Trends in Neurosciences - October 2013

October 2013 Volume 36, Number 10 pp.
555620
Editor Andrew M. Clark Executive Editor, Neuroscience Katja Brose Journal Manager Olaf Meesters Journal Administrators Ria Otten Patrick Scheffmann Advisory Editorial Board Silvia Arber, Basel, Switzerland Anders Bjrklund, Lund, Sweden J. Paul Bolam, Oxford, UK Sarah W. Bottjer, Los Angeles, CA, USA Barry J. Dickson, Vienna, Austria Chris Q. Doe, Eugene, OR, USA Craig C. Garner, Stanford, CA, USA Yukiko Goda, Wako, Japan Kenneth D. Harris, London, UK Nancy Ip, Kowloon, Hong Kong Meyer Jackson, Madison, WI, USA Maria Karayiorgou, New York, NY, USA Robert C. Malenka, Stanford, CA, USA Mark P. Mattson, Baltimore, MD, USA Freda D. Miller, Toronto, Canada Richard G.M. Morris, Edinburgh, UK Maiken Nedergaard, Valhalla, NY, USA Eric Nestler, New York, NY, USA Hitoshi Sakano, Tokyo, Japan Greg Stuart, Canberra, Australia Wolf Singer, Frankfurt, Germany Marc Tessier-Lavigne, New York, NY, USA Chris A. Walsh, Boston, MA, USA Editorial Enquiries Trends in Neurosciences Cell Press 600 Technology Square Cambridge, MA 02139, USA Tel: +1 617 3972823 Fax: +1 617 3972810 E-mail: tins@cell.com
Editorial
555
Looking back and looking forward

Science & Society
Andrew M. Clark
557
The Brain Prize 2013: the optogenetics revolution

Opinions
Andreas Reiner and Ehud Y. Isacoff
561
Molecular nexopathies: a new paradigm of neurodegenerative disease A developmental ontology for the mammalian brain based on the prosomeric model Challenges of understanding brain function by selective modulation of neuronal subpopulations
Reviews
Jason D. Warren, Jonathan D. Rohrer, Jonathan M. Schott, Nick C. Fox, John Hardy, and Martin N. Rossor Luis Puelles, Megan Harrison, George Paxinos, and Charles Watson
570
579
Arvind Kumar, Ioannis Vlachos, Ad Aertsen, and Clemens Boucsein
587
Sugar for the brain: the role of glucose in physiological and pathological brain function Control of neuronal voltage-gated calcium ion channels from RNA to protein Blurring the boundaries: developmental and activity-dependent determinants of neural circuits
Philipp Mergenthaler, Ute Lindauer, Gerald A. Dienel, and Andreas Meisel
598 610
Diane Lipscombe, Summer E. Allen, and Cecilia P. Toro Verena Wolfram and Richard A. Baines
Cover: The mammalian brain expends tremendous energy for its physiological function. On pages 587597 of this issue, Mergenthaler, Lindauer, Dienel, and Meisel review the tight regulation of glucose metabolism as the foundation for normal brain function, as well as the role of disturbed glucose metabolism in the pathophysiology of diverse brain disorders. The sweets and the cupcake-like brain on the cover signify the reliance of the brain on sugar-derived energy, and the background images, including the angel from Albrecht Drers Apocalypsis cum figuris, motifs from Jules Vernes adventure novels, and Daniela Nickaus explorer, illustrate the infinite intellectual and imaginative capacity of the brain. Cover design by Malte Nickau (www.graco-berlin.de).
Editorial
Looking back and looking forward

Andrew M. Clark
Trends in Neurosciences, Cell Press, 600 Technology Square, Cambridge, MA 02139, USA
Trends in Neurosciences (TiNS) was launched over 35 years ago with the aim of publishing insightful reviews, opinions, and commentaries covering all disciplines of the neurosciences. TiNS started strong; among the articles appearing in the journal that rst year include those penned by a Nobel Laureate, previous and future chairs of the Society for Neuroscience, and numerous other leading experts. Throughout its ensuing tenure, TINS has continued to publish timely and authoritative pieces with the goal of synthesizing, interpreting, and drawing novel insights into an increasingly fractured and complex literature. Although the neuroscientists toolkit has greatly expanded over recent years, the subjects covered remain surprisingly consistent: neuropsychiatric diseases, cellular and molecular studies of synapse development and plasticity, and systems investigations of learning and memory, to name but a few. The success of TINS in providing highquality coverage of such topics is reected in its consistent ranking among the best reviews journals in what has grown from a handful of similar titles at its inception to a plethora today. Despite its long run, TiNS, until now, has had only four editors, in chronological order: David Bouseld, Gavin Swanson, Sian Lewis, and Rachel Jurd. Together, these previous editors oversaw the launch and growth of the journal, shepherded it from print to primarily electronic distribution in the internet age, developed new means for engaging and building a community of readers, and commissioned countless articles that, one hopes, not only reviewed, but also helped set trends in the eld. I am humbled to follow in their footsteps and I aim to continue in the tradition that they started; this latest issue reects both the breadth of coverage and the high standard of quality that I aim to maintain at TINS. The neurosciences are one of the preeminent elds in biology today, as reected in both the amount of resources devoted to investigations of the nervous system and the increasing number of major prizes being awarded for breakthroughs in this eld. In this issue, in a Science and Society article, Reiner and Isacoff highlight the work that led to the recent receipt of one of these signicant awards, the Brain Prize 2013, which was awarded to Ernst Bomberg, Edward Boyden, Karl Deisseroth, Peter Hegemann, Gero Miesenbock, and Georg Nagel for their development and renement of optogenetics. Science and Society pieces are intended to spotlight subjects of wide interest, or topics at the intersection of the bench and the wider world. Readers can look forward to continuing to see more of both types in the future.
Corresponding author: Clark, A.M. (tins@cell.com).
One of the key features of TiNS opinion articles is that they present a personal viewpoint on a particular subject, thus advancing a novel perspective or hypothesis. In this issue, three such opinion articles present novel perspectives on important issues in neurodegenerative diseases, neural development, and neural coding, respectively. Warren, Rohrer, Scott, Fox, Hardy, and Rossor hypothesize that specic neural networks are differentially susceptible to particular pathogenic proteins and propagating protein abnormalities, while Puelles, Harrison, Paxinos, and Watson promulgate a hierarchical classication of brain structures based upon recent ndings concerning differential gene expression during development. Finally, Kumar, Vlachos, Aersten and Boucsein note that the parallel, recurrent architecture of most real neural networks limits the utility of selectively inactivating only particular elements (e.g., a given cell class) one-by-one, suggesting careful computational consideration of network structure will be a necessary precondition of experiments seeking to unravel network function. TiNS readers can expect to continue to see creative, novel, and insightful opinion articles covering other equally important topics in the future. Brief, pointed reviews that go beyond simply reciting the literature, to integrate and synthesize recent results into alternative interpretations, and to suggest productive areas for future research are the hallmark of TiNS. In this issue, one can nd three such reviews from leading groups in molecular, cellular, and systems neuroscience. Mergenthaler, Lindauer, Dienel, and Meisel review the role of glucose in fueling brain function and the role of breakdowns in this process in neurological disorders, while Lipscombe, Allen, and Toro cover the mechanisms controlling the expression of neuronal voltage-gated calcium channels (CaV). Finally, Wolfram and Baines delve into recent ndings suggesting neurotransmitter phenotype and expression of specic ion channels in neurons are not always hard and soft wired, respectively. In some form or another, all of my predecessors have expressed in this space the statement that these are exciting times to be a neuroscientist. That such a statement can be repeated, in all earnestness, over such a long period of time speaks to what a fascinating, dynamic, and exciting endeavor the study of the nervous system is. Currently, the eld is faced with many challenges. According to the World Health Organization (WHO), the socioeconomic burdens of psychiatric and neurological diseases are only expected to continue to increase [1]. Meanwhile, large scientic funding bodies in many countries continue to see their budgets decline in real or ination-adjusted terms. However, there are also numerous possibilities and potentials for great discoveries. The ongoing development of methods for controlling the activity of identied neuronal subtypes
555
0166-2236/$ see front matter 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.tins.2013.09.002 Trends in Neurosciences, October 2013, Vol. 36, No. 10
Editorial
on a fast timescale offers the potential to dissect circuit function in exquisite detail, next-generation sequencing offers the hope to better understand the genetic basis of neurodegenerative and neuropsychiatric diseases, and new computational tools for mining large and complex data sets are constantly being rened. As always, the journal welcomes feedback on our effort to cover this exciting ground; it would be impossible to identify and report on all the newest trends without invaluable input from our Advisory Editorial Board, authors, and readers. In teaming with our colleagues at other Cell Press journals, be they other Trends titles such as Trends in Cognitive Sciences, Trends in Endocrinology and Metabolism, or Trends in Pharmacological Sciences,
Trends in Neurosciences October 2013, Vol. 36, No. 10
or premier research journals, such as Neuron and Cell, to provide up-to-date and authoritative coverage of the neurosciences, TiNS hopes to be able to serve the increasingly diverse needs of this broad community. I will be at the upcoming Society for Neuroscience conference in San Diego, please stop by the Cell Press/Elsevier booth to chat and pick up a free copy of TiNS, or stop me in a poster aisle or at a symposium, to let me know your thoughts on how the journal can best continue to strive towards achieving this goal.
Reference
1 World Health Organization (2011) Mental Health Atlas 2011, WHO
556
Science & Society
The Brain Prize 2013: the optogenetics revolution

Andreas Reiner1 and Ehud Y. Isacoff1,2
1
Department of Molecular and Cell Biology and Helen Wills Neuroscience Institute, University of California Berkeley, Berkeley, California 94720, USA 2 Physical Bioscience Division, Lawrence Berkeley National Laboratory, Berkeley, California 94720, USA
The 2013 Grete Lundbeck European Brain Research Prize was awarded to Ernst Bamberg, Edward Boyden, Karl ck, and Deisseroth, Peter Hegemann, Gero Miesenbo Georg Nagel for their invention and renement of optogenetics. Why optogenetics? And why this sextet? To appreciate why, we turn rst to some of the core questions of neuroscience and the technical difculties that long obstructed their resolution.
The beauty and the agony of brain circuits How does the brain process sensory information, drive behavior, mediate perception, thought, and memory? Since n y Cajal used the the late 1880s, when Santiago Ramo mysterious tendency of the Golgi stain to label only a scattershot of neurons, and could thereby visualize the shapes of individual cells and trace their connections, neuroscientists have recognized that an essential part of the explanation lies in the neuronal circuitry. Modern morphological approaches allow the staining of entire cell populations and visualization of the synaptic connections therein. However, although these modern techniques, like Cajals work originally, help in providing the physical wiring diagram of the neural computer, they also illuminate the difculty of the neuroscience enterprise. They demonstrate that every part of the nervous system contains a multitude of neurons with distinct shapes, axonal and dendritic arbors, and contact partners. The challenge then is not only to nd a way to determine who is actually connected to whom, but how these individual synapses transmit information, and how the many types of neurons and connections give rise to functional networks that perform diverse tasks. Over the hundred years following Cajal the standard approach to mapping this activity was to identify the neurons active during specic sensory, motor, or cognitive events, with the assumption that those neurons are directly involved in processing the relevant information. In these studies, electrical signatures and cell location provided some information about neuron type, but available approaches could not discern the enormous variety of cell types which was becoming evident from emerging genetic analysis. In addition, the eld lacked a method
Corresponding author: Isacoff, E.Y. (ehud@berkeley.edu).
for selectively stimulating or inhibiting activity in specic cell types or transmission at specic synaptic connections. The solution for discerning cell types and selectively detecting their activity over an extended area came in the form of genetically encoded uorescent markers and optical reporters of action potential ring. This came from the discovery of green uorescent protein (GFP) as well as its subsequent cloning and genetic engineering to generate many color variants and to make optical reporters of cell activity, work which was awarded the 2008 Nobel Prize in Chemistry. A major remaining challenge has been to develop the means to manipulate activity in the nervous system with the spatial, temporal, and cell-type resolution that is provided by the optical detection of activity using uorescent protein reporters. The 2013 Brain Prize was awarded this year for the solution to this challenge (Figure 1). Light as a force for exciting and inhibiting neuronal activity Having seen the power of light for visualization of neural ring across a distributed network, one wonders if the ability to focus light might also be used for pinpoint stimulation of neurons. Since 1970 efforts were made to directly photostimulate neurons with light. The development of light-sensitive protecting groups and their application to photo-uncaging of ATP inspired the development of photolabile calcium chelators by Roger Tsien, Bob Zucker, and Graham Ellis-Davis, and of caged neurotransmitters, starting in the group of Peter Hess. These methods could be used to stimulate specic cells that were genetically identied by the selective expression of a uorescent protein reporter, providing the desired cell type specicity. However, they are best suited to stimulation of one cell at a time, a limitation when one considers the great power both for circuit mapping and behavioral analysis of an approach that would enable stimulation or inhibition of multiple cells of a genetically select type. To achieve such a goal, a genetically encoded system that can be reversibly controlled by light seemed ideal. Where can one turn to look for light-sensitive proteins? The obvious place is in the vertebrate eye to the natural worlds photoreceptor proteins, the opsins. But rhodopsin and its cone cousins are G protein-coupled receptors that signal to ion channels in the complex biochemical and cellular context of the photoreceptor cell. There are several factors which prohibit mammalian opsins from being attractive candidates. However, the photoreceptor system in ck the Drosophila eye is simpler. In 2002 Gero Miesenbo
557
Science & Society
TRENDS in Neurosciences
Figure 1. The award ceremony for the Brain Prize 2013. From the left: Ernst Bamberg, Edward Boyden, Karl Deisseroth, Peter Hegemann, Gero Miesenbo ck, Georg Nagel, Crown Prince Frederick, and former chairman of the board of the Grete Lundbeck Foundation Nils Axelsen. Reproduced with permission from the Grete Lundbeck Foundation.
and Boris Zemelman [1] made a critical conceptual and experimental step by expressing a trio of Drosophila proteins (arrestin-2, rhodopsin, and its cognate Ga subunit: dubbed chARGe) in cultured hippocampal neurons, enabling them to be excited by light. If one had to identify the paper that launched the thousand ships of optogenetics, this is it. Beautiful as chARGe was, it had two shortcomings: excitation by light was modest in strength and the approach could not be easily transferred to other systems because of the difculty of carrying out triple transfection. ck combined photo-uncaging The following year, Miesenbo of ATP or capsaicin with the heterologous cell-specic expression of either the excitatory P2X or TrpV1 receptor channels [2]. This approach also had limitations: It was conned for use in animals that lack receptors for ATP and capsaicin, in other words, it could not be used in vertebrates. At roughly the same time, Ed Callaways lab developed the reverse logic, expressing the Drosophila allatostatin receptor in specic mammalian neurons and activating it with allatostatin, which has no mammalian receptor, to open G protein-gated inwardly rectifying K+ channels (GIRKs), and thereby inhibit action potential ring. But allatostatin, like engineered GPCRs that are activated only by synthetic ligands (e.g., the RASSLs described by Bruce Conklin in 1998), had no light-activated variant, and thus lacked the rapid switching and spatial precision that is enabled by optical control. Was there, then, a simpler opsin to which one could cks conceptual turn, allowing for exploitation of Miesenbo breakthrough, but with fewer components, faster speed, and sufcient strength? The answer is yes. Opsins from
558
bacteria and single-cell algae had been studied for decades. But they were known to be different from their relatives in the eye and were so foreign phylogenetically that one wondered if they could ever be imported in a functional form into the mammalian brain. Starting in the lab of Oesterhelt in the mid-1980s, Ernst Bamberg and Peter Hegemann worked on two microbial opsins from Halobacterium salinarium: bacteriorhodopsin (bR) and halorhodopsin (hR). These opsins, although related to the seven transmembrane helix G protein-coupled receptor opsins of the eye, are self-contained light-driven ion pumps. In response to light, bacteriorhodopsin pumps H+ and halorhodopsin pumps Cl across the membrane, thereby changing the ionic balance and membrane potential. Like rhodopsin, bR and hR use retinal as a chromophore. In the late 1980s Hegemann branched out and characterized the phototaxis in the single-cell algae Chlamydomonas reinhardtii, which appeared to depend on another rhodopsin relative. The work culminated in two seminal publications in 2002 and 2003 from a joint effort of the Bamberg and Hegemann labs, both with Georg Nagel as a rst author, which described the cloning, heterologous expression, and functional characterization in Xenopus oocytes of these opsins, which turned out to be cationselective ion channels that were named channelrhodopsin 1 and 2 (ChR1 and ChR2) [3,4]. Importantly, Nagel et al. demonstrated that ChR2 could be expressed to achieve light-induced depolarization in non-neuronal mammalian cell lines. Here one had, it seemed, the perfect and complementary set of genetically encoded light-driven pumps and channels. But could they be used in neurons? The reputation
Science & Society

of microbial proteins was that they are notoriously difcult to express, and they normally operate in extreme ionic environments very different from those of the central nervous system. Indeed, for various reasons bR and hR from Halobacterium salinarium and ChR1 were never successfully transferred to neurons. But in 2005 Karl Deisseroth and Ed Boyden, in a collaboration with Nagel and Bamberg, and Stephan Herlitzes lab in collaboration with Hegemann and Lynn Landmesser, successfully expressed ChR2 in mammalian neurons and demonstrated robust ring of action potentials in response to blue light [5,6]. In his study, Herlitze was also able to inhibit neuronal ring using vertebrate rhodopsin, coupling it to GIRKs or P/Q-type Ca2+ channels, and performed the rst in vivo experiments in intact spinal cord and living chicken embryo [6]. In the same year, Alexander Gottschalk, in collaboration with Nagel and Bamberg, succeeded in optically altering the behavior of the worm Caenorhabditis elegans [7], Hiromu Yawo demonstrated optical control of ring in acute hippocampal slice following sterotatic injection of ChR2-coding viruses in mouse [8], and Zhuo-Hua Pan demonstrated functional ChR2 expression in the mouse retina [9]. In 2007 both Deisseroth, continuing the collaboration with Nagel and Bamberg, and Boyden used a halorhodopsin from another archaeon, Natronomonas pharaonis, which lives in soda lakes in Egypt, to pump Cl ions into neurons in response to light of a different wavelength from that used to activate ChR2 and thereby to hyperpolarize neurons and inhibit ring [10,11]. Later, Boyden added pumps which extrude H+ to hyperpolarize neurons. In parallel to the development of opsins to control neuronal ring, chemical optogenetics was developed, thus enabling direct light-control of the signaling proteins of the mammalian synapse. This approach was inspired by foundational studies by Bernard Erlanger and coworkers in the 1970s, who developed synthetic photoswitchable tethered and diffusible ligands for muscle nicotinic receptors. The rst such control of neural activity, reported in 2004, was a light-blocked K+ channel, SPARK, which could be used to inhibit neuronal ring [12]. This was followed by an excitatory light-activated ionotropic kainate-type glutamate receptor, LiGluR [13], an inhibitory version of LiGluR known as HyLighter, light-activated and lightblocked neuronal nicotinic acetylcholine receptors [14], and light-activated and light-blocked G protein-coupled metabotropic glutamate receptors that inhibit spiking and neurotransmitter release [15]. Although these lightgated channels and neurotransmitter receptors can be used in the manner of the opsins to excite and inhibit neuronal ring, and to inject calcium into neurons or glia, their unique promise is for gating synaptic transmission and plasticity. Expansion of tools, explosion of neuroscience Since 2005, when ChR2 was rst used to excite neurons in response to light, efforts in the eld, led by the sextet, culminated in the isolation of new microbial opsins and opsin engineering to generate excitatory and inhibitory versions with larger currents and different ion selectivity, kinetics, and excitation spectra. As envisioned, light
as a minimally invasive stimulus allowed exquisite spatiotemporal patterning of neuronal activity, whereas the combination with elaborate genetic methods allowed a specicity of targeting to cell type that was never possible before. Soon, these methods were taken from cellular studies in vitro to behavioral studies in awake animals. The spread of optogenetics with microbial opsins has been nothing short of spectacular, a product of the combined power of the method and its simplicity of use. Because retinal is bioavailable in the vertebrate nervous system and because, unlike in visual rhodopsins, it binds irreversibly to the microbial opsins, the user is only a transfection, a viral infection, or genetic cross away from an experiment. Consequently, the opsins have been deployed across a wide range of organisms, from worm to y to sh to mouse to non-human primates. They have been used to study brain waves, sleep, memory, hunger, addiction, aggression, courtship, sensory modalities, and motor behavior. They have also been used as models for clinical treatment: for deep brain stimulation, on one hand, and as a retinal prosthetic on the other. This revolution led to hundreds of papers a veritable deluge that even swept away entertaining qualms of ck about the optogenetic catechism. The simMiesenbo plicity of the term and the power of the methodology have conquered all. There is little question in the neuroscience community that optogenetics deserves awards of the magnitude of the 2013 Brain Prize. It is a great thing that the prize has recognized the people who launched the eld both those whose basic science on opsins paved the way, as well as those who engineered them into the nervous system.
Acknowledgments
This work was supported by the National Institutes of Health (NIH) Nanomedicine Development Center for the Optical Control of Biological Function (PN2EY018241).
References
1 Zemelman, B.V. et al. (2002) Selective photostimulation of genetically chARGed neurons. Neuron 33, 1522 2 Zemelman, B.V. et al. (2003) Photochemical gating of heterologous ion channels: remote control over genetically designated populations of neurons. Proc. Natl. Acad. Sci. U.S.A. 100, 13521357 3 Nagel, G. et al. (2002) Channelrhodopsin-1: a light-gated proton channel in green algae. Science 296, 23952398 4 Nagel, G. et al. (2003) Channelrhodopsin-2, a directly light-gated cation-selective membrane channel. Proc. Natl. Acad. Sci. U.S.A. 100, 1394013945 5 Boyden, E.S. et al. (2005) Millisecond-timescale, genetically targeted optical control of neural activity. Nat. Neurosci. 8, 1263 1268 6 Li, X. et al. (2005) Fast noninvasive activation and inhibition of neural and network activity by vertebrate rhodopsin and green algae channelrhodopsin. Proc. Natl. Acad. Sci. U.S.A. 102, 1781617821 7 Nagel, G. et al. (2005) Light activation of channelrhodopsin-2 in excitable cells of Caenorhabditis elegans triggers rapid behavioral responses. Curr. Biol. 15, 22792284 8 Ishizuka, T. et al. (2006) Kinetic evaluation of photosensitivity in genetically engineered neurons expressing green algae light-gated channels. Neurosci. Res. 54, 8594 9 Bi, A. et al. (2006) Ectopic expression of a microbial-type rhodopsin restores visual responses in mice with photoreceptor degeneration. Neuron 50, 2333
559
Science & Society

10 Han, X. and Boyden, E.S. (2007) Multiple-color optical activation, silencing, and desynchronization of neural activity, with single-spike temporal resolution. PLoS ONE 2, e299 11 Zhang, F. et al. (2007) Multimodal fast optical interrogation of neural circuitry. Nature 446, 633639 12 Banghart, M. et al. (2004) Light-activated ion channels for remote control of neuronal ring. Nat. Neurosci. 7, 13811386 13 Volgraf, M. et al. (2006) Allosteric control of an ionotropic glutamate receptor with an optical switch. Nat. Chem. Biol. 2, 4752

14 Tochitsky, I. et al. (2012) Optochemical control of genetically engineered neuronal nicotinic acetylcholine receptors. Nat. Chem. 4, 105111 15 Levitz, J. et al. (2013) Optical control of metabotropic glutamate receptors. Nat. Neurosci. 16, 507516
0166-2236/$ see front matter 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.tins.2013.08.005 Trends in Neurosciences, October 2013, Vol. 36, No. 10
560
Opinion
Molecular nexopathies: a new paradigm of neurodegenerative disease

Jason D. Warren1, Jonathan D. Rohrer1, Jonathan M. Schott1, Nick C. Fox1, John Hardy2, and Martin N. Rossor1
1
Dementia Research Centre, Department of Neurodegenerative Disease, UCL Institute of Neurology, University College London, London, UK 2 Reta Lilla Weston Laboratories and Department of Molecular Neuroscience, UCL Institute of Neurology, University College London, London, UK
Neural networks provide candidate substrates for the spread of proteinopathies causing neurodegeneration, and emerging data suggest that macroscopic signatures of network disintegration differentiate diseases. However, how do protein abnormalities produce network signatures? The answer may lie with molecular nexopathies: specic, coherent conjunctions of pathogenic protein and intrinsic network characteristics that dene network signatures of neurodegenerative pathologies. Key features of the paradigm that we propose here include differential intrinsic network vulnerability to propagating protein abnormalities, in part reecting developmental structural and functional factors; differential vulnerability of neural connection types (e.g., clustered versus distributed connections) to particular pathogenic proteins; and differential impact of molecular effects (e.g., toxic-gain-of-function versus loss-of-function) on gradients of network damage. The paradigm has implications for understanding and predicting neurodegenerative disease biology. Introduction Neural networks are a key theme in contemporary neuroscience [13]. Operationally, a neural network can be dened as a complex system comprising nodes and links represented by neurons and their connections [4]. Neural networks extend over scales ranging from microscopic (neurons and synapses) to macroscopic (anatomical regions and bre tracts), and may be structural (dened by physical connections; e.g., bre tracts) or functional (dened by physiological connections). Neuroimaging techniques, such as functional MRI (fMRI) and diffusion tensor tractography [4,5], coupled with methodologies such as graph theory [2,6], have delineated intrinsic, distributed neural networks supporting cognitive functions in the healthy brain [79]. Neural network models have been successfully applied to common neurodegenerative syndromes [35,715], building
Corresponding author: Warren, J.D. (jason.warren@ucl.ac.uk). Keywords: neurodegeneration; dementia; neural network; nexopathy. 0166-2236/$ see front matter 2013 Published by Elsevier Ltd. http://dx.doi.org/10.1016/j.tins.2013.06.007
on the key insight that neurodegenerative diseases, such as Alzheimers disease (AD) and frontotemporal lobar degeneration (FTLD), produce distinctive clinical syndromes with regular patterns of evolution due to the spread of pathogenic protein abnormalities via large-scale brain networks. To date, work in neurodegenerative disease has mainly focussed on linking clinical phenotypes to network alterations. However, it remains unclear how molecular (protein) abnormalities translate to network damage and, thus, clinical phenotypes; and whether pathological substrates can be predicted reliably from macroscopic network signatures. Recent advances in genetics and histopathology have enabled the detailed mapping of neurodegenerative clinico-anatomical phenotypes onto specic proteinopathies, transcending broad categories such as tauopathy or ubiquitinopathy. Histopathological patterns of protein deposition reect underlying molecular (biochemical or conformational) characteristics in a range of neurodegenerative diseases, most strikingly in the FTLD spectrum [1624]. Although the concept requires further substantiation and qualication, the diffusive intercellular or prionlike spread of pathogenic misfolded proteins holds promise as a general mechanism for the evolution of the neurodegenerative process in a wide range of diseases [8,9,2528]. Various candidate mechanisms that might link protein pathophysiology with intercellular miscommunication and local circuit disruption have been identied [3,29,30]. However, the mechanisms that translate local effects of proteinopathies to specic patterns of large-scale network disintegration remain largely unknown. Here, we address this problem. We propose the term molecular nexopathy (Latin nectere, tie) to refer to a coherent conjunction of pathogenic protein and intrinsic neural network characteristics expressed as a macroanatomical signature of brain network disintegration. We argue that improved understanding of the molecular mechanisms of network disintegration will constitute a new paradigm of neurodegenerative disease. The essential features of the paradigm that we propose are presented in Box 1 and Figures 13. We now consider potential mechanisms whereby molecular dysfunction might be linked to neural circuit disruption. We then assess the extent to which the molecular
Trends in Neurosciences, October 2013, Vol. 36, No. 10
561
Opinion
Box 1. The molecular nexopathy paradigm: key substrates and constraints
Molecular, microstructural, and functional substrates Pathogenic proteins, including misfolded aggregates and toxic oligomers, could promote neural network disintegration by disrupting synaptic function or maintenance [57], axonal transport or repair [79], or via downstream trophic [58,80] or cellcell signalling abnormalities [3,10,29,62,8185]. Disease effects would exploit intrinsic network vulnerabilities: in particular, developmental patterns of protein expression and structural and functional interactions across neural circuits [38,39] (Figure 1, main text). Shorter-range or clustered dendritic and interneuronal connections appear particularly vulnerable to some tauopathies [48,51], whereas longer-range, more widely distributed (axonal) projections may be relatively more vulnerable to other molecular insults (e.g., toxic oligomers derived from amyloid precursor protein [58] or deficiency of trophic or protective factors such as GRN [57]). The balance between molecular net toxic gain-of-function and loss-of-function effects [57,62] might help to determine the extent to which proteinopathies exert uniform or graded effects across neural circuits (Figure 1, main text). Softwired alterations in synaptic function, in part reflecting pervasive network activity patterns, might give way to subsequent (irreversible) hard-wired structural damage and cell loss. Disease propagation Transcellular and, in particular, synaptically mediated diffusion of misfolded proteins and permissive templating of protein misfolding appear to be general principles of disease evolution across a range of neurodegenerative proteinopathies [26,86], including prion [85], betaamyloid [3133], tau [8790], alpha-synuclein [91,92], TDP-43 [71,72], and superoxide dismutase 1 [27]. Inoculation with beta-amyloid and
tau triggers uptake by cells and templated conformational alteration of normally folded protein to a potentially pathogenic, misfolded state [26,86]. Trans-synaptic spread of tau-containing tangle pathology from entorhinal cortex neurons occurs in transgenic mice that only express mutant MAPT in those neurons [89,90]. Macroanatomical signatures Particular structural and functional neuroimaging profiles of network disintegration have specific molecular associations [7,16,17,2023,68,69,74,75,9395] (Figure 2, main text). In the FTLD spectrum, structural neuroimaging evidence suggests a scheme for partitioning pathologies according to whether the macroanatomical brain atrophy profile is localised or distributed, and whether atrophy is relatively symmetric or strongly asymmetric between the cerebral hemispheres [18,23] (Figure 2, main text). In the case of AD, early (even presymptomatic) functional and structural alterations occur within a specific distributed parieto temporofrontal network mediating default mode processing or stimulus-independent thought in healthy individuals [4,11] and network disintegration tracks pathological disease staging [11,14,96]. Variant AD phenotypes, such as posterior cortical atrophy and logopenic progressive aphasia, may reflect differential involvement of corticocortical projection zones that together constitute a distributed AD-vulnerable network [15,97], although the precise pathophysiological roles of the two major candidate pathogenic proteins (beta-amyloid and phosphorylated tau) and the factors that modulate the profile of network damage remain contentious [4,11,15]. Convergence of syndromes as disease evolves may hold a solution to the problem of phenotypic heterogeneity (Figure 3, main text).
nexopathy paradigm can be reconciled with the central problems of disease evolution and phenotypic heterogeneity, and propose experimental tests of the paradigm in future work. How do pathogenic molecules produce specic brain network disintegration? Networks show variable intrinsic vulnerability to proteinopathies The molecular nexopathy paradigm makes no assumptions about the instigating event that triggers the neurodegenerative process, which might be stochastic but which results in the creation of a potentially pathogenic molecule. However, once initiated, the topography of neurodegeneration preferentially targets network elements that are vulnerable to the instigating molecular species (Figure 1). Emerging evidence, including inoculation experiments in animals [3133] (Box 1), implies that a neurodegenerative process may home to a brain region (or regions) based on intrinsic vulnerability to the pathogenic protein. Neurodegeneration may propagate by prion-like seeding or templating of the protein abnormality (e.g., conformational misfolding) across neural connections, in addition to physical transfer of instigating pathogenic proteins. The presence of a specic pathogenic abnormality that propagates across a network would distinguish a neurodegenerative molecular nexopathy from other diseases that disrupt brain networks (for example, stroke and traumatic brain injury): one important corollary is that compensatory or homeostatic responses are ultimately inadequate in neurodegenerative nexopathies. Regional neural vulnerability to a proteinopathy could reect anatomically restricted expression of the culprit protein by cell populations or additional epigenetic factors
562
that direct the expression of the protein to particular brain areas [3436] or determine neuronal susceptibility to toxic events [10]. Regional differentiation of protein expression is a fundamental feature of normal brain development [37], establishing intrinsic specicities of connections within neural circuits [38,39] and thereby in turn directing the function of the circuit. Therefore, local proles of protein expression could confer selective vulnerability or resistance of particular network elements to particular neurodegenerative diseases, and would also help drive the functional phenotypic signature of the disease. For example, differential expression of neuroprotective factors has been linked to the relative vulnerability of particular neuronal populations in the basal ganglia in Parkinsons disease [40], and regional expression of genes involved in inammatory signalling may modulate disease onset with progranulin (GRN) mutations [41]. By contrast, epigenetic effects during brain development alter the regulation and expression of amyloid precursor protein and potentially inuence the later development of AD [42]. Brain areas that are highly neuroplastic, more specialised, or phylogenetically more recent (for example, the language system) may be relatively more vulnerable to proteinopathies [43], whereas primary motor and sensory cortex both show relative resistance. The process of neurodegeneration might selectively unravel the sequence of normal ontogeny within the vulnerable network (for example, by withdrawing essential trophic support, repair mechanisms, or physiological signalling across damaged synaptic connections) [29]. Regional specicity might also arise from cellular morphological factors, particularly at synaptic connections: even if a protein is widely expressed in the brain, its effects
Opinion
(A) Clustered no gradient
N6 I N1 N2 H N4 N3 N5 I N1 N2 H N4 N3
(B)
Clustered gradient
N6 N5
MAPT
(A)
TDPC
(B)
(C) Distributed no gradient

N6 I N1 N2 H N4 N3 N5
(D)
Distributed gradient
N6
CBD
(C)
I N1 N2 H N4
N5
GRN
(D)
N3
Figure 1. A taxonomy of molecular nexopathy mechanisms. Here, we model putative templates of neurodegenerative network damage at a given arbitrary time point following introduction of a pathogenic protein. In each panel, the stylised local neural circuit comprises nodes (N; e.g., neuronal somas) with links (e.g., axons or dendrites) that behave as either shorter-range clustered (unfilled lines) or longer-range distributed (filled lines) connections; the most highly connected node behaves as a local hub (H), whereas node I is the site of an instigating insult (wavy arrow) associated with a pathogenic protein. Grey symbols represent unaffected or minimally affected network elements; pathogenic effects are coded in red (filled circles representing deleted network elements and half-tone circles representing dysfunctional network elements) and pathogenic effects are assumed to be potentially bidirectional across network connections. The taxonomy shown assumes two basic, interacting dichotomies arising from the conjunction of pathogenic protein and intrinsic network characteristics: selective targeting of shorter-range clustered neural connections [e.g., IN1N2HN3N4 N5 (A,B)] versus longer-range distributed neural connections [e.g., I-H-N5-N6, (C,D)]; and effects that are relatively uniform [no gradient, (A,C)] or strongly graded [gradient (B,D)] across the network. In each case, the hub H is intrinsically relatively more vulnerable due to its high connectedness with the rest of the network [60]. Targeting of connection types could reflect subcellular compartmentalisation of pathogenic proteins and/or local synaptic properties or other morphological characteristics (e.g., targeting of dendritic versus axonal compartments). Gradients of effects could be established by intrinsic polarities in network protein expression and/or the net functional effect of a pathogenic molecular cascade (e.g., uniform toxic gain-of-function versus graded loss-of-function effects). The model that we propose requires propagation of disease effects across network elements, but does not specify the precise nature of those effects: for example, propagation could occur by direct protein transfer, protein seeding or templating in contiguous elements, or deleterious pathophysiological signalling, all potentially operating at different stages of disease evolution. The model predicts coherence between culprit molecule, neural connection types predominantly targeted, and functional (e.g., cognitive) phenotype as the neurodegenerative process scales to the level of the whole brain (Figure 2, main text).
may propagate only in particular cell types [44] or across specic patterns of connections [29]. Animal models have demonstrated exquisite microanatomical and biochemical specicity of intercellular connections in key vulnerable structures (such as hippocampus) [45,46]. Age-related neuronal resprouting may enhance local deposition of amyloid precursor protein in entorhinal cortex early during the course of AD [47]. Protein expression and morphological specicity at receptors and synapses would interact with subcellular molecular factors. For example, tau isoforms show distinctive subcellular distributions [48,49]. Deranged microtubular transport of abnormal tau facilitates accumulation of aggregated tau in somatic and dendritic rather than axonal compartments [50,51], whereas diffusible tau may further focus pathogenic effects of the protein on local synaptic and glial connections [51]. The role of glial elements is poorly understood, but may inuence local expression and development of network damage, reected
Figure 2. Scaling nexopathies to large-scale brain networks. The inset cartoon (left) shows a stylised axial view of the cerebral hemispheres in a normal brain. Circles represent neural network elements and colours code large-scale functional networks associated with generic clinical syndromes in previous connectivity work [7]: the anterior temporal lobe semantic network (green); the frontoinsular salience network implicated in behavioural variant frontotemporal dementia (yellow); the dominant hemisphere speech production network implicated in progressive nonfluent aphasia (magenta); the frontoparietal network associated with corticobasal syndrome (light blue); and the temporoparietal default mode network implicated in Alzheimers disease (dark blue). Putative shorter-range clustered (unfilled black lines) and longer-range distributed (filled black lines) connections between network elements are shown; connections between major functional networks are also represented (grey lines). The middle panels show proposed cross-sectional schemas of network breakdown (red, following Figure 1, main text), after an instigating insult (wavy arrow) associated with a pathogenic protein. Alongside each panel, axial and coronal MRI brain sections show corresponding observed atrophy profiles in patients with representative, canonical, pathologically confirmed, proteinopathies (CBD, corticobasal degeneration associated with 4-repeat tau pathology; GRN, mutation in progranulin gene; MAPT, mutation in microtubule-associated protein tau gene; TDPC, TDP-43 type C pathology [19], associated with the clinical syndrome of semantic dementia; the left hemisphere is on the left in all sections). These atrophy profiles illustrate macroanatomical scaling of the nexopathy templates proposed in Figure 1 (main text): (A) predominant involvement of clustered (shorter-range) connections with uniform extension, leading to relatively focal (temporal lobe) atrophy that is relatively symmetrically distributed between the cerebral hemispheres; (B) predominant involvement of clustered (shorter-range) connections with a strong gradient of network damage, leading to relatively focal (temporal lobe), strongly asymmetric atrophy; (C) predominant involvement of distributed (longer-range) connections with uniform extension, leading to distributed, relatively symmetrical atrophy; and (D) predominant involvement of distributed (longer-range) connections with a gradient of network damage, leading to distributed, strongly asymmetric atrophy. A particular proteinopathy here affects network connections with particular characteristics (e.g., clustered versus distributed synaptic linkages); functional networks will be targeted according to their specific network characteristics, but the effects of a particular nexopathy will in general tend to spread between functional networks, while continuing to target connections with similar properties across these networks. This would account for empirical variability in the closeness with which proteinopathies map onto particular functional networks (e.g., the mapping is relatively close for TDPC pathology with the semantic network, whereas most proteinopathies involve the salience network). The scheme makes specific predictions about the sequence of regional involvement with particular proteinopathies (e.g., sequential involvement of homologous contralateral temporal lobe regions with TDPC pathology) (see also Figure 3, main text).
macroscopically in the relative extent of grey matter versus white matter damage [52]. Proteinopathies are tted to neural circuits Central to the molecular nexopathy paradigm is the t between the pathogenic molecule and local neural circuit (and, ultimately, large-scale network) characteristics (Figures 1 and 2). Although data on specic local interactions of neural circuits with proteins remain limited, a
563
Opinion
t0 t1 t2
bvFTD
PNFA
CBS
Figure 3. Temporal evolution of disease and phenotypic heterogeneity. This schematic illustrates the application of the molecular nexopathy concept to the problem of phenotypic heterogeneity in neurodegenerative disease, using the example of corticobasal degeneration. The inset cartoon (left) shows major functional networks in a stylised normal dominant cerebral hemisphere, colour coded as in Figure 2 (main text). The panels illustrate evolving network involvement shortly after onset of the neurodegenerative insult (t0) and at two arbitrary later time points (t1 and t2). The initial location of the insult (wavy arrow) determines the clinical presentation (behavioural variant frontotemporal dementia, bvFTD; progressive nonfluent aphasia, PNFA; or classical corticobasal syndrome, CBS). The core corticobasal functional network (light blue in inset) is involved with disease evolution in each case; however, variable additional involvement of other contiguous functional networks (the salience network, speech production network or default mode network) modulates the phenotype. Each of these phenotypes arises from a common template of network involvement determined by the type of neural connection predominantly involved (here, represented as longer-range distributed intrahemispheric projections); the common nexopathy signature of corticobasal degeneration is revealed in the temporal profile of disease evolution.
proteinopathy might spread between brain regions by causing connected regions to develop the same intracellular protein abnormality or, less directly, by affecting the function of those connected regions [3]. The coupling between functional and structural connectivity in neurodegenerative diseases remains poorly dened: however, initial dysfunction could promote subsequent molecular alterations and destruction of network elements, as shown in lesion and tract-tracing studies in humans and nonhuman primates [5,6,53]. In more theoretical terms, the effects of a proteinopathy on a network might be regarded as a form of information ow, where information signies a change in network function (in particular, alterations in synaptic properties) associated with the introduction of the abnormal protein [3]. The characteristics of information exchange across articial neural networks have attracted considerable interest in computational neurobiology [2,54]: this theoretical framework might be adapted to the case of proteinopathies. Pathogenic molecular effects often have time constants that are much longer than those typically associated with information ow in neural networks. However, dynamic downstream alterations in synaptic function across circuits would occur over much shorter timescales. A further, potentially related, factor is the role of pervasive patterns of activity in circuit function and in predisposing networks to the effects of neurodegenerative disease [55]: examples include the differential and possibly usedependent susceptibility of particular motor pools to amyotrophic lateral sclerosis (ALS) [27], or the altered
564
trafcking of amyloid and tau in the isodendritic core associated with perturbations of the sleepwake cycle in AD [56]. Putative behavioural and activity-related factors are likely to interact with underlying genetic and epigenetic predisposing factors, and the causal sequence in general remains to be determined. An important theoretical motivation for applying the network information-processing framework to the neurodegenerative proteinopathies is the rich taxonomy of network activity patterns that follow from relatively simple starting assumptions when modelling the evolution of articial neural networks [54]. In particular, it has been shown in such articial networks that patterns of neural activity in local network elements can, under certain conditions, scale up to the entire network; that the characteristics of local microcircuits strongly inuence the activity pattern produced by the network as a whole; and, furthermore, that these patterns may be highly polarised. All these are network properties predicted in the case of the neurodegenerative proteinopathies, for which an activity pattern could be interpreted as the cumulative effect of protein-associated network damage integrated over time [55]. Relevant network and protein characteristics have yet to be worked out in detail for neurodegenerative proteinopathies. However, it has been shown that brain networks have small-world properties expressed as a high degree of clustering among network elements and short average path lengths between clusters [2]. These small-world properties suggest a basic dichotomy between shorter-range neural connections within clusters (local neural circuits, with relatively long neural path length) and relatively sparse longer-range neural connections (with relatively short neural path length) between clusters, in line with highly segregated and hierarchical brain network architectures observed empirically [6] and with limited evidence concerning the cellular pathophysiology of certain proteinopathies [48,51,57,58] (Box 1). This putative dichotomy between clustered and distributed network connections suggests a morphological basis for partitioning neurodegenerative diseases according to whether they produce relatively localised versus more distributed proles of macroscopic brain atrophy [18,20] (Figures 1 and 2). Clustered and distributed connections could be modelled in synthetic neural circuits and could be dened in brain networks using anatomical methods that can measure effective path lengths between network elements [6]; for example, dynamic causal modelling. Alternative morphological dichotomies might also operate: for example, selective targeting of excitatory versus inhibitory projections [44]. The degree of macroanatomical asymmetry of network damage within and between cerebral hemispheres appears to be a further partitioning characteristic of many neurodegenerative diseases (Figure 2), although to demonstrate interhemispheric asymmetries, it may be necessary to retain individual hemispheric asymmetry proles when pooling data in group neuroimaging studies [18,2022]. Network asymmetries could be determined, in part, by congurational features of the host network that might tend to polarise network activity. Such polarity has been demonstrated in a computational model of Huntingtons
Opinion
disease [44]. Asymmetries might be predicted based on extrapolation from network architectures in the brains of other species. For example, the nodes of large-scale networks in the macaque brain have a highly nonuniform distribution within the cortex and the connections between nodes are hierarchically organised [6]. Given that the human brain shares network homologies with that of the macaque [53], this architecture would tend to focus the effects of neurodegenerative disease at particular vulnerable hub regions (for example, in prefrontal cortex and posterior cingulateprecuneus [1,59]). Much more generally, it has been shown that highly connected network elements are intrinsically more vulnerable to extinction following perturbing events in a variety of hierarchical systems, ranging from ecology to economics [60]. In the context of neurodegenerative disease, extinction might be equated to destruction of highly connected network elements after introduction of a pathogenic protein and susceptibility from connectedness might, for example, predict disproportionate vulnerability of dominant hemisphere language hubs in the progressive aphasias [43] and medial parietal hubs binding the default mode network in AD [1,4,15,55,59]. If functional connections between brain regions are dened based on the strength and direction of spontaneous activity correlations, fMRI data suggest a fundamental dichotomy between positive connections that are dominant within a cerebral hemisphere versus negative connections that are dominant between hemispheres [61]: negative interhemispheric functional correlations will tend to establish intrinsically asymmetric interhemispheric interactions that could be exploited by neurodegenerative pathologies (Figure 2). It is unlikely a priori that any set of neuronal or neural network features would confer vulnerability uniquely to a single molecular species. However, further specicity in the prole of network involvement (in particular, whether strong polarity of damage is expressed across the brain) may be driven, in part, by functional characteristics of the pathogenic protein itself. Directional protein dysfunction drives network asymmetries Across the spectrum of potentially pathogenic proteins, there is a basic distinction between toxic-gain-of-function (deleterious effects of protein accumulation) and loss-offunction (impaired physiological, signalling or trophic) molecular effects [57,62]. The loss of function of a key protein is likely to lead ultimately to the loss of function of the affected network element and, therefore, might be regarded in computational terms as inhibiting the affected element; the net computational effect of a toxic gain of function is more difcult to predict. Large-scale network asymmetries (i.e., asymmetric macroscopic atrophy proles) might result from interaction of intrinsic connectivity structure with a gradient of molecular effects across the vulnerable network. We envisage that, within an affected network, an overall toxic gain of function will spread relatively uniformly, whereas an overall loss-of-function effect will establish a gradient of tissue loss due to attenuation of downstream
synaptic inputs. Such polarising network-level effects of loss-of-function proteinopathies would be in line with a net inhibitory action on damaged connections, because selective inhibition of network elements can generate highly polarised network structures and self-amplifying network activity patterns in computational models [54,61,63,64]. Proteinopathic effects would interact with (and may, in part, be driven by) intrinsic, ontogenetic network gradients [38,39]. Trophic effects modulate intercellular gradients in normal morphogenesis and developmental disorders [65] as well as in computational models [66]. Certain loss-offunction effects could become self amplifying due to additional, catastrophic mechanisms that might be specic to particular protein alterations: an example is GRN mutations, which may inhibit neuronal repair processes leading to accelerated collapse of network architecture [67]. Although it is unlikely that polarised protein effects operate in pure form in the brain [57,62], for a given disease process and disease stage, toxic gain-of-function or loss-of-function effects may dominate at the network level (Figure 1). Intracellularly, particular pathogenic proteins have complementary loss-of-function and toxic-gainof-function effects [62]. However, the overall primary balance of those effects across a neural network may depend on specic molecular actions at key network elements (e.g., synapses) that act as the nal common pathway for network damage. Additional specicity may be conferred by biochemical characteristics and conformational signatures of protein subtypes within broad categories, such as tau and Tar DNA-binding protein 43 (TDP-43) [24,49]. We currently lack such specic information for most key pathogenic proteins in the neurodegenerative spectrum [62]. There is further substantial potential for interactions among pathogenic proteins (for example, between tau and beta-amyloid in AD [28]). Protein-specic effects might modulate intrinsic network connectivity properties, contributing to phenotypic variation associated with particular proteins within a common network architecture [for example, the relatively symmetric atrophy prole associated with microtubule-associated protein tau (MAPT) mutations versus the strongly asymmetric prole associated with TDP-43 type C (TDPC) pathology [19] within anterior temporal lobe networks [18]]. Temporal evolution and the problem of heterogeneity A critical feature of neurodegenerative molecular nexopathies is likely to be their pattern of evolution in time as well as spatially within the brain. The rapidity of network breakdown might depend on the relative proportions of connection types affected by the pathological process, the predominant involvement of longer-range connections corresponding to rapid spread and involvement of clustered connections corresponding to slower spread, respectively. This would t with available data for certain neurodegenerative disorders. For example, patients with MAPT mutations and relatively focal anterior temporal lobe damage have, on average, slower rates of overall brain atrophy and survive substantially longer compared with patients with GRN mutations associated with widespread intrahemispheric damage [68]; interhemispheric asymmetry increases with advancing disease in association with
565
Opinion
GRN mutations [17]; but MAPT and GRN mutations produce similar local rates of atrophy within key structures such as the hippocampus [69]. Taken together, such evidence suggests that disease effects are preferentially amplied if long intrahemispheric bre tracts are implicated. The temporal evolution of atrophy proles associated with a particular proteinopathy may reveal a characteristic signature of network involvement that unites apparently disparate phenotypes (Figure 3). For example, tauopathies in the FTLD spectrum (such as corticobasal degeneration) may present with a behavioural syndrome due to frontal lobe involvement, with a language syndrome due to involvement of peri-Sylvian cortices in the dominant hemisphere, with a parietal lobe syndrome or with atypical parkinsonism: the nexopathy paradigm predicts phenotypic convergence over time due to progressive erosion of core frontoparietal, frontotemporal, or frontosubcortical networks implicated in particular tauopathies [18]. There is substantial evidence for such phenotypic convergence in the FTLD spectrum [18,70] and related overlap syndromes such as FTD-ALS [71]; however, precise correlations with
particular brain networks have yet to be widely established. Similarly, variant AD phenotypes have been interpreted as modulating a core temporoparietal-prefrontal default mode network [15]. Phenotypic convergence implies that initial stochastic insults anywhere in a vulnerable network will lead ultimately to a common signature of network breakdown, although the precise sequence of network involvement will tend to reect the initial locus of pathology within the network (Figure 3). This concept of the differential involvement of a core vulnerable network (with increasingly complete involvement of the network over time) suggests one possible solution to the apparent paradox of individual phenotypic variation associated with particular proteinopathies [70]. The clinico-anatomical expression of a given proteinopathy often varies between individuals as well as between syndromic subgroups [15,27,43]. The molecular nexopathy paradigm requires that the neuroanatomical prole of disease evolution is not random, but adheres to a spatiotemporal template of network damage: the location of disease onset within the vulnerable network may vary between
Box 2. Testing the paradigm: outstanding questions and future directions

Is diffusive protein spread a general mechanism of neurodegeneration? The strength of evidence for prion-like spread, permissive templating, and direct cellcell transmission varies among proteinopathies, and has not been established for some (e.g., fused-in-sarcoma protein) [98]. Protein spread could occur via mechanisms other than network pathways (e.g., extracellular diffusion [26,27]). Nonprion diseases have not been shown to have prion-like spontaneous infectivity and, typically, evolve more slowly [86]. Hypothesis Diffusive protein spread is a general mechanism of network disintegration. Key experiments Inoculation and tracer studies in animal models [3133] and the development of novel molecular model systems [77]. How do protein properties relate to profiles of network degeneration? Protein properties and expression profiles remain to be related in detail to cell morphological features, dynamics of cellular transport mechanisms, and intercellular interactions. For various pathogenic proteins, the final net direction of effects (loss-of-function versus toxic gain-of-function effects) remains contentious [62]. For several neurodegenerative diseases, the pathogenic protein species or the pathogenic form of the protein has not been unambiguously determined [25]. Hypothesis Network degeneration arises predictably from interactions of pathogenic protein properties with specific network morphological and functional characteristics. Key experiments Computational modelling of artificial neural networks with biologically plausible properties [44]; development of model neural circuits in vitro [99] and in vivo [38,39]; histomorphometric and immunolabelling analyses of subcellular protein targets and putative connection vulnerabilities (e.g., dendritic versus axonal) [48,51]; correlative phenotypic-histological studies; and cognitive science paradigms to characterise network activity and architectures [27,30,100]. Are empirical patterns of disease evolution consistent with nexopathy models? Certain proteinopathies have specific macroanatomical signatures of network disintegration [4,7,11,1318,2023], but disease overlap and heterogeneity are substantial. The problem of heterogeneity may be resolved, in part, by mapping longitudinal profiles of disease evolution, using neuroimaging tools that can capture convergent structural and/or functional changes across the brain and over time [7,8,11]. The molecular nexopathy paradigm makes specific predictions about the sequence of regional evolution with particular proteinopathies (Figures 2 and 3, main text). fMRI reflects synaptic function and can assess functional connectivity of networks when active and at rest [1,7], including potentially compensatory and homeostatic responses [3,10,73]; in addition, diffusion tensor imaging can assess structural connectivity of white matter pathways that bind large-scale networks [4,5]. Hypothesis Macroanatomical spatiotemporal network signatures arise predictably from underlying molecular lesions. Key experiments Longitudinal quantitative tracking of regional tissue damage in different neurodegenerative diseases in large patient cohorts (in particular, those with defined molecular lesions) derived from multicentre collaborative platforms [101]; connectivity metrics (e.g., dynamic causal modelling) to assess relations between network elements (e.g., putative shorter- versus longer-range connections) and dynamic, physiologically motivated techniques, such as magnetoencephalography (MEG) [12]; development of new molecular ligands (e.g., phosphorylated tau); and practical quantitative metrics of network function and connectivity [1,2]. What are the implications for disease diagnosis, prognosis, and treatment? Besides predicting underlying proteinopathies, improved understanding of determinants of network disintegration would enable the prediction and more accurate tracking of disease evolution. Identification of specific neural network dysfunction pre-dating the loss of structural integrity may enable modulation of culprit molecular deficits. Specific molecular treatments could be directed at modifying protein mechanisms that sustain network integrity [102] or bolstering inhibitory processes to counteract excitotoxic effects [103]. Interventions that reduce local protein concentrations might ameliorate diffusive protein spread for many diseases. Hypothesis Profiles of network disintegration provide biomarkers for tracking disease evolution and treatment response. Key experiments Detailed natural history studies and incorporation of network-level metrics in clinical trials.
566
Opinion
individuals, but progression of the particular disease in individuals, over time, would tend to recapitulate a characteristic pattern of network involvement. Therefore, to establish the disease template conclusively will entail detailed natural history studies: such studies in ALS have exploited the well-understood and highly regular organisation of the cerebral motor pools and their connections [27,71,72]. This example has also underlined the considerable functional reserve inherent in many brain networks, implying that noisy information transfer by surviving elements can support network functions until a critical stage of network failure is reached [3,10]. Both neuroimaging and behavioural metrics will be required to capture the prodromal phase of early network alterations as well as compensatory or homeostatic responses [3,11,73]. The effects of a particular proteinopathy need not and generally will not be restricted to a single vulnerable largescale network (Figures 2 and 3). Rather, nexopathy inheres in the type of network connections affected. To the extent that connections with particular properties are concentrated in a single functional network, the nexopathy paradigm would predict that proteinopathies targeting those connections should principally affect that network: this may explain the existence of neurodegenerative diseases (such as those associated with MAPT mutation and TDP-C pathology) that preferentially target the anterior temporalinferior frontal lobe semantic network [7,18,21,69,74,75]. In general, however, connection types will be represented in more than one functional network (Figure 2), providing a mechanism for the spread of proteinopathies between networks, with further phenotypic variation and potential overlap of clinicoanatomical proles among proteinopathies [18,70]. Functional interactions between large-scale brain networks will also tend to obscure network specicities [76]. Ultimately, disease spread via secondarily connected systems throughout the brain implies that network and connection specicity will be most evident earlier during the evolution of a particular disease. As a nal important caveat on the differentiation of nexopathies, it is unlikely that complete specicity will apply across the entire gamut of pathogenic proteins implicated in neurodegenerative disease. Rather, we envisage a taxonomy of predictable proles of network disintegration: within the taxonomy, particular proles of nexopathy might be common to different pathogenic proteins to the extent that those proteins share key properties that promote network damage or dysfunction. Different proteins might, for example, participate in a common, multicomponent pathogenic cascade (perhaps best characterised at present for AD [15,57]). Future directions: testing the molecular nexopathy paradigm The molecular nexopathy paradigm requires substantiation drawing on diverse molecular, cellular, and systems neuroscience (behavioural and neuroimaging) approaches, including synthetic and in vitro neural circuits, transgenic and other animal models, and dynamic macroanatomical techniques [12]. Clinical studies will continue to have a key role in delineating the sometimes counterintuitive phenotypes that dene brain network disintegration (Figure 2);
and phenotyping should be supported by detailed correlative histological studies. If the molecular nexopathy concept can be substantiated, it would hold great potential for understanding, tracking, and predicting the expression of neurodegenerative proteinopathies. Indeed, the concept need not be restricted to proteins: for example, abnormal cellular signalling linked to carbohydrate moieties could in principle give rise to sugar nexopathies [77]. Several specic, testable questions follow (Box 2) that collectively could direct future work. Beyond the principled evaluation and monitoring of candidate therapies, if mapping network breakdown is equivalent to mapping the expression of a molecular lesion, then delineating such a network could be regarded as a direct in vivo assay of the function of the protein: a concept analogous to that proposed for the theoretical neural nets of computational neuroscience [78]. If neural network dysfunction were sufciently well specied, this could in turn help identify (or discriminate between) candidate molecular mechanisms driving the neurodegenerative process and suggest rational candidate therapies.
Acknowledgements
We thank Jane Warren for proposing the term nexopathy. This work was undertaken at University College London Hospital (UCLH)/UCL and the National Institute for Health Research (NIHR) Queen Square Dementia Biomedical Research Unit. The Dementia Research Centre is an Alzheimers Research Trust Co-ordinating Centre. J.M.S. is a Higher Education Funding Council for England Clinical Senior Lecturer. M.N.R. is an NIHR senior investigator. N.C.F. is an Medical Research Council (MRC) Senior Clinical Fellow. J.D.W. is supported by a Wellcome Trust Senior Clinical Fellowship (Grant No 091673/Z/10/Z).
References
1 Hagmann, P. et al. (2008) Mapping the structural core of human cerebral cortex. PLoS Biol. 6, e159 2 Bullmore, E. and Sporns, O. (2009) Complex brain networks: graph theoretical analysis of structural and functional systems. Nat. Rev. Neurosci. 10, 186198 3 Palop, J.J. et al. (2006) A network dysfunction perspective on neurodegenerative diseases. Nature 443, 768773 4 Pievani, M. et al. (2011) Functional network disruption in the degenerative dementias. Lancet Neurol. 10, 829843 5 Catani, M. (2007) From hodology to function. Brain 130, 602605 6 Modha, D.S. and Singh, R. (2010) Network architecture of the longdistance pathways in the macaque brain. Proc. Natl. Acad. Sci. U.S.A. 107, 1348513490 7 Seeley, W.W. et al. (2009) Neurodegenerative diseases target largescale human brain networks. Neuron 62, 4252 8 Raj, A. et al. (2012) A network diffusion model of disease progression in dementia. Neuron 73, 12041215 9 Zhou, J. et al. (2012) Predicting regional neurodegeneration from the healthy brain functional connectome. Neuron 73, 12161227 10 Small, D.H. (2008) Network dysfunction in Alzheimers disease: does synaptic scaling drive disease progression? Trends Mol. Med. 14, 103108 11 Filippi, M. and Agosta, F. (2011) Structural and functional network connectivity breakdown in Alzheimers disease studied with magnetic resonance imaging techniques. J. Alzheimers Dis. 24, 455474 12 Hughes, L.E. et al. (2013) Reorganisation of brain networks in frontotemporal dementia and progressive supranuclear palsy. Neuroimage Clin. 2, 459468 13 Gardner, R.C. et al. (2013) Intrinsic connectivity network disruption in progressive supranuclear palsy. Ann. Neurol. 73, 603616 14 Scahill, R.I. et al. (2002) Mapping the evolution of regional atrophy in Alzheimers disease: unbiased analysis of uid-registered serial MRI. Proc. Natl. Acad. Sci. U.S.A. 99, 47034707 15 Warren, J.D. et al. (2012) The paradox of syndromic diversity in Alzheimer disease. Nat. Rev. Neurol. 8, 451464
567
Opinion
16 Rohrer, J.D. et al. (2010) TDP-43 subtypes are associated with distinct atrophy patterns in frontotemporal dementia. Neurology 75, 2204 2211 17 Rohrer, J.D. et al. (2010) Distinct proles of brain atrophy in frontotemporal lobar degeneration caused by progranulin and tau mutations. Neuroimage 53, 10701076 18 Rohrer, J.D. et al. (2011) Clinical and neuroanatomical signatures of tissue pathology in frontotemporal lobar degeneration. Brain 134, 25652581 19 Mackenzie, I.R. et al. (2011) A harmonized classication system for FTLD-TDP pathology. Acta Neuropathol. 122, 111113 20 Whitwell, J.L. and Josephs, K.A. (2012) Neuroimaging in frontotemporal lobar degeneration; predicting molecular pathology. Nat. Rev. Neurol. 8, 131142 21 Whitwell, J.L. et al. (2012) Neuroimaging signatures of frontotemporal dementia genetics: C9ORF72, tau, progranulin and sporadics. Brain 135, 794806 22 Whitwell, J.L. et al. (2012) Frontal asymmetry in behavioral variant frontotemporal dementia: clinicoimaging and pathogenetic correlates. Neurobiol. Aging 34, 636639 23 Agosta, F. et al. (2012) Neuroimaging ndings in frontotemporal lobar degeneration spectrum of disorders. Cortex 48, 389413 24 Tsuji, H. et al. (2012) Molecular analysis and biochemical classication of TDP-43 proteinopathy. Brain 135, 33803391 25 Collinge, J. and Clarke, A.R. (2007) A general model of prion strains and their pathogenicity. Science 318, 930936 26 Frost, B. and Diamond, M.I. (2010) Prion-like mechanisms in neurodegenerative diseases. Nat. Rev. Neurosci. 11, 155159 27 Bak, T.H. and Chandran, S. (2012) What wires together dies together: verbs, actions and neurodegeneration in motor neuron disease. Cortex 48, 936944 28 Nussbaum, J.M. et al. (2012) Prion-like behaviour and tau-dependent cytotoxicity of pyroglutamylated amyloid-b. Nature 485, 651655 29 Garden, G.A. and La Spada, A.R. (2012) Intercellular (mis)communication in neurodegenerative disease. Neuron 73, 886 901 30 Warren, J.D. et al. (2012) Disintegrating brain networks: from syndromes to molecular nexopathies. Neuron 73, 10601062 31 Baker, H.F. et al. (1994) Induction of beta (A4)-amyloid in primates by injection of Alzheimers disease brain homogenate. Comparison with transmission of spongiform encephalopathy. Mol. Neurobiol. 8, 2539 32 Meyer-Luehmann, M. et al. (2006) Exogenous induction of cerebral beta-amyloidogenesis is governed by agent and host. Science 313, 17811784 33 Eisele, Y.S. et al. (2010) Peripherally applied Abeta-containing inoculates induce cerebral beta-amyloidosis. Science 330, 980982 34 Schott, J.M. et al. (2006) Apolipoprotein E genotype modies the phenotype of Alzheimer disease. Arch. Neurol. 63, 155156 35 Agosta, F. et al. (2009) Apolipoprotein E epsilon4 is associated with disease-specic effects on brain atrophy in Alzheimers disease and frontotemporal dementia. Proc. Natl. Acad. Sci. U.S.A. 106, 2018 2022 36 Lace, G. et al. (2009) Hippocampal tau pathology is related to neuroanatomical connections: an ageing population-based study. Brain 132, 13241334 37 Grove, E.A. and FukuchiShimogori, T. (2003) Generating the cerebral cortical area map. Annu. Rev. Neurosci. 26, 355380 38 Jessell, T.M. and Sanes, J.R. (2000) Development. The decade of the developing brain. Curr. Opin. Neurobiol. 10, 599611 39 Dasen, J.S. et al. (2005) A Hox regulatory network establishes motor neuron pool identity and target-muscle connectivity. Cell 123, 477 491 40 Double, K.L. et al. (2010) Selective cell death in neurodegeneration: why are some neurons spared in vulnerable regions? Prog. Neurobiol. 92, 316329 41 Milanesi, E. et al. (2013) Molecular signature of disease onset in Granulin mutation carriers: a gene expression analysis study. Neurobiol. Aging 34, 18371845 42 Zawia, N.H. and Basha, M.R. (2005) Environmental risk factors and the developmental basis for Alzheimers disease. Rev. Neurosci. 16, 325337 43 Mesulam, M.M. (2009) Dening neurocognitive networks in the BOLD new world of computed connectivity. Neuron 62, 13
568
44 Wickens, J.R. et al. (1995) Effects of local connectivity on striatal function: stimulation and analysis of a model. Synapse 20, 281298 45 Zappone, C.A. and Sloviter, R.S. (2001) Commissurally projecting inhibitory interneurons of the rat hippocampal dentate gyrus: a colocalization study of neuronal markers and the retrograde tracer Fluoro-gold. J. Comp. Neurol. 441, 324344 46 Somogyi, P. and Klausberger, T. (2005) Dened types of cortical interneurone structure space and spike timing in the hippocampus. J. Physiol. 562, 926 47 Roberts, G.W. et al. (1993) On the origin of Alzheimers disease: a hypothesis. Neuroreport 4, 79 48 McMillan, P. et al. (2008) Tau isoform regulation is region- and cellspecic in mouse brain. J. Comp. Neurol. 511, 788803 49 Hara, M. et al. (2013) Isoform transition from four-repeat to threerepeat tau underlies dendrosomatic and regional progression of neurobrillary pathology. Acta Neuropathol. 125, 565579 50 Zhang, B. et al. (2004) Retarded axonal transport of R406W mutant tau in transgenic mice with a neurodegenerative tauopathy. J. Neurosci. 24, 46574667 51 Ikeda, M. et al. (2005) Accumulation of lamentous tau in the cerebral cortex of human tau R406W transgenic mice. Am. J. Pathol. 166, 521531 52 Lobsiger, C.S. and Cleveland, D.W. (2007) Glial cells as intrinsic components of non-cell-autonomous neurodegenerative disease. Nat. Neurosci. 10, 13551360 53 Mesulam, M.M. (1998) From sensation to cognition. Brain 121, 1013 1052 54 Shimizu, T. (2004) Information ow and emergence of the global patterns in connected neural networks. Phys. A 333, 478496 55 De Haan, W. et al. (2012) Activity dependent degeneration explains hub vulnerability in Alzheimers disease. PLoS Comput. Biol. 8, e1002582 56 Kang, J.E. et al. (2009) Amyloid-beta dynamics are regulated by orexin and the sleep-wake cycle. Science 326, 10051007 57 Winklhofer, K.F. et al. (2008) The two faces of protein misfolding: gain- and loss-of-function in neurodegenerative diseases. EMBO J. 27, 336349 58 Wirths, O. (2006) Axonopathy in an APP/PS1 transgenic mouse model of Alzheimers disease. Acta Neuropathol. 111, 312319 59 Fransson, P. and Marrelec, G. (2008) The precuneus/posterior cingulate cortex plays a pivotal role in the default mode network: Evidence from a partial correlation network analysis. Neuroimage 42, 11781184 60 Saavedra, S. et al. (2011) Strong contributors to network persistence are the most vulnerable to extinction. Nature 478, 233235 61 Gee, D.G. et al. (2011) Low frequency uctuations reveal integrated and segregated processing among the cerebral hemispheres. Neuroimage 54, 517527 62 Halliday, G. et al. (2012) Mechanisms of disease in frontotemporal lobar degeneration: gain of function versus loss of function effects. Acta Neuropathol. 124, 373382 63 Li, Z. and Dayan, P. (1999) Computational differences between asymmetrical and symmetrical networks. Network 10, 5977 64 Smeal, R.M. et al. (2010) Phase-response curves and synchronized neural networks. Philos. Trans. R. Soc. B 365, 24072422 65 Devenport, D. and Fuchs, E. (2008) Planar polarization in embryonic epidermis orchestrates global asymmetric morphogenesis of hair follicles. Nat. Cell Biol. 10, 12571268 66 Posada, A. and Clarke, P.G. (1999) The role of neuronal death during the development of topographically ordered projections: a computational approach. Biol. Cybern. 81, 239247 67 Piscopo, P. et al. (2010) Hypoxia induces up-regulation of progranulin in neuroblastoma cell lines. Neurochem. Int. 57, 893898 68 Beck, J. et al. (2008) A distinct clinical, neuropsychological and radiological phenotype is associated with progranulin gene mutations in a large UK series. Brain 131, 706720 69 Whitwell, J.L. et al. (2011) Trajectories of brain and hippocampal atrophy in FTD with mutations in MAPT or GRN. Neurology 77, 393 398 70 Kertesz, A. et al. (2005) The evolution and pathology of frontotemporal dementia. Brain 128, 19962005 71 Brettschneider, J. et al. (2013) Stages of pTDP-43 pathology in amyotrophic lateral sclerosis. Ann. Neurol. http://dx.doi.org/10.1002/ ana.23937
Opinion
72 Polymenidou, M. and Cleveland, D.W. (2011) The seeds of neurodegeneration: prion-like spreading in ALS. Cell 147, 498508 73 Agosta, F. et al. (2012) Resting state fMRI in Alzheimers disease: beyond the default mode network. Neurobiol. Aging 33, 15641578 74 Whitwell, J.L. et al. (2011) Altered functional connectivity in asymptomatic MAPT subjects: a comparison to bvFTD. Neurology 77, 866874 75 Rohrer, J.D. et al. (2008) Tracking progression in frontotemporal lobar degeneration: serial MRI in semantic dementia. Neurology 71, 1445 1451 76 Chiong, W. et al. (2013) The salience network causally inuences default mode network activity during moral reasoning. Brain 136, 19291941 77 Jessell, T.M. et al. (1990) Carbohydrates and carbohydrate-binding proteins in the nervous system. Annu. Rev. Neurosci. 13, 227255 78 Seung, H.S. (2009) Reading the book of memory: sparse sampling versus dense mapping of connectomes. Neuron 62, 1729 79 Salehi, A. et al. (2006) Increased App expression in a mouse model of Downs syndrome disrupts NGF transport and causes cholinergic neuron degeneration. Neuron 51, 2942 80 Woo, N.H. and Lu, B. (2006) Regulation of cortical interneurons by neurotrophins: from development to cognitive disorders. Neuroscientist 12, 4356 81 Grant, S.G. (2003) Synapse signalling complexes and networks: machines underlying cognition. Bioessays 25, 12291235 82 Paraoanu, L.E. and Layer, P.G. (2008) Acetylcholinesterase in cell adhesion, neurite growth and network formation. FEBS J. 275, 618 624 83 Pietri, M. et al. (2006) Overstimulation of PrPC signaling pathways by prion peptide 106-126 causes oxidative injury of bioaminergic neuronal cells. J. Biol. Chem. 281, 2847028479 84 Shen, W. et al. (2008) Dichotomous dopaminergic control of striatal synaptic plasticity. Science 321, 848851 85 Aguzzi, A. and Rajendran, L. (2009) The transcellular spread of cytosolic amyloids, prions, and prionoids. Neuron 64, 783790 86 Hardy, J. and Revesz, T. (2012) The spread of neurodegenerative disease. N. Engl. J. Med. 366, 21262128 87 Clavaguera, F. et al. (2009) Transmission and spreading of tauopathy in transgenic mouse brain. Nat. Cell Biol. 11, 909913
88 Frost, B. et al. (2009) Propagation of tau misfolding from the outside to the inside of a cell. J. Biol. Chem. 284, 1284512852 89 Liu, L. et al. (2012) Transsynaptic spread of tau pathology in vivo. PLoS ONE 7, e31302 90 De Calignon, A. et al. (2012) Propagation of tau pathology in a model of early Alzheimers disease. Neuron 73, 685697 91 Volpicelli-Daley, L.A. et al. (2011) Exogenous a-synuclein brils induce Lewy body pathology leading to synaptic dysfunction and neuron death. Neuron 72, 5771 92 Dunning, C.J. et al. (2012) Can Parkinsons disease pathology be propagated from one neuron to another? Prog. Neurobiol. 97, 205219 93 Whitwell, J.L. et al. (2007) Rates of cerebral atrophy differ in different degenerative pathologies. Brain 130, 11481158 94 Acosta-Cabronero, J. et al. (2011) Atrophy, hypometabolism and white matter abnormalities in semantic dementia tell a coherent story. Brain 134, 20252035 95 Goll, J.C. et al. (2012) Nonverbal sound processing in semantic dementia: a functional MRI study. Neuroimage 61, 170180 96 Braak, H. and Del Tredici, K. (2011) Alzheimers pathogenesis: is there neuron-to-neuron propagation? Acta Neuropathol. 121, 589595 97 Von Gunten, A. et al. (2006) Neural substrates of cognitive and behavioral decits in atypical Alzheimers disease. Brain Res. Rev. 51, 176211 98 Guest, W.C. et al. (2011) Generalization of the prion hypothesis to other neurodegenerative diseases: an imperfect t. J. Toxicol. Environ. Health A 74, 14331459 99 Shi, Y. et al. (2012) Human cerebral cortex development from pluripotent stem cells to functional excitatory synapses. Nat. Neurosci. 15, 477486 100 Lambon Ralph, M.A. et al. (2010) Coherent concepts are computed in the anterior temporal lobes. Proc. Natl. Acad. Sci. U.S.A. 107, 27172722 101 Rohrer, J.D. et al. (2012) GENFI: The GENetic Frontotemporal dementia Dement Initiative. Dement. Geriatr. Cogn. Disord. 34 (Suppl. 1), 1289 102 Feigin, A. et al. (2007) Modulation of metabolic brain networks after subthalamic gene therapy for Parkinsons disease. Proc. Natl. Acad. Sci. U.S.A. 104, 1955919564 103 Graf, R.A. and Kater, S.B. (1998) Inhibitory neuronal activity can compensate for adverse effects of beta-amyloid in hippocampal neurons. Brain Res. 786, 115121
569
Opinion
A developmental ontology for the mammalian brain based on the prosomeric model
Luis Puelles1, Megan Harrison2, George Paxinos3,4, and Charles Watson3,4,5
1 2
Department of Human Anatomy, University of Murcia, Murcia 30003, Spain Fremantle Hospital, Fremantle 6160, Australia 3 Neuroscience Research Australia, Sydney 2031, Australia 4 University of New South Wales, Sydney 2052, Australia 5 Faculty of Health Sciences, Curtin University, Perth 6845, Australia
In the past, attempts to create a hierarchical classication of brain structures (an ontology) have been limited by the lack of adequate data on developmental processes. Recent studies on gene expression during brain development have demonstrated the true morphologic interrelations of different parts of the brain. A developmental ontology takes into account the progressive rostrocaudal and dorsoventral differentiation of the neural tube, and the radial migration of derivatives from progenitor areas, using fate mapping and other experimental techniques. In this review, we used the prosomeric model of brain development to build a hierarchical classication of brain structures based chiey on gene expression. Because genomic control of neural morphogenesis is remarkably conservative, this ontology should prove essentially valid for all vertebrates, aiding terminological unication. What is ontology? The concept of ontology (see Glossary) was borrowed from the realm of philosophy by information scientists, who now use it as a way to represent an existing domain of knowledge in the form of a hierarchical taxonomy [1]. The availability of a brain ontology is vital for the eld of neuroinformatics. There have been several attempts to create a brain ontology, the most notable of which are NeuroNames [24], the Biomedical Information Research Network (BIRN) [5], and the Brain Architecture Management System (BAMS) [6,7]. However, these ontologies are largely based on traditional topographic classication of parts of the adult brain, whereas the discovery of gene targeting in mice [8] has revealed details of gene expression, lineage mapping, and causal inductive mechanisms during development, leading to a new form of hierarchical classication based on ontogeny. Conventional ontologies have the weight of tradition, but they do not include fundamental ontogenetic data, such
Corresponding author: Watson, C. (c.watson@curtin.edu.au). Keywords: ontology; gene expression; neuromeres; forebrain; midbrain; hindbrain. 0166-2236/$ see front matter . Crown Copyright 2013 Published by Elsevier Ltd. All rights reserved. http:// dx.doi.org/10.1016/j.tins.2013.06.004
as neuromeric developmental units, genoarchitectonic evidence for natural boundaries between brain parts, or their inner subdivisions (as opposed to many arbitrary classical divisions unrelated to causal mechanisms). Therefore, they have little power to encapsulate emergent understanding of brain development and structural evolution. By contrast, a developmental ontology connects adult neuroanatomy with the tradition of comparative embryology; it contemplates developmental structural units shared Glossary
Diencephalon: the caudal subdivision of the forebrain that joins the midbrain to the secondary prosencephalon; it contains three major alar domains (pretectum, thalamus, and prethalamus), as well as the corresponding tegmental regions. Evo-devo: an approach to the analysis of brain structure based on the merging of concepts drawn from evolution and embryonic development. Hodology: the study of connections within the central nervous system (odos is Greek for a road). Neuromeres: transverse unitary subdivisions of the neural tube that share a common dorsoventral structure (floor, basal, alar, and roof plates), but each have differential molecular identities and fates; they comprise the secondary prosencephalon, diencephalon (prosomeres), the midbrain (mesomeres), and the hindbrain (rhombomeres). Ontogeny: Greek for the genesis of being: the process of development. Ontology: a formal conceptualization of the structure of a knowledge base, usually in the form of a hierarchical classification. Pallium: major subdivision of the telencephalon, usually visualized as covering and surrounding the subpallium; in mammals, it gives rise to the cerebral cortex and several claustroamygdaloid pallial nuclei. Prosencephalon: Greek for forebrain: the part of the brain that appears at the rostral end of the neural tube. Secondary prosencephalon: the rostral major subdivision of the developing forebrain that separates from the diencephalon caudally (early in development, both are encompassed within the primary prosencephalon); the secondary prosencephalon includes the telencephalon, the eye, and the hypothalamus. Subpallium: a major subdivision of the telencephalon usually visualized topographically as lying under the pallium, at the brain baseit generates the so-called basal ganglia, including the striatum, pallidum, diagonal-basal area, and preoptic area. Tagma: a meaningful higher-level unit of biological structure, comprising segments that share a general character (e.g., the Drosophila thorax tagma as opposed to the abdominal tagma). Telencephalon: a dorsal subdivision of the secondary prosencephalon that forms the pallium and subpallium. Topography: a system for describing and representing position relative to external references. Topology: a system for describing the relative position of the components of a structure irrespective of external references and any nondisruptive deformations; topology attends exclusively to the invariant neighborhood relations between the components.
570
Opinion
(A)
NP
(B) M F H SpC
(C) M D SP PPH PH PMH MH
SpC
(D) p2 CSP p3 m2 r2 p1 m1 is r1 r3 r9 r4 r5 r6 r7 r8 r10 r11
RSP
SpC
(E) MTt Pt Th Tel PtTg MTg ThTg PTh PlsTg r1 PThTg r2 PedHy r3 PPHy r4 is PlsTt
SpC r11 r9 r10 r5 r6 r7 r8 Alarbasal boundary
POTel
Alarbasal boundary
Figure 1. A series of diagrams of lateral views of the developing mouse brain. (A) The neural primordium (NP), which is a hollow tube with no subdivisions. In (B), the rostral (left) part of the neural tube shows the appearance of the forebrain (F), midbrain (M), and hindbrain vesicles (H), with the developing spinal cord (SpC) on the right. In (C), the forebrain vesicle has two divisions, the secondary prosencephalon (SP) and the diencephalon (D), and the hindbrain is divided into four regions: the prepontine hindbrain (PPH), the pontine hindbrain (PH), the pontomedullary hindbrain (PMH), and the medullary hindbrain (MH). In (D) from the top, more subdivisions appear in the forebrain [caudal secondary prosencephalon (CSP or hp1); rostral secondary prosencephalon (RSP or hp2); and prosomeres 13 of the diencephalon (p1, p2, and p3)], midbrain [mesomere 1 and 2 (m1 and m2)], and hindbrain [isthmus (is) and rhombomeres 111 (r1 to r11)]. In (E), some parts of the forebrain have become further differentiated: the caudal prosencephalon has formed the main part of the telencephalon; the rostral secondary prosencephalon has formed the preoptic telencephalon (POTel), the terminal hypothalamus (THy), and the peduncular hypothalamus (PedHy)); and prosomeres 13 have formed the pretectum (Pt), thalamus (Th), and prethalamus (PTh), respectively. In this diagram, the diencephalon and midbrain are further subdivided by the alarbasal boundary, which bounds distinct tegmental regions [prethalamic tegmentum (PThTg); thalamic tegmentum (ThTg); pretectal tegmentum (PtTg); midbrain tegmentum (MTg); and preisthmic tegmentum (PIsTg)]. The dorsal part of the midbrain is divided into the main midbrain tectum (MTt) and smaller preisthmic tectum (PIsTt). Created by L. Puelles for the Allen Brain Institute (http:// developingmouse.brain-map.org).
among vertebrates, as revealed by fate-mapping studies, and is consistent with evolutionarily conserved gene patterns. Because of this, a developmental ontology has the capacity to stimulate insights into causation. Our proposal of a developmental ontology for the adult mouse brain is a simplied version of the extended version designed by L. Puelles for the Allen Developing Mouse Brain Atlas (http:// developingmouse.brain-map.org). The new ontology is consistent with the most recent version of the prosomeric model [9], a known paradigm for structural and molecular
analysis of vertebrate brains (Figure 1). Note that this review centers on the ontology and does not aim to explore the prosomeric model itself. Comparing traditional and developmental ontologies All adult brain ontologies start with the recognition of three basic elements: forebrain, midbrain, and hindbrain. Even at this level, the developmental ontology is distinctive, in that it includes the isthmus within the hindbrain, rather than in the midbrain, as found in traditional
571
Opinion
(A) (B)
Pallium
Roofplate
Roofplate
Thalamus
Pretectum p1
Telencephalon
Striatum
Diencephalon
p2 zli Prethalamus p3
Midbrain Telencephalon
Cerebellum
m1 m2
Diencephalon
Midbrain
Cerebellum
Substana nigra
is
Pallidum Dg Dg Septal roofplate

Peduncular hypothalamus
STh
Cephalic exure
M
r1 r2 r3
Hi
Hi
nd
r4
nd
Preopc area Anterior commissure
Terminal hypothalamus
Cervical exure
br ain
br ain
Hypothalamus
r11
Spinal cord
Cervical exure
Spinal cord
Eye
Neurohypophysis
Eye
Neurohypophysis
Figure 2. The segmental organization of the developing brain. (A) A diagram of a lateral view of developing mouse brain at a stage later than the last element shown in Figure 1 (main text). The telencephalon is now divided into the pallium and subpallial regions [striatum, pallidum, diagonal domain (Dg), and preoptic area]. The septal roofplate (gray shading) extends from the telencephalic roof to the developing anterior commissure (ac). Within the terminal hypothalamus, the eye vesicle, the neurohypophysis, and the mamillary bodies (M) are differentiating. Within the peduncular hypothalamus, the subthalamic nucleus (STh) is developing. The red line represents the alarbasal boundary, also in the midbrain and diencephalon. In the diencephalon, this molecular boundary is for a short distance pulled to the diencephalic roof as the zona limitans (ZLi), which largely separates p2 and p3 at alar plate levels. The gray area above the cephalic flexure represents the most rostral area of Sonic hedgehog (Shh) expression in the floor plate. The diagram shows that the developing substantia nigra extends rostrally from the midbrain into the diencephalon. Other abbreviations are as in Figure 1. (B) This is another version of (A), to show the alarbasal boundary from the spinal cord to the rostral hypothalamus. The basal plate is in green and the alar plate is in pink.
ontologies. The importance of transgenic fate mapping to the understanding of such delimitation has been emphasized previously [1012]. There are many novel features at the next hierarchical level of the developmental ontology. In the forebrain, three components are recognized: the telencephalon, the hypothalamus (including the eye vesicle), and the diencephalon (Figure 1). The telencephalon and hypothalamus constitute the secondary prosencephalon [12] (Figure 1C), which can be separated from the diencephalon proper on the basis of gene expression patterns [1315]. This is further supported by experimental evidence showing that the holoprosencephalic malformation syndrome selectively affects the telencephalon and the hypothalamus [16], and the Otx2/ Emx1+/ double mutant mouse selectively loses the diencephalon, but retains the hypothalamustelencephalon complex [17,18]. In traditional ontologies, the hypothalamus was incorporated within the diencephalon, because it was incorrectly assumed to be the equivalent of the basal and oor plates of the thalamus [19]. Recent fate maps and molecular ndings [14,15] indicate that the hypothalamus is topologically rostral to the thalamus and other parts of the diencephalon, and that the telencephalic and eye vesicles represent bilateral alar evaginations. The neural tube ends rostrally at the terminal wall (between the mamillary body and the anterior commissure) with a distinct bow-like specialized median domain, named the acroterminal area, across the tuberal (basal) and chiasmatic (alar) hypothalamus and the preoptic area [16]. Although it does not evaginate, the preoptic area belongs to the telencephalic subpallium on the basis of shared molecular proles [20]. This classication is further supported by the nding that the preoptic area contributes tangentially migrating neurons to the telencephalic pallium, similar to the other parts of the subpallium, but not to the hypothalamus [2124].
572
In the hindbrain, the developmental ontology recognizes 12 segments (the isthmus and 11 rhombomeres) (Figure 1D,E), principally on the basis of broblast growth factor 8 (Fgf8) [10,25] and homeobox (Hox) gene expression [26], whereas the traditional ontology divides the hindbrain into pons and medulla oblongata, classic gross divisions that are not consistent with the rhombomeric subdivisions [12]. In the developmental ontology, the cerebellum is separated from the basilar pons, and properly classied as a dorsal outgrowth of the isthmus and the rst rhombomere, a result of several fate-mapping studies [27 29] (Figure 1D,E). At the third level of the developmental ontologic hierarchy, the telencephalon is divided into pallium and subpallium, as it is in traditional ontologies (Figure 2A). However, as noted above, the new subpallium includes the preoptic area [20], parallel to three other distinct radial subpallial domains, identied in mediolateral order as diagonal domain (old substantia innominata, covered supercially by the diagonal band formation), pallidum, and striatum (Figure 2A). Each of these domains extends radially from the ependyma to the pial surface and has diverse stratied derivatives. The striatum contacts the pallium across the palliosubpallial boundary [30]. The four subpallial domains are elongated along the septoamygdaloid axis, and they form at their ends composites known as the subpallial amygdala and the subpallial septum [30] (Figure 3). The hypothalamus is divided into terminal (rostral) and peduncular (caudal) neuromeric portions, continuous dorsally with the preoptic area (unevaginated telencephalon) and the evaginated telencephalic hemisphere, respectively [16] (hypothalamic prosomeres hp2 and hp1; Figure 2A). At this third level, the diencephalon is divided caudorostrally into three diencephalic prosomeres: the pretectum (prosomere 1), the thalamus (prosomere 2), and the prethalamus (prosomere 3),
Opinion
(A) (B)
Striat
Striat
Pall Diag Preopt
Sept Parasept Preopt
Pall Diag Central Amygd
Figure 3. A diagram to show the main subdivisions of the subpallium examined in a coronal plane. Panel (A) represents the four main histogenetic domains of the subpallium: striatum (Striat), pallidum (Pall), diagonal domain (Diag), and preoptic area (Preopt). These extend radially from the ventricle to the pial surface. Panel (B) shows the secondary subdivisions of these domains along the septoamygdaloid axis. Each subpallial histogenetic domain shows diversely differentiated septal (Sept), paraseptal (Parasept), central, and amygdaloid regions (Amygd). This concept recapitulates and expands the idea of an extended subpallial amygdala.
including corresponding tegmental components [14,16, 3133] (Figure 2A). Conventional ontologies assign the diencephalic tegmental territory to either midbrain or posterior hypothalamus. The diencephalic tegmentum contains rostral parts of the ventral tegmental area and substantia nigra, which are in fact isthmo-meso-diencephalic plurisegmental entities [11,16,3436]. Subsequent hierarchical levels of subdivision of the forebrain, midbrain, and hindbrain are based on the well-known dorsoventral regionalization of the neural tube into roof plate, alar plate, basal plate, and oor plate, introduced by His [37]. These longitudinal zones can now be more precisely dened by their differential molecular proles. The longitudinal extent of the oor plate, ending at the mamillary oor, is marked by genes such as Sonic hedgehog (Shh), netrin 1 (Ntn1), and LIM homeobox transcription factor 1-beta (Lmx1b), although these markers are not unique to this zone [15]. The basal plate of the forebrain, midbrain, and hindbrain is dened by the presence of either NK2 homeobox 1 (Nkx2.1) or 6 (Nkx6.1) [38]. At early developmental stages, expression of paired box (Pax) genes (Pax6 in the forebrain; Pax 3 and Pax7 in the midbrain and hindbrain) broadly characterizes the alar plate [39,40]. Genes of the Zic family are expressed at the dorsal part of the alar plate, next to the roof plate [41]. Consistent with a specic set of gene expression patterns [e.g., Nkx2.2, patched 1 (Ptc1), or gamma glutamylcysteine synthetase 1 (Gsh1)], the developmental ontology includes a distinct liminal band at the ventral rim of the alar plate. The liminal region is dened as the junction of basal and alar plates. This broad-brush picture of the developmental ontology is based upon the redenition of brain organization suggested by the prosomeric model, which attempts to connect developmental discoveries with adult neuroanatomy
across the whole brain. This picture needs more detailed explanation in three areas: (i) the issue of the now obsolete columnar forebrain axis of Herrick [42]; (ii) the denition of neuromeres; and (iii) the organization of the pallial and subpallial telencephalic territories. Herrick and the columnar forebrain axis Twentieth-century ideas on forebrain subdivisions were dominated by the columnar model of the forebrain put forward by Herrick [42]. Contradicting the earlier analysis of His [43], Herrick proposed that the brainstem axis simply extended rostrally as a straight line into the forebrain, so that the thalamushypothalamus complex (then called the diencephalon) stood in direct axial continuity with the telencephalon rostrally and the midbrain caudally. This concept was not based on developmental ndings, and it disregarded the ostensible cephalic exure, present in all vertebrates. It was based instead on an ill-fated ad hoc attempt to extrapolate brainstem functional columns into the forebrain. The columnar hypothalamus was held to be exclusively basal (and connected directly with the midbrain tegmentum), matched with an exclusively alar thalamus. The columnar concept of forebrain organization has not stood the test of developmental gene expression patterns and related causal mechanisms, whereas these recent discoveries have strongly endorsed the earlier axial conception of His, in which the telencephalon is dorsal to the hypothalamus, and the diencephalon proper lies caudal to it [16,4447] (Figure 2B). This fundamental error has had a major impact on the interpretation of the mutual topographic relations of the telencephalon, eyes, hypothalamus, and diencephalon by many authors [19,48]. In the updated revised model of the forebrain that underpins our developmental ontology, the term diencephalon accordingly includes only the
573
Opinion
prethalamus, thalamus, and pretectum, each with a piece of basal tegmentum, and excludes the hypothalamus (Figure 2B). Within the developmental ontology, the hypothalamus is divided rst into a rostral part (terminal hypothalamus) and a caudal part (peduncular hypothalamus) [16] (Figure 2A). The intrahypothalamic boundary separating these parts runs parallel to the subsequent course of the fornix tract; the boundary lies just rostral to the fornix and the neighboring medial forebrain bundle and lateral forebrain bundle (cerebral peduncle) (Figure 2A). This boundary also separates the preoptic area from the diagonal region in the evaginated telencephalon, and also divides the mamillary and retromamillary areas one from another [16] (Figure 2A). The evaginated eyes and surrounding supraoptic and suprachiasmaticanterior-hypothalamic areas are all alar derivatives of the terminal hypothalamus, whereas the tuberal region, the attached neurohypophysis, and the mamillary bodies are corresponding basal derivatives (see [16]). The peduncular hypothalamus (marked by its role as bed of the medial and lateral forebrain bundles and the fornix) contains the main part of the deep paraventricular nucleus and the radially migrated entopeduncular complex as its principal alar derivatives, and includes retrotuberal and retromamillary formations in its basal subregion. The basal subregion contains the subthalamic nucleus (a migrated retromamillary derivative), which was previously not considered to be a part of the hypothalamus [16,49] (Figure 2A). Neuromeres Periodic transverse outpouchings in the neural tube wall were rst recognized over a century ago ([12,50,51]). Orr [51] called them neuromeres (prosomeres in the prosencephalon, mesomeres in the midbrain, and rhombomeres in the hindbrain), thus viewing them as neural segments within the general plan of head segmentation (Figure 1D). The neuromere concept fell into disuse with the rise in popularity of Herricks columnar paradigm, basically because brain segments did not offer at that time a recognizable functional signicance. However, interest in neuromeres has returned in the molecular era because distinct molecular proles characterize these developmental units and their fate-mapped derivatives, thus dispensing with the old myth that neuromeres were transient early phenomena. The molecular segmental scenario provides an opportunity for causal explanations of brain structure, an endeavor that completely collapsed with the old columnar theory. Moreover, physiologists have identied various examples of brainstem functional circuitry that relate to neuromeric compartments [52,53]. In a similar way, functionally distinct prethalamic, thalamic, pretectal, and midbrain circuits are now recognized to be neuromerically organized, after their interpretation as longitudinal zones was abandoned. Therefore, any developmental ontology is obliged to take account of these transverse components of the neuraxis. The boundary between diencephalon and midbrain is marked by the caudal boundary of the forebrain expression of Pax6 [54]. The boundary between midbrain and hindbrain is sharply dened by the interface of the expression
574
territories of orthodenticle homeobox 2 (Otx2) and gastrulation brain homeobox 2 (Gbx2) [55,56]. The midbrain proper can be subdivided into rostral and caudal mesomeres (m1 and m2; Figure 1) [11]. The existence of a thin m2 component of the midbrain, separating the isthmus from the inferior colliculus, was previously recognized by early embryologists [57]. Gene-mapping studies support the existence of m2, dened by coexpression of Otx2 and Pax2 [58]. Its alar subregion has been called the preisthmic domain (which includes the nucleus sagulum, the cuneiform gray, and subbrachial nucleus) and its basal structures include the retrorubral A8 catecholaminergic cells and the rostral-most part of the dorsal raphe nucleus [11,12,58]. The developing hindbrain is overtly segmented in its central part (r2r6). Its rostral (isthmus, r1) and caudal (r7r11) parts are not overtly segmented, but are differentially coded molecularly into cryptic neuromeric compartments [12,26,5961] (Figure 1D,E). The rostral hindbrain is inuenced by the isthmic organizer [6264] and can be divided into the isthmus and rhombomere 1, which contribute to the formation of the cerebellum in the same general way as the hypothalamus builds the telencephalon [6466]. The isthmus itself is dened by the early expression of Fgf8 [25]. The remainder of the hindbrain is marked by the diversied expression of Hox genes and ephrins [66 71]. Rhombomeres r2r6 are usually recognized as overt bulges separated by constrictions in the embryonic hindbrain [51,66,67,71]. However, the boundaries of cryptorhombomeres r7r11 were distinguished in the chick on the basis of fate mapping and differential Hox gene expression, and the same subdivisions exist in the mouse [72]. In fact, the series of rhombomeres seems to be conserved among all vertebrates [73,74]. These segments in the hindbrain are molecularly distinct developmental units [26,67,75,76]. For example, we can now recognize 11 neuromeric parts of the trigeminal sensory column across r1r11. The vestibular column can be similarly subdivided segmentally across r1r9. Raphe nuclei have been recently analysed in the rhombomeric context, identifying some 45 separable cell groups across the whole hindbrain [77]. Differential histogenetic behavior due to regional changes in the molecular identity of both progenitors and derived neurons correlates with the development of differentiated transverse blocks of hindbrain structure [26,60,66,71]. Because of shared dorsoventral patterning, many rhombomeres contain similarly placed nuclei and, because early observable boundaries largely become invisible as development proceeds, such serial nuclei form plurineuromeric sensory columns, but the unique molecular identities and consequent differential hodological or functional properties of individual segments often persist [52]. Fate- mapping has shown that the neuromeric developmental units are still separated by cryptic boundaries as the hindbrain matures [60,66,71] and, therefore, represent necessary conceptual levels for structural classication in any modern ontology. Three major tangential migrations cause signicant alterations to the anatomy of the hindbrain region. Pontine neurons migrate from the caudal rhombic lip (r6r7) to the
Opinion
ventral surface of r3 and r4, where they form the pontine nuclei [71,72,7781]. More caudal rhombic lip derivatives (r8r11) migrate via an extramural tangential stream into the lateral reticular nucleus and external cuneate nucleus, and via an intramural stream into the inferior olive [60,81 83]. The mammalian facial motor nucleus migrates from the medial part of r4 to reach its familiar supercial position in r6 [18,84]. Recently, the interpeduncular nucleus has been shown to form by tangential migratory convergence of diverse alar and basal cell populations at the isthmic and r1 midline [85,86]. The new developmentally based hindbrain ontology purposefully avoids using the classic subdivision of pons and medulla. The term pons is particularly misleading as a regional descriptor. The pons of the human brain mistakenly refers to a region that apparently extends from the midbrain to the caudal border of the sixth rhombomere (the caudal end of the facial nucleus). In fact, the basilar pontine formation arises from the rhombic lip in r6 and r7, and migrates to the ventral margin of r3 and r4 [29,72,81,87]. In the mouse, the crossed pontocerebellar bers of the middle cerebellar peduncle grow across r2 into the r1 entrance into the cerebellum, thus encircling the trigeminal root; this gives the impression that r2 also forms part of the pons. Therefore, the term pons as a regional descriptor of a zone ventral to the cerebellum reaching from the midbrain to the medulla must be abandoned, although the r3r4 region can be called pontine hindbrain. The word pons sensu stricto can be properly applied to the nuclei and crossing bers of the basilar pontine formation. Note that there is a substantial prepontine hindbrain, often misidentied as a caudal midbrain domain, which includes isthmus and rhombomeres 1 and 2. The trapezoid body and neighboring trapezoid and superior olivary and periolivary nuclei of the auditory pathway, as well as the nucleus abducens, are strictly retropontine, being associated with r5. The pyramidal decussation lies in r11. Pallium and subpallium The telencephalic hemisphere is not an axial vesicle, because there are two of them. It is formed by an idiosyncratic patterning mechanism that generates pallial and subpallial regions that are already at neural plate stages [8891]. The traditional columnar misconception that the longitudinal axis of the brain ends in the telencephalon has resulted in the widespread assumption that the embryonic pallium lies dorsal to the embryonic subpallium; in reality, the pallium is topologically caudal to the subpallium, as shown conclusively by neural plate fate maps [16,89,90,92]. The rostral end of the brain axial dimension is identied by landmarks in the roof, the oor, and the alarbasal boundary. The rostral end of the roof plate has been fate mapped to the locus of the anterior commissure [16,89]; the oor plate ends at the mamillary pouch [16]; the alarbasal boundary is continuous from left to right just under the suprachiasmatic nucleus [16]. These three points are equally rostral-most and can be traced molecularly from early neural plate stages onwards. The median wall interconnecting them builds the terminal wall of the neural tube. The telencephalon sits alongside the dorsal
alar subregion of this wall, with the preoptic area extending into the bed of the anterior commissure. Differential gene expression denes the telencephalic palliosubpallial boundary [9294] (Figure 2A). The presence of specic regional molecular proles has profound histogenetic and functional consequences, with the subpallium becoming the almost exclusive source of inhibitory (GABAergic) telencephalic neurons [95], and the pallium becoming the main source of excitatory (glutamatergic) neurons [96,97]. The GABAergic subpallial neurons that migrate tangentially into the developing pallium express distal-less homeobox (Dlx), LIM homeobox 6 (Lhx6), and various other genes [95]. The amniote pallium can be divided on anatomical and molecular grounds into septal, medial, dorsal, lateral, ventral, and amygdaloid zones [97]. In mammals, the medial pallium forms hippocampal and parahippocampal structures; the dorsal pallium forms the neocortical mantle, and the lateral and ventral pallium are held to form the olfactory cortical structures with attached pallial claustroamygdaloid nuclei [97,98]. The intimate relation of the mammalian claustrum with the insular cortex is still not well understood. Non-mammalian vertebrates show the same pallial domains, but mammals are the only animals that have a fully developed neocortex [9,99]. Further developmental ontologic subdivisions of each of the pallial and subpallial zones can be recognized [100]. The medial pallium differentiates into the diverse components of the hippocampal formation: dentate gyrus, CA elds, subiculum, presubiculum, parasubiculum, entorhinal cortex, and retrosplenial cortex [101]. The dorsal pallium includes the typical central group of neocortical areas (orbitofrontal, parietal, occipital, and temporal) and the mesocortical areas at its periphery (cingulate, perirhinal, periorbital, and insular cortex) [102]. The lateral pallium and ventral pallium each contribute to the olfactory areas, the insular claustrum, and the amygdala [98]. The subpallium is adjacent to the ventral pallium along the palliosubpallial boundary. It is organized bilaterally as several bands of differentially specied radial histogenetic territories, stretching along the septoamygdaloid axis (Figure 3). Each subpallial band has septal, paraseptal, central, and amygdaloid differentiated portions, wherein the major derivatives are the central ones (striatum, pallidum, diagonal-basal domain, and preoptic area; Figure 3). The best-known paraseptal element is the striatal nucleus accumbens, which is complemented by analogous paraseptal parts within the pallidal, diagonal, and preoptic zones. The concept of the extended amygdala [103] antedated the realization that some amygdaloid nuclei share properties with subpallial sectors beyond the apparent limits of the traditional amygdala, extending supraand infracapsularly all the way to the septum. The present ontology is consistent with this view, with its medial (diagonal) and lateral (pallidal) portions of the stria terminalis complex and central amygdaloid striatal elements [30]. The other major components of the diagonal-basal domain are the substantia innominata, the basal magnocellular nucleus, and the nuclei of the diagonal band. Ontological representation of these regions is complicated by the fact that some classic anatomic entities (such as the
575
Opinion
Box 1. The future of brain ontology
Is the new ontology flexible enough to incorporate future discoveries in gene expression during development? How will future changes to the ontology be moderated and approved for inclusion? Who will host the current ontology and its future derivatives? What are the barriers to extending the ontology to include all vertebrate classes? In cases where a neuron group migrates from its place of origin to another brain region, should it be classified according to the region of origin, the final location, or both?
septum and the amygdala) comprise developmentally distinct sectors deriving from the pallium, striatum, pallidum, and diagonal area, respectively and also encompassing tangentially migrated cells derived from the alar hypothalamus and the preoptic area [22,30,96,97]. Concluding remarks A new brain ontology based on modern developmental criteria and models provides a more accurate and complete view of the natural morphologic interrelations of different parts of the brain. Neuronal populations have been classied on the basis of progressive rostrocaudal and dorsoventral differentiation of the neural tube. Radially migrated (stratied) derivatives can be traced from progenitor areas based on fate mapping, gene mapping, and other experimental evidence. Overall, this demonstrates a consistent relation between embryonic patterns and adult structure, which is expressed in the form of a developmental ontology. Because genomic control of neural morphogenesis is remarkably conservative, this ontology should prove to be essentially valid for all vertebrates, aiding terminological unication. The new ontology will undoubtedly be contested, and many details are still subject to revision that will be based on further discoveries (Box 1).
References
1 Gruber, T.R. (1993) A translation approach to portable ontology specications. Knowl. Acquisit. 5, 199220 2 Bowden, D.M. (1995) NeuroNames brain hierarchy. Neuroimage 2, 6383 3 Bowden, D.M. et al. (2007) Creating neuroscience ontologies. Methods Mol. Biol. 401, 6787 4 Bowden, D.M. (2012) NeuroNames: an ontology for the Braininfo portal to neuroscience on the web. Neuroinformatics 10, 97114 5 Bug, W.J. et al. (2008) The NIFSTD and BIRNLex vocabularies: building comprehensive ontologies for neuroscience. Neuroinformatics 6, 175194 6 Bota, M. et al. (2005) Brain architecture management system. Neuroinformatics 3, 1548 7 Bota, M. and Swanson, L.W. (2008) BAMS neuroanatomical ontology: design and implementation. Front. Neuroinform. 2, 2 8 Capecchi, M.R. (1989) The new mouse genetics: altering the genome by gene targeting. Trends Genet. 5, 7076 9 Puelles, L. (2001) Brain segmentation and forebrain development in amniotes. Brain Res. Bull. 55, 695710 10 Zervas, M. et al. (2004) Cell behaviors and genetic lineages of the mesencephalon and rhombomere 1. Neuron 43, 345357 11 Puelles, E. et al. (2012) Midbrain. In The Mouse Nervous System (Watson, C. et al., eds), pp. 337359, Elsevier Academic Press 12 Alonso, A. et al. (2012) Development of the serotonergic cells in murine raphe nuclei and their relations with rhombomeric domains. Brain Struct. Funct. http://dx.doi.org/10.1007/s00429-012-0456-8
13 von Kupffer, C. (1906) Die Morphogenie des Centralnervensystems. In Handbuch der vergleichenden Entwicklungslehre der Wirbeltiere (Hertwig, O., ed.), pp. 1272, (Bd 2 Teil 3), Gustav Fischer 14 Conte, I. et al. (2005) Comparative analysis of Six3 and Six6 distribution in the developing and adult mouse brain. Dev. Dyn. 234, 718725 15 Shimogori, T. et al. (2010) A genomic atlas of mouse hypothalamic development. Nat. Neurosci. 13, 767775 16 Puelles, L. et al. (2012) Hypothalamus. In The Mouse Nervous System (Watson, C. et al., eds), pp. 221313, Elsevier Academic Press 17 Lagutin, O.V. et al. (2003) Six3 repression of Wnt signaling in the anterior neuroectoderm is essential for vertebrate forebrain development. Genes Dev. 17, 368379 18 Studer, M. (2001) Initiation of facial motoneurone migration is dependent on rhombomeres 5 and 6. Development 128, 37073716 19 Swanson, L.W. (2010) Brain Architecture, Oxford University Press 20 Kimura, J. et al. (2005) Emx2 and Pax6 function in cooperation with Otx2 and Otx1 to develop caudal forebrain primordium that includes future archipallium. J. Neurosci. 25, 50975108 21 Flames, N. et al. (2007) Delineation of multiple subpallial progenitor domains by the combinatorial expression of transcriptional codes. J. Neurosci. 27, 96829695 22 Hirata, T. et al. (2009) Identication of distinct telencephalic progenitor pools for neuronal diversity in the amygdala. Nat. Neurosci. 12, 141149 23 Gelman, D.M. et al. (2009) The embryonic preoptic area is a novel source of cortical GABAergic interneurons. J. Neurosci. 29, 93809389 24 Flandin, P. et al. (2010) The progenitor zone of the ventral medial ganglionic eminence requires Nkx2-1 to generate most of the globus pallidus but few neocortical interneurons. J Neurosci 30, 28122823 25 Watson, C. (2010) The presumptive isthmic region in a mouse as dened by fgf8 expression. Brain Behav. Evol. 75, 315 26 Marin, F. et al. (2008) Hox gene colinear expression in the avian medulla oblongata is correlated with pseudorhombomeric domains. Dev. Biol. 323, 230247 27 Otero, R.A. et al. (2004) Chick/quail chimeras with partial cerebellar grafts: an analysis of the origin and migration of cerebellar cells. J. Comp. Neurol. 333, 597615 28 Alvarado-Mallart, R-M. (1993) Fate and potentialities of the avian mesencephalic/metencephalic neuroepithelium. J. Neurobiol. 24, 13411355 29 Hallonet, M. and Alvarado-Mallet, R-M. (1997) The chick/quail chimeric system: a model for early cerebellar development. In Perspectives on Developmental Neurobiology (Bronner-Fraser, M. et al., eds), pp. 1731, Gordon and Breach Science Publishers 30 Bardet, S.M. et al. (2010) Ontogenetic expression of sonic hedgehog in the chicken subpallium. Front. Neuroanat. 4, 116 31 Puelles, L. and Rubenstein, J.L. (2003) Forebrain gene expression domains and the evolving prosomeric model. Trends Neurosci. 26, 469476 32 Lim, Y. and Golden, J.A. (2007) Patterning the developing diencephalon. Brain Res. Rev. 53, 1726 33 Puelles, L. (2001) Thoughts on the development, structure and evolution of the mammalian and avian telencephalic pallium. Philos. Trans. R. Soc. Lond. B: Biol. Sci. 356, 15831598 34 Puelles, L. and Medina, L. (1994) Development of neurons expressing tyrosine hydroxylase and dopamine in the chicken brain: a comparative segmental analysis. In Phylogeny and Development of Catecholamine Systems in the CNS of Vertebrates (Smeets, W.J.A.J. and Reiner, A., eds), pp. 381404, Cambridge University Press 35 Verney, C. et al. (2001) Structure of longitudinal brain zones that provide the origin for the substantia nigra and ventral tegmental area in human embryos, as revealed by cytoarchitecture and tyrosine hydroxylase, calretinin, calbindin, and GABA immunoreactions. J. Comp. Neurol. 429, 2244 36 Smits, S. et al. (2006) Developmental origin and fate of mesodiencephalic dopamine neurons. Prog. Neurobiol. 78, 116 ge zur Einteilung des Gehirns. Arch. Anat. 37 His, W. (1893) Vorschla Entwickelungsgesch. 3, 172179 38 Puelles, L. et al. (2004) Gene maps and related histogenetic domains in the forebrain and midbrain. In The Rat Nervous System (3rd edn) (Paxinos, G., ed.), pp. 324, Elsevier Academic Press
576
Opinion
39 Manuel, M. and Price, D.J. (2005) Role of Pax6 in forebrain regionalization. Brain Res. Bull. 66, 387393 40 Matsunaga, E. et al. (2001) Role of Pax3/7 in the tectum regionalization. Development 128, 40694077 41 Aruga, J. et al. (1994) A novel zinc nger protein, Zic, is involved in neurogenesis, especially in the cell lineage of cerebellar granule cells. J. Neurochem. 63, 18801890 42 Herrick, C.J. (1910) The morphology of the forebrain in Amphibia and Reptilia. J. Comp. Neurol. 20, 413547 ber das frontale Ende des Gehirnrohrs. Arch. Anat. 43 His, W. (1893) U Entwickelungsgesch. 3, 157171 44 His, W. (1892) Zur allgemeinen Morphologie des Gehirns. Arch. Anat. Entwickelungsgesch. 2, 346383 45 His, W. (1895) Die Anatomische Nomenclatur. Nomina Anatomica. Arch. Anat. Entwickelungsgesch. 1895 (Suppl.), 1180 46 His, W. (1904) Die Entwicklung des menschlichen Gehirns wa hrend der ersten Monate, Hirzel 47 Martinez, S. et al. (2012) Molecular regionalization of the developing neural tube. In The Mouse Nervous System (Watson, C. et al., eds), pp. 218, Elsevier Academic Press 48 Kuhlenbeck, H. (1973) Overall morphological pattern. In The Central Nervous System of Vertebrates (Vol. 3, part II), pp. 1777, Karger 49 Skidmore, J.M. et al. (2012) A novel TaulacZ allele reveals a requirement for Pitx2 in formation of the mammillothalamic tract. Genesis 50, 6773 50 von Baer, K.E. (1828) Entwicklungsgeschichte der Thiere: Beobachtung und Relfexion, Borntrager 51 Orr, H.J. (1887) Contribution to the embryology of the lizard. J. Morphol. 1, 311372 52 Chatonnet, F. et al. (2006) Ontogeny of central rhythm generation in chicks and rodents. Respir. Physiol. Neurobiol. 154, 3746 53 Diaz, C. and Glover, J.C. (2002) Comparative aspects of the hodological organization of the vestibular nuclear complex and related neuron populations. Brain Res. Bull. 57, 307312 54 Schwarz, M. et al. (1999) Pax2/5 and Pax6 subdivide the early neural tube into three domains. Mech. Dev. 82, 2939 55 Millet, S. et al. (1996) The caudal limit of Otx2 gene expression as a marker for the midbrain/hindbrain boundary: a study using in situ hybridization and chick/quail homotopic grafts. Development 122, 37853797 56 Millet, S. et al. (1999) A role for Gbx2 in repression of Otx2 and positioning the mid/hindbrain organizer. Nature 401, 161164 57 Palmgren, A. (1921) Embryological and morphological studies on the mid-brain and cerebellum of vertebrates. Acta Zool. 2, 194 nchez, M. et al. (2005) Distinct pre-isthmic domain, dened 58 Hidalgo-Sa by overlap of Otx2 and Pax2 expression domains in the chicken caudal midbrain. J. Comp. Neurol. 483, 1729 59 Puelles, L. et al. (2007) The Chick Brain in Stereotaxic Coordinates. An Atlas Featuring Neuromeric Subdivisions and Mammalian Homologies, Elsevier Academic Press 60 Cambronero, F. and Puelles, L. (2000) Rostrocaudal nuclear relationships in the avian medulla oblongata: fate-map with quailchick chimeras. J. Comp. Neurol. 427, 522545 61 Puelles, L. (2013) Plan of the developing vertebrate nervous system (prosomere model, overview of brain organization). In Comprehensive Developmental Neuroscience (Rakic, P. and Rubenstein, J.L.R., eds), pp. 187210, Elsevier Academic Press 62 Joyner, A. et al. (2000) Otx2, Gbx2 and Fgf8 interact to position and maintain a mid-hindbrain organizer. Curr. Opin. Cell Biol. 12, 736741 nez, S. (2001) The isthmic organizer and brain regionalization. 63 Mart Int. J. Dev. Biol. 45, 367371 64 Aroca, P. and Puelles, L. (2005) Postulated boundaries and differential fate in the developing rostral hindbrain. Brain Res. Rev. 49, 179190 65 Martinez, S. and Alvarado-Mallart, R.M. (1989) Rostral cerebellum originates from the caudal portion of the so-called mesencephalic vesicle: a study using chick/quail chimeras. Eur. J. Neurosci. 1, 549560 66 Wingate, R.J.T. and Lumsden, A. (1996) Persistence of rhombomeric organisation in the postsegmental hindbrain. Development 122, 2143 2152 67 Lumsden, A. and Krumlauf, R. (1996) Patterning the vertebrate neuraxis. Science 274, 11091115
68 Xu, Q. et al. (2000) Roles of Eph receptors and ephrins in segmental patterning. Philos. Trans. R. Soc. Lond. B: Biol Sci. 355, 9931002 69 Capecchi, M. (1997) The role of hox genes in hindbrain development. In Molecular and Cellular Approaches to Neural Development (Cowan, W.M., ed.), pp. 334355, Oxford University Press 70 OLeary, D.M. and Wilkinson, D.G. (1999) Eph receptors and ephrins in neural development. Curr. Opin. Neurobiol. 9, 6573 n, F. and Puelles, L. (1995) Morphological fate of rhombomeres in 71 Mar quail/chick chimeras: a segmental analysis of hindbrain nuclei. Eur. J. Neurosci. 7, 17141738 72 Farago, A.F. et al. (2006) Assembly of the brainstem cochlear nuclear complex is revealed by intersectional and subtractive genetic fate maps. Neuron 50, 205218 73 Nieuwenhuys, R. (2009) The structural organization of the forebrain: a commentary on the papers presented at the 20th Annual Karger Workshop Forebrain evolution in shes. Brain Behav. Evol. 74, 77 85 74 Nieuwenhuys, R. (2011) The structural, functional, and molecular organization of the brainstem. Front. Neuroanat. 5, 33 75 Wilkinson, D.G. et al. (1989) Segment-specic expression of a zincnger gene in the developing nervous system of the mouse. Nature 337, 461464 76 Krumlauf, R. (1994) Hox genes in vertebrate development. Cell 78, 191201 77 Jensen, P. et al. (2008) Redening the serotonergic system by genetic lineage. Nat. Neurosci. 11, 417419 78 Bulfone, A. et al. (2000) Barhl1, a gene belonging to a new subfamily of mammalian homeobox genes, is expressed in migrating neurons of the CNS. Hum. Mol. Genet. 9, 14431452 79 Geisen, M.J. et al. (2008) Hox paralog group 2 genes control the migration of mouse pontine neurons through Slit-Robo signaling. PLoS Biol. 6, e142 80 Zhu, Y. et al. (2009) SDF1/CXCR4 signalling regulates two distinct processes of precerebellar neuronal migration and its depletion leads to abnormal pontine nuclei formation. Development 136, 19191928 81 Altman, J. and Bayer, S.A. (1996) Development of the Cerebellar System: In Relation to Its Evolution, Structure, and Functions, CRC Press 82 Bloch-Gallego, E. et al. (2005) Development of precerebellar nuclei: instructive factors and intracellular mediators in neuronal migration, survival and axon pathnding. Brain Res. Rev. 49, 253266 83 Ray, R.S. and Dymecki, S.M. (2009) Rautenlippe Redux: toward a unied view of the precerebellar rhombic lip. Curr. Opin. Cell Biol. 21, 741747 84 Vivancos, V. et al. (2009) Wnt activity guides facial branchiomotor neuron migration, and involves the PCP pathway and JNK and ROCK kinases. Neural Dev. 11, 47 novas, B. et al. (2012) Pax7 and Nkx6.1 identify 85 Lorente-Ca differentially originated neuronal populations that migrate into the interpeduncular nucleus. Dev. Biol. 361, 1226 86 Moreno-Bravo, J.A. et al. (2013) Role of Shh in the development of molecularly-characterized tegmental nuclei in mouse rhombomere 1. Brain Struct. Funct. http://dx.doi.org/10.1007/s00429-013-0534-6 87 Zervas, M. et al. (2005) Classical embryological studies and modern genetic analysis of midbrain and cerebellum development. Neural Dev. 69, 101138 88 Inoue, T. et al. (2000) Fate mapping of the mouse prosencephalic neural plate. Dev. Biol. 219, 373383 89 Cobos, I. et al. (2001) Fate map of the avian anterior forebrain at the four-somite stage, based on the analysis of quail-chick chimeras. Dev. Biol. 239, 4667 90 Sanchez-Arrones, L. et al. (2009) Incipient forebrain boundaries traced by differential gene expression and fate mapping in the chick neural plate. Dev. Biol. 335, 4365 91 Sanchez-Arrones, L. et al. (2012) Sharpening of the anterior neural border in the chick by rostral endoderm signaling. Development 139, 10341044 92 Shimamura, K. et al. (1995) Longitudinal organization of the anterior neural plate and neural tube. Development 121, 39233933 93 Shimamura, K. and Rubenstein, J.L. (1997) Inductive interactions direct early regionalization of the mouse forebrain. Development 124, 27092718
577
Opinion
94 Rubenstein, J.L. et al. (1998) Regionalization of the prosencephalic neural plate. Annu. Rev. Neurosci. 21, 445477 95 Anderson, S.A. et al. (1997) Interneuron migration from basal forebrain to neocortex: dependence on Dlx genes. Science 278, 474476 96 Swanson, L. and Petrovich, G.D. (1998) What is the amygdala? Trends Neurosci. 21, 323331 97 Puelles, L. et al. (2000) Pallial and subpallial derivatives in the embryonic chick and mouse telencephalon, traced by the expression of the genes Dlx-2, Emx-1, Nkx-2. 1, Pax-6, and Tbr-1. J. Comp. Neurol. 424, 409438 98 Medina, L. et al. (2004) Expression of Dbx1, Neurogenin 2, Semaphorin 5A, Cadherin 8, and Emx1 distinguish ventral and lateral pallial histogenetic divisions in the developing claustroamygdaloid complex. J. Comp. Neurol. 474, 504523
99 Puelles, L. (2011) Pallio-pallial tangential migrations and growth signaling: new scenario for cortical evolution? Brain Behav. Evol. 78, 108127 100 Northcutt, R.G. and Kaas, J.H. (1995) The emergence and evolution of mammalian neocortex. Trends Neurosci. 18, 373379 101 Witter, M.P. and Amaral, D.G. (2004) Hippocampal formation. In The Rat Nervous System (3rd edn) (Paxinos, G., ed.), pp. 635704, Elsevier Academic Press 102 Palomero-Gallagher, N. and Zilles, K. (2004) Isocortex. In The Rat Nervous System (3rd edn) (Paxinos, G., ed.), pp. 728760, Elsevier Academic Press 103 Heimer, L. et al. (2007) Anatomy of Neuropsychiatry: The New Anatomy of the Basal Forebrain and Its Implications for Neuropsychiatric Illness, Elsevier Academic Press
578
Opinion
Challenges of understanding brain function by selective modulation of neuronal subpopulations

Arvind Kumar, Ioannis Vlachos, Ad Aertsen, and Clemens Boucsein
Bernstein Center Freiburg (BCF), and Neurobiology and Biophysics, Faculty of Biology, University of Freiburg, Freiburg, Germany
Neuronal networks confront researchers with an overwhelming complexity of interactions between their elements. A common approach to understanding neuronal processing is to reduce complexity by dening subunits and infer their functional role by selectively modulating them. However, this seemingly straightforward approach may lead to confusing results if the network exhibits parallel pathways leading to recurrent connectivity. We demonstrate limits of the selective modulation approach and argue that, even though highly successful in some instances, the approach fails in networks with complex connectivity. We argue to rene experimental techniques by carefully considering the structural features of the neuronal networks involved. Such methods could dramatically increase the effectiveness of selective modulation and may lead to a mechanistic understanding of principles underlying brain function. Introduction Structural features of networks on multiple scales The mammalian brain is often referred to as the most complex biological system, not only because of the large number of cells (1011) but mainly because of the multitude of interactions between them. Even if all connections (1015) were known, a system with such complexity could not be successfully treated at all levels of detail simultaneously. Instead, it is advisable to focus on an appropriate spatial scale and to reduce the lower levels to putative functional units, approximating their intrinsic ne structure with the help of fewer variables, if not ignoring them entirely. At the macroscopic scale (1 cm), the brain can be described as a network of different areas or regions. Together with this anatomical modularity, neurophysiological evidence indicates that these different regions in the brain are to some extent specialized to process specic sensory, motor, or cognitive information, and that they are interconnected in a non-trivial fashion [1,2]. At mesoscopic scales (1 cm), brain regions are themselves often anatomically segregated and can be subdivided into layers, subelds, or nuclei. Examples are the hippocampal formation, the interconnected cortical layers or columns, and the networks
Corresponding authors: Kumar, A. (arvind.kumar@biologie.uni-freiburg.de); Boucsein, C. (clemens.boucsein@biologie.uni-freiburg.de). 0166-2236/$ see front matter 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.tins.2013.06.005
in the limbic system such as the basal ganglia and amygdala. Thus, at macro and mesoscopic levels the brain is best considered as a network of networks (NoN). Finally, at microscopic scales (1 mm), brain tissue can be dened as a network of neurons, which can be broadly categorized as excitatory and inhibitory. At this spatial resolution, brain tissue conforms most appropriately to the term network because the interconnected nodes are naturally dened (individual neurons), and their connectivity can, to some extent, be measured experimentally. Below this scale, specialized, spatially extended cells can add additional degrees of freedom through compartmentalization and non-linear dendritic integration [3,4] which might affect computations performed at higher levels. However, although different spatial scales of network organization render the system complex, the presence of specic network motifs (Figure 1) raises the hope that different spatial scales can be described using common principles. Such conceptual reduction of complexity is commonly referred to as model-building. Their mathematical formulation makes them accessible for theoretical studies, opening the opportunity to study their behavior beyond the limitations that often constrain experiments on real brains. Commonly used models typically fall into one of three classes: feedforward models, ring models, and models with recurrent connectivity (Figure 1). The simplest model is the feedforward network, which allows an intuitive understanding of its working principles merely from visual inspection of its graphical description (Figure 2A). In such a feedforward network it is straightforward to characterize the separate contributions of all populations by controlled modulation of their activity. Several networks in the sensory and motor periphery may conform (at least approximately) to such simple architecture regarding their intra- and inter-network connectivity hence they can be treated as effectively feedforward. Examples include cortico-striatal interactions as well as many networks in the sensory and motor periphery, and the hippocampus with its largely feedforward, trisynaptic pathway. In fact, at short time-scales (such that the neuronal activity cannot travel over the full loop), even the entorhinal cortex-hippocampal and the thalamo-cortical loop (thalamus and layer IV) can be studied as an effectively feedforward network. Finally, particular random recurrent networks can also be reduced to an effective feedforward structure [5].
579
Opinion
(A) (D)
successful approach to networks that show a slightly more complex connectivity structure beyond simple feedforward circuitry. Limitations of the selective activity modulation approach Beyond the rst stages of sensory processing or the penultimate stages of motor processing, most networks in the brain cannot be approximated by a feedforward structure. Higher brain areas exhibit more recurrency for which it is non-trivial to reveal the specic activity patterns that implement a presumed function and to identify the elements involved. To demonstrate the problems that may occur in recurrent networks, we converted our feedforward network into a recurrent network by adding one more connection from the third to the second population (Figure 2A,B). In this model it is necessary to modulate the activity of P2 and P3 both separately and simultaneously to understand their functional roles in the network. In the case of more interacting neuronal populations or neurons (both referred to as nodes in the sequel), modulating the activity of single nodes within the network yields only limited information about network function [6,7] when activity in one node inhibits a second and excites a third, and these in turn are coupled and feed back onto the rst, even brief consideration of the network behavior reveals its complexity. Thus, for a network of ve interacting nodes, to assess the functional contributions of its elements and their possible interactions it would be necessary to modulate the activity of one, two, three, and four nodes simultaneously, and to record the corresponding network activity in each case. In general, for a network of N nodes, we would need to selectively modulate all single nodes, and all combinations of two to (N 1) nodes, separately to identify the relevant subset(s) of nodes (cf. [8]). For a network of N interacting elements, the number of all possible subsets of nodes is given by Bells number [9]: BN
N 1 X
Feedforward e.g., rena to LGN Inhibitory recurrent network e.g., striatum, central amygdala (B) (E)
Apparent feedforward e.g., entorhinal cortex, and hippocampus No exc. recurrence e.g., CA1 (C) (F)
Recurrent exc. inh. network e.g., CA3, neocorcal layers
Fully recurrent
Figure 1. Organization of neuronal networks at different spatial scales. (AC) Neuronal network at microscopic scales (1 mm). There are three main motifs at this scale: (A) a recurrent network composed solely of inhibitory neurons; for example, the striatum or the central amygdala. The neurons are maintained in their spiking state by excitation from other networks. (B) A recurrent network composed of both excitatory and inhibitory neurons but with little or no recurrent connectivity among excitatory neurons; for example, the CA1 subfield of the hippocampus. (C) A recurrent network with both excitatory and inhibitory neurons and mutual connectivity between the two populations; for example, the individual layers of the neocortex and the CA3 subfield of the hippocampus. (DF) Beyond the scale of 1 mm, networks in the brain are in fact networks of networks (NoN). These NoNs can be classified into three different categories as follows. (D) Feedforward connection between two networks is the simplest NoN motif. Such motifs are most common in the sensory and motor periphery. (E) Next, NoNs such as the entorhinal and hippocampus loop are effectively feedforward, although there may be multiple projections from one field to other fields (e.g., entorhinal cortex connects to dentate gyrus, CA3, and CA1). (F) Finally, in fully recurrent NoNs the constituent neuronal networks (nodes) form both feedforward and feedback connections. Such network motifs are commonly observed both at anatomical (e.g., the inter-layer connectivity in the neocortex) and functional levels (e.g., the network of various subnetworks involved in visual information processing). Abbreviations: exc., excitatory; exc. inh., excitatory and inhibitory; LGN, lateral geniculate nucleus.
N 1! Bk k!N 1 k! k0
[1]
With recent improvements in brain stimulation technology it has become possible to modulate neuronal networks and/or groups of neurons with unprecedented specicity. Because selective stimulation of neuronal networks has been very informative regarding the functional interactions in early sensory and peripheral motor circuits (essentially feedforward structures), it seems only natural to use these tools in the same approach to other brain areas as well, and to all levels of structural detail, even down to the single cell level. However, several considerations concerning the dynamics of network activity reveal fundamental problems in extending this seemingly
580
Unfortunately, BN grows faster than exponentially with N. For a system consisting of six populations (the minimal population size to study thalamo-cortical interactions), B6 = 203, and for a system with ten populations, B10 = 115 975. Thus, as the number of interacting nodes in a network grows, the number of node combinations that need to be modulated rapidly becomes prohibitively large. For example, even considering only ve layers of a neocortical column (those containing the majority of all cell bodies), to comprehend the contribution of each layer to a particular function we would need B5 = 52 different combinations of single and multi-layer stimulations. Obviously, to treat neocortical networks that way is an oversimplication, and ignores that a large fraction of the inputs originate from the surround [10]. Moreover, interlayer connectivity [11] implies that activity dynamics and layer-specic responses are inuenced by the activity of various neuron types and networks in different layers [12,13]. Thus, modulation of a specic neuron type in a
Opinion
(A)
P1 1
P2 2
(C) P3 3 P1 1 P2 2
(B) P1 1 P2 2 P3 3
P5 5 P4 4
P3 3
Figure 2. Only simple networks of interacting neuronal populations can be studied using selective modulation of individual populations. (A) In a feedforward network it is straightforward to characterize and understand the contribution of each population onto the next. (B) Any deviation from the strict feedforward motif requires control of the activity of more than one population. In this three-population network we need to stimulate, in addition to single populations, pairs of populations to understand network function. (C) In a recurrent network of networks (NoN) (Figure 1D) with N populations, we need BN (Equation 1) different combinations of simultaneous stimulation of multiple populations to understand the impact of one population upon the others. Unfortunately, the number BN grows with N faster than exponentially. In the toy example with five populations, this requires 52 different stimulation patterns involving 15 populations. Abbreviation: l15, firing rates of respective neuron populations P1P5.
given layer [14] is not sufcient to understand the dynamics [12] and stimulus response [13] of a cortical column. Indirectly, this issue was already encountered when Lee et al. [15] attempted to extract the contribution of specic pyramidal neurons to the fMRIBOLD (blood oxygen leveldependent) signal, a problem complicated even more by recurrent inter- and intra-layer connections [16]. This combinatorial problem becomes even more intractable when nodes interact in a non-linear fashion (as is most often the case in brain networks) or when couplings are dynamic and activity-dependent. In both cases, one would need to linearize the system and perform the experiments and associated analyses around a specic operating point for each of which Bells number applies and, hence, to understand the full system, Bells number would
need to be multiplied by the number of operating points chosen. Finally, we note that most experiments thus far rely on steady-state responses upon stimulation of one or more nodes. However, when dealing with interactions among dynamical systems it is important to consider both transient and steady-state responses to identify the functional subset of the network because steady-state responses alone can be misleading [17]. Taken together, these considerations demonstrate that, even if a thorough understanding of the functional network mechanisms is not mandatory (e.g., in medical applications), the sheer number of possibilities to affect network behavior by selective modulation of its elements renders a successful outcome of such experiments rather unlikely. Common techniques for selective activity modulation Assuming that the above-mentioned combinatorial issues with selective modulation could be resolved (see below for new perspectives), another important issue is the availability of experimental methods for selectively modulating the activity of neuronal networks. Deciding which methods to apply for studying a given network partly depends on the research goals: if we wish to understand thoroughly the working principles underlying both, the function and the functioning of the network, we need to identify the functional role of each element and characterize their interactions. However, for the technical task of controlling of prosthetic devices by brain activity, local eld potentials (LFPs) can be utilized [18], even though the mechanisms underlying their generation are still under debate [19] and the LFP is unlikely to affect brain activity (but how LFP might inuence spiking see [20,21]). The earliest successful selective modulation experiments probably date back to early Greco-Roman times when Galen (AD 129199) discovered that mental impairments of gladiators were associated with specic head injuries. Since those early, involuntary brain lesion studies, experimenters have sought to rene loss-of-function approaches and, indeed, lesions of brain structures by targeted ablations at different levels of detail have provided important insights into the functional organization of
581
2 3
4
Figure 3. Scheme of the dynamical space of a five-population network of networks (NoN). The population activity represents the average activity of neurons in the given neuronal population. In a dynamical system description, the function of the NoN, in an abstract sense, can be described as the transition from one state to another (i.e., control of the network activity dynamics). Here the states are schematically indicated by colored circles; arrows mark the transitions between states. Indeed, often more than one transition from any given state is possible. Controllability provides the necessary criteria to determine whether a system is controllable; in other words, whether it can be steered from one dynamical state to another in finite time. Abbreviation: l15, firing rates of different neuron populations.
Opinion
(A) 12 10 Populaon acvity (sp/s) 8 6 4 2 (B)
12 10 8 6 4 2 0 500 550
Populaon acvity (sp/s)
100
Key:
Out-degree k-shell-out
600
650
700
Network ID 2000 2500 3000
50
0 0 0 500 1000 1500 Out-degree 0.0 0.2 0.4

0.6
0.8
1.0
Figure 4. Estimation of embeddedness in numerical simulations of spiking neuronal networks. (A, inset) Network response (peri-stimulus time histogram, PSTH) for identical stimulation of 30 different subpopulations of 250 neurons each in an example network. (A) We estimated the sum of the PSTHs for 100 different networks (see [42] for details) upon stimulation of different subpopulations (250 neurons). Two networks with small-world properties are highlighted (dark blue, orange dots). The random networks which do not show much out-degree variance are indicated by triangles. The network response increases with the out-degree, but the same out-degree can also give different responses. (B) Average correlation coefficient (r, sorted) between the sum of PSTHs as a function of out-degree and k-shell-out indices. Both these measures predict the network response; however, neither alone describes it completely. (Figure adapted from [42]). Abbreviation: sp/s, spikes per second.
the CNS [22]. First, pieces of brain tissue were physically removed and, later, silenced by injection of toxic chemicals or local tissue cooling (Table 1). A major renement in selective modulation came with electrical stimulation. Since 1870 [23] it is used both to identify the function of brain areas and as a therapy to intervene with aberrant activity dynamics associated with brain disorders. Moreover, electrical stimulation also allows activation rather than silencing of nerve tissue. However, the effects of electrical brain stimulation (including behavioral effects) are generally not well understood, partly because they may involve a multitude of pathways [24]. All these classical approaches affect the activity of more or less the entire cell population in a locally conned volume of brain tissue, most likely (through axonal stimulation) together with cells downstream and upstream. This implies that they operate on multiple levels simultaneously, involving a hierarchy of networks. With recent developments in optogenetic methods these caveats have been largely eliminated, leading to an unprecedented richness of tools for selective activation/inactivation. Transgenic or viral transfection with light-gated ion channels now allow a specic group of neurons to be stimulated while recording neuronal activity without stimulus artifacts [25,26]. Thus, it seems only natural to combine the formerly successful modulation approaches with these new capabilities. In the light of the aforementioned theoretical considerations, however, we should appreciate that the simple modulation approach will rarely (only in structures reducible to feedforward networks) lead to reliable guidelines for clinical intervention, let alone to a thorough understanding of the neuronal mechanisms involved. What is a network, and how can its function be investigated? Dening networks Even assuming we could deal with the combinatorial explosion associated with selective modulation, another
582
fundamental issue needs to be resolved: how to dene the network which should be stimulated or whose function needs to be understood? Classical methods used to modulate brain activity activate/inactivate cells in a spatially localized fashion. Hence, it seems intuitive to consider anatomically proximate neurons as a functional unit. Indeed, neighboring neurons often appear to be responsive to similar stimulus features, or are co-active during similar tasks as expressed in the concept of functional maps. However, experimental data show that neurons involved in cognitive and motor functions are distributed throughout the brain [27]. Likewise, primary sensory areas can be modulated by other stimulus modalities [2831], and spatial grouping of neurons with similar stimulus preferences does not seem to be a general feature of cortical networks [32]. The applicability of proximity-based grouping breaks down entirely if cellspecic markers are employed, as in many optogenetic approaches. Here, only a subpopulation of cells within a given volume is modulated based on the expression of particular biomarkers. At the level of such subpopulations, feedforward circuits have been identied [33], but the respective cells are also involved in other circuits within the same tissue volume and, thus, modulating their activity may have confounding effects. In addition, it can be problematic to demarcate functionally specialized neuronal subpopulations from the outset. Various classication schemes based on neuron location, morphology, gene expression and spike patterns have been suggested [3437]. Undoubtedly, many of these features will relate to the functional role of the respective populations: activating excitatory cells has very different effects on the network than activating inhibitory neurons. Also, applying different criteria may lead to overlapping boundaries between putative neuron types, and differentiation is still limited for key neuronal populations. For instance, the subdivision of neocortical pyramidal cells based on their ring pattern seems to be of limited
Opinion
Table 1. Selective activity modulation approaches
Stimulation method Electrical microstimulation Chemical injection, e.g., neurotransmitter agonists and antagonists Magnetic stimulation (e.g., transcranial magnetic stimulation, TMS) Cryogenic (cooling of brain tissue) Optogenetic stimulation Temporal resolution Good Poor
Spatial resolution Poor, because axons from distal regions can also be stimulated Depends on the diffusion of the chemical Poor, because axons from distal regions can also be stimulated Poor, because axons from distal regions can also be affected Moderate. Depends on the penetration of light and expression pattern of the optogenetic tools. Potential for improvement by two-photon techniques
Specicity Poor. To some extent it is possible to select between stimulating somata and axons [62] Good
Good
Poor
Good Good. However, thus far all transfected neurons are stimulated simultaneously [63]
Poor Good. Even specic neurons and their components can be targeted
Experimental methods for selective modulation of the activity of particular parts of the brain have been developed for various spatial and temporal scales.
usefulness in specic network types: propagation of spiking activity in feedforward networks is not inuenced by the type of neuron model, and feedforward networks with integrate-and-re neurons [38,39] behave similarly to those with detailed neuron models [40]. Likewise, in a study of the response of different types of neuron models (ring-rate model, conductance-based leaky-integrateand-re neurons) to higher-order input activity correlations, the input statistics proved much more important than the neuron type [41]. At the microscopic scale, it is instructive to investigate how a given neuron inuences local and downstream network activity, given its other properties (morphology, gene expression, ring pattern, neurotransmitter type) [42]. Recent experiments suggest that neurons indistinguishable by existing classication schemes might form functional subpopulations [43]. Activity-dependent labeling showed that a comparatively small group of neocortical pyramidal cells might provide the major background drive in their respective networks [44]. Similarly, long-range projection targets of cortical output cells are predicted by their intracortical connectivity [45], suggesting that it might be informative to classify them by their afferents and efferents. The density of outgoing projections, together with their overall activity, could be combined into a single variable, quantifying how strongly a cell (cell group) inuences the network activity. This, together with the type of neurotransmitter and, to some extent, the synaptic properties, may be used to dene the embeddedness of a neuron (Box 1) [42]. Selectively modulating individual neurons and recording the network response allows the effective embeddedness of a neuron in vivo to be estimated experimentally. In addition, selectively visualizing the presynaptic sources [46,47] or postsynaptic targets [48] of a neuron may provide useful insights into its structural embeddedness. The structural embeddedness of a neuron in its local microcircuit may also be estimated by juxtacellular recording in vivo and labeling afterwards [49]. Next to these indirect methods, new approaches for measuring neuronal embeddedness need to be established. Indeed, optogenetic tools rank among the most promising candidates in this respect.
Controllability as a concept for selective modulation in complex networks When considering the brain as a dynamical system, ring rates of neurons or average population rates of participating networks are often used as dynamic variables. Traditional approaches such as eigenvalue or Schur decomposition, transforming a network into a new system with only selfinteractions or feedforward interactions, respectively, are extremely successful in helping to understand physical systems. However, it has thus far not been possible to map these concepts to experimental studies of neuronal networks [5]. Although it could be held that the ultimate goal of neuroscience is to understand the functional role of each neuronal network and neuron type in the brain, there are fundamental problems with this. We discuss here a new
Box 1. Embeddedness
The concept of embeddedness was initially coined for socioeconomic networks to understand the effect of social relations on economic decision-making [64]. Within the context of neuroscience, embeddedness measures the impact of the spiking activity of a neuron on the spiking activity of the surrounding network. It can be experimentally estimated by stimulating a neuron and counting the total number of extra spikes in the network [65] (Figure 4). When a neuron is activated, it sends spikes to its postsynaptic neurons; some of which will produce spikes and send those to their postsynaptic neurons, and so on. Thus, to estimate the impact of activating a neuron we need to know all possible direct and indirect (involving multiple synapses) paths from the stimulated neuron to other neurons in the network. That is, the sum of the powers of the connectivity matrix (SNAN) can be used to estimate the number of extra spikes induced by stimulating a neuron or a group of neurons [66]. When the embeddedness is estimated using the connectivity matrix, we refer to it as structural embeddedness. Although the connectivity matrix is an important determinant, so far it has not been possible to associate a particular graph property alone with embeddedness because synaptic and cellular properties, ongoing activity, neuromodulators, etc. also shape the impact of a neuron on the dynamics of the network. The experimental estimate of extra spikes implicitly includes all these other factors as well. When the effective connectivity matrix Aeff (measured from the neuronal activity) is known, the effective embeddedness could be estimated using the powers of Aeff. Finally, graph-theoretical measures such as out-degree, k-shell index, eigenvalue centrality etc. are also related to the embeddedness (Figure 4). For details see [42].
583
Opinion
Box 2. Controllability
Consider a network of neurons or networks: Il Al BU j l [I]
where l is the column vector of ring rates of the N neurons (neuronal populations in an NoN), A is the connectivity matrix (size N N), representing the connectivity among the neurons or the neuronal populations, respectively, I is the identity matrix describing the fact that the activity in a network would change in the absence of inputs from other nodes and external sources, U (size R 1) represents the R external inputs, B is the connectivity of the neuronal populations with the external inputs (size N R), and j is the noise in each of the neuronal populations. All activity states of an N-node network can be described in an Ndimensional space (Figure 3). Controllability determines whether a linear system (Equation I) allows a transition from one dynamical state to another in finite time [50]. The N NR controllability matrix (C) is defined as: C B AB A2 B . . . AN1 B [II]
For a fully controllable system, the matrix C should have a full rank, in other words, Rank(C) = N. This ensures that all dynamical states of the system are accessible. Because the exact values of the matrices A and B are not known for most neuronal systems, C cannot be evaluated. However, the notion of controllability can be extended to the concept of structural controllability [51]. To estimate the structural controllability matrix CS, all non-zero entries in the matrices A and B are replaced with 1 s, reducing the connectivity matrices to binary adjacency matrices [67]. If Rank(CS) = N, the system is controllable [51] and the controllability of a network can be determined, even if the exact values of the connection strengths are not known. Furthermore, the concepts of driver nodes [52] and power dominant set [53] allow us to calculate the minimal set of nodes that need to be stimulated to control the network dynamics and suffices to visit all activity states of the network. Driver nodes: the set of the minimum number of nodes to achieve full control over the network dynamics. To identify the driver nodes we need to find the maximum matching in the network, which refers to the maximum set of links that do not share starting or end nodes [52]. A node is matched if a link in the maximum matching set points towards it; otherwise, it is unmatched. The unmatched nodes are the driver nodes that need to be directly stimulated to control the network dynamics. For instance, in Figure 2C the first node P1 is unmatched. Power dominant set (PDS): when the nodes have self-couplings (Figure 2C, dotted arrows), every node could become a driver node [53]. For such systems the control nodes are the minimum number of nodes from which the rest of the nodes are only one link away [53]. In Figure 2C, the nodes P1, P3, P5 or P1, P2, P4 constitute the PDS.
paradigm based on the concept of controllability (Box 2), which might be a promising candidate to assist in understanding the function and dynamics of neuronal networks. Interestingly, in this paradigm we could potentially avoid the combinatorial explosion. Based on the dynamical description of neuronal network activity (Equation I in Box 2), the function of the brain could be considered as the ability to nd desired trajectories of the network activity state (Box 2) because different behavioral states can be mapped onto these activity states. Following this interpretation, a more pertinent question in understanding brain function would be to know how the overall system dynamics could be controlled, rather than to determine how one particular neuron population affects the activity of the others. However, even in this more utilitarian view, the problem still arises concerning which specic nodes should be
584
stimulated, and how to steer the network to a desired state, even when the connectivity matrix A is known. A trivial, but biologically non-implementable, solution to control the network dynamics is to stimulate all nodes externally. By contrast, a systematic search for the set of nodes to stimulate for controlling the system leads to the afore-described combinatorial explosion. Engineers have successfully applied the concept of controllability to predict whether a given system can be steered from one dynamical state to another in nite time [50] (Box 2). For a dynamical system, controllability depends on the matrices A and B. Fortunately, in the absence of exact values of A and B, the notion of structural controllability (Box 2) [51] can be effectively used to determine the controllability of a system, and it is possible to identify a minimum set of nodes, the stimulation of which is sufcient to control the full network state [52]. Thus, equipped with the knowledge of A, we can bypass the brute force approach and identify the driver nodes whose stimulation can control the dynamical repertoire of the neuronal network. The estimation of the driver nodes explicitly assumes that the nodes/neurons do not have self-connections, which is true for most neurons. However, at the system dynamics level the self-connections capture the intrinsic dynamics of the nodes in other words, node dynamics in the absence of inuences from other nodes and neuron refractoriness. Ignoring the intrinsic dynamics implicitly means that the activity state of the node does not change when there are no inuences from other nodes (i.e., l = 0 or innite time constant) [53]. Typically, the activity of biological neurons/networks decays to zero in the absence of external inputs. Thus, the absence of self-connections can only be justied in restricted cases, for example, when the dynamics of the constituent networks of a NoN exhibit attractor dynamics and can exhibit asynchronous-irregular self-sustained activity [54]. Furthermore, it has been shown that the power dominant set (PDS) (Box 2) is the set of driver nodes when intrinsic dynamics and self-connections are included in the dynamics [53]. In this context it is important to note that hubs, which are typical choices for external brain stimulation, usually are not part of the driver nodes and PDS [52]. Extending the framework of controllability to neuronal networks, we argue that the concepts of driver nodes [52] and the PDS [53] have emerged as interesting alternatives for choosing the nodes to be stimulated. Once the appropriate set of driver nodes is known, the controllability Gramian [55] can be used to determine the input pattern needed to drive the system to a desired state. Because the number of neurons in a set of driver nodes or PDS is typically substantially less than the number of nodes in the network, the total number of stimulations required will always be signicantly less than Bells number. However, the framework of controllability still requires simultaneous manipulations of multiple nodes. Therefore, we will certainly need technology that is capable of varying the activity of more than few nodes independently. In addition to the notion of controllability, other graph-theoretical measures [56,57] may also be used to make educated choices of the stimulation sites.
Opinion
Concluding remarks The considerations outlined above identied two major pitfalls when applying selective activity modulation to understanding brain function: (i) the need for simultaneous control of the activity of multiple neuronal populations leads to a severe combinatorial explosion; (ii) using gene expression, ring patterns, morphological or spatial location criteria alone to dene functional groups of neurons is highly ambiguous. Interestingly, in other scientic disciplines it has been realized that the seemingly intuitive approach of selective activity modulation of parts of a complex system to study its function can be misleading. For instance, systems biologists now agree that genes themselves act as a complicated network of dependencies, and that single-gene/single-function mapping is the exception rather than the rule [58]. We argue that the dynamic and nonlinear nature of interactions within large neuronal networks in the mammalian brain makes it unlikely that single-neuron-type/single-function relationships hold for these systems, either. Recent cross-modal studies even suggest that the separation between areas concerning their involvement in different functions may be less clear than previously thought [2831]. Instead, brain function is more likely to be the result of coordinated activity distributed over multiple brain areas. Assuming that brain function can be understood as an interaction of subnetworks of different types of neurons ignores this insight, as well as the rich repertoire of activity dynamics that can be displayed by neuron subtypes depending on the activity of the embedding network [59]. Finally, the approach of dening cell populations based on currently available markers can only be a starting point and needs to be complemented by more sophisticated criteria. Embeddedness, graph-theoretical measures, and criteria derived from linear and nonlinear control theory [60] may be promising candidates in this respect. Future directions We have shown that knowing the network connectivity of neurons and neuronal populations can help in choosing the most appropriate network node(s) for activity modulation to help understand the function and dynamics of networks in the brain. Instead of needing to go painstakingly through all possible combinations, such approaches (e.g., based on controllability) could potentially evade some of the inherent limitations of selective activity modulation approaches. Indeed, recent developments of optogenetic tools provide interesting avenues in this direction. Hence, future research on the development of experimental tools should be driven by the following goals: (i) To dene tools that can be used to control the activity of multiple neuronal populations simultaneously. (ii) To identify biomarkers for classifying neurons according to their recent spiking activity [43,61]. (iii) To extract the functional and anatomical connectivity of different types of neurons and neuronal populations, and to use these to measure their embeddedness and the controllability of the network [42]. (iv) To design selective activity modulation experiments based on the network-related features extracted with the above-mentioned approaches.
In parallel, it is also important to develop new computational models, where selective activity modulation-based experimental approaches can be tested under fully controlled conditions. As a starting point, we provide an online system (Neural System Prediction and Identication Challenge, nuSPIC) for extracting the function of relatively small networks of spiking neurons designed to perform a specic task. Its web interface provides several tools for performing a variety of experiments, including selective activity modulation. We believe that nuSPIC provides useful models for calibrating the efcacy of experimental approaches and to help interpret their results. Once we have tools to control selectively the activity of multiple neuronal populations simultaneously, and the knowledge of their mutual connectivity, we will be in a position to develop a deeper understanding of their contribution to brain function.
Acknowledgments
We thank Nikos Logothetis, Moshe Abeles, Stefan Rotter, and Yexica Aponte for helpful discussions and comments on the initial version of the manuscript. Partial funding by German Federal Ministry of Education and Research (BMBF) grant 01GQ0420 to BCCN Freiburg, 01GQ0830 to the Bernstein Focus Neurotechnology (BFNT) Freiburg/Tuebingen, and the BrainLinks-BrainTools Cluster of Excellence funded by the German Research Foundation (DFG) grant EXC 1086, is gratefully acknowledged.
References
1 Fellman, S.J. and Van Essen, D.C. (1991) Distributed hierarchical processing in the primate cerebral cortex. Cereb. Cortex 1, 146 2 Modha, D.S. and Singh, R. (2010) Network architecture of the longdistance pathways in the macaque brain. Proc. Natl. Acad. Sci. U.S.A. 107, 1348513490 3 Murayama, M. et al. (2009) Dendritic encoding of sensory stimuli controlled by deep cortical interneurons. Nature 457, 11371141 4 Xu, N.L. et al. (2012) Nonlinear dendritic integration of sensory and motor input during an active sensing task. Nature 492, 247251 5 Goldman, M.S. (2009) Memory without feedback in a neural network. Neuron 61, 621634 6 Johannesma, P.I.M. and Aertsen, A. (1987) Conservation and dissipation in neurodynamics. In Physics of Cognitive Processes (Caianiello, E.R., ed.), pp. 228257, World Scientic Publishing 7 Johannesma, P.I.M. et al. (1986) From synchrony to harmony: Ideas on the function of neural assemblies and on the interpretation of neural synchrony. In Brain Theory (Palm, G. and Aertsen, A., eds), pp. 2547, Springer 8 Koch, C. (2012) Modular biological complexity. Science 337, 531532 9 Rota, G.-C. (1964) The number of partitions of a set. Am. Math. Monthly, 71, 498504 10 Boucsein, C. et al. (2011) Beyond the cortical column: abundance and physiology of horizontal connections imply a strong role for inputs from the surround. Front. Neurosci. 5, 32 11 Binzegger, T. et al. (2004) A quantitative map of the circuit of cat primary visual cortex. J. Neurosci. 24, 84418453 12 Kremkow, J. et al. (2007) Emergence of population synchrony in a layered network of the cat visual cortex. Neurocomputing 70, 20692073 13 Wagatsuma, N. et al. (2011) Layer-dependent attentional processing by top-down signals in a visual cortical microcircuit model. Front. Comput. Neurosci. 5, 31 14 Olsen, S.R. et al. (2012) Gain control by layer six in cortical circuits of vision. Nature 482, 4752 15 Lee, J.H. et al. (2010) Global and local fMRI signals driven by neurons dened optogenetically by type and wiring. Nature 465, 788792 16 Logothetis, N.K. (2010) Bold claims for optogenetics. Nature 468, E3E4 17 Janusonis, S. (2012) Relationships among variables and their equilibrium values: caveats of time-less interpretation. Biol. Rev. Camb. Philos. Soc. 87, 275289
585
Opinion
18 Mehring, C. et al. (2003) Inference of hand movements from local eld potentials in monkey motor cortex. Nat. Neurosci. 6, 12531254 19 Nunez, P. (2006) Electric Fields of the Brain: the Neurophysics of EEG. Oxford University Press 20 Anastassiou, C.A. et al. (2011) Ephaptic coupling of cortical neurons. Nat. Neurosci. 14, 217223 hlich, F. and Mccormick, D.A. (2010) Endogenous electric elds may 21 Fro guide neocortical network activity. Neuron 67, 129143 22 Luria, A.R. (1973) The Working Brain: An Introduction to Neuropsychology. Allen Lane 23 Fritsch, G. and Hitzig, E. (1870) Uber die elektrische Erregbarkeit des Grosshirns. Arch. Anat. Physiol. 37, 300332 24 Logothetis, N.K. et al. (2010) The effects of electrical microstimulation on cortical signal propagation. Nat. Neurosci. 13, 12831291 25 Boyden, E.S. et al. (2005) Millisecond-timescale, genetically targeted optical control of neural activity. Nat. Neurosci. 8, 12631268 ck, G. and Kevrekidis, I.G. (2005) Optical imaging and control 26 Miesenbo of genetically designated neurons in functioning circuits. Annu. Rev. Neurosci. 28, 533563 27 Cisek, P. and Kalaska, J.F. (2010) Neural mechanisms for Interacting with a world full of action choices. Annu. Rev. Neurosci. 33, 269298 28 Bizley, J.K. et al. (2007) Physiological and anatomical evidence for multisensory interactions in auditory cortex. Cereb. Cortex 17, 21722189 29 Ghazanfar, A.A. et al. (2005) Multisensory integration of dynamic faces and voices in rhesus monkey auditory cortex. J. Neurosci. 25, 50045012 30 Iurilli, G. et al. (2012) Sound-driven synaptic inhibition in primary visual cortex. Neuron 73, 814828 31 Kayser, C. et al. (2008) Visual modulation of neurons in auditory cortex. Cereb. Cortex 18, 15601574 32 Horton, J.C. and Adams, D.L. (2005) The cortical column: a structure without a function. Philos. Trans. R. Soc. Lond. Ser. B: Biol. Sci. 360, 837862 33 Kremkow, J. et al. (2010) Functional consequences of correlated excitatory and inhibitory conductances in cortical networks. J. Comput. Neurosci. 28, 579594 ki, G. (1996) Interneurons of the hippocampus. 34 Freund, T.F. and Buzsa Hippocampus 6, 347470 35 Luo, L. et al. (2008) Genetic dissection of neural circuits. Neuron 57, 634660 36 Markram, H. et al. (2004) Interneurons of the neocortical inhibitory system. Nat. Rev. Neurosci. 5, 793807 37 Migliore, M. and Shepherd, G.M. (2005) An integrated approach to classifying neuronal phenotypes. Nat. Rev. Neurosci. 6, 810818 38 Diesmann, M. et al. (1999) Stable propagation of synchronous spiking in cortical neural networks. Nature 402, 529533 39 Kumar, A. et al. (2010) Spiking activity propagation in neuronal networks: reconciling different perspectives on neural coding. Nat. Rev. Neurosci. 11, 615627 40 Shinozaki, T. et al. (2007) Controlling synre chain by inhibitory synaptic input. J. Phys. Soc. Jpn. 76, 044806 41 Kuhn, A. et al. (2003) Higher-order statistics of input ensembles and the response of simple model neurons. Neural Comput. 15, 67101
42 Vlachos, I. et al. (2012) Beyond statistical signicance: implications of network structure on neuronal activity. PLoS Comput. Biol. 8, e1002311 43 Liu, X. et al. (2012) Optogenetic stimulation of a hippocampal engram activates fear memory recall. Nature 484, 381385 44 Yassin, L. et al. (2010) An embedded subnetwork of highly active neurons in the neocortex. Neuron 68, 10431050 45 Brown, S.P. and Hestrin, S. (2009) Intracortical circuits of pyramidal neurons reect their long-range axonal targets. Nature 457, 11331136 46 Bock, D.D. et al. (2011) Network anatomy and in vivo physiology of visual cortical neurons. Nature 471, 177182 47 Rancz, E.A. et al. (2011) Transfection via whole-cell recording in vivo: bridging single-cell physiology, genetics and connectomics. Nat. Neurosci. 14, 527532 48 Komiyama, T. et al. (2010) Learning-related ne-scale specicity imaged in motor cortex circuits of behaving mice. Nature 464, 11821186 49 Burgalossi, A. et al. (2011) Microcircuits of functionally identied neurons in the rat medial entorhinal cortex. Neuron 70, 773786 50 Kalman, R.E. (1963) Mathematical description of linear dynamical systems. J. Soc. Ind. Appl. Math. Ser. A 1, 152192 51 Lin, C-T. (1974) Structural controllability. IEEE Trans. Automatic Control 19, 201208 52 Liu, Y-Y. et al. (2011) Controllability of complex networks. Nature 473, 167173 53 Cowan, N.J. et al. (2012) Nodal dynamics, not degree distributions, determine the structural controllability of complex networks. PLoS ONE 7, e38398 54 Kumar, A. et al. (2008) The high-conductance state of cortical networks. Neural Computation 20, 143 55 Rugh, W.J. (1996) Linear System Theory (2nd edn), Prentice-Hall 56 Dorogovtsev, S.N. and Mendes, J.F.F. (2003) Evolution of Networks: From Biological Nets to the Internet and WWW. Oxford University Press 57 Newman, M.E.J. (2003) The structure and function of complex networks. SIAM Rev. 45, 167256 58 Chouard, T. (2008) Beneath the surface. Nature 456, 300303 59 Wang, X-J. (2010) Neurophysiological and computational principles of cortical rhythms in cognition. Physiol. Rev. 90, 11951268 60 Slotine, J-J. and Li, W. (1991) Applied Nonlinear Control. Prentice Hall 61 Garner, A.R. et al. (2012) Generation of a synthetic memory trace. Science 335, 15131516 62 Ranck, J.B. (1975) Which elements are excited in electrical stimulation of mammalian central nervous system: a review. Brain Res. 98, 417440 63 Yizhar, O. et al. (2011) Neocortical excitation/inhibition balance in information processing and social dysfunction. Nature 477, 171178 64 Granovetter, M. (1985) Economic action and social structure: the problem of embeddedness. Am. J. Sociol. 91, 481510 65 London, M. et al. (2010) Sensitivity to perturbations in vivo implies high noise and suggests rate coding in cortex. Nature 466, 123127 66 Pernice, V. et al. (2011) How structure determines correlations in neuronal networks. PLoS Comp. Biol. 7, e1002059 67 Newman, M.E.J. (2010) Graph Theory and Complex Networks: An Introduction. Maarten van Steen
586
Review
Sugar for the brain: the role of glucose in physiological and pathological brain function
Philipp Mergenthaler1, Ute Lindauer2, Gerald A. Dienel3, and Andreas Meisel1
1
Department of Experimental Neurology and Department of Neurology, Center for Stroke Research, NeuroCure Cluster of Excellence, Charite University Medicine Berlin, Berlin, Germany 2 Experimental Neurosurgery, Department of Neurosurgery, TUM-Neuroimaging Center, Technical University Munich, Munich Cluster for Systems Neurology (SyNergy), Munich, Germany 3 Department of Neurology, Department of Physiology and Biophysics, University of Arkansas for Medical Sciences, Little Rock, AR, USA
The mammalian brain depends upon glucose as its main source of energy, and tight regulation of glucose metabolism is critical for brain physiology. Consistent with its critical role for physiological brain function, disruption of normal glucose metabolism as well as its interdependence with cell death pathways forms the pathophysiological basis for many brain disorders. Here, we review recent advances in understanding how glucose metabolism sustains basic brain physiology. We synthesize these ndings to form a comprehensive picture of the cooperation required between different systems and cell types, and the specic breakdowns in this cooperation that lead to disease. Nobody realizes that some people expend tremendous energy merely to be normal. Albert Camus, Notebooks 19421951.
Glossary
Autophagy: an intracellular recycling pathway that can be activated under conditions of metabolic stress to inhibit cell death. It involves the lysosomal degradation of cytoplasmic proteins or entire organelles for catabolic regeneration of nutrient pools [61]. Bloodbrain barrier (BBB): the permeability barrier arising from tight junctions between brain endothelial cells, restricting diffusion from blood to brain. Entry into the brain is limited to molecules that can diffuse across membranes (e.g., oxygen and other gases, or lipid-permeable compounds) or have transporter molecules (e.g., glucose transporters). Neuroactive compounds (e.g., glutamate or adrenalin) in the blood are restricted from entering into the brain. Functional activation: a response by the brain to a specific stimulus (e.g., sensory stimulation) that increases cellular activity and metabolism above the resting and/or baseline value before onset of the stimulus. Brain activation has the same meaning but is a more general term that includes increased activity during abnormal or disease states. Glutamateglutamine cycle: the release of the neurotransmitter glutamate from excitatory neurons, its sodium-dependent uptake by astrocytes, its conversion to glutamine by glutamine synthetase in astrocytes, the release of glutamine and uptake into neurons followed by the conversion to glutamate by glutaminase and its repackaging into synaptic vesicles. Glyceraldehyde-3-phosphate dehydrogenase (GAPDH): a glycolytic enzyme that reduces NAD+ to NADH and converts D-glyceraldehyde-3-phosphate to 1,3bisphospho-D-glycerate, an intermediary metabolite in the generation of pyruvate. Glycolysis: a cytoplasmic pathway for metabolism of one molecule of glucose to produce two molecules of pyruvate, with phosphorylation of 2 ADP to form 2 ATP and reduction of 2 NAD+ to 2 NADH. Cytoplasmic oxidation of NADH can be achieved by conversion of pyruvate to lactate by the LDH reaction or via the MAS (see Figure 2A in main text). The MAS is required to generate pyruvate for oxidation in the TCA cycle, whereas LDH removes this substrate from the cell. Net production of lactate in the presence of adequate levels and delivery of oxygen is sometimes termed aerobic glycolysis, contrasting the massive production of lactate under hypoxia or anoxia (anaerobic glycolysis). Hexokinase (HK): the enzyme catalyzing the first step in glucose metabolism: the irreversible conversion of glucose to Glc-6-P in an ATP-dependent reaction. The brain has different HK isoforms that have specific functions. HKI is the major isoform in brain for the glycolytic pathway; it has a broad substrate specificity and is feedback-inhibited by Glc-6-P. HKII is a minor, hypoxia-regulated isoform in the brain that controls neuronal survival depending on the metabolic state. HKIV (glucokinase, GK) is a minor isoform of hexokinase in the brain that has an important role in glucose-sensing neurons; it is specific for glucose and is not inhibited by Glc-6-P. Ketogenic diet: a diet that has a high fat and low carbohydrate content so that plasma levels of ketone bodies (acetoacetate and b-hydroxybutyrate) increase and serve as alternative oxidative fuel. Metabolic coupling: a synergistic interaction between different cells or cell types in which compounds produced in one cell are used by another cell. Neurovascular unit: groups of neurons, astrocytes, endothelial cells, vascular smooth muscle cells, and pericytes that are involved in local signaling activities, metabolic interactions, and regulation of blood flow. Tricarboxylic acid (TCA) cycle: a mitochondrial pathway for oxidation of pyruvate to produce 3 CO2 and generate FADH2 and NADH that are oxidized via the electron transport chain with conversion of oxygen to water and formation of approximately 32 ATP per glucose molecule. This ATP yield is less than the theoretical maximum due to proton leakage across the mitochondrial membrane.
Glucose metabolism: fueling the brain The mammalian brain depends on glucose as its main source of energy. In the adult brain, neurons have the highest energy demand [1], requiring continuous delivery of glucose from blood. In humans, the brain accounts for approximately 2% of the body weight, but consumes approximately 20% of glucose-derived energy, making it the main consumer of glucose (approximately 5.6 mg glucose per 100 g human brain tissue per minute [2]). Glucose metabolism provides the fuel for physiological brain function through the generation of ATP, the foundation for neuronal and non-neuronal cellular maintenance, as well as the generation of neurotransmitters. Therefore, tight
Corresponding author: Mergenthaler, P. (philipp.mergenthaler@charite.de). Keywords: glucose metabolism; metabolic coupling; apoptosis; brain-body axis; metabolic brain disease. 0166-2236/$ see front matter 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.tins.2013.07.001
587
Review
(E) (D)
Cell fate
Oligodendrocytes
(C)
Neurons Glc Lac
Hypothalamus AgRP
Astrocytes
(A)
uro e s i g ndoc n a rin ls e
POMC
Pericytes
(B)
Liver Gastrointesnal tract Pancreas
GLUT1
Endothelial cells (BBB)
Ne
Glucose
Aerent and eerent vagus
Figure 1. The role of glucose in brain function. Glucose (Glc) is the main source of energy for the mammalian brain. (A) Specialized centers in the brain, including proopiomelanocortin (POMC) and agouti-related peptide (AgRP) neurons in the hypothalamus, sense central and peripheral glucose levels and regulate glucose metabolism through the vagal nerve as well as neuroendocrine signals. (B) Glucose supply to the brain is regulated by neurovascular coupling and may be modulated by metabolismdependent and -independent mechanisms. Glucose enters the brain from the blood by crossing the bloodbrain barrier (BBB) through glucose transporter 1 (GLUT1), and (C) glucose and other metabolites (e.g. lactate, Lac) are rapidly distributed through a highly coupled metabolic network of brain cells. (D) Glucose provides the energy for neurotransmission, and (E) several glucose-metabolizing enzymes control cellular survival. Disturbed glucose metabolism on any of these levels can be the foundation for the development of a large variety of disorders of the brain (see section on Disease mechanisms).
regulation of glucose metabolism is critical for brain physiology and disturbed glucose metabolism in the brain underlies several diseases affecting both the brain itself as well as the entire organism. Here, we provide a comprehensive overview of the functional implications and recent advances in understanding the fundamental role of glucose metabolism in physiological and pathological brain function. Although brain energy metabolism has been investigated for decades, certain aspects remain controversial, in particular in the eld of energy substrate consumption and utilization. It is beyond the scope of this review to resolve these controversies; rather, it is our aim to highlight conicting concepts and results to stimulate discussion in key areas. To this end, we review the bioenergetics of neurotransmission, the cellular composition of a metabolic network, the regulation of cerebral blood ow (CBF), how peripheral glucose metabolism and energy homeostasis are sensed and controlled by the central nervous system (CNS), and the tight regulation of cellular survival through glucose-metabolizing enzymes. Glucose is required to provide the precursors for neurotransmitter synthesis and the ATP to fuel their actions, as well as the energy demands of the brain that are not related to signaling. Cellular compartmentation of glucose transport and metabolism is intimately related to local regulation of blood ow, and glucose-sensing neurons govern the brainbody nutrient axis. Glucose metabolism is connected to cell death pathways by glucose-metabolizing
588
enzymes. Thus, disruption of pathways of glucose delivery and metabolism leads to debilitating brain diseases. We highlight the multifaceted role and complex regulation of glucose metabolism in the CNS as well as the physiological and pathophysiological consequences of balanced and disturbed glucose metabolism (Figure 1). Glucose metabolism: the bioenergetic basis for neurotransmission The largest proportion of energy in the brain is consumed for neuronal computation and information processing [3]; for example, the generation of action potentials and postsynaptic potentials generated after synaptic events (Figure 1D), and the maintenance of ion gradients and neuronal resting potential [1,4]. Additionally, glucose metabolism provides the energy and precursors for the biosynthesis of neurotransmitters (for a comprehensive overview, see [5]). Importantly, astrocytic glycogen seems to be directly relevant for learning [6]. Furthermore, the glycolytic end-product lactate appears to have a role in long-term memory formation [7], although the exact mechanism has not yet been established. Lactate injections [7] alter the intracellular redox state and pH due to co-transport of H+ with lactate, and lactate receptors may also have a role in linking brain energy metabolism and neurotransmission [8,9]. However, oxidative metabolism in both neurons and astrocytes appears to contribute to sustained learning effects after training, and glycogen can supply carbon for synthesis of glutamate during learning [6].
Review
(A) (B)
GLUT
Glc
Glc GLUT1 Cytosol Glc Pentose-phosphate-pathway Glc-6-P PPP Fru-6-P NADPH Glc Glc-6-P NAD+ MAS NADH Pyr NADH NAD+ Release from cell LDH TCA & ETC MAS Acetyl CoA TCA Glu Gln Gal-3-P Pyr Lac MCT2 Glyc Glc Glc-6-P NADP+ Glc GLUT3 MCT1 Lac release from brain Glc GLUT1
Glycolysis
BBB
Glyc (astrocytes)
NALS
Lac Extracellular uid Glu-Gln cycle
ANLS
Lac MCT1, 4 Glu Gln Pyr
TCA
Neuron
Astrocyte
Lac
Oxidave metabolism
Figure 2. Generation of energy in the brain and three models for the fate of lactate derived from glucose metabolism in the brain. (A) Major pathways of glucose metabolism. Hexokinase uses ATP to phosphorylate glucose (Glc) to glucose-6-phosphate (Glc-6-P) in the first irreversible step of the glycolytic pathway. Glc-6-P regulates hexokinase activity by feedback inhibition [19], and it is a branch-point metabolite that has alternative metabolic fates. Glc-6-P can continue down the glycolytic pathway to generate pyruvate that can then be used in mitochondria by oxidative metabolism via the tricarboxylic acid (TCA) cycle. It can also enter the pentose phosphate shunt pathway (PPP) to generate NADPH for management of oxidative stress and precursors for nucleic acid biosynthesis and, in astrocytes, Glc-6-P is a precursor for glycogen. Most of the glucose carbon derived from the PPP re-enters the glycolytic pathway downstream of Glc-6-P. The glycolytic pathway produces a net of 2 ATP per molecule of glucose and oxidation of pyruvate via acetyl coenzyme A (acetyl CoA) in the TCA cycle produces approximately 30 ATP for a total of approximately 32 ATP. Formation of pyruvate from glucose requires regeneration of NAD+ from NADH produced by the glyceraldehyde-3-phosphate dehydrogenase reaction by the malate-aspartate shuttle (MAS). NADH cannot cross the mitochondrial membrane, and the MAS transfers cytoplasmic NADH to the mitochondria, where it is oxidized via the electron transport chain (ETC). When glycolytic flux exceeds that of the MAS or the TCA cycle rate, or during hypoxic or anoxic conditions, NAD+ is regenerated by the lactate dehydrogenase (LDH) reaction that converts pyruvate to lactate. Because intracellular accumulation of lactate would cause reversal of the LDH reaction, lactate must be released from the cell by monocarboxylic acid transporters (MCT). Exit of lactate eliminates pyruvate as an oxidizable substrate for that cell and limits the ATP yield of glycolysis per molecule glucose to two. (B) Three models for the fate of lactate generated in the brain from blood-borne glucose or astrocytic glycogen. The astrocyte-to-neuron lactate shuttle (ANLS) was proposed on the basis of glutamate-evoked increases in glucose utilization and lactate release by cultured astrocytes (reviewed in [29]). In brief, the model states that Na+-dependent uptake of the neurotransmitter glutamate from the synaptic cleft by astrocytes generates a demand for 2 ATP in astrocytes, one to extrude Na+ and one to convert glutamate into glutamine in the glutamateglutamine (GluGln) cycle. The model states that this ATP is generated by the glycolytic pathway and is associated with release of lactate from astrocytes and its uptake by nearby neurons, where it is oxidized. Thereby, astrocyteneuron metabolic coupling is linked with the glutamateglutamine cycle and excitatory neurotransmission. Thus, during brain activation, glycolytic upregulation is stated to occur in astrocytes, with astrocyte-derived lactate providing the major fuel for neurons. The neuron-to-astrocyte lactate shuttle (NALS) is based on the kinetics of glucose uptake into brain cells in response to increased metabolic demand and different model assumptions compared with the ANLS [27]. Here, glucose is predicted to be predominantly taken up into neurons due to their high energy demand and the higher transport rate of the neuronal glucose transporter 3, GLUT3, compared with the astrocytic glucose transporter 1, GLUT1 [16]. Lactate is posited to be generated by neurons and taken up by astrocytes. The lactate release model [5] is based on the observed mismatch between total glucose utilization and oxidative metabolism and measured lactate release from brain during brain activation in vivo. If lactate were produced and locally oxidized, total and oxidative metabolism would be similar in magnitude. However, the increase in oxidative metabolism varies with experimental condition and pathways stimulated, and it is much less than that of total glucose utilization [5]. Astrocytes have a faster and greater capacity for lactate uptake from extracellular fluid, and for lactate dispersal among gap junctioncoupled astrocytes compared with neuronal lactate uptake and shuttling of lactate to neurons [17]. Astrocytic endfeet surround the vasculature, and can discharge lactate to perivascular fluid for efflux from brain.
It has been suggested that action potentials have been rendered highly efcient through evolution [10] and, thus, most of the energy consumed in the brain is used on synaptic activity [3,10,11]. The human cortex alone requires approximately 31023 ATP/s/m3 [1], and the energy expenditure to release one synaptic vesicle is calculated to be approximately 1.64105 molecules ATP [3]. Consequently, a model of energy use in the brain suggests that a considerably larger amount of energy is spent in the gray matter compared with the white matter [12]. In essence, the brain increases its utilization of glucose upon activation [13]. Glucose uptake in the brain: how are neurons and astrocytes fed? Dependence of the brain on glucose as its obligatory fuel derives mainly from the bloodbrain barrier (BBB; see Glossary), and its selective permeability for glucose in the adult brain. Glucose cannot be replaced as an energy source,
but it can be supplemented, as during strenuous physical activity when blood lactate levels are elevated [14] or during prolonged starvation [15] when blood levels of ketone bodies are elevated and BBB monocarboxylic acid transporter (MCT) levels are upregulated. Because entry of neuroactive compounds (e.g., glutamate, aspartate, glycine, or D-serine) into the brain is restricted by the BBB, these compounds must be synthesized from glucose within the brain. The BBB and its transport properties sharply contrast with muscle and liver that do not have tight junctions between their vascular endothelial cells and have different transporter levels for various compounds, enabling these organs to metabolize glucose, monocarboxylic acids, fatty acids, amino acids, and ketone bodies. The large blood-to-brain concentration gradient drives the facilitative transport of glucose across the endothelial membranes via glucose transporter 1 (GLUT1) into extracellular uid (Figures 1B and 2). The steady-state brain
589
Review
tissue glucose concentration is approximately 20% of that in arterial plasma. GLUT1 further mediates glucose uptake from extracellular uid into astrocytes, oligodendroglia, and microglia, whereas GLUT3, which has a higher transport rate than GLUT1, facilitates neuronal glucose uptake (Figures 1C and 2B) [16]. Glucose transport capacity exceeds demand over a wide range, and the higher transport rate of GLUT3 ensures that neurons have sufcient glucose supplies under varying glucose levels and different activity states [5]. Although astrocytes are generally believed to be involved in the uptake and distribution of brain metabolites [3,17,18], modeling predicts that most glucose diffuses from endothelial cells through the gaps between the surrounding astrocytic endfeet, and throughout the extracellular uid to more distant brain cells, facilitating rapid GLUT3-mediated uptake into neurons [16]. However, some glucose may also be taken up into astrocytic endfeet, followed by its diffusion down its concentration gradients to other gap junction-coupled astrocytes, with release to extracellular uid at sites more distant from the capillary [3,17,18]. Local rates of glucose utilization are driven by functional activities (Figure 1D) that consume ATP and generate ADP, which is an obligatory co-substrate for energy-producing reactions. Intracellular glucose is phosphorylated by hexokinase I (HKI) to form glucose-6-phosphate (Glc-6P), thereby trapping the molecule in the cell and, thus, creating a sink that draws more glucose into the cell (Figure 2A). The intracellular glucose pool size is maintained as the net balance between rates of its inux, efux, and metabolism. The Km (half-saturation constant) of HKI for glucose is low [19] and, therefore, HKI can operate at maximal velocity as long as the intracellular glucose exceeds approximately 0.81 mmol/L. Glc-6-P governs HKI activity by feedback inhibition, such that the in vivo activity of HKI in resting, awake brain is only approximately 5% of its maximal capacity measured in vitro. Thus, dis-inhibition of HKI by consumption of Glc-6-P can stimulate HKI ux by up to 20-fold, a capacity that greatly exceeds the four-to sixfold rise in the cerebral metabolic rate of glucose (CMRglc) during seizures and ischemia [20,21]. Glc-6-P is not only metabolized via the glycolytic pathway to generate ATP, but is also the substrate for the pentose phosphate shunt pathway (PPP) that generates NADPH to manage oxidative stress and to synthesize nucleic acid precursors (Figure 2A). Phosphofructokinase is considered to be the major regulator of the glycolytic pathway due to its allosteric regulation by many metabolites (e.g., inhibition by ATP, citrate, H+, and activation by ADP, AMP, fructose-6-P, fructose-1,6,-P2, fructose-2,6,-P2, and ribose-1,5,-P2) that act in concert to integrate the uxes of the glycolytic and tricarboxylic acid (TCA) cycle pathways. Glucose metabolism is also the source for biosynthesis of other compounds required by the brain, including complex carbohydrates that are components of glycoproteins and glycolipids, amino acids, one-carbon donors for methylation reactions, and the supply of neurotransmitter precursors [5,22]. To summarize, CMRglc is controlled in each cell by the rate of ADP production (i.e., ATP demand) and regulation of rate-controlling enzymes by metabolites.
590
In astrocytes, Glc-6-P is the precursor for glycogen, a polymer comprising glucose residues. Glycogen is the only energy reserve in the brain (Figure 2A,B). In normal brain, glycogen turnover occurs at normal glucose levels, consistent with its role as an important local energy buffer for astrocytes, and it is mobilized by functional activation or energy decits [23,24]. Modeling predicts that glycogenolysis reduces astrocytic glucose utilization by maintaining levels of Glc-6-P sufcient to sustain high feedback inhibition of HKI, thereby sparing glucose for neurons [25]. During severe hypoglycemia or aglycemia, low rates of glycogenolysis equivalent to only a few percent of normal glucose utilization rates are sufcient to prolong neuronal functions [5,22]. To maintain glycolytic ux, NADH produced by glyceraldehyde-3-phosphate dehydrogenase (GAPDH) must be oxidized. Regeneration of NAD+ can occur by two mechanisms: the malateaspartate shuttle (MAS) or the lactate dehydrogenase (LDH) reaction. The MAS is required for generation of pyruvate as an oxidative fuel because LDH activity is associated with lactate release (Figure 2A, see gure legend for details). In resting awake brain, most of the glucose is completely oxidized to CO2 and water, and nearly a stoichiometric amount of O2 (i.e., 6 O2 per glucose) is consumed. Due to biosynthetic reactions and slight efux of lactate from the brain, the oxygen:glucose consumption ratio is generally approximately 5.55.8. Under different conditions ranging from deep anesthesia to the conscious state, the rate of neuronal glucose oxidation is approximately proportional to glutamatergic neurotransmission [26], indicating that the rate of the major energy-producing pathway in neurons (the TCA cycle, Figure 2) is directly related to the energy demands associated with the ux through the glutamate glutamine cycle (Figure 2B) [5]. During brain activation, glycolysis is usually preferentially upregulated compared with oxygen consumption [5], and the oxygen:glucose utilization ratio falls. Because this phenomenon occurs in normal, normoxic subjects that have excess oxygen delivery to the brain, it is sometimes called aerobic glycolysis to distinguish it from the large rise in glycolysis during hypoxia or anoxia. Stimulation of glycolysis generates increased amounts of lactate that cannot only be released from brain (Figure 2), but can also have various important functions, including serving as supplementary oxidative fuel for astrocytes and neurons, modulating redox signaling of metabolic state, regulating blood ow [5], or functioning as a mediator of metabolic information [9]. Metabolic interactions among astrocytes and neurons, and lactate shuttling Both neurons [16,27,28] and astrocytes [18,29] have been described as the main consumers of glucose. The cellular contributions to overall glucose utilization has been a controversial issue for decades because current technology does not have adequate spatiotemporal resolution to quantify metabolic activity in single cells in vivo. Two conicting concepts describe the predominant cellular fate of glucose during brain activation and propose different directions and magnitudes of shuttling of lactate among neurons and
Review
astrocytes. A third model is derived from demonstration of substantial lactate release from brain, irrespective of the originating cell type (Figure 2B) [5,17]. The astrocyte-to-neuron lactate shuttle (ANLS; Figure 2B) claims that glutamatergic neurotransmission stimulates astrocytic lactate production that serves as an important neuronal fuel during activation [29]. However, this notion remains controversial because glutamate does not stimulate glycolysis in most astrocyte preparations, the cellular origin of lactate in vivo is unknown, substantial lactate oxidation by neurons has not been demonstrated during brain activation, and studies supporting this model [29] have been challenged [5,22]. Furthermore, the neurotransmitter glutamate itself may directly support energetics of perisynaptic astrocytes, because the glial glutamate aspartate transporter (GLAST) forms a macromolecular complex linking glutamate uptake with its oxidation [30] which can provide ATP to meet the astrocytic energy demands. Glutamate oxidation at the site of its uptake eliminates the need for glycolysis to generate ATP and the ANLS [31]. The notions that glycogen-derived lactate is necessary as the bioenergetic basis for neuronal memory consolidation [7,32] and that glucose-derived lactate from oligodendrocytes is required to support axons [33,34] demand direct experimental proof of the magnitude and contribution of lactate shuttling compared with other energy sources. The neuron-to-astrocyte lactate shuttle (NALS, Figure 2B) is based on different assumptions than the ANLS and accounts for the kinetics of glucose transporters in neurons and astrocytes. The NALS model predicts predominant neuronal glucose uptake during activation, with transfer of lactate to astrocytes such that the direction of metabolite ux can be context-dependent [16,27]. Astrocytes have key roles in lactate uptake from extracellular uid and lactate dispersal to other astrocytes via gap junctional communication; these processes occur at rates that are two- to fourfold faster than lactate uptake by neurons or astrocytic transfer of lactate to neurons [17]. Thus, astrocytes are poised to take up lactate from interstitial uid and release lactate from their endfeet to perivascular uid for discharge to lymphatic drainage systems and venous blood [5,17]. The total glucose utilization substantially exceeds oxidative metabolism of glucose, with release of sizeable quantities of lactate released from activated brain (Figure 2B) [5,22]. The use of lactate as a supplemental fuel varies with its availability and physiological state of the subject. In sedentary subjects, the brain lactate level exceeds that in blood, facilitating the efux of lactate from activated brain regions to blood. By contrast, strenuous physical activity increases glycolysis in muscles and increases blood lactate levels, reversing the direction of the lactate gradient from blood to brain and ooding the entire brain with lactate. Under these conditions, lactate is oxidized in the brain in amounts that increase with blood lactate level. However, lactate oxidation in brain during exercise accompanies increased CMRglc and release of brain-derived lactate to blood [35], suggesting separate routes for lactate efux and inux. Thus, increased blood lactate level represents a glucose-sparing physiological state in which use of a
supplemental oxidative fuel helps maintain availability of glucose for the glycolytic and pentose phosphate shunt pathways that provide critical functions for the brain. Glucose metabolism and the regulation of CBF Under resting conditions, local CBF is highest in brain regions with the highest local glucose metabolism. All brain regions are metabolically active at all times, but there is a large heterogeneity among various brain structures. During functional activation, the increase in local CBF usually parallels the increase in CMRglc, whereas the increase in oxygen metabolism is lower [36]. However, there is at least one example where, under peripheral somatosensory stimulation, local CBF in the ipsilateral cortex can decrease despite increased CMRglc [37]. This close correlation of CBF and CMRglc (and, to a lesser extent, the cerebral metabolic rate of oxygen, CMRO2) demands highly dynamic and ne-tuned mechanisms to adapt local glucose and oxygen delivery and carbon dioxide removal via the blood to the actual demand of active brain regions. The traditional metabolic hypothesis of neurovascular coupling [38], mediated by vasoactive metabolic products such as lactate, CO2/H+, or adenosine, has recently been replaced by the currently favored neuronal hypothesis. It suggests that neuronal energy demand is communicated to the vasculature (either directly or indirectly by astrocytes) within the neurovascular unit in an anticipatory, feed-forward manner by vasoactive neurotransmitters or products of synaptic signaling, and that vasodilation occurs independently of glucose metabolism-induced signaling (Figure 1B; reviewed in [39]). The proposed feed-forward regulation is a reliable basis for fast adaptation of the regional blood ow to the actual local level of neuronal activity, avoiding risky drops in glucose and oxygen concentrations, which might occur during exclusive metabolic regulation. However, recent studies suggest that changes in the lactate:pyruvate ratio and, therefore, the cytosolic NADH:NAD+ ratio [40] or increased lactate production [41,42] are at least partially responsible for vasodilation during neuronal activation. Thus, neurovascular coupling regulated by feed-forward signaling may be supplemented or modulated by metabolism-dependent mechanisms [43]. Experimental studies show that direct glucose-sensing mechanisms are unlikely to be involved in the activityinduced regulation of CBF. Neither hyperglycemia nor mild-to-moderate hypoglycemia signicantly changes the blood ow responses to functional activation [44,45]. In addition, during acute hypoglycemia, resting CBF only increases signicantly when blood and brain glucose are dramatically reduced (for a detailed review, see [46]). The consequences of impaired adaptation of CBF to CMRglc are under active investigation. Articial reduction of the CBF response during functional activation had no impact on evoked neuronal activity in an acute experimental setting [47]. However, it is assumed that chronic global hypoperfusion of the brain may be not only a consequence, but also an early cause of neurodegeneration in vascular dementia and Alzheimers disease (AD) [48] (see below). Thus, ne-tuned CBF-CMRglc-CMRO2 regulation is indispensable for healthy brain.
591
Review
Brainbody axis: central control over peripheral glucose metabolism Given that the brain relies on exogenous nutrient supplies, it is not surprising that it can increase these supplies, especially glucose, by regulating systemic homeostasis and food intake [49,50] (Figure 1A). Specialized neuronal networks in the hypothalamic arcuate nucleus and in the hindbrain sense, integrate, and regulate energy homeostasis and glucose levels, and signal to the periphery through a dedicated neuronal network [4951]. Indeed, central glucose sensing and peripheral regulation of glucose metabolism are tightly linked [52]. In addition to their peripheral action, hormones [4951], including insulin [53] and glucagon-like peptide-1 (GLP-1) [54,55], mediate peripheral glucose uptake through neuronal signaling cascades. Furthermore, brain insulin receptors [56] and other metabolic receptors and transporters, such as glucose transporters [57,58], mediate metabolic signaling in the brain. In the hypothalamus, pro-opiomelanocortin (POMC) [59], melanin-concentrating hormone (MCH) [60], and neuropeptide Y (NPY)/agouti-related peptide (AgRP) neurons sense peripheral glucose levels and regulate energy metabolism in an antagonistic fashion [49]. Defective neuronal maintenance in these cells has severe consequences for
peripheral metabolism. Defective autophagy [61] in POMC neurons can lead to lifelong metabolic defects, such as peripheral glucose intolerance and obesity [62,63]. However, disrupted autophagy in glucose-sensing AgRP neurons promotes leanness and reduced food intake [64]. Interestingly, glucokinase (GK) is expressed in select neuronal populations in hypothalamic glucose-sensing formations [65], and a protein complex containing GK and Bcl-2 antagonist of cell death (BAD) might regulate peripheral [66], as well as central glucose sensing. Besides hormones and nutrients, both afferent and efferent metabolic signals link hindbrain nuclei and the gastrointestinal tract through the vagal nerve [51]. Thus, a complex interplay between the brain, in particular the hypothalamus, and peripheral systems control glucose supply to the brain [49,51], peripheral nutrient uptake [49,51] and utilization [67], as well as feeding [6769]. Notably, the regulation of energy homeostasis through the brain is not limited to glucose metabolism, but also includes most other major energy-producing systems with close links between these systems [49,51,64,70]. The mechanisms of the brainbody interaction in the regulation of glucose metabolism have recently been reviewed in greater detail [49,50].
(A)
(B)
Glc Glc GLUT
HKII PEA15 TIGAR
x
O2
Cytosol Glc
OMM IMM
VDAC
HKII
Glc-6-P
Cell survival
PEA15
Fru-6-P
PPP
NADPH
(C)
HKII
TIGAR
Glc Gal-3-P
x
OMM IMM TCA
PEA15 O2
TIGAR VDAC Cell death
GAPDH
Pyr Lac
Figure 3. The connection between glucose metabolism and cell death. (A) Glucose metabolism and cell death regulation intersect at several levels. Glucose-metabolizing enzymes, including hexokinase II (HKII), glucokinase (GK), the fructose-2,6-bisphosphatase Tp53-induced glycolysis and apoptosis regulator (TIGAR), glyceraldehyde-3phosphate dehydrogenase (GAPDH), and others, are involved in the regulation of cell death through different mechanisms. Phosphoprotein-enriched in astrocytes (PEA15) might function as a molecular linker between HKII and TIGAR under certain conditions. Flux through the pentose phosphate pathway (PPP) generates NADPH, which is important for neuronal redox environment and inhibits cell death. (B,C) The expression of HKII in neurons is upregulated under hypoxic conditions. Together with PEA15, it functions as a molecular switch to regulate neuronal viability depending on the metabolic state [72]. HKII and PEA15 interact and bind to mitochondria through the outermitochondrial membrane voltage-dependent anion channel (VDAC). During hypoxia, HKII protects cells from cell death, whereas during glucose deprivation, where HKII detaches from mitochondria and the interaction with PEA15 is destabilized, HKII promotes cell death [72]. HKII also interacts with TIGAR under hypoxic conditions [77]. Similar to PEA15, which increases the capacity of HKII to protect neurons, TIGAR increases the glycolytic activity of HKII. However, the exact mechanistic link is presently unknown. Abbreviations: Gal-3-P, glyceraldehyde-3-phosphate; Glc, glucose; Glc-6-P, glucose-6-phosphate; GLUT, glucose transporter; Fru-6-P, fructose-6-phosphate; Lac, lactate; NADPH, reduced nicotinamide adenine dinucleotide phosphate; Pyr, pyruvate; TCA, tricarboxylic acid cycle; OMM, outer mitochondrial membrane; IMM, inner mitochondrial membrane. HKII was rendered in PyMOL using structure 2nzt (RCSB Protein Data Bank).
592
Review
Glucose metabolism and the regulation of cell death Glucose metabolism is evolutionarily linked to the regulation of cell death [71] (Figures 1E and 3A), and this link is tightly controlled in a similar fashion in many cell types, arguing for a universal role of coregulated metabolic and apoptotic pathways. Neurons and cancer cells are among the cell types that rely almost exclusively on glucose metabolism for energy generation, and recent evidence suggests that these cells use similar mechanisms to adapt to substrate deprivation and promote survival [72,73]. Hexokinase II (HKII), a hypoxia-regulated HK isoform in the brain, has been demonstrated to control neuronal survival depending on the metabolic state [72] (Figures 1E and 3). HKII restricts or inhibits apoptosis in a variety of different cell types depending on whether it is bound to mitochondria [72,74] and on the availability of glucose [72]. Furthermore, the capacity of HKII to phosphorylate glucose is involved in sensing the metabolic state of the cell (Figure 3). In addition, HKII elicits its antiapoptotic function through a molecular interaction with phosphoprotein enriched in astrocytes/phosphoprotein enriched in diabetes (PEA15/PED) [72]. HKII protects against neuronal cell death after hypoxia [72] and in the presence of oxidative stress [75]. However, HKII increases neuronal cell death under glucose deprivation, thereby functioning as a molecular switch that regulates neuronal survival depending on the metabolic state. Importantly, the capacity of HKII to phosphorylate glucose and its interaction with PEA15 both mediate this effect [72] (Figure 3B,C). Intriguingly, controlled regulation of glucose metabolism protects both neurons and cancer cells from apoptosis through related mechanisms [72,73], highlighting the universal importance of sensing the metabolic state of a cell. In both cell types, HKII protects from cell death under hypoxic conditions [72], and increased activity of the PPP provides the reducing environment for inhibiting cytochrome c-mediated apoptosis, thereby preventing cellular damage through oxidative stress [73]. Furthermore, PPP inhibition after selective experimental activation of glycolysis and the concomitant deprivation of NADPH triggers apoptosis in neurons [76]. However, it has not been established whether the fructose-2,6-bisphosphatase Tp53-induced glycolysis and apoptosis regulator (TIGAR), which promotes ux through the PPP and interacts with HKII under hypoxic conditions [77], promotes preferential PPP ux in neurons and thereby cooperates with HKII to prevent cell death. Nevertheless, increased HKII activity upon interaction with TIGAR [77] is reminiscent of the increased capacity of HKII to inhibit neuronal apoptosis when it is bound to PEA15 [72] (Figure 3B,C). Finally, other enzymes of the glycolytic cascade, including GAPDH, have also been demonstrated to inhibit cell death under certain conditions [78]. However, GAPDH has also been suggested to mediate neuronal apoptosis after DNA damage [79]. Thus, glycolytic enzymes can regulate neuronal cell death in a context-dependent fashion, exerting both pro- and antiapoptotic effects. Control of apoptosis by glucose-metabolizing enzymes appears not to be a one-way street. As an example, the Bcl2 family member BAD interacts with GK (also known as hexokinase IV) in liver and pancreatic b cells to modulate
apoptosis in response to changes in glucose levels [66]. However, it is presently unknown whether GK has a role in regulating neuronal viability depending on the metabolic state in select glucose-sensing neuronal populations (see above). Furthermore, BclXL, an antiapoptotic Bcl-2 family member, has been suggested to increase the metabolic efciency of neurons by decreasing a proton leak within the F1FO ATPase and across the inner mitochondrial membrane [80,81]. Disturbed metabolism is closely linked to cell death pathways and autophagy [61]. Indeed, GAPDH [78], as well as many players in the apoptotic cascades, regulate autophagy [61]. However, whether these mechanisms involve (dysfunctional) glucose sensing through glycolytic enzymes or if they are controlled by coregulated apoptotic and/or autophagic pathways (e.g., by the Bcl-2 family [82]), remains to be established. Thus, disturbed signaling through these pathways is thought to be the pathophysiological basis for a variety of diseases (Box 1). Disease mechanisms Neurons are largely intolerant of an inadequate energy supply and, thus, the high energy demand of the brain predisposes it to a variety of diseases if energy supplies are disrupted. Various CNS pathologies are the consequence, and sometimes also the cause, of disturbed central or peripheral glucose energy metabolism, which can be affected at almost every level of the cellular or biochemical metabolic cascades (Figure 1). Changes in the glucose metabolism of affected patients can efciently be measured by positron emission tomography (PET) imaging in a large variety of clinical scenarios (Box 2). Neuroglycopenia is a neurodevelopmental syndrome characterized by mental retardation or developmental delay, anomalous coordination and muscle tone, as well as thalamocortical hypometabolism [83]. This syndrome is further characterized by infantile drug-resistant seizures, developmental retardation, and microcephaly in many cases [84]. It can be caused by persistent hypoglycemia during development or by deciency in the major glucose
Box 1. Glucose metabolism, cell death, and neurodegeneration

Glucose metabolism and the regulation of cell death are tightly coupled [66,71,72] (Figure 3, main text). Autophagy can be activated upon metabolic stress (e.g., starvation) of cells, as well as upon other stressors, including hypoxia and inflammation, to promote cellular survival under these conditions [61]. Autophagy, in turn, can be regulated by cell death and metabolic pathways [61], including key regulators of glucose metabolism [78]. Defective autophagy, oxidative stress, and bioenergetic stress have been linked to the development of neurodegenerative diseases [61,96,106]. Disrupted axonal nutrient supply and a defective metabolic network are associated with neurodegeneration in the CNS and peripheral nervous system [33,34]. Future research will elucidate the role of defective glucose metabolism and the extent of the involvement of members of the glycolytic cascade [72,7678] in the pathophysiology of neurodegenerative diseases.
593
Review
Box 2. Glucose metabolism and functional brain imaging
Under physiological conditions, cerebral glucose metabolism is tightly correlated with neuronal activity [13]. Therefore, imaging of local CMRglc can be used to visualize areas of increased neuronal activity. The most frequently used methods of brain metabolic imaging are the detection of radiolabeled glucose by PET in vivo or for diagnostic imaging (Figure I), and by ex vivo autoradiography [107]. The glucose analogs [18F]fluoro-2-deoxyglucose or 2-deoxy[14C]glucose are injected intravenously, transported into the brain, and phosphorylated by HK to 2-deoxyglucose-6-phosphate (2-DG-6P). Fluorescent 2-DG derivatives can be used for fluorescence imaging in animal models [108], but quantitative determination of CMRglc with the fluorescent analogs requires evaluation of the kinetics for competition with glucose for transport and phosphorylation [5]. Labeled 2-DG-6-P is trapped in the tissue and then detected. In an experimental setting, autoradiography of previously sliced tissue provides a higher spatial resolution than does external imaging. To achieve further information on the metabolic pathways of glucose, nuclear magnetic resonance spectroscopy after [13C]glucose infusion can be applied as a powerful tool to measure a large number of metabolic fluxes of glucose noninvasively [109]. Due to low temporal resolution, these methods do not enable measuring fast dynamic changes in glucose metabolism during neuronal activation.
-66
-62
-58
-54
-50
-46
-42
-38
-34
-30
-26
-22
-18
-14
-10
-6
-2
+2
+6
+10
+14
+18
+22
+26
+30
+34
+38
+42
+46
+50
Hypometabolism Hypermetabolism
Figure I. Example of a diagnostic [ F]fluoro-2-deoxyglucose PET-imaging study. As illustrated in this 23-year-old female patient after a 2-month course of severe anti-NMDA-receptor encephalitis, these patients typically show widespread frontotemporal cortical hypermetabolism as well as bioccipital and cerebellar cortical hypometabolism [103]. For visualization, hyper- and hypometabolism are color coded across the entire brain as depicted in the lookup table. Images are from superior (top left) to inferior (bottom right). For details on the voxel-based statistical analysis to demonstrate hyper- and hypometabolism and the corresponding cohort study see [103]. Images courtesy of R. Buchert.
18
transporter of the BBB, GLUT1 (Figure 1B). Indeed, more than 10% of early-onset absence epilepsy [85] and up to 1% of the common idiopathic generalized epilepsies [86] have been ascribed to GLUT1 deciency caused by mutations in
594
the solute carrier family 2 (SLC2A1) gene. Early diagnosis of the GLUT1 deciency syndrome is important because adherence to a ketogenic diet [15] is an effective treatment for most patients [83]; in general, a ketogenic diet efciently suppresses epileptic seizures in childhood drug-resistant epilepsy [87]. The anticonvulsant effect of restricted dietary carbohydrate intake further underscores the relevance of glucose-derived energy for neuronal excitability [15]. Furthermore, inhibition of glycolysis with the glucose analog 2-deoxyglucose in experimental seizures is an effective treatment and illustrates a role for glycolysis-derived NADH in the metabolic regulation of genes involved in epilepsy [88]. Interestingly, BAD appears to be involved in the regulation of neuronal energy substrate utilization independent of its apoptotic function [89]. However, it remains unclear whether GK [66], which has a restricted expression pattern in the brain, or another HK isoform has a role in facilitating glucose sensing by BAD in this context. A thromboembolic occlusion of a brain-supplying artery leads to an acute disruption in the blood supply to a specic brain territory, causing cerebral ischemia (Figure 1B). Within minutes, glucose depletion and associated compromised bioenergetic pathways cause extensive neuronal death in the core of the infarction and, over time, in the surrounding tissue [90,91]. Cellular models suggest that increased levels of the mitochondrial-bound glycolytic enzyme HKII protect neurons from cell death in ischemia [72] (Figure 1E). Spreading depression is a self-propagating wave of neuronal depolarization in the cortex, which is associated with a variety of neurovascular diseases, including stroke, subarachnoid hemorrhage, traumatic brain injury [92], and migraine [93]. Spreading depolarizations (SD) disturb cortical glucose metabolism [94], although interestingly it has been demonstrated that hyperglycemia increases the cortical resistance against SD [95]. Although neurodegenerative diseases are not classically thought to be caused by disturbed metabolism, bioenergetic defects are emerging as important pathophysiological mechanisms (Box 1) [3] in several disorders. One of the earliest signs of AD is a reduction in cerebral glucose metabolism, and both human studies and animal models suggest that disturbed glucose metabolism is associated with AD progression [96]. In a mouse model of AD, GLUT1 expression was reduced both at the BBB as well as in astrocytes, which was paralleled by impaired glucose transport and reduced cerebral lactate release during neuronal activation [97] (Figure 1B). Dysregulated glucose metabolism in metabolic disorders such as obesity or type 2 diabetes mellitus has been linked to AD progression and cognitive impairment [96]. However, a large clinical trial could not demonstrate a benecial effect of aggressive glucose lowering on cognitive outcome in patients with diabetes [98]. In patients with Parkinsons disease (PD), widespread cortical hypometabolism is accompanied by glucose hypermetabolism in the external pallidum and possibly other subcortical structures [99]. In one model of PD, HKII, which regulates neuronal viability depending on the metabolic state [72] (Figure 1E), has been suggested to inhibit degeneration of dopaminergic neurons [100]. Disturbed metabolism in myelin-producing cells is associated with axonal degeneration. In the brain, defective
Review
lactate transporter levels in oligodendrocytes are linked to axonopathy [34] (Figure 1C) and, in the peripheral nervous system, disrupted oxidative phosphorylation in Schwann cells is related to severe neuropathy [33]. However, demyelination without extensive axonal loss in an animal model of multiple sclerosis [101] hints to a complex underlying mechanism. Autoantibodies against the NR1 subunit of the N-methyl-D-aspartate receptor (NMDAR) may inhibit glutamatergic transmission (Figure 1D) by blocking, crosslinking, and initiating the internalization of the receptor. Patients with anti-NMDAR encephalitis have distinct symptoms that occur over weeks and months. The symptoms start with fever, psychosis, and seizures, and progress to abnormal movements, respiratory failure, dysautonomia, and coma [102]. As a consequence of impaired NMDAR function, and correlating with disease activity, a characteristic pattern of [18F]uoro-2-deoxyglucose (FDG)-PET abnormalities (Box 2, Figure I) includes increased frontotemporal and decreased occipitoparietal glucose metabolism. Despite these severe and long-lasting symptoms and changes in metabolism, most patients do not have pathological ndings in diagnostic magnetic resonance imaging (MRI) studies. Furthermore, normalization in cerebral glucose metabolism accompanies recovery [103]. Finally, disrupted central glucose sensing, insulin signaling, and defective hypothalamic circuits have been implicated in the pathophysiological mechanism of type 2 diabetes mellitus and obesity [49,59,6264,67] (Figure 1A). At the same time, dysregulated glucose metabolism in diabetes mellitus can injure the brain through both hypoand hyperglycemia [104]. Furthermore, cachexia, a severe complication after cerebral ischemia, has been ascribed in
part to the dysregulation of the hypothalamuspituitary adrenal axis and perturbed efferent signaling [105]. Given the role of hypothalamic structures in glucose and nutrient sensing (see above and [49,51]), disturbed central glucose sensing and impeded central regulation of peripheral metabolism (see above) may contribute to the development of cachexia after CNS damage. Concluding remarks Glucose metabolism is closely integrated with brain physiology and function. Although recent studies have shed light on the complex regulation of biochemical, cellular, and systemic pathways, many features of the exact regulation remain controversial or elusive (Box 3). The advent of novel and rened biochemical or genetic tools, screening methods, imaging technologies, and systems analyses will enable the study of cellular, subcellular, and even biochemical mechanisms in the cell or in vivo with unprecedented temporal and spatial resolution. In addition to studying individual biochemical or cellular pathways and their control over intracellular signaling cascades (e.g., programmed cell death), peripheral homeostasis or brain activity, future challenges lie in integrating the parts of the puzzle to form a conclusive picture of the cooperation between different systems and cell types. Ultimately, a thorough understanding of these mechanisms will lead to better insight into the pathophysiology of multiple diverse disorders of the brain and enable the development of novel treatment strategies.
Acknowledgments
We are grateful to the members of our labs for their contribution to our underlying research. We would like to thank Ralph Buchert, Department , for kindly providing and analyzing the PET of Nuclear Medicine, Charite images. This work was supported by the Seventh Framework Programme of the European Union (FP7/2008-2013) under Grant Agreements 201024 and 202213 (European Stroke Network), the Deutsche Forschungsgemeinschaft (NeuroCure Cluster of Excellence, Exc 257; SyNergy, Munich Cluster for Systems Neurology, Exc 1010), the Bundesministerium fu r Bildung und Forschung (Center for Stroke Research Berlin, 01 EO 08 01), a Canadian Stroke NetworkEuropean Stroke Network Transatlantic Collaboration grant, and the National Institutes of Health (DK081936). Disclaimer statement P.M. has received research funding from Sano. A.M. has received research funding and speaker honoraria from Bayer Vital GmbH, Sano, and Wyeth Pharma GmbH.
Box 3. Outstanding questions

Does modeling accurately predict actual energy use by the brain? Can metabolic substrates be interchanged? Are there functional consequences of using glucose and other metabolites? Future studies need to address the consequences of oxidative versus glycolytic metabolism for different brain functions. What metabolic substrates support neurons and other cell types under different functional states of the brain? What is the relevance of shuttling of metabolic substrates between different cell types? Does the direction of metabolic shuttling between cells depend on the (physiological or experimental) context? How does metabolic coupling among brain cells and shuttling of metabolites sustain brain activity? Are there additional anatomical or cellular networks in the brain that control peripheral glucose metabolism? Do cell death pathways contribute to central glucose sensing? How does disturbed peripheral metabolism influence central glucose sensing and regulation? How does disrupted central glucose sensing cause systemic metabolic disorders? What is the functional connection in the regulation of cell death pathways through different members of the glycolytic cascade? What is the functional impact of this connection for different cells of the brain (e.g., neurons, astrocytes, oligodendroglia, etc.)? How does dysbalanced metabolism and subsequent dysregulation of cell death pathways contribute to neurodegenerative diseases or other acute or chronic disorders of the brain? How can knowledge about cerebral glucose metabolism be exploited for refined therapies of neurodegenerative disorders or other diseases of the brain?
References
1 Howarth, C. et al. (2012) Updated energy budgets for neural computation in the neocortex and cerebellum. J. Cereb. Blood Flow Metab. 32, 12221232 2 Erbsloh, F. et al. (1958) [The glucose consumption of the brain & its dependence on the liver]. Arch. Psychiatr. Nervenkr. Z. Gesamte Neurol. Psychiatr. 196, 611626 3 Harris, J.J. et al. (2012) Synaptic energy use and supply. Neuron 75, 762777 4 Ivannikov, M.V. et al. (2010) Calcium clearance and its energy requirements in cerebellar neurons. Cell Calcium 47, 507513 5 Dienel, G.A. (2012) Fueling and imaging brain activation. ASN Neuro 4, e00093 6 Hertz, L. and Gibbs, M.E. (2009) What learning in day-old chickens can teach a neurochemist: focus on astrocyte metabolism. J. Neurochem. 109 (Suppl. 1), 1016 7 Suzuki, A. et al. (2011) Astrocyte-neuron lactate transport is required for long-term memory formation. Cell 144, 810823
595
Review
8 Lauritzen, K.H. et al. (2013) Lactate receptor sites link neurotransmission, neurovascular coupling, and brain energy metabolism. Cereb. Cortex http://dx.doi.org/10.1093/cercor/bht136 9 Bergersen, L.H. and Gjedde, A. (2012) Is lactate a volume transmitter of metabolic states of the brain? Front. Neuroenerget. 4, 5 10 Alle, H. et al. (2009) Energy-efcient action potentials in hippocampal mossy bers. Science 325, 14051408 11 Liotta, A. et al. (2012) Energy demand of synaptic transmission at the hippocampal Schaffer-collateral synapse. J. Cereb. Blood Flow Metab. 32, 20762083 12 Harris, J.J. and Attwell, D. (2012) The energetics of CNS white matter. J. Neurosci. 32, 356371 13 Sokoloff, L. (1999) Energetics of functional activation in neural tissues. Neurochem. Res. 24, 321329 14 van Hall, G. et al. (2009) Blood lactate is an important energy source for the human brain. J. Cereb. Blood Flow Metab. 29, 11211129 15 Lutas, A. and Yellen, G. (2013) The ketogenic diet: metabolic inuences on brain excitability and epilepsy. Trends Neurosci. 36, 3240 16 Simpson, I.A. et al. (2007) Supply and demand in cerebral energy metabolism: the role of nutrient transporters. J. Cereb. Blood Flow Metab. 27, 17661791 17 Gandhi, G.K. et al. (2009) Astrocytes are poised for lactate trafcking and release from activated brain and for supply of glucose to neurons. J. Neurochem. 111, 522536 18 Rouach, N. et al. (2008) Astroglial metabolic networks sustain hippocampal synaptic transmission. Science 322, 15511555 19 Wilson, J.E. (2003) Isozymes of mammalian hexokinase: structure, subcellular localization and metabolic function. J. Exp. Biol. 206, 20492057 20 Borgstrom, L. et al. (1976) Glucose consumption in the cerebral cortex of rat during bicuculline-induced status epilipticus. J. Neurochem. 27, 971973 21 Lowry, O.H. et al. (1964) Effect of ischemia on known substrates and cofactors of the glycolytic pathway in brain. J. Biol. Chem. 239, 1830 22 Dienel, G.A. (2012) Brain lactate metabolism: the discoveries and the controversies. J. Cereb. Blood Flow Metab. 32, 11071138 23 Dienel, G.A. et al. (2007) A glycogen phosphorylase inhibitor selectively enhances local rates of glucose utilization in brain during sensory stimulation of conscious rats: implications for glycogen turnover. J. Neurochem. 102, 466478 24 Walls, A.B. et al. (2009) Robust glycogen shunt activity in astrocytes: effects of glutamatergic and adrenergic agents. Neuroscience 158, 284292 25 Dinuzzo, M. et al. (2012) The role of astrocytic glycogen in supporting the energetics of neuronal activity. Neurochem. Res. 37, 24322438 26 Hyder, F. and Rothman, D.L. (2012) Quantitative fMRI and oxidative neuroenergetics. Neuroimage 62, 985994 27 Mangia, S. et al. (2011) Response to comment on recent modeling studies of astrocyte-neuron metabolic interactions: much ado about nothing. J. Cereb. Blood Flow Metab. 31, 13461353 28 Hall, C.N. et al. (2012) Oxidative phosphorylation, not glycolysis, powers presynaptic and postsynaptic mechanisms underlying brain information processing. J. Neurosci. 32, 89408951 29 Pellerin, L. and Magistretti, P.J. (2012) Sweet sixteen for ANLS. J. Cereb. Blood Flow Metab. 32, 11521166 30 Bauer, D.E. et al. (2012) The glutamate transporter, GLAST, participates in a macromolecular complex that supports glutamate metabolism. Neurochem. Int. 61, 566574 31 Dienel, G.A. (2013) Astrocytic energetics during excitatory neurotransmission: what are contributions of glutamate oxidation and glycolysis? Neurochem. Int. 63, 244258 32 Newman, L.A. et al. (2011) Lactate produced by glycogenolysis in astrocytes regulates memory processing. PLoS ONE 6, e28427 33 Funfschilling, U. et al. (2012) Glycolytic oligodendrocytes maintain myelin and long-term axonal integrity. Nature 485, 517521 34 Lee, Y. et al. (2012) Oligodendroglia metabolically support axons and contribute to neurodegeneration. Nature 487, 443448 35 Overgaard, M. et al. (2012) Hypoxia and exercise provoke both lactate release and lactate oxidation by the human brain. FASEB J. 26, 30123020 36 Fox, P.T. et al. (1988) Nonoxidative glucose consumption during focal physiologic neural activity. Science 241, 462464
596
37 Devor, A. et al. (2008) Stimulus-induced changes in blood ow and 2deoxyglucose uptake dissociate in ipsilateral somatosensory cortex. J. Neurosci. 28, 1434714357 38 Roy, C.S. and Sherrington, C.S. (1890) On the regulation of the bloodsupply of the brain. J. Physiol. 11, 85158 39 Attwell, D. et al. (2010) Glial and neuronal control of brain blood ow. Nature 468, 232243 40 Vlassenko, A.G. et al. (2006) Regulation of blood ow in activated human brain by cytosolic NADH/NAD+ ratio. Proc. Natl. Acad. Sci. U.S.A. 103, 19641969 41 Vafaee, M.S. et al. (2012) Oxygen consumption and blood ow coupling in human motor cortex during intense nger tapping: implication for a role of lactate. J. Cereb. Blood Flow Metab. 32, 18591868 42 Gordon, G.R. et al. (2008) Brain metabolism dictates the polarity of astrocyte control over arterioles. Nature 456, 745749 43 Gordon, G.R. et al. (2011) Bidirectional control of arteriole diameter by astrocytes. Exp. Physiol. 96, 393399 44 Wolf, T. et al. (1997) Excessive oxygen or glucose supply does not alter the blood ow response to somatosensory stimulation or spreading depression in rats. Brain Res. 761, 290299 45 Powers, W.J. et al. (1996) Effect of stepped hypoglycemia on regional cerebral blood ow response to physiological brain activation. Am. J. Physiol. 270, H554H559 46 Nehlig, A. (1997) Cerebral energy metabolism, glucose transport and blood ow: changes with maturation and adaptation to hypoglycaemia. Diabetes Metab. 23, 1829 47 Leithner, C. et al. (2010) Pharmacological uncoupling of activation induced increases in CBF and CMRO2. J. Cereb. Blood Flow Metab. 30, 311322 48 Nicolakakis, N. and Hamel, E. (2011) Neurovascular function in Alzheimers disease patients and experimental models. J. Cereb. Blood Flow Metab. 31, 13541370 49 Lam, C.K. et al. (2009) CNS regulation of glucose homeostasis. Physiology 24, 159170 50 Grayson, B.E. et al. (2013) Wired on sugar: the role of the CNS in the regulation of glucose homeostasis. Nat. Rev. Neurosci. 14, 2437 51 Grill, H.J. and Hayes, M.R. (2012) Hindbrain neurons as an essential hub in the neuroanatomically distributed control of energy balance. Cell Metab. 16, 296309 52 Lam, T.K. et al. (2005) Regulation of blood glucose by hypothalamic pyruvate metabolism. Science 309, 943947 53 Filippi, B.M. et al. (2012) Insulin activates Erk1/2 signaling in the dorsal vagal complex to inhibit glucose production. Cell Metab. 16, 500510 54 Hayes, M.R. et al. (2011) Intracellular signals mediating the food intake-suppressive effects of hindbrain glucagon-like peptide-1 receptor activation. Cell Metab. 13, 320330 55 Sandoval, D.A. et al. (2008) Arcuate glucagon-like peptide 1 receptors regulate glucose homeostasis but not food intake. Diabetes 57, 2046 2054 56 Bruning, J.C. et al. (2000) Role of brain insulin receptor in control of body weight and reproduction. Science 289, 21222125 57 Chari, M. et al. (2011) Glucose transporter-1 in the hypothalamic glial cells mediates glucose sensing to regulate glucose production in vivo. Diabetes 60, 19011906 58 Mounien, L. et al. (2010) Glut2-dependent glucose-sensing controls thermoregulation by enhancing the leptin sensitivity of NPY and POMC neurons. FASEB J. 24, 17471758 59 Parton, L.E. et al. (2007) Glucose sensing by POMC neurons regulates glucose homeostasis and is impaired in obesity. Nature 449, 228232 60 Kong, D. et al. (2010) Glucose stimulation of hypothalamic MCH neurons involves K(ATP) channels, is modulated by UCP2, and regulates peripheral glucose homeostasis. Cell Metab. 12, 545552 61 Kroemer, G. et al. (2010) Autophagy and the integrated stress response. Mol. Cell 40, 280293 62 Coupe, B. et al. (2012) Loss of autophagy in pro-opiomelanocortin neurons perturbs axon growth and causes metabolic dysregulation. Cell Metab. 15, 247255 63 Kaushik, S. et al. (2012) Loss of autophagy in hypothalamic POMC neurons impairs lipolysis. EMBO Rep. 13, 258265 64 Kaushik, S. et al. (2011) Autophagy in hypothalamic AgRP neurons regulates food intake and energy balance. Cell Metab. 14, 173183
Review
65 Lynch, R.M. et al. (2000) Localization of glucokinase gene expression in the rat brain. Diabetes 49, 693700 66 Danial, N.N. et al. (2003) BAD and glucokinase reside in a mitochondrial complex that integrates glycolysis and apoptosis. Nature 424, 952956 67 Joly-Amado, A. et al. (2012) Hypothalamic AgRP-neurons control peripheral substrate utilization and nutrient partitioning. EMBO J. 31, 42764288 68 Aponte, Y. et al. (2011) AGRP neurons are sufcient to orchestrate feeding behavior rapidly and without training. Nat. Neurosci. 14, 351 355 69 Wu, Q. et al. (2009) Loss of GABAergic signaling by AgRP neurons to the parabrachial nucleus leads to starvation. Cell 137, 12251234 70 Yi, C.X. and Tschop, M.H. (2012) Brain-gut-adipose-tissue communication pathways at a glance. Dis. Model Mech. 5, 583587 71 King, A. and Gottlieb, E. (2009) Glucose metabolism and programmed cell death: an evolutionary and mechanistic perspective. Curr. Opin. Cell Biol. 21, 885893 72 Mergenthaler, P. et al. (2012) Mitochondrial hexokinase II (HKII) and phosphoprotein enriched in astrocytes (PEA15) form a molecular switch governing cellular fate depending on the metabolic state. Proc. Natl. Acad. Sci. U.S.A. 109, 15181523 73 Vaughn, A.E. and Deshmukh, M. (2008) Glucose metabolism inhibits apoptosis in neurons and cancer cells by redox inactivation of cytochrome c. Nat. Cell Biol. 10, 14771483 74 Majewski, N. et al. (2004) Hexokinase-mitochondria interaction mediated by Akt is required to inhibit apoptosis in the presence or absence of Bax and Bak. Mol. Cell 16, 819830 75 Gimenez-Cassina, A. et al. (2009) Mitochondrial hexokinase II promotes neuronal survival and acts downstream of glycogen synthase kinase-3. J. Biol. Chem. 284, 30013011 76 Herrero-Mendez, A. et al. (2009) The bioenergetic and antioxidant status of neurons is controlled by continuous degradation of a key glycolytic enzyme by APC/C-Cdh1. Nat. Cell Biol. 11, 747752 77 Cheung, E.C. et al. (2012) Mitochondrial localization of TIGAR under hypoxia stimulates HK2 and lowers ROS and cell death. Proc. Natl. Acad. Sci. U.S.A. 109, 2049120496 78 Colell, A. et al. (2007) GAPDH and autophagy preserve survival after apoptotic cytochrome c release in the absence of caspase activation. Cell 129, 983997 79 Ishitani, R. and Chuang, D.M. (1996) Glyceraldehyde-3-phosphate dehydrogenase antisense oligodeoxynucleotides protect against cytosine arabinonucleoside-induced apoptosis in cultured cerebellar neurons. Proc. Natl. Acad. Sci. U.S.A. 93, 99379941 80 Alavian, K.N. et al. (2011) Bcl-xL regulates metabolic efciency of neurons through interaction with the mitochondrial F1FO ATP synthase. Nat. Cell Biol. 13, 12241233 81 Chen, Y.B. et al. (2011) Bcl-xL regulates mitochondrial energetics by stabilizing the inner membrane potential. J. Cell Biol. 195, 263276 82 Galluzzi, L. et al. (2012) Non-apoptotic functions of apoptosisregulatory proteins. EMBO Rep. 13, 322330 83 Pascual, J.M. et al. (2007) Brain glucose supply and the syndrome of infantile neuroglycopenia. Arch. Neurol. 64, 507513 84 Leen, W.G. et al. (2010) Glucose transporter-1 deciency syndrome: the expanding clinical and genetic spectrum of a treatable disorder. Brain 133, 655670 85 Arsov, T. et al. (2012) Early onset absence epilepsy: 1 in 10 cases is caused by GLUT1 deciency. Epilepsia 53, e204e207 86 Arsov, T. et al. (2012) Glucose transporter 1 deciency in the idiopathic generalized epilepsies. Ann. Neurol. 72, 807815 87 Neal, E.G. et al. (2008) The ketogenic diet for the treatment of childhood epilepsy: a randomised controlled trial. Lancet Neurol. 7, 500506
88 Garriga-Canut, M. et al. (2006) 2-Deoxy-D-glucose reduces epilepsy progression by NRSF-CtBP-dependent metabolic regulation of chromatin structure. Nat. Neurosci. 9, 13821387 89 Gimenez-Cassina, A. et al. (2012) BAD-dependent regulation of fuel metabolism and K(ATP) channel activity confers resistance to epileptic seizures. Neuron 74, 719730 90 Dirnagl, U. et al. (1999) Pathobiology of ischaemic stroke: an integrated view. Trends Neurosci. 22, 391397 91 Mergenthaler, P. et al. (2004) Pathophysiology of stroke: lessons from animal models. Metab. Brain Dis. 19, 151167 92 Dreier, J.P. (2011) The role of spreading depression, spreading depolarization and spreading ischemia in neurological disease. Nat. Med. 17, 439447 93 Vecchia, D. and Pietrobon, D. (2012) Migraine: a disorder of brain excitatory-inhibitory balance? Trends Neurosci. 35, 507520 94 Sakowitz, O.W. et al. (2013) Clusters of spreading depolarizations are associated with disturbed cerebral metabolism in patients with aneurysmal subarachnoid hemorrhage. Stroke 44, 220223 95 Hoffmann, U. et al. (2013) Glucose modulation of spreading depression susceptibility. J. Cereb. Blood Flow Metab. 33, 191195 96 Kapogiannis, D. and Mattson, M.P. (2011) Disrupted energy metabolism and neuronal circuit dysfunction in cognitive impairment and Alzheimers disease. Lancet Neurol. 10, 187198 97 Merlini, M. et al. (2011) Vascular beta-amyloid and early astrocyte alterations impair cerebrovascular function and cerebral metabolism in transgenic arcAbeta mice. Acta Neuropathol. 122, 293311 98 Launer, L.J. et al. (2011) Effects of intensive glucose lowering on brain structure and function in people with type 2 diabetes (ACCORD MIND): a randomised open-label substudy. Lancet Neurol. 10, 969977 99 Borghammer, P. et al. (2012) Glucose metabolism in small subcortical structures in Parkinsons disease. Acta Neurol. Scand. 125, 303310 100 Corona, J.C. et al. (2010) Hexokinase II gene transfer protects against neurodegeneration in the rotenone and MPTP mouse models of Parkinsons disease. J. Neurosci. Res. 88, 19431950 101 Skripuletz, T. et al. (2011) De- and remyelination in the CNS white and grey matter induced by cuprizone: the old, the new, and the unexpected. Histol. Histopathol. 26, 15851597 102 Dalmau, J. et al. (2011) Clinical experience and laboratory investigations in patients with anti-NMDAR encephalitis. Lancet Neurol. 10, 6374 103 Leypoldt, F. et al. (2012) Fluorodeoxyglucose positron emission tomography in anti-N-methyl-D-aspartate receptor encephalitis: distinct pattern of disease. J. Neurol. Neurosurg. Psychiatry 83, 681686 104 Scheen, A.J. (2010) Central nervous system: a conductor orchestrating metabolic regulations harmed by both hyperglycaemia and hypoglycaemia. Diabetes Metab. 36 (Suppl. 3), S31S38 105 Scherbakov, N. et al. (2011) Body weight after stroke: lessons from the obesity paradox. Stroke 42, 36463650 106 Harris, H. and Rubinsztein, D.C. (2012) Control of autophagy as a therapy for neurodegenerative disease. Nat. Rev. Neurol. 8, 108117 107 Sokoloff, L. et al. (1989) The [14C]deoxyglucose method for measurement of local cerebral glucose utilization. In Carbohydrates and Energy Metabolism (Boulton, A. et al., eds), pp. 155193, Humana Press 108 Yao, J. et al. (2012) Noninvasive photoacoustic computed tomography of mouse brain metabolism in vivo. Neuroimage 64C, 257266 109 de Graaf, R.A. et al. (2011) State of the art direct 13C and indirect 1H[13C] NMR spectroscopy in vivo. A practical guide. NMR Biomed. 24, 958972
597
Review
Control of neuronal voltage-gated calcium ion channels from RNA to protein

Diane Lipscombe, Summer E. Allen, and Cecilia P. Toro*
Department of Neuroscience, Brown University, Providence, RI 02912, USA
Voltage-gated calcium ion (CaV) channels convert neuronal activity into rapid intracellular calcium signals to trigger a myriad of cellular responses. Their involvement in major neurological and psychiatric diseases, and importance as therapeutic targets, has propelled interest in subcellular-specic mechanisms that align CaV channel activity to specic tasks. Here, we highlight recent studies that delineate mechanisms controlling the expression of CaV channels at the level of RNA and protein. We discuss the roles of RNA editing and alternative premRNA splicing in generating CaV channel isoforms with activities specic to the demands of individual cells; the roles of ubiquitination and accessory proteins in regulating CaV channel expression; and the specic binding partners that contribute to both pre- and postsynaptic CaV channel function. Ten CaV genes, thousands of different CaV mRNAs, and many more functionally different proteins . . .different Ca currents show so many differences in fundamental properties that we nd it easier to assume that there are more than one type. [1] CaV channels are present at every critical step of information transfer in the nervous system, from signal detection to perception, and from neuronal development to programmed apoptosis. Strategically located at points of sensory detection and on both sides of the synapse, CaV channels have a dening role in integrating signals and inuencing synaptic strength. All but one of the ten mammalian genes that encode the pore-forming a1 subunit of CaV channels are expressed in the nervous system (Box 1). Each CaV gene has a distinct expression prole reecting its functional specialization to support specic cellular tasks. For more than 50 years, the functional diversity of CaV currents among different cell types has been attributed to the expression of multiple channel types [1]. Indeed, individual neurons can express multiple CaV genes, consistent with the varied subcellular distribution of calcium channel
Corresponding author: Lipscombe, D. (Diane_Lipscombe@brown.edu). Keywords: voltage-gated calcium channels; alternative splicing; splicing factors; synaptic transmission; synaptic proteins; ubiquitination. * Current address: Oregon Hearing Research Center, Oregon Health & Science University, Portland, OR 97239, USA. 0166-2236/$ see front matter 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.tins.2013.06.008
subtypes controlling a range of neuronal functions. The value of highly selective toxins and drugs used to tease out the relative contributions of CaV channels within single cells is hard to overstate. Yet this is not always straightforward: for example, until recently, the absence of pharmacological tools to distinguish CaV1.2 and CaV1.3 channels [2], which collectively give rise to most dihydropyridine-sensitive Ca currents in neurons, has stalled progress in dening their individual contributions. Even when specic toxins are available, the challenges associated with isolating the functional contributions of closely related CaV channels are magnied several-fold when considering the thousands of channel isoforms potentially generated from single genes by alternative pre-mRNA splicing and RNA editing. All ten mammalian calcium channel, voltage-dependent, alpha 1A through 1I, and 1S genes (CACNA1A through CACNA1I and CACNA1S) contain multiple sites of alternative splicing; each gene has the potential to generate thousands of unique splice isoforms controlled by the expression and relative activities of cell specic splicing factors (Box 1) (Figures 1 and 2). Typically, these sites are located in coding regions that tolerate sequence variation and that are not essential for structural integrity: N termini, cytoplasmic loops III and IIIII, extracellular linkers that connect transmembrane a-helices S3 and S4 in domains III and IV, and the C termini (Figures 2A,B and 3). Consequently, these hypervariable regions are hotspots for cell specic posttranslational modication, second messenger action on, and protein binding to, CaV channels (Figures 3 and 4). The tethering of appropriate CaV channels to presynaptic active zones or to postsynaptic specializations depends on specic nanodomain-level protein interactions (Figures 5 and 6), as discussed below for the most recently discovered of these interactions. A common feature across these studies is the ability to resolve molecular-level detail of specic RNAs and proteins within subpopulations of neurons and functionally distinct subcellular compartments. Novel techniques have been developed and applied to identify the following: splicing factor binding sites on CaV channel pre-mRNAs; CaV channel RNAs in specic populations of neurons; and CaV proteins at active zones. These methodologies marry disciplines to yield a more comprehensive, multidimensional view of CaV channels and their regulation from RNA to protein. Here, we highlight these and other recent studies elucidating cell specic control of CaV RNA
598
Review
Box 1. By any other name
CaV channels are highly specialized plasma membrane proteins that convert small changes in the membrane potential into rapid, localized increases in intracellular calcium. Only a limited number of ion channels, including several ligand-gated ion channels, allow the passage of calcium. However, CaV channels are exquisitely selective for calcium over other cations as long as extracellular calcium concentrations exceed approximately 10 mM. In the absence of extracellular calcium, CaV channels lose their high selectively to calcium and readily conduct a range of cations. Most CaV channels have a dual function: they support ionic current that changes the membrane potential, and they permit the flow of calcium across the plasma membrane that serves as an intracellular second messenger. In mammals, ten genes encode the ten major a1 subunits of CaV channels (CACNA1A through CACNA1I, and CACNA1S). They have unique expression profiles, cellular function, and pharmacology, and are associated with various diseases. The numerous aliases used to identify and distinguish the different CaV channel genes, proteins, and currents are shown for seven different species (Table I). The pore-forming a1 subunits of CaV channels fall into three classes based on sequence homology: CaV1 (1.1, 1.2, 1.3, and 1.4), CaV2 (2.1, 2.2, and 2.3) and CaV3 (3.1, 3.2, and 3.3). The classes are functionally distinguished by a combination of pharmacological tools and
biophysical properties, although pharmacological differences across CaV1 (dihydropyridine sensitive), CaV2 (v-agatoxin-IVA inhibits CaV2.1; v-conotoxin GVIA inhibits CaV2.2), and CaV3 channels more reliably distinguish among CaV subtypes. Pharmacological, genetic, and functional studies have helped delineate the different cellular roles controlled by CaV channels. Presynaptic calcium entry, and subsequent transmitter release, are mediated by CaV2 channels (mainly CaV2.1 and CaV2.2) throughout the nervous system, by CaV1.3 channels in IHCs [73] and by CaV1.4 channels in retina [74,75]. Dendritic postsynaptic CaV1 (mainly CaV1.2) channels regulate calcium entry that controls gene expression [76,77]. Dendritic CaV2.3 channels [32] are implicated in acquisition of dendritic phenotype [78] and oscillatory burst discharge in the reticular thalamus [79]. Postsynaptic CaV1.3 channels underlie pacemaking in certain cells [80], are important for brain stem neuron development [63], and are implicated in calcium-dependent death of dopaminergic neurons [2]. The roles of presynaptic CaV2 channels in synaptic transmission and short-term plasticity [58] and the role of postsynaptic CaV1 channels in activitydependent gene expression in neurons have been reviewed elsewhere [81]. A functional, correctly targeted CaV channel depends on its association with other calcium channel subunits, including CaVb, CaVa2d, and other proteins; some of these aspects of CaV channel regulation are detailed in other recent reviews [82].
Table I. Voltage-gated calcium channel nomenclature

Protein name CaV1.1 CaV1.2 CaV1.3 CaV1.4 CaV2.1 CaV2.2 CaV2.3 CaV3.1 CaV3.2 CaV3.3 Old a1S a1C a1D a1F a1A a1B a1E a1G a1H a1I Current type L Gene name Human CACNA1S CACNA1C CACNA1D CACNA1F CACNA1A CACNA1B CACNA1E CACNA1G CACNA1H CACNA1I Mouse, Rat Cacna1s Cacna1c Cacna1d Cacna1f Cacna1a Cacna1b Cacna1e Cacna1g Cacna1h Cacna1i Zebrash cacna1sa,b cacna1c cacna1da,b cacna1f cacna1aa,b cacna1ba,b cacna1g cacna1i Puffersh 1S-a,b,c 1C 1D-a,b,c,d 1F-a,b,c 1A-a,b 1B 1E-a,b 1G-a,b 1H-a,b 1I Drosophila Ca-a1D Caenorhabditis elegans egl-19
P/Q N R T
cac
unc-2
Ca-a1T
cca-1
processing, posttranslational CaV protein modications, and CaV proteinprotein interactions. Collectively, this work demonstrates exciting progress made in dening the molecular mechanisms underlying the expression of CaV channel isoforms in specic neurons, and in linking their expression to critical cell functions. Cell specic alternative pre-mRNA splicing and RNA editing regulate CaV channel function . . .the more you look, the more you see. (Robert M. Pirsig, 1974) Alternative pre-mRNA splicing is particularly prevalent in the mammalian brain, and is essential for normal neuronal development, axon targeting, neuronal excitability, and neural circuit formation. This form of pre-mRNA processing occurs in the nucleus of the cell, and is controlled by the concerted actions of a set of cell specic, RNAbinding proteins called splicing factors. Trans-acting splicing factors bind to consensus cis-sequence motifs in premRNAs to inuence the action of the spliceosome, promoting or repressing the inclusion of alternatively spliced exons, and promoting or repressing the use of alternative splice acceptor or donor sites at intronexon boundaries
(Figure 1). Nova, rbFox, and polypyrimidine tract binding (PTB) families of splicing factors direct inclusion or skipping of alternatively spliced exons in several neural gene pre-mRNAs, including CACNA1 genes (Box 1) (Figure 2). These splicing factor protein families have the capacity to either repress (silence) or enhance exon inclusion, in different target pre-mRNAs or within the same pre-mRNA, depending on the position of their consensus binding motifs, either upstream (50 ) or downstream (30 ) relative to the target exon [37] (Figures 1 and 2C). Genome-wide maps of splicing factor-binding sites reveal certain common features in the mechanism of action of certain splicing factors and bring us closer to dening a working splicing code [8]. These splicing factor genomic maps have advanced understanding of cell specic regulation of channel structure and function. In addition, several studies have shown that alternative splicing can be controlled by epigenetic factors [9], signaling factors including cellular protein kinases and phosphatases [10], and by cell specic alternative splicing of splicing factors themselves [5,6]. Splicing factors that target CACNA1 pre-mRNAs CACNA1 pre-mRNAs are targets of splicing factors from the Nova, rbFox, PTB, and Muscleblind-like 2 (Mbnl2)
599
Review
pre-mRNA (A)
Repressor
mRNA
+
(B)
Enhancer
(C)
Repressors (D)
Enhancer
+
Figure 1. Splicing factors repress and enhance alternative exon inclusion during pre-mRNA processing. Splicing factors bind to consensus nucleotide sequences in introns or target exons of pre-mRNAs to enhance or repress (silence) recognition of the exon by the spliceosome during pre-mRNA processing. Alternatively spliced exons (colored), constitutive exons (gray), and introns (black connecting lines) are illustrated. The position of splicing factor binding, relative to the target exon, is often predictive of splicing factor action. For example, members of the Nova, rbFox, and polypyrimidine tract binding (PTB) splicing factor protein families tend to repress or silence exon inclusion when they bind their respective nucleotide sequence-binding motifs upstream (A) or within (D) the target exon, and enhance exon inclusion when they bind their respective binding motifs downstream of the target exon (B). Alternatively spliced cassette exons are excluded or included during pre-mRNA splicing (A,B). Mutually exclusive alternative splicing involves two or more exons, only one of which is selected during pre-mRNA processing (C,D). Mutually exclusive exons often start or end with an incomplete codon. In this case, there is a shift in the reading frame in mRNAs that either lack both mutually exclusive exons or contain both, and early protein termination (not shown). Often, exon selection is influenced by the concerted action of several splicing factors (D). The expression levels and activities of individual splicing factors depend on many cellular factors, including those involved in development, neuronal activity, and defining neuronal subtype.
RNA-binding protein families [47] (Figure 2D). Moreover, multiple rbFox splice isoforms act on the same CACNA1 pre-mRNA exon, but with different consequences [11]. Importantly, developmental changes in the patterns of CaV splice isoform expression can parallel the changes in expression levels of key splicing factors. For example, as PTB expression levels fall in the developing brain, levels of neuronal PTB (nPTB) increase [12,13]. PTB controls the choice between a pair of mutually exclusive exons in CACNA1C (CaV1.2), e8 and e8a, via binding to motifs in CACNA1C pre-mRNA upstream of e8a [14] (Figure 2C). As PTB levels decrease during development, the ratio of e8a- to e8containing CaV2.2 mRNAs increases because PTB no longer represses e8a. In this case, nPTB does not substitute for PTB as a repressor of e8a [14]. E8a and e8 encode alternate forms of transmembrane S6 in domain I of CaV1.2 in a region that is critical for normal channel function (Figure 2B). E8a and e8 are of special interest because certain mutations in either exon cause the severe multi-organ hereditary disorder, Timothy Syndrome [15]. The severity of neurological symptoms in individuals with Type I and Type II Timothy Syndrome depends on which exon carries the mutation, presumably reecting the different expression patterns of e8 and e8a in the brain [16]. CACNA1B (CaV2.2) pre-mRNAs are targets of Nova2, and several Nova2-binding sites align with previously identied alternatively spliced exons. For example, highthroughput sequencing combined with cross-linking immunoprecipitation methods (HiTS-CLIP) showed that Nova2binding sites are located within the intron immediately
600
upstream, and overlapping, a short six-nucleotide exon e31a in CACNA1B pre-mRNAs [17] (Figure 2C). The positioning of RNA-binding motifs for Nova2 upstream of e31a predicts Nova2 repression of e31a. The HiTS-CLIP data are consistent with earlier reports that e31a-containing CaV2.2 mRNAs are found at very low levels in the rodent brain where Nova2 is expressed. By contrast, e31a-containing CaV2.2 mRNAs dominate in peripheral ganglia, which do not express Nova2 [18]. The same splicing factor can act differently on multiple alternatively spliced exons within the same pre-mRNA. This is the case for Nova2; in contrast to its silencing action on e31a, it binds motifs downstream of e24a to augment e24a expression in brain (Figure 2C,D) [18]. Intriguingly, there are clusters of Nova2-binding sites in other regions of CACNA1B that do not map to known alternatively spliced exons, and the functional signicance of these is not yet known. RbFox and Nova splicing factor protein families have overlapping networks of synaptic protein targets, consistent with the collective action of multiple splicing factors on several alternative CACNA1 exons [17] (Figures 1 and 2C,D). Loss of rbFox2 results in reduced expression of Nova1, perhaps reecting the regulation of Nova1 premRNA splicing by rbFox2 [17]. Several important studies have shown profound consequences, including gross neurodevelopmental abnormalities, of disrupting or eliminating splicing factors that regulate alternative splicing of multiple targets [5,7,19]. However, because the splicing of hundreds of neuronal pre-mRNAs, including those that encode other splicing factors, are disrupted in these studies, the relative
Review
(A) Transmembrane-spanning (S1S6), domains (IIV), and intracellular regions of CaV channels
S1
S6
S1
S6
S1
S6
S1
S6
I
III N terminus (B) Locaon of alternave exons in CaV channels
8/8a
II IIIII
III IIIIV
IV
C terminus
33 21/22 12 24a
Key:
CaV1.2 CaV1.3 CaV1.4 CaV2.1 CaV2.2 CaV2.3 CaV3.1 CaV3.2 CaV3.3
31a 31a 32
8a/8b out 14 16 in 2/2a 9/9a 9 9* 11 9b 10* 10a10/10a 1a/b/c 1 18a 14 17 20a 17
25a
31a 31a/b 31/32 34 35 37a/b 35a 40b 42 46 47 37 42a 43 44
25c
26 26
45a 45 45
(C) Splicing factors target exons in CaV pre-mRNAs

Brain
CaV2.1 pre-mRNA
24 24a 25 31 31a 32
(D) Known exons regulated by splicing factors

Channel Gene CaV2.1 Cacna1a
32
Exon
Locaon
Splicing factor Nova Nova Nova Nova rbFox PTB rbFox rbFox rbFox rbFox rbFox
Refs
Nova Nova
24a 31a 24a 31a 18a 8a 9* 33 8b 11 29
IIIS3-IIIS4 IVS3-IVS4 IIIS3-IIIS4 IVS3-IVS4 II-III IS6 I-II IVS3-IVS4 IS6 I-II IVS3-IVS4
[18] [18] [4] [18] [91] [14] [90] [90] [5] [5] [5]
CaV2.2 pre-mRNA
+
24 24a 25 31
31a
CaV2.2 Cacna1b
E12 corcal neurons

CaV1.2 pre-mRNA
PTB
CaV1.2
7 8a 8 9 9* 10
31 32 33
34
35
Cacna1c CaV1.3
Adult corcal neurons

CaV1.2 pre-mRNA
PTB
rbFox
rbFox
7 8a 8 9
9* 10
31 32 + 33
Cacna1d
35
34
CaV1.1 Cacna1s
Figure 2. Alternative splicing is extensive in voltage-gated calcium ion (CaV) channels and is regulated by the action of cell-specific splicing factors. (A) Major domains and regions located on a schematic of the a1 subunit of a CaV channel. There are four structurally homologous domains: I, II, III, and IV. Each domain comprises six transmembrane spanning a-helices (S1S6). The intracellular regions that link the domains are labeled III, IIIII, and IIIIV. The N terminus and C terminus are intracellular. This naming system is used to identify the major regions of CaV channels. For example, the third transmembrane spanning a-helix in domain IV is referred to as IVS3. (B) Location of alternative exons mapped on to a schematic of the a1 subunit of a CaV channel. The 24 transmembrane-spanning domains are shown (gray) as well as extracellular and intracellular regions. The approximate locations of alternative exons are shown (circles) and color coded to indicate specific CaV channel subtypes. Exon numbers are indicated and the numbering system used follows mouse calcium channel, voltage-dependent, alpha 1A-1I, 1S (Cacna1a-Cacna1i, Cacna1s) gene numbering. Alternate first exons are shown at the start of the N termini, and mutually exclusive alternative exons are designated as X/Y. The following references were used to compile information for CaV1.2 [14]; CaV1.3 [83]; CaV1.4 [84]; CaV2.1 [85]; CaV2.2 [83], accession # FJ609386; CaV2.3 [83,86]; CaV3.1 [87]; CaV3.2 [88]; CaV3.3 [89]. (C) Splicing factor target exons in CaV pre-mRNAs. Two splicing patterns are shown for three splicing factor RNA-binding protein families that have similar mechanisms of action: Nova, rbFox and polypyrimidine tract binding (PTB). The first example (top) illustrates the action of Nova on alternatively spliced exon cassettes, e24a and e31a, in CaV2.1 and CaV2.2 pre-mRNAs. Nova enhances inclusion of e24a during pre-mRNA splicing of CaV2.1 and CaV2.2 by binding to elements in the respective introns downstream of the target exons (intronic splicing enhancer, ISE), whereas Nova represses (or silences) inclusion of e31a during pre-mRNA splicing by binding element upstream and within or overlapping the target exon (intronic splicing silencer, ISS; exonic splicing silencer, ESS). Nova is expressed in brain where CaV2.1 and CaV2.2 mRNAs containing e24a and lacking e31a dominate [18]. In the second example (lower), PTB represses inclusion of mutually exclusive e8a during splicing of CaV1.2 pre-mRNA. PTB is expressed in embryonic brain and, therefore, CaV1.2 mRNAs lacking e8a and containing e8 dominate early during development. In adult cortical neurons, PTB levels are reduced and it no longer represses e8a. By contrast, rbFox is upregulated during development and represses inclusion of e9* and enhances inclusion of e33 [14,90]. In adult cortical neurons, CaV1.2 mRNA containing e8a, lacking e9*, and containing e33 dominate [14]. (D) Alternatively spliced exons of CaV channel genes are shown together with splicing factors known to regulate their expression. Cassette exons are either included or excluded. E8a/e8 of CaV1.2, and e8b/e8a of CaV1.3 are mutually exclusive (A,B). Therefore, repression of e8a during pre-mRNA splicing of CaV1.2 promotes inclusion of e8, and repression of e8b during pre-mRNA splicing of CaV1.3 promotes inclusion of e8a. CaV1.2 e8a is strongly repressed by PTB but more weakly repressed by the neuronal homolog, nPTB, despite their similar RNA-binding motifs [14]. Based upon [4,5,14,18,90,91].
601
Review
e17a I K727
Key:
Acidic Basic Hydrophobic Polar Dierent
S E I P Start End Inseron Phosphorylaon
e17 E K727 Synprint S L718
1 del. S L754
Channel amino acid #

CaV2.2 583 CaV2.1 590
2 del. S A738 e16 S I667
Synprint S A722 e17 S Y705 e16 E N704 CaMKII P S784 (2.2) 2 del. S R793 (2.1)
CaV2.2 683 CaV2.1 690
PKC P S774 1 del. S R756
IIS6 IIIII linker

CaV2.2 781 CaV2.1 790
e18a I A755
CaMKII P S896,S898 PKC P S898
CaV2.2 871 CaV2.1 884
1 del. E P947 2 del. E P947
Synprint E H963 Synprint E R984
CaV2.2 963 CaV2.1 968
2 del. E P1001
CaV2.2 1048 CaV2.1 1062 1 del. E L1139
CaV2.2 1114 CaV2.1 1162
IIIII linker IIIS1

Figure 3. The IIIII intracellular loop regions of voltage-gated calcium (CaV) 2.1 and CaV2.2 have divergent sequences. Amino acid alignments for CaV2.1 (rat sequence: NM_012918) and CaV2.2 (rat sequence: AF055477) are shown for approximately 700 amino acids encoding the IIIII intracellular loops as well as sequence in IIS6 and IIIS1 (Clustal Omega software [92]). The approximate locations of boundaries between transmembrane and intracellular domains are indicated. The chemical nature of each amino acid is coded as follows: acidic (green) D, E; basic (purple) K, R, H; hydrophobic (orange) A, F, I, L, M, P, V, W; and polar (blue) C, G, N, Q, S, T, Y. The black solid bars below the alignments indicate regions of differences in amino acid type between CaV2.1 and CaV2.2. White circles indicate amino acids of interest, the amino acid numbers are shown and they are coded as follows: (S) alternative exon, deletion (del.), or synprint area start; (E) alternative exon del., or synprint area end; (I) alternative exon insertion; (P) phosphorylation. References: Alternative e16, e17, and e17a of CaV2.1 [85]; CaV2.2 e18a [93,94]; CaV2.2 synprint [95]; CaV2.1 synprint [96]; PKC and CaMKII phosphorylation [97]; CaV2.1 D1 and D2 [98]; CaV2.
contribution of splicing events within a single gene family, such as CACNA1, cannot be distinguished. Individual alternatively spliced exons have measurable cellular and behavioral consequences Many individual splice sites in CACNA1 genes are evolutionarily conserved, and it is often assumed that each adds functional value and tness to the cells in which it is expressed. Another possibility is that the functional impact of any individual splicing change may not be discernible at the cellular or behavioral level. Alternative pre-mRNA splicing of genes may have evolved because essential proteins operate over a broader dynamic range than would be possible from the activity of a single protein. This feature of alternative splicing could be critical during development, particularly in neurons that adapt to, and continue to function in the face of, dramatic changes in morphology and activity [20]. Yet, there are several notable examples showing that exon choice in an inuential gene can indeed modulate behavior. For example, reproductive dominance in honeybees is controlled by alternative splicing of exon 5 in a gene homologous to the gemini transcription factor of Drosophila [21], and cell specic expression of a transient receptor potential cation channel, subfamily V, member 1 (TRPV1) splice isoform confers unique infrared sensing capabilities to vampire bats [22].
602
Recently, the cellular and behavioral consequences of an alternatively spliced exon in the mouse Cacna1b gene were demonstrated. Enhanced expression of the nociceptorspecic exon 37a splice isoform of CaV2.2 increased cellular sensitivity to inhibition by activated m-opioid receptors and behavioral sensitivity to spinal morphine-induced analgesia [23] (Figure 4). Cell specic expression of e37a in wild type mice may help explain why CaV2.2 channels in nociceptors are particularly sensitive to inhibition by drugs, transmitters, and hormones that act through G protein-coupled receptors. Additional alternatively spliced exons in Cacna1b and in the closely related Cacna1a (CaV2.1) and Cacna1e (CaV2.3) genes modify amino acid sequences in the cytoplasmic domains of CaV channels. The actions of G proteins and second messengers on other ion channels may also be inuenced by cell specic alternative splicing of ion channel pre-mRNAs in other parts of the nervous system. Cell specic inclusion of exons in Cacna1 genes can also impact disease states. Already mentioned are the different symptoms in Type I and Type II Timothy Syndrome determined by which exon, e8a or e8, in CACNA1C carries the mutation [16]. Other examples include the different consequences of Cacna1a (CaV2.1) mutations associated with familial hemiplegic migraine on the short (D47) and long (+47) splice isoforms of CaV2.1 [24], and an epilepsycausing mutation in mouse Cacna1h (CaV3.2) that results
Review
Key: RIM src TK e37b GPCR

CaV2 CaV2
e37a e37a
i/o
i/o
Only G G + src
Only src
Figure 4. Alternative splicing influences G protein-coupled receptor (GPCR) inhibition of voltage-gated calcium (CaV) 2.2 channels. Illustration showing a Gi/o protein-coupled receptor (green) inhibiting presynaptic CaV2.2 channels (blue) by different signaling mechanisms. The GPCR is activated by neurotransmitter released from a neighboring neuron. Following activation, the Gbg dimer dissociates from Gi/o, binds to and directly inhibits both CaV2.2 channel splice isoforms [e37a (orange) and e37b (pink)]. A second G protein-dependent pathway, that requires src tyrosine kinase (TK) activation but that is independent of Gbg, inhibits e37aCaV2.2. This schematic is based on data published in [23,99]. A prediction that follows from these studies for e37aCaV2.2 channels, which are more distant from GPCRs, is that they will not be inhibited by membrane-delimited Gbg, but may still be inhibited by src TK. This is because the latter mechanism involves a diffusible second messenger.
Figure 5. The density of voltage-gated calcium (CaV) 2 channels at presynaptic nerve terminals is controlled by RIM and the CaVa2d-1 subunit. The figure shows that CaV2 channel trafficking to, or retention at, presynaptic nerve terminals depends on the association of CaV2 proteins with both RIM and CaVa2d-1 [30,38,45,46,49]. The details of where RIM and CaVa2d-1 interact with CaV2, in which intracellular compartment, are not known but the figure is intended to capture the finding that overexpression of CaV2 channels does not lead to an increase in functional presynaptic CaV2 channels (left) unless CaVa2d-1 is also overexpressed (right) [45,46]. These data suggest that CaVa2d-1 limits the insertion into, or the stabilization of, CaV2 channels at the active zones of the presynaptic nerve terminal [45].
in different channel phenotypes depending on whether alternative exon 25 is expressed [25]. RNA editing of CACNA1D mRNA Additional cell specic alterations in amino acid sequences can arise from RNA editing, introducing potentially even ner specialization of CaV activities. In RNA editing, posttranscriptional deamination of adenosine to inosine (read as guanosine during translation) is catalyzed in RNA by the adenosine deaminase acting on RNA (ADAR) family of enzymes. Evidence for different RNA-edited versions of CaV channel mRNAs was rst reported for the Drosophila CaV2 homolog cacophony (Dmca1A) (Box 1, Table I) [26]. Recently, central nervous system (CNS)-specic editing of mammalian CaV1.3 RNAs by the ADAR protein ADR2 has been carefully documented and shown to generate four distinct sequences within the C terminus IQ domain: IQDY, MQDY, IRDY, and MRDY [27]. By contrast, only one IQ sequence, IQDY, is found in CaV1.3 channels expressed outside of the CNS [27]. The IQ domain forms part of a critical calmodulin-binding site in CaV channels that mediates calcium-dependent inhibition. The latter is an important feedback control on subsequent calcium entry; cytoplasmic calcium inhibits the further gating of the ion channel through which it owed. Editing of the IQ site (e.g., to MQ) reduces calcium-dependent inactivation and, as a consequence, is thought to participate in shaping the rhythmicity of action potential ring in neurons of the suprachiasmatic nucleus [27]. Cell specic RNA editing could exert similar control over CaV channels in other populations of neurons, and
could alter amino acid sequences of other critical domains that regulate specic functions. Because of the potential importance of CaV1.3 channels in promoting calciumdependent cell death in dopaminergic neurons in the substantia nigra pars compacta and the proposed connection between CaV1.3 activity and Parkinsons disease [28], it will be interesting to know which edited versions of CaV1.3 RNAs dominate in these neurons. Mapping the cell specic expression patterns of the ADAR2 protein and mapping ADAR2-binding sites to specic RNAs would provide valuable information in this regard. However, as is the case for splicing factors, ADAR2 activity depends on several factors, including phosphorylation-dependent prolyl-isomerase Pin1 that controls its nuclear localization [29], the WW domain containing E3 ubiquitin protein ligase 2 (WWP2) that promotes its proteasomal degradation [29], as well as other second messengers. Mechanisms that control numbers of CaV channels at the cell surface CaV channel activity depends not only on the pattern of expression of functionally different splice and RNA edited isoforms, but also on the overall expression level of CaV channel proteins in specic subcellular compartments. Counting CaV2 channels at active zones of different synapses by quantitative molecular and ultrastructural analyses recently demonstrated a tight correlation between presynaptic CaV2.1 and CaV2.2 channel number and vesicle release probability [30,31], whereas the third member of the CaV2 gene family, CaV2.3, was shown (with the exception of the interpeduncular nucleus) to be mostly restricted to postsynaptic structures, particularly dendritic shaft plasma membrane [32]. Thus, the overall activities of presynaptic CaV2.1 and CaV2.2 channels have a major inuence on the efcacy of transmission at mammalian synapses. We know a great deal about G protein-coupled receptor-mediated control of CaV2 channels, particularly CaV2.2. These well-studied signaling pathways terminate
603
Review
Presynapc
Presynapc
CaV2.2
CaV1.3
Vesicle
Synapsin
V-ATPase
CSP NSF Ub CRMP
CaN PP2A
Munc18
Dynamin
Ribbon Vesicle
AP-2
Endophilin
SV2
VAMP
SNAP-25
MAP1A
Vesicle
Ribeye CaBP4
CaM
RIM1 CaSK
Syntaxin
CaV
Dynacn
RIM2
Ub
Harmonin
14-3-3 Clathrin
Whirlin
CaV
Laminin
CaV2
HA
P AKA 0 5 1 / 9 7
A PP2 Ub PKA
P2B MA
eIF
3
Erb De nsi n CaM KII in
CaV1.2
Postsynapc
CaV1.3
Sha
nk
eIF3
Figure 6. Presynaptic and postsynaptic voltage-gated calcium ion (CaV) channels are associated with several membrane-anchoring and signaling proteins that define their function. Illustrated are a subset of proteins shown to bind presynaptic CaV2.2, postsynaptic CaV1.2, and pre- and postsynaptic CaV1.3 channels. Not all proteins reported to bind to these CaV channels are shown, but those illustrated have been confirmed by more than one proteomics screen and/or by a functional assay. In addition, some G proteins, chaperone, and cytoskeletal proteins are not included. The proteins shown are approximately grouped by function, but beyond that there is no significance to location in the synapse or relative to each other. Data were collected using the following references: CaV2.2: cysteine string protein (CSP) [100]; collapsin response mediator protein 2 (CRMP) [101]; laminin [102]; microtubule-associated protein 1A (MAP1A) [103]; ubiquitin (Ub) [35,36]; other proteins were identified by [54,55,104]. CaV1.2: protein phosphatase 2A (PP2A) [105]; protein kinase A (PKA) [106]; A kinase (PRKA) anchor protein 79/150 (AKAP79/150) [107109], HA [110], Ub [34,111], eukaryotic translation initiation factor 3 (eIF3) [112]; MAP2B [108,113]; and dynamin [112]. CaV1.3 presynapse: whirlin [114]; regulating synaptic membrane exocytosis 2 (RIM2) [48]; harmonin [42]; Ub [42], ribeye [60,115]; and calcium-binding protein 4 (CaBP4) [116,117]. CaV1.3 postsynapse: Shank [118]; erbin [81]; densin [64]; calcium/calmodulin-dependent protein kinase II (CaMKII) [64]; and eIF3 [112].
on CaV2.2 channels and account for the short-term presynaptic effects of many neurotransmitters, hormones, and drugs that modulate synaptic transmission. By contrast, the cellular mechanisms that control the overall number of CaV channels at active zones, which likely contribute to long-term changes in synaptic plasticity, were only recently elucidated [31]. Not surprisingly, cellular mechanisms analogous to those critical in setting the overall expression levels of postsynaptic receptors, including the ubiquitin proteasome system (UPS) [33], also control surface expression of presynaptic CaV2.1 and CaV2.2 channels [3436]. In addition, proteinCaV2 channel interactions described recently appear critical in establishing the number of CaV2 channels specically at presynaptic terminals [30,37,38] (Figure 5). Such molecular interactions are dynamic and subject to modulation, thereby regulating the density and number of presynaptic CaV2 channels during synaptic plasticity [30,31]. Ubiquitin regulates CaV1 and CaV2 channels Ubiquitination of several neuronal proteins is a critical posttranslational mechanism to modify the trafcking, endocytosis, and activity of synaptic receptors and ion
604
channels to adjust synaptic strength [34,39]. Ubiquitin covalently attaches to intracellular lysines of target proteins and, depending on the type of conjugate (i.e., monoubiquitination or polyubiquitination), can promote internalization, modify protein function, or target protein for degradation via the UPS [40]. Activity-dependent ubiquitination of postsynaptic AMPA receptors is described and known to regulate synaptic plasticity. Until recently, CaV channels were rarely considered important targets of ubiquitin despite functional evidence that ubiquitin-dependent changes in synaptic efcacy involved presynaptic components [41]. Four reports now show that CaV1.3, CaV1.2, and CaV2.2 channels are all targets of posttranslational modication by ubiquitin [3436,42]. These reports implicate the UPS in controlling activity-dependent pre- and postsynaptic calcium levels. Ubiquitination of CaVa1 subunits is inuenced by their association with CaVb subunits, known to regulate plasma membrane targeting of CaV, and by cell specic alternative splicing. CaVb subunit binding to CaV1.2 and CaV2.2 reduces CaVa1 ubiquitination and protects channels from proteasomal degradation [34,36]. Similarly, CaV2.2 channel splice isoforms with restricted expression (e37a containing) have reduced ubiquitination compared with other broadly
Review
expressed isoforms (e37b containing) and are associated with increased channel current densities [35]. Thus, the inuence of the UPS on CaV2.2 channels is cell specic, reecting the particular expression prole of certain alternative splice forms. Cav2.2 was recently identied as a target of ubiquitin based on a large-scale proteomics analysis of diGly-modied lysine residues of proteins expressed in a human colorectal carcinoma cell line [43]. Modied lysines in the region of Cav2.2 encoded by e37a and e37b were not identied in this screen, suggesting that ubiquitination at this site is cell specic. Future large-scale proteomics analyses that enrich for neuronal ion channels, and other membrane proteins, hold promise for mapping major sites of posttranslational modications of critical synaptic proteins. CaVa2d and RIM regulate CaV2 expression Two protein families have recently grabbed the limelight as being essential for trafcking or tethering CaV2 channels to nerve terminals: CaVa2d and regulating synaptic membrane exocytosis (RIM) (Figures 5 and 6). CaVa2d-1, rst shown to interact with CaV channels approximately 25 years ago, is a glycosylphosphatidylinositol (GPI)-anchored extracellular protein [44]. This auxiliary subunit of CaV2 channels promotes membrane trafcking, speeds channel activation and inactivation, and binds the analgesics gabapentin and pregabalin. CaVa2d-1 has now been shown to act as a rate-limiting factor controlling the number of CaV2.1 channels at presynaptic terminals [45]. Evidence for a limiting factor in CaV2 channel trafcking to presynaptic nerve terminals emerged from studies in cultured hippocampal neurons. Whereas somatic Ca currents were reliably enhanced in cultured neurons transiently expressing exogenous CaV2.1 or CaV2.2 channels, synaptic transmission was not [45,46]. However, coexpressing CaVa2d-1 along with CaV2.1 or CaV2.2 channels in these neurons augmented channel function, facilitated CaV2 coupling to exocytosis, and enhanced synaptic transmission [45] (Figure 5). RIM proteins are also critical CaV2 channel partners at nerve terminals. The reported actions of RIM proteins on presynaptic CaV channels are diverse: they augment CaV2.1 currents in presynaptic calyx of Held terminals [37], decrease voltage-dependent inactivation (VDI) of CaV2.2 channels, interfere with the inhibitory actions of m-opioid receptors on CaV2.2 channels expressed in HEK293 cells [47], and slow CaV1.3 channel inactivation in inner hair cells [48]. However, their importance in tethering CaV2 channels to active zones is central to explaining their in vivo function [38,49]. Meticulous high-resolution analyses of the molecular and morphological architecture of glutaminergic synapses show that RIM protein numbers scale proportionately with presynaptic CaV2.1 channel numbers at active zone areas [30]. Moreover, the numbers of these complexes scale with, and predict, vesicle release probability, consistent with critical function [30] (Figure 6). Controlling subcellular targeting of CaV channels Synapse-specic location of CaV2 channels CaV2.1 and CaV2.2 channels couple differentially to neurotransmitter vesicle release machinery according to synapse
type. In hippocampal interneurons, CaV2.1 channels mediate the release of neurotransmitter from parvalbumin (PV)expressing fast-spiking basket cells, whereas CaV2.2 channels mediate the release of neurotransmitter from cholecystokinin (CCK)-expressing basket cells [50,51]. The physical distance between presynaptic channels and calcium sensors of exocytosis are predicted to differ between CaV2.1 and CaV2.2. These predictions as based on the differential ability of BAPTA and EGTA, which are fast and slow calcium chelators, respectively, to inhibit synaptic events mediated by CaV2.1 and CaV2.2 channels [50,51]. CaV2.1 channels at PV nerve terminals are predicted to be within nanometers of the calcium sensor that leads to synchronous GABA release, whereas CaV2.2 channels at CCK terminals are predicted to be within micrometers of the Ca sensor and are thought to underlie the highly asynchronous GABA release of CCK neurons [5052]. However, the precise molecular identity of CaV2.2 and CaV2.1 splice isoforms at these synapses is not yet determined. Therefore, it is possible that different splice isoforms of CaV2, which are known to inuence binding to release machinery, contribute to synapse-specic differences in calcium-dependent exocytosis. Alternative splicing of auxiliary CaV subunits, as well as other interacting proteins, is also likely to contribute to the functional diversity among different synapses. For example, the four CACNA2D1-4 genes that encode mammalian CaVa2d14 subunits are each subject to alternative premRNA processing. Similarly, the regulating synaptic membrane exocytosis 1 and 2 (Rims1 and Rims2) genes generate ve principal RIM proteins by the use of independent promoters (RIM1a, RIM1b, RIM2a, RIM2b, and RIM2g) and both Rims genes contain sites of alternative splicing [49]. This provides substantial capacity for synapse-specic differences in Ca-dependent transmitter release. Finally, asynchronous transmission can occur as a consequence of prolonged CaV2 channel openings, a phenomenon that is signicantly augmented in CaV2.1 channels associated with CaVb2a subunits [53]. CaVb subunits strongly inuence the rate of CaV2 channel inactivation, and CaVb2a subunits in particular slow channel inactivation signicantly compared with CaVb1, CaVb3, and CaVb4 subunits (e.g., [53]). Proteins in nanodomains with synaptic CaV2 channels Although lacking resolution at the level of specic synapses, a proteomics screen combined with afnity purication and high-resolution quantitative mass spectroscopy has identied over 200 proteins that interact closely with CaV2 proteins in rodent brain [54]. This valuable data set includes a subset of proteins isolated from brain synaptosomes that were previously identied to interact with CaV2.2 channels [55], as well as cytoskeletal and chaperone proteins and a complex of novel proteins involved in both exocytosis and endocytosis (Figure 6). Functional analyses, validating the signicance of several of these proteinprotein interactions, suggest the interactions are critical to the subcellular specialization of CaV2 channels. For example, the collapsin response mediator protein 2 (CRMP2)CaV2.2 interaction controls CaV2.2 channel current densities in sensory neurons [56,57]. Disrupting this interaction using a cell permeable blocking peptide
605
Review
reduced CaV2.2 trafcking to presynaptic terminals, spontaneous excitatory postsynaptic currents in the spinal dorsal horn, and inammatory and neuropathic pain [56]. The absence of gross behavioral or motor effects in these mice supports a unique role for CRMP2CaV2.2 in maladaptive responses in sensory neurons to noxious stimuli. Other protein interactions that, similar to CRMP2, are mediated via C termini of CaV2 channels may be important in stabilizing the presynaptic molecular architecture (Figure 6). Perhaps the best-studied region of CaV2 channels is the synprint site located in the IIIII intracellular linkers of CaV2.1 and CaV2.2 (Figure 3). The synprint region binds synaptic soluble N-ethylmaleimide-sensitive factor activating protein receptor (SNARE) proteins syntaxin 1A and SNAP-25 and synaptotagmin (Figure 6). This tripartite interaction (CaV2SNAREsynaptotagmin), which is so critical to the release process, is reviewed in detail elsewhere [58]. Less studied is the interaction between endocytotic proteins and presynaptic CaV2 channels, most notably adaptor-related protein complex 2 (AP-2; adaptor protein for clathrin-mediated endocytosis) [54]. AP-2 was shown to associate with CaV2.2 and CaV2.1, but not CaV1.2, via the synprint region to control endocytosis, but not exocytosis, at calyx of Held nerve terminals [59]. The IIIII linker sequence of CaV channels is highly variable among CACNA1 genes (Figure 3) and contains several sites of alternative pre-mRNA splicing (Figure 3). Therefore, alternative splicing of CaV2.2 and CaV2.1 pre-mRNAs generates IIIII splice isoforms that have different capacities to interact with SNAREs [58] and potentially with AP2. Proteins that bind and modulate pre- and postsynaptic CaV1.3 CaV1.3 channels are found both pre- and postsynaptically in cochlear inner hair cells (IHCs). Presynaptic CaV1.3 channels regulate transmitter release, interact with ribeye (the ribbon synapse protein that promotes channel clustering), and negatively regulate the size of the ribbon body [60,61] (Figure 6). Harmonin also associates with presynaptic CaV1.3 channels at the onset of hearing in 2-week-old mice. This interaction promotes CaV1.3 ubiquitination and leads to a decrease in channel surface expression. In the deaf circler mouse, dfcr, a mutant allele of the harmonin gene disrupts the harmonin interaction with CaV1.3, leading to abnormally high CaV1.3 currents in IHCs [42]. Postsynaptic CaV1.3 channels also have a critical role in the auditory brainstem, where they are required for the normal development of the inhibitory medial nucleus of the trapezoid body (MNTB) to the lateral superior olive (LSO) projection [62,63]. There is a substantial reduction in the number of MNTBLSO axons over the rst 2 weeks of life, concomitant with strengthening of the remaining synapses. Thus, postsynaptic CaV1.3 channels in the LSO neurons are thought to be vital for the development and maturation of inhibitory MNTBLSO projections [62]. Postsynaptic CaV1.3 channels isolated from brain are also found in a complex with densin and calcium/calmodulin-dependent protein kinase II (CaMKII), where they localize to dendritic spines [64]. Densin promotes calcium-dependent facilitation of CaV1.3 channels
606
[64]. Alternative pre-mRNA splicing of CACNA1D premRNA must generate functionally distinct presynaptic and postsynaptic CaV1.3 channels. Dening the major splicing patterns of CaV1.3 isoforms, their differential expression patterns, and functional properties should provide valuable insight into structural domains essential for presynaptic and postsynaptic function. Concluding remarks and challenges The number of molecularly distinct CaV proteins that can be generated from CACNA1 genes is stunning. Cell specic gene expression, alternative pre-mRNA splicing, RNA editing, posttranslational modications including ubiquitination, miRNA targeting, and subcellular specic proteinprotein interactions are all used according to the demands of the cell. At the level of RNA, we still lack cell specic information about which mRNA is expressed and when. Several cell type-specic transcriptome data sets have been generated from the combined application of methods, including highthroughput mRNA sequencing, pooling of genetically homogenous cells, and enriching for polysomal RNAs [13,65]. These methods are cataloguing CACNA1 transcripts according to neuronal subtype and developmental stage, although substantial variability in the composition of certain transcripts has been observed across different studies and stochastic variations in transcriptomes are evident even in genetically similar cell types [13]. Genetically targeted and afnity-based miRNA proling (miRAP) has also generated the rst comparative analysis of cell specic miRNAs in glutamatergic and GABAergic neurons of neocortex and cerebellum [66], giving insights into miRNA-dependent control of mRNA translation and stability in different classes of neuron. In addition, recent genome-wide mapping of splicing factor-binding sites by CLIP-seq, as already discussed, locates sites of splicing factor binding relative to target exons in pre-mRNAs that, combined with other information, could eventually lead to a splicing code [8]. These, and other sequence-based data sets not mentioned here, are publicly accessible (http://genome.ucsc.edu/) and recently integrated with the encyclopedia of DNA elements project (ENCODE, http://www.genome.gov/10005107). At the protein level, major challenges remain but exciting technological advances are being made that combine high-resolution imaging with proteomic analyses of single synapses [67]. It should soon be possible to visualize and distinguish among CaV isoforms with the spatial resolution needed to place them at individual active zones and synapses. Similarly, it may be possible to dene unique patterns of posttranslational modications, and characterize unique proteinCaV interactions that occur in highly localized, specialized subcellular regions of neurons. Such information should reveal the molecular mechanisms that underlie functional specialization at the level of individual synapses, and may suggest new therapeutic strategies to target specic regions or neural circuits within the brain. At the gene level, several hereditary diseases and disorders originate from point mutations or trinucleotide expansions in CACNA1 genes, but a major, as yet unexplained nding implicates CACNA1C (CaV1.2) in determining an individuals susceptibility to a range of
Review
psychiatric illnesses. In the most recent and largest genome-wide association study of its kind, four risk loci were identied with signicant and overlapping links to autism spectrum disorder, attention-decit/hyperactivity disorder, bipolar disorder, major depressive disorder, and schizophrenia [68]. Two of these risk single nucleotide polymorphisms (SNPs) are located in calcium channel genes, CACNA1C (CaV1.2) and CACNB2 (CaVb2) [69 72]. The widespread expression of CaV1.2 channels in the body, including muscle cells of the cardiovascular system, endocrine cells, and smooth muscle, adds to the intrigue. The cellular mechanisms described here that individualize CaV channel function according to cell type and alternative pre-mRNA splicing, in particular, could be selectively disrupted in individuals exhibiting psychiatric illness without involvement of non-CNS systems.
References
1 Hagiwara, S. and Byerly, L. (1981) Membrane biophysics of calcium currents. Fed. Proc. 40, 22202225 2 Kang, S. et al. (2012) CaV1.3-selective L-type calcium channel antagonists as potential new therapeutics for Parkinsons disease. Nat. Commun. 3, 1146 3 Licatalosi, D.D. et al. (2008) HITS-CLIP yields genome-wide insights into brain alternative RNA processing. Nature 456, 464469 4 Ule, J. et al. (2006) An RNA map predicting Nova-dependent splicing regulation. Nature 444, 580586 5 Gehman, L.T. et al. (2011) The splicing regulator Rbfox1 (A2BP1) controls neuronal excitation in the mammalian brain. Nat. Genet. 43, 706711 6 Yeo, G.W. et al. (2009) An RNA code for the FOX2 splicing regulator revealed by mapping RNA-protein interactions in stem cells. Nat. Struct. Mol. Biol. 16, 130137 7 Charizanis, K. et al. (2012) Muscleblind-like 2-mediated alternative splicing in the developing brain and dysregulation in myotonic dystrophy. Neuron 75, 437450 8 Irimia, M. and Blencowe, B.J. (2012) Alternative splicing: decoding an expansive regulatory layer. Curr. Opin. Cell Biol. 24, 323332 9 Luco, R.F. et al. (2011) Epigenetics in alternative pre-mRNA splicing. Cell 144, 1626 10 Zhou, Z. et al. (2012) The Akt-SRPK-SR axis constitutes a major pathway in transducing EGF signaling to regulate alternative splicing in the nucleus. Mol. Cell 47, 422433 11 Dredge, B.K. and Jensen, K.B. (2011) NeuN/Rbfox3 nuclear and cytoplasmic isoforms differentially regulate alternative splicing and nonsense-mediated decay of Rbfox2. PLoS ONE 6, e21585 12 Boutz, P.L. et al. (2007) A post-transcriptional regulatory switch in polypyrimidine tract-binding proteins reprograms alternative splicing in developing neurons. Genes Dev. 21, 16361652 13 Tang, F. et al. (2011) Development and applications of single-cell transcriptome analysis. Nat. Methods 8, S6S11 14 Tang, Z.Z. et al. (2011) Regulation of the mutually exclusive exons 8a and 8 in the CaV1.2 calcium channel transcript by polypyrimidine tract-binding protein. J. Biol. Chem. 286, 1000710016 15 Splawski, I. et al. (2004) Ca(V)1.2 calcium channel dysfunction causes a multisystem disorder including arrhythmia and autism. Cell 119, 1931 16 Splawski, I. et al. (2005) Severe arrhythmia disorder caused by cardiac L-type calcium channel mutations. Proc. Natl. Acad. Sci. U.S.A. 102, 80898096 discussion 8086 17 Zhang, C. et al. (2010) Integrative modeling denes the Nova splicingregulatory network and its combinatorial controls. Science 329, 439 443 18 Allen, S.E. et al. (2010) The neuronal splicing factor Nova controls alternative splicing in N-type and P-type CaV2 calcium channels. Channels 4, 483489 19 Gehman, L.T. et al. (2012) The splicing regulator Rbfox2 is required for both cerebellar development and mature motor function. Genes Dev. 26, 445460
20 Turrigiano, G. (2012) Homeostatic synaptic plasticity: local and global mechanisms for stabilizing neuronal function. Cold Spring Harb. Perspect. Biol. 4, a005736 21 Jarosch, A. et al. (2011) Alternative splicing of a single transcription factor drives selsh reproductive behavior in honeybee workers (Apis mellifera). Proc. Natl. Acad. Sci. U.S.A. 108, 1528215287 22 Gracheva, E.O. et al. (2011) Ganglion-specic splicing of TRPV1 underlies infrared sensation in vampire bats. Nature 476, 8891 23 Andrade, A. et al. (2010) Opioid inhibition of N-type Ca2+ channels and spinal analgesia couple to alternative splicing. Nat. Neurosci. 13, 12491256 24 Adams, P.J. et al. (2009) Ca(V)2.1 P/Q-type calcium channel alternative splicing affects the functional impact of familial hemiplegic migraine mutations: implications for calcium channelopathies. Channels 3, 110121 25 Powell, K.L. et al. (2009) A Cav3.2 T-type calcium channel point mutation has splice-variant-specic effects on function and segregates with seizure expression in a polygenic rat model of absence epilepsy. J. Neurosci. 29, 371380 26 Smith, L.A. et al. (1996) A Drosophila calcium channel alpha1 subunit gene maps to a genetic locus associated with behavioral and visual defects. J. Neurosci. 16, 78687879 27 Huang, H. et al. (2012) RNA editing of the IQ domain in Ca(v)1.3 channels modulates their Ca(2)(+)-dependent inactivation. Neuron 73, 304316 28 Ilijic, E. et al. (2011) The L-type channel antagonist isradipine is neuroprotective in a mouse model of Parkinsons disease. Neurobiol. Dis. 43, 364371 29 Marcucci, R. et al. (2011) Pin1 and WWP2 regulate GluR2 Q/R site RNA editing by ADAR2 with opposing effects. EMBO J. 30, 4211 4222 30 Holderith, N. et al. (2012) Release probability of hippocampal glutamatergic terminals scales with the size of the active zone. Nat. Neurosci. 15, 988997 31 Sheng, J. et al. (2012) Calcium-channel number critically inuences synaptic strength and plasticity at the active zone. Nat. Neurosci. 15, 9981006 32 Parajuli, L.K. et al. (2012) Quantitative regional and ultrastructural localization of the Ca(v)2.3 subunit of R-type calcium channel in mouse brain. J. Neurosci. 32, 1355513567 33 Henley, J.M. et al. (2011) Routes, destinations and delays: recent advances in AMPA receptor trafcking. Trends Neurosci. 34, 258268 34 Altier, C. et al. (2011) The Cavbeta subunit prevents RFP2-mediated ubiquitination and proteasomal degradation of L-type channels. Nat. Neurosci. 14, 173180 35 Marangoudakis, S. et al. (2012) Differential ubiquitination and proteasome regulation of Ca(V)2.2 N-type channel splice isoforms. J. Neurosci. 32, 1036510369 36 Waithe, D. et al. (2011) Beta-subunits promote the expression of Ca(V)2.2 channels by reducing their proteasomal degradation. J. Biol. Chem. 286, 95989611 37 Han, Y. et al. (2011) RIM determines Ca(2)+ channel density and vesicle docking at the presynaptic active zone. Neuron 69, 304316 38 Kaeser, P.S. et al. (2011) RIM proteins tether Ca2+ channels to presynaptic active zones via a direct PDZ-domain interaction. Cell 144, 282295 39 Rotin, D. and Staub, O. (2011) Role of the ubiquitin system in regulating ion transport. Pugers Arch. 461, 121 40 MacGurn, J.A. et al. (2012) Ubiquitin and membrane protein turnover: from cradle to grave. Annu. Rev. Biochem. 81, 231259 41 Rinetti, G.V. and Schweizer, F.E. (2010) Ubiquitination acutely regulates presynaptic neurotransmitter release in mammalian neurons. J. Neurosci. 30, 31573166 42 Gregory, F.D. et al. (2011) Harmonin inhibits presynaptic Cav1.3 Ca(2)(+) channels in mouse inner hair cells. Nat. Neurosci. 14, 11091111 43 Kim, W. et al. (2011) Systematic and quantitative assessment of the ubiquitin-modied proteome. Mol. Cell 44, 325340 44 Davies, A. et al. (2010) The alpha2delta subunits of voltage-gated calcium channels form GPI-anchored proteins, a posttranslational modication essential for function. Proc. Natl. Acad. Sci. U.S.A. 107, 16541659
607
Review
45 Hoppa, M.B. et al. (2012) alpha2delta expression sets presynaptic calcium channel abundance and release probability. Nature 486, 122125 46 Cao, Y.Q. and Tsien, R.W. (2010) Different relationship of N- and P/Qtype Ca2+ channels to channel-interacting slots in controlling neurotransmission at cultured hippocampal synapses. J. Neurosci. 30, 45364546 47 Weiss, N. et al. (2011) Rim1 modulates direct G-protein regulation of Ca(v)2.2 channels. Pugers Arch. 461, 447459 48 Gebhart, M. et al. (2010) Modulation of Cav1.3 Ca2+ channel gating by Rab3 interacting molecule. Mol. Cell. Neurosci. 44, 246259 49 Kaeser, P.S. et al. (2012) RIM genes differentially contribute to organizing presynaptic release sites. Proc. Natl. Acad. Sci. U.S.A. 109, 1183011835 50 Goswami, S.P. et al. (2012) Miniature IPSCs in hippocampal granule cells are triggered by voltage-gated Ca2+ channels via microdomain coupling. J. Neurosci. 32, 1429414304 51 Hefft, S. and Jonas, P. (2005) Asynchronous GABA release generates long-lasting inhibition at a hippocampal interneuron-principal neuron synapse. Nat. Neurosci. 8, 13191328 52 Daw, M.I. et al. (2009) Asynchronous transmitter release from cholecystokinin-containing inhibitory interneurons is widespread and target-cell independent. J. Neurosci. 29, 11112 11122 53 Few, A.P. et al. (2012) Asynchronous Ca2+ current conducted by voltage-gated Ca2+ (CaV)-2.1 and CaV2.2 channels and its implications for asynchronous neurotransmitter release. Proc. Natl. Acad. Sci. U.S.A. 109, E452E460 54 Muller, C.S. et al. (2010) Quantitative proteomics of the Cav2 channel nano-environments in the mammalian brain. Proc. Natl. Acad. Sci. U.S.A. 107, 1495014957 55 Khanna, R. et al. (2007) The presynaptic CaV2.2 channel-transmitter release site core complex. Eur. J. Neurosci. 26, 547559 56 Brittain, J.M. et al. (2011) Suppression of inammatory and neuropathic pain by uncoupling CRMP-2 from the presynaptic Ca(2)(+) channel complex. Nat. Med. 17, 822829 57 Wilson, S.M. et al. (2011) Further insights into the antinociceptive potential of a peptide disrupting the N-type calcium channel-CRMP-2 signaling complex. Channels 5, 449456 58 Catterall, W.A. and Few, A.P. (2008) Calcium channel regulation and presynaptic plasticity. Neuron 59, 882901 59 Watanabe, H. et al. (2010) Involvement of Ca2+ channel synprint site in synaptic vesicle endocytosis. J. Neurosci. 30, 655660 60 Sheets, L. et al. (2011) Ribeye is required for presynaptic Ca(V)1.3a channel localization and afferent innervation of sensory hair cells. Development 138, 13091319 61 Sheets, L. et al. (2012) Presynaptic CaV1.3 channels regulate synaptic ribbon size and are required for synaptic maintenance in sensory hair cells. J. Neurosci. 32, 1727317286 62 Hirtz, J.J. et al. (2012) Synaptic renement of an inhibitory topographic map in the auditory brainstem requires functional Cav1.3 calcium channels. J. Neurosci. 32, 1460214616 63 Hirtz, J.J. et al. (2011) Cav1.3 calcium channels are required for normal development of the auditory brainstem. J. Neurosci. 31, 82808294 64 Jenkins, M.A. et al. (2010) Ca2+-dependent facilitation of Cav1.3 Ca2+ channels by densin and Ca2+/calmodulin-dependent protein kinase II. J. Neurosci. 30, 51255135 65 Okaty, B.W. et al. (2011) Cell type-specic transcriptomics in the brain. J. Neurosci. 31, 69396943 66 He, M. et al. (2012) Cell-type-based analysis of microRNA proles in the mouse brain. Neuron 73, 3548 67 Micheva, K.D. et al. (2010) Single-synapse analysis of a diverse synapse population: proteomic imaging methods and markers. Neuron 68, 639653 68 Smoller, J.W. et al. (2013) Identication of risk loci with shared effects on ve major psychiatric disorders: a genome-wide analysis. Lancet 381, 13711379 69 Schizophrenia Psychiatric Genome-Wide Association Study Consortium (2011) Genome-wide association study identies ve new schizophrenia loci. Nat. Genet. 43, 969976 70 Nyegaard, M. et al. (2010) CACNA1C (rs1006737) is associated with schizophrenia. Mol. Psychiatry 15, 119121
608
71 Ferreira, M.A. et al. (2008) Collaborative genome-wide association analysis supports a role for ANK3 and CACNA1C in bipolar disorder. Nat. Genet. 40, 10561058 72 Lee, M.T. et al. (2011) Genome-wide association study of bipolar I disorder in the Han Chinese population. Mol. Psychiatry 16, 548556 73 Johnson, S.L. and Marcotti, W. (2008) Biophysical properties of CaV1.3 calcium channels in gerbil inner hair cells. J. Physiol. 586, 10291042 74 Bech-Hansen, N.T. et al. (1998) Loss-of-function mutations in a calcium-channel alpha1-subunit gene in Xp11.23 cause incomplete X-linked congenital stationary night blindness. Nat. Genet. 19, 264 267 75 Morgans, C.W. (2001) Localization of the alpha(1F) calcium channel subunit in the rat retina. Invest. Ophthalmol. Vis. Sci. 42, 24142418 76 Dolmetsch, R.E. et al. (2001) Signaling to the nucleus by an L-type calcium channel-calmodulin complex through the MAP kinase pathway. Science 294, 333339 77 Wheeler, D.G. et al. (2008) CaMKII locally encodes L-type channel activity to signal to nuclear CREB in excitation-transcription coupling. J. Cell Biol. 183, 849863 78 Nishiyama, M. et al. (2011) Semaphorin 3A induces CaV2.3 channeldependent conversion of axons to dendrites. Nat. Cell Biol. 13, 676 685 79 Zaman, T. et al. (2011) Cav2.3 channels are critical for oscillatory burst discharges in the reticular thalamus and absence epilepsy. Neuron 70, 95108 80 Vandael, D.H. et al. (2013) Cav1.3 and Cav1.2 channels of adrenal chromafn cells: Emerging views on cAMP/cGMP-mediated phosphorylation and role in pacemaking. Biochim. Biophys. Acta 1828, 16081618 81 Calin-Jageman, I. and Lee, A. (2008) Ca(v)1 L-type Ca2+ channel signaling complexes in neurons. J. Neurochem. 105, 573583 82 Dolphin, A.C. (2012) Calcium channel auxiliary alpha2delta and beta subunits: trafcking and one step beyond. Nat. Rev. Neurosci. 13, 542 555 83 Lipscombe, D. and Castiglioni, A.J. (2004) Alternative splicing in voltage gated calcium channels. In Calcium Channel Pharmacology (McDonough, S.I., ed.), pp. 369409, Kluwer Academic/Plenum 84 Boycott, K. et al. (2001) A summary of 20 CACNA1F mutations identied in 36 families with incomplete X-linked congenital stationary night blindness, and characterization of splice variants. Hum. Genet. 108, 9197 85 Soong, T.W. et al. (2002) Systematic identication of splice variants in human P/Q-type channel alpha1(2.1) subunits: implications for current density and Ca2+-dependent inactivation. J. Neurosci. 22, 1014210152 86 Pereverzev, A. et al. (2002) Alternate splicing in the cytosolic II-III loop and the carboxy terminus of human E-type voltage-gated Ca(2+) channels: electrophysiological characterization of isoforms. Mol. Cell. Neurosci. 21, 352365 87 Ernst, W.L. and Noebels, J.L. (2009) Expanded alternative splice isoform proling of the mouse Cav3.1/alpha1G T-type calcium channel. BMC Mol. Biol. 10, 53 88 Zhong, X. et al. (2006) A prole of alternative RNA splicing and transcript variation of CACNA1H, a human T-channel gene candidate for idiopathic generalized epilepsies. Hum. Mol. Genet. 15, 14971512 89 Gomora, J.C. et al. (2002) Cloning and expression of the human T-type channel Ca(v)3.3: insights into prepulse facilitation. Biophys. J. 83, 229241 90 Tang, Z.Z. et al. (2009) Developmental control of CaV1.2 L-type calcium channel splicing by Fox proteins. Mol. Cell. Biol. 29, 4757 4765 91 Minovitsky, S. et al. (2005) The splicing regulatory element, UGCAUG, is phylogenetically and spatially conserved in introns that ank tissue-specic alternative exons. Nucleic Acids Res. 33, 714724 92 Sievers, F. et al. (2011) Fast, scalable generation of high-quality protein multiple sequence alignments using Clustal Omega. Mol. Syst. Biol. 7, 539 93 Coppola, T. et al. (1994) Molecular cloning of a murine N-type calcium channel alpha 1 subunit. Evidence for isoforms, brain distribution, and chromosomal localization. FEBS Lett. 338, 15
Review
94 Pan, J.Q. and Lipscombe, D. (2000) Alternative splicing in the cytoplasmic II-III loop of the N-type Ca channel alpha 1B subunit: functional differences are beta subunit-specic. J. Neurosci. 20, 4769 4775 95 Sheng, Z.H. et al. (1996) Calcium-dependent interaction of N-type calcium channels with the synaptic core complex. Nature 379, 451 454 96 Rettig, J. et al. (1996) Isoform-specic interaction of the alpha1A subunits of brain Ca2+ channels with the presynaptic proteins syntaxin and SNAP-25. Proc. Natl. Acad. Sci. U.S.A. 93, 73637368 97 Yokoyama, C.T. et al. (2005) Mechanism of SNARE protein binding and regulation of Cav2 channels by phosphorylation of the synaptic protein interaction site. Mol. Cell. Neurosci. 28, 117 98 Rajapaksha, W.R. et al. (2008) Novel splice variants of rat CaV2.1 that lack much of the synaptic protein interaction site are expressed in neuroendocrine cells. J. Biol. Chem. 283, 1599716003 99 Raingo, J. et al. (2007) Alternative splicing controls G proteindependent inhibition of N-type calcium channels in nociceptors. Nat. Neurosci. 10, 285292 100 Swayne, L.A. et al. (2006) The cysteine string protein multimeric complex. Biochem. Biophys. Res. Commun. 348, 8391 101 Chi, X.X. et al. (2009) Regulation of N-type voltage-gated calcium channels (Cav2.2) and transmitter release by collapsin response mediator protein-2 (CRMP-2) in sensory neurons. J. Cell Sci. 122, 43514362 102 Nishimune, H. et al. (2004) A synaptic laminin-calcium channel interaction organizes active zones in motor nerve terminals. Nature 432, 580587 103 Leenders, A.G. et al. (2008) The role of MAP1A light chain 2 in synaptic surface retention of Cav2.2 channels in hippocampal neurons. J. Neurosci. 28, 1133311346 104 Khanna, R. et al. (2007) A proteomic screen for presynaptic terminal N-type calcium channel (CaV2.2) binding partners. J. Biochem. Mol. Biol. 40, 302314 105 Davare, M.A. et al. (2000) Protein phosphatase 2A is associated with class C L-type calcium channels (Cav1.2) and antagonizes channel phosphorylation by cAMP-dependent protein kinase. J. Biol. Chem. 275, 3971039717
106 Johnson, B.D. et al. (1994) Voltage-dependent potentiation of L-type Ca2+ channels in skeletal muscle cells requires anchored cAMPdependent protein kinase. Proc. Natl. Acad. Sci. U.S.A. 91, 1149211496 107 Oliveria, S.F. et al. (2007) AKAP79/150 anchoring of calcineurin controls neuronal L-type Ca2+ channel activity and nuclear signaling. Neuron 55, 261275 108 Davare, M.A. et al. (1999) The A-kinase anchor protein MAP2B and cAMP-dependent protein kinase are associated with class C L-type calcium channels in neurons. J. Biol. Chem. 274, 3028030287 109 Hall, D.D. et al. (2007) Critical role of cAMP-dependent protein kinase anchoring to the L-type calcium channel Cav1.2 via A-kinase anchor protein 150 in neurons. Biochemistry 46, 16351646 110 Kochlamazashvili, G. et al. (2010) The extracellular matrix molecule hyaluronic acid regulates hippocampal synaptic plasticity by modulating postsynaptic L-type Ca(2+) channels. Neuron 67, 116128 111 Kawaguchi, M. et al. (2006) Essential role of ubiquitin-proteasome system in normal regulation of insulin secretion. J. Biol. Chem. 281, 1301513020 112 Green, E.M. et al. (2007) The tumor suppressor eIF3e mediates calcium-dependent internalization of the L-type calcium channel CaV1.2. Neuron 55, 615632 113 Davare, M.A. and Hell, J.W. (2003) Increased phosphorylation of the neuronal L-type Ca(2+) channel Ca(v) 1.2 during aging. Proc. Natl. Acad. Sci. U.S.A. 100, 1601816023 114 Kersten, F.F. et al. (2010) Association of whirlin with Cav1.3 (alpha1D) channels in photoreceptors, dening a novel member of the usher protein network. Invest. Ophthalmol. Vis. Sci. 51, 2338 2346 115 Knirsch, M. et al. (2007) Persistence of Ca(v)1.3 Ca2+ channels in mature outer hair cells supports outer hair cell afferent signaling. J. Neurosci. 27, 64426451 116 Yang, P.S. et al. (2006) Switching of Ca2+-dependent inactivation of Ca(v)1.3 channels by calcium binding proteins of auditory hair cells. J. Neurosci. 26, 1067710689 117 Cui, G. et al. (2007) Ca2+-binding proteins tune Ca2+-feedback to Cav1.3 channels in mouse auditory hair cells. J. Physiol. 585, 791803 118 Zhang, H. et al. (2005) Association of CaV1.3 L-type calcium channels with Shank. J. Neurosci. 25, 10371049
609
Review
Blurring the boundaries: developmental and activity-dependent determinants of neural circuits

Verena Wolfram and Richard A. Baines
Faculty of Life Sciences, University of Manchester, Manchester, UK
The human brain comprises approximately 100 billion neurons that express a diverse, and often subtypespecic, set of neurotransmitters and voltage-gated ion channels. Given this enormous complexity, a fundamental question is how is this achieved? The acquisition of neurotransmitter phenotype was viewed as being set by developmental programs hard wired into the genome. By contrast, the expression of neuron-specic ion channels was considered to be highly dynamic (i.e., soft wired) and shaped largely by activity-dependent mechanisms. Recent evidence blurs this distinction by showing that neurotransmitter phenotype can be altered by activity and that neuron type-specic ion channel expression can be set, and perhaps limited by, developmental programs. Better understanding of these early regulatory mechanisms may offer new avenues to avert the behavioral changes that are characteristic of many mental illnesses. Neurons express diverse signaling properties Neural circuits in organisms as diverse as worms, ies, and humans exhibit remarkably similar design and developmental principles [13]. Circuit function depends on the concerted action of distinct classes of sensory neuron, regulatory interneuron and motor neuron. The function of each neuronal subtype is dened by its position, axon trajectory, synaptic connectivity, neurotransmitter expression, and electrophysiological properties. An important but unanswered question is how do neurons acquire subtypespecic properties? The answer undoubtedly depends on the relative contributions of both developmental programs (e.g., the type-specic transcription factor expression) and activity-dependent mechanisms. Although the identication of the developmental and activity-dependent mechanisms that shape axon trajectory and neurotransmitter phenotypes has progressed [46], the same is not true for the regulation of expression of ionic currents in early embryonic neurons. Progress has been hampered by the lack of suitable model systems in which genetics and electrophysiology can be combined at the level of identiable neurons. Recent developments in the fruit y, Drosophila melanogaster, and zebrash,
Corresponding author: Baines, R.A. (Richard.Baines@manchester.ac.uk). 0166-2236/$ see front matter 2013 Published by Elsevier Ltd. http://dx.doi.org/10.1016/j.tins.2013.06.006
Danio rerio, which allow such recordings, have started to yield important rst clues. In this review, we consider these recent ndings that directly relate to the question of how to build a neural circuit. We present an updated view of the potential contribution of differing regulatory mechanisms operative during development that determine active signaling properties of embryonic neurons. Neuronal specication Neuronal specication occurs early during embryogenesis and, for simplicity, can be considered to comprise three broad processes: proliferation, migration, and differentiation. Although all of these processes are crucial for the proper formation of neural circuits, we concentrate on neuronal differentiation in this review. It is during neuronal differentiation that neurons rst acquire their specic, and often unique, properties; these include axon projection, dendrite arborization, neurotransmitter specication, and ion channel expression. Until recently, the three former properties were considered to be highly stereotypic and were described as being hard wired. In support of this, many developmental transcription factors have been identied as important determinants that specify axon pathnding and neurotransmitter specication. By contrast, much less is known with regard to the developmental determinants that orchestrate ion channel gene expression. Indeed, this aspect of neuronal function has been shown to be dynamic and under extensive control of extrinsic neuronal activity [7] and activity-dependent homeostatic mechanisms [8]. Implicit in these studies is the concept that, unlike specication of neurotransmitters, the emergence of electrical properties in embryonic neurons is dependent on network activity and, as such, could be considered to be soft wired. However, the distinction between hard- and soft-wired properties is blurred by recent evidence that shows that neurotransmitter phenotype is inuenced by activity and that ion channel expression can be set by subtype-specic intrinsic developmental mechanisms. Neurotransmitter specication In comparison to ion channels, our understanding of the regulatory mechanisms that specify neurotransmitter expression is more comprehensive. The choice of neurotransmitter not only determines the functional modality of any given neuron, but is also critical for the functionality of the
610
Review
circuit to which each neuron contributes. As such, the specication of neurotransmitter phenotype is a key step for each and every neuron during embryonic development. Several studies have shown that neurotransmitters are set by neuron subtype-specic transcriptional programs and, as such, could be considered to be a hard wired characteristic. However, as we describe here, recent experiments overturn this view by showing that the choice of neurotransmitter can be respecied by neural activity [5]. The setting of neurotransmitter phenotype through cell type-specic developmental programs could be considered consistent with both the robustness and stability of neurotransmitter expression throughout the life of a neuron. There are several considerations that strengthen such a hypothesis. First, many neuron types generally express only one classical neurotransmitter [most likely acetylcholine (ACh), glutamate, GABA, serotonin, noradrenaline (nAdr), or dopamine (Da)] and often neurons with different classical transmitters develop from distinct pools of neuronal precursors [9]. Second, the restricted number of neurotransmitters and, in many cases [e.g., ACh, Da, nAdr, or 5hydroxytryptamine (5-HT)], the requirement of gene cassettes to produce that neurotransmitter is indicative of tight transcriptional regulation. Third, many circuits, such as sensory input circuitry or the motor network output, must maintain reliable transmission, which would be ensured by an early and stable encoding of neurotransmitter phenotype. Indeed, several studies (described below) describe genetic programs, active during development, that specify neurotransmitter expression. Transcription factor specication of neurotransmitter phenotype In different phyla, neurotransmission at the peripheral neuromuscular junction (NMJ) is mediated by different neurotransmitters. For example, in the nematode Caenorhabditis elegans, NMJs comprise a mixture of excitatory cholinergic and inhibitory GABAergic synapses [1013]. The excitatory NMJs of dipteran insects (i.e., Drosophila) and chordates (i.e., vertebrate) utilize glutamate and ACh, respectively [1416]. The expectation that neurotransmitter expression is tightly regulated at NMJs has been partly met by the identication of transcription factors of the LIM domain and homeodomain (HD domain) families, which are differentially expressed in motor neurons, where they orchestrate the developmental decisions of which neurotransmitter to express [1719]. Neurotransmitter specication is arguably best understood within the eight classes of C. elegans motor neuron (AS, DA, DB, DD, VA, VB, VC, and VD). For example, expression of the UNC-3 transcription factor is sufcient to specify a cholinergic phenotype in type A and B motor neurons (VA, VB, DA, DB, and AS) (Figure 1A) [17], whereas the HD transcription factor UNC-30 is required for the GABAergic phenotype of D motor neurons (VD, DD) [20,21]. In addition, AST-1, an E-twenty six (ETS) domain transcription factor, is sufcient to coordinate expression of genes required to synthesize Da [22]. Such observations are consistent with a simple, perhaps even a one factorone transmitter code. Acquisition of an appropriate neurotransmitter phenotype often requires coordinated expression of
several genes, including enzymes that are essential for the synthesis of transmitters, vesicular transporters, and, in some cases, autoreceptors. Coregulation of such gene cassettes by transcription factors is facilitated in one of two ways: either members of gene cassettes are organized within a single transcriptional unit or operon [23,24] or, when dispersed across the genome, are coordinately regulated by means of common cis-regulatory elements [25]. It seems unlikely, even in the relatively simple central nervous system (CNS) of C. elegans, that a one factorone transmitter code is sufcient for all neurotransmitter choices. Indeed, although UNC-3 species a cholinergic phenotype in A and B motor neurons [17], all cholinergic neurons are not specied by UNC-3; neither are all UNC-3 neurons cholinergic. For example, UNC-3 is not required for the cholinergic phenotype of the AIY interneuron, which is, instead, governed by the interplay of TTX-3 and CEH-10 [26]. UNC-3 expression is also observed in the noncholinergic ASI chemosensory neuron that releases neuropeptide-like proteins, such as N-acetylneuraminate pyruvate lyase (NPL1) [27,28]. These latter observations support the existence of more complicated and contextspecic transcription codes. Developmental studies in Drosophila motor neurons have made important contributions to understanding the mechanisms of neuronal differentiation. Conserved transcription factors, such as Even-skipped (Eve), Islet, Lim3, and Hb9, have been shown to have pivotal roles in neuronal subtype specication [2932]. These transcription factors are differentially expressed between motor neurons and subsets of interneurons, supporting a concept of combinatorial activity [30,31,33]. Interestingly, it is in interneurons where the potential to specify neurotransmitter phenotypes has been shown. For example, Islet is required for both serotonergic and dopaminergic interneuron phenotypes and, moreover, ectopic expression is sufcient to initiate expression of the Da-synthesizing enzyme tyrosine hydroxylase in some, but not all, neurons. Importantly, ectopic expression must occur early during neuronal development to alter transmitter phenotype, suggestive of the presence of a critical period [33]. In vertebrates, mature NMJs are cholinergic and most, if not all, motor neurons express Islet-1, Islet-2, Lim3 (Lhx3), and Hb9 (MNR2/MNX1), at some stage during their development. Expression of Islet-1, as well as MNR2 and Lhx3, has been associated with a cholinergic phenotype. Thus, ectopic expression of MNR2 in interneurons, normally expressed in paired box 6 (PAX6+) motor neuron progenitors, is sufcient to activate a motor neuron-like developmental program including the expression of choline acetyltransferase (ChAT), the rate-limiting enzyme in the synthesis of ACh [18]. However, it is clear that MNR2 (Hb9) alone is insufcient to determine a cholinergic phenotype because it is also expressed in noncholinergic neurons, such as mouse ventral spinal glutamatergic interneurons [34]. Transmitter choice can also be achieved by the active suppression of alternative transmitter phenotypes. Islet-1 participates in an early fate decision between zebrash primary motor neurons and interneurons by repressing the interneuronal GABAergic phenotype (Figure 1A) and, hence, the motor neuron phenotype is established [19].
611
Review
(A)
Caenorhabdis elegans
UNC-3
Zebrash
Islet
Chicken
GABA
MNR2
ACh
ChAT
Drosophila
Islet
Mouse
5-HT
DBX1
GABA
(B)
_ _ _
+ + +
Low acvity
+ + +
_ _ _ _ _ _
High acvity
+ + +
_ _ _
+ + + + +
_ _ _ _ _
+ + +
Figure 1. Specification of neurotransmitter phenotype. (A) Neurotransmitter phenotypes are genetically specified early during development by expression of developmental transcription factors. Transcription factors act by activating or repressing the transcription of proteins that are important for the synthesis and transport of neurotransmitters. Often, transcription factors are part of combinatorial codes and act together in a complex. Examples of transcription factors are given for neurons of representative species across different phyla. Arrows indicate that the transcription factors are required for expression of the respective neurotransmitter, whereas T-bars indicate a repressive effect. (B) Activity-dependent switching of neurotransmitter phenotypes. Studies in Xenopus demonstrate that enhanced (shown on the right) as well as reduced (shown on the left) neuronal activity is sufficient to induce a respecification of the neurotransmitter phenotype in neurons of the spinal cord to maintain an appropriate excitationinhibition balance. Decreased activity favors an increase in neurons expressing excitatory neurotransmitters [acetylcholine (ACh) and glutamate, orange circles], whereas an increase in activity leads to increased numbers of GABA-expressing neurons (blue circles) [48]. Abbreviations: 5-HT, 5-hydroxytryptamine; ChAT, choline acetyltransferase; DBX1, developing brain homeobox 1; UNC-3, uncoordinated-3.
Studies of vertebrate interneurons have also provided substantial clues to the complexity of neurotransmitter specication. Interneurons are either inhibitory, expressing GABA or glycine, or excitatory, mostly expressing glutamate. In mouse dorsal horn neurons, two transcription factors, T cell leukemia, homeobox 3 (TLX3) and ladybird homeobox 1 (LBX1), determine whether glutamate or GABA is expressed. Whereas TLX3 promotes the glutamatergic phenotype and suppresses the GABAergic phenotype [35], LBX1 promotes the GABAergic phenotype and suppresses the glutamatergic phenotype [36]. By manipulation of either TLX3 or LBX1, a neuron can be pushed toward excitatory or inhibitory neurotransmission, respectively. In neurons where both TLX3 and LBX1 are coexpressed, TLX3 antagonizes the function of LBX1, thus
612
ensuring that, ultimately, only one of these two transmitters, with diametrically opposite effects, is specied [36]. The above example demonstrates nicely how two transcription factors compete to specify neurotransmitter phenotypes. However, often multiple transcription factors work together to regulate the same neurotransmitter phenotype in different neuronal subsets and, as such, form a combinatorial code. Evidence for combinatorial activity includes the observations that Islet-1 represses the interneuron-specic GABAergic phenotype in zebrash motor neurons [19]. However, when Islet-1 is misexpressed in GABAergic interneurons, not all go on to acquire a motor neuron identity. Indeed, only interneurons that express the transcription factor Lhx3 seemingly have this potential. Lhx3 is also expressed in motor neurons, suggesting
Review
(A) Acvity (C)
(B)
Figure 2. Specification of ion channel expression in embryonic neurons. To acquire the unique electrical properties that will underpin the contribution of a neuron to a circuit, there are two more probable scenarios. (A) The neuron expresses a common default set of ion channels, making it almost indistinguishable from other neurons (as indicated by the consistency in color). Once part of a network (C), activity-dependent mechanisms shape the final cocktail of ion channels expressed. (B) Individual neuron subtypes express distinct sets of ion channels before the formation of networks, regulated by differential developmental programs (indicated by the different colors of the surrounding neurons). These differences in ion channel repertoire convey distinct functionality to each neuron that might be postulated to reduce the time required to produce functional networks. Similarly, once part of a functional network (C), activity-dependent mechanisms act to fine-tune those electrical properties.
that it is the coexpression of Islet-1 and Lhx3 that facilitates the manifestation of a motor neuron identity [37] and suppression of a GABAergic phenotype [19]. Similarly, in mouse, the GABAergic phenotype in excitatory interneurons is repressed by developing brain homeobox 1 (DBX1) [38]. Intriguingly, both DBX1 and Islet-1 are also expressed in GABAergic interneurons [3941], indicating that their ability to repress the GABAergic phenotype is context dependent. Similar to that observed in C. elegans [22,26], consensus-binding sequences have also been described for selected vertebrate transcriptional regulators. For example, the mouse ETS domain transcription factor Pet-1 binds a cisregulatory element that directs expression of serotonin pathway genes, including tryptophan hydroxylase 2 (Tph2), the rate-limiting enzyme in the synthesis of serotonin, and solute carrier family 6 (neurotransmitter transporter, serotonin), member 4 (Slc6a4), the serotonin transporter [42]. Moreover, Pet-1 is required throughout development and in to adult life to establish and maintain the serotonergic phenotype. Similarly, the homeodomain transcription factor paired-like homeodomain 3 (Pitx3), together with its interactor Nurr1, regulates genes required for Da synthesis, again through binding to specic promoter elements [43]. These types of observation are consistent with these transcription factors acting as terminal selectors of neurotransmitter phenotypes [44]. Activity-dependent respecication of neurotransmitter phenotype When considering motor neurons and sensory neurons, a permanent and stable neurotransmitter phenotype might impart stability for function, which would be consistent with the role of these types of neuron. By contrast, within more complex central interneuron networks, neurotransmitter plasticity could offer a mechanism to maintain the important balance between neuronal excitation and inhibition (E/I). An appropriate E/I balance has been hypothesized to be critical for neuronal development, and disturbance has been linked to an increased probability
for neurological disorders, such as seizure, autism, and schizophrenia [45]. However, it recently became apparent that both central interneurons and motor neurons can undergo activity-dependent respecication of neurotransmitter phenotype [46,47]. These experiments were carried out in Xenopus embryos where different classes of spinal cord neuron show distinct patterns of spontaneous Ca2+ spike activity [46]. Experimentally decreasing this activity, by expressing the Kir2.1 K+ channel, resulted in increased expression of excitatory neurotransmitters, glutamate and ACh, over the inhibitory transmitters, GABA and glycine. By contrast, increasing Ca2+ spike activity, by overexpressing a voltage-gated Na+ channel, was sufcient to induce a compensatory increase in inhibitory transmitter expression [46]. Such activity-dependent neurotransmitter respecication is achieved through a step in which excitatory and inhibitory neurotransmitters are coexpressed [46], but whether it always results in a complete replacement of transmitters remains to be evaluated (Figure 1B). Activity can also regulate the occurrence of the neuromodulator serotonin, with increasing activity resulting in a reduction of serotonergic neurons [48]. Where examined, neurotransmitter respecication by activity is transduced via an activity-dependent regulation of transcription factors, such as Tlx3 and Lmx1b [47,48]. As discussed above, in chick and mouse spinal cord, Tlx3 acts as molecular switch that favors glutamatergic over GABAergic neurotransmission. Ectopic expression of Tlx3 is sufcient to increase glutamatergic neurons at the expense of GABAergic cells, whereas loss-of-function mutants show the opposite effect [36]. Neurotransmitter switching, which might be important to maintain an appropriate E/I balance [46,47], is conned to a brief critical period before synapse formation. The details of how electrical activity acts to respecify neurotransmitter phenotype remain to be determined. However, it is clear that activity can alter transmitter phenotype at a time when genetically determined programs were widely believed to predominate.
613
Review
Box 1. Development of electrical properties of Drosophila embryonic central neurons

The first appearance of each current, recorded in motor neurons, and its continuity throughout development are shown in Figure I. Important behavioral outputs, such as first muscle movements to hatching, are also indicated. Time is counted from initial egg laying in hours. The first currents to express are the voltage-activated delayed rectifier K+ conductance and ligand-gated ACh currents. These are followed by voltage-activated Ca2+ and Na+ currents. A-type K+ currents, including Shaker and Shal, coincide with the appearance of the first action potentials, which are preceded in these neurons by excitatory postsynaptic potentials (EPSCs). Ionic currents similarly develop in a time-dependent manner in Xenopus neurons; however, here, voltagegated Ca2+ currents precede other currents [53]. Similar to Drosophila, the A-type K+ current shows a late onset in Xenopus [52].
ACh response
ICa INa IA
CNS
IK EPSCs 12 13 14 15 16 17
Acon potenals
18
19 20 Hours AEL
21
First Movements
Onset of coordinated movement
Hatching
Figure I. Time line for development of motor neuron electrical properties. Abbreviations: ACh, acetylcholine; AEL, after egg laying; CNS, central nervous system. Adapted from [51].
Ion channel expression Just as neurotransmitters underpin neuronal communication within a CNS, so a well-tuned set of voltage-gated ion channels is essential for the ability of the receiving neuron to integrate synaptic information and to instigate an appropriate neuronal output. In contrast to neurotransmitters, the identication of transcription factors capable of regulating ion channel gene expression is more limited. Questions also remain as to whether the same transcription factors coregulate both aspects of neuronal differentiation. As we describe below, early indications suggest that this is indeed so, raising the possibility of coregulation of multiple aspects of neuronal signaling through common developmental mechanisms. Ion channel expression, plasticity, and homeostasis Electrophysiological analysis of mature neurons has generated a paradox. This is because, although it is possible to distinguish, and even identify, neuron subtypes by their characteristic expression of ion channels, the same electrophysiological behavior can, in silico, be induced by multiple disparate sets of underlying ion channels [49,50]. Therefore, a key question is the extent to which ion channel gene expression is regulated by developmental, as opposed to activity-dependent, mechanisms. The former might be expected to result in xed expression levels of ion channel genes, whereas the latter might achieve appropriate circuit outputs through more varied expression patterns. It could be envisaged that the ion channel repertoire of embryonic neuron subtypes is indistinguishable from one another, representing a default, developmentally determined, state (Figure 2). The stereotypic and sequential expression of specic ion channels by developing neurons is
614
consistent with this view [5153]. By contrast, numerous studies have highlighted the importance of activity-dependent homeostatic regulation of ion channel expression in both developing and mature neurons [5456]. Where observed, homeostatic regulation of neuronal electrical properties is considered important for network stability by maintaining a constant neuronal output (i.e., action potential ring) in response to changes in network synaptic activity, rapid turnover of ion channels, or unpredictable network perturbations [57]. A key substrate for homeostatic plasticity is the repertoire of voltage-dependent conductances expressed by neurons, in particular voltage-gated Na+ conductances [5862]. Because of the presumed importance that activity has in establishing, rening, and maintaining the ion channel repertoire of a given neuron, the role of developmental transcriptional programs has only very recently been appreciated. Now a new picture emerges in which developmental factors specify neuron subtype-specic ion channel expression proles before circuit formation. Of course, these properties are likely open to subsequent activity-dependent modication once network activity is established. Ion channel specication by developmental factors Computational modeling has suggested that a xed neuronal output can be achieved by an almost random combination of many ion channels, creating an almost indenite parameter space [49]. Electrophysiological analysis of biological neurons shows that the actual parameter space is more restricted. Elegant work from the Marder laboratory and colleagues has shown that the expression of ion channels, in a mature neuron, does not vary randomly but rather in a coordinated fashion, with pairs or sets of channels exhibiting either transcriptional correlation or
Review
(A) * **
dMN incl. * RP2 and ** aCC vMN incl. RP1, RP3, RP4 and RP5
IKv
WT dMN
AC PC
200 pA/pF WT vMN
10 ms Dorsal muscles Ventral muscles 90 mV 40 mV 60 mV
(B)
MiP
MiP (dMN) CaP (vMN)
CaP
500 pA 10 ms
20 mV 80 mV 40 mV
Figure 3. Embryonic motor neurons show distinct electrophysiological properties before active synapse formation. (A) Diagram of the ventral nerve cord in a late-stage Drosophila embryo. Dorsal motor neurons (dMN, in magenta) express the transcription factor Even-skipped and project to dorsal muscles. Ventral motor neurons (vMN, in green) express the transcription factor Islet and project to ventral muscles. Voltage clamp experiments show differences in outward K+ currents between the two motor neuron populations that are independent of synaptic activity. Magnitude of currents are normalized to cell capacitance. (B) Diagram of the zebrafish spinal cord depicting two distinct primary motor neurons, the dorsal projecting, Islet-1 expressing MiP (magenta) and the ventral projecting, Islet-2 expressing CaP (green). Voltage clamp recordings show that K+ currents are larger in ventral motor neurons compared with dorsal motor neurons. Adapted and modified from [67] (A) and [66] (B).
anticorrelation [63,64]. This may be indicative that, similar to neurotransmitters, ion channels are also regulated in a gene battery-like manner. Transcription linkage of ion channels might also explain why neurons of the same type or origin can be distinguished by their electrophysiological properties (i.e., by the ion channels they express). In addition, it points toward single neurons expressing a restricted ion channel repertoire that may be acquired through early developmental mechanisms. Where it has been investigated, ion channel expression, similar to the expression of neurotransmitters, is detected very early during embryonic development and before synapse formation [51,53] (Box 1). It is tempting to speculate that this is because both are coregulated by the same developmental mechanisms. This early expression is independent of external factors, such as synaptic activity, and is seen specically for ion channels required for basic neuronal excitation, such as voltage-gated Na+, Ca2+, and K+ channels [52]. However, the majority of these studies were conned to mostly unidentied neurons and one cannot exclude the possibility that neurons express a default set of ion channels in a time-dependent manner. What has been missing is a comparative analysis of the acquisition of electrical conductances within
individual neurons of particular subtypes (e.g., motor neurons). One of the rst model organisms where such a comparative analysis was possible is Drosophila due, in greater part, to the stereotypic position of identiable embryonic motor neurons and their accessibility to electrophysiology. Within the past few years, it has been established in both Drosophila and zebrash that distinct motor neurons exhibit different electrical properties before synapse formation (Figure 3) [6567]. These initial studies provided the rst clear indications that the expression of at least some ion channels is regulated by early developmental mechanisms and that these are specic to particular neuronal subtypes. In accordance with this hypothesis, ion channels have been identied as potential targets of differentially expressed transcription factors in worms, ies, and vertebrates [17,65,68]. Specic examples include Ca2+ and K+ voltage-activated channels in C. elegans that contain a motif (COE motif), which is recognized by the UNC-3 transcription factor, a determinant of the cholinergic phenotype of type A and B motor neurons [17]. However, no electrophysiological data are available for these motor neurons. By contrast, the terminal selector gene TTX3 is required for the normal expression of
615
Review
Box 2. Slowpoke and Shaker K+ currents contribute to action potential firing
Slowpoke Slowpoke is the Drosophila homolog of vertebrate Ca2+-gated K+ channels (BKs) [77,78]. BKs are activated by membrane depolarization and simultaneous increase in intracellular Ca2+ and contribute to the repolarizing phase of the action potential and the afterhyperpolarization [79]. Depending on cell context, BKs can both dampen or increase action potential firing [65,80]. Shaker Shaker is the Drosophila homolog of vertebrate Kv1.1, a voltagedependent K+ channel, carrying A-type current (fast activating and inactivating) [8183]. Shaker and its homologs are activated by membrane depolarization. They are involved in repolarization of the action potential and, as such, can modulate firing frequency [84] and spike propagation [85]. Often, the presence of the A-type channel and its localization leads to a reduction or dampening of action potential firing [86].
transcription factors during early development. Such mechanisms would facilitate the acquisition of subtypespecic ion channel repertoires on which plasticity and homeostasis can subsequently act during later postembryonic stages. Concluding remarks Coordinated gene expression, coupled with activity-dependent renement, underpins the formation of functional neural circuits. The examples described above illustrate the many interactions so far documented between developmental mechanisms (i.e., transcription factors) and neuronal activity in the specication of neurotransmitter phenotype and ion channel expression in developing neurons. Although these are early days, several themes are beginning to emerge that will hopefully be explored over coming years. The rst is that the canonical view of properties such as specication of neurotransmitters being developmentally hard wired whereas expression of ion channels is set, largely, as part of activity-dependent feedback (i.e., soft wired) is blurred by recent experiments. The demonstration that neuron subtype-specic electrical properties are set by developmental mechanisms is particularly important because it is consistent with the viewpoint that neural circuit function and, therefore, behavior, are encoded to some degree within the genome. Although not a new concept [74], these rst glimpses of developmental determination of ion channel gene expression offer substantial evidence to support this view. Much might be gained from re-evaluating whether some neurological disorders arise from incorrect developmental specication of ion channel gene expression during early neurogenesis. By contrast, activity-dependent respecication of neurotransmitter content overturns the long-held view that this neuronal property is developmentally xed. Interesting questions include the precise timing during which the expression of both neurotransmitter-associated and ion channel genes are available to modication during neurogenesis and whether change to one might instigate obligatory change to the other. A second important theme to emerge is that multiple aspects of neuronal differentiation are regulated by common factors. Perhaps one of the best examples is from Drosophila, where the transcription factor Islet is seemingly able to regulate axon guidance, neurotransmitter phenotype, and expression of electrical properties. Might we consider Islet to be a terminal selector and will all of its target genes be identiable by the presence of conserved cis-regulatory motifs similar to what has been observed for some C. elegans transcription factors (e.g., ttx-3 and unc-3) [17,22,26]? Related questions include whether there are additional terminal selectors that orchestrate separate or overlapping cassettes of gene targets. Although recent studies in C. elegans support this view [75,76], whether this will also hold true for other organisms remains to be determined. Taken to the extreme, this might mean that we will be able to predict key neuronal properties based on the prole of such transcription factor expression. Although these, and other, important questions remain, the prospect is bright. The recent combination of
outward K+ currents in AIY interneurons [69], although in an earlier study, no specic voltage-gated ion channel target was identied [26]. In Drosophila dorsally projecting motor neurons, Eve has been shown to regulate the expression of Slowpoke, a Ca2+ and voltage-gated K+ channel of the BK family [65]. In addition, Islet-1 knockout was shown to inuence the transcription of several ion channel genes in zebrash motor neurons, as evidenced by microarray studies [68]. Perhaps the most conclusive studies to date for the early developmental regulation of ion channel expression come from Drosophila, where a direct link between the differential expression of transcription factors, ion channels, and electrophysiological properties has been recently demonstrated. In the ventral nerve cord of the Drosophila larva, motor neurons can be readily identied by cell body position, axonal projection, and muscle targets [7072]. Differences in axonal targeting, to either dorsal or ventral body wall muscles, have been linked to subtype-specic expression of developmentally important transcription factors, in particular expression of Eve being required for dorsal axon targeting and Islet for ventral axon targeting [29,73]. Whereas Eve is expressed in motor neurons innervating dorsal muscles, so-called dorsal motor neurons [29], Islet, Lim3, and Hb9 are expressed in motor neurons that target ventral muscles, termed ventral motor neurons [30,32,33]. Recently, it has been shown that dorsal and ventral motor neurons differ in their electrophysiological properties (Figure 3A) [65,67]. Specically, dorsal motor neurons exhibit larger outward K+ currents (Figure 3) and re fewer action potentials (Figure 4) [67]. Two transcription factors, Eve and Islet, have been linked to subtype-specic ion channel gene expression. In dorsal motor neurons [29] Eve is sufcient to downregulate, but not abolish, the expression of slowpoke [65]. By contrast, in ventral motor neurons, Islet is both necessary and sufcient to repress completely the Shaker K+ current, an A-type K+ current analogous to vertebrate Kv1.1 [67] (Box 2). Whereas Slowpoke is expressed in both motor neuron types, albeit to varying amounts, Shaker is absent from ventral motor neurons. This is indicative that not only type, but also the relative level, of ion channels are regulated by
616
Review
(A) dMN (B)
vMN
+ Islet Islet Sh - Islet
Sh
Figure 4. Islet is deterministic for motor neuron subtype electrical properties in Drosophila. (A) A dorsal (aCC) motor neuron labeled by DiI applied to its neuromuscular junction (NMJ) with its target muscle (muscle 1). The dorsal motor neuron expresses the Shaker K+ channel, which conducts a fast activating-inactivating potassium current akin to the A-type current (Box 2, main text). The traces show typical recordings of action potential firing obtained by current injection at the soma (10 pA for 500 ms) via whole-cell current clamp. (B) A ventral motor neuron labeled by DiI applied to its NMJ with its target muscle (muscle 6). The ventral motor neurons do not express the Shaker K+ channel due to transcriptional repression mediated by Islet. In the absence of Shaker, the neuron fires comparatively more action potentials during 500 ms of 10 pA current injection [compare to (A)]. When Islet is ectopically expressed in dorsal motor neurons, their endogenous Shaker K+ current is diminished. By contrast, loss of function of Islet in ventral motor neurons results in a pronounced Shaker K+ current [67]. This demonstrates that, at least in Drosophila, Islet forms part of an developmental decision-making process that is critical to specifying subtype-specific electrical properties in developing motor neurons before neural circuit formation.
electrophysiology and molecular genetics, in worms, ies, and sh (to name just a few), offers the prospect of making progress to answer these and other related questions to determine the relative contribution of developmental versus activity-dependent mechanisms for the formation of neural circuits.
Acknowledgments
We thank Matthias Landgraf, Stefan Pulver, and Gino Poulin for their comments. We also thank Carlo Giachello for his help in constructing the gures. We are also grateful to Rosa Moreno and Angeles Ribera for permission to reproduce their work in this review. The experiments described from the Baines group were supported by the Wellcome Trust.
References
1 Eisen, J. (1998) Genetic and molecular analyses of motoneuron development. Curr. Opin. Neurobiol. 8, 697704 2 Jurata, L.W. et al. (2000) Transcriptional mechanisms in the development of motor control. Curr. Opin. Neurobiol. 10, 7279 3 Grillner, S. and Jessell, T.M. (2009) Measured motion: searching for simplicity in spinal locomotor networks. Curr. Opin. Neurobiol. 19, 572586 4 Polleux, F. et al. (2007) Transcriptional regulation of vertebrate axon guidance and synapse formation. Nat. Rev. Neurosci. 8, 331340 5 Spitzer, N. (2012) Activity-dependent neurotransmitter respecication. Nat. Rev. Neurosci. 13, 94106 6 Landgraf, M. and Thor, S. (2006) Development and structure of motoneurons. Int. Rev. Neurobiol. 75, 3353 7 Spitzer, N.C. et al. (2000) Differentiation of electrical excitability in motoneurons. Brain Res. Bull. 53, 547552
8 Baines, R.A. et al. (2001) Altered electrical properties in Drosophila neurons developing without synaptic transmission. J. Neurosci. 21, 19 9 Anderson, R.L. et al. (2001) Development of electrophysiological and morphological diversity in autonomic neurons. J. Neurophysiol. 86, 12371251 10 del Castillo, J. et al. (1963) The physiological role of acetylcholine in the neuromuscular system of Ascaris lumbricoides. Arch. Int. Physiol. Biochim. 71, 741757 11 del Castillo, J. et al. (1964) Inhibitory action of gamma-aminobutyric acid (GABA) on Ascaris muscle. Experientia 20, 141143 12 del Castillo, J. et al. (1967) The initiation of action potentials in the somatic musculature of Ascaris lumbricoides. J. Exp. Biol. 46, 263279 13 Hobert, O. (2010) Neurogenesis in the nematode Caenorhabditis elegans. In WormBook (The C. elegans Research Community, ed.), pp. 124, WormBook 14 Ladle, D.R. et al. (2007) Assembly of motor circuits in the spinal cord: driven to function by genetic and experience-dependent mechanisms. Neuron 56, 270283 15 Fatt, P. and Katz, B. (1951) An analysis of the end-plate potential recorded with an intracellular electrode. J. Physiol. 115, 320370 16 Jan, L.Y. and Jan, Y.N. (1976) L-glutamate as an excitatory transmitter at the Drosophila larval neuromuscular junction. J. Physiol. 262, 215236 17 Kratsios, P. et al. (2012) Coordinated regulation of cholinergic motor neuron traits through a conserved terminal selector gene. Nat. Neurosci. 15, 111 18 Tanabe, Y. et al. (1998) Specication of motor neuron identity by the MNR2 homeodomain protein. Cell 95, 6780 19 Hutchinson, S.A. and Eisen, J.S. (2006) Islet1 and Islet2 have equivalent abilities to promote motoneuron formation and to specify motoneuron subtype identity. Development 133, 21372147
617
Review
20 Jin, Y. et al. (1994) Control of type-D GABAergic neuron differentiation by C. elegans UNC-30 homeodomain protein. Nature 372, 780783 21 Eastman, C. et al. (1999) Coordinated transcriptional regulation of the unc-25 glutamic acid decarboxylase and the unc-47 GABA vesicular transporter by the Caenorhabditis elegans UNC-30 homeodomain protein. J. Neurosci. 19, 62256234 22 Flames, N. and Hobert, O. (2009) Gene regulatory logic of dopamine neuron differentiation. Nature 458, 885889 23 Alfonso, A. et al. (1994) Alternative splicing leads to two cholinergic proteins in Caenorhabditis elegans. J. Mol. Biol. 241, 627630 24 Treinin, M. et al. (1998) Two functionally dependent acetylcholine subunits are encoded in a single Caenorhabditis elegans operon. Proc. Natl. Acad. Sci. U.S.A. 95, 1549215495 25 Hobert, O. et al. (2010) The molecular and gene regulatory signature of a neuron. Trends Neurosci. 33, 435445 26 Wenick, A.S. and Hobert, O. (2004) Genomic cis-regulatory architecture and trans-acting regulators of a single interneuronspecic gene battery in C. elegans. Dev. Cell 6, 757770 27 Nathoo, A.N. et al. (2001) Identication of neuropeptide-like protein gene families in Caenorhabditis elegans and other species. Proc. Natl. Acad. Sci. U.S.A. 98, 1400014005 28 Kim, K. et al. (2005) The UNC-3 Olf/EBF protein represses alternate neuronal programs to specify chemosensory neuron identity. Dev. Biol. 286, 136148 29 Landgraf, M. et al. (1999) even-skipped determines the dorsal growth of motor axons in Drosophila. Neuron 22, 4352 30 Odden, J.P. et al. (2002) Drosophila HB9 is expressed in a subset of motoneurons and interneurons, where it regulates gene expression and axon pathnding. J. Neurosci. 22, 91439149 31 Broihier, H.T. and Skeath, J.B. (2002) Drosophila homeodomain protein dHb9 directs neuronal fate via crossrepressive and cellnonautonomous mechanisms. Neuron 35, 3950 32 Thor, S. et al. (1999) A LIM-homeodomain combinatorial code for motor-neuron pathway selection. Nature 397, 7680 33 Thor, S. and Thomas, J.B. (1997) The Drosophila islet gene governs axon pathnding and neurotransmitter identity. Neuron 18, 397409 34 Wilson, J.M. et al. (2005) Conditional rhythmicity of ventral spinal interneurons dened by expression of the Hb9 homeodomain protein. J. Neurosci. 25, 57105719 35 Cheng, L. et al. (2004) Tlx3 and Tlx1 are post-mitotic selector genes determining glutamatergic over GABAergic cell fates. Nat. Neurosci. 7, 510517 36 Cheng, L. et al. (2005) Lbx1 and Tlx3 are opposing switches in determining GABAergic versus glutamatergic transmitter phenotypes. Nat. Neurosci. 8, 15101515 37 Thaler, J.P. et al. (2002) LIM factor Lhx3 contributes to the specication of motor neuron and interneuron identity through celltype-specic proteinprotein interactions. Cell 110, 237249 38 Pierani, A. et al. (2001) Control of interneuron fate in the developing spinal cord by the progenitor homeodomain protein Dbx1. Neuron 29, 367384 39 Lacin, H. et al. (2009) dbx mediates neuronal specication and differentiation through cross-repressive, lineage-specic interactions with eve and hb9. Development 136, 32573266 40 Liem, K.F., Jr et al. (1997) A role for the roof plate and its resident TGFbeta-related proteins in neuronal patterning in the dorsal spinal cord. Cell 91, 127138 41 Muroyama, Y. et al. (2002) Wnt signaling plays an essential role in neuronal specication of the dorsal spinal cord. Genes Dev. 16, 548553 42 Liu, C. et al. (2010) Pet-1 is required across different stages of life to regulate serotonergic function. Nat. Neurosci. 13, 11901198 43 Jacobs, F. et al. (2009) Pitx3 potentiates Nurr1 in dopamine neuron terminal differentiation through release of SMRT-mediated repression. Development 136, 531540 44 Hobert, O. (2008) Regulatory logic of neuronal diversity: terminal selector genes and selector motifs. Proc. Natl. Acad. Sci. U.S.A. 105, 2006720071 45 Yizhar, O. et al. (2011) Neocortical excitation/inhibition balance in information processing and social dysfunction. Nature 477, 171178 46 Borodinsky, L.N. et al. (2004) Activity-dependent homeostatic specication of transmitter expression in embryonic neurons. Nature 429, 18
618
47 Marek, K.W. et al. (2010) cJun integrates calcium activity and tlx3 expression to regulate neurotransmitter specication. Nat. Neurosci. 13, 18 48 Demarque, M. and Spitzer, N.C. (2010) Activity-dependent expression of Lmx1b regulates specication of serotonergic neurons modulating swimming behavior. Neuron 67, 114 49 Marder, E. and Taylor, A.L. (2011) Multiple models to capture the variability in biological neurons and networks. Nat. Neurosci. 14, 133 138 50 Schulz, D.J. et al. (2006) Cellular excitability and the regulation of functional neuronal identity: from gene expression to neuromodulation. J. Neurosci. 26, 1036210367 51 Baines, R.A. and Bate, M. (1998) Electrophysiological development of central neurons in the Drosophila embryo. J. Neurosci. 18, 46734683 52 Ribera, A.B. and Spitzer, N.C. (1990) Differentiation of IKA in amphibian spinal neurons. J. Neurosci. 10, 18861891 53 ODowd, D.K. et al. (1988) Development of voltage-dependent calcium, sodium, and potassium currents in Xenopus spinal neurons. J. Neurosci. 8, 792805 54 Turrigiano, G. et al. (1994) Activity-dependent changes in the intrinsic properties of cultured neurons. Science 264, 974977 55 Turrigiano, G.G. and Nelson, S.B. (2004) Homeostatic plasticity in the developing nervous system. Nat. Rev. Neurosci. 5, 97107 56 Baines, R.A. (2005) Neuronal homeostasis through translational control. Mol. Neurobiol. 32, 19 57 Turrigiano, G. (2012) Homeostatic synaptic plasticity: local and global mechanisms for stabilizing neuronal function. Cold Spring Harb. Perspect. Biol. 4, a005736 58 Ransdell, J.L. et al. (2012) Rapid homeostatic plasticity of intrinsic excitability in a central pattern generator network stabilizes functional neural network output. J. Neurosci. 32, 96499658 59 Muraro, N.I. et al. (2008) Pumilio binds para mRNA and requires Nanos and Brat to regulate sodium current in Drosophila motoneurons. J. Neurosci. 28, 20992109 60 Desai, N.S. et al. (1999) Plasticity in the intrinsic excitability of cortical pyramidal neurons. Nat. Neurosci. 2, 515520 61 Mee, C.J. et al. (2004) Regulation of neuronal excitability through pumilio-dependent control of a sodium channel gene. J. Neurosci. 24, 86958703 62 Driscoll, H.E. et al. (2013) Pumilio-2 regulates translation of nav1.6 to mediate homeostasis of membrane excitability. J. Neurosci. 33, 9644 9654 63 Schulz, D.J. et al. (2006) Variable channel expression in identied single and electrically coupled neurons in different animals. Nat. Neurosci. 9, 356362 64 Schulz, D.J. et al. (2007) Quantitative expression proling of identied neurons reveals cell-specic constraints on highly variable levels of gene expression. Proc. Natl. Acad. Sci. U.S.A. 104, 15 65 Pym, E.C. et al. (2006) The homeobox transcription factor Evenskipped regulates acquisition of electrical properties in Drosophila neurons. Neural Dev. 1, 116 66 Moreno, R.L. and Ribera, A.B. (2009) Zebrash motor neuron subtypes differ electrically prior to axonal outgrowth. J. Neurophysiol. 102, 24772484 67 Wolfram, V. et al. (2012) The LIM-homeodomain protein islet dictates motor neuron electrical properties by regulating K+ channel expression. Neuron 75, 663674 68 Sun, Y. et al. (2008) A central role for Islet1 in sensory neuron development linking sensory and spinal gene regulatory programs. Nat. Neurosci. 11, 12831293 69 Faumont, S. et al. (2006) Developmental regulation of whole cell capacitance and membrane current in identied interneurons in C. elegans. J. Neurophysiol. 95, 36653673 70 Landgraf, M. et al. (1997) The origin, location, and projections of the embryonic abdominal motorneurons of Drosophila. J. Neurosci. 17, 96429655 71 Hoang, B. and Chiba, A. (2001) Single-cell analysis of Drosophila larval neuromuscular synapses. Dev. Biol. 229, 5570 72 Baines, R.A. et al. (1999) Postsynaptic expression of tetanus toxin light chain blocks synaptogenesis in Drosophila. Curr. Biol. 9, 15 73 Thor, S. and Thomas, J.B. (2002) Motor neuron specication in worms, ies and mice: conserved and lost mechanisms. Curr. Opin. Genet. Dev. 12, 558564
Review
74 Wilson, D. (1968) Inherent asymmetry and reex modulation of the locust ight motor pattern. J. Exp. Biol. 48, 631641 75 Hobert, O. (2011) Regulation of terminal differentiation programs in the nervous system. Annu. Rev. Cell Dev. Biol. 27, 681696 76 Doitsidou, M. et al. (2013) A combinatorial regulatory signature controls terminal differentiation of the dopaminergic nervous system in C. elegans. Genes Dev. 27, 13911405 77 Elkins, T. et al. (1986) A Drosophila mutation that eliminates a calciumdependent potassium current. Proc. Natl. Acad. Sci. U.S.A. 83, 84158419 78 Atkinson, N.S. et al. (1991) A component of calcium-activated potassium channels encoded by the Drosophila slo locus. Science 253, 551555 79 Adams, P.R. et al. (1982) Intracellular Ca2+ activates a fast voltagesensitive K+ current in vertebrate sympathetic neurones. Nature 296, 746749 80 Montgomery, J.R. and Meredith, A.L. (2012) Genetic activation of BK currents in vivo generates bidirectional effects on neuronal excitability. Proc. Natl. Acad. Sci. U.S.A. 109, 1899719002
81 Tempel, B.L. et al. (1988) Cloning of a probable potassium channel gene from mouse brain. Nature 332, 837839 82 Papazian, D.M. et al. (1987) Cloning of genomic and complementary DNA from Shaker, a putative potassium channel gene from Drosophila. Science 237, 749753 83 Tempel, B.L. et al. (1987) Sequence of a probable potassium channel component encoded at Shaker locus of Drosophila. Science 237, 770775 84 Connor, J.A. and Stevens, C.F. (1971) Prediction of repetitive ring behaviour from voltage clamp data on an isolated neurone soma. J. Physiol. 213, 3153 85 Debanne, D. et al. (1997) Action-potential propagation gated by an axonal I(A)-like K+ conductance in hippocampus. Nature 389, 286289 86 Brewster, D.L. and Ali, D.W. (2013) Expression of the voltage-gated potassium channel subunit Kv1.1 in embryonic zebrash Mauthner cells. Neurosci. Lett. 539, 5459
619

Trends in Neurosciences - October 2013

Diunggah oleh

Informasi Dokumen

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Trends in Neurosciences - October 2013

Diunggah oleh

Hak Cipta:

Format Tersedia

October 2013 Volume 36, Number 10 pp.

Looking back and looking forward

The Brain Prize 2013: the optogenetics revolution

Andreas Reiner and Ehud Y. Isacoff

Arvind Kumar, Ioannis Vlachos, Ad Aertsen, and Clemens Boucsein

Philipp Mergenthaler, Ute Lindauer, Gerald A. Dienel, and Andreas Meisel

Looking back and looking forward

Corresponding author: Clark, A.M. (tins@cell.com).

Trends in Neurosciences October 2013, Vol. 36, No. 10

Science & Society

The Brain Prize 2013: the optogenetics revolution

Science & Society

Trends in Neurosciences October 2013, Vol. 36, No. 10

Science & Society

Trends in Neurosciences October 2013, Vol. 36, No. 10

Science & Society

Trends in Neurosciences October 2013, Vol. 36, No. 10

Molecular nexopathies: a new paradigm of neurodegenerative disease

Trends in Neurosciences October 2013, Vol. 36, No. 10

Trends in Neurosciences October 2013, Vol. 36, No. 10

(C) Distributed no gradient

Trends in Neurosciences October 2013, Vol. 36, No. 10

Trends in Neurosciences October 2013, Vol. 36, No. 10

Trends in Neurosciences October 2013, Vol. 36, No. 10

Box 2. Testing the paradigm: outstanding questions and future directions

Trends in Neurosciences October 2013, Vol. 36, No. 10

Trends in Neurosciences October 2013, Vol. 36, No. 10

Trends in Neurosciences October 2013, Vol. 36, No. 10

Trends in Neurosciences, October 2013, Vol. 36, No. 10

Trends in Neurosciences October 2013, Vol. 36, No. 10

(C) M D SP PPH PH PMH MH

(D) p2 CSP p3 m2 r2 p1 m1 is r1 r3 r9 r4 r5 r6 r7 r8 r10 r11

SpC r11 r9 r10 r5 r6 r7 r8 Alarbasal boundary

Trends in Neurosciences October 2013, Vol. 36, No. 10

Pallidum Dg Dg Septal roofplate

Preopc area Anterior commissure

Trends in Neurosciences October 2013, Vol. 36, No. 10

Pall Diag Preopt

Sept Parasept Preopt

Pall Diag Central Amygd

Trends in Neurosciences October 2013, Vol. 36, No. 10

Trends in Neurosciences October 2013, Vol. 36, No. 10

Trends in Neurosciences October 2013, Vol. 36, No. 10

Trends in Neurosciences October 2013, Vol. 36, No. 10

Trends in Neurosciences October 2013, Vol. 36, No. 10

Challenges of understanding brain function by selective modulation of neuronal subpopulations

Trends in Neurosciences October 2013, Vol. 36, No. 10

Recurrent exc. inh. network e.g., CA3, neocorcal layers

Trends in Neurosciences October 2013, Vol. 36, No. 10

Trends in Neurosciences October 2013, Vol. 36, No. 10

Populaon acvity (sp/s)

Network ID 2000 2500 3000

0 0 0 500 1000 1500 Out-degree 0.0 0.2 0.4

Trends in Neurosciences October 2013, Vol. 36, No. 10

Trends in Neurosciences October 2013, Vol. 36, No. 10

Trends in Neurosciences October 2013, Vol. 36, No. 10

Trends in Neurosciences October 2013, Vol. 36, No. 10

Trends in Neurosciences October 2013, Vol. 36, No. 10

Neurons Glc Lac

Liver Gastrointesnal tract Pancreas

Endothelial cells (BBB)

Aerent and eerent vagus

Trends in Neurosciences October 2013, Vol. 36, No. 10

Trends in Neurosciences October 2013, Vol. 36, No. 10