JOURNAL OF THE OPTICAL SOCIETY OF AMERICA

Integrative Processes in Central Visual Pathways of the Cat*

DAvID H. HUBELNeurophiysiology Laboratory, Harvard Medical School, Boston, Massachusetts

One may study the visual system by stimulating the retina with spots or patterns of light and recordingfrom single cells at successive stages in the visual pathway. By comparing response properties of the cells in agiven structure with those of the fibers feeding into it we can attempt to learn something of how the structuremodifies the visual information it receives. A description is given of responses of single cells in the opticnerve, the lateral geniculate body, and the visual cortex of the cat.

AN image falling upon the retina exerts an influenceon millions of receptors. It is the task of the cen-

tral nervous system to make sense of the spatial andtemporal patterns of excitation in this retinal mosaic.Unless we know something of how the nervous systemhandles the messages it receives, we cannot easily cometo grips with the problems of perception of form, move-ment, color, or depth.

For a study of integrative sensory mechanisms thevisual system of mammals offers the advantage of acomparatively direct anatomical pathway. At eachstage, from bipolar cells to the striate cortex, we cancompare activity of cells with that of the incomingfibers, and so attempt to learn what each structurecontributes to the visual process. In this paper Isummarize a series of studies on the cat visual systemmade by Torsten Wiesel and myself. I concentratemainly on experiments related to form and movement.

It is often contended that in studying a sensorysystem we should first learn to understand thoroughlythe physiology of receptors, and only then proceed toexamine more central processes. In the visual systemone should presumably have a firm grasp of rod andcone physiology before looking at bipolar and retinalganglion cells; one should thoroughly understand retinalmechanisms before taking up studies of the brain.Unfortunately it is not always possible to be so system-atic. In the case of the visual system, orderly progressis impeded by the great technical difficulties in recordingfrom single retinal elements, especially from the rodsand cones and from bipolar cells. At the single-cell level,knowledge of the electrophysiology of these structuresis consequently almost entirely lacking. If we wish tolearn how the brain interprets information it receivesfrom the retina we must either struggle with retinalproblems of formidable difficulty or else skip over thefirst two stages and begin at a point where the appro-priate single-unit techniques have been worked out, i.e.,the retinal ganglion cell.

The subject of retinal ganglion-cell physiology iscomplicated by the fact that studies have been made in

* Invited paper presented at the Symposium on PhysiologicalOptics, Joint Session of the Armed Forces-NRC Committee onVision, the Inter-Society Color Council, and the Optical Societyof America, 14-15 March 1962, Washington, D. C.

a wide variety of vertebrates and under a number ofdifferent experimental conditions. Here I only describethe receptive field organization of retinal ganglion cellsin the cat. This is necessary for an understanding of theintegrative function of the lateral geniculate body,since the geniculate receives its main visual inputdirectly from the retina.

Because there is convergence of a number of afferentfibers onto each cell, both for bipolar cells and for retinalganglion cells, we are not surprised to learn that a singleganglion cell may receive its input ultimately from alarge number of rods and cones, and hence from aretinal surface of considerable extent. At first glance itmight seem that a progressive increase in the size ofreceptive fields as we follow the visual pathway centrallymust lead to a wasteful and pointless blurring of de-tailed information acquired by the exquisitely finereceptor mosaic. To understand why fineness of dis-crimination is not necessarily blunted we must realizethat all retinal connections are not necessarily excita-tory. The existence of inhibitory connections meansthat when we shine a spot of light on the receptive fieldof a given cell we may decrease, rather than increase,the cell's rate of firing. The effect of the stimulus willdepend on the part of the receptive field we illuminate.The fineness of discrimination of a cell is determined notby the over-all receptive field size, but by the arrange-ment of excitatory and inhibitory regions within thereceptive field.

With the experiments of Kuffler (1953) it becameapparent that in the light-adapted cat, retinal ganglioncells did not necessarily respond uniformly throughouttheir receptive fields: their discharges might be acti-vated or suppressed by a spot of light, depending uponwhere on the retina the spot fell. The receptive field ofa ganglion cell could thus be mapped into distinctexcitatory and inhibitory regions. Two types of cellswere distinguished by Kuffler: those with fields havinga more or less circular excitatory center with an annularinhibitory surround, and those having the reversearrangement. These two concentrically arranged fieldtypes were called "on"-center and "off"-center fields.The terms "off" center and "off" response refer to theempirical finding that when a spot of light suppresses acell's firing, turning the spot off almost always evokes

58

VOLUME 53, NUMBER JANUARY 1963

CENTRAL VISUAL PATHWAYS OF CAT

a discharge, termed the "off" response. Converselywhen we see an "off" discharge we usually find thatduring the stimulus the maintained firing of a cell issuppressed.

Within the excitatory or inhibitory region of areceptive field one can demonstrate summation, i.e., fora given intensity of stimulus the response increases(number of spikes and frequency of firing increase,latency and threshold decrease) as the area stimulatedis increased. On the other hand, when both types ofregion are included in a stimulus their separate effectstend to cancel. If the entire receptive field is illuminated,for example by diffuse light, a relatively weak responseof the center type is usually obtained: an "on"-centercell thus gives a weak "on" response, and an "off"-center cell a weak "off" response. I shall use the term"peripheral suppression" to refer to this antagonisticinteraction between center and periphery.

Retinal ganglion cells differ from one another inseveral ways besides those related to field-center type.Obviously they vary in the location of their receptivefields on the retina. In the cat (Wiesel 1960) and themonkey (Hubel and Wiesel 1960) there are considerabledifferences in the sizes of receptive-field centers, recep-tive fields near the area centralis or fovea showing amarked tendency to have smaller centers than fields inthe peripheral retina. Even for a given region of theretina there is a large variation in the size of fieldcenters. In the monkey the smallest field center so farmeasured had a diameter of 4 minutes of arc; this wassituated 4 from the fovea. It is likely that the centersof foveal fields are much smaller than this. The totalextent of a field is more difficult to determine, since theeffect of a spot of light upon a cell decreases graduallywith increasing distance from the field center. Measure-ments made by constructing area-threshold curves(Wiesel 1960) suggest that receptive fields may notgreatly differ in their total size despite wide variationsin center sizes.

Retinal ganglion cells differ also in the effectivenesswith which the receptive-field periphery antagonizes thecenter response. This may be measured by determiningthe difference between the threshold intensity of a spotcovering the receptive-field center and that of a largespot covering the entire field. The difference tends to begreater for cells with small field centers (and hencelarge peripheral zones) than for cells with large fieldcenters. Since cells with small field centers are especiallycommon in the area centralis, this ability to discriminateagainst diffuse light is particularly pronounced in thatpart of the retina.

In the cat we know of no functional retinal ganglioncell types besides the "on"-center and "off"-center cellsdescribed by Kuffler. Occasionally diffuse light evokesa discharge both at "on" and at "off." This may occurin either "on"-center cells or "off"-center cells. It

depends to some extent on the state of light adaptation,stimulus intensity, and other variables. The receptivefields of cells showing "on-off" responses to large spotsdo not seem to differ in any fundamental way fromordinary "on"-center or "off"-center cells. There thusseems to be no reason for regarding "on-off" retinalganglion cells of cat as a distinct type.

In the cat the arrangement of excitatory and in-hibitory regions within a given receptive field remainsthe same for all effective stimulus wavelengths. Thefields thus seem to be very different from the morecomplex opponent-color fields described by Wolbarsht,Wagner, and MacNichol (1961) for goldfish retinalganglion cells (see discussion of Barlow, in Wolbarshtet al., 1961, p. 176). In the monkey optic nerve andlateral geniculate body there are two types of neurones,one resembling cells of the cat in having receptive-fieldcharacteristics that are independent of wavelength, theother showing color-specific responses in many wayssimilar to those seen in the goldfish (Hubel and Wiesel1960; DeValois 1960).

In the frog, Maturana, Lettvin, McCulloch, andPitts (1960) have described retinal ganglion cells withhighly complex response properties. Their records weremade from unmyelinated axons or their terminalarborizations. If such axons exist in the optic nerves ofcats, they have probably escaped detection in physio-logical studies. Unfortunately cat optic nerves have notyet been examined with the electron microscope, and itis not known whether or not they contain unmyelinatedfibers.

LATERAL GENICULATE BODY

Anatomically, the dorsal lateral geniculate bodydiffers from most other structures in the central nervoussystem, and certainly from the retina and cortex, in itsrelative simplicity. In a sense it is a one-synapse waystation, since its cells receive their major input directlyfrom the optic tract, and since most of them send theiraxons directly to the visual cortex. It has often beenasked whether the lateral geniculate body serves anyintegrative purpose besides that of relaying incomingmessages to the cortex for further elaboration. Althoughin some ways the lateral geniculate body is a simplestructure, an anatomist would hardly contend that it isnothing but a one-to-one relay station. The existence ofconvergence and divergence, complex dendritic arbori-zations, and, in the cat at least, cells with short axonsterminating within the nucleus itself, all seem to beagainst such a supposition.

A strong case was made for the presence of one-to-onesynapses in the geniculate by the microelectrodestudies of Bishop, Burke, and Davis (1958). By elec-trically stimulating the severed proximal stump of theoptic nerve and recording extracellularly from lateralgeniculate cells they were able to record excitatory

January 1963 59

DAVID H. HUBEL

postsynaptic potentials (or the associated extracellularcurrents) and show that they were not continuouslygraded, but, at least for two of the cells they studied,were all-or-nothing. Most excitatory postsynapticpotentials were followed by geniculate spikes. The au-thors concluded that some lateral geniculate cells canbe excited by one impulse in a single optic-nerve fiber.They were inclined to attribute the fact that the lateralgeniculate cell occasionally failed to fire to the effectsof anaesthesia, rather than to variation in possibleadditional inputs not detected by their electrode. Sinceboth Bishop, Burke, and Davis (1958) and Freygang(1958) observed lateral geniculate cells for which theexcitatory postsynaptic potentials were graded inseveral discrete steps, it is clear that not all geniculatesynapses are of a simple one-to-one type. At least somemust have several excitatory inputs.

If there is a "straight through" connection betweensome optic-nerve fibers and lateral geniculate cells, asBishop's findings suggest, there should be no differencesin receptive fields at the two levels. A study of lateralgeniculate cells in the cat (Hubel and Wiesel 1961)showed that lateral geniculate fields indeed have thesame concentric center-periphery organization, and likeretinal ganglion cells, are of two types, excitatory centerand inhibitory center. It is clear enough, then, that inthe lateral geniculate body there is no very profoundreorganization of the incoming messages. Nevertheless,there was a suggestion that the ability of a receptive-field periphery to antagonize the center response wasmore marked in geniculate cells than in optic-nervefibers. This was true even when variations in peripheralsuppression with position of receptive fields on theretina (referred to above) were taken into account.

The fact that one can record geniculate spikes to-gether with excitatory synaptic potentials of an all-or-none type suggested the possibility of making a moredelicate test of geniculate function, namely, a compari-son of the responses and receptive fields of a particulargeniculate cell with those of its own excitatory post-synaptic potential (Hubel and Wiesel 1961). When thiswas done for cells with all-or-none synaptic potentials,it was found that while almost all lateral geniculatespikes were triggered by an optic-nerve impulse, theconverse was not true; each synaptic potential did notnecessarily trigger a postsynaptic spike. The successrate of the optic-nerve impulse varied widely, dependingon how the retina was stimulated. The receptive fieldcenters of the optic nerve fiber and the lateral geniculatecell were, as far as one could judge, precisely super-imposed. If one shone a restricted spot of light over thecommon receptive-field center, the likelihood that animpulse would trigger a postsynaptic spike was veryhigh. If, on the other hand, the entire receptive fieldincluding the periphery was illuminated, very few of thesynaptic potentials were followed by geniculate-cell

spikes. For small spots in the center portion of the fieldthe thresholds of the two units were apparently iden-tical, but for large spots, including diffuse light, theywere often several log units apart. Sometimes thegeniculate cell would not respond to diffuse light at anyintensity.

It was thus possible not only to confirm the impres-sion that peripheral suppression is enhanced by lateralgeniculate cells, but to obtain some notion of how thechange is brought about.

This result shows clearly that even when we record asingle all-or-none excitatory synaptic potential alongwith a geniculate cell spike, the synaptic potential weobserve does not represent the only input to the cell.There must be other inputs which are influenced byilluminating the periphery of the common receptivefield. Illuminating the periphery might activate retinalganglion cells whose "on" centers were distributed overthis annular region; if these neurons made inhibitorysynaptic connections with the geniculate cell we couldexplain the cell's failure to be triggered when diffuselight was used. We might equally well suppose thatlighting the periphery suppressed the firing of a set of"off"-center retinal ganglion cells making excitatoryconnections with the geniculate cell. Now inclusion ofthe receptive-field periphery would suppress these cells,removing the tonic asynchronous activation needed toenable the geniculate cell to follow the triggering im-pulses. The important point is that the geniculate cellmust be receiving input from not one, but a largenumber of optic-nerve fibers. In a cell bound to anoptic nerve fiber by a synapse having a "straightthrough" property, the property is a conditional one,depending on activity of other optic nerve fibers.

I have mentioned the possibility of suppressing acell's firing by withdrawal of tonic excitation rather thanby direct synaptic inhibition. The synapse that we arediscussing gives us a vivid example of just that, for inilluminating the field center of an "off"-center genicu-late cell we suppress firing by suppressing activity inthe main optic-nerve fiber feeding into it. Of course thecell may at the same time be actively inhibited by otheroptic-nerve fibers (we have no evidence for or againstthis), but this inhibition would not be the main reasonfor the cessation of firing. To suppress the firing of anycell in the visual system there need only be one in-hibitory link in the entire chain beginning with andincluding the receptor. In the case of the center of an"off"-center cell in the lateral geniculate we do notknow at what stage this inhibitory link occurs. It isapparently not in the geniculate, and there is no evi-dence for or against its being at the retinal ganglion-celllevel.

We may sum up the implications of these experimentsas follows: (1) all cat geniculate cells apparently havemultiple visual inputs; (2) there is often a particular

60 Vol. 53

January 1963 CENTRAL VISUAL

relationship between a cell and one optic-nerve fiberwith which it makes a powerful excitatory synapse; (3)when such a relationship exists, the receptive-fieldcenters of the incoming fiber and the geniculate cell areof the same type, i.e., both are "on"-center or both are"off"-center; (4) the lateral geniculate body has thefunction of increasing the disparity, already present inthe retinal ganglion cell, between responses to a smallcentered spot and to diffuse light.

The lateral geniculate body may have other functionsbesides that of increasing the effects of the receptive-field periphery. Cells in which the synaptic potential isgraded in several steps must have more than oneexcitatory afferent. This kind of convergence mightproduce a geniculate receptive-field center larger thanany of the field centers of the afferents. So far this hasnot been tested experimentally.

Some electrophysiological studies have suggested thatthe lateral geniculate body receives afferent fibersbesides those of the optic tract (Hubel 1960; Widen andAjmone-Marsan 1960; Arden and Sderberg 1961).Nauta and Bucher (1954) have observed a cortico-geniculate projection in the rat, and recently Nauta(personal communication) and Beresford (1961) havefound in the cat a topographically precise reciprocalpathway from the striate cortex to the lateral geniculatebody of the same side. So far we have found no genic-ulate cells with the complex properties typical ofcortical cells, but fibers with these properties arefrequently recorded just dorsal to the lateral geniculatebody. A knowledge of the presence of a reciprocal path-way is important if we are to avoid including these unitsin a study of geniculate cells, particularly if there is anychance that the recording electrode is not in the genic-ulate but just above it.

A problem that has attracted considerable attentionconcerns the amount of binocular interaction in thelateral geniculate body (Bishop, Burke, and Davis 1959;Erulkar and Fillenz 1960; Griisser and Sauer 1960;Hubel and Wiesel 1961). While there is evidence thatsome geniculate cells can be influenced from the twoeyes, it seems to be agreed that the proportion ofbinocularly influenced cells in the lateral geniculatebody is small. This is certainly in keeping with theanatomical findings (Silva 1956; Hayhow 1958). Wehave so far not succeeded in mapping out, for anygeniculate cell, two receptive fields, one in each eye.The marked contrast between the scarcity of binocularinteraction in the cat's geniculate and its preponderancein the visual cortex does not argue for any major roleof the geniculate in binocular vision.

On anatomical grounds it is well established thatalternate layers of the lateral geniculate body receivetheir input from alternate eyes. This has been con-firmed in the cat by physiological methods (Cohn,1956); cells in a given layer can be driven from one of

PATHWAYS OF CAT 61

the two eyes, but not from the other. A precise topo-graphical representation of the contralateral half-fieldsof vision on each geniculate layer, the maps in thedifferent layers being in register, has been establishedanatomically for the rhesus monkey (Polyak 1957) bynoting trans-synaptic atrophy following small retinallesions. Although a similar anatomical study in the cathas not been made, the physiological evidence for a pre-cise topographical representation in this animal is clear(Hubel and Wiesel 1961). The receptive fields of simul-taneously recorded cells are near to one another andoften overlap almost completely. The receptive fields ofcells recorded in sequence by an electrode passingnormal to the layers are close together or almost super-imposed, whereas in an oblique or tangential penetra-tion, fields of successively recorded cells move system-atically along the retina. Finally, the maps in successivelayers are in register.

From what I have said about the lateral geniculatebody it will be apparent that the physiological prop-erties of even that simple structure are far from simple.The fact that a number of incoming optic-tract fibersconverge upon one cell presents us with a number ofpossibilities. Any particular geniculate cell will haveits own receptive field with center and surround. Eachfiber converging upon the cell will have its own centerlocated in the center or surround of the geniculate cell'sfield: the incoming fiber may have an "on" center oran "off" center; the synapse it makes may be excitatoryor inhibitory. If excitatory, the synapse may be power-ful, capable of setting up a spike in the geniculate cell;or it may be weak, contributing to the summed effectsof a large number of other incoming fibers. Somehowthese and perhaps other possibilities are made use of,to produce a mechanism in which individual incomingimpulses may trigger individual postsynaptic impulses,but in which the coupling between the incoming andoutgoing signals is varied. Such an ingenious piece ofmachinery would surely have great appeal to a me-chanical or an electrical engineer. It may be worthstressing how different this synapse seems to be fromthat of the anterior horn cell of the spinal cord, which,because it has been so extensively studied by modernelectrophysiological methods, is apt to be taken as aprototype of synapses in the central nervous system.

VISUAL CORTEX

If the lateral geniculate body is anatomically astructure of relative histological simplicity, the primaryvisual cortex is in contrast one of very great complexity.There is considerable order to the architectural plan ofthe cortex, yet our knowledge of the connectionsbetween cells gives us very little notion of how thisstructure functions. Of course, it has been known foryears that the striate cortex is concerned with vision,and that in most mammals it is indispensible for form

DAVID H. HUBEL

vision. What we have not known is how cortical cellshandle the messages they receive from the lateralgeniculate body. We have had insufficient evidence evento decide whether the messages are modified at all, orjust handed on to some still higher centers for furtherelaboration (cf. Brindley 1960, p. 122).

As long as methods for single-cell recording were notavailable to neurophysiologists, this question of inte-grative cortical mechanisms could only be approachedin a limited way. Since gross electrodes record onlysynchronous activity one could only examine attributesshared by all or most cells in a relatively large volumeof tissue (the order of 1 mm 3). We know now that theone important physiological quality shared by cellsover such a large area of striate cortex has to do withthe regions of visual field from which cells receive theirprojections. It is therefore not surprising that topo-graphy was one aspect of visual cortical function to beextensively explored with gross electrodes.

In a series of studies by Talbot and Marshall (1941),Talbot (1942), and Thompson, Woolsey, and Talbot(1950) the cortex was mapped in the cat, rabbit, andmonkey according to the retinal areas projecting to it.These authors were able to go well beyond what wasknown from anatomical studies by showing that in thecat and rabbit there is a double representation of thevisual half-field on the cortex of the contralateralhemisphere. The two maps lie adjacent to each other,bounded by a line which Talbot and Marshall termedthe "line of decussation." This line receives projectionsfrom the vertical meridian. Any retinal region (besidesthe vertical meridian) projects to two regions on thecortex, one medial to the line of decussation and theother lateral to it. There has been some tendency toassume that the medial representation, called VisualArea I, is the classical striate cortex, whereas Area IIis nonstriate. There is nothing in the literature to sup-port the latter assumption, though to my knowledge ithas never been questioned except by Bard (1956).

The mapping experiments of Talbot and Marshalland of Thompson, Woolsey, and Talbot have since beenconfirmed for the cat by single-unit techniques. Wehave confirmed the topographical projection scheme inthe cat (Hubel and Wiesel 1962), including the presenceof a second visual representation lateral to the first.Daniel and Whitteridge (1961) have repeated theexperiments in the monkey and have extended the mapto buried parts of the cortex. Although Talbot andMarshall did not describe a second visual area in themonkey, Wiesel and I have recently found electro-physiological evidence for a precise retinotopic projec-tion to nonstriate visual cortex.

The introduction of microelectrodes supplies uswith a powerful means of studying properties ofindividual cortical cells, especially those properties thatare not common to cells in a large volume of nervous

tissue. To learn what kinds of transformations thevisual cortex makes on the incoming visual signals wemay compare responses of single cortical cells withthose of afferent fibers from the lateral geniculate body.If we were to find no differences in receptive fields ofcells in these two structures we would indeed be disap-pointed, for it would mean either that in spite of itsanatomical complexity the striate cortex did virtuallynothing, or else that our present microelectrode tech-niques were not equal to the problem. The secondalternative is a possible one, since the elaborativefunctions of the cortex might be discernable only byexamining simultaneously large numbers of cells andcomparing their firing patterns, perhaps with the helpof computers. As it turns out, there are differences inreceptive fields, differences which give us a fair idea ofsome of the functions of the cortex. Here I only attemptto summarize some of our own work (see Hubel andWiesel 1959, 1962); for other microelectrode investiga-tions of the visual cortex the reader may refer to severalrecent symposia (Rosenblith 1961; Jung and Kornhuber1961).

In the striate cortex we have found no cells withconcentric "on"- or "off"-center fields. Instead therehas been an astonishing variety of new response types.These differ one from another in the details of distribu-tion of excitatory and inhibitory regions, but they haveone thing in common: that areas giving excitation andinhibition are not separated by circles, as in the retinaand geniculate, but by straight lines. Some cells, forexample, have receptive fields with a long narrowexcitatory area flanked on either side by inhibitoryareas, whereas others have the reverse arrangement, aninhibitory area flanked on the two sides by excitatoryareas (Hubel and Wiesel 1962, Text-Fig. 2). Some fieldshave only two regions of opposite type separated by asingle straight line. Summation occurs just as in theretina and geniculate, and the most effective stationaryretinal stimulus for a cortical cell is one falling oneither the excitatory parts of a receptive field or theinhibitory parts, but not on both simultaneously. Con-sequently stimuli such as long narrow dark or lightrectangles, or boundaries with light to one side anddarkness to the other ("edges"), are likely to be themost potent for cortical cells. Each cell will have itsown optimal stimulus. Moreover the stimulus thatworks best in influencing a cell, exciting or inhibiting it,will do so only when shone on the appropriate part ofthe retina and in the correct orientation. Some cellsprefer one inclination, vertical, horizontal, or oblique,others prefer another; and all inclinations seem to beabout equally well represented. We have termed theinclination of the most effective stimulus the "receptive-field axis orientation" and have come to realize thatthis is one of a cell's most important properties. Forexample, if a stimulus such as a long narrow rectangle

62 Vol. 53

January1963 CENTRAL VISUAL

of light is shone at right angles to the optimum orienta-tion it has little or no effect. Here the light coversportions of both the excitatory and inhibitory regions,and the two effects oppose each other.

We have already seen that turning on or off a diffuselight is not an ideal stimulus for a retinal ganglion cell.It evokes a response, but a much weaker one than thatproduced by a centered circular spot of just the rightsize. I have described how cells in the lateral geniculatebody are influenced even less than retinal ganglion cellsby diffuse light. In the cortex the process is apparentlycarried a step further. Here many cells give no responseat all when one shines light on the entire receptive field.How the cat detects diffuse light and distinguishesdifferent levels of diffuse illumination is something wedo not know. Perhaps the mechanism is subcortical; itis known that the cat can make discriminations ofintensity of diffuse light even when it lacks a visualcortex (Smith 1937). The information that a large patchof retina is evenly illuminated may be supplied only bycells that are activated by the boundaries of the patch;the fact that cells with fields entirely within the illumin-ated area are uninfluenced presumably signals the ab-sence of contours within the patch of light-in otherwords, that the region is diffusely lit.

One may ask why a diffuse flash of light evokes sucha large cortical slow wave, if only a small proportionof cells respond to the stimulus, and these only relativelyweakly. Too little is known about slow waves to permitan entirely satisfactory answer. It is possible that alarge slow wave may be produced by a small proportionof cells firing weakly but synchronously. It is interest-ing, however, that the visual evoked response is maximaloutside the cortical area commonly accepted as striate(Doty 1958), and that within the primary visual areait is maximal well in front of the area centralis repre-sentation. Indeed, the area representing central visiongives only a relatively feeble response to a diffuse flash(Doty 1958). We have not thoroughly explored corticalareas representing the far periphery of the retinas: itmay be that compared with cells receiving projectionsfrom centralis, those with receptive fields in the farperiphery respond more actively to diffuse light. This is,in fact, the case with retinal ganglion cells (Wiesel 1960)and with geniculate cells (Hubel and Wiesel 1961).

The amazing selectivity with which cortical cellsrespond to a highly specific stimulus and ignore almostanything else is explained by the existence of excitatoryand inhibitory receptive-field subdivisions. While thesemechanisms clearly make use of inhibition, it must bestressed that we have no direct evidence that thecortex contains inhibitory synapses, just as we havenone in the case of geniculate or retinal ganglion cells(see discussion of Bremer, in Jung 1960, p. 233). When-

ever we suppress firing by turning on a stimulus, theeffect may be produced by withdrawing tonic excitation

PATHWAYS OF CAT

as easily as by directly inhibiting, and so far the appro-priate methods of distinguishing the two possibilitieshave not been used in the visual system.

In their behavior cells whose receptive fields can bedivided into excitatory and inhibitory regions are proba-bly the simplest of the striate cortex. It is thereforereasonable to suppose that at least some cells withsimple fields receive their projections directly from thegeniculate (Hubel and Wiesel 1962; Text-Fig. 19). Inthe striate cortex we find cells of a second type whoseproperties we have called "complex." Cells with com-plex receptive fields do not respond well to small spotsof light, and it has not been possible to map their fieldsinto separate excitatory and inhibitory regions. Theybehave as though they received their afferents from alarge number of cortical cells with simple fields, all ofthese fields having the same axis orientation, but vary-ing slightly from one to the next in their exact retinalpositions. A complex cell thus responds to an appro-priately oriented slit, edge, or dark bar, not just whenit is shone in one highly critical retinal position, as wefind with simple cells, but over considerable regions ofretina, sometimes up to 5°-10 or more. Presumablywhenever the properly oriented stimulus is appliedwithin this area, it activates some cells with simple fields(different ones for different positions of the stimulus)and these in turn activate the complex cell. For example,a typical complex cell might be activated by a horizontalslit of light regardless of its exact position within aregion several degrees in diameter. For such a cellchanging the orientation by more than 5°-10 rendersthe stimulus ineffective, as does making it wider thansome optimum width (e.g., more than 4°). It is asthough such a cell had the function of responding to theabstract quality "horizontal," irrespective of the exactretinal position.

The idea that a complex cell receives its input froma large number of simple cells all having the samereceptive-field axis orientation has a remarkable parallelin the functional anatomy of the cat striate cortex.Cells that are close neighbors almost always havereceptive-field axis orientations that are, as far as onecan tell, identical. By making long penetrations in themanner of Mountcastle (Mountcastle 1957; Powell andMountcastle 1959) one can show that the regions ofconstant axis orientation extend from surface to whitematter, with walls perpendicular to the cortical layers(Hubel and Wiesel 1962). Within one of these regions,or "columns," there occur all functional types of cell,including simple and complex. All the cells in a columnhave their receptive fields in the same general region ofretina, but there is a slight variation in exact receptive-field position from one cell to the next. If we assumethat a complex cell receives its input from cells withsimple fields in the same column, this constancy ofreceptive-field axis orientation together with the slight

63

DAVID H. HUBEL

differences in position of fields is sufficient to accountfor all of the complex cell's properties. A column is thusconsidered to be a functional unit of cortex, to whichgeniculate axons project in such a way as to producesimple cortical fields all with the same axis orientation,and within which simple cells converge upon complexones.

From the standpoint of cortical physiology it isinteresting that these visual columns are in many waysanalogous to the columns in the cat somatosensorycortex, described in 1957 by Mountcastle, and confirmedfor the monkey by Powell and Mountcastle (1959). Acolumnar organization may be a feature of manycortical areas. It seems surprising that this type oforganization, which must depend primarily on anato-mical connections, should have no known anatomicalcorrelate.

As far as we know all striate cortical cells in the catcan be categorized as simple or complex; there do notseem to be still higher orders of cells in this part of thebrain. We are inclined to think of complex cells as repre-senting a stage in the process of form generalization,since we can displace an image by several degrees on theretina, as long as we do not rotate it, and the populationof complex cells that is influenced by the borders of thestimulus will not greatly change. The same is true if wedistort the image, for instance by making it smaller orlarger. As far as we know, this is the first stage in themammalian visual pathway in which such an abstract-ing process occurs.

It is important to realize again that the size of areceptive field does not have any necessary bearing ona cell's ability to discriminate fine stimuli. In the cat atypical cortical receptive field in or near the areacentralis may have a diameter of 1-20, and complexfields range in size from 2°-3° up to 100 or more. Never-theless the optimum stimuli for these cells are likely tobe of the order of 10 minutes of arc in width. In asimple field this corresponds to a dimension such as thewidth of a long narrow receptive-field center. Thepresence of convergence at each stage of the visualpathway does lead to increased receptive-field size, butnot to a loss of detail. This is the result of an interplaybetween inhibitory and excitatory processes.

So far I have not made any reference to one of themost important aspects of vision, namely movement.A moving stimulus commands attention more than astationary one; clinically, movement is generally one ofthe first types of visual perception to return after acortical injury (for references, see Teuber, Battersby,and Bender 1960, p. 19); even for the perception ofstationary objects, eye movements are probably neces-sary (Ditchburn and Ginsburg 1952; Riggs, Ratliff,Cornsweet, and Cornsweet, 1953). It is not surprising,then, to find that a moving spot or pattern is in generala powerful stimulus for cortical cells. To understand

why this is so we must return for a moment to aconsideration of cells with simple fields. If we bring aspot from a neutral region of retina into a cell's excita-tory area we produce an "on" response; if we remove aspot from the "off" region of a cell we evoke an "off"discharge. If we combine the two maneuvers by movinga spot from an "off" area into an "on" area, the twomechanisms work together to produce a greatly en-hanced response. Of course, the cortical cell is mostefficiently activated by the stimulus if it is a slit, darkbar, or edge, and if it is oriented in the direction appro-priate for the cell. If the receptive field of the cell is notsymmetrical (if one flank is smaller or produces lesspowerful effects than the other), the responses to twodiametrically opposite directions of movement may bedifferent. For example, a cell may fire when a spot ismoved from left to right across the retina, but not whenit is moved from right to left.

Now let us consider how a moving stimulus influencesa complex cell. According to the scheme proposedabove, a cell with complex properties receives its inputfrom a number of cells with simple fields whose positionsare staggered. Because of these differences in fieldposition, a moving stimulus will activate first one simplecell and then another. The complex cell will thus becontinuously bombarded and will fire steadily as thestimulus moves over a relatively wide expanse of retina.A stationary stimulus shone into the receptive field ofa complex cell evokes as a rule only a transient responsebecause of the adaptation which presumably occurs atthe receptors and at subsequent synapses. The movingpattern would bypass much of this adaptation by ac-tivating many cells in sequence.

The same mechanism may play a part in the percep-tion of stationary objects by making use of the saccadiceye movements which, at least in man, seem necessaryfor the persistence of a visual impression. A visualimage as it passes across the moving retina presumablyactivates numbers of simple cells briefly and in sequence,leads to a more steady activation of a much smallernumber of complex cells.

From what has been said so far it will be apparentthat the striate cortex has a rich assortment of func-tions. It rearranges the input from the lateral geniculatebody in such a way as to make lines and contours themost important stimuli. Directionality of stimuli mustbe accurately specified; the presence of a columnarsystem based on receptive-field axis orientation testifiesto the importance of this variable. What appears to bea first step in the process of perceptual generalizationresults from a cell's responding to a property of aboundary (its orientation) apart from its exact position.Movement also becomes an important stimulus param-eter, whose rate and direction both must be specified ifa cell is to be effectively driven.

To this list one more function must be added, that of

64 Vol. 53

January 1963 CENTRAL VISUAL

combining the pathways from the two eyes. In con-trast to the lateral geniculate body, most cells in thecat cortex (probably at least 85%) receive input fromthe two eyes (Hubel and Wiesel 1962, Part II). Bymapping out receptive fields in each eye separately andcomparing them we can begin to learn about themechanisms of binocular vision, and perhaps ultimatelysomething about binocular depth perception andbinocular rivalry.

The primary visual, or striate, cortex is probablyonly an early stage of the visual pathway. Yet, un-fortunately, we have very little knowledge of thepathway from this point on. Except in the rat (Nautaand Bucher 1954) and cat (Beresford 1961) the pointsto which the striate cortex projects are not known.Even less is known about the connections of the neigh-boring nonstriate visual cortex, called 18 and 19, orparastriate and peristriate; we have no accurate de-scription of what areas project to them, or of wherethey send their projections. There even seem to bedoubts as to the validity of the distinction between thetwo areas (Lashley and Clark 1946). Clearly, more willhave to be learned about the anatomy before neuro-physiologists can make much progress in parts of thepathway beyond the striate cortex.

The work I have described may help to show howvisual messages are handled by the brain, at least in theearly stages of the process. The analysis takes us towhat are probably at least sixth-order neurons in thevisual pathway. Our understanding of cells with com-plex fields will be incomplete until we know how theseproperties are used at the next stage of integration,just as our grasp of the significance of retinal and genicu-late receptive-field organization was incomplete withouta knowledge of cortical receptive fields. There is noway of foreseeing what the next transformations willbe, but to judge from what we have learned so far onewould guess that the process of abstraction will go on,and that response specificity will increase. But it is wellto remember that central nervous physiology is in adescriptive and exploratory phase. Our ignorance ofCNS processes is such that the best predictions standa good chance of being wrong.

REFERENCES

G. Arden and U. Sderberg, "The Transfer of Optic Informationthrough the Lateral Geniculate Body of the Rabbit," in SensoryCommunication, edited by W. A. Rosenblith (MIT Press andJohn Wiley & Sons, Inc., New York, 1961), pp. 521-544.

P. Bard, Medical Physiology (C. V. Mosby Company, St. Louis,Missouri, 1956), p. 1176.

W. A. Beresford, "Fibre Degeneration following Lesions of theVisual Cortex of the Cat," in Neurophysiologie und Psychophysikdes visuellen Systems, edited by R. Jung and H. Kornhuber(Springer-Verlag, Berlin, 1961).

P. 0. Bishop, W. Burke, and R. Davis, "Synapse Discharge bySingle Fibre in Mammalian Visual System," Nature 182,728-730 (1958).

, "Activation of Single Lateral Geniculate Cells by Stimula-tiQn of Either Optic Nerve," Science 130, 506-507 (1959),

PATHWAYS OF CAT 65

G. S. Brindley, Physiology of the Retina and the Visual Pathway(Edward Arnold, Ltd., London, 1960).

R. Cohn, "Laminar Electrical Responses in Lateral GeniculateBody of Cat," J. Neurophysiol. 19, 317-324 (1956).

P. M. Daniel and D. Whitteridge, "The Representation of theVisual Field on the Cerebral Cortex in Monkeys," J. Physiol.(London) 159, 203-221 (1961).

R. L. De Valois, "Color Vision Mechanisms in the Monkey,"J. gen. Physiol. 43, Pt. 2, 115-128 (1960).

R. W. Ditchburn and B. L. Ginsborg, "Vision with StabilizedRetinal Image," Nature 170, 36-37 (1952).

R. W. Doty, "Potentials Evoked in Cat Cerebral Cortex byDiffuse and by Punctiform Photic Stimuli," J. Neurophysiol.21, 437-464 (1958).

S. D. Erulkar and M. Fillenz, "Single-Unit Activity in the LateralGeniculate Body of the Cat," J. Physiol. (London) 154, 206-218(1960).

W. H. Freygang, Jr., "An Analysis of Extracellular Potentialsfrom Single Neurons in the Lateral Geniculate Nucleus of theCat," J. gen. Physiol. 41, 543-564 (1958).

O.-J. GrUsser and G. Sauer, "Monoculare und binoculare Licht-reizung einzelner Neurone im Geniculatum laterale der Katze,"Pflug. Arch. ges. Physiol. 271, 595-612 (1960).

W. R. Hayhow, "The Cytoarchitecture of the Lateral GeniculateBody in the Cat in Relation to the Distribution of Crossed andUncrossed Optic Fibers," J. comp. Neurol. 110, 1-64 (1958).

D. H. Hubel, "Single Unit Activity in Lateral Geniculate Bodyand Optic Tract of Unrestrained Cats," J. Physiol. (London)150, 91-104 (1960).

D. H. Hubel and T. N. Wiesel, "Receptive Fields of SingleNeurones in the Cat's Striate Cortex," J. Physiol. (London)148, 574-591 (1959).

-- , "Receptive Fields of Optic Nerve Fibres in the SpiderMonkey," J. Physiol. (London) 154, 572-580 (1960).

,"Integrative Action in the Cat's Lateral Geniculate Body,"J. Physiol. (London) 155, 385-398 (1961).

, "Receptive Fields, Binocular Interaction and FunctionalArchitecture in the Cat's Visual Cortex," J. Physiol. 160, 106-154 (1962).

R. Jung, "Microphysiologie corticaler Neurone: Ein Beitrag zurKoordination der Hirnrinde und des visuellen Systems," inStructure and Function of the Cerebral Cortex, edited by D. B.Tower and J. P. Schad6 (Elsevier Publishing Company,Amsterdam, 1960).

R. Jung and H. Kornhuber, Editors, Neurophysiologie undPsychophysik des Visuellen Systems (Springer-Verlag, Berlin,1961).

S. W. Kuffler, "Discharge Patterns and Functional Organizationof Mammalian Retina," J. Neurophysiol. 16, 37-68 (1953).

K. S. Lashley and G. Clark, "The Cytoarchitecture of the CerebralCortex of Ateles: a Critical Examination of ArchitectonicStudies," J. comp. Neurol. 85, 223-305 (1946).

H. R. Maturana, J. Y. Lettvin, W. S. McCulloch, and W. H. Pitts,"Anatomy and Physiology of Vision in the Frog (Ranapipiens)," J. Gen. Physiol. 43, Pt. 2, 129-176 (1960).

V. B. Mountcastle, "Modality and Topographic Properties ofSingle Neurons of Cat's Somatic Sensory Cortex," J. Neuro-physiol. 20, 408-434 (1957).

J. H. Nauta and V. M. Bucher, "Efferent Connections of theStriate Cortex in the Albino Rat," J. comp. Neurol. 100, 257-286 (1954).

S. Polyak, The Vertebrate Visual System, edited by H. Kluver (TheUniversity of Chicago Press, Chicago, 1957).

T. P. S. Powell and V. B. Mountcastle, "Some Aspects of theFunctional Organization of the Cortex of the Postcentral Gyrusof the Monkey: a Correlation of Findings Obtained in a SingleUnit Analysis with Cytoarchitecture," Johns Hopkins HospitalBull. 105, 133-162 (1959).

L. A. Riggs, F. Ratliff, J. C. Cornsweet, and T. N. Cornsweet,"The Disappearance of Steadily Fixated Visual Test Objects,"J. Opt. Soc. Am. 43, 495-501 (1953).

W. A. Rosenblith, Editor, Sensory Communication (MIT Pressand John Wiley & Sons, Inc., New York, 1961).

P. S. Silva, "Some Anatomical and Physiological Aspects of theLateral Geniculate Body," J. comp. Neurol. 106, 463-486(1956).

DAVID H. HUBEL

K. U. Smith, "Visual Discriminations in the Cat: V. The Post-operative Effects of Removal of the Striate Cortex uponIntensity Discrimination," J. genet. Psychol. 51, 329-369(1937).

S. A. Talbot, "A Lateral Localization in the Cat's Visual Cortex,"Federation Proc. 1, 84 (1942).

S. A. Talbot and W. H. Marshall, "Physiological Studies onNeural Mechanisms of Visual Localization and Discrimination,"Am. J. Ophthalmol. 24, 1255-1263 (1941).

H.-L. Teuber, W. S. Battersby, and M. B. Bender, Visual FieldDefects after Penetrating Missile Wounds of the Brain (HarvardUniversity Press, Cambridge, Massachustts, 1960).

JOURNAL OF THE OPTICAL SOCIETY OF AMERICA

J. M. Thompson, C. N. Woolsey, and S. A. Talbot, "Visual Areas Iand II of Cerebral Cortex of Rabbit," J. Neurophysiol. 13,277-288 (1950).

L. Widen and C. Ajmone-Marsan, "Effects of Corticipetal andCorticifugal Impulses upon Single Elements of the DorsolateralGeniculate Nucleus," Exptl. Neurol. 2, 468-502 (1960).

T. N. Wiesel, "Receptive Fields of Ganglion Cells in the Cat'sRetina," J. Physiol. (London) 153, 583-594 (1960).

M. L. Wolbarsht, H. G. Wagner, and E. F. MacNichol, Jr.,"Receptive Fields of Retinal Ganglion Cells: Extent andSpectral Sensitivity," in Neurophysiologie und Psycho physik desvisuellen System~zs, edited by R. Jung and H. Kornhuber(Springer-Verlag, Berlin, 1961).

VOLUME 53, NUMBER JANUARY 1963

Functional Basis for "On"-Center and "Off"-Center Receptive Fields in the Retina*tHENRY G. WAGNER

Naval Medical Research Institute, Bethesda 14, Maryland

EDWARD F. MACNICHOL, JR.The Johns Hopkins University, Baltimore 18, Maryland

AND

MYRON L. WOLBARSHTNaval Medical Research Institute, Bethesda 14, Maryland

The ganglion cells in the goldfish retina may have either "on"-center or "off"-center receptive fields.Evidence is presented to show that for any cell: (1) Under suitable conditions either pure "on" or pure "off"responses can be evoked by small spot stimuli at any point within the receptive field. (2) The sensitivities ofboth "on" and "off" processes are maximal in the center of the field. (3) The relative sensitivity of theseprocesses is not constant but changes as a function of position in the field. (4) The response evoked by astimulus of any size and at any location within the receptive field represents the sum of the contributionsfrom both the "on" and "off" processes.

INTRODUCTION

EVEN before the first electrophysiological recordingswere made from the vertebrate retina, histological

studies' had revealed that many photoreceptive ele-ments were connected to the same ganglion cell. Thusit should not have been unexpected when Hartlineobserved2 that illumination at any point over a rela-tively large area of the retina which he termed thereceptive field would evoke a response in the axon of asingle ganglion cell. His work' on the frog revealed thatthe intensity of the light necessary to evoke a threshold

* Invited paper presented at the Symposium on PhysiologicalOptics, Joint Session of the Armed Forces-NRC Committee onVision, the Inter-Society Color Council, and the Optical Societyof America, 14-15 March 1962, Washington, D. C.

t The opinions or assertions expressed herein are the privateones of the author and are not to be construed as official orreflecting the views of the Navy Department or the Naval serviceat large. The research was supported in part by National ScienceFoundation Grant G-7086.

1 S. R. Cajal, "La Retine des vertbr6s," Cellule 9, 119 (1892).2 H. K. Hartline, "The response of Single Optic Nerve Fibers

of the Vertebrate Eye to Illumination of the Retina," Am. J.Physiol. 121, 400 (1938).

3H. K. Hartline, "The Receptive Field of the Optic NerveFibers," Am. J. Physiol. 130, 690 (1940).

response varied with the position of the exploring spot,generally being lowest near the center of the field andprogressively higher towards the periphery. The exactedge of the field was difficult to define since the signifi-cance of scattered light in causing a response could notbe evaluated directly. The diameter was estimated tobe of the order of 1 mm.

Some years later Kuffler4 reported the observationthat there were to be found in the cat retina, ganglioncells which changed their response character from "on"to "off" depending upon whether the stimulus was inthe center or in the periphery of the receptive field.These were called "on"-center and "off"-center types.The mutual antagonism between the central andperipheral portions of the field was clearly evident inthe studies of Barlow, Fitzhugh, and Kuffler.5 Thesefields have since been found in other vertebrate retinasand appear to be of fundamental significance in theorganization of the retina. Further study of these

4 S. W. Kuffler, "Discharge Patterns and Functional Organiza-tion of Mammalian Retina," J. Neurophysiol. 16, 37 (1953).

5 H. B. Barlow, R. Fitzhugh, and S. W. Kuffler, "Change ofOrganization in the Receptive Fields of the Cat's Retina duringDark Adaptation," J. Physiol. (London) 137, 338 (1957),

66 Vol. 53