calcium. Cells were voltage-clamped at 70mV during the LFS to minimize any contribution of voltage-gated calcium channels. LTD was not affected by chelation of postsynaptic calcium when BAPTA was introduced through whole-cell electrodes (46% 6, p > 0.05). We confirmed the ability of BAPTA to buffer dendritic calcium transients by imaging calcium in the dendritic tree (Fig. 4). Under control conditions, action potentials result in a rapid rise in intracellular calcium, which decays to baseline in both the soma and the dendrite. Similar calcium rises could be recorded over the entire dendritic tree (n = 3). Inclusion of 10 mM BAPTA in the internal solution produced considerably reduced calcium rises. No changes in internal calcium were found in the dendrites in any of the three cells studied. In addition, action potentials were broader after loading with BAPTA, providing electrophysiological evidence that BAPTA had diffused into the postsynaptic cell17 (Fig. 4, inset). In seven cells, large-scale disruption of postsynaptic processes by recording with whole-cell electrodes filled with cesium fluoride also failed to block the induction of LTD (data not shown). Fluoride strongly binds calcium ions on its own, inhibits calcium currents19,20 and disrupts intracellular enzymatic events mediated by G proteins and protein phosphatases21,22. These data indicate that metabolic processes in the postsynaptic cell are unlikely to be important for the induction of LTD at these synapses. Therefore, we next studied the presynaptic terminal. At the mossy-fiber synapse in area CA3 of the mammalian hippocampus, LFS reduces transmitter release by a mechanism involving a presynaptic metabotropic glutamate receptor,
379

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

a
% EPSP slope

c
% EPSC amplitude

Time (min)

Time (min)

b
% EPSP slope

d
% EPSC amplitude

Fig. 3. A rise in postsynaptic calcium is not required for LTD. (a) Raw data from a single cell that was impaled with a microelectrode containing 200 mM BAPTA. (b) There is no statistical difference between LTD in control hippocampal slices and LTD recorded from cells filled with BAPTA (n = 5). (c) Raw data of LTD recorded from a cell loaded with the calcium chelator BAPTA (10 mM) through a whole-cell patch pipette. The cell was voltage-clamped at 70 mV throughout the experiment. (d) There is no difference in LTD between BAPTA-filled and control cells (46% 6 versus 41% 8; n = 4). Sample sweeps at the times indicated (1 and 2) are shown between the graphs. The short, solid horizontal lines (ad) show the time during which LFS (0.5 Hz) was applied. The cross-hatched bars indicate the time of BAPTA application.

1998 Nature America Inc. http://neurosci.nature.com

Time (min)

Time (min)

mGluR223. Because mGluR2 is negatively coupled to cAMP turnover, and LTP at these synapses requires increases in cAMP24, LTD in area CA3 might result from the mGluR-mediated downregulation of cAMP25. Like mammalian mossy fibers, glutamatergic inputs in the chicken hippocampus also show potentiation in response to forskolin application17. Thus, a mechanism similar to that acting at mossy fibers may also apply to the avian hippocampus. We therefore applied the mGluR antagonist MCPG (1 mM) for 10 minutes before delivery of the LFS. LTD in the presence of MCPG was not significantly different from LTD under control conditions (52% 12 versus 41% 8; n = 5; p > 0.05; Fig. 5a and b). To determine if a rise in presynaptic calcium was necessary for the induction of LTD, we superfused hippocampal slices with the membrane-permeant calcium chelator BAPTA-AM. BAPTAAM is hydrolyzed by cytosolic esterases and trapped intracellu-

larly as the active chelator BAPTA. Superfusion of BAPTA-AM (5 M) produced a small reduction in the amplitude of the EPSP (21% 4; n = 5; p < 0.05), confirming that BAPTA had gained access to the presynaptic terminals. LTD was completely blocked in the presence of BAPTA-AM (Fig. 6a and b), showing that buffering of intra-terminal calcium during LFS prevents the downregulation of transmitter release. Because BAPTA-AM completely blocked LTD, whereas basal synaptic transmission was only reduced by 20%, the calcium requirements of these two processes must be distinct. As elevation of calcium in the presynaptic terminal during the LFS is required for LTD induction, calcium-binding protein(s) probably play an important part in the mechanism underlying LTD. Calmodulin has a high affinity for calcium and, after calcium binding, is responsible for activation of a number of important enzymes26. To investigate the involvement of calmodulin in LTD, we superfused hippocampal slices with the Control BAPTA membrane-permeable calmodulin antagonist, calmidazolium (20 M) for 15 minutes before LFS. Calmidazolium had no effect on basal synaptic transmission but significantly reduced LTD (n = 4; p < 0.05; Fig. 7a and b), indicating the involvement of a calmodulin-dependent step in the reduction of transmitter release. A main target of calcium/calmodulin in neurons is CaMKII. CaMKII is present in chick forebrain synaptosomes and can phosphorylate proteins in presynaptic terminals4,27. To determine if CaMKII is involved in LTD, we used KN-93, a specific inhibitor of CaMKII that acts by competing for the calmodulin binding site on the enzyme28,29. Bath-application of KN-93 (5 M) had no effect on basal synaptic transmission, but prevented Fig. 4. BAPTA buffers calcium in the dendritic tree. Calcium transients evoked in induction of LTD (Fig. 7c). Indeed, in the presence response to current injection (300 ms; 0.3 nA) are shown recorded from the soma of KN-93, a small potentiation of synaptic trans(top row) and 100 m away in the dendritic tree (middle row). Cells were loaded mission was apparent after LFS (127% 20 of with 50 M Oregon green with (right) or without (left) 10 mM BAPTA. Action baseline; n = 4). KN-92, which exerts the nonspepotentials evoked in response to current injection are shown in the bottom panels. cific actions of KN-93 but does not block activaThe inset (lower right corner) shows superimposed action potentials recorded with tion of CaMKII, had no effect on LTD compared an electrode containing 10 mM BAPTA immediately after gaining intracellular access with control hippocampal slices (43% 9 versus and six minutes later (asterisk). 41% 8, respectively; n = 4; p > 0.05; Fig. 6d).
380 nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

1998 Nature America Inc. http://neurosci.nature.com

Fig. 5. Activation of metabotropic glutamate receptors is not required for LTD induction. (a) Raw data from an experiment in which LTD was tested in the presence of mGluR antagonist MCPG (1 mM). Averaged traces corresponding to the times indicated (1 and 2) are shown between the graphs. (b) LTD in the presence of MCPG was not significantly different from control (52% 12 versus 41 8%; n = 5; p > 0.05). The short, solid horizontal lines show the time during which LFS (0.5 Hz) was applied. The cross-hatched bars indicate the time during which MCPG was superfused over the hippocampal slices.

a
% EPSP slope

Time (min)

b
% EPSP slope

Time (min)

Loading the postsynaptic cell with KN-93 (10 M) had no significant effect on LTD (n = 4; p > 0.05; data not shown), indicating that the effects of bath-applied KN-93 were presynaptic.

Discussion Here we have shown that low-frequency stimulation of glutamatergic synapses in the avian hippocampus causes a long-lasting depression of synaptic transmission. This LTD does not require the activation of either NMDA receptors or metabotropic glutamate receptors and is unaffected by preventing increases in postsynaptic calcium. However, LTD is blocked by buffering presynaptic calcium. These data indicate that induction of LTD at these synapses is presynaptic and is triggered by a rise in intraterminal calcium. LTD was inhibited by bath-applicaa tion of a specific CaMKII inhibitor, whereas its postsynaptic application was without effect. These results indicate that activation of a presynaptic CaMKII can produce a long-lasting reduction in synaptic efficacy. CaMKII is a multifunctional protein kinase with a number of substrates in the presynaptic terminal, including the synapsins, rabphillin3A30, synaptotagmin31 and calcium channels32. Injection of CaMKII into rat brain synaptosomes 16 or the presynaptic terminal of the squid giant synapse15 causes an increase in transmitter release. Synapsin I, an abundant b synaptic vesicle protein, is a known target of CaMKII, and injection of synapsin I phosphorylated at the CaMKII site mimics the effect of CaMKII. It has therefore been proposed that activation of CaMKII might increase transmitter release by phosphorylating synapsin I14. However, analyses of mice with targeted deletion of the CaMKII gene indicate that presynaptic CaMKII activity may either potentiate
% EPSP slope % EPSP slope

or depress synaptic transmission depending on the pattern of synaptic activity6,33. The reduction in transmitter release may be secondary to trapping of calmodulin by autophosphorylated CaMKII34 and the subsequent reduction in calmodulindependent processes involved in transmitter release. The differential effects of large-scale introduction of CaMKII15,16 into the presynaptic terminal and its activation by physiological stimuli may result from a more selective activation of specific substrates by endogenous CaMKII. Calcium channels involved in transmitter release contain an intracellular structural element (the synprint site) that interacts with syntaxin and synaptosome-associated protein of 25 kDa (SNAP-25)35,36, which are proteins involved in vesicle docking and fusion. These interactions are disrupted by phosphorylation of the synprint site by CaMKII32, which results in a reduction of transmitter release37,38. Thus, the reduction in synaptic efficacy after low-frequency stimulation may result from phosphorylation of the synprint site on presynaptic calcium channels by CaMKII. Although the exact mechanism of action of CAMKII in reducing transmitter release remains to be elucidated, it is obvious that CaMKII is capable of modulating synaptic transmission to produce either a longterm potentiation11,12 or long-term depression, depending on the cell type or the subcellular pool involved. This adds a new dimension to the recently recognized ability of CaMKII to act as a bidirectional switch in synaptic plasticity33.
Methods Chickens (12 weeks old) were anesthetized with intraperitoneal sodium pentobarbitone (10 mg/kg) and decapitated. The brain was rapidly removed and immersed in ice-cold Ringer solution (119 mM NaCl, 2.5 mM KCl, 1.3 mM MgCl2, 4.5 mM CaCl2, 1.0 mM Na2PO4, 26.2 mM

Time (min)

Time (min)

Fig. 6. Induction of LTD requires a sustained rise in presynaptic calcium. (a) Raw data from a single cell recorded from a hippocampal slice that was superfused with the membrane-permeable calcium chelator BAPTA-AM (5 M). Application of BAPTA-AM has only a small effect on basal transmission but blocks the induction LTD. Sample traces at the times indicated (1 and 2) are shown below the graph. (b) The average response of cells treated with BAPTA-AM 20 minutes after LFS was not different from baseline (96% 8; n = 5; p > 0.05). The short, solid horizontal lines show the time during which LFS (0.5 Hz) was applied. The cross-hatched bars indicate the time during which BAPTAAM was superfused over the hippocampal slices.
381

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

1998 Nature America Inc. http://neurosci.nature.com

a c Fig. 7. Induction of LTD requires activation of presynaptic calcium/calmodulin-dependent protein kinase II. (a) Raw data from a single cell showing that bath application of the calmodulin antagonist calmidazolium (20 M) inhibits LTD. (b) Average data from cells recorded in the presence of calmidazolium (n = 4; filled circles) has been superimposed on Time (min) Time (min) data obtained in control conditions (open circles). Calmidazolium significantly inhibited LTD (82% 11 versus 41% 8; p < 0.05). (c) The response of a single cell to LFS in the presence of the specific CaMKII inhibitor KNd b 93. (d) Average data from cells recorded in the presence of KN-93 (n = 4; filled circles) has been superimposed on data recorded from cells superfused with the inactive control compound KN-92 (n = 4; open circles). KN-93 abolished LTD (average EPSP slope after LFS was 127% 20), whereas KN-92 had no effect on LTD (43% 9; p > 0.05). Time (min) Time (min) Sample traces at the times indicated (1 and 2) are shown between the graphs. The short, solid horizontal lines (ad) show the time during which LFS (0.5 Hz) was applied. The dotted bars indicate the times during which calmidazolium (a,b), K-92 (c) or KN-92/93 (d) were superfused over the hippocampal slices.
% EPSP slope % EPSP slope

NaHCO3, 11 mM glucose). Coronal slices containing the hippocampus were then prepared as described17. These procedures were in accordance with the guidelines of University of Newcastle Animal Ethics Committee. Slices were superfused with Ringer solution equilibrated with 95% O2, 5% CO2 to maintain a pH of 7.4. Intracellular recordings were made with sharp microelectrodes filled with 1 M or 3 M KCl or 1 M KMeSO4 (50120 M) using an Axoclamp2A (Axon Instruments, Foster City, California). For loading cells with BAPTA, electrode tips were loaded with 200 mM KBAPTA in 1 M KCl and then back-filled with KCl. Recording and stimulating electrodes were placed as described17. Afferent fibers were stimulated at 0.1 Hz during the baseline period, and LTD was induced with a single low-frequency train at 0.5 Hz for either 150 or 300 pulses. When two inputs were stimulated to test for input specificity, the independence of the two inputs was verified by testing for crossfacilitation between the inputs. Only inputs that showed no cross-facilitation were accepted for analysis. Cell input resistance was monitored on-line throughout the experiment. There were no consistent changes in input resistance or membrane potential after LFS. Experiments were done in both the presence and the absence of the GABAA channel blocker picrotoxin (50 M). In experiments in which no picrotoxin was added, electrodes were filled with KMeSO4. As picrotoxin had no effect on the ability to induce LTD, these data were pooled. Wholecell recordings were made with patch pipettes (24 M) filled with a solution of 117 mM CsGluconate, 10 mM CsCl, 8 mM NaCl, 10 mM HEPES, 2 mM ATP, 0.2 mM GTP and 10 mM BAPTA (pH 7.3). To load cells with KN-93, 10 M KN-93 was added to the above solution and BAPTA was omitted. BAPTA was also omitted from the internal solution for the experiments shown in Fig. 1d. Recordings were obtained using the blind approach39 and amplified using an Axopatch 2C amplifier (Axon Instruments, Foster City, California). The ability of postsynaptic BAPTA to buffer calcium was verified by imaging calcium transients in dendrites in response to action potentials. Calcium imaging was done using a monochromator-based illumination system (T.I.L.L. Photonics, Planegg, Germany), and frames were captured at a frequency of 33 Hz. Cells were loaded with the calcium indicator Oregon green (50 M; Molecular Probes, Eugene, Oregon). After the whole-cell configuration was achieved, dye was allowed to diffuse into the dendrites for at least five minutes. In response to action potentials evoked by somatic current injection, distinct rises in free calcium were recorded in dendrites at distances greater than 100 m from the
382

soma. No changes in free calcium could be detected in the dendrites when 10 mM BAPTA was included in the internal solution. Currents were filtered at 5 kHz and digitized at 10 kHz (ITC-16; Instrutech, Long Island, New York) using custom software running on a Macintosh computer under Igor (Wavemetrics, Lake Oswego, Oregon). Series resistance (620 M) was monitored on-line, and cells in which resistance changed by more than 15% during the experiment were discarded. In some experiments, cells were loaded with a cesium fluoride internal solution (117 mM CsF, 10 mM CsCl, 8 mM NaCl, 10 mM HEPES, 10 mM BAPTA, pH 7.3). All experiments were done at room temperature. All values are expressed as means standard error of EPSP slopes (or EPSC amplitude) measured 25 minutes after the end of LFS (unless otherwise stated), and levels of statistical significance were determined by paired Students t-test. Drugs used were D-APV (RBI Research Biochemicals, Natick, Massachusetts); MCPG, calmidazolium, KN-93, KN-92 (Calbiochem, La Jolla, California), BAPTA and BAPTA-AM (Molecular Probes, Eugene, Oregon).

Acknowledgements
This work was supported by grants from the National Health and Medical Research Council of Australia (J.R. and P.S.). P.S. is a Charles and Sylvia Viertel Senior Medical Research Fellow. T.M. was supported by an Australian Postgraduate Award.

RECEIVED 5 MARCH: ACCEPTED 17 JULY 1998


1. Rose, S. P. How chicks make memories: the cellular cascade from c-fos to dendritic remodelling. Trends Neurosci. 14, 390397 (1991). 2. Sherry, D. F., Jacobs, L. F. & Gaulin, S. J. C. Spatial memory and adaptive specialization of the hippocampus Trends Neurosci. 15, 298303 (1992). 3. Rostas, J. A. P. in Neural and Behavioral Plasticity: The Use of the Domestic Chick as a Model (ed. Andrew, R. J.) 177211 (Oxford University Press, New York, 1991). 4. Rostas, J. A. P. & Dunkley, P. R. Multiple forms and distribution of calcium/calmodulin stimulated protein kinase II in brain. J. Neurochem. 59, 11911202 (1992). 5. Braun, A. P. & Schulman, H. The multifunctional calcium/calmodulin dependent protein kinase: from form to function. Annu. Rev. Physiol. 57, 417445 (1995). 6. Mayford, M., Wang, J., Kandel, E. R. & Odell, T. J. CaMPKII regulates the frequency-response function of hippocampal synapses for the production of both LTD and LTP Cell 81, 891904 (1995).

nature neuroscience volume 1 no 5 september 1998

% EPSP slope

% EPSP slope

1998 Nature America Inc. http://neurosci.nature.com

articles

1998 Nature America Inc. http://neurosci.nature.com

7. Miller, S. G. & Kennedy, M. B. Relulation of brain type II Ca2+/calmodulindependent protein kinase by autophosphorylation: a Ca2+-triggered molecular switch. Cell 44, 861870 (1986). 8. Lisman, J. E. A mechanism for memory storage insensitive to molecular turnover: A bistable autophosphorylating kinase. Proc. Natl Acad. Sci. USA 82, 30553057 (1985). 9. Lledo, P.-M. L. et al. Calcium/calmodulin-dependent protein kinase II and long term potentiation enhance synaptic transmission by the same mechanism. Proc. Natl Acad. Sci. USA 92, 1117511179 (1995). 10. Wang, J.-H. & Kelly, P. T. Postsynaptic injection of Ca2+/CaM induces synaptic potentiation requiring CaMPKII and PKC activity. Neuron 15, 443452 (1995). 11. Malenka, R. C. et al. An essential role for postsynaptic calmodulin and protein kinase activity in long-term potentiation. Nature 340, 554557 (1989). 12. Malinow, R., Schulman, H. & Tsien, R. W. Inhibition of postsynaptic PKC or CaMKII blocks induction but not expression of LTP. Science 245, 862866 (1989). 13. Silva, A. J., Stevens, C. F., Tonegawa, S. & Wang, Y. Deficient hippocampal longterm potentiation in -calcium-calmodulin kinase II mutant mice. Science 257, 201206 (1992). 14. Llinas, R., Gruner, J. A., Sugimori, M., McGuiness, T. L. & Greengard, P. Regulation by synapsin I and Ca2+-calmodulin-dependent protein kinase II of transmitter release in squid giant synapse. J. Physiol (Lond.) 436, 257282 (1991). 15. Llinas, R. R., McGuinness, T. L., Leonard, C. S., Sugimori, M. & Greengard, P. Intraterminal injection of synapsin I or calcium/calmodulin-dependent protein kinase II alters neurotransmitter release. Proc. Natl Acad. Sci. USA 82, 30353039 (1985). 16. Nichols, R. A., Sihra, T. S., Czernik, A. J., Nairn, A. C. & Greengard, P. Calcium/calmodulin-dependent protein kinase II increases glutamate and noradrenaline release from synaptosomes. Nature 343, 647651 (1990). 17. Margrie, T. W., Rostas, J. A. P. & Sah, P. Long term potentiation of synaptic transmission in the avian hippocampus. J. Neurosci. 18, 12071216 (1998). 18. Lynch, G., Larson, J., Kelso, S., Barrionuevo, G. & Schottler, F. Intracellular injections of EGTA block induction of hippocampal long-term potentiation. Nature 305, 719721 (1983). 19. Kostyuk, P. G., Krishtal, D. A. & Pidoplichko, V. I. Effects of internal fluoride and phosphate on membrane currents during intracellular dialysis of nerve cells. Nature 257, 691693 (1975). 20. Kay, A. R., Miles, R. & Wong, R. K. S. Intracellular fluoride alters the kinetic properties of calcium currents facilitating the investigation of synaptic events in hippocampal neurons. J. Neurosci. 6, 29152920 (1986). 21. Wiseman, A. in Handbook of Experimental Pharmacology (ed. Smith, F. A.) 4897 (Springer-Verlag, New York, 1970). 22. Sternweis, P. C. & Gilman, A. G. Aluminum: a requirement for activation of the regulatory component of adenylate cyclase by fluoride. Proc. Natl Acad. Sci. USA 79, 48884891 (1982).

23. Yokoi, M. et al. Impairment of hippocampal mossy fiber LTD in mice lacking mGluR2. Science 273, 645647 (1996). 24. Weisskopf, M. G., Castillo, P. E., Zalutsky, R. A. & Nicoll, R. A. Mediation of hippocampal mossy fiber long-term potentiation by cyclic AMP. Science 265, 18781882 (1994). 25. Kobayashi, K., Manabe, T. & Takahashi, T. Presynaptic long-term depression at the hippocampal mossy fiber-CA3 synapse. Science 273, 648650 (1996). 26. Klee, C.B. in Calmodulin (eds Cohen, P. & Klee, C.B.) (Elsevier, Ameterdam, 1988). 27. Weinberger, R. P. & Rostas, J. A. Developmental changes in protein phosphorylation in chicken forebrain. II. Calmodulin stimulated phosphorylation. Brain Res. 471, 259272 (1988). 28. Sumi, M. et al. The newly synthesized selective Ca2+/calmodulin dependent protein kinase II inhibitor KN-93 reduces dopamine contents in PC12h cells. Biochem. Biophys. Res. Commun. 181, 968975 (1991). 29. Mamiya, N. et al. Inhibition of acid secretion in gastric parietal cells by the Ca2+/calmodulin-dependent protein kinase II inhibitor KN-93. Biochem. Biophys. Res. Commun. 195, 608615 (1993). 30. Sudhof, T. C. The synaptic vesicle cycle: a cascade of protein-protein interactions. Nature 375, 645653 (1995). 31. Pompoli, M. Synaptotagmin is endogenously phosphorylated by Ca2+/calmodulin kinase II in synaptic vesicles. FEBS Lett. 317, 8588 (1992). 32. Yokoyama, C. T., Sheng, Z.-H. & Catterall, W. A. Phosphorylation of the synaptic protein interaction site in N-type calcium channels inhibits interaction with SNARE proteins. J. Neurosci. 17, 69296938 (1997). 33. Chapman, P. F., Frenguelli, B. G., Smith, A., Chen, C.-M. & Silva, A. J. The Ca2+/calmodulin kinase II: A bidirectional modulator of presynaptic plasticity. Neuron 14, 591597 (1995). 34. Meyer, T., Hanson, P. I., Stryer, L. & Schulman, H. Calmodulin trapping by calcium-calmodulin-dependent protein kinase. Science 256, 11991202 (1992). 35. Sheng, Z. H., Rettig, J., Cook, T. & Catterall, W. A. Calcium-dependent interactions of N-type calcium channels with the synaptic core complex. Nature 379, 451454 (1996). 36. Sheng, Z. H., Retting, J., Takahashi, M. & Catterall, W. A. Identification of a syntaxin-binding site on N-type calcium channels. Neuron 13, 13031313 (1994). 37. Mochida, S., Sheng, Z.-H., Baker, C., Kobayashi, H. & Catterall, W. A. Inhibition of neurotransmission by peptides containing the synaptic protein interaction site of N-type Ca2+ channels. Neuron 17, 781788 (1996). 38. Rettig, J. et al. Alteration of Ca2+ dependence of neurotransmitter release by disruption of Ca2+ channel/syntaxin interaction. J. Neurosci. 17, 66476656 (1997). 39. Blanton, M. G., Lo Turco, J. J. & Kriegstein, A. R. Whole cell recording from neurons in slices of reptilian and mammalian cerebral cortex J. Neurosci. Methods 30, 203210 (1989).

nature neuroscience volume 1 no 5 september 1998

383

1998 Nature America Inc. http://neurosci.nature.com

articles

Mechanoelectrical transduction assisted by Brownian motion: a role for noise in the auditory system
Fernn Jaramillo1 and Kurt Wiesenfeld2
1 2

Dept. Physiology, 1648 Pierce Dr., Emory University School of Medicine, Atlanta, Georgia 30322, USA School of Physics, Georgia Institute of Technology, Atlanta, Georgia 30332, USA Correspondence should be addressed to F.J. (fjaram@physio.emory.edu)

1998 Nature America Inc. http://neurosci.nature.com

The organs of the vestibular, auditory and lateral line systems rely on a common strategy for the stimulation of their primary receptors, the hair cells: stimuli induce shear between hair cell epithelia and accessory structures to which hair bundles, the hair cells mechanosensitive organelles, are attached. The inner hair cells of the cochlea, whose hair bundles are not attached to the overlying tectorial membrane, are a notable exception. Because their hair bundles are not restrained, they undergo significant Brownian motion, a characteristic traditionally thought to blunt the sensitivity of hearing. Contrary to this view, the work reported here indicates that Brownian motion of the hair bundle serves to enhance the sensitivity of mechanoelectrical transduction.

The auditory system has traditionally been considered to be limited by noise13. Noise, in the form of random collisions of air molecules against the eardrum or of Brownian motion of cochlear components, was therefore thought to limit the sensitivity of the ear. However, such views were presented at a time when it was not known that random noise can increase a systems sensitivity to weak signals through stochastic resonance (SR). Although initially it seems counter-intuitive as a physical phenomenon, SR is by now well understood and has been documented in a wide variety of experiments in mechanical, electronic and optical systems4. One of the classical situations in which SR is known to occur is the switching between two states that are separated by an energy barrier. Such a situation, familiar to neuroscientists, can be used to model the behavior of many systems, including ion channels. A channels gate can be modeled as being in one of two low-energy states, open or closed, which are separated by an activation energy barrier. The channels natural stimulus (for example, voltage) serves to modulate the potential by alternately raising the wells relative to the barrier that separates them. Although a very weak signal will not excite the gate over the barrier, the presence of random noise will occasionally induce transitions between the wells and, as expected, increasing the noise induces more switching events. Surprisingly, the weak signal can entrain this hopping, resulting in regular transitions. Moreover, the regularity can improve with the addition of more noise, but this has an upper limit: excessive noise results in frequent transitions whose timing bears no relation to the weak signal5. Thus, optimal signal transduction occurs at some non-zero level of input noise. Not surprisingly, SR has been reported in several biological systems, including mechanoreceptors, ion channels and model neurons69. We thought that SR might have a role in the auditory system10,11, because of the unique way that inner hair cells work. Outnumbered nearly four to one by outer hair cells, inner hair cells nevertheless gather and transmit most of the auditory infor384

mation that reaches the brain. Because their hair bundles are not attached to the tectorial membrane, their stimulation probably results from the motion of surrounding fluid1214. However, an unrestrained hair bundle undergoes Brownian motion whose range is at least ten times larger than the motion resulting from near-threshold stimuli1517. How can transduction of small signals be effective in the presence of such large noise? The work presented here indicates that Brownian motion of the hair bundle provides an optimal noise level that enhances the sensitivity of mechanoelectrical transduction to weak signals.

Results Because hair bundles in auditory hair cells are thought to undergo Brownian motion, we sought to determine whether this random motion can enhance the mechanoelectrical transduction caused by very small hair bundle displacements. This requires that the small glass probes used to displace a hair bundle be controlled to a degree comparable to that of hair-bundle Brownian motion (that is, a few nanometers). We were able to control probe motion with sub-nanometer precision in the range of frequencies relevant to our experiments (Fig. 1). To find SR in mechanoelectrical transduction, we used the whole-cell recording technique18 to measure the transduction current signal-to-noise ratio (SNR) as a function of noise amplitude in hair cells of the frog sacculus. To generate noise with spectral characteristics matching those of hair-bundle Brownian motion, we used low-pass filtered (corner frequency, at which the signal is attenuated by 3 dB, 1000 Hz) white noise. Filtering the noise limits the high frequencies, which are absent from bundle Brownian motion because of viscous drag. The amplitude of the noise was chosen to encompass the hair bundles range of Brownian motion: interferometric measurements in amphibian hair cell bundles place the root-mean-square (r.m.s) of this motion at 23 nm (refs. 16,17). These estimates are consistent with what is expected, given thermal energy and hair bundle stiffnature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

ness, as well as with estimates a b derived from detailed modeling studies11,19.20. Noise was added to a small (amplitude, 2 nm; frequency, 300 or 500 Hz) periodic stimulus. The sum of stimulus and noise was conveyed to the hair bundle by a stiff glass probe, and the resulting mechanoelectrical transduction current was Deviation from rest (nm) Time (ms) subjected to Fourier analysis (Fig. 2). The results indicate that addition of a small d c amount of noise to the bundles motion leads to an enhancement of the SNR, as evidenced by the emergence of the signal peak from the background noise (Fig. 2b). As a control, we verified that current spectra generated when the stimulus probe was immediately adjacent to but Frequency (Hz) Probe noise (nm) uncoupled from the hair bundle caused no detectable peak (and therefore no SR) at 300 Fig. 1. Control of the stimulus probe. The noise of a stimulus probe at rest is low, and its position can be Hz. Similarly, SR was not controlled well at frequencies up to 1 kHz. (a) Representative sample of an unattached probes mechanical observed in mechanically noise. The probe tip was positioned a few micrometers above the bottom of a fluid-filled glass chamber. stimulated transducing cells After optical calibration, the position of the probe was monitored by sampling the differential photodiode when the probes noise was output. The data for this probe were fitted by a single Gaussian curve with a standard deviation of 0.89 nm. relatively high (> 5 nm) Noise for other probes ranged between 0.7 and 1.2 nm. (b) To verify the responsiveness of the stimulus because of uncontrolled apparatus, the piezoelectric driver was driven by a 10 nm (peak), 1601000 Hz sweeping stimulus. The mechanical vibrations, as in thin line shows the digitized command, whereas the symbols (connected by a dotted line) represent the experiments before the probe motion as estimated from the photodiodes differential output. (For clarity, only a small portion of improved mechanical stabi- the sweep is shown (580600 Hz).) (c) The difference between command (Y) and motion (X) for the lization of our recording setup. entire sweep is shown here. Deviations from the command fitted Gaussian distributions with standard deviations from about 1.0 nm at low frequencies to as high as 1.4 nm at high frequencies. (d) The SNR of The noise dependence of optically determined probe motion, as a function of noise added to a 2 nm, 500 Hz stimulus, which was the transduction SNR (Fig. 3a used to drive the piezoelectric stimulator. The SNR declines as a power function of the type y = a+xb, and b) is characteristic of SR. showing no evidence of SR. All noise levels are r.m.s. We can predict this dependence (Fig. 3a and b, continuous lines) from the gating spring model of mechanoeletrical transduction2. In this successbetween these states. For this transduction model, the SNR is ful model, elastic elements (gating springs) act directly on transrelated to noise (RT) (unpublished results) according to duction channels to open and close them. The gating spring 2 -G0/RT model assumes that the transduction channels exist in either a 2 G1 SNR (1) 8 (RT) f e closed or an open state (although the possibility of several closed states is allowed), with transitions being governed by the opening In its simplest form this model is an example of the two-state systems described above. However, SR is not limited to two-state (-(G0 - (t) G1)/RT) k+ = f e systems, and thus its presence in mechanoelectrical transduction and closing is compatible with multi-state models of gating. The noise level that leads to a maximum SNR ranged between (-(G0 + (t) G1)/RT) k- = f e 1.0 and 2.7 nm (1.9 0.61, n = 8 cells), which is comparable to rates. Here levels of Brownian motion in amphibian hair cells and to the 2nm estimate for Brownian motion of inner hair cells11,17. ThereR T f= fore, our data indicate that SR for mechanoelectrical transduction (L h) occurs within a physiologically relevant noise range, that due to corresponds to the pre-exponential term of Arrhenius rate law2,21, Brownian motion of inner hair cell bundles. The small amplitude of these fluctuations can be appreciated by comparing them where R is the gas constant, T is the temperature, L is Avogadros with the dimensions of a single stereocilium (one of the hair bunconstant, h is Plancks constant, is proportional to the perioddles individual hairs, several hundred nm in diameter) or with ic stimulus amplitude, G0 is the energy barrier separating the the bundles displacementresponse relationship (Fig. 3c). The closed and open states and G1 is the intrinsic energy difference
Frequency (samples per bin) Y-X (nm)

1998 Nature America Inc. http://neurosci.nature.com

nature neuroscience volume 1 no 5 september 1998

SNR

nm

385

1998 Nature America Inc. http://neurosci.nature.com

articles

Current spectral density (pA2/Hz)

nature of the stimulus is also relevant: stochastic resonance occurs most readily for limitingly low-amplitude stimuli. The amplitude we used (2 nm) was dictated by our ability to experimentally resolve the transduction signal in the presence of the whole-cell recording background noise. This amplitude, although small, is considerably larger than that believed to arise in response to threshold auditory stimulation15. Thus, SR might be more effective at near-threshold stimulus levels.

a
hair bundle noise 1.0 nm r.m.s.

Discussion How does this dependence of the SNR on noise occur? The weak periodic displacement affects the transition rates between the open and closed states of the transduction channel through its influence on the hair bundle (Fig. 3d). For very weak displacements, the transition rates are small and the channels can become trapped in the closed state, hampering signal transduction. Brownian motion serves to enhance hopping between the states, thus increasing the transition rates. Furthermore, the coherence of the transitions also increases over a certain range of added noise. Over this range, the signal (channel opening) rises more steeply than the noise, resulting in the increasing portion of the SNR curve. For still larger noise levels, the coherence is degraded, and so there is a peak in the SNR. Experiments such as these are particularly difficult to do in the more vulnerable inner hair cells. However, the fundamental features of mechanoelectrical transduction in amphibian saccular hair cells are similar to those found in the cochlea22,23. Thus, amphibian hair cells are an appropriate model in which to explore these issues. However, transduction differences between mammalian and amphibian hair cells, or the mechanics or hydrodynamics of inner hair cell bundles, could result in reduced or enhanced effectiveness of SR in the auditory system. It will therefore be important to establish the extent to which SR is present in the mammalian cochlea. How could the mechanical properties of mammalian cochlear hair bundles affect SR? The most important properties are a bundles stiffness and viscous drag coefficient. The stiffness of inner hair cell bundles in the mouse is similar to that measured in amphibians19,20,24. However, because hair bundle height (and therefore stiffness) varies along the mammalian cochlea, it is important to consider how stiffness could affect the extent to which Brownian motion modulates transitions between the open and closed states of the channel. As a first approximation, bundle stiffness is inversely proportional to the square of bundle height (assuming a constant number of stereocilia)20. Thus, a bundle half as long as a distant neighbor would be four times as stiff. Because Brownian motion is inversely proportional to the square root of stiffness, this stiffer bundles motion would be reduced only by one-half. Moreover, because of the geometrical arrangement of the transduction channels gating springs (thought to be the fine linkages that join the tips of adjacent stereocilia), a reduction in height would lead to a proportional increase in the bundles mechanical sensitivity25,26. Therefore, small variations in hair bundle height would have only have a minor effect on the bundles Brownian motion ability to influence channel gating. Differences in viscous drag in the cochlea could affect the range of stimulus frequencies within which SR is effective. In vestibular hair cells, viscous drag acting on the bundle limits Brownian motion to relatively low frequencies (~ 500 Hz)17. Noise with this frequency composition probably would not be effective in enhancing the SNR for higher frequency signals. However, theoretical considerations suggest a corner frequency for thermal fluctuations in cochlear hair bundles of about
386

Frequency (Hz)

Current spectral density (pA2/Hz)

hair bundle noise 1.4 nm r.m.s.

1998 Nature America Inc. http://neurosci.nature.com

Frequency (Hz)

Current spectral density (pA2/Hz)

hair bundle noise 10.0 nm r.m.s.

Frequency (Hz)

Fig. 2. Spectral density of mechanoelectrical transduction at different noise levels. (a) In the absence of added noise, the only source of noise is the probes own uncontrolled vibrations (1.0 nm). In these conditions, the transduction signal resulting from the stimulation of the hair bundle (2 nm, 300 Hz) barely emerges from the background noise. (b) The addition of noise (to a total of 1.4 nm) to the stimulus probe leads to a distinct peak at 300 Hz (arrow). (c) This peak gradually disappears as noise is further increased to 10.0 nm. For a typical cell, the average stimulus (one cycle of the 2 nm sinusoid) does not open a single transduction channel. (Typically a channel opens every other cycle.) Thus, the spectral characteristics of whole-cell current noise are not dominated by white noise. All noise levels are r.m.s.

4 kHz (ref. 11), which means SR is effective in the mid-frequency range of audition. The auditory system has an enormous dynamic range, sensing sound pressure levels which vary over five orders of magnitude (100 dB). Added random motion, however, increases the SNR of transduction in our experiments by up to twofold.
nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

1998 Nature America Inc. http://neurosci.nature.com

Fig. 3. Noise depena b dence of SNRs for 300 Hz 500 Hz mechanoelectrical transduction. (a) SNR of mechanoelectrical transduction as a function of the noise added to the stimulus probe for a hair cell stimulated at 300 Hz. The continuous line represents the predicted SNR versus noise level for Hair bundle noise (nm) Hair bundle noise (nm) the two-state model. The data were fitted to a simc d plified version of equation a -b/ x 1: SNR = x e where a and b are constants and x is the hair bundle noise. (Note that the pre-exponential term in equation 1 includes RT.) The range of SNRs varied from cell to cell (see b), depending on the whole-cell recording noise and on the number of intact transduction channels. (b) SNR of X mechanoelectrical transBundle displacement (nm) duction as a function of the noise added to the stimulus probe for a hair cell (different from that in a) stimulated at 500 Hz. (c) The relation of simulated Brownian motion (low-pass filtered, 2 nm white noise applied to a hair bundle) to a cells displacementresponse relationship, which relates the open probability of transduction channels (left y-axis) to the hair bundle displacement (x-axis) (ref. 2). The relationship was fitted by a single Boltzmann function. A 250-ms segment of noise (right y-axis) is superimposed for comparison. (d) A transduction channel is represented as existing in one of two mutually exclusive states, open (O) or closed (C), which are separated by an energy barrier. For simplicity, the energy profile is plotted as a symmetrical double-well potential. Displacement of the bundle (exaggerated for illustration purposes) serves to modulate the energy profile, favoring one of the two states. At very low displacement amplitudes, an optimal noise level serves to facilitate channel transitions. All noise levels are r.m.s.
SNR SNR Time (s) Popen U(X)

How significant is such an enhancement is in view of the dynamic range of hearing? This dynamic range is compressed into hair bundle displacements, that range between 0.2 and 200 nm (60 dB) 15. Although this compression increases the importance of a twofold gain, the more relevant question is how effective would transduction be in the absence of random bundle motion. This question cannot be answered at present because it is not possible to entirely suppress noise in the stimulus probe (Fig. 1a). However, our interpretation suggests that in the complete absence of random fluctuations the transduction of weak signals would almost cease. It might seem, a priori, that hair bundle stimulation would be most effective if driven by a rigid structure such as the tectorial membrane. However, by detaching from accessory structures at some point in evolutionary history, inner hair cells would have gained an increased sensitivity to weak sounds by exploiting SR, an acquisition with an obvious selective advantage.
Methods ELECTROPHYSIOLOGY. Hair cells were obtained from the maculae sacculi of leopard frogs (Rana pipiens) using published methods27, and were transferred to the stage of an Axiovert 100 inverted microscope (Carl Zeiss, Oberkochen, Germany) equipped with a 63, 1.4 N.A (numerical aperture), oil objective. Experiments were done at room temperature (2025C) in a perilymph-like saline (110 mM NaCl, 2 mM KCl, 4 mM CaCl2, 3 mM glucose, 5 mM HEPES, pH 7.25, with 1 N NaOH). For nature neuroscience volume 1 no 5 september 1998

whole-cell recording, electrodes (3-5 M) were tip-filled with a solution of 102 mM CsCl, 2 mM NaCl, 1 mM ATP (Mg2+ salt), 5 mM HEPES, titrated to pH 7.3 with CsOH. Cells were voltage clamped to -70 mV using an Axopatch-200B amplifier (Axon Instruments, Foster City, California). MECHANICAL STIMULATION. Mechanical stimulation, imaging and optical calibration were done as described19,25. Mechanical stimuli were conveyed to a cell by a stiff mechanical probe firmly attached to the bundles kinociliary bulb. (A clean probe sticks tightly to the bulb.) Immediately after obtaining the whole-cell recording configuration, the cell was stimulated by a single train of 20 mechanical pulses to assess the state of the transduction apparatus. A weak periodic stimulus (typically a 2 nm (peak), 300 or 500 Hz sine wave) was generated with a digital function generator (33120 Hewlett Packard, Palo Alto, California). White noise was generated with a second digital function generator (DS335, Stanford Research Systems, Sunnyvale, California). The two signals were added on a differential amplifier (AM502, Tektronix, Wilsonville, Oregon) and low-pass filtered to 1 kHz with a single pole filter, to resemble a weak periodic signal with added Brownian motion. (All noise amplitudes here refer to the post-filtered signal.) The output of the differential amplifier was used to drive the piezoelectric stimulator, whose resonant frequency with a probe mounted was roughly 2 kHz. DATA ACQUISITION AND ANALYSIS. Mechanoelectrical transduction current was low-pass filtered at 1.0 kHz and digitized at 2.5 kHz with a Lab-NB board (National Instruments, Austin, Texas), operating under LabView.
387

1998 Nature America Inc. http://neurosci.nature.com

articles

Data were collected in 512 sample segments and subjected to Fourier analysis. Spectra from 1224 data segments were averaged to improve resolution. The SNR was obtained by dividing the integral of the peak corresponding to the signal (300500 Hz) by the noise level surrounding the peak in an equal bandwidth.

12. 13. 14. 15. 16. 17. 18.

Acknowledgements
We thank Tim Cope and Ray Dingledine for critically reading the manuscript, Richard Jacobs for technical advice, and Mario Ruggero, Ille Gebeshuber and Peter Jung for discussions. This work was supported by the NIDCD and by a grant from the Emory/Georgia Tech. Consortium.

RECEIVED 30 APRIL: ACCEPTED 13 JULY 1998


1. de Vries, H. L. Brownian movement and hearing. Physica 14, 4860 (1948). 2. Corey, D. P. & Hudspeth, A. J. Kinetics of the receptor current in bullfrog saccular hair cells. J. Neurosci. 3, 962976 (1983). 3. Harris, G. G. Brownian motion in the cochlear partition. J. Acoust. Soc. Am. 44, 176186 (1968). 4. Wiesenfeld, K. & Moss, F. Stochastic resonance and the benefits of noise: from ice ages to crayfish and SQUIDs. Nature 373, 3336 (1995). 5. McNamara, B. & Wiesenfeld, K. Theory of stochastic resonance. Phys. Rev. A 39, 48544869 (1989). 6. Bezrukov, S. M. & Vodyanov, I. Noise-induced enhancement of signal transduction across voltage-dependent ion channels. Nature 378, 362364 (1995). 7. Collins, J. J., Imhoff, T. T. & Grigg, P. Noise-enhanced information transmission in rat SA1 cutaneous mechanoreceptors via aperiodic stochastic resonance. J. Neurophysiol. 76, 642645 (1996). 8. Levin, J. E. & Miller, J. P. Broadband neural encoding in the cricket cercal sensory system enhanced by stochastic resonance. Nature 380, 165168 (1996). 9. Douglass, J. K., Wilkens, L., Pantazelou, E. & Moss, F. Noise enhancement of information transfer in crayfish mechanoreceptors by stochastic resonance. Nature 365, 337340 (1993). 10. Narins, P. M., Benedix, J. H., Jr. & Moss, F. Can increasing temperature improve information transfer in the anuran peripheral auditory system? Aud. Neurosci. 3, 389400 (1997). 11. Svrcek-Seiler, W. A., Gebeshuber, I. C., Rattay, F., Biro, T. S. & Markum, H.

19. 20. 21. 22. 23. 24. 25. 26. 27.

1998 Nature America Inc. http://neurosci.nature.com

Micromechanical models for the Brownian motion of hair cell stereocilia. J. Theor. Biol. (in press). Lindeman, H. H., Ades, H. W., Bredberg, G. & Engstrm, H. The sensory hairs and the tectorial membrane in the development of the cats organ of Corti. Acta Otolaryngol. 72, 229242 (1972). Dallos, P. Response characteristics of mammalian cochlear hair cells. J. Neurosci. 5, 15911608 (1985). Dallos, P., Billone, M. C., Durrant, J. D., Wang, C. & Raynor, S. Cochlear inner and outer hair cells: functional differences. Science 177, 356358 (1985). Ruggero, M. Responses to sound of the basilar membrane of the mammalian cochlea. Curr. Opin. Neurobiol. 2, 449456 (1992). Denk, W. & Webb, W. W. Forward and reverse transduction at the limit of sensitivity studied by correlating electrical and mechanical fluctuations in frog saccular hair cells. Hear. Res. 60, 89102 (1992). Denk, W., Webb, W. W. & Hudspeth, A. J. Mechanical properties of sensory hair bundles are reflected in their Brownian motion measured with a laser differential interferometer. Proc. Natl Acad. Sci. USA 86, 53715375 (1989). Hamill, O. P., Marty, A., Neher, E., Sakmann, B. & Sigworth, F. J. Improved patch-clamp techniques for high-resolution current recording from cells and cell-free membrane patches. Pflugers Arch. 391, 85100 (1981). Howard, J. & Hudspeth, A. J. Mechanical relaxation of the hair bundle mediates adaptation in mechanoelectrical transduction by the bullfrogs saccular hair cell. Proc. Natl Acad. Sci. USA 84, 30643068 (1987). Howard, J. & Ashmore, J. F. Stiffness of sensory hair bundles in the sacculus of the frog. Hear. Res. 23, 93104 (1986). Atkins, P. W. in Physical Chemistry, (Oxford University Press, Oxford, 1978). Howard, J., Roberts, W. M. & Hudspeth, A. J. Mechanoelectrical transduction by hair cells. Ann. Rev. Biophys. and Biophys. Chem. 17, 99124 (1988). Russell, I. J., Richardson, G. P. & Cody, A. R. Mechanosensitivity of mammalian auditory hair cells in vitro. Nature 321, 517519 (1986). Russell, I. J., Kossl, M. & Richardson, G. P. Nonlinear mechanical responses of mouse cochlear hair bundles. Proc. R. Soc. Lond. B 250, 217227 (1992). Howard, J. & Hudspeth, A. J. Gating compliance associated with gating of mechanoelectrical transduction channels in the bullfrogs saccular hair cell. Neuron 1, 189199 (1988). Pickles, J. O., Comis, S. D. & Osborne, M. P. Cross-links between sterocilia in the guinea pig organ of Corti, and their possible relation to sensory transduction. Hear. Res. 15, 103112 (1984). Assad, J. A., Hacohen, N. & Corey, D. P. Voltage dependence of adaptation and active bundle movement in bullfrog saccular hair cells. Proc. Natl Acad. Sci. USA 86, 29182922 (1989).

388

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

A new intrathalamic pathway linking modality-related nuclei in the dorsal thalamus


John W. Crabtree, Graham L. Collingridge and John T.R. Isaac
Department of Anatomy, School of Medical Sciences, University of Bristol, Bristol BS8 1TD, UK Correspondence should be addressed to J.W.C. (J.W.Crabtree@bris.ac.uk)

1998 Nature America Inc. http://neurosci.nature.com

Transmission of sensory information through the dorsal thalamus involves two types of modalityrelated nuclei, first order and higher order, between which there are thought to be no intrathalamic interactions. We now show that within the somatosensory thalamus, cells in one nucleus, the ventrobasal complex, can influence activity in another nucleus, the medial division of the posterior complex. Stimulation of ventrobasal complex cells evoked inhibitory postsynaptic currents in cells of the medial division of the posterior complex. These currents exhibited the reversal potential and pharmacology of a GABAA receptor-mediated chloride conductance, indicating that they result from the activation of a disynaptic pathway involving the GABAergic cells of the thalamic reticular nucleus. These findings provide the first direct evidence for intrathalamic interactions between dorsal thalamic nuclei.

Nearly fifty years ago, Rose and Woolsey1 introduced their nowclassic view of the organization of the mammalian dorsal thalamus, which contained two types of nuclei, extrinsic and intrinsic. Extrinsic nuclei were thought to receive their main afferents from extrathalamic sources, such as the ascending somatosensory, auditory and visual pathways and to send efferents to primary cortical areas (sensory cortex). In contrast, intrinsic nuclei were thought to receive their main afferents from extrinsic thalamic nuclei and to send efferents to secondary cortical areas (association cortex). Thus, the organization of the dorsal thalamus would parallel that of the cerebral cortex in that the dorsal thalamus would also contain association areas that represented further elaborations of its sensory areas. Accordingly, the activity of cells in the intrinsic nuclei would constitute a higher functional level than the activity of cells in the extrinsic nuclei. However, because evidence for direct connections between dorsal thalamic nuclei failed to materialize, this aspect of Rose and Woolseys scheme was eventually abandoned. Instead, evidence soon emerged that intrinsic thalamic nuclei also received afferents from extrathalamic sources2,3, giving rise to the idea of parallel sensory pathways through the thalamus4. A revised thalamic scheme has been proposed57 in which sensory nuclei in the dorsal thalamus are classified as first order (formerly extrinsic) or higher order (formerly intrinsic). Both first-order and higher-order thalamic nuclei send efferents to cortex, receive modulatory inputs from cortical layer VI and are reciprocally connected with the thalamic reticular nucleus (TRN), which is a ventral thalamic derivative8. However, whereas firstorder nuclei receive their main driving inputs through ascending sensory pathways, higher-order nuclei are thought to receive their main driving inputs through descending pathways from cortical layer V. Thus, first-order nuclei would transmit information about stimuli in the sensory periphery to the cerebral cortex, whereas higher-order nuclei would transmit information
nature neuroscience volume 1 no 5 september 1998

about sensory processing in one cortical area to another cortical area. Such transmission of information through modality-related first-order and higher-order nuclei is thought to involve separate pathways for each nucleus. Recent results from studies of the cats TRN9,10 revive the possibility of intrathalamic interactions between nuclei in the dorsal thalamus, although not involving direct excitatory connections as originally envisaged1. For example, in the somatosensory part of the cats thalamus, cells in TRN that project to the ventrobasal complex (VB, a first-order nucleus) or to the medial division of the posterior complex (POm, a higher-order nucleus) occupy overlapping territories. Because of reciprocal connections between TRN and the dorsal thalamus, such overlap suggests that TRN cells with outputs to POm could receive inputs from VB and vice versa. These connections would allow intrathalamic interactions between two modality-related nuclei in the dorsal thalamus through disynaptic circuits involving TRN. If they exist, such interactions would take the form of inhibitory postsynaptic events, because TRN is composed entirely of GABAergic cells1113, whose axons make F-type synapses in dorsal thalamic nuclei1416. Here we show that TRN cells in the rat that project to VB or POm occupy overlapping territories. Furthermore, using a thalamic slice preparation, we present evidence for intrathalamic connections between VB and POm involving an intermediary pathway through TRN. Our findings are the first direct evidence for intrathalamic interactions between two nuclei in the dorsal thalamus.

Results

DISTRIBUTIONS OF TRN CELLS PROJECTING TO VB OR POM


In a preliminary examination of the rats thalamus, we analyzed the distributions of TRN cells that project to VB or POm to determine whether the rats thalamus would be a suitable choice in
389

1998 Nature America Inc. http://neurosci.nature.com

articles

Fig. 1. Horizontal sections showing the locations of labeled cells in TRN after an injection of WGA-HRP in VB or POm in the juvenile rat. (a) A tracer-injected VB showing reaction product. The arrows point to retrogradely labeled cells in TRN. (b) Labeled cells in TRN after VB injection shown in (a). The arrows point to the outer border of TRN. (c) A tracer-injected POm showing reaction product. The arrows point to retrogradely labeled cells in TRN. (d) Labelled cells in TRN after POm injection in (c). The arrows point to the outer border of TRN. Scale bars represent 500 m (a,c) or 200 m (b,d). Rostral, top; medial, right (ad).

d
to monitor their response to VB stimulation. By cutting slices in the horizontal plane, intrathalamic connections between TRN and both VB and POm were preserved, whereas connections between the thalamus and somatosensory cortex were severed. The slices were transilluminated from below, which clearly showed the somatosensory nuclei of the thalamus and enabled the appropriate positioning of the recording and stimulating electrodes (Fig. 2a). VB neurons were stimulated using local application of glutamate so that cell bodies were activated but axons of passage were not. The inclusion of a blue dye in the stimulating solution enabled the location of the stimulation site in VB and the spread of the stimulating solution to be visualized. In 29 experiments, robust inhibitory postsynaptic currents (IPSCs; Fig. 2b) were recorded in POm cells in response to glutamate stimulation in VB. These IPSCs rose rapidly with a latency typically between 50 and 100 ms relative to the end of the glutamate pulse. A component of this latency included the time required for the glutamate to reach and activate the appropriate VB cells. There was a general topographic relationship between the location of a responsive cell in POm and the location of the stimulating electrode in VB that evoked the response. That is, stimulating in more rostral regions of VB evoked IPSCs in more rostrally located POm cells (Fig. 2c). After an experiment, recovery of biocytin-filled cells (Fig. 2d) confirmed the location of recorded cells in the POm.

1998 Nature America Inc. http://neurosci.nature.com

which to study possible intrathalamic interactions between dorsal thalamic nuclei. In both adult (n = 3) and juvenile (n = 3) rats, tracer injections restricted to either VB or POm resulted in labeled cells and terminals that occupy a centroventral part, or somatosensory sector17,18, of TRN. After a large injection in the VB (Fig. 1a), cells and VB terminals across the entire thickness of TRN were labeled (Fig. 1b). A large injection in POm (Fig. 1c) also labeled cells and POm terminals across the entire thickness of TRN (Fig. 1d). These examples of labelling are from 19-day-old rats. Thus, as in the cat9, TRN cells in the rat that project to VB or POm occupy overlapping territories. Furthermore, the terminal field from VB (or POm) overlaps the cells that project to POm (or VB). These anatomical results indicated that the rats somatosensory thalamus was a promising system in which to test our hypothesis.

RESPONSE OF POM CELLS TO VB STIMULATION


An in vitro slice preparation was developed to allow whole-cell voltage-clamp recordings to be made from neurons in POm and

CHARACTERIZATION OF THE RESPONSE EVOKED IN POM


The currentvoltage (IV) relationship of the response evoked by VB stimulation was examined by collecting responses at

Fig. 2. IPSCs were evoked in POm cells by glutamate stimulation in VB. (a) The thalamic slice preparation as it appears in the recording chamber. Transilluminating the slice clearly reveals VB, POm, TRN (R) and the internal capsule (asterisk). A glutamate- and dye-filled stimulating electrode is positioned in VB and a recording electrode is positioned in POm. The dashed lines indicate the borders of VB and POm. Scale bar represents 500 m. (b) An example of an IPSC recorded from a POm cell in response to glutamate stimulation (solid bar) in VB. (c) Horizontal section (schematic) through the thalamus showing the topographic relationship between stimulation sites in the VB (upper left numbers) and recording sites in the POm (lower right numbers). The corresponding numbers indicate the general locations of the stimulating electrode in VB and of POm cells in which a response was evoked. (d) A montage of the biocytin-filled POm cell from which the recording in Fig. 4c was obtained. The cell has a large soma, approximately 25 m in diameter. Rostral, top; medial, right (a,c).
nature neuroscience volume 1 no 5 september 1998

390

1998 Nature America Inc. http://neurosci.nature.com

articles

a
IPSC amplitude (pA)

b
IPSC amplitude (pA) Membrane potential (mV)

Membrane potential (mV)

1998 Nature America Inc. http://neurosci.nature.com

Fig. 3. Currentvoltage relationship of the IPSCs evoked in POm cells by glutamate stimulation in VB. (a) Graph of IPSC amplitude versus holding potential. Inset, superimposed average IPSCs at the different holding potentials from the same cell. (b) Summary graph of IPSC amplitude versus holding potential for five cells. Diagonal line, regression analysis of the data (a,b).

different holding potentials. As shown for an example cell (Fig. 3a) and in a pool of five experiments (Fig. 3b), IPSCs had an approximately linear IV relationship over the potentials examined and reversed at a potential (67 2 mV; n = 5) close to the expected reversal potential for a Cl conductance under our recording conditions. The IPSCs were reversibly antagonized by the competitive GABAA antagonist bicuculline (10 M). After the collection of a stable baseline of IPSCs, bicuculline was bath applied for ten

minutes, which blocked the response, and on washout the IPSCs returned (Fig. 4a). Summary data for five experiments (Fig. 4b) show that bicuculline caused a reduction in IPSC amplitude to 14 6% of baseline, and on washout the amplitude recovered to 76 32%. In two addditional experiments, the non-competitive GABAA receptor anatgonist picrotoxin (50 M) was bath applied; this procedure also blocked the IPSCs (Fig. 4c and d). Because picrotoxin blockade is only slowly reversible, no washout was attempted in these experiments.

c
IPSC amplitude (%)

IPSC amplitude (%)

Time (min)

d b
IPSC amplitude (%)

Fig. 4. Bicuculline or picrotoxin blocks the IPSCs evoked in POm cells by glutamate stimulation in VB. (a) IPSC amplitude versus time from an example experiment in which bicuculline was bath applied (solid bar). Top, Individual traces taken at the times indicated (1,2,3) during the experiment; glutamate Time (min) stimulation is indicated by the solid bars. (b) Summary graph of the effect of bicuculline (solid bar) on the evoked IPSCs in five cells. (c) IPSC amplitude versus time from an example experiment in which picrotoxin was bath applied (solid bar). Dashed line indicates the baseline response level (100%) in (ac). (d) Individual traces taken at the times indicated (1,2) during the experiment (c). Glutamate stimulation is indicated by the solid bar.
nature neuroscience volume 1 no 5 september 1998 391

1998 Nature America Inc. http://neurosci.nature.com

articles

Fig. 5. Horizontal section (schematic) through the thalamus showing an intrathalamic pathway linking VB with POm through TRN (R). The cell in VB projects to somatosensory cortex (arrow) and provides a collateral input to a reticular cell that projects to a cell in POm. The cell in POm also projects to somatosensory cortex (axon not shown).

Discussion Here we show that TRN cells projecting to VB or POm occupy overlapping territories. Using electrophysiological recordings in a thalamic slice preparation, we then show that stimulation of VB cells evokes IPSCs in POm cells and that these IPSCs have the reversal potential and pharmacological characteristics of GABAA receptor-mediated currents. Our data provide strong evidence for an intrathalamic pathway linking modality-related nuclei in the dorsal thalamus. The IPSCs evoked in POm cells can be entirely accounted for by the activation of a disynaptic pathway involving the GABAergic cells of TRN. The IPSCs cannot be explained by the activation of polysynaptic pathways involving the cerebral cortex or the GABAergic cells of the substantia nigra19, because the thalamic slices were cut in a plane that would sever such long-range circuits. Furthermore, activation of disynaptic circuits between POm and TRN is unlikely to account for the IPSCs, because the glutamate stimulation sites were restricted to VB, as indicated by the dye included in the stimulation solution. The IPSCs are also unlikely to result from the activation of a monosynaptic pathway from TRN to POm because the glutamate stimulation selectively activated VB cells and not axons of passage, such as those from TRN passing through VB on their way to POm. Furthermore, the stimulating electrode was routinely placed closer to the recording site in POm than to the inner border of TRN. However, inward currents were never recorded in POm cells, indicating that the glutamate did not spread into POm and, therefore, was very unlikely to have spread into TRN. Finally, because VB and POm conspicuously lack GABAergic interneurons11,20,21, cells in TRN are the only source of GABAergic afferents to POm in our slice preparation. Collectively, these considerations strongly indicate that the IPSCs recorded in this study result from the activation of a functional pathway from VB to POm through TRN (Fig. 5). We have also collected preliminary data from four VB cells indicating a functional pathway from POm to VB through TRN. The reciprocal VBTRNPOm and POmTRNVB pathways will tentatively be defined here as forming closed-loop circuits, but only in the sense that the same thalamic nuclei are involved. Whether or not these pathways form true closed-loop circuits6 involving the same neurons remains to be determined. Intrathalamic pathways involving more than one dorsal thalamic nucleus could be a prominent feature of the mammalian thalamus. Further demonstrations of these pathways could be anticipated in those thalamic systems in which TRN cells project to more than one dorsal thalamic nucleus9,10,22,23 or where TRN cells projecting to first-order or higher-order dorsal thalamic nuclei occupy overlapping territories. Such overlapping popula392

tions of cells in TRN are found in the somatosensory sector of the cat9 and in the auditory sector of the bushbaby24 and the cat10. The rostral sector of TRN, which is mainly connected to motor- and limbic-related thalamic nuclei, is particularly rich in overlapping cell populations in the rat23,2530. Even where there is marginal territorial overlap between populations of TRN cells, as in the visual sector of the bushbaby31,32 and the rat33, pathways connecting dorsal thalamic nuclei could still involve a considerable fraction of TRN cells, such as those representing the central visual field. Intrathalamic pathways involving two dorsal thalamic nuclei would link together the activity of cells in modality-related first-order (for example, VB) and higher-order (for example, POm) nuclei. Our results here indicate that this linkage is not widespread or random. Instead, the locations of a stimulation site in VB and a responsive cell in POm were in somatotopic register in accordance with the maps contained in VB 34,35 and POm 36. The topography exhibited by this intrathalamic pathway would reflect the somatotopic organization of the inputs3739 and outputs9,40,41 of TRN. One physiological role of intrathalamic pathways involving TRN cells could be the modulation of neuronal firing patterns in one dorsal nucleus by activity in another modality-related dorsal nucleus. In the dorsal thalamus there are two main patterns of neuronal firing, the tonic and burst modes6,42,43, and neurons switch between these modes in response to changes in membrane potential. During each response mode, a different type of sensory information is thought to be transmitted through the thalamus6, one type that is related to sensory analysis (tonic mode) and the other type that is related to sensory detection (burst mode). Inputs from the brainstem parabrachial region44 and from the cerebral cortex45 depolarize dorsal thalamic neurons and are implicated in switching the mode of response from burst to tonic. By their hyperpolarizing effects, inputs from TRN could cause dorsal thalamic cells to switch from tonic to burst mode. Such effects would be expected from the outputs of TRN, if this nucleus is involved in sensory attentional processes46. Thus, in the dorsal thalamus, the type of sensory information transmitted through one nucleus could be selectively and topographically modulated by another modality-related nucleus through an intrathalamic pathway involving TRN. The identification of this pathway signals an expansion in our understanding of how the dorsal thalamus processes and transmits information.
Methods NEUROANATOMY. All experimental procedures were in accordance with U.K. Home Office regulations (Animals Scientific Procedures Act of 1986). DA or Wistar rats 50 to 70 days old (adults) and 13 to 20 days old (juveniles) were used. Adults were anesthetized with sodium pentobarbital (40 mg/kg, intraperitoneally), and juveniles were anesthetized with a mixture of xylazine (2.5 mg/kg, intraperitoneally) and ketamine (20 mg/kg, intraperitoneally). The rats were then mounted in a stereotaxic apparatus. The cells of origin of the thalamic reticular pathways to VB and POm were defined by injecting these nuclei with the tracer wheatgerm agglutinin conjugated to horseradish peroxidase (WGA-HRP, Type VI; Sigma). To inject the tracer into the thalamus, trephine holes were made in the skull at appropriate stereotaxic coordinates47, and the dura was excised. WGA-HRP (a 0.40.5% solution; <0.2 l) was injected bilaterally into VB or POm using micropipettes with a tip diameter of 3040 m. Injections were made by pulses of pressure (312 psi; <1 second in duration) over a 510 min period using a pressure-injection unit.

1998 Nature America Inc. http://neurosci.nature.com

HISTOLOGY. After a postinjection interval of 1820 h, the rats were deeply anesthetized with sodium pentobarbital (60 mg/kg, intraperitoneally).
nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

They were then perfused through the heart with a 0.9% saline rinse followed by a mixture of 1% paraformaldehyde and 2% glutaraldehyde in 0.1 M phosphate buffer (pH 7.4). The thalamus was then dissected away from the rest of the brain and was stored overnight in a phosphate-buffered 30% sucrose solution. Thalamic tissue was cut frozen into sections, 50 m in thickness, in the horizontal plane. The sections were then processed using the tetramethyl benzidine method48. After labelling had been drawn and photographed, sections were counterstained with thionin. ELECTROPHYSIOLOGY. The brains of 13- to 20-day-old Wistar rats were removed and placed in ice-cold extracellular solution. Horizontal thalamic slices (500 m in thickness) were cut on a vibratome and were allowed to recover for 12 h at room temperature (23C25C). They were then transferred to a recording chamber where they were submerged beneath a continuously superfusing solution saturated with 95% O2 and 5% CO2. The standard extracellular solution contained 119 mM NaCl, 2.5 mM KCl, 1.0 mM NaH2PO4, 26.2 mM NaHCO3 and 11.0 mM glucose (pH 7.4). Whole-cell voltage-clamp recordings were made from POm cells using glass electrodes (35 M). The whole-cell solution contained 130 mM Cs methane sulphanate, 10.0 mM HEPES, 0.5 mM EGTA, 3.0 mM NaCl, 5.0 mM QX-314Cl, 4.0 mM Mg-ATP, 0.3 mM Na-GTP and 10.7 mM biocytin (pH 7.2 with CsOH, 275 mOsm). During recordings, cells were held at 40 mV unless otherwise indicated. A glass electrode (tip diameter, 2.53.0 m) was placed in VB and was used to evoke intrathalamic responses in POm cells. This stimulating electrode was filled with 5 mM L-glutamate monosodium salt in extracellular solution (pH 7.4) and 5.0% methylene blue and was connected to a pressure-injection unit. Glutamate was ejected at a frequency of 0.1 or 0.05 Hz by pulses of pressure (5-15 psi; 50 ms duration). After an optimal evoked response was obtained, stimulation was maintained without interruption at a constant frequency and intensity for the duration of the experiment. After the recordings, tracer-filled cells were processed for biocytin histochemistry using the ABC-diaminobenzidine method49 and their locations in POm confirmed by thionin counterstaining. Recordings were made using an Axopatch-ID amplifier. Data were filtered at 2 kHz, digitized at 5 kHz and stored on computer. IPSC amplitudes, input resistance and series resistance were analyzed on-line50 and displayed using custom software (W. W. Anderson & G. L.Collingridge, Soc. Neurosci. Abstr. 23, 665, 1997). IPSC peak amplitudes were measured by averaging across a 2530 ms time window centered on the peak of the response. For IV experiments, reversal potentials were estimated using regression analysis. Data are expressed as percentages of the average baseline amplitude (baseline, 100%). Pooled data are expressed as means standard error.

Acknowledgements
We thank the U.K. Medical Research Council (G.L.C.) and The Wellcome Trust (J.T.R.I.) for their support.

RECEIVED 23 JUNE: ACCEPTED 28 JULY 1998


1. Rose, J. E. & Woolsey, C. N. Organization of the mammalian thalamus and its relationships to the cerebral cortex. Electroenceph. Clin. Neurophysiol. 1, 391404 (1949). 2. Altman, J. & Carpenter, M. B. Fiber projections of the superior colliculus in the cat. J. Comp. Neurol. 116, 157177 (1961). 3. Jones, E. G. & Powell, T. P. S. An analysis of the posterior group of thalamic nuclei on the basis of its afferent connections. J. Comp. Neurol. 143, 185215 (1971). 4. Schneider, G. E. Two visual systems. Brain mechanisms for localization and discrimination are dissociated by tectal and cortical lesions. Science, 163, 895902, (1969). 5. Guillery, R. W. Anatomical evidence concerning the role of the thalamus in corticocortical communication: a brief review. J. Anat. 187, 583592 (1995). 6. Sherman, S. M. & Guillery, R. W. Functional organization of thalamocortical relays. J. Neurophysiol. 76, 13671395 (1996). 7. Guillery, R. W., Feig, S. L. & Lozsdi, D. A. Paying attention to the thalamic reticular nucleus. Trends Neurosci., 21, 2832 (1998). 8. Rose, J. E. The ontogenetic development of the rabbits diencephalon. J. Comp. Neurol. 7, 61129 (1942).

9. Crabtree, J. W. Organization in the somatosensory sector of the cats thalamic reticular nucleus. J. Comp. Neurol. 366, 207222 (1996). 10. Crabtree, J. W. Organization in the auditory sector of the cats thalamic reticular nucleus. J. Comp. Neurol. 390, 167182 (1998). 11. Houser, C. R., Vaughn, J. E., Barber, R. P. & Roberts, E. GABA neurons are the major cell type of the nucleus reticularis thalami. Brain Res. 200, 341354 (1980). 12. Hendrickson, A. E., Ogren, M. P., Vaughn, J. E., Barber, R. P. & Wu, J.-Y. Light and electron microscopic immunocytochemical localization of glutamic acid decarboxylase in monkey geniculate complex: Evidence for GABAergic neurons and synapses. J. Neurosci. 3, 12451262 (1983) 13. Oertel, W. H. et al. Coexistence of glutamic acid decarboxylase- and somatostatin-like immunoreactivity in neurons of the feline nucleus reticularis thalami. J. Neurosci. 3, 13221332 (1983). 14. Montero, V. M. & Scott, G. L. Synaptic terminals in the dorsal lateral geniculate nucleus from neurons of the thalamic reticular nucleus: A light and electron microscope autoradiographic study. Neuroscience 6, 25612577 (1981). 15. Montero, V. M. Ultrastructural identification of axon terminals from the thalamic reticular nucleus in the medial geniculate body in the rat: An EM autoradiographic study. Exp. Brain Res. 51, 338342 (1983). 16. Peschanski, M., Ralston, H. J. & Roudier, F. Reticularis thalami afferents to the ventrobasal complex of the rat thalamus: An electron microscope study. Brain Res. 270, 325329 (1983). 17. Jones, E. G. Some aspects of the organization of the thalamic reticular complex. J. Comp. Neurol. 162, 285308 (1975). 18. Sugitani, M. Electrophysiological and sensory properties of the thalamic reticular neurones related to somatic sensation in rats. J. Physiol. (Lond.) 290, 7995 (1979). 19. MacLeod, N. K., James, T. A., Kilpatrick, I. C. & Starr, M. S. Evidence for a GABAergic nigrothalamic pathway in the rat. II. Electrophysiological studies. Exp. Brain Res. 40, 5561 (1980). 20. Barbaresi, P., Spreafico, R., Frassoni, C. & Rustioni, A. GABAergic neurons are present in the dorsal column nuclei but not in the ventroposterior complex of rats. Brain Res. 382, 305326 (1986). 21. Ohara, P. T. & Lieberman, A. R. Some aspects of the synaptic circuitry underlying inhibition in the ventrobasal thalamus. J. Neurocytol. 22, 815825 (1993). 22. Steriade, M., Parent, A. & Hada, J. Thalamic projections of nucleus reticularis thalami of cat: A study using retrograde transport of horseradish peroxidase and fluorescent tracers. J. Comp. Neurol. 229, 531547 (1984). 23. Kolmac, C. I. & Mitrofanis, J. Organisation of the reticular thalamic projection to the intralaminar and midline nuclei in rats. J. Comp. Neurol. 377, 165178 (1997). 24. Conley, M., Kupersmith, A. C. & Diamond, I. T. The organization of projections from subdivisions of the auditory cortex and thalamus to the auditory sector of the thalamic reticular nucleus in Galago. Eur. J. Neurosci. 3, 10891103 (1991). 25. Cornwall, J. & Phillipson, O. T. Afferent projections to the dorsal thalamus of the rat as shown by retrograde lectin transport. I. The mediodorsal nucleus. Neuroscience 24, 10351049 (1988). 26. Cicirata, F., Angaut, P., Serapide, M. F. & Panto, M. R. Functional organization of the direct and indirect projection via the reticularis thalami nuclear complex from the motor cortex to the thalamic nucleus ventralis lateralis. Exp. Brain Res. 79, 325337 (1990). 27. Gonzalo-Ruiz, A. & Lieberman, A. R. Topographic organization of projections from the thalamic reticular nucleus to the anterior thalamic nuclei in the rat. Brain Res. Bull. 37, 1735 (1995) 28. Lozsdi, D. A. Organization of connections between the thalamic reticular and the anterior thalamic nuclei in the rat. J. Comp. Neurol. 358, 233246 (1995). 29. Lizier, C., Spreafico, R. & Battaglia, G. Calretinin in the thalamic reticular nucleus of the rat: Distribution and relationship with ipsilateral and contralateral efferents. J. Comp. Neurol. 377, 217233 (1997). 30. Pinault, D. & Deschnes, M. Projection and innervation patterns of individual thalamic reticular axons in the thalamus of the adult rat: A threedimensional, graphic and morphometric analysis. J. Comp. Neurol. 391, 180203 (1998). 31. Conley, M. & Diamond, I. T. Organization of the visual sector of the thalamic reticular nucleus in Galago. Evidence that the dorsal lateral geniculate and pulvinar nuclei occupy separate parallel tiers. Eur. J. Neurosci. 2, 211226 (1990). 32. Harting, J. K., Van Lieshout, D. P. & Feig, S. Connectional studies of the primate lateral geniculate nucleus: Distribution of axons arising from the thalamic reticular nucleus of Galago crassicaudatus. J. Comp. Neurol. 310, 411427 (1991). 33. Coleman, K. A. & Mitrofanis, J. Organization of the visual reticular thalamic nucleus of the rat. Eur. J. Neurosci. 8, 388404 (1996). 34. Emmers, R. Organization of the first and second somesthetic regions (SI and SII) in the rat thalamus. J. Comp. Neurol. 124, 215228 (1965). 35. Waite, P. M. E. Somatotopic organization of vibrissal responses in the ventrobasal complex of the rat thalamus. J. Physiol. (Lond.) 228, 527540 (1973). 36. Diamond, M. E., Armstrong-James, M. & Ebner, F. F. Somatic sensory responses in the rostral sector of the posterior group (POm) and in the ventral posterior medial nucleus (VPM) of the rat thalamus. J. Comp. Neurol.

1998 Nature America Inc. http://neurosci.nature.com

nature neuroscience volume 1 no 5 september 1998

393

1998 Nature America Inc. http://neurosci.nature.com

articles

318, 462476 (1992). 37. Shosaku, A., Kayama, Y. & Sumitomo, I. Somatotopic organization in the rat thalamic reticular nucleus. Brain Res. 311, 5763 (1984). 38. Crabtree, J. W. The somatotopic organization within the rabbits thalamic reticular nucleus. Eur. J. Neurosci. 4, 13431351 (1992). 39. Crabtree, J. W. The somatotopic organization within the cats thalamic reticular nucleus. Eur. J. Neurosci. 4, 13521361 (1992). 40. Pinault, D., Bourassa, J. & Deschnes, M. The axonal arborization of single thalamic reticular neurons in the somatosensory thalamus of the rat. Eur. J. Neurosci. 7, 3140 (1995). 41. Cox, C. L., Huguenard, J. R. & Prince, D. A. Heterogeneous axonal arborizations of rat thalamic reticular neurons in the ventrobasal nucleus. J. Comp. Neurol. 366, 416430 (1996). 42. Jahnsen, H. & Llins, R. Electrophysiological properties of guinea-pig thalamic neurones: An in vitro study. J. Physiol. (Lond.) 349, 205226 (1984). 43. Steriade, M. & Deschnes, M. The thalamus as a neuronal oscillator. Brain Res. Rev. 8, 163 (1984). 44. Lu, S.-M., Guido, W. & Sherman, S. M. The brainstem parabrachial region

45. 46. 47. 48. 49. 50.

controls mode of response to visual stimulation of neurons in the cats lateral geniculate nucleus. Visual Neurosci. 10, 631642 (1993). Godwin, D. W., Vaughan, J. W. & Sherman, S. M. Metabotropic glutamate receptors switch visual response mode of lateral geniculate nucleus cells from burst to tonic. J. Neurophysiol. 76, 18001816 (1996). Crick, F. Function of the thalamic reticular complex: The searchlight hypothesis. Proc. Natl Acad. Sci. USA 81, 45864590 (1984). Paxinos, G. & Watson, C. in The Rat Brain in Stereotaxic Coordinates (Academic Press, Sydney, 1986). Mesulam, M. M. The blue reaction product in horseradish peroxidase neurohistochemistry: Incubation parameters and visibility. J. Histochem. Cytochem. 74, 12731280 (1976). Horikawa, K. & Armstrong, W. E. A versatile means of intracellular labeling: injection of biocytin and its detection with avidin conjugates. J. Neurosci. Methods 25, 111 (1988). Benke, T. A., Lthi, A., Isaac, J. T. R. & Collingridge, G. L. Modulation of AMPA receptor unitary conductance by synaptic activity. Nature 393, 793797 (1998).

1998 Nature America Inc. http://neurosci.nature.com

394

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

Functional connectivity between simple cells and complex cells in cat striate cortex
Jose-Manuel Alonso and Luis M. Martinez
Laboratory of Neurobiology, The Rockefeller University, New York, New York 10021, USA Both authors contributed equally to this work. Correspondence should be addressed to J.-M.A (alonsoj@rockvax.rockefeller.edu)

1998 Nature America Inc. http://neurosci.nature.com

In the cat primary visual cortex, neurons are classified into the two main categories of simple cells and complex cells based on their response properties. According to the hierarchical model, complex receptive fields derive from convergent inputs of simple cells with similar orientation preferences. This model received strong support from anatomical studies showing that many complex cells lie within the range of layer IV simple-cell axons but outside the range of most thalamic axons. Physiological evidence for the model, however, has remained elusive. Here we demonstrate that layer IV simple cells and layer II and III complex cells show correlated firing consistent with monosynaptic connections. As expected from the hierarchical model, all connections were in the direction from the simple cell to the complex cell, most frequently between cells with similar orientation preferences.

The orientation of a line at a given position in visual space is represented by a small volume of primary visual cortex, known as an orientation column. Although cells within such a column share some response properties (for example, orientation selectivity), they are functionally diverse. There are two main functional cell types, simple cells and complex cells. Simple cells have receptive fields made of separate and elongated subregions that evoke responses when turning a light spot on (on-subregion) or off (off-subregion). In contrast, complex cells tend to produce similar responses throughout the entire receptive field (on or off responses, onoff responses or responses to moving stimuli only). Although the specific roles of simple cells and complex cells in vision are still unknown, an attractive hypothesis is that they represent two different stages in hierarchical processing. In the first stage, simple cells become selective to the orientation of a line in a specific location. In the second stage, complex cells become selective to the orientation of a line regardless of its exact position within the receptive field. According to this hierarchical model, the convergence of geniculate inputs from the thalamus generates simple receptive fields in layer IV and, in turn, the

convergence of simple cell inputs generates complex receptive fields in the superficial layers1. The hierarchical model, appealing in its simplicity, has been subject of extensive research over the last 30 years. Recently, the first part of the model concerning the origin of simple receptive fields has received substantial support24 (although the relative contribution from corticocortical and geniculocortical inputs is debatable5). The second part of the model, however, has been challenged based on four main criticisms. First, previous studies did not find physiological evidence for direct excitatory connections from simple cells to complex cells 68. Second, the demonstration that some complex cells received direct geniculate input913 led to the proposal that thalamic inputs generate both simple and complex receptive fields9,12,14,15. Third, some stimuli seemed to drive complex cells but not simple cells16. Fourth, blockades of layer A in the lateral geniculate nucleus (LGN) were found to inactivate layer IV simple cells but not layer II and III complex cells17. In favor of a hierarchical model, layer IV simple cells send strong projections to the superficial layers of the cortex13,1820, where most cells are complex21(although this has been ques-

Table 1. Correlation strength as a function of the difference in orientation between the simple cell and the complex cell
Orientation on difference Number of correlated cell pairs Correlation on strength 0 degrees 21 (n = 38) 11.0% 4.0 (n = 17) 22 degrees 17 (n = 34) 7.7% 5.3 (n = 12) 45 degrees 10 (n = 50) 3.7% 2.0 (n = 8) 67 degrees 5 (n = 16) 3.9% 2.9 (n = 5) 90 degrees 0 (n = 3)

Number of asymmetric correlations and correlation strength as a function of the difference in orientation between the simple cell and the complex cell (only for CA > 0.4). Correlations consistent with monosynaptic connections were more frequently found and were stronger between cells with similar orientation preferences. Correlation strength was calculated, after subtracting the baseline, only for cell pairs with overlapping receptive fields (overlap > 60%).

nature neuroscience volume 1 no 5 september 1998

395

1998 Nature America Inc. http://neurosci.nature.com

articles

1998 Nature America Inc. http://neurosci.nature.com

Fig. 1. Coronal section (thickness, 40 m) of cat visual cortex showing five electrode tracks in the medial bank. All five tracks can be seen in one section, which proves the high precision of the electrode alignment. Because each of these electrodes could be moved independently, we could record from cells located at the position represented by the dotted line at different cortical layers (track thickness, 100 m; electrode thickness, 80 m; vertical misalignment for this case, < 100 m).

tioned22) and do not receive direct geniculate input10,11,13. Here we provide evidence for the functional connectivity predicted by these anatomical studies and other theoretical models based on a hierarchical organization2326.

Results We recorded simultaneously from 141 pairs of layer IV simple cells and layer II and III complex cells in vertical alignment within cat visual cortex (misalignment, < 200 m). This precise electrode alignment was made possible by using a multielectrode matrix of thin and independently movable probes27.

Five or six electrodes were coated with DiI or DiO28 and introduced into the medial bank of the cortex. At the end of the experiment, histological analysis demonstrated that all electrodes ran parallel to each other and parallel to the cortical surface (within almost the same coronal plane). For example, in some experiments we were able to place five electrode tracks in a section of 40 m thickness (Fig. 1) and record cells within the same orientation column (Fig. 1, dotted line). Because of the high precision of the electrode alignment, most of the 141 cell pairs studied had overlapping receptive fields and similar orientation preferences (average orientation difference, 30.7 degrees; average receptive field overlap, 77.8%; Table 1). In these 141 cell pairs, cross-correlation analysis showed strong correlated firing (a positive correlogram) in more than half of the cases (n = 82). The most representative examples of these positive correlograms are shown in Figs 2, 3 and 4. The first example is from a layer IV simple cell simultaneously recorded with a layer II complex cell (Fig. 2). The receptive field of the simple cell was mapped by reverse correlation with white noise (random checkerboards of black and white squares29,30), and the field from the complex cell was mapped with a moving bar. These simultaneously recorded cells had overlapping receptive fields ( Fig. 2a) and similar orientation preferences ( Fig. 2b ). Cross-correlation analysis showed a strong positive peak displaced from zero, indicating that the simple cell tended to fire before the complex cell (Fig. 2c). The fast rise time and peak time of this displaced positive peak is consistent with a monosynaptic excitatory connection 32 (Fig. 2c, inset). Because the complex cell was recorded outside of the range of most geniculate afferents, these two cortical cells could not share strong thalamic inputs. (Within the superficial layers, only cells near the layer III/IV border have been demonstrated to receive direct geniculate input10,11,13.) The second example is from a layer IV simple cell simulta-

% max. resp.

Orientation

Spikes

ms

Fig. 2. A layer IV simple cell simultaneously recorded with a layer II complex cell. (The electrode track for the complex cell ran within the top 100 m of layer II.) (a) Superimposed receptive fields of the simple cell, mapped with white noise29,30, and the complex cell, mapped with a moving bar. Solid lines, on responses; dashed lines, off responses. The darkest gray corresponds to the strongest responses. As expected from previous studies21, vertically aligned layer IV simple cells and layer II and III complex cells had similar receptive field sizes (complex/simple, 1.18 0.41). (b) Orientation tuning curves of the simple cell and the complex cell obtained with a moving bar. Solid line, simple cell; dotted line, complex cell. (c) Cross-correlation between the firing patterns of the simple cell and the complex cell. The ordinate gives the number of simple-cell spikes that occurred before (right) or after (left) a complex-cell spike. This correlogram shows a peak displaced to the right of zero, indicating that the simple cell tended to fire before the complex cell. The characteristics of the positive peak are consistent with a direct excitatory connection from the simple cell to the complex cell (inset). The dip on the left side of the correlogram matched the autocorrelogram of the simple cell, as would be expected from a monosynaptic connection31. The gray line represents the shuffle correlogram (or shift predictor), a correlation produced by the stimulus, not by direct neural connections 32. Both cells were simultaneously stimulated by a bar moved at the orientation preferred for the complex cell. Strength, 3.8%; correlogram asymmetry, 1.
396 nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

1998 Nature America Inc. http://neurosci.nature.com

Fig. 3. A layer IV simple cell b a simultaneously recorded with a complex cell located near the layer III/IV border. (a) Receptive fields of the simple cell (left) and the complex cell (right) both mapped with white noise. The dotted circle superimposed on both panels represents the receptive field of the multi-unit activity that was simultaneously Orientation recorded from layer A of LGN. (b) Orientation tuning curves c of the simple cell (solid line) and the complex cell (dashed line) obtained with a moving bar. (c) Cross-correlations between the simple cell and the complex cell (right), the geniculate multiunit and the complex cell (middle), and the geniculate multi-unit and the simple cell (left). All cells were simultaneously stimulated with a moving ms ms ms bar. (Gray line is the shuffle correlogram.) The geniculocortical correlations (middle and left) show a peak displaced from zero toward the right, indicating that both the simple cell and the complex cell received direct geniculate input. The intracortical correlation (right) indicates that the simple cell and the complex cell tended to fire simultaneously, probably because they received input from a common source (inset, right). Strength, 21.0% (right correlogram), 19% (middle correlogram), 16% (left correlogram). Simplecomplex cell correlogram asymmetry, 0.16.
Spikes Spikes % max. resp. Spikes

neously recorded with a complex cell located at the layer III/IV border (Fig. 3). Again, the cells had overlapping receptive fields (Fig. 3a; mapped with white noise) and similar orientation preferences (Fig. 3b). Unlike the previous example, however, the correlation between the simple cell and the complex cell had a positive peak centered at zero. This centered peak indicates that both cells tended to fire simultaneously, probably because they received input from a common source (Fig. 3c, right correlogram). Because the two cells were recorded from within the range of geniculate afferents, the common source of input could originate in the thalamus. To test this hypothesis, we recorded multi-unit activity in layer A of the LGN (from a region in retinotopic alignment with the cortical receptive fields; Fig. 3a). As expected, the geniculocortical correlograms indicated that both the simple cell and the complex cell received direct geniculate input (Fig. 3c, left and middle). Other correlograms between simple cells and complex cells had shapes between those depicted in Figs 2 and 3. In the third example (Fig. 4), we recorded simultaneously from a layer IV simple cell and a layer III complex cell. In this case, cross-correlation analysis showed a strong positive peak displaced towards the right, as in Fig. 2, as well as a slight correlation at the zero bin, as in Fig. 3 (Fig. 4c, right). The displaced positive peak is consistent with a direct excitatory connection from the simple cell to the complex cell. In addition, the slight correlation at zero may originate from shared common inputs (Fig. 4c, inset). To determine a possible source of common input, we again recorded multi-unit activity from geniculate layer A. Unlike the previous example, however, a positive geniculocortical correlation was found only for the simple cell and not for the complex cell (Fig. 4c, left and middle). This finding indicates that other intracornature neuroscience volume 1 no 5 september 1998

tical (for example, simple cells) or thalamic sources may be able to generate a slight correlation at time zero between simple cells and complex cells. The correlograms between simple cells and complex cells were usually wider than the geniculocortical correlograms3,12,34. This may reflect differences in the relative size of the excitatory postsynaptic potential (EPSP)35, the firing properties of the postsynaptic neuron 36 or the contribution from a compound polysynaptic EPSP added to a monosynaptic EPSP37. The main differences among the correlograms between simple cells and complex cells, however, were in their asymmetry with respect to zero. To quantify these correlogram asymmetries (CA), we measured the right side (R) and the left side (L) of the correlogram: CA = (R-L)/(R+L). The distribution of correlograms according to their symmetry had two distinct characteristics (Fig. 5): no correlation was strongly displaced (CA < 0.3) in the direction from the complex cell to the simple cell, and the distribution had two peaks (at zero and at one) that correlated well with the characteristics of the complex cell recorded. Complex cells with asymmetric correlograms tended to be located superficially in the cortex, whereas complex cells with symmetric correlograms were found closer to the layer III/IV border (Fig. 6a). Within the superficial layers, only cells near the layer III/IV border are known to receive direct geniculate input10,11,13; our data therefore indicate that complex cells with asymmetric correlograms did not receive common geniculate input along with simple cells (Fig. 6a). This idea is also supported by experiments using cross-correlation analysis between multi-unit activity in geniculate layer A and cortical single units. We were able to record simultaneously from correlated simple cells, complex cells and geniculate multi-unit activity in 32 cases (Fig. 6b). In these exper397

1998 Nature America Inc. http://neurosci.nature.com

articles

1998 Nature America Inc. http://neurosci.nature.com

Fig. 4. A layer IV simple cell a b Simple cell Complex cell simultaneously recorded with a layer III complex cell. (a) Receptive fields of the simple cell and the complex cell, each mapped with sparse noise (light or dark squares on a gray background)33. Each receptive field is represented by two panels. The left panel shows responses to light Orientation spots (solid lines, on c responses); right panel, responses to dark spots (dashed lines, off responses). The simple cell had an on-subregion that only responded to the presentation of light spots (left panel), and an off-subregion that responded only to the presentation of dark spots (right panel). The complex cell produced onoff responses; it responded ms ms weakly to light spots (left ms panel) and much more strongly to dark spots (right panel). The dominant response sign of simple cells and complex cells (on or off ) was not a good predictor of connectivity. (That is, the simple cell could produce mainly on-responses and the complex cell mainly off-responses and still be connected.) This finding is consistent with models in which complex cells are constructed from convergent inputs of simple cells with receptive fields in counter-phase1,2426. (b) Orientation tuning curves of the simple cell (solid line) and the complex cell (dashed line) obtained with a moving bar. (c) Cross-correlations between the simple cell and the complex cell (right), the geniculate multi-unit and the complex cell (middle), and the geniculate multi-unit and the simple cell (left). All cells were simultaneously stimulated with a moving bar. (Gray lines are shuffle correlograms.) Only the simple cell (left), not the complex cell (middle), seems to receive a direct geniculate (layer A) input, as estimated by crosscorrelation analysis. The correlation between the simple cell and the complex cell (right) shows a displaced peak, which slightly crosses the zero of the correlogram. This correlation is consistent with the circuit diagram illustrated in the inset on the right. Strength, 9.6% (right correlogram) or 2.8% (left correlogram). Asymmetry of the simplecomplex cell correlogram, 0.84.
Spikes Spikes Spikes % max. resp.

iments, we found evidence for direct geniculate input to every simple cell studied but not to most complex cells that had very asymmetric correlograms with simple cells. Therefore, the pronounced asymmetric correlograms between simple cells and complex cells may be generated by a monosynaptic connection and not by shared common inputs from the thalamus (Figs 5 and 6a and b). Consistent with this interpretation, most correlograms that were asymmetric with respect to zero had also asymmetric peak shapes or peak asymmetries (PA) (Fig. 6c). In a monosynaptic connection, the shape of the correlogram peak should be determined by the fast rise time and the slow decay time of the monosynaptic excitatory postsynaptic potentical (Fig. 6c). In contrast, shared common inputs should tend to generate correlograms with rise and decay times similar to each other. We plotted the correlogram asymmetry (CA) as a function of the peak asymmetry: PA = |RTDT|/(RT+DT), where RT is rise time and DT is decay time (Fig. 6c) As expected, PA and CA were significantly correlated (correlation coefficient, 0.7; p < 0.001). In addition to their noticeable asymmetry, the correlograms between simple cells and complex cells had many characteristics consistent with monosynaptic connections. The peak times of asymmetric correlograms (3.65 2.09 ms) were consistent with the peak times from intracortical monosynaptic EPSPS (latency + rise time = 2.364.96 ms) averaged from five studies: 1.47.8 ms, 2.14.1 ms, 45 ms, 1.33.7 ms and 3.04.2 ms; refs 4044,
398

respectively. Disynaptic intracortical EPSPS are expected to have much longer peak times, with latencies longer than 5 ms (refs 4345). The rise times of the symmetric correlograms (-0.3 < CA < 0.3) were much slower (8.94 5.04 ms) than the rise times of asymmetric correlograms (particularly for correlograms with little zero crossing, 2.23 1.03 ms for CA > 0.7; 4.34 3.56 for CA > 0.4; rise time and CA were inversely correlated; r = -0.54, p < 0.001). Asymmetric correlograms were most frequently found and were stronger between cells that were in the same column (orientation difference less than 22 degrees, total receptive field overlap; Table 1). Some of the strongest asymmetric correlograms (Figs 2 and 4) showed features on the left side of zero that matched the autocorrelogram of the presynaptic cell (as would be expected with a monosynaptic connection31). This finding that many layer II and III complex cells receive inputs from simple cells but not from the thalamus still leaves open an important question. To determine if superficial complex cells can share convergent inputs from simple cells (layer IV) and first-order complex cells (cells in layer III/IV border that receive thalamic input), we recorded from triplets of simple cells, firstorder complex cells and layer II and III complex cells, all vertically aligned within the cortex (n = 8; the quadruplet in Fig. 7 is two triplets). In all these cases, we found a correlation consistent with a monosynaptic connection between the simple cell and the superficial complex cell (CA = 0.74 0.14). However, we found a similar correlation between complex cells (layer III/IV border
nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

1998 Nature America Inc. http://neurosci.nature.com

Fig. 5. Distribution of correExcitatory connection Common input Non-correlated a lations between simple cells and complex cells according to their symmetry with respect to zero. (a) Four different types of correlograms; from left to right: a flat correlogram, a symmetric correlogram, a displaced peak that Time (ms) Time (ms) Time (ms) Time (ms) crosses the zero of the correlogram and a displaced n = 59 peak without zero crossing. The gray lines are the shuffle b correlograms. Different types of correlograms are consistent with different types of connectivity. The interpretan = 82 tion (labelled horizontal lines at top ) is based on the correlogram asymmetry. This asymmetry has been quantified by comparing the right or positive side of the correlogram (R) with the left or negCorrelogram asymmetry (CA) ative side (L) within 5 ms (each side). This ratio (CA) is defined as LR/L+R. (b) Right, histogram showing the distribution of 82 correlograms between simple cells and complex cells according to their symmetry; left, total number of flat correlograms. Many correlograms were flat despite the precise vertical alignment of the recordings, which may result from the specificity of the projection (cells in different layer IV sublaminae may project to specific depths within layer II and III; ref. 38) and/or differences in the circuitry among several complex cell types 17,39.
Number of cell pairs

and superficial) in only two cases (Fig. 7, quadruplet). In one experiment, we were able to record from a layer IV simple cell, two layer III/IV border complex cells and a superficial complex cell simultaneously (Fig. 7, left). The receptive fields of the layer IV simple cell (Fig. 7, B) and the two first-order complex cells (Fig. 7, A and C) were mapped with white noise. The receptive field of the superficial complex cell (Fig. 7, D) was mapped with a moving bar. In this case, both the layer IV simple cell and the two layer III/IV border complex cells showed a positive correlation with the superficial complex cell. Although the simple cell was not the best match for orientation preference, it still had the strongest correlation (Fig. 7, orientation tuning curves at top; the shaded curve is from the superficial complex cell). In the remaining six triplets studied, only the simple cell showed a positive correlation with the superficial complex cell (Fig. 7, right).

Discussion Our results indicate that layer IV simple cells send monosynaptic input to layer II and III complex cells with similar orientation preferences. However, as the evidence for monosynaptic connections is based on the cells correlated firing, it is important to consider other alternative circuits that could make a simple cell fire before a complex cell. There are four possibilities. (A disynaptic circuit is not included because intracortical disynaptic correlograms are expected to be weak and broad and have much longer peak times4345.) First, the simplest case is a monosynaptic connection (Fig. 8a); this circuit is supported by the evidence presented here. Second, both the complex cell and the simple cell might receive input from common geniculate cells (from axons with different conduction velocities; Fig. 8b). This possibility is unlikely because asymmetric correlograms are predominantly found in complex cells
nature neuroscience volume 1 no 5 september 1998

recorded from outside the range of the vast majority of geniculate afferents (Fig. 6b). Also, as asymmetric correlograms are always displaced in the direction from the simple cell to the complex cell (Fig. 5), such a circuit would require that simple cells receive inputs from faster geniculate afferents than complex cells. This is not supported by many studies using electrical stimulation and intracellular recordings 11,13 , electrical stimulation and extracellular recordings9,10,46, and cross-correlation analysis12. Third, a geniculate cell could connect monosynaptically to the simple cell and disynaptically to the complex cell (Fig. 8c). This possibility is unlikely because it includes an intracortical disynaptic connection 6,7,12,35,47 . Because each intracortical connection produces relatively weak correlated firing (10% of correlated spikes on average3,6,7,12,34; Table 1), a disynaptic connection should produce barely significant correlations (1% or smaller). Moreover, the large jitter generated by intracortical disynaptic connections should smear the correlogram4345. In support of this line of reasoning, the disynaptic connections between geniculate cells and complex cells tend to generate flat correlograms (Figs 4 and 6b), and 40% of vertically aligned simple cells and complex cells do not show a prominent correlation (although many of these cells should be disynaptically connected; Fig. 5). Fourth, a retinal ganglion cell might connect to a fast and a slow geniculate cell (for example, Y cell and W cell; Fig. 8d). The W cell connects to the complex cell and the Y cell to the simple cell. This circuit is unlikely for reasons similar to those noted for the second circuit possibility (Fig. 8b). Also, W and Y geniculate cells do not usually receive input from common retinal afferents22,48. Moreover, the distance between LGN and striate cortex is too short (~15 mm) and the differences in conduction velocities too small (< 50 m/s between Y and W cells22,48) to generate a large spread in the
399

1998 Nature America Inc. http://neurosci.nature.com

articles

1998 Nature America Inc. http://neurosci.nature.com

Fig. 6. Complex cells with asymmetric correlograms were a b usually located superficially in the cortex and were not correlated with layer A geniculate multi-unit activity. In these comr = 0.44 plex cells, the correlogram was asymmetric not only with p < 0.01 respect to the zero bin, but also with respect to the peak. n = 40 (a) Laminar location of the complex cell as a function of the asymmetry in the simplecomplex cell correlogram. Complex cells with asymmetric correlograms tended to be located superficially in the cortex and those with symmetric correlograms near the layer III/IV border. The laminar position was calculated by measuring the distance from the electrode track to the layer III/IV border and normalizing by the cortical thickness. Only those electrode tracks almost parallel to the cortical surCorrelogram asymmetry (CA) Correlogram asymmetry (CA) face were plotted. Measurements were made in steps of 100 m (the average width of the DiI/DiO electrode track). The open circles are the cells from Fig. 2 (CA = 1), Fig. 3 (CA = 0.16) and Fig. 4 (CA = 0.84). n = 40; r = 0.44; p < 0.01. c (b) Strength of the geniculocortical correlations as a function of the correlogram asymmetry between simple cells (filled squares) and complex cells (open squares). When the correlor = 0.70 p < 0.001 gram between the simple cell and the complex cell was symn = 51 metric, both cells showed positive geniculocortical correlations. However, when the simple-complex cell correlogram had a pronounced asymmetry, only the simple cell and not the complex cell seemed to receive direct geniculate input. Some direct geniculate connections to complex cells may have been missed because of the limitations of extracellular recordings11 or because the multi-unit activity did not contain any geniculate cell connected to the complex cell. However, this technical limitation should not be important, as we always Correlogram asymmetry (CA) found direct geniculate connections for all simple cells (n = 32) and all complex cells that gave strong responses to static stimuli (n = 7) (receptive field overlap between geniculate multi-unit and cortical cells > 50%). Vertical lines connecting squares represent cells recorded simultaneously. (c) Correlograms that were asymmetric with respect to zero (CA) were also asymmetric with respect to the peak (PA), and vice versa. Right, two examples of correlograms with different peak asymmetry. Each correlogram is split into two halves (continuous and dotted lines) that are shown superimposed. The two halves are nearly identical for symmetric correlograms (bottom) but not for asymmetric correlograms (top). PA = |RT-DT|/(RT+DT) where RT is rise time and DT is decay time. Left, PA and CA were strongly correlated. Because most spikes are produced in the EPSP rising phase 35,37, part of the correlogram decay is near the baseline. Therefore, only correlograms with large positive peaks and low baseline noise were measured. Open circles are cells from Figs 2, 3 and 4. n = 51; r = 0.70; p < 0.001.
Peak asymmetry (PA) Correlation strength (%) Cortical layer

correlation peak47. In general, if the mechanisms for the last three circuits (Fig. 8bd) were able to generate asymmetric correlograms, these correlograms should be found over a large region of cortex (a geniculate axon covers 11.5 mm2 of cortical surface 49 ), in particular between different orientation columns. However, this is obviously not true for either asymmetric correlograms between simple cells and complex cells or asymmetric correlograms between complex cells6,7,50. Our results (Fig. 7) also indicate that some superficial complex cells may receive convergent input from simple cells and first-order complex cells. The relative strength of each type of input, however, cannot be determined here. In the eight triplets examined, the correlation between first-order complex cells and superficial complex cells was weak or flat. This may indicate either weak convergence, a sampling problem (for example, in two of eight triplets, the complex cells had different orientation preferences) or that the connections from first-order complex cells were too weak to be detected by crosscorrelation analysis. The idea that the inputs from simple cells and these first-order complex cells converge into the superficial layers seems reasonable, as these cell types have many features in common. Both produce strong responses to stationary stimuli, and the responses are all either on or off and show
400

rather linear spatial summation. In fact, these first-order complex cells were classified by some as simple cells22 and by others as a separate class (E-on or E-off)6,7,12. Three previous studies did not find evidence for monosynaptic connections from simple cells to complex cells by using cross-correlation analysis68. In two cases 6,7, monosynaptic connections may have been missed because of small electrode misalignments or the small sample size (between vertically aligned simple cells and complex cells). A subsequent report8 overcame the alignment problem by doing cross-correlation analysis between cells recorded with the same electrode; however, this approach limited the sample to neighboring cells, most likely recorded within the same cortical layer. Our results provide physiological evidence for monosynaptic connections from layer IV simple cells to layer II and III complex cells with spatially overlapping receptive fields and similar orientation preferences. In addition, we have shown that some complex cells in lower layer III share geniculate inputs (and perhaps cortical targets) with simple cells. Therefore, both serial and parallel circuits are involved in the generation of complex receptive fields, which could provide the basis for the differences observed in complex cell responses to visual stimulation.
nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

Fig. 7. Convergence of layer IV Superficial complex cell simple cells and layer III/IV border complex cells onto layer II and III complex cells. Left, a layer IV simple cell simultaneously recorded with two complex cells located at the layer III/IV border and Orientation another complex cell located in layers II and III. The receptive Superficial fields of the layer IV simple cell complex cell and the two layer III/IV border complex cells are shown at the bottom of the figure (boxes A, B and C). The receptive field of the superficial complex cell (box D) is shown as a shaded square at the top of the figure (superimposed with receptive field drawings from the other three cells). Receptive fields were mapped with white noise by reverse correlation (layer IV simple cell and layer III/IV border complex cells) and with ms ms moving bars (superficial complex ms cell). All cells had similar orientation preferences and overlapping receptive fields (top of the figure; the receptive field and orientation tuning curve for the superficial complex cell are shaded). The ms ms correlation between the simple Complex cell Complex cell Simple cell A Complex cell B cell and the superficial complex (III-IV border) (III-IV border) (Layer IV) (III-IV border) cell (BD) had a positive peak Simple cell (Layer IV) displaced from zero consistent with a monosynaptic connection (correlation strength, 4%). In this case, weaker but prominent peaks were also found between the layer III/IV border complex cells and the superficial complex cell (correlation strength for AD, 2%; for CD, 2%). Right, examples of correlograms from the remaining six triplets (a layer IV simple cell, a layer III/IV border complex cell and a superficial complex cell simultaneously recorded). All correlations between the simple cell and the superficial complex cell were positive. In contrast, there were no prominent positive correlations between the layer III/IV border complex cells and the superficial complex cell. A distinct negative correlation was found in one of the cases, which indicates that some of these cells may be inhibitory neurons6,7. Inset vertical and crossed lines show the differences in orientation between first-order cells (layer IV simple cell, layer III/IV border complex cell) and the superficial complex cell. In the two cases at the bottom, the absence of a positive correlation between complex cells may be due to differences in orientation preferences.
spikes spikes spikes

1998 Nature America Inc. http://neurosci.nature.com

Methods CORTICAL RECORDINGS. Cats (2.53 kg) were anesthetized with ketamine (10 mg/kg) and thiopental sodium (20 mg/kg; maintenance 2 mg/kg per hour; administered intravenously) and paralyzed with Norcuron (0.2 mg/kg per hour, administered intravenously). Temperature (37.538C), EKG, EEG, and expired CO2 were monitored throughout the experiment. Pupils were dilated with 1% atropine sulfate and the nictitacting membranes retracted with 10% phenylephrine.

A multi-electrode matrix was used for the cortical recordings 27. Recorded signals were collected by a computer running Datawave Systems software (Broomfield, Colorado). Usually, five or six electrodes were introduced into the medial bank of the cortex parallel to each other and parallel to the cortical surface (electrode thickness, 80 m). The tips of the electrodes were coated with DiI or DiO, and the tracks were reconstructed at the end of the experiment28. The electrode tracks were always within the same or four contiguous 40-m sections (Fig. 1). Each cells laminar location was estimated by combining depth readings and the location of electrode tracks. Because the electrodes ran parallel to each other and parallel to the cortical surface, the measurement error was mostly imposed by the thickness of the DiI/DiO track (100 m).
nature neuroscience volume 1 no 5 september 1998

MAPPING FIELDS. Receptive fields were mapped by reverse correlation using two different stimuli: binary dense white noise, and ternary sparse noise. White noise29,30 consisted of a series of 16 16-pixel random checkerboards (0.4 degrees/pixel) presented each during 20 ms. Sparse noise33 was made of large (1.2-degree) light and dark spots, each presented during 40 ms in a gray background. Layer IV simple receptive fields could always be mapped with all types of stimuli (white noise, sparse noise and moving bars). In contrast, layer II and III complex receptive fields could be mapped only with sparse noise and/or moving bars in most cases (Figs 2 and 4, compared with Fig. 3). Receptive fields are shown as contour plots (concentric lines representing 10% decrements in response with respect to the maximum). DATA ANALYSIS. The correlations were obtained by using a moving bar as a stimulus. Cells with overlapping receptive fields showed a slow stimulus-dependent correlation that was calculated by separating the cells responses to each period of the repetitive visual stimulus and shuffling these fragments32. The significance level for a positive correlation was set to 2.5 times the standard deviation of the shuffle correlogram (or baseline). For correlograms between cortical neurons, the correlation strength was calculated, after subtracting the baseline, as A/B, where A
401

1998 Nature America Inc. http://neurosci.nature.com

articles

Sonia and Claudia. This study was supported by the NIH and Human Frontier Science Program Organization.

RECEIVED 29 MAY; ACCEPTED 28 JULY


1. Hubel, D. H. & Wiesel, T. N. Receptive fields, binocular interaction and functional architecture in the cats visual cortex. J. Physiol. (Lond.) 160, 106154 (1962). 2. Chapman, B., Zahs, K. R. & Stryker, M. P. Relation of cortical cell orientation selectivity to alignment of receptive fields of the geniculocortical afferents that arborize within a single orientation column in ferret visual cortex. J. Neurosci. 11, 13471358 (1991). 3. Reid, R. C. & Alonso, J. M. Specificity of monosynaptic connections from thalamus to visual cortex. Nature 378, 281284 (1995). 4. Ferster, D., Chung, S. & Wheat, H. Orientation selectivity of thalamic input to simple cells of cat visual cortex. Nature 380, 249252 (1996). 5. Sompolinsky H. & Shapley R. New perspectives on the mechanisms for orientation selectivity. Curr. Opin. Neurobiol. 7, 514522 (1997). 6. Toyama, K., Kimura, M. & Tanaka, K. Cross-correlation analysis of interneuronal connectivity in cat visual cortex. J. Neurophysiol. 46,191201 (1981). 7. Toyama, K., Kimura, M. & Tanaka, K. Organization of cat visual cortex as investigated by cross-correlation analysis. J. Neurophysiol. 46, 202214 (1981). 8. Ghose, G. M., Freeman, R. D. & Ozhawa, I. Local intracortical connections in the cats visual cortex: postnatal development and plasticity. J. Neurophysiol. 72, 12901303 (1994). 9. Hoffman, K. P. & Stone, J. Conduction velocity of afferents of cat visual cortex: a correlation with cortical receptive field properties. Brain Res. 32, 460466 (1971). 10. Bullier J. & Henry, G. H. Laminar distribution of first order neurons and afferent terminals in cat striate cortex. J. Neurophysiol. 42, 12711281 (1979). 11. Ferster, D. & Lindstrom, S. An intracellular analysis of geniculo-cortical connectivity in area 17 of the cat. J. Physiol. (Lond.) 342, 181215 (1983). 12. Tanaka, K. Cross-correlation analysis of geniculostriate neuronal relationships in the cat. J. Neurophysiol. 49, 13031318 (1983). 13. Martin, K. A. C. & Whitteridge, D. Form, function and intracortical projections of spiny neurons in the striate visual cortex of the cat. J. Physiol. (Lond.) 353, 463504 (1984). 14. Heggelund, P. Receptive field organization of complex cells in cat striate cortex. Exp. Brain. Res. 42, 99107 (1981). 15. Mel, B. W., Ruderman, D. L. & Archie, K. A. Translation-invariant orientation tuning in visual complex cells could derive from intradendritic computations. J. Neurosci. 18, 43254334 (1998). 16. Hammond, P. & MacKay, D. M. Differential responsiveness of simple and complex cells in cat striate cortex to visual texture. Exp. Brain Res. 30, 275296 (1977). 17. Malpeli, J. G. Activity of cells in area 17 of the cat in absence of input from layer A of lateral geniculate nucleus. J. Neurophysiol. 49, 595610 (1983). 18. Gilbert, C. D. & Wiesel, T. N. Morphology and intracortical projections of functionally characterized neurons in the cat visual cortex. Nature 280, 120125 (1979). 19. Lund, J. S., Henry, G. H., Macqueen, C. L. & Harvey, A. R. Anatomical organization of the primary visual cortex (area 17) of the cat. A comparison with area 17 of the macaque monkey. J. Comp. Neurol. 184, 599618 (1979). 20. Hirsch, J. A., Alonso, J. M. & Reid, R. C. Visually evoked calcium action potentials in cat striate cortex. Nature 378, 612616 (1995). 21. Gilbert, C. D. Laminar differences in receptive field properties of cells in cat primary visual cortex. J. Physiol. (Lond.) 268, 391421 (1977). 22. Orban, G. A. Neuronal Operations in the Visual Cortex. (Springer-Verlag, Berlin, 1984). 23. Movshon J., Thompson I. & Tolhurst D. Receptive field organization of complex cells in cats striate cortex. J. Physiol. (Lond.) 283, 7999 (1978). 24. Adelson, E. H. & Bergen, J. R. Spationtemporal energy models for the perception of motion. J. Opt. Soc. Am. A2, 284299 (1985). 25. Ohzawa, I., DeAngelis, G. C. & Freeman, R. D. Stereoscopic depth discrimination in the visual cortex: neurons ideally suited as disparity detectors. Science 249, 10371041 (1990). 26. Heeger, D. J. Normalization of cell responses in cat striate cortex. Visual Neurosci. 9, 181197 (1992). 27. Eckhorn, R. & Thomas, U. Guidance of thin-shaft microprobes by rubber tubes: a new method for the insertion of multiple microprobes into neuronal and muscular tissue, including fiber-electrodes, fine wires, needles and microsensors. J. Neurosci. Methods. 49, 175185 (1993). 28. Snodderly D. M. & Gur, M. Organization of striate cortex of alert, trained monkeys (Macaca fascicularis): ongoing activity, stimulus selectivity, and widths of receptive field activating regions. J. Neurophysiol. 74, 21002125 (1995). 29. Sutter, E. in Advanced Methods of Physiological Systems Modeling Vol. 1, 303315 (Univ. Southern California, 1987). 30. Reid, R. C., Victor, J. D. & Shapley, R. M. The use of m-sequences in the analysis of visual neurons: linear receptive field properties. Visual Neurosci. 14, 10151027 (1997). 31. Moore, G. P., Segundo, J. P., Perkel, D. H. & Levitan, H. Statistical signs of synaptic interaction in neurons. Biophys. J. 10, 876900 (1970).

d
Retina LGN Visual cortex

1998 Nature America Inc. http://neurosci.nature.com

Fig. 8. Four different circuits that could make a simple cell fire before a complex cell. According to our results, distinctly asymmetric correlograms are probably generated by a monosynaptic connection (circuit A). S, simple cell; C, complex cell.

is the number of complex cell spikes preceded by a simple cell spike within 110 ms; B = ((S2+C2)/2), where S is the total number of spikes produced by the simple cell and C is the total number of spikes produced by the complex cell. Because the firing rates of simple cells and complex cells can differ sometimes by several orders of magnitude, B is also a correction factor. The correlation strength between geniculate multi-unit and cortical units was calculated as C/D, where C is the number of cortical spikes preceded by a geniculate spike within 16 ms, minus baseline, and D is the total number of cortical spikes. The correlogram asymmetry (CA) between simple cells and complex cells was calculated, after subtracting the baseline, as R-L/R+L, where R is the number of paired spikes at the right side of the correlogram and L is those at the left side, within 5 ms on each side. Correlograms with CAs larger than 0.4 were taken as an indication for a monosynaptic connection for three reasons: we never found correlograms with CA < 0.3 (Fig. 5), most complex cells with CA > 0.4 were recorded superficially in the cortex (Fig. 6a), and we could not demonstrate a prominent geniculate layer A input for most complex cells with CA > 0.4 (Fig. 6b). The peak asymmetry (PA) was calculated as |RT-DT|/(RT+DT), where RT is rise time and DT is decay time. Rise times and decay times were measured from the baseline to 70% of the peak (Fig. 6c). (These are the ranges where the largest differences between RT and DT were found.) Of the positive correlograms (peak > 2.5 standard deviation of baseline), we included in the analysis of PA only those with large peaks and low noise, as determined arbitrarily with the equation peak > 0.1 baseline + (280 standard deviation (baseline)/(peakbaseline)). For this purpose, the baseline was defined as the average value in the correlogram, 10 ms on each side. Simple cells and complex cells were classified using the original Hubel and Wiesel1 criteria. According to these criteria, simple receptive fields have separate and elongated on- and off-subregions whose spatial organization matches the cells optimal orientation. In contrast, complex cells respond to variously shaped stationary or moving forms in a way that is not predictable from maps made with small circular spots1. Using these criteria, layer III/IV border cells with round receptive fields were classified as complex cells (Figs 3 and 7). These complex cells may have been classified by others as simple cells with one subregion22 or as E-on or E-off cells6,7,12.

Acknowledgements
We thank T.N. Wiesel for discussion and suggestions. We also thank C. Reid, D. Ferster, R. Freeman and J. Hirsch for criticisms and comments. Technical assistance was provided by K. Desai, J. Kornblum, and K. McGowan. Thanks to

402

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

1998 Nature America Inc. http://neurosci.nature.com

32. Perkel, D. H., Gerstein, G. L. & Moore, G. P. Neuronal spike trains and stochastic point processes. II. Simultaneous spike trains. Biophys. J. 7, 419440 (1967). 33. Jones, J. P. & Palmer, L. A. The two-dimensional spatial structure of simple receptive fields in cat striate cortex. J. Neurophysiol. 58, 11871211 (1987). 34. Alonso, J. M., Usrey, W. M. & Reid, R. C. Precisely correlated firing in cells of the lateral geniculate nucleus. Nature 383, 815819 (1996). 35. Fetz, E. E. & Gustafsson, B. Relation between shapes of post-synaptic potentials and changes in firing probability of cat motoneurons. J. Physiol. (Lond.) 341, 387410 (1983). 36. Surmeier, D. J. & Weinberg, R. J. The relationship between cross-correlation measures and underlying synaptic events. Brain Res. 331, 180184 (1985). 37. Matsumura, M., Chen, D., Sawaguchi, T., Kubota, K. & Fetz, E. E. Synaptic interactions between primate precentral cortex neurons revealed by spiketriggered averaging of intracellular membrane potentials in vivo. J. Neurosci. 23, 77577767 (1996). 38. Fitzpatrick, D. The functional organization of local circuits in visual cortex: insights from the study of tree shrew striate cortex. Cereb. Cortex 6, 329341 (1996). 39. Sillito, A. M. Inhibitory processes underlying the directional specificity of simple, complex and hypercomplex cells in the cats visual cortex. J. Physiol.(Lond.) 271, 699720 (1977). 40. Markram, H., Lubke, J., Frotscher, M., Roth, A. & Sakmann, B. Physiology and anatomy of synaptic connections between thick tufted pyramidal neurones in the developing rat neocortex. J. Physiol. (Lond.) 500, 409440 (1997). 41. Stratford, K. J., Tarczy-Hornoch, K., Martin, K. A. C., Bannister, N. J. & Jack J. J. B Excitatory synaptic inputs to spiny stellate cells in cat visual cortex.

Nature 382, 258261 (1996). 42. Deuchars, J., West, D. C. & Thomson, A. M. Relationships between morphology and physiology of pyramid-pyramid single axon connections in rat neocortex in vitro. J. Physiol. (Lond.) 478, 423435 (1994). 43. Mason, A., Nicoll, A. & Stratford, K. Synaptic transmission between individual pyramidal neurons of the rat visual cortex in vitro. J. Neurosci. 11, 7284 (1991). 44. Ferster, D. & Lindstrom, S. Synaptic excitation of neurons in area 17 of the cat by intracortical axon collaterals of cortico-geniculate cells. J. Physiol. (Lond.) 367, 233252 (1985). 45. Fitzsimonds, R. M., Song, H. J. & Poo, M. M. Propagation of activity-dependent synaptic depression in simple neural networks. Nature 388, 439448 (1997). 46. Singer, W., Tretter, F. & Cynader, M. Organization of cat striate cortex: a correlation of receptive-field properties with afferent and efferent connections. J. Neurophysiol. 38, 10801098 (1975). 47. Kirkwood, P. A. On the use and interpretation of cross-correlation measurements in the mammalian central nervous system. J. Neurosci. Methods 1, 107132 (1979). 48. Sherman, S. M. & Guillery, R. W. Functional organization of thalamocortical relays. J. Neurophysiol. 76, 13671395 (1996). 49. Humphrey, A. L., Sur, M., Uhlrich, D. J. & Sherman, S. M. Projection patterns of individual X- and Y-cell axons from the lateral geniculate nucleus to cortical area 17 in the cat. J. Comp Neurol. 233, 159189 (1985). 50. Tso, D. Y., Gilbert C. D. & Wiesel T. N. Relationships between horizontal interactions and functional architecture in cat striate cortex as revealed by cross-correlation analysis. J. Neurosci. 6, 11601170 (1986).

nature neuroscience volume 1 no 5 september 1998

403

1998 Nature America Inc. http://neurosci.nature.com

articles

A neural correlate for vestibulo-ocular reflex suppression during voluntary eyehead gaze shifts
Jefferson E. Roy and Kathleen E. Cullen
Aerospace Medical Research Unit, McGill University, Montreal, Quebec H3G 1Y6, Canada Correspondence should be addressed to K.E.C. (cullen@medcor.mcgill.ca)

1998 Nature America Inc. http://neurosci.nature.com

The vestibulo-ocular reflex (VOR) is classically associated with stabilizing the visual world on the retina by producing an eye movement of equal and opposite amplitude to the motion of the head. Here we have directly measured the efficacy of VOR pathways during voluntary combined eyehead gaze shifts by recording from individual vestibular neurons in monkeys whose heads were unrestrained. We found that the head-velocity signal carried by VOR pathways is reduced during gaze shifts in an amplitude-dependent manner, consistent with results from behavioral studies in humans and monkeys. Our data support the hypothesis that the VOR is not a hard-wired reflex, but rather a pathway that is modulated in a manner that depends on the current gaze strategy.

The vestibulo-ocular reflex (VOR) produces a compensatory eye movement of equal and opposite amplitude to the motion of the head. This reflex effectively stabilizes the visual world on the retina for the wide range of head motions that are generated during daily activities1,2. However, primates can also voluntarily redirect their visual axes to new targets in their environment. Under natural conditions, in which the head is not restrained, a combination of eye and head movements (a gaze shift) is commonly used by humans39 and monkeys1015 to redirect the visual axis to a new point in space. During such voluntary orienting behaviors, however, the eye movements produced by the VOR would be counterproductive; the VOR would produce an eye-movement command in the direction opposite to that of the intended shift of the axis of gaze. Bizzi and colleagues1012 proposed that the VOR remains functional during eyehead gaze shifts, such that it eliminates the contribution of the head movement to the change in gaze. In this schema, the linear summation hypothesis6, the actual eye movement generated at the level of the extraocular motoneurons reflects the summation of two signals: an intact VOR mediated by vestibular pathways (that is, a compensatory eye-movement signal) and a separate saccadic eye-movement signal that redirects the fovea in space. More recent studies have provided results that argue against the linear summation hypothesis. This hypothesis cannot account for the observation that accurate gaze shifts can be generated to targets beyond the mechanical limits of ocular motility46,14. Moreover, a number of behavioral studies have shown that the gain of the VOR (eye velocity/head velocity) is considerably attenuated during gaze shifts, and the amount of attenuation increases as gaze shift amplitude increases from approximately 10 to 100 degrees57,13,1518. Previous studies have not, however, determined a neural correlate for this on-line suppression of the VOR during voluntary gaze shifts. A simple three-neuron arc (vestibular afferents to interneurons in the vestibular nuclei to extraocular motoneurons) mediates the most direct pathway of the VOR 19 . Studies in
404

head-restrained monkeys have shown that there are several classes of vestibular nuclei neurons that discharge in relation to horizontal passive whole-body rotation and/or eye movements2028. Position-vestibular-pause (PVP) neurons are thought to constitute most of the intermediate leg of the direct VOR pathway; PVP neurons receive a strong monosynaptic connection from the ipsilateral vestibular afferents and, in turn, project directly to contralateral extraocular motoneurons20,26,28. Here we investigated whether VOR suppression measured behaviorally during voluntary eye-head gaze shifts could be accounted for by a comparable reduction in the activity of the direct VOR pathways (that is, PVP neurons). First, we characterized the discharges of PVP neurons in the head-restrained condition, and then we released the head to record the discharges of these neurons during voluntary eyehead gaze shifts.

Results

NEURONAL ACTIVITY IN THE HEAD-RESTRAINED CONDITION


Figure 1 shows discharge patterns of a typical PVP neuron, unit 46_1, during saccadic eye movements and passive wholebody rotation. This neuron was representative of the PVP neurons in our sample (n = 9) in that its firing rate increased as a function of contralateral eye position during spontaneous eye movements (Fig. 1a), and its firing rate increased during ipsilateral whole-body sinusoidal rotation in the dark (VORd; Fig. 1b). We also did a second whole-body rotation experiment in which the monkey canceled its VOR by fixating on a headcentered visual target that moved with the vestibular turntable (VORc; Fig. 1c). This design allowed us to dissociate a neurons sensitivity to vestibular stimuli from its modulation related to eye movement. The activity of PVP neurons increased during ipsilateral whole-body sinusoidal rotation regardless of whether or not the animal was canceling its VOR (compare Fig. 1b and c). In addition, each of the neurons stopped firing (paused) during ipsilaterally directed saccades and vestibular nystagmus quick phases (Fig. 1a and b, arrows). A minority
nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

1998 Nature America Inc. http://neurosci.nature.com

of the PVP neurons (22%) also completely paused for saccades in the contralateral direction. To quantify neuronal discharges, we calculated each neurons resting discharge (bias), sensitivity to eye position (k x, (spikes/s)/degree), and sensitivity to passive head velocity (gx, (spikes/s)/(degrees/s)). For our sample of PVP neurons, the mean head-velocity sensitivity was significantly lower (~25%) when the monkey canceled its VOR compared with VORd (mean, gxc = 1.23 0.79 versus gxd = 1.61 0.52; p < 0.05). A similar decrease in sensitivity has been reported for PVP neurons in the squirrel monkey20,29. The discharges of 70% of the neurons in this study were also recorded during rapid passive head rotations in the dark on a fixed body (which elicited velocities as high as 300 degrees per s, with trajectories comparable to those generated in large gaze shifts described below). During these rotations, the firing rate of each cell was accurately predicted when gaze was stable (that is, during compensatory VOR), using a model based on the linear addition of each neurons bias and head-velocity sensitivity during VORd.

Saccades

Mean firing rate (spikes/s) Horizontal eye position (degrees)

1s

VORd

NEURONAL ACTIVITY DURING VOLUNTARY EYE-HEAD GAZE SHIFTS


After a neuron was fully characterized during head-restrained experiments, the monkeys head was slowly and carefully released to allow full freedom of head motion. During this critical transition, the unit activity and waveform were monitored on an oscilloscope to ensure that the cell remained undamaged and well isolated. Visual targets were presented to the monkey30 to elicit large gaze shifts of up to 75 degrees in amplitude. We characterized the head-velocity signal carried by each PVP neuron during voluntary gaze shifts of two amplitude ranges: small, 2030 degrees, and large, 5060 degrees. Because these cells increased their activity for ipsilateral head movement during passive whole-body rotation (Fig. 1b), only ipsilateral gaze shifts were considered. Typical gaze shifts are illustrated in Figs 2b and c and 3 for the example PVP neuron unit 46_1. As previously demonstrated31,32, eye and gaze velocity reached their peak values early during large gaze shifts; however, subsequent re-accelerations were often observed (Figs 2b and c and 3b). Most PVP neurons showed a cessation of activity (pause) that preceded the onset of ipsilateral gaze shifts by approximately 10 ms, similar to that previously reported29,32. None of the PVP neurons in our sample paused for the entire duration of ipsilaterally directed gaze shifts. In fact, for many neurons, activity resumed before the completion of the ocular saccade component of a gaze shift. For each cell, we first investigated whether the eye-position and head-velocity sensitivities estimated during head-restrained saccade and VORd experiments could predict the firing rate (fr) of the neuron during combined eyehead gaze shifts. The headrestrained model was given by the following equation (model 1): fr = bias + (kx eye position) + (gxd head velocity) (1)

1s

VORc

1s

We analyzed the example PVP neuron unit 46_1 using this model (Figs 2 and 3). Model 1 consistently overestimated the discharges of PVP neurons for ipsilaterally directed vestibular quick phases (VORd, Fig. 2a), during which all PVP neurons paused. However, model 1 provided an excellent estimate of PVP neuron activity before and after vestibular quick phases (Fig. 2a). Similarly, model 1 overestimated the discharges of each PVP neuron during large (5060 degree) gaze shifts, but provided a reliable estimate of PVP neuron discharges immediately before and after gaze shifts, even though the head was still moving ( Fig. 2b and c). In addition, the model did not overestimate the discharge of the neuron during the postnature neuroscience volume 1 no 5 september 1998

Fig. 1. Activity of an example PVP neuron (unit 46_1) during the head-restrained condition. (a) The PVP cell increased its discharge for contralateral eye movements (kx = -1.55 spikes/s/degree; sample mean, kx = -1.18 0.7 spikes/s/degree) and paused for ipsilateral saccades (arrows). Inset, Mean neuronal firing rate was well correlated with horizontal eye position during periods of steady fixation (r = -0.87, n = 129). (b, c) Passive whole-body rotation (0.5 Hz, 80 degrees/s peak velocity) was used to study the neurons response to head movements during VOR in the dark (VORd, (b)), and head movements during an experiment in which the monkey canceled its VOR by fixating a target that moved with the table (VORc, (c)). The headvelocity sensitivity coefficient of the PVP during VORd (gxd) and VORc (gxc) were 1.49 and 1.14 (spikes/s)/(degree/s), respectively. Unit 46_1 was typical of most PVP neurons in our study in that its modulation was less during VORc than during VORd. The arrows (b) indicate pauses in the activity of the PVP neurons during ipsilateral vestibular quick phases. Traces directed upward are in the ipsilateral direction. E, horizontal eye position; H, horizontal head , horizontal gaze velocity; E , horizontal position; FR, firing rate; G eye velocity; H , horizontal head velocity. deg, degree; sp, spikes.
405

1998 Nature America Inc. http://neurosci.nature.com

articles

1998 Nature America Inc. http://neurosci.nature.com

Fig. 2. Activity of an a b c example PVP neuron (unit 46_1) before, during and after 50 deg vestibular quick phases 25 deg (a) and large gaze shifts (b,c). The firing rate (shaded trace) is plotted above the output of the spike threshold discriminator for each gaze shift. A model based on headrestrained eye and passive head-movement 400 deg/s sensitivities (arrow, 100 deg/s model 1, dark trace) is superimposed on the model 1 firing rate traces. This 200 sp/s 200 sp/s model estimated accurately the discharge of PVP neurons before and immediately after quick phases and gaze 200 ms 500 ms shifts. In contrast, this model overestimated the discharge of this neuron during vestibular quick phases and gaze shifts. Because PVP neurons typically discharge at rates less than ~400 spikes/s, the firing rates predicted by model 1 were limited to values below this maximum. For example, this units firing rate was limited to 350 spikes/s, and accordingly the prediction of model 1 was not allowed to take on values exceeding this limit. G, horizontal gaze position; E, , horizontal gaze velocity; E , horizontal eye velocity; H , horizontal horizontal eye position; H, horizontal head position; FR, firing rate; G head velocity; deg, degree; sp, spikes. Dotted vertical lines indicate the onset and offset of vestibular quick phases (a) and gaze shifts (b,c), using a criterion of 25 degrees/s.

saccadic interval even when followed by a second corrective saccade (Fig. 2c). Model 1 consistently overestimated PVP neuron activity during small (2030 degree), as well as large gaze shifts (Fig. 3). To quantify this observation, we determined the best estimate of each neurons head-velocity sensitivity (gfs) during small and large gaze shifts using the equation (model 2): fr = bias + (kx eye position) + (gfs head velocity) (2)

where the values of the bias and kx were taken from the headrestrained analysis (above), and the model was fit to a data set that contained 20 or more gaze shifts within each amplitude range. A percent attenuation index, 1 - (gfs / gxd), was then calculated to compare the values that were estimated for gfs to those estimated for gxd in the head-restrained analysis. The example neuron here was typical in that its head-velocity sensitivity attenuation (compare model 1 and model 2) was greater for large gaze shifts (Fig. 3b) than for small ones (Fig. 3a). Figure 4a further illustrates the amplitude-dependent attenuation of gfs that we observed during gaze shifts for the example neuron. Gaze shifts (range, 2065 degrees) were sorted into separate data sets, each spanning 5 degrees and containing 15 or more examples. The percent attenuation index of the head-velocity sensitivity during the gaze shift (Fig. 4a, filled symbols) was calculated for each amplitude range. During the gaze shift, the amount of attenuation of the head-velocity signal carried by PVP neurons tended to increase with gaze-shift amplitude. A particularly notable example of this relationship is shown for a second neuron, unit 39_1 (Fig. 4b). For our entire sample of PVP neu406

rons, the percent attenuation of head-velocity sensitivity was significantly greater for large gaze shifts (5060 degrees) than for small gaze shifts (2030 degrees) ( Fig. 4c , filled columns; 58% 10 versus 35% 15; p <0.05). We also used model 2 to obtain an estimate of head-velocity sensitivity over the interval of 1080 ms that immediately followed each gaze shift. A post-movement drift in gaze often follows gaze shifts (Fig. 3b, left side), suggesting that the VOR is not fully operational during this interval31. We found attenuation of gfs immediately after gaze shifts for our example PVP neurons (Fig. 4a and b , open symbols). However the headvelocity signal was much less attenuated after the gaze shift than during the gaze shift, and the level of attenuation did not vary systematically with gaze-shift amplitude (Fig. 4a and b; compare open symbols with filled symbols). For our entire sample of PVP neurons, the percent attenuation was significantly less (p > 0.001) during the post-gaze-shift interval (7% 8 and 16% 4 for small and large gaze shifts, respectively) than during the gaze shift itself. This result indicates that the head-velocity sensitivity of PVP neurons increased immediately after gaze shifts (Fig. 4c, open columns) and achieved a value only slightly less than that observed during VOR in the dark. The residual post-gaze-shift attenuation of the PVP neurons responses was accompanied by a similar attenuation in the behavioral VOR gain (gain attenuation = 1(eye velocity/head velocity); mean, 9% 5 and 7% 6 for small and large gaze shifts, respectively; Fig. 4c, shaded columns). In contrast, the behavioral VOR gain during VORd was 0.98 0.06, corresponding to an attenuation of only ~2% from perfect gaze stabilization. After we completed our characterization of the neuron durnature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

1998 Nature America Inc. http://neurosci.nature.com

Fig. 3. Example gaze shifts of 2030 degrees (a) and 5060 degrees (b) amplitude for the example PVP neuron, unit 46_1. The firing rate (shaded trace) is plotted above the output of the spike threshold discriminator for each gaze shift. Superimposed on the firing rate trace are two model fits: models 1 and 2 (indicated by arrows). Model 1 overestimates the firing rate during gaze shifts of both amplitudes. G, horizontal gaze velocity; E, horizontal eye position; H, horizontal head position; FR, firing rate; , horizontal gaze velocity; G , horizontal eye velocity; E , horizontal head velocity; H deg, degree; sp, spikes. Dotted vertical lines indicate the onset and offset of gaze shifts using a criterion of 25 degrees/s.

2030 deg gaze shifts

5060 deg gaze shifts

50 deg

400 deg/s

model 1

200 sp/s
model 2

100 ms

ing active gaze shifts, the monkey was returned to the headrestrained condition. Forty-four percent of the neurons in our sample remained well isolated after the monkeys head had been rerestrained. We confirmed that these neurons had not been lost or damaged by repeating the passive whole-body rotation experiments (VORd and VORc); neural discharges did not dif-

fer considerably in the second head-restrained characterization, compared to the initial head-restrained characterization.

Discussion Here we compared the efficacy of the direct VOR pathways during whole-body rotation with that during voluntary orienting

a
Percent attenuation of head velocity sensitivity

b
Percent attenuation of head velocity sensitivity

nature neuroscience volume 1 no 5 september 1998

Percent attenuation of head velocity sensitivity or attenuation of VOR gain

Fig. 4. Amplitude-dependent attenuation of head-velocity sensitivity during gaze shifts. (a) Relationship between the percent attenuation of the head-velocity sensitivity and gaze-shift amplitude for the example PVP neuron (unit 46_1) during Gaze-shift amplitude (deg) Gaze-shift amplitude (deg) gaze shifts ( ; r = 0.57) and immediately (1080 ms) after gaze shifts 1080 ms 1-behavioral VOR ( ; r = 0.32). (b) Relationship Gaze shift after gain (1080 ms after) between the percent attenuation of the head-velocity sensitivity and gaze-shift amplitude for a second PVP neuron (unit 39_1) during gaze shifts ( ; r = 0.87) and immediately after gaze shifts ( ; r = -0.35). (c) During gaze shifts, the percent attenuation of head-velocity sensitivity for our sample of PVP neurons was significantly different for gaze shifts of 2030 degrees and 5060 degrees (filled columns; 35% 15 versus 58% 10; p < 0.05, asterisk). In contrast, there was little attenuation in the postgaze-shift interval (open columns; 7% 8 and 16% 4, respectively). The behavioral VOR gain (eye velocity/head velocity) was also slightly less than one during this interval (shaded columns; 9% 5 and 7% 6 for small and large gaze shifts, respectively). Gaze shift amplitude (deg)
407

1998 Nature America Inc. http://neurosci.nature.com

articles

Percent attenuation of head velocity sensitivity or attenuation of VOR gain

Gaze shift amplitude (deg)

c
1998 Nature America Inc. http://neurosci.nature.com

400 deg/s

model 3

200 sp/s

Fig. 5. Mechanisms for VOR suppression during gaze shifts. (a) Mean percent attenuation of PVP () head-velocity sensitivity during gaze shifts compared with percent attenuation of the VOR gain as measured in previous behavioral experiments during gaze shifts in humans (b) and monkeys (). (b) Brainstem mechanism that could produce an apparent reduction in VOR gain during gaze shifts. PVP neurons receive a strong monosynaptic connection from the ipsilateral vestibular ) and, in turn, project directly to primary extraocular afferents (H aff motoneurons (heavy lines). During steady fixation and slow eye movements (VORd), PVP neurons also carry an eye position signal (E), which is provided by the intrinsic properties of these vestibular neurons and/or their interconnections with the nucleus prepositus hypoglossi and cerebellum. During saccades, vestibular quick phases and gaze shifts, brainstem burst neurons (BNs) become active, denoted by the gate. BNs provide an +H ) to type I PVP neurons (studied here) during gaze inhibitory input -(E BN shifts through type II VN neurons (vestibular nuclei neurons that are activated during the slow-phase vestibular nystagmus generated by contralateral head rotation). (c) The eye- and head-velocity signals carried by BNs could effectively reduce the signal carried by PVP neurons during gaze shifts. Accordingly, discharge of unit 46_1 during gaze shifts was estimated using model 3, which provided an excellent fit of the unit discharge (model 3, gfs = 0.45 (spikes/s)/(degree/s) and bfs = -0.47 (spikes/s)/(degree/s)).

100 ms

eyehead movements in the horizontal plane. PVP neurons constitute most of the intermediate leg of the simple three-neuron arc that mediates the direct VOR pathway. Our main conclusions are that head-velocity information is less effectively transmitted from the vestibular afferents to the extraocular motoneurons by these neurons during active orienting gaze shifts than during the compensatory VOR elicited by passive whole-body rotation, and that the head-velocity signal carried by the direct VOR pathway is more attenuated for large- than for small-amplitude gaze shifts. Indeed, we found that the attenuation of the head-velocity signal carried by PVP neurons during gaze shifts corresponds with the results of behavioral studies that have characterized the VOR gain during gaze shifts in monkeys and humans57,13,1518. We plotted the percent attenuation of the VOR versus gaze-shift amplitude based on previous studies in the rhesus monkey13 and in humans18 (Fig. 5a). For comparison, we superimposed a plot of percent attenuation of PVP head-velocity sensitivity during gaze shift versus gaze-shift amplitude from this study. The variability between the two behavioral data sets demonstrates the uncertainty of previous estimates of VOR gain. However, the variabilities within our data set and that of the human study18 are similar (15% and 10% versus 19%, 11% and 13%, respectively). Although our data do not exactly match either of the two behavioral data sets, the similarity in trends is strong evidence that the amplitude-dependent reduction of the head-velocity signal carried by direct VOR pathways is at least partly responsible for the amplitude-dependent decrease in VOR gain that has been reported in behavioral studies.
408

We also found that the average head-velocity sensitivity of PVP neurons recovered significantly within the first 1080 ms after the end of a gaze shift. The residual attenuation of the PVP head-velocity signal during this period was minimal (~12%). This discharge attenuation is in agreement with the observation that the actual gain of the behavioral VOR immediately after a gaze shift is, on average, slightly less than one during this interval (~8%). Our results do not agree with those of the recent behavioral study in humans18, which reported that the VOR returns to a value greater than unity at the end of gaze shifts. This discrepancy may have arisen from a difference in the head-velocity profiles used in the two studies. These authors18 used head perturbations of 1220 Hz to investigate the VOR gain, whereas we studied natural head movements, which have very little frequency content above 5 Hz1. Several behavioral studies have attempted to characterize the time course of the suppression of the VOR during gaze shifts. The VOR was first reported to be switched from an on to an off state during gaze shifts6. A gradual restoration of the VOR was also proposed5, furthered by findings that the gain of the VOR fell off exponentially as gaze shift amplitude increased7. Other authors suggest that the VOR gain increases from zero to unity during the last 40 ms of a gaze shift16. The most recent proposal is that VOR gain decreases exponentially (time constant, ~50 ms) from the onset of the gaze shift18. In our analysis, we estimated average headvelocity sensitivities for each neuron in two intervals: during the gaze shift and immediately after the gaze shift. By estimating average head-velocity sensitivities, we have not attempted to account
nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

1998 Nature America Inc. http://neurosci.nature.com

for possible dynamic modulation of the gain of VOR pathways within each of these intervals; the precise time course of attenuation of VOR pathways during gaze shifts remains to be determined. The attenuation that we observe during active gaze shifts can be accounted for by known brainstem mechanisms (Fig. 5b). Burst neurons in the brainstem paramedian pontine reticular formation generate a burst of spikes to drive horizontal saccadic eye movements, but do not discharge during either steady fixation or slow eye movements, including VORd slow phases3340. Burst neurons project directly to neurons in the vestibular nuclei (VN), which are activated during the slowphase vestibular nystagmus generated by contralateral head rotation (type II VN neurons)36. These type II VN neurons then send inhibitory projections to VOR interneurons, which are activated during slow-phase vestibular nystagmus generated by ipsilateral head rotation (type I PVP neurons)41. The activity of brainstem burst neurons (inhibitory burst neurons and presumably the excitatory burst neurons) carry head- as well as eye-velocity signals during gaze shifts30. The eye- and headvelocity signals carried by excitatory burst neurons could have an inhibitory effect on the activity of type I PVP neurons (the subject of this study) through type II VN neurons. We propose that the signals related to eye and head movement carried by brainstem burst neurons (which are only active during saccades, vestibular nystagmus quick phases and gaze shifts) contribute to decreasing the efficacy of the direct VOR pathways during gaze shifts. Indeed, we found that a model with separate terms for eye- and head-velocity could produce an excellent fit to PVP neuron discharges during gaze shifts (model 3, Fig. 5c): fr = bias + (kx eye position) + (gfs head velocity) + (bfs eye velocity) (3)

nals. Gaze-, head- and eye-positions were differentiated to produce velocity signals. Locations of the medial and lateral vestibular nuclei were determined relative to the abducens nucleus, and extracellular recordings were made in these nuclei. The abducens nucleus was identified based on its stereotypical discharge patterns during eye movement and whole-body rotation experiments21. Neural spike trains were determined using a windowing circuit (BAK) that was set manually to generate a pulse coincident with the rising phase of each action potential. The neural discharge was represented using a spike density function in which a Gaussian function was convolved with the spike train43. A neurons eye-position sensitivity coefficient, kx, spikes/s/degree, was determined by plotting its mean firing rate against eye position during periods of steady fixation in the saccade experiment and calculating the slope of the relationship. The eye-position contribution to the neuronal discharge (the product of kx and eye position) was subtracted from the firing rate before head-velocity sensitivities during VORd and VORc were determined. Statistical significance was determined by using a paired Students t-test. A least-squared regression analysis was then used to determine each units phase shift relative to head velocity, resting discharge (bias, spikes/s) and head-velocity sensitivity (gx, in (spikes/s)/(degrees/s)). This analysis was done during VORd and VORc to obtain two estimates of a cells head-velocity sensitivity (gxd and gxc respectively). Only unit data from periods of slow-phase vestibular nystagmus (VORd) or steady fixation (VORc) that occurred between quick phases of vestibular nystagmus and/or saccades were included in the analysis.

Acknowledgements
We thank D. Guitton for discussions and comments on the manuscript, P. A. Sylvestre and G. A. Wellenius for critically reading the manuscript and W. Kucharski and A. Smith for technical assistance. This study was supported by the Medical Research Council of Canada (MRC).

RECEIVED 23 MARCH: ACCEPTED 17 JULY 1998


1. Grossman, G. E., Leigh, R. J., Abel, L. A., Lanska, D. J. & Thurston, S. E. Frequency and velocity of rotational head perturbations during locomotion. Exp. Brain Res. 70, 470476 (1988). 2. Grossman, G. E., Leigh, R. J., Bruce, E. N., Huebner, W. P. & Lanska, D. J. Performance of human vestibuloocular reflex during locomotion. J. Neurophysiol. 62, 264271 (1989). 3. Andr-Deshays, C., Berthoz, A. & Revel, M. Eye-head coupling in humans. I. Simultaneous recording of isolated motor units in dorsal neck muscles and horizontal eye movements. Exp. Brain Res. 69, 399406 (1988). 4. Barnes, G. R. Vestibulo-ocular function during co-ordinated head and eye movements to acquire visual targets. J. Physiol. 287, 127147 (1979). 5. Guitton, D. & Volle, M. Gaze control in humans: eye-head coordination during orienting movements to targets within and beyond the oculomotor range. J. Neurophysiol. 58, 427459 (1987). 6. Laurutis, V. P. & Robinson, D. A. The vestibulo-ocular reflex during human saccadic eye movements. J. Physiol. 373, 209233 (1986). 7. Plisson, D., Prablanc, C. & Urquizar, C. Vestibulo-ocular reflex inhibition and gaze saccade control characteristics during eye-head orientation in humans. J. Neurophysiol. 59, 9971013 (1988). 8. Zangemeister, W. H. & Stark, L. Gaze latency: variable interactions of head and eye latency. Exp. Neurol. 75, 389406 (1982). 9. Zangemeister, W. H. & Stark, L. Types of gaze movement: variable interactions of eye and head movements. Exp. Neurol. 77, 563577 (1982). 10. Bizzi, E., Kalil, R. E. & Tagliasco, V. Eye-head coordination in monkeys: evidence for centrally patterned organization. Science 173, 452454 (1971). 11. Dichgans, J., Bizzi, E., Morasso, P. & Tagliasco, V. Mechanisms underlying recovery of eye-head coordination following bilateral labyrinthectomy in monkeys. Exp. Brain Res. 18, 548562 (1973). 12. Morasso, P., Bizzi, E. & Dichgans, J. Adjustment of saccade characteristics during head movements. Exp. Brain Res. 16, 492500 (1973). 13. Tomlinson, R. D. Combined eye-head gaze shifts in the primate. III. Contributions to the accuracy of gaze saccades. J. Neurophysiol. 64, 18731891 (1990). 14. Tomlinson, R. D. & Bahra, P. S. Combined eye-head gaze shifts in the primate. I. Metrics. J. Neurophysiol. 56, 15421557 (1986). 15. Tomlinson, R. D. & Bahra, P. S. Combined eye-head gaze shifts in the primate. II. Interactions between saccades and the vestibulo-ocular reflex. J. Neurophysiol. 56, 15581570 (1986).

Other brainstem mechanisms could also contribute to modulating the vestibular sensitivity of vestibular neurons during active gaze shifts. For example, it has been proposed that a centrally generated copy of the head efference signal could be used to regulate the vestibular sensitivity of PVP neurons during active head movements29. Future efforts will need to consider the complex interactions between the brainstem circuits that generate voluntary eye movements (such as saccades) and vestibularly driven eye movements to realistically model the control of gaze during voluntary orienting behaviors.
Methods Three monkeys (two Macaca mulatta and one Macaca fasicularis) were prepared for chronic extracellular recording. The surgical preparation and extracellular recording techniques have been described30. All experimental protocols were approved by the McGill University Animal Care Committee and complied with the guidelines of the Canadian Council on Animal Care. During the experiments, the monkey was seated in a primate chair, which was fixed to the suprastructure of a vestibular turntable. Monkeys were trained to track a target (a HeNe laser spot projected onto a cylindrical screen) for a juice reward. The target was stepped along the horizontal plane over a range of 30 degrees in the saccade experiment and 40 degrees in the gaze shift experiment. Behavioral experiments, target and turntable motion and the storage of data were controlled by a UNIX-based real-time data acquisition system (REX)42. Gaze- and head-position signals, target position, vestibular turntable velocity and unit activity were recorded on a DAT tape for later playback and analysis. Off-line, the position signals were low-pass filtered at 250 Hz (8 pole Bessel filter) and sampled at 1000 Hz. The gaze- and head-position signals were digitally filtered at 125 Hz. Eye position was calculated from the difference between the recorded gaze- and head-position signature neuroscience volume 1 no 5 september 1998

409

1998 Nature America Inc. http://neurosci.nature.com

articles

1998 Nature America Inc. http://neurosci.nature.com

16. Lefvre, P., Bottemanne, I. & Roucoux, A. Experimental study and modeling of vestibulo-ocular reflex modulation during large shifts of gaze in humans. Exp. Brain Res. 91, 496508 (1992). 17. Plisson, D. & Prablanc, C. Vestibulo-ocular reflex (VOR) induced by passive head rotation and goal-directed saccadic eye movements do not simply add in man. Brain Res. 380, 397400 (1986). 18. Tabak, S., Smeets J. B. J. & Collewijn, H. Modulation of the human vestibuloocular reflex during saccades: probing by high-frequency oscillation and torque pulses of the head. J. Neurophysiol. 76, 32493263 (1996). 19. Lorente De No, R. Vestibular-ocular reflex arc. Arch. Neurol. Psychiatry 30, 245291 (1933). 20. Cullen, K. E. & McCrea, R. A. Firing behavior of brain stem neurons during voluntary cancellation of the horizontal vestibuloocular reflex I. secondary vestibular neurons. J Neurophysiol. 70, 828843 (1993). 21. Cullen, K. E., Chen-Huang, C. & McCrea, R. A. Firing behavior of brainstem neurons during voluntary cancellation of the horizontal vestibulo-ocular reflex. II. eye-movement related neurons. J. Neurophysiol. 70, 844856 (1993). 22. Fuchs, A. F. & Kimm, J. Unit activity in vestibular nucleus of the alert monkey during horizontal angular acceleration and eye movement. J. Neurophysiol. 38, 11401161 (1975). 23. Keller, E. L. & Daniels, P. Oculomotor related interaction of vestibular and visual stimulation in vestibular nucleus cells in the alert monkey. Exp. Neurol. 46, 187198 (1975). 24. King, W. M., Lisberger, S. G. & Fuchs, A. F. Responses of fibers in medial longitudinal fasciculus (MLF) of alert monkeys during horizontal and vertical conjugate eye movements evoked by vestibular or visual stimuli. J. Neurophysiol. 39, 11351149 (1976). 25. Lisberger, S. G. & Miles, F. A. Role of the primate vestibular nucleus in longterm adaptive plasticity of the vestibulo-ocular reflex. J. Neurophysiol. 43, 17251745 (1980). 26. McCrea, R. A., Strassman, E. M. & Highstein, S. M. Anatomical and physiological characteristics of vestibular neurons mediating the horizontal vestibulo-ocular reflex of the squirrel monkey. J. Comp. Neurol. 264, 547570 (1987). 27. McFarland, J. L. & Fuchs, A. F. Discharge patterns of nucleus prepositus hypoglossi and adjacent vestibular nucleus during horizontal eye movement in behaving macaques. J. Neurophysiol. 41, 319332 (1992). 28. Scudder, C. A. & Fuchs, A. F. Physiological and behavioural identification of vestibular nucleus neurons mediating the horizontal vestibuloocular reflex in trained rhesus monkeys. J. Neurophysiol. 68, 244264 (1992). 29. McCrea, R. A., Chen-Huang, C., Belton, T. & Gdowski, G. T. Behavior contingent processing of vestibular sensory signals in the vestibular nuclei. Ann. NY Acad. Sci. 781, 292303 (1996).

30. Cullen, K. E. & Guitton, D. Analysis of primate IBN spike trains using system identification techniques. II. relationship to gaze, eye, and head movement dynamics during head-free gaze shifts. J. Neurophysiol. 78, 32833306 (1997). 31. Freedman, E. G. & Sparks, D. L. Eye-head coordination during headunrestrained gaze shifts in rhesus monkeys. J. Neurophysiol. 77, 23282348 (1997). 32. Phillips, J. O., Ling, L., Siebold, C. & Fuchs, A. F. Behavior of primate vestibulo-ocular reflex neurons and vestibular neurons during head-free gaze shifts. Ann. NY Acad. Sci. 781, 276291 (1996). 33. Hikosaka, O. & Kawakami, T. Inhibitory neurons related to the quick phase of vestibular nystagmus - their location and projections. Exp. Brain Res. 27, 377396 (1977). 34. Hikosaka, O., Igusa, Y., Nakao, S. & Shimazu, H. Direct inhibitory synaptic linkage of pontomedullary reticular burst neurons with abducens motoneurons in the cat. Exp. Brain Res. 33, 337352 (1978). 35. Igusa, Y., Sasaki, S. & Shimazu, H. Excitatory premotor burst neurons in the cat pontine reticular formation related to the quick phase of vestibular nystagmus. Brain Res. 182, 451456 (1980). 36. Sasaki, S. & Shimazu, H. Reticulovestibular organization participating in generation of horizontal fast eye movement. Ann. NY Acad. Sci. 374, 130145 (1981). 37. Scudder, C. A., Fuchs, A. F. & Langer, T. P. Characteristics and functional identification of saccadic inhibitory burst neurons in the alert monkey. J. Neurophysiol. 59, 14301454 (1988). 38. Strassman, A., Highstein, S. M. & McCrea, R. A. Anatomy and physiology of saccadic burst neurons in the alert squirrel monkey. I. Excitatory burst neurons. J. Comp. Neurol. 249, 337357 (1986). 39. Strassman, A., Highstein, S. M. & McCrea, R. A. Anatomy and physiology of saccadic burst neurons in the alert squirrel monkey. II. Inhibitory burst neurons. J. Comp. Neurol. 249, 358380 (1986). 40. Yoshida, K., Berthoz, A., Vidal, P. P. & McCrea, R. A. Morphological and physiological characteristics of inhibitory burst neurons controlling rapid eye movements on the alert cat. J. Neurophysiol. 48, 761784 (1982). 41. Nakao, S., Sasaki, S., Schor, R. H. & Shimazu, H. Functional organization of premotor neurons in the cat medial vestibular nucleus related to slow and fast phases of nystagmus. Exp. Brain Res. 45, 371385 (1982). 42. Hays, A. V., Richmond, B. J. & Optican, L. M. A UNIX-based multiple process system for real-time data acquisition and control. W ESCOM Conf. Proc. 2, 110 (1982). 45. Cullen, K. E., Rey, C. G., Guitton, D. & Galiana, H. L. The use of system identification techniques in the analysis of oculomotor burst neuron spike train dynamics. J. Comput. Neurosci. 3, 347368 (1996).

410

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

Expectation of reward modulates cognitive signals in the basal ganglia


Reiko Kawagoe, Yoriko Takikawa and Okihide Hikosaka
Department of Physiology, School of Medicine, Juntendo University, 2-1-1 Hongo, Bunkyo-ku, Tokyo 113-8421, Japan Correspondence should be addressed to O.H. (hikosaka@med.juntendo.ac.jp)

1998 Nature America Inc. http://neurosci.nature.com

Action is controlled by both motivation and cognition. The basal ganglia may be the site where these kinds of information meet. Using a memory-guided saccade task with an asymmetric reward schedule, we show that visual and memory responses of caudate neurons are modulated by expectation of reward so profoundly that a neurons preferred direction often changed with the change in the rewarded direction. The subsequent saccade to the target was earlier and faster for the rewarded direction. Our results indicate that the caudate contributes to the determination of oculomotor outputs by connecting motivational values (for example, expectation of reward) to visual information.

Visual responses of neurons can be modulated by changes in behavioral contexts. Many widely used behavioral tasks require the subject to respond by choosing one stimulus among many or by choosing one feature (for example, color) of a stimulus among many features13. In this type of procedure, the same reward is given for all correct trials, so that the motivational state of the subject is assumed to be the same no matter which stimulus (or stimulus feature) represents the correct response. This model is therefore ideal for investigating the cognitive aspect of action or attention, but not the motivational aspect. Action is controlled by both cognition and motivation4,5, and motivational states vary considerably. The same action can lead to different reward outcomes in different behavioral contexts. Both neural and behavioral responses (for example, speed of action) may co-vary with such motivational changes, which may have different consequences in the subsequent decision-making processes. However, there have been few physiological studies that manipulated the outcome of an action in terms of reward (its amount or kind) while keeping the subjects actions constant6. Consequently, little is known about neural mechanisms of the motivational aspect of attention or action selection. To investigate how expectation of reward affects cognitive information processing, we devised a memory-guided saccade task in which the subject had to make a saccade to a remembered cue location. However, correct performance was only rewarded when the cue had appeared at one of the four possible locations.

The cognitive requirement was always the same, in that the subject had to attend to the cue stimulus, remember its location and make a saccade to the location, but the motivational significance varied. Using this model, we studied single-neuron activity in the monkey caudate nucleus, a major input zone of the basal ganglia, as the basal ganglia may be involved in control of action based on motivation5,79. We found that visual or memory-related responses of presumed projection neurons were frequently modulated by expectation of reward, either as an enhancement or as a reduction of the response.

Results We trained two monkeys on a memory-guided saccade task in two reward conditions: all directions rewarded (ADR) and one direction rewarded (1DR). In ADR, which is the conventional reward schedule, the monkeys were rewarded each time they made a memory-guided saccade to the cued location for that trial. In 1DR, which we devised specifically for this study, the monkeys were rewarded for making correct memory-guided saccades only in one direction, termed the rewarded direction (Fig. 1). Monkeys were not rewarded (exclusive 1DR) or were rewarded with a smaller amount (relative 1DR) when they made a correct response in one of the other three directions, but they had to make a correct saccade to proceed to the next trial. The rewarded direction was fixed in a block of 60 successful trials, and a total of four blocks were done, with four different rewarded directions. Thus, the cue

Eye position Fixation point Target point Tone Reward No reward

No reward

Reward

No reward 1s

Fig. 1. Memory-guided saccade task in the one direction rewarded condition (1DR). Throughout a block of experiment (60 trials), only one direction was rewarded. (Here the right direction was rewarded.) Different directions were rewarded in different blocks.
411

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

1DR

Rewarded direction

ADR

1998 Nature America Inc. http://neurosci.nature.com

Fig. 2. Reward-dependent visual response (reward-facilitated type) of a neuron recorded in the right caudate nucleus. The data obtained in one block of ADR (right) and four blocks of 1DR (left) are shown in columns. In the histogram/raster display, the neuron discharge aligned on cue onset is shown separately for different cue directions (R, right; U, up; L, left; D, down). For each cue direction, the sequence of trials was from bottom to top. The rewarded direction is indicated by a bulls eye mark. Polar diagrams (top row) show the magnitudes of response for four cue directions. Target eccentricity was ten degrees. The neurons response was strongest for the rewarded direction in any block of 1DR, whereas its preferred direction was to the left in ADR.

stimulus signified two things: the direction of the saccade to be most vigorous for the rewarded direction. The response dependmade later, and whether or not a reward (or a larger reward) was ed strongly on the reward condition (two-way ANOVA (reward to be obtained after the saccade. condition cued direction), main effect of reward condition; Among 241 neurons that we recorded in the caudate nucleus, p < 0.0001). Another type of caudate neuron also depended on there were neurons showing phasic visual responses to the cue reward expectation, but in the opposite manner (Fig. 3). In ADR, stimulus (visual response; n = 114), sustained activity during the this neuron showed almost no response to any of the four cue delay period (memory-related response; n = 79), saccadic responsstimuli. In 1DR, however, it showed vigorous responses to the es (n = 92) and activity preceding the cue stimulus (n = 89). Here cue that indicated no reward, whereas it showed no response to we studied 87 neurons with visual or memory-related responses, the rewarded cues, no matter which direction was rewarded. A in which four blocks of 1DR and one block of ADR were fully examined. 1DR ADR Among the fully examined neurons, 27 Rewarded direction of 45 neurons (60%) with visual ALL responses and 20 of 50 neurons (40%) with memory-related responses showed clear direction selectivity when tested in ADR (one-way ANOVA (cued direction), p < 0.01; eight neurons showed both visual and memory responses). The preferred direction was usually contralateral (70%), as reported10. We found, however, that such spatial selectivity depended on the reward condition. A typical neuron in the right caudate nucleus (Fig. 2) responded to the left (contralateral) cue stimulus most vigorously in ADR, whereas the response to the right cue was meager. The neurons direction selectivity is shown as a polar diagram (Fig. 2, top row). In 1DR, however, the neurons direction selectivity changed. For example, when the rewarded direction was right, this neuron responded to the right cue stimulus much better than to the other directions. Similarly, the neu- Fig. 3. Reward-dependent visual response (reward-suppressed type) of a neuron recorded in ron changed its preferred direction in the left caudate nucleus. Target eccentricity was 20 degrees. The neuron showed vigorous other blocks so that its response was responses exclusively to the non-rewarded cues. See legend to Fig. 2 for explanation of layout.
412
Cued direction

Cued direction

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

The cells shown in Figs 24 were not exceptional. Most caudate neurons showed either a strong enhancement (data points close to the ordinate) or a reduction (data points close to the abscissa) of response by expectation of reward (Fig. 5a). A statistically significant modulation was found in 76 of 87 neurons (87%) in visual or memoryrelated responses: visual response, 36/45 (80%); memory response, 43/50 (86%; two-way ANOVA (reward condition cued direction), main effect of reward condition; p < 0.01). Among the 76 modulated neurons, 64 neurons (visual, 31; memory, 36) showed an enhancement (reward-facilitated neurons), whereas 12 neurons (visual, 5; memory, 7) showed a reduction of response (reward-suppressed neurons). The results were the same for exclusive 1DR and relative 1DR. The caudate contributes to the initiFig. 4. Reward-dependent memory response (reward-facilitated type) of a neuron recorded in ation of saccades with its connection to the left caudate nucleus. See legend to Fig. 2 for format. Here the neuron discharge was aligned the superior colliculus through the subon saccade onset. Cued/rewarded direction: RU, right-up; LU, left-up; LD, left-down; RD, right- stantia nigra11. The modulation of caudown. Target eccentricity was 20 degrees. The neuron showed sustained memory-related date neuron activity by reward activity and phasic saccadic activity for the right-up direction, both of which were stronger expectation might therefore produce when this direction was rewarded. changes in the characteristics of the subsequent saccade to the remembered cue location. Our data confirmed this prediction; the latencies were shorter (Fig. 5b) and the peak velocities were higher (Fig. 5c) when the third type of response is illustrated by a neuron in the left causaccades were followed by reward than when they were not (paired date nucleus (Fig. 4), which showed sustained activity in ADR t-test, p < 0.0001). The saccade latencies were significantly difafter the cue was presented in the right-up (RU) direction; the ferent in the two monkeys, but the difference between the rewardactivity reached its peak at the time of the saccade. In 1DR, the ed and non-rewarded conditions was evident for each monkey. neurons activity for the RU direction was reduced considerably In addition, saccades to the rewarded direction were more accuwhen this direction was not rewarded (columns 24), whereas rate than those to the non-rewarded directions; the monkeys occasome activity appeared in the right-down (RD) and left-up (LU) sionally made incorrect saccades on non-rewarded trials. directions when they were rewarded.
1DR Rewarded direction ADR Cued direction

1998 Nature America Inc. http://neurosci.nature.com

a
Cell activity, rewarded (Hz)

b
Saccade latency, rewarded (ms)

c
Saccade velocity, rewarded (deg/s) Saccade latency, non-rewarded (ms) Saccade velocity, non-rewarded (deg/s)

Cell activity, non-rewarded (Hz)

Fig. 5. Effects of reward expectation on caudate neuron activity (a), saccade latency (b), and saccade velocity (c). Values in the rewarded (ordinate) and non-rewarded (abscissa) conditions are compared. After determining the preferred direction for each neuron, we calculated the mean magnitude (test-control, Hz) of the neurons response to its preferred cue in two conditions: when the preferred direction was rewarded (one block) and when the preferred direction was not rewarded (three blocks). Data from two monkeys are shown with different symbols. Both visual and memory-related responses are included. Arrows 24 indicate the data for the neurons shown in Figs 24, respectively. The saccade parameters were obtained for each neuron by averaging across saccades to the neurons preferred direction separately for the rewarded and non-rewarded conditions (b and c).
nature neuroscience volume 1 no 5 september 1998 413

1998 Nature America Inc. http://neurosci.nature.com

articles

ron in Figs 2 and 6a. For each block, the rewardsuppressed neuron initially showed almost no response to any direction, but then started responding to the three directions that indicated no reward ( Fig. 6b ). A similar time course of Trial number Trial number response modulation was observed in some of the b other reward-contingent caudate neurons. For the non-rewarded cues, 27 of 64 reward-facilitated neurons significantly decreased their responses, whereas 4 of 12 reward-suppressed neurons significantly Trial number increased their responsTrial number es, when the initial 15 triFig. 6. Change in direction selectivity within one block of 1DR trials. Data for two sequential blocks are shown for two neurons: (a) neuron from Fig. 2; (b) neuron from Fig. 3. Discharge rates for four cue directions are als and the subsequent trials were compared plotted individually against the trial number. The rewarded cue is indicated by a filled symbol. ( t -test, p < 0.01). Reward-contingent neurons were distributed in the caudate nucleus from its head to the body (Fig. 7). There was We next determined how quickly the caudate neurons no distinct tendency for differential distribution of different types changed their response when the rewarded direction was changed of neurons: reward-facilitated versus reward-suppressed (Fig. 7) (Fig. 6). In the first block of 1DR for the reward-facilitated neuor visual versus memory (not shown). ron shown in Fig. 2, the rewarded direction was to the left, which was the neurons preferred direction in ADR (Fig. 6a, left). The responses were initially strong for all directions except the right, Discussion but the responses to the left cue gradually increased, whereas the These data show that visual and memory responses of caudate responses to the other cues decreased rapidly and remained near neurons were strongly modulated by the reward schedule in the zero. In the next block (Fig. 6a, right), the rewarded direction memory-guided saccade task. The monkeys were required to was changed to the right, which was the non-preferred direction complete the same tasks of spatial attention and motor proin ADR. Again, the responses were initially strong for all direcgramming on each trial, yet caudate neurons responses dependtions but decreased gradually, with only the response to the right ed selectively on whether a successful trial would be rewarded cue persisting. The time course for the reward-suppressed neuron immediately. This reward-dependent modulation of neural shown in Fig. 3 was opposite to that of the reward-facilitated neubehavior can be viewed as reflecting a form of motivation.
Cell activity (Hz)

1998 Nature America Inc. http://neurosci.nature.com

Fig. 7. Recording sites of reward-contingent caudate neurons plotted on coronal sections in one monkey. Open circles, reward-facilitated neurons; filled triangles, reward-suppressed neurons. AC indicates the level of the anterior commissure; the sections anterior and posterior to the AC are indicated by plus and minus numbers (distances in mm), respectively. Inset, a mid-sagittal view, indicating the levels of the coronal sections (anterior to left); the position of the AC is indicated by a dot. To reconstruct the recording sites based on MR images, recordings were made for selected penetrations through implanted guide tubes, which were then visualized on MR images (sections AC 2 and AC 0, right side). One neuron in the section AC 2 (left side) was judged to be inside the neuron cluster bridging the caudate and the putamen.
414 nature neuroscience volume 1 no 5 september 1998

Cell activity (Hz)

Cell activity (Hz)

Cell activity (Hz)

1998 Nature America Inc. http://neurosci.nature.com

articles

The modulation of caudate neural activity could instead be considered a kind of attentional modulation. However, this is conceptually different from the type of attention investigated in previous studies. Thus, the previous studies on attention13 were based on the attend-versus-ignore comparison, whereas our study was based on the rewarded-versus-nonrewarded comparison. In the former comparison, cognitive processing was allocated to the to-be-attended location or object, and reward was given consistently. Here the required cognitive processing was identical for different target locations, but the reward outcome was different. The basal ganglia may direct attention to items associated with reward, whereas the cerebral cortex, especially the parietal cortex12, may direct attention based on task requirements. The neurons we recorded had low spontaneous activity and were presumably projection neurons, which are GABAergic13. They are thought to modulate the final inhibitory outputs of the basal ganglia, either by disinhibition or by enhancement of inhibition1416. Anatomically, the striatal projection neurons are characterized by many spines on their dendrites 17,18 , to which glutamatergic cortico-striatal axons and dopaminergic axons make synaptic contacts19,20. Dopaminergic neurons in the substantia nigra show responses to sensory stimuli that predict the upcoming reward21,22. Thus, a caudate neuron could receive spatial information through the corticostriatal inputs23 and rewardrelated information through the dopaminergic input21. Given these considerations, our findings are consistent with the view that the efficacy of the corticostriatal synapses is modulated by the dopaminergic input22,24,25. The co-activation of these two inputs should produce synaptic enhancement and depression, respectively, in reward-facilitated neurons and reward-suppressed neurons. Such opposing processes might be mediated by different dopaminergic receptors, such as D1 and D226,27. Alternatively, the reward-contingent modulation may occur in the cerebral cortex, especially in the prefrontal cortex6. Memory-related sustained activity in prefrontal neurons is modulated by dopaminergic inputs28,29. It is thus possible that the rewardcontingent activity of caudate neurons may simply reflect the plasticity of the cerebral cortex. Conversely, the caudate neurons may influence the activity in the cerebral cortex through the output nuclei of the basal ganglia and the thalamus30. The reward-contingent modulation of caudate neuron activity was correlated with the changes in saccade latency and velocity. A mechanism underlying the changes may be the serial inhibitory connections from the caudate to the superior colliculus through the substantia nigra pars reticulata15,31. An enhancement of caudate neuron activity when reward is expected (Fig. 2) would produce an enhanced disinhibition of the superior colliculus and consequently a reduction of saccade latency and an increase in saccade velocity, especially for memory-guided saccades32, which we observed here. In contrast, an enhancement of caudate neuron activity when reward was not expected (Fig. 3) might affect the indirect pathway (including the globus pallidus external segment 33 and subthalamic nucleus34), which would lead to the suppression of saccades to the non-rewarded cues, as seen here. Consistent with this, dopaminergic denervation in the caudate of monkeys leads to deficits in spontaneous saccades35 and memoryguided saccades36. Dopamine-deficient monkeys also showed spatial hemineglect 37 . Similar oculomotor and attentional deficits have been reported in patients with Parkinsons disease38,39. Abulia, lack of will, is a symptom that often occurs after a lesion in the caudate40,41. The basal ganglia may contribute to the selection of action42,43.
nature neuroscience volume 1 no 5 september 1998

Our study indicates that the caudate nucleus contributes to the control of oculomotor action by associating motivational values, such as the expectation of reward, to a visual target.
Methods GENERAL. We used two male Japanese monkeys (Macaca fuscata). After each monkey was sedated by general anesthesia, we implanted a head holder, chambers for unit recording and a scleral search coil31. All surgical and experimental protocols were approved by the Juntendo University Animal Care and Use Committee and are in accordance with the National Institutes of Health Guide for the Care and Use of Animals. The monkeys were trained to perform saccade tasks, especially a memoryguided saccade task44. Eye movements were recorded using the search coil method. We recorded extracellular spike activity of presumed projection neurons, which showed very low spontaneous activity (< 3 Hz)11, but not of presumed interneurons, which showed irregular tonic discharge45 .

1998 Nature America Inc. http://neurosci.nature.com

Here we studied cells that showed visual and/or memory responses. We defined a visual response as phasic activity that started within 200 ms after onset of the cue stimulus and reached its peak within another 200 ms, and a memory-related response as sustained activity that started at least 200 ms after the cue onset and ended before or with the saccade (A neuron could have both types of responses.) For each neuron, we used a set of four target locations of equal eccentricity (either 10 degrees or 20 degrees), arranged at either normal or oblique angles, depending on the neurons receptive field. The recording sites were verified by MRI (Hitachi, AIRIS, 0.3T). TASK PROCEDURES. The monkeys did the memory-guided saccade task in two different reward conditions: all-directions-rewarded condition (ADR) and one-direction-rewarded condition (1DR). For every caudate neuron recorded, we required the monkeys to do one block of ADR and four blocks of 1DR (that is, four different rewarded directions). The use of the memory-guided saccade task allowed us to dissociate visually evoked activity from motor-related activity. In both conditions, a task trial started with the onset of a central fixation point that the monkeys had to fixate (Fig. 1). A cue stimulus (spot of light) came on 1 s after onset of the fixation point (duration, 100 ms), and the monkeys had to remember its location. After 11.5 s, the fixation point turned off, and the monkeys were required to make a saccade to the previously cued location. The target came on 400 ms later for 150 ms at the cued location. The saccade was judged to be correct if the eye position was within a window around the target (usually within 3 degrees) when the target turned off. The correct saccade was indicated by a tone stimulus. The next trial started after an inter-trial interval of 3.54 s. In ADR, every correct saccade was followed by the tone stimulus and a liquid reward. In 1DR, an asymmetric reward schedule was used (Fig. 1) in which correct responses in only one of the four directions was rewarded, but correct responses in the other directions were either not rewarded (exclusive 1DR) or rewarded with a smaller amount (about 1/5) (relative 1DR). The highly rewarded direction was fixed in each block of experiments, which consisted of 60 successful trials. Even for the nonrewarded or less-rewarded direction, the monkeys had to make a correct saccade. If the saccade was incorrect, the trial was repeated. The average amount of reward per trial was approximately the same for 1DR and ADR. The target cue was chosen pseudo-randomly such that the four directions were randomized in every sub-block of four trials; thus, one block (60 trials) consisted of 15 trials for each direction. 1DR testing was done in four blocks, each with a different rewarded direction. Other than the actual reward, no indication was given to the monkeys as to which direction was currently rewarded. DATA ANALYSIS. For each neuron responding to the cue stimulus, we first determined the duration of the response (test duration) based on cumulative time histograms, usually based on the most robust response. A control duration (usually 500 ms) was set just before the onset of the fixation point. The neurons response was calculated for each trial as the spike frequency during the test duration minus the spike frequency during the control duration.
415

1998 Nature America Inc. http://neurosci.nature.com

articles

Acknowledgements
We thank Masamichi Sakagami, Johan Lauwereyns, Katsuyuki Sakai, Hiroyuki Nakahara, Thomas Trappenberg and Brian Coe for comments, Makoto Kato for designing the computer programs and Masashi Koizumi for technical support. This work was supported by CREST (Core Research for Evolutional Science and Technology) of Japan Science and Technology Corporation (JST) and JSPS (Japan Society for the Promotion of Science) Research for the Future program.

RECEIVED 28 MAY: ACCEPTED 24 JULY 1998


1. Wurtz, R. H., Goldberg, M. E. & Robinson, D. L. Brain mechanisms of visual attention. Sci. Am. 246, 124135 (1982). 2. Hillyard, S. A. Electrophysiology of human selective attention. Trends Neurosci. 8, 400405 (1985). 3. Desimone, R. & Duncan, J. Neural mechanisms of selective visual attention. Annu. Rev. Neurosci. 18, 193222 (1995). 4. Konorski, J. Integrative Activity of the Brain (Univ. Chicago Press, Chicago, 1967). 5. Mogenson, G. J., Jones, D. L. & Yim, C. Y. From motivation to action: functional interface between the limbic system and the motor system. Progress Neurobiol. 14, 6997 (1980). 6. Watanabe, M. Reward expectancy in primate prefrontal neurons. Nature 382, 629632 (1996). 7. Robbins, T. W. & Everitt, B. J. Neurobehavioural mechanisms of reward and motivation. Curr. Opin. Neurobiol. 6, 228236 (1996). 8. Schultz, W., Apicella, P., Scarnati, E. & Ljungberg, T. Neuronal activity in monkey ventral striatum related to the expectation of reward. J. Neurosci. 12, 45954610 (1992). 9. Bowman, E. M., Aigner, T. G. & Richmond, B. J. Neural signals in the monkey ventral striatum related to motivation for juice and cocaine rewards. J. Neurophysiol. 75, 10611073 (1996). 10. Hikosaka, O., Sakamoto, M. & Usui, S. Functional properties of monkey caudate neurons. II. Visual and auditory responses. J. Neurophysiol. 61, 799813 (1989). 11. Hikosaka, O., Sakamoto, M. & Usui, S. Functional properties of monkey caudate neurons. I. Activities related to saccadic eye movements. J. Neurophysiol. 61, 780798 (1989). 12. Bushnell, M. C., Goldberg, M. E. & Robinson, D. L. Behavioral responses in monkey cerebral cortex. I. Modulation in posterior parietal cortex related to selective visual attention. J. Neurophysiol. 46, 755772 (1981). 13. Ribak, C. E., Vaughn, J. E. & Roberts, E. The GABA neurons and their axon terminals in rat corpus striatum as demonstrated by GAD immunocytochemistry. J. Comp. Neurol. 187, 261284 (1979). 14. Chevalier, G. & Deniau, J. M. Disinhibition as a basic process in the expression of striatal functions. Trends Neurosci. 13, 277280 (1990). 15. Hikosaka, O. & Wurtz, R. H. in The Neurobiology of Saccadic Eye Movements (eds. Wurtz, R. H. & Goldberg, M. E.) 257281 (Elsevier, Amsterdam, 1989). 16. Alexander, G. E. & Crutcher, M. D. Functional architecture of basal ganglia circuits: neural substrates of parallel processing. Trends Neurosci. 13, 266271 (1990). 17. Preston, R. J., Bishop, G. A. & Kitai, S. T. Medium spiny neuron projection from the rat striatum: an intracellular horseradish peroxidase study. Brain Res. 183, 253263 (1980). 18. Kawaguchi, Y., Wilson, C. J. & Emson, P. C. Projection subtypes of rat neostriatal matrix cells revealed by intracellular injection of biocytin. J. Neurosci. 10, 34213438 (1990). 19. Smith, A. D. & Bolam, J. P. The neural network of the basal ganglia as revealed by the study of synaptic connections of identified neurones. Trends Neurosci. 13, 259265 (1990). 20. Groves, P. M., Linder, J. C. & Young, S. J. 5-Hydroxydopamine-labeled dopaminergic axons: Three dimensional reconstructions of axons, synapses and postsynaptic targets in rat neostriatum. Neuroscience 58, 593604 (1994).

21. Schultz, W., Apicella, P. & Ljungberg, T. Responses of monkey dopamine neurons to reward and conditioned stimuli during successive steps of learning a delayed response task. J. Neurosci. 13, 900913 (1993). 22. Schultz, W., Dayan, P. & Montague, P. R. A neural substrate of prediction and reward. Science 275, 15931599 (1997). 23. Parthasarathy, H. B., Schall, J. D. & Graybiel, A. M. Distributed but convergent ordering of corticostriatal projections: analysis of the frontal eye field and the supplementary eye field in the macaque monkey. J. Neurosci. 12, 44684488 (1992). 24. Houk, J. C., Adams, J. L. & Barto, A. G. in Models of Information Processing in the Basal Ganglia (eds. Houk, J. C., Davis, J. L. & Beiser, D. G.) 249270 (MIT Press, Cambridge, Massachusetts, 1995). 25. Wickens, J. & Ktter, R. in Models of Information Processing in the Basal Ganglia (eds. Houk, J. C., Davis, J. L. & Beiser, D. G.) 187214 (MIT Press, Cambridge, Massachusetts, 1995). 26. Gerfen, C. R. et al. D1 and D2 dopamine receptor-regulated gene expression of striatonigral and striatopallidal neurons. Science 250, 14291432 (1990). 27. Calabresi, P., De Murtas, M. & Bernardi, G. The neostriatum beyond the motor function: Experimental and clinical evidence. Neuroscience 78, 3960 (1997). 28. Sawaguchi, T., Matsumura, M. & Kubota, K. Effects of dopamine antagonists on neuronal activity related to a delayed response task in monkey prefrontal cortex. J. Neurophysiol. 63, 14011412 (1990). 29. Williams, G. V. & Goldman-Rakic, P. S. Modulation of memory fields by dopamine D1 receptors in prefrontal cortex. Nature 376, 572575 (1995). 30. Alexander, G. E., DeLong, M. R. & Strick, P. L. Parallel organization of functionally segregated circuits linking basal ganglia and cortex. Annu. Rev. Neurosci. 9, 357381 (1986). 31. Hikosaka, O., Sakamoto, M. & Miyashita, N. Effects of caudate nucleus stimulation on substantia nigra cell activity in monkey. Exp. Brain Res. 95, 457472 (1993). 32. Hikosaka, O. & Wurtz, R. H. Modification of saccadic eye movements by GABA-related substances. II. Effects of muscimol in the monkey substantia nigra pars reticulata. J. Neurophysiol. 53, 292308 (1985). 33. Kato, M. & Hikosaka, O. in Age-Related Dopamine-Deficient Disorders (eds Segawa, M. & Nomura, Y.) 178187 (Karger, Basal, 1995). 34. Matsumura, M., Kojima, J., Gardiner, T. W. & Hikosaka, O. Visual and oculomotor functions of monkey subthalamic nucleus. J. Neurophysiol. 67, 16151632 (1992). 35. Kato, M. et al. Eye movements in monkeys with local dopamine depletion in the caudate nucleus. I. Deficits in spontaneous saccades. J. Neurosci. 15, 912927 (1995). 36. Kori, A. et al. Eye movements in monkeys with local dopamine depletion in the caudate nucleus. II. Deficits in voluntary saccades. J. Neurosci. 15, 928941 (1995). 37. Miyashita, N., Hikosaka, O. & Kato, M. Visual hemineglect induced by unilateral striatal dopamine deficiency in monkeys. NeuroReport 6, 12571260 (1995). 38. Crawford, T. J., Henderson, L. & Kennard, C. Abnormalities of nonvisuallyguided eye movements in Parkinsons disease. Brain 112, 15731586 (1989). 39. Hikosaka, O., Imai, H. & Segawa, M. in Vestibular and Brain Stem Control of Eye, Head and Body Movements (eds. Shimazu, H. & Shinoda, Y.) 405414 (Japan Scientific Society Press, Tokyo, 1992). 40. Caplan, L. R. et al. Caudate infarcts. Arch. Neurol. 47, 133143 (1990). 41. Bhatia, K. P. & Marsden, C. D. The behavioural and motor consequences of focal lesions of the basal ganglia in man. Brain 117, 859876 (1994). 42. Hikosaka, O. in The Basal Ganglia IV: New Ideas and Data on Structure and Function (eds. Percheron, G., McKenzie, J. S. & Feger, J.) 589596 (Plenum Press, New York, 1994). 43. Graybiel, A. M. Building action repertoires: memory and learning functions of the basal ganglia. Curr. Opin. Neurobiol. 5, 733741 (1995). 44. Hikosaka, O. & Wurtz, R. H. Visual and oculomotor functions of monkey substantia nigra pars reticulata. III. Memory-contingent visual and saccade responses. J. Neurophysiol. 49, 12681284 (1983). 45. Aosaki, T. et al. Responses of tonically active neurons in the primates striatum undergo systematic changes during behavioral sensorimotor conditioning. J. Neurosci. 14, 39693984 (1994).

1998 Nature America Inc. http://neurosci.nature.com

416

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

Relearning sound localization with new ears


Paul M. Hofman, Jos G.A. Van Riswick and A. John Van Opstal
University of Nijmegen, Department of Medical Physics and Biophysics, Geert Grooteplein 21, NL-6525 EZ Nijmegen, The Netherlands Correspondence should be addressed to J.V.O (johnvo@mbfys.kun.nl)

1998 Nature America Inc. http://neurosci.nature.com

Because the inner ear is not organized spatially, sound localization relies on the neural processing of implicit acoustic cues. To determine a sounds position, the brain must learn and calibrate these cues, using accurate spatial feedback from other sensorimotor systems. Experimental evidence for such a system has been demonstrated in barn owls, but not in humans. Here, we demonstrate the existence of ongoing spatial calibration in the adult human auditory system. The spectral elevation cues of human subjects were disrupted by modifying their outer ears (pinnae) with molds. Although localization of sound elevation was dramatically degraded immediately after the modification, accurate performance was steadily reacquired. Interestingly, learning the new spectral cues did not interfere with the neural representation of the original cues, as subjects could localize sounds with both normal and modified pinnae.

Human auditory localization is remarkably accurate, even in complete darkness under open-loop conditions14 (Fig. 2; see also Methods). This performance may rival that of the barn owl, a well studied nocturnal hunter that relies heavily on its auditory system to capture prey. Yet, quite different localization mechanisms are used by each species. In barn owls, sound azimuth (horizontal direction) is derived from interaural timing differences, whereas sound elevation cues are provided by the interaural level differences that result from the updown asymmetry of its facial ruff5. In humans, however, both interaural cues only relate to the sound-source azimuth. In contrast, sound elevation and front-back direction are determined on the basis of spectral cues generated by the direction-dependent filtering of the pinnae69. Sounds entering the ear via the pinna aperture have some frequencies amplified and some attenuated, with an effect that can be described mathematically by linear transfer functions (pinna filters, Fig. 1)10. Sound elevation detection in humans and in many other mammalian species may therefore be considered as a spectral pattern-recognition problem. The importance of the pinnae in sound localization has been demonstrated in both pinna-occlusion experiments11, and in narrow-band sound localization studies1. Moreover, when localization is attempted through another persons ears (using virtual sound-source synthesis techniques), localization errors increase dramatically12. However, accurate localization on the basis of spectral cues poses constraints on the sound spectrum. A sound needs to be broad-band in order to yield sufficient spectral shape information, and the acoustic signal at the eardrum comprises the original source spectrum as modified by the linear pinna filter. Both spectral functions however, are a priori unknown to the auditory localization system, and the extraction of sound elevation and frontback direction is therefore not trivial1,2,13. It has been suggested that the auditory system may resolve this problem by assuming that real-life sounds do not contain the prominent peaks and notches of the different pinna filters. (Fig. 1a)1,2,13. Because sound localization relies on implicit physical cues, the auditory system must somehow transform the binaural difnature neuroscience volume 1 no 5 september 1998

ferences and monaural spectral pinna cues into consistent spatial information. It is thought that the auditory system acquires these spatial relations through learning, and that the visual system may train and calibrate the acoustic localization process by providing accurate spatial feedback. Indeed, behavioral experiments with young barn owls have shown that the integrity of the visual system guides acoustic localization performance 14,15 : When reared with prisms, the owls auditory localization response shifts in the same direction as the altered visual representation, although the acoustic cues remain unchanged. Likewise, acoustic perceptual shifts that are induced by one-sided ear plugs are resolved by the availability of visual feedback, although some adjustment of the optic tectums spatial map also occurs after eyelid suture16. In addition, the formation of the auditory space map in the owls inferior colliculus has been shown to rely mostly on visual experience in early life17. Comparable results have been obtained for the spatial representation of sound in the midbrain superior colliculus of newborn mammals18,19. Conceivably, the human auditory localization system may develop by a similar learning process, as the subtle acoustic cues vary substantially during growth. However, no data are available that clearly demonstrate an adaptive capability of the human auditory localization system20. We therefore tested whether human subjects would be able to adapt to a consistent change in the spectral localization cues.

Results To investigate the processes underlying the formation of the acoustic spatial percept in humans, four adult subjects continuously wore well fitting, custom-made molds within the concha of both ears for a period of up to six weeks. Subjects did not receive any specific localization training during this period. Although the molds dramatically altered the subjects spectral shape cues, they still provided consistent spectral information about stimulus elevation (Fig. 1a). The undisturbed ear of this example subject contained prominent spectral peaks and notches at different frequency bands for each sound elevation. For
417

1998 Nature America Inc. http://neurosci.nature.com

articles

Frequency (kHz) (kHz) example, at elevation -40 degrees, there was a proa Frequency found notch at about 6 kHz and a peak near 12 kHz. 1 2 4 8 16 The position and presence of such peaks and notches 40 varies in a systematic, albeit complex, manner with 69 elevation . There were marked changes after insert20 ing the mold (Fig. 1b). 0 During the adaptation period, each subjects local20 ization performance in two dimensions (2D) was quantified several times a week under open-loop con40 ditions. Because of the method used to measure localization performance, the target domain was confined 15 10 5 0 5 10 15 b 40 to positions within the oculomotor range ( 40 degrees straight ahead in all directions)13. We deter20 mined baseline behavior in response to white-noise 0 sounds presented at random locations for all four subjects (Fig. 2a). Both azimuth and elevation compo20 nents of the measured eye-movement vectors correlate 40 well with the actual stimulus directions, and the firstTransfer (dB) Transfer saccade end points accurately capture the spatial struc15 10 5 0 5 10 15 ture defined by the stimulus locations. Perfect behavior would require that the averaged saccade responses Fig. 1. Effect of molds on pinna acoustic transfer. The human pinna filters the align with the stimulus matrix. The data were also acoustic spectrum in a direction-dependent way. (a) Normal pinna transfer quantified by determining linear regression lines functions of the right ear of one subject. (b) Pinna transfer functions of the between target and response coordinates, for eleva- same ear after application of the mold (see also photograph). The linear tion and azimuth components, respectively. The fit- acoustic transfer functions are shown as a function of frequency (ordinate) and ted slopes, together with their standard deviations, are sound direction (abscissa) in the midsagittal plane (azimuth zero degrees; eleprovided in Fig. 3. Here, the range for the mean vation from -40 to +50 degrees). Color encodes the amplitude (in dB) of the absolute errors across target positions ( for eleva- transfer function. A value of zero dB indicates that the presence of the head tion, and for azimuth components) as well as Pear- and pinna does not change the sound pressure amplitude of a tone at that parsons linear correlation coefficients ( r and r , ticular frequency and elevation. Light colors correspond to sound amplificarespectively) are provided for each row. Both the tion; dark areas refer to sound attenuation. azimuth and elevation components of saccadic responses are quite accurate for all four subjects before application of the molds ( , 4.27.7 degrees; , 2.96.6 degrees; r, 0.920.96; r, 0.970.98) (Fig. 2a). the subjects localization accuracy with undisturbed ears was Immediately after application of the molds, however, localstill as high as before the start of the experiment (, 4.59.6 ization performance was dramatically disrupted, as far as the degrees; , 4.55.6 degrees; r, 0.950.96; r, 0.98) (Fig. 2e). detection of sound elevation was concerned. All subjects perThe largest absolute elevation errors in the control conditions ceived the sound at a roughly fixed, eccentric elevation angle are for subject PH, who displayed a systematic downward (given by the offset value of the regression line, which fell below response bias. Apparently, the auditory system had acquired a -20 degrees for subjects PH, JO and JR, and was about +20 new representation of the pinna transfer functions, without degrees for MZ), regardless of the actual stimulus elevation (Fig. interfering with the old set. 2b). (Saccade responses scatter around an approximately horiWe made adaptation curves (azimuth and elevation gain zontal line.) The azimuth component of the responses, howevas a function of time) for all four subjects (Fig. 3). Qualitaer, seemed to be almost unaffected and still accurate ( , tively, the results are very similar, although idiosyncratic differences in the learning behavior, as well as in the overall 16.323.0 degrees; , 3.88.3 degrees; r, 0.0-0.3; r, 0.970.98). performance accuracy may also be noted. As long as several This result underlines the importance of the spectral cues in days after the adaptation session was concluded, subjects could sound elevation localization in humans; it also demonstrates the still localize adequately with the molds, although the overall existence of independent neural mechanisms for the detection elevation gain slowly decreased with time (tested in three subof sound source azimuth and elevation. jects, data not shown). However, we cannot exclude any conAfter wearing the molds for several days, localization accuratribution of subtle differences in the spectral localization cues cy steadily improved over time in all subjects (Fig. 2c). There was provided by the individual molds. some stimulus-related structure in the elevation responses, which is apparent from significant values of the gains and decreasing offset values for the regression lines (, 11.1-16.5 degrees; , Discussion 3.38.3 degrees; r, 0.630.82; r, 0.970.98). This improvement, These experiments show that the adult human auditory system is depicted by the systematic unfolding of the response matrices, capable of considerable adaptation in response to altered speccontinued for about three to six weeks, after which the learning tral cues. This finding corroborates earlier results from barn owls process seemed to stabilize (, 7.412.6 degrees; , 4.27.0 in which the ability to adjust the auditory space map in the optic tectum in response to altered acoustic cues (induced by cutting degrees; r, 0.780.89; r, 0.98) (Fig. 2d). the facial ruff) persists into adulthood21. It differs qualitatively After localization performance had reached a stable level, the molds were removed and the subjects behavior was tested from findings showing that changes in the auditory space map without the molds. Interestingly, immediately after adaptation, in response to prisms are strongly limited by age in this species22. Elevation (deg) Original Original Modified Modified

1998 Nature America Inc. http://neurosci.nature.com

418

Elevation (deg)

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

PH MZ JO JR Fig. 2. Adaptation to altered spectral cues. Localization behavior of all four subjects (from left to right) before, during and immediately after the adapa tation period. Day zero marks the start of the adaptation experiment. The panels show, for each subject, the individual saccade vector endpoints in the azimuthelevation plane ( symbol ). In addib tion, the saccade vectors were also averaged for targets belonging to similar directions by dividing the target space into sixteen half-overlapping sectors. Averaged data points ( symbol ) from neighboring stimulus sectors are c connected by thick lines. In this way, a regular response matrix indicates that the subjects saccade endpoints capture the actual spatial distribution of the applied target positions. The target matrix, computed in the d same way as the saccade matrix, has been included for comparison (thin lines). (a) Results of the preadaptation control experiment on day zero, immediately preceding the application of the molds. (b) Localization responses e immediately after inserting the molds (day 0). Note the dramatic deficit in elevation responses for all subjects. (c) Results during Response azimuth (deg) the adaptation period after twelve (PH), six (MZ), five (JO) and 29 (JR) days of continuously wearing the ear molds. (d) Results near the end of the adaptation period. Stable and reasonably accurate localization behavior has been established in all subjects. (e) Results of the control condition, immediately after removal of the molds. All subjects localized sounds with their original ears equally well as before the start of the experiment several weeks earlier. Pre control Response elevation (deg) Modified pinnae
b

1998 Nature America Inc. http://neurosci.nature.com

The existence of a critical period for the formation of a spatial auditory map in the mammalian superior colliculus has also been demonstrated for newborn guinea pigs and ferrets23. Our finding that the molds remained effective after their removal indicates that the newly acquired pinna representations may have a semipermanent basis. Interestingly, the presence of experienceinduced auditory memory traces has recently been demonstrated in the adult barn owl localization system24. Another finding is the ability of the human sound localization system to cope with different representations of the spectral pinna cues. We believe that this property does not reflect the involvement of higher cognitive processes, as subjects never received feedback about their performance during the recording
nature neuroscience volume 1 no 5 september 1998

sessions. Moreover, response latencies were typically well below 300 milliseconds, and subjects were not aware of a difference in perceived sound quality for the mold and no-mold conditions. Apparently, both filter sets (see Fig. 1) are simultaneously represented within the human auditory system. The learning of a new set of pinna transfer functions therefore resembles more the acquisition of a new language than other forms of sensory adaptation. As a consequence, both neural representations will always be activated by sounds, and the question then is how the auditory system could select the correct pinna filter. It is possible to show, however, that when the two pinna sets are sufficiently disjunct from each other (that is, the correlations between different filter functions within and among the two sets are close
419

Post control

1998 Nature America Inc. http://neurosci.nature.com

articles

azimuth azimuth
1.25

1.00
Response gain Response Gain elevation elevation

0.75 C 0.50

bration of the auditory localization system, especially for those spatial regions where vision has a poor resolution (that is, in the far retinal periphery) or is even absent (for example, for rear stimulus positions and in darkness). Indeed, recent evidence indicates that active head movements are necessary to resolve frontback ambiguities in localization26. At present, it is not known which pathways of the human auditory system are responsible for the spectral analysis in elevation detection, nor where the learning in the human auditory system may take place. Future experiments will aim to determine the sensorimotor mechanisms and sites that are involved in these processes.
Methods SUBJECTS AND EAR MOLDS. Subjects were four adults (the authors and one nave subject), ages 2240. Pinna filter functions were measured by recording the sound pressure level in the ear canal, at a location 12 mm from the eardrum, with a thin silicone tube attached to a miniature microphone (Knowles, EA1842). Sounds were presented at many different locations surrounding the subject (speaker positioned at five-degree intervals, both in azimuth [-90, +90] degrees, and in elevation [-40, +75] degrees). The sound used to measure the pinna filters was a minimumpeak broad-band FM sweep (0.2-20 kHz, flat amplitude spectrum, Schrder phase10,13,27).

0.25

PH MZ JO JR

1998 Nature America Inc. http://neurosci.nature.com

0.00 0 10 20
Time (days) (days) Time

30

40

Fig. 3. Summary of the results for all four subjects. Adaptation curves for azimuth and elevation gain (defined as the slope of the best-fit regression line through the data of each recording day), as a function of time from the start of the adaptation experiment (in days). Standard deviations in the gains were obtained by bootstrapping the data 100 times30,31. Results for the pre- and post-adaptation control conditions are also shown for comparison (symbols C, elevation data only, for clarity; control azimuth gains fell in the range 1.01.2 pre- and 1.21.3 post-adaptation).

Concha molds were prepared by making a negative imprint of both ears by filling both pinnae with plaster. From the hardened negative plaster images, silicone positive replicas of both ears were manufactured. Subsequently, the concha molds were precisely shaped by applying a thin polyester layer (about 0.51 mm in thickness) within the concha replicas. Then, a thin layer of wax (about 2 mm) was applied, finished with skin-colored paint (see Fig. 1). We verified that the modified ear still received specific elevation-dependent spectral features by training a two-layer feedforward neural network to map the pinna filter functions of the modified ear (input layer) onto the elevation domain (one output unit). The trained network yielded a high correlation between required and predicted target elevation. EXPERIMENTAL CONDITIONS. Auditory spatial locations, as well as firstsaccade vector endpoints are described in a double-pole azimuth () and elevation () coordinate system5,13; and are the directions relative to the vertical median plane and horizontal plane respectively. Auditory stimuli used in the localization experiments consisted of broad-band white-noise sound bursts (0.220 kHz, 500 ms, 60 dB SPL), and were presented in complete darkness in an echo-free room at randomly chosen locations within the 2D oculomotor range ([, ] within [-30, 30] degrees). The speaker was moved by a two-link robot system that could rapidly position the stimulus at any point on a virtual hemisphere surrounding the subject at a radius of 0.9 m (refs 4,13). Subjects kept their head immobile against a head rest and were instructed to generate a rapid and accurate saccadic eye movement from an initial, centrally presented light-emitting diode toward the perceived location of the sound. No feedback was given about performance. Eye movements were measured with the magnetic search coil technique28,29. Earlier studies from our group4,13 have indicated excellent localization behavior under these open-loop testing conditions (no dynamic spatial cues, as the head was static; see also Fig. 2a).

to zero1,2,13), a broad-band white-noise sound will typically yield a high response for only one particular filter13. Not only does this filter belong to the actually applied pinna set, it also corresponds to the correct target elevation13. If the sound localization percept is ultimately produced by a winner-take-all mechanism, this maximally activated filter determines the actual localization response. In this way, the subject may rely entirely on the available acoustic input, provided the sound-source spectrum itself is both broad-band and unrelated to any of the stored pinna filters of either set1,2,13. The adaptation mechanism reported in this study is very different from adaptation to prisms in humans, for which a strong after-effect is usually obtained, resulting in systematic localization errors immediately after prism removal. Also in owls, adaptation to prisms, as well as to monaural ear plugs, results in a remapping of the auditory spatial map such that, when the prism or earplug is removed, the owl exhibits systematic errors in sound localization. These errors are resolved only by providing adequate (visual) feedback14. All four subjects were unaware of the anomaly in their acoustic localization ability while wearing the molds in daily life. This emphasizes once more the power of visual spatial information over the perception of auditory location (also known as the ventriloquist effect25) and may have provided the driving force underlying the learning response. We propose that the adaptive capability of the human auditory localization system is contingent on the availability of a sufficiently rich set of spectral cues, as well as on visual feedback about actual performance in daily life. Although not tested here, active head movements may also have contributed to the cali420

Acknowledgements
This research was supported by the Dutch Foundation for the Life Sciences (SLW 805-33.705-P; PMH), the University of Nijmegen, the Netherlands (AJVO), and the Human Frontiers Science Program (RG0174/1998-B; AJVO).

RECEIVED 28 MAY: ACCEPTED 7 JULY 1998


1. Middlebrooks, J. C. Narrow-band sound localization related to external ear acoustics. J. Acoust. Soc. Am. 61, 26072624 (1992).

nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

1998 Nature America Inc. http://neurosci.nature.com

2. Zakarouskas, P. & Cynader, M. S. A computational theory of spectral cue localization. J. Acoust. Soc. Am. 94, 13231331 (1993). 3. Oldfield, S. R. & Parker, S. P. Acuity of sound localization: a topography of auditory space. I. Normal hearing conditions. Perception 13, 581600 (1994). 4. Frens, M. A. & Van Opstal, A. J. A quantitative study of auditory-evoked saccadic eye movements in two dimensions. Exp. Brain Res. 107, 103117 (1995). 5. Knudsen, E. I. & Konishi, M. Mechanism of sound localization in the barn owl (Tyto alba). J. Comp. Physiol. A 133, 1321 (1979). 6. Batteau, D. W. The role of pinna in human localization. Proc. R. Soc. Lond. B 168, 158180 (1967). 7. Blauert, J. Spatial Hearing. The Psychophysics of Human Sound Localization. (MIT Press, Cambridge, Massachusetts, 1996). 8. Teranishi, R. & Shaw, E. A. G. External-ear acoustic models with simple geometry. J. Acoust. Soc. Am. 44, 257263 (1968). 9. Lopez-Poveda, E. A. & Meddis, R. A physical model of sound diffraction and reflections in the human concha. J. Acoust. Soc. Am. 100, 32483259 (1996). 10. Wightman, F. L. & Kistler, D. J. Headphone simulation of free-field listening. I. Stimulus synthesis. J. Acoust. Soc. Am. 85, 858867 (1989). 11. Oldfield, S. R. & Parker, S. P. Acuity of sound localization: a topography of auditory space. II. Pinna cues absent. Perception 13, 601617 (1984). 12. Wenzel, E. M., Arruda, M., Kistler, D. J. & Wightman, F. L. Localization using nonindividualized head-related transfer functions. J. Acoust. Soc. Am. 94, 111123 (1993). 13. Hofman, P. M. & Van Opstal, A. J. Spectro-temporal factors in two-dimensional human sound localization. J. Acoust. Soc. Am. 103, 26342648 (1998). 14. Knudsen, E. I. & Knudsen, P. F. Vision guides the adjustment of auditory localization in young barn owls. Science 230, 545548 (1985). 15. Knudsen, E. I. & Knudsen, P. F. Vision calibrates sound localization in developing barn owl. J. Neurosci. 9, 33063313 (1989). 16. Knudsen, E. I. & Mogdans, J. Vision-independent adjustment of unit tuning to sound localization cues in response to monaural occlusion in developing owl optic tectum. J. Neurosci. 12, 34853493 (1992). 17. Brainard, M. S. & Knudsen, E. I. Experience-dependent plasticity in the

18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31.

inferior colliculus: a site for visual calibration of the neural representation of auditory space in the barn owl. J. Neurosci. 13, 45904608 (1993). King, A. J., Hutchings, M. E., Moore, D. R. & Blakemore, C. Developmental plasticity in the visual and auditory representation in the mammalian superior colliculus. Nature 332, 7376 (1988). Withington-Wray D. J., Binns, K. E. & Keating, M. J. The maturation of the superior collicular map of auditory space in the guinea pig is disrupted by developmental visual deprivation. Eur. J. Neurosci. 2, 682692 (1990). Javer, A. R. & Schwarz, D. W. F. Plasticity in human directional hearing. J. Otolaryngol. 24, 111117 (1995). Knudsen, E. I., Esterly, S. D. & Olsen, J. F. Adaptive plasticity of the auditory space map in the optic tectum of adult and baby barn owls in response to external ear modification. J. Neurophysiol. 71, 7994 (1994). Knudsen, E. I. & Knudsen, P. F. The sensitive period for auditory localization in barn owls is limited by age, not by experience. J. Neurosci. 6, 19181924 (1986). King, A. J. & Moore, D. R. Plasticity of auditory maps in the brain. Trends Neurosci. 14, 3137 (1991). Knudsen, E. I. Capacity for plasticity in the adult owl auditory system expanded by juvenile experience. Science 279, 15311533 (1998). Stein, B. E. & Meredith, A. M. The Merging of the Senses (MIT Press, Cambridge, Massachusetts, 1993). Perrett, S. & Noble, W. The contribution of head motion cues to localization of low pass-noise. Percept. Psychophys. 59, 10181026 (1997). Schrder, M. R. Synthesis of low-peak factor signals and binary sequences with low autocorrelation. IEEE Trans. Inform. Theory 16, 8589 (1970). Robinson, D. A. A method of measuring eye movement using a scleral search coil in a magnetic field. IEEE Trans. Biomed. Eng. 10, 137145 (1963). Collewijn, H., Van der Mark, F. & Janssen, T. J. Precise recording of human eye movements. Vision Res. 15, 447450 (1975). Press, W. H., Flannery, B. P., Teukolsky, S. A. & Vettering, W. T. Numerical Recipes in C, 2nd edn (Cambridge Univ. Press, Cambridge, 1992). Efron, B. & Tibshirani, R. Statistical analysis in the computer age. Science Wash. D.C. 253, 390395 (1991).

nature neuroscience volume 1 no 5 september 1998

421

1998 Nature America Inc. http://neurosci.nature.com

articles

Analysis of temporal structure in sound by the human brain


Timothy D. Griffiths1,2,3, Christian Bchel2, Richard S.J. Frackowiak2 and Roy D. Patterson1
1 2 3

Centre for the Neural Basis of Hearing, Physiology Department, University of Cambridge, Downing Street, Cambridge, CB2 3EG, UK Wellcome Department of Cognitive Neurology, Institute of Neurology, 12 Queen Square, London, WC1N 3BG, UK Department of Physiological Sciences, Newcastle University Medical School, Newcastle upon Tyne, NE2 4HH,UK Correspondence should be addressed to T.D.G. (t.griffiths@fil.ion.ucl.ac.uk)

1998 Nature America Inc. http://neurosci.nature.com

For over a century, models of pitch perception have been based on the frequency composition of the sound. Pitch phenomena can also be explained, however, in terms of the time structure, or temporal regularity, of sounds. To locate the mechanism for the detection of temporal regularity in humans, we used functional imaging and a delay-and-add noise, which activates all frequency regions uniformly, like noise, but which nevertheless produces strong pitch perceptions and tuneful melodies. This stimulus has temporal regularity that can be systematically altered. We found that the activity of primary auditory cortex increased with the regularity of the sound. Moreover, a melody composed of delay-and-add notes produced a distinct pattern of activation in two areas of the temporal lobe distinct from primary auditory cortex. These results suggest a hierarchical analysis of time structure in the human brain.

SPECTRAL MODELS OF PITCH PERCEPTION


Traditionally, pitch perception is explained in terms of spectral analysis, and in particular the spectral analysis performed by the inner ear, or cochlea, where the frequency components of a sound produce activity in different places along the cochlea in accordance with their frequencies1. Sinusoidal sounds are the easiest to explain; they contain only one frequency component and produce a pitch corresponding to that frequency. Musical notes and the vowels of speech are composed of sets of harmonically related sinusoids, and they produce a pitch corresponding to the fundamental of the harmonic series. A short sample of the neural activity pattern2 (NAP) produced by the cochlea in response to a musical note composed of all the harmonics of 62.5 Hz is shown in Fig. 1a; the pitch of the note is 62.5 Hz, which is close to the C two octaves below middle C on the keyboard. The ordinate in the figure is place along the length of the cochlea, and it is essentially a logarithmic frequency scale. The abscissa is time in milliseconds, so the figure shows the temporal microstructure of the neural response. The repetition rate of the neural pattern corresponds to the pitch of the sound. Figure 1b shows the auditory spectrum of the sound 3 at three points in time: 200, 300 and 400 ms after the onset of the note. The auditory spectrum is the running temporal average of activity along the cochlea. In the lowfrequency region, there are peaks corresponding to the first eight harmonics of the note. As frequency increases, however, the resolution of the spectral analysis decreases, and the auditory spectrum becomes smooth above the eighth harmonic. These auditory spectra show that the first eight harmonics in the series are resolved (that is, they produce individual peaks in the auditory spectrum). In spectral models of pitch perception 4, the low-frequency peaks are identified by the brain, and their frequency spacing is calculated to extract the pitch of the sound.
422

TEMPORAL MODELS OF PITCH PERCEPTION


The problem with the traditional model of pitch is that the sound in Fig. 1 still produces the same pitch, (albeit with reduced strength or salience) when the sound is filtered to remove all the low-frequency components that produce individual peaks in the auditory spectrum. In this case, the lower half of the NAP and auditory spectrum are empty, and there are no spectral peaks with which to calculate the pitch. In spectral models, the temporal information in the NAP is averaged out in the construction of the auditory spectrum. In contrast, single-unit recordings show that temporal regularity in a stimulus is followed with extreme (sub-millisecond) accuracy by auditory nerve impulses5, and this has prompted the development of temporal models that associate pitch with the dominant time interval in the NAP6,7. To emphasize the importance of time-interval information, the modelers developed a stimulus in which a noise is delayed and added to itself repeatedly6,7. It is perceived as a tone with a pitch corresponding to the reciprocal of the delay, mixed with a background noise, a bit like a cracked bassoon playing in the wind. The strength of the pitch increases with the number of iterations of the delay-and-add process. Figure 1c and d shows the NAP and auditory spectra produced when the delay is 16 ms and there are eight iterations of the delay-and-add process. The sound has the same pitch as the sound represented in the upper panel, and the pitch is strong relative to the background noise; moreover, the pitch remains when the stimulus is high-pass filtered. For this stimulus, neither the auditory spectrum nor the NAP provides a good explanation of the perceived pitch. There are no stable, harmonically related peaks in the auditory spectra, and although the NAP does contain an excess of time intervals at 16 ms, it is not apparent in the NAP because the envelope does not repeat regularly in time. In temporal models of hearing, it is assumed that the pitch is extracted with autocorrelation, in which the neural pattern is cornature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

ferent neurons detect and accumulate activity about specific time intervals in the NAP. Animal studies indicate that the temporal integration may occur in the inferior colliculus9; the current study was carried out to delimit the location of the temporal integration mechanism in humans. In models of temporal integration, regular and irregular sounds with the same total energy and the same auditory spectra produce the same firing rate in units located prior to temporal integration but distinctly different rates of firing after the regularity is detected. Specifically, regular sounds Magnitude Time (ms) will produce greater activity than irregular sounds in units of the temporal array that c d detect the regularity. Delay-and-add noise enables us to create a set of stimuli, all of which have the same total energy and the same auditory spectra, but which nevertheless, have varying degrees of temporal regularity. These stimuli should produce uniform levels of activity in nuclei prior to temporal integration and very different levels in the temporal integration area and subsequent areas. In this study, we used positron emission tomography (PET) to address the question of where in the auditory system temporal integration occurs. We have identified Time (ms) Magnitude brain areas that show increased activation Fig. 1. Spectral representations of complex sounds that produce a pitch perception of 62.5 as a function of the degree of temporal Hz. The right panels show the activity produced by the sounds across the auditory spec- structure of the stimulus, which must, trum, averaged over time. (a) Activity pattern in the auditory nerve in response to a musi- therefore, lie at a higher level in the audical note. (b) The musical note produces a clear harmonic structure at low frequencies. (c) tory pathway than the temporal-integraActivity pattern in response to the synthetic stimulus used in this experiment. A stimulus tion mechanism. We also investigated the with eight delay-and-add cycles, or iterations, is represented. (d) The experimental stimulus effect of presenting these stimuli as produces no consistent spectral pattern, especially at the relatively high frequencies corre- melodies. This allowed us to define areas sponding to the passband used in the experiment. The simulation is based on a conventional that are involved in the analysis of tempoauditory filter bank implemented in the AIM software2. ERB, equivalent rectangular bandral structure at the level of sound sequences width (absolute frequency scale based on the ears frequency analysis. (seconds) rather than the level of pitch (milliseconds). We conclude that temporal integration for the representation of time structure at the millisecond level occurs before primary auditorelated with a delayed version of itself 6,7. When autocorrelation is ry cortex, whereas analysis of longer-term time structure in sound applied to each channel of the NAPs in Fig. 1, the result is the two occurs beyond primary auditory cortex. autocorrelograms in Fig. 2a and b; they shows that there is a concentration of time intervals at 16 ms in both stimuli in most channels of the NAP. Figure 2c and d show the average autocorrelation Results across frequency channels; the heights of the peaks at 16 ms show FINE TIME-STRUCTURE ANALYSIS that the pitch is relatively stronger in the harmonic sound, but it First, we defined brain areas activated by the stimulus after the is still prominent in the delay-and-add noise. The height of the temporal integration mechanism had stabilized the representapeak in the summary autocorrelogram rises and falls with the tion of the fine time structure within the individual. Ten nortemporal regularity of the NAP, and it predicts the pitch of commal subjects underwent 12 scans using positron emission plex sounds and the relative salience of the pitch6,7. tomography to measure regional cerebral blood flow as an index of neural activity while attending to the stimuli. Subjects were required to listen for any changes in the sound but did not perTHE ANATOMICAL BASIS FOR TIME-INTERVAL PROCESSING form any output task. Over the 80-second acquisition period of Temporal models of pitch do not address the question of where any scan, the subject was presented with a sequence of one-secthis time-interval processing might occur. Physiological studies ond sounds. The pitch of the individual sounds in each scan, show that the accuracy of timing information decreases at highdetermined by the delay in the delay-and-add process, was varer levels of the auditory system8, which means that the temporal ied for all scans over the same pitch range. All sounds in any microstructure must be converted into a more stable code (probscan contained the same degree of temporal structure, deterably spatial9) in the auditory pathway up to the cortex. The mined by the number of iterations. Between scans, the number process involves short-term temporal integration in which difCenter frequency (ERBs) Center frequency (ERBs)

1998 Nature America Inc. http://neurosci.nature.com

nature neuroscience volume 1 no 5 september 1998

Center frequency (ERBs)

Center frequency (ERBs)

423

1998 Nature America Inc. http://neurosci.nature.com

articles

a
Center frequency (ERBs)

Autocorrelation lag (ms)

Autocorrelation lag (ms)

c
1998 Nature America Inc. http://neurosci.nature.com
Magnitude

Autocorrelation lag (ms)

Autocorrelation lag (ms)

Fig. 2. Time-interval representations (autocorrelograms) of complex sounds that produce a pitch perception of 62.5 Hz. (a) The autocorrelogram of activity in the auditory nerve corresponding to the musical note in Fig. 1. (b) The autocorrelogram of activity in the auditory nerve corresponding to the synthetic stimulus used in this experiment. In both cases temporal regularity is apparent at 16 ms, which is clearly shown by the peaks in the summary autocorrelation for the musical note (c) and the experimental stimulus (d).

of iterations was varied from 016 in a parametric design. Thus, although the scans were 80 seconds long, the only differences between the scans were in the fine temporal structure at the millisecond level. This allowed us to seek brain areas that occur after temporal integration; models of temporal integration2 predict increasing activation as a function of temporal regularity after integration has occurred. This experiment was therefore specifically designed to delimit the location of brain regions occurring after temporal integration, rather than to demonstrate pitch mapping as in previous experiments10. The first analysis sought the predicted increase in activity with increasing temporal regularity in three areas for which there is reasonable anatomical specification: the inferior colliculus, medial geniculate body and primary auditory cortex. Figure 3 is a statistical parametric map showing areas where there was a main effect of temporal structure (number of iterations) on activity. That is, the map shows areas where regional brain activity increases with increased temporal regularity in the sound stimulus. Taking the accepted position of the human primary auditory cortices on each side11 as our prior anatomical hypothesis, the relationship in both areas was significant using the spatial extent method of Friston12. The areas of activation show anterior posterior asymmetries, with the right area of activation being placed more anteriorly, which is consistent with previous reports of the location
Table 1. Areas showing significant iterationmelody interaction
Area Right posterior temporal Left anterior temporal Left posterior temporal Right anterior temporal 424 Talairach coordinates 72 -40 6 -54 10 -18 -58 -42 -2 58 12 -26 Z score 5.68 4.99 4.73 4.71

of the primary auditory cortex11. The area of activation on the left is situated more laterally and may also correspond to secondary auditory cortex rather than primary auditory cortex . No corresponding effects were demonstrated in the inferior colliculus or medial geniculate body. It was noteworthy that no significant activation as a fuction of temporal regularity was found in any other cortical area.

LONGER-TERM TIME-STRUCTURE ANALYSIS


Second, we defined areas where the pattern of activation depends on the temporal structure of the sound at the level of sound sequences (seconds), rather than at the level of milliseconds, by presenting the individual sounds as different patterns of pitch (as melodies and as simpler pitch patterns). Specifically, we sought areas where the pattern of activation varied with the degree of sound-sequence complexity. Each subject underwent two scans for each level of fine-structure regularity (number of iterations). The pitch range in each scan was the same. However, the pitch was varied in two distinct ways at each level (Fig. 4). In the nomelody condition, the pitch was increased in a systematic staircase pattern, whereas in the melody condition, the same variation in pitch over the scan was achieved by presenting novel diatonic melodies composed specifically for the experiment (Methods). This second analysis sought brain areas involved in processing the longer-term temporal structure of the sound, determined by the type of sound sequence. We hypothesized that areas involved in the analysis of sound sequences would show greater activation as a function of iteration for the melody condition than for the no-melody condition. Sound-sequence analysis depends on detecting the pitches of individual sounds, and areas involved in sequence analysis will show a more marked dependence on pitch strength for more complex pitch sequences like the melody condition. Such an interaction was shown in four areas (Table 1; Fig.
nature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

b
Adjusted rCBF

Left Adjusted rCBF

Right

Temporal structure 10 Log (n+1)

Temporal structure 10 Log (n+1)

Fig. 3. Main effect of temporal structure within the individual sounds (iteration). (a) Statistical parametric map showing areas where brain activity increased with increasing temporal structure of the sound, as determined by the number of delay-and-add cycles (iterations) for the noise stimulus. The areas of significant effect are shown in color, rendered onto axial and coronal MRI brain sections. Activation is demonstrated in the left superior temporal planum (Talairach coordinates -70 -16 -2, Z = 3.45) and in the right superior temporal planum (58 -2 -4, Z = 2.8) at the level of the primary auditory cortices. (b) Relationship between activity and temporal structure in the two areas showing a significant effect in (a). Temporal structure is measured by 10 log (n + 1), where n is the number of iterations.
1998 Nature America Inc. http://neurosci.nature.com

5a and b): the right and left posterior superior temporal gyri and the right and left anterior temporal lobes. These areas are not contiguous with primary auditory cortex, as are most of the primate auditory association cortices1315. Figure 5c shows the relationship between activation and iteration for the no-melody and melody conditions considered separately, in the four areas of interaction. In all areas, the blood flow decreased with iteration in the no-melody condition and increased with iteration in the melody condition.

cephalography are also consistent with auditory cortex subserving perception based on timing information10,20, but they used stimuli with peaks in the auditory spectrum as well as temporal regularity. The latter study, additionally, suggested orthogonal coding of timing and spectral information in the auditory cortices, given the constraint on the temporal interpretation above. The current study indicates that iterated rippled noise in combination with parametric functional imaging provides a powerful method for studying auditory temporal processing in the absence of spectral cues.

Discussion

FINE TIME-STRUCTURE ANALYSIS


This is the first demonstration of activity in the human auditory cortex that varies with the degree of temporal regularity independently of the auditory spectrum, indicating that temporal integration occurs at or before the auditory cortex. In contrast, most previous functional imaging studies of human auditory cortex have demonstrated a mapping of spectral features1618. Our data show that the human primary auditory cortices subserve both timing and spectral analyses. This is in accord with animal work suggesting orthogonal representation of both timing and spectral information in the ascending pathway in the inferior colliculus9. Although the data are consistent with temporal integration at or before the auditory cortex, we suspect that this process has occurred before the cortex; lesions of the human auditory cortex produce deficits at longer temporal intervals (tens of milliseconds) than those involved in pitch perception (milliseconds)19. Two previous human studies using magnetoen-

LONGER-TERM TIME-STRUCTURE ANALYSIS

The analysis of longer-term time structure demonstrated strikingly symmetrical areas in the anterior and posterior temporal lobes where the pattern of activity varies according to the complexity of the sound sequence. These areas are hypothesized to be involved in sound-sequence analysis. The differences in sequence complexity occur at the level of seconds, which is relevant both to melodic perception and to the suprasegmental (prosodic) analysis of speech. We suggest that the areas shown are analyzing longer-term time structure in a manner that is not specific to music or speech. Human lesion studies21 also support bilateral involvement of the superior temporal cortices in both melodic and prosodic analysis, although melodic analysis deficits without dysprosody can occur with right unilateral lesions22,23. The areas shown are likely to be activated via cortico-cortical projections from primary auditory cortex. We suggest that they are acting on pitch tokens (symbolic representation of pitch) and are not involved in the primary processing of the pitch itself. a Stimuli such as these where the spectrum is carefully controlled may be useful in future studies of sound sequencing and melodic perception, as they allow manipulation of the long-term temporal structure without altering the auditory spectrum over time. In previous b imaging studies of pitch processing in speech and music2426, the output tasks have been varied with identical stimuli to demonstrate different neural subFig. 4. Pitch sequences used as stimuli. Excerpts from the beginning and end of (a) the no- strates for different aspects of speech melody pitch sequence and (b) one of the melody pitch sequences. The greater sequence and music processing. This represents a more cognitive approach to musical complexity in the melody condition is apparent.
nature neuroscience volume 1 no 5 september 1998 425

1998 Nature America Inc. http://neurosci.nature.com

articles

c
Adjusted rCBF

Left anterior temporal Adjusted rCBF

Right anterior temporal

1998 Nature America Inc. http://neurosci.nature.com

Left posterior temporal Adjusted rCBF Adjusted rCBF

Right posterior temporal

Temporal structure 10 Log (n+1)

Temporal structure 10 Log (n+1)

Fig. 5. Iterationmelody interaction. Statistical parametric map showing areas where brain activity increased more with iteration for the melody condition than for the no-melody condition. (a, b) Bilateral activation of the anterior (a) and posterior (b) temporal cortices is demonstrated. The rostrocaudal (z) and anteroposterior (y) levels for the sections are z = -20, y = 0 (a) and z = -10, y = -40 (b). (c) Patterns of brain activity (measured by adjusted regional cerebral blood flow) for the two conditions with different long-term temporal structure, in the four areas showing significant interaction in (a) and (b). For each condition, regional cerebral blood flow is plotted against the temporal structure in the individual sounds. Temporal structure is measured by 10 log (n + 1), where n is number of iterations. Squares show average regional cerebral blood flow for ten subjects for the no-melody condition, and circles show average regional cerebral blood flow for the melody condition. Adjusted regional cerebral blood flow is the value relative to a defined mean of 50 ml per dl per min. Error bars show one standard error. The regression line for each condition is plotted.

analysis than in the present study; we have demonstrated activation beyond primary auditory cortex without varying the output task. This suggests that we are observing a low-level mechanism for sound sequencing, and we would argue that the present study shows activation that is relevant to the temporal analysis of sound in general.

HIERARCHICAL ANALYSIS OF TEMPORAL STRUCTURE


This work suggests that the processing of temporal structure in the human brain occurs at different anatomical levels. Fine temporal structure associated with the perception of pitch is processed in the auditory pathway up to and including primary auditory cortex, whereas the analysis of pitch sequences involves cortical areas distinct from primary auditory cortex. Processing of emergent temporal properties, such as pitch sequences in sound, is allowed by cortico-cortical connections from the primary auditory cortex and suggests analysis in the human brain based on a hierarchy of temporal levels.
Methods STIMULI AND PET SCANNING. Twelve scans were carried out on each subject using the oxygen-15-labeled water bolus technique. A Siemens scanner was used in three-dimensional mode. Stimuli were add-same iterated rippled noise7 (IRNS) presented over insert earphones at 75 dB SPL total power. Sounds were 1 s long and presented with a repetition period of 1.1 s. Within any given scan, the number of iterations for all the sounds was fixed at one of six values: 0, 1, 2, 4, 8 or 16. The passband for the sounds was 13.5 kHz. High-pass filtering at 1 kHz precludes spectral processing in the auditory system for these lowpitched stimuli. At each of the six values of iteration, subjects underwent two scans, during which the pitch of the sounds was varied 426

between 50 and 250 Hz in one of two different ways. In the melody condition, pitch variation was used to produce novel diatonic melodies consisting of six four-bar phrases that presented a theme, development and recapitulation in the traditional way (Fig. 4b). The sounds all had the same duration, and so there was no rhythmic information except in the last bar of each phrase, where there was a double gap before the first note and no gap between the last two notes. This modification suggests an end-of-phrase pause, which is important to the sense of music, without actually varying the mean rate of sounds. In the nomelody condition (Fig. 4a), pitch variation was strictly limited; the pitch was fixed for 12 notes and then it was increased by two notes on the diatonic scale. The initial pitch was always A1 (55 Hz), and over the 8 sets of 12 sounds, the pitch rose slowly over the full range used in the melodies. There was no rhythm information whatsoever in the no-melody condition; the sounds were delivered at a constant rate. ANALYSIS. Data were analyzed with Statistical Parametric Mapping software [SPM96, http//:www.fil.ion.ucl.ac.uk/spm]. Scans were realigned and spatially normalized 27 to the standard stereotactic space of Talairach28. The standard template of the Montreal Neurological Institute was used29. The data were smoothed with a Gaussian filter (full width at half maximum of 16 mm). All analyses were also carried out using a Gaussian filter with a full width at half maximum of 8 mm, to maximize the chance of demonstrating activation in small subcortical structures. Analysis of covariance was used to correct for differences in global blood flow between the scans and implement a regression analysis to find areas where blood flow increased linearly with temporal structure (represented by log (n + 1), where n is number of iterations). The significance of this regression was assessed with the t statistic at each voxel. These statistics (after transformation to a Z score) constitute an SPM{Z}30. Implementation of second- and third-order polynomial fit did not demonstrate a better fit to the data. The activations demonstrated in the primary auditory cortices as a function of iteration were signifinature neuroscience volume 1 no 5 september 1998

1998 Nature America Inc. http://neurosci.nature.com

articles

cant at p < 0.05 using the spatial extent method of Friston12 to take the prior anatomical hypotheses into account. The interaction analysis sought areas where the relationship between regional cerebral blood flow and iteration had a greater slope for the melody than for the no-melody condition. The interactions in the areas listed were significant at the level of p < 0.05 after correction for multiple comparisons using Gausssian field theory, in the absence of a specific prior anatomical hypothesis. A main effect of melody was also demonstrated involving the primary auditory cortices and extending along the superior temporal lobes bilaterally. However, the interaction analysis is more informative than the simple main effect: the main effect demonstrates areas involved in sequence analysis and areas with an augmented response to sounds that simply vary more from sound to sound. The consistent difference in blood flow between the zero-iteration conditions [10 log (n + 1) = 0] in each panel of Fig. 5c is due to the simple rhythm cue used to indicate end of phrase in the melody condition. Although the rhythm was constant across iterations within the melody and no-melody conditions, it could still contribute to the iteration.
1998 Nature America Inc. http://neurosci.nature.com

Acknowledgements
T.D.G., C.B. and R.S.J.F. are supported by the Wellcome Trust. R.D.P. is supported by the MRC (UK).

RECEIVED 29 APRIL: ACCEPTED 16 JULY 1998


1. von Helmholtz, H. L. F. On the Sensations of Tone (Longmans, London, 1885). 2. Patterson, R. D., Allerhand, M. H. & Giguerre, C. Time-domain modeling of peripheral auditory processing: a modular architecture and a software platform. J Acoust. Soc. Am. 98, 18901894 (1995). 3. Patterson, R. D. The sound of a sinusoid: Spectral models. J. Acoust. Soc. Am. 96, 14091418 (1994). 4. Cohen, M. A., Grossberg, S. & Wyse, L. A spectral network model of pitch perception. J. Acoust. Soc. Am. 98, 862879 (1995). 5. Rose, J. E., Brugge, J. F., Anderson, D. J. & Hind, J. E. Phase-locked response to low-frequency tones in single auditory nerve fibres of the squirrel monkey. J. Neurophysiol. 30, 769793 (1967). 6. Patterson, R. D., Handel, S., Yost, W. A. & Datta, A. J. The relative strength of the tone and the noise components in iterated rippled noise. J. Acoust. Soc. Am. 100, 32863294 (1996). 7. Yost, W. A., Patterson, R. & Sheft, S. A time domain description for the pitch strength of iterated rippled noise. J. Acoust. Soc. Am. 99, 10661078 (1996). 8. Rouilly, E., deRibaupierre, Y. & deRibaupierre, F. Phase-locked responses to low frequency tones in the medial geniculate body. Hear. Res. 213226 (1979). 9. Langner, G. & Schreiner, C. E. Periodicity coding in the inferior colliculus of the cat. I. neuronal mechanisms. J. Neurophysiol. 60, 17991822 (1988).

10. Pantev, C., Hoke, M., Lutkenhoner, B. & Lehnertz, K. Tonotopic organization of the auditory cortex: pitch versus frequency representation. Science 242, 486488 (1989). 11. Penhune, V. B., Zatorre, R. J., MacDonald, J. D. & Evans, A. C. Interhemispheric anatomical differences in human primary auditory cortex: probabalistic mapping and volume measurement from magnetic resonance scans. Cereb. Cortex 6, 661672 (1996). 12. Friston, K. J. Testing for anatomically specified regional effects. Hum. Brain. Mapp. 5, 133136 (1997). 13. Merzenich, M. M. & Brugge, J. F. Representation of the cochlear partition on the superior temporal plane of the macaque monkey. J. Neurophysiol. 24, 193202 (1973). 14. Rauschecker, J. F., Tian, B., Pons, T. & Mishkin, M. Serial and parallel processing in rhesus monkey auditory cortex. J. Comp. Neurol. 382, 89103 (1997). 15. Pandya, D. N. Anatomy of the auditory cortex. Rev. Neurologique 151, 486494 (1995). 16. Lauter, J. L., Herscovitch, P., Formby, C. & Raichle, M. E. Tonotopic organisation in the human auditory cortex revealed by positron emission tomography. Hear. Res. 20, 199205 (1985). 17. Talavage, T. M., Ledden, P. J., Sereno, M. I., Rosen, B. R. & Dale, A. M. Multiple phase-encoded tonotopic maps in human auditory cortex. Neuroimage 5, S8 (1997). 18. Wessinger, C. M., Buonocore, M. H., Kussmaul, C. L. & Mangun, G. R. Tonotopy in human auditory cortex examined with functional magnetic resonance imaging. Hum. Brain Mapp. 5, 1825 (1997). 19. Phillips, D. P. & Farmer, M. E. Acquired word deafness and the temporal grain of sound representation in the primary auditory cortex. Behav. Brain Res. 40, 8594 (1990). 20. Langner, G., Sams, M., Heil, P. & Schulze, H. Frequency and periodicity are represented in orthogonal maps in the human auditory cortex; evidence from magnetoencephalography. J. Comp. Physiol. 181, 665676 (1997). 21. Peretz, I. et al. Functional dissociations following bilateral lesions of auditory cortex. Brain 117, 12831301 (1994). 22. Zatorre, R. J. & Halpern, A. R. Effect of unilateral temporal lobe excision on perception and imagery of songs. Neuropsychologia 31, 221232 (1993). 23. Griffiths, T. D. et al. Spatial and temporal auditory processing deficits following right hemisphere infarction. A psychophysical study. Brain 120, 785794 (1997). 24. Zatorre, R. J., Evans, A. C., Meyer, E. & Gjedde, A. Lateralization of phonetic and pitch discrimination in speech processing. Science 256, 846849 (1992). 25. Zatorre, R. J., Evans, A. C. & Meyer, E. Neural mechanisms underlying melodic perception and memory for pitch. J. Neurosci.14, 19081919 (1994). 26. Platel, H. et al. The structural components of musical perception. A functional anatomical study. Brain 120, 229243 (1997). 27. Friston, K.J. et al. Spatial registration and normalisation of images. Hum. Brain Map. 2, 165169 (1995). 28. Talairach, P. & Tournoux, J. A Stereotactic Coplanar Atlas of the Human Brain (Thieme, Stuttgart, 1988). 29. Evans, A. C., Kamber, M., Collins, D. L., & Macdonald, D. in Magnetic Resonance Scanning and Epilepsy (eds Shorvon, S., Fish, D. Andermann, F., Bydder, G. M. & Stefan, H.) 263274 (Plenum, 1994). 30. Friston, K.J. et al. Statistical parametric maps in functional imaging: a general linear approach. Hum. Brain Mapp. 2, 189210 (1995).

nature neuroscience volume 1 no 5 september 1998

427

Anda mungkin juga menyukai