Influence of spatial frequency on tuning and bias for orientation and direction in the cat's striate cortex

Influence of spatial frequency on tuning and bias for orientation and direction in the cat's striate cortex

l’~kw Res. Vol. 30. No. 3. Printed in Great Britain. 0042-6989/90S3.00+0.00 pp.359-369.1990 Copyrighte 1990Pqamon Presspk All rightsreserved INF...

1MB Sizes 0 Downloads 67 Views

l’~kw Res. Vol. 30. No. 3.

Printed in Great Britain.

0042-6989/90S3.00+0.00

pp.359-369.1990

Copyrighte 1990Pqamon Presspk

All rightsreserved

INFLUENCE OF SPATIAL FREQUENCY ON TUNING AND BIAS FOR ORIENTATION AND DIRECTION IN THE CAT’S STRIATE CORTEX P. HAMMONDand C. J. D. POMFRETT Department of Communication and Neuroscience, University of Keele. Keele, Staffordshire ST5 SBG. England (Receiued 2 January 1989; revised I8 April 1989)

Abstract-Directionality, orientation and spatial frequency tuning were determined for 108 neurones recorded extracellularly from the striate cortex of anaesthetixed cats. Significant sharpening of orientation selectivity with increasing spatial frequency was seen in all simple neurones and the overwhelming majority of complex neurones. Orientation selectivity sharpened in 90 and broadened in only 10 of 100 fully characterized neurones. At least four distinct classes of neurone could be characterixed on the basis of their directionality at optimal spatial frequency, and the presence or absence of changes in directionality over a range of spatial frequencies: in two classes, directionality was spatial-frequency dependent; in the remaining two it was invariant. With two exceptions Type A neurones (23 4s) were direction-selective; they were narrowly tuned for orientation and spatial frequency, and their directionality was invariant with spatial-frequency. The majority of neurones (52 cells) were Type B, most of which were direction-biased; their bias for direction varied systematically with spatial frequency. Type C were direction-biased and spatial-frequency selective (5 cells), but showed a clear reuersol of bias with change in spatial fmquency. Type D, a subset of direction-biased cells, were bidirectional and spatial-frequency invariant (8 cells), with comparable response strengths to motion in two opposing directions at all spatial frequencies. These response types crossed traditional boundaries between categories of simple and complex neurones, assigned on the basis of spatial summation, presence or absence of end-inhibition, and receptive field size. Feline

Visual cortex

Area 17

Spatial frequency

INTRODUCTION

Recordings of the neural activity in the striate cortex of the cat have concentrated on orientation and direction specificity, primarily to moving bar or grating stimuli which elicit optimal responses from single neurones (Hubel & Wiesel, 1962). Campbell, Cooper and EnrothCugell(1969) have shown that cortical neurones are additionally bandbass-tuned for spatial and temporal frquency, and independent studies have revealed that most cells demonstrate bandpass characteristics for velocity (Orban, Kennedy & Maes, 1981a,b). Whilst it has generally been assumed that the orientation and directional bias of these neurones are invariant properties (Hammond, Andrews & James, 1975), one study has hinted that the orientation bandwidth of certain neurones in area 17 of the cat narrows with increasing spatial frequency (Vidyasagar & Sigiienza, 1985). Another has shown that the directional bias in many neurones alters with stimulus length (Hammond & Mouat, 1986).

Tuning

We here report the changing patterns of response in striate cortical neurones to a broad range of spatial frequencies with respect to alterations in orientation selectivity and directionality. A preliminary abstract has already been published elsewhere (Hammond & Pomfrett, 1988). MATERIALS AND METHODS

Preparation The physiological preparation and recording techniques were identical to those described in detail in previous papers (Hammond & MacKay, 1983, 1985) and are described only briefly here. Adult cats were anaesthetized surgically with 2-5% halothane in a 72.5% : 27.5% N20:02 mixture. With aseptic precautions they were chronically implanted with a closed recording chamber, incorporating terminals for EEG recording and a stainless steel peg for atraumatic stereotaxic head restraint. End-tidal CO2 levels, ECG, heart rate, EEG, and body temperature were monitored both during this

359

360

P. HAMMOND and C. J. D. POUFRETT

procedure and during short sessions on subsequent occasions in which recordings were made from area 17 under light anaesthesia (N,0/02 supplemented with halothane concentrations of between 0.2 and 0.7%). The eyes were protected with unpowered contact lenses, applied with wetting solution on the inner surface only. Pupils were dilated with 1% atropine sulphate (Minims: Smith & Nephew); eyelids and nictitating membranes were retracted with 10% phenylephrine hydrochloride (Minims: Smith & Nephew). 5 mm artificial pupils were placed as close as possible to the eyes and centred over the dilated natural pupil. Focus was corrected by trial lenses for a viewing distance of 57 cm (1 cm = 1” visual angle). Neurones were recorded extracellularly with low-impedance micropipettes, external tip diameter 1.5-2.0 pm, filled with 4 M NaCl. Electrodes were advanced stereotaxically in 2 pm steps through minute punctures in the otherwise intact dura. The craniotomy was sealed with warm, 2% agar in 0.9% saline to impose stability. Recording tracks were vertical, 3.5-6.5 mm behind the interaural plane and within 2.0mm of the midline in either hemisphere, to a maximum depth of 5.0mm. Receptive fields of recorded neurones lay in the contralateral hemifield, close to the midline and O-15” below the horizontal plane passing through the projections of the areae centrales. Stimulation and analysis Visual stimuli were generated on a Hewlett-Packard 1304A high-resolution electrostatic display (with white P4 phosphor) under computer control. The display was positioned 57cm in front of the animal and was usually centred on zero azimuth. Receptive fields and positions of back-projected areae centrales were mapped on acetate transparencies attached to an identical display. Average luminance was 1.1 log cd/m’; contrast was 0.4. Square- or sine-wave gratings were presented at different orientations, spatial frequencies and sweep velocities with real-time monitoring and analysis of the responses of extracellularlyrecorded cortical neurones. All measurements were made during stimulation of the dominant eye, with the other eye occluded. The usual sequence was to centre the grating on the receptive field, then to determine the optimal orientation, spatial frequency, and velocity of movement. Thereafter velocity was maintained

constant. Orientation tuning and directionality were determined quantitatively for an optimal spatial frequency grating, sweeping twice backand-forth (1 set in each direction, with a 1 set pause between each back-and-forth sweep), before stepping direction 10” through 360”. Responses of neurones were archived to 1 msec accuracy on computer disc. Typically, eight round-the-clock sequences of directional tuning were averaged at each spatial frequency tested (a practice which yields reliable tuning functions with little hysteresis or sequence dependence; Hammond, 1978). In a number of instances (to be reported elsewhere) we subsequently also determined the spatial frequency tuning for the non-dominant eye. Tuning and selectivity for a range of at least two, but usually five or more, different spatial frequencies were established in random sequence, ranging from a low spatial frequency to a high value which was just below the neurone’s upper spatial frequency cut-off point. For each tuning curve the optimal orientations and strengths of response for preferred and opposite directions of motion were derived from the intersection of pairs of regression lines fitted to each flank of each peak in the tuning curve. Total tuning width was derived from the intercepts between regression fits and resting discharge level. Tuning symmetry was assessed from the ratio of the half-widths of tuning. The direction-selectivity index was assessed by the relationship: KP - S) - (N -

wp

- 9

where P and N are the response strengths in preferred and opposite directions of motion and S is the resting discharge level, measured in impulses/set (Hammond & Mouat, 1986)--see also Fig. 4. Neurones were assigned to one of two directionality groups, direction-biased or directionselective, on the basis of their response patterns to stimuli of optimal orientation, spatial frequency and velocity of motion, according to whether the ratio of responses to preferred: opposite directions was less than or exceeded 2O:l. Receptive fields were then mapped at the resultant optimal orientation and velocity for the best spatial frqucncy, using a modified version of the minimum response field method: a grating patch, within which the grating drifted continuously in the neurone’s preferred direction, presented against a uniform

Spatial frequency tuning in feline striate cortex

Change

of Tuning

Width

with

0

Cell Type

Increasing

361

Spatial

Broadening

Frequency:

-Narrowing

Simple, Diiection Biased _ Simple.

Direction Selective

Standard Complex, Direction Biased _ Standard Complex, Direction Selective Intermediate Complex, Direction Biased _ Intermediate Complex. Direction Selective _ Special Complex, Direction Biased _ Special Complex. Direction Selective 40

30

20

Number

lo

0

lb

of Cells

Fig. 1. Distribution of neurones demonstrating narrowing or broadening of orientation tuning width with increasing spatial frequency.

grcy background, was moved into the receptive with orientation on the X-axis, average spike field from either end and from either side in turn frequency on the Y-axis, and spatial frequency until a response was just elicited. Permanent on the Z-axis (see Fig. 3). This presentation records of the rectangular maps obtained were permitted rapid identification of the response retained for later reference. pattern to changing spatial frequency at both Length summating properties of the cell were the preferred and non-preferred stimulus orienthen determined with a dark bar of optimal tations, which were difficult to identify in other orientation, width and velocity, moving alter- types of representation. nately in preferred and opposite directions (1 set in each direction). Twenty responses to RESULTS each bar length were averaged. Length was varied between 0.25” or 0.5“ and 10” in a preDirectional and spatial frequency tuning data determined pseudo-random sequence. Neuronal were obtained for 108 neurones in the striate type (simple or complex) was assessed using cortex of 13 adult cats. All of the neurones standard criteria established by Hubel & Wiesel responded to velocities within the range (1962), elaborated for discreteness of light- and 0.5-8.O”/sec, which places the entire sample of dark-edge discharge centres by Bishop, Coombs cells within the velocity broad-band or velocity and Henry (1971). The presence or absence of low-pass classes described by Orban et al. endstopping was determined from the length (1981a). summation profile. Complex neurones were further subdivided: standard complex neurones summated to lengths matching or exceeding the Reduction in tuning width at high spatial frequen mapped height of their minimum response field; ties Tuning width was measured over a range of special complex neurones showed little summation and responded to contours much shorter spatial frequencies in 100 of the total sample of than their minimum response field; intermediate 108 neurones. All subsequent comprisons are complex neurones exhibited substantial length confined to these 100 neurones. Figure 1 shows summation, but with a limit value smaller than the results for simple and complex cells, group the minimum response field height (Hammond by group, divided into direction-selective and direction-biased categories. Orientation selec& Ahmed, 1985). Off-line analysis included the detailed exami- tivity sharpened at high spatial frequencies in 90 nation of spatial frequency tuning charac- neurones, including all the simple cells; only teristics using microcomputers. The spatial a small minority of 10 complex neurones (5 frequency and orientation tuning profiles of standard and 5 special complex cells) became neurones were reconstructed as 3-D displays less selective.

362

P. HAMMOND and C. J. D.

Different neurones possessed different spatial frequency bandpass characteristics; thus we applied the most appropriate range of spatial frequencies in each case, which varied from cell to cell. In order to compare neurones within each class, for each neurone we therefore normalized the data for total orientation tuning width at each spatial frequency tested. Next we pooled all neurones of a given type for each of the eight ciasses illustrated in Fig, 2, readjusting the normalized averages at each spatial frequency so that the highest value was set at 100% in order to compare the different classes. The error bars in Fig. 2 indicate the standard errors of the adjusted averages. In every class, overall, there was a trend towards greater selectivity for orientation at high spatia1 frequencies, despite the inclusion of the minority of neurones with poorer selectivity at high spatial frequencies, and even though every class included neurones of similar type but dissimilar spatial frequency bandpass cha~cteristics. All 14 simple cells sampled responded to spatial frequencies within the range 0.25-2.0~~. The 3 direction-selective cells were all sharply tuned, with a modest reduction in tuning width from around 50-70” to 40-60” at high spatial frequencies, With the exception of one unusually broadly tuned member, orientation tuning widths in the 11 direction-biased simple cells covered a broad range between 40-140” at optimal spatial frequency, declining to between IO--to0 at high spatial frequency. Standard complex cells were encountered most frequently (64 cells, of which 41 were direction-biased and 23 direction-selective at optimal spatial frequency). They responded over a broad range of spatial frequencies up to at least 4c/“. All but 3 direction-biased and 2 direction-selective examples demonstrated a reduction in tuning width with increasing spatial frequency. The orientation tuning widths spanned a broad range from 40” to over 180” at low spatial frequencies, reducing to 25-120” at high spatial frequency. No significant differences were seen between neurones exhibiting end-inhibition and those with none. Only 6 complex neurones were classified as intermediate (after Hammond and Ahmed, 1985; and see Methodology), All 6 (4 directionbiased, 2 direction-selective at optimal spatial frequency) showed a very marked reduction in tuning width with increasing spatial frequency,

POMFRETT

from 1IO to 210” at low spatial frequencies to around 35-80” at high spatial frequency. The sample of special complex neurones was small (I 6 cells) and the incidence of endstopping high (7 cells). Orientation tuning width decreased with increasing spatial frequency in all 6 direction-biased examples. Comparable reductions in tuning width occurred in 5 of the direction-selective members, but increased in the remaining 5. In those whose tuning tuning width ranged between sharpened, 60480” at low spatial frequencies, declining to 50-150” at high.

DIRECTIONAL TUNING CHARACI’ERISTICS VS SPATIAL FREQUENCY

Complete directional tuning profiles over a range of spatial frequencies were obtained in 88 of the total sample of neurones {see Table I ). Our designation of each ~urone’s dir~tionaIity ~dir~tion-~l~tive~di~tion-bias) follows the standard convention of assessing its directional tuning and properties under optimai conditions, for stimuli of bat orientation and spatial frequency, direction and velocity of motion. Dfreclion-selective neurones responded preferentially to one unique direction, with little or no response in the opposite direction; directionbiased neurones responded differentially or with comparable strength to opposing directions of motion ~H~mmond & Pomfrett, 1989, 1990). We designated the dir~tionality of each cell numerically according to its diction-~tivity index (see Methodology). On this criterion total direction-selectivity achieves a vaiue of unity (or exceeds unity for cells with null suppression); bidirectionality a value of zero. We arbitrarily set the divide between directionselective and direction-biased cells at 0.95 (i.e. a response ratio for opposing directions of motion >20: 1). At least four types of variation in directional tuning characte~sti~ (Types A to D below) were observed. The dist~bution of different categories of simple and complex neurones amongst the four defined Types is shown in Tabie 1. As is clearly evident from the examples illustrated in the 3-D plots of Fig. 3, in. most neurones systematic changes occurred in the tuning width and directional bias, and in the appearance of the tuning peaks and troughs. Alterations in spatial frequency affected both the excitation and suppression of driven activity

t 1~ 1

Spatial frequency iuning in f&e

100



50

25

*

1

1 Simple cells

1 Simple cells direction selective (~31

k

0, 100

1

$1

75

s

1W

4

363

striate cortex

s

, 12

I

1

1

I

direction biased (n.11)



t2

Standard complex cells direction biased (n&41 1

75

t 50 1;1\

1

I Intermediate ComrHex Cells direction biased h4)

Special complex cells dirrotion blared (nr6) 0.2

I

I

I

,

I

0.4

0.6

1.0

2.0

4.0

Ofatlng ePatial freQuency Wdeg

)

fig. 2. Changes in orientation tuning width of direction-scktiw and di~ion-~ simpk and compkx ncuroncswith increasingspatial frequency.Beforepooling the data for each clam of ncurones,individual data for each ncuront were normalized. For comparison bctwctn classes. averagesof normaiii pookd data werethen readjustedso that the highest value was set at 100%.Errorbarsindiitc the standard mrs of the adjustedaverages.Numbers of cells in ash pled set of data are indicated in pare&mm. Values alongside each point indiitc number of neuronm contributing to that datum.

relative to spontaneous resting activity, as for example in the Type A complex neurone shown in Fig. 3 (centre row, Ieft), where the response

peak varied in width and was composed of several discrete sub-peaks whose characteristics varied with spatial frequency.

P. HAMMOND and C. J. D. POMFRETT

364

Table I. Distribution of spatial frequency response types Type

A

B

C

D

Totals

Simple Direction-selective Direction-biased

1 1

I 6

0 0

0

I

2 8

Standard complex Direction-selective Direction-biased

14 0

7 26

0 5

0 7

21 38

Intermediate compIex Direction-selective Dir~tion-bias

I 0

1 4

0 0

0 0

2 4

Special complex Direction-selective Direction-biased

5 1

2 5

0 0

0 0

7 6

23

52

5

8

88

Totals

Type A: directionality frequency

invariant with spatial

Twenty-three neurones fitted this category. The two simple-cells in this group were bandpass spatial-frequency tuned about a single best spatial frequency. In the dir~~on-~l~tive example illustrated in Fig. 3 (upper row) the orientation tuning and directional bias characteristics did not change with variation of spatial frequency. Twenty-one complex neurones (14 standard, 1 intermediate and 6 special) also responded to changing spatial frequencies of stimuli with a single peak response and no change of directionality. However, whereas the simple cells were tightly bandpass tuned for spatial frequency, the complex neurons in this category responded over a broad spatial frequency range. A set of responses for one such neurone (a directionselective special complex cell with null suppression) is shown in Fig. 3 (amtre row, left). Type B: directionality frequency

dependent

on spatial

This was by far the commonest category. Of 52 neurones, a strong majority (41 neurones) were direction-biased, only 11 were directionselective, at optimal spatial frequency. The characteristic feature of these neurones was that their directional bias changed systematically with spatial frequency. A typical example, a direction-biased standard complex neurone, is shown in Fig. 3 (centre row, right). All Type B neurones exhibited a single preferred direction of motion at all spatial frequencies, whereas the response peak for the opposite direction waxed and waned with variation in spatial frequency, Unlike the obviously

direction-biased example in Fig. 3, all the neurones direction-selective at the optimal spatial frequency and orientation were clearly direction-biased for coarser or finer gratings. Simple and complex cells alike were represented although, as with Type A, the range of spatial frequencies over which simple cells responded was characteristically narrower than that seen for complex neurones. Type C: reversal of directionaiity with spatial frequency

Ail 5 members of this highly interesting minority group were direction-biased standard complex neurones, when assessed at optimal spatial frequency. However, the directional bias reversed with change in spatial frequency (Fig. 3, lower row, left; and see also Fig. 4). One direction of motion elicited a relatively stronger response at higher spatial frequencies, whilst the opposing direction induced greater drive at lower spatial fr~uencies. This was in marked contrast to the responses seen for Type B neurones where, despite variations in directional bias, the preferred direction of motion remained constant at every spatial frequency. Opposing directions of motion in Type C neurones exhibited different spatial frequency tuning characteristics (e.g. Fig. 4). In Type A and D neurones, by contrast, spatial frequency characteristics were either identical or closely similar for either direction of motion. In the illustrated example (Fig. 3) it is interesting to note that the peak at around 220”, preferred at spatial frequencies above approx. lSc/“, was still present at low spatial frequencies but was broader and appeared to be composed of several small subpeaks (e.g. at OSc/“). Type D: bidirectional, directionality with spatial frequency

invariant

Bidirectional neurones represented an extreme case of direction-biased neurones, responding more-or-less equally to two opposing directions, nominally 180”apart. Arbitrarily we set the divide between dir~~on-~a~ and bidirectional at a direction-selectivity index of 0.05 (i.e. responses equal in strength to within 5%), with reversal of bias expressed as negative values. Eight neurones (1 simple, 7 standard complex) fitted this criterion at optimal spatial frequency. in every case (e.g. Fig. 3, lower row, right) they remained bidirectional throughout the spatial frequency range. Peak firing

Spatial frequency tuning in feline striate cortex

Fig. 3.3-D comparisons of orientation/direction selectivity vs spatial frequency tuning in different types of simple and complex naurones. The upper row illustrates a Type A simple cell (VC-2I8- 17: direction-sekctivc, dimctionality invariant with spatial frequency). Centre and lower rows illustrate four types of behaviour amongst compkx neuronesz Type A spatial-frequency invariant response in a ditection-selective special compkx neurone (W-234-2) with null suppression for directions of motion opposite those prefe&, Type 3 di~ion-~~ standard compkx neurone (W-236-24). bias changing with spatiaf frequency; Type C directionbiased standard compkx neurone (K-2344 bias reversing with spatial frequency; Type D bidirectional, endstopped standard complex neurone (VC-228-I). directionality invariant with spatial frequency. Direction of stimulus motion (degrees, measured clockwise from upward motion, o’, in lo” incmments). neuronal responss (imp/se4 and spatial frequency (c/” ) are plotted on X. Yand Zaxes respa%ively, averaged over eight presentations per direction of motion, for each spatial frequency. In each case the perspective view and ordering of spatial frequencks (low-high or high-low) have been chosen to give the best visualiition of changes in directional bias, preferred di~tion/o~n~tion and shape of each tuning peak.

frequency varied with spatial frequency, but in tandem for both directions of stimulus movement, in marked contrast to Type B and C neurons,

however, we compared response strengths for preferredand opposite directions of motion of grating patterns at each spatial frequency, cvaluatcd from the interceptbetween regressionfits to the two flanks of each peak in the tuning QuunrtJTcation of response types curves, with due allowance for levels of resting By their nature, the data do not lend themselves discharge. The upper part of Fig. 4 compares to rigorous statisticalanalyses. In each neurone, these manipulations on representativeexamples

P. HAMMOND and C.J.D.

366

POMFRETT

3 60

TYDe

c

160-

Type D

20 10 1

ooao

O‘----x-x--;b----2;,----4!o 0.2

Qrrtlng *Datirl fregwncy

k/dog

1

Type

:.

-l{ 0.2

I

0.4 &atlng

I

I

0.6 1.0 maIlal froqwncy

I

t

2.0 Wdeg

C

4.0

I

Fig. 4. Variation of directionality with spatial frequency. In the upper and centre rows response strengths in preferred and opposite dimetions of motion {solid 8nd opea circks mspectively) are comp8red at each spatial frequency tested, for the Type A simpk cell, and the Types B-D cotnpkx c&s illustrated in Fig. 3. Firins kveis are derived from the intercept of rqression fks to the two flanks of each peak in the tuning curves for each spatial frequency, based on the average of 8 responses per dir&on of motion. The horizontal broken line indicates the resting discharge kvel (the type A cell was siknt). In the lower row the direction sekctivity index (see Methodology) for Types A-D is plotted a&at spatial frequancy: a value of unity indicates total direction-sekctivity; xero, biiirectionality; negative values, reversal of bias.

from each of the four directionality types of neurones (the Type A simple cell, and the Types B-D complex ceils, already illustrated in Fig. 3). In each case the optimal spatial frequency, on the basis of which a neurone was designated as either direction-selective or direction-biased, is shown as an unfilled symbol. These data are

re-expressed as direction-selectivity indices in the lower graph, from which it is self-evident that Types A and D are clearcut e&k with spatial-frequency invariant directionality. Type C, also, is clearly de&d, with a frank reversal of bias at around l&r, in favour of the three highest spatial frequencies tested. The definitive

Spatial frcquetcy tuning in feline striate cortex

characteristic of Type B, as shown from the response vs spatial frequency plot (Fig. 4, upper row, right), is that the optimal spatial frequency and bandpass characteristics are d@erent for opposing directions of motion, whereas in Type A and Type D they are comparable. We cannot, however, rule out the possibility that Type B embraces a continuum of cells with differing shades of directional bias (assessed at optimal spatial frequency), from direction-selective cells at one extreme to bid~tional cells at the other. Sine-wave us square-wave grating stimuli We used square-wave grating stimuli extensively in this study, in order to retain coherence with previous observations. However, 5 neurones (1 simple and 3 standard complex, direction-biased neurones; 1 direction-selective special complex cell) were also tested with sinewave gratings over a range of spatial frquenties. As anticipated, we observed littIe difference except at extreme low spatial frequencies, a difference p~u~bly due to the hip-f~quen~ harmonic content of low spatial-frequency square-wave gratings. Both grating configurations induced comparable sharpening of orientation tuning. DISCUSSION

One of the major findings presented in this paper, that orientation tuning width decreases with increasing spatial frequency, agrees with one previously publish~ report for simple cells (Vidyasagar & SigGenza, 1985). However, Vidyasagar and SigCenza failed to recognize that a similar decrease in tuning width occurs for complex cells. This is surprising, since we have seen a decrease in tuning width for almost all cells tested, whether they wcrc simple or complex. The discrepancy is probably due to Vidyasagar and Sigiienza’s reconstruction of most of their directional tuning plots from successive spatial frequency tunings over a restricted range of directions of motion progressively to either side of the preferred direction. Our method differed sign&a&y in that we performed successive complete directional tunings over sequences of spatial frequencies presented in pseudo-random order. Orban et al. (1981a,b) suggested that velocity low-pass neurones in are 17 of the cat might be responsible for detecting stationary objects during visual fixation. They also commented that the low-pass neurones encoded such a small

367

dynamic range of velocities that they were probably not responsible for coding velocity as such. This observation is of particular significance because all of the neurones which we have described here fell within the velocity low-pass or broad-band classes described by Orban and co-workers, and so were responsive to a similarly low dynamic range of velocities. It may be inappropriate for such neurones to encode small changes in velocity if they are responsible for maint~ning some aspect of visual fixation, However, it would be highly appropriate for such neurones to respond to specific spatial frequencies while encoding fixation cues from the visual environment, whereas neurones sensitive to higher velocities may be less suitably equipped to encode spatial frequency. Profiles of response to changing spatial frequency show that there are at least four types of neurones with distinct levels of complexity. These types are not just subdivisions of established cortical cell cIasses based on receptive iieId, resting discharge and length summating properties, but encompass several classical cell groupings within each type. For example, simple cells show predominantly sharp bandpass spatial-frequency tuning profiles, to orientations ranged closely about a single preferred direction or opposing pair of directions of motion. Indeed, the sharp spatial frequency tuning of simple cells, in particular, has led to speculation that they may act as spatial frquency filters (Maffei & Fiorentini, 1973). We have observed that certain (especially standard) complex neurones also show a similarly sharply orientationtuned response, but are responsive to a much wider range of spatial frequencies than are simple cells. Adhering to hierarchical principles (Hubel & Wiesel, 1962) it is possible that these neurones receive the pooled input from several simple cells or groups of simple cells with the same orientation preference but a range of overlapping spatial frequency tunings. In particular, a bidirectional cell could combine two such subjects of neuronal inputs (perhaps from two pools of direction-selective neurones with opposite preferred directions of motion) to give two peaks of similar or identical spatial frequency response but with preference for opposite directions of motion. Whilst the concept of neuronal properties consequent upon pooling of lower order inputs encompasses some elements of. classical serial processing theories for the operation of Area 17 (Hubel & Wiesel, 1%2), the single, relatively

368

P. HAMMOND and C. J. D. POMFIWT

broad tuning peak for orientation shown by some neurones during cursory examination is frequently composed of several spatialfrequency dependent smaller peaks. Stimulation at high spatial frequency results in one or more peaks dropping out from the most broadly peaked overall response, with consequent narrowing of orientation tuning. Such cells may be receiving pooled, parallel and serial input from neurones with narrow spatial frequency band-pass characteristics and also from neurones with broad-band spatial frequency characteristics. The input to these cells (whether simple or complex) may, therefore, be from lower-order simple as well as complex neurones. Undoubtedly the most interesting patterns of variation in spatial frequency sensitivity were the variations in directional bias shown by a majority of neurones (Type B), and the radical reversal of bias which occurred in a minority of neurones (Type C). In both groups the strength of the response peak to one direction of motion declined over part of the spatial frequency range as that to the other increased, in the extreme case (Type C) changing the characteristics of the cell from direction-biased for one preferred direction of motion, through bidirectional, to direction-biased for the opposite direction. Such marked changes in neuronal characteristics with changing spatial frequency have profound consequences, since most of the earlier descriptions of cell classes in the literature are based solely on responses to the optimal spatial frequency. In this context, the present report shows that a number of cells direction-biased at the optimal spatial frequency become increasingly directionselective or bidirectional at supra- and suboptimal spatial frequencies. Our data cannot distinguish unequivocally between two possibilities: the existence of discrete types of directionality dependence on spatial frequency, or a continuum from one extreme to another. However, we feel that Types A and D, which together constitute 35% of the sample, are discrete. Both types are conventional in being directionality invariant; one class is direction-selective, the second bidirectional, and there is good evidence on grounds of differences in orientation tuning width (Hammond & Pomfrett, 1989, 1990) for regarding these as separate entities. Likewise, the minority group of cells whose directionality reuerses with spatial frequency (Type C) constitutes a clearcut group, The greatest uncertainty is whether the majority group (Type B), whose directionality varies with

spatial frequency, constitutes one (or more) discrete groups or represents a continuum, from direction-selective cells at one extreme to bidirectional cells, or cells with only rather weak directional bias, at the other. The changes in directional tuning seen with variation in spatial frequency are similar to changes in neuronal directionality induced by the micro-injection of GABA in the vicinity of neurones (Eysel, Muche & Worgiitter, 1988). In particular, Eysel’s group showed that GABA injection changed the fundamental directional preference, but not the preferred orientation of motion. If the functional significance of GABA is indeed related to alteration of the connections responsible for spatial frequency tuning, the work of Eysel and co-workers suggests that such interaction is predominantly lateral rather than columnar in the cortex. This would support the presence of lateral rather than columnar organization for spatial frequency within cortical laminae, as suggested by Berardi, Bisti. Cattaneo, Fiorentini and Maffei (1982), although this arrangement has been challenged by Tolhurst and Thompson (1981). Our findings suggest that changes in spatial frequency, as well as affecting orientation selectiuity, may shift the preferred orientation (slightly) and also the directional bias (significantly) in a majority of neurones. It is additionally possible that previous studies may have missed the optimal orientation and/or spatial frequency of a neurone by varying either orientation or spatial frequency separately, whilst keeping the other parameter fixed. This would critically affect the assessment of optimal spatial frequency and/or orientation during sampling from large groups of neurones. Tolhurst and Thompson (198 I) suggested, without providing any convincing demonstration, that end-stopped neurones preferred higher spatial frequencies than cells which did not exhibit end-inhibition. We were unable to demonstrate any such difference between endstopped and non-endstopped standard complex neurones which were therefore pooled in the current analysis. This in no way implies that there might not be a difference for the other cell types, for which our sample size was too small. We found little difference in the response of neurones when square-wave or sine-wave grating stimuli were employed. In the 5 cells which were examined with both classes of stimuli, they responded equally well to square- or sine-wave gratings. This is in marked contrast to the striate cortical neurones of the primate,

Spatial frequency tuning in feline striate cortex

which have been shown to respond more selectively to spatial frequency wge with .sinewave gratings, whereas square-wave gratings elicited little spatial frequency sensitivity (Schiller, Finlay & Volman, 1976). Schiller and co-workers proposed that edge-effects from square-wave gratings may mask spatial frequency selectivity. It is highly likely that the cat’s visual system possesses broadly comparable mechanisms for detecting edges, albeit over a much lower spatial frequency range. The most plausible explanation for the marginal difference in sensitivity to sine- and square-wave gratings rests with the poorer optical resolution of the cat’s eye, which is designed more for high sensitivity than for acuity. Thus high spatial frequency square-wave gratings register as sinewave grating images on the retina and are, therefore, interpreted in the same way as sinewave gratings by the neural mechanisms from retina to striate cortex. This suggestion inevitably needs further work with more detailed comparisons between sine-wave and squarewave stimuli at the level of the striate cortex. Notwithstanding, the sharpening of orientation selectivity is consistently evoked by gratings of high spatial frequency, whether sine- or squarewave, and appears to be an almost universal feature of cortical neuronal processing. Ackno&cigetnenrs-Dr Bashir Ahmed collaborated in some experiments during a period of study leave from the University of Kuwait. The authors wish to acknowledge the expert technical assistance of David Glover. Supported by grants G8509721N and G8614611N from the Medical Research Council to P. H. We gratefully acknowledge continued and generous support from the following companies for the provision of ophthalmic and pharmaceutical preparations and sterile disposable sundries: Smith & Nephew Pharmaceuticals Ltd. Smith & Nephew Medical Ltd: Monoject Division of Sherwood Medical; Abbott Laboratories Ltd; Roche Products Ltd; Imperial Chemical Industries Ltd; The Wellcome Foundation Ltd; CNS-Respiratory Division of Astra Pharmaceuticals Ltd. REFERENCES Berardi. N.. Bisti, S.. Cattaneo. A., Fionntini. A. & MatTei, L. (I 982). Correlation between the preferred orientation and spatial frequency of neurones in areas 17 and 18 of the cat. Journal of Physiology, London. 323. 603-618. Bishop. P. 0.. Coombs. J. S. & Henry, G. H. (1971). Responses to visual contours: Spatio-temporal aspects of excitation in the receptive fields of simple striate neurones. Journal qf Physiology, London, 219, 625-657.

Campbell. F. W.. Cooper. G. F. & Enroth-Cugell. C. (I 969). The spatial selectivity of the visual cells of the cat. Journal qf Physiology, London, 203, 223-235.

369

Eysel. U.T., Muche. T. & Wi5r#tter. F. (1988). Lateral interactions at direction-selective striate neurones in the cat’&m&u!tiat&d bjl @!&-ortical inactivation. Joumul qf Physiology, L.ondon 399, 657-675.

Hammond, P. (1978). Directional tuning of complex cells in area I7 of the feline visual cortex. Journal of Physiology. London, 285, 47849

I,

Hammond, P. & Ahmed, B. (1985). Length summation of complex cells in cat striate cortex: A reappraisal of the “special”/“standard” classification. Neuroscience, 15, 639-649. Hammond, P. & MacKay, D. M. (1983). Influence of luminance gradient reversal on simple ceils in feline striate cortex. Journal of Physiology, London, 337, 69-87.

Hammond, P. & MacKay. D. M. (1985). Influence of luminance gradient reversal on complex cells in feline striate cortex. Journal of Physiology, London, 359, 315-329.

Hammond, P. & Mouat. G. S. V. (1986). Influence of stimulus length on directional bias of complex cells in cat striate cortex. Neuroscience, 18, 25-30. Hammond, P. & Pomfrett, C. J. D. (1988). Visual cortical neurones in the anaesthetized cat: Influence of spatial frequency for orientation and direction. Journal of Physiology, London, 407, 66P. Hammond, P. & Pomfrett, C. J. D. (1989). Directional and orientational tuning of feline striate cortical neurones: Correlation with neuronal class. Vision Reseurch. 29, 653-662.

Hammond, P. & Pomfrett. C. J. D. (1990). Directionality of cat striate cortical neurones: contribution of suppression. Experimenru/ Brain Research (submitted). Hammond, P., Andrews. D. P. & James, C. R. (1975). Invariance of orientational and directional tuning in visual cortical cells of the adult cat. Brain Research, 96, 56-59.

Hubel, D. H. & Wiesel. T. N. (1962). Receptive fields, binocular interaction and functional architecture in the cat’s visual cortex. Journal of Physiology, London. 160, M-154.

Maffei, L. & Fiorentini. A. (1973). The visual cortex as a Vision Research, 13, spatial frequency analyser. 1255-1267.

Orban. G. A., Kennedy, H. & Maes, H. (198 la). Response to movement of neurones in areas 17 and 18 of the cat: Velocity sensitivity. Journal of Neurophysiology. 45, 1043-1058.

Orban, G. A., Kennedy, H. & Maes, H. (1981b). Response to movement of neurones in areas I7 and 18 of the cat: Direction selectivity. Journal of Neurophysiology. 45, 1059-1073.

Schiller, P. H.. Finlay, B. L. & Volman, S. F. (1976). Quantitative studies of single-cell properties in monkey striate cortex. III. Spatial frequency. Journal of Neurophysiology. 39, 1334-i 351. Tolhurst, D. J. & Thompson, 1. D. (1981). On the variety of spatial frequency sekctivities shown by neurones in area 17 of the cat. Proceedings of the Royal Society B, 213, 183-199.

Vidyasagar, T. R. & Sigiienza. J. A. (1985). Relationship between orientation tuning and spatial frequency in neurones of cat area 17. Experimental Brain Research, 57. 628-63

I.