R OB E R T S HAP L E Y AND C HR I S T I NA E NR OT H- C UGE L L Laboratory o f Biophysics, Rockefeller University, New York, New York 10021, USA and Departments of Neurobiology & Physiology, and Engineering Sciences & Applied Mathematics, Northwestern University, Evanston, lllinois 60201, USA CONTENTS 1. I n t r o d u c t i o n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264 1. 1. T e r mi n o l o g y . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264 1. 2. Th e P u r p o s e o f Ad a p t a t i o n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267 2. P s y c h o p h y s i c a l La ws o f Li g h t Ad a p t a t i o n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276 2. 1. Th e Ro d Sy s t e m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277 2. 2. Th e Co n e Sy s t e m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290 3. Ga i n a n d Co n t r a s t Ga i n i n Re t i n a l Ga n g l i o n Cel l s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294 3. 1. Ga i n Co n t r o l i n t he S c o t o p i c Ra n g e . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296 3. 2. Ga i n Co n t r o l i n t he P h o t o p i c Ra n g e . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302 3. 3. Co n t r a s t Ga i n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303 3. 4. Ef f e c t o f Ad a p t a t i o n o n t he Si ze o f t he Re c e p t i v e Fi e l d Ce n t e r . . . . . . . . . . . . . . . . . . . . . 304 3. 5. Ad a p t a t i o n a l P o o l i n g a n d Re c e p t i v e Fi e l d Si ze . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306 3. 6. Ga i n Co n t r o l a n d Re c e p t i v e Fi e l d Si ze Ac r o s s t he P o p u l a t i o n o f Ga n g l i o n Cel l s . . . . . . 312 3. 7. Th e Ef f e c t o f t he Re c e p t i v e Fi e l d S u r r o u n d o n t he Ga i n o f t he Ce n t e r . . . . . . . . . . . . . . 314 3. 8. A Di f f e r e n t k i n d o f Ga i n Co n t r o l : Th e Co n t r a s t Ga i n Co n t r o l . . . . . . . . . . . . . . . . . . . . . 316 3. 9. Ti me Co u r s e o f Ga i n Ad j u s t me n t i n Re t i n a l Ga n g l i o n Cel l s . . . . . . . . . . . . . . . . . . . . . . . . 318 4. Ga i n as a F u n c t i o n o f I l l u mi n a t i o n i n Ama c r i n e Cel l s , Bi p o l a r Cel l s , a n d Ho r i z o n t a l Cel l s . 320 4. 1. Ama c r i n e Cel l s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320 4. 2. Bi p o l a r Cel l s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321 4. 3. Ho r i z o n t a l Cel l s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323 5. Ga i n Co n t r o l i n P h o t o r e c e p t o r s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324 5. 1. P h o t o r e c e p t o r s whi c h Ad a p t a l ot a n d S a t u r a t e . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325 5. 2. P h o t o r e c e p t o r s whi c h S a t u r a t e b u t d o n o t Ad a p t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328 6. Th e o r i e s o f Re t i n a l Ga i n Co n t r o l a n d t h e De t e r mi n a n t s o f Vi s ua l Se ns i t i vi t y . . . . . . . . . . . . . . 329 6. 1. Th e o r i e s o f t he Re t i n a l Ga i n Co n t r o l . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330 6. 2. Re t i n a l Ga i n a n d Vi s ua l Se ns i t i vi t y . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335 7. Re t r o s p e c t i v e a n d Co n c l u s i o n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337 Ap p e n d i x 1 - - Re t i n a l Ne u r o n s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337 Ap p e n d i x 2 - - Re c e p t i v e Fi e l ds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340 Ac k n o wl e d g e me n t s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342 Re f e r e n c e s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342 263 264 R. M. SHAPLEY AND C. ENROTH-CUGELL 1. I NTRODUCTI ON Vision is our pr i mar y sensory channel f or i nt eract i on with t he out si de worl d. It allows us t o recognize familiar faces and creat ures, and objects; it allows us t o ori ent ourselves in space and t o navigate f r om place to place. It is a pat hway f or e s t he t i c e n j o y me n t a nd f or i n f o r ma t i o n t ransmi ssi on. The visual system is one of t he many miracles of nat ure. Intensive st udy of t he visual process has revealed t hat t he ret i na must per f or m several oper at i ons on t he image delivered by t he eye' s optics in or der t o make manageabl e t he di f f i cul t j obs of t he brain : pat t ern recogni t i on and spatial localization. One of t he basic oper at i ons t he ret i na per f or ms is the subj ect of this chapt er: ret i nal adapt at i on. The retinal neur ons adapt t o vari at i ons in i l l umi nat i on by changi ng t hei r gain and response t i me course. The pur pose of adapt at i on is t o keep t he retinal response t o visual obj ect s appr oxi mat el y t he same when t he level of i l l umi nat i on changes. Thus, cent ral visual processing may pr oceed wi t hout the brai n having to at t end t o changes in t he average light level caused by t he daily solar cycle, by shading, by artificial i l l umi nat i on, or by ot her , perhaps unpredi ct abl e, events. We will demonst rat e the visual significance of the retinal regulation of cont rast sensitivity at di fferent levels of illumination. This basic funct i on of retinal adapt at i on is so i mpor t ant f or vision t hat t here is a hi erarchy of retinal adapt at i on mechani sms at several di fferent sites within t he retinal net wor k. Ther e is clear evidence for adapt at i on in individual phot or ecept or s, in some species. Ther e is evidence f or adapt at i on at t he level of t he out er pl exi f or m layer of t he retina, in bi pol ar cells. Ther e is also evidence for anot her stage of adapt at i on at the inner pl exi f or m layer, in amacri ne cells. Adapt at i on per f or med by the ret i nal net wor k t hus appears to involve at least t hree mechani sms in most retinas. Such evidence leads to t he concept of a hi erarchy of mechani sms which may be engaged at di fferent backgr ound levels and with di f f er ent t i me courses. In individual cells or in the retinal net wor k, t he neural signals sent on t o t he next stage in neural processing usually are fed back t o regul at e the response to new or persistent inputs. The t heori es for adapt at i on which we shall discuss requi re feedback in or der t o explain t he phenomena associated with visual adapt at i on. One can t herefore state as an overview t hat visual adapt at i on is achieved by a hi er ar chy of feedbacks designed t o regul at e cont rast sensitivity. It is impossible within t he space of a review chapt er t o deal with visual adapt at i on t o changes in i l l umi nat i on and also t o do j ust i ce to t he subj ect of recovery of sensitivity in t he dar k af t er all light has been t urned off. The latter phenomenon usually goes by t he name dark adaptation. Dark adapt at i on is in some ways similar t o light adapt at i on but is di f f er ent in such significant ot her ways t hat it deserves a chapt er all its own. It is not covered in t he fol l owi ng pages. 1.1. Terminology Because t he fact s and t heor i es of visual adapt at i on are compl i cat ed enough, one ought t o be clear about t he meani ngs of words which are used t o describe the facts, and so we will defi ne several words which are critical f or t he ensuing discussion. It is most i mpor t ant t o defi ne what adaptation means, but some prel i mi nary t erms requi re defi ni t i on first. 1.1.1. SENSITIVITY AND GAIN Unf or t una t e l y, " s e ns i t i vi t y" has di f f er ent meani ngs in di f f er ent fields. In psychophysi cs it means l / t hr eshol d or, in ot her words, the reciprocal of t he stimulus st rengt h requi red f or t he stimulus t o be perceived reliably. Accordi ng to this meaning, "sensi t i vi t y" is rel at ed t o t he si gnal / noi se rat i o inside the psychophysical observer (cf. f or example, Barlow and Levick, 1969; or Rose, 1948, 1973). The " noi s e" in this case is caused by all the physiological fl uct uat i ons in t he ret i na and brai n, fl uct uat i ons which make it difficult f or an observer t o be certain t hat a stimulus has been present ed. This " noi s e " is caused, in t he dar k, by t her mal br eakdown of phot opi gment in phot or ecept or s, spont aneous r a n d o m r el ease o f n e u r o t r a n s mi t t e r s , and fl uct uat i ons in t he physi ol ogi cal state of the ret i na and brai n. When t he r et i na is i l l umi nat ed, addi t i onal noise is caused by t he retinal response t o the r andoml y arriving st ream of light quant a. Psychophysi cal "sensi t i vi t y" can be i nfl uenced by VISUAL ADAPTATION AND RETINAL GAIN CONTROLS processes which change t he magni t ude of t he noise f r om which t he signal must be pi cked out as well as by t hose which af f ect t he size of t he signal. I n phys i ol ogi c a l e x p e r i me n t s , t he wor d "sensi t i vi t y" is usual l y used t o mean t he reci procal of t he stimulus requi red t o pr oduce a neural response of a cri t eri on size. In this meani ng of t he wor d, noise is i gnor ed because neural responses are usual l y averaged over a number of identical stimulus cycles t o el i mi nat e noise as much as possible. Ther e has been conf usi on in t he l i t erat ure when a result on psychophysi cal sensitivity has been t a k e n t o i mpl y s o me t h i n g de f i ni t e a b o u t physiological sensitivity or vice versa, because in one case noise has an ef f ect and in t he ot her it does not . We will t r y t o avoi d conf usi on by calling psychophysi cal sensi t i vi t y, "'sensitivity", and physiological sensitivity, "'gain". Gai n we define as t he r at i o of t he magni t ude of t he physi ol ogi cal response t o t he stimulus magni t ude, in t he small- signal range in which response is pr opor t i onal t o t he stimulus. Gai n t hus has units like mV/ qua nt um in phot or ecept or s, or i mpul s e/ quant um in ret i nal g a n g l i o n cel l s. Al t h o u g h we di s t i ngui s h physiological gain f r om psychophysi cal sensitivity, t he t wo are rel at ed. I f t he i nt ernal noise were mor e or less unaf f ect ed by backgr ound light, a reduct i on in ret i nal gai n as a consequence of an increase in backgr ound woul d pr oduce a cor r es pondi ng r educt i on in visual sensitivity. 1.1.2. ADAPTATION Consi der ed in t erms of sensitivity and gain, t he one unambi guous t r adi t i onal t er m which describes t he adapt at i on st at e is total dark adaptation, t he st at e of highest sensitivity reached by an observer and highest ret i nal gai n r eached by a living ret i na t hat has been left in t ot al darkness f or a few hours. We defi ne light adaptation as t hose vari at i ons in t he pr oper t i es of t he visual system f r om t he t ot al l y dar k adapt ed state which are pr oduced by vari at i on in t he level of light. For i nst ance, light may change the gain of the retina, or its time course of response, or its spectral sensitivity, or its spatial summat i on propert i es. However , an i ncrease of ret i nal noise due t o t he r a ndom times of arrivals of light quant a is not strictly speaki ng a pr oper t y of t he ret i na whi ch is changed by light. But it is a f act or whi ch coul d lead t o an increase of t he psychophysi cal 265 t hreshol d. Ther ef or e, in or der t o under st and how light adapt at i on cont r i but es t o vision, specifically t o t he vari at i on of visual sensitivity with mean level of i l l umi nat i on, one must distinguish bet ween adapt at i on and the effects of increased noise caused by light. Similarly, as we will show, gain can be reduced by sat ur at i on, e.g. t he l i mi t at i on on t he ampl i t ude of response i mposed by a response ceiling. We wish t o distinguish this ki nd of gain reduct i on due to sat urat i on, or as it has been called "r es pons e compr es s i on" , and t he gain cont r ol of adapt at i on which involves a change in the properties of t he ret i na with t i me duri ng i l l umi nat i on. 1.1.3. BRIGHTNESS, LUMINANCE, AND RETINAL ILLUMINATION Thr oughout our discussion, brightness means apparent brightness, the subjective sensation of how light or dar k an obj ect is. The obj ect i ve measure of t he amount of light emanat i ng f r om a l umi nous source or refl ect i ng obj ect , weighted by t he obser ver ' s spectral sensitivity f unct i on, is called luminance (Wyszecki and Stiles, 1967, p. 372). The illumination falling ont o a surface f r om a di st ant l umi nous source is pr opor t i onal t o t he l umi nance of t he source mul t i pl i ed by t he square of t he numeri cal aper t ur e of t he opt i cal system bet ween source and surface. The numeri cal aper t ur e of an opt i cal system is t he reci procal of its f - number . For t he eye, t he numeri cal aper t ur e is pr opor t i onal t o t he pupi l di amet er, and t he ret i nal i l l umi nat i on is t her ef or e pr opor t i onal t o t he area of t he pupil. Lumi nance can be expressed in t erms of effect i ve quant a of light per unit solid angle per unit t i me per uni t area of t he source. Il l umi nat i on can be expressed in t erms of effect i ve quant a of light per uni t t i me per uni t ar ea of t he surface on which t he light is falling. We will write L f or stimulus l umi nance and I f or retinal i l l umi nat i on, with t he under st andi ng t hat in most experi ment s t he pupil area is fixed and t her ef or e t he t wo quant i t i es are si mpl y pr opor t i onal . Evi dence will be present ed bel ow about t he i mpor t ance of t he l umi nous f l ux of light falling on a recept i ve field. Fl ux is illumination multiplied by area and can be expressed in units of effect i ve quant a per unit time. 266 R. M. SHAPLEY AND C. ENROTH- CUGELL 1. 1. 4. CONTRAST, CONTRAST SENSITIVITY, AND CONTRAST GAIN Cont rast is a physi cal pr oper t y of t he visual stimulus; it is the magni t ude of l umi nance vari at i on in t he st i mul us rel at i ve t o t he aver age l umi nance. We will show t hat t he per cept i on of cont r ast depends upon retinal adapt at i on. There is a pr obl em with defi ni ng cont r ast preci sel y because t here are t wo obvi ous defi ni t i ons which di ffer appr oxi mat el y by a f act or of t wo. In st udyi ng t he visibility of aperi odi c obj ect s like uni f or m disks or bar s or rect angl es on a backgr ound, t he nat ur al defi ni t i on of cont r ast is C = ( L o - L B ) / ( L B ) ( l a) where Lo is the l umi nance of the obj ect and LB the l umi nance of t he backgr ound, as i ndi cat ed in Fig. l(a). Lo - - LB is usually called M_~ and so equat i on (1) is usual l y wri t t en as: C = AL/ LB. ( l b) It is well known t hat f or t est st i mul i of l arge ar ea the psychophysi cal sensitivity fol l ows Weber ' s Law cont rast , which are each appr opr i at e for a part i cul ar ki nd of st i mul us, t he t wo di f f er ent defi ni t i ons are related because t hey refer to a single physical reality, namel y t he rel at i ve var i at i on of a modul at ed c ompone nt r ef er r ed t o a st eady st at e, or average, component . Cont rast sensitivity we defi ne as one di vi ded by the psychophysi cal t hreshol d cont rast , either Weber or Rayl ei gh cont r ast as t he case may be. Contrast gain is neural response divided by st i mul us cont r ast (Weber or Rayleigh) and will have units mV per unit cont r ast , or ( i mpul ses/ s) per uni t cont r ast . At low cont r ast , t he Rayl ei gh cont r ast of a grat i ng is appr oxi mat el y one hal f t he Weber cont r ast , as can be seen by compar i ng equat i on (1) with equat i on (3) as appl i ed to Fig. 1. The Rayleigh ( a ) APERIODIC STIMULUS I Background [L, I Lo A L T / L B = k (2a) CT = k (2b) where AL T is t he t hr eshol d l umi nance i ncr ement , and k is a const ant , t he t hr eshol d cont r ast . C is r ef er r ed t o in t he psychophysi cal l i t erat ure as the Weber f r act i on, but we pr ef er t o call it t he Weber contrast, Cw. Ther e is a second def i ni t i on of cont r ast whi ch is used f or peri odi c spat i al pat t erns like sine grat i ngs. Thi s is t he defi ni t i on used i mpl i ci t l y by Rayl ei gh (1889) and mor e explicitly by Mi chel son (1927) t o express t he visibility of i nt erference fringes: (b) PERIODIC STIMULUS LMEAN C R = ( t ma x - L mi n ) / ( L ma x +L mi n ) ( 3 ) = ( Zma x - Lmi n ) / ( 2 L . . . . )- As seen in Fi g. l ( b) , Lma x is t he ma x i mu m l umi nance and Lmi n the mi ni mum l umi nance in t he spat i al l y peri odi c pat t er n. We call CR the Rayleigh contrast. Bot h defi ni t i ons of cont r ast have been used in t he l i t erat ure of adapt at i on, so we make t hem explicit here. It is i mpor t ant t o realize t hat t hough t here are t hese t wo di f f er ent defi ni t i ons of FIG. 1. The t wo ki nds of cont r as t . (a). We be r cont r as t is i l l ust r at ed by a one - di me ns i ona l pl ot o f t he l umi na nc e profi l e o f a br i ght obj ect on a ba c kgr ound. The We be r c ont r a s t is def i ned as ( Lo- LB) / LB. (b). Rayl ei gh c ont r a s t is i l l ust r at ed wi t h t wo di f f er ent l umi na nc e pr of i l es o f gr at i ng pat t er ns : t he upper pr of i l e is o f a s qua r e wave gr at i ng, t he l ower is t he pr of i l e of a si ne gr at i ng. For each gr at i ng, t he Rayl ei gh c ont r a s t is def i ned as (Lmax-Lmin)/(Lmax+Lrnin). Thi s is equi val ent t o t he a mpl i t ude o f t he gr at i ng di vi ded by t he me a n level, i . e. ( Lmax-Lmean) / Lmean . VISUAL ADAPTATION AND RETINAL GAIN CONTROLS 267 cont r ast is t he mean- t o- peak ampl i t ude of t he grat i ng di vi ded by t he mean; t he Weber cont r ast is t he pe a k- t o- pe a k ampl i t ude di vi ded by t he l umi nance at t he t r ough of t he l umi nance profi l e, whi ch f or low cont rast s is appr oxi mat el y t he same as t he mean luminance. One may t herefore compar e measur ement s of Rayl ei gh and Weber cont r ast sensitivities when t he stimuli are at low cont r ast , as in most psychophysi cal experi ment s. One shoul d expect t he Rayl ei gh cont r ast sensitivity t o be about twice t hat of the Weber cont rast sensitivity, because of t he appr oxi mat e f act or of t wo di f f er ence in t he def i ni t i on of t he t wo kinds of cont r ast . Ther ef or e, when we write about (Weber) cont r ast sensitivity or (Rayleigh) cont r ast sensitivity, we distinguish t hem because of t he way t he measur ement s were made, but we concei ve of t hem as t wo di f f er ent measures of one underl yi ng pr oper t y of t he visual syst em, its sensitivity f or cont r ast . 1.2. The Purpose of Adaptation 1.2.1. To MAKE THE MAGNITUDE OF THE RETINAL RESPONSE DEPENDENT ON CONTRAST In science as in ot her activities it is a good strategy t o ask " Wh y ? " Thi s quest i on put s pr obl ems in perspect i ve, and t he at t empt s t o answer " Wh y ? " are nearl y always frui t ful . So one ought t o ask, " Wh y is t here light a da pt a t i on? " We believe t he answer is t hat adapt at i on keeps t he retinal response t o cont r ast i nvar i ant with changes of i l l umi nat i on, and t her eby achieves one maj or goal of vision: const ancy of t he visual per cept i on of refl ect i ng obj ect s. Ani mal s i ncl udi ng man have evol ved in a worl d of reflecting surfaces: water, eart h, leaves, flowers, and ot her ani mal s. What charact eri zes a refl ect i ng obj ect opt i cal l y is its r ef l ect ance, since t he r e f l e c t a nc e is d e t e r mi n e d by t he phys i cal charact eri st i cs of t he surface of t he obj ect . The reflectance is i nvari ant with respect t o illumination, within r easonabl e limits. The l umi nance of an obj ect , L o, is pr opor t i onal t o t he pr oduct of t he obj ect ' s refl ect ance, Ro, and i l l umi nat i on, I. L o = K' Ro ' I (4) We know f r om experi ence and experi ment (e.g. Land and McCann, 1971) t hat over a wide range of i l l umi nat i on t he brightness of a reflecting obj ect does not change and t hat t he brightnesses of an ar r ay of refl ect i ng surfaces are percei ved in t he order of their reflectances. This leads t o the (correct) i nference t hat the visual system must have a met hod f or est i mat i ng somet hi ng about t he refl ect ance of refl ect i ng obj ect s (cf. Land and McCann, 1971; Mar r , 1982). We believe this " s ome t hi ng" is t he cont rast (either Weber or Rayleigh contrast), which depends onl y on t he refl ect ance, as shown below. Fur t her mor e, one may i nfer t hat t he const ancy of per cept i on of an obj ect as its l umi nance varies i ndi cat es an under l yi ng const ancy of ret i nal response t o the cont rast of t he obj ect in spite of t he l umi nance vari at i on. For i nst ance, t he brightness of this pri nt does not appear t o var y as t he page is br ought closer t o or f ur t her away f r om a light source, even t hough t he l umi nance of the pri nt may var y by fact ors of ten. We believe t he r eason t he pri nt appears t he same brightness is t hat t he magni t ude of our retinal responses is the same when our eyes sweep across t he pr i nt at each of t he di f f er ent levels of i l l umi nat i on, because t he ret i nal response magni t ude depends on t he cont r ast of t he pri nt . We will now present t he ar gument t hat t he const ancy of ret i nal response is achi eved by means of t he mechani sms of ret i nal adapt at i on which pr oduce a dependence of t hat response on cont r ast ( Robson, 1975). Thi s concept arises nat ur al l y out of a cri t i que of a st at ement of t he pr obl em by Er nst Mach, t he great philosopher-physicist (Mach, 1865, t r ansl at ed in Rat l i ff, 1965). At t he end of his first paper on t he visual illusion now called " Ma c h Bands , " Mach pr oposed a new pri nci pl e f or psychol ogy: uni que psychol ogi cal events must correspond to uni que physical processes inside t he brai n. As an i l l ust rat i on of this new principle, he of f er ed t he fol l owi ng exampl e: "Let us examine another phenomenon with the help of our principle, which to my knowledge no one has yet discussed. White of a lesser intensity appears gray next to a brighter white. On the other hand, we are never in doubt whether we have before us a white or gray paper even under quite different conditions of illumination: in bright sunshine, overcast sky, in twilight, or by candle light, we have always almost the same sensation. What might be the cause of this? If the light intensity is 2-,3-, 2 6 8 R . M. SHAPLEY or n- f ol d bri ght er, so t hen is t he ret i nal i mage of t he white paper 2-, 3-, or n- f ol d br i ght er , but so al so is t he rest of t he vi sual field a nd t he ent i re r et i na recei ves t he 2-, 3-, or n- f ol d i l l umi nat i on. The r at i o o f t he qua nt i t y of light on t he ent i re r et i na a nd t he i mage of t he paper r emai ns c ons t a nt under ot her wi se equal condi t i ons , l t hi nk, t her ef or e, t hat a pr ocess is i ni t i at ed whos e i nt ensi t y depends on t hi s r at i o, and whi ch caus es t he s ens at i on of whi t e for t he ret i nal i mage. The br i ght nes s of t he ret i nal i mage is, so to speak, bei ng eval uat ed in t er ms of t he t ot al exci t at i on. Thi s is a j udge me nt , t he psychol ogi cal side of t he mat t er . The physi cal si de is t he pr oces s me nt i one d. It has not yet been di s c ove r e d. " Mach' s exampl e shows a deep insight i nt o t he per cept i on of bri ght ness and t he pur pose of visual adapt at i on. However , t here are some details of his br i ef anal ysi s which are not quite right. In appr eci at i ng what is still val i d and what needs cor r ect i on, one may begi n t o see t he poi nt of light adapt at i on. We will discuss t wo related but distinct comment s in t he quot ed par agr aph. The first is, " Whi t e of a lesser i nt ensi t y appear s gr ay next t o a bri ght er whi t e" . Thi s obser vat i on can be expl ai ned by mode r n psychophysi cs which has est abl i shed t hat the bri ght ness of an obj ect is det er mi ned, to a great ext ent , by t he Weber cont r ast bet ween t he obj ect and its sur r oundi ngs ( Hei nemann, 1955, 1972; Whi t t l e and Chal l ands, 1969). However , it is al so known t hat bri ght ness is mai nl y det er mi ned by the contrast near the border bet ween an obj ect and its sur r oundi ngs. Thi s surpri si ng concl usi on is forced by t wo experi ment s. First, the brightness di fference bet ween the t wo regi ons of unequal l umi nance vari es di rect l y with t he shar pness of the bor der bet ween t hem, bei ng maxi mal f or t he st eepest bor der ( Thomas and Kovar , 1965; Shapl ey and Tol hur st , 1973). Second, t wo regions of equal l umi nance appear of unequal bri ght ness when a local l umi nance di fference is i nt roduced as a bor der bet ween t he m ( Rat l i f f , 1965; Cr a i k, 1966; Cornsweet , 1970; Land and McCann, 1971; Shapley and Tol hur st , 1973). See Fig. 2 f or an i l l ust rat i on of this effect . Thus, the dependence of bri ght ness on cont r ast , which Mach referred t o in his 1865 paper , is now known t o be mai nl y a dependence on bor der cont r ast . The rel at i on bet ween bri ght ness, cont r ast , and light adapt at i on emerges f r om a critical exami nat i on of a second r emar k made by Mach, which concerns bri ght ness const ancy, " . . . we are never in doubt AND C. ENROTH- CUGELL whet her we have bef or e us a whi t e or gr ay paper even u n d e r qui t e d i f f e r e n t c o n d i t i o n s o f i l l u mi n a t i o n . . . " . Thi s obs e r va t i on can be expl ai ned by the fact t hat the cont r ast of an obj ect on a backgr ound is not changed by var i at i on in the level of i l l umi nat i on. Thus, if the visual syst em has the ability to derive brightness f r om cont rast , it will t hereby achieve brightness constancy. We will at this poi nt demons t r at e t hat cont r ast is i nvar i ant wi t h i l l umi nat i on, and t hen show how the cal cul at i on of cont r ast by the visual syst em can be expl ai ned in t er ms of retinal adapt at i on. The i nvari ance of cont r ast wi t h changes in t he level of i l l umi nat i on can be demons t r at ed by an exampl e. Consi der as the si mpl est case a uni f or ml y i l l umi nat ed scene wi t h an obj ect on a backgr ound. The si t uat i on is i l l ust rat ed in Fig. 3. The light comi ng f r om t he obj ect side of t he bor der is pr opor t i onal to I . Ro where I is the light falling on the scene f r om the source of illumination, and Ro is the reflectance of the object. The light reflected f r om the backgr ound side of the bor der is I . R B. The Weber cont rast is (IRo-IRB)/IR B. Dividing numer - at or and denomi nat or by I yields (Ro-RB)/RB as the cont r ast of an obj ect of refl ect ance Ro upon a backgr ound of refl ect ance Ra. Thus, t he cont r ast is independent of the level of illumination I and depends onl y on the reflectances of obj ect and back- gr ound. Now we must show t hat retinal adapt at i on provi des the mechani sm by which the visual syst em responds to cont rast . Consi der what happens when the recept i ve field of a retinal cell (see Appendi x 2) crosses the bor der bet ween an obj ect and a backgr ound, as in t he exampl e of Fig. 3. Suppose t hat the recept i ve field is " l o o k i n g " at the backgr ound j ust bef or e an eye movement occurs [Fig. 3(a)], and t hat the eye movement causes t he recept i ve field to cross t he bor der [Fig. 3(b)]. The change in the amount of light falling on the recept i ve field of t he neural uni t is the st i mul us which elicits a neural signal whi ch identifies t he bor der . The st i mul us is t hus IRo-IR w Now we must consi der the rol e of adapt at i on. A neur on crossing t he bor der onl y has been " l ooki ng" at the backgr ound side of t he bor der . So t he neur on is adapt ed to IR~. As a r easonabl e hypot hesi s about what adapt at i on does, to be j ust i fi ed by dat a l at er, we pr opos e it adj ust s t he gai n of the neur on t o be VI SUAL ADAPTATI ON AND RETI NAL GAI N CONTROLS 269 FIG. 2. Br i ght ness depends on bor der cont r as t . Thi s is a n i l l ust rat i on o f t he Cr ai k - O' Br i e n - Cor ns weet i l l usi on (cf. Rat l i f f , 1965; Cor ns we e t , 1970). The ent i r e r i ght ha l f o f t he fi el d is a ppa r e nt l y br i ght er t h a n t he l eft hal f , yet t he l umi na nc e s o f t he t wo ha l f fi el ds ar e equal a wa y f r o m t he bor de r bet ween t h e m, as can be seen by cover i ng t he bor der wi t h a n opa que st r i p. Near t he bor der , t he l umi na nc e is st eepl y decreasing t owa r ds t he bor der on t he da r k si de, a n d it is st eepl y increasing t owa r ds t he bor de r on t he br i ght si de, as can be seen in t he l u mi n a n c e pr of i l e dr a wn u n d e r n e a t h t he p h o t o g r a p h . VISUAL ADAPTATION AND RETINAL GAIN CONTROLS 271 (a) / I - R o = L o R e c e p tiv e ~ Field ~) , i ~ /I'RB=LB Bockground (b) Receptive F i e l d , ) B a c k g ro u n d FIG. 3. The invariance of contrast with changes in level of illumination, and how this is sensed by a visual cell's receptive field. This picture shows the luminance profile of an object (with reflectance R o) on a background (with reflectance RB) when both are illuminated uniformly, with illumination L The position of the receptive field of an individual retinal cell is indicated by the shaded rectangle. In (a), the visual cell is "looking" at the background. In Co), an eye movement has carried the receptive field of the cell onto the object. Equation (5) in the text demonstrates that this cell will respond to the Weber contrast of the object on the background, i.e. (Ro-RB)/RB, which is independent of illumination L As explained in the text, in order for this calculation to work the cell must be influenced by a gain control which adjusts the cell's gain to be reciprocal with background illumination. pr opor t i onal t o t he reci procal of t he val ue of t he i l l umi nat i on at which t he cell' s recept i ve field has been " l o o k i n g " up t o t he moment it crosses t he border. This inference is suggested by Weber' s Law, equat i on (2). Thus t he signal in our hypot het i cal ne ur on will equal t he st i mul us I' Ro-I' RB multiplied by t he gain, K/(I. RB), where K is a pro- por t i onal i t y const ant . The concl usi on is t hat what t he visual system uses t o charact eri ze an obj ect is a signal S =K" (I 'Ro-I " RB)/I " RB (5) which is pr opor t i onal t o the Weber cont rast . In this case t he neural response t o t he obj ect is i nvari ant with respect t o changes in t he i l l umi nat i on, I. Thi s chai n of reasoni ng onl y hol ds i f t he gain of t he ret i na is reci procal with backgr ound i l l umi nat i on, in anal ogy with Weber ' s Law. When ret i nal gain is not reci procal with backgr ound, one shoul d expect devi at i ons f r om brightness const ancy. This line of reasoni ng leads t o t he concl usi on t hat t he det ect i on of cont r as t depends on light adapt at i on. A di f f er ent view is t hat t he dependence of ret i nal responses on cont r ast is a consequence of t he spatially ant agoni st i c c e n t e r - s u r r o u n d organi zat i on of receptive fields (discussed bel ow in Appendi x 2), but we believe this view t o be i ncorrect . I f t her e are eye movement s, spatial cont r ast will be " s e e n " by t he highly localized recept i ve fields of ret i nal neur ons as successive cont rast , or in ot her words as t empor al modul at i on of t he amount of light falling on t hose recept i ve fields ( Hel mhol t z, 1909). As shown above, in this si t uat i on, t he mechani sms of adapt at i on t oget her with t he compact nat ur e of t he recept i ve fields pr oduce a response dependent on cont r ast wi t hout any r equi r ement f or a spat i al l y ant agoni st i c sur r ound. The pur pose of havi ng recept i ve field surrounds pr obabl y is t o make retinal neurons mor e sensitive t o pat t erns with nar r ow spatial gradi ent s, or high spatial frequenci es, t han t o coarse pat t erns or di ffuse light. Put anot her way, with light adapt at i on but wi t hout recept i ve field sur r ounds, r et i nal neur ons woul d r es pond t o cont r ast . However, with receptive field surrounds but wi t hout adapt at i on, t he ret i na woul d be unabl e t o r espond t o cont rast ; it woul d sat urat e qui ckl y (see Sect i on 1.2.2.). This poi nt of view is similar t o t hat put f or war d by Whi t t l e and Chal l ands (1969). Consi derat i on of Mach' s st at ement t oget her with mor e recent wor k on brightness const ancy under condi t i ons of non- uni f or m i l l umi nat i on leads t o an i nt erest i ng concl usi on about t he spatial spread of adapt at i on. Mach, and ot hers af t er hi m (Heri ng, 1920; Hel son, 1964), have supposed t hat t he eye compared, "t he quant i t y of light on the entire retina and t he i mage of t he pa pe r " , in or der t o cal cul at e the brightness of t he paper. This is not correct . The first correct i on is t hat brightness depends on a local phenomenon, t he cont r ast near t he bor der of an obj ect , not on t he quant i t y of light in t he i mage of t he obj ect (see Fig. 2). Second, t he ret i na does not compar e t he quant i t y of light " o n the entire r et i na" with t he amount of light in t he obj ect , as Mach suggested. As Davi dson and Fr eeman (1965) and Land and McCann (1971) have shown, brightness is const ant under condi t i ons of spatially non- uni f or m i l l umi nat i on. We have argued t hat ret i nal 272 adapt at i on is t he basis of bri ght ness const ancy. I f ret i nal adapt at i on aver aged light " o n t he ent i re r et i na" as Mach supposed, it coul d not pr oduce bri ght ness const ancy when t he i l l umi nat i on was spat i al l y non- uni f or m. Ret i nal gai n woul d be the same at all poi nt s on t he ret i na. Neur al signals at one poi nt on t he ret i na woul d be at t enuat ed t o t he same ext ent as neural signals f r om anot her regi on of unequal l ocal i l l umi nat i on. The cor r ect cal cul at i on of the bor der cont r ast by t he ret i na would br eak down. This implies t hat the retinal gain cont r ol mechani sm must be somewhat localized, as i ndeed phys i ol ogi c a l evi dence d e mo n s t r a t e s (Cl el and and Enr ot h- Cugel l , 1968 among ot hers). Mach and ma ny ot hers later, including Land and McCann (1971), pr opos ed t hat t he vi sual syst em cal cul at ed t he r at i o I Ro/ I RB = Ro/ RB directly, but this is i ncorrect , pr obabl y f or the fol l owi ng reason. The rat i o Ro / Rs is nearl y al ways close t o 1 in nat ur e because t he ( achr omat i c) refl ect ances of most nat ur al obj ect s are so nearl y the same. I f Ro/ RB were the quant i t y bei ng measur ed, correct i dent i fi cat i on of whet her an obj ect is bri ght er or dar ker t han t he backgr ound woul d become a pr obl em of accurat e measur ement and compar i s on of t he rat i o wi t h 1. The Weber cont r ast as defi ned earlier is the difference between the reflectance ratio and 1. That is, (Ro-RB)/RB = ( Ro/ Ra) - l . The Weber cont r ast , ( Ro- RB) / R B, changes sign f or obj ect s bri ght er t han the backgr ound (posi t i ve cont r ast ) compar ed t o obj ect s dar ker t han t he backgr ound (negative cont rast ). The change of sign l eads t o a mu c h eas i er , l ess e r r o r - p r o n e di scr i mi nat i on of dar k f r om bri ght obj ect s t han would comput at i on of the refl ect ance rat i o Ro/ RB. Fur t her mor e, t he vi sual syst ems of ver t ebr at es put the basic meas ur ement of t he sign of t he cont r ast i nt o t he f unct i onal archi t ect ure of t he ret i na (cf. Her i ng, 1920). Cells whi ch are excited by posi t i ve cont r ast (the " o n " cells of Har t l i ne, 1938; and t he " on- c e nt e r " cells of Kuf f l er , 1953) are segregat ed f r om the cells which are excited by negative cont rast ( " o f f " cells of Har t l i ne and " o f f - c e n t e r " cells of Kuf f l er ) . As di scussed in Appe ndi x 1, t he el abor at i on of these " o n " and " o f f " pat hways involves separ at i on of ret i nal synapt i c connect i ons in t he inner pl exi f or m l ayer of t he ret i na, as suggested by Fami gl i et t i and Kol b (1976), and t hen pr oven by Fami gl i et t i et al. (1977) and Nel son et R. M. SHAPLEY AND C. ENROTH-CUGELL al. (1978). These physi ol ogi cal and anat omi cal obs er vat i ons r ei nf or ce t he pur el y f unct i onal hypot hesi s t hat the ret i na is designed t o measure the cont r ast of obj ect s in or der to pr ovi de t o t he br ai n an i l l umi nat i on- i nvar i ant descri pt i on of the worl d of obj ect s. Spat i al l y localized light adapt at i on is a cruci al f act or in this retinal f unct i on. 1.2.2. TO HANDLE THE LARGE RANGE OF ILLUMINATION LEVELS A rel at ed r eason f or the necessity of light adapt at i on is t he very extensive r ange of aver age light levels pr esent ed t o t he eye by nat ure. A white paper (refl ect ance = 1) in moonl i ght has a l umi nance of about 3 x 10 -2 cd m-L A white paper in sunlight is six orders of magni t ude bri ght er, about 3 x 104 cd m- L Backgr ounds whi ch af f ect vi si on ext end t hree log units bel ow refl ect ed moonl i ght and one log unit above reflected sunlight, a t ot al r ange of about 10'% Par t of this enor mous r ange is t aken care of by paral l el processi ng in separ at e r od and cone pat hways. In humans f or exampl e, the highly sensitive rod syst em handles the t hree l owest decades of backgr ounds . The less sensitive cone system handles the upper six log units, and the decade of order 0.1 cd m -2 is shared (Riggs, 1965). However , in ot her ani mal s the overl ap of the ranges of backgr ound handl ed by r od and cone systems may di ffer because of a di fferent rod - cone weighting. For i nst ance, in t he cat in which the r od - cone rat i o is about one hundr ed times great er t han in man (St ei nberg et al., 1973), t he r od syst em handl es five log units r at her t han t hree and the cones onl y domi nat e visual r esponses above 10 cd m -2 in backgr ound. A t abl e which expresses the l umi nances in st andar d units and in t er ms of quant a of light per second per degree squar ed in area, is of f er ed as Tabl e 1. A human is so sensitive when dar k adapt ed t hat he can detect (wi t hout any false posi t i ve responses) t hat a light fl ash has been pr esent ed when onl y about one hundr ed quant a of light are i nci dent on his cor nea (Nagel, 1909; Hecht et al., 1942; cf. Cor nsweet , 1970). When one t akes i nt o account l osses in t he eye and t he i nef f i ci ency of t r ansduct i on, this implies t hat about t went y retinal responses t o quant a of light are requi red f or such ul t ra-rel i abl e visual per f or mance (this is a hi gher numbe r t han est i mat ed by Hecht et al., 1942, and VISUAL ADAPTATION AND RETINAL GAIN CONTROLS TABLE I. Quantal Equivalents of Photometric Units and some Other Useful Equivalences P h o t o me t r i c Un i t E q u i v a l e n t i n Qu a n t a 1 c d m -2 t h r o u g h 1 mm ~ p u p i l , s c o t o p i c (1 s c o t o p i c t d, h u ma n ) 1 c d m -2 t h r o u g h 1 mm ~ p u p i l , p h o t o p i c (1 p h o t o p i c t d , h u ma n ) 1 l u me n ( s c o t o p i c ) 1 l u me n ( p h o t o p i c ) Ot h e r Us e f u l Eq u i v a l e n c e s 4. 46. l 0 s q u a n t a ( 5 0 7 n m) ( de g 2 s) -t 1. 26. 106 q u a n t a ( 5 6 0 n m) ( deg 2 s) -1 1. 4. 1015 q u a n t a ( 5 0 7 n m) s -1 4 . 2 . 1 0 is q u a n t a ( 5 6 0 n m) s -1 Un i t Us e f u l E q u i v a l e n t 1 q u a n t u m( 5 0 7 n m) s - ' 4 . 1 0 -' 9 wa t t s 1 q u a n t u m( 5 6 0 n m) s -I 3. 5" 10 -' 9 wa t t s 1 d e g 2 o n h u ma n r e t i n a 8. 5" 10 -4 c m 2 1 d e g 2 c a t r e t i n a 4 . 8 - 1 0 -4 c m 2 273 is based on mor e recent estimates of quant um efficiency by Barlow, 1977). However, Sakitt (1972) has shown t hat i f one relaxes t he st ri ngent requi rement of no false positives, human observers can do bet t er t han chance when on t he average onl y a single quant um of light excites t he ret i na. Thi s must mean t hat a single quant um response is compar abl e in magni t ude t o t he i nt ra-ret i nal dar k noise. Compar abl e sensitivity is possessed by ot her ani mal s. The high sensitivity of a dar k adapt ed eye poses a pr obl em when t he obser ver moves i nt o bri ght er surroundi ngs. Al most all neur ons have a limited response range, f r om small signals t o the peak levels set by bi ol ogi cal cons t r ai nt s such as i oni c equilibrium potentials. The range of responses is no great er t han a f act or of one hundr ed f r om noise t o ceiling. The pr obl em is obvi ous. How can t he r od pat hway in t he ret i na encode t hr ee t o five log units of stimulus level when it onl y has a f act or of one hundr ed in response t o wor k with? The answer is it cannot , and it does not . The ret i na adapt s (reduces its gain) in t he presence of large average inputs in or der t o represent onl y modul at i ons ar ound t he average level when the average becomes t oo large t o handle with t he high dar k adapt ed gain. This pr obl em of sat ur at i on of neural responses is rel at ed t o t he need f or stable cont r ast sensitivity, which we have argued is t he mai n pur pose of adapt at i on. I f t he ret i na did not adapt , t he cont rast gain and cont rast sensitivity woul d pl ummet at high light levels. Thus we woul d become blind t o reflecting objects in bright daylight. This woul d not be a stable survival st rat egy, and t her ef or e t here is a biological need f or adapt at i on. In fact , as our review of t he psychophysi cal and physi ol ogi cal results will show, t he cont r ast sensitivity and cont r ast gai n of humans and ani mal s general l y increase as t he i l l umi nat i on increases, fi nal l y levelling of f t o asympt ot i c values in bri ght light. At this poi nt we will demonst r at e how t he sat ur at i on of neural response woul d lead t o a decline in cont r ast gain. Then we will present a suggestion f or an adapt at i on mechani sm whi ch of f er s an escape f r om t he " s at ur at i on cat as t r ophe" . Fi rst let us consi der how t he Na k a - Ru s h t o n equat i on, whi ch appr oxi mat el y describes t he i nt e ns i t y- r e s pons e f unct i on of distal ret i nal n e u r o n s , woul d l ead t o a " s a t u r a t i o n - cat as t r ophe" if t he ret i na did not adapt . This is t he Na k a - Rusht on equat i on: R = ( I / ( I + Is)) Rmu (6) where R is t he response of t he neur on measur ed as t he change in membr ane pot ent i al f r om its t ot al l y dar k adapt ed level, I is t he i l l umi nat i on of t he stimulus. Is is t he semi -sat urat i on const ant also equal t o the illumination at which R reaches its half- maxi mal value. This equat i on ( Naka and Rusht on, 1966) is called t he Mi c ha e l i s - Ment en equat i on by Bayl or and colleagues (Bayl or and Hodgki n, 1973; 274 Ba y l o r e t al . , 1974; Ba y l o r e t al . , 1979) a f t e r a s i mi l a r e q u a t i o n whi c h ar i s es i n t he t h e o r y o f e n z y me ki net i cs . Th e e q u a t i o n de s c r i be s a s ys t e m whi ch s a t ur a t e s . Fi gur e 4 s hows a g r a p h o f R/ Rma x vs / p l o t t e d on l i n e a r - l i near c oor di na t e s , a n d it can be seen t h a t R is a s a t u r a t i n g f u n c t i o n o f I : a b o v e t he va l ue I s , t he l i ght c a n i nc r e a s e by s ever al o r d e r s o f ma g n i t u d e b u t t he r e s p o n s e , R, c a n i nc r e a s e at mo s t b y a f a c t o r o f t wo. R. M. SHAPLEY AND C. ENROTH-CUGELL t he a mo u n t o f r e s p o n s e pe r a mo u n t o f c o n t r a s t , t hus : Gco n = dR/(dI /I a) = dR/ dl og/ = ~/ ( ~r / &) ( 8 ) a n d t o c a l c ul a t e t he c o n t r a s t ga i n f or a n e u r o n whi c h obe ys t he Na k a - R u s h t o n r e l a t i o n we s ubs t i t ut e e q u a t i o n (7) i nt o e q u a t i o n (8) t o o b t a i n : g E or- 1 . 0 - 0.8 - ~ 0.6 0.4 0.2 0 I I 0 200 400 I I 600 800 ILLUMINATION FIG. 4. The Naka-Rusht on relation. This is a plot of ( R/ Rmax) =I / ( I +I s) , the Na k a - Rushton relation, on l i near-l i near coordinates. As can be seen, the relation produces a compressive nonlinear curve; the response saturates when I > I s. In this example, the semi-saturation illumination, I s, was chosen to be 100 in arbitrary units. As a r g u e d by Wi l l i a ms a n d Ga l e (1977), Na k a e t al. (1979), No r ma n n a n d P e r l ma n (1979c), a n d Va l e t o n a n d v a n No r r e n ( 1 9 8 3 ) , t h e Na k a - R u s h t o n e q u a t i o n c a n be us ed t o de s c r i be ga i n a n d c o n t r a s t ga i n i n e x p e r i me n t s i n whi c h t he s t i mul us is a n i n c r e me n t o n a b a c k g r o u n d , b y e xa mi ni ng how an i nc r e me nt a l r e s pons e h R de pe nds on a n i n c r e me n t a l s t i mul us h / s u p e r i mp o s e d on a b a c k g r o u n d I B. I n t h i s c a s e t h e I i n t h e Na k a - R u s h t o n e q u a t i o n is IB + h i , t he r e s pons e is R B + AR, a n d t he ga i n is d R / d I ~- A R / A I , when h i is s mal l . The ga i n c a n be c a l c u l a t e d by d i f f e r e n t i a t i n g e q u a t i o n (6) ( Wi l l i a ms a n d Ga l e , 1977; Na k a e t al . , 1979): d R / d I = [ l / ( / a + / s ) ] 2 / s R m a x ~ { 1/ (&2)}. (/sRmax) f o r / a > > / s (7) Th e We b e r c o n t r a s t ga i n Gcon c a n be d e f i n e d as Gcon = / s R ma x " I a / ( / B + / s ) 2 "-~ ( / s R ma x ) " 1 / ( [ a ) a s / B > > I s . (9) Wh a t t hi s me a n s is t h a t on a c c o u n t o f s a t u r a t i o n t he We b e r c o n t r a s t ga i n wo u l d de c l i ne as t he a mo u n t o f s t e a d y b a c k g r o u n d l i ght i nc r e a s e d, i n i nve r s e p r o p o r t i o n t o t he a v e r a g e l i ght l evel . Th u s t he " s a t u r a t i o n - c a t a s t r o p h e " is i mp l i c i t i n e q u a t i o n (9). Fi g u r e 5 i l l us t r a t e s t he ga i n a n d t he 1.0 ~ / - - ~ , N \ x - - Goin --- Cont r osf Get n D 0.8 x x \ a . 0 . 6 "-.~ ~ 0 . 4 ~ ~ 0 . 2 o " 7 - - - ~ 1 1 -.. o 200 4 0 0 6 0 0 8 0 0 ILLUMINATION FIG. 5. The gain and (Weber) contrast gain of the Na ka - Rushton relation. Gain is drawn as the continuous curve and contrast gain as the dashed curve. The saturating nature of the Naka - Rushton relation causes the gain to be a monotonically decreasing function of illumination [cf. equation (7)], while the contrast gain has a peak at I = I s (which was chosen to be 100 as in Fig. 4). c ont r a s t gai n ( equat i ons 7 a nd 9) f or a n e u r o n whi ch obe ys t he Na k a - R u s h t o n r e l a t i o n . I f t he r e t i n a wo u l d not a d a p t , t he ( We b e r ) c o n t r a s t ga i n wo u l d d r o p as t he l i ght l evel i nc r e a s e d. Th e r e t i n a a d a p t s t o a v o i d s a t u r a t i o n by h a v i n g an " a u t o ma t i c ga i n c o n t r o l " ( Ros e, 1948); t he gai n o f t he r et i na is r e duc e d af t er t he r e c e pt i on o f e nough l i ght , so t ha t t he n e u r a l r e s p o n s e u s u a l l y d o e s n ' t s a t u r a t e i n t he p h y s i o l o g i c a l r a n g e o f i l l u mi n a t i o n . We wi l l ha ve a l ot t o s a y a b o u t t he me c h a n i s m o f t hi s a u t o ma t i c ga i n c o n t r o l , wh e n we di s c us s t he VISUAL ADAPTATION AND RETINAL GAIN CONTROLS experi ment al psychophysi cal and physiological ~00 dat a. It is wort h observing how an aut omat i c gai n "~ t~ cont rol solves the probl em of the " s at ur at i on x 300. cat ast r ophe" formal l y. One requires t hat R is not t;x simply a funct i on of the i l l umi nat i on I but depends 200 on I times a gai n fact or, g, which depends on the value of I at present and in the recent past. Thus " g i s a " f u n c t i o n a l " o f I ( t ) , g = g { t , I ( t ) } , b u t w e w i l l ~ loo write it simply as g{IB} t o si mpl i fy the algebra. This si mpl i fi cat i on is reasonabl e since we are onl y 0 anal yzi ng the case of an i ncrement on a st eady backgr ound, in which case g woul d be fixed at a st eady state level set by the backgr ound. The Na k a - Ru s h t o n equat i on modi fi ed to include adapt at i on becomes: R/Rmm, = g{ I B} ' I / ( g{ I B} ' I + Is) (lOa) which can be expressed anot her way by dividing numer at or and denomi nat or by g{Is}, as follows: (10b) R/Rm~,= I /[I + (I s/g{I B})] = I / ( I + I s ' ) in which I s' = Is/g{Ia}. The effect of adapt at i on can therefore be t hought of, t o a first appr oxi mat i on, as changi ng the value o f t he s e mi - s a t u r a t i o n c o n s t a n t i n t he Na k a - Ru s h t o n equat i on, Is (Dawis and Purpl e, 1982). This change in t he semi -sat urat i on const ant has been observed in retinal neurons; of t en it is referred t o as "cur ve- shi f t i ng" because when e qua t i on (10b) is gr a phe d on l i n e a r - l o g coordi nat es, the curve translates t o the right as I s' increases ( Nor mann and Werbl i n, 1974; Werbl i n, 1977; Nor mann and Perl man, 1979c among others). Fi gure 6 illustrates the phenomenon of "cur ve- shi ft i ng" in cat retinal ganglion cells, from the work of Sakmann and Creut zfel dt (1969). Sat urat i on is post poned by this strategy. Wi t h the appropri at e choice of t he g funct i onal , one can obt ai n Weber' s Law wi t h such a modi fi ed Naka - Rusht on model (Dawis and Purpl e, 1982; Valeton and van Norren, 1983). It can also be shown t hat such a model can avoid the probl em of cont rast gain falling at high light levels, as follows. In anal ogy with equat i on (9) we can derive the cont rast gain of the modi fi ed 275 I I 9.10 -s I l I I 9.10 ~ 9 10 -3 9 10 "z 9.10 -I 9 t est spot t umi nonce ( c d / r n 2) Fr o. 6. " Cu r v e - s h i f t i n g " i n cat r et i nal ga ngl i on cell r es pons es . Pe a k r e s pons e o f a cat r et i nal ga ngl i on cell (in spi kes s -1) t o a t est s pot f l as hed on a b a c k g r o u n d , as a f unc t i on o f t est s pot l umi na nc e , at six di f f er ent b a c k g r o u n d l umi nances . The ba c kgr ound l umi na nc e s ar e i ndi cat ed in t he f i gur e above a n d t o t he l eft o f each cur ve, i n uni t s o f l og cd m -2 (di l at ed pupi l ) . The t est s pot was 0. 5 dog i n di amet er a nd was directed t o t he receptive field cent er. The backgr ound was 30 dog i n di amet er , a n d was o n c ont i nuous l y. The s t i mul us s pot was on f or 500 ms , a n d was pr es ent ed repet i t i vel y at 0. 2 Hz. The r es pons e me a s ur e was t he n u mb e r o f i mpul s e s i n t he 50 ms af t er s t i mul us ons e t mi n u s t he n u mb e r o f i mpul s es i n a c ompa r a bl e cont r ol per i od. Twe nt y r es pons es were aver aged f or each poi nt . Fr o m S a k ma n n a n d Cr eut zf el dt (1969). Na k a - Rusht on model: Gco.' = I s " Rm,~,'IB/(IB + I s ' ) 2 and since adapt at i on keeps In<< I s ' , the cont rast gai n is approxi mat el y: Gcon' ~Rmax Is/Is'= R m a , , ' g { I a } I a / I s ( l l ) whi ch can be kept al ways i ncr easi ng and approxi mat el y const ant at high backgrounds by the correct choice of the dependence of the g functional, and t herefore I s ' , on t he backgr ound. As will be seen, the ret i na usually makes this correct choice. Summi ng up, we may say t hat adapt at i on is necessary t o prevent sat urat i on which woul d otherwise depress the cont rast sensitivity (cf. Adel son, 1982). In discussing psychophysical and electrophysiological results on sensitivity and gain, we will show how the processes of light adapt at i on count eract the t endency of visual neurons to sat urat e. It may be instructive t o consider Fechner' s integration of Weber' s Law (equation 2) as anot her proposed answer t o the probl em of providing 276 R. M. SHAPLEY AND C. ENROTH- CUGELL const ant cont rast sensitivity (Boring, 1950). Fechner pr opos ed t hat a cert ai n fixed a mount of change in i nt ernal sensat i on was requi red f or t hreshol d. We can call this val ue ArT. He t hen pr opos ed t hat Ar T was pr opor t i onal t o M/IB which is also fixed at t hr eshol d in t he Weber ' s Law range. The crucial as s umpt i on was t hat r, the i nt ernal a mount of sensat i on, was si mpl y a f unct i on of t he a mount of light I; t hus r = r(/). Wi t h this assumpt i on, Fechner coul d t hen t reat Weber ' s Law as an appr oxi mat i on to a di fferent i al equat i on f or r. Thus, A/ / I coul d be appr oxi mat ed by a di fferent i al dI/I, while the i ncr ement in sensat i on Ar coul d be appr oxi mat ed by the di fferent i al dr. Thi s led to Fechner ' s di fferent i al equat i on: dr = a. dI / I (12a) obj ect i ons, one may say t hat Fechner arri ved at t oo shal l ow an i n t e n s i t y - r e s p o n s e curve because his t heor y was t oo si mpl e (see Sperl i ng and Sondhi , 1968 f or a si mi l ar poi nt of view). Havi ng discussed the funct i onal role of the retinal gai n cont r ol in st avi ng o f f s at ur at i on and mai nt ai ni ng hi gh cont r ast sensitivity, we will now consi der psychophysi cal results on t he dependence of visual sensitivity on the backgr ound or mean level of i l l umi nat i on. A critical issue is, t o what ext ent is sensitivity det ermi ned by noise and to what extent by retinal adapt at i on? Then we will consi der the pr oper t i es and mechani sms of retinal adapt at i on. 2. P S YCHOP HYS I CAL LAWS OF L I GHT ADAP T AT I ON where a is a pr opor t i onal i t y const ant . I nt egr at i ng this equat i on, he obt ai ned, r = a. l og(/ ) + rl (12b) where r, is the response at I = 1. Fechner ' s Law (12b) is not ri ght , as can be seen bot h t heoret i cal l y and empi ri cal l y. First, the t heoret i cal obj ect i on is t hat r may not be vi ewed, wi t hout adequat e j ust i f i cat i on, as a si mpl e f unct i on of the val ue of /, t he light, i ndependent of t he past hi st ory of i l l umi nat i on. Because it is now well known t hat t he ret i na and br ai n have a finite t i me course of response, and r esponse ampl i t ude and t i me course ar e mo d i f i e d by t he p r e v i o u s h i s t o r y o f i l l umi nat i on, we now must consi der r to be a t r ans f or mat i on of /, or, t o be t echni cal , a functional. Ther ef or e, t he mat hemat i cal f or ms of equat i ons (12a, b) are not j ust i fi ed. Fur t her mor e, t her e are empi r i cal consequences of Fechner ' s t heor y. Fechner ' s r easoni ng leads us t o believe t hat We be r ' s Law implies t hat r must be pr opor t i onal t o log L Thi s is a ver y shal l ow sat ur at i ng f unct i on of L The much st eeper Na k a - Rus ht on equat i on is act ual l y obs er ved. I n essence, Fe c hne r ' s expl anat i on of Weber ' s Law is in t er ms of response c o mp r e s s i o n on a v e r y ge nt l y s a t u r a t i n g i n t e n s i t y - r e s p o n s e f unct i on, r at her t han t he cor r ect expl anat i on in t er ms of aut omat i c cont r ol of gai n on a much st eeper i n t e n s i t y - r e s p o n s e cur ve. Combi ni ng t heor et i cal and empi r i cal As pr opos ed in t he Introduction, we believe light adapt at i on t o be an essential retinal mechani sm f or a l l owi ng e f f o r t l e s s , i l l u mi n a t i o n - i n v a r i a n t eval uat i on of the optical characteristics of reflecting objects. One har dl y ever is awar e of adapt at i on when it is per f or mi ng this f unct i on. However , the effect s of adapt at i on become not i ceabl e when one observes self-luminous sources of light: st ars, l amps, candles, and televisions. These visual objects change in bri ght ness when t he ambi ent level of i l l umi nat i on changes. When the light level is high enough, t hey di sappear f r om view compl et el y; t hey fall bel ow t he per cept ual t hr eshol d f or i ncrement s. Thus t he act i on of light adapt at i on is exposed by t he dai l y cycle of t he fadi ng and r eappear ance of t he stars, t hough adapt at i on is wor ki ng unobt r u- sively all t he t i me t o keep t he per cept i on of refl ect i ng obj ect s unaf f ect ed by t hat same cycle. The i nfl uence of ambi ent i l l umi nat i on on t he sensitivity f or l umi nous sources like st ars and candl es has been not i ced since ant i qui t y. A poet i c descr i pt i on and psychol ogi cal expl anat i on was of f er ed by Shakespeare in The Merchant of Venice. As Por t i a and Neri ssa ret urn t o Por t i a' s villa Bel mont at night af t er t he t r i umph over Shyl ock in Venice, t hey speak about sensat i on and percept i on: P ORT I A Th a t l i ght we see is bur ni ng in my hal l . How f ar t hat little candl e t hr ows hi s be a ms , So shi nes a good deed in a n a u g h t y wor l d. VI SUAL ADAPTATI ON AND RETI NAL GAI N CONTROLS 2 7 7 NERISSA When the moon shone we did not see the candle. PORTIA So doth the greater glory dim the less. A substitute shines brightly as a king Until a king be by, and then his state Empties itself, as doth an inland brook Into the main of waters . . . . (Act V, Scene 1, II. 89-97). Th e ba s i c p h e n o me n o n is a c c u r a t e l y d e s c r i b e d he r e , t he f a d i n g t o d i s a p p e a r a n c e o f a we a k i n c r e me n t o n a b a c k g r o u n d . Si nce t he b a c k g r o u n d was mo o n l i g h t i n S h a k e s p e a r e ' s " e x p e r i me n t " , Ne r i s s a mu s t ha ve b e e n o b s e r v i n g l i ght a d a p t a t i o n o f t he r o d s ys t e m. Thi s we n o w c o n s i d e r mo r e q u a n t i t a t i v e l y . 2.1. The Rod System 2.1.1. SENSITIVITY AS A FUNCTION OF BACKGROUND 2.1.1.1. Aguilar and Stiles' experiment. A ba s i c f act o f vi s ua l e xpe r i e nc e is t h a t t he s e ns i t i vi t y f or i nc r e me nt s o f l i ght is r e duc e d b y s t e a dy b a c k g r o u n d i l l u mi n a t i o n . We wi l l r evi ew t he p s y c h o p h y s i c a l i n v e s t i g a t i o n s o f t hi s p h e n o me n o n i n o r d e r t o a n s we r t he q u e s t i o n , " t o wh a t e xt e nt is t he l os s o f vi s ual s ens i t i vi t y d u r i n g i l l u mi n a t i o n c a us e d by t r ue l i ght a d a p t a t i o n o f t he vi s ual s ys t e m, a n d t o wh a t e xt e nt is i t c a u s e d b y i n c r e a s e d " n o i s e " i n t he s t i mu l u s i t s e l f ? " The c a n o n i c a l d a t a c o n c e r n i n g t he d e p e n d e n c e o f t he p s y c h o p h y s i c a l t h r e s h o l d o n t he l evel o f b a c k g r o u n d i l l u mi n a t i o n a r e t h o s e o f Ag u i l a r a n d St i l es (1954; Fi g. 7). Th e s t i mul i wer e c h o s e n t o s t i mu l a t e t he r o d s ys t e m mo s t e f f e c t i ve l y. Th e s t i mul us was l a r ge (9 d i a me t e r di s k) , p r e s e n t e d i n t he p e r i p h e r y o f t he vi s ua l f i el d, a n d wa s o f a b l u e - g r e e n c o l o r o n a r e d b a c k g r o u n d . Th e r e d b a c k g r o u n d de s e ns i t i z e d t he c one s ys t e m r e l a t i ve l y mo r e t h a n wo u l d a whi t e b a c k g r o u n d , a n d t he b l u e - g r e e n t es t d i s k s t i mu l a t e d t he r o d s mo r e e f f e c t i ve l y c o mp a r e d t o c one s t h a n wo u l d a whi t e s t i mul us . Th e s t i mu l u s was p r e s e n t e d f or 0. 2 s whi c h a l l o we d t he r o d s t o s u mma t e t he i r s i gnal s t e mp o r a l l y . I n t hi s way, Ag u i l a r a n d St i l es wer e abl e _ _ g 0 -- 4 g 4-' t~ ~ .g g 0 lxl ,I- ra 0 / ,' 2 ~ 4-" E - I - I1) b B ~ - 2 - - - - I Scare 06 - 3 1 - - - - 0. 4 ~ ~ - - 0 . 2 - 4 - 4 - 2 0 2 4 L o g ( f i e L d r a d i a n c e , s c o t o p i c r o L a n d s ) FIG. 7. Human " r o d " threshold as a function of background illumination. The stimulus was a 9 deg disk on a 20 deg background presented 9 deg from the fovea. The disk was blue - green and the background was red in color, in order to enhance rod responses compared to cone responses. The stimulus was presented for 200 ms. The background was continuous. Curve A is the threshold illumination vs background. Curve B is what we would call the Weber contrast at threshold, and is labeled the "Fechner fraction" in this graph. From Wyszecki and Stiles (1967); original data replotted from Aguilar and Stiles (1954). t o i s ol a t e t he r o d s ys t e m a n d s t u d y i t s p r o p e r t i e s ove r a wi de r r a nge o f b a c k g r o u n d s t h a n ha s us ua l l y be e n pos s i bl e . On e uni t mu s t be d e f i n e d i n o r d e r t o u n d e r s t a n d t he i r r e s ul t s , t he uni t o f r e t i na l i l l u mi n a t i o n , t he t r o l a n d ( a b b r e v i a t e d t d) . I n h u ma n s , a l u mi n a n c e o f 1 cd m -2 vi e we d t h r o u g h a p u p i l wi t h a n a r e a o f 1 mm 2 i s s ai d t o g e n e r a t e a r et i nal i l l umi na t i on o f 1 t d. The val ues o f t r o l a n d s i n mo r e p h y s i c a l uni t s is gi ve n i n g r e a t de t a i l b y Wy s z e c k i a n d St i l es (1967). As Ag u i l a r a n d St i l es d e s c r i b e d t he i r d a t a , t he f i r s t ma j o r f e a t u r e s h a r e d b y s e ve r a l o b s e r v e r s was t h a t wh e t h e r t h e r e was z e r o b a c k g r o u n d o r a b a c k g r o u n d b e l o w 2 . 1 0 -3 t d , t h e t h r e s h o l d 278 r emai ned const ant and equal t o t he absol ut e t hr eshol d [Fig. 7 (curve A)]. In t he human, 2. 10 -a td of ret i nal i l l umi nat i on is equi val ent t o about 1000 quant a(507 nm) (deg 2 s) -1 i nci dent on t he cor nea (Wyszecki and Stiles, 1967). Thi s is t he range of backgr ounds in whi ch t he sensitivity is l i mi t ed by i nt ernal noise in t he obser ver , t he so- called " d a r k - l i g h t " (Barl ow, 1957, 1965). Then Agui l ar and Stiles observed t hat the t hreshol d began t o rise and t he sensitivity began t o fall when t he backgr ound exceeded 2. 10 -3 td. At backgr ounds above 10 -2 td (equi val ent t o r oughl y t he or der of 10 -3 cd m -2 wi t h a 3 mm pupi l ), t he t hr eshol d was equal t o a const ant t i mes t he backgr ound: Weber ' s Law. MT = kwlB R. M. SHAPLEY AND C. ENROTH-CUGELL t han 10-' td, consi st ent wi t h t he val ue r epor t ed by Fuor t es et al. (1961) and slightly hi gher but close t o Bl ackwel l ' s (1946) val ue. The r o d - cone br eak depends on retinal locus since, f or exampl e, t here can be no r o d - c o n e br eak in t he al l -cone f ovea. The number s cited above have been t aken f r om psychophysi cal i nvest i gat i on of t he near peri phery, 5 - 1 0 f r om fovea. 2. 1. 1. 2. The square root law and "' noise". As was first made clear by Rose (1948), the r od- dr i ven visual syst em is of t en st arved f or light quant a and works right up against the limit i mposed by quant al noise in t he st i mul us and the backgr ound. Under such condi t i ons, t he ul t i mat e l i mi t at i on on visual per f or mance will be neural noise caused ei t her by f l uct uat i ons in the light, or by f l uct uat i ons in the (13) sensor y pr oper t i es of the obser ver , or bot h. Thi s has defi ni t e i mpl i cat i ons about dependence of sensitivity on backgr ound l umi nance. The f or m of t he human sensitivity versus backgr ound curve depends on exper i ment al condi t i ons (e.g. st i mul us size, dur at i on, wavel engt h) and does not al ways (14) l ook like the curve obt ai ned by Agui l ar and Stiles (1954; Fig. 7, curve A). It is likely t hat over much of the visual range in t he real worl d of white stimuli on a white backgr ound, the rod-dri ven visual syst em does not at t ai n t he ideal condi t i on of adapt at i on, We be r ' s Law. The effect s of quant al noise and i nt ernal noise on the dependence of sensitivity on backgr ound are i l l ust rat ed especi al l y dr amat i cal l y in Fig. 9 f r om Bar l ow (1965). He ar r anged a r od- i s ol at i on i ncrement t hreshol d experi ment to illustrate the four di f f er ent regi mes whi ch can be seen in the human sensi t i vi t y' s dependence on backgr ound. The st i mul us was a bl ue green spot 0.75 in di amet er , of 8 ms dur at i on, pr esent ed 10 f r om t he f ovea on a 10 or ange backgr ound. I f t he st i mul us had been smal l er in area, the sensitivity woul d have been f l uct uat i on- l i mi t ed f r om the regi on l abel ed " d a r k l i ght " to " r o d s at ur at i on" . I f it had been of l onger dur at i on and l arger in area, t he curve woul d have l ooked like t he results of Agui l ar and Stiles (1954). The l i mi t at i on on sensitivity at the low end of t he backgr ound scale is usual l y at t r i but ed to " d a r k l i ght " r at her t han t o the quant al f l uct uat i ons f r om the backgr ound (Barl ow, 1957, 1964, 1977). I n this regi on of backgr ounds it is supposed t hat quant al f l uct uat i ons f r om t he backgr ound light are smal l Thus t he sensi t i vi t y S = 1/(AIT) is i nver sel y pr opor t i onal t o backgr ound in t he We be r ' s Law range: S = l / ( kw' I B) = k ' w' l / ( I n) . Above 102 td (equi val ent t o 5.107 q(507 nm) (deg 2 s) -1 i nci dent on t he cornea), the i ncr ement t hreshol d rose mor e steeply t han Weber ' s Law. This is what woul d be expect ed f or a syst em whi ch sat urat es. For exampl e, a syst em whi ch obeyed t he Naka - Rusht on equat i on (6), whi ch is one ki nd of sat ur at i on, ul t i mat el y has a t hr eshol d which depends on t he squar e of t he backgr ound level, as shown in our equat i on (7) above. Ther ef or e, t he sensitivity above 102 td in backgr ound i l l umi nat i on can be expl ai ned by sat ur at i on in one of t he neural el ement s in t he vi sual pat hway l eadi ng f r om rods t o t he cent ral ner vous syst em. Thi s phe nome non is of t en called " r o d s a t ur a t i on" , but t hat is a shor t hand descri pt i on. The psychophysi cal dat a do not e s t a bl i s h t h a t t he r e s p o n s e s o f r od phot or ecept or s t hemsel ves are sat ur at i ng duri ng " r o d s at ur at i on" . In humans, the rods nor mal l y yield the i ncrement t hr eshol d t o the cones at a backgr ound ar ound 10-' t d, as shown in Fig. 8 (Wyszecki and Stiles, 1967). Her e backgr ounds are gi ven in log erg (deg 2 s)-' , and the r od - cone br eak occurs at - 6.7 log erg (deg 2 s) -1. Thi s cor r esponds to appr oxi mat e- ly 6.104 quant a(deg 2 s) -1 which is j ust a little great er VISUAL ADAPTATI ON AND RETI NAL GAI N CONTROLS 279 - 5 - 4 - 5 - 6 I I I I : 5 P a r a ' [ - o v e o ;% = 5 8 0 n m # =500 nm I I I I / , ~. . o/ I I I 1 I I I I - - 8 ( Z e r o ) - 9 - - 8 - - 7 - - 6 - - ~ - - 4 - - 3 - - 2 Log M,~ - I FIG. 8. The r od - cone t r a ns i t i on i n h u ma n vi si on. Th e l og o f t he ener gy per uni t t i me per uni t ar ea o f a t est s pot is pl ot t ed on t he or di nat e. The abs ci s s a is t he l og ener gy per uni t t i me per uni t a r e a o f a concent r i c b a c k g r o u n d di sk. Th e uni t s f or b o t h ar e er gs (deg 2 s) -t me a s u r e d at t he c or ne a ( bef or e l osses i n t he eye a n d r et i na) . The s e can be expr es s ed equi val ent l y as qua nt a ( 507 n m) ( deg 2 s) -I a n d t he r e f or e c o mp a r e d t o r et i nal i l l umi na t i ons i n t r ol a nds , t d (cf. Ta bl e 1). Th e r o d - c one br e a k t a ke s pl ace at a b o u t - 6. 7 l og er gs (deg 2 s) -1 whi c h is equal t o 6. 10" quant a( 507 n m) (deg 2 s) -1, a n d a ppr oxi ma t e l y equi val ent t o 0.1 t d. I n t hi s e xpe r i me nt t he t est s t i mul us was a 1 deg s pot pr es ent ed f or 60 ms . Th e t est wavel engt h was 580 n m. Th e ba c kgr ound wavel engt h was 500 n m, pr es ent ed cont i nuous l y. Ba c kgr ound size was not gi ven but was pr oba bl y a b o u t 10 deg i n di a me t e r , concent r i c wi t h t he t est s pot . Te s t a n d b a c k g r o u n d s pot s were pl aced 5 deg f r o m t he f ovea. F r o m Wys zecki a n d Stiles (1967). compar ed t o t he vari ance whi ch is i nt ernal t o t he r et i na caused by t he r ma l i s omer i zat i on of phot opi gment , spont aneous openi ng and closing of phot or ecept or membr ane channels, or spont aneous neur ot r ansmi t t er release. These sources of noise have been shown t o act in a way compar abl e t o i l l umi nat i on, summi ng over distance and t i me, and t her ef or e are concei ved of as an equi val ent " d a r k l i ght " (Barl ow, 1957). The val ue of " d a r k l i ght " was est i mat ed by Barl ow (1957) as t he ret i nal i l l umi nat i on at t he i nt ersect i on of a line dr awn t hr ough t he initial fl at por t i on of t he i n c r e me n t - t h r e s h o l d curve wi t h a line ext r apol at ed down f r om t he st rai ght sl opi ng por t i on ( on l o g - l o g coor di nat es) . Let us make explicit why this i nt ersect i on gives an est i mat e of t he " d a r k l i ght ". The i n c r e me n t - t h r e s h o l d cur ve may be expressed in an equat i on: MT = (ID + IB) p. P is t he slope of t he rising por t i on of t he curve on l o g - log coordinates, and varies f r om 1 for Weber' s Law t o 0.5 f or t he square r oot law. The initial flat por t i on of t he curve is where Io is much l arger t han IB and t her ef or e in this range AI x = ( I o ) P. The st rai ght sl opi ng por t i on is where IB is much l arger t han Its, and t here M x = (IB) P. At t he i nt ersect i on of t he t wo lines ( on log - log coor di nat es) , t he t wo expressions for t he threshold illumination are equal, and t hus at this poi nt t he backgr ound i l l umi nat i on Ia equal s t he " d a r k l i ght " Io. It appears t hat t he " d a r k l i ght " limits visual sensitivity by being the domi nant t er m in t he retinal noi se r at her t han by setting t he val ue of t he ret i nal gain cont r ol . Barl ow (1957) surveyed t he l i t erat ure on measur ement s of " d a r k l i ght " and f ound t hat P R R 3 - H 280 most investigators agreed that the "dark light" was represented by an equivalent retinal illumination of 2- 10 -3 td. This is equivalent to 1000 quanta(507 nm) (deg 2 s)-' incident on the cornea. As a source of noise, such a value of "dar k light" is enough to explain the magnitude of the absolute threshold of about one hundred quanta at the cornea (Hecht et al . , 1942), with reasonable assumptions about integration time, area of summation, and quantum efficiency of the eye, as the following calculation demonstrates (cf. Barlow, 1957). If the integration time is about 0.1 s, and the integration area is about 1 deg 2, and the eye' s quant um efficiency is about 0.25, then one can calculate that the "dar k light" has a mean and variance of 25 events per integration time per integration area, and therefore a standard deviation of 5 events per integration time per integration area. If one assumes a threshold signal to noise ratio of 4 (Rose, 1973), the absolute threshold predicted from the "dar k noise" caused by "dar k light" would be about 20 quanta effectively absorbed by the rods, which is close to the observed 25 (the measured 100 at the cornea multiplied by the quantum efficiency of 0.25). Thus, "dar k light" appears to limit absolute sensitivity by causing "dar k noise". >. n 5 8 o ._J R. M. SHAPLEY AND C. ENROTH-CUGELL Ascending the scale of backgrounds in Fig. 9, we next encounter the region in which the increment threshold increases like the square root of the background. This is the range in which Rose (1948) proposed that the visual threshold would be limited by quantal fluctuations. In this range one observes the square root law; A~T ~--- ~Q ~B 1/2 ( 1 5 a ) S = kQ' 1 B - 1 / 2 (15b) The square root law follows from Rose' s explanation because the standard deviation of a neural shot-noise (Dodge e t al . , 1968), resulting from the temporal summation of the neural responses to randomly arriving quanta, would grow like the square root of the background light level. The response to the increment on this background would have to be picked out in the presence of the neural shot noise induced by the background, and the stimulus strength would have to be increased in proportion with the standard deviation of that noise. This argument is valid with the assumption that the signal/noise ratio at threshold is kept i / " FIXATION POINT x ~ RO0 SATURATION - i ' " AOAPTINGBACKGROUNO / W ~ / QUANTUM FLUCTUATIONS / / DURATION 8 MSEC I 1 I I 1 l I I 2 3 4 5 6 7 8 9 LOG ( QU A N T A 5 0 7 mp DEG. "= SEC " t ) BACKGROUND FIG. 9. Sensitivity vs background: dark light, quantal fluctuations, Weber's Law, and rod saturation. The full repertoire of psychophysical laws is illustrated in this experimental curve which is the graph of sensitivity for a blue -green test spot, 0.75 deg in diameter, 8 ms in duration, presented 10 deg from the fovea on an orange colored background which was 10 deg in diameter. The different regimes are labeled in the figure. From Barlow(1965). VISUAL ADAPTATI ON AND RETI NAL GAI N CONTROLS const ant (Rose, 1948). Ther e are good reasons t o believe in this assumpt i on, because human subjects can keep t hei r reliability (e.g. percent age of false positive responses) pr et t y const ant at t hr eshol d across backgr ounds, and this is consi st ent with a const ant si gnal / noi se r at i o across backgr ounds (Rose, 1973). The expl anat i on of t he square r oot law as a result of quant al fl uct uat i ons and t he resul t i ng neur al shot noi se seems i nt ui t i vel y implausible t o some people because we do n o t " see" t he q u a n t a l f l u c t u a t i o n s whi ch l i mi t our per f or mance as we can see supr at hr eshol d visual noise, e.g. the " s n o w" on a poor television picture. However , i f one comput es t he number of quant a avai l abl e in cont r ast det ect i on tasks, as Rose did ori gi nal l y, one is f or ced t o t he concl usi on t hat quant al noise is t he limiting f act or in such per f or mance. We will do this cal cul at i on bel ow in t he cont ext of t he dependence of cont rast sensitivity on mean level of i l l umi nat i on. Not e t hat t he decrease in sensitivity caused by quant al noi se is not st ri ct l y speaki ng l i ght adapt at i on. Rat her , it is a loss in sensitivity caused by t he propert i es of t he stimulus. The visual system woul d not have t o adapt in or der t o fol l ow t he square r oot law. Next in Fig. 9 are t he t ransi t i on t o Weber ' s Law, and fi nal l y r od sat ur at i on, bot h of whi ch we consi dered when we discussed t he results of Agui l ar and Stiles. 2. 1. 1. 3. Square root to Weber transition. The t ransi t i on f r om t he square r oot law t o Weber ' s Law deserves special consi der at i on. Thi s occurs at t he backgr ound illumination at which the visual system switches f r om being quant um-l i mi t ed t o being gain- c o n t r o l l i mi t e d. As s uch, i t is t he f i r s t psychophysi cal i ndi cat i on of real light adapt at i on. Cr awf or d' s (1947) results were t he first t o i ndi cat e t ha t t he t r a n s i t i o n o c c u r r e d at d i f f e r e n t backgr ounds f or test t arget s of di f f er ent size. Barl ow (1957) expl or ed t he condi t i ons whi ch i nfl uenced t he t ransi t i on. The concl usi ons he r eached were t hat bri ef, small test spots pr oduced increment thresholds which rose like the square r oot of t he backgr ound over t he ent i re scot opi c range. The i ncr ement t hr eshol d curves f or spots of large area, present ed f or l onger times, had slopes close t o 1, i.e. Weber ' s Law, over t he whol e r od range as in Agui l ar and Stiles' wor k. Fi gure 9 illustrates 281 t hat i ncrement t hr eshol d curves f or spots of i nt er medi at e area, pr esent ed bri efl y, have a transition f r om square r oot t o Weber laws. It would be i nt erest i ng t o know exact l y how this t ransi t i on depends on t he spatial and t empor al fact ors, but Barl ow' s dat a f or t he r od system are i ncompl et e on this poi nt . 2.1.2. CONTRAST SENSITIVITY AS A FUNCTION OF ILLUMINATION LEVEL Under t he condi t i ons t hat Weber ' s Law hol ds, the threshold increment of illumination is a constant t i mes t he backgr ound, i.e. (MT)=kw lB. Thi s implies t hat t he t hr eshol d cont r ast and its inverse, t he cont rast sensitivity, are const ant in t he Weber ' s Law regi me, as can be seen in t he fol l owi ng equat i on Scon = 1 / ( AI T / / B ) Sco n = 1/kw = k' w. (16) In Fig. 7 this is t he range f r om 10 -2 t o 102 t d. Remember t hat const ant cont r ast sensitivity is t he ideal whi ch light adapt at i on must strive f or i f it is t o achi eve its mai n goal: percept ual i nvari ance of refl ect i ng surfaces with changes in backgr ound. The r e f or e , t o t he ext ent t he ps ychophys i cs appr oaches Weber ' s Law, this maj or goal is met . The threshold Weber contrast shows a clear break bet ween r od and cone f unct i on as can be seen in Fig. 10, f r om Bl ackwel l ' s (1946) dat a (see also St ei nhardt , 1936; Crai k, 1938). The mi ni mal (Weber) cont r ast t hr eshol d f or t he rods is about 0.08 cor r espondi ng appr oxi mat el y t o a cont r ast sensitivity, as we have def i ned it above, of 12. However , smal l er t arget s requi re hi gher cont rast ; an 18' spot has as its least r od cont r ast t hreshol d a (Weber) cont r ast of 0.67 wi t h a cor r espondi ng (Weber) cont r ast sensitivity of onl y 1.5. Ther e is a clear j ump in per f or mance when the cones come in, wi t h We b e r c o n t r a s t t h r e s h o l d s decl i ni ng asympt ot i cal l y t o 0.008 and cont r ast sensitivities cl i mbi ng t o 125 or mor e. At t he t ransi t i on f r om square r oot t o Weber ' s Law, t he cont r ast t hr eshol d undergoes a t ransi t i on f r om declining like t he inverse square r oot of backgr ound t o becomi ng a const ant i ndependent of backgr ound. Thi s can be seen f r om t he expressions 282 R. M. SHAPLEY AND C. ENROTH-CUGELL 3 ~ 5 . 6 2 -7 6 -5 q -3 2 ] 0 1 5 Log o d a p t o t i o n b r , g h t n e s s { f oo~ t o mb e r ~ s, - I f o o t L a mb e r ~ = 5 4 5 cd m ) FIG. 10. The dependence of the Weber contrast threshold on background illumination. The test targets were circular disks on a large concentric background, viewed continuously. The test and background were neutral in color, and thresholds were determined by the method of forced choice. The points are the average thresholds of ten subjects. Different sized test targets generated the different curves, and the size of the target, in minutes of visual angle, is indicated for each curve. The straight line drawn in at the lower left corner of the graph has a slope of - 1/2, which is the predicted slope of the contrast threshold against background for the square root law. Modified from Blackwell (1946). f or the square r oot law equat i on (15) and Weber ' s Law equat i on (14). The cont rast threshold, MT/ I B, becomes Square r oot law (MT/Ia) = kQ I n- l/2 (17a) Weber ' s Law (AIT/Ia) = kw. (17b) Behavior of this sort can be seen in Blackwell' s dat a Fig. 10. The contrast threshold falls roughl y like the inverse square r oot over most of t he scotopic range, t hough it is begi nni ng t o level of f t owar ds Weber ' s Law [equat i on (17b)] f or t he largest targets. Under t he condi t i ons of Blackwell' s experi ment s, t he r od system does not ent er i nt o t he Weber Law regime bef or e t he cones take over. However , the results of Koender i nk et al. (1978) on cont r ast sensitivity as a f unct i on of retinal eccent ri ci t y and mean level suggest t hat , in t he far per i pher y of t he ret i na, Weber ' s Law is achieved in the scot opi c range even wi t hout t he t wo- col or pr ocedur e of Agui l ar and Stiles (1954). Mor e recent studies of t hreshol d cont r ast have used a stimulus with a sinusoidal l umi nance profi l e [Fig. 1 (B)]. Ther e are t hree extensive studies on t he (Rayleigh) cont rast sensitivity of t he r od system, by van Nes and Bouman (1967), by Dai t ch and Gr een (1969), and by Smi t h (1973). These are consi st ent in finding a maxi mal (Rayleigh) cont rast sensitivity of bet ween 20 and 30 (which is equi val ent t o a Weber cont r ast sensitivity of 1 0 - 15, in agreement with Blackwell). Fi gure 11 is a pl ot of some of Dai t ch and Gr een' s dat a showi ng t he dependence of scot opi c cont r ast sensitivity on mean level. The dependence on mean level of t he (Rayleigh) cont r ast sensitivity f or gratings conf i r ms the concl usi on t hat t he square r oot law r at her t han Weber ' s Law domi nat es r od vision in humans (Dai t ch and Green, 1969; Smi t h, 1973). The cont r ast sensitivities f or very low spatial f r equency gratings (less t han 0.5 c/ deg) level of f t o a Weber Law limit, but cont r ast sensitivities f or higher spatial frequenci es increase with t he average level of i l l umi nat i on, fol l owi ng t he predi ct i on of t he square r oot law, equat i on (17a). Ther ef or e, the shape of t he " cont r as t sensitivity vs spatial f r equency" curves f or st eady viewing change with the average level. At low levels t hey are "l ow- pass" funct i ons which have t hei r peak sensitivity at low spatial f r equency. At higher levels of i l l umi nat i on, t here is a defi ni t e peak at an i nt ermedi at e spatial frequency, and a low spatial f r equency " c u t o f f " (Dai t ch and Green, 1969; Smith, 1973). An anal ogous effect is seen in phot opi c human vision (van Nes and Bouman, 1967). I f one calculates how many quant a are available t o detect t hreshol d cont rast , one is forced t o admi t t hat quant al fl uct uat i ons must pl ay a significant VISUAL ADAPTATION AND RETINAL GAIN CONTROLS 283 > . . t - , - 1 / 1 z ~ t . ) 10.0 8.0 6,0 4.0 2 . 0 0 . 0 I 2 C/DEG o o o o 0.5 ClDEG 1 L I l I I I -2.50 -2.00 -I.50 -I.00 -0.50 0.00 0.50 MEAN I LLUMI NATI ON (LOG TROLANDS) FIG. 11. The dependence of the Rayleigh contrast sensitivity on mean level. Results for two spatial frequencies of test target are shown: 0.5 c deg-' as empty circles, and 2 c deg -1 as filled circles. An artificial pupil of 1 mm di ameter was used. The test targets were sinusoidal grating patterns generated on the face of a c at hode ray t ube ( CRT) di s pl ay which subtended 8 deg by 3 deg, centered 12 deg from the fovea. The phosphor of the CRT was P31(yellow-green) but a Wratten 65 filter was interposed between subject and screen to shift the dominant wavelength of the phos phor t owards the peak of the rod spectral sensitivity function. The target was presented for 200 ms every second. Control experiments established that the rods were the photoreceptors which determined threshold up to 0 log t d. Pl ot t e d from Daitch and Green (1969). par t in limiting cont r ast sensitivity in t he scot opi c r ange. Suppos e we ma ke t he c ons e r va t i ve assumpt i on t hat no less t han one quar t er of t he quant a i nci dent on t he cor nea may be used f or cont r ast det ect i on. Fur t her we assume t hat t he i nt egrat i on t i me over which quant a may be count ed is about 0.1 s. Fi nal l y, and most crucially, we assume t hat t he i nt egr at i on area f or t he neur ons which are i nvol ved in t he det ect i on task is at most 1 deg 2. Thi s last assumpt i on is based on t he measur ement s of recept i ve field cent er sizes of ret i nal gangl i on cells in cats and monkeys at compar abl e ret i nal loci t o t he ones used in t he human psychophysi cal experi ment s ( Hubel and Wiesel, 1960; de Monast er i o, 1978; Cl el and et al., 1979), and on t he not i ons of t he Channel Hypot hesi s expanded in Sect i on 2. 1. 3. Gi ven these assumpt i ons, one may cal cul at e what t he highest cont r ast sensitivity mi ght be in t he scot opi c range, at t he highest backgr ound in this range whi ch is about 10-' td. This backgr ound retinal illumination cor r esponds t o 5" 104 quant a( deg ~ s) -' i nci dent on t he cor nea. I f one quar t er of t hese quant a may be used by t he ret i na, in an area of 1 deg 2, over a t i me of 0.1 s, t hen t he t ot al avai l abl e quant a f r om such a backgr ound are 1250. I f quant al fl uct uat i ons are t he limiting f act or , t hen t he st andar d devi at i on of t he quant al count with this mean number of 1250 must be t he limiting source of noise. The st andar d devi at i on of t he quant al noise must be t he square r oot of 1250, whi ch is about 35. In most det ect i on experi ment s t he s i gna l - noise rat i o is t aken t o be about 4, so t hat t he t hr eshol d amount of quant a woul d have t o be about 4-35 = 140. This is 140 neural l y triggered quant al events; we must mul t i pl y by 40 (the reci procal of 1/ 40 whi ch is t he pr oduct of the quant al efficiency of 1/ 4 times the integration t i me of 0.1 s) t o obt ai n t he stimulus i l l umi nat i on, measur ed at t he cor nea, t o pr oduce this many neural events. Thus we get t hat t he stimulus i l l umi nat i on must be equi val ent t o 5. 6. 103 quant a(deg 2 s)-' . The best quant um-l i mi t ed (Weber) cont r ast sensi t i vi t y one coul d expect woul d t her ef or e be (Ia/MT) = (5" 104)/(5.6 103) or about 9, whi ch is about as high as t he human (Weber) cont r ast sensitivity reaches in t he scot opi c range. It is i nt erest i ng t hat Agui l ar and Stiles (1954) went t hr ough an anal ogous cal cul at i on in t he Discussion of t hei r famous paper, and concl uded t hat t hey had discredited the idea t hat quant al fluctuations limited det ect i on of cont rast . However , in t hei r cal cul at i on t hey made t he assumpt i on t hat , because t hey used a disk 9 deg in diameter as a stimulus, all the quant a in t he 64 deg 2 ar ea of t he disk were available f or t he det ect i on t ask. We now know this assumpt i on is unr easonabl e. Pr esent l y avai l abl e knowl edge about t he di st ri but i on of recept i ve field sizes suppor t s Rose' s (1948) hypot hesi s about t he i mpor t ance of quant um fl uct uat i ons in scot opi c vision. The great i mpr ovement in cont r ast sensitivity associated with the shift f r om r od t o cone pat hways is compl i cat ed by t he effect s of ret i nal i nhomo- genei t y studied by Koender i nk et al. (1978). They measur ed cont r ast sensitivity funct i ons with sine gratings which subt ended a fixed area of 4 deg by 4 deg, at several l ocat i ons in t he visual field. Fi gure 12 shows t hei r fi ndi ng t hat in t he far per i pher y t here is no increase of cont r ast sensitivity bet ween scot opi c and phot opi c backgr ound levels, and in t he near per i pher y t here is onl y an increase of about a f act or of t wo in t he phot opi c cont r ast 284 R. M. SHAPLEY AND C. ENROTH- CUGELL 10 I I I 1 I I - / ecc l I I I L I 0 o 10 5 0 FIG. 12. Lowes t t hr e s hol d mo d u l a t i o n dept h as a f unc t i on of r et i nal eccent ri ci t y a n d me a n level o f i l l umi nat i on. Thes e da t a ar e all f r o m c ont r a s t t hr e s hol d me a s u r e me n t s f or si ne gr at i ng pat t er ns whi ch s ubt e nde d 4 deg by 4 deg a nd were dr i f t i ng at a t e mpor a l f r equency of 4 Hz. The st i mul i were pr es ent ed on a Hewl et t - Pa c ka r d 1310A di s pl ay wi t h a P 15 phos phor ( domi nant wavel engt h at 510 nm) , and were viewed by t he s ubj ect t h r o u g h a 2 mm di amet er art i fi ci al pupi l . Cont r as t sensi t i vi t y cur ves were me a s ur e d at each eccentricity a nd i l l umi nat i on, a n d t he mi n i mu m t hr e s hol d mo d u l a t i o n dept h (t he r eci pr ocal o f t he peak c ont r a s t sensi t i vi t y acr os s spat i al f r equency) for each condi t i on was denot ed m*, whi ch is t he qua nt i t y pl ot t ed i n t he f i gur e. The di f f er ent s ymbol s ar e f or di f f er ent me a n i l l umi nat i ons . Emp t y circles are for 10 t d; filled circles ar e f or 1 t d; circles wi t h dot s in t he cent er ar e f or 0.1 t d. Fr o m Koender i nk et al . (1978). sensitivity over the scot opi c. It seems t hat t he foveal phot opi c cont rast sensitivity is especially high when compar ed with t he phot opi c (and scotopic) cont rast sensitivity of t he peri pheral retina. They also f ound t hat , f or a st i mul us of a gi ven size and spat i al frequency, t he t r ansi t i on f r om squar e r oot law to Weber ' s law depended on the ret i nal eccentricity; t he t r ansi t i on occur r ed at l ower l umi nances t he mor e per i pher al was t he st i mul us. 2.1.3. THE CHANNEL HYPOTHESIS Ther e exists a wor ki ng hypot hesi s one can use t o i nt er pr et Ba r l ow' s , Bl ackwel l ' s , Dai t ch and Gr een' s, and Koenderi nk e t a l . ' s , findings about the i nfl uence of size or spat i al f r equency or retinal eccent ri ci t y on t he t r ansi t i on f r om squar e r oot law t o Weber ' s Law. Thi s hypot hesi s is t hat t he vi sual syst em is compos ed of a set of size or spat i al f r equency channel s ( Campbel l and Robs on, 1968; Enr ot h- Cugel l and Robson, 1966; Robs on, 1975; Gr a ha m, 1980). It is supposed t hat a t est st i mul us sifts t hr ough t he channel s and excites onl y t hose channel s which are " t u n e d " t o t he st i mul us. Such an hypot hesi s can be used in conj unct i on wi t h research fi ndi ngs on t he adapt at i on pr oper t i es of ret i nal gangl i on cells wi t h di fferent -si zed recept i ve fields to make sense out of t he t r ansi t i on f r om square r oot law t o Weber ' s law (Enrot h-Cugel l and Shapl ey, 1973b). In essence, t he expl anat i on is t hat t he channel s whi ch r es pond t o l ow spat i al frequenci es suf f er mor e loss of sensitivity wi t h increase in backgr ound t han do the channel s which r espond t o high spat i al frequenci es because ret i nal gai n cont rol s depend on areal s ummat i on of backgr ound i l l umi nat i on. The channel s which are most sensitive t o low spat i al frequenci es, accor di ng to this view, sum adapt i ve signals over a l arger area t han do t he channel s mos t sensitive t o high spat i al frequenci es. A mor e det ai l ed expl anat i on of this poi nt is of f er ed in t he Theor y section. 2 . 1 . 4 . HUMAN AND FELINE CONTRAST SENSITIVITIES COMPARED Since mos t of t he el ect rophysi ol ogi cal results on retinal adapt at i on and gai n cont r ol come f r om exper i ment s on ani mal s, it is rel evant t o compar e h u ma n p s y c h o p h y s i c a l and cat b e h a v i o r a l meas ur ement s of cont r ast sensitivity at di f f er ent mean levels of i l l umi nat i on. Recent l y, this has become possible because of the results of Past er nak and Mer i gan (1981). Thei r results show t hat t here are not i ceabl e di fferences bet ween t he cont r ast sensitivities of cat and man. As can be seen in Fig. 13, the cat ' s cont r ast sensitivity i mpr oves with mean level in a way anal ogous t o man, but at all mean levels t he cat ' s peak cont r ast sensitivity is l ower. Thi s may be because t he hi gh mean l umi nances were phot opi c f or t he human, but all mean l umi nances were pr obabl y scot opi c f or t he cat. The r o d - c o n e t r ansi t i on in cat s is at a much hi gher mean l umi nance t han in man, pr es umabl y because of t he much hi gher r od - cone rat i o in t he cat (Lennie e t a l . , 1976). Cont r ast sensitivity grows much mor e steeply for high spatial frequencies t han f or low, in cat as in man. However , t he acui t y of t he cat is much worse t han t hat of man, as can be seen in Fig. 13 by the separ at i on bet ween t he high spat i al f r equency cut of f s of t he cont r ast sensitivity curves at all mean levels except t he lowest. However , t he basi c si mi l ari t i es bet ween t he dependences of fel i ne and h u ma n c ont r a s t sensitivities on mean level of illumination encourage us t o at t empt a synt hesi s of huma n psychophysi cal studies of light adapt at i on wi t h t he body of ioo io LU t.~ ~0 0 2 io VI S UAL ADAP T AT I ON AND RE T I NAL GAI N CONT ROL S ~ 0 0 CO/m2 1 0 - 2 cclln~- ~ 16 10 acdlm2 16 x 1 0 _ 6 c d / r n 2 S P A T I A L F R E Q U E N C Y ( c / d e g ) FIG. 13. The de pe nde nc e o f c ont r a s t sensi t i vi t y on me a n l umi na nc e in ma n a nd cat . The cat r esul t s were obt ai ned wi t h a t wo- al t er nat i ve f or ced choi ce p a r a d i g m a n d t hr e s hol d was t a ke n as 75% cor r ect . The h u ma n r esul t s were obt a i ne d in a si mi l ar ma n n e r . For bot h h u ma n a n d cat , t he nat ur al pupi l openi ng det er mi ned ret i nal i l l umi nat i on. The gr at i ng pat t er ns were di s pl ayed on t wo CRTs wi t h P3 1 ( y e l l o w- g r e e n ) phos phor s . The me a n l umi na nc e was cont r ol l ed wi t h neut r al dens i t y fi l t ers pl aced i n f r ont o f t he CRT scr een. Fr o m Pa s t e r n a k a n d Me r i ga n (1981). knowl edge a bout ret i nal adapt at i on in t he cat . 2.1.5. THE TIME COURSE OF ADAPTATION The change in sensi t i vi t y af t er a ba c kgr ound has been t ur ned on is not i mmedi at e. The t i me cour se of this change has been st udi ed using t he t echni que of " Cr a wf o r d ma s k i n g " , in whi ch a backgr ound is fl ashed on f or a fi ni t e t i me, and a br i ef t est fl ash is super i mposed on t he backgr ound at vari ous times af t er t he onset of t he backgr ound. The mos t useful 285 resul t s obt ai ned wi t h this met hod on t he r od vi sual pat hways are t hose of Adel son (1982). Fi gure 14 summar i zes his mai n fi ndi ngs. I t can be seen t hat when t he t est fl ash is super i mposed on a st eady backgr ound, t he t hr eshol d f or t he t est fol l ows a curve like t hat obt ai ned by Agui l ar and Stiles (1954; Fi g. 7A). However , when t he t est is appl i ed j ust at t he onset of t he backgr ound, t he dat a show evidence f or sat ur at i on, a st eep upt ur n in t he t hr eshol d vs backgr ound curve, at a backgr ound i l l umi nat i on mor e t han t wo log units l ower t han t he st eady-st at e curve. One can t her ef or e i nfer t hat t he process of adapt at i on t akes t i me t o rescue t he r esponse f r om t hi s sat ur at i on. Ho w much t i me can be i nf er r ed f r o m t he ot her curves in Fi g. 14. At t est - backgr ound offset s of 2 0 0 - 1000 ms the sat ur at i on occurs at about one log uni t hi gher t han at zero of f set . Thi s means t hat a first st age of adapt at i on is basi cal l y compl et e wi t hi n 200 ms. Then bet ween 1 s and st eady st at e a second sl ower process moves t he sat ur at i on out anot her log uni t . Adel son (1982) i nt er pr et s these results in t er ms of a cascade of processi ng, in which an initial linear st age is f ol l owed by a sat ur at i ng t r ansducer . I f adapt at i on act s at or bef or e t he sat ur at i on, sat ur at i on of t he r esponse of t he cascade can be avoi ded. Thi s is a si mi l ar scheme t o t he one we out l i ned in Sect i on 1.2.2. However , Adel son (1982) consi ders t wo di f f er ent t ypes of adapt at i on whi ch he calls " s ubt r a c t i ve " and " mul t i pl i c a t i ve " . I n experi ment s in which he used pr e- adapt at i on bef or e t he " Cr a wf o r d ma s ki ng" , he was abl e t o show t hat t he r api d adapt at i on, whi ch has a t i me const ant of about 100 ms, is multiplicative, i.e. it acts like a gai n cont r ol . However , t he sl ow adapt at i on, whi ch has a t i me const ant of a bout 30 s under t he condi t i ons of Ad e l s o n ' s e xpe r i me nt s , a p p e a r e d t o be subt r act i ve. The nat ur e of t he slow subt r act i ve adapt at i on di scovered in these experi ment s remai ns t o be det er mi ned. The r api d gai n cont r ol has been obs er ved in physi ol ogi cal exper i ment s t o be described bel ow. It is of t heoret i cal i mpor t ance t hat t he gai n cont r ol mechani s m, while fast , is not t ur ned on i nst ant aneousl y when t he backgr ound goes on. 2.1.6. WHAT PSYCHOPHYSI CS I MPLI ES ABOUT RETI NAL MECHANI SMS 2.1.6.1. Psychophysical sensitivity loss: gain 286 R. M. SHAPLEY AND C. ENROTH-CUGELL 3 2 S o o a~ - - I k - 2 --5 ( a) Del / / / ! ~ 1 200 9J I L 5 o o . , | L ,ooo d / / " cgr--~: g ~; o 0 5 i 5 2 2 ~ t DeLay sec , g, " 1 , " , - : i - : , 0 20 40 60 80 ] O0 -3 -2 - I 0 l 2 3 DeLay sec Background, Log scot t d FIG. 14. Transient saturation and adaptation in the human rod system. A square test spot, 4.5 deg on a side and of 480 nm wavelength, was superimposed on an 11 deg red background centered 12 deg from the fovea in the nasal direction of the visual field (temporal retina). In panel (a) are shown increment threshold curves taken at different times after the onset of the background flash. The test was 30 ms in duration, and the background was either continuous (empty squares) or was flashed on at various times before the test and was left on for 400 ms after the test. The delay between background onset and test onset is shown in (a) as an inset. Also graphed there (lower right) are the Weber contrasts at threshold for the different conditions, though they are labeled Weber fractions here. In panels (b) and (c) the time course of adaptation is shown by plotting the value of the threshold for the test spot presented with different delays after the onset of the background (3 scotopic td) which remains on. Panels (b) and (c) are the same data plotted on two different time scales, to illustrate the rapid and slow phases of adaptation. From Adelson (1982). reduction or noise increase? Wh a t d o t he p s y c h o p h y s i c a l r es ul t s t el l us a b o u t t he u n d e r l y i n g me c h a n i s ms o f t he c o n t r o l o f s e ns i t i vi t y i n h u ma n r o d vi s i on? I n e x p e r i me n t s whi c h s i mu l a t e d o u r e v e r y d a y vi s ual p e r c e p t i o n o f o b j e c t s ( Bl ackwel l , 1946), t he r o d - d r i v e n h u ma n vi s ual s ys t e m d i d not qui t e r e a c h t he goa l o f We b e r ' s La w. Up t o t he t r a ns i t i on f r o m t he s quar e r o o t l aw t o We b e r ' s Law, t he t he or e t i c a l e x p l a n a t i o n r e qui r e d is onl y i n t e r ms o f c ha nge s i n t he n e u r a l noi s e l evel c a us e d b y q u a n t u m f l u c t u a t i o n s i n t he b a c k g r o u n d l i ght . At hi gher b a c k g r o u n d s t h a n t he l evel s r e qui r e d f or t he r od - cone t r a ns i t i on i n t he whi t e - on- whi t e s t i mul us s i t u a t i o n , t he r o d s ys t e m mu s t a d a p t , f or i t s s e ns i t i vi t y d r o p s mo r e t h a n c a n be a c c o u n t e d f or s i mp l y i n t e r ms o f a n i nc r e a s e o f noi s e due t o q u a n t u m f l uc t ua t i ons . Al s o , t he n e u r o n s whi c h a r e t he me d i u m f or pe r c e i vi ng gr a t i ngs o f ve r y l ow s pa t i a l f r e q u e n c y mu s t ha ve t he i r s ens i t i vi t i es c o n t r o l l e d b y mo r e t h a n q u a n t a l noi s e, be c a us e t he s e ns i t i vi t y d r o p s mo r e s t e e pl y t h a n t he s q u a r e r o o t l aw. Thus noi s e a n d ga i n c o n t r o l mu s t be i nvol ve d i n t he v a r i a t i o n o f vi s ual s e ns i t i vi t y wi t h c ha nge s i n b a c k g r o u n d o r a ve r a ge l evel o f i l l u mi n a t i o n . 2. 1. 6. 2. The site of adaptation: the retina. On e q u e s t i o n we ha ve not a s k e d or a n s we r e d i s, wh a t d o t he p s y c h o p h y s i c a l r es ul t s i mp l y a b o u t t he si t e o f a d a p t a t i o n ? I t c o u l d be i n t he r e t i n a or i n t he b r a i n o r b o t h . Se ve r a l p s y c h o p h y s i c a l e x p e r i me n t s i mp l y t ha t i t is r e t i na l . F o r e x a mp l e , Ba t t e r s b y a n d Wa g ma n (1962) d e mo n s t r a t e d t h a t t he r e is l i t t l e i n t e r o c u l a r t r a n s f e r o f s t e a d y l i ght a d a p t a t i o n . I f t he r e wer e a l ar ge c e nt r a l c o mp o n e n t o f l i ght a d a p t a t i o n , one mi ght e xpe c t a l a r ge a mo u n t o f i n t e r o c u l a r t r a n s f e r . Ot h e r s i mi l a r e x p e r i me n t s a r e t hos e o f He i n e ma n n (1955) a n d Wh i t t l e a n d Cha l l a nds (1969) b o t h o f whi ch i nvol ved i nt e r oc ul a r br i ght ne s s ma t c h i n g . I n b o t h set s o f e x p e r i me n t s , a di m ma t c h i n g l i ght on an even d a r k e r b a c k g r o u n d was ma t c h e d t o mu c h b r i g h t e r l i ght s on b r i g h t b a c k g r o u n d s . Th e t wo l i ght s wer e pr e s e nt e d t o d i f f e r e n t eyes a n d t he r e was e v i d e n t l y l i t t l e or no i n t e r o c u l a r t r a n s f e r o f t he a d a p t i n g ef f ect o f t he b r i g h t b a c k g r o u n d . Th e ga i n o f t he h u ma n e l e c t r o r e t i n o g r a m ( ERG) i n r e s p o n s e t o a l a r ge t est t a r g e t c ha nge s wi t h b a c k g r o u n d i l l u mi n a t i o n i n a wa y ve r y s i mi l a r t o t he ps yc hophys i c a l sensi t i vi t y f or t ar get s o f t he s a me VISUAL ADAPTATION AND RETINAL GAIN CONTROLS size (Biersdorf et al., 1965). This experiment indicates that the gain of the retina is affected at the same light levels as the visual threshold for large targets. A further question one might raise is, where in the retina is the site of this gain change? Much of the physiological work we present below was motivated by this question. Prior to the physiological investigation, this question of the site in the retina of gain control was investigated psychophysically and this work led to the two concepts of gain control by pigment depletion, which has been shown to be only a minor factor in the control of retinal gain, and the adaptation pool, which still motivates research. 2.1.6.3. Pi gment depletion. Vertebrate visual pigments are bleached by light into non-receptive intermediates which must be regenerated to a receptive state enzymatically. It was thought for some time that visual pigment depletion, resulting from bleaching by the background light, was the main cause of light adaptation (Hecht, 1924). This would put the site of adaptation right at the initial transduction stage in vision. However, it is now known that the pigment regenerates so quickly that there is a rather small amount of pigment bleached 1.0 0.8 r, o "~ 0.6 o L- ~ 0.4 0.2 287 for either rods or cones over much of the visual range of backgrounds (Rushton, 1962). This can be seen in Figs. 15 and 16. Figure 15 is from Alpern and Pugh (1974) and shows the fraction of human rhodopsin, the rod pigment, bleached as a function of a steady background. Even up to rod saturation (5" 108 quanta(507 nm) (deg 2 s) -~) the rod pigment has been bleached away less than 270, but the visual threshold would have risen by a factor of more than 107, a discrepancy in prediction. Figure 16 shows a similar result for human cones, from Alpern et ai. (1971). The cone pigment has been only bleached by half at about 104s trolands which is near the top of the photopic range. So we can rule out pigment depletion as playing a major role in human light adaptation, though it has some effect on photopic (cone-driven) sensitivity (Boynton and Whitten, 1970; Valeton and van Norren, 1983). It is interesting that photopigment depletion began to be doubted as an explanation of light adaptation quite a long time ago, from experiments on increment threshold and photopigment concentration in the frog retina conducted by Granit et al. (1939). In those experiments ERG thresholds were increased by log units by backgrounds which had no . - . \ . : , \ ! \ . ' \ ; 5 7 5 n m " ! \ . 5 0 0 n m . ~ . 0 , I I I I I I 1 2 1 3 1 4 L o g q u a n t a s e c - I c m - ~ o f r e t i n a a t t h e c o r n e a ~ 1 0 s e c W h i t e I 1 5 FIG. 15. Pigment depletion in rods caused by steady illumination. The amount of the rhodopsin in a 5 deg field located 18 deg peripheral to the fovea in the temporal retina was measured with a retinal densitometer (described in Alpern et a l . , 1971 and in Alpern and Pugh, 1974) Plotted vertically i s t h e fractional amount of rhodopsin left in the steady state at the background illumination plotted on the horizontal scale. The three curves are for steady lights of 500 nm, 575 nm, and white light The 500 and 575 nm lights were obtained with the white source and interference filters The results to concentrate on are the 500 nm results, since 500 nm is near the peak wavelength of absorption of rhodopsin and so will give the greatest pigment depletion at the lowest retinal illumination. The half bleaching point for the 500 nm curve is at about 13.2 log quanta(cm 2 s)-' which is equivalent to about 10.2 log quanta(deg 2 s) -1, the more familiar unit of retinal illumination [also equivalent to about 4.3 log (scotopic) td]. These values of quanta are referred to the cornea and do not take into account losses in the eye. From Alpern and Pugh 0974). PRR3-H* 288 R. M. SHAPLEY AND C. ENROTH-CUGELL I I ~ " ( i _ p ) 0 0 ) 4 . . ~ = p I 2 06- - 0"4 i ' ~ ~ 0"2 O'O I 1 l 1 4 45 50 5~ log I 2 photopic trofands FIG. 16. Pigment depletion in cones caused by steady illumination. Plotted in the graph are retinal densitometry measurements of the human fovea at different levels of steady illumination of a yellow - green bleaching light. P is the fraction of cone pigment remaining in the steady state. The half-bleaching illumination is about 4.5 log photopic td, which is equivalent to approximately 10.6 log quanta(deg ~ s)-' of quanta at 560 nm, the peak wavelength for the long-wavelength cones. Note that the half-bleaching point for the cones is within half a log unit of the half bleaching point for rods, when expressed in units of quanta of the optimal wavelength for the pigment. From Alpern et al. (1971). measurable effect on the extractable rhodopsin concentration. Furthermore, Rose (1948, 1973) has presented an elegant argument which disposed of pigment depletion as the sole explanation of light adaptation. If pigment depletion were significant in the scotopic range, the quantal-noise limited contrast sensitivity would be held down to a very low level. This would be the consequence of depleted pigment; light would be absorbed less efficiently. Since the contrast sensitivity does rise in the quantal-noise limited range, and it rises to a degree consistent with a relatively constant quantum-efficiency, there cannot be a significant amount of pigment depletion in the scotopic range of vision. Figure 15 proves the correctness of this argument. 2.1.6.4. Adapt at i on pools. Since one of the critical issues in this field is the actual site in the retina of the process of light adaptation, a natural question to ask is, how far does adaptation spread in space? We have already argued in the Introduction that light adaptation has to be localized to some extent, for functional reasons. But how localized? Perhaps the photoreceptor is the site of adaptation, or perhaps the retinal ganglion cell, or perhaps one of the retinal interneurons. William Rushton, in his well known Ferrier lecture, summarized his years of work on this problem (Rushton, 1965). He concluded that there must be adaptation pools more proximal in the retina than the photoreceptors, and that within these pools light evoked signals were added up to set the gain of the retina by negative feedback. Rushton used two pieces of psychophysical evidence to support his contention that there were adaptation pools. His arguments from these two bits of evidence are vulnerable to criticism, and yet the idea of adaptation pools has been extremely fruitful and provocative. Rushton' s first bit of evidence for the concept of adaptation pools was the extremely low level of light required to raise the psychophysical threshold by 0.5 log units above the absolute threshold. According to Rushton, the number of light quanta which adapted the retina was only one tenth of the number of rods upon which the background fell, i.e. about 5-10' quanta s -1 over a circular back- ground 10 deg in diameter. Thus, he concluded rod VISUAL ADAPTATION AND signals were reduced in magnitude, even those from rods which had never seen the adapting light. How- ever, as we have stated repeatedly, a decline in psychophysical sensitivity does not necessarily mean a decline in retinal gain but may simply be a rise in the noise level against which the signal must be discriminated. There would be no change in retinal gain in this latter case, even though the psycho- physical sensitivity had declined. When we repeated this experiment of Rushton' s on retinal ganglion cells, and found a significant effect on retinal gain at such low backgrounds that only a fraction of the rods received a photon each second, Rushton' s argument was made much tighter (Enroth-Cugell and Shapley, 1973a). Rushton' s second argument was based on an experiment with gratings. The idea was to present a background grating and then a test grating in phase, and another test grating 180 deg out of phase, with the background grating. All gratings were square waves of 2 c deg -1 spatial frequency. Rushton reported no difference in threshold for the in-phase and out-of-phase condition, and then argued this meant that the background signals were pooled in an adaptation pool which he supposed was larger in diameter than the period of the background grating, 0.5 deg. There is something wrong with the reasoning behind this experiment, and with the actual result, which was not replicated by Barlow and Andrews (1967). The weakness of the argument is interesting because it relates to the degree of localization of adaptive effects, to the concept of spatial channels, and to results on adaptation pools in retinal neurons. Adaptation pools ought to be roughly the same size as the receptive fields of the retinal neurons they adapt. This is consistent with our argument in the Introduction (Section 1) about the function of adaptation in providing brightness constancy when there are gradients of illumination, as was observed by Davidson and Freeman (1965) and Land and McCann (1971). Electrophysiology indicates that the size of an adaptation pool is equal to (Cleland and Enroth-Cugell, 1968; Enroth-Cugell and Shapley, 1973b) or less than (Green e t al . , 1977; Harding, 1977) the size of the receptive field of a retinal ganglion cell it regulates. Therefore, functional reasoning and electrophysiological evidence support the inference that if a retinal cell RETINAL GAIN CONTROLS 289 can resolve a grating pattern, the adaptation pool which controls that cell's adaptation state should also be able to resolve the grating. Now the channel hypothesis implies that the 2 c deg -1 grating that Rushton (1965) used stimulated just those cells whose receptive fields could resolve 2 c deg -1. Daitch and Green' s (1969) data indicate that these cells are towards the high end of the spatial frequency range for rod-driven visual cells in the peripheral retina. We must conclude that if these cells resolved the 2 c deg -1 test pattern their adaptation pools must also have been able to resolve the pattern. Therefore, one ought to have observed that the threshold for a superimposed test pattern should have depended on its position relative to the adapting grating. In-phase with the adapting grating should have been the position of highest threshold. This is just what Barlow and Andrews (1967) found, and the failure of Rushton (1965) to obtain this result probably was a consequence of image blur or light scatter. Barlow and Andrews concluded that their results cast doubt on the existence of adaptation pools. However, by the argument above, their result does not disprove the existence of adaptation pools at all. It just confirms what is now known from other experiments, that the adaptation pools "see", i.e. resolve patterns, as well as the receptive fields they serve. A different kind of experiment on adaptive summation was performed by Westheimer (1965). The test spot was small (0.1 deg) and the adapting field was a circular disk concentric with the stimulus. The adapting disk was constant in luminance but varied in diameter. Under these conditions, threshold increased as a function of adapting disk diameter up to a diameter of about 0.75 deg. In this range, threshold was approxi- mately proportional to the area of the adapting disk. Surprisingly, for disk diameters larger than 0.75 deg the threshold dropped, suggesting that light falling beyond 0.375 deg from the test spot acted to "sensitize" the response to the test. At first the "sensitization" discovered by Westheimer was thought to be due to adaptation produced by a neural element which had a spatially- opponent cent er - surround receptive field. Such a neuron might have a steady state response to the large adapting disk which was much smaller than its steady response to the small disk. 290 I f the gain of the cell were related to this steady state response, t he gai n r educt i on woul d be less f or t he large disk. However , physi ol ogi cal results on s c o t o p i c a d a p t a t i o n do not s u p p o r t t hi s i nt erpret at i on of the psychophysi cal result (Enrot h- Cugell e t al . , 1975; Barl ow and Levi ck, 1976). That is, the gai n of the receptive field center in cat retinal gangl i on cells in t he scot opi c r ange decreases and t hen levels of f as backgr ound di amet er is increased, and is not a decreasing and t hen increasing funct i on as woul d be supposed f r om t he convent i onal i nt er pr et at i on of sensi t i zat i on. Fur t he r mor e , psychophysi cal sensi t i zat i on in t he scot opi c r ange has not been convi nci ngl y demons t r at ed under stabilized i mage condi t i ons ( MacLeod, 1978). Thi s suggests t hat ret i nal or cent ral st i mul at i on by t he movi ng bor der of t he backgr ound disk may be needed f or sensi t i zat i on t o be obser ved ( MacLeod, 1978). Mor eover , t he rel at i on of sensi t i zat i on t o retinal adapt at i on has been quest i oned by Lenni e and MacLeod (1973) who showed t hat the key to sensitization was the u n i f o r mi t y in l umi nance of the 0.75 deg di amet er desensitizing disk and t he out er annul us. I f t he out er annul us were ei t her l ower o r h i g h e r in l umi nance t han t he cent ral disk it woul d reduce sensi t i zat i on. 2.2. The Cone System 2.2.1. SENSITIVITY AS A FUNCTION OF BACKGROUND The cone syst em ( Phot opi c syst em) is like t he r od syst em in yielding We be r ' s Law under some circumstances and t he square r oot law under others. As is evi dent f r om Fig. 8 f r om Wyszecki and Stiles (1967), in huma n vi si on t he cones t ake over at t hr eshol d f r om t he r ods above 0.1 t o 1 td in backgr ound ret i nal i l l umi nat i on. The i ncr ement t hr eshol d curves in t he l i t erat ure t end t o all show a cone pl at eau (cone " d a r k l i ght "; Barl ow, 1958) f r om 0.1 up t o about 10 t d. The phot opi c Weber ' s Law f or a moder at e- si zed t est spot ( di amet er > 0.5 deg) t akes over f r om t he cone " d a r k l i ght " l i mi t ed behavi or at 10 td very rel i abl y. Al t hough t he phot opi c t hr eshol ds f or spot s fol l ow We be r ' s Law up t o t he poi nt where t he pi gment is bl eached away, it has recent l y been shown t hat sat ur at i on can be demons t r at ed in t he R. M. SHAPLEY AND C. ENROTH-CUGELL cone syst em wi t h adapt i ng backgr ounds which are t hemsel ves small in ar ea (Buss e t al . , 1982). Thi s sat ur at i on can begi n at as low a backgr ound as 10 t d, in the low phot opi c range. On a small di amet er (0.25 deg) backgr ound t he t hr eshol d of a smal l spot cl i mbs much mor e st eepl y t han Weber ' s Law ( obt ai ned wi t h a large backgr ound) woul d predi ct . Thi s effect may be obser ved under stabilized i mage condi t i ons ( Tul unay- Keesey and Vassilev, 1974). The phot opi c sat ur at i on is t hought to be due t o sat ur at i on of a cone-dri ven i nt erneuron in the ret i na, one which has a cent er - sur r ound i nt er act i on so t hat a backgr ound of large ar ea will release t he put at i ve i nt er neur on f r om sat ur at i on. Electrophysiological evidence on sensitization in the phot opi c range is present l y scant and equivocal, but t here are some i ndi cat i ons it may exist (see Section 3.7.). 2. 2. 2. CONTRAST SENSITIVITY AND MEAN LEVEL For most t arget s, and in par t i cul ar moder at e- sized spots with a sharp edge on a large backgr ound, We be r ' s Law hol ds f r om 10 td to l 0 s td, i.e. t hr oughout the phot opi c r ange of backgr ounds. This is shown very clearly in the dat a of Whittle and Chal l ands (1969), Fig. 17. The l ower curve for each subj ect is t he i ncr ement t hr eshol d curve; above it are curves of const ant bri ght ness as mat ched t o a test fl ash in t he cont r al at er al eye. It is i nt erest i ng t ha t t he c o n s t a n t b r i g h t n e s s c ur ve s ar e appr oxi mat el y paral l el t o the i ncr ement t hr eshol d curve. Thi s means t hat t he appar ent bri ght ness of t wo di f f er ent i ncr ement s was pr opor t i onal t o t hei r (Weber) cont r ast s. Thus, Whi t t l e and Chal l ands' results i mpl y t hat (Weber) cont r ast r at her t han l umi nance det er mi ned the appar ent bri ght ness. They also poi nt out t hat ot her, cent ral , fact ors may cont r i but e t o appar ent bri ght ness. Under t he condi t i ons of t hei r experi ment s, t he mechani sms which det ermi ned brightness were purel y monocul ar and t her ef or e pr oba bl y retinal - - as evi denced by t he fact t hat st i mul us cont r ast agai nst a fai rl y hi gh backgr ound in one eye was needed t o mat ch a much weaker flash l umi nance agai nst zero backgr ound in t he ot her eye. These dat a of Whi t t l e and Chal l ands suppor t t he mai n hypot heses we st art ed out wi t h about t he f unct i onal si gni fi cance of adapt at i on in est abl i shi ng bri ght ness const ancy cont i ngent on cont rast , which cont rast depends on the reflectances VISUAL ADAPTATION AND RETINAL GAIN CONTROLS 4 3 2 I 0 -I -ao - 2 I . x Log Z _ . L I L i I ! I I I - I 0 I 2 3 4 5 291 4 3 2 ! 0 - I - - dO _ - I t - o , 2 - " ~ / - " ' I i i i i " ; i - I 0 I 2 3 4 $ FIG. 17. Increment threshold and constant brightness curves for two human subjects. A test patch subtending 56' by 28' on a uniform background was presented to one eye, and a comparison patch, on zero background, was presented to the other eye, except for the bottom curve in each panel, which is a monocular increment threshold curve for each subject. For each of the constant brightness curves, the luminance of the comparison patch was held fixed, and the luminances of the background and test patch were varied so that the test matched the comparison in brightness. Therefore, each such curve represents test and comparisons matched at a constant fixed (subjective) brightness. The test and comparison patches were flashed for 200 ms, every 1.5 s at the low backgrounds up to every 8 s at the high background levels. The curves are similar at all brightness levels, having a flat portion and then a sloping portion with slope = 1. At the higher brightnesses, the transition from flat to sloping portions was somewhat broader than at low brightness or at threshold brightness. Each cross ( x ) marks the value of I at the intersection of the flat and sloping portions of the curve above it. Units are in log (photopic) td. From Whittle and Challands (1969). o f o b j e c t s . T h e r e h a s b e e n a l o t o f wo r k o n s p a t i a l v i s i o n o f t h e c o n e s ys t e m, i n p a r t i c u l a r f o v e a l c o n e v i s i o n . T h e i d e a o f s p a t i a l c h a n n e l s is d e r i v e d f r o m t hi s wo r k . T h e o u t s t a n d i n g r e s e a r c h o n t h e e f f e c t o f me a n l u mi n a n c e s o n s p a t i a l c o n t r a s t s e n s i t i v i t y h a s b e e n d o n e b y v a n Ne s a n d B o u ma n ( 1967) . T h e i r wo r k wa s p r e s e n t e d b o t h as c o n t r a s t t h r e s h o l d vs s p a t i a l f r e q u e n c y p a r a me t r i c i n me a n l e ve l s h o wn i n Fi g . 18, a n d a l s o as t h r e s h o l d c o n t r a s t vs me a n l e ve l , p a r a me t r i c i n s pa t i a l f r e q u e n c y , as i n Fi g. 19. Wh a t c a n b e s e e n, e s p e c i a l l y f r o m Fi g. 19, is t h a t t h e t r a n s i t i o n f r o m s q u a r e r o o t l a w b e h a v i o r t o We b e r L a w b e h a v i o r d e p e n d s a l o t o n s p a t i a l f r e q u e n c y . I f we c a l l t h e s p a t i a l f r e q u e n c y k, t h e n t h e i r r e s ul t s i mp l y t h a t t h e t r a n s i t i o n i l l u mi n a t i o n IBT is p r o p o r t i o n a l t o t h e s q u a r e o f t h e s p a t i a l f r e q u e n c y I aT=A "k 2 ( 18) wh e r e A is s i mp l y a p r o p o r t i o n a l i t y c o n s t a n t . Th i s 292 R . M. SHAPLEY AND C. ENROTH- CUGELL I 00 IC 8 0. 0009 0. 009 . / 0.09 9 90 O I I I 0 I 0 0 Spati al frequency in cycles per degree FIG. 18. The dependence of cont r as t sensi t i vi t y on spat i al f r equency at di f f er ent me a n ret i nal i l l umi nat i ons in t he phot opi c r ange. Act ual l y what is pl ot t ed is t hr e s hol d modul a t i on M ( what we woul d t e r m t he t hr e s hol d Rayl ei gh cont r as t ) of a si ne gr at i ng as a f unc t i on o f spat i al f r equency at each of sever al levels o f me a n i l l umi nat i on, i ndi cat ed at t he t op o f t he f i gur e. Li ght f r om a mo n o c h r o ma t o r was t r a ns mi t t e d t h r o u g h a phot ogr a phi c t r a ns pa r e nc y o f a si ne gr at i ng whi ch was i maged by a Maxwel l i an vi ew s ys t em on t he s ubj e c t ' s r et i na. The gr at i ng s ubt e nde d 8. 25 deg by 4. 5 deg a nd was f i xat ed, t her ef or e cent ered on t he fovea. The cont r ast of t he pat t er n was reduced by addi ng a cons t ant level of light f r om anot her monoc hr oma t or . The t hr e s hol d mo d u l a t i o n is t he r eci pr ocal o f wha t we have cal l ed t he (Rayl ei gh) cont r as t sensi t i vi t y. I n t hi s f i gur e me a n i l l umi nat i ons ar e i ndi cat ed in phot opi c t d. The wavel engt h was 525 n m. Co mp a r e t hes e da t a wi t h t hos e of Pa s t e r na k a nd Mer i gan (1981), Fi g. 13. Fr o m va n Nes a nd Bo u ma n (1967). has interesting consequences for model s of phot opi c light adapt at i on and in part i cul ar for a model which uses t he ideas of spat i al f r equency channel s and adapt at i on pool s, pr esent ed l at er on. An i nt erest i ng sidelight on Fig. 18 is t hat it i ndi cat es t hat cont r ast sensitivity cont i nues t o i mpr ove f or grat i ngs of high spat i al f r equency until well i nt o t he phot opi c range, and Fig. 19 conf i r ms t hat fine grat i ngs obey t he squar e r oot law even up to quite high levels of i l l umi nat i on. I f one accept s t he qua nt um f l uct uat i on expl anat i on f or this behavi or , one must concl ude t hat vi si on of fine pat t er ns is qua nt um limited up to qui t e hi gh light levels. As before, we must add the qualification t hat this result does not i ndi cat e defi ni t el y whet her or not t he retinal gai n is af f ect ed at t hese light levels. It is i mpor t ant t hat t he cones " s e e " so much bet t er t han the rods when t he cri t eri on is t hr eshol d cont r ast . Thi s can be seen in Figs. 8 - 1 0 , and especially by compar i ng Fig. 11 f r om Dai t ch and Gr een wi t h Fig. 18 f r om van Nes and Bouman. Ther e' s a j ump of about a f act or of 10 or so f r om the opt i mal rod-dri ven cont rast sensitivity of Dai t ch and Gr een, at about 1 td backgr ound, t o t he best cont r ast sensitivity of van Nes and Bouma n wi t h peak c ont r a s t sensi t i vi t i es a bove 200. Thi s cor r esponds t o a Weber f r act i on (AIT/ IB) of less t han 0.01, in agr eement with t he sensitivities of Bl ackwel l ' s subjects who had an asympt ot i cal l y low Weber f r act i on of 0. 008 (Bl ackwel l , 1946). However , the si t uat i on ma y be compl i cat ed by retinal i nhomogenei t y. Koender i nk e t al . (1978) VISUAL ADAPTATI ON AND RETI NAL GAI N CONTROLS 293 M i n % ' / ' * " J ' " ' " - L ' ' . ' " ' ' " ' ' ' " ' 4 c o ~ * ~o * 4 8 c ml 4 0 . 3 = . ' l - : , , : 1 o ~ I I , . 1 . , I . . I , , 1 , , I . . I 1 10"3 t O t 10"1 1 10 10 2 10 3~ 10 4 Bo i n t el FIG. 19. The dependence o f t hr e s hol d mo d u l a t i o n on r et i nal i l l umi nat i on f or di f f er ent spat i al f r equenci es. Wa ve l e ngt h was 525 n m i n t he s a me e xpe r i me nt s as i n Fi g. 18. B 0 is wha t v a n Nes a n d Bo u ma n cal l ed me a n r et i nal i l l umi nat i on, a n d i t is gi ven i n t d. M is wha t we woul d cal l t hr e s hol d ( Rayl ei gh) c ont r a s t . The s l opi ng por t i ons o f t hes e cur ves ar e t her ef or e r oughl y consi st ent wi t h t he s quar e r oot l aw, a nd t he fi at por t i ons , wher e cont r as t sensi t i vi t y is c ons t a nt as me a n level changes , ar e cons i s t ent wi t h We b e r ' s La w. The t r a ns i t i on bet ween t hes e t wo par t s o f t he cur ve a ppe a r s t o s hi f t t o t he r i ght , t o hi gher i l l umi na t i ons , f or hi gher spat i al f r equenci es . F r o m va n Nes a n d Bo u ma n (1967). measur ed cont r ast sensi t i vi t y in cent r al and peri pheral r et i na (Fig. 12) as a f unct i on of mean level. They f ound much less i mpr ovement in cont r ast sensitivity with t he shift f r om r od t o cone vision in t he per i pher al ret i na. Thi s result suggests t here may be somet hi ng special about foveal cone vision whi ch allows especially high cont r ast sensitivities. 2 . 2 . 3 . TEMPORAL FREQUENCY RESPONSES AND MEAN ILLUMINATION As wi t h spat i al f r e que nc y, t he t e mpor a l f r equency of a stimulus i nfl uences t he dependence of sensitivity on mean level. Thi s has been shown by Kel l y (1972). Fi gur e 20 f r om his wor k demonst rat es the sensitivity f or gratings of di fferent spatial frequenci es over a range of t empor al frequenci es at several mean levels. At low spatial f r equency and l ow t o i nt er medi at e t empor al f r equency, he obt ai ned Weber ' s Law. At high spatial f r equency and low t o i nt ermedi at e t empor al frequenci es he observed t he square r oot law ( i ndi cat ed as D- R f or d e Vr i e s - Ro s e ) . He di scovered t hat , at ver y high t empor al frequenci es and low spatial frequencies, sensitivity was mor e or less i ndependent of mean level. Thi s is t he " l i ne a r " regi on, so-called because t he visual system appears t o be behavi ng in a linear manner in t hat t he sensitivity f or a modul at ed stimulus is not af f ect ed by t he presence of di f f er ent steady levels. Figure 21 also f r om Kelly (1972) illustrates t hese results in t erms of t wo s p a t i o - t e mp o r a l adapt at i on maps, one at a mean i l l umi nat i on of 50 t d and t he ot her at 200 t d. A compl et e t heoret i cal expl anat i on f or all these results of Kelly is not available, but we can i ndi cat e some basic ideas whi ch may account f or t hem. The effect s of spatial f r equency have been discussed above. The t endency t owards Weber Law be ha vi or of l ower t e mpor a l f r equenci es is compar abl e t o Bar l ow' s (1957) results on steeper i ncr ement t hr eshol d curves with l onger dur at i on stimuli. Bot h results suggest t hat t he adapt at i on mechani sms in t he ret i na are somewhat sluggish in t i me course and fail t o be as effect i ve on t he responses t o stimuli which are hi gher in t empor al f r equency t han 8 Hz as t hey are on responses t o stimuli l ower in f r equency. Thi s is consi st ent with Adel son' s (1982) observat i ons about t he onset of light adapt at i on and with el ect rophysi ol ogi cal measurement s (Enrot h-Cugel l and Shapley, 1973a; Bayl or and Hodgki n, 1974; Derri ngt on and Lennie, 1982). The " l i ne a r " range observed by Kelly (1972) is somewhat har der t o under st and. Suppose t he gain of t he r et i na f or high f r equency stimuli is i ndependent of mean level. Still, one woul d expect t hat t he great er quant al fl uct uat i ons at high light levels woul d cause a r educt i on in sensitivity. Thi s leads t o t he i nference t hat t he noise whi ch limits per f or mance at high t empor al frequencies must not 294 | I ' \ ] i o o o " ' D - R ~oo [ ' - 2 cpd l g 7 0 o 2OO <3 500 O.S c:pd 2 [ ] r I I i~ RECTI LI NEAR PATTERNS Subj . TZ I C ) ~ a ; " ~ , ~ I ( d ) 1 0 o o 20 __ ~> 5 o r o o 100 ; 200 8 pd D- R 12 cpd O- R ~ T l , ] [ i J I I r 1 2 5 10 20 50 1 2 5 10 20 50 TEMPORAL FREQUENCY (Hz) FIG. 20. Thr e s hol d i l l umi nat i on as a f unc t i on o f t e mpor a l f r equency at di f f er ent spat i al f r equenci es. The t hr e s hol d i l l umi nat i on AB was t he a mpl i t ude o f si ne gr at i ng whi ch coul d j us t be seen on a me a n i l l umi nat i on o f B t d. I n each o f t he f our panel s o f t he f i gur e, t he f our di f f er ent cur ves ar e f r o m me a s u r e me n t s at t h e f ol l owi ng me a n r e t i na l i l l umi nat i ons : 36 t d ( ar r owheads ) , 114 t d (t ri angl es), 360 t d ( di amonds ) , a nd 1140 t d (circles). The t est t ar get was a pal e- green CRT whi ch subt ended 7 deg; it was viewed monocul ar l y a nd f i xat ed. The gr at i ngs were modul a t e d i n t i me wi t h a s i nus oi dal wa ve f or m; t he t e mpor a l f r equency o f t he mo d u l a t i o n is pl ot t ed on t he hor i zont al axi s. Me a s u r e me n t s f r o m f our spat i al f r equenci es ar e s hown: 0. 5 c deg -1, 2 c deg -1, 8 c deg -~, a nd 12 c deg-L Thr e s hol ds , whi ch ar e separ at ed f r om t hos e at ot her me a n i l l umi nat i ons by a f act or whi ch is equal t o t he r at i o o f t he me a n l evel s, c o n f o r m t o We b e r ' s Law, a nd ha ve a W wr i t t en next t o t he cur ve. Thr e s hol ds whi ch ri se like t he s qua r e r oot o f t he me a n level are labeled D- R for t he de Vries-Rose law, s ynonymous wi t h t he s qua r e r oot l aw. Fr o m Kel l y (1972). be quant al noise and must be i ndependent of mean light level. This is not implausible. Most sources of noise in the visual system, e.g. channel opening and closing in neur onal membr anes, or spont aneous t r ans mi t t er r el ease, s houl d have wi de- band component s. These component s may be relatively larger at high t empor al frequenci es t han at low, compar ed t o t he light evoked neural shot noise R. M. SHAPLEY AND C. ENROTH-CUGELL caused by quant al fl uct uat i ons. One must suppose t hat this high f r equency noise is i mmune to t he a da pt a t i ona l ef f ect s of t he mean level of i l l umi nat i on. Per haps the noise which limits det ect i on of high f r equency responses is post- retinal. In any case, a compl et e and adequat e expl anat i on f or Kelly' s "l i near r ange" requires f ut ur e research. 3. GAI N AND CONTRAST GAI N IN RETI NAL GANGLI ON CELLS At t he out set of this section on t he physi ol ogy of ret i nal adapt at i on we concent r at e on retinal gangl i on cells, t he out put stage of t he retina. All i nf or mat i on which flows f r om t he ret i na t o the brai n about t he visual appear ance of t he out si de world passes along the axons of these ganglion cells. The evidence of ret i nal adapt at i on in t he activity of these neur ons allows us t o establish a link bet ween t he visual, percept ual f unct i on of light adapt at i on and t he underl yi ng retinal mechani sms. We will f ur t her concent r at e our at t ent i on on t wo kinds of retinal ganglion cells in t he cat ' s retina, t he X and Y cells (see Appendi x 2), because most is known about t hem. Compar i son with t he retinas of ot her species and with human vision will be made frequent l y. As in t he I nt r oduct i on, we stress t he i mpor t ance of a hi er ar chy of gai n cont r ol mechani sms at di f f er ent sites in t he ret i na. Fur t her mor e, t he role of ret i nal gain cont rol s in maki ng t he ret i na r espond t o cont r ast will be made evident. One can speak about the g a i n of retinal ganglion cells because t hei r i mpul se rat e vari at i on caused by i ncrement s (or decrement s) of i l l umi nat i on are pr opor t i onal t o the magni t ude of the i ncrement (or decrement ) over a consi derabl e range of response ampl i t ude. Thi s is illustrated by Fig. 22 (Shapl ey and Kapl an, unpubl i shed). The stimuli were fine gratings which stimulated the center of the receptive field. (In this initial discussion we will be dealing with t he gain of t he cent er onl y, but will consi der t he gain of t he sur r ound bel ow. ) The rat i o of t he change in impulse rat e with change in stimulus magni t ude is t he gain, G G = d R / d I . (19a) VISUAL ADAPTATION AND RETINAL GAIN CONTROLS 295 50 20 10 5 2 1 0.5 0.2 >- (J Z O UL. --I <C I-- 'K 50 20 , 0 >- 3 Z 5 IAJ O u. 2 ..J k- ,< 1 0.5 0.2 I 2 5 10 20 50 1 2 5 10 20 50 TEMPORAL FREQUENCY (Hzl TEMPORAL FREQUENCY (Hz) LI NEAR (const) FIG. 21. So-called Spat i o- Temporal Adaptation Maps. The adaptation behavior of a human observer is mapped on the plane defined by spatial frequency and temporal frequency, as vertical and horizontal coordinates. Each of these maps only applies to one mean retinal illumination: 50 td for the one on the left, and 200 td on the right. They indicate the slopes of the curves which relate increment threshold to mean illumination, at the different points in the spatio - temporal frequency plane, de Fries-Rose is equivalent to the square root law, as explained in connection with Fig. 20. Weber means that, in this spatio - temporal region, increase of mean illumination will cause a corresponding and proportional increase in the luminance required to reach threshold: in other words, contrast sensitivity will be constant in the region marked Weber. Li near means that mean illumination does not affect the modulation threshold; in other words, in the Linear region, contrast sensitivity grows in proportion with mean illumination. The shaded areas are transition zones between the other regions. From Kelly (1972). As Cl e l a n d a n d En r o t h - Cu g e l l (1968) s h o we d , t he r e s p o n s e o f t he c e nt e r me c h a n i s m o f g a n g l i o n cel l s a c t ua l l y d e p e n d s on t he l u mi n o u s f l ux f al l i ng on t he c e nt e r o f t he r e c e pt i ve f i el d ( see Se c t i on 3. 5. 1. ) . Th e r e f o r e , t he ga i n is mo r e p r o p e r l y e xpr e s s e d as, G= d R / d F ( 19b) a n d ha s uni t s i mp u l s e s / q u a n t u m ( i / q ) o f l i ght . Th e f l ux F is t he s t i mul us i l l umi na t i on t i mes t he s t i mul us a r e a , or , i f t he s t i mu l u s is l a r ge r t h a n t he r e c e pt i ve f i e l d c e nt e r , s t i mu l u s i l l u mi n a t i o n t i me s t he s u mmi n g a r e a o f t he c e nt e r . Th e l i nk b e t we e n vi s ua l s e ns i t i vi t y a n d r e t i na l ga i n i s ve r y s t r ong. Ba r l o w a n d Le vi c k (1969) pr e s e nt e d t he a r g u me n t t ha t t hr e e f a c t or s d e t e r mi n e t he a b i l i t y o f r e t i n a l g a n g l i o n cel l s t o de t e c t a s t i mul us : (i) t he v a r i a n c e i n t he d i s c h a r g e o f ne r ve i mpul s e s ; (ii) t he t i me c our s e o f t he r e s pons e whi c h d e t e r mi n e s t he o p t i ma l i n t e g r a t i o n t i me ove r whi c h ne r ve i mpul s e s s h o u l d be c o u n t e d b y t he n e r v o u s s ys t em i n or de r t o de t e r mi ne i f a s t i mul us is pr es ent ; a n d (i i i ) t he r e t i n a l ga i n. The ga i n wi l l be di s c us s e d at l e ngt h s u b s e q u e n t l y . Th e f i r s t o f Ba r l o w a n d Le v i c k ' s t hr e e f a c t o r s , t he v a r i a n c e o f t he d i s c h a r g e o f ne r ve i mpul s e s , c o u l d c o n c e i v a b l y be i n f l u e n c e d b y p h o t o r e c e p t o r " n o i s e " , o r b y f l u c t u a t i o n s i n n e u r a l r e s p o n s e due t o f l u c t u a t i o n s i n t he n u mb e r o f q u a n t a a r r i v i n g f r o m t he b a c k g r o u n d , or b y p o s t - r e c e p t o r a l , i n t r a r e t i n a l " n o i s e " , e. g. s y n a p t i c " n o i s e " . The s e c o n d f a c t o r , t he o p t i ma l i n t e g r a t i o n t i me i n Ba r l ow a nd Le v i c k ' s s cheme, is t he d u r a t i o n o f t i me ove r whi c h ner ve i mpul s e s s h o u l d be c o u n t e d t o o b t a i n t he be s t s e p a r a t i o n o f s i gnal f r o m noi s e. I t is gr e a t e r t h a n zer o b e c a u s e t he p h o t o r e c e p t o r s a n d 296 5 0 . . " " ~ 5 0 / ~ 2o g I I I I ( o , o ) t 2 3 4 I I I I b ( ) I O 0 2 0 0 5 0 0 4 0 0 F L U X ( x ] O 4 q u o n t o / s e c ) FIG. 22. The proportionality of response with stimulus flux at different mean levels of illumination in cat retinal ganglion cells. These data were taken from an X on-center retinal ganglion cell in a decerebrate, paralyzed cat. The stimuli were drifting sine grating patterns. The response measure was the amplitude of the response at 4 I-Iz, the temporal frequency of the drift. The spatial frequency was chosen to stimulate the center of the receptive field only; it was 1 c deg -~ (cf. Linsenmeier et al . , 1982, p. 1175 for a discussion of how gratings can pick out the center). The stimulus was produced on a CRT with a white P4 phosphor and subtended 8 deg by 10 deg on the retina. The center of the receptive field had a summing area of 0.02 deg 2, approximately, and the stimulus display was centered on the middle of the receptive field with a mirror. The mean retinal illumination was controlled with neutral density filters interposed between the CRT and the cat. The responses plotted as filled circles were obtained with a mean retinal illumination of approximately 5.10 s quanta(507 nm) (deg 2 s)-'. The other two curves were obtained at ten (open circles), and one thousand (diamonds) times higher mean retinal illuminations. The stimulus contrast ranged from 0.02 up to 0.64, but only part of this range is shown for each background level. The stimulus flux was estimated by multiplying the amplitude of the stimulus retinal illumination by the central summing area of the cell. At the highest mean retinal illumination, the stimulus fluxes would be off the scale for the two lower means, and so the data are plotted on a reduced horizontal scale, as indicated in the figure. The two scales were chosen so that equal horizontal distance is equal contrast for the open circles and diamonds, the results from the experiments with the two higher mean illuminations. The approximate equality of the slopes of the response curves for these two sets of results implies equal contrast gain for these two mean levels. However, the main point of the figure is the linear range of response. Shapley and Kaplan, unpublished results. the fol l owi ng ret i nal stages have a pr ol onged response to each quant um of light (see for exampl e Fig. 55), and t hus i nt egrat i ng over some finite t i me allows one to add up the neural consequences of each qua nt um abs or pt i on. But t he opt i mal R. M. SHAPLEY AND C . E N R O T H - C U G E L L i nt egr at i on t i me is less t han i nf i ni t y because the r et i na' s response to each quant um event ual l y dies away, so there is no poi nt i nt egrat i ng the neural response beyond the t i me at which the signal out wei ghs the noise. Therefore, there is some opt i mum t i me, about 100 ms, over which the nervous system shoul d count nerve i mpul ses in order to detect t hat a st i mul us has been present ed to the gangl i on cell. Barl ow and Levick measured how these three factors depended on backgr ound light i n order to discover whi ch was most i mpor t ant in cont rol l i ng the abi l i t y of gangl i on cells to detect a st i mul us- initiated signal. The mot i vat i on for this analysis was t hat if these t hree factors cont rol l ed gangl i on cell per f or mance t hey shoul d al so cont r i but e to behavi oral sensitivity. They demonst r at ed t hat the gangl i on cell gai n was reduced dr amat i cal l y by backgr ound i l l umi nat i on even t hough the vari ance of the impulse discharge and the opt i mal i nt egrat i on t i me for retinal responses changed rat her little with backgr ound. Thi s is i l l ust rat ed i n Fig. 23. They t herefore f ound compel l i ng evi dence for the pr oposi t i on t hat the mai n l i nk bet ween visual adapt at i on and ret i nal adapt at i on is the cont r ol of ret i nal gai n by steady backgr ound i l l umi nat i on. Fur t her work by Der r i ngt on and Lenni e (1982), on the relative const ancy of the vari abi l i t y of the ma i nt a i ne d di schar ge wi t h me a n level, has st rengt hened this concl usi on. 3.1. Gain Control in the Scotopic Range 3. I. I. GAIN AND BACKGROUND Al l of the X and Y cat gangl i on cells st udi ed so far have received i nput from ret i nal rod and cone pat hways (Daw and Pear l man, 1969). Thi s gives an experi ment er the oppor t uni t y to st udy the gai n cont rol s for the rod and cone pat hways i n an i ndi vi dual cell. Represent at i ve results on the gai n of the receptive field cent er as a f unct i on of backgr ound i n the scotopic range are gi ven i n Fig. 24 from Enr ot h- Cugel l and Shapl ey (1973a). The st i mul i were smal l spots placed in the cent er of the receptive field and modul at ed with a slow square wave t i me course. The gai n [the i / q rat i o equi val ent to the gai n as i n equat i on (19)] is pl ot t ed on doubl e- l ogar i t hmi c coordi nat es vs backgr ound VISUAL ADAPTATION AND RETINAL GAIN CONTROLS 297 9 o / e Log (AT o'A'T) l o g s Ao g - I " 7 / / I I I I I I I I I 0 I 2 5 4 5 6 7 8 l o g q u o n o ( 5 0 7 ) p e r s q . d e g r e e p e r s e c FIG. 23. Three factors limit the reliable detection of light by retinal ganglion cells in the cat, and the dominant factor is the gain. The incremental "threshold" (labeled M 0. A. T, and having units of quanta) of a ganglion cell is plotted here as filled circles against the background retinal illumination. The open circles are the quant um- spike ratios at different backgrounds. The quantum - spike ratio is the reciprocal of what we have called gai n. The empty squares are the estimated optimal averaging time; the empty triangles are the standard deviations of the impulse number distributions. The vertical coordinates were chosen so that changes in log "threshold" are the sum of the changes in the logarithms of the underlying factors. Clearly the quantum - spike ratio, the reciprocal of the gain, is the dominant factor. From Barlow and Levick (1969). i l l umi nat i on. It is seen t hat the gain is const ant for backgrounds below a critical level, and t hat above this level the gai n declines al most inversely with backgr ound. Act ual l y, the typical behavi or of retinal gangl i on cells is described by the following equat i on: GR = GRO / ( 1 +I B/I Ro) P (20) G R is the gain for rod-driven ganglion cell activity. GRO is the dark-adapt ed gain for the rod pat hway. IRO is the i l l umi nat i on at which the gai n has dr oppe d by 2 -P and is r ef er r ed t o as t he "t r ans i t i on i l l umi nat i on". P is the exponent of the t erm in the denomi nat or which depends on the act i on of the gai n cont rol ; P is also the slope of the gain vs background curve on log - log cooordinates. For the cells in the Enrot h-Cugel l and Shapley (1973a) st udy, the average value of P was 0.9. This is very close t o Weber' s Law which woul d have an exponent of 1 instead of 0.9 in the denomi nat or . At mean ret i nal i l l umi nat i ons above 5.108 quanta(507 nm) (deg 2 s)-' the gain of the rod pat hway declines more steeply t han equat i on 20 woul d indicate, because of " r o d s at ur at i on" (Lennie e t al . , 1976). The value of the t ransi t i on i l l umi nat i on IRO, varied over a range of two log units, f r om about 300 quant a( 507 nm) (deg 2 s)-I up t o 3.104 quant a(deg 2 s)-' . The ganglion cells with the largest centers, presumabl y large peripherally located Y cells, had the lowest values of the t ransi t i on i l l umi nat i on. Barlow and Levick (1976) and Barlow (1977) have suggested t hat the transition illumination, IRO, may be anal ogous to the " da r k l i ght " inferred from the pl at eau in the human psychophysi cal increment sensitivity at low backgrounds. There is a similarity, 298 R. M. SHAPLEY AND C. ENROTH-CUGELL 2 . 5 x 1 0 s ~ 2 . 5 x 10 4 E 2"5 x 103 I I I I I i iS I I I I 1 10 2 10 4 10' 0 . 0 ~ 4 0 . ~ 4 0-04 Backgr . r et . i l i um, q(507 nm) / sec, deg 2 o" o v ,,J m 2"5 x 105 2 5 x 10" 2.5 x 102 I I I I I _ I i , 10 = 10 4 10' Backgr. rec. i l i um, q(507 nm) / sec, deg 2 0.0004 0 . 0 0 4 0 . 0 4 o FIG. 24. Gai n vs backgr ound i l l umi nat i on, in t he scot opi c range, f or cat ret i nal gangl i on cell cent ers. Ret i nal fl ux [in uni t s o f quant a(507 nm) s-' ] r equi r ed f or a smal l cent r al st i mul us t o elicit a cr i t er i on r esponse (30 i mpul ses s -~) is pl ot t ed vs t he ret i nal i l l umi nat i on o f a 12 deg concent ri cal l y l ocat ed backgr ound. The gai n in i mpul s e s / qua nt um ( i / q) is i ndi cat ed al so on t he ri ght hand vert i cal scales. The l at t er quant i t y was cal cul at ed by mul t i pl yi ng t he st i mul us ret i nal i l l umi nat i on by t he ar ea o f t he st i mul us, 0.03 deg 2, and by a f act or o f 1/ 3, t he est i mat ed f r act i on o f qua nt a i nci dent on t he ret i na whi ch were abs or bed. I n t he l eft ha nd panel , t he filled circles are f or whi t e st i mul i on a whi t e backgr ound; t he open circles ar e f or b l u e - gr een st i mul i on a r ed backgr ound, t o demons t r at e r od i sol at i on. I n t he ri ght panel all t he poi nt s ar e f or whi t e on whi t e. The resul t s in t he l eft panel ar e f r om an on- cent er gangl i on cell; t he resul t s in t he r i ght panel ar e f r om an of f - cent er cell. Fr om Enr ot h- Cugel l and Shapl ey (1973a). because bot h t he " d a r k l i ght " and t he t ransi t i on i l l umi nat i on are needed t o account f or observed plateaus: the backgr ound i l l umi nat i on must exceed IRO f or t he gain t o dr op f r om its dar k adapt ed val ue in gangl i on cells, while t he backgr ound illumination must exceed I D, the " da r k l i ght ", f or the psychophysi cal sensitivity t o dr op f r om its dar k adapt ed value. However , t he f unct i onal di fference out wei ghs t he appar ent similarity. The t ransi t i on i l l umi nat i on is i nvol ved in gain cont rol ; t he " da r k l i ght " (either estimated f r om psychophysics or f r om physiological experiments) sets the noise level of the ret i na in t he dark. This ar gument is suppor t ed by the est i mat ed values of " d a r k l i ght " and t he t ransi t i on i l l umi nat i on, which are qui t e di fferent . The " dar k l i ght " of cat ganglion cells was estimated by Barl ow et al. (1971), as follows. Based on t he value of t he mai nt ai ned discharge in t he dark, and the slope of t he stimulus - response curve obt ai ned in t he dark, these aut hor s est i mat ed t he magni t ude of t he light flux which woul d have been requi red t o generat e t he mai nt ai ned di scharge in t he dark, and called this val ue t he " d a r k l i ght ". They f ound a wide variation in this estimate of the " dar k l i ght ". However , t aki ng t hei r highest value, t he " d a r k l i ght " was equivalent to about 100 quant a s -1 retinal flux. For a cell with a small cent er, say about 0.1 deg 2 in area, this woul d be pr oduced by 103 quant a(deg 2 s) -t retinal illumination; f or the largest cells it woul d be pr oduced by about 3 q(deg 2 s) -1. These values for the feline " da r k l i ght " are t oo low, by at least a f act or of t en, f or t he " d a r k l i ght " t o be equi val ent t o t he t ransi t i on i l l umi nat i on in cat ret i nal gangl i on cells. Rat her, some cri t eri on amount of voltage or current or subst ance in a retinal cell, much larger t han t hat caused by " d a r k l i ght ", must be exceeded, and t hen t he gain cont rol of adapt at i on begins t o act. As argued in Sect i on 2.1.1.2., " da r k l i ght " pr obabl y limits sensitivity by provi di ng a noise, t he " da r k noi se", against which a signal must be pi cked out , r at her t han by setting t he gain. 3. 1. 2. GAIN AND DYNAMICS Ther e are dynami c consequences of adapt at i on which are hi dden in t he simple pi ct ure of Fig. 24. As shown in Fig. 25, t he t i me course of response of t he recept i ve field cent er varies with adapt at i on level, as f ound bot h by Yoon (1972) and by Enr ot h- Cugell and Shapl ey (1973a). The nat ure of t he change is t hat t he response of t he cent er t o an i ncrement al step of i l l umi nat i on becomes mor e t ransi ent , t he mor e light adapt ed t he cell is in t he VISUAL ADAPTATION AND RETINAL GAIN CONTROLS 299 q Q t/Q r a t i o - J . e - - 1 . 2 5 sec ~ l [ O n r a t i o - - - I O n I I ~ 0 0 2 . 0 0 0 6 1 5 0 i m p u l s e s ] s e c ( a ) 0 .0 8 0-01 I 5 0 im p u ls e s / s e e ( b ) FIG. 25. Ti me cour se o f ( r od- dr i ven) squar e wave r esponses at di f f er ent adapt at i on levels i n cat ret i nal gangl i on cells. (a) The r esponses (averaged over many st i mul us cycles and s moot hed) o f t he recept i ve fi el d cent er o f an on- cent er cell at fi ve di f f er ent levels o f ba c kgr ound ret i nal i l l umi nat i on: (a) l - - 5 0 quant a( deg 2 s) "1, ( a ) 2- - 6. l & quant a( deg a s) - ' , ( a) - - 6" l & quant a( deg 2 s) -1, (a)4---9.104 quant a( deg 2 s) -t , a nd ( a) 5- - 5" l & qua nt a (deg 2 s) -1. (b) A similar result is shown for an off-center cell at three background illuminations: (b)l--100 quanta (deg 2 s) -t, (b)2--3. l& quanta(deg 2 s) -l, (b)3--103 quanta (deg 2 s)-'. Note that in the cat 1 deg2= 0.048 mm a, approximately. For each response the i / q ratio, the gain, is shown at the left of the averaged response histogram. The time courses of the responses and the i / q ratios at the top of each column were the same as those obtained in total dark adaptation. The drop in gain with increase in background illumination goes hand in hand with the change in time course observed in the histograms. From Enroth-Cugell and Shapley (1973a). scotopic range. This finding applies to off-center as well as to on-center cells, and to X as well as Y cells (Jakiela e t al . , 1976). This change in time course of response with adapt at i on is related to another observation: the steepness of the decline in gain with background depends on the temporal pattern of the stimulus (and response). For example, the results in Fig. 24 were derived from measurements of the peak response to an incremental step of illumination on a background. One would obtain rather similar data from measurements of the gain of the response to a sinusoidally modul at ed small spot for temporal frequencies of 2 Hz or less. In this case, either peak- to-peak impulse rate modul at i on or the amplitude of the sinusoid which is the best approxi mat i on to the neural response are two response measures which would give the same dependence on background. Similar measurements at 8 Hz or above have a shallower dependence on background, i.e. the exponent P in equation (20) would be around 0.6 for intermediate temporal frequencies of modul at i on (between 3 and 10 Hz; Enroth- Cugell and Shapley, 1973a; Derrington and Lennie, 1982). The responses to temporal frequencies above 16 Hz suffer almost no attenuation in amplitude with increases in mean level; for such high temporal frequencies, the exponent P in equation (20) is near zero (Shapley e t al . , 1983). Previously, Sakmann and Creutzfeldt (1969) and Barlow and Levick (1976) also observed shallow slopes of gain vs background curves with brief incremental stimuli. The responses to such brief pulses of light, which contain a wide range of temporal frequency components, follow a gain vs background curve which obeys equation (20) with an exponent of about 0.6 on the average. However, as presented earlier, the gain of the peak of the response to a prolonged flash has a steeper dr opof f of gain with background; the exponent in this case is about 0.8, according to Barlow and 300 Levi ck (1976; however, cf. Cl el and and Enr ot h- Cugell, 1970; and Enr ot h- Cugel l and Shapl ey, 1973a in whi ch P = 0.9). The st eeper sl ope f or t he l onger fl ash makes sense because t he pr ol onged flash can be vi ewed as t he sum of pr edomi nant l y l ow f r equency component s , which t end t o be at t enuat ed mor e by i ncreases in backgr ound ( adapt i ng) light. Ther e is a good r eason t o consi der t he effect s of a da pt a t i on on di f f er ent t e mpor a l f r equency component s . As ment i oned in Sect i on 1.2.2., t he response of the ret i na t o st i mul at i on by light shoul d be considered t o be a f unct i onal , or t ransf ormat i on, of t he st i mul us. I f L ( t ) is t he st i mul us, t hen t he r esponse is R = R{ ( t , L( t ) } . That is, t he response R depends on t i me and, f ur t her mor e, t he response at t i me t depends on t he val ue of t he st i mul us at t i me t and also the values of the stimulus in the past. This is a f or mal way of expressi ng what is well known about the retina: it has a finite i nt egrat i on time, and there are sluggish gai n cont rol s, in the recept ors and t he net wor k, which modi f y the r et i na' s r esponse cont i ngent on t he past hi st or y of i l l umi nat i on. Ther e is a st andar d mat hemat i cal appar at us for anal yzi ng f unct i onal s; it is called syst ems anal ysi s. A par t i cul ar l y useful subset of this appar at us is l i near syst ems anal ysi s, a mat hemat i cal t echni que f or anal yzi ng f unct i onal s whi ch are linear, i.e. systems in which the response to t wo separat e inputs is si mpl y t he sum of the responses t o each of the inputs presented alone. In analyzing a linear system, si nusoi ds are t he st i mul us of choi ce because t hey pass t hr ough such a s ys t em unc ha nge d in wavef or m, t hough scaled in ampl i t ude and shi ft ed in phase. Whi l e t he ret i na is deci dedl y not l i near under all condi t i ons, it may behave like a l i near syst em ar ound an oper at i ng poi nt set by t he mean level of i l l umi nat i on, and t her ef or e the retinal responses to si nusoi ds of di f f er ent t empor al f r equency serve t o pr ovi de a good quant i t at i ve descri pt i on of how t he retinal f unct i onal behaves at di f f er ent mean levels of i l l umi nat i on. The di f f er ent effect s of backgr ound on t he responses of gangl i on cells t o di f f er ent t empor al frequenci es can be expl ai ned by a t heor y in which the centerpiece is a gai n cont rol which is a nonl i near negat i ve f eedback ( Enr ot h- Cugel l and Shapl ey, 1973a). In or der t o account f or t he i ncreased t ransi ence of squar e wave responses in t he light R. M. SHAPLEY AND C. ENROTH-CUGELL adapt ed st at e (Fig. 25), one must assume in this t heor y t hat t he gai n cont r ol has a s omewhat mor e prol onged i nt egrat i on t i me t han the phot orecept ors. Accor di ng t o this view, t he decay seen in t he gangl i on cel l ' s st ep r esponse is due to t he i ncrement al increase in the val ue of the gai n cont rol signal due t o t he st ep st i mul us. 3.1.3. ADAPXAXION or THE SURROUNO MECHANISM The sur r ound of t he recept i ve field also adapt s to backgr ound light in the scotopic range. Figure 26 o" g E __= E 10 5 10 4 10 3 Su r r o u n d X Z ( . ~ e ~ / ~ ' C e n t r e 1 ;I ] I I I 0 10 3 10 4 10 "~ 10 6 10 5 10 4 10 ~ - - o - - - / ~ - - - 0 - - / ; sur 7 Ce n t r e j = j I 4 k , , I I I | 0 10 3 10 4 qO ~ 10 6 B a c k g r o u n d i l l u m i n a t i o n ( q u a n t . a / d e g 2 s e c ) FIG. 26. Comparison of center and surround gains in cat retinal ganglion cells. Plotted vertically is the stimulus retinal illumination required to evoke criterion responses as a function of the background retinal illumination plotted on the horizontal axis. Upper and lower panels are from two different on-center X ceils. For the centre responses (filled symbols) the criterion was ten extra impulses evoked during a one second presentation of an increment of the illumination of an opt i mum spot which just filled the center. For the surrounds (empty symbols) the criterion was determined from st i mul us- r esponse curves for opt i mum spots and diffuse illumination. The criterion illumination for the surround was such that the response to diffuse illumination of this amount was ten impulses less than the response to an opt i mum spot of the same retinal illumination. Retinal illuminations are given as quanta (507 nm) (deg 2 s)-' incident on the retina, under the assumption that 75% of the quant a incident on the cornea reach the retina. Rod isolation was insured by the use of bl ue- green stimuli on red backgrounds. From Enroth- Cugell and Lennie (1975). VISUAL ADAPTATION AND RETINAL GAIN CONTROLS from Enroth-Cugell and Lennie (1975) reveals the kinds of surround adaptation which are seen. In these experiments, the surround response was estimated by subtracting the center response from that obtained by illumination of both center and surround. The data are the incremental illumination required to elicit a criterion surround response; thus these curves may be interpreted as indications of changes in surround gain with background illumination. In the upper panel, the surround and center begin to reduce their gains together at roughly the same level of background. In the lower panel the surround starts off with a lower dark- adapted gain, but reduces its gain with background at a higher level than the center. The result is that in both cases the center and surround have roughly the same gains (integrated over their respective total summing areas) in the Weber-Law, light adapted scotopic range. But there are differences across the population of retinal ganglion cells in the relative gain of center and surround in the low scotopic range (see also Kaplan et al . , 1979). The results of the well-known investigation of Barlow et al. (1957) are sometimes interpreted to mean that the gain of the surround goes to zero in total dark adaptation. This, however, is not precisely what Barlow et al. found. Their results on the dependence of gain on area at different backgrounds implied that the ratio of the gain of the center to the gain of the surround increased in total dark adaptation. However, the extent of the increase in the cent er - sur r ound ratio was not determined in their study. It was later shown by Enroth-Cugell and Lennie (1975) and by Kaplan et al. (1979) that this ratio may increase from a value of 1.2 in the light adapted state to as much as 3 in the dark adapted state. That is, the surround is relatively weaker compared to the center in the dark, but it is not gone. In fact, the gain of the surround is always maximal when the retina is dark adapted (see Fig. 26). In other words, as the background level is increased from total darkness, the center gain usually is reduced at a lower level of background than is the surround' s gain. There is a methodological reason for the differences in conclusions about the strength of the surround in the dark. Barlow et al. (1957) used auditory threshold for an "off-response" to measure the magnitude of the response of the 301 surround in on-center cells. This method has the disadvantage that, in these cells, the surround response consists of sustained inhibition when the surround mechanism is dark adapted (Enroth- Cugell and Lennie, 1975). The presence and magnitude of the "off-response" is associated with the adaptation level, just as the magnitude of the center's transient overshoot depends on the center being somewhat light-adapted (see Fig. 25). Enroth- Cugell and Lennie (1975) and Kaplan et al. (1979) measured magnitude of inhibition as an indicator of surround response strength, and they did it with objective averaging techniques. Their work reveals that there are marked variations across the population of ganglion cells in the degree to which background illumination affects the ratio of the center and surround gains (see Fig. 26). The finding of variability in the degree of cent er-surround balance in the dark has also been reported by Barlow and Levick (1976). The fact that background light in the scotopic range can affect the gains of center and surround differently implies that, in this range, the gains of these receptive field mechanisms are controlled at a site (or sites) in the retina more proximal than the photoreceptors. If the only site of gain reduction were the photoreceptors, one would observe that the gains of center and surround would begin to drop at the same background level. There is good evidence that at higher backgrounds some of the reduction in gain in the mammalian retina is due to photoreceptor adaptation (Sakmann and Filion, 1972; Valeton and van Norren, 1983). These observations are consistent with our assertion in the Introduction that there is a hierarchy of gain controls. The ratio of the total integrated gains of center and surround in the light-adapted state is approximately 1.2 for a large population of retinal ganglion cells (Linsenmeier et al . , 1982). There is considerable variance in this ratio among the ganglion cell population. In any one cell, the ratio is approximately constant from 10-2-5cd m -2 back- ground luminance on up (Enroth-Cugell and Lennie, 1975). Thus, in cat ganglion cells the balance between center and surround is established in the low- to mid- scotopic range and is invariant with adaptation level throughout the mid- to high scotopic and photopic ranges. 302 3.2. Gain Control in the Photopic Range The i nput s f r om cones t o cat gangl i on cells can be isolated f r om t hose f r om rods by means of Stiles' t wo-col or t echni que (see Wyszecki and Stiles, 1967, p. 572). When this is done, it is f ound t hat cone signals have a const ant gai n over t he r ange t hat t he gai n of r od signals dr ops by a f act or of one t hous and or mor e ( Enr ot h- Cugel l e t a l . , 1977a), f r om t ot al dar k adapt at i on t o t he high scot opi c range. Thi s can be seen in Fig. 27. Pl ot t ed t here are the t hreshol d illuminations for criterion cone-dri ven and r od- dr i ven responses t o be elicited by a test stimulus on a bl ue backgr ound, as a funct i on of the level of backgr ound. For a single gangl i on cell one obt ai ns a t wo- br anched gai n vs backgr ound curve whi ch is r emi ni s cent of t he t wo- br a nc he d, psychophysi cal sensitivity vs backgr ound curves (Fig. 8). This indicates t hat the separat e gai n cont rol of r od and cone signals is achi eved by t he ret i na pr i or t o t he gangl i on cells, and t hat r od and cone signals are kept segregat ed at least up t o t he poi nt s R. M. SHAPLEY AND C. ENROTH-CUGELL at which t he gains are set in t he paral l el " r o d " and " c o n e " pat hways t hr ough the retina. These results are f ound for receptive field centers of bot h on- and of f - cent er cells, and f or bot h X and Y cells. Thi s has the added i mpl i cat i on t hat r od and cone signals are segregat ed in bot h t he X and Y pat hways until the gai n is set. The results in Fig. 27 i mpl y i ndependence of adapt at i on mechani sms for the r od and cone pat hways t o the gangl i on cells in the cat. However , Nel son (1977) f ound t hat r od signals are coupl ed i nt o cones. Hi s wor k led hi m to the concl usi on t hat t he mai n pat hway of r od signals t o hor i zont al cells was t hr ough t he cones. Pr es umabl y a si mi l ar concl usi on woul d appl y t o t he bi pol ar cells. That is, r od signals shoul d t ravel t hr ough cone bi pol ar cells because of t he large amount of coupl i ng of rod signals i nt o cones. Thi s poses a pr obl em, namel y how can the cone and r od signals adapt separ at el y when t hey are carri ed by t he same i nt er neur ons? One possi bl e expl anat i on is t hat all adapt at i on may t ake pl ace in r ecept or s, but t hat expl anat i on has 1 0 - ' ~ L _ ~s - o 8 A E C o o 6 O . 4 On-centre X ~ , , " / L - - . / , , l I I I I I I I I --o0 3 5 7 9 11 L o g background (photons (500 nm) deg -2 sec -1) _ ---- 11 ? " V E c 9 e 7 ~ FIG. 27. Gain for the rod and cone systems in a cat retinal ganglion cell. The illumination required to evoke a criterion response is graphed vs the background illumination. The points plotted as filled symbols were obtained with a blue - green stimulus light on a bl ue - green background, and were taken to represent the gain of the rod pathway. Their values a r e given by their heights on the left vertical axis, in equivalent quanta of a monochromatic 507 nm light. The empty symbols are the criterion illuminations for a red spot of light on the same blue - green background, and are interpreted as indicating the cone pathway' s gain. The response criterion was just-audible synchrony of the cell's firing rate with the 4 Hz square wave modulation of the small stimulus spot (0.2 deg diameter), which was located in the middle of the receptive field. In this figure, retinal illuminations are referred to quanta at the cornea, before losses in the eye. From Enroth-Cugell e t al . (1977a). VISUAL ADAPTATION AND RETINAL GAIN CONTROLS ot her pr obl ems wi t h t he spat i al i nt er act i ons whi ch are i nvol ved in adapt at i on. At present we can onl y rai se this issue as one whi ch mus t be resol ved by f ut ur e r esear ch. The gai n in t he phot opi c r ange is pr opor t i onal t o 1 / I B, on t he a ve r a ge , a b o v e a cr i t i cal ba c kgr ound i l l umi nat i on/ co ( Daw and Pear l man, 1969; Enr ot h- Cugel l e t al . , 1977a). Thus a descri pt i on of t he gai n f or t he cone syst em in Fig. 27 is (with Gc0 t he dar k adapt ed gai n of t he cone si gnal s, /co t he i l l umi nat i on at whi ch t he gai n is hal ved): 303 respect t o t hei r dar k adapt ed val ues ( Jaki el a e t al . , 1976; Enrot h-Cugel l e t al. 1977a). I f the t i me course of t he st ep r esponses are a resul t of t he act i on of ret i nal gai n cont r ol s, t hese resul t s i ndi cat e t hat t he gai n cont r ol mechani s m f or X cells is not t he same as f or Y cells. I t suggests t hat t he X cel l ' s gai n is cont r ol l ed by a mechani s m wi t h a l onger t i me const ant t han t he gai n cont r ol f or t he Y cell. Such a concept of t wo gai n cont r ol s, one f or X and one f or Y, arises also in t he consi der at i on of t he spat i al pr oper t i es of ret i nal gai n cont r ol s, and will be di scussed bel ow. G c = Gc0/(1 +I a/ I c o) (21) and t he gai n of t he gangl i on cell woul d be G = G c + G R (22) i f we assume t hat r od and cone signals are si mpl y added ( Enr ot h- Cugel l e t al . , 1977b). The t i me cour se of t he gangl i on cel l ' s r esponse under goes anot her change at t he r od - cone br eak. Thi s is i l l ust rat ed in Fig. 28 f r om Enr ot h- Cugel l e t al. (1977a). When t he backgr ound put s t he cell i nt o t he me s opi c r ange, but is less t ha n /co, t he response t o a small i ncrement of i l l umi nat i on which onl y st i mul at es t he cones is sust ai ned, like t he r esponse t o a st i mul us whi ch onl y st i mul at es t he r ods in t he dar k adapt ed st at e. As t he backgr ound is i ncr eased a bove / c o, t he st ep r esponse becomes mor e t ransi ent , recapi t ul at i ng the r od results. Thus, t he r esponse t o a st ep of light in a cat gangl i on cell is sust ai ned in t ot al dar k adapt at i on, becomes mor e t r ansi ent t hr oughout t he scot opi c range, becomes sustained agai n j ust above the r od - cone t ransi t i on, t hen becomes mor e t r ansi ent agai n in t he mi d- t o hi gh- phot opi c range. The r emar ks above about t he expl anat i on of t he effects of adapt at i on on response t i me cour se in t he scot opi c r ange shoul d al so appl y t o t he phot opi c range. One shoul d expect t hat a nonl i near f eedback is t he mechani s m whi ch links gai n r educt i on wi t h t he change in r esponse t i me course. Bot h in t he scot opi c and phot opi c ranges, t he i ncr ement al st ep r esponses of Y cells decay at a f ast er rat e t han X cells when bot h cell t ypes are in t he Weber Law regi ons of t he gai n vs. backgr ound curve and have suf f er ed t he same dr op in gai n wi t h 3 . 3 . Co n t r a s t Ga i n We have pr esent ed t he ar gument earlier t hat one pur pose, per haps t he mos t i mpor t a nt pur pose, of light adapt at i on is t o maxi mi ze t he vi sual cont r ast sensitivity and to keep it const ant as the backgr ound or mean level vari es. One of t he det er mi nant s of cont r ast sensi t i vi t y is cont r ast gai n, or how l arge a r esponse is pr oduced by a gi ven a mount of cont r ast . The cont r ast gai n Gcon is t he backgr ound i l l umi nat i on I B t i mes t he Gai n G as defi ned in equat i on (19a), so Gcnn = ]B" (GR + Gc) (23a) Gon = GR0"IB/ ( 1 + I a/I Ro) P + Gco" I B/ (1 + I s / Ico). (23b) I f t he gai n is expressed as i mpul s e s / qua nt um as in equat i on (19b), t hen t he cont r ast gai n is equal t o ga i n. ba c kgr ound fl ux. I n ei t her case, t he cont r ast gai n has uni t s of [i mpul ses s- ' ] [ cont r ast ] -1, or ma y somet i mes be expressed as [i mpul ses s- ' ] [ per cent cont r ast ] -1 i f cont r ast is gi ven in per cent age r at her t han as a f r act i on. A gr aph of t he cont r ast gai n in single gangl i on cells is shown in Fig. 29 (Shapl ey e t al . , 1983). The cont r ast gai n i ncreases st eadi l y in t he scot opi c range, t hen levels of f (or somet i mes may even dr op s omewhat ) as t he cell ent ers t he phot opi c range. X and Y cells have a si mi l ar dependence of cont r ast gai n on mean level. The dependence of cont r ast gai n on mean level is rel at ed t o t he dependence of gai n on mean level. I n t he r ange of backgr ounds in whi ch I a is much 304 R. M. SHAPLEY AND C. ENROTH-CUGELL i E c o 0 1 3 - - 11 On-centre X / I 5 0 i m p u l s e s s e c - 1 I , I , I , I , I , I , I , I , I 5 7 9 11 13 Log b a c k g r o u n d ( p h o t o n s (560 nm) deg - I sec -~) FIG. 28. Change of time course and gain in cone-driven X ganglion cell responses as a function of increasing background illumination. The curve plots the retinal illumination (referred to the cornea) required to evoke a criterion peak response of 35 impulses s -~ above the mean impulse rate. Stimuli were 0.2 diameter red spots located at the midpoint of the receptive field center, and were modulated in a square wave manner at 0.5 Hz. The background was blue - green, in order to suppress the rods and yield an isolated cone-driven response. From Enroth-Cugell et al. (1977a). l arger t han t he t ransi t i on i l l umi nat i on IRO, and the response is dri ven by rods, t he slope of the cont r ast gai n curve on l o g - log coor di nat es is 1 - P , when t he slope of t he gai n curve is - P. Thus, Weber ' s Law, when the slope of the gain curve is - 1, implies a sl ope of t he cont r ast gai n curve of zero. When P is great er t han 1, in t he regi on of r od sat ur at i on, t he slope of t he cont r ast gai n curve becomes negat i ve, and t he cont r ast gai n act ual l y dr ops. 3. 4. Effect of Adaptati on on the Size of the Receptive Field Center Ther e is evi dence t hat t he size of the recept i ve field center in cat retinal gangl i on cells is pract i cal l y const ant over wi de ranges of mean level or backgr ound level of i l l umi nat i on. Smal l er and larger spots which fall compl et el y within the central- summi ng ar ea of a recept i ve field have al most i dent i cal gai n vs backgr ound curves, as i mpl i ed in Fig. 30 (Cl el and and Enr ot h- Cugel l , 1968). I n t he fi gure, gai n vs ar ea was measur ed f or a single ga ngl i on cell at sever al di f f e r e nt s c ot opi c backgrounds. The parallelism of the curves suggests t hat all these spot s of di f f er ent sizes were af f ect ed t o t he same ext ent by the i ncrease in backgr ound illumination. Cleland and Enrot h-Cugel l (1968) also showed t hat t he di st r i but i on of l umi nous flux a mong several spot s pr oduced exact l y t he same response in magni t ude and t i me course as t he same l umi nous flux concent r at ed in a single spot , as l ong as all stimulus spots were pl aced at equally sensitive poi nt s in t he recept i ve field cent er. Thi s led t o t he concept of a single cent er - mechani sm or cent ral s umma t i on pool wi t hi n which neural signals are added; the evidence of Fig. 30 suggests t hat , at least under some exper i ment al condi t i ons, t he recept i ve field cent er adapt s as a uni t at a site in t he ret i na at which t he cent er' s signals have been pool ed. Thi s finding applies t o the receptive field in the phot opi c as well as t he scot opi c r ange ( Enr ot h- Cugel l e t al . , 1977a). However , t here are some dat a i ndi cat i ng some var i at i on of recept i ve field cent er size wi t h mean level. All t hese results have been obt ai ned f r om exper i ment s which used si nusoi dal grat i ng stimuli t o est i mat e t he size of t he center. The first result VI SUAL ADAPTATI ON AND RETI NAL GAI N CONTROLS 305 o 5 o 4 -% ~3 E 2 z c9 1 z 8 O o I - 5 2 A I I I I - 2 -1 0 1 LOG MEAN LUMINANCE ( c d / mZ ) FIG. 29. Cont r ast gain in cat retinal gangl i on cells as a funct i on of mean luminance. The stimuli were drifting sine gratings, presented at (Rayleigh) cont rast s f r om 0.02 (2%) up to 0.64 (64%). The t emporal frequency of t he dri ft was 4 Hz. Mean stimulus l umi nance was varied over a four to five log unit range with neutral density f'dters. The CRT screen had a white P4 phosphor. The phot opi c l umi nance of t he screen was measured; t he scotopic equivalent l umi nance is approxi mat el y twice t he phot opi c for t hi s light. Artificial pupils, 3 mm in diameter, were used. With such a pupil, 1 cd m -2 produces a retinal i l l umi nat i on of approxi mat el y 4.106 quanta(507 nm) (deg 2 s) -' on t he cat ' s retina. The slope of t he linear por t i on of t he response - cont rast curve provided an estimate of t he cont rast gain in impulses s -1 [contrast]-L These are t he results from t hree different on-cent re X cells; off-cent er and Y cells gave essentially similar results. The responses at t he highest mean l umi nance used, 100 cd m -s, were cone-driven, but responses at lower mean levels were rod-driven. Fr om Shapley et al. (1983). i s t h a t o f E n r o t h - C u g e l l a n d R o b s o n ( 1 9 6 6 ) wh o f o u n d a t wo - f o l d r e d u c t i o n i n t h e r e c e p t i v e f i e l d c e n t e r d i a me t e r o f a n X c e l l wh e n t h e me a n l e ve l wa s v a r i e d f r o m 5" 10 .4 t o 16 c d m -2, a p p a r e n t l y mo s t o f t h e c h a n g e t a k i n g p l a c e a t t h e l o w e n d o f t h e r a n g e o f me a n l e ve l s , a s s e e n i n Fi g . 31. S i mi l a r r e s u l t s h a v e r e c e n t l y b e e n o b t a i n e d b y De r r i n g t o n a n d L e n n i e ( 1 9 8 2 ) wh o r e p o r t a 32070 r e d u c t i o n i n c e n t e r d i a me t e r wh e n t h e a v e r a g e i l l u mi n a t i o n wa s i n c r e a s e d f r o m 2 . 1 0 -3 t o 2 0 0 c d m- L S i n c e t h e s e me a n l e v e l s s p a n t h e r o d - c o n e t r a n s i t i o n i n t h e c a t wh i c h o c c u r s a r o u n d 1 0 - 100 c d m -2 i n wh i t e l i g h t ( wh e n a 3 mm p u p i l d i a me t e r i s u s e d ; c f . E n r o t h - Cu g e l l et al . , 1 9 7 7 a ) , o n e wo u l d wi s h t o h a v e mo r e d e t a i l e d k n o wl e d g e a b o u t h o w t h e c e n t e r s i ze v a r i e d wi t h a v e r a g e l i g h t l e ve l b e f o r e f o r mu l a t i n g a n e x p l a n a t i o n o f t h e c e n t e r ' s c o n t r a c t i o n i n t h e l i g h t . Ho we v e r , t h e r e s u l t s o f E n r o t h - C u g e l l et al. ( 1 9 7 7 b ) o n t h e r e l a t i v e s i z e s o f t h e r e c e p t i v e f i e l d c e n t e r 7 6 ~ i . e | .~_ - ~ 1,e ~ p ~ 2 2"6 1 0 t t I I I I I 0.1 0. 2 0-3 O . S 0 . 7 1 2 3 Bt i mu] us c ~ me t e r , degrees FIG. 30. Ar ea- s ens i t i vi t y curves at several backgr ound luminances in cat ganglion cells. In this figure t he label on t he ordi nat e, "sensi t i vi t y", is used t o mean what we have called " ga i n" . The diameter of a circular spot centered on t he receptive field was varied in steps and t he l umi nance required to produce a criterion response was measured at each value of t he diameter. The t emporal modul at i on was a 4 Hz si newave. The cr i t er i on r esponse was j us t - audi bl e synchroni zat i on of t he cell' s discharge with t he stimulus modul at i on. The sloping and flat port i ons of these curves intersect at a diameter denot ed D, which is t aken t o be t he di amet er of t he central summing area of t he receptive field cent er. In t hese experi ment s t he val ue of D, was approximately constant across background level. Zero on t he "Rel at i ve sensitivity" scale corresponds t o a stimulus amplitude modul at i on of approximately 109 quanta(507 nm) (deg 2 s)-' retinal illumination. The stimulus dept h of modulation was 0.6. The background was a circular spot with a di amet er of 8.5 deg. The retinal i l l umi nat i ons of t he backgr ound are given in log td, but these are " cat t d " , and 1 " c a t t d " is equi val ent t o appr oxi mat el y 6. 105 quanta(507 nm) (deg 2 s)-' on t he retina. Thus t he lowest backgr ound illumination used in these experiments was approxi mat el y 3.6. los quant a(deg 2 s) -1 on t he retina. The l umi nance required, t hr ough t he 4 mm di am artificial pupil used, t o produce this lowest backgr ound i l l umi nat i on was 5.10-4 cd m 2. The brightest background was l 0 s brighter. This was an on-center cell, presumabl y a Y-cell. Fr om Cleland and Enrot h-Cugel l (1968). wh e n i t i s d r i v e n b y r o d s o r d r i v e n b y c o n e s s u g g e s t t h a t t h e t r a n s i t i o n f r o m r o d s t o c o n e s d o e s n o t p r o d u c e t h e c e n t e r ' s c o n t r a c t i o n . R a t h e r , i t a p p e a r s mo r e l i k e l y t h a t t h e c e n t e r s i ze i s r e l a t i v e l y c o n s t a n t f r o m mi d - s c o t o p i c t o mi d - p h o t o p i c l e ve l s , a n d o n l y b e c o me s l a r g e r i n t h e l o w s c o t o p i c r a n g e . T h i s i n f e r e n c e s h o u l d b e t e s t e d f u r t h e r s i n c e i t ma y b e 306 100 10 I I I I 1 1 i I I I I I I I I 1111111 0 0 1 0+1 1 S p a t i a l f r e q u e n c y ( c / d e g ) R. M. SHAPLEY AND C. ENROTH- CUGELL o f r e t i na l a d a p t i v e s i gnal s has be e n i nf e r r e d f r o m p s y c h o p h y s i c a l e x p e r i me n t s . F u r t h e r mo r e , que s t i ons a b o u t t he i nf l ue nc e o f b a c k g r o u n d l i ght at one pl a c e i n t he vi s ual f i el d on t he r e s p o n s e t o a t es t l i ght at a n o t h e r pl a c e ha ve be e n r a i s e d i n t he or i e s o f vi s i on by ( a mo n g ot he r s ) He r i n g (1920), He l s o n (1964), Spe r l i ng (1970), a n d Gr o s s b e r g (1981). A we a l t h o f p h y s i o l o g i c a l r es ul t s s u p p o r t s t he c o n c e p t o f a d a p t a t i o n a l p o o l i n g , b u t i ndi c a t e s t h a t t he p o o l s a r e s ma l l e r t h a n mo s t t he or i s t s ha ve i expect ed. Fu r t h e r mo r e , t her e is evi dence f or pool i ng o f a d a p t i v e s i gnal s ove r t he e nt i r e c e nt e r o f t he r e c e pt i ve f i el d, a n d al s o mo r e l oc a l i z e d a d a p t i v e p o o l i n g i n s u b - r e g i o n s o f t he r e c e pt i ve f i el d c e nt e r a n d s u r r o u n d . Ho we v e r , i n t he cat , s t e a d y i l l u mi n a t i o n o f t he s u r r o u n d has l i t t l e or no ef f ect on t he ga i n o f t he cent er . I n t he cat r et i na, t he gai n o f t he cent er me c ha ni s m o f t he r e c e pt i ve f i el d o f a r e t i na l g a n g l i o n cel l is d e t e r mi n e d by t he s um o f al l t he s t e a d y l i ght f al l i ng on t he cent er , a n d onl y on t he cent er . Thi s has been p r o v e n by a n u mb e r o f di f f e r e nt e xpe r i me nt s whi ch a r e c ons i s t e nt wi t h each ot he r . Th e f i r st was t he e x p e r i me n t o f Cl e l a n d a n d En r o t h - Cu g e l l (1968), t he r es ul t s o f whi c h a r e s h o wn i n Fi g. 32. I n t hi s e x p e r i me n t , t he signal summat i on area was me a s u r e d wi t h s t i mul us di s ks o f v a r i o u s a r e a s . I l l u mi n a t i o n was a d j u s t e d t o gi ve a c o n s t a n t cr i t er i on r es pons e. Ri c c o ' s La w hel d a p p r o x i ma t e l y f or di s k a r e a s l ess t h a n t he s i gnal s u mma t i o n a r e a , i . e . I . A- - k R s , whe r e kRs is t he c o n s t a n t f or Ri c c o s i gnal s u mma t i o n . F o r a r e a s l a r ge r t h a n t he s i gnal s u mma t i o n a r e a , Ri c c o ' s La w no l o n g e r hel d a n d a c o n s t a n t i l l u mi n a t i o n was r e q u i r e d t o el i ci t a cr i t er i on r es pons e. The a r e a over whi ch Ri c c o' s La w hel d was e q u a t e d wi t h t he c e n t e r ' s a r e a o f s i gnal s u mma t i o n . F o r t he s a me cel l t he adaptive summat i on area was d e t e r mi n e d , a g a i n wi t h a c o n s t a n t r e s p o n s e c r i t e r i on. I n t hi s case, a r e a a n d i l l u mi n a t i o n o f a n a d a p t i n g di s k wer e v a r i e d r e c i p r o c a l l y i n o r d e r t h a t t he g a n g l i o n cel l wo u l d p r o d u c e a c o n s t a n t r e s pons e t o t he t est s pot . The t est s pot was c o n s t a n t i n a r e a a n d i l l u mi n a t i o n ; it was p l a c e d i n t he cent er o f t he r e c e pt i ve f i el d. Ad a p t i v e s u mma t i o n f ol l owe d Ri c c o ' s La w al so f or di s ks wi t h a r e a s l ess t h a n t he a d a p t i v e s u mma t i o n ar ea. Tha t is, f or a d a p t a t i o n , I . A = kRA, wher e kRA is t he c o n s t a n t f or Ri c c o a d a p t i v e s u mma t i o n . Fo r l a r ge r di s ks , s u mma t i o n o f a d a p t i v e s e ns i t i vi t y FIG. 31. The effect of mean luminance on the dependence of a ganglion cell's contrast gain on spatial frequency. Four mean luminance levels were used: 16 (empty circles), 0.5 (f'dled triangles), 1.6" 10 -2 (filled circles), and 5" 10 -4 cd m -2 (empty triangles). The highest mean luminance corresponds to approximately 108 quanta(507 nm) (deg 2 s) -1 retinal illumination. What we have called "contrast gain" is labeled "contrast sensitivity" in the figure. A 3.5 mm diameter artificial pupil was used. The criterion response was audible impulse rate modulation at the drift rate of the grating, 4 Hz. The data were fit with smooth curves calculated from a Difference of Gaussians model, where the smaller Gaussian represents the receptive field center. The estimated diameter of the center changed by about a factor of two over the range of mean luminances studied. The cell studied here was an on-center X cell. From Enroth-Cugell and Robson (1966). a wa y t o f or ge a s t r o n g e r l i nk b e t we e n t he mi c r o c i r c u i t r y o f t he r e t i na a n d i t s f unc t i on. I n a n y case, t he d a t a on t he r e l a t i ve l y s ma l l c ha nge s o f r e c e pt i ve f i el d c e nt e r si ze wi t h b a c k g r o u n d or me a n l evel ser ve t o r e i n f o r c e even mo r e s t r o n g l y t he c o n c l u s i o n t h a t s i gnal s f r o m d i f f e r e n t pa r t s o f t he r e c e pt i ve f i el d c e nt e r a d a p t t o g e t h e r a n d wi t h a p p r o x i ma t e l y t he s a me s l ope o n t he ga i n vs b a c k g r o u n d cur ve. 3. 5. Adaptati onal Pool i ng and Receptive Field Size 3.5.1. SIGNAL POOLS AND ADAPTATION POOLS I n v e s t i g a t i o n o f t he s p a t i a l s u mma t i o n o f d e s e n s i t i z a t i o n by a d a p t i n g l i ght s is i mp o r t a n t f or an u n d e r s t a n d i n g o f t he f unc t i ons a n d me c h a n i s ms o f l i ght a d a p t a t i o n . As we di s c us s e d i n Se c t i ons 1. 2. 1. a n d 2. 1. 6. i n c o n n e c t i o n wi t h t he i de a s o f Wh i t t l e a n d Ch a l l a n d s , L a n d a n d Mc Ca n n , Ru s h t o n , a n d We s t h e i me r , t he s p a t i a l s u mma t i o n VISUAL ADAPTATION AND RETINAL GAIN CONTROLS 307 3 i f i ' : f b r j 0.1 0 2 0 3 0 5 0.7 I 2 3 0 0 I 0 2 03 0 5 0 7 I 2 3 Di a me t e r o f a d a p t i n g s p o t , d e g r e e s St i muLus d i a me t e r , degrees F[6. 32. Compar i son of adaptive (a) and signal (b) summat i on areas of an on-center cat ganglion cell. (a) The hori zont al axis gives t he log of t he di amet er of unmodul at ed adapting spots. The vertical axis gives t he log of t he reciprocal of t he adapt i ng spot i l l umi nat i on in relative units. The adapt i ng i l l umi nat i on was set so t hat t he cell produced a criterion response t o a small (0.13 deg) centrally located spot of fixed luminance, which was sinusoidally modulated (4 Hz, 0.6 contrast). (b) The horizontal axis gives the log of the diameter of the stimulus, which was again a spot modulated at 4 Hz, sinusoidally, at 0.6 contrast. But in this experiment the stimulus diameter was varied. The vertical axis in (10) gives the log of the relative gain for the spots of different size; it is also the log of the reciprocal of the illumination required to elicit a constant response. In 0a), the spots were presented on a steady background of 6.10 ~ quanta(deg a s)-'. The criterion response in both (a) and (b) was just audible synchronization of the cell's discharge with the stimulus modulation. From Cleland and Enroth-Cugell (1968). st opped. The i mpor t a nt resul t of this exper i ment is t hat t he signal s umma t i on ar ea and t he adapt i ve s umma t i on ar ea were t he same. Pr oba bl y mos t i f not all of t hese exper i ment s were on Y ret i nal gangl i on cells, but l at er wor k i ndi cat es si mi l ar results hol d f or X cells ( Har di ng, 1977). The results are t he s ame f or on- and off-cel l s. Very si mi l ar resul t s have been obt ai ned wi t h t hi s exper i ment al design on rat opt i c t ract fibers by Green et al. (1977) and Tong and Gr een (1977). Resul t s of this ki nd have also been obt ai ned in l ower vert ebrat es: in frog gangl i on cells (Reut er, 1969; Bur khar dt and Ber nt son, 1972), and in gol dfi sh gangl i on cells (Schel l art and Spekrei j se, 1972). Rel at ed exper i ment s by Shapl ey et al. (1972) and by Enrot h-Cugel l and Shapl ey (1973b) indicate t hat , as t he ar ea of an adapt i ng spot of f i xed l umi nance is i ncreased, t he gai n declines. Thi s is i l l ust rat ed in Fig. 33. The (fixed) l umi nance of t he adapt i ng spot was chosen so t hat f or t he smal l est adapt i ng spot the gai n had not been reduced f r om its dar k adapt ed ma xi mum. As t he ar ea of t he adapt i ng spot was i ncreased, gai n declined and t he response t o t he test st i mul us became mor e t r ansi ent . Resul t s si mi l ar t o t hese were obt ai ned by Schel l art and Spekrei j se (1972) in t he gol df i sh ret i na. They f ound t hat t he t empor al i mpul se r esponse was speeded up and made mor e di phasi c by i ncreasi ng t he ar ea of a backgr ound spot of const ant l umi nance. Anot he r exper i ment whi ch demons t r at es t he si gni fi cance of t he ar ea as well as t he l umi nance of t he backgr ound is i l l ust rat ed in Fig. 34. The fi gure shows t he act ual responses of a gangl i on cell t o t ur ni ng on adapting spot s of qui t e di f f er ent ar ea and l umi nances. The l umi nance of t he l arger one had been adj ust ed unt i l it pr oduced t he same gai n r educt i on as t he smal l er adapt i ng spot , as i ndi cat ed by the response t o t he super i mposed br i ef test flash. Equal adapt i ve effect is associ at ed wi t h equal responses of t he gangl i on cell t o t he " a d a p t i n g " spot s. The concl usi on f r om all t hese exper i ment s is t hat t he gai n of t he gangl i on cell cent er mechani s m is set by t he sum of all t he st eady st at e i nput t o t he cent er. Thi s in t ur n i mpl i es t hat t he adapt i ve effect of a backgr ound on t he cent er mechani s m of a gangl i on cell is, under t he condi t i ons of these experi ment s, det er mi ned by t he total effective f l ux: t he sum of all t he light per uni t t i me which falls on t he center f r om t he backgr ound, wei ght ed by t he spat i al "sensi t i vi t y pr of i l e " (Cl el and and Enr ot h- Cugel l , 1968; Enr ot h- Cugel l and Shapl ey, 1973b). The meani ng of these results in a psychophysi cal cont ext is t hat t he s umma t i on pool of a gangl i on cell is t he same size as t he adapt at i on pool . The meani ng in a neur oanat omi cal cont ext is t hat what ever i nt er neur on det er mi nes t he size of t he recept i ve field cent er is al so i mpl i cat ed in t he gai n cont r ol process. I n t he light o f t hese results on t he dependence of 308 R, M. SHAPLEY AND C. ENROTH- CUGELL Area a d a p t a t i o n f~eLd Gain { i mpul se/ quant um) (a) ~ i ' ~ 014 degrees 2 _ _ @ o2 {b) 0 O05 I 5 0 imp s p FIG. 33. Re duc t i on in gai n wi t h i ncr ease i n ar ea o f a n a da pt i ng s pot of c ons t a nt r et i nal i l l umi nat i on. Thr ee aver aged r es pons es were el i ci t ed by a smal l (0. 18 deg) cent r al l y l ocat ed s pot ( a, b, c) . I n each r un, t he a r e a o f a n a da pt i ng di sk o f c ons t a nt r et i nal i l l u mi n a t i o n - - 6 - 10 a quant a( 507 nm) (deg ~ s) -1 was var i ed; t he ar ea in deg 2 is gi ven in t he f i gur e. Ga i n was cal cul at ed by di vi di ng r es pons e in i mpul s es s " by t he r et i nal s t i mul us f l ux in q u a n t a s- ' . Thi s exper i ment was done on a n on- cent er cat gangl i on cell, pr oba bl y a Y cell. Fr o m Shapl ey et a/ . (1972). i m p u l s e s / s e c 0 impuh;es/sec ~ - Dt ~- I ! 0 impulses/see I 50 i m p u l s e s / s e e FIG. 34. Ga i n r educt i on wi t h t wo a da pt i ng s pot s of di f f er ent ar ea but t he s a me effect i ve f l ux. The t est s t i mul us was a 50 ms pul s e o f l i ght , 0.1 deg i n di amet er , 2. 107 qua nt a ( de g 2 s) -t r et i nal i l l umi nat i on. (a) Th e condi t i oni ng s t i mul us ( hat ched) was 0.1 deg s pot s upe r i mpos e d on t he t est s pot , modul a t e d i n a s quar e wave ma n n e r at 0. 4 Hz. Th e i l l umi nat i on i n t he l i ght pha s e was 3" 107 qua nt a ( de g 2 s) -t . (b) The condi t i oni ng s t i mul us was a 1.57 deg di amet er di sk, concent r i c wi t h t he t est s pot . It s i l l umi nat i on was 2. 5" 104 qua nt a ( de g 2 s) -1, a n d al so 0. 4 Hz. (c) Tes t s pot was pr es ent ed wi t hout a ny condi t i oni ng s t i mul us . The t wo condi t i oni ng st i mul i mu s t have ha d t he s a me ef f ect i ve f l ux, by def i ni t i on, becaus e t he y pr oduc e d i dent i cal r es pons es . Thes e r esul t s ar e f r o m a n on- cent er gangl i on cell, pr e s uma bl y a Y cell. Fr o m Enr ot h- Cuge l l a n d Shapl ey (1973b). VISUAL ADAPTATION AND gai n cont r ol on t he pr oduct of i l l umi nat i on and area, we must revise equat i ons (20) and (21) whi ch expressed t he gai n in t erms of i l l umi nat i on. The earlier equat i ons were cor r ect under t he condi t i ons of full field i l l umi nat i on when t he ent i re recept i ve field woul d be covered by t he uni f or m backgr ound. In this case t he gain woul d j ust be scaled down by a f act or equal t o t he area of t he cent er of t he recept i ve field. But a mor e general equat i on can be written which applies t o backgr ounds of any spatial conf i gur at i on. For t he r od pat hway: GR = GR0/(1 + FB/ FRo) p (24) and a similar equat i on describes the gain of the cone pat hway, Go Fl ux is i l l umi nat i on mul t i pl i ed by area, in this case t he t ot al summi ng area of t he cent er of the recept i ve field. For step responses, t he exponent P has t he val ue 0.9 f or t he r od pat hway and is somewhat higher f or the cone pat hway. What equat i on (24) means is t hat t he gain depends on backgr ound fl ux, F B, not retinal i l l umi nat i on, flux added up over t he entire cent er of t he receptive field and wei ght ed by t he di st ri but i on of sensitivity of t he recept i ve field cent er. Al so, t he adapt i ng flux has t o exceed a critical value, denot ed FRO f or t he scot opi c system, in or der f or t he cell t o under go t he transition f r om dark adapt at i on t o light adapt at i on. It may help t o concei ve of this critical fl ux as t he backgr ound flux requi red t o pr oduce a critical level of D. C. neural signal which, when exceeded, t urns on t he retinal gai n cont r ol . The val ue of t he critical fl ux, FRO, is about 104 quant a s -1 at t he r et i na on t he average ( Enr ot h- Cugel l and Shapl ey, 1973b). 3.5.2. LOCAL AND GLOBAL GAIN CONTROLS WHICH DEPEND ON FLUX Fur t her investigation of t he spatial summat i on of adapt i on in cat gangl i on cells was per f or med by Har di ng (1977). One of his maj or results is illustrated in Fig. 35. The experi ment was designed t o measure t he spatial weighting of adapt at i on with a t wo- spot par adi gm: one test and one adapt i ng spot . The test spot was fixed in posi t i on in t he mi ddl e of t he recept i ve field. Then t he posi t i on of t he adapt i ng spot was vari ed and its l umi nance adj ust ed so t hat t he gain of t he response t o t he test was reduced by a criterion amount . In the same cells t he sensitivity pr of i l e f or eliciting a response was RETINAL GAIN CONTROLS 309 also measured. This latter profi l e is called the signal sensitivity profi l e. As can be seen f r om t he results in Fig. 35, t he signal sensitivity pr of i l e and t he adapt at i on profi l e were appr oxi mat el y t he same in X cells. In some Y cells t he gain r educt i on f or a test spot near t he adapt i ng spot was great er t han f or a test spot f ar t her f r om t he adapt i ng spot , by mor e t han woul d be predi ct ed f r om t he Y cell' s signal sensitivity profi l e. That is, t her e were i ndi cat i ons of local adapt at i on in t he Y cell cent er. A similar sort of ef f ect was seen in an i nvest i gat i on of light adapt at i on of t he receptive field sur r ound of Y cells (Cleland et al . , 1973). There is thus evidence for t wo di fferent gain controls in the Y cell center: one local, one mor e global. Ther e is also evi dence f or local adapt at i on in some X cells ( Har di ng, 1978). A discussion of t he i mpl i cat i ons of these experi ment s f or X/ Y recept i ve field or gani zat i on woul d car r y us t oo far f r om t he cent ral issues of this paper. But a br i ef comment about these di f f er ent adapt at i on profiles in X and Y cells may pr ovoke some t hought a bout t he r et i nal mi c r oc i r c ui t r y under l yi ng adapt at i on. It is known t hat at any ret i nal locus t he receptive field cent ers of X cells are about t en times smaller in area t han t hose of Y cells ( Hochst ei n and Shapl ey, 1976b; Cl el and e t al . , 1979; So and Shapley, 1979; Linsenmeier e t al . , 1982). There exist subuni t s of t he Y cell' s recept i ve field which are r oughl y t he same size as X cell cent ers at t he same retinal eccentricity (Hochst ei n and Shapl ey, 1976b; So and Shapl ey, 1979; cf. Appendi x 2). It has been suggested t hat X cell centers and Y cell subunits are appr oxi mat el y det er mi ned by t he spatial summi ng areas of bi pol ar cells ( Hochst ei n and Shapl ey, 1976b; Vi ct or and Shapl ey, 1979). While recent neuroanat omi cal investigation of the retina suggests this is only an approxi mat i on to the actual situation, it seems now t o be an appr oxi mat i on r at her t han mere specul at i on (see Sterling, 1983). Thus, i f one accepts our previ ous assert i on t hat t he i nt er neur on which determines t he size of the X cell' s center must set t he gain of t he center, t hen t he bi pol ar cells must cont r ol t he gai n of t he X cent er, in some way. I f these same bi pol ar cells feed i nt o t he Y cells' subuni t s, one must suppose t hey cont r ol t he gain of t he subuni t s. The local effect s of gain r educt i on in Y cells seen in Fig. 35 coul d be expl ai ned by this " b i p o l a r " gain cont r ol , whi ch we i nfer t o be 310 R. M. SHAPLEY AND C. ENROTH-CUGELL X-cells Y-cells Y-cells , o x I 2 9 2 : ; " 1 9 / 2 } d '~ 18/1 ~/ ~ , _3' ' 0 2O/ 2 3 / 2 4 / 4 "/ 3 1 / 6 - 3 ' 6 ' 3 d i s t a n c e on r e t i n a (deg.) j d , ~ . 2 3 / 4 ' ~ 31/3 5' :~) 29/5 v, , ~ 3 2 / 3 0 - 3 0 3 FIG. 35. Adaptive and signal gain profiles from four X and eight Y cells. The stimulus used to obtain each profile was a spot 0.2 deg in diameter, modulated by a 2 Hz squarewave. The response criterion was 3 extra impulses over 100 ms at the beginning of each response. Signal gain: the circles indicate the relative gain at each position with respect to the peak gain at position zero, the middle of the receptive field center. The vertical scale is logarithmic and a vertical calibration for 10x is given in the figure. Adaptive effectiveness: The horizontal position of the triangles indicates the location within the receptive field center of a steady adapting spot of 0.2 deg diameter. The modulated test stimulus was fixed in the middle of the receptive field center. The test stimulus was set to be 3x brighter than required to reach criterion. Then the luminance of the adapting spot was adjusted to reduce the response to the criterion level, at each position of the adapting spot. The log, of the ratio of the adapting luminance at position zero divided by the adapting luminance at each of the other positions tested, is given by the vertical position of the triangles. Signal and adaptive profiles have been superimposed at position zero to allow comparison of their respective spread. From Harding (1977). c o mmo n t o X cel l cent er s a nd Y cel l s ubuni t s . The r e mu s t al s o be a s e c o n d ga i n c o n t r o l f or t he mu c h l a r ge r Y cel l cent er ; a l i ke l y c a n d i d a t e is one o f t he l a r ge f i el d a ma c r i n e cel l s whi c h s h o u l d ha ve a d e n d r i t i c s p r e a d r o u g h l y t e n t i me s l a r ge r i n a r e a t h a n t he d e n d r i t i c s p r e a d o f an X g a n g l i o n cel l . Loc a l a d a p t a t i o n ef f ect s l i ke t hos e i n Fi g. 35 have been seen p r e v i o u s l y i n g a n g l i o n cel l s f r o m o t h e r speci es: r a t ( Gr e e n et al . , 1977; To n g a n d Gr e e n , 1977), f r o g ( Bu r k h a r d t a n d Be r n t s o n , 1972), gol df i s h ( Eas t er , 1968). The f unc t i ona l a na l ogi e s o f t hes e cel l s wi t h cat X or Y cel l s is obs c ur e , b u t i n e ve r y cas e t he cel l s s t u d i e d b a d r e l a t i ve l y l a r ge r e c e pt i ve f i el ds c o mp a r e d t o t he si ze o f d e n d r i t i c s p r e a d s o f b i p o l a r cel l s. Mo r e o v e r , e vi de nc e f or a s e c o n d ga i n c o n t r o l wi t h a l ar ge s u mmi n g a r e a c o mp a r a b l e t o t he cent er size has been f o u n d i n each o f t he s e r e t i na s : i n r a t ( Gr e e n et al . , 1977), i n f r og ( Re ut e r , 1969; Bu r k h a r d t a n d Be r nt s on, 1972), a nd i n g o l d f i s h ( Sc he l l a r t a n d Spe kr e i j s e , 1972). Thus , l ocal a d a p t a t i o n ef f ect s i n each o f t hese cases mi ght be due t o l oc a l i z e d " b i p o l a r " ga i n c o n t r o l s whi c h a r e p r e s e n t i n a d d i t i o n t o a s e c ond, " a ma c r i n e " , ga i n c o n t r o l whi c h s ums l i ght e v o k e d ne ur a l s i gnal s ove r a l a r ge r a r e a . Thi s s h o u l d be c l e a r l y l a be l l e d as a c ha i n o f i nf e r e nc e r a t he r t h a n e s t a bl i s he d f act . Yet i t is i nt er es t i ng be c a us e i t ser ves t o r e i nf or c e t he i de a t ha t t he r e is a h i e r a r c h y o f ga i n c o n t r o l s , a " b i p o l a r " ga i n c o n t r o l a n d an " a ma c r i n e " ga i n c ont r ol as wel l as p h o t o r e c e p t o r gai n c ont r ol s whi ch wi l l be di s c us s e d be l ow. 3.5.3. OPTICAL AND NEURAL FACTORS IN ADAPTATIONAL POOLING One mus t c ons i de r t he p r o p o s i t i o n t ha t al l r et i nal ga i n c o n t r o l by l i ght is a p u r e l y l oc a l , r e c e p t o r a l p h e n o me n o n . I t ma y s eem an a b s u r d p r o p o s i t i o n i n t he f ace o f al l t he e vi de nc e ci t ed a b o v e a b o u t t he a d a p t i v e s u mmi n g a r e a be i ng e qua l i n si ze t o t he r e c e pt i ve f i el d c e nt e r , b u t i t de s e r ve s s e r i ous VISUAL ADAPTATION AND consi der at i on because of t he pr obl em of light scat t eri ng wi t hi n t he eye and of opt i cal aber r at i ons in t he lens and cor nea. I f t he measur ed size of t he recept i ve field cent er and adapt i ve summi ng area were great l y i nfl uenced by such i mperfect i ons in physiological optics, one woul d have to discount the concl usi ons about t he site or sites of gain cont rol (s) whi ch were based on t he cor r el at i on of t he spat i al ext ent s of t hese t wo di f f er ent mechani sms. However , direct measurement s of t he physiological optics of t he cat ' s eye (Bonds, 1974; Robs on and Enr ot h- Cugel l , 1978) i ndi cat e t hat , except f or rare cells with t he smallest recept i ve fields, t he opt i cal effect on the measured size of the receptive field and adapt i ve summi ng area is small. It is i nt erest i ng t o not e t ha t t he exi s t ence o f t he pos t ul a t e d " amacr i ne" gain cont rol of t he Y cells is not subject t o any doubt s based on opt i cal bl ur or scat t er, because t he adapt i ve summi ng ar ea and recept i ve field cent er of Y cells are so large. The presence of nei ghbor i ng X cells with fields t en t i mes smal l er in ar ea serves as a cont r ol on t he opt i cal cont r i but i on t o t he size of t he Y cells' fields. However , t here coul d be some quest i on about whet her t he size of t he recept i ve fields of X cells, whi ch can be qui t e small in area, might be due t o optical blur or scatter. It becomes a quant i t at i ve quest i on in t he case of t he X cells. However , even f or X cells, t he opt i cal bl ur seems t o be less t han t he neural summat i on ar ea ( Robson and Enr ot h- Cugel l , 1978) when t he physi ol ogi cal opt i cs are opt i mi zed wi t h best r ef r act i on and a small artificial pupil. Thr oughout this paper , ar gument s based on recept i ve fields' sizes have onl y cited as evi dence t he results of experi ment s in whi ch t he physi ol ogi cal optics were opt i mi zed. That some of t he ret i nal gai n - - setting mechani sms must i nvol ve pool i ng of signals f r om many recept ors is t he concl usi on of a physiological ext ensi on of Rus ht on' s reasoni ng about t he low l evel of b a c k g r o u n d s whi ch pr oduc e l i ght adapt at i on (Enrot h-Cugel l and Shapley, 1973a). As di s cus s ed a bove , Ru s h t o n f o u n d t ha t t he psychophysi cal scot opi c t hr eshol d was raised by a fact or of t wo f r om its dark adapt ed value when onl y one r od in a hundr ed act ual l y caught a quant um of light. However , as st at ed bef or e, this result implies not hi ng about gai n cont r ol in t he ret i na. It coul d be expl ai ned in t erms of an i ncrease in PRR3-I RETINAL GAIN CONTROLS 311 " noi s e " f r om t he backgr ound. Appl yi ng Rusht on' s reasoni ng t o ret i nal gain, we measur ed how much backgr ound light is requi red t o r educe t he gain of a gangl i on cell by a f act or of t wo. In several cells with large receptive fields, t he requi red backgr ound light yields one quant um absor bed per second per hundr ed rods. The i nt egrat i on t i me of feline r ods is al most cert ai nl y shor t er t han a second; i ndeed, ot her wor k indicates a maxi mum i nt egr at i on t i me of 0.1 s. Thus, one r od in a t housand per integration time receiving a quant um of background light is enough t o reduce t he gain by hal f a log unit. Thi s is solid suppor t f or Rusht on' s concl usi on t hat signals f r om a pool of rods must set t he gain f or signals f r om rods which have not themselves received light quant a f r om t he backgr ound. In ceils with smal l er recept i ve fields, t he l umi nance requi red t o reduce gai n was hi gher, so t hat in t he worst case appr oxi mat el y one r od in t en received a quant um of light per i nt egr at i on time. We now presume t hat these ganglion cells with small fields were X cells, on t he basis of t he receptive field cent er size. So agai n t he ver y l ow level of backgr ound l umi nance whi ch is requi red t o reduce t he gain may suggest t hat phot or ecept or signals must be pool ed to set the gain in the cat retina. Not e t ha t l i ght s c a t t e r woul d not a f f e c t t hes e measur ement s because t hey were made wi t h large uni f or m backgr ounds whi ch were at least one t housand times larger in area t han t he opt i cal poi nt spread f unct i on at hal f hei ght . Fur t her mor e, t he f or ce of these ar gument s is not af f ect ed by t he fact t hat not all phot or ecept or s which lie within t he spread of t he dendritic tree of a ganglion cell proj ect t o t hat gangl i on cell. The statistical r andomness of t he quant um cat ch makes t he f r act i on of recept ors hit by quant a t he same whet her one consi ders t he ent i re popul at i on of phot or ecept or s, or onl y t hat popul at i on which projects t o the ganglion cell under st udy, as l ong as t he backgr ound is t rul y uni f or m. That t here are di f f er ent t r ansi t i on levels, f r om dar k t o light adapt ed, f or scot opi c recept i ve field cent er and sur r ound mechani sms (Fig. 26) also suppor t s t he idea of a gain cont r ol pr oxi mal t o t he phot orecept ors. The same rods must drive the center and sur r ound. I f onl y t he rods adapt ed, t he cent er and sur r ound woul d have t o lose t hei r gain in parallel. Since this is not observed, we must concl ude t hat gain must be cont r ol l ed at a site in 312 t he ret i na af t er cent er and s ur r ound have been segregated. The neur oanat omy of the cat retina tells us t hat all post - r ecept or al i nt er neur ons pool t he act i vi t y of ma ny phot or ecept or s. Ther ef or e, t he p o s t - r e c e p t o r a l ga i n c o n t r o l mu s t p o o l phot or ecept or signals. R. M. S H A P L E Y A N D C . E N R O T H - C U G E L L 3.6. Gain Control and Receptive Field Size Across the Popul ati on of Gangl i on Cells Since l i g h t - e v o k e d neural signals are s ummed over t he recept i ve field cent er t o set t he gai n of the cent er of a par t i cul ar gangl i on cell, one nat ur al l y woul d guess t hat gangl i on cells wi t h cent ers of di f f er ent sizes woul d be light adapt ed t o di f f er ent extents by a large uni f or m backgr ound. The results on the spatial s ummat i on of adapt i ve effect suggest t hat gai n depends on the total, st eady state, effective flux, i.e. i l l umi nat i on mul t i pl i ed by ar ea wei ght ed by the cent er' s gain per unit area. The t ot al effective fl ux falling on the cent er of a cell wi t h a small recept i ve field will be less t han t he flux falling on a l arge recept i ve field center; t he gai n shoul d be reduced less in t he smal l recept i ve field cent er. The initial test of this i dea is of f er ed in Fig. 36, f r om Enr ot h- Cugel l and Shapl ey (1973b). What is pl ot t ed in Fig. 36 is t he t r ansi t i on i l l umi nat i on at the knee of t he curve rel at i ng gai n and backgr ound i l l umi nat i on. The t ransi t i on i l l umi nat i on is defi ned here empi r i cal l y as t he i l l umi nat i on at which t he gai n has dr opped by a f act or of t wo f r om t he dar k adapt ed gain (Not e the slight di fference between this def i ni t i on and t he mor e r i gor ous defi ni t i on of t ransi t i on i l l umi nat i on in connect i on with equat i on (20)). Because t he dat a in Fig. 36 were col l ect ed f r om ma ny di f f er ent cats, possi bl y in di f f er ent physi ol ogi cal states, the t ransi t i on i l l umi nat i on was mul t i pl i ed by t he dar k adapt ed gai n t o obt ai n a cor r ect ed t r ansi t i on i l l umi nat i on. Thi s cor r ect ed t r ansi t i on i l l umi nat i on is pl ot t ed agai nst cent er summi ng area, det ermi ned f r om an area - t hreshol d curve (Cl el and and Enr ot h- Cugel l , 1968). I t can be seen in Fig. 36 t hat the cells with l arger centers have a l ower effect i ve t r ansi t i on level, and t hat t he t ransi t i on level is appr oxi mat el y pr opor t i onal to t he cent er s ummi ng ar ea. These dat a mi ght be compat i bl e wi t h ot her f unct i ons of cent er size besides area, because of the large vari ance. The cells U E c E c c o c 10 10 ~ 1 0 2 ] l I I I I I o U ~ o o O o o o o o o o o o ~ o o o ~o o o I I I I 1 1 I 025 05 10 25 50 10 25 Total summing area of centr resp mech (deg ~) FIG. 36. Transition level as a function of center area for a population of cat retinal ganglion cells. The value of the transition level from the horizontal to the sloping portions of the gain vs background illumination curves (as in Fig. 24) is plotted against center summing area. Empty circles are from on-center cells, while filled circles are from off-center cells. The cells were not grouped into X and Y classes. From Enroth-Cugell and Shapley (1973b). in this st udy were not classified as X or Y. Ther ef or e, a quest i on unr esol ved by these results is whet her X and Y cells have t he same dependence of effect i ve t r ansi t i on level on cent er area. Fur t her evi dence on t he dependence of gai n setting on recept i ve field cent er size across t he pop- ul at i on of gangl i on cells comes f r om t he concor dant studies of Fi scher and Ma y (1970) and Cl el and et al. (1973). The results of bot h studies i mpl i ed t hat t he cent er ' s gai n, def i ned as G = d R / d F where F is l umi nous fl ux (i l l umi nat i on t i mes area), is inversely pr opor t i onal t o t he cent er' s summi ng area when t he gangl i on cell is well light adapt ed in t he mi d- scot opi c t o mesopi c r ange by large uni f or m backgr ounds . Thi s is consi st ent wi t h t he result of Enr ot h- Cugel l and Shapl ey (1973b) on t he " ef f ec- t i ve t r ansi t i on l evel " and wi t h t he appr oxi mat el y i nverse r el at i onshi p bet ween gai n and backgr ound above the t ransi t i on level, equat i on (21). Thus, three studies seem consistent in suppor t i ng t he hypot hesi s t hat gangl i on cells with larger centers are mor e light- adapt ed t han t hose wi t h smal l er cent ers under t he same fixed uni f or m backgr ound conditions, because of spat i al s umma t i on of adapt i ng signals. RETINAL GAIN CONTROLS 313 VISUAL ADAPTATION AND Recently, Linsenmeier et al . (1982) have taken a fresh look at this question by measuring the gain of X and Y ganglion cells in the cat in response to drifting gratings on high scotopic or mesopic backgrounds. They also gauged the size of the center of each ganglion cell by fitting the observed dependence of contrast gain on spatial frequency with a "Di fference of Gaussi ans" model. The spatial sensitivity profiles of receptive field center and surround are approximated by Gaussian functions in this model. The spatial spread of the center' s Gaussian is a measure of the effective radius of the center' s distribution of sensitivity (or, more precisely, gain). Figure 37 is their graph of the peak gain of the center plotted vs the center' s effective radius, for a large population of cat retinal ganglion cells. The figure demonstrates that the cells with the largest centers had the lowest peak gain, and that the gain was approximately the inverse of the center' s r adi us . While this result is qualitatively like the earlier results of Enroth-Cugell and Shapley, Fischer and May, and Cleland et al . , it is quantitatively different in that gain in the light adapted state is inverse to the radius and not the area of the center. However, there is quite a lot of variance of gain across the population of ganglion cells, so much so that Linsenmeier et al . ' s results do not conclusively disprove the a r e a - ga i n relation. Furthermore, since their measurements were made at backgrounds which might be in the high scotopic or in the mesopic range, the precise value of the slope of the gain vs area line might be influenced by the degree to which rods or cones are the predomi nant phot orecept or input for cells of different sizes. These qualifying remarks suggest that the book is not closed on the dependence of gain on area of the receptive field center. As suggested below, the hypothesis of Enroth-Cugell and Shapley (1973b), that gain varies inversely with center ar ea in the light-adapted state, is useful in rat i onal i zi ng psychophysi cal results on the dependence of sensitivity vs background curves on target size. The interpretation of these area effects in adapt at i on must be modified by the discovery that there is not a wide variation in receptive field center size among ganglion cells of one type, X or Y, at a given retinal locus (Cleland et al . , 1979; So and Shapley, 1979). The coefficient of variation of the 1 0 0 0 ~ - O 7 o -~ 3 0 0 o O ~ , , ' , Z~ A Z ~
I n 1 0 0 ~ ~,,v~ f... o 30 2~ 0 ix a. 1 0 z~ O, 0 , 3 1 3 C e n T e r - r - e d i u s ( d e g ) FiG. 37. Relation between t he center's size and its peak gain in the middle of the receptive field. The dependence of (Rayleigh) contrast gain on spatial frequency was determined for X and Y cells by adjustment of contrast to reach a constant response criterion. The experimental curves were fit with a Difference of Gaussians model as in Fig. 31, from which both the center's radius and its gain at the peak of its sensitivity profile could be determined. These two values are plotted against each other to show that, at the same mean luminance, cells with larger centers have lower gain. The empty symbols are for on-center cells, the filled symbols are for off-center cells. Circles denote X cells; triangles denote Y cells. The pupillary area was 16 mmL The mean luminance was around 14 cd m-L From Linsenmeier et al. (1982). center-diameter distribution at any one retinal locus is at most 0.25 (So and Shapley, 1979) and is probabl y less in an individual animal. There is a marked increase in receptive field center-diameter at retinal loci away from the ar ea cent r al i s ; the di amet er of t he cent er is appr oxi mat el y proport i onal to the distance from ar ea cent r al i s . This is true for both X and Y cells, and for bot h on- and off-center cells. On- and off-cells have approximately the same size at any one locus on the retina. As stated above, Y cells have an approximately ten times larger area than X cells at each locus. The combination of these facts with the preceding results on the effects of area on adaptation leads to the following conclusions. First, cells with larger receptive fields in the periphery of the retina ought to be more light-adapted than central ganglion cells with smaller centers, under condi t i ons of uni form const ant backgr ound illumination. Second, Y ganglion cells ought to be more light-adapted than X ganglion cells at the same retinal locus. By the degree of light-adaptation we mean the degree to which gain has been reduced 314 r el at i ve t o i t s d a r k - a d a p t e d val ue. Thes e concl us i ons have t o be q u a l i f i e d t o i nc l ude t he p r o v i s o t ha t t he b a c k g r o u n d c o n d i t i o n s ha ve t o be s uch t ha t al l t he g a n g l i o n cel l s a r e d r i v e n b y t he r o d p a t h wa y . At pr e s e nt t he r e a r e no f i r m d a t a a b o u t t he ef f ect o f t he si ze o f cent er s o n t he de gr e e o f a d a p t a t i o n o f t he g a n g l i o n cel l p o p u l a t i o n i n t he p h o t o p i c r a nge . R. M. S H A P L E Y A N D C . E N R O T H - C U G E L L 3. 7. The Effect of the Receptive Field Surround on the Gai n of the Center 1 0 ' 1 0 5 o 1 0 ~ 1 0 4 ( a ) The s u r r o u n d o f t he r e c e pt i ve f i el d pl a ys no r ol e, or at mos t a mi n o r r ol e , i n s et t i ng t he ga i n o f t he cent er . Thi s is t he c o n c l u s i o n whi c h is i mpl i c i t i n t he p r o f u s i o n o f r es ul t s p r e s e n t e d a b o v e whi c h d e mo n s t r a t e t h a t t h e s u mma t i o n a r e a f o r a d a p t a t i o n is ei t her e qua l t o or s ma l l e r t h a n t he s i gnal s u mma t i o n a r e a o f t he r e c e pt i ve f i el d c e nt e r i t sel f . Ho we v e r , t hi s c onc l us i on is c ount e r - i nt ui t i ve t o a n u mb e r o f t h e o r i s t s who have p r o p o s e d t h a t t he c e n t e r ' s ga i n o u g h t t o be r e g u l a t e d b y t he s u r r o u n d ( He l s on, 1964; Sper l i ng, 1970; Gr os s be r g, 1981). Suc h a n i mp o r t a n t p o i n t de s e r ve s di r e c t s c r ut i ny. Th e i nf l ue nc e o f t he s u r r o u n d on t he c e nt e r i n t he s c o t o p i c r a nge was me a s u r e d b y En r o t h - Cu g e l l et al . (1975). The i r r es ul t s a r e g r a p h e d i n Fi g. 38. Th e ga i n a n d t he me a n i mp u l s e r a t e as a f u n c t i o n o f b a c k g r o u n d a r e a a r e s hown f or t wo cel l s, one X o n - c e n t e r cel l , a n d one Y on- c e nt e r cel l . The gai n d r o p s as t he a r e a o f a b a c k g r o u n d s pot , o f c ons t a nt l umi na nc e , is i ncr eas ed so t ha t it j us t fills t he cent er . I nc r e a s e o f a r e a b e y o n d t hi s va l ue p r o d u c e s no f ur t he r i ncr eas e or decr eas e i n gai n. Ho we v e r , t her e is s t e a d y s t at e i n p u t f r o m t he s u r r o u n d c a u s e d by t he l a r ge b a c k g r o u n d s b e c a u s e t he me a n i mp u l s e r a t e does de c l i ne whe n t he b a c k g r o u n d s pot gr ows l a r ge r i n si ze t h a n t he g a n g l i o n c e l l ' s c e nt e r a n d i n t r u d e s i nt o t he s u r r o u n d . Th u s s t e a d y s u r r o u n d i n p u t doe s not r e g u l a t e c e nt e r ga i n i n t hi s e x p e r i me n t . Th e q u e s t i o n o f wh e t h e r t he s u r r o u n d mi ght r e gul a t e t he gai n o f t he cent er i n t he p h o t o p i c r a nge c a n n o t be a n s we r e d so de f i ni t i ve l y i n t he ne ga t i ve . En r o t h - Cu g e l l et al . ( 1977a) f o u n d t h a t Y cel l s b e h a v e d t he s a me wa y i n t he p h o t o p i c a n d s c ot opi c r a nge s , i . e. t he r e was no c h a n g e i n ga i n whe n t he a r e a o f a b a c k g r o u n d s p o t o f f i xe d l u mi n a n c e was 10 ~ t 0.01 i i ~ , h H l t ] i l HI l [ ~ , i ~ JJ~l ~ i ~[ : I l d 0 1 1 0 10 100 Background area (deg ~) ( b) Mai ntai ned f requency g 4 0 - ? _E 2 0 0 01 0 1 1 0 10 100 Background area (deg z) F1G. 38. No ef f ect of the sur r ound on the gai n o f the center in cat retinal ganglion cells in the scotopic range. Red steady spots, centered on the receptive field middle, and of diameters indicated by their horizontal coordinates, were used as backgrounds. Their illuminations were 4. 104 quanta(507 nm) (deg a s) -1 for the Y cell (filled symbols) and 2.5. l0 s for the X cell (empty symbols). For both cells the stimulus was a blue - green test spot (0.18 deg diameter for the Y cell, 1 deg diameter for the X cell). The stimulus illumination was adjusted to produce an approximately constant peak response, and the gain calculated from response magnitude and stimulus flux. The gain (i/q ratio) fell as the background diameter was increased so long as the background diameter was less than the diameter of the center (marked with the empty arrow for the X cell, with the filled arrow for the Y cell). To determine whether light falling outside the center did indeed activate the surround, the maintained firing was measured and is displayed in the lower panel. The cells' maintained firing rates were reduced by the light failing beyond the center, indicating sustained surround antagonism of the center, even though the gains of the centers were not affected by this same light. From Enroth-Cugell et al. (1975). i nc r e a s e d b e y o n d t ha t o f t he c e nt r a l s u mmi n g a r e a , as i l l us t r a t e d i n Fi g. 39 ( Not e t ha t i n Fi gs 39 a n d 40 t he ve r t i c a l axi s is l og t h r e s h o l d f or a c r i t e r i o n r e s pons e , whi c h is e qui va l e nt t o t he l og o f t he r eci pr ocal o f t he gai n) . Howe ve r , t he d a t a on X cel l s wer e i nc ompl e t e . On l y one X cel l was s t udi e d i n t he p h o t o p i c r a nge , a n d i t s h o we d a s ma l l but s i gni f i cant i ncr eas e o f cent er gai n when t he a d a p t i n g VISUAL ADAPTATION AND RETINAL GAIN CONTROLS 315 s p o t wa s e n l a r g e d t o c o v e r t h e s u r r o u n d , a s s h o wn t o i n c l u d e t h e s u r r o u n d wa s s ma l l e r t h a n t h e i n Fi g . 4 0 ( a ) f r o m u n p u b l i s h e d r e s u l t s o f L e n n i e , f r a c t i o n a l l o s s o f g a i n wh e n a s t i mu l u s s p o t wa s He r t z a n d E n r o t h - C u g e l l . T h e f r a c t i o n a l r e c o v e r y e n l a r g e d t o i n c l u d e t h e s u r r o u n d a s i s i n d i c a t e d i n o f c e n t e r g a i n wh e n t h e b a c k g r o u n d wa s e n l a r g e d Fi g . 4 0 ( b ) . T h i s c o u l d b e b e c a u s e t h e s u r r o u n d A T 7 E c 0 " ( a ) 0 9 On- cent r e Y 0 0 8 0 O O 0 0 OO o O ( b) 0 I 10 On-centre Y N g ~ 0 ~ ~,~. 9 0 o 0 00000 O 8 I J I J l l l l Jl I J i i l i ; l I I OI I 1 J l l Jl l l t I J I l l l i J : 0.1 1. 0 10 1-0 10 Test di amet er (deg) Background di amet er (deg) FIG. 39. Signal and adaptive summat i on in a Y gangl i on cell in t he phot opi c range: no effect of surround on adaptive summat i on. (a) The phot opi c cent er' s size was det ermi ned as in Figs 30 and 32, by obt ai ni ng t hreshol d i l l umi nat i on for a const ant response, as a funct i on of area. Thi s was done with red spots (modul at ed at 4 Hz) on a steady b l u e - green backgr ound which had a retinal i l l umi nat i on of 8.108 quanta(507 nm) (deg ~ s) -1. (b) The size of t he " adapt at i on pool " of t he center was det ermi ned with an a r e a - a d a p t a t i o n curve as in Fig. 32(b). A fixed 0.2 deg di amet er stimulus was set t o be at t hreshol d on a 15 deg background. The backgr ound was reduced in area, in discrete steps, and t he i l l umi nat i on of t he backgr ound was adj ust ed t o keep t he response t o t he central test spot at criterion. The required i l l umi nat i on of t he backgr ound is pl ot t ed against its diameter. Fr om Enrot h-Cugel l et al. (1977a). ON-CENTER X (b) - ( a) o I00 B O ~. ~ , o -o o o O 0 p 9.0 o o 70 O o g o B O 60 S I I I I / l l I I I I 0.1 0. 5 1.0 ,5.0 I O0 O1 0. 5 1.0 5. 0 I 0. 0 BACKGROUND DI AMETER ( deg) TEST DI AMETER (deQ) FIG. 40. Signal and adapt i ve summat i on i n an individual X gangl i on cell in t he phot opi c range. (a) The left panel shows t he backgr ound i l l umi nat i on required t o keep a 0.2 deg di amet er red test stimulus (modul at ed at 4 hz) at audi t ory t hreshol d for modul at i on, as t he ( bl ue - gr e e n) backgr ound di amet er was varied f r om 0.2 t o 10 deg. (b) The ri ght panel shows t he test i l l umi nat i on required for a 4 Hz modul at ed, red, 0.2 deg diameter, spot t o produce a t hreshol d response, on a b l u e - green 15 deg backgr ound of retinal i l l umi nat i on 8-10 a quanta(507 nm) (deg 2 s)-' . Fr om Lennie, Hert z and Enroth-Cugell, unpubl i shed results. 316 mi ght be rel at i vel y less sensitive t o st eady st at e i l l umi nat i on t han is the center; the measur ement of surround - center gai n in (b) was made at a stimulus modul at i on r at e of 4 Hz, while t he i l l umi nat i on of cent er and sur r ound in t he adapt at i on exper i ment , plotted in (a), was const ant in time. This experi ment needs t o be repl i cat ed on a l arger popul at i on of X cells t han one, in or der t o be abl e t o eval uat e the s ur r ound' s cont r ol of cent er gain. However , it is an indication t hat the surround may pl ay a di fferent rol e in X cells in t he phot opi c r ange f r om its i neffect ual per f or mance in the scot opi c range. Thi s resul t is t ant al i zi ng f r om t he s t andpoi nt of s t r u c t u r e - f u n c t i o n correl at i ons. The absence of either sensi t i zat i on or desensi t i zat i on (in the cat retina) caused by the surround in the scot opi c range implies t hat the i nt erneuron which cont rol s the gain of t he gangl i on cell cent er has a recept i ve field the size of the cent er, with no spat i al l y ant agoni st i c i nput . The absence of s ur r ound gai n cont r ol of the cent er in the phot opi c range f or Y cells carries the same i mpl i cat i on. The presence of a smal l but si gni fi cant sensi t i zat i on of t he cent er by t he sur r ound in t he phot opi c r ange in X cells suggests t hat t he gai n of t he X cent er in t he phot opi c range mi ght be set with an i nt er neur on wi t h a cent er the size of the X cel l ' s phot opi c cent er but with an addi t i onal sur r ound. One possi bl e specul at i on is t hat X cells al ways have their gai n set by the bi pol ar cells f r om which t hey receive di rect i nput , and t hat t he r od bi pol ar cells have a weak or non-exi st ent surround mechani sm while the cone bi pol ar cells do possess a s ur r ound (cf. Nel son et al. , 1981). The Y cells whi ch receive mos t of t hei r i nput vi a the amacr i ne r out e woul d al ways have t hei r gai n set by i nt e r ne ur ons whi ch do not ha ve s u r r o u n d ant agoni sm. Ther e is a compl et el y di f f er ent expl anat i on f or sensi t i zat i on which shoul d be consi dered. Suppose t hat t he sensi t i zat i on exper i ment is done agai nst a low di ffuse backgr ound. Scat t ered light f r o m t he test spot may excite the sur r ound when the adapt i ng spot is smal l , and not when t he adapt i ng spot is l arge, and effect i ve in desensitizing t he sur r ound. Thus, t he " c o n t r o l " response t o a test spot on t he small backgr ound may be a " mi xe d" response f r om cent er and sur r ound, which is cl eansed of sur r ound c o n t a mi n a t i o n by t he e n l a r g e me n t o f t he backgr ound. Occasi onal l y, such mi xed responses t o R. M. SHAPLEY AND C. ENROTH-CUGELL test stimuli on small adapt i ng spot s were seen, bot h in t he Enr ot h- Cugel l et al. (1975) st udy (Figs 2, 3 and 4 in t hei r paper ) and in l at er wor k on sensitization of the cone pat hway by Lenni e, Her t z and Enr ot h- Cugel l (unpubl i shed). Removal of sur r ound cont ami nat i on pr oduces a l arger peak response, and a much l arger sust ai ned response, t o a test spot pl aced in the center of the receptive field. The l ar ge s ens i t i zat i on ef f ect s obs e r ve d in psychophysi cal experi ment s may be due mor e to this "r el ease f r om sur r ound c ont a mi na t i on" t han f r om t he possi bl e but pr obabl y weaker i nfl uences of sur r ound signals on t he gai n of t he cent er. 3 . 8 . A D i f f e r e n t Ki n d o f Ga i n Co n t r o l : T h e Co n t r a s t Ga i n Co n t r o l Up to this poi nt we have onl y consi dered the effect of st eady i l l umi nat i on on t he cont r ol of retinal gai n and dynami cs. However , t here is anot her gai n cont r ol which depends not on t he st eady light level but r at her on the aver age modul at i on of opt i cal stimuli over a wide regi on of visual space. This is what Shapl ey and Vi ct or (1978, 1979, 1980, 1981) have t er med t he cont rast gain control. It is pr oba bl y equi val ent t o the "si l ent s ur r ound" di scovered by H. B. Barl ow (1953), and t o t he "s uppr es s i ve s ur r ound" f ound in t he ret i na by Cl el and and Levi ck (1974). Anal ogous retinal mechani sms have been f ound in pre-gangl i oni c i nt er neur ons in t he mudpuppy ret i na by Werbl i n and Copenhagen (1974) and Thi bos and Werbl i n (1978b). Pl ease not e t hat t he t er mi nol ogy may be conf usi ng in this case; t he cont r ast gai n cont r ol adj ust s the gai n of the ret i na cont i ngent on cont rast r at her t han flux. Per haps we shoul d call it t he cont rast gai ncont rol , t o distinguish it f r om t he flux gai ncont r ol s. Thi s new ki nd of gai n cont r ol was ori gi nal l y di scovered in exper i ment s in whi ch t he t empor al frequency response of cat retinal gangl i on cells was measur ed as a f unct i on of cont r ast . The results of such an exper i ment are shown in Fig. 41. The dat a are di spl ayed in a Bode pl ot , wi t h log ampl i t ude vs log t empor al frequency in t he upper gr aph and with linear phase vs log t empor al f r equency in the l ower panel . I f t here were no cont r ast gai ncont r ol , t he ampl i t ude curves woul d be the same shape, i.e. they coul d be super i mposed by means of a vertical shift. VISUAL ADAPTATI ON AND Fur t her mor e, the phase curves at di fferent contrasts woul d be superi mposabl e. This is not t he case. The ampl i t ude curve is shi ft ed t owar ds hi gher t empor al frequenci es at hi gher cont rast s, and t he phases of i nt ermedi at e frequenci es are advanced at hi gher cont rast s. At an r . m. s, average cont r ast of 0.2, t he ampl i t ude of the response t o 0.5 Hz may be reduced by 507o on account of t he act i on of t he cont r ast gai ncont r ol . The phase may be shi ft ed at 8 Hz by as much as 60 deg. Anot her way t o l ook at t he cont rast gai ncont rol is t hat it produces what appears t o be a f r e q u e n c y - d e p e n d e n t sat ur at i on. The ampl i t udes of r esponses t o hi gher t empor al frequencies grow approxi mat el y proport i onal l y with cont r ast . The ampl i t udes of responses t o l ower t empor al frequenci es grow less t han pr opor t i onal l y with cont r ast , and appear t o sat urat e at a l ower cont r ast . That this ef f ect is not simple sat ur at i on is pr oven by t he t empor al f r equency dependence. The spatial and t empor al dependence of t he cont rast gai ncont rol ' s act i on indicate t hat it has t he same characteristics as t he nonl i near recept i ve field subuni t s whi ch feed exci t at i on t o Y cells (Shapl ey and Vi ct or, 1978; cf. Appendi x 2). It appear s t hat t he subuni t s precede t he cont r ast gai ncont r ol in ret i nal processi ng; t he subuni t s comput e t he t ot al average cont r ast whi ch is t hen used t o cont r ol t he r et i na' s dynami c responses. The way t hat t he cont r ast gai ncont r ol modi fi es t he t i me cour se of ret i nal responses appear s t o be by modul at i on of t he st rengt h of pre-existing f eedback pat hways. That is, increase of cont r ast t ends t o t ur n on t he cont r ast gai ncont r ol whi ch t hen increases t he st rengt h of negative f eedback in bot h X and Y ret i nal pat hways (Shapl ey and Vi ct or, 1981). The cont r ast gai ncont r ol calculates a cont r ast signal by averagi ng t he cont r ast modul at i on over a wide expanse of ret i na. The spatial ext ent of t he area of cont r ast averagi ng has not been det ermi ned precisely, but it must be consi der abl y l arger t han t he ext ent of t he cent er mechani sm of Y gangl i on cells. Rough estimates of its spatial ext ent may be based on t he area over which t he pr oduct , cont r ast times area, pr oduces a given amount of low f r equency suppressi on or mi d- f r equency phase advance. Thi s area is on t he or der of t en degrees, i.e. about 2 mm on t he ret i na. Fur t her mor e, t he cont r ast gai ncont r ol may receive weaker but still si gni fi cant i nput f r om still f ur t her reaches of t he RETI NAL GAI N CONTROLS 3 1 7 ret i na, in this way also resembl i ng t he exci t at or y nonl i near subuni t s of Y cells. Thus t he cont r ast gai ncont rol must involve a t hi rd type of i nt erneuron di fferent f r om t he previ ousl y i nferred i nt erneurons whi ch are needed t o account f or t he f l ux gaincontrols of the X and Y cell center mechanisms. Because of its wide summi ng area, t he cont r ast gai ncont r ol can account f or nonl i near spatial summat i on of responses pr oduced by cont r ast modul at i on in t he cent er and per i pher y of X and Y ganglion cell receptive fields (Shapley and Victor, 1979). Thi s has been observed with sinusoidal test stimuli by Shapley and Victor and with square wave Y CELL C o n f r a s t o 0 . I 3 2 . 0 0 o 0 0 1 2 5 ,o.oo _ E 3 2 o 1 , 0 0 < 0 3 2 I ] [ I I I ~ -0. 5 E -1.0 < -1.5 I - 2. 0 T E M P O R A L F R E Q U E N C Y FIG. 41. The c ont r a s t ga i nc ont r ol r eveal ed i n t e mpor a l f r equency r e s pons e s o f r et i nal ga ngl i on cells. The s e ar e da t a f r om a n on- cent er Y cell. The me a n l umi nance was 20 cd m -2, a n d t he pupi l di amet er was 3 mm. The s t i mul us was a 0. 25 c deg- ' si ne gr at i ng whi ch was modul a t e d i n a mpl i t ude by a s u m o f s i nus oi ds , wi t h t e mpor a l f r equenci es f r o m 0. 2 up t o 32 Hz. The e mp t y circles were obt ai ned whe n t he cont r as t was 0. 0125 per si nusoi d, i. e. i f any one o f t he ei ght s i nus oi ds ha d been pr esent ed al one, t he peak cont r as t o f t he modul a t e d gr a t i ng woul d have been 0. 0125. The filled circles were obt ai ned wi t h 0. 1/ s i nus oi d cont r as t . The ampl i t udes pl ot t ed ar e t he Four i er a mpl i t ude s i n t he cel l ' s i mpul s e t r ai n whi ch wer e at t he t e mpor a l f r equenci es pr es ent i n t he s t i mul us ; t he pha s e s ar e t he pha s e s hi f t s o f t hos e Four i er c o mp o n e n t s i n t he i mpul s e t r ai n wi t h r espect t o t he c or r e s pondi ng c o mp o n e n t in t he i nput si gnal . I f t he r et i na were l i near , or i f t he nonl i near i t y were a si mpl e sat ur at i on, t he t wo ampl i t ude cur ves s houl d be paral l el , a n d t he pha s e cur ves s houl d be paral l el , a nd t he pha s e cur ves s houl d s upe r i mpos e . The accent uat i on o f r es pons es at hi gh f r equenci es, a nd t he pha s e a dva nc e at mi d- r a nge f r equenci es , as c ont r a s t i ncr eases, is t he s i gna t ur e o f t he c ont r a s t gai ncont r ol . Fr o m Shapl ey a n d Vi ct or (1979). 318 test stimuli by Enr ot h- Cugel l and Jaki el a (1980). The effect of t he cont r ast gai ncont r ol on square wave responses is t o make t hem smaller and mor e t ransi ent , as can be seen in Fig. 42 f r om Enr ot h- Cugell and Jaki el a (1980). Thi s is an exampl e of nonl i near spatial summat i on because the peripheral stimulus generat es no response when present ed al one but suppresses t he response t o t he cent ral l y pl aced bar stimulus when bot h are present ed t oget her. This ki nd of suppressive ef f ect f r om peri pheral st i mul at i on has been observed in many gangl i on cell types in several di f f er ent species. -.~u I001 50] ~ 0 ~ 100 ~ 5 0 ] L o ON-CENTERY-CELL grating R. M. SHAPLEY AND C. ENROTH- CUGELL 3 . 9 . T i me C o u r s e o f Ga i n Ad j u s t me n t i n Re t i n a l Ga n g l i o n Ce l l s drifting ( a ) (b) t ( c ) ~ ( d ) 0 l 2 3 sec FIG. 42. Re duc t i on o f t he c e nt e r ' s ga i n by a movi ng pat t er n in t he per i pher y o f t he recept i ve field in a cat ret i nal gangl i on cell. Res pons es o f an on- cent er Y cell t o a cent er ed bar , 1 deg wi de, 13 deg hi gh, o f l umi na nc e 0. 25 cd m "2 f l as hi ng at 0. 4 Hz. (a) The bar was s upe r i mpos e d on a b a c k g r o u n d o f l umi nance 5 c d m -~. Bot h bar a nd backgr ound were generat ed on a CRT whi ch s ubt e nde d 13 deg by 16 deg. (b) A dri ft i ng si ne gr at i ng wi t h a cont r ast o f 0. 5, and spat i al f r equency 1.25 c deg -1 cover ed t he CRT scr een except f or a 2 deg wi de b a n d down t he cent er o f t he scr een wher e t he s t i mul us ba r was pr es ent ed. The gr at i ng was dr i f t i ng wi t h a t e mpor a l f r e que nc y o f 1.14 Hz. (c) Shows t he t i me cour s e o f t he bar s t i mul us . (d) I ndi cat es t he t i me cour s e o f t he l umi na nc e c ha nge at a poi nt in t he field over whi ch t he gr at i ng was dr i f t i ng. Fr o m Enr ot h- Cuge l l a n d Jaki el a (1980). The ret i nal gain as measured in gangl i on cells is readj ust ed rapi dl y but not i nst ant aneousl y by sudden changes in backgr ound level. This has been det ermi ned in physi ol ogi cal experi ment s which are anal ogous t o Cr awf or d' s psychophysi cal experi- ment on t he t i me course of light adapt at i on in humans ( Cr awf or d, 1947). = c o22: E 2 1 0 N 0.6 ~02 ~ I 1 I I I 1 I [ t }- 0 1 O0 300 500 700 Del ay bet ween ' on" of condi t i oni ng and ' o n ' of t est l i ght (rnsec) FtG. 43. Ti me cour s e of gai n change in cat gangl i on cells: modi f i e d Cr a wf or d exper i ment . On a 12 deg s t eady ba c kgr ound of r et i nal i l l umi nat i on 5' l os quant a( 507 nm) (deg 2 s) -1 were cent er ed: (a) a s qua r e wave modul a t e d condi t i oni ng spot (1.5 s on, 3.5 s off), di am. 0. 57 deg, ret i nal i l l umi nat i on 2. 5. los qua nt a ( de g 2 s)-' ; (b) a t est spot , f l as hed on f or 20 ms in one cell (filled circles) a n d for 50 ms in a not he r cell ( empt y circles), of di am. 0.1 deg, wi t h r et i nal i l l umi nat i on 3. l 0 s quant a( 507 nm) (deg ~ s)-' . The t est st i mul i were pr es ent ed wi t h var yi ng del ays wi t h r espect t o t he ons et o f t he condi t i oni ng l i ght . Pl ot t ed in t he f i gur e f or t wo di f f er ent on- cent er cells ar e t he ma gni t ude s o f t he r es pons es t o t he t est pul ses of l i ght at di f f er ent del ay t i mes bet ween onset of condi t i oni ng light and onset of t est pulse. The arrows poi nt t o t he r es pons es to t he t est st i mul i whi ch gener at ed r es pons es whi ch coi nci ded in t i me wi t h t he peak o f t he r es pons e t o t he condi t i oni ng l i ght . Fr o m Enr ot h- Cuge l l a nd Shapl ey (1973a). In this experi ment retinal gain is pr obed by measur ement of t he magni t ude of t he response t o a br i ef test pulse of light. The test pulse is appl i ed in the dar k and at various times aft er a conditioning light is t ur ned on. The results of an experi ment of this sort are shown in Fig. 43 f r om Enr ot h- Cugel l and Shapl ey (1973a). The response t o t he test pulse of light is t he same a hal f second af t er t he condi t i oni ng light is of f as it is when t he condi t i oni ng light has been of f f or several minutes. The response is defi ni t el y reduced within 200 ms f r om t he time t he condi t i oni ng light is t ur ned on. VISUAL ADAPTATION AND RETINAL GAI N CONTROLS Bet ween zero and 200 ms, t he response declines uni f or ml y in magni t ude. In rel at ed experi ment s, it was shown t hat t he gain st ayed at t he same new val ue f r om 200 ms until 5 s af t er t he condi t i oni ng light was t ur ned on. These experi ment s were done in Y ceils under low scot opi c condi t i ons. Si mi l ar results have been obt ai ned in X cells of t he cat, by Sai t o and Fukada (1975), who st udi ed t he t i me course of gain adj ust ment in X and Y cells under mesopi c or l ow phot opi c condi t i ons. Ther e is a gap in our knowl edge about t he t i me course of gai n adj us t ment in X cells under scot opi c condi t i ons, and a quest i on about whet her t he t i me course of adapt at i on changes mar kedl y bet ween scot opi c and phot opi c levels. Sai t o and Fukada (1975) f ound t hat Y cells showed a much sharper, mor e t ransi ent , gain r educt i on i mmedi at el y af t er t he condi t i oni ng light was t ur ned on, and t hen agai n a r educt i on i mmedi at el y af t er it was ext i ngui shed, as i ndi cat ed in Fig. 44. Thi s was at hi gher levels of backgr ound XON Co) (c) k . . . . I 319 illumination t han t he earlier experiments of Enr ot h- Cugell and Shapl ey (1973a), which may explain t he di fferences in findings. The gain r educt i on at bot h " o n " and " o f f " of t he condi t i oni ng light seen by Sai t o and Fukada may possi bl y have been due t o one of t he " a ma c r i n e " gain cont r ol s we have post ul at ed, ei t her t he cont r ast gai ncont r ol or t he st eady-st at e fl ux gai ncont r ol of t he Y cell cent er. It seems mor e likely t hat t he cont r ast gai ncont r ol is i nvol ved in this phenomenon. The cont r ast gai ncont rol contains t he kind of nonl i neari t y which woul d cause r educt i on at bot h " o n " and " o f f " ; it is an even- or der nonl i near i t y which generat es responses of t he same sign at " o n " and " o f f " , responses like t hose seen in amacr i ne cells and Y cells. It is i nt erest i ng t hat similar gain reduct i ons at " o n " and " o f f " are seen psychophysically in t he ori gi nal Cr awf or d (1947) exper i ment and in i nt raret i nal recordi ng f r om t he amacr i ne cell l ayer ( Gor don and Gr aham, 1973; also unpubl i shed results). It may be t hat t ransi ent gain reduct i ons in YON (e) 1.0 0 . 5 - (d) S ' - - / . [ ! i i I 2 0 i 2 Time, sec | ( h ) I - 0 ] i o i 2 3 3 4 0 Ti me, f - / J 2 3 sec FIG. 44. Cr a wf or d exper i ment f or X a nd Y cells under l ow phot opi c or hi gh scot opi c condi t i ons. X cell dat a ar e in (a) - (d); Y cell dat a ar e ( e ) - (h). PST hi s t ogr ams s how aver aged r esponses t o t he t est s pot ( shor t bar under hi s t ogr ams i ndi cat es st i mul us t i me) and t o t he condi t i oni ng spot (l onger bar under hi st ograms). The backgr ound l umi nance was 5.9 cd m -2 t hr ough a 4 mm di amet er art i fi ci al pupi l . The unat t enuat ed st i mul i wer e : t est , 3.9" 103, a nd condi t i oni ng, 6. 8- 10 ~ cd m "2. They each s ubt ended 5' di amet er . The t est s pot was pl aced i n t he mi ddl e o f t he recept i ve fi el d cent er , whi l e t he condi t i oni ng s pot was pl aced 15' t o t he side, still wi t hi n t he cent er. (a) and (e): t he t est spot was pr esent ed 2 s af t er onset o f t he condi t i oni ng spot . (b) a nd (f): t est s pot al one, as a cont r ol . (c) a nd (g): t est s pot pr es ent ed 2 s af t er of f s et o f t he condi t i oni ng spot . The dur at i on o f t he t est s pot was 200 ms f or t he X cell, 100 ms f or t he Y cell. The dur at i on o f t he condi t i oni ng s pot wa s 3 s. The vert i cal cal i br at i on is 50 i mpul ses s - ' . (d) a nd (h): The t i me courses o f t he change o f magni t ude o f t he r esponse t o t he t est caus ed by pr es ent at i on o f t he condi t i oni ng spot . The rel at i ve magni t ude o f t he r es pons e is pl ot t ed agai nst t he del ay bet ween t he onset t i mes o f t est and condi t i oni ng spot s. Fr om Sai t o and Fuka da (1975). PRR3-1* 320 R. M. SHAPLEY AND Y cells, due t o subst ant i al amacri ne i nput t o these cells, are t he basis f or t he ps ychophys i cal " o n - o f f " t ransi ent sensitivity losses. While t he above results indicate t he rapi di t y of some gain cont r ol processes in the retina, t hey do not rule out t he existence of slower gain cont rol s. These slower processes have not been studied with t he Cr awf or d paradi gm, but t hey shoul d be. Adel son' s (1982) results on t he t i me course of human light adapt at i on i ndi cat e t he presence of slower adapt at i on processes t han have been studied in gangl i on cells up t o now. It is well known but poor l y document ed t hat when t he mean level of i l l umi nat i on is st epped up or down by t wo or mor e log units in an adapt at i on experi ment , the gain may t ake several mi nut es t o settle down or up to its new steady value. Such slow adapt at i on processes requi re f ur t her research. 4. GAI N AS A FUNCTI ON OF I LLUMI NATI ON IN AMACRI NE CELLS, BI POLAR CELLS, AND HORI ZONTAL CELLS The phenomena of gain cont r ol in gangl i on cells requi re an expl anat i on in t erms of t he funct i onal connect i ons of t he retinal net wor k a nd/ or t he i nt r i ns i c p r o p e r t i e s of t he r e c e p t o r s a nd i nt er neur ons in t he retina. The f undament al quest i on is, where does t he cont r ol of gain begin? C. ENROTH-CUGELL Ideally, one woul d want t o have t he answer t o this quest i on f or t he cat ret i na f r om whi ch so much of t he results on gain cont r ol in gangl i on cells have been obt ai ned. While t here are some f r agment ar y results on the cat, t he t echni cal difficulties of i n t r a c e l l u l a r r e c o r d i n g ha ve p r e v e n t e d a compr ehensi ve st udy of gain cont r ol in cat ret i nal i nt erneurons. We will t her ef or e concent r at e on t he results f r om t he retinas of t wo col d- bl ooded vert ebrat es whi ch have been studied t he most : t he mudpuppy ( Nor mann and Werblin, 1974; Werblin, 1974; Werbl i n and Copenhagen, 1974; Thi bos and Werbl i n, 1978a, b) and t he channel cat fi sh ( Naka et al . , 1979). 4.1. Amacrine Cells In the mudpuppy t he amacri ne cells have a steep response versus i l l umi nat i on curve which can be shi ft ed al ong t he log i l l umi nat i on axis by st eady backgr ound i l l umi nat i on. This is illustrated in Fig. 45 f r om the wor k of Werbl i n and Copenhagen (1974). This " cur ve shi ft i ng" is evidence f or a gain cont r ol at or pr i or t o t he amacr i ne cell (see Sect i on 1 . 2 . 2 . ) . In t he catfish, Naka et al. (1975) divided t he amacri ne cells into t wo classes, t he t ype N and t ype C cells. The type C cells correspond t o what Werblin and colleagues (Werbl i n and Dowling, 1969; Werblin and Copenhagen, 1974) call amacrine cells; m i l l l v o l t s I o 8 5 . 0 ~ . ~ 4 . 5 ~ - / / .s 4 - / I I / " - " ~ - ~ - ~ - 2 . 5 ~" I / ~ z . o 2 - I I ~ " - - , . ~ - - - - , . 5 I 0 ~ ' = l ' I J I t n L o 2 . 0 3 . 0 4 . 0 .... J [ ] I o mv LOG I STIMULUS Is Flo. 45. Curve-shifting in mudpuppy amacrine cells, recorded intracellularly. Diffuse illumination was used as a test stimulus. The dashed curves are inferred in regions where no data were taken. The peak value of the response is plotted against log flash illumination. The retinal illumination from the tungsten source used was about one quant um per rod per second for log I = 0. This works out to roughly 6" 103 quanta (522 nm) (mm 2 s) -~ for log I = 0. From Werblin and Copenhagen (1974). VISUAL ADAPTATION AND RETINAL GAIN CONTROLS they produce " on- off" responses to spots or full- field flashes. The type N cells are like the so-called "sustained" amacrine cells found by Kaneko (1973) and Naka and Ohtsuka (1975). The change of gain of these two types of cell with background is illustrated in Fig. 46 from Naka et al. (1979). They clearly are affected similarly by backgrounds and also greatly resemble the ganglion cells in the way they adapt. That is, the gain is approximately the reciprocal of background, approximately Weber's Law. We have to make a brief digression to make clear that the results in Figs 45 and 46 were obtained in compl et el y di fferent ways. Werblin and Copenhagen (1974) used standard rectangular increments of light on a steady background, and measured the peak of the change in intracellularly 4.2. Bipalar Cells recorded membrane potential. Naka et al . (1979) measured first-order Wiener kernels by cross- 1 0 4 z ~ 10 3 ~ 10 2 b I-- _ ~ i O + io o ~ . / - - THEORETICAL IO 4 " ~ " \ HORJZONTAL o, \XX~, IO 2 " G A N G L I O N C ~ BIPOLAR CELL I I I I IO 0 IO I i0 z iO 3 MEAN INTENSITY I I I I0 0 I01 10 2 OFF CENTER ~BIPOLAR CELL f T Y P E N CELL 1 0 4 FIG. 46. I nc r e me nt a l ga i n o f cat f i s h ne ur ons , r ecor ded i nt r aeel l ul ar l y. The or di na t e is t he i nc r e me nt a l gai n, t he a mpl i t ude o f t he fi rst or der Wi ener ker nel nor ma l i z e d by t he ma g n i t u d e o f t he whi t e- noi s e mo d u l a t i o n . Th e hor i z ont a l c oor di na t e is t he me a n i l l umi nat i on. Th e dot t e d l i ne is a pr edi ct i on f r o m t he Na k a - Ru s h t o n r el at i on f or t he ga i n i f t her e were no a da pt a t i on, j us t s at ur at i on. I ns et is a n aver age o f t en s i mul t aneous r ecor di ngs f r om hor i zont al a n d gangl i on cells. Res ul t s f or va r i ous cell t ypes ar e l abel ed i n t he f i gur e. We bel i eve t he l owest me a n i nt ens i t y was a r et i nal i r r adi ance o f 0.1 ~ wat t s crn -2 o n t he ret i na. The light sour ce was i n mos t c a s e s a g l o w- mo d u l a t o r t ube whi c h was a t t e nua t e d by neut r al dens i t y fi l t ers. F r o m Na k a e t al . (1979). 321 correlating a white-noise modulated light stimulus with the resulting "noi sy" modulation of the cells' membrane potential. The gain for the type N cells, and for all the other cells the results of which are shown in Fig. 46, is the gain at the peak of the first order Wiener kernel. The peak value of the kernel (first or second order) at each background level, divided by the standard deviation of the white noise stimulus, may be taken as a measure of gain at that background level. The bipolar cells form the main link between photoreceptors and ganglion cells and thus the dependence of gain on background for these cells is of crucial importance for understanding the site of adaptation. The available data on gain control in bipolars come mainly from Naka et al. (1979) and from Thibos and Werblin (1978a). The results of Naka et al . (1979) on catfish bipolars are displayed in Fig. 46, on the same graph as the amacrine cell data. Clearly, the bipolars' gain begins to drop at about the same level of illumination as the amacrines, and falls with a similar but somewhat shallower slope. Since bipolar cells are the input to amacrine cells in catfish (Naka and Ohtsuka, 1975) as in other animals (Dowling and Boycott, 1965), it is reasonable to suppose that the dependence of gain on background observed in the amacrines is already largely determined at the bipolar level. The results on light adaptation in bipolars in mudpuppy (Thibos and Werblin, 1978a) have concentrated on the way in which the surround of the bipolar receptive field sets the gain of the center. This is illustrated in Fig. 47, which shows response vs log illumination under two conditions: (i) no background i l l umi nat i on; (ii) background illumination falls on the bipolar's surround. In case (i), the i nt ens i t y- r es pons e funct i on is approximately fit by the Naka - Rushton relation, equation (6). In case (ii), the response vs illumination curve of the same form is shifted to 322 t he right as woul d fol l ow f r om gain r e- adj ust ment by illumination of t he surround (see Section 1.2.2.). Cont r ol experi ment s established t hat this was not a result of light scat t er ont o t he cent er (Werbl i n 1974). This is evidence t hat , in t he mudpuppy ret i na, t he gain of t he bi pol ar cell is set by signals comi ng f r om its sur r ound, whi ch in this species is believed t o be medi at ed by hori zont al cells (Werblin and Dowl i ng, 1969; Thi bos and Werbl i n, 1978a). Such results suggest t hat , in t he mudpuppy, horizontal cells act as a gain cont rol on bipolar cells. Two poi nt s of compar i s on wi t h pr evi ousl y present ed psychophysi cal and physi ol ogi cal results are needed here, to prevent the (probabl y erroneous) inference t hat this conclusion is generally applicable t o all vert ebrat es. The first compar i son of Fig. 47 is with t he sensi t i zat i on p h e n o me n o n in huma n vi si on di scovered by West hei mer (1965). The results in Fi g. 47 ar e t he opposi t e o f We s t h e i me r sensitization. Il l umi nat i on in t he per i pher y of t he mudpuppy bi pol ar' s receptive field desensitizes the cent er in Wer bl i n' s (1974) and Thi bos' and Wer bl i n' s (1978a) experi ment s. Ther e is, however, a puzzling and unresol ved cont r adi ct i on with Bur khar dt ' s (1974) r epor t of sensitization in mudpuppy bi pol ar cells. Per haps it has t o do with di f f er ent r ecept or i nput t o t he bi pol ar cells in the t wo sets of experi ment s. Thi bos and Werbl i n (1978a) were worki ng at low backgr ounds at which rods were t he pr edomi nant i nput while Bur khar dt was worki ng with a highly light adapt ed mudpuppy ret i na in which it is pr obabl e t hat cones were t he pr edomi nant phot or ecept or inputs t o t he proxi mal retinal neurons. I f this is t he expl anat i on f or t he opposi t e results, it woul d be an i nt erest i ng and unusual exampl e of a reversal in funct i onal characteristics because of t he t ransi t i on f r om r od t o cone pat hways. A second compar i son wor t h pursui ng is t hat bet ween t he st rong effect on t he gain of the cent er exert ed by t he sur r ound of mudpuppy bi pol ar cells (Fig. 47) vis-a-vis negligible or, at most , weak effect of t he sur r ound on t he gain of t he cent er in X and Y cat ret i nal gangl i on cells (Figs 3 8 - 4 0 ) . This di fference bet ween results on t he spatial ext ent of " adapt at i on pool s " suggests t hat qui t e di f f er ent mechani sms are i nvol ved in t he cont r ol of gain in t he mudpuppy and cat retinas. In a sense such a R. M. SHAPLEY AND C. ENROTH- CUGELL di fference woul d not be surprising because t he mudpuppy and cat have evol ved quite di f f er ent l y with widely di f f er ent visual capacities. The details of exact l y how t he retinal net wor k is connect ed spatially to regul at e gain mi ght well di f f er bet ween t wo such di st ant l y rel at ed animals. Ashmore and Falk (1980) have demonst rat ed t hat t he gain of bi pol ar cells, in t he al most all-rod ret i na of t he dogfi sh, begins t o dr op at ext remel y low backgr ounds because of sat ur at i on in t he bi pol ar cell itself. In the dogfi sh ret i na, t here is a very high ampl i fi cat i on of r od signals at t he r od- bi pol ar synapse, and as a result t he bi pol ar cells appr oach t hei r response ceilings at very low backgr ounds. Ther e is no evi dence f or a gain cont r ol , and t h e r e f o r e t r ue l i ght a d a p t a t i o n , in t hes e experiments. However, the results of Werblin (1974) and Naka et al. (1979) illustrate how an aut omat i c gain cont rol , acting on signals f r om phot orecept ors t o bi pol ar cells, staves of f sat ur at i on in mudpuppy and cat fi sh bi pol ars. In Fig. 46 f or exampl e, the catfish bi pol ar' s gain begins t o dr op at a lower mean level t han the hori zont al cell' s gain, presumabl y due -17- E -6 "- --i 8- 4-' o - 19- E - - 2 0 ,,~ T I i " I I ~ I I - 7 - - 6 5 4 FIG. 47. The ef f ect of a st eady i l l umi nat i on in t he per i pher y o f t he recept i ve fi el d on t he s t i mul us - r e s pons e f unc t i on o f mu d p u p p y bi pol ar cells; i nt racel l ul ar recordi ng. The s t i mul us was a f l as hi ng s pot cent er ed on t he r ecept i ve fi el d o f t he bi pol ar cell. The s pot ' s di amet er , 0. 4 mm, was chos en t o s t i mul at e t he cent er opt i mal l y. St i mul us dur a t i on was 1 s. The r e s pons e me a s ur e was t he s t eady s t at e pl at eau o f t he 1 s response. Filled circles ar e f or no i l l umi nat i on i n t he peri phery o f t he r ecept i ve fi el d; e mp t y circles ar e r e s pons e ma gni t ude s whe n a - 4 l og uni t st eadi l y i l l umi nat ed a n n u l u s was pl aced i n t he per i pher y o f t he fi el d. A modi f i e d Na k a - Ru s h t o n f unc t i on was f i t t ed t o t he da t a i n t he condi t i on o f no per i pher al i l l umi nat i on, a n d t he n t r ans l at ed l at er al l y t o fi t the r esul t s obt a i ne d wi t h t he a nnul us . The l i ght s our ce was a t ungs t e n l a mp, a t t e nua t e d by neut r al fi l t ers. The una t t e nua t e d r et i nal i l l umi nat i on was cal cul at ed t o be a bout 10 '3 quant a( 522 nm) (cm ~ s) -t . I l l umi nat i ons ar e gi ven as l og a t t e nua t i on rel at i ve t o t hi s val ue. Fr o m Thi bos a n d Wer bl i n (1978a). VI SUAL ADAPTATI ON AND RETI NAL GAI N CONTROLS t o a neural gain cont rol acting on the bi pol ar and not on the hori zont al cell. But at higher levels of mean i l l umi nat i on, the hori zont al cell pot ent i al starts sat urat i ng and its gain pl ummet s, failing below the bi pol ar' s gain. Since bipolar cells appear t o have steeper i nt e ns i t y- response funct i ons t han horizontal cells (Werblin, 1974), t hey might saturate at even lower mean levels t han the hori zont al cells, were it not for the saving action of the gain control. The aut omat i c gai n cont rol causes the bipolar cells (and the more proxi mal retinal neurons t hey feed) to lose some gain at lower levels in order t o preserve gain at higher levels which woul d otherwise be lost because of sat ur at i on (see Section 1.2.2.). 4. 3. Hori zont al Cells There is quite a lot of i nf or mat i on on the control of gai n in hori zont al cells because t hey are easier to record f r om intracellularly t han the other retinal interneurons. Since horizontal cells are anatomically one synapse away f r om phot orecept ors, t hey of t en mimic the receptors' adapt at i onal properties. One of the significant out comes of this fact is t hat the regul at i on of gain in phot orecept ors may be i nferred, wi t h caut i on, f r om studies of the regul at i on by light of the gain of hori zont al cells. - 5 0 - 4 O E - 3 0 == - 2 0 - i o 10 ~ / j " , I I Dar k adapt ed - 6 3 BG O 5. 3 BG - 4 . 2 BG - 3 2 BG ~ f I I f - 7 . 0 - 6 . 0 - 5 0 323 We can begin t o see some of the diversity in phot or ecept or a da pt a t i on in t he var i et y of hori zont al cells' adapt at i onal behavi or (cf. also Section 5). Perhaps the best studied receptor - horizontal cell system is in the t urt l e retina. Results on adapt at i on in t urt l e hori zont al cells are quite clear, as shown in Fig. 48 ( Nor mann and Perl man, 1979b). The i nt ens i t y- r es pons e curves shift t o the right with backgr ound, and the gain follows Weber' s Law. These horizontal cells are driven exclusively by long- wavelength cones, and their adapt at i onal properties are mai nl y det ermi ned by their cone inputs ( Nor mann and Perl man, 1979a). The curve- shifting in Fig. 48 is clear evidence for adapt at i on of the t ype suggested in equat i on (10), with an aut omat i c gain cont rol located in the cones. To illustrate inter-species diversity we compare the t urt l e hori zont al cell with the hori zont al cell of the catfish retina, the gain vs backgr ound curve of which is illustrated in Fig. 46. Here the hori zont al cell curve is not like the curve of Weber' s Law. Rat her, it follows the dashed curve, which is a predi ct i on based on the Na k a - Ru s h t o n rel at i on [equations (6) and (7)]. That is, the loss of gai n is caused by saturation. Under the conditions of these experiments, the catfish horizontal cells were driven exclusively f r om long-wavelength cones (Naka, I I I - 4 0 ~ 3 0 - 2 0 LOg rel at i ve intL~r~Slly F{G. 48. Ad a p t a t i o n i n t ur t l e hor i z ont a l cells; i nt r acel l ul ar r ecor di ng. The st i mul i were 0. 5 s i nc r e me nt s or de c r e me nt s on a b a c k g r o u n d , except f or t he r es ul t s on t he da r k a da pt e d eye, whi ch were i nc r e me nt s onl y. The st i mul i wer e l ar ge 3. 2 mm di a me t e r s pot s on t he r et i na. Pe a k r e s pons e s me a s u r e d f r o m t he da r k- a da pt e d r est i ng pot ent i al ( dot t ed line) were pl ot t ed as a f unc t i on o f t est i l l umi nat i on. The cur ves were dr a wn by eye. The hor i zont al bar t h r o u g h each cur ve i ndi cat es t he st eady me mb r a n e pot ent i al me a s u r e d at l east t wo mi nut e s af t er b a c k g r o u n d ons et . The i l l umi na t i ons ar e gi ven as l og a t t e nua t i on. Th e u n a t t e n u a t e d t est s t i mul us (0 l og) was 6. 4- 10 ~ qua nt a ( 640 nm) ( cm 2 s) -1 on t he r et i na. Th e u n a t t e n u a t e d b a c k g r o u n d i l l umi na t i on was 9 . 1 . 1 0 TM qua nt a ( 640 n m) ( cm 2 s) -1. Fr o m No r ma n n a n d Pe r l ma n (1979b). 324 per sonal communi cat i on) . Fi gure 46 i ndi cat es t hat hor i zont al cells, and by i nference t he cones, of the cat fi sh ret i na do not adapt but mer el y sat urat e, but t hat bi pol ar cells, amacr i ne, and gangl i on cells do adapt . In t he cat fi sh, much of t he adapt at i on t o st eady light mus t t ake pl ace bet ween cones and bi pol ar cells, t hough Fig. 46 also implies t hat t here are addi t i onal stages of gai n cont r ol in t he i nner pl exi f or m l ayer. Ther e are dat a on t he dependence of gai n on backgr ound in cat hori zont al cells, under conditions such t hat t he responses were due t o phot or ecept or i nput f r om rods only. The gr aph in Fig. 49 indicates t hat cat hori zont al cells, driven by rods, have a gai n vs backgr ound dependence whi ch appr oxi mat el y follows Weber ' s Law above 1 td [Not e this is a " cat t d " and t her ef or e is equi val ent t o about 8. 10 s quant a(507 nm) (deg 2 s)-' ; St ei nberg, 1971]. Not e t hat t hese hor i zont al cells t ypi cal l y receive mi xed phot or ecept or i nput (Steinberg, 1971), even t hough their maj or direct synapt i c cont act is with cones (see Appendi x 1). Thi s suggests t hat r od - cone coupl i ng is i ndeed i mpor t ant f or det er mi ni ng t he response pr oper t i es of hor i zont al cells. In compar i ng hor i zont al cell r esponses wi t h gangl i on cell responses (Fig. 24), one notices t hat hor i zont al cell gai n does not begi n t o dr op until t he backgr ound is 1 td, which is t wo t o f our log units hi gher t han the t r ansi t i on level of i l l umi nat i on f or gangl i on cells. Thi s suggests t hat in the cat , as in t he cat fi sh, t here must be gai n cont r ol mechani sms mor e pr oxi mal in t he ret i na t han t he rods or the hori zont al cells. However , the Weber Law behavi or of hor i zont al ceils in the cat suggests t hat the rods may also adapt when the level of illumination is high enough (but see Section 5). Gai n versus backgr ound curves are not avai l abl e f or cone- dr i ven hor i zont al cell responses in cat . The di versi t y of hor i zont al cell gai n changes due to backgr ounds i ndi cat ed so far is represent at i ve of t hat seen general l y. For i nst ance, in mu d p u p p y t he hori zont al cells behave as if influenced by sat urat i on and adapt at i on in t he scot opi c range, but resembl e turtle hor i zont al cells - - al most pur e adapt at i on - - in t he phot opi c r ange ( Nor ma nn and Werbl i n, 1974). However , car p hor i zont al cells show onl y sat urat i on in the phot opi c range (Wi t kovsky, 1967). But the r od- dr i ven skat e hor i zont al cells do adapt to light by gai n r educt i on af t er an initial peri od of R. M. SHAPLEY AND C. ENROTH-CUGELL sat ur at i on (Dowl i ng and Ri pps, 1971). Ther e is not one st or y f or all ver t ebr at es. The pr oxi mi t y of hor i zont al cells t o phot or ecept or s suggests t hat t he recept ors mi ght also be very diverse in how t hey deal wi t h changes in mean or backgr ound i l l umi nat i on. Thi s expect ed di versi t y is f ound. 5. G A I N C O N T R O L I N P H O T O R E C E P T O R S Some phot or ecept or s adj ust t hei r gai n by adapt i ng in t he presence of st eady i l l umi nat i on and onl y sat ur at e in very bri ght light, and ot her phot or ecept or s adapt ver y little bef or e t hey sat ur at e. Thi s di versi t y of r ecept or f unct i on wi t h respect to gai n cont r ol and sat ur at i on has been suggest ed al ready in t he discussion of gai n cont rol s in hor i zont al cells. In order to organi ze this diverse mat er i al , we will present t he dat a on r ecept or s as fol l ows: (a) phot or ecept or s which adapt a lot a n d al so sat urat e; and (b) phot or ecept or s which adapt very little and mai nl y sat ur at e. 30 .~_ o_ ~ 25 m ~20 <1 o ~ P5 4O 55 l ii I 0 05 - 05 0 0.5 ~0 r 5 20 25 Log l~z, ,~ ( scol : opi c t d) 1 30 FIG. 49. Log test illumination (at 440 nm) plotted as a function of log background illumination (620 nm) for a constant amplitude peak horizontal cell response intracellularly recorded in the cat retina. The background duration was 5.5 - 6.5 s; the stimulus was 0.5 s in duration applied after the response to the background had settled to a constant value. The stimulus was a spot on a larger background. The criterion response was 2.25 mV, and was rod-driven. From Steinberg (1971). VISUAL ADAPTATION AND RETINAL GAIN CONTROLS 325 5. 1. Pho t o r e c e pt o r s whi ch Adapt a l ot and Sat urat e The photoreceptors which actually adapt, i.e. more or less rapidly adjust their gain to avoid 10 saturation, include the following r o d s : skate, (Dowling and Ripps, 1972); gecko, (Kleinschmidt and Dowling, 1975); toad (Fain, 1976; Baylor e t al . , 1979). The adapting c o n e s include: turtle (Baylor and Hodgkin, 1974; Normann and Perlman, 1979a); mudpuppy (Normann and Werblin, 1974); frog (Hood and Hock, 1975); ground squirrel (Dawis and Purple, 1982); perhaps monkey (Valeton and van Norren, 1983). The results on cone adaptation in ground squirrel, frog, and monkey are from massed potential recordings which in one way or another isolated the cones. Figure 50 illustrates photoreceptor adaptation in rods from the toad B u f o m a r i n u s (Fain, 1976). There is evidence of saturation (in the decline of the peak response on background illumination), but there is also clear curve-shifting of the V-log/curve, consistent with the idea of an automatic gain control, equation (10) (cf. Normann and Werblin, 1974; Dawis and Purple, 1982). Furthermore, the gain of the toad rod follows Weber's Law (Fain, lo 1976), and this can even be seen in the photocurrent recorded from the outer segment by the suction electrode technique (Baylor e t al . , 1980) as in '~ Fig. 51. In Fig. 51 the vertical axis is labeled ~ lo "sensitivity", but it is equivalent to what we have _ defined as "gai n". Log gain vs log background has a flat portion and then a declining portion with a 01 slope about - 1. The slope of unity suggests an adaptive gain control rather than the steeper slope associated with pure saturation (see Section 1.2.2.). The transition from flat to sloping is at about an illumination of 0.28 quanta/am -2 s -1, which is about forty times greater than the "dark light" of the rods, the spontaneous current fluctuations in the dark, measured independently. Baylor e t al. (1980) make the point that this implies that the gain control transition level in rods is not determined by the "dark light". Similarly, we have pointed out in connection with gain reduction in ganglion cells, that the "dark light" in the cat retina is too small to be responsible for determining the illumination at which gain starts to fall (see Section 3.1.1.). The gain reduction of the rod photocurrent is accompanied by a very significant speeding up of 1 O 5 I I 7 8 s o o . l o g I I I I 1 0 1 1 1 2 1 3 ( q u a n t a c r n - 2 f l a s h 1 ) FIG. 50. Effect of background light on the intensity- response curves of toad rods; intracellular recording. Peak response amplitude to a brief (around 100 ms) flash is plotted against log flash energy per unit area (quanta cm-2), for the dark adapted rod and for several background levels of illumination. The stimulus and background were diffuse. The illumination of the backgrounds is given in log units to the left of each curve, and the units are log quanta(505 nm) (cm 2 s) -1. From Fain (1976). 1 0 0 - - ~ s d 7 " . " I ! I I I 0-01 0"I I 0 1 0 I(}(3 l e ( p h o t o n s ~ m - 2 s e c - I ) FIG. 51. Gain vs background in the photocurrent of a toad rod outer segment; recorded with suction electrode. The gain is plotted on the vertical axis with units picoamps/(photon /~m-2), but could really be given in picoamps/photon since the rods are of fixed cross-sectional area. The background illumination is plotted on the horizontal axis. The transition from horizontal to sloping portions of the curve occurs at 0.28 quanta(507 nm) (/am 2 s) -1. The arrow points to this illumination. Background and test were diffuse. From Baylor e t al . (1980). the rod response. The gain declines by about a factor of one hundred over a range of backgrounds which speed up the time to peak of the photocurrent (in response to a brief flash) by a factor of seven. 326 R . M. S H A P L E Y A N D C . E N R O T H - C U G E L L - 1 5 - 1 0 g -5 g. 0 I I I D a r k a d a p t e d - 4 " 4 B G D - 3 . 2 B G - 2 . 1 B G 121 I I I - - 6- 0 - - 5"0 - 4 . 0 I I I I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . I I I I - 3 " 0 - 2 . 0 - 1 . 0 0 . 0 Log r e l a t i v e i nt e ns i t y FIG. 52. Gain adjustment in turtle cones (recorded intracellularly) caused by background illumination. The stimuli were 0.5 s increments or decrements on a steady background (except for the curve for the dark adapted cone which only is for increments). The stimulus spot was 3.2 mm in diameter on the retina. Peak responses measured from the dark-adapted resting potential (dotted line) were plotted as a function of test illumination. The curves are Naka- Rushton functions. The horizontal bar through each curve indicates the steady membrane potential measured at least two minutes after background onset. The illuminations are given as log attenuation. The unattenuated test stimulus (0 log) was 6.4.10" quanta(640 nm) (cm 2 s)-' on the retina. The unattenuated background illumination was 9.1.10 is quanta(640 nm) (cm 2 s)-L From Normann and Perlman (1979a). Bayl or e t al . (1980) f ound t hat G R ~ (tpeak) 25, 2- wher e G R is t he gai n of t he r od i n pi c oa mp/ qua nt um and t p e a k is the t i me from the onset of the bri ef flash st i mul us to the peak of the 1 response. Adj us t ment of gai n i n cones by adapt at i on and sat ur at i on is represent ed i n Fig. 52 from the dat a of Nor ma nn and Per l man (1979a). The paral l el ~ 0 - I curves are the response t empl at e f r om the dark ,-I adapt ed cone, and are descri bed by the Naka- Rus ht on r el at i on, e qua t i on (6). Whe n t he _o backgr ound is raised, the response to i ncrement s is somewhat compressed (consi st ent wi t h sat ur at i on) - 2 because the steady state response increases t owards the maxi mal pot ent i al the cone is capabl e of produci ng, as i ndi cat ed by the short hori zont al bars - 3 i nt ersect i ng each operat i ng curve. Fr om the shift of 2 the operat i ng curves, one may i nfer Weber ' s Law since there seems t o be about a one log uni t shift for each log uni t increase in background. That turtle cones appr oxi mat el y fol l ow Weber ' s Law is i ndi cat ed i n Fig. 53 from Bayl or e t al . (1974b). The solid line is Weber ' s Law up t o about log Is = 4.5, and t hen rises due t o sat ur at i on. It is based on a t heory for the cone' s gai n cont r ol mechani sm by Bayl or e t al . (1974b). Act ual l y the cone' s gai n follows Weber ' s Law for about a log uni t more t han the Bayl or e t al . t heory predi ct s, and sat urat es rat her less t han the t heory predicts. One possibility not consi dered by Bayl or e t al . is t hat some of the / / ~ 0 I I I I I 3 4 , 5 6 7 l o g I , FIG. 53. Turtle cones' gain as a function of background; intracellular recording. The ordinate of this unique plot is the logarithm of the difference between the reciprocal of the gain in the light adapted cone and the reciprocal of the gain in the dark adapted cone, i.e. Iog{1/SF-I/SFD}. This number unfortunately goes to minus infinity when the background is so low that the gain of the cone is the same as in the dark adapted state, but that region of the curve is not plotted here anyway. The gain S F is in gV quantum-' effectively absorbed. The horizontal axis is log background flux; the units are quanta/sec effectively absorbed by the cone. The stimuli were 10 ms flashes, and the gain is calculated for the peak of the flash response. These are data from a red turtle cone. The smooth curve is from Baylor et al.' s theory for cone adaptation. From Baylor et aL (1974). VISUAL ADAPTATION AND RETINAL GAIN CONTROLS 327 deviation at the high backgrounds might be due to pigment bleaching, which is not included in their theory. Cones also speed up their response with background and for turtle cones, Gc ~" (tpeak) 4, that is the gain of a cone decreased like the fourth power of its time to peak for the response to a brief flash. The total amount of "speeding up" of the cone response is about a factor of two over a range of backgrounds in which the gain drops by a factor of two hundred. The speeding up of rod and cone responses with the change in gain is illustrated in Fig. 54 (Baylor and Hodgkin, 1974; Baylor e t al . , 1980). Steady illumination affects different parts of the cone's response in different ways. This is illustrated in Fig. 55 from Baylor and Hodgkin (1974). The I I I 0 _o --1 impulse response of the cone, i.e. the response to a brief, weak flash of light in the linear range, changes as the steady background increases. The gain for later parts of the impulse response drops, but the gain for the initial rising phase of the response is relatively unaffected by steady backgrounds. The peak of the response becomes smaller at higher backgrounds and occurs at earlier times after the flash, but the rising phases of impulse responses at several levels of background superimpose. Although this feature may not be completely convincing in Fig. 55, it is true. Recent preliminary experimental results involving the measurements of temporal frequency responses in turtle horizontal cells and cones are consistent with the resistance of the rising phase of the impulse I 1 . 6 2 . 2 - - 2 . . y- O - , I I 1 . 8 2 . 0 l o g t m a x E E O O f ~ 100 - 5 0 - 2 0 - 1 0 - 5 . 0 - 2 . 0 - 1.0 0-5 0.2 0.1 0-1 I I I 0-2 0. 5 1.0 2.0 f'a~ak (see) I I 5.0 10 FIG. 54. The speeding up of cone and rod responses in the turtle and toad retinas. As mean illumination increases, the gain and the time to peak decrease. In the left panel, the log of the gain is plotted against the log of the time between the peak response to a brief flash and the onset of the flash, for turtle cones (intracellular recording). The gain is plotted against peak time for toad rods (suction electrode recording) in the right panel. For about two log units log tma x is approximately proportional to log gain (here denoted SF). In the left panel, the different symbols are for different red cones, eight in all. On the right are graphed data points from seven different rod outer segments. Note that the cone time to peak stops decreasing when the gain drops below 0 log. The units of the left ordinate are log[#V (quanta gm-~)-l]; on the right the units are picoamps (quanta ~m-2)-L On the left the units of the abscissa are log[ms]; on the right they are simply s. From (left) Baylor and Hodgkin (1974), and (right) Baylor et al. (1980). 328 R. M. SHAPLEY AND C. ENROTH-CUGELL r e s p o n s e t o a d a p t a t i o n ( T r a n c h i n a , Go r d o n a n d S h a p l e y , u n p u b l i s h e d ) . m s e c 0 1 O 0 2 0 0 3 O 0 I I I 3 0 - 10 0 4"52 ~ ~ " " "; 2 I - E o 1 0 ~. 0 - ~ > 0. 15 0- 10 0. 05 5.79 1 ~ I I oo 200 300 msec FIG. 55. Effect of increasing illumination of conditioning step on the time course of the flash response in a turtle cone (intracellular), The flash was I1 ms in duration, and was applied 1.1 s after the onset of a conditioning step of illumination. The horizontal axis is the time after flash onset. The ver t i cal axes are all in uni t s of gai n: /aV/photoisomerization, which is equivalent to taV/quantum effectively absorbed. The numbers near each curve are the illuminations of the conditioning steps, in units of log[effective quanta(cone s)-q. These data are from a red cone. The stimulus was a 150/am white spot. From Baylor and Hodgkin (1974). 5 . 2 . P h o t o r e c e p t o r s w h i c h S a t u r a t e b u t d o n o t A d a p t T h e be s t s t u d i e d e x a mp l e s o f r e c e p t o r s wh i c h s a t u r a t e wi t h o u t a d a p t a t i o n a r e t h e r o d s o f t h e mu d p u p p y s t u d i e d b y No r ma n n a n d We r b l i n ( 1974) . F i g u r e 56 s h o ws t h e i r r e s ul t s . I t c a n be s e e n t h a t t he r o d o p e r a t i n g c ur ve s s hi f t t o t he r i ght a ve r y l i t t l e wi t h i n c r e a s i n g b a c k g r o u n d , b u t ma i n l y c o l l a p s e d u e t o s a t u r a t i o n . S i mi l a r r e c e p t o r b e h a v i o r is i mp l i e d b y t h e h o r i z o n t a l cel l r e s ul t s o f Wi t k o v s k y ( 1967) i n c a r p a n d Na k a et al. ( 1979) i n c a t f i s h . Re c e n t l y , Nu n n a n d Ba y l o r ( 1982) h a v e r e p o r t e d t h a t r o d s i n t h e mo n k e y Macaca fascicularis ma i n l y s a t u r a t e wi t h o u t a d a p t i n g . Th e r e is s u p p o r t i v e e v i d e n c e f r o m ma s s r e c e p t o r p o t e n t i a l r e c o r d i n g i n t h e r a t t h a t a t l e a s t s o me ma mma l i a n r o d s s a t u r a t e wi t h o u t a d a p t i n g ( P e n n a n d Ha g i n s , 1972; Gr e e n , 1973). Th i s is a l s o c o n s i s t e n t wi t h t h e v e r y s ma l l a mo u n t o f a d a p t a t i o n i n t h e a - wa v e o f t he r a t ' s E R G, t h o u g h t t o be d e t e r mi n e d b y r e c e p t o r a n d h o r i z o n t a l cel l r e s p o n s e s ( Do wl i n g , 1967), a n d b y ma s s r e c e p t o r p o t e n t i a l r e c o r d i n g i n t h e c a t e ye i n wh i c h t h e i n n e r n u c l e a r l a y e r a n d t h e g a n g l i o n cel l l a y e r we r e r e n d e r e d a n o x i c b y o c c l u d i n g t h e o p h t h a l mi c a r t e r y ( S a k ma n n a n d F i l i o n , 1972). S a k ma n n a n d F i l i o n a l s o s h o we d t h a t t h e g a i n o f t h e i s o l a t e d r o d r e c e p t o r p o t e n t i a l wa s E = 2 Q ~ 0 - I , , I ' : 0 4 Fro. 56. Rod operating curves on backgrounds. Plotted is the magnitude of the rod response (recorded intracellularly) measured as the difference between the peak of the response and the steady polarization just before the response. The response to test flashes brighter than the background are shown below the horizontal axis, and to test flashes dimmer than the background above the axis. The stimuli were diffuse 2 s flashes. The background illuminations can be read as the test stimuli which produced zero response, i.e. the intersection of the curve with the horizontal axis. Circles are from a rod in a normal retina, and squares are from a rod in a retina treated with aspartate. The units of the horizontal axis, which indicates the test flash illumination, are as follows: 6. l0 s quanta(522 nm) (cm 2 s) -1 when log I = 0. From Normann and Werblin (1974). V I S U A L A D A P T A T I O N A N D R E T I N A L G A I N C O N T R O L S 329 I 0 ' " ~ 10 4 "0 i 0 I "E I 0 ' I 0 Dotk I 0 0 ( a) E ._E MO) f I I I I 0 I 0 z IO s 104 I 0 " I 0 " Adapt i nt ensi t y ( t r ol ands) ( b ) / / / / / o~ ~3 / / / / / ~5 02 , / / / . 4 .~o,,I.,.~ I I I I I I I IO I 01 J O I IO* IO I IO e IO v Test i nt ens i t y ( t r ol ands) FIG. 57. Adaptation and saturation in the monkey cone receptor potential (extracellular recording of a mass response). (a) The retinal illumination required to reach a criterion response in the monkey's late receptor potential is plotted against background illumination. The criterion was 10/aV. Also shown are human psychophysical thresholds under the same stimulus conditions. In both cases, the stimulus was a yellow (580 nm) light presented on a yellow background. The background was on steadily for at least 5 min before test flashes were presented. The test stimuli were 150 ms in duration, presented in the fovea. The axes are given in photopic td. (b) Stimulus - response curves under dark adapted conditions and for five background levels for the monkey late receptor potential. The adapting and test fields were both 1 mm spots, centered on the fovea. The adapting illuminations are indicated in the figure in log td(photopic). The data are the mean values from six animals. They have been normalized for between animal comparisons, as per cent of the maximum response for each animal. From Boynton and Whitten (1970). a p p r o x i ma t e l y c o n s t a n t ove r s e ve r a l l og uni t s whe n g a n g l i o n cel l g a i n was d r o p p i n g l i ke We b e r ' s La w. Th u s , ma mma l i a n r o d s a n d t e l e os t c one s s e e m t o s a t u r a t e wi t h o u t a d a p t i n g . Th e a n a t o mi c a l l y l a r ge r r ods a n d cones o f a mp h i b i a n s a n d r ept i l es d o a d a p t t o l i ght . F a i n (1976) ha s s ugge s t e d t he ne c e s s i t y f or a d a p t a t i o n i n t he mo r p h o l o g i c a l l y l a r g e r r e c e p t o r s be c a us e o f t hei r l a r ge r q u a n t u m c a t c hi ng ar eas , a n d c o n s e q u e n t o v e r l o a d i n g at l owe r l evel s o f r e t i na l i l l u mi n a t i o n . We ha ve p u t o f f a d i s c u s s i o n o f ma mma l i a n c one s t o l as t b e c a u s e t he e vi de nc e a b o u t t h e m is i n d i r e c t a n d s o me wh a t e q u i v o c a l . Fi g u r e 57 f r o m Bo y n t o n a n d Wh i t t e n (1970) s h o ws , i n t he r i ght - h a n d p a n e l , r e s p o n s e - l og I c ur ve s o f t he i s o l a t e d c o n e r e c e p t o r p o t e n t i a l r e c o r d e d wi t h g r o s s e l e c t r o d e s i n t he f o v e a o f a mo n k e y ' s eye i n whi c h t he o p h t h a l mi c a r t e r y h a d be e n b l o c k e d . The s e cur ves ma y be a c c o u n t e d f or s ol el y i n t e r ms o f a mo d i f i e d Na k a - Ru s h t o n r el at i on. On l y s a t ur a t i on, c a l l e d b y B o y n t o n a n d Wh i t t e n " r e s p o n s e c o mp r e s s i o n , " was us ed t o d r a w t he cur ves t h r o u g h t he p o i n t s whi c h wer e a ve r a ge s f r o m r es ul t s o n si x mo n k e y s . On t he l ef t , d a t a f r o m a n o t h e r mo n k e y a r e seen t o f o l l o w We b e r ' s La w. As we k n o w f r o m o u r p r e v i o u s d i s c u s s i o n , i n S e c t i o n 1 . 2 . 2 . , s a t u r a t i o n a n d We b e r ' s La w a r e i n c o mp a t i b l e , a n d i n f act u p o n i ns pe c t i on, Fi g. 57(a) is n o t c ons i s t e nt wi t h Fi g. 57( b) . I f one c o n s t r u c t s a n i n c r e me n t a l g a i n vs b a c k g r o u n d c ur ve f r o m t he d a t a i n Fi g. 57( b) , b y c h o o s i n g a c o n s t a n t r e s p o n s e c r i t e r i o n , one o b t a i n s a c u r v e q u i t e u n l i k e Fi g. 57( a) ; i t is mo r e s h a l l o w at l ow b a c k g r o u n d s a n d s t e e pe r at hi gh b a c k g r o u n d s . I t is c o n s i s t e n t wi t h a n e x p l a n a t i o n i n t e r ms o f s a t u r a t i o n . So we d o n ' t k n o w whi c h c one r e c e p t o r p o t e n t i a l ( t hos e i n Fi g. 57( a or b) is r e p r e s e n t a t i v e o f ma mma l i a n c one s . Mo r e r e c e nt i n v e s t i g a t i o n o f t he s a me p r e p a r a t i o n ( Va l e t o n a n d v a n No r r e n , 1983) i ndi cat es t ha t a n a u t o ma t i c ga i n c o n t r o l r a t he r t h a n s a t u r a t i o n is t he p r e d o mi n a n t f a c t o r i n c o n t r o l l i n g ga i n i n p r i ma t e cones . But t hi s is a n i mp o r t a n t t opi c whi c h wi l l u n d o u b t e d l y r ecei ve mo r e a t t e n t i o n . 6. T HE OR I E S OF R E T I NAL GAI N C ONT R OL AND T HE DE T E R MI NANT S OF VI S UAL S E NS I T I VI T Y We ha ve d e v o t e d t hi s c h a p t e r up t o t hi s p o i n t t o t he q u e s t i o n s " wh y " a n d " wh a t " c o n c e r n i n g t he c o n t r o l o f r e t i n a l ga i n a n d vi s ua l s ens i t i vi t y. Ho we v e r , t o d e a l wi t h t he q u e s t i o n " Ho w" i t is d o n e , we wi l l at t hi s p o i n t r evi ew t he or i e s o f r e t i na l 330 and visual f unct i on. Theori es of light adapt at i on can be divided i nt o t wo categories: (a) t heori es of t he retinal gain cont r ol and (b) t heori es of how t he gain cont rol and noise combi ne t o det ermi ne visual sensitivity in psychophysi cal or behavi oral experi- ment s. These will be discussed in t ur n. 6. 1. Theories of the Retinal Gai n Control All t heori es of ret i nal gain cont r ol share t he common f eat ur e t hat t hey at t empt t o account f or a basic nonl i neari t y of vision, namel y t hat t he gain and time course of responses t o time varying stimuli are dependent on t he mean level of i l l umi nat i on, or t he backgr ound i l l umi nat i on, t o which t he t i me varyi ng stimuli are added. The nonl i neari t y of light adapt at i on is a gentle one, allowing a linear range of vision ar ound t he operat i ng poi nt of the mean level. Anot her way of saying this is t hat t he mean level must be changed a lot t o have a big effect on gain and time course. For i nst ance, in t he Weber Law range, in or der t o reduce t he gain t o hal f its value at one mean level, the value of the mean must be doubl ed. Thi s is a large change in mean level. Thus for normal vision in the real world of (Weber) cont rast s, which are in t he range 0 - 0 . 5 , t he response of the visual system will not be t hrown into a very nonl i near range by t he processes of retinal light adapt at i on. In t he model s t o be consi dered, adapt at i on acts like a nonl i near negative feedback: a negative feedback because increasing t he out put reduces the gain of earlier stages; nonl i near because t he f eedback signal is not added or subt ract ed but is used as a gain cont rol and a cont rol l er of the time const ant s of response. R. M. SHAPLEY AND C. ENROTH- CUGELL gain and time course of response in phot or ecept or s of t he horseshoe crab, L i m u l u s p o l y p h e m u s . We have excl uded the t opi c of light adapt at i on in i nvert ebrat es f r om our review, f or obvi ous reasons of length cont r ol . The phot or ecept or s of inverte- brates are quite di fferent f r om those of vertebrates; most of t hem depol ari ze in response t o increases in i l l umi nat i on r at her t han hyper pol ar i ze as do verte- brat e rods and cones. However , t he F u o r t e s - Hodgki n ( FH) model has crossed phyl ogenet i c boundar i es. Sui t abl y modi f i ed, it has been used by Bayl or e t al . (1974a, b) and Bayl or e t al . (1980) t o analyze t he gain and time course of turtle cones and t oad rods, respectively. It is also rel at ed t o model s devel oped f or t he cat ret i na by Enr ot h- Cugel l and Shapley (1973a) and Shapley and Victor (1981), and t o a model f or human vision pr oposed by Sperling and Sondhi (1968), and t her ef or e is a useful starting poi nt . The FH model is composed of a sequence of stages of t empor al i nt egrat i on connect ed in series, as di agrammed in Fig. 58. Each stage is represented as an RC circuit in t he original model . However , a t heor y with t he same f or mal st ruct ure can be devised for stages which are chemical reaction states reached by repeat ed catalytic react i ons (Borsellino e t al . , 1965). The t i me course of t he response of stage j in t he FH cascade is descri bed by a di fferent i al equat i on which says t hat t he bui l dup of response (or s u b s t a n c e ) j is pr opor t i onal t o a react i on rat e const ant a~ times t he amount of response (or substance) in stage j - I , minus the decay f r om s t agej which is equal t o anot her rate const ant a2 times t he amount of subst ance in state j . 6 . 1 . 1 . FUORTES AND HODGKIN' S MODEL FOR PHOTORECEPTORS Fuort es and Hodgki n (1964) originally developed a t heoret i cal model t o account f or t he rel at i on of d y j ( t ) / d t = a~v j _~( t ) - az yj ( t ) . (25) In this scheme onl y t he first stage is driven by light, d y ~ ( t ) / d t = Y . I ( t ) - az. vl(t) (26) V o V 1 V , _ I Vr Vn Inp Out put C O 0 C - ' - ' - ~ I~ C R o FIG. 58. Di agr am of t he Fuor t es - Hodgki n ( FH) model . The el ement s l abel ed/ a r epr esent i sol at i on st ages wi t h gai n g. Li ght a da pt a t i on is a s s u me d t o af f ect onl y t he val ue o f R. Fr o m Fuor t es a nd Hodgki n (1964). VISUAL ADAPTATION AND RETINAL GAIN CONTROLS BuiLd up o f p a r t i c l e s 331 ~ ( - t ) m , , , ~ l ~ - y , Li ght Q u ( i (~ 1 Y2 D Y3 l l l l l ] l ~ Y4 ]1 Y5 " Removal of par t - i ct es Y 6 ( =Z, ) m ~ m d l l ~ Z ~ Open channels ChanneLs cl osed by Z , K 1 2 ~ " ~ K23 K34 Z2 . Z3 K21 K32 I Z 4 FIG. 59. This is a diagram of the Bayl or- Hodgki n- Lamb (BHL) model. The blocking substance is denoted y~ or z,. Y is used for build-up, z for destruction of the blocking substance. Note that this scheme is fundamentally different from the FH model (Fig. 58) in that the rate constant for production of each y stage is the same as for its decay, except for the first and last stages. In the FH model, the rate constant for growth of each stage is different from the rate of decay, and only the decay rate is subject to change by adaptation. Here adaptation only works by changing the rate constant of decay from state z,. From Baylor e t al . (1974). where Yis t he gai n of t he initial t r ansduct i on f r om fight t o y and I(t) is the fight stimulus. The last stage, t he t i me evol ut i on of whi ch is given by y, ( t ) , is pr esumed in t he model t o cont r ol t he membr ane conduct ance and t her eby t he el ect rophysi ol ogi cal r esponse t o l i ght of t he phot or e c e pt or . In compari ng equat i ons (25) and (26) with Fig. 58, one shoul d make t he i dent i f i cat i on a , = ~ / C and a2 = 1/ RC. Thi s model was devel oped t o expl ai n first of all t he many stages of t empor al i nt egr at i on in t he process of visual t r ansduct i on, as refl ect ed in t he slow rise t o t he peak of t he i mpul se response of phot or ecept or s. It does this ver y well. In t urt l e cones and t oad r ods t he model indicates t hat t he number of stages is f r om f our t o six. Fr om t he experi ment al evidence of the t oad r od phot ocur r ent recordi ngs ( Bayl or et ai . , 1980) pr obabl y all of t he t empor al i nt egr at i on is in t he out er segment of t he phot or ecept or . The FH model as r epr esent ed in equat i ons (25) and (26) is a l i near model . That is, t he sum of t wo i nput s L( t ) + 12(0 yields a response y. ~(t) + y. 2(t) whi ch is t he sum of t he responses t o t he t wo i nput s present ed al one - - t he pri nci pl e of Superposi t i on. However , Fuor t es and Hodgki n (1964) descri bed a modi fi cat i on of t he model which woul d help explain observed depart ures f r om linearity, in particular t he change of gain wi t h changes in mean level of i l l umi nat i on which is t he basis of phot or ecept or light adapt at i on. They suggested t hat t he rat e const ant of decay, a2, was i ncreased by increase of st eady level of t he out put stage y, but t hat t he rat e const ant f or bui l dup al was i ndependent of i l l umi nat i on. Thus, t he scheme t hey pr oposed was: a2 = ao2(1 +f 0' . ) ) (27) where t he f unct i on fLy.) is chosen t o fit t he dependence of gain on mean level. The poi nt of t he FH model f or adapt at i on is t hat t he gain and t i me course of response are t her eby l i nked, whi ch may be seen as follows. Say T2 = 1/a2. The i mpul se response (t hat is, t he response t o a br i ef fl ash in t he linear range) of t he simple n-stage FH model is, [cf. equat i on (7) of Fuor t es and Hodgki n, 1964] Q . A . T2"-'. (t l T2)"-' e -'IT2 (28) where Q is t he amount of light in t he fl ash and A is a const ant pr opor t i onal t o (al)". The time t o peak of this i mpul se response is ( n - 1)T2 and t he gai n is pr opor t i onal t o /'2 "-1. I f onl y T2 is r educed by adapt at i on, as assumed by Fuor t es and Hodgki n, t he t i me t o peak will short en and t he gain will drop, much mor e steeply t han t he t i me t o peak. 332 Ther ef or e, in t he ori gi nal FH model , t he rel at i on bet ween gai n and t i me t o peak of t he i mpul se response is a power l aw where t he exponent is one less t han t he numbe r of stages whi ch are af f ect ed by adapt at i on. R. M. SHAPLEY AND C. ENROTH-CUGELL di fferent i al equat i ons f or the BHL model , up to the z~ stage is: 6. 1. 2. THE BAYLOR -- HODGKIN -- LAMB MODEL I n the t r eat ment of t urt l e cone responses, Bayl or et al . (1974b) modi f i ed t he ori gi nal FH model by setting dy, (t ) dt dyj(t) dt - I ( t ) - a y, ( t ) - - - a [Yj_l(t) - y~(t)] (31) a, = as = a = 1 / T . (29) dy. (t ) dz, (t ) Several ot her kinetic schemes were f ound t o fit the wavef or ms of cone responses bet t er (Bayl or et al . , 1974a) but t he modi f i ed FH model was chosen adequat e f or an expanded t heor y of adapt at i on and responses t oo l arge t o be in t he l i near range. It al so has a much si mpl er i nt er pr et at i on t han t he FH model as a chai n of chemi cal react i ons ( compar e Borsel l i no et al . , 1965, wi t h Bayl or et al . , 1974b). Fi gure 59 illustrates the model used by Bayl or et al. (1974b), which we will call t he BHL model . There are i mpor t ant differences between the BHL model and t he ori gi nal FH model . Since, in t he BHL model , t he rat e const ant s f or bui l dup and decay of t he y subst ance are t he same, t here is no change in t he gai n when t he val ue of t he rat e const ant s is changed, as t here is in t he FH model . Thi s can be seen by consi deri ng t he i mpul se r esponse of t he first n stages of t he BHL model : y , (t ) = Q B ( t / T ) " - ' e t / T (30) dt dt - a Y n ~( t ) - {K, 2 + v z , ( t ) } z , ( t ) + K21 A {1 + v Z2(t)} z2(t ) dz2(t) - g , ~ {1 + v z 2 ( t ) } z , ( t ) - dt {K,2 [1 + vZ2(t)] + K23}Z~(t). In t hese equat i ons v is t he aut ocat al yt i c const ant implicitly defi ned by the equat i on K,2 = /(,2 + vz2. A is t he equi l i br i um const ant of zx, z2 and is equal t o K, 2 / K2 , . Thi s is a compl i cat ed set of di f f er ent i al equat i ons and an i nt ui t i ve feeling of what is goi ng on is di ffi cul t . Some insight may be gai ned by exami ni ng a special case of interest: a weak, si nusoi dal l y modul at ed i ncr ement which pr oduces a r esponse in t he l i near range. In this case, where T = 1 / a , and B is a const ant . I n this case t he peak of Yn is at ( n- 1) T and t he magni t ude of t he peak is i ndependent of T. Gai n cont r ol cannot be achi eved in this model si mpl y by an i ncrease of t he rat e, a. Ther ef or e, t he BHL model has t o have a new f eat ur e, an ext r a pat hway f or decay of t he fi nal st age of t he cascade where t he decay rat e depends on t he level of i l l umi nat i on. Thi s is shown in Fig. 59 as t he pat hway denot ed z, , z,, z3, z,. z, is equi val ent t o Yn and is assumed t o be t he concent r at i on of " bl ocki ng par t i cl es" which lead t o closing of t he sodi um channel s in t he phot or e c e pt or ' s membr ane. The decay of z, t o the next state, z2, is subj ect t o a u t o c a t a l y s i s by the level of z2 present (see Fig. 59). Thus, t he syst em of / (t ) = Io + I, e i~t yj ( t ) = Yjo + Yil ei~' z~(t) = Z~o + z~le ~t. (32) We make t he a ppr oxi ma t i on t hat 0>>, and t hat Io, t he mean level of i l l umi nat i on which ma y al so be vi ewed as t he mean arri val r at e of light quant a, is l arge c ompa r e d t o t he rat e const ant s of t he cascade. Then we obt ai n [cf. equat i on (40) of Bayl or et al. (1974b)] f or the cone' s f r equency r esponse in t he light adapt ed range: VISUAL ADAPTATION AND RETINAL GAIN CONTROLS 333 z , , ( c o ) y . l ( c o ) I , ( o ~ ) I , ( c o ) where T = l / a (33 1 + A 1 + io~B2 B, = [g,2 + K2,]/[K23"K,2] Bz = 1/{J~2,[1 + vlo/(Kz3K,2)]} A = K,2/K2,. Thus, t he BHL model predi ct s no ef f ect of light adapt at i on in t he limit of high f r equency. Thi s is qual i t at i vel y in agr eement with t he measur ement s of t he cone' s i mpul se response by Bayl or and Hodgki n (1974). They f ound t hat t he rising phase of t he i mpul se response was not af f ect ed by backgr ound level. Thi s result is also qual i t at i vel y in agr eement with Kel l y' s (1972) psychophysi cal dat a on t he l i neari t y of high f r equency response in h u ma n vi s i on. Re c e nt di r e c t i n t r a c e l l u l a r measur ement s of t he f r equency responses of t urt l e hor i zont al cells also conf i r m this pr edi ct i on of t he BHL t heor y ( Tr anchi na, Gor don and Shapl ey unpubl i shed). Not i ce t hat t he f r equency response of t he cone consists of t hr ee t erms, t he first t wo of which are stages of t empor al i nt egr at i on whi ch are i ndepen- dent of t he mean level, Io. The onl y place where io enters in t he expressi on f or t he cone' s light adapt ed frequency response is in the time const ant B~ in t hi rd t er m of equat i on (33). Thi s t er m has t he f or m of t he f r equency response of a negative feedback l oop whi ch has a gai n A and t i me const ant B2. The t i me const ant of the feedback is what is affect ed by mean level in t he BHL model . As Bayl or et al. (1974b) demonst r at ed, this coul d account qual i t at i vel y f or t he ef f ect of backgr ounds on t he gai n and t i me course of t he cone' s i mpul se response [the Four i er t r ans f or m of t he f r equency response in equat i on (33)]. However , as i ndi cat ed above in Fig. 53, t he predi ct ed gain vs backgr ound curve is st eeper t han t he real dat a, while t he dependence of t he t i me t o peak of t he response on mean level is t oo shallow. In a l at er paper , concer ni ng t oad rods, Bayl or et al. (1980) pr oposed t hat f eedback t o earlier stages of processi ng, r at her t han t o j ust t he last stage as in t he BHL model , woul d pr ovi de a bet t er fit t o t he dat a. However , t he t heor y of such a system has not been anal yzed and woul d cert ai nl y be mor e compl i cat ed t han t he BHL model . A very i mpor t ant consequence of t he BHL model is t hat t he amplitudes of responses t o high t empor al frequenci es of modul at i on are unaf f ect ed by mean level. This can be seen in equat i on (33) in t he limit as co~oo. Then t he onl y t er m whi ch cont ai ns Io appr oaches t he val ue uni t y, and it does so at l ower values of co t han t he ot her t er ms appr oach zero. 6.1.3. ENROTH-CUGELL AND SHAPLEY'S MODEL OF ADAPTATION IN GANGLION CELLS Anot her scheme f or obt ai ni ng Weber ' s Law behavi or linked t o dynami c changes is di agr ammed in Fig. 60 ( Enr ot h- Cugel l and Shapl ey, 1973a). In this model f or t he rod-dri ven retinal net wor k of t he cat , t he r od signal is subj ect ed t o a mul t i pl i cat i ve gai n cont r ol at t he level of t he r o d - b i p o l a r synapse. In t he original model it was pr oposed t hat t he f eedback signal which mul t i pl i ed t he r od signal was pr oduced by t he hor i zont al cell. Lat er wor k reviewed above indicates r at her t hat t he f eedback signal may arise in bi pol ar cells. In any case t he f or mal expressi on of t he model is: r(t ) = P(t)/exp{H(t)/Htrig} P( t ) = f ~ ( t - t ' ) p ( t ' ) dt ' p ( t ' ) = g - ( t ' / r , ) 3 exp { - t ' / x p } H( t ) = f ~ r ( t - t ' ) h ( t ' ) dt ' h ( t ' ) = exp ( - - t ' / TH) . (34) Wher e r(t ) is t he r od signal t o t he bi pol ars, H( t ) is t he hor i zont al cell pot ent i al (now t hought of as t he bi pol ar cell' s pot ent i al ) and T. is t he t i me const ant of t he f eedback neur on. Htri, is a critical level this signal must exceed, Tp is t he t i me const ant of t he r od, I ( t ) is t he light stimulus, P( t ) is t he phot ocur r ent of t he r od, ~ is t he gai n in t he dark. The f r equency response of this model has been wor ked out and it is (Enrot h-Cugel l and Shapl ey, 1973a): 3 3 4 R . M . SHAPLEY AND C. ENROTH- CUGELL r,(o) - g-exp { - r o / H t r i g } l,((o) (35) n : E p j ( t ) j = o f l T t 4 ~ 1 t + i o) + l ' n t rig 1 + i m ' r n . ~ where ro is the st eady state value of r ( t ) . In t he model , ro appr oxi mat el y increases like log Io f or g-Io>>H~r~g. Two feat ures are not ewor t hy. The general " s h a p e " of t he f r equency r esponse resembles t hat of the BHL model ' s f r equency response, equat i on (33). However , t he effect of mean level on dynami cs is on t he gain of t he feedback t er m r o H t r i g , rat her t han on the t i me const ant r,,. Mor eover , t he gain of high f r equency responses in this model does dr op at high backgr ounds by t he f act or e x p { - r o / H, ng}, t hat is appr oxi mat el y r eci pr ocal t o Io in t he hi gh backgr ound limit. This predi ct i on is not correct f or single cone responses, but it has not been tested in t he system for which this model was designed, t he cat ret i na in t he scot opi c range. Measur ement of t he t empor al f r equency response of cat retinal gangl i on cells at several di f f er ent backgr ound or mean levels woul d be a crucial test of t he di fferent predi ct i ons of t he BHL model and t he E n r o t h - Cugell - Shapley model. Prel i mi nary experiments of this t ype appear t o conf i r m a model of t he BHL t ype and t o excl ude mul t i pl i cat i ve model s like t he E n r o t h - Cu g e l l - Shapl ey model (Shapl ey e t a l . , 1983). 6.1.4. THE CONTRAST GAINCONTROL MODEL The cont r ast gai ncont r ol adj ust s t he t i me course of response cont i ngent on t he average level of cont rast rat her t han simply cont i ngent on t he mean flux. Shapl ey and Vi ct or (1981) pr oposed a t heor y f or t he way in whi ch t he t i me course of response was changed by cont rast , a t heor y which resembles the BHL model . Thei r model is di agrammed in Fig. 61. Ther e are n stages of t empor al i nt egr at i on and one negative f eedback stage. The t empor al fre- quency response of this system is: n ( t ) = ~ r j ( l ) ( t l ~ "-I r(~ ;; r e l l ( t ) I e x H o t c e l t J FIG. 60. A model for adaptation in the cat retina. The p(t)s are the photocurrents in individual rods. The amount of signal transmission from rods to bipolar and horizontal cells is called r(t), with subscripts for each rod. H ( t ) is the horizontal cell potential due to the sum of its inputs. An exponential function of the horizontal cell's potential is what controls the sensitivity of receptor transmission. Thus, r(t) =p(t )/exp{H(t )}. From Enroth-Cugell and Shapley (1973a). f 1 r~(o)) A 1 - + 11(o) + io 1 +- iwT (36) Shapl ey and Vi ct or (1981) f ound t hat increase of cont rast affect ed onl y the rat i o K / t H and had onl y negligible affects on A , x~ or n. Formal l y, equat i on (36) resembles equat i on (33) f or t he BHL model . The ef f ect of cont r ast is exert ed on t he same t erm, t he f eedback t erm, and in a r oughl y similar way. Whet her this similarity of the consequences of these model s f or i ncrement al responses has any deep VISUAL ADAPTATI ON AND RETI NAL GAI N CONTROLS 335 meani ng about mechani sms, it does reveal t hat functionally t he t wo processes seem t o af f ect t i me cour se of responses in similar ways. Not e t hat in none of t hese model s is response simply a gai n f act or times an unadapt ed signal. Rat her , in each case, adj ust ment of gai n and adj ust ment of t i me const ant s, a n d / o r st rengt h of negative feedback, appear t o be necessary, in or der t o expl ai n t he associ at i on of gai n and t i me cour se in t he dat a f r om visual neur ons. The response of retinal neur ons is t hus a functional of t he light now and t he past hi st or y of i l l umi nat i on. The ret i nal f unct i onal is under t he cont r ol of mean level and mean cont r ast . FI o. 61. A di a gr a m o f a mode l f or t he way t he c ont r a s t ga i nc ont r ol modi f i es fi rst or der r e s pons e s o f ga ngl i on cells. I c is t he c ont r a s t si gnal . The mode l cons i s t s o f a ga i n s t age A, una f f e c t e d by c ont r a s t , N L st ages o f l ow pa s s f i l t er i ng, and one stage of high pass filtering. Each low pass stage has a time constant "r,. The high pass stage is a feedback loop with one stage with a time constant TH, and with feedback gain K. Contrast mainly affects K and T H. From Shapley and Victor (1981). 6. 2. Ret i nal Gai n and Vi sual Sensi t i vi t y The psychophysi cal consequences of retinal gain cont r ol , and t he i mpl i cat i ons of psychophysi cal results about ret i nal gain cont rol s, requi re f or t hei r i nt er pr et at i on a t heor y linking t he psychophysi cs and neur ophysi ol ogy of vision. In a previ ous publ i cat i on (Enrot h-Cugel l and Shapley, 1973b) we suggested a sket ch of such a t heor y and here will make it explicit. The aims of t he t heor y are: (i) t o show how gain and noise i nt eract in det ermi ni ng sensitivity in r od vision; (ii) t o show how t he dependence of gain setting on receptive field size can be used t o explain t he observed spatial effect s on t he psychophysi cal laws of light adapt at i on (Barl ow, 1957, 1965; van Nes and Bouman, 1967; Koender i nk et al. 1978). In our proposed model , psychophysical t hreshol d is pr esumed t o be gover ned by a signal-to-noise rat i o. The signal is pr esumed t o be t he neural response of a population of ret i nal gangl i on cells; t he noise is t he vari abi l i t y of firing in t hose cells. The signal is subj ect t o gain cont r ol which is supposed, in t he t heor y, t o behave in t he way t hat t he retinal gain cont rol has been observed t o behave in cat ret i nal gangl i on cells. We pr esume t hat t here are a few (at least) ret i nal spatial channel s at any par t i cul ar ret i nal locus. We post ul at e t hat r od and cone signals are i ndependent , in t hei r noise gener at i on and in t hei r gain cont r ol , unt i l late in the retinal stages of signal processing. Rod and cone pat hways must have i ndependent dar k noises, and di f f er ent dar k adapt ed gains. Ther e are t wo mai n f act or s which det er mi ne t he sensitivity of gangl i on cells: gain and noi se (Barl ow and Levi ck, 1969; Rose, 1948). Fr om wor k cited above, t he gai n of a gangl i on cell' s response t o a step of light on a backgr ound in t he scot opi c range, G R is: G R = GRO/(1 + FB/FRo) P (37) where GR0 is t he dar k adapt ed gain, FB is t he backgr ound fl ux effect i vel y absor bed, FRO is t he cri t eri on amount of flux which must be exceeded t o t ur n on t he gain cont r ol of t he r od pat hway, and P is an exponent , usually measured t o be 0.9, which we will appr oxi mat e t o 1 (Cl el and and Enr ot h- Cugell, 1970; Enr ot h- Cugel l and Shapl ey, 1973a). The noise will be t he dar k noise summed t oget her with t he backgr ound fl ux over t he recept i ve field. Thus the variance o 2 of the noise f r om the r od pat h- way in a gangl i on cell' s act i vi t y will be oR 2 -- [ a d "(FB + FRo)] FRD = IRD fl l t FB = IB " A, (38) Wher e IB is t he backgr ound ret i nal i l l umi nat i on, A, is t he summi ng area of t he gangl i on cell cent er, FRD is t he dar k noise in t he r od pat hway (in equi val ent quant a s-X). It is assumed t hat t he 336 R. M. SHAPLEY AND C. ENROTH-CUGELL psychophysical threshold will be reached when the cell's si gnal - noi se ratio reaches 4. Thus, if Is is the retinal illumination of a light stimulus and As is its area in squared degrees, then: In the scotopic range we can make the approxima- tion that FB < < FCD < Fc0 ( 44) 4 = I s " A s " ( GR) / OR (39) Solving for the threshold stimulus illumination, one obtains: Thus, Gc ~ Gco and, oc 2 ~ Gco 2 F c o . (45) 1 s = 4 OR/ [ A s GR] = 4 . [GR = ( F . + F R D ) ] ' / 2 / [ A s I GR o / ( 1 + F B I F . o ) } ] . (40) In order to obtain the correct value for the (Weber) scotopic contrast sensitivity, which approximately equals 10, one can calculate that Oc = Gco FCD ' ' 2 = 0.02 FRO (46) When FB, the background flux, is small, the threshold is approximately constant depending only on the values of the dark adapted gain GR0 and the dark noise FRD. However, as FB becomes large compar ed to FRD, the t hreshol d becomes proportional to Fa 1/2. That is, it follows the square root law. This happens even though the retinal gain is dropping like 1/(1 + FB/ FRo ) , because the variance of the noise is dropping also, being proport i onal to FB-' when FB becomes large compared to FRO. In order to obtain Weber' s Law one must put a clamp on the noise; it cannot keep declining in variance as the retinal gain declines. Indeed, as Barlow and Levick (1969) and Derrington and Lennie (1982) have found, the noise at the ganglion cell does not decline as the gain declines. How can this be so? One idea is that noise from the c o n e p a t h w a y is added into the ganglion cell, and is more or less unaffected by changes of gain in the rod pathway. Thus, if we change equation (39) to include cone pathway noise with variance Oc 2 we obtain 4 = I s A s " GR/ ( OR 2 + OC2) ' / 2 (41) where OC 2 = Gc2[FB + FCD ] ( 42) and in analogy with the rod pathway: Gc = Gc0 (1 + F B / F c o ) . (43) Since we found FRO tO be equivalent to 3000 phot on absorptions s-' , this would mean the cone dark noise would have to produce a variance of about 60 phot on events s-' at the ganglion cell, a 106 / ~ t 0 I deg . J . / / 1 0 4 0 5 o k - 1 0 3 2 x l O -5 2 x l O -4 2 x i O 3 2 x l O -2 2 x l O -I 2 2 0 Tr ol onds I 0 I 02 103 I 04 105 106 107 I B G ( q / d e g 2 s e c ) FXG. 6 2 . T h e p r e d i c t i o n s o f a t h e o r y o f l i g h t a d a p t a t i o n concerning the effects of stimulus area on the slope of the increment threshold curve. The curves are the predicted increment threshold behaviour for retinal ganglion cells which have the areas indicated in the figure. The psychophysical data plotted for comparison are: empty circles from Barlow's (1957) data for a 19 deg 2 target, filled circles for Barlow's (1957) 0.0077 deg 2 target, and crosses for Aguilar and Stiles' (1954) 64 deg z target. See text. VISUAL ADAPTATION AND RETINAL GAIN CONTROLS not - unr easonabl e number whi ch must be checked experi ment al l y in t he f ut ur e. Not i ce t hat , in this model , t he dar k noises and t he cri t eri on fluxes f or t he t ur ni ng on of t he gain cont r ol are distinctly di f f er ent . Thi s is in accor d with actual measurement s on ganglion cells (Enrot h- Cugell and Shapl ey, 1973a; Barl ow, 1977) whi ch show cl earl y t hat t he dar k light is t oo small t o account f or the behavi or of t he gain vs backgr ound relationship. A di fferent f undament al const ant , t en t o one hundr ed times l arger t han t he dar k noise must be hypot hesi zed. The poi nt of t he model is reveal ed in Fig. 62 whi ch shows predi ct i ons of t he model f or t he i ncrement t hreshol d curves of t wo types of cell, one with a small cent er and one with a l arger cent er. The cells' sensitivities are pl ot t ed on t he same coor di nat es as psychophysi cal dat a f r om Agui l ar and Stiles (1954) and Barl ow (1957) whi ch exhi bi t characteristic steep and shallow slopes, respectively. Cl earl y, t he model account s adequat el y f or t he t ransi t i on f r om square r oot t o Weber behavi or cont i ngent on t arget size. It provi des a way t o under st and why this t r ansi t i on is a f unct i on of t he square of the spatial frequency of a sinusoidal target (van Nes and Bouman, 1967). It also provi des a f r amewor k with whi ch t o under st and why noise sometimes controls sensitivity and why, under ot her ci rcumst ances, t he gain det ermi nes sensitivity. 7. RETROSPECTI VE AND CONCLUSI ON We have considered t he " wh y " , the " wh a t " , and t he " h o w" of visual adapt at i on, and even at t emp- t ed t o say " wh e r e " and " wh e n " it t akes place. At t he out set we demons t r at ed how aut omat i c adj ust ment of ret i nal gain woul d cause bri ght ness c o n s t a n c y u n d e r d i f f e r e n t c o n d i t i o n s o f illumination, by maki ng retinal responses dependent on cqnt rast . Then we exami ned what happens t o t he visual per f or mance of humans under di f f er ent condi t i ons of backgr ound or mean level of i l l umi nat i on. In this section, we poi nt ed out t hat " n o i s e " f r om t he stimulus may somet i mes limit per f or mance, and t hat at ot her times ret i nal adapt at i on det ermi nes t he limits of sensitivity. The vi sual r es pons es o f r et i nal ga ngl i on cel l s, particularly cat ganglion cells, as a funct i on of mean 337 or backgr ound level, were consi dered next. In these cells, gain cont r ol seems t he domi nant f act or in limiting t hei r capaci t y t o detect stimuli as t he mean level of i l l umi nat i on varies. The rel at i on of this wor k t o human vision, where bot h gain and " n o i s e " seem significant, is an open quest i on f or f ut ur e research. In seeking t o answer t he quest i on, " wh e r e " t he gain cont r ol is l ocat ed, we next consi dered what is known about adapt at i on in retinal i nt erneurons. Then we reviewed the t opi c of adapt at i on in phot orecept ors. Finally, we discussed some t heori es of how t he gain of ret i nal neur ons is cont r ol l ed, and how such a gain cont r ol mi ght cont r i but e t o visual per f or mance. Ther e are several neural gain cont rol s in t he ret i na. Surpri si ngl y, what evidence we have about t he t i me course of gain r educt i on af t er an increase in mean i l l umi nat i on, t oget her with t he change in response t i me cour se with change in gain, suggests t hat all these gain cont r ol s share t he pr oper t y of at t enuat i ng t he component s of response t o slow var i at i on of t he st i mul us, and spar i ng t he component s of t he response t o rapi d vari at i ons of t he stimulus light. However , t he di f f er ent gain cont r ol s do seem t o have di f f er ent spat i al i nt egrat i on areas, and this allows one t o distinguish t hem. Fut ur e wor k is requi red t o establish t he answer t o " wh e r e " in t he ret i na some of t he gain cont r ol s live. However , it will be even mor e i mpor t ant t o di scover or i nvent model s f or " h o w" t he gain cont rol s manage t o make t he ret i na except i onal l y sensitive t o cont r ast and yet insensitive t o large swings in t he mean level. A P P E N D I X 1 - - R E T I N A L N E U R O N S The details of ret i nal ci rcui t ry var y f r om species t o species al t hough many of t he general principles, at least in hi gher mammal s, are t he same. Much of what we have wri t t en about t he physi ol ogi cal mechani sms of ret i nal adapt at i on stems f r om experi ment s on t he cat ret i na and any f r ui t f ul consi der at i on of these mechani sms requi res some familiarity with retinal funct i onal connectivity. The bri ef comment s about retinal circuitry which follow draw heavily on t he detailed studies of the cat retina under t aken over t he past t en years or so ( f or 338 summar y see Sterling, 1983). I nf or mat i on in the vert ebrat e retina flows mainly in a r adi al di rect i on f r om the receptors, in the out er ret i na, t he r od and cone cells, t hr ough bi pol ars t owar ds gangl i on cells in t he i nner ret i na, with lateral interaction in bot h pl exi form layers, and with feedback within and bet ween layers. Increasing t he amount of light t hat falls on vertebrate receptors causes the inside of the receptor cell t o become mor e negative. That is, ver t ebr at e receptors respond t o light by h y p e r p o l a r i z i n g . Since t ransmi t t ers are released in response t o neur on depol ari zat i on it must be concl uded t hat vert ebrat e recept ors release t he maxi mum amount of t hei r as yet uni dent i fi ed t r ansmi t t er or t ransmi t t ers in complete darkness. Direct evidence for this has been obt ai ned by Schacher e t al. (1974) in the frog retina, and by Ripps e t al. (1976) in t he skate ret i na, using horseradi sh peroxi dase ( HRP) upt ake as a measure of synapt i c vesicle t ur nover . Rod signals leave t he r od cell' s di st ri but i ng end f or one t ype of bi pol ar cell while cone signals leave t he di st ri but i ng end of t he cone cells f or ot her bi pol ar cells. The st at ement t hat t he cat has r od bi pol ars and cone bi pol ars means onl y t hat at present t here is no a n a t o mi c a l evi dence t hat bot h rods and cones cont act t he same bi pol ar cell (Kolb e t al. 1981). This does not necessarily mean t hat rod and cone signals are strictly separated in the bi pol ar layer. For example, Nelson (1977) f ound t hat mixed rod - cone signals t ravel in some cone bi pol ars (see also Nel son, 1980). Structurally, t here is onl y one kind of r o d bi pol ar in the cat, one t hat receives signals in invaginations of the proxi mal surface of t he r od spherules. Ther e are two structurally different kinds of cone bipolars: the invaginating ones which cont act cones in pits in the proxi mal surface of cone pedicles while ot her cone bi pol ars onl y f or m superficial cont act s with the pedicle bases. These are t he " f l a t " cone bi pol ars. It seems t hat all r ecept or cells in t he cat do not " t a l k " to bi pol ars in exact l y t he same language because t he ul t rast ruct ural synaptic specializations are not t he same in the r od and cone pits as in t he cont act s bet ween t he base of t he cone pedicles and t he flat cone bi pol ar dendri t es (Sterling, 1983), suggesting t hat t he mode of signal transmission is di fferent in the t wo types of synapse. This is also consistent with the recent physiological R. M. SHAPLEY AND C. ENROTH-CUGELL wor k of Saito e t al. (1979) in t he carp ret i na which suggests t hat rods and cones af f ect f undament al l y di fferent ionic channels in their respective bipolars. Al t hough, as far as is known, all vert ebrat e recept ors r espond t o light by hyperpol ari zi ng, t he same does not hol d t rue f or all bi pol ar cells. This has been known since 1969 when Werbl i n and Dowl i ng showed t hat some mudpuppy bi pol ars hyper pol ar i ze whi l e ot her s depol ar i ze when i l l umi nat ed. Nel son and co-workers (Nelson and Kol b, 1982) have r ecor ded i nt racel l ul arl y f r om anatomically identified bipolars in the cat retina and t hey f ound t hat in t he cat ret i na t oo some bipolars hyperpolarize while others depolarize when the light level is increased. It is in t he i nner pl exi f or m l ayer ( I PL) t hat t he t hi rd or der neur ons, t he gangl i on cells, communi - cate with ot her ret i nal cells. Two feat ures of this connect i vi t y are of special interest. First, some ganglion cells' dendri t i c trees are l ocat ed within t he a-l ami na of t he I PL where t hey receive i nf or mat i on f r om t hose bipolars whi ch ext end onl y as far as t he a-lamina. Ot her ganglion cell dendritic trees branch out in t he b-l ami na of t he I PL and are fed by bi pol ars whose di st ri but i ng end ext ends t o t he b- l ami na. One of t he maj or recent achi evement s wi t hi n r et i nal a n a t o my - p h y s i o l o g y is t he demonst r at i on by Fami gl i et t i and co-workers (Famiglietti e t al . , 1977; Nelson e t al . , 1978) t hat when bi pol ars and gangl i on cells talk t o each ot her in l ami na-a of t he i nner pl exi f or m layer, t he gangl i on cell is of t he of f - cent er t ype, i.e. an i ncrement of i l l umi nat i on within t he cent er of the gangl i on recept i ve field (see Appendi x 2) causes it to become hyperpol ari zed. But when t he t wo talk to each ot her in l ami na-b t he gangl i on cell is of t he on-cent er t ype, i.e. it depol ari zes in response to increased illumination of the receptive field center. Sterling (1983) and his colleagues have f ound t hat at least t wo di fferent types of bi pol ar cells converge on each retinal gangl i on cell of t he mor phol ogi cal class called bet a (Boycot t and W~issle, 1974), which is equi val ent t o t he funct i onal class, X. Thus each of f - cent er X gangl i on cell receives direct synaptic i nput f r om one kind of flat cone bi pol ar and an i nvagi nat i ng cone bi pol ar, as well as indirect i nput f r om r od bipolars. A similar state of affai rs applies t o on-cent er X cells: t wo cone bi pol ar direct inputs, and an indirect r od bi pol ar i nput . Ther e are thus VISUAL ADAPTATION AND RETINAL GAIN CONTROLS 339 several cone bi pol ar t ypes, since on- and of f - X cells each recei ve i nput f r om t hei r own speci al invaginating bi pol ars, and t hei r own special t ype of fl at bi pol ar (Sterling, 1983). Lat er al spread of signals wi t hi n t he ret i na is i mpor t ant f or adapt i ve mechani sms. Lat er al i nt er act i on is possible al r eady at t he r ecept or cell level. In t he cat t here are gap j unct i ons bet ween pedicles of i ndi vi dual cone cells and also bet ween rod-spherul es and processes ext endi ng f r om cone- pedicles (see e.g. Nel son e t al . , 1981). Nel son (1977) has demonst r at ed t hat i ndi vi dual cat cones receive r od signals whi ch pr esumabl y ent er vi a t he r o d - c o n e gap j unct i ons. To what ext ent t he c o n e - c o n e j unct i ons also represent f unct i onal cont act s in t he cat is not known at present and we are not aware t hat anybody has shown t hat t he cat ret i na has any f unct i onal or anat omi cal r o d - r od cont act s, but , in some col d- bl ooded vert ebrat es, funct i onal coupl i ng bet ween rods via gap j unct i ons is a pr omi nent f eat ur e (e.g. Det wi l er e t al . , 1980). The mor e distal of t he t wo l at eral l y spreadi ng neur on t ypes is t he h o r i z o n t a l cel l of whi ch t he cat has t wo kinds ( Boycot t , 1974; Boycot t e t al . , 1978). The t ype A and t ype B cell bot h have a soma and a dendri t i c ar bor i zat i on in whi ch all synapt i c cont act s are with c o n e s . The t ype B cell in addi t i on has a l ong, ver y t hi n ( pr obabl y non-i mpul se carryi ng) axon whi ch leaves t he soma- dendr i t e compl ex and at a distance of about hal f a millimeter " e xpl ode s " i nt o an el abor at e ar bor i zat i on whi ch cont act s onl y r o d s . In t he cat, hor i zont al cells onl y hyperpol ari ze t o increases in illumination, and most receive bot h cone and r od signals f r om phot o- r ecept or s (St ei nberg, 1971; Ni emeyer and Gour as, 1973; Nel son e t a L, 1975). The ot her i nt er neur ons whi ch medi at e lateral i nt eract i on are t he amacri ne cells which spread their processes in t he inner pl exi form layer. The amacrine cell f ami l y is made up of at least t went y- t wo di f f er ent types in t he cat ret i na, as distinguished by dendritic branchi ng pat t erns in t he I PL (Kolb e t aL, 1981). The most well studied of these cells is t he Al l amacr i ne cell whi ch receives di rect synapt i c i nput f r om r od bi pol ars in l ami na-b of t he I PL, and gap- j unct i on cont act s f r om cone-bi pol ars in l ami na-b also. It also receives cone bi pol ar i nput f r om synapses in lamina-a. So far onl y r od bipolars which hyper pol ar i ze t o i ncrement s of light have been descri bed (Nelson, 1980) and yet AI I amacri nes depol ari ze t o i ncrement s, suggesting t hat t he r o d - b i p o l a r t o AI I amacr i ne synapse is sign- i nvert i ng. AI I amacri nes make synapt i c cont act ont o pr es umed of f - c e nt e r (i . e. d e c r e me n t - exci t at ory) gangl i on cells, whi ch suggests t hat t he synapse between t he AI I and the off-cent er ganglion cell is sign-inverting also. None of t he ot her amacri ne cells' connect i vi t y is as well charact eri zed as t hat of t he AI I cell, yet cert ai n feat ures of t hei r st ruct ure are not ewor t hy. As in ot her species, t here are numer ous amacr i ne ~ amacr i ne synapt i c cont act s in t he I PL of t he cat ret i na. Fur t her mor e, t here is a preferent i al i nput of amacri ne cells t o Y gangl i on cells, while t her e is much mor e di rect bi pol ar i nput t o t he X cells (Kol b, 1979; Sterling, 1983). Since most is known about t he processes of ret i nal adapt at i on in t he X and Y gangl i on cells in t he cat, t he similarities and di fferences in t hei r anat omi cal connect i ons are i mpor t ant f or under - st andi ng t he site in t he ret i na where ret i nal adapt at i on t akes place. One of the curious features of retinal mor phol ogy is t hat t here appears t o be onl y one r od bi pol ar t ype compar ed t o several cone bi pol ar types (Sterling, 1983). Since recept i ve field or gani zat i on is j ust as rich in t he scot opi c range as in t he phot opi c, one mi ght have expect ed as much anat omi cal el abor- at i on of t he r od bi pol ar f ami l y as of t he cone bi pol ars. Because of t he great degree of r od - cone i ndependence, we guess t hat r o d - c o n e coupl i ng (Nel son, 1977) is of s econdar y i mpor t ance. However , Sterling (1983) believes t hat r o d - cone coupl i ng is t he mai n pat hway f or r od i nput t o gangl i on cells over most of t he scot opi c range. This is not really a cont r over sy, since bot h our view and Sterling' s are equally speculative at present. The AII amacri ne, whi ch gets its pr edomi nant i nput f r om r od bi pol ar cells, couples into the phot opi c circuitry when it makes gap j unct i ons with t he cone bi pol ar cells in l ami na-b of t he I PL. The AI I amacr i ne seems t o be a crucial link in scotopic receptive fields. One of t he most i nt erest i ng i nt er neur ons in t he retina is t he i nt erpl exi form cell (Boycot t et al . , 1975; Kol b and West, 1977; Nakamur a e t al . , 1980). This neur on is per haps t he most likely candi dat e t o be i nvol ved in gain cont r ol in t he ret i na. It has dendri t i c arbori zat i ons in t he I PL, where it receives synapses f r om amacr i ne cells of so-far i ndet er- 340 R . M. SHAPLEY AND C. ENROTH- CUGELL I P L b C B o 1 ~ a 2 ~ ~ All ~ F ~ G ON-Beta G I l l , o N FIG. 63. Di a gr a m o f connect i ons i n t he i nner pl exi f or m l ayer o f t he cat r et i na f r om serial EM r econs t r uct i ons . Chemi cal s ynaps es ar e i ndi cat ed by t he ar r ows : gap j unc t i ons ar e i ndi cat ed by t he s ymbol -I . CB i ndi cat es a cone bi pol ar cell; f our subt ypes are labeled. RBP s t ands f or r od bi pol ar cell. Al l is a subcl ass of amacr i ne cells. The Bet a gangl i on cells are pr es umed t o be t he anat omi cal equi val ent of t he X cells. The i nner pl exi f or m l ayer is subdi vi ded i nt o s ubl ami nae- a a nd -b, as i ndi cat ed i n t he f i gur e. On gangl i on cells recei ve i nput in s ubl a mi na - b, whi l e o f f gangl i on cells recei ve i nput in s ubl a mi na - a . Fr o m St erl i ng (1983). mi nat e t ype. Then t he i nt er pl exi f or m cell makes synapses on all t ypes of bi pol ar cells at t he level of t he I PL but mor e especially at the level of t he out er pl exi f or m layer ( OPL) . It is t he perfect exampl e of a f eedback neur on. Mor eover , it coul d af f ect all bi pol ars which is a pr oper t y one woul d want f r om a gain cont r ol since on- and of f - pat hways must be adapt ed in t he same way in or der t o keep cont r ast sensitivity const ant . However , t he wide t ermi nal br anchi ng of t he i nt er pl exi f or m cell in t he OPL is a pr obl em i f this cell is t o pl ay a role in regulating gain on st eady backgr ounds, because, at least f or X cells, t he area of t he ret i na over which adapt i ve signals are summed for the center is less t han 1 deg 2, i.e. less t han 200 /a by 200/ a. However , t here is evidence f or a di f f er ent kind of gain cont r ol which depends on stimulus cont r ast r at her t han average flux (see Sect i on 3.8; Shapl ey and Vi ct or, 1978; cf. Werbl i n and Copenhagen, 1974) and perhaps the i nt erpl exi form cell' s propert i es may mat ch t hose of this ot her gain cont r ol . A schematic di agram of the neur onal circuitry in t he cat ' s I PL, where gangl i on cells receive t hei r i nput s, is of f er ed in Fig. 63, f r om a recent review article (Sterling, 1983). APPENDI X 2 - - RECEPTI VE FI ELDS The neural signal which leaves the ret i na consists of trains of impulses carri ed by t he axons of retinal ganglion cells. The response of these cells t o a visual stimulus may be defi ned as a change in the rat e of firing of impulses. An adequat e stimulus t o cause such a change in firing is some change in the i l l umi nat i on on t he ret i na. The wor k of Har t l i ne (1940), Barl ow (1953), and Kuf f l er (1952, 1953) showed t hat each retinal gangl i on cell generat ed responses, as defi ned above, t o st i mul at i on over a limited area of t he retina, and this area was defi ned as the receptive field of t hat ganglion cell. Wor ki ng in t he cat retina, Kuffl er (1953) f ound t hat ganglion cell recept i ve fields consisted of t wo concent ri c zones which he called t he center and surround. The cent er and sur r ound were mut ual l y antagonistic. In on-cent er cells in which the center caused excitatory VISUAL ADAPTATION AND RETINAL GAIN CONTROLS 3 4 1 ON-CENTER RECEPTIVE FIELD , , , , , , , , , , , , , , , , , , , , , , , , , I I I I I I I I I I I I I I I I I I I I I I l l l l l I I I l l l l l l l l I @. I I I I I I IIIIIII IIIIII I I I I I I IIIIIII IIIIII OFF-CENTER r f f ~- ] F- - - ' q III1 IIIIIIIII I I I l l l I l l l l l l l l I I ~ l - - - - " q I I I I I I IIIIIII I l l l l I I I I I I IIIIIII IIIII FIG. 64. Schemat i c r epr es ent at i on o f t he mu t u a l a nt a goni s m bet ween cent er a n d s u r r o u n d i n on a n d of f - cent er cells. I n on- cent er cells ( a, b, c) , a n i nc r e me nt i n t he cent er i ncr eases t he cell' s fi ri ng rat e while an i ncr ement o f l i ght i n t he s ur r ound decreases t he firing rate. Wh e n t he t wo i ncr ement s are appl i ed t oge t he r in s ync hr ony, as i n (c), t he r e s pons e t o l i ght " o n " is l ess t h a n i n (a) a n d t he r e s pons e t o l i ght " o f f " is l ess t h a n i n (b). Thi s me a n s t he t wo r egi ons ar e mut ua l l y ant agoni s t i c. I n of f - cent er cells ( d, e, f ) , t he cel l s' r e s pons e pa t t e r n at l i ght " o n " i n t he cent er a nd i n t he per i pher y ar e r ever sed, but t he mu t u a l a n t a g o n i s m is t he s a me . responses t o increments of light, the sur r ound woul d cause i nhi bi t ory responses t o increments. In off-cent er cells in which the central region was inhibitory during an increment, the surround would be excitatory during an increment. The on- and off- center cells and their cent er-surround organi zat i on are illustrated in Fig. 64. Rodi eck (1965) made a maj or advance by developing a model for the cat gangl i on cell receptive field in terms of overlapping center and sur r ound mechani sms. Each mechani sm may be conceived of as the receptors and i nt erneurons, the signals of which are pool ed t oget her to influence the firing of the gangl i on cell. Wi t hi n the center mechanism all light evoked signals generated within the pool of these receptors and i nt erneurons are summed, according t o the model , and similarly for the surround. Then center and surround signals are summed at the gangl i on cell. The signals f r om di fferent positions within each pool are weighted by what Rodi eck called the "sensi t i vi t y pr of i l e", and which we will refer t o as the "spat i al distribution of gai n". The center has its own narrow spatial di st ri but i on of gain, and the surround has a rat her broader spatial di st ri but i on. Rodi eck proposed t hat these t wo spatial di st ri but i ons could be appr oxi mat ed by Gaussi an surfaces wi t h di fferent extents of spread. The spatial resol ut i on and opt i mal spatial t uni ng of retinal gangl i on cells can be rat i onal i zed in terms of Rodi eck' s model (Enrot h-Cugel l and Robson, 1966). The spatial resol ut i on of the cell is due t o the finite size of the center, and in fact can be predicted from knowledge of the magni t ude of the spread of the cent er' s Gaussian spatial distribution of gain (Cleland et eL, 1979; So and Shapley, 1979; Linsenmeier et el. , 1982). Figure 65 shows the Rodi eck model for the receptive field of cat retinal gangl i on cells. ~ R-MECHANISM __ ~ D - M E C H A N I S M C ~ " E N ' I - - - I 1 " - - ~ I ' , ~ SURROUND j _ L RESPONS~ ~ CENTER TYPE NO RESPONSE ON-OFF TYPE RESPONSE RESPONSE FIG. 65. The t wo spat i al l y over l appi ng me c h a n i s ms i n t he Rodi eck mode l o f r et i nal ga ngl i on cell r ecept i ve fi el ds. I n t he s ket ch at t he t op, t he hor i zont al axi s r epr es ent s di s t ance on t he r et i na. The hei ght s o f t he t wo cur ves r epr es ent t he ga i n o f t he cent er a n d o f t he s u r r o u n d , as l abel ed, as a f unc t i on o f pos i t i on on r et i na. Bot h ar e Ga us s i a n f unc t i ons o f posi t i on; t he cent er ' s Ga us s i a n has a nar r ower spr ead t ha n t he s u r r o u n d ' s . The cent er a nd s u r r o u n d ha ve oppos i t e si gn i n t hi s model . Thi s r esul t s i n mu t u a l a n t a g o n i s m. I n t hi s mode l t he cent er a nd s u r r o u n d c o mp o n e n t s c ombi ne by s i mpl e addi t i on, i . e. l i near l y. Th u s , r e s pons e t o s t i mul a t i on a nywhe r e wi t hi n t he fi el d i s, accor di ng t o t he model , s i mpl y a s u m o f t he cent er a n d s u r r o u n d c o mp o n e n t s i n r e s pons e t o t he s t i mul us , as is i l l ust r at ed i n t he f i gur e. Fr o m Rodi eck (1973). SURROUNO TYP~ RESPONSE 342 There are several ganglion cell classes in each vertebrate retina (see Rodieck, 1979). This is a significant complication, because not all of these cell types conform to the simple elegance of the Rodieck model. As far as we know, the ganglion cell classes in the cat retina which have the highest contrast gain are the X and Y cells discovered by Enroth-Cugell and Robson (1966). While X cells do behave in a way approximately predictable from the Rodieck model, Y cells exhibit nonlinear summation of visual signals (Enroth-Cugell and Robson, 1966; Hochstein and Shapley, 1976a, b). It seems that while X cells have two pools or mechanisms as in Rodieck's model, Y cells have at least three different types of neuronal mechanism. The new type of mechanism is what Hochstein and Shapley have dubbed the "nonlinear subunits", small spatial pools within which neural signals are summed in a linear manner, but between which a nonlinearity is interposed before signal summation. There have been several excellent reviews of the physiology and anatomy of cat ganglion cells (Robson, 1975; Rodieck, 1979; Lennie, 1980; Wfissle, 1982; Levick and Thibos, 1983, among others). Ac k n o wl e d g e me n t - - We t ha nk our col l eagues in New Yor k a nd Eva ns t on for hel pf ul c omme nt s a nd cri t i ci sm dur i ng t he wri t i ng of t hi s review paper . We t h a n k Dr . E. N. Pu g h for hi s hel pf ul advi ce. For per mi s s i on t o cite unpubl i s he d wor k we t ha nk Dr s. J. Gor don, T. Har di ng, E. Kapl an, P. Lenni e, and D. Tr anchi na. Pr epar at i on of this chapt er has been partially suppor t ed by gr ant s f r om t he Nat i onal I nst i t ut es of Heal t h: RO1 EY 01472, and ROI EY00206. R E F E R E N C E S ADELSON, E. (1982) Sa t ur a t i on a n d a da pt a t i on in t he r od s ys t em. Vision Res. 22: 1 2 9 9 - 1 3 1 2 . AGU1LAR, M. a nd STILES, W. S. (1954) Sa t ur a t i on o f t he r od me c h a n i s m o f t he r et i na at hi gh levels o f s t i mul at i on. Optica Act a 1: 5 9 - 6 5 . ALPERN, M. and PUGH, JR. , E. N. (1974) The dens i t y and phot osensi t i vi t y o f h u ma n r hodops i n in t he living ret i na. J. Physiol. 237: 3 4 1 - 370. ALPERN, M. , MAASEIDVAAG, F. a nd OHBA, N. (1971) The ki net i cs of cone vi sual pi gme nt s i n ma n . Vision Res. 11: 539 - 549. ASHMORE, J. F. a nd FALK, G. (1980) Re s pons e s of r od- bi pol ar cells in t he d a r k - a d a p t e d r et i na o f t he dogf i s h, Scyliorhinus canicula. J. Physiol. 300: 1 1 5 - 150. BARLOW, H. B. (1953) S u mma t i o n a nd i nhi bi t i on i n t he f r og' s r et i na. J. Physiol. 119: 6 9 - 88. BARLOW, H. B. (1957) I nc r e me nt t hr e s hol ds at l ow i nt ensi t i es consi dered as signal noi se di scri mi nat i ons. J. Physiol. 136: 4 6 9 - 488. R. M. SHAPLEY AND C. ENROTH- CUGELL BARLOW, H. B. (1958) I nt r i nsi c noi se of cones. In: Visual Problems of Col our, Vol. II, pp. 6 1 7 - 630. Her Maj es t y' s St at i oner y Of f i ce, Lo n d o n . BARLOW, H. B. (1964) Da r k- a da pt a t i on: a new hypot hes i s . Vision Res. 4: 4 7 - 58. BARLOW, H. B. (1965) Opt i c ner ve i mpul s es a nd We he r ' s Law. Col d Spring Harb. Syrup. quant. Biol. 30: 5 3 9 - 5 4 6 . BARLOW, H. B. (1977) Ret i nal a nd cent r al f act or s in h u ma n vision limited by noise. In: Vertebrate Photoreception (H. B. Bar l ow a nd P. Fat t , eds). Ac a de mi c Pr ess, Lo n d o n . BARLOW, H. B. a nd ANDREWS, D. S. (1967) Sensi t i vi t y of r ecept or s and r ecept or " p o o l s " . J. opt. Soc. Am. 57: 837 - 838. BARLOW, H. B. and LEVICK, W. R. (1969) Thr ee f act or s l i mi t i ng t he rel i abl e det ect i on o f l i ght by r et i nal gangl i on cells o f t he cat . J. Physiol. 200: l - 2 4 . BARLOW, H. B. and LEVICK, W. R. (1976) Thr e s hol d set t i ng by t he s ur r ound of cat ret i nal gangl i on cells. J. Physiol. 259: 737 757. BARt.OW, H. B. , FITZHUGH, R. a nd KUFFLER, S. W. (1957) Cha nge of or gani zat i on in t he recept i ve fields of t he cat ' s ret i na dur i ng dar k adapt at i on. J. Physiol. 137: 338 - 354. BAkLOW, H. B. , LEVICK, W. R. and YooN, M. (1971) Responses t o si ngl e q u a n t a o f l i ght in r et i nal gangl i on cells of t he cat . Vision Res. 11: Suppl . , 8 7 - 101. BATTERSBY, W. a nd WA6MAN, I. (1962) Neur al l i mi t s of vi sual exci t abi l i t y. IV. Spat i al de t e r mi na nt s o f r et r ochi as mal i nt er act i on. Am. J. Physiol. 203: 3 5 9 - 3 6 5 . BAYLOR, D. A. a nd HODGKIN, A. L. (1973) Det ect i on a nd r es ol ut i on o f vi sual st i mul i by t ur t l e phot or ecept or s . J. Physiol. 234: 1 6 3 - 198. BAYLOR, D. A. a nd HODGKIN, A. L. (1974) Cha nge s in t i me scal e a nd sensi t i vi t y in t ur t l e phot or ecept or s . J. Physiol. 242: 7 2 9 - 758. BAYLOR, D. A. , HODGKIN, A. L. a nd LAMB, T. D. (1974a) The electrical r es pons e o f t ur t l e cones t o f l ashes a nd st eps of l i ght . J. Physiol. 242: 6 8 5 - 727. BAYLOR, O. A. , HODGKIN, A. L. , and LAMB, T. D. (1974b) Re c ons t r uc t i on o f t he electrical r es pons es o f t ur t l e cones t o f l ashes a n d st eps of l i ght . J. Physiol. 242: 7 5 9 - 791. BAYLOR, D. A. , LAMB, T. D. and YAU, K. - W. (1979) The me mb r a n e cur r ent o f si ngl e r od out er s egment s . J. Physiol. 2811:589 - 61 I. BAYLOR, O. A. , MATTHEWS, G. and YAU, K. - W. (1980) Two c o mp o n e n t s of electrical dar k noi s e in t oa d r et i nal r od out er s egment s . J. Physiol. 309: 5 9 1 - 621. BIERSDORF, W. R. , GRANDA, A. M. a nd LAWSON, H. F. (1965) El ect ri cal me a s u r e me n t o f i ncr ement al t hr es hol ds in t he h u ma n eye. 3. opt. Soc. Am. 55: 4 5 4 - 4 5 5 . BLACKWELL, H. R. (1946) Cont r a s t t hr es hol ds of t he h u ma n eye. J. opt. Soc. Am. 36: 6 2 4 - 6 4 3 . BONDS, A. B. (1974) Opt i cal qual i t y of t he l i vi ng cat eye. J. Physiol. 243: 7 7 7 - 7 9 5 . BORING, E. G. (1950) The Hi st ory o f Experimental Psychology. Appl e t on, Ce nt ur y, Cr of t s , New Yor k. p. 288. BORSELLINO, A. , FUORTES, M. G. F. a nd SMITH, T. G. (1965) Vi sual r es pons es in Li mul us . Col d Spring Harb. Syrup. quant. Biol. 30: 4 2 9 - 4 4 3 . BOYCOTT, B. B. (1974) Aspect s of t he compar at i ve a na t omy and phys i ol ogy of t he ver t ebr at e r et i na. In: Essays on the Nervous System: A Festschrift f or Professor J. Z. Young (R. Bellairs and E. G. Gr ay, eds) pp. 223 - 257. Cl ar endon Pr ess, Oxf or d. BOYCOTT, B. B. a nd WASSLE, H. (1974) The mor phol ogi cal t ypes of ga ngl i on cells o f t he domes t i c c a t ' s r et i na. J. Physiol. 2 4 0 : 3 9 7 - 419. VISUAL ADAPTATION AND RETINAL GAIN CONTROLS 343 BOYCOTT, B. B., DOWLING, J. E., FISHER, S. K., KOLB, H. and LATIES, A. M. (1975) Interplexiform cells of the mam- malian retina and their comparison with catecholamine- containing retinal cells. Proc. R. Soc. Lond. B, 191: 353 - 368. BOYCOTT, B. B., PEICHL, L. and W~,SSLE, H. (1978) Morphological types of horizontal cell in the retina of the domestic cat. Proc. R. Soc. Lond. B, 203:229 - 245. BOYNTON, R. M. and WHITTEN, D. N. (1970) Visual adaptation in monkey cones: recordings of late receptor potentials. Science 170:1423 - 1426. BURKHARDT, D. A. (1974) Sensitization and cent re- surround antagonism in Necturus retina. J. Physiol. 236:593 - 610. BURKHARDT, D. A. and BERNTSON, G. G. (1972) Light adaptation and excitation: lateral spread of signals within the frog retina. Vision Res. 12: 1095-1111. Buss, C. M., HAYHOE, M. M. and STROMEYER, C. F., III (1982) Lateral interactions in the control of visual sensitivity. Vision Res. 22: 693- 709. CAMPBELL, F. W. and ROBSON, J. G. (1968) Application of Fourier analysis to the visibility of gratings. J. Physiol. 197: 551-566. CLELAND, B. G. and ENROTH-CUGELL, C. (1968) Quantitative aspects of sensitivity and summation in the cat retina. J. Physiol. 198: 17-38. CLELAND, B. G. and ENROTH-CUGELL, C. (1970) Quantitative aspects of gain and latency in the cat retina. J. Physiol. 2 0 6 : 7 3 - 9 1 . CLELAND, B. G. and LEVICK, W. R. (1974) Brisk and sluggish concentrically organized ganglion cells in the cat' s retina. J. Physiol. 240: 421- 456. CLELAND, B. G., LEVlCK, W. R. and SANDERSON, K. J. (1973) Properties of sustained and transient ganglion cells in the cat retina. J. Physiol. 228: 649-680. CLELAND, B. G. , HARDING, T. H. and TULUNAY-KEESEY, O. (1979) Visual resolution and receptive field size: Examination of two kinds of cat retinal ganglion cell. Science 205:1015 - 1017. CORNSWEET, T. (1970) Visual Perception. p. 273, also Chapter 2. Academic Press, New York. CRAIK, K. J. W. (1938) The effect of adaptation on differential brightness discrimination. J. Physiol. 92: 406-421. CRAIK, K. J. W. (1966) The Nature o f Psychology (S. Sherwood, ed.). Cambridge University Press, Cambridge, p. 96. CRAWFORD, B. H. (1947) Visual adaptation in relation to brief conditioning stimuly. Proc. R. Soc. Lond. B, 134: 283 - 302. DAITCH, J. i . and GREEN, D. G. (1969) Contrast sensitivity of the human peripheral retina. Vision Res 9: 947- 952. DAVIDSON, E. H. and FREEMAN, R. B., JR. (1965) Brightness constancy under a gradient of illumination. Psychonom. Sci. 2: 349- 350. DAW, N. W. and PEARLMAN, A. L. (1969) Cat colour vision: One cone process or several? J. Physiol. 201:745 - 764. DAWlS, S. M. and PURPLE, R. L. (1982) Adaptation in cones: a general model. Biophys. J. 39:151 - 155. DERRINGTON, A. M. and LENNIE, P. (1982) The influence of temporal frequency and adaptation level on receptive field organization of retinal ganglion cells in cat. J. Physiol. 333:343 - 366. DETWILER, P. B., HODGKIN, A. L. and MCNAUGHTON, P. A. (1980) Temporal and spatial characteristics of the voltage response of rods in the retina of the snapping turtle. J. Physiol. 300:213 - 250. DO~E, F. A., KNIGHT, B. W. and TOYODA, J. (1968) Voltage noise in Limulus visual cells. Science 160: 88. DOWLING, J. E. (1967) Site of visual adaptation. Science 155: 273 - 279. DOWLING, J. E. and BOYCOTT, B. B. (1965) Neural connections of the retina: fine structure of the inner plexiform layer. Cold Spring Harb. Symp. quant. Biol. 30: 393- 402. DOWLING, J. E. and RIPPS, H. (1971) S-potentials in the skate retina: Intracellular recordings during light and dark adaptation. J. gen. PhysioL 58: 163-189. DOWLING, J. E. and RIPPS, H. (1972) Adaptation in skate photoreceptors. J. gen. Physiol. 60: 698-719. EASTER, S. (1968) Adaptation in goldfish retina. J. Physiol. 195: 273 - 281. ENROTH-CUGELL C. and JArIELA, H. G. (1980) Suppression of cat retinal ganglion cell responses by moving patterns. J. PhysioL 3tl2: 49-72. ENROTH-CUGELL C. and LENNIE, P. (1975) The control of retinal ganglion cell discharge by receptive field surrounds. J. PhysioL 2,47:551-578. ENROTH-CUGELL C. and ROBSON, J. G. (1966) The contrast sensitivity of retinal ganglion cells of the cat. J. Physiol. 157: 517- 552. ENROTH-CUGELL C. and SHAPLEY, R. M. (1973a) Adaptation and dynamics of cat retinal ganglion cells. J. Physiol. 233: 271 - 309. ENROTH-CUGELL C. and SHAPELY, R. M. (1973b) Flux, not retinal illumination, is what cat retinal ganglion cells really care about. J. Physiol. 233:311-326. ENROTH-CUGELL C., LENNIE, P. and SHAPELY, R. M. (1975) Surround contribution to light adaptation in cat retinal ganglion cells. J. Physiol. 247: 579-588. ENROTH-CUGELL C., HERTZ, B. G. and LENNIE, P. (1977a) Cone signals in the cat's retina. J. Physiol. 269:273 - 296. ENROTH-CUGELL, C., HERTZ, B. G. and LENNIE, P. (1977b) Convergence of rod and cone signals in the cat' s retina. J. PhysioL 269: 297-318. FAIN, G. L. (1976) Sensitivity of toad rods: dependence on wave-length and background illumination. J. Physiol. 261:71 - 101. FAMIGLIETrI, E. V., JR. and KOLB, H. (1976) Structural basis for ON- and OFF-center responses in retinal ganglion cells. Science 194:193 - 195. FAMIGLIETTI, E. V., KANEKO, A. and TACHIBANA, M. (1977) Neuronal architecture of on and off pathways to ganglion cells in carp retina. Science 198: 1267- 1269. FISCHER, B. and MAY, H. U. (1970) lnvarianzen in der Katzenretina: Gesetzm~issige Beziehungen zwischen Empfindlichkeit, Grosse und Lage receptiver Felder von Ganglienzellen. Expl Brain Res. 11:448 - 464. FOORTES, M. G. F. and HODGKIN, A. L. (1964) Changes in time scale and sensitivity in the ommatidia of Limulus. J. Physiol. 172:239 - 263. FUORTES, M. G. F., GUNKEL, R. D. and RUSHTON, W. A. H. (1961) Increment thresholds in a subject deficient in cone vision. J. PhysioL 156: 179- 192. GOROON, J. and GRArtAM, N. (1973) Early light and dark adaptation in frog on - off retinal ganglion cells. Vision Res. 13: 647-659. GRAHAM, N. (1980) Spatial-frequency channels in human vision: detecting edges without edge detectors. In: Visual Coding and Adaptability (C. S. Harris, ed.). Lawrence Erlbaum, Hillsdale, New Jersey. GRANIT, R., MUNSTERHJELM, A. and ZEWl, M. (1939) The relation between concentration of visual purple and retinal PRR3-3 344 R. M. SHAPLEY AND C. ENROTH-CUGELL sensitivity t o light duri ng dark adapt at i on. J. Physiol. 96: 31 - 4 4 . GREEN, D. G. (1973) Scot opi c and phot opi c component s o f t he r at el ect r or et i nogr am. J. Physiol. 228: 7 8 1 - 797. GREEN, D. G. , TONG, L. and CICERONE, C. M. (1977) Lat eral spr ead o f l i ght adapt at i on in t he rat ret i na. Vision Res. 17: 4 7 9 - 486. GROSSBERG, S. (1981) Adapt i ve r esonance in devel opment , per cept i on and cogni t i on. In: Mat hemat i cal Psychology and Psychophysiology (S. Gr ossber g, ed. ) pp. 107 - 156. Er l baum, Hi l l sdal e, New Jersey. HARDING, T. H. (1977) Fi el d adapt at i on and signal s ummat i on wi t hi n t he recept i ve fi el d cent er of cat ret i nal gangl i on cells. Thesi s, Pur due Uni versi t y. HARDING, T. H. (1978) Frontiers in Visual Science (S. J. Cool and E. L. Smi t h, III, eds) pp. in Enroth-Cugel1483 - 489. Spri nger, Berlin, Hei del ber g, New Yor k. HARTLINE, H. K. (1938) The r esponse o f single opt i c nerve fi bres o f t he ver t ebr at e eye t o i l l umi nat i on o f t he ret i na. Am. J. Physiol. 121: 4 0 0 - 4 1 5 . HARTLINE, H. K. (1940) The recept i ve fields of opt i c nerve fi bers. Am. J. Physiol. 130: 6 9 0 - 6 9 9 . HECHT, S. (1924) The visual di scri mi nat i on o f i nt ensi t y and t he Weber - Fechner Law. J. gen. Physiol. 7: 235 - 267. HECHT, S. , SHLAER, S. and PIRENNE, M. H. (1942) Ener gy, quant a, and vision. J. gen. Physiol. 25: 8 1 9 - 8 4 0 . HEINEMANN, E. G. (1955) Si mul t aneous bri ght ness i nduct i on as a f unct i on of i nduci ng- and t est -fi el d l umi nances. J. exp. Psychol. 50: 8 9 - 9 6 . HEINEMANN, E. G. (1972) Si mul t aneous br i ght ness i nduct i on. In: Handbook o f Sensory Physiology (D. J ames on and L. M. Hurvi ch, eds) Vol. VI I / 4, pp. 146- 169. Spri nger, Berlin. HELMHOLTZ, H. VON (1909) Cont r as t . In: Treatise on Physio- logical Optics 3rd edn (J. P. C. Sout hal l , ed. ) Vol . 2, Ch. 24, 1924; r pt . 1962, Dover , New Yor k. HELSON, H. (1964) Adapt at i on Level Theory. Har per & Row, New Yor k. HERING, E. (1920) In: Outline o f a Theory o f the Li ght Sense (L. M. Hur vi ch and D. J ames on, t r ans. 1964). Har var d Uni versi t y Press, Cambr i dge. HOCHSTEIN, S. and SHAPLEY, R. M. (1976a) Quant i t at i ve analysis of retinal gangl i on cell classifications. J. Physiol. 262: 237 - 264. HOCHSTEIN, S. and SHAPLEY, R. M. (1976b) Li near and nonl i near spat i al subuni t s in Y cat ret i nal gangl i on cells. J. Physiol. 262: 2 6 5 - 284. HOOD, D. and HOCK, M. (1975) Li ght adapt at i on of t he r ecept or s: i ncr ement t hr eshol d f unct i ons f or t he f r og' s r ods and cones. Vision Res. 15: 5 4 5 - 5 5 3 . HUBEL, n . H. and WIESEL, T. N. (1960) Recept i ve fields o f opt i c nerve fi bers in t he spi der monkey. J. PhysioL 154: 572 - 580. JAKIELA, H. G. , ENROTH-CUGELL, C. and SHAPLEY, R. M. (1976) Ada pt a t i on and dynami cs in X-ceils and Y-cells o f t he cat ret i na. Expl Brain Res. 24: 335 - 342. KANEKO, A. (1973) Recept i ve fi el d or gani zat i on o f bi pol ar and amacr i ne cells in t he gol df i sh ret i na. J. Physiol. 235: 133 - 153. KAPLAN, E. , MARCUS, S. and So, Y. T. (1979) Ef f ect s of dar k adapt at i on on spatial and t emporal propert i es of receptive fields in cat l at eral geni cul at e nucl eus. J. Physiol. 294: 561 - 580. KELLY, D. H. (1972) Adapt at i on ef f ect s on s p a t i o - t e mp o r a l si ne-wave t hr eshol ds. Vision Res. 12: 8 9 - 102. KLEINSCHMIDT, J. and DOWLING, J. E. (1975) Int racel l ul ar r ecor di ngs f r om Gecko phot or ecept or s dur i ng light and dar k adapt at i on. J. gen. Physiol. 66: 6 1 7 - 6 4 8 . KOENDERINK, J. J . , BOUMAN, M. A. , BUENO DE MESQUITA, A. E. and SLAPPENDEL, S. (1978) Per i met r y of cont r ast det ect i on t hreshol ds of movi ng spatial sine wave pat t erns. IV. The i nf l uence o f mean ret i nal i l l umi nance. J. opt. Soc. Am. 68: 8 6 0 - 8 6 5 . KOLB, H. (1979) The i nner pl exi f or m layer in t he ret i na of t he cat: El ect r on mi cr oscopi c obser vat i ons. J. Neurocytol. 8 : 2 9 5 - 329. KOLB, H. and WEST, R. W. (1977) Synapt i c connect i ons o f t he i nt erpl exi form cell in t he ret i na of t he cat . J. Neurocytol. 6: 1 5 5 - 170. KOLB, H. , NELSON, R. and MARIANI, A. (1981) Amacr i ne cells, bi pol ar cells and gangl i on cells of t he cat ret i na: A Golgi st udy. Vision Res. 21: 1081 - 1114. KUFFLER, S. W. (1952) Neur ons in t he ret i na: or gani zat i on, i nhi bi t i on and exci t at i on pr obl ems. Col d Spring Harb. Syrup. quant. Biol. 1 7 : 2 8 1 - 2 9 2 . KUFFLER, S. W. (1953) Di scharge pat t er ns and f unct i onal or gani zat i on o f mammal i an ret i na. J. Neurophysiol. 16: 37 - 68. LAND, E. H. and MCCANN, J. J. (1971) Li ght ness and ret i nex t heor y. J. opt. Soc. Am. 61: 1 - 11. LENNIE, P. (1980) Paral l el visual pat hways: a review. Vision Res. 20: 5 6 1 - 594. LENNIE, P. and MACLEOD, D. I. A. (1973) Backgr ound conf i gur at i on and r od t hr eshol d. J. Physiol. 233: 143 - 156. LENNIE, P. , HERTZ, B. G. and ENROTH-CUGELL, C. (1976) Sat urat i on of r od pool s in cat . Vision Res. 16:935 - 940. LEVICK, W. R. and TmBOS, L. N. (1983) Recept i ve fields o f cat gangl i on cells: cl assi fi cat i on and const r uct i on. In: Progress in Ret i nal Research (N. N. Os bor ne and G. J. Chader , eds) Ch. 11, Vol . 2, Pe r ga mon Press, Oxf or d. LINSENMEIER, R. A. , FRISHMAN, L. J. , JAKIELA, H. G. and ENROTH-CUGELL, C. (1982) Recept i ve field pr oper t i es of X and Y cells in t he cat ret i na deri ved f r om cont r ast sensi t i vi t y measur ement s. Vision Res. 22: 1173- 1183. MACLEOD, D. I. A. (1978) Visual sensitivity. A. Rev. Psychol. 29: 613 - 645. MARR, D. (1982) Vision pp. 2 5 0 - 2 6 1 . W. H. Fr eeman, San Fr anci sco. MICHELSON, A. A. (1927) Studies in Optics p. 31. Uni v. of Chi cago Press, Chi cago. de MONASTERIO, F. M. (1978) Cent er and sur r ound mechani sms of opponent - col or X and Y gangl i on cells o f ret i na o f macaques. J. Neurophysiol. 41: 1418- 1434. NAGEL, W. (1909) Appendi x: Adapt at i on, t wi l i ght vision, and t he dupl i ci t y t heor y. In: Hel mhol t z, H. von, Treatise on Physiological Optics, 3rd ed. (J. P. C. Sout hal l , ed. ) Vol. 2, p. 342. 1924; r pt . 1962, Dover , New Yor k. NAKA, K. - I . and OHTSUKA, T. (1975) Mor phol ogi cal and f unct i onal i dent i f i cat i ons of cat fi sh ret i nal neur ons. II. Mor phol ogi cal i dent i f i cat i on. J. Neurophysiol. 38: 7 2 - 9 1 . NAKA, K. -I. and RUSHTON, W. A. H. (1966) S- pot ent i al s f r om l umi nosi t y uni t s in t he ret i na o f fish (Cypri ni dae). J. Physiol. 185: 5 8 7 - 5 9 9 . NAKA, K. - I . , MARMARELIS, P. Z. and CHAN, R. Y. (1975) Mor phol ogi cal and f unct i onal i dent i fi cat i ons o f cat fi sh ret i nal neur ons . III. Funct i onal i dent i f i cat i on. J. Neurophysiol. 38: 92 - 131. NAKA, K.-I., CHAN, R. Y. and YASUI, S. (1979) Adapt at i on in VISUAL ADAPTATION AND RETINAL GAIN CONTROLS 345 catfish retina. J. Neurophysiol. 42: 441-454. NAKAMURA, Y., MCGUIRE, B. A. and STERLING, P. (1980) Interplexiform cell in cat retina: identification by uptake of gamma-(3H)ami nobut yri c acid and serial reconstruction. Proc. natn. Acad. Sci. U.S.A. 7 7 : 658- 661. NELSON, R. (1977) Cat cones have rod input: A comparison of the response properties of cones and horizontal cell bodies in the retina of the cat. J. comp. Neurol. 172:109 - 135. NELSON, R. (1980) Functional stratification of cone bipolar cell axons in the cat retina. Invest. Ophthal. vis. Sci. Suppl. 1 9 : 130. NELSON, R. and KOLB, H. (1982) Sypaptic patterns and response properties of bipolar and ganglion cells in the cat retina. Invest. Ophthal. vis. Sci. Suppl. 22: 175. NELSON, R., LOTZOW, A. V., KOLB, H. and GOURAS, P. (1975) Horizontal cells in cat retina with independent dendritic systems. Science 189: 137- 139. NELSON, R., FAMIGLIETTI, E. V., JR. and KOLB, H. (1978) Intracellular staining reveals different levels of stratification for on- and off-center ganglion cells in cat retina. J. Neurophysiol. 41: 472-483. NELSON, R., KOLB, H., ROBINSON, M. M. and MARIANI, A. S. (1981) Neural circuitry of the cat retina: Cone pathways to ganglion cells. Vision Res. 21: 1527- 1536. VAN NES, F. L. and BOUMAN, M. A. (1967) Spatial modulation transfer in the human eye. J. opt. Soc. Am. 57:401 - 406. NIEMEYER, G. and GOURAS, P. (1973) Rod and cone signals in S-potentials of the isolated perfused cat eye. Vision Res. 13: 1603- 1612. NORMANN, R. A. and PERLMAN, I. (1979a) The effects of background illumination on the photoresponses of red and green cones. J. Physiol. 286: 491- 507. NORMANN, R. k. and PERLMAN, I. (1979b) Signal transmission from red cones to horizontal cells in the turtle retina. J. Physiol. 286: 509-524. NORMANN, R. A. and PERLMAN, I. (1979C) Evaluating sensitivity changing mechanisms in light-adapted photoreceptors. Vision Res. 19: 391-394. NORMANN, R. A. and WERBLIN, F. S. (1974) Control of retinal sensitivity. I. Light and dark adaptation of vertebrate rods and cones. J. gen. Physiol. 63: 37-61. NUNN, B. J. and BAYLOR, D. A. (1982) Visual transduction in retinal rods of the monkey Macacafascicularis. Nature 299: 726- 728. PASTERNAK, T. and MERIGAN, W. (1981) The luminance dependence of spatial vision in the cat. Vision Res. 21: 1333 - 1340. PENN, R. and HAGINS, W. (1972) Kinetics of the photocurrent of retinal rods. Biophys. J. 12:1073 - 1094. RATLIFF, F. (1965) Mach Bands: Quantitative Studies on Neural Networks in the Retina. p. 75, p. 270. Holden-Day, San Francisco. RAYLEIGH, J. W. S. (1889) On the limit to interference when light is radiated from moving molecules. Phil. Mag. 2 7 : 298 - 304. REUTER, T. (1969) Visual pigments and ganglion cell activity in the retinae of tadpoles and adult frogs (Rana temporaria L.) Acta zool. Fennica 122:1 - 64. RIC_,Os, L. A. (1965) Light as a stimulus for vision. In: Vision and Visual Perception (C. H. Graham, ed.) pp. 1 - 38. Wiley, New York. RIPPS, H., SHAKm, M. and MACDONALD, E. D. (1976) Peroxidase uptake by photoreceptor terminals of the skate retina. J. cell Biol. 70: 86- 96. ROBSON, J. G. (1975) Receptive fields: neural representation of the spatial and intensive attributes of the visual image. In: Seeing (E. C. Carterette and M. S. Friedman, eds.), Vol. 5 of Handbook of Perception. Academic Press, New York. ROBSON, J. G. and ENROTH-CUGELL, C. (1978) Light distribution in the cat' s retinal image. Vision Res. 1 8 : 159- 173. RODIECK, R. (1965) Quantitative analysis of cat retinal ganglion cell response to visual stimuli. Vision Res. 5:583 - 601. RODIECK, R. (1973) The Vertebrate Retina W. H. Freeman, San Francisco. RODIECK, R. (1979) Visual pathways. A. Rev. Neurosci. 2: 193 - 226. ROSE, A. (1948) The sensitivity performance of the human eye on an absolute scale. J. opt. Soc. Am. 38: 196- 208. RosE, A. (1973) Vision, Human and Electronic. Plenum Press, New York. RUSHTON, W. A. H. (1962) Visual pigments in man. Sci. Am. 139: 2- 10. RUSHTON, W. A. H. (1965) The Ferrier Lecture, 1962. Visual adaptation. Proc. R. Soc. Lond. B 162: 20-46. SAITO, H.-A. and FUKADA, Y. (1975) Research note: Gain control mechanisms within the receptive field center of cat' s retinal ganglion cells. Vision Res. 15:1407 - 1410. SAITO, T., KONr)O, H. and TOYODA, J.-I. (1979) Ionic mechanisms of two types of on-center bipolar cells in the carp retina. I. The responses to central illumination. J. gen. Physiol. 73: 73- 90. SAKITT, B. (1972) Counting every quantum. J. Physiol. 223: 131 - 150. SArMANN, B. and CREUTZEELDT, O. D. (1969) Scotopic and mesopic hght adaptation in the cat's retina. Pflugers Arch. 313: 168- 185. SAKMAm4, B. and FILION, M. (1972) Light adaptation of the late receptor potential in the cat retina. In: Advances in Experimental Medicine and Biology, Vol. 24, The Visual System - - Neurophysiology, Biophysics, and their Clinical Applications (G. B. Arden, ed.) pp. 87- 93. Plenum Press, New York. SCHACHER, S. M., HOLTZMAN, E. and HOOD, D. C. (1974) Uptake of horseradish peroxidase by frog photoreceptor synapses in the dark and the light. Nature 249:261 - 263. SCHELLART, N. A. M. and SPEKREIJSE, H. (1972) Dynamic characteristics of retinal ganglion cell responses in goldfish. J. gen. Physiol. 59: 1- 21. SHAPLEY, R. M. and TOLHURST, D. J. (1973) Edge detectors in human vision. J. Physiol. 229:165 - 183. SHAPLEY, R. M. and VICTOR, J. D. (1978) The effect of contrast on the transfer properties of cat retinal ganglion cells. J. Physiol. 285: 275- 298. SHAI'LEY, R. M. and VICTOR, J. D. (1979) The contrast gain control of the cat retina. Vision Res. 19: 431-434. SHAPLEY, R. M. and VICTOR, J. D. (1980)The effect of contrast on the non-linear response of the Y cells. J. Physiol. 302: 535 - 547. SHAPLEY, R. M. and VICTOR, J. D. (1981) How the contrast gain control modifies the frequency responses of cat retinal ganglion ceils. J. Physiol. 318:161 - 179. SHAPLEY, R. M., ENROTH-CUGELL, C., BONDS, A. B. and KIRBY, A. (1972) Gain control in the retina and retinal dynamics. Nature 236: 352-353. SHAPLEY, R. M., KAPLAN, E. and TRANCHINA, E. (1983) The effects of ambient illumination on contrast sensitivity and dynamics of the cat retina and LGN. Neuroscience 3 4 6 R. M. SHAPLEY AND C. ENROTH-CUGELL Abst r act s, 9: 811. SMITH, R. A. (1973) Lumi nance dependent changes in mesopi c visual cont r as t sensitivity. J. PhysioL 230: 115- 131. So, Y. -T. and SHAPLEY, R. M. (1979) Spat i al pr oper t i es o f X and Y cells in t he l at eral geni cul at e nucl eus. Expl Brain Res. 36: 5 3 3 - 550. SPERLING, G. (1970) Model o f visual adapt at i on and cont r ast det ect i on. Perception & Psychophysics 8: 143 - 157. SPERLING, G. and SONDHI, M. (1968) Model f or visual l umi nance di scr i mi nat i on and flicker det ect i on. J. opt. Soc. Am. 58: 1133- 1145. STEINBERG, R. H. (1971) Increment al responses t o light recorded f r om pi gment epi t hel i al cells and hor i zont al cells o f t he cat ret i na. J. Physiol. 217: 93 - 110. STEINBERG, R. H. , REID, M. and LACY, P. L. (1973) The distribution o f rods and cones in t he retina of t he cat (Felis domesticus). J. comp. Neurol. 148: 2 2 9 - 2 4 8 . STEINHARDT, J. (1936) Int ensi t y di scr i mi nat i on in t he huma n eye. I. The rel at i on o f A I / I t o i nt ensi t y. J. gen. Physiol. 20: 185- 209. STERLING, P. (1983) Mi croci rcui t ry of t he cat ret i na. A. Rev. Neurosci. 6: 149- 185. THmOS, L. N. and WERBLIN, F. S. (1978a) The r esponse pr oper t i es o f t he st eady ant agoni st i c s ur r ound in t he mudpuppy ret i na. J. Physiol. 278: 7 9 - 9 9 . THIBOS, L. N. and WERBLIN, F. S. (1978b) The pr oper t i es of s ur r ound ant agoni s m elicited by spi nni ng wi ndmi l l pat t er ns in t he mudpuppy ret i na. J. PhysioL 278: 101 - 116. THOMAS, J. S. and KOVAR, C. W. (1965) The ef f ect o f cont our shar pness on percei ved br i ght ness. Vision Res. 5: 5 5 9 - 5 7 1 . TONG, L. and GREEN, D. G. (1977) Ada pt a t i on pool s and exci t at i on recept i ve fi el ds of rat ret i nal gangl i on cells. Vision Res. 17: 1233- 1236. TULUNAY-KEESEV, U. and VASSlLEV, A. (1974) Foveal spatial sensi t i zat i on wi t h st abi l i zed vision. Vision Res. 14: 101 - 105. VALETON, M. J. and VAN NORREN, D. (1983) Li ght adapt at i on o f pr i mat e cones: an anal ysi s based on extracellular dat a. Vision Res. 23: 1539- 1547. VICTOR, J. D. and SHAPLEY, R. M. (1979) The nonl i near pat hway o f Y gangl i on cells in t he cat ret i na. J. gen. Physiol. 7 4 : 6 7 1 - 6 8 9 . W~,SSLE, H. (1982) Morphol ogi cal types and central proj ect i ons of gangl i on cells in t he cat ret i na. In: Progress in Retinal Research (N. Os bor ne and G. Chader , eds) Vol . 1, pp. 125 - 152. Pe r ga mon Press, Oxf or d. WERBLIN, F. S. (1974) Cont r ol o f retinal sensitivity. II. Lat eral i nt eract i ons at t he out er pl exi f or m layer. J. gen. Physiol. 63: 6 2 - 87. WERBLIN, F. S. (1977) Synapt i c i nt er act i ons medi at i ng bi pol ar r esponse in t he ret i na o f t he t i ger sal amander . In: Vertebrate Photoreception (H. B. Barl ow and P. Fat t , eds) pp. 2 0 5 - 230. Academi c Press, New Yor k. WERBLIN, F. S. and COPENHAGEN, D. R. (1974) Cont r ol of ret i nal sensitivity, l I I . Lat eral i nt er act i ons at t he i nner pl exi f or m layer. J. gen. Physiol. 63: 8 8 - 110. WERBLIN, F. S. and DOWLING, J. E. (1969) Or gani zat i on of t he ret i na of t he mudpuppy, Nect urus maculosus. I1. Int racel l ul ar r ecor di ng. J. Neurophysiol. 32: 339 - 354. WESTHEIMER, G. (1965) Spat i al i nt er act i on in t he human ret i na dur i ng scot opi c vi si on. 3. Physiol. 181: 8 8 1 - 894. WHITTLE, P. and CHALLANDS, P. D. C. (1969) The effect of background luminance on t he brightness of flashes. Vision Res. 9: 1095 - l 110. WILLIAMS, T. S. and GALE, J. G. (1977) A cri t i que o f an incremental t hreshol d funct i on. Vision Res. 17:881 - 882. WITKOVSKY, P. (1967) A compar i s on of gangl i on cell and S- pot ent i al r e s pons e pr ope r t i e s in car p r et i na. J. Neurophysiol. 30: 5 4 6 - 5 6 1 . WYSZECKI, G. and STILES, W. S. (1967) Color Science: Concepts and Met hods, Quantitative Data and Formulas. pp. 2 1 2 - 2 2 7 , Wiley, New Yor k. YOON, M. (1972) I nf l uence o f adapt at i on level on r esponse pat t ern and sensitivity o f gangl i on cells in t he cat ' s retina. J. Physiol. 221: 93 - 104.