Experimental Brain Research
Exp. Brain Res. 31, 163-177 (1978)
9
Springer-Verlag 1978
Intensity Coding in Primate Visual System R.B. Barlow, Jr., 1 D.M. Snodderly and H.A. Swadlow 2 Eye Research Institute of Retina Foundation, 20 Staniford Street, Boston, Massachusetts 02l 14, USA
Summary. The pupil reflex and the discharge of L G N cells of the awake macaque were measured under stimulus conditions that yielded evidence for wide-range intensity coding in human psychophysical experiments. Ganzfeld flashes of white light were delivered under dark-adapted conditions to the surgically immobilized eye of the monkey while the other eye was observed in the infrared. Three-sec flashes elicited a consensual pupil reflex that was graded from - 8 to 0 log Lamberts (L), indicating that the optic nerve fibers are capable of coding at least an 8 log-unit range of light intensity. In the physiological experiments, shorter flashes ( 0 . 1 - 0 . 5 s e c ) but otherwise identical conditions elicited monotonically graded responses from one type of L G N cell over the photopic range of - 5 to 0 log L. Responses from other types of L G N cells were also graded over wide ranges but had different thresholds and, in some cases, nonmonotonic intensity-response functions. Latency of the excitatory L G N responses decreased with increasing intensity according to a power function with slope o f - 0 . 0 8 . The pupil reflex and the L G N cell excitatory responses approximate power functions of light intensity with exponents of 0.22 and 0.14-0.29 respectively. The range of intensity coding found for single L G N cells is the widest yet reported for diffuse stimuli.
Key words: Intensity coding - Ganzfeld illumination - Lateral geniculate nucleus - Pupil reflex - Response latencies
The human visual system has an effective operating range of more than 10 log units of light intensity. Throughout a wide part of this range the visual system can perform as a photometer as indicated by the pupil reflex (Alpern and Ohba, 1 Present address: Institute for Sensory Research, Syracuse University, Syracuse, New York 13210, USA 2 Present address: Department of Neurology, Harvard Medical School, Beth Israel Hospital, 330 Brookline Avenue, Boston, Massachusetts 02215, USA
0 0 1 4 - 4 8 1 9 / 7 8 / 0 0 3 1 / 0 1 6 3 / $ 3.00
164
R.B. Barlow, Jr. et al.
1972; Spring and Stiles, 1948) and by brightness sensation (Stevens and Stevens, 1963; Barlow and Verrillo, 1976). Measurements of both types of responses show that neurons in the visual system must be capable of transmitting an intensity code that is graded over at least an 8 log-unit range. What is the neural code for light intensity? Physiological experiments have not yet provided a satisfactory answer. The general view is that the effect of retinal illumination on neural response is greatest in the photoreceptors and diminishes as one proceeds to higher levels in the visual system (for discussion, see Levick, 1973). This view is based primarily on the responses recorded from cells in excised tissue and in anesthetized preparations. The situation is complicated by the fact that the stimuli used in the physiological experiments differ in many ways from the stimuli used to measure pupil responses and brightness sensation. Thus apparent discrepancies may, in fact, be due to uncontrolled differences in the experimental situations. We have tried to eliminate some of the confounding variables by studying pupil responses, physiology, and psychophysics under as nearly identical conditions as possible. Our approach was to determine the operating range of the consensual pupil reflex and of neurons in the lateral geniculate nucleus (LGN) in the awake monkey in response to ganzfeld illumination of the retina. We chose to study these responses because they are both influenced directly by the activity of optic nerve fibers and thus the results should reflect intensity coding in the optic nerve. Measurements of both types of responses were made under the same stimulus conditions that yielded psychophysical evidence for wide-range intensity coding in humans (Barlow, 1976; Barlow and Verrillo, 1976). Our results show that both the pupil reflex and the L G N cell responses are graded over wide intensity ranges. These data are consistent with the results of recent studies of maintained discharge rates showing that central visual neurons can code incident diffuse stimuli over at least several log units of light intensity (Bartlett and Doty, 1974; Papaioannou and White, 1972).
Methods Three monkeys (Macaca fascicularis, two females and one male) ranging in weight from 2.5-5.5 kg had one eye surgicallyimmobilized and a stainless steel recording well and headbolts implanted as described in the preceding paper (Snodderly et al., 1978).
Ganffeld Optical System A monocular ganzfeld optical system designed for human psychophysicalexperiments was modified for use with monkeys (Barlow and Verrillo, 1976). The present system differed from the earlier one primarily in having 10 x greater light intensity. Briefly, a ping-pongball was cut in haft and trimmed to an ellipsoid of 35 x 32 mm to fit snugly over the eye of the monkey. The cut edge of the ping-pongball was glued to the open tip of an opaque white cone; the inside surface of the cone and the convexsurface of the ping-pong ball were illuminated with light from a tungsten filament lamp (600 W BVE, General Electric, 100 V, color temperature 2800~K as previouslydescribed (Barlow and Verrillo, 1976). Broadband color filters (Wratten Nos. 29, 47B, and 58) or neutral density filters could be inserted in the beam. The maximum intensity of white light (400-700 nm) incident on the cornea was 1.0 Lamberts (L) at the arbitrary zero setting, indicated in the figures of this paper by 0 log L.
Intensity Coding in Primate Visual System
165
Pupil Reflex Measurements An awake animal was placed in a standard primate chair and clamps from a stable superstructure were attached to bolts implanted in the skull. The ganzfeld optical system was aligned directly in front of the immobilized eye with the speculum and contact lens in place. The pupil of the unoperated eye was focused on the image plane of a 35-ram camera (Medical Nikor) equipped with high-speed infrared film (HIE-135, Kodak) and flash (87C filter, Kodak). After the animal had dark adapted for 30 rain, a 3-sec flash of visible light was delivered to the immobilized eye and the pupil of the unoperated eye was photographed with infrared. The animal was allowed to dark adapt according to a schedule of human dark-adaptation times measured in the same apparatus.The test flash was then repeated. The intensity of the test flash was varied at random over a 9 log-unit range (10 -9 to 1 L). A flash duration of 3 sec was long enough to elicit a nearly complete consensual reflex, yet short enough to avoid excessive amounts of light adaptation. The consensual reflex was measured at each flash intensity by projecting the image of each negative onto a large grid (magnification factor -= 25). This technique yielded reliable measurements of changes in the pupil diameter of 0.1 mm or more.
MicroelectrodeRecording On a separate day, single-unit recordings were made from LGN cells while the immobilized eye was stimulated by ganzfeld flashes as in the pupil reflex measurements. Before taking measurements on intensity coding, cells were first classified into one of three general groups: (I) those excited by photopically balanced blue, red, and green flashes, (II) those inhibited by the flashes, and (III) those giving opponent responses to the flashes. All cells reported in this study were recorded from the parvocellular layers of the LGN. After a cell was classified, its response was recorded to flashes of white light. The long flash duration (1-3 sec) used in the pupil measurements and in some of the human psychophysical experiments were not practical because the cumulative effects of light adaptation made the experimental protocol too lengthy for single unit recording. As a compromise, we used flash durations of 100 to 500 msec. Since human brightness sensation is little changed when the flash is shortened to even 10 msec, this is probably not an important variable (Barlow and Verrillo, 1976). An experimental run consisted of nine test flashes covering an intensity range of 8 or more log units in 1 log-unit steps. Blanks were inserted periodically. The sequence of flash intensities in each run was random with the restriction that a test flash could not be 3 log units dimmer than the preceding flash. Between flashes the animal was dark adapted according to a schedule based on human psychophysical measurements (Barlow and Verrillo, 1976). The times between flashes were sufficiently long to permit human thresholds to return to within 0.3 log units of the dark-adapted level. This required longer interstimulus intervals than are normally used in single-unit studies. For example, the protocol for a typical series, listing the filter density paired with the time in minutes until the next stimulus was as follows: 6, 0.5; 5, 1; 2, 1.5; 4, 1; 8, 0.5; 3, 1.5; 1, 2; 0, 2; ~, 0.5; 7, 0.5; 9, 0.5. A single run was complete in about 25 rain, and an entire session required a recording time of one hour which rarely occurred. Spike data from each test flash were recorded on magnetic tape and processed at a later time on digital computers (Linc-8 and PDP-12) with a resolution of 100 ~sec. Before each stimulus presentation, the unoperated eye of the monkey was observed by one of the experimenters using an infrared viewer. We tried to present stimuli only when the monkey appeared alert. If the monkey became drowsy and the eyelid began to close, the animal was awakened by auditory stimuli. When this was ineffective and the animal clearly remained inattentive during a test flash, that trial was deleted from the data processing.
Results
Pupil Reflex Figure 1 shows the consensual pupil reflex of the awake macaque for different levels of ganzfeld illumination. Each infrared photograph of the right eye was
166
R.B. Barlow, Jr. et al.
0
-3
-6
Fig. 1. The consensual pupil reflex of the right eye of an awake macaque photographed in the infrared. Each photograph was taken just prior to the termination of a 3-sec flash of white light delivered to the dark-adapted, immobilized left eye. No changes in pupil diameter were detected for flash intensities of less than - 9 log Lamberts (L)
-9 LOG L A M B E R T S
IO --
I
I
I
I
h
O
-
/
-
8, .,"8 / " o
/
It.u
/
O
o,1." " / ~
.J
o=. z
,'8/
9
o,e .
IJJ
"r
~
I I
-8
I
-6
I
I
-4 -2 LOG LAMBERTS
I
0
Fig. 2. Magnitude of the consensual pupil reflex as a function of log flash intensity. Filled and unfilled circles give the results for two animals. Dashed and solid lines connect the geometric means of the data for each monkey. Broken line gives comparable data for the human pupil (Barlow, 1976)
taken just before the termination of a 3-sec flash of white light delivered to the dark-adapted, immobilized left eye. The magnitude Of the consensual reflex was determined with this technique for flash intensities ranging from - 9 to 0 log L. Shifts in gaze of the mobile right eye could impair the measurements of pupil diameter when the direction of gaze deviated from the optic axis of the camera at the instant the picture was taken. In Figure 1 the direction of gaze was straight into the camera for the test at -3, - 6 , and - 9 log L but was about 10 ~ off axis for the test at 0 log L. Tests in which the direction of gaze was within 10 ~ of the axis of the camera generally yielded consistent results and only those within this range were used for the quantitative analysis. The direction of gaze was judged by the position of the bright spot in the pupil which is the reflected image of the ring flash.
IntensityCodingin Primate VisualSystem
167
Figure 2 summarizes the results of the pupil reflex measurements for two monkeys. Plotted on the ordinate is the change in pupil diameter on a log scale and plotted on the abscissa is the log of the flash intensity. Curves are fitted to the geometric means at each intensity. The lower portions of the curves roughly approximate a power function relationship between flash intensity and pupil reflex. A least squares fit to the data from -8 to -2 log L yields power functions with exponents of 0.22 for the filled circles and 0.21 for the unfilled circles. The value of the exponent will be considered in the Discussion. Both sets of data in Figure 2 indicate that the magnitude of the consensual pupil reflex increased monotonically over a range of at least 8 log units of light intensity. Since much of the earlier work on intensity-response functions (e.g. Jacobs, 1965; Marrocco, 1975) in the primate visual system has utilized animals anesthetized with barbiturates, we attempted to establish a comparison situation by eliciting the pupil reflex in a barbiturate-anesthetized monkey. One set of measurements with the same stimulus conditions was made on an animal anesthetized with 35 mg/kg sodium pentobarbital. When this was done the pupil remained about 50% constricted in the dark (4.1 mm diam.) and further constriction occurred only at the three highest test intensities.
Responses of LGN Cells Figure 3 gives a computer reconstruction of the spike trains recorded from an LGN cell in response to 100-msec flashes of white light in the ganzfeld. The flash intensities were randomized in the experimental protocol (see Methods), but for purposes of illustration the responses are arranged in order of increasing intensity. The data were obtained from the same monkey whose pupil reflex measurements are given by the filled circles and solid curve in Figure 2. Intense flashes elicited strong bursts of impulses and pronounced afterdischarges from the cell in Figure 3. The afterdischarges for the three highest test intensities persisted for at least 3 sec after the flash. Flashes of intermediate intensity elicited weaker responses with no detectable afterdischarge. No clear responses were evoked by the dimmest flashes. These data were recorded from an LGN cell that we classified in Group I, that is, a cell that was excited by photopically balanced blue, green, and red test flashes. As indicated below, other cells in Group I yielded similar results. Figure 4 gives instantaneous frequency plots of segments of six records in Figure 3. For each of six intensities the impulse discharge of the LGN cell is plotted as a sequence of instantaneous firing rates (the inverse of the interspike intervals) for a 400-msec period. Dark bars indicate the duration of the test flash. This type of display demonstrates several characteristics of the impulse discharge that are not readily apparent in the spike trains in Figure 3. First, the discharge often contains bursts of spikes as, for example, seen in the responses to -3 and -1 log L. Second, the peak frequency (minimum interspike interval) within a burst increases with flash intensity. At -4 log L the minimum interspike interval was about 2.5 msec whereas at 0 log L it was less than 1.5 msec. The instantaneous frequency plots also show the prominent aflerdischarge that is characteristic of responses to intense flashes.
168
R.B. Barlow, Jr. et al. i
I
I d l ldl l li
-I
~1 erie
l=l[llli
-2
idntlrl,n, lllll tltll
-3
j,,
0
i
"5
-6
-7
-~
II
,J
I|11111 l l i J IIm ~m t Jlllll l
I HIIHIUtiIIIIII IIIdEIHIIIIII III IJ
ill = jll~l~IFIIHmN J iHnl m~ rlr === I r==m= I= fill= rl=
,i I n(rltlll, i,lt~'d
,nuluHuUu hldJ'i')v ~,llunUdnuh
I, ,',' ,',HI,~7,11~7,
,l,,'~,~,h'NlT?','ll,
I I I III I I IlflI II / II II / I IIl[lllll I [I I ir Ii l I II E I IllilMIII II I IIII Ill111 I I I III II I IIIIIII I I II III I l I I l fill IIlI Illl , I1'1 I 'l l ' 111 I I i[ll I 11 I I 1 II I III I II I Ill III IIII IIII I I IIII I I II ill ~ I ii Iii i Ii ~, I I IiIlll I I tl / I I I I II I I I
"4
I lIJl l l | l l l l lIl
l[ IIIII
1
I
" ,t, ,",,,, , I' " III1f ' I II [ I I tll l/ I II I[ I IY I II II II II I I IJ I I I I I IIIlt irl~itl i III Ill III II tl II I 11 l I i '171 IIII If t I Illl Illtiirr i rllrll Ill , t lllllllltl I
Fig. 3. Spike trains fired by a single LGN cell at different test-flash intensities. Each trace is a computer reconstruction in which vertical fines denote the times of occurrence of the spikes recorded i sec before a 0.1-sec flash, during the flash, and 0.9 sec after the flash. Flash is indicated by short horizontal bars. Numbers at the left give the flash intensity in log Lamberts. Two test runs are displayed for each of the four highest intensities, and up to eight runs are given for lower intensities. Each test flash was presented under dark-adapted conditions by the ganzfeld optical system. This cell was classified in Group I (see Methods)
600 taJ tJd~
400
la. 03
o~ coo
z O
i
i
-5
-4
n
-5 LOG
-2 LAMBERTS
-7
Fig. 4. Effect of light intensity on the firing pattern of the LGN cell in Figure 3. Shown are six 400-msec segments of the instantaneous firing rate (inverse of interspike interval) at the six flash intensities indicated on the abscissa
Response Latencies We considered several possible methods for relating the characteristics of the responses to light intensity. From Figures 3 and 4 one obvious measure is the total number of spikes fired in response to the test flash. Since increases in intensity caused decreases in the latency of response, counting the number of spikes required specification of the latency of the discharge as a function of light intensity. We did this by calculating a cumulative function for the spike train similar to that used by Freund et al. (1972). All the trials at a given intensity were summed together and an average prestimulus discharge rate was calculated based on a 2-sec period preceding the stimulus. This average rate was designated the "expected" rate in the absence of a response, when the stimulus was below
Intensity Coding in Primate Visual System I00
I
I
I
I
I
I -:3
169 I
~
E
I
I 0
50
20 I -6
I -5
-4 LOG
I -2
-I
LAMBERTS
Fig. 5. Effect of flash intensity on the latency of response of five Group I cells. Each point is the mean response latencyfor a single cell at one test intensity. Data were collected for all five cells at 0, -2, and -4 log L. Latencies could not be detected at flash intensities of less than -5 log L. Measurements were made only after the animal was dark adapted. Response latency (T) can be related to intensity by the method of least squares, givingthe power function T = ToL-'~ where To = 28.6 msec is the mean latency at 1 Lambert
threshold. By integrating over time the expected spike rate, we obtained an expected cumulative function that also aplied to the case of a subthreshold stimulus. This expected cumulative function was subtracted from the actual average cumulative function to judge when more than the expected number of spikes had arrived. The time at which the average cumulative function exceeded the expected number of spikes by one was taken as the latency of the response. This latency measurement gave consistent results when applied to excitatory responses. Figure 5 summarizes data from five Group I cells tested with intensities ranging over 5 or more log units. The mean latency decreased about 2.5 times from 70 msec at - 5 log L to 28 msec at 0 log L. A least squares fit to the data yields a power function with an exponent -0.08. These data are not adjusted for the much shorter shutter delay of about 3 msec.
Spike Counts The relationship between response latency and stimulus intensity was used as a guideline for establishing a count window to determine the number of spikes elicited by each flash. In order to minimize the chance of missing part of the response, the count window began slightly earlier than any of our recorded latencies. Counts for stimuli below - 5 log L, where no latency could be measured, all began with the same delay of 60 msec. The duration of the count window was 100 msec longer than the stimulus; this ensured that weak, long-latency responses would be detected. It also incorporated a small part of the afterdischarge at higher stimulus luminances. Because of the time needed ro run a complete series, we were able to obtain intensity functions covering the full range from threshold to maximum stimulus
170
R.B. Barlow, Jr. et al.
IOO
I
I
I
I
I
I
tll I00~
9 9
to
,~10 QC hX uJ
Y'
?_ t/d > l0
V--
to i
i
I
I
T
I
I
~,
g
7
T
o
LOG
LAMBERTS
Fig. 6. Intensity functions for four LGN cells classified in Group I. The unfilled circles give the mean number of extra spikes fired in response to 500-msec test flashes for three cells. The location of the 600-msec count window for -6 to 0 log L was delayed 60, 50, 40, 30, 25, 20 and 20 msec from the beginning of the stimulus to allow for response latency. The number of extra spikes is calculated as the number of spikes in the count window minus the expected number of spikes based on the firing rate in two secs preceding the stimulus. Each of the unfilled circles is the average of 5 to 7 stimulus presentations. The filled circles summarize the responses from the cell of Figures 3 and 4 for 100-msec flashes. Each point is the average of 2 to 6 presentations using a 200-msec count window with the same latency correction as above. The solid line gives the medians of the estimates of brightness by human subjects for 1-sec test flashes in the same ganzfeld apparatus (Barlow and Verrillo, 1976)
l u m i n a n c e for only five excitatory cells. Results from four of these cells are s h o w n in F i g u r e 6 along with h u m a n b r i g h t n e s s j u d g m e n t s m a d e u n d e r similar stimulus conditions. T h e fifth cell was n o t i n c l u d e d in the average b e c a u s e s o m e i n t e r m e d i a t e l u m i n a n c e s were n o t tested. T h e s e excitatory L G N cells h a v e o p e r a t i n g r a n g e s of five or m o r e log units. T h e i r responses differ m o s t strikingly f r o m the psychophysical b r i g h t n e s s f u n c t i o n in having higher thresholds. F o r example, the m e a n r e s p o n s e of the cell of F i g u r e 3 was n o t significant at - 5 log L ( C h i - s q u a r e test), whereas h u m a n d a r k - a d a p t e d t h r e s h o l d is - 9 . 4 log L ( B a r l o w a n d Verrillo, 1976). H o w e v e r , h u m a n p h o t o p i c t h r e s h o l d 3 in the ganzfeld a p p a r a t u s is - 5 log L (Barlow a n d Verrillo, 1976), which suggests t h a t o u r l i m i t e d s a m p l e of i n t e n s i t y f u n c t i o n s 3 Color identification of ganzfeld stimuli by dark-adapted human observers becomes chance below -5 log L (S. Bolanowski, S. Grufferman and R. Barlow, unpublished observations). This luminance level characterizes the mesopic region in human vision (LeGrand, 1947) and the photopic threshold of squirrel monkeys (Jaeobs, 1973)
'
Intensity Coding in Primate Visual System
171
A -I -2 -3 -4 -5 -6 -7
-8 -9
IIIIII IIIg I IIIIIIIIIIII
I IIII IIIII !11 IIII I II II IIIIIII II IIIlalll I I IIII I I IIIII IIII D II III I II III II II IIIIII II II II I II I II II II IIIIIIIIIII II II IIIIII III IIIII IIIII I IIIIII I I II Itll III IIII II
t
I
II I
I
I I
I
I
III III I I I If I gill
Ill I I II I It I I I II IIHII Illllllllll 11 II I I II IIIIIII IIIIIIR I IIII IIIIII Iflllllll IIIIIll I nl Ill III IIII1[I
I IIII I I IIIIIII I f
I I I
II
I lI III III IIII IIII I IIIIIII III llllllllll I II I I II I II fl# IIII II lllllllll III I II IIII II IIIII III IIIIII III IIIIII I gll IIIII IIII II I I II I III IIIIII IIIHII l III lllll III lllllllll Illllll I IIIIIII II III II IIIIIIIII gll IIII I I I IIIII Ill I II Ihl DIIIIIIIIIIIII III tl IIII I II I I I Igll I III II I I IIIII fill II IllUl III IIII I
I IIIIIIIIII Igllllll II I gl I III I I IIII t11t Hill II1[I II II Ill II]flll l]f IIIII I I I1 III I II II IIII III II II IIII I II IIIIIIIIII I IIII I llal I III IIII IHI I II III II I III II II III II IIII IlUlIII IUIIII
I //'t
I
II
B o
9
OQ
%
0
9
-I0
Exp. Brain Res. 31, 163-177 (1978)
9
Springer-Verlag 1978
Intensity Coding in Primate Visual System R.B. Barlow, Jr., 1 D.M. Snodderly and H.A. Swadlow 2 Eye Research Institute of Retina Foundation, 20 Staniford Street, Boston, Massachusetts 02l 14, USA
Summary. The pupil reflex and the discharge of L G N cells of the awake macaque were measured under stimulus conditions that yielded evidence for wide-range intensity coding in human psychophysical experiments. Ganzfeld flashes of white light were delivered under dark-adapted conditions to the surgically immobilized eye of the monkey while the other eye was observed in the infrared. Three-sec flashes elicited a consensual pupil reflex that was graded from - 8 to 0 log Lamberts (L), indicating that the optic nerve fibers are capable of coding at least an 8 log-unit range of light intensity. In the physiological experiments, shorter flashes ( 0 . 1 - 0 . 5 s e c ) but otherwise identical conditions elicited monotonically graded responses from one type of L G N cell over the photopic range of - 5 to 0 log L. Responses from other types of L G N cells were also graded over wide ranges but had different thresholds and, in some cases, nonmonotonic intensity-response functions. Latency of the excitatory L G N responses decreased with increasing intensity according to a power function with slope o f - 0 . 0 8 . The pupil reflex and the L G N cell excitatory responses approximate power functions of light intensity with exponents of 0.22 and 0.14-0.29 respectively. The range of intensity coding found for single L G N cells is the widest yet reported for diffuse stimuli.
Key words: Intensity coding - Ganzfeld illumination - Lateral geniculate nucleus - Pupil reflex - Response latencies
The human visual system has an effective operating range of more than 10 log units of light intensity. Throughout a wide part of this range the visual system can perform as a photometer as indicated by the pupil reflex (Alpern and Ohba, 1 Present address: Institute for Sensory Research, Syracuse University, Syracuse, New York 13210, USA 2 Present address: Department of Neurology, Harvard Medical School, Beth Israel Hospital, 330 Brookline Avenue, Boston, Massachusetts 02215, USA
0 0 1 4 - 4 8 1 9 / 7 8 / 0 0 3 1 / 0 1 6 3 / $ 3.00
164
R.B. Barlow, Jr. et al.
1972; Spring and Stiles, 1948) and by brightness sensation (Stevens and Stevens, 1963; Barlow and Verrillo, 1976). Measurements of both types of responses show that neurons in the visual system must be capable of transmitting an intensity code that is graded over at least an 8 log-unit range. What is the neural code for light intensity? Physiological experiments have not yet provided a satisfactory answer. The general view is that the effect of retinal illumination on neural response is greatest in the photoreceptors and diminishes as one proceeds to higher levels in the visual system (for discussion, see Levick, 1973). This view is based primarily on the responses recorded from cells in excised tissue and in anesthetized preparations. The situation is complicated by the fact that the stimuli used in the physiological experiments differ in many ways from the stimuli used to measure pupil responses and brightness sensation. Thus apparent discrepancies may, in fact, be due to uncontrolled differences in the experimental situations. We have tried to eliminate some of the confounding variables by studying pupil responses, physiology, and psychophysics under as nearly identical conditions as possible. Our approach was to determine the operating range of the consensual pupil reflex and of neurons in the lateral geniculate nucleus (LGN) in the awake monkey in response to ganzfeld illumination of the retina. We chose to study these responses because they are both influenced directly by the activity of optic nerve fibers and thus the results should reflect intensity coding in the optic nerve. Measurements of both types of responses were made under the same stimulus conditions that yielded psychophysical evidence for wide-range intensity coding in humans (Barlow, 1976; Barlow and Verrillo, 1976). Our results show that both the pupil reflex and the L G N cell responses are graded over wide intensity ranges. These data are consistent with the results of recent studies of maintained discharge rates showing that central visual neurons can code incident diffuse stimuli over at least several log units of light intensity (Bartlett and Doty, 1974; Papaioannou and White, 1972).
Methods Three monkeys (Macaca fascicularis, two females and one male) ranging in weight from 2.5-5.5 kg had one eye surgicallyimmobilized and a stainless steel recording well and headbolts implanted as described in the preceding paper (Snodderly et al., 1978).
Ganffeld Optical System A monocular ganzfeld optical system designed for human psychophysicalexperiments was modified for use with monkeys (Barlow and Verrillo, 1976). The present system differed from the earlier one primarily in having 10 x greater light intensity. Briefly, a ping-pongball was cut in haft and trimmed to an ellipsoid of 35 x 32 mm to fit snugly over the eye of the monkey. The cut edge of the ping-pongball was glued to the open tip of an opaque white cone; the inside surface of the cone and the convexsurface of the ping-pong ball were illuminated with light from a tungsten filament lamp (600 W BVE, General Electric, 100 V, color temperature 2800~K as previouslydescribed (Barlow and Verrillo, 1976). Broadband color filters (Wratten Nos. 29, 47B, and 58) or neutral density filters could be inserted in the beam. The maximum intensity of white light (400-700 nm) incident on the cornea was 1.0 Lamberts (L) at the arbitrary zero setting, indicated in the figures of this paper by 0 log L.
Intensity Coding in Primate Visual System
165
Pupil Reflex Measurements An awake animal was placed in a standard primate chair and clamps from a stable superstructure were attached to bolts implanted in the skull. The ganzfeld optical system was aligned directly in front of the immobilized eye with the speculum and contact lens in place. The pupil of the unoperated eye was focused on the image plane of a 35-ram camera (Medical Nikor) equipped with high-speed infrared film (HIE-135, Kodak) and flash (87C filter, Kodak). After the animal had dark adapted for 30 rain, a 3-sec flash of visible light was delivered to the immobilized eye and the pupil of the unoperated eye was photographed with infrared. The animal was allowed to dark adapt according to a schedule of human dark-adaptation times measured in the same apparatus.The test flash was then repeated. The intensity of the test flash was varied at random over a 9 log-unit range (10 -9 to 1 L). A flash duration of 3 sec was long enough to elicit a nearly complete consensual reflex, yet short enough to avoid excessive amounts of light adaptation. The consensual reflex was measured at each flash intensity by projecting the image of each negative onto a large grid (magnification factor -= 25). This technique yielded reliable measurements of changes in the pupil diameter of 0.1 mm or more.
MicroelectrodeRecording On a separate day, single-unit recordings were made from LGN cells while the immobilized eye was stimulated by ganzfeld flashes as in the pupil reflex measurements. Before taking measurements on intensity coding, cells were first classified into one of three general groups: (I) those excited by photopically balanced blue, red, and green flashes, (II) those inhibited by the flashes, and (III) those giving opponent responses to the flashes. All cells reported in this study were recorded from the parvocellular layers of the LGN. After a cell was classified, its response was recorded to flashes of white light. The long flash duration (1-3 sec) used in the pupil measurements and in some of the human psychophysical experiments were not practical because the cumulative effects of light adaptation made the experimental protocol too lengthy for single unit recording. As a compromise, we used flash durations of 100 to 500 msec. Since human brightness sensation is little changed when the flash is shortened to even 10 msec, this is probably not an important variable (Barlow and Verrillo, 1976). An experimental run consisted of nine test flashes covering an intensity range of 8 or more log units in 1 log-unit steps. Blanks were inserted periodically. The sequence of flash intensities in each run was random with the restriction that a test flash could not be 3 log units dimmer than the preceding flash. Between flashes the animal was dark adapted according to a schedule based on human psychophysical measurements (Barlow and Verrillo, 1976). The times between flashes were sufficiently long to permit human thresholds to return to within 0.3 log units of the dark-adapted level. This required longer interstimulus intervals than are normally used in single-unit studies. For example, the protocol for a typical series, listing the filter density paired with the time in minutes until the next stimulus was as follows: 6, 0.5; 5, 1; 2, 1.5; 4, 1; 8, 0.5; 3, 1.5; 1, 2; 0, 2; ~, 0.5; 7, 0.5; 9, 0.5. A single run was complete in about 25 rain, and an entire session required a recording time of one hour which rarely occurred. Spike data from each test flash were recorded on magnetic tape and processed at a later time on digital computers (Linc-8 and PDP-12) with a resolution of 100 ~sec. Before each stimulus presentation, the unoperated eye of the monkey was observed by one of the experimenters using an infrared viewer. We tried to present stimuli only when the monkey appeared alert. If the monkey became drowsy and the eyelid began to close, the animal was awakened by auditory stimuli. When this was ineffective and the animal clearly remained inattentive during a test flash, that trial was deleted from the data processing.
Results
Pupil Reflex Figure 1 shows the consensual pupil reflex of the awake macaque for different levels of ganzfeld illumination. Each infrared photograph of the right eye was
166
R.B. Barlow, Jr. et al.
0
-3
-6
Fig. 1. The consensual pupil reflex of the right eye of an awake macaque photographed in the infrared. Each photograph was taken just prior to the termination of a 3-sec flash of white light delivered to the dark-adapted, immobilized left eye. No changes in pupil diameter were detected for flash intensities of less than - 9 log Lamberts (L)
-9 LOG L A M B E R T S
IO --
I
I
I
I
h
O
-
/
-
8, .,"8 / " o
/
It.u
/
O
o,1." " / ~
.J
o=. z
,'8/
9
o,e .
IJJ
"r
~
I I
-8
I
-6
I
I
-4 -2 LOG LAMBERTS
I
0
Fig. 2. Magnitude of the consensual pupil reflex as a function of log flash intensity. Filled and unfilled circles give the results for two animals. Dashed and solid lines connect the geometric means of the data for each monkey. Broken line gives comparable data for the human pupil (Barlow, 1976)
taken just before the termination of a 3-sec flash of white light delivered to the dark-adapted, immobilized left eye. The magnitude Of the consensual reflex was determined with this technique for flash intensities ranging from - 9 to 0 log L. Shifts in gaze of the mobile right eye could impair the measurements of pupil diameter when the direction of gaze deviated from the optic axis of the camera at the instant the picture was taken. In Figure 1 the direction of gaze was straight into the camera for the test at -3, - 6 , and - 9 log L but was about 10 ~ off axis for the test at 0 log L. Tests in which the direction of gaze was within 10 ~ of the axis of the camera generally yielded consistent results and only those within this range were used for the quantitative analysis. The direction of gaze was judged by the position of the bright spot in the pupil which is the reflected image of the ring flash.
IntensityCodingin Primate VisualSystem
167
Figure 2 summarizes the results of the pupil reflex measurements for two monkeys. Plotted on the ordinate is the change in pupil diameter on a log scale and plotted on the abscissa is the log of the flash intensity. Curves are fitted to the geometric means at each intensity. The lower portions of the curves roughly approximate a power function relationship between flash intensity and pupil reflex. A least squares fit to the data from -8 to -2 log L yields power functions with exponents of 0.22 for the filled circles and 0.21 for the unfilled circles. The value of the exponent will be considered in the Discussion. Both sets of data in Figure 2 indicate that the magnitude of the consensual pupil reflex increased monotonically over a range of at least 8 log units of light intensity. Since much of the earlier work on intensity-response functions (e.g. Jacobs, 1965; Marrocco, 1975) in the primate visual system has utilized animals anesthetized with barbiturates, we attempted to establish a comparison situation by eliciting the pupil reflex in a barbiturate-anesthetized monkey. One set of measurements with the same stimulus conditions was made on an animal anesthetized with 35 mg/kg sodium pentobarbital. When this was done the pupil remained about 50% constricted in the dark (4.1 mm diam.) and further constriction occurred only at the three highest test intensities.
Responses of LGN Cells Figure 3 gives a computer reconstruction of the spike trains recorded from an LGN cell in response to 100-msec flashes of white light in the ganzfeld. The flash intensities were randomized in the experimental protocol (see Methods), but for purposes of illustration the responses are arranged in order of increasing intensity. The data were obtained from the same monkey whose pupil reflex measurements are given by the filled circles and solid curve in Figure 2. Intense flashes elicited strong bursts of impulses and pronounced afterdischarges from the cell in Figure 3. The afterdischarges for the three highest test intensities persisted for at least 3 sec after the flash. Flashes of intermediate intensity elicited weaker responses with no detectable afterdischarge. No clear responses were evoked by the dimmest flashes. These data were recorded from an LGN cell that we classified in Group I, that is, a cell that was excited by photopically balanced blue, green, and red test flashes. As indicated below, other cells in Group I yielded similar results. Figure 4 gives instantaneous frequency plots of segments of six records in Figure 3. For each of six intensities the impulse discharge of the LGN cell is plotted as a sequence of instantaneous firing rates (the inverse of the interspike intervals) for a 400-msec period. Dark bars indicate the duration of the test flash. This type of display demonstrates several characteristics of the impulse discharge that are not readily apparent in the spike trains in Figure 3. First, the discharge often contains bursts of spikes as, for example, seen in the responses to -3 and -1 log L. Second, the peak frequency (minimum interspike interval) within a burst increases with flash intensity. At -4 log L the minimum interspike interval was about 2.5 msec whereas at 0 log L it was less than 1.5 msec. The instantaneous frequency plots also show the prominent aflerdischarge that is characteristic of responses to intense flashes.
168
R.B. Barlow, Jr. et al. i
I
I d l ldl l li
-I
~1 erie
l=l[llli
-2
idntlrl,n, lllll tltll
-3
j,,
0
i
"5
-6
-7
-~
II
,J
I|11111 l l i J IIm ~m t Jlllll l
I HIIHIUtiIIIIII IIIdEIHIIIIII III IJ
ill = jll~l~IFIIHmN J iHnl m~ rlr === I r==m= I= fill= rl=
,i I n(rltlll, i,lt~'d
,nuluHuUu hldJ'i')v ~,llunUdnuh
I, ,',' ,',HI,~7,11~7,
,l,,'~,~,h'NlT?','ll,
I I I III I I IlflI II / II II / I IIl[lllll I [I I ir Ii l I II E I IllilMIII II I IIII Ill111 I I I III II I IIIIIII I I II III I l I I l fill IIlI Illl , I1'1 I 'l l ' 111 I I i[ll I 11 I I 1 II I III I II I Ill III IIII IIII I I IIII I I II ill ~ I ii Iii i Ii ~, I I IiIlll I I tl / I I I I II I I I
"4
I lIJl l l | l l l l lIl
l[ IIIII
1
I
" ,t, ,",,,, , I' " III1f ' I II [ I I tll l/ I II I[ I IY I II II II II I I IJ I I I I I IIIlt irl~itl i III Ill III II tl II I 11 l I i '171 IIII If t I Illl Illtiirr i rllrll Ill , t lllllllltl I
Fig. 3. Spike trains fired by a single LGN cell at different test-flash intensities. Each trace is a computer reconstruction in which vertical fines denote the times of occurrence of the spikes recorded i sec before a 0.1-sec flash, during the flash, and 0.9 sec after the flash. Flash is indicated by short horizontal bars. Numbers at the left give the flash intensity in log Lamberts. Two test runs are displayed for each of the four highest intensities, and up to eight runs are given for lower intensities. Each test flash was presented under dark-adapted conditions by the ganzfeld optical system. This cell was classified in Group I (see Methods)
600 taJ tJd~
400
la. 03
o~ coo
z O
i
i
-5
-4
n
-5 LOG
-2 LAMBERTS
-7
Fig. 4. Effect of light intensity on the firing pattern of the LGN cell in Figure 3. Shown are six 400-msec segments of the instantaneous firing rate (inverse of interspike interval) at the six flash intensities indicated on the abscissa
Response Latencies We considered several possible methods for relating the characteristics of the responses to light intensity. From Figures 3 and 4 one obvious measure is the total number of spikes fired in response to the test flash. Since increases in intensity caused decreases in the latency of response, counting the number of spikes required specification of the latency of the discharge as a function of light intensity. We did this by calculating a cumulative function for the spike train similar to that used by Freund et al. (1972). All the trials at a given intensity were summed together and an average prestimulus discharge rate was calculated based on a 2-sec period preceding the stimulus. This average rate was designated the "expected" rate in the absence of a response, when the stimulus was below
Intensity Coding in Primate Visual System I00
I
I
I
I
I
I -:3
169 I
~
E
I
I 0
50
20 I -6
I -5
-4 LOG
I -2
-I
LAMBERTS
Fig. 5. Effect of flash intensity on the latency of response of five Group I cells. Each point is the mean response latencyfor a single cell at one test intensity. Data were collected for all five cells at 0, -2, and -4 log L. Latencies could not be detected at flash intensities of less than -5 log L. Measurements were made only after the animal was dark adapted. Response latency (T) can be related to intensity by the method of least squares, givingthe power function T = ToL-'~ where To = 28.6 msec is the mean latency at 1 Lambert
threshold. By integrating over time the expected spike rate, we obtained an expected cumulative function that also aplied to the case of a subthreshold stimulus. This expected cumulative function was subtracted from the actual average cumulative function to judge when more than the expected number of spikes had arrived. The time at which the average cumulative function exceeded the expected number of spikes by one was taken as the latency of the response. This latency measurement gave consistent results when applied to excitatory responses. Figure 5 summarizes data from five Group I cells tested with intensities ranging over 5 or more log units. The mean latency decreased about 2.5 times from 70 msec at - 5 log L to 28 msec at 0 log L. A least squares fit to the data yields a power function with an exponent -0.08. These data are not adjusted for the much shorter shutter delay of about 3 msec.
Spike Counts The relationship between response latency and stimulus intensity was used as a guideline for establishing a count window to determine the number of spikes elicited by each flash. In order to minimize the chance of missing part of the response, the count window began slightly earlier than any of our recorded latencies. Counts for stimuli below - 5 log L, where no latency could be measured, all began with the same delay of 60 msec. The duration of the count window was 100 msec longer than the stimulus; this ensured that weak, long-latency responses would be detected. It also incorporated a small part of the afterdischarge at higher stimulus luminances. Because of the time needed ro run a complete series, we were able to obtain intensity functions covering the full range from threshold to maximum stimulus
170
R.B. Barlow, Jr. et al.
IOO
I
I
I
I
I
I
tll I00~
9 9
to
,~10 QC hX uJ
Y'
?_ t/d > l0
V--
to i
i
I
I
T
I
I
~,
g
7
T
o
LOG
LAMBERTS
Fig. 6. Intensity functions for four LGN cells classified in Group I. The unfilled circles give the mean number of extra spikes fired in response to 500-msec test flashes for three cells. The location of the 600-msec count window for -6 to 0 log L was delayed 60, 50, 40, 30, 25, 20 and 20 msec from the beginning of the stimulus to allow for response latency. The number of extra spikes is calculated as the number of spikes in the count window minus the expected number of spikes based on the firing rate in two secs preceding the stimulus. Each of the unfilled circles is the average of 5 to 7 stimulus presentations. The filled circles summarize the responses from the cell of Figures 3 and 4 for 100-msec flashes. Each point is the average of 2 to 6 presentations using a 200-msec count window with the same latency correction as above. The solid line gives the medians of the estimates of brightness by human subjects for 1-sec test flashes in the same ganzfeld apparatus (Barlow and Verrillo, 1976)
l u m i n a n c e for only five excitatory cells. Results from four of these cells are s h o w n in F i g u r e 6 along with h u m a n b r i g h t n e s s j u d g m e n t s m a d e u n d e r similar stimulus conditions. T h e fifth cell was n o t i n c l u d e d in the average b e c a u s e s o m e i n t e r m e d i a t e l u m i n a n c e s were n o t tested. T h e s e excitatory L G N cells h a v e o p e r a t i n g r a n g e s of five or m o r e log units. T h e i r responses differ m o s t strikingly f r o m the psychophysical b r i g h t n e s s f u n c t i o n in having higher thresholds. F o r example, the m e a n r e s p o n s e of the cell of F i g u r e 3 was n o t significant at - 5 log L ( C h i - s q u a r e test), whereas h u m a n d a r k - a d a p t e d t h r e s h o l d is - 9 . 4 log L ( B a r l o w a n d Verrillo, 1976). H o w e v e r , h u m a n p h o t o p i c t h r e s h o l d 3 in the ganzfeld a p p a r a t u s is - 5 log L (Barlow a n d Verrillo, 1976), which suggests t h a t o u r l i m i t e d s a m p l e of i n t e n s i t y f u n c t i o n s 3 Color identification of ganzfeld stimuli by dark-adapted human observers becomes chance below -5 log L (S. Bolanowski, S. Grufferman and R. Barlow, unpublished observations). This luminance level characterizes the mesopic region in human vision (LeGrand, 1947) and the photopic threshold of squirrel monkeys (Jaeobs, 1973)
'
Intensity Coding in Primate Visual System
171
A -I -2 -3 -4 -5 -6 -7
-8 -9
IIIIII IIIg I IIIIIIIIIIII
I IIII IIIII !11 IIII I II II IIIIIII II IIIlalll I I IIII I I IIIII IIII D II III I II III II II IIIIII II II II I II I II II II IIIIIIIIIII II II IIIIII III IIIII IIIII I IIIIII I I II Itll III IIII II
t
I
II I
I
I I
I
I
III III I I I If I gill
Ill I I II I It I I I II IIHII Illllllllll 11 II I I II IIIIIII IIIIIIR I IIII IIIIII Iflllllll IIIIIll I nl Ill III IIII1[I
I IIII I I IIIIIII I f
I I I
II
I lI III III IIII IIII I IIIIIII III llllllllll I II I I II I II fl# IIII II lllllllll III I II IIII II IIIII III IIIIII III IIIIII I gll IIIII IIII II I I II I III IIIIII IIIHII l III lllll III lllllllll Illllll I IIIIIII II III II IIIIIIIII gll IIII I I I IIIII Ill I II Ihl DIIIIIIIIIIIII III tl IIII I II I I I Igll I III II I I IIIII fill II IllUl III IIII I
I IIIIIIIIII Igllllll II I gl I III I I IIII t11t Hill II1[I II II Ill II]flll l]f IIIII I I I1 III I II II IIII III II II IIII I II IIIIIIIIII I IIII I llal I III IIII IHI I II III II I III II II III II IIII IlUlIII IUIIII
I //'t
I
II
B o
9
OQ
%
0
9
-I0
