Mechanisms of Visual Sensitivity: Backgrounds and Early Dark Adaptation

i,%on Res. Vol. 23. No. I?. pp. 1423-3432.1983 0042-6989X3 53.00 + 0.00 Printed III Great Britain. Ail rights reserved Copyright .c 1983Pergamon Press Ltd

MECHANISMS OF VISUAL SENSITIVITY: BACKGROUNDS AND EARLY DARK ADAPTATION

WILSON S. GEISLER Department of Psychology, University of Texas, Austin, TX 78712. U.S.A.

(Received 2 October 1980; in raised form 14 June 1983)

Abstract-There is substantial physiological and psychophysical evidence for an adaptation mechanism whose effect, under many circumstances, is equivalent to placing a neutral density filter in front of the eye. Furthermore. this mechanism is of sufficient strength to predict the generalized Weber‘s law for increment thresholds on steady backgrounds. However, it was shown that an additional transient mechanism (with a time-course of around 100 msec) is also needed to account for the inc~ment-threshold results. The effect of this mechanism on increment thresholds during early dark adaptation was parametrically examined. Several models for the transient mechanism were considered. The one best able to account for the results consists of a subtractive inhibitory stage operating prior to a saturating nonlinearity.

Light adaptation Dark adaptation Increment thresholds Lateral inhibition Weber’s law

INTRODUCTION adaptation” model. The evidence comes from experi- In the cone system, increment and decrement thresh- ments in which flashed-background increment- olds measured against steady backgrounds are ap threshold functions were measured in the fovea under proximately described by Weber’s law over a wide various fixed states of light adaptation. For example. range of background intensities and test-field param- Geisler (1981) measured flashed-background eters. Substantial exceptions to the law occur for test increment-threshold functions 500 msec after the stimuli that are very small and brief (Barlow, 1958; offset of adapting backgrounds. Figure I shows the Geisfer, 1979a) or of high spatial or temporal fre- pattern of results obtained in the study. The solid quencies (Kelly, 1972); but. even under these condi- curve labelled - co is the increment-threshold functions the thresholds converge toward Weber’s law as tion obtained for flashed backgrounds in the dark- background intensity is raised. The widespread valid- adapted fovea. The other solid curves represent ity of Weber’s law across both stimulus conditions and species may be due to its usefulness for pattern recognition. In particular, Weber’s law implies that a visual system exhibit pattern constancy-the prop- erty that the set of visible lines and edges within the visual field remains constant across all illumination levels (Hemila, 1977; Geisler. 1978a). Many models for the adaptation processes underlying Weber’s law have been proposed. These include logarithmic response compression coupled with a constant criterion for detection (Cornsweet and Pin- sker, 1965), power-law response compression with increasing internal variability (Mansfield, 19X), linear transduction with quanta1 fluctuations and sources of correlated neural noise (Triesman, 1966), power-law transduction followed by a multiplicative adaptation factor (Thijssen and Vendrik. 1971), and 0 I 2 3 4 5 nonlinear transduction preceded by a multiplicative LO6 BICKMOUNO adaptation factor (Lythgoe. 1936: Craik, 1938; Geis- Fig. I. The pattern of increment-threshold functions ler. 1975: 1981: Dawis and Purple. 1980). Recently, reported in Geisler (1981). The solid curves are psychophysical evidence has been accumulating for washy-background inc~ment-th~hold functions obtained the latter type of model (Geisler. 197Xb. 1981: Hood, at various fixed levels of adaptation. The dashed curve is the optimal function that can be obtained with multiplicative 1978; Hood ef ui.. 1978). which will be referred to adaptation. The dotted curve shows the actual function that here as the “dark-glasses” or “multiplicative- is obtained.

1423 1424 WILSON S. GEISLEK flashed-background increment-thr~shoId functions steady backgrounds that would be obtained if only obtained for the adapting-background intensities in- multiplicative adaptation were operating. The dotted dicated on the left. The functions obtained at all curve shows the function that is obtained: it is 0.7-l .O adapting-background intensities are roughly identical log units lower than the dashed curve at moderate to in shape and continuously accelerating when plotted high background levels. Changes in the maximum of in log-log coordinates. Furthermore, a function ob- the intensity-response function, as least of the sort tained at any given adapting-background intensity observed in electrophysiological research. are also can be brought into coincidence with a function unable to account for the results. Typically. light obtained at any other intensity by translation along adaptation decreases the maximum of the inten- a 45 line. This is exactly the prediction made by the sity-response function which should cause increment simple multiplicative-adaptation model (Geisler, thresholds to increase rather than decrease. 1978b). There appears to be at least one other factor or Figure 1 is also useful for explaining how multi- mechanism responsible for the increment thresholds plicative adaptation accounts for Weber’s iaw. The obtained with steady adapting backgrounds. Experi- X’s indicate the point on each curve where the ments in which the adapting background is momen- minimum Weber fraction (AZ/l) is reached. AS can be tarily switched off and on just before presentation of seen. the points where the Weber fraction reaches its the increment field (Ikeda and Boynton. 1965: Gei- minimum trace out a straight line of slope one (the sler, 1981) provide further evidence for this additional dashed line). In the standard increment-threshold mechanism. These studies indicate that this mech- experiment the adapting-background intensity and anism has a relatively quick decay time---on the order the background intensity against which threshold is of 100 msec. The purpose of the present experiments measured are identical. Now, if the adapting back- was to further explore this mechanism which is grounds produce just enough multiplicative adapta- apparently involved in the very early phases of light tion for the flashed-background increment-threshold and dark adaptation. curves to shift so that their minimum Weber fractions occur at the adapting-background intensity, then the increment-thresholds must be given by the X’s. In THEOR\ other words. Weber’s law is predicted. The results in Electrophysiological studies. including those listed Geisler ( 198 1) confirm that the increment-threshold in the introduction. have often found that the curves shift sufficiently to yield Weber’s law. flash-response characteristics of retinal neurons in the There is also considerable electrophysiological evi- vertebrate retina are described by the equation dence for the multiplicative adaptation model (Naka I” and Rushton, 1966; Dowling and Ripps, 1971, 1972; R = R,,, -. I” + I,” Glantz. 1972: Grabowski ef ai., 1972; Baylor and Hodgkin. 1974: Norman and Werblin, 1974; Werblin, where I is the intensity of the flash, R,,, is the 1974: Kleinschmidt and Dowling, 1975; Green ef al., maximum electrical response of the neuron. I, is the 1975: Fain, 1976; Norman and Perlman, 1979; Dawis half-saturation constant. and n is an exponent with a and Purple. 1980; Valeton and Norren, 1983). Most value usually between 0.7 and 1.0. As mentioned of these studies measured the effects of either bleaches above, changes in the flash-response characteristic or backgrounds on the intensity-response functions due to bleaching and backgrounds can usually be of retinal neurons. especially receptors. Typically it described by changes in I, (that is. by multiplicative was found that most of the changes in the intensity- adaptation) and by changes in R,,, (the maximum response functions are attributable to multiplicative response). In a previous paper (Geisler. 1981). a adaptation (i.e. to changes in the half-saturation simple model based on equation (I) was shown to constant). It is also typically found that light adapta- account fairly well for the effects of bleaches and tion produces some relatively small decreases in the backgrounds on flashed-background increment- maximum of the intensity-response function. Psycho- threshoId functions. It was assumed that threshold is physically. there is little evidence of such decreases reached when the response R , to the background- (Geisler. 1981); however. one would not expect the plus-increment exceeds the response R to the back- small effects observed in most electrophysiological ground alone by some fixed criterion amount (5.That studies to have much influence on the psychophysical is. threshold is reached when results. Although the multiplicative adaptation observed in R, - R=ii. (2) electrophysiological recordings, and in the appropri- The arguments in equation (1) for R + and R are, ate psychophysical experiments, is sufficient to respectively account for Weber’s law. the actual increment- threshold function obtained for steady backgrounds AI + s’ I i- ;: I, (3) cannot be explained by multiplicative adaptation alone (Geisler. 1981). The dashed curve in Fig. 1 and shows the (optimal) increment-threshold function for Mechanisms of visual sensitivity: backgrounds and early dark adaptation 1425 where I, is the intensity of the adapting background, intensity-r~ponse functions ~flash-response charac- and < and 1’are parameters which take into account teristics). and the upper plots (e.g. Fig. ZA) show temporal and spatial integration (more generally, the the corresponding flashed-background increment- weighting function) in order to adjust for the fact that threshold functions. The curves labelled (a) give the the adapting background, gashed background. and intensity-response or increment-threshold function in increment may be of different durations and dia- the absence of any adaptation or inhibition. The meters. Note that if the increment field is presented curves labelled (b) show the functions obtained with at a sufficient temporal or spatial distance from the some fixed level of adaptation or inhibition. In all adapting or flashed backgrounds then the corre- cases. these predictions were obtained under the sponding parameter, 4 or T, must be zero. assumption that the adapting background has been In this paper, only simple static models, like the switched off prior to presentation of the flashed one described above, will be considered in analyzing background and increment field [i.e. ;’ = 0.0 in terms the experimental results. Such models are commonly (3) and (4)]. Notice that each adaptation or inhibition used to describe the effects of bleaches and back- mechanism leads to a qualitatively different predic- grounds on neural and psychophysical responses. Of tion. course, they are extremely oversimplified represent- ations of the underlying dynamic processes, but there EXPERIMENT 1 are two justifications for considering them, at least as a starting point. The first is that their predictions are The pattern of results shown in Fig. 1 cannot he very easily derived and understood. Secondly, some explained by multiplicative adaptation alone. Given of the adaptation processes are slow with respect to that the flashed-background increment-threshold the time-course of response to the test stimuli. There- function is described by the solid curve labelled - 3~. fore, on any given test trial it is reasonable to view the lowest thresholds for steady backgrounds that the visual system as static with respect to these slow can be obtained by multiplicative adaptation are processes. Static models must, of course. be viewed limited by the dashed curve. As mentioned earlier, the with greater caution if the time-course of the under- actual increment-threshold function obtained is given lying dynamic processes are similar to the durations by the dotted curve. However. if the adapting back- of the test stimuli. For example, equation (1) has been ground is switched off and on for a couple of hundred utilized to describe how the peak response of a msec just before presentation of the increment field, neuron, or perhaps response at some fixed delay thresholds jump up to the dashed curve. thereby (Baylor er al., 1974). changes with flash intensity. becoming consistent with the predictions of multi- Thus, when applying the above model to flashed- plicative adaptation. Indeed, all the increment- background increment-threshold data it is implicitly threshold data that we have measured a couple of assumed that detection of the increment always oc- hundred msec or more after offset of adapting back- curs at the peak of the response or at some fixed grounds can be explained by multiplicative adapta- delay. However. at least under some conditions the tion alone (Geisler, 1981). The purpose of the present point in time when the increment is detected changes experiment was to examine the transient mechanism drastically with flashed-background intensity level by parametrically measuring the changes on thresh- (Geisler, 1981). In order to minimize this problem, old that occur during the first several hundred msec the stimulus conditions under consideration here after background offset. were picked so as to force detection to occur near the point in time when the peak response of the retina Method occurs (Geisler. 1981). All stimuli were generated in a 3-channel The models to be considered here consist of equa- Maxwellian-view system that was under control of a tions (1) and (2) in conjunction with one or more of PDP-I 1 computer. The three channels were derived the following six types of adaptation or inhibition: from the same tungsten source. The images of the (A, B) muItiplication of the total input intensity by a tungsten filament were 2 mm in diameter. and were factor between 0 and 1.O (multiplicative adaptation), superimposed at the center of the pupil. (C. D) multiplication of the output of the nonlinear The stimulus display was comprised of either a 45’ transducer by some factor (change in R,,,). (E, F) or 3’ increment field that was centered in a 5 subtraction of some value from the total input adapting background field. Suh.jects fixated the center intensity (subtractive inhibition before the non- of the imaginary cross formed by four red fixation linearity), (G, H) subtraction of some value from the lights that surrounded the background field. For ail OUtpUt of the nonlinear transducer (subtractive in- the experiments reported here. the stimuli consisted hibition following the nonline~Irity). (f. J) addition of of white tungsten light (3000 K). some value to the total input intensity (a simple “dark Ail wedges and filters were periodically calibrated light” model), and (K. L) addition of some value to with a PIN IO photodiode (United Detector Tech- the output of the nonlinear transducer. The eIIects of nologies, Santa Barbara, CA) placed in the position these six types of adaptation or inhibition are shown normally occupied by the subject’s pupil. The photo- in Fig. 2. The lower plots (e.g. Fig. 2B) show the diode was also used to set the maximum intensities of 1426

012345 012345 0 / 2 3 4 5 LOG FLASHED BACK~~~?WND

Fig. 2. Predicted intensity-response functions and flashed-background increment-threshold functions for various adaptation and inhibition mechanisms. The curves labelled (a) show the predictions in the absence of any adaptation or inhibition. The curves labelled (b) show the predictions for a fixed level of adaptation or inhibition. (A, B) The “dark glasses” mechanism. Multiplicative gain control prior to the nonlinearity. (C, D) The “change-in-R,,” mechanism. Multiplicative gain control following a nonlinearity. (E, F) Subtractive inhibition prior to a nonlinearity. (G, W) Subtractive inhibition following a nonlinearity. (K, L) Additive signals following a nonlinearity.

% , &‘,.‘“‘.‘.“““ -m 0 ; 2 3 4 0 OIO.20.30405 10305.0 7090 LOG GACKGROUNGtro) GAP DURATIONISECONDS )

Fig. 3. (Left) Increment-threshold functions for gap durations of 0, I50 and 9500 msec. with an intertrial interval of IO sec. For the gap of 9500 msec the eye remains essentially dark adapted at all background intensity levels. (Right) Increment threshold as a function of gap duration for a background of 3.94 log td. Note that the time scale changes from tenths of seconds to seconds after the hatch marks. Mechanisms of visual sensitivity: backgrounds and early dark adaptation 1427 the channels prior to each experimental session, and to check the intensities after each session. Absolute A o--o 3’ e-.-a 45’ calibrations were obtained with the method recom- mended by Westheimer (1965). The stimulus presentation sequence used in this experiment is shown in the inset of Fig. 3. The increment field was always flashed for 50msec once every IOsec. When the gap in the background was present, onset of the background and increment fields was simultaneous. Thresholds were measured as a function of gap duration, test field diameter, and background intensity. Note that measuring threshold t following a brief gap in the background is equivalent to measuring increment threshold during early dark adaptation. There were two subjects, the author and an experi- enced paid subject. All thresholds were obtained by the method of adjustment. A switch on the side of the response box allowed the subjects to deliver blank trials whenever they wished. This tended to increase the subjects’ confidence in the settings and to increase the reliability of the thresholds. Each data point in the figures below reflects the average of from 3 to 6 threshold settings.

Results 0 loo 200 300 400 500 Some of subject W.G.‘s results for the 45’ in- GAP DURAT1ON t YSEC) crement field are shown in Fig. 3. The open triangles are the increment thresholds obtained with a 9.5 set Fig. 4. ~nc~~t-th~hoId functions for increment field diameters of 3’ and 45’ and gap durations of 0 and 1SO msec. gap. Since the inter-trial interval was IOsec, this (A) Subject W.G. (8) Subject S.S. means that the background was only flashed for SOOmsec. Under these conditions the cone system remains essentially dark adapted at all background ning with the largest gap duration, working down- intensity levels (Geisler. 1981). The open squares are ward to 1.O sec. After each decrease in gap duration the increment thresholds obtained with steady adapt- the subject adapted to the new duration for at least ing backgrounds. and the open circles are the thresh- 2 min. olds obtained with a gap of 150 msec. As can be seen, Further data on the two subjects are shown in Figs the pattern of results in Fig. 3 is just like the one that 4 and 5. Figure 4 shows th~hold as a function of gap was obtained earlier (see Fig. 1). duration for background intensities of 3.94 and The solid circles show the thresholds obtained for 1.98 log td and for test sizes of 3’ and 45’. Figure 5 a 4 log td background, as the gap varied from 0 to shows increment-threshold functions obtained with 9.5 sec. Threshold increased very rapidly as the gap gaps of 0 and 150 msec and for test spot sizes of 3’ and duration was increased from 0 to IOOmsec, but 45’. The gap causes increases in threshold for both further increases to I.0 set produced little further sizes of test field but the effect is a little smaller for the changes in threshold. With yet further increases in 3’ spot size. For both test spot sizes and both back- gap duration. threshold slowly climbed up to its ground intensities the time course of the initial phase dark-adapted level. (Note that the scale changes from seems to be about the same, with the thresholds tenths of seconds to seconds after the hatch marks.) reaching an asymptotic value within the first 50-I 50 Although the initial phase of the curve up to at least msec of dark adaptation. For clarity standard errors 500msec truly shows the time course of increment- have not been plotted in any of the figures presented threshold changes during early dark adaptation, it here. For most data points the standard error of the should be noted that this is JIM so for the second mean is less than 0.02 log units and for no data point phase of the curve because of the IOsec repetition it is greater than 0.1 log unit. cycle. Actually, it takes many minutes of dark adaptation for the thresholds to rise from the plateau value observed with a gap of 500msec to the dark- The pattern of results on the right in Fig. 3 clearly adapted value observed with a gap of 9.5 set (Geisler, suggests that at least two distinct mechanisms are 1981). In order to obtain the data points for gap underlying the increment-threshold functions shown durations above I.0 sec. the subject was first dark on the left in Fig. 3. Previous results (Geisler. 1978b. adapted. then the thresholds were measured begin- 1981) have already shown that the adaptation efTects 1428 W~~solriS. GEISLER

citation would mean that the signals produced by the 4 A P - onset of an increment or background would not o--i3 150 MSEC 0 partially cancel themselves; at least, not as much. - 0 MSEC Therefore sensitivity should be greater against a 3 steady adapting background than against an adapting background with a brief gap preceding 2 presentation of the increment field. The relatively slow decay of inhibition would imply that the cancel- 3’ ling effect of a background would not decay instantly :::::“-: when the background was extinguished; thus. threshold would not rise instantly as gap duration is p 0 45’ increased [Fig. 2 (E, Ffj show that a fixed level of subtractive inhibition before the nonIinearity c - I3 1 raises threshold at low background levels. lowers “I 4 I- threshold at intermediate background levels and has little or no effect at high background levels. The :: 3 J almost constant difference in log threshold between the with- and without-gap conditions at moderate to 2 high background levels (see Figs 3 and 4) requires that the relative strength of excitation to inhibition be constant over this range. The fast rise in threshold as gap duration is increased would also occur if the maximum response. 0 R,,,, were to decrease during the first IOOmsec of dark adaptation. This sort of adaptation mechanism 0 I 2 3 4 is not consistent with receptor adaptation effects LOG BACKGROUND(TO) which typically show an increase in R,,, during dark adaptation; nonetheless, this mechanism co&d arise Fig. 5. Increment threshold as a function of gap duration for later in the visual system [Fig. 2 (C. D)] show increment field diameters of 3’ and 45’ and background levels of 3.98 and 1.98log td. (A) Subject W.G. (B) Subject that a fixed decrease in R,,,., causes a general increase S.S. in threshold at all background intensity levels. The other four possible mechanisms presented in remaining after the first 100 msec of dark adaptation the theory section cannot account for the increase in are explained by multiplicative adaptation (see Fig. threshold with gap duration. It was already shown in I). Actually it is highly likely that there are at least the introduction that multiplicative adaptation [Fig. three mechanisms governing the adaptation effects 2 (A, B)] cannot account for the effects. Subtractive observed after the first lOOmsec, with each con- inhibitjon following the nonlinearity can also be tributing its own approximately multiplicative effect. ruled out. Figure 2 (G. H) show that a fixed level of They are. ( I ) a short-term neural or network mech- subtractive inhibition raises threshold at very low anism. (2} a long-term receptor or bleaching mech- background levels. but has no effect at higher back- anism and (3) photopigment depletion (Dowling, ground levels. Furthermore, the increased threshold 1960. 1977: Hollins and Alpern. 1973; Geisler, 1979a, at very low background levels only occurs if the 19X1). inhibition exceeds the level necessary to completely What is the mechanism responsible for the fast rise suppress all response to the background. If this in threshold observed during the first IOOmsec of mechanism were to account for the rise in threshold dark adaptation? Of the six m~hanisms discussed in with gap duration. any persisting inhibition would the theory section. the most likely is a subtractive have to completely suppress the response to the inhibitory mechanism, iike those inferred from flicker background when it comes back on following the studies (Kelly. 1971; Kelly and Savoie, 1978; Roufs, gap. This is an absurd prediction that is contradicted 1972. 1974). operating prior to the major compressive by the subjective appearance of the background nonlinear stages responsible for threshold elevation which, if anything, appears brighter following the (Gersler. IYXI ). This explanation requires that the gap. An additive signal before the nonlinearity [Fig. inhibition build up and decay more slowly than 2 (I, J)] also seems extremely unlikely since the excitation an assumption that is also needed to additive signal would have to grow during the first explain the low-frequency roll-off seen in temporal IOOmsec of dark adaptation then remain at a con- modulation transfer functions (Kelly, 1961, 1971). stant high level for 1.O set or more. and it must be of Subtractive inhibrtion would cause the steady back- greater intensity than the background itself. The tinal ground stgnals to partially cancel themselves prior to mechanism. an additive signal following the non- the major nonlinearities. On the other hand, the linearity, can be trivially ruled out because it cannot relativjely slow build-up of inhibition relative to ex- affect increment threshold. Mechanisms of visual sensitivity: backgrounds and early dark adaptation 1429

- Iu) MSEC .-. 0 MSEC

Fig. 6. Flashed-background increment-threshold functions measured against an adapting background of 2.9 log td for gap durations of 0 and 150 msec. (A) Subject W.G. (B) Subject S.S.

EXPERIMENT 2 maximum of the nonlinearity, then the curves for the two conditions should be vertically separated by The purpose of this experiment was to determine approximately the same amount at all but the highest which of the two viable alternative hypotheses dis- flashed background levels, where they should become cussed above best accounts for the effects of gap even more separated. duration on threshold. The two alternatives, subtractive inhibition before the nonlinearity and changes in the maximum output after the non- Method Iinearity, make strong and very different predictions The stimulus con~guration and the procedure for for the experiment which is shown in the inset of Fig. obtaining thresholds were the same as in the first 6(A). The experiment involved measuring experiment. Increment-threshold functions for flashed-background increment threshold functions flashed backgrounds were measured against an against a axed-intensity adapting background under adapting background whose intensity was fixed at two conditions. In one. the adapting background was 2.9 log td. In one condition the adapting background presented continuously, and in the other, a gap was left on continuously and in the other a gap of occurred just before presentation of the flashed back- 150msec preceded presentation of the flashed background. ground. The gashed backgrounds were presented for When the adapting background is not switched off 500 msec once every IOsec. The duration of the during presentation of the flashed background, the increment field was always 50 msec and its onset was prediction of the subtractive inhibition hypothesis is simultaneous with that of the flashed background. different from that shown in Fig. 2(E. F). If, in Following 12min of dark adaptation, the subject accord with the hypotbesis, introducing a gap results adapted to the 2.9 log td background for 2-3 min. In in the decay of a subtractive inhibitory signal then the each experimental session, thresholds were measured effect of the gap must be just like increasing the in both conditions and at all flashed-background adapting background intensity by some fixed levels. amount. This should cause the curves for the two conditions (in log-log coordinates) to be separated at Results and discussion low flashed background levels and come together at high flashed background levels. The reason is that a The results for two subjects are shown in Fig. 6. fixed added intensity makes a large contribution to The solid circles are the thresholds obtained without the total background intensity at low flashed back- the gap, the open circles with the gap. Note that when ground levels but a negligible contribution at high the flashed background is not presented the stimulus flashed background levels. The predicted curves conditions are just like those in the first experiment. actually look like those in Fig. 2 (I), except that the Thus, the two thresholds plotted at - n; in Fig. 6 labels, (a) and (b), on the curves must be inter- were obtained under the same conditions as the open changed. circles and squares plotted at about 3.0 log td in The prediction of the other hypothesis is not Fig. 3. qualitatively different than that shown in Fig. 2 Of the two hypotheses, it is clear that the (C, D). If introducing the gap causes a decrease in the subtractive-inhibition hypothesis is best able to ac- 1430 WILSON S. GEISLER

count for the results. although the predictions are still duration flashes) that are 3-5 log td. which is at least not perfect. In particular. the dashed curves show the i-3 log units above the levels estimated from the increment-threshold function predicted from the psychophysical results (Geisler. 1981). It could be solid circles under the assumption that the gap added argued that the LRP and LERG studies overestimate a fixed intensity to the flashed background. the half-saturation constant of most receptors be- In addition to Fig. 6, there is some further evidence cause the measurement is a sum across many recep- for a subtractive inhibitory mechanism. First of tors whose half-saturation constants and photo- all, in a previous paper (Geisler. l981) flashed- pigments may differ. However. this ma!’ not be a background increment-threshold functions were mea- strong objection since it can also be argued that the sured against steady adapting backgrounds of vari- signal driving any postreceptor neurons 1s itself the ous intensities. It was shown that all of the data are sum of signals from a number of receptors. accurately predicted by equations (I) and (2) and Adelson (1982) has obtained evidence for sub- terms (3) and (4) with only changes in the half- tractive and multiplicative adaptation in the rod saturation constant (i.e. with only multiplicative ad- system. Like the present studies on the cone system. aptation) as long as 7 in terms (3) and (4) is allowed his data imply that these adaptation mechanisms to be a constant much less than 1.0 (c was set to 1.0). operate prior to the saturating nontinearities. How- But if ;’ is less than 1.0 it implies that the steady- ever, Adelson found that the subtractive adaptation background quanta are less effective than effect has a very slow time course instead of the fast dashed-background quanta in driving the non- time courses we found in the cone system. Since linearity. This partial cancellation of the background Adetson used a tight-adaptation paradigm and the signals. which seems to be analogous to the back- present study a dark-adaptation paradigm. it is pos- ground cancellation effects observed in studies of sible that the two studies measured different aspects color contrast (Walraven, l976), could be accom- of essentially the same sort of mechanism. However. plished by subtractive inhibition before the non- as yet unpublished experiments in our laboratory linearity. In fact, allowing 7 to be a constant less than make this unlikely. In rod-isolation experiments anal- 1.0 is exactly equivalent to subtractive inhibition ogous to the present cone experiments. we found prior to the nonlinearity where the value being sub- strong multiplicative-adaptatjon effects, but almost tracted is a fixed fraction of the steady background negligible subtractive-inhibition effects. It will be intensity, I, (i.e. it is equivalent to a constant ratio of interesting to see if Adelson’s slow subtractive- excitation to inhibition across all steady-background inhibition effect occurs in the cone system. levels). Despite the above successes of the simple sub- A second piece of suggestive evidence is that tractive-inhibitjon model. it was soon realized that it subtractive inhibition models have been able to ac- makes some counter-intuitive predictions. These pre- count for many of the results from small perturbation dictions were tested in the experiment described studies (Kelly. 1971: Kelly and Savoie, 1978; Roufs, below and were found to be incorrect. 1972, 1974). Furthermore, the time courses of in-

hibition derived from these models are at least of the EXPERIMENT 3 same order of magnitude as the decay rate (of around 50-IOOmsec) implied from the present data (e.g. see We have seen that subtractive inhibition preceding Fig. 4f. It is impo~ant to note that in the small a nontinearjty is abie to account for the effects of gap perturbation studies the eye usually remains adapted duration on increment threshold if the intensity of the to the background level around which the per- background against which threshold is measured is turbations are occurring. Furthermore, the per- greater than or equal to that of the adapting back- turbations tend to be small and quick relative to I sec. ground. However. what does this model predict if the Thus. the experimental results they obtained at a background against which threshold is measured is given background level should only reflect those less intense than the adapting background. This mechanisms with fast enough time constants to oper- situation is depicted in the inset of Fig. 7. Suppose ate within the very early stages of light and dark first that the flashed background is kept off. adaptation. In other words. their paradigms would Following offset of the adapting background the not tap into the relatively stower multiplicative subtractive inhibitory signal will begin to decay. Since mechanisms. there is no background intensity the inhibitory signal If the transient threshold effects reported here are wilt, of course, greatly exceed the excitatory signal due to inhibition prior to the major nontinearities in from the background so that threshold will decline as the cone system, then these nonlinearities are unlikely gap duration is increased (i.e. over time. the in- to be in the photoreceptors. There is some support for hibition will release its suppression of the signal from this conclusion from measurements of the late recep- the test flash). Next, suppose that the Ilashed back- tor potential in macaque monkeys (Boynton and ground is set to a moderate intensity. As the sub- Whitten. 1970; Baron and Boynton, 1975; Valeton tractive-inhibitory signal decays. a point in time will and Noren, 1983). These studies find half-saturation be reached such that the excitatory signal from the constants in the dark-adapted retina (for fairly long background is just cancelled. Threshold should be at M~hanisms of visual sensitivity: back~ounds and early dark adaptation 1431

level for gap durations exceeding about IOOmsec 1 (perhaps a little later for the low flashed-background levels). Thus, the fast time-course of threshold change observed in the first experiment has been confirmed for a wider range of stimulus conditions. The lower- most curve, labelled - xc, shows the changes in absolute threshold during very early dark adaptation. These results are in basic agreement with the results obtained by Baker (1963) and others. In agreement with the subtractive-inhibition model, threshold rises with gap duration for intense flashed backgrounds and falls when the flashed background is kept off. However, the predictions regarding the intermediate flashed-background intensities were not upheld. For example, the curve obtained with the 2.64log td flashed background (solid triangles) should dip down to somewhere around 1.0 log td before rising to its asymptotic level.

CONCLUSION Of the six adaptation or inhibition mechanisms listed in the theory section, subtractive inhibition prior to a saturatoring nonIinearity is best able to

L account for the changes in increment thresholds 0 100 200 300 400 500 during very early dark adaptation. If to this is added GAP DURATION (MSEC) the multiplicative-adaptation mechanism to represent Fig. 7. Threshold as a function of gap duration with an the short- and long-term adaptation processes, and adapting background of 3.94 log td. (A) Subject W.G. (B) photopigment depletion, then the model is able subject D.U. Flashed background intensity in log td: to account for a wide range of increment-threshold @---.+ - cc; o-0, 1.78; A-A, 2.61; A-A, 3.65; data. However, there are some trouble spots, the m---m, 4.24. most serious being the inability of the model to predict increment thresholds during very early dark a minimum at this point. since beyond it the in- adaptation for flashed backgrounds less intense hibitory signal will not be strong enough to totally than the adapting background. Even though sub- cancel the background. Furthermore, this minimum tractive-inhibition mechanisms cannot account for all threshold should be the same as the final level reached the present results, it is quite possible that the in- (at IOO-200 msec) when the flashed background is hibitory mechanisms that were shown here to be kept off. operating during the first lOOmset of dark adaptation are the same inhibitory mechanisms that are Method revealed in temporal-modulation-threshold and The stimulus configuration and the procedure for 2-flash-threshold studies (Kelly, 197 1; Kelly and obtaining thresholds were the same as in the first Savoie, 1978; Roufs, 1972, 1974; Rashbass, 1970). experiment. Increment thresholds were measured as a function of gap duration for flashed backgrounds of Ackno~,~e~~~~enrs~harles Horton, Carrie Barris and - XV, 1.78, 2.64, 3.65, 4.24logtd. Following 12min David Ussery helped with data collection and analysis. Tim Fister did much of the computer programming. This re- of dark adaptation the subject adapted to a 3.94 log search was supported by NIH grant EY02688. td background for several minutes. In each session thresholds were measured at all gap durations for one flashed-background intensity level. The flashed back- REFERENCES grounds were then presented for 500 msec once every 10 sec. The increment flash was presented for 50 msec Adelson E. A. (1982) Saturation and adaptation in the rod and its onset was always simultaneous with that of system. Vision Res. 22, 1299-I 3 12. Baker H. D. (1963) Initial stages of light and dark adapta- the flashed background. As shown in the inset of Fig. tion. J. opl. Sot. Am. 53, 98-103. 7, the adapting background was always switched Barlow H. B. (1958) Intrinsic noise of cones. In Visual back on at the offset of the flashed background. Problems of Co&r, Vol. 2, pp. 617-630. H.M.S.O., London. Results and discussion Baron W. S. and Boynton R. M. (1974) The primate fovea1 local electroretinogram: an indicator of photoreceptor The results for two subjects are shown in Fig. 7. activity. Vkion Res. 14, 495-50 I. First, note that all the curves reach an asymptotic Baylor D. A. and Hodgkin A. L. (1974) Changes in time 1432 WIt.$QONS. &3SLEK

scale and sensitivity in turtle photoreceptors. j. Pll,rsiol. Hood D. C.. flves T.. Maurer E.. ~ndell B. and Buck- 242, 729-758. ingham E. i i47X) Human cone saturation as a function of Boynton R. M. and Whitten D. N. (1970) Visual adaptation ambient intensity: a test of modelsof shifts in the dynamic in monkey cones: recordings of late receptor potentials. range. l’rsiorr Rcs. 19, Y83-993. S&-we 170, I4?3- 1426. Ikeda M. and Boynton R. M. (1965) Negative Ilashes. Cornsweet T. N. and Pinsker H. M. (1965) Luminance positive flashes. and flicker examined by increment djs~~minatjon of brief flashes under various conditions of threshold technique. 3. r>nz.Strc,. .&,I. S5, 560-566. adaptation, J. Ph.wittl. 176, 294-3 IO. Kelly D. H. (1961) Visual responses to time-dependent Dawis S. M. and Purple. R. I_. (1980) Asperate PI11 stimuli. I. Amplitude sensitivity measurements. J. t~pr. evidence for optimal intensity discrimination. Inrwr. SOCK.A,,? I_, 51 4”4_79.I- Op~f~iu~. tisuaf Sci. Suppi. 280. Kefty D. H. (1971) Theory of Bicker and tramlent responses. Dowiing J. E. (1963) Neural and photochemjcal mech- I. tiniform tields. f. 0pr. Sot,. .4/n. 61, 537-546. anisms of visual adaptation in the rat. J. gert. PIirsiof. 46, Kelly D. H. (1972) Adaptation effects on spatl~)-temporal 1287-1301. sine-wave thresholds. I’i.s;cutRnv. 12, 8Y- 101. Dowling J. E. (1977) Receptoral and network mechanisms Kelly D. H. and Savoie R. E. t 1978) Theory of lhckcr and of visual adaptation. .~e!(r[).~~je~?re.~Res. Pmg. Bull. 15, transient responses. III. An essential nonhnearit~. J. r>p/. 397-407. St><. An?. 68, 1481.-14YO. Dowling J. E. and Ripps H. (1971 f S-potentials in the skate Kiei!~schmidt J. and Dowhng J. E. (1975) Intra~eilular retina: intracellular recordings during light and dark recordings from Gecko photoreceptors during light and adaptation. J. gen. Ph,rsiol. !!& 163-189. dark adaptation, J. ,~t”. PIi~~.sio/.66. 617-648. Dowling f. E. and Ripps H. (1972) Adaptation in skate Mansfield R. J. W. (1976)Visual adaptation: retinal frans- photoreceptors. J. g&. Ph,mioi. 50, 69%719. duction. brightness and sensitivity. Miri!z Res. 16, Fain G. L. (1976) Sensitivitv of toad rods: deuendence on 679690. vvave-length and background illumination. 1. P/tmio(. Naka K. I. and Rushton W. A. Il. (lY66) Potentials from 261, 7i-iol. - luminosity units in the retina of tish (cyprmidae). Geisier W. S. { 1978af Effects of photopjgm~nt depletion on J, PIirsioj. 185, 587-SYY. brightness and threshold. l&&r Rps. 18, 269-278. Normann R. A. and Perlman I. (1939) The rlRcts of Geisler W. S. (1978b) Adaptation. afterimages and cone background illumination on the photoresponses of red saturation. Flision Rex 18. 279-289. and green cones. J. P/r,mio/. 286, 491-507. Geisler W. S. (1979) Evidence for the equivalent- Normann R. A. and Webiin F. S. (1974) Control of retinal background hypothesis in cones. Vi~siotr Rex 19, 799- sensitivity. I. Light and dark adaptation of vertebrate 805. rods and cones. 3. gerr. ~l~~,~~t~l.63, 37-61. Geisler W. S. (1980) Increment-threshold and detection Roufs J. A. J. (1972) Dynamic properties of vision. 1. latency in the human rod and cone systems. Vision Res. Experimental relationships between flicker and flash 20, 98 l-994. threshold. I’ision Rex 12, 24 l-278. Geisler W. S. 41981j Efl’eets of bleaching and backgrounds Roufs J. A. J. (1974) Dynamic properties of vision. IV. on the gash response of the cone system. J. Plz~&t., Land. Thresholds of decremental flashes. incremental Rashes, 312,413-434. and doubiets in relation to flicker fusion. Vision Rm. 14, Glantz R. M. (1971) Visual adaptation: a case of nonlinear 831-851. summation. L’ision Rrs. 12, 103-109. Thijssen J. M. and Vendrik A. J. H. (1971) Differential Grabowski S. R.. Pinto L. H. and Pak W. L. (1972) luminance sensitivitv of the human visual svstem. Percenr. Adaptation in retinal rods of axolotl: intracellular record: ings.. Scirnce 176, I 24@-1243 Valeton J. M. and Norren D. van ( 1983) Light adaptation Green D. G.. Dowiine J. E.. Steeel 1. M. and Rioos H. of primate cones: an analvsis based on extracellular data. Retinal mechanisms-of visual adaptation in the’ skate. In ‘Rcwrding und Anulwi.~ of Conr Re.~pon.ws in the Intuct J. pm Phjxol. 6.5, 483-502. Monkey Eye (Edited by Valeton T.). Omslay. Joop Stalk, Hollins M. and Alpern M. (1973) Dark adaptation and Utrerht. visual pigment regeneration in human cones. 2. EL???. Walraven .I. (1976) Discounting the background: the Piwsin(. 62, 430-k. missing link in the explanation of chromatic induction. Hood D. C. (1978) Psvchonhvsical and nhvsioloeical tests t:isinn Rex. 16. 289-295. of proposed physiofogic;l mechanisms of lighi adapta- Werblin F. S. and Copenhagen D. R. (1974) Controi of tion In 1‘krnd P.s~r~~~)~~~~,,~~~.~: frs ~~?~.s~#~~~j~ff~ Brrst retinal sensitivity. III. Lateral interactions at the inner (Edited by A~ln~ton J.. Krauskopf J. and Wooten B.). plexiform layer. J. ,cetr. ~~i~sjo/. 63, 88-l IO. Academic Press. New York.