Cerebral Cortex Advance Access originally published online on September 8, 2005
Cerebral Cortex 2006 16(6):888-895; doi:10.1093/cercor/bhj032
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
High Response Reliability of Neurons in Primary Visual Cortex (V1) of Alert, Trained Monkeys
1 Department of Biomedical Engineering, Technion, Israel Institute of Technology, Haifa 32000, Israel and 2 Department of Ophthalmology and School of Graduate Studies, Medical College of Georgia, Augusta, GA 30912-3402 and Schepens Eye Research Institute Boston, MA 02114, USA
Address correspondence to Moshe Gur, Department of Biomedical Engineering Technion, Israel Institute of Technology Haifa, 32000 Israel. Email: mogi{at}bm.technion.ac.il.
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The reliability of neuronal responses determines the resources needed to represent the external world and constrains the nature of the neural code. Studies of anesthetized animals have indicated that neuronal responses become progressively more variable as information travels from the retina to the cortex. These results have been interpreted to indicate that perception must be based on pooling across relatively large numbers of cells. However, we find that in alert monkeys, responses in primary visual cortex (V1) are as reliable as the inputs from the retina and the thalamus. Moreover, when the effects of fixational eye movements were minimized, response variability (variance/mean — Fano factor, FF) in all V1 layers was low. When presenting optimal stimuli, the median FF was 0.3. High variability, FF
1, was found only near threshold. Our results suggest that in natural vision, suprathreshold perception can be based on small numbers of optimally stimulated cells.
Key Words: behaving monkey • Fano factor • primary visual cortex • V1 layers • response variability
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
How reliably do sensory neurons respond to repeated input? Many authors have recognized that the answer to this question is of fundamental importance in determining the neuronal resources needed to represent the external world (Werner and Mountcastle, 1963
; Heggelund and Albus, 1978; Tolhurst et al., 1983; Bradley et al., 1987; Scobey and Gabor, 1989; Vogels et al., 1989; Movshon, 2000). Understanding the nature and origins of neuronal variability is a key component of relating activity of sensory neurons to the perceptual capacities of the organism (cf. Heggelund and Albus, 1978; Vogels, 1990; Shadlen et al., 1996; Purushotaman and Bradley, 2005). The variability of neuronal responses also constrains the nature of the neural code; information can be sparsely represented by a few spikes only if spikes are reliably generated when a stimulus is repeated (Mainen and Sejnowski, 1995; Gallant and Vinje, 2001).
Spike count variability is often characterized by the ratio of the variance to the mean (Fano factor; FF) where the variability expected of a Poisson process has an FF of 1. From studies of anesthetized animals, it has been concluded that the variability of neurons in the visual pathway increases with each synaptic relay as information passes from the retina through the lateral geniculate nucleus (LGN) to the visual cortex (Kara et al., 2000; Demb et al., 2004). Since membrane potential fluctuations are similar in subcortical and cortical cells (Carandini, 2004; Demb et al., 2004), the greater variability of cortical neurons has been puzzling, and it has generated explanations ranging from shorter refractory periods for cortical neurons (Kara et al., 2000) to a nonlinear transformation between generator potential and firing rate of the neurons (Carandini, 2004).
From a theoretical perspective, the conviction that cortical neurons are highly variable has had a profound influence on thinking about mechanisms of behavior, including cognitive functions of humans (Glimcher, 2005). In addition, the observed unreliability of individual neurons has led to various schemes of population coding (Tolhurst et al., 1983; Vogels, 1990; Shadlen et al., 1996) and to suggestions why variability may be beneficial (Anderson et al., 2000; Carandini, 2004). Biophysical models derived from noncortical neurons that predict regular firing have had to be modified to account for the high variability found in cortical neurons (Softky and Koch, 1993).
There are conflicting reports concerning the issue of variability throughout the cell's response range; Tolhurst et al. (1983) reported that the ratio of variance to the mean (FF) was constant throughout the response range in both monkeys and cats, while Kara et al. (2000) showed that the FF decreased at higher firing rates. Differing from the above two studies, Carandini (2004) found that variability can be a complex function of firing rate, with low variability near threshold and in some cases at high response rates as well.
One factor that complicates measures of response variability in anesthetized animals is uncontrolled fluctuations in responsiveness and spontaneous activity (Tolhurst et al., 1983; Villeneuve and Casanova, 2003). These fluctuations may become more evident at more central stages of processing, leading to inflated estimates that do not represent the real reliability of the alert, perceiving brain. In fact, the lowest variability reported for V1 neurons in anesthetized cats (mean FF = 0.55) has been attributed to sampling from layer 4, where cells are receiving a strong, reliable driving input from the thalamus (Kara et al., 2000; Movshon, 2000).
We have previously reported that responses in V1 of alert, trained monkeys are very reliable, as long as the effects of eye movements are taken into account (Gur et al., 1997). With optimal stimuli that elicited strong responses, we obtained FF values of 0.24 (intercept in Fig. 3 of that paper), which is the lowest value reported for V1 cells. However, it has been suggested that this low value is due to sampling many cells from layer 4 (Kara et al., 2000; Movshon, 2000), and hence might not be representative of further processing stages in the cortex.
|
Given the fundamental importance of this issue, we have conducted a new series of experiments to determine the reliability of V1 neurons in alert monkeys. We have separately analyzed responses from each of the cortical layers to show that high reliability is not restricted to the input layers. We also measured reliability as a function of response strength and found that responses are most reliable when they are robust and clearly above threshold. We conclude that in real life, V1 neurons are reliably processing information needed by the cortical network for perception and action.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Data were collected from six adult female monkeys (two Macaca fascicularis and four M. mulatta). Monkeys were trained to fixate on a light-emitting diode (LED) for water reward. Once the monkey learned the task, a head-holding post and a recording well were implanted under deep anesthesia. All procedures complied with National Institutes of Health guidelines and were approved by the Animal Care and Use Committees of the Schepens Eye Research Institute and the Medical College of Georgia.
Nerve-spike and Eye-movement Recording
Fiber electrodes made from quartz-insulated platinum-tungsten alloy (Eckhorn and Thomas, 1993) with bare tip lengths 
5 µm and impedance at 1 kHz of 3–4 M
were most frequently used to record single-unit activity. In some experiments, glass-insulated platinum–iridium electrodes (Snodderly, 1973) with a tip diameter of 1–1.5 µm, and bare tip length of 5–7 µm, were used. In the initial experiments, the position of the dominant eye was monitored by a double Purkinje image eye tracker (2–3 minarc resolution; 100 Hz sampling rate) and recorded in a computer file, together with spike arrival times (0.1 ms time resolution) and spike shapes collected at 10–25 kHz (Gur et al., 1999). In recent experiments, eye position was recorded with an implanted scleral coil (1–2 minarc resolution; 200 Hz sampling rate) (Robinson, 1963; Judge et al., 1980). The trial started when the monkey correctly pressed the lever in response to the LED and continued for 5 s provided that the gaze remained within a predefined fixation window, between ±0.5° and ±1.5°.
Stimulus Presentation
Bar stimuli were initially displayed on a Barco 7351 monitor at a 60 Hz noninterlaced frame rate, with a Truevision ATVista video graphics adapter. In recent experiments, stimuli were displayed on a Sony 500 PS monitor at a 160 Hz noninterlaced frame rate with a Cambridge Research Systems VSG2/3F video board. Bars were optimized for orientation, length, velocity and color (green or red), 0.9 or 1 log units brighter or darker than the background of 1 or 5 cd/m2. Incremental (bright) bars were presented on a neutral gray background; decremental (dark) bars were presented on a background of a single color (Snodderly and Gur, 1995). After the ocular dominance was established, stimuli were viewed binocularly, unless responses during monocular viewing were substantially stronger. The eye position signal was added to the stimulus position signal at the beginning of each video frame (Gur and Snodderly, 1987, 1997a,b; Snodderly and Gur, 1995; Snodderly et al., 2001; Kagan et al., 2002; Gur et al., 2005). This was done to compensate for changes in eye position during intersaccadic intervals. Note that the maximum delay between shifts in eye position and subsequent corrections could be as long as 28 ms for the 60 Hz frame rate, and 10 ms at the 160 Hz frame rate; thus this procedure was not intended to compensate for the fast saccadic eye movements. Saccades were automatically detected using a velocity threshold of 10°/s (Snodderly et al., 2001; Fig. 1) and epochs where saccades occurred <100 ms before stimulus onset were excluded during data analysis so all data were collected only during inter-saccadic, drift periods. Note that changes in eye position during drifts were typically <10'/s (Fig. 1A) so that even the largest delay (28 ms) in eye position compensation would have caused an insignificant position error (<0.3'). This procedure enabled us to accurately map receptive fields (Snodderly and Gur, 1995; Kagan et al., 2002; Gur et al., 2005) and to produce consistent measures of neuronal responses (Gur et al., 1997) in alert monkeys.
|
Assignment to Layers
From our total sample of 118 cells it was possible to assign 83 cells to the various V1 laminae. In chronic recordings such as these it is not possible to use electrolytic lesions as anatomical markers since such lesions are only detectable for a few days. As detailed in Gur et al. (2005) and Snodderly and Gur (1995), we used information from several sources to locate the layer of origin: (i) distance from the cortical surface; (ii) physiological properties — an alternating sequence of layers with high ongoing multiunit background activity (4A, 4C and 6), and layers with very low background activity, (2/3, 4B and 5), in addition to the clusters of direction-selective cells found in layer 4B (Livingstone and Hubel, 1984; Orban et al., 1986; Hawken et al., 1988); and (iii) dye marking, anatomically indicating the depth of the electrode penetration. It is important to note that most of our penetrations were nearly normal to the cortical surface and they usually were not more than 1.5–2 mm long. It is easier to recognize the transitions between layers with these short, perpendicular penetrations than in longer tangential ones. Also, the fact that the distances were short helped to minimize the effects of any distortions that might occur during tissue processing.
For 15 cells we had cortical depth, and physiological and anatomical data. These cells were assigned confidence level 1. For the rest of the cells we did not have dye marking. For 38 of these cells, the depth and physiological data could lead to only one interpretation — a particular layer; those cells were assigned confidence level 2. For the remaining 30 cells, the assigned layer was based on the most likely interpretation of the depth and physiological evidence; those cells were assigned confidence level 3. An important strength of our assignment scheme is that it correctly places recording sites with high spontaneous multiunit activity in anatomical locations with high cytochrome oxidase activity, as shown independently by three different laboratories (Livingstone and Hubel, 1984; DeYoe et al., 1995; Snodderly and Gur, 1995).
Data Analysis
The total count of spikes generated during one sweep of a bar across the RF was taken as a measure of the response strength. This measure allows for comparison with many previous studies, including those using alert monkeys (Tolhurst et al., 1983; Vogels et al., 1989; Vogels, 1990; Kara et al., 2000; Carandini, 2004; Demb et al., 2004). For each cell studied, at least six, and usually ten or more responses contributed to the calculation of response mean, response variance and variance/mean (FF; to minimize bias the sample variance, with n – 1 in the denominator, was used). For different cells, different sweep durations were used ranging from 33 to 250 ms and data were collected over periods ranging from 10 to 120 s.
Statistical Analyses
Correlations were calculated using the nonparametric Spearman r, since most variables were not normally distributed. Pairwise comparisons utilized the Mann–Whitney U-test or the Wilcoxon matched-pairs signed-ranks test. Values reported for individual parameters are medians and the associated interquartile ranges (IQR) unless otherwise stated.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Data Collected
Responses of V1 single cells to repeated optimal stimuli were measured for 118 cells. Receptive fields were located at eccentricities of 4.1 ± 1.9° (mean ± SD). Most cells (83/118) were assigned to specific cortical layers. In 102/118 cells the stimulus was an optimally oriented sweeping bar; flashing bars, centered on the receptive field, were used in 16/118 cells. For 64/118 cells, we also measured response variability to non-optimal stimuli.
Effects of Eye Movements
We have previously shown that fixational eye movements influence neural responses (Gur and Snodderly, 1987, 1997a; Snodderly and Gur, 1995; Snodderly et al., 2001) by moving the receptive field with respect to the stimulus. Thus, to achieve repeated identical stimulation of the receptive field, it is essential to minimize the effects of fixational eye movements (Gur et al., 1997) as illustrated in Figure 1. Experimental records from three behavioral fixation trials with different interactions between eye movements and receptive fields are shown. The stimulus is repeated multiple times during each 5 s trial. In Figure 1A, the stimulus sweeps forward and back across the receptive field. In Figure 1B,C the stimulus is stabilized on the receptive field and turned on and off many times in each trial. In all cases, the monkey is attempting to maintain steady fixation. The fixational eye movements during the trials consist of slow drifts interposed with small (<30 min) fixational saccades and occasional blinks (Fig. 1C, arrows). Next to the trial displays are raster plots showing spike occurrence times for each stimulus presentation.
We compensate for the position offsets from one trial to another and the slow drifts during the trial by feeding the eye position signal to the stimulus pattern generator as described in the Materials and Methods. Because of inherent delays in the system, we can not completely compensate for saccades, so their impact is minimized by excluding epochs affected by saccades as shown in Figure 1A. Only some segments (rectangles) were selected for analysis while others were excluded where saccades occurred near (<100 ms) response onset. After exclusion of segments affected by fixational saccades, responses were quite consistent, with a FF = 0.25. For final analyses, data were combined across trials to obtain sufficient repeats of identical stimuli as previously described (Snodderly and Gur, 1995).
There were cases where saccades had only a small impact on response variability. Figure 1B shows a trial recorded from a cell with a large receptive field where the stimulus was a bar about twice the width of the receptive field and flashing at 15 Hz. It is obvious that drifts and small saccades did not have much effect on response consistency. Response variability for the whole trial, while ignoring the effects of eye movements, was very low — FF = 0.18, quite close to response variability found after excluding responses immediately following saccades (FF = 0.15). For cells like this one, which was not very selective for orientation or size, it was possible to use a stimulus larger than the receptive field, so fixational eye movements only moved the receptive field within the stimulus, resulting in a fairly constant light flux within the receptive field.
The third cell (Fig. 1C) was recorded in a monkey that was able to maintain fixation with very few saccades. In the example shown, there were two blinks (arrows) and only two saccades during the 5 s trial. Since responses of this cell were very transient, and were not affected by either blinks or saccades, all responses in the trial were selected. Note that in addition to the low response variability (FF = 0.28) response latency was also very consistent (56.9 ± 2.3 ms, mean ± SD).
Response Reliability in Different V1 Layers
To check whether the low variability we have found in alert monkey V1 (Gur et al., 1997) is related to sampling cells from the thalamic input layers (Kara et al., 2000; Movshon, 2000), we compared the variability of cells located in different V1 layers. Eighty-three cells were assigned to layers following a procedure using three levels of confidence as described in the Materials and Methods. Since the pattern of responses for each of the three levels of confidence was very similar, all data were combined to compute median values for each layer. The counting windows were, with the exception of layer 4A, quite similar. Median values (ms) were: layer 2/3, 60; layer 4A, 32.5; layer 4B, 60; layer 4C, 75; layer 5, 67.5; layer 6, 92.5. Figure 2 shows the median and IQR of the FF for each layer. The median FF was quite similar across layers, and values for the main input layer 4C were not significantly different from FF values in other V1 layers (Mann–Whitney test). In fact, with the exception of layer 4A cells where the FF was significantly different (P < 0.01) from the FF in layers 2/3, and 5, FF values in other layers were not significantly different from each other. As can be surmised from the interquartile range bars, not only the median FF, but also the distribution of FF values was quite similar in all layers except 4A, where four of five cells had very low variability.
|
Response Variability for Optimal and Suboptimal Stimuli
There have been conflicting reports from experiments conducted with anesthetized animals whether FF increases (Carandini, 2004), decreases (Kara et al., 2000) or stays constant (Tolhurst et al., 1983) as response amplitude increases. To explore this issue we analyzed responses to optimal stimuli, to suboptimal stimuli and to near-threshold stimuli. For 64 cells we were able to record responses to optimal stimuli and to a range of suboptimal ones. In 44 cells, responses were recorded as a function of orientation; in 17 cells we changed contrast and in three cells the width of the stimulating bar was varied. The dependency of the FF on response strength was similar for the different stimulus conditions so results were combined. Figure 3 shows experimental records from two 5 s fixation trials. The cell was stimulated by an optimally oriented sweeping bar (Fig. 3A) and by a bar 60°-away-from-optimum (Fig. 3B). The robust responses evoked by the optimally oriented bar were quite consistent (FF = 0.26) while the near-threshold responses evoked by the non-optimal bar were highly variable (FF = 1.6). The trials presented in Figure 3 depict a rare occasion where eye position was not compensated for. Due to this monkey's exceptionally stable fixation we were able to select responses in all segments. Those responses are shown in the raster plots next to the trial displays. It is interesting to note that since the lower-left to upper-right drift (Fig. 3A) was not compensated for, response latency gradually increased (Fig. 3A, lower raster plot). When the position shift due to the drift was calculated off-line and response latency was corrected accordingly, response latency was quite similar in all segments (Fig. 3A, upper raster plot). In Figure 3B the difference between compensated and uncompensated eye position is seen in the last segment where a small saccade to the left and down has lengthened response latency (lower raster plot) compared with latency corrected for eye position (upper raster plot).
To compare the relationship of variance to stimulus amplitude across cells, suboptimal responses were normalized as fractions of maximal responses. We compared variability of responses at four levels of response strength: maximal response, 61–90%, 31–60% and <30% of maximal response (Fig. 4A). While there was a small increase in median FF as the response decreased from maximal to 61–90% of maximal (P < 0.05), there was a dramatic, >3-fold increase in FF as the response approached threshold. The difference in the FF between the maximal or near-maximal response groups and either of the two low-response groups was highly significant (P < 0.001). There was no significant difference between the two low-response groups.
|
In contrast to the strong reduction in the FF with response amplitude, there was no clear trend in the variance as a function of response amplitude (Fig. 4B). The median variance was quite similar for the four groups and only the two low-amplitude groups differed (P < 0.01). The fact that variance did not increase while response mean increased, explains the sharp decrease in FF as response amplitude increased.
Estimating ‘Best’ Reliability
The distribution of FF for our total sample for responses elicited by optimal stimuli is shown in Figure 5. The distribution is quite tight with 72% of the cells having FF < 0.4. We recognize, however, that these values still may not represent the best reliability that the cortex can achieve. Even after our best efforts there are residual movements of the receptive field relative to the stimulus due to calibration errors, system noise and small artifacts caused by muscular movements. We estimate that these errors may be about 3–4' and have found (unpublished results) that such small motions can modulate responses in many cells.
|
To determine whether some of the variability of our sample could be due to uncompensated receptive field movements, we analyzed separately the data from cells with different receptive field properties. Of the 27 cells with FF
0.2 (median = 0.15, IQR = 0.06; mean ± SD = 0.14 ± 0.04), 10 (37%) had receptive fields >55' and were not very selective for orientation. The stimulus was usually larger than the receptive field so the responses of these cells were not very sensitive to small receptive field movements generated by small position errors. This situation was illustrated in Figure 1B. In contrast to the large proportion of cells with large receptive fields among the highly reliable cells there were only 5/81 (6%) among the less reliable cells.
The remainder of the highly reliable cells (17/27) had very small receptive fields (median = 10', IQR = 8'; mean ± SD = 16.7 ± 12.5'), which allowed little time for the sweeping stimulus (used in 15/17 cells) to interact with the receptive field. Thus these responses were quite immune to small, relatively slow changes in eye position. In two cells, where flashed stimuli were used, responses were very transient (cf. Fig. 1C) leading, again, to short response durations that were less likely to be perturbed by position errors. To test this second explanation for high reliability, we compared, for all cells with receptive field width <55', the time during which 90% of the spikes were generated. There was a weak but significant correlation (r = 0.28; P < 0.02) between FF and response duration period. What is more telling, however, is the comparison between the most and least reliable cells; in the 17 cells with FF 0.2, response duration (median = 50, IQR= 25 ms; mean ± SD = 59.7 ± 36.1 ms) was practically half of that found for the 17 least reliable (FF = 0.5) cells (median = 100; IQR = 40 ms ; mean ± SD = 93.2 ± 36.2). These results are consistent with the possibility that residual errors in compensation for eye position contributed to the higher FF of the less reliable cells and that our values are a minimum estimate of the reliability that cortical cells can achieve.
FF and the Length of the Counting Interval
Because our stimulus duration varied from cell to cell, our counting interval varied between 10 and 200 ms. Although >50% of all intervals were within the relatively small range of 50–110 ms, it is appropriate to consider whether the length of the counting interval might have influenced our estimate of the FF (see Warzecha and Egelhaaf, 1999; Wiener et al., 2001; Middleton et al., 2003). Consideration of this issue is important also to justify comparison with published data where various durations of counting intervals have been used (e.g. Kara et al., 2001; Wiener et al., 2001; Demb et al., 2004).
It is possible that for spatially selective cells, the effects of residual position errors could lead to over-estimating the response variance in large counting intervals, as discussed above. To avoid this confounding factor, we analyzed the relation between the length of the counting interval and the FF for two categories of cells. Either they were spatially non-selective (n = 11; receptive field width, 37–140'; half bandwidth at 70% of peak, 62–90°) or they were spatially selective (n = 9, receptive field width 13–27'; half bandwidth at 70% of peak, 13–27') but generating very short responses (<50 ms) so that residual position errors were not likely to contribute much to response variance.
Two analyses were performed on data gathered with stimuli generating strong responses. For nine spatially non-selective cells that were stimulated by stimuli of both long and short duration, the FF was computed for long (median = 200 ms) and for short (median = 16 ms) counting intervals. Even though the counting intervals differed by almost an order of magnitude, the FFs were almost identical; the median FFs for the long and the short counting intervals were 0.133 and 0.141, respectively (not significantly different; P > 0.2, Mann–Whitney test).
In another analysis, for 13 cells (four spatially non-selective; nine spatially selective) a large counting interval (median = 500 ms) was generated by concatenating four or five consecutive responses and counting all spikes within this interval. The comparison short intervals (median = 40 ms) were simply one of the component responses, chosen at random, from the large intervals. As in the first analysis, the median FFs of the long and short intervals were not significantly different (0.37 and 0.31, respectively; P > 0.4).
We conclude that under our conditions, with strong responses, the FF is not critically dependent on the counting interval in the range that we have used.
Can Long Refractory Periods Account for the High Reliability of Cortical Neurons?
Because refractory periods tend to regularize neural responses (Teich et al., 1978), it has been proposed (Kara et al., 2000) that the higher variability usually found in V1 cells than in thalamic or retinal cells results from a shorter refractory period in cortical cells. If refractory period is an important determinant of reliability, we might expect to find a relatively long refractory period in alert monkeys that regularized responses by prohibiting short intervals. To examine this possibility, we looked at the relationship between inter-spike intervals (ISIs) and FF. Figure 6 shows that, for our total sample, there was no significant correlation between the median ISI and FF for responses generated by optimal stimuli (r = 0.09; P > 0.3).
|
For a more detailed look at the issue of absolute and relative refractory periods, we examined the ten cells in our sample with the shortest ISI (total median range: 1.4–3 ms) and four cells not included in the present sample that had even shorter ISIs (<1.1 ms). As an upper estimate of the absolute refractory period we took the shortest recorded interval, and as a measure of the relative refractory period we examined pairs of spikes for the shortest interval for which the amplitude of the second spike was reduced. We found that the shortest interval recorded was 0.8 ms and for intervals between 0.8 and 1 ms there was a reduction in the amplitude of the second spike. In no case was a reduction in amplitude seen when the interval was >1 ms. For the ten most rapidly firing cells in our present sample, all but two ISIs were >1.1 ms and >95% of all ISI were larger than 1.4 ms. Therefore, neither an absolute refractory period of 0.8 ms nor a relative refractory period of 1 ms could have affected the ISI distribution. For the rest of our sample, 108/118 cells, >95% of all ISIs were >2 ms. Thus we can safely conclude that the high response reliability of our sample cannot be explained by regularization by long refractory periods.
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
We have shown that in alert monkeys, V1 cells respond very consistently when stimulated by repeated identical inputs. This high reliability (median FF = 0.3) is much better than expected for a Poisson process (FF = 1). High reliability was not restricted to input layer 4C but was evident in all V1 layers and was comparable to the reliability of the LGN and the retina. We also showed that variability is not constant throughout the neuron's response range; around threshold, variability is that expected from a Poisson process (FF
1), but for suprathreshold responses to optimal stimuli, variability is low (cf. Kara et al., 2000; Warzecha et al., 2000). Conditions Leading to High Response Reliability
Cortical cells receive inputs from many sources and the distribution of post-synaptic potentials is Poisson-like (Anderson et al., 2000; Carandini, 2004) — similar to that found in retinal ganglion cells (Demb et al., 2004). We suggest that the key to transforming the stochastic synaptic activity into a consistent output is an effective stimulus-driven input that strongly depolarizes the cell's membrane so that the fluctuations in resting potential are much smaller than the stimulus-driven depolarization (cf. Demb et al., 2004). The other requirement is that the gain of the neurons should be reasonably stable so that identical inputs lead to similar responses.
Our data showing high reliability of V1 cells differ from results from other laboratories. To achieve this high reliability, we have established data collection and analysis procedures that minimize variations in stimulation of the receptive field when stimuli are repeated. We have shown (Snodderly et al., 2001) that fixational eye movements move the receptive field with respect to the stimulus and generate unpredictable responses. Thus, in the presence of these eye movements, repeated presentation of the same external stimulus is not sufficient to ensure identical receptive field stimulation. As demonstrated in Snodderly et al. (2001), keeping a tight fixation window, or having low variance of eye position, does not guarantee that fixational eye movements will have a small impact on neuronal responses. How cells are affected depends on the interaction between the spatio-temporal receptive field, stimulus spatiotemporal properties, and the speed, amplitude and direction of the eye movements. The effects are sufficiently large that minimizing the influence of fixational eye movements removes an important source of variability in alert monkeys (Gur et al., 1997).
In our experiments, the monkeys perform a fixation task that controls their behavioral state. The stability of the behavioral state presumably contributes to the low response variance in two ways. First, the ongoing spontaneous activity is stable and fluctuations in resting potential contributed by the network are similar from trial to trial so the ‘background input noise’ to the cell is relatively constant. Consistent with this idea, optical imaging studies show that fluctuations in ongoing activity in awake monkeys are reduced compared to fluctuations in anesthetized monkeys (A. Grinvald, personal communication). The second contributor to low variance is stability of the gain of the neuron, so that for a repeated input, the neuron gives a similar output. If these conditions are satisfied, we should expect the results we obtained, namely similar responses of retinal and LGN cells (Croner et al., 1993; Edwards et al., 1995; Demb et al., 2004) the variance of the response of V1 cells is relatively constant at different response levels, and the variability (FF) declines with stronger responses. These conditions probably are not satisfied in experiments with anesthetized animals, where fluctuations in state are very difficult to avoid. Changes in the level of anesthesia strongly affect cortical cells' responsiveness: for example, an increase in the level of Halothane from 0.3 to 0.5% resulted in a 2-fold reduction of responses of cat V1 neurons (Villeneuve and Casanova, 2003). This change in responsiveness presumably represents a change in the gain of the neurons. Thus it is to be expected that in anesthetized animals the variance in cortical cells' responses would increase at higher response levels, as is often reported (Tolhurst et al., 1983; Vogels, 1990; Edwards et al., 1995; Carandini, 2000). We also note that behavioral state is not controlled even in unanesthetized animals unless they are performing a specific and demanding task. Monkeys become drowsy in a boring laboratory environment, and their state fluctuates in ways that can influence neuronal responses (Snodderly and Gur, 1995).
Basing Behavioral Decisions on Neuronal Responses
Our perception of an unchanging scene is very consistent, even though cortical neurons are often reported to be very unreliable. The usual suggestion for bridging the gap between the performance of individual neurons and perception is that perception relies on pooled activity of several neurons to average out the noise present in individual responses. Estimates of the pool size range from several cells where no correlation between cells is assumed (Tolhurst et al., 1983) to 100 cells where a weak (0.2) correlation is assumed (Shadlen et al., 1996). These estimates are based on highly variable neuronal responses with FF 3 for the non-correlated estimate and 1.7 for the correlated one. However, given that the FF found for our data (0.3) is an order of magnitude lower than that used by Tolhurst et al. (1983), it is likely that for optimal stimuli, decisions could be based on responses of only one or two cells. The situation is different for threshold stimuli where the FF for our data is 1. There, a few cells are needed if responses are not correlated (Tolhurst et al., 1983) and many more (Shadlen et al., 1996) if there is even a small degree of correlation.
Implications for Natural Vision
In natural viewing, eye movements are not a source of noise but part of the mechanism inducing neural responses. When we inspect an object, eye movements cause cortical receptive fields to cross, or briefly land on object elements. During these brief epochs a few spikes are generated by those cells with receptive fields that match the spatial properties of the object elements and the dynamics of the retinal image motions (Snodderly et al., 2001). Our results show that the high reliability of cortical cells enables the spikes generated by responsive cells to be sufficient for reliable perception.
|
Acknowledgments |
|---|
Supported by NIH EY12243 and US-Israel BSF grant 2003252.
|
References |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Anderson JS, Lampl I, Gillespie DC, Ferster D (2000) The contribution of noise to contrast invariance of orientation tuning in cat visual cortex. Science 290:1968–1972.
Bradley A, Skottun BC, Ohzawa I, Sclar G, Freeman RD (1987) Visual orientation and spatial frequency discrimination: a comparison of single neurons and behavior. J Neurophysiol 57:755–772.
Carandini, M (2004) Amplification of trial-to-trial response variability by neurons in visual cortex. PLOS Biol 2:1483–1493.[ISI]
Croner LJ, Purpura K, Kaplan E (1993) Response variability in retinal ganglion cells of primates. Proc Natl Acad Sci USA 90:8128–8130.
Demb JB, Sterling P, Freed MA (2004) How retinal ganglion cells prevent synaptic noise from reaching the spike output. J Neurophysiol 92:2510–2519.
DeYoe EA, Trusk TC, Wong-Riley MTT (1995) Activity correlates of cytochrome oxidase-defined compartments in granular and supragranular layers of primary visual cortex of the macaque monkey. Vis Neurosci 12:629–639.[ISI][Medline]
Eckhorn R, Thomas U (1993) A new method for the insertion of multiple microprobes into neural and muscular tissue, including fiber electrodes, fine wires, needles and microsensors. J Neurosci Methods 49:175–179.[CrossRef][ISI][Medline]
Edwards DP, Purpura KP, Kaplan E (1995) Contrast sensitivity and spatial frequency response of primate cortical neurons in and around the cytochrome oxidase blobs. Vision Res 35:1501–1523.[CrossRef][ISI][Medline]
Gallant JL, Vinje WE (2001) Reverse physiology: predicting single spikes. Neuron 30:646–647.[CrossRef][ISI][Medline]
Glimcher PW (2005) Indeterminacy in brain and behavior. Annu Rev Psychol 56:25–56.[CrossRef][ISI][Medline]
Gur M, Snodderly DM (1987) Studying striate cortex neurons in behaving monkeys: benefits of image stabilization. Vision Res 27:2081–2087.[CrossRef][ISI][Medline]
Gur M, Snodderly DM (1997a) Visual receptive fields of neurons in primary visual cortex (V1) move in space with the eye movements of fixation. Vision Res 37:257–265.[CrossRef][ISI][Medline]
Gur M, Snodderly DM (1997b) A dissociation between brain activity and perception: chromatically opponent cortical neurons signal chromatic flicker that is not perceived. Vision Res 37:377–382.[CrossRef][ISI][Medline]
Gur M, Beylin A, Snodderly DM (1997) Response variability of neurons in primary visual cortex (V1) of alert monkeys. J Neurosci 17:2914–2920.
Gur M, Beylin A, Snodderly DM (1999) Physiological properties of macaque V1 neurons are correlated with extracellular spike amplitude, duration, and polarity. J Neurophysiol 82:1451–1464.
Gur M, Kagan I, Snodderly DM (2005) Orientation and direction selectivity of neurons in V1 of alert monkeys: functional relationships and laminar distributions. Cereb Cortex 15:1207–1221.
Hawken MJ, Parker AJ, Lund JS (1988) Laminar organization and contrast sensitivity of direction-selective cells in the striate cortex of the Old World monkey. J Neurosci 8:3541–3548.[Abstract]
Heggelund P, Albus K (1978) Response variability and orientation discrimination of single cells in striate cortex of cat. Exp Brain Res 32:197–211.[ISI][Medline]
Judge SJ, Richmond BJ, Chu FC (1980) Implantation of magnetic search coils for measurement of eye position: an improved method. Vision Res 20:535–538.[CrossRef][ISI][Medline]
Kagan I, Gur M, Snodderly DM (2002) Spatial organization of receptive fields of V1 neurons of alert monkeys: comparison with responses to gratings. J Neurophysiol 88:2557–2574.
Kara P, Reinagel P, Reid RC (2000) Low response variability in simultaneously recorded retinal, thalamic, and cortical cells. Neuron 27:635–646.[CrossRef][ISI][Medline]
Livingstone MS, Hubel DH (1984) Anatomy and physiology of a color system in the primate visual cortex. J Neurosci 4:309–356.[Abstract]
Mainen ZF, Sejnowski TJ (1995) Reliability of spike timing in neocortical neurons. Science 268:1503–1506.
Middleton JW, Chacron MJ, Linder B, Longtin A (2003) Firing statistics of a neuron model driven by long-range correlated noise. Phys Rev E 68:21920–21928.[CrossRef]
Movshon JA (2000) Reliability of neuronal responses. Neuron 27:1412–1414.
Orban GA, Kennedy H, Bullier J (1986) Velocity sensitivity and direction selectivity of neurons in areas V1 and V2 of the monkey: influence of eccentricity. J Neurophysiol 56:462–480.
Purushotaman G, Bradley DC (2005) Neural population code for fine perceptual decisions in area MT. Nat Neurosci 8:99–106.[CrossRef][ISI][Medline]
Robinson DA (1963) A method of measuring eye movements using a scleral search coil in a magnetic field. IEEE Trans Biomed Engng BME- 10:137–145.
Scobey RP, Gabor AJ (1989) Orientation discrimination sensitivity of single units in cat primary visual cortex. Exp Brain Res 77:398–406.[CrossRef][ISI][Medline]
Shadlen MN, Britten KH, Newsome WT, Movshon JA (1996) A computational analysis of the relationship between neuronal and behavioral responses to visual motion. J Neurosci 16:1486–1510.
Snodderly DM (1973) Extracellular single unit recording. In: Bioelectric recording techniques, Part A: Cellular processes and brain potentials (Thompson RF, Patterson MM, eds), pp 137–163. New York: Academic Press.
Snodderly DM, Gur M (1995) Organization of striate cortex (V1) of alert, trained monkeys (Macaca fascicularis): ongoing activity, stimulus selectivity, and widths of receptive field activating regions. J Neurophysiol 74:2100–2125.
Snodderly DM, Kagan I, Gur M (2001) Selective activation of visual cortex neurons by fixational eyemovements: implications for neural coding. Vis Neurosci 18:259–277.[CrossRef][ISI][Medline]
Softky WR, Koch C (1993) The highly irregular firing of cortical cells is inconsistent with temporal integration of random EPSPs. J Neurosci 13:334–350.[Abstract]
Teich MC, Matin L, Cantor BI (1978) Refractoriness in the maintained discharge of the cat's retinal ganglion cell. J Opt Soc Am 68:386–402.[Medline]
Tolhurst DJ, Movshon JA, Dean AF (1983) The statistical reliability of signals in single neurons in cat and monkey visual cortex. Vision Res 23:775–785.[CrossRef][ISI][Medline]
Villeneuve MY, Casanova C (2003) On the use of isoflurane versus halothane in the study of visual response properties in primary visual cortex. J Neurosci Methods 129:19–31.[CrossRef][ISI][Medline]
Vogels R (1990) Population coding of stimulus orientation by striate cortical cells. Biol Cybern 64:25–31.[CrossRef][ISI][Medline]
Vogels R, Spileers W, Orban GA (1989) The response variability of striate cortical neurons in the behaving monkey. Exp Brain Res 77:432–436.[CrossRef][ISI][Medline]
Warzecha A-K, Egelhaaf M (1999) Variability in spike trains during constant or dynamic stimulation. Science 283:1927–1930.
Warzecha A-K, Kretzberg J, Egelhaaf M (2000) Reliability of fly motion sensitive neurons depends on stimulus parameters. J Neurosci 20:8886–8896.
Weiner MC, Oram MW, Richmond BJ (2001) Consistency of encoding in monkey visual cortex. J Neurosci 21:8210–8221.
Werner G, Mountcastle VB (1963). The variability of central neural activity in a sensory system, and its implications for the central reflection of sensory events. J Neurophysiol 26:958–977.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  M. Gur and D. M. Snodderly Physiological differences between neurons in layer 2 and layer 3 of primary visual cortex (V1) of alert macaque monkeys J. Physiol., May 1, 2008; 586(9): 2293 - 2306. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. Gur and D. M. Snodderly Direction selectivity in V1 of alert monkeys: evidence for parallel pathways for motion processing J. Physiol., December 1, 2007; 585(2): 383 - 400. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. P. J. de Kock, R. M. Bruno, H. Spors, and B. Sakmann Layer- and cell-type-specific suprathreshold stimulus representation in rat primary somatosensory cortex J. Physiol., May 15, 2007; 581(1): 139 - 154. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. C. Muresan and C. Savin Resonance or Integration? Self-Sustained Dynamics and Excitability of Neural Microcircuits J Neurophysiol, March 1, 2007; 97(3): 1911 - 1930. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||