1. Neuroscience
Download icon

Optogenetically induced low-frequency correlations impair perception

  1. Anirvan Nandy  Is a corresponding author
  2. Jonathan J Nassi
  3. Monika P Jadi
  4. John Reynolds
  1. The Salk Institute for Biological Studies, United States
  2. Yale University, United States
Short Report
  • Cited 1
  • Views 2,057
  • Annotations
Cite this article as: eLife 2019;8:e35123 doi: 10.7554/eLife.35123

Abstract

Deployment of covert attention to a spatial location can cause large decreases in low-frequency correlated variability among neurons in macaque area V4 whose receptive-fields lie at the attended location. It has been estimated that this reduction accounts for a substantial fraction of the attention-mediated improvement in sensory processing. These estimates depend on assumptions about how population signals are decoded and the conclusion that correlated variability impairs perception, is purely hypothetical. Here we test this proposal directly by optogenetically inducing low-frequency fluctuations, to see if this interferes with performance in an attention-demanding task. We find that low-frequency optical stimulation of neurons in V4 elevates correlations among pairs of neurons and impairs the animal’s ability to make fine sensory discriminations. Stimulation at higher frequencies does not impair performance, despite comparable modulation of neuronal responses. These results support the hypothesis that attention-dependent reductions in correlated variability contribute to improved perception of attended stimuli.

https://doi.org/10.7554/eLife.35123.001

Introduction

Neurons exhibit responses that are highly variable (Shadlen and Newsome, 1998), with nearby neurons in the cortex exhibiting correlated variability in their spiking output (Cohen and Kohn, 2011; Smith and Kohn, 2008; Smith and Sommer, 2013; Zohary et al., 1994). It has been estimated on theoretical grounds that even weak correlations substantially reduce the information coding capacity of a population (Zohary et al., 1994). Spatial attention can reduce correlated variability (often referred to as noise correlations) among neurons in macaque visual area V4 (Cohen and Maunsell, 2009; Mitchell et al., 2009), an area that is strongly modulated by spatial attention (Reynolds and Chelazzi, 2004). Mitchell et al., 2009 found that this reduction is restricted to low frequencies below 10 Hz. These studies have estimated, on theoretical grounds, that the reduction in correlated variability accounts for a large fraction (about 80%) of the perceptual benefit due to attention. However, these estimates rely on specific assumptions about the relationship between noise and signal correlations, and thereby, on how population signals are read out in the brain (Abbott and Dayan, 1999; Averbeck et al., 2006; Moreno-Bote et al., 2014; Panzeri et al., 1999). Theoretical studies using heterogeneous tuning curves and optimal readout have concluded that correlated variability does not necessarily limit information (Ecker et al., 2011; Shamir and Sompolinsky, 2006). Consistent with the interpretation that attention-dependent reductions in correlated variability improve perception, a recent study of the effects of naturally occurring fluctuations in neural correlations found improved sensory discrimination when neurons in Area V4 were desynchronized (Beaman et al., 2017). However, other studies have posited that correlations themselves may be induced by fluctuations in attention (Goris et al., 2014), resulting in variation in response gain that is shared across neurons, and other experiments have shown that attention can, under some conditions, also increase correlated neural response variability (Ruff and Cohen, 2014). Taken together, these studies call into question the simple idea that attention reduces correlations so as to improve sensory discrimination. Importantly, all of these studies are correlative in nature. The causal role of correlated variability in perception has not been tested and thus the proposal that low-frequency correlated variability is detrimental for perception has remained purely hypothetical.

Here, we sought to directly test the effects of correlated variability on sensory discrimination by using optogenetic activation to induce correlations in Area V4 as monkeys performed an orientation discrimination task near perceptual threshold. We exploited the fact that attentional modulation of correlated variability is both spatially- and frequency selective: attention-dependent reductions in correlation are restricted to low frequencies (<10 Hz (Mitchell et al., 2009)). We reasoned that the correlations that impair perception may have an inherent time scale, with low- but not high-frequency correlations impairing perception. If so, we would predict that the effects of correlations on perception should be specific to this low-frequency range.

Results and discussion

We took advantage of a novel approach to primate optogenetics and electrophysiology (Nassi et al., 2015; Ruiz et al., 2013) in which the native dura mater is replaced by a silicone based artificial dura (Figure 1A, B). This approach provides an optically clear window into the awake-behaving primate brain and allows precise opto-electrophysiology. We injected a lenti-viral construct (lenti-CaMKIIa-C1V1E162T-ts-EYFP) to preferentially drive expression of the depolarizing opsin C1V1 in excitatory neurons (Yizhar et al., 2011) in a restricted portion (200-300 µm diameter) of dorsal V4 of two macaque monkeys (Figure 1C). Despite some heterogeneity in orientation tuning width at each injection site, overall there was similar tuning among neurons within a site (Figure 2—figure supplement 1).

Surface Optogenetics and electrophysiology through an artificial dura.

(A) Schematic of an artificial dura (AD) chamber. A portion of the native dura mater (red) is resected and replaced with a silicone based optically clear artificial dura (AD). The optical clarity of the AD allows precisely targeted injections of viral constructs and subsequent optical stimulation and electrophysiological recordings. (B) An AD chamber is shown over dorsal V4 in the right hemisphere of Monkey A. sts = superior temporal sulcus, lu = lunate sulcus, io = inferior occipital sulcus. Area V4 lies on the pre-lunate gyrus between the superior temporal and lunate sulci. Scale bar = 5 mm; M = medial, A = anterior (C) EYFP expression at the first injection site (lenti-CaMKIIα-C1V1-ts-EYFP) after 4 weeks.

https://doi.org/10.7554/eLife.35123.002

We trained two monkeys to perform an attention-demanding orientation-change detection task (Figure 2A). The monkeys were spatially cued to attend to one of two spatial locations. In the ‘attend in’ condition, the monkeys were instructed to covertly attend to a spatial location within the receptive fields of neurons at the viral injection site, while maintaining fixation at a central fixation point. In the ‘attend-away’ condition attention was directed to a location of equal eccentricity across the vertical meridian. On each trial, a sequence of oriented Gabor stimuli simultaneously flashed on and off at both spatial locations (200 ms on, variable 200–400 ms off). At an unpredictable time (minimum 1 s, maximum 5 s), one of the two stimuli (95% probability at cued location; 5% probability at uncued location, ‘foil trials’) briefly changed in orientation (200 ms) and the monkey was rewarded for making a saccade to the location of orientation change. If no change occurred within 5 s (‘catch trials’, 13% of trials), the monkey was rewarded for holding fixation. We controlled task difficulty by varying the degree of orientation change and thereby obtained behavioral performance curves (psychometric functions) for each recording session (Figure 2B, Figure 2—figure supplement 2A,B). Impaired performance (Figure 2—figure supplement 2A, left panel, square symbol) and slower reaction times (Figure 2—figure supplement 2A, right panel, square symbol) were observed for the foil trials, indicating that the monkey was indeed using the spatial cue in performing the task.

Figure 2 with 5 supplements see all
Optogenetically-induced low-frequency correlations cause a frequency- and spatially-selective impairment in an attention-demanding orientation discrimination task.

(A) Attention task: While the monkey maintained fixation, two oriented Gabor stimuli (schematized as oriented bars) flashed on and off simultaneously at two spatial locations: one at the RF of the opsin injection site, the other at a location of equal eccentricity across the vertical meridian. The monkey was cued to covertly attend to one of the two locations. At an unpredictable time, one of the two stimuli changed in orientation. The monkey was rewarded for making a saccade to the location of orientation change at either location (95% probability of change at cued location; 5% probability at un-cued location [foil trials]). If no change occurred (catch trials), the monkey was rewarded for maintaining fixation. On a random subset of trials, the opsin site was optically stimulated using a low-frequency (4-5 Hz) sinusoidally modulated laser light (λ=532nm). (B) Psychometric functions for an example behavioral session showing performance (hit rate) as a function of task difficulty (size of orientation change) for the baseline (no optical stimulation) condition in gray and low-frequency optical stimulation condition in blue. Top, monkey was instructed to attend to the site of optical stimulation; Bottom, monkey was instructed to attend to the contralateral hemifield. Error bars are std. dev. obtained by a jackknife procedure and corrected for the number of jackknives (20). The data has been fitted with a smooth logistic function. (C) The perceptual discrimination modulation index (PDMI; change in psychometric function threshold due to optical stimulation) in the low-frequency optical stimulation condition when the monkey was attending in to the site of optical stimulation across all behavioral sessions. The solid line represents the mean of the distribution. The PDMI distribution is significantly different from zero. (D–E) No significant change in PDMI either when the monkey was attending away from the site of optical stimulation (D) or due to high-frequency optical stimulation (E).

https://doi.org/10.7554/eLife.35123.003

To test if low-frequency correlations impair discrimination we optically stimulated neurons at the opsin injection site with 4-5Hz sinusoidally modulated low-power laser stimulation, on a randomly chosen subset of trials (‘low-frequency stimulation’ condition). Our goal was to induce correlations without significantly altering the mean firing rates by using low-power stimulation. Significant changes in mean firing rate could have unknown effects such as masking of the stimulus evoked response. Equating firing rates also avoids any indirect effects of mean firing rate changes on spike-count correlations (Cohen and Kohn, 2011). We find that low-frequency optical stimulation modulates the timing of the neural response (Figure 4D) but does not alter the overall magnitude of the population response (Figure 4A, B, C). We replicate previous findings that attention reduces low-frequency spike-count correlations in the baseline (no optical stimulation) condition (Mitchell et al., 2009) (Figure 3A, left panel; gray versus white bars, p=0.02, t-test). As predicted, low-frequency optical stimulation increases low-frequency correlations (Figure 3A, left panel; blue versus gray bar, p=0.045, t-test). The induced correlations were at a level comparable in strength to that observed when attention was directed away from the RF location in the baseline condition (Figure 3A, left panel; blue versus white bar). Optogenetic activation is accompanied by a period of reduced activity following stimulation. By careful titration of laser intensity (amplitude of sinusoidal modulation) we were able to alter the timing of spiking without altering mean firing rate. This is shown in Figure 4: we see a robust increase in firing rate due to attention in both the low-frequency stimulation and baseline conditions (Figure 4A, B), but there is no significant rate increase due to optical stimulation either during the pre-stimulus blank period (Figure 4C, left panel, p=0.49, t-test; Figure 4—figure supplement 1, top-left panel) or during the stimulus presentation period (Figure 4C, right panel, p>0.1, t-test; Figure 4—figure supplement 1, bottom-left panel). Rather, unit activity shows phase locking to optical stimulation (Figure 4D,E; pre-stimulus period). The distributions of spiking activity with respect to the phase of the optical stimulation show significant deviation from what would be expected from a null distribution (Figure 4D, example units, p0.01; Figure 4E, population, p<0.01, Rayleigh test). The null distributions were derived from a rate-matched Poisson process. We see a similar phase locking due to optical stimulation during the stimulus presentation period (Figure 4F, p<0.01, Rayleigh test), although the peak of the phase-lock distribution for the stimulus presentation period occurs earlier (around 120°) compared to that for the pre-stimulus period (around 210°). This would be expected, as the neurons are depolarized by the visual stimulus and hence more easily pushed to spiking threshold by optogenetic depolarization, as compared to when no stimulus is present. Thus, the physiology data shows that we successfully induced correlated activity among neurons at the opsin site without affecting the response rates.

Optical stimulation at low- and high-frequencies induces low- and high-frequency correlated activity.

(A) Consistent with earlier reports (Mitchell et al., 2009), attention reduces baseline spike-count correlations at low frequencies (200ms counting window, p=0.02; left panel, white versus gray bar) but not at high frequencies (50ms window; right panel, white versus gray bar). Low-frequency optical stimulation increases low-frequency correlations (p=0.045; left panel, gray versus blue bar) but not high-frequency correlations (p>0.1; right panel, gray versus blue bar). (B) High-frequency optical stimulation increases high-frequency correlations (p<0.05). n=79 pairs for baseline and low-frequency stimulation, n=27 pairs for high-frequency stimulation, collapsed across attention conditions. Mean +/- s.e.m. in all plots.

https://doi.org/10.7554/eLife.35123.009
Figure 4 with 5 supplements see all
Low-frequency stimulation induces phase-locking without increasing firing rates.

(A) Peri-stimulus time histograms (PSTH) of two example units for the different experimental conditions. Both units show a robust firing rate modulation due to attention (solid versus dashed lines) but no rate increase due to low-frequency optical stimulation (blue versus gray lines). Horizontal bars represent stimulus duration. (B) Population data showing the same rate increase due to attention, but no significant increase due to optical stimulation (n=94). Same convention as in (A). (C) Distribution of rate modulation indices for the low-frequency stimulation attend-in condition compared to the baseline attend-in condition for a 200ms pre-stimulus period (left panel) and 200ms stimulus presentation period (60-260ms after stimulus onset; right panel). The arrowheads depict the median of the distributions. Neither distribution is significantly different from zero (p>0.1). (D) Phase plots for two example units showing the distribution of spiking activity with respect to the phase of the optical stimulation, during the pre-stimulus period. In gray is the null distribution obtained from a rate-matched Poisson process. Both units show significant deviations from the null distribution (p0.01 for both, Rayleigh test), indicative of phase locking. (E) Population phase-locking plot illustrating the bias in spiking activity to the downswing of optical stimulation during the pre-stimulus period (n=68). Same convention as in D). The distribution of spiking phase is significantly different from null (p<0.01, Rayleigh test). (F) Same as in (E), but for the stimulus presentation period (n=74). The distribution of spiking phase is significantly different from null (p<0.01, Rayleigh test).

https://doi.org/10.7554/eLife.35123.010

Behaviorally, we find that low-frequency stimulation impairs the monkey’s ability to detect fine orientation changes, and does so only at the opsin location, as indicated by impairment in the attend-in condition (Figure 2B, upper panel; Figure 2—figure supplement 2B), not in the attend-away condition (Figure 2B, lower panel), in which the monkey discriminated orientation at the contralateral location. To quantify this behavioral deficit, we estimated the threshold of the monkeys’ psychometric functions and calculated the change in threshold due to optical stimulation as a modulation index (perceptual discrimination modulation index, PDMI; see Materials and methods). We find a significant increase in PDMI due to low-frequency stimulation in the attend-in condition (p=0.02, t-test; Figure 2C), indicating impaired detection of fine orientation changes. Unexpectedly, we also found a significant increase in slope (p=0.007, t-test; Figure 2—figure supplement 5A), suggesting that the shift from non-detection to detection occurs over a narrower range of orientations in the laser stimulation condition. In a large fraction of individual behavioral sessions (Figure 2—figure supplement 5B), both the changes in threshold and slope were significant (20/42 sessions). 11/42 sessions had significant threshold change only, while 5/42 sessions had significant slope change only.

The impairment due to optical stimulation is location specific: there was no significant change in PDMI on trials when the monkey was cued to detect the target at the unstimulated location (attend-away condition, Figure 2D). Importantly, the impairment is also frequency specific. When we stimulate the neurons with 20Hz sinusoidally modulated low-power laser stimulation (‘high-frequency stimulation’ condition), we observed no significant change in PDMI (Figure 2E), despite a significant increase in high-frequency spike-count correlations (Figure 3B) and phase locking comparable to the low-frequency stimulation condition (Figure 4—figure supplement 2). As in the low-frequency stimulation condition, we find no significant changes in mean firing rates with the high-frequency stimulation condition (Figure 4—figure supplement 1, right panels). The stimuli in each sequence were presented with irregular timing to ensure that any impairment did not stem from stimuli appearing at a particular phase of the laser stimulation, such as the phase at which neural sensitivity was at its nadir. To verify that such phase alignment did not nonetheless occur by chance, we measured the phase of target stimulus onset for both low- and high-frequency stimulation and found no phase preference (Figure 4—figure supplement 3A). Nor was behavioral performance phase dependent (Figure 4—figure supplement 3B). A two-way ANOVA of normalized performance with factors 'laser-phase' (two different phase bin arrangements are shown in Figure 4—figure supplement 3B) and 'delta-orientation' (the trial-by-trial difference between target and non-target orientation) revealed no significant main effect of laser phase (Flaser-phase=1.53, p=0.2, left bin arrangement in Figure 4—figure supplement 3BFlaser-phase=0.73, p=0.53, right bin arrangement in Figure 4—figure supplement 3B), a significant main effect of orientation and no significant interaction between the factors. A second potential concern is that the laser might impair orientation discrimination by distorting or flattening orientation tuning curves. We find that orientation tuning curves were not significantly altered by the laser at either frequency (Figure 4—figure supplement 4), at least in the range of orientations used in the experiment. A two-way ANOVA of normalized firing rates with factors 'laser-condition' (low-frequency laser, no laser) and 'orientation' revealed no significant main effect of laser condition (Flaser-condition=0.8, p=0.37), a significant main effect of orientation and no significant interaction between the two factors. Similarly, we found no significant main effect of the high-frequency laser condition (Flaser-condition=0.05, p=0.82), a significant main effect of orientation and no significant interaction between the two factors. A third potential concern is that the rhythmic laser stimulation might cause a sort of frequency-dependent adaptation that would cause neurons to be less sensitive to visual stimuli presented at a similar frequency. If so, the low-frequency (4-5Hz) laser stimulation could reduce the responses evoked by 200ms visual stimuli, impairing the monkey’s ability to discriminate the stimuli, while the high-frequency laser stimulation might not cause this effect, explaining the observed impairment. To test this, we measured the firing rates evoked by the first four non-target stimulus flashes, on no-laser, low-frequency laser and high-frequency laser trials, in the 'attend in' condition (Figure 4—figure supplement 5). Though the first stimulus in the sequence evoked a stronger response than the subsequent stimuli (reflecting a form of visual stimulus-driven adaptation), we find no evidence that the addition of the laser at either frequency caused a change in mean firing rate. A two-way ANOVA of normalized firing rate with factors 'flash-position' (1,2,3 or 4) and 'laser condition' (low-frequency laser, high-frequency laser, no laser) revealed a significant main effect of flash position (Fflash-position=3.33, p=0.02), no significant main effect of laser condition (Flaser-condition=0.65, p=0.52) and no significant interaction between the two factors. For the small number of sessions (n=15) over which we could do this analysis, the PDMI trends toward significance for the low-frequency condition (p=0.07), but is highly non-significant for the high-frequency condition (p=0.7). Additionally, we did not find significant changes in false-alarm rates with either low- or high-frequency stimulation (p>0.1, t-test; Figure 2—figure supplement 2E). This was true for false alarms made during catch trials as well as on non-catch trials. Thus, we find no evidence that laser stimulation caused our subjects to mis-perceive a non-target as a target.

V4 has patchy organization for orientation tuning, so simultaneously recorded neurons tended to prefer similar orientations (see Figure 2—figure supplement 1). Under these conditions, where signal correlations are positive, positive noise correlations should reduce discriminability (Averbeck et al., 2006) by increasing the overlap between neural responses evoked by discriminanda. In the present experiment, where the task was to discriminate target from non-target, this would predict that in sessions where we observed laser-induced perceptual impairment, we should observe laser-induced decreases in discriminability at the level of pairs or populations of neurons, especially on miss trials, when the monkey was unable to discriminate target from non-target. For each recording session, we calculated a neural measure of discriminability between non-target and target stimuli across simultaneously recorded neurons (neural discriminability modulation index, NDMI; Figure 5; see Materials and methods, (Cohen and Maunsell, 2010)). We then examined whether laser-induced increases in perceptual threshold were correlated with laser-induced reductions in neural discriminability on trials in which the target appeared at the opsin location (the 'attend in' condition). Figure 5B shows NDMI for each experimental session (calculated from miss trials where the animal failed to detect a target) against the corresponding PDMI. We examined this in two ways: by measuring the correlation in a N-dimensional space where N is the number of neurons recorded in a given session (left panel) or by measuring the average NDMI across neural pairs (right panel). As predicted, there is a strong and significant negative correlation in both analyses (Figure 5Bρ=-0.29 p=0.05, NDMI from all simultaneous neurons, left panel; ρ=-0.42 p=0.01, average NDMI across all simultaneous pairs, right panel; robust correlation). In both analyses, the sessions in which low-frequency laser-induced correlations caused the strongest perceptual suppression all showed negative NDMIs. In other words, laser-induced reductions in discriminability at the neural level corresponded to increased perceptual thresholds. There was no significant correlation between NDMI and changes in the slope of the psychometric function. NDMI calculated from hit trials had no correlation with threshold or slope changes.

Optical modulation of neural discriminability correlates with behavioral perturbations.

(A) Schematic of neural discriminability analysis. The responses of a hypothetical set of 3 neurons to target (green) and non-target (magenta) stimuli are depicted as point clouds. Each dot represents a stimulus presentation. The discriminability (D) between the two response categories is defined as the Mahalanobis distance between the centroid of the target responses and the non-target point cloud. (B) Neural Discriminability Modulation Index (NDMI) due to optical stimulation is plotted against the corresponding PDMI (behavioral threshold change) for each experimental session. NDMI is calculated either from multi-dimensional clouds from all simultaneously recorded neurons (left panel; n = 42 sessions) or as the average of two-dimensional clouds from all pairs of simultaneously recorded neurons (right panel; n = 35 sessions). NDMI is negatively correlated with PDMI. Since both NDMI and PDMI are dependent measures, the data were fitted with a line whose slope was obtained from a Model II regression.

https://doi.org/10.7554/eLife.35123.016

To confirm whether it is possible to induce coherent activity in a neuronal ensemble due to sub-threshold rhythmic stimulation, we examined the consequences of such stimulation on a conductance-based model of excitatory and inhibitory neurons (Figure 6A; see Materials and methods). We calculated the strength of coherent activity in the network (spike-spike coherence, SSC) both with and without sub-threshold stimulation (Figure 6B,C). We quantified the change in coherence due to stimulation as a modulation index (SSC MI; Figure 6D). We find that it is indeed possible to induce coherent activity in the network at a desired frequency (Figure 6D, Figure 6—figure supplement 1D) and that this induction is robust to a wide range of network (Figure 6—figure supplement 1C, Figure 6—figure supplement 2A) and stimulation parameters (Figure 6—figure supplement 2B).

Figure 6 with 2 supplements see all
Low-frequency sub-threshold stimulation induces coherent activity in a computational model of E-I neurons.

(A) Schematic of a local conductance-based E-I network with mutually coupled excitatory (E) and inhibitory (I) units. A fraction (50%) of the E units are sensitive to 'optical' stimulation. Wee, self-excitation among E units;Wii, self-inhibition among I units;Wie, excitation provided by E to I; Wei, inhibition provided by I to E. (B) Simulation of a network of 800 E and 200 I units (Wee=16, Wii=-1, Wie=4, Wei=-18). The raster plot shows the activity of all units in the model (blue, I; green, E without opsin; magenta, E with opsin) to a step input (Ie,Ii) and 4Hz sinusoidal optical stimulation (Iopto). (C) Population spiking rate averaged across 1000 simulations of the scenario in (B) with and without optical stimulation. (blue, I; orange, all E; green, E without opsin; magenta, E with opsin. solid lines, with optical stimulation; dashed lines, without optical stimulation) (D) Spike- spike coherence (SSC) among E units was calculated for the two conditions with and without optical stimulation and the change in SSC across the two conditions was calculated as a modulation index (SSC MI). SSC MI exhibits a peak at 4Hz due to optical stimulation.

https://doi.org/10.7554/eLife.35123.017

The location specificity of the impairment also suggests that the impairment is not due to a phosphene effect (Jazayeri et al., 2012). If attention were drawn away from the unstimulated location by a phosphene we would expect impaired performance in the attend-away condition. We also verified that the impairment was not due to a thermal effect by stimulating a location in the chamber a few millimeters from the opsin site (Figure 2—figure supplement 4A) and not due to visual distractions caused by the laser light by stimulating outside the brain (Figure 2—figure supplement 4B). In both cases, we did not observe any changes in behavior.

One of the two animals used in this study was euthanized to verify that, as we have previously found in macaque V1 (Nassi et al., 2015), lenti/CAMKII leads to selective expression in pyramidal neurons. The dura mater adhered to V4 bilaterally and we were unable to perform histology. The second animal (Monkey A) is currently in use in another study. Thus, we do not have a measure of lenti/CAMKII selectivity in macaque V4. However, it is reasonable to assume that the opsin was strongly biased toward pyramidal neurons. The viral constructs and injection protocol used in the present study were identical to those used previously in V1 (Nassi et al., 2015). In that study, lenti/CAMKII expressing C1V1/EYFP led to expression that was highly selective for excitatory neurons (Figure 1B of Nassi et al., 2015): of 119 neurons imaged in five different fields of view, only 2 (1.7%) were double labeled for both EYFP and parvalbumin/calretinin/calbindin, indicating that expression was heavily biased toward excitatory neurons. CaMKII has also been used in other macaque brain areas including perirhinal cortex (Tamura et al., 2017), where it also led to opsin expression primarily in excitatory neurons. It is possible that lenti/CAMKII leads to less selective expression in V4 pyramidal neurons, but whatever the degree of selectivity we were able to activate neuronal responses in a phase-dependent manner with low intensity laser stimulation. as needed to test the effect of correlated variability on perceptual discrimination in the present study.

Our results establish the first causal link between correlated variability and perception. The optogenetic stimulation protocol in our study, using sinusoidal modulation of laser irradiance, induces the kind of correlations in a local population of the cortex that might not be physiologically realistic. It nevertheless establishes the causal relevance of low-frequency correlated variability in perception and supports the hypothesis that attention-dependent reductions in correlated variability enhance perception. Recently, studies have theorized that only certain correlations – those that are indistinguishable from stimulus-induced correlations – are information limiting (Moreno-Bote et al., 2014). We speculate that the correlations induced in our study included such information-limiting correlations, resulting in the observed impairment. The timescale of these low-frequency correlations is consistent with inter-saccadic intervals (200–300 ms), which may be a relevant timeframe for gathering visual information (Yarbus et al., 1967). Decreases in correlated variability at this timescale could therefore be critical for perception. Our study paves the way for investigating the laminar and cell-class specific components of the cortical circuit that determine this critical component of perception.

Materials and methods

Surgical procedures

Request a detailed protocol

Surgical procedures have been described in detail previously (Nandy et al., 2017; Nassi et al., 2015; Ruiz et al., 2013). In brief, an MRI compatible low-profile titanium chamber was placed over the pre-lunate gyrus, on the basis of preoperative MRI imaging in two rhesus macaques (right hemisphere in Monkey A, left hemisphere in Monkey C). The native dura mater was then removed and a silicone based optically clear artificial dura (AD) was inserted, resulting in an optical window over dorsal V4 (Figure 1A,B). All procedures were approved by the Institutional Animal Care and Use Committee and conformed to NIH guidelines.

Viral injections

Request a detailed protocol

Viral injection procedures have been described in detail previously (Nassi et al., 2015). In brief, we injected a VSVg-pseudotyped lentivirus carrying the C1V1-EYFP gene behind the 1.3kb CaMKIIα promoter (lenti-CaMKIIa-C1V1E162T-ts-EYFP; titer = 3×1010 TU/ml) at 2 cortical sites in monkey A and 1 cortical site in monkey C while they were anesthetized and secured in a stereotactic frame. The viral constructs were chosen to preferentially drive expression of the depolarizing opsin C1V1 in excitatory neurons local to the injection site (Han et al., 2009). We injected approximately 0.5μl of virus at each depth in 200μm increments across the full 2mm thickness of cortex. All injections were targeted to para-foveal regions of V4 with eccentricities between 5 and 8 degrees of visual angle. Expression of the fluorescently tagged opsin was confirmed using epifluorescence goggles (BLS Ltd., Budapest, Hungary) after about 4-6 weeks of viral injection (Figure 1C).

Opto-Electrophysiology

Request a detailed protocol

At the beginning of each recording session, a plastic insert, with an opening for targeting electrodes and for optical stimulation, was lowered into the chamber and secured. This served to stabilize the site against cardiac and respiratory pulsations. The opening was centered at the site of viral injection. A single tungsten microelectrode (FHC Inc) was mounted on an adjustable X-Y stage attached to the recording chamber and advanced into the injection site using a micromanipulator (Narishige Inc) until a spike (single neuron or multi-unit) could be reliably isolated from background voltage fluctuations. Site targeting was done under microscopic guidance (Zeiss Inc) using the microvasculature as reference. A single optical fiber (600 μm multimode fiber, 0.37NA, Thorlabs Inc) was mounted on the same X-Y stage and positioned over the injection site perpendicular to the calvarium. The microelectrode was positioned at an angle of 20 degrees with respect to the optical fiber (see schematic in Figure 1A).

We used a 532 nm diode-pumped solid-state (DPSS) laser (OEM Laser Systems Inc) as the light source for optical stimulation. The laser was placed on an optical breadboard in-line with a Uniblitz mechanical shutter (Vincent Associates), electro-optical modulator (‘EOM’, ConOptics Inc) and an optical fiber collimator/coupler (Thorlabs, Inc) attached to the optical fiber. A beam-splitter between the EOM and collimator directed approximately 1% of the light toward a high-speed photo-detector (Thorlabs, Inc). The EOM allowed us to control the intensity of laser light entering the fiber and was controlled using custom-written Labview software and a National Instruments digital acquisition board. Before each experiment we calibrated the output of the high-speed photodetector to the full range of intensities (irradiance units) measured at the fiber tip using an integrating sphere photodiode power sensor and a digital power meter (Thorlabs, Inc). This enabled real-time, calibrated irradiance measurements during all experiments.

Neuronal signals were recorded extracellularly, filtered, and stored using the Multichannel Acquisition Processor system (Plexon Inc). Neuronal signals were classified as either multi-unit clusters or isolated single units using Plexon Offline Sorter software. Single units were identified based on two criteria: (a) if they formed an identifiable cluster, separate from noise and other units, when projected into the principal components of waveforms recorded on that electrode and (b) if the inter-spike interval (ISI) distribution had a well defined refractory period.

Data was collected over 42 sessions (24 sessions in Monkey A, 18 in Monkey C), yielding a total of 94 units. Frequently, multiple units could be identified while recording from the single tungsten electrodes. Data was collected over an additional three sessions for control analyses (Figure 2—figure supplement 4).

Task and stimuli

Stimuli were presented on a computer monitor placed 57 cm from the eye. Eye position was continuously monitored with an infrared eye tracking system (ISCAN ETL-200). Trials were aborted if eye position deviated more that 1° (degree of visual angle, ‘dva’) from fixation. Experimental control was handled by NIMH Cortex software (http://www.cortex.salk.edu/).

Receptive Field (RF) Mapping

Request a detailed protocol

At the beginning of each recording session, neuronal RFs were mapped using subspace reverse correlation (Ringach et al., 1997) in which Gabor (eight orientations, 80% luminance contrast, spatial frequency 1.2 cycles/degree, Gaussian half-width 2°) or ring stimuli (80% luminance contrast) appeared at 60 Hz while monkeys maintained fixation. Each stimulus appeared at a random location selected from an 11 × 11 grid with 1° spacing in the appropriate visual quadrant. All RFs were in the lower visual quadrant (lower-left in Monkey A, lower-right in Monkey C) and with eccentricities between 5 and 8 dva.

Irradiance response curves

Request a detailed protocol

After estimating the RF of a single-unit or multi-unit cluster, we assessed its sensitivity to optical stimulation. While the monkey maintained fixation, we measured the neuronal response to visual (achromatic Gabor stimulus, spatial frequency 1.2 cycles/degree, 20% luminance contrast) and optical stimulation. The visual stimulus was flashed at the RF for 200 ms with a simultaneous step laser pulse chosen from one of several irradiance values (typically 0, 10, 30, 50 and 70 mW/mm2) (Figure 2—figure supplement 5).

Attention task

Request a detailed protocol

In the main experiment, monkeys had to perform an attention-demanding orientation change-detection task (Figure 2A). While the monkey maintained fixation, two achromatic Gabor stimuli (spatial frequency 1.2 cycles/degree, 6 contrasts randomly chosen from an uniform distribution of luminance contrasts, c=[10, 18, 26, 34, 42, 50%]) were flashed on for 200ms and off for a variable period chosen from a uniform distribution between 200-400ms. One of the Gabors was flashed in the center of the RFs, the other at a location of equal eccentricity across the vertical meridian. At the beginning of a block of trials, the monkey was spatially cued (‘instruction trials’) to covertly attend to one of these two spatial locations. During these instruction trials, the stimuli were only flashed at the spatially cued location. At an unpredictable time (minimum 1s, maximum 5s, mean 3s), one of the two stimuli changed in orientation. The time of orientation change was chosen by sampling from an exponential distribution (thus leading to a flat hazard function of wait times till orientation change). If the sampled change time exceeded 5s, the trial was treated as a catch trial (see below), in which the change did not actually occur during the trial and the monkey was rewarded for maintaining fixation. If the orientation change did occur, the monkey was rewarded for making a saccade to the location of orientation change. However, the monkey was rewarded for only those saccades where the saccade onset time was within a window of 100-400ms after the onset of the orientation change. The location of orientation change was chosen with 95% probability at the cued location and with 5% probability at the uncued location (‘foil trials’). We controlled task difficulty by varying the degree of orientation change (Δori), which was randomly chosen from one of 8 orientations in the range 1-15°. The orientation change in the foil trials was fixed at 4°. These foil trials allowed us to assess the extent to which the monkey was using the spatial cue, with the expectation that there would be an impairment in performance and slower reaction times (Figure 2—figure supplement 2A) compared to the case in which the change occurred at the cued location. If no change occurred before 5s, the monkey was rewarded for maintaining fixation (‘catch trials’, 13% of trials). We will refer to all stimuli at the baseline orientation as ‘non-targets’ and the stimulus flash with the orientation change as the ‘target’. If the monkey made a saccade to a non-target stimulus at any time, it was treated as a false alarm and the monkey was not rewarded.

On a random subset of trials (50% of trials in experimental sessions with low-frequency stimulation only; 33% of trials in experimental sessions with both low- and high-frequency stimulation conditions), neurons at the injection site were stimulated with 4–5 Hz sinusoidally modulated low-power laser stimulation (‘low-frequency stimulation’ condition). The sinusoidal modulation had excursions from a minimum irradiance close to 0 mW/mm2 to a maximum irradiance, chosen such that the equivalent root-mean-squared intensity elicited a firing rate either 10% above (Figure 2—figure supplement 5, left example unit) or 10% below (Figure 2—figure supplement 5, right example unit) the firing rate in the zero-irradiance condition. The optical stimulation lasted the entire duration of the trial. On a subset of experimental sessions (n = 15), neurons at the injection site were also stimulated with 20 Hz sinusoidally modulated low-power laser stimulation (‘high-frequency stimulation’ condition; 33% of trials).

Data analysis

Behavioral Analysis

Request a detailed protocol

For each orientation change condition Δori, we calculated the hit rate as the ratio of the number of trials in which the monkey correctly identified the target with a saccade over the number of trials in which the target was presented. The hit rate as a function of Δori, yields a behavioral psychometric function (Figure 2B, Figure 2—figure supplement 1, Figure 2—figure supplement 2). Psychometric functions were fitted with a smooth logistic function (Palamedes MATLAB toolbox). Error bars were obtained by a jackknife procedure (20 jackknives, 5% of trials left out for each jackknife). Performance for the foil trials were calculated similarly as the hit rate for trials in which the orientation change occurred at the un-cued location (Figure 2—figure supplement 2A, left panel, square symbol). Performance for the catch trials was calculated as the fraction of trials in which the monkey correctly held fixation for trials in which there was no orientation change (Figure 2—figure supplement 2A, left panel, star symbol). Psychometric functions were obtained separately for the baseline (no laser stimulation) and the optical stimulation conditions.

Psychophysical studies have found that human observers are better able to discriminate stimulus orientations near the cardinal than oblique orientations (Girshick et al., 2011; Heeley and Timney, 1988; Appelle, 1972; Orban et al., 1984; Campbell et al., 1966). Electrophysiological and imaging studies in humans, monkeys, cats and ferrets have found that cardinal orientations are overrepresented in V1 (De Valois et al., 1982; Furmanski and Engel, 2000; Li et al., 2003; Wang et al., 2003; Chapman and Bonhoeffer, 1998). Consistent with this, we find that monkeys performed the task better during sessions in which they were required to discriminate orientation changes from cardinal (0°, 90°) non-target orientations, as reflected in elevated performance in detecting the smallest orientation change (p=0.05, t-test) and elevated threshold (p=0.002, t-test) for cardinal compared to non-cardinal orientations (Figure 2—figure supplement 2D). Threshold is the stimulus condition at which performance was mid-way between the lower and upper asymptotes of the fitted psychometric function.

We characterized the change in behavioral performance due to optical stimulation as a modulation index (PDMI, perceptual discrimination modulation index):

PDMI=thresholdopto-thresholdbaseline/thresholdbaseline

In addition, we assessed any changes in psychometric function slope (steepness of the curve at threshold) due to optical stimulation as the change over the baseline (no laser) condition normalized by the slope at baseline (Figure 2—figure supplement 3A). Significant changes in threshold and slope for each individual behavioral session (Figure 2—figure supplement 3B) were calculated by comparing the distributions of threshold and slope values estimated from the jackknife procedure between the optical stimulation and baseline conditions (t-test, α=0.05, corrected for the number of jackknifes).

Peri-stimulus time-histograms

Request a detailed protocol

For this and subsequent analyses of neuronal data, we restricted our analyses to non-target flashes from correct trials (hit trials in which the monkey correctly detected a target or correct catch trials). Neuronal responses were binned using a sliding window of width 30 ms that was shifted by 10 ms increments to obtain the time-varying firing rates, also known as the peri-stimulus time-histograms (PSTH), of the recorded units (Figure 4A). Population PSTH plots (Figure 4B) were obtained after normalizing the responses of each neuron to the peak across the four experimental conditions (two attention conditions [attend-in, attend-away] x two stimulation conditions [no stimulation, laser stimulation]).

Spike-phase distributions

Request a detailed protocol

We calculated the phase of each spike with respect to the sinusoidal laser stimulation during a 200 ms blank period before a non-target stimulus flash. We only considered those inter-stimulus periods where the inter-stimulus interval was greater than 500 ms (in other words, the interval between onset of the stimulus and the offset of the previous stimulus was greater than 300 ms), so as to minimize artifacts due to stimulus offset. Polar plots in Figure 4C show the distributions of spiking phases. To see if these distributions were significantly different from chance, we calculated a null distribution by generating spike times from a rate-matched Poisson process (gray polar plots in Figure 4C). To obtain reliable estimates for spike-phase distributions, we restricted our analysis to units with a minimum firing rate of 5 spikes/s (n = 68; firing rate averaged over the 200 ms stimulus-evoked period between 60–260 ms after non-target onset).

Spike-count correlations (rSC)

Request a detailed protocol

We calculated the Pearson correlation of spike counts across trials for every pair of simultaneously recorded units. In order to remove the influence of confounding variables like stimulus strength, spike counts were z-scored using the mean and standard deviation for repetitions of each stimulus type. Ordered pairs of z-scored spike counts were collapsed across contrast conditions and the Pearson correlation was calculated from these ordered pairs. This was done separately for the different attention and optical stimulation conditions and also for different sized counting windows (50ms for high-frequency correlations, 200ms for low-frequency correlations) during the stimulus-evoked period between 60-260ms after non-target onset (Figure 3). Multiple non-overlapping windows were used for those counting windows that were smaller than the 200ms stimulus evoked period.

Neural Discrimination Modulation Index (NDMI)

Request a detailed protocol

For each neuron in our population, we extracted spike counts to repeated presentations of non-target and target stimuli (60-260ms after stimulus onset) for the baseline and low-frequency stimulation conditions when the animals were attending in to the RF. Response rates for each neuron were normalized by the maximum response across conditions after first subtracting the spike rates during a 200ms pre-stimulus period (pre-stimulus rates calculated separately for baseline and low-frequency stimulation conditions). We thus obtained two response clouds for each experimental condition: one for the non-target stimuli and the other for the target stimuli (schematic in Figure 5A). We calculated the neural discriminability between the two response clouds as the Mahalanobis distance (D) between the centroid of the target responses and the non-target response cloud. The modulation of this discriminability measure due to optical stimulation was quantified as an index (Neural Discriminability Modulation Index, NDMI):

NDMI= Dopto-DbaselineDopto+Dbaseline

NDMI for each experimental session was calculated in two ways: either from multi-dimensional clouds from all simultaneously recorded neurons or as the average of two-dimensional clouds across all pairs of simultaneously recorded neurons.

Computational model

Request a detailed protocol

A similar model has been described previously (Nandy et al., 2017). We set up a conductance-based model of NE excitatory and NI inhibitory neurons with 80% connection probability (both within and across the two populations) and with the following synaptic weights (Figure 6):

E to E: wEE=WEENE; I to I: wII=WIINI; E to I: wIE=WIENE; I to E: wEI=WEINI

We simulated models of NE=800 excitatory and NI=200 inhibitory spiking units. The spiking units were modeled as Izhikevich neurons (Izhikevich, 2003) with the following dynamics:

dvdt=0.04v2+5v+140-u+I
dudt=abv-u
if v=30mV, then vc and uu+d

v is the membrane potential of the neuron and u is a membrane recovery variable. I is the current input to the neuron (synaptic and injected DC currents). The parameters ab, c and d determine intrinsic firing patterns and were chosen as follows:

Excitatoryunits:a=0.02, b=0.2, c=65, d=8
Inhibitoryunits:a=0.1, b=0.2, c=65, d=2

Presynaptic spikes from excitatory units generated fast (AMPA) and slow (NMDA) synaptic currents, while presynaptic spikes from inhibitory units generated fast GABA currents:

Isyn=igAMPAtvt-VAMPA+jgNMDAtvt-VNMDA+kgGABAtvt-VGABA

where VAMPA=0,VNMDA=0,VGABA=-70 are the respective reversal potentials (mV). The synaptic time courses g(t) were modeled as a difference of exponentials:

gt=1τd-τrexp-t-τlτd-exp-t-τlτr where τl,τr and τd are the latency, rise and decay time constants with the following parameter values (ms) (Brunel and Wang, 2003):

τlτrτd
AMPA10.52
NMDA1280
GABA10.55

The NMDA to AMPA ratio was chosen as 0.45 (Myme et al., 2003). The network was stimulated by a DC step current (IE=2.8, II=2.3) of duration 1.5s (Figure 6B). Synaptic noise was simulated by drawing from a normal distribution (Isyn-noiseN(μ=0,σ=3)). To simulate the laser stimulation in the main experiment, we chose a random subset (50%) of excitatory units to which we injected a 4Hz sinusoidally modulated current (Iopto;mean current=0.5;peak to trough range=0.7). Such a current by itself did not produce spiking activity in the network.

We computed the spike-spike coherence between all pairs of excitatory units in the model (irrespective of whether the units were subjected to the additional sinusoidally modulated current) using multi-taper methods (Mitra and Pesaran, 1999), over a 400ms window for both simulation conditions: with and without Iopto. Spike trains were tapered with a single Slepian taper, giving an effective smoothing of 2.5Hz for the 400ms window (NW=1, K=1). To control for differences in firing rate between the two conditions, we adopted a rate matching procedure similar to (Mitchell et al., 2009). Induction of coherent activity in the network due to sub-threshold sinusoidal stimulation was calculated as a modulation index of coherence across the two conditions: SSC MI=(SSCwith-SSCwithout)/(SSCwith+SSCwithout). In order to obtain a baseline for the coherence expected solely due to trends in firing time-locked to network stimulation, we also computed coherence in which trial identities were randomly shuffled (Figure 6—figure supplement 1C-D).

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
    Correlations and the encoding of information in the nervous system
    1. S Panzeri
    2. SR Schultz
    3. A Treves
    4. ET Rolls
    (1999)
    Proceedings of the Royal Society of London. Series B: Biological Sciences 266:1001–1012.
    https://doi.org/10.1098/rspb.1999.0736
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
    Eye Movements and Vision
    1. AL Yarbus
    2. B Haigh
    3. LA Rigss
    (1967)
    New York: Plenum press.
  40. 40
  41. 41

Decision letter

  1. Tatiana Pasternak
    Reviewing Editor; University of Rochester, United States
  2. Eve Marder
    Senior Editor; Brandeis University, United States

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Optogenetically induced low-frequency correlations impair perception" for consideration by eLife. Your article has been reviewed by two peer reviewers, and the evaluation has been overseen by a Reviewing Editor and a Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

The paper examined whether optogenetically induced increases in spike count correlations in V4 neurons affect the ability of macaque monkeys to detect small orientation changes of Gabor gratings. The authors report that the optogenetic stimulation in the low but not high frequency range affected performance and this effect was limited to the attended location, concluding that the attention induced reduction of low frequency spike count correlations results in improved performance.

The reviewers commented on the ambitious nature of the work and its potential importance, but raised a number of serious reservations that must be addressed for the paper to be further considered for publication in eLife. These are summarized below.

Essential revisions:

1) Anatomical evidence confirming that the neurons were actually transfected should be provided. This includes the information about the affected cell type, their layer locations and variability of the expression.

2) Variable performance on the orientation change task makes it difficult to interpret the effects of optogenetic stimulation. This problem should be addressed.

3) The presence of on- and off- transients in visual stimulation occurring at a frequency of 3-5Hz creates a potential problem in the study aimed at detecting correlations occurring at similarly low frequency. It is important to rule out that optogenetic stimulation may be affecting the detection of the onsets and offsets of visual stimuli. Reviewer 1 suggests a control condition that would eliminate such transients to see whether optogenetic manipulation is still selective for low frequencies.

A related question (reviewer 2) concerns the phase of optical stimulation relative to stimulus presentation whether the presence of the behavioral deficits depended on the phase of optical stimulation. Also, provide information about the phase of optical stimulation used to compute noise correlations

4) Data presentation limited to averages and distributions does not allow the evaluation of significant effects in each experiment. The data suggest that in some cases optogenetic stimulation resulted in elevated thresholds and slopes and in some cases in opposite effects. This is a problem that needs to be addressed and discussed.

5) Please provide a direct comparison between firing rates with and without optical stimulation, showing effects on individual neurons. Rate modulation index does not allow the reader to assess such effects directly.

6) Was orientation tuning affected by optical stimulation? If they were, the disruption of orientation representation in V4 by stimulation could potentially explain the behavioral effects. This should be addressed.

7) Please explain cued and uncued locations vs. "catch trials" and their randomization during the experiment (reviewer 1).

Reviewer #1:

The goal of this paper is to causally test the idea that low-frequency oscillations amongst cortical neurons are a major limiting factor in visual perception and are actively controlled for visual attention. The strategy is a bold one – use optogenetics to introduce low-frequency correlations into the activity of cortical neurons in extrastriate area V4 and document the effects on both visual perception and neuronal activity. If the idea is correct, then the introduction of low-frequency correlations should impair performance. But there are many reasons that performance might be impaired when you alter visual cortical activity. So it is also necessary to show that visual activity has not simply been disrupted by the optogenetic manipulation, and that it is specifically the temporal restructuring of the activity that is crucial. To this end, the paper shows that overall firing rates are not changed, and that the effects are found only with low-frequency and not high-frequency stimulation, among other controls.

This is an ambitious set of experiments to attempt to pack into a short paper, especially given all of the potential pitfalls and controls that need to be considered. In the end, I was not fully convinced that the results as presented support the conclusions. There are several basic issues that need to be addressed to make the case more convincing. I also think that a more detailed unpacking of the data is need in order to understand the results.

The most obvious missing piece is histology. Figure 1C shows a surface view of the cortex illustrating the expression of EYFP. This is not sufficient to answer the questions relevant for interpreting the behavioral and neuronal data. Can you confirm that neurons were transfected? Which types of neurons, and in which layers? How variable was the expression across layers? Can you rule out retrograde transfection from axon terminals? Without histology to verify what was stimulated, I find it difficult to interpret the results.

Performance in the orientation change task seems extremely variable, to the point that it raises concerns about how to interpret changes in task performance. In some cases, the thresholds for detecting orientation changes seem to be in the expected range (a few degrees) – for example, Figure 2B and Figure 2—figure supplement 2A. But in other cases, the thresholds are unusually high, 10 degrees or more (Figure 2—figure supplement 2C). There are also unexpected elevations in% correct for signal values that should be below threshold – for example, 40% correct for the lowest orientation change value in Figure 2—figure supplement 2B. How could the monkey get 40% correct for an undetectable signal? In order to interpret the optogenetic manipulation, I would need to be reassured about the reliability of task performance in the absence of optogenetic manipulation.

There is a very basic aspect of the experimental design that seems like a problem, but perhaps the authors have a very well-reasoned explanation for this approach. The hypothesis is that low-frequency (4-5 Hz) correlations play a central role in cortical processing, and this guides the choice of sinusoidal optogenetic stimulation. But the stimulus itself is flashed on for 200 ms and then left off for 200-400 ms, which means that there were visual transients (on or off) also occurring at a frequency of 3-5 Hz. If the goal is to test the importance of intrinsic low-frequency correlations, why use a stimulus that includes transients in this same frequency range? It seems that an alternative explanation for the perturbations in performance might be that the optogenetic stimulation masks the ability to detect the onsets and offsets of the visual stimulus. Can this be ruled out? If the stimulus were simply left on for an extended period of time (seconds), and then changed orientation, would the optogenetic manipulation still be effective and still selective for low frequencies?

The data were compressed for presentation in ways that made it difficult to understand what additional significant effects might be present in each experiment. Figure 2C is a good example of this. The histogram illustrates that, overall, both the threshold and slope increased in trials with low-frequency optical stimulation, by showing that the population-level ratio of stimulation relative to baseline is significantly greater than 1. First, please clarify the stats (here and elsewhere). This is reported as a t-test in the fourth paragraph of the Results and Discussion. Which type – paired sample? Is the distribution normal, or should a non-parametric test be used? Second, please report the data in a way that allows us to appreciate what happened on individual experiments. Given the scale on the x-axis, it looks like optical stimulation may have caused significant increases in threshold or slope in some experiments and significant decreases in others. If so, it would be misleading to base the conclusion on the average effect. This point is especially relevant for the attend away condition (Figure 2D), in which the average threshold is not different, but the spread in the histogram suggests that many individual experiments were significant but roughly equally split between increases and decreases.

I had similar concerns about the neuronal data in Figure 4. In addition to plotting the rate modulation index, please directly compare the firing rates with and without optical stimulation. Was the firing rate significantly changed for individual neurons? The effects of attend-in and attend-away should be similarly documented across the population of neurons. Did optical stimulation significantly change this modulation for individual neurons? The concern here is that there may be no difference on average, but a more or less balanced combination of significant increases and significant decreases.

Beyond changes in firing rate, it is also possible that the optical stimulation disrupted the tuning properties of the neurons. Do you have data to confirm that the orientation tuning and receptive field properties of the neurons was not changed during optical stimulation? If the claim is specifically about the role of low-frequency correlations, it is important to rule out the possibility that behavioral effects were due simply to disrupting the representation of orientation information in V4.

Reviewer #2:

Nandy and colleagues investigate whether optogenetically induced increases in spike count correlations in V4 neurons affect the ability of macaque monkeys to detect small orientation changes of Gabor gratings. The find that this is indeed the case, but only if the optogenetic stimulation is in the low frequency range (4-5Hz, sinusoidal modulation), not when it occurs at higher frequencies (tested were 20 Hz). This only occurred when the neurons stimulated represented the attended (instructed) location, not when the represented the non-instructed location. The authors conclude that the attention induced reduction of low frequency spike count correlations indeed convey behavioural benefits.

This is an interesting study, but I have a few questions:

I was unable to determine whether the phase of stimulation was fixed relative to stimulus presentation. Probably not as stimulus presentation varied within a trial. This means that orientation changes would occur at variable phases of optogenetic stimulation. If so, it will be important to know whether behavioural deficits occurred equally across different phases, or whether they were co-modulated.

The histograms show and the noise correlations also seem to be calculated over different optogenetic phases (if my assumption from above is correct). It will be important to determine whether there were changes in either when calculated for different optogenetic phases. If so, these need to be documented in detail.

The authors state that no effect was seen in the attend away condition. Was that even true on catch trials (i.e. attend away, but change occurs at RF)? If so, it needs to be explained why.

Why was the higher frequency in the beta range, not in the gamma range, where some people might expect to see behavioural benefits to occur?

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Thank you for submitting your article "Optogenetically induced low-frequency correlations impair perception" for consideration by eLife. Your article has been reviewed by two peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Eve Marder as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

The reviewers acknowledged your efforts to address the comment raised in their initial review. However, a number of issues remain and those undermine their confidence in the main conclusions of the paper. These must be addressed for the manuscript to be considered for publication in eLife. The points listed below are based not only on the reviews attached below, but also on the discussion among the reviewers that followed.

Essential revisions:

1) Lack of histology

The reviewers felt that the lack of histology seriously complicates the interpretation of the results but are willing to accept a possibility that the expression patterns of the viral vector in V4 are the same as those shown for V1 in the previous publication from the lab. However, you should add appropriate disclaimers about the assumptions about transfected neurons (based on the published V1 data), and discuss how your interpretation would be affected if these assumptions turn out to be incorrect.

2) Variability in orientation thresholds

The reviewers felt that the explanation attributing variability to the difference in thresholds for oblique and cardinal orientation, needs stronger documentation. To that end, rather than providing 8 example psychometric functions, the relationship between base orientation and performance should be documented by plotting thresholds for each of the sessions as a function of orientation.

3) Potential confound of visual and optogenetic stimulation delivered at similar low frequencies

Reviewers did not feel that the phase analysis included in the revision adequately addressed the problem. Reviewer 2 commented that while the overall effect may not be phase-specific, it may be specific "to cases where the optogenetic frequency is similar to the stimulus onset/offset frequency". Please provide additional information, to address this.

Please address the absence of label for the x-axis in Figure 4—figure supplement 3B and different sign of effects on threshold and sensitivity. One of the reviewers suggests sorting the experiments into subsets based on how the psychometric curves change (for example, based on where the experiments fall in Figure 2—figure supplement 5).

4) Changes in thresholds and slopes

The summary plot (Figure 2—figure supplement 5) reveals not only increases in threshold (22/42 cases) but also decreases in threshold (~14 cases). Please address the apparently opposite effects produced by stimulation.

In addition, it appears that stimulation increased the slope more often than decrease it. Since the increase in slope is likely to be associated with an increase in sensitivity, the proposed role of decreased correlations in improving the separability of neuronal activity is puzzling. This effect appears inconsistent with the proposed interpretation of the changes in spike-count correlations with stimulation. The reviewers suggest that you link the measured changes in correlations with the observed changes in psychometric functions by using a decoding scheme similar to that used by Cohen and Maunsell, 2009. This approach would allow you to predict the changes in psychometric curves given the changes in spike-count correlations.

5) Orientation tuning

Please provide additional information concerning the Figure 4—figure supplement 4 scaling on the x-axis and pooling data across neurons. How were the neurons used for this analysis selected? (see comment from reviewer 1).

6) Please provide documentation of false alarm rates from catch trials with and without optogenetic stimulation (see comment from reviewer 1).

7) The phase locking of spikes to the phase of stimulation should also be shown during the stimulus, in addition in the absence of stimulation.

8) The effect of optogenetic phase on behavior should be aligned with the preferential population activity, which seems to peak/trough at 240° and 60° respectively.

9) Please clarify what statistical test was used to determine that the phase of optical stimulation did not affect behavior or orientation tuning

Reviewer #1:

In my comments below, for clarity I repeat the original comments in quotes, with my new comments inserted after each item.

Overall, the authors have made efforts to address each major comment, but have been prevented from fully settling several of the points due to technical limitations. Some of these are substantial points that affect my confidence in the main conclusions.

"The most obvious missing piece is histology. Figure 1C shows a surface view of the cortex illustrating the expression of EYFP. This is not sufficient to answer the questions relevant for interpreting the behavioral and neuronal data. Can you confirm that neurons were transfected? Which types of neurons, and in which layers? How variable was the expression across layers? Can you rule out retrograde transfection from axon terminals? Without histology to verify what was stimulated, I find it difficult to interpret the results."

The authors do not have histology for the two monkeys used in the study but they show a figure from a previously published study using the same viral vector injected into cortical area V1.

Will the expression patterns be the same in V4, the area targeted in this study? I don't know, and I don't know of any published work using this vector in macaque V4. Apparently, there is tissue from one of the monkeys (that was sacrificed) from this study; even if the tissue is damaged due to penetrations it should be possible to identify transfected cells, and the approximate size and layer distribution. It is not clear what steps were taken along these lines, or if that tissue was unfortunately discarded.

"Performance in the orientation change task seems extremely variable, to the point that it raises concerns about how to interpret changes in task performance. In some cases, the thresholds for detecting orientation changes seem to be in the expected range (a few degrees) – for example, Figure 2B and Figure 2—figure supplement 2A. But in other cases, the thresholds are unusually high, 10 degrees or more (Figure 2—figure supplement 2C). There are also unexpected elevations in% correct for signal values that should be below threshold – for example, 40% correct for the lowest orientation change value in Figure 2—figure supplement 2B. How could the monkey get 40% correct for an undetectable signal? In order to interpret the optogenetic manipulation, I would need to be reassured about the reliability of task performance in the absence of optogenetic manipulation."

The authors explain that this is probably due to differences in detection performance – in particular, thresholds were lower for baseline orientations near the cardinal (i.e., horizontal and vertical) orientations. They show 8 sample psychometric curves that are consistent with this explanation.

This seems plausible, but the author should show it holds true for the other 34 sessions as well. If you simply plot threshold for each session as a function of the baseline orientation, this would show whether the variance was indeed systematically related to the baseline orientation.

"There is a very basic aspect of the experimental design that seems like a problem, but perhaps the authors have a very well-reasoned explanation for this approach. The hypothesis is that low-frequency (4-5 Hz) correlations play a central role in cortical processing, and this guides the choice of sinusoidal optogenetic stimulation. But the stimulus itself is flashed on for 200 ms and then left off for 200-400 ms, which means that there were visual transients (on or off) also occurring at a frequency of 3-5 Hz. If the goal is to test the importance of intrinsic low-frequency correlations, why use a stimulus that includes transients in this same frequency range? It seems that an alternative explanation for the perturbations in performance might be that the optogenetic stimulation masks the ability to detect the onsets and offsets of the visual stimulus. Can this be ruled out? If the stimulus were simply left on for an extended period of time (seconds), and then changed orientation, would the optogenetic manipulation still be effective and still selective for low frequencies?"

The authors agree these are important potential confounds but for technical reasons, they are not able to do the control experiment in which the stimulus is simply left on, to test whether the optogenetic stimulus might be masking the visual onset and offsets that occur in the same frequency range.

The phase analysis is interesting, but does not address the same point exactly. I wouldn't necessarily expect the effect to be phase-specific, but I do suspect it might be specific to cases where the opto frequency is similar to the stimulus onset/offset frequency. This question remains open.

Some comments about Figure 4—figure supplement 3B: What is the x-axis and why is it unlabeled? There seem to be some interesting possible mixed effects at low delta orientations. If you pool across all sessions (with different sign of effects on threshold and sensitivity) perhaps some effects are getting averaged out. Have you tried sorting the experiments into subsets based on how the psychometric curves change (for example, based on where the experiments fall in Figure 2—figure supplement 5)?

"The data were compressed for presentation in ways that made it difficult to understand what additional significant effects might be present in each experiment. Figure 2C is a good example of this. The histogram illustrates that, overall, both the threshold and slope increased in trials with low-frequency optical stimulation, by showing that the population-level ratio of stimulation relative to baseline is significantly greater than 1. First, please clarify the stats (here and elsewhere). This is reported as a t-test in the fourth paragraph of the Results and Discussion. Which type – paired sample? Is the distribution normal, or should a non-parametric test be used? Second, please report the data in a way that allows us to appreciate what happened on individual experiments. Given the scale on the x-axis, it looks like optical stimulation may have caused significant increases in threshold or slope in some experiments and significant decreases in others. If so, it would be misleading to base the conclusion on the average effect. This point is especially relevant for the attend away condition (Figure 2D), in which the average threshold is not different, but the spread in the histogram suggests that many individual experiments were significant but roughly equally split between increases and decreases."

The authors now provide a summary plot (Figure 2—figure supplement 5) that summarizes the changes in threshold and slope. This is helpful. It shows that in addition to the main effect reported in the paper – the increase in threshold seen in 22/42 cases – there is also sometimes a significant decrease in threshold (~14 cases). Any ideas about why the effect flips sign in these cases?

More curiously, and harder to understand, the stimulation also tends to increase the slopes more often than it significantly decreases the slopes. An increase in slope would imply that the sensitivity of the monkey during the stimulation had increased. Given the proposed role of decreased correlations in improving the separability of neuronal activity, shouldn't the main effect have been a decrease in slope?

"I had similar concerns about the neuronal data in Figure 4. In addition to plotting the rate modulation index, please directly compare the firing rates with and without optical stimulation. Was the firing rate significantly changed for individual neurons? The effects of attend-in and attend-away should be similarly documented across the population of neurons. Did optical stimulation significantly change this modulation for individual neurons? The concern here is that there may be no difference on average, but a more or less balanced combination of significant increases and significant decreases."

The authors have added a Figure 4—figure supplement 1 that compares the spike rates for all neurons with and without optogenetic stimulation, and report no significant change in firing rate due to stimulation. I find this set of figures very convincing.

"Beyond changes in firing rate, it is also possible that the optical stimulation disrupted the tuning properties of the neurons. Do you have data to confirm that the orientation tuning and receptive field properties of the neurons was not changed during optical stimulation? If the claim is specifically about the role of low-frequency correlations, it is important to rule out the possibility that behavioral effects were due simply to disrupting the representation of orientation information in V4."

The authors respond that they did not measure tuning curves. However, they do have data from the non-target orientation and some target orientations, which they report in Figure 4—figure supplement 4. I find it difficult to evaluate this plot because I don't understand the scaling on the x-axis or how data were pooled across neurons. It is also not clear that data from all neurons should be included in this analysis, unless their activity was strongly modulated across the range of orientations used (i.e., the data indicate direction tuning over the domain tested). And then the data might be aligned on the x-axis so that 1 value corresponded to the "best" direction.

The issue of possible changes in neuronal tuning is critical for interpreting the results. Perhaps the authors can do more to address this.

"There was some ambiguity in the description of how the orientation change in the stimulus was managed. The paper describes a 95% probability at the cued location and 5% at the uncued location. But then there were also "catch trials" without any change. It's not clear how these add up. Were the 95% and 5% independent? What was the probability of a catch trial? Were these truly randomized, or were they presented in as a fixed fraction of the trials?"

The authors now explain their definition of 'catch' trials.

Are the FA rates on from catch trials documented somewhere in the paper? I did not see it except for Figure 2—Figure supplement 2A, which curiously appears to show a false alarm rate on catch trials of about 50% Is this correct? How can the FA rate be that high when the hit rate drops well below that for small orientation changes? I would expect the FA rate to be the floor for the curve.

Aside from trying to understand the plots, the other reason for asking about FAs is to know whether the FAs also changed with optogenetic stimulation. This would be important for assessing possible changes in response criterion, which would also be important to nail down, since changes in criterion could also shift the psychometric curves

Reviewer #2:

The authors have addressed some of my previous points, but I have a few issues remaining:

They describe the phase locking of spikes to the phase of stimulation when no stimulus was present. However, it would be important to see this also for the stimulus period.

The authors use 4 bins to calculate the effect of optogenetic phase on behaviour, but these are not aligned with the preferential population activity alignment, which seems to peak/trough at 240° and 60° respectively. This needs to be done.

It is unclear what statistical test was used to determine whether behaviour was unaffected by the phase of optical stimulation?

The same is true for the effect on orientation tuning.

In general, statistical reporting should be checked and adequately improved.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Thank you for submitting your article "Optogenetically induced low-frequency correlations impair perception" for consideration by eLife. Your article has been reviewed by two peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Eve Marder as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

The revised manuscript has addressed most of the reservations raised by the two reviewers. There remain only a few issues that need the authors attention. These are listed below.

Essential revisions:

1) Please substitute multiple t-tests with ANOVA to reveal potential interactions.

2) Please address the point concerning correction analysis, raised by reviewer 2.

3) Please include the figure from Essential Revisions #3 as a supplementary figure. Also include: a) the average firing rate traces for each of the 4 flashes, and b) confirm the behavioral effects for the low-frequency but not high-frequency stimulation.

Reviewer #2:

The authors have also performed the analysis relating to the NDMI and PDMI. They report a significant negative correlation, between the two. Looking at Figure 5 this seems to be driven by 5/42 experiments. They do not report what type of correlation was calculated (Pearson? Spearman? Robust correlation to control for outliers?). I think a robust correlation would be appropriate, while given the distribution of data I assume Pearson is inappropriate. They also present a line, which I assume is a slope of a linear regression? Given that NDMI and PDMI are dependent variables, slopes need to be calculated of x vs. y and y vs. x, and then the average slope needs to be taken.

https://doi.org/10.7554/eLife.35123.024

Author response

Essential revisions:

1) Anatomical evidence confirming that the neurons were actually transfected should be provided. This includes the information about the affected cell type, their layer locations and variability of the expression.

Unfortunately we cannot provide post mortem histology on the two animals used in this study. Monkey A is still being used for an ongoing experiment. Euthanizing Monkey A for histology would halt that experiment. Monkey C had to be sacrificed due to health complications and unfortunately the brain tissue at the recording site, which was penetrated multiple times during the experiment (and another subsequent experiment), was not in suitable condition for post mortem histology.

However, we have performed histological analysis in a third monkey, at a site not used for recording and this confirmed that expression of C1V1 and EYFP was biased almost exclusively to excitatory neurons at and around the injection site. See Author response image 1.

Author response image 1

White arrows in the upper left panel indicate neurons that were immuno-positive for EYFP (FP+; green, top-left panel). White arrows in the upper right panel show neurons that were immune-positive for a cocktail of inhibitory neuron markers (parvalbumin, calretinin, or calbindin, red, top-right panel). The lower left panel shows a merged image illustrating that in this sample there was little or no overlap. Lower right panel shows quantification of this over 119 neurons imaged in five different fields of view. Only 1.7% of all EYFP-positive neurons counted of the 119 neurons were double labeled for both EYFP and PV/CR/CB (n = 2) (bottom-right panel), indicating that expression was heavily biased toward excitatory neurons. Scale bar, 20 mm. This is Figure 1B from Nassi et al., 2015, Neuron, which was conducted in our laboratory just prior to conducting this study using the identical viral construct used in the present study. We would be happy to include a version of this figure as a supplementary figure if the reviewers and Editor feel that this would strengthen the manuscript.

Reviewer 1: “Can you confirm that neurons were transfected?”

In the current study, we used the same viral constructs and the same viral injection protocol as was used in Nassi et al., where we quantified selective expression in pyramidal neurons. As in Nassi et al., 2015, in the present experiment we used fluorescent goggles to visualize protein expression at the injection site (See Figure 1C, main manuscript). We were able to see a bright fluorescent spot in both monkeys, throughout the experiment, indicating continued expression of GFP-tagged opsin. Additional evidence that cells were not only transfected but that they expressed opsin at a level sufficient to drive neurons is that we found modulation of firing that was tightly time locked to laser stimulation (see e.g., phase locking with laser intensity at high and low frequency: Figure 4—figure supplement 2) and firing rates that varied with levels of optogenetic stimulation (irradiance response curves, Figure 2—figure supplement 4). Taken together, all these lines of evidence give us a high level of confidence that not only were neurons transfected but also they continued to do so through the months of the study in both animals.

Reviewer 1: “Which types of neurons, and in which layers? How variable was the expression across layers?”

As we mentioned above, we find that, using this viral construct > 97% of transfected neurons are excitatory. There was no bias in expression toward any particular layer (Figure 1C of Nassi et al.).

Reviewer 1: “Can you rule out retrograde transfection from axon terminals?”

Lentivirus is primarily a locally infecting virus. Retrograde labeling with sufficient levels of opsin to be functional in vivo is very challenging and has been difficult to accomplish in the primate. So this possibility seems very unlikely.

Thus, in conclusion, based on our published data in V1 and similar in vivo epifluorescence profiles we observed in V4, we conclude that expression of opsin is primarily in excitatory neurons throughout all layers with a similar lateral spread (200300𝜇m). We would be happy to include supplemental figures from the data collected in the Nassi et al., 2015 study, showing this in the present paper if the reviewer and editors feel that this would strengthen the paper.

2) Variable performance on the orientation change task makes it difficult to interpret the effects of optogenetic stimulation. This problem should be addressed.

The monkeys used in our experiments were highly trained in the orientation change detection task and participated in multiple experiments. As a result, they could achieve performance levels that were very high even for the smallest orientation changes. The smallest orientation change that we tested was 1-degree orientation. This is a substantially lower threshold than has been achieved in primate studies (See e.g., Figure 1 from Cohen and Maunsell, 2009). However, we know from the psychophysical literature (Campbell and Green, 1965; Cowey and Rolls, 1974) that humans can detect orientation changes that are much lower than 1-degree, and that human observers are most sensitive to orientation changes for gratings presented near the cardinal orientations. We looked back at our data and found indeed a pattern consistent with this. For sessions in which the baseline (non-target) orientation was horizontal or vertical, monkeys consistently performed better in detecting orientation changes of 1 degree, the smallest change we tested. Please see Author response image 2, which shows% correct in detecting the target over eight sessions, including three (red stars) in which the non-target was presented at a cardinal orientation. This was a likely source of variation in animal performance in our study. Despite this source of variability, we find that low- but not high- frequency optogenetic stimulation systematically impaired performance for fine orientation discriminations at the stimulation site.

Author response image 2
Examples of behavioral sessions without optogenetic stimulation.

The red asterisks mark behavioral sessions where the baseline orientation (orientation of the non-target stimuli) was either horizontal or vertical.

3) The presence of on- and off- transients in visual stimulation occurring at a frequency of 3-5Hz creates a potential problem in the study aimed at detecting correlations occurring at similarly low frequency. It is important to rule out that optogenetic stimulation may be affecting the detection of the onsets and offsets of visual stimuli. Reviewer 1 suggests a control condition that would eliminate such transients to see whether optogenetic manipulation is still selective for low frequencies.

A related question (reviewer 2) concerns the phase of optical stimulation relative to stimulus presentation whether the presence of the behavioral deficits depended on the phase of optical stimulation. Also, provide information about the phase of optical stimulation used to compute noise correlations

We agree that these are important potential confounds that need to be addressed. Though as noted above, it was, unfortunately, not possible to do the very thoughtfully considered control experiments in these animals, as reviewer 1 had suggested (one animal had to be humanely euthanized due to health concerns and the V4 chamber of the second animal has been removed and replaced with a V1 chamber). We have, however, undertaken analyses that we believe address these concerns. We reasoned that if the change in performance were due to a phase alignment between the visual stimuli and the optical stimulation, we would expect to find variation in behavioral performance that varied with laser phase (a point raised by reviewer 2). These analyses are described below, after each reviewer’s individual comments.

Reviewer 2: “it will be important to know whether behavioural deficits occurred equally across different phases, or whether they were co-modulated”

We have added a supplementary figure (Figure 4—figure supplement 3) in which we first show (panel A) that stimulus onset times were not phase locked to the laser, at either frequency. The distribution of target stimulus onset times is plotted as a function of laser phase for low-frequency (left panel) and high-frequency (right panel) optical simulation. This is due to the aperiodic nature of the stimulus flashes, with variable inter-stimulus-intervals between each non-target and between the final non-target and the target. So, targets were equally likely to occur at all phases of laser stimulation and there was therefore no tendency for targets to phase lock to the laser, at either frequency.

We next examined whether discrimination performance varied as a function of laser phase and target onset. See Panel B, which shows percent correct performance as a function of the phase of the laser at the time of target stimulus onset, grouped into 4 phase bins. We find no significant difference in performance as a function of laser phase either on low-frequency trials (Panel B) or high frequency trials (data now shown).

Reviewer 2: “The histograms show and the noise correlations also seem to be calculated over different optogenetic phases […] it will be important to determine whether there were changes in either when calculated for different optogenetic phases”.

As the reviewer fully appreciates, induction of correlation by optical stimulation is the method by which we introduced correlations in the experiment. However, if we are interpreting the reviewer’s concern correctly, s/he is making a more subtle point: that there might be phase-dependent variation in the degree of endogenously generated correlation. This would not be surprising, given that increases in luminance contrast have been found to reduce pairwise “noise” correlations. So, one might imagine that, as with contrast elevation, noise correlations might similarly fall off with laser intensity. We have therefore undertaken an analysis to test this directly, in which we compute pairwise correlations as a function of laser phase. This is shown in Author response image 3.

In this analysis, we examined the phase dependence of spike-count correlations for both low- and high-frequency correlations (200ms and 50ms counting windows, respectively) on low- and high-frequency optical stimulation trials. That is, for each pair of neurons, we computed the spike rate for each cycle of laser stimulation, divided into four phase bins. We then computed correlations across multiple cycles of laser stimulation. The results of this analysis appear in Author response image 3. Though it is not statistically significant, there does appear to be a tendency for spike count correlations to be lower, on average, at the peak of laser intensity. However, we find no evidence that perceptual performance varied with phase and so we do not believe this change in correlation, even if it were significant, influenced the monkey’s performance. Nor can it account for the observed perceptual impairment that was caused by low frequency laser stimulation, which was phase independent (Figure 4—figure supplement 3B).

Author response image 3
Phase dependence of spike-count correlations for both low- (left panel, 200ms counting window) and high-frequency correlations (right pane, 50ms counting window) for the low-frequency optical stimulation condition.

4) Data presentation limited to averages and distributions does not allow the evaluation of significant effects in each experiment. The data suggest that in some cases optogenetic stimulation resulted in elevated thresholds and slopes and in some cases in opposite effects. This is a problem that needs to be addressed and discussed.

We do see evidence of both changes in threshold and changes in slope. To give the reviewer and reader a sense for the data, we have added a supplementary figure (Figure 2—figure supplement 5) in which we show the changes in behavioral performance for each experimental session. This figure shows that, consistent with our main conclusion, low frequency laser induced correlations did tend to increase threshold (preponderance of red and green points to the right of zero. There were also changes in slope, which tended to increase with laser stimulation (red and blue points above zero). In total, for a large fraction of the sessions, both the changes in threshold and slope were significant (20/42). 11/42 sessions had significant threshold change only, while 5/42 sessions had significant slope change only.

5) Please provide a direct comparison between firing rates with and without optical stimulation, showing effects on individual neurons. Rate modulation index does not allow the reader to assess such effects directly.

We have added a supplementary figure (Figure 4—figure supplement 1) in which we compare the spike rates for all neurons in our population between the optical stimulation conditions (both low- and high-frequency) and the baseline (unstimulated condition). We find no significant modulation of firing rate due to optical stimulation, as indicated by the points falling along the line of unity.

6) Was orientation tuning affected by optical stimulation? If they were, the disruption of orientation representation in V4 by stimulation could potentially explain the behavioral effects. This should be addressed.

The reviewer is correct to point out that if the laser altered tuning, this could affect the monkey’s performance and possibly account for the observed impairment. In particular, if the differences in firing rate between target and non-target orientations were reduced by low frequency optical stimulation, this could explain our findings. While we did not record tuning curves, we do have data from the non-target orientation and from multiple target orientations, so we could examine the effect of laser stimulation over the range of orientations spanned by targets and non-targets.

The results of this analysis are shown in a new supplementary figure (Figure 4—figure supplement 4), where we plot the response rates to the non-target and target orientations for the optical stimulation (for both low- and high-frequency) and unstimulated baseline conditions. To control for variation in mean firing rate across neurons, we have normalized the firing rates by the maximum rate for each neuron. We do not see any significant differences in firing rates to the different orientations due to optical stimulation.

7) Please explain cued and uncued locations vs. "catch trials" and their randomization during the experiment (reviewer 1).

We thank the reviewer for pointing out that we were not clear in our use of these terms. We have now revised the Materials and methods section to make this clear. To summarize: The time of target onset was drawn from an exponential distribution with time constant of 3 seconds, with the constraint that the target could not appear sooner than 1 sec after the start of the trial. If the target time drawn from the exponential distribution was >5 seconds, the trial was deemed a ‘catch’ trial and the target never appeared (13% of trials). On these catch trials, the monkey was rewarded for holding fixation to the 5 second mark. If the time drawn from the distribution fell between 1 second and 5 seconds, the target appeared at that time and the monkey was rewarded if it detected the target (“hit”) with an eye movement. Failure to detect the target was considered a “miss” and no reward was given. Trials were blocked according to which location was cued (left or right of fixation). When the target appeared, 95% of the time it appeared at the cued location. The remaining 5% of trials were deemed ‘foil’ trials.

The decision whether or not to stimulate (50/50), and the frequency of stimulation, were completely randomized from trial to trial.

Reviewer #1:

[…] This is an ambitious set of experiments to attempt to pack into a short paper, especially given all of the potential pitfalls and controls that need to be considered. In the end, I was not fully convinced that the results as presented support the conclusions. There are several basic issues that need to be addressed to make the case more convincing. I also think that a more detailed unpacking of the data is need in order to understand the results.

The most obvious missing piece is histology. Figure 1C shows a surface view of the cortex illustrating the expression of EYFP. This is not sufficient to answer the questions relevant for interpreting the behavioral and neuronal data. Can you confirm that neurons were transfected? Which types of neurons, and in which layers? How variable was the expression across layers? Can you rule out retrograde transfection from axon terminals? Without histology to verify what was stimulated, I find it difficult to interpret the results.

Please see our responses to Essential revisions #1 above.

Performance in the orientation change task seems extremely variable, to the point that it raises concerns about how to interpret changes in task performance. In some cases, the thresholds for detecting orientation changes seem to be in the expected range (a few degrees) – for example, Figure 2B and Figure 2—figure supplement 2A. But in other cases, the thresholds are unusually high, 10 degrees or more (Figure 2—figure supplement 2C). There are also unexpected elevations in% correct for signal values that should be below threshold – for example, 40% correct for the lowest orientation change value in Figure 2—figure supplement 2B. How could the monkey get 40% correct for an undetectable signal? In order to interpret the optogenetic manipulation, I would need to be reassured about the reliability of task performance in the absence of optogenetic manipulation.

Please see our responses to Essential revisions #2 above.

There is a very basic aspect of the experimental design that seems like a problem, but perhaps the authors have a very well-reasoned explanation for this approach. The hypothesis is that low-frequency (4-5 Hz) correlations play a central role in cortical processing, and this guides the choice of sinusoidal optogenetic stimulation. But the stimulus itself is flashed on for 200 ms and then left off for 200-400 ms, which means that there were visual transients (on or off) also occurring at a frequency of 3-5 Hz. If the goal is to test the importance of intrinsic low-frequency correlations, why use a stimulus that includes transients in this same frequency range? It seems that an alternative explanation for the perturbations in performance might be that the optogenetic stimulation masks the ability to detect the onsets and offsets of the visual stimulus. Can this be ruled out? If the stimulus were simply left on for an extended period of time (seconds), and then changed orientation, would the optogenetic manipulation still be effective and still selective for low frequencies?

Please see our responses to Essential revisions #3 above.

The data were compressed for presentation in ways that made it difficult to understand what additional significant effects might be present in each experiment. Figure 2C is a good example of this. The histogram illustrates that, overall, both the threshold and slope increased in trials with low-frequency optical stimulation, by showing that the population-level ratio of stimulation relative to baseline is significantly greater than 1. First, please clarify the stats (here and elsewhere). This is reported as a t-test in the fourth paragraph of the Results and Discussion. Which type – paired sample? Is the distribution normal, or should a non-parametric test be used? Second, please report the data in a way that allows us to appreciate what happened on individual experiments. Given the scale on the x-axis, it looks like optical stimulation may have caused significant increases in threshold or slope in some experiments and significant decreases in others. If so, it would be misleading to base the conclusion on the average effect. This point is especially relevant for the attend away condition (Figure 2D), in which the average threshold is not different, but the spread in the histogram suggests that many individual experiments were significant but roughly equally split between increases and decreases.

Please see our response to Essential revisions #4 above.

I had similar concerns about the neuronal data in Figure 4. In addition to plotting the rate modulation index, please directly compare the firing rates with and without optical stimulation. Was the firing rate significantly changed for individual neurons? The effects of attend-in and attend-away should be similarly documented across the population of neurons. Did optical stimulation significantly change this modulation for individual neurons? The concern here is that there may be no difference on average, but a more or less balanced combination of significant increases and significant decreases.

Please see our response to Essential revisions #5 above.

Beyond changes in firing rate, it is also possible that the optical stimulation disrupted the tuning properties of the neurons. Do you have data to confirm that the orientation tuning and receptive field properties of the neurons was not changed during optical stimulation? If the claim is specifically about the role of low-frequency correlations, it is important to rule out the possibility that behavioral effects were due simply to disrupting the representation of orientation information in V4.

Please see our response to Essential revisions #6 above.

Reviewer #2:

[…] I was unable to determine whether the phase of stimulation was fixed relative to stimulus presentation. Probably not as stimulus presentation varied within a trial. This means that orientation changes would occur at variable phases of optogenetic stimulation. If so, it will be important to know whether behavioural deficits occurred equally across different phases, or whether they were co-modulated.

Please see our responses to Essential revisions #3 above.

The histograms show and the noise correlations also seem to be calculated over different optogenetic phases (if my assumption from above is correct). It will be important to determine whether there were changes in either when calculated for different optogenetic phases. If so, these need to be documented in detail.

Please see our responses to Essential revisions #3 above.

The authors state that no effect was seen in the attend away condition. Was that even true on catch trials (i.e. attend away, but change occurs at RF)? If so, it needs to be explained why.

We think that the reviewer is referring to what we termed “foil” trials in our manuscript (i.e. change occurs in uncued spatial location). Since the foil trials were roughly 5% of the trials, which were then split between the stimulated and unstimulated conditions, we did not have sufficient statistical power to evaluate the effect of optical stimulation during the foil trials in the attend away condition.

Why was the higher frequency in the beta range, not in the gamma range, where some people might expect to see behavioural benefits to occur?

One of our initial hypotheses was that gamma range stimulation might produce behavioral benefits. However, in our pilot experiments, we tried gamma range stimulation using both sinusoidal stimulation and ramp stimulation (Adesnik and Scanziani, 2010, Nature) protocols, but we did not find any noticeable change in behavior. For our main experiment, we used the beta frequency range for our control condition, since it was above the 10Hz frequency range where a previous study from our lab (Mitchell, Sundberg and Reynolds, 2009) did not find an appreciable change in attention dependent coherent activity.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

1) Lack of histology

The reviewers felt that the lack of histology seriously complicates the interpretation of the results but are willing to accept a possibility that the expression patterns of the viral vector in V4 are the same as those shown for V1 in the previous publication from the lab. However, you should add appropriate disclaimers about the assumptions about transfected neurons (based on the published V1 data), and discuss how your interpretation would be affected if these assumptions turn out to be incorrect.

We acknowledge that it would have been more compelling to perform histology in order to demonstrate the viral expression pattern. As we have indicated in the previous round of review, one of our animals who had to be euthanized due to health complications but we could not perform histology since the dura mater adhered to V4 bilaterally. The second animal is still in use for experiments. However, we feel reasonably confident that the expression pattern in V4 would be similar to what has been seen in V1 (Nassi et al., 2015) and perirhinal cortex (Tamura et al., 2017), for example.

We have added a paragraph (Results and Discussion, ninth paragraph) in the revised manuscript outlining these details.

2) Variability in orientation thresholds

The reviewers felt that the explanation attributing variability to the difference in thresholds for oblique and cardinal orientation, needs stronger documentation. To that end, rather than providing 8 example psychometric functions, the relationship between base orientation and performance should be documented by plotting thresholds for each of the sessions as a function of orientation.

We have added new analysis which shows that monkeys performed the task better during sessions in which they were required to discriminate orientation changes from cardinal (0°, 90°) non-target orientations, as reflected in elevated performance in detecting the smallest orientation change and elevated threshold for cardinal compared to non-cardinal orientations (Figure 2—figure supplement 2D).

We have added additional text to the second paragraph of the subsection “Data analysis”.

3) Potential confound of visual and optogenetic stimulation delivered at similar low frequencies

Reviewers did not feel that the phase analysis included in the revision adequately addressed the problem. Reviewer 2 commented that while the overall effect may not be phase-specific, it may be specific "to cases where the optogenetic frequency is similar to the stimulus onset/offset frequency". Please provide additional information, to address this.

We apologize for misunderstanding the reviewer’s initial concern, which we took to be centered on phase. It does seem plausible that neural and perceptual sensitivity might vary with the depolarization state of the neurons, and might therefore vary with laser phase. From the reviewer’s comments, he or she appears to agree that we do show that the target stimuli are not locked to any particular phase of the laser (Figure 4—figure supplement 3A) and that behavioral performance does not depend on when the target stimulus appeared in the laser cycle (Figure 4—figure supplement 3B). We have carefully considered the reviewer’s suggestion that the effect of low frequency stimulation might selectively impair stimuli presented at a similar frequency. Although we are unaware of any studies that have examined this, one might imagine that presenting the laser at a particular frequency might cause a sort of frequency-dependent adaptation that would cause neurons to be less sensitive to visual stimuli presented at a similar frequency. If so, the low (4-5Hz) frequency laser could reduce the responses evoked by 200 msec stimuli, impairing the monkey’s ability to discriminate the stimuli, while the high frequency laser might not cause this effect, explaining the observed impairment. To test this, we measured the firing rates evoked by the first four non-target stimulus flashes, on no-laser, low-frequency-laser and high-frequency laser trials, all on trials when the monkeys were attending in to the stimuli appearing at the location corresponding to the opsin site. Though the first stimulus in the sequence evoked a stronger response than the subsequent stimuli (reflecting a form of visual-stimulus-driven adaptation), we find no evidence that the addition of the laser at either frequency caused a change in mean firing rate. Please see Author response image 4 (n=38 units; mean +/- s.e.m.). We hope we have correctly understood the nature of the reviewer’s concern. If the reviewer and the editors feel that this analysis will strengthen the paper, we will be happy to include this as a supplementary figure.

Author response image 4

Please address the absence of label for the x-axis in Figure 4—figure supplement 3B and different sign of effects on threshold and sensitivity. One of the reviewers suggests sorting the experiments into subsets based on how the psychometric curves change (for example, based on where the experiments fall in Figure 2—figure supplement 5).

We fixed the label in Figure 4—figure supplement 3B.

The second part of the comment is addressed as part of #4 below.

4) Changes in thresholds and slopes

The summary plot (Figure 2—figure supplement 5) reveals not only increases in threshold (22/42 cases) but also decreases in threshold (~14 cases). Please address the apparently opposite effects produced by stimulation.

In addition, it appears that stimulation increased the slope more often than decrease it. Since the increase in slope is likely to be associated with an increase in sensitivity, the proposed role of decreased correlations in improving the separability of neuronal activity is puzzling. This effect appears inconsistent with the proposed interpretation of the changes in spike-count correlations with stimulation. The reviewers suggest that you link the measured changes in correlations with the observed changes in psychometric functions by using a decoding scheme similar to that used by Cohen and Maunsell, 2009. This approach would allow you to predict the changes in psychometric curves given the changes in spike-count correlations.

We thank the reviewers for this excellent suggestion. We have now done this analysis and find that for recording sessions where perceptual discrimination was impaired by low frequency laser stimulation, the laser reduced discriminability of target and nontarget at the neural level, as well. In this analysis, which appears in new Figure 5, we calculated a neural measure of discriminability (neural discriminability modulation index, NDMI) that was modeled on the analysis used by Cohen and Maunsell, 2009. We computed this measure in two ways: (1) using the full N-dimensional space that was defined by the responses across the N neurons recorded on each recording session and (2) a second pairwise analysis. Each of these analyses provided a measure of the degree of overlap between the neural responses evoked by the discriminanda. We repeated this analysis for no-laser trials and laser trials, and computed an index whose value corresponded to the change in discriminability caused by the addition of the laser. For each session, we then compared the change in perceptual threshold caused by the laser (we have termed this change the perceptual discrimination modulation index, PDMI, in the revised manuscript), with these measures of laser-induced changes in neural discriminability. This analysis revealed that on sessions for which the laser impaired perceptual discrimination at the behavioral level, there was a marked reduction in discriminability at the neural level, for both neural measures. The NDMI and PDMI are also significantly negatively correlated, using either NDMI measure. In other words, laser-induced reductions in discriminability at the neural level corresponded to increased perceptual thresholds.

We also repeated the analysis for laser induced changes in slope. For the purpose of this experiment, the psychometric function threshold is the measure of interest. It is a measure of the change in orientation needed to reliably discriminate target from nontarget. Slope does not measure this, and though we do see laser dependent changes in slope, they were not correlated with the NDMI. A change in slope means that the shift from detecting to not-detecting occurs over a narrower range of orientation changes, which is not really the point of the paper. We also note that the slope is very sensitive to the form of the psychometric function that is fit to the data. Instead of fitting a logistic function to the data as we have, if we fit a Weibull function to the data, we see a

similarly significant increase in threshold, but no significant change in slope.

Thus, we have focused on the threshold change in the manuscript and report the slope change in a supplementary figure (Figure 2—figure supplement 5A).

The NDMI analysis is detailed in the sixth paragraph of the Results and Discussion and in the subsection “Neural Discrimination Modulation Index (NDMI)”.

5) Orientation tuning

Please provide additional information concerning the Figure 4—figure supplement 4 scaling on the x-axis and pooling data across neurons. How were the neurons used for this analysis selected? (see comment from reviewer 1).

We realized that our analysis in the previous round was incorrect in the sense that the tuning curves across neurons were not aligned to peak response. We have corrected this (see updated Figure 4—figure supplement 4). The spike rates were normalized to peak response across stimulation conditions (optical stimulation and baseline no-stimulation conditions). The population tuning curves were plotted after aligning the peak response of each neuron to zero and binning the data into orientation bins.

6) Please provide documentation of false alarm rates from catch trials with and without optogenetic stimulation (see comment from reviewer 1).

We now report false alarm rates for both optical stimulation conditions and compare them to the false alarm rates to the baseline condition (Figure 2—figure supplement 2E). We have added text to document this: “Nor did we find significant changes in false-alarm rates with either low- or high-frequency stimulation (𝑝 > 0.1, t-test; Figure 2—figure supplement 2E). This was true for false alarms made during catch trials as well as on non-catch trials. Thus, we find no evidence that laser stimulation caused our subjects to mis-perceive a non-target as a target.”

As mentioned, false alarm rates on catch trials are similar to those on non-catch trials: ~6% of trials. The ~50% performance rate in catch trials reported in the previous version of the figure (Figure 2—figure supplement 2A, ‘*’ symbol) arises mostly due to the monkey breaking fixation toward the end of the long catch trials. We realize that this way of reporting performance in catch trials could be a potential source of confusion. We have rectified this by calculating catch trial performance after first excluding errors due to fixation breaks and have updated the figure.

7) The phase locking of spikes to the phase of stimulation should also be shown during the stimulus, in addition in the absence of stimulation.

We have done this additional analysis and included the result in Figure 4F, with additional text: “We see a similar phase locking due to optical stimulation during the stimulus presentation period (Figure 4F, 𝑝 < 0.01, Rayleigh test), although the peak of the phase-lock distribution for the stimulus presentation period occurs earlier (around 120°) compared to that for the pre-stimulus period (around 210°). This would be expected, as the neurons are depolarized by the visual stimulus and hence more easily pushed to spiking threshold by optogenetic depolarization, as compared to when no stimulus is present.”

8) The effect of optogenetic phase on behavior should be aligned with the preferential population activity, which seems to peak/trough at 240° and 60° respectively.

We have added another sub-panel in Figure 4—figure supplement 3B where we have done the phase contingent analysis using a different bin arrangement as per the review’s suggestion.

9) Please clarify what statistical test was used to determine that the phase of optical stimulation did not affect behavior or orientation tuning

We have added the details of the statistical test in the fifth paragraph of the Results and Discussion.

Reviewer #1:

The authors do not have histology for the two monkeys used in the study but they show a figure from a previously published study using the same viral vector injected into cortical area V1.

Will the expression patterns be the same in V4, the area targeted in this study? I don't know, and I don't know of any published work using this vector in macaque V4. Apparently, there is tissue from one of the monkeys (that was sacrificed) from this study; even if the tissue is damaged due to penetrations it should be possible to identify transfected cells, and the approximate size and layer distribution. It is not clear what steps were taken along these lines, or if that tissue was unfortunately discarded.

Please see our response to Essential revisions #1.

The authors explain that this is probably due to differences in detection performance – in particular, thresholds were lower for baseline orientations near the cardinal (i.e., horizontal and vertical) orientations. They show 8 sample psychometric curves that are consistent with this explanation.

This seems plausible, but the author should show it holds true for the other 34 sessions as well. If you simply plot threshold for each session as a function of the baseline orientation, this would show whether the variance was indeed systematically related to the baseline orientation.

Please see our response to Essential revisions #2.

The authors agree these are important potential confounds but for technical reasons, they are not able to do the control experiment in which the stimulus is simply left on, to test whether the optogenetic stimulus might be masking the visual onset and offsets that occur in the same frequency range.

The phase analysis is interesting, but does not address the same point exactly. I wouldn't necessarily expect the effect to be phase-specific, but I do suspect it might be specific to cases where the opto frequency is similar to the stimulus onset/offset frequency. This question remains open.

Please see our response to Essential revisions #3.

Some comments about Figure Figure 3—figure supplement 3B: What is the x-axis and why is it unlabeled? There seem to be some interesting possible mixed effects at low delta orientations. If you pool across all sessions (with different sign of effects on threshold and sensitivity) perhaps some effects are getting averaged out. Have you tried sorting the experiments into subsets based on how the psychometric curves change (for example, based on where the experiments fall in Figure 2—figure supplement 5)?

Please see our responses to Essential revisions #3 and #4.

The authors now provide a summary plot (Figure 2—figure supplement 5) that summarizes the changes in threshold and slope. This is helpful. It shows that in addition to the main effect reported in the paper – the increase in threshold seen in 22/42 cases – there is also sometimes a significant decrease in threshold (~14 cases). Any ideas about why the effect flips sign in these cases?

More curiously, and harder to understand, the stimulation also tends to increase the slopes more often than it significantly decreases the slopes. An increase in slope would imply that the sensitivity of the monkey during the stimulation had increased. Given the proposed role of decreased correlations in improving the separability of neuronal activity, shouldn't the main effect have been a decrease in slope?

Please see our response to Essential revisions #4.

The authors respond that they did not measure tuning curves. However, they do have data from the non-target orientation and some target orientations, which they report in Figure 4—figure supplement 4. I find it difficult to evaluate this plot because I don't understand the scaling on the x-axis or how data were pooled across neurons. It is also not clear that data from all neurons should be included in this analysis, unless their activity was strongly modulated across the range of orientations used (i.e., the data indicate direction tuning over the domain tested). And then the data might be aligned on the x-axis so that 1 value corresponded to the "best" direction.

The issue of possible changes in neuronal tuning is critical for interpreting the results. Perhaps the authors can do more to address this.

Please see our response to Essential revisions #5.

The authors now explain their definition of 'catch' trials.

Are the FA rates on from catch trials documented somewhere in the paper? I did not see it except for Figure 2—figure supplement 2A, which curiously appears to show a false alarm rate on catch trials of about 50% Is this correct? How can the FA rate be that high when the hit rate drops well below that for small orientation changes? I would expect the FA rate to be the floor for the curve.

Aside from trying to understand the plots, the other reason for asking about FAs is to know whether the FAs also changed with optogenetic stimulation. This would be important for assessing possible changes in response criterion, which would also be important to nail down, since changes in criterion could also shift the psychometric curves.

Please see our response to Essential revisions #6.

Reviewer #2:

They describe the phase locking of spikes to the phase of stimulation when no stimulus was present. However, it would be important to see this also for the stimulus period.

Please see our response to Essential revisions #7.

The authors use 4 bins to calculate the effect of optogenetic phase on behaviour, but these are not aligned with the preferential population activity alignment, which seems to peak/trough at 240° and 60° respectively. This needs to be done.

Please see our response to Essential revisions #8.

It is unclear what statistical test was used to determine whether behaviour was unaffected by the phase of optical stimulation?

Please see our response to Essential revisions #9.

The same is true for the effect on orientation tuning.

Please see our response to Essential revisions #9.

In general, statistical reporting should be checked and adequately improved.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

1) Please substitute multiple t-tests with ANOVA to reveal potential interactions.

We have included ANOVA analyses to show:

a) Behavioral performance was not dependent on laser phase (no main effect of laser phase, no significant interaction between laser phase and delta orientation (the trial-by-trial difference between target and non-target orientation); Results and Discussion, fifth paragraph).

b) Orientation tuning curves were not altered by laser stimulation (no main effect of laser condition, no significant interaction between laser condition and orientation; Results and Discussion, fifth paragraph).

c) Visual stimulus driven adaptation is not modulated by laser simulation (no main effect of laser condition, no significant interaction between laser condition and stimulus position within trial; Results and Discussion, fifth paragraph).

2) Please address the point concerning correction analysis, raised by reviewer 2.

As suggested by the reviewer, we have repeated the analysis using robust regression which excludes outliers in the data. In our dataset, the approach did not find any outliers, so the results were unchanged (see Author response image 5; red and green lines completely overlap). The reviewer has also rightly pointed out that since both NDMI and PDMI are dependent measures, a line fitted from a Model II regression is the appropriate approach. We have updated the fitted lines in Figure 5B with those obtained from a Model II regression (also shown in Author response image 5; black line).

Author response image 5

3) Please include the figure from Essential Revisions #3 as a supplementary figure. Also include: a) the average firing rate traces for each of the 4 flashes, and b) confirm the behavioral effects for the low-frequency but not high-frequency stimulation.

We have included a new supplementary figure (Figure 4—figure supplement 5) as suggested by the reviewer. The results show that optogenetic stimulation at either frequency altered neither mean firing rates (as we intended) nor had a measurable effect on adaptation. We have included the firing rate traces for each flash. We have commented on the behavioral effects in the main text (Results and Discussion, fifth paragraph).

Reviewer #2:

The authors have also performed the analysis relating to the NDMI and PDMI. They report a significant negative correlation, between the two. Looking at Figure 5 this seems to be driven by 5/42 experiments. They do not report what type of correlation was calculated (Pearson? Spearman? Robust correlation to control for outliers?). I think a robust correlation would be appropriate, while given the distribution of data I assume Pearson is inappropriate. They also present a line, which I assume is a slope of a linear regression? Given that NDMI and PDMI are dependent variables, slopes need to be calculated of x vs. y and y vs. x, and then the average slope needs to be taken.

We thank the reviewer for raising these issues. We repeated the analysis using robust regression, which excludes outliers. Robust regression did not identify and exclude any data points, so the results are unchanged. We also repeated the regression analysis using Model II regression, which treats both variables as dependent variables. Please see our response to Essential revision #2 above.

https://doi.org/10.7554/eLife.35123.025

Article and author information

Author details

  1. Anirvan Nandy

    1. Systems Neurobiology Laboratories, The Salk Institute for Biological Studies, La Jolla, United States
    2. Department of Neuroscience, Yale University, New Haven, United States
    Contribution
    Conceptualization, Data curation, Software, Formal analysis, Funding acquisition, Investigation, Methodology, Writing—original draft, Writing—review and editing, Designed the experiments, Collected and analyzed the data, Developed the computational model, Ran the simulations
    For correspondence
    anirvan.nandy@yale.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-4225-5349
  2. Jonathan J Nassi

    Systems Neurobiology Laboratories, The Salk Institute for Biological Studies, La Jolla, United States
    Present address
    Inscopix Inc, Palo Alto, United States
    Contribution
    Conceptualization, Investigation, Designed the experiments
    Competing interests
    No competing interests declared
  3. Monika P Jadi

    1. Department of Neuroscience, Yale University, New Haven, United States
    2. Department of Psychiatry, Yale University, New Haven, United States
    Contribution
    Formal analysis, Visualization, Writing—review and editing
    Competing interests
    No competing interests declared
  4. John Reynolds

    Systems Neurobiology Laboratories, The Salk Institute for Biological Studies, La Jolla, United States
    Contribution
    Conceptualization, Supervision, Writing—original draft, Writing—review and editing, Designed the experiments, Wrote the manuscript
    Competing interests
    No competing interests declared

Funding

Brain and Behavior Research Foundation

  • Anirvan Nandy
  • Jonathan J Nassi

National Institutes of Health (R01 EY021827)

  • John Reynolds
  • Anirvan Nandy

National Institutes of Health (NIH T32 EY020503)

  • Anirvan Nandy

National Institutes of Health (R00EY025026)

  • Monika P Jadi

NIH Blueprint for Neuroscience Research

  • John Reynolds

Gatsby Charitable Foundation

  • John Reynolds

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

This research was supported by NIH R01 EY021827 to JHR and ASN, the Gatsby Charitable Foundation to JHR, fellowship from the NIH T32 EY020503 training grant and the NARSAD Young Investigator Grant to ASN, The Salk Institute Excellerators Fellowship Program and the NARSAD Young Investigator Grant to JJN, NIH R00 EY025026 Pathways to Independence Award to MPJ, and by a NEI core grant for vision research P30 EY019005 to the Salk Institute. We would like to thank Ed Callaway and Euiseok Kim for help with optogenetic reagents, Rob Teeuwen for assistance with animal training and Catherine Williams and Mat LeBlanc for excellent animal care.

Ethics

Animal experimentation: This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All of the animals were handled according to approved institutional animal care and use committee (IACUC) protocols of the Salk Institute. All procedures were approved by the Institutional Animal Care and Use Committee at the Salk Institute (Protocol #14-00014) and conformed to NIH guidelines.

Senior Editor

  1. Eve Marder, Brandeis University, United States

Reviewing Editor

  1. Tatiana Pasternak, University of Rochester, United States

Publication history

  1. Received: January 16, 2018
  2. Accepted: February 6, 2019
  3. Accepted Manuscript published: February 22, 2019 (version 1)
  4. Version of Record published: February 26, 2019 (version 2)
  5. Version of Record updated: March 7, 2019 (version 3)

Copyright

© 2019, Nandy et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 2,057
    Page views
  • 338
    Downloads
  • 1
    Citations

Article citation count generated by polling the highest count across the following sources: PubMed Central, Crossref, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)