All experimental procedures were conducted in accordance with the guidelines of the Canadian Council on Animal Care and were approved by the Dalhousie University Committee on Laboratory Animals. Electrophysiological recordings were collected from 6 VGAT-ChR2(H134R)-EYFP transgenic mice (JAX stock # 014548, Jackson Laboratories), and 13 Pvalb-IRES-Cre;Ai32 mice, which were produced by cross-breeding Pvalb-IRES-Cre (JAX stock # 008069) and Ai32 (JAX stock # 012569) animals. For brevity VGAT-ChR2(H134R)-EYFP and Pvalb-IRES-Cre;Ai32 transgenic mice will henceforth be referred to as VGAT and PvAi32, respectively. Mice were 2–8 months old, and weighed between 20 and 33 g. The neuronal orientation tuning data presented here for the first time was part of a larger set of experiments performed in these mice, parts of which have already been published (King et al., 2016).

Mice were first pre-medicated with chlorprothixene (5 mg/kg I.P.; Sigma Aldrich), then placed in a custom face-mask and anesthetized with isoflurane in oxygen for the remainder of the experiment (2.5% isoflurane during induction, 1.5% during surgery, and 0.5% during recording; Pharmaceutical Partners of Canada). Additional doses of chlorprothixene were given every 4 h. Anesthetized mice were maintained at a body temperature of 37.5°C with a heating pad. The skull was stabilized with a head-post secured using dental epoxy. V1 was exposed for recording and optogenetic photostimulation with a small craniotomy (~1 mm²) 0.8 mm anterior and 2.3 mm lateral from lambda (Paxinos and Franklin, 2001). A wall of petroleum jelly surrounding the craniotomy was filled with saline to prevent dehydration of the cortex. The corneas were protected by frequent application of optically neutral silicone oil (30,000 cSt, Sigma Aldrich). The pupils were not dilated so as to maintain a large depth of focus, and the eyes were not immobilized because eye movements under anesthesia have been shown to be negligible in mice (Wang and Burkhalter, 2007; Niell and Stryker, 2008; Gao et al., 2010).

Extracellular recordings were obtained with glass micropipettes containing 2M NaCl with a tip diameter of 2–5 μm. Electrode depth along vertical penetrations was controlled using a micromanipulator (FHC, Bowdoin, ME). Signals were bandpass filtered between 50 and 2,000 Hz, and sampled at 40 kHz with a CED 1401 digitizer and Spike2 software (Cambridge Electronic Designs, Cambridge, UK). Online analyses were performed in Spike2 from triggered transistor-transistor logic (TTL) pulses from a window discriminator (Cornerstone by Dagan). Spike sorting was performed offline with Spike2 software, which searched for and sorted spikes using a supervised template-matching algorithm. We then used a principle components analysis to check the clustering of spike waveforms.

Optogenetic activation of V1 interneurons came about in different ways in the two kinds of transgenic mice. The VGAT transgenic mice express Channelrhodopsin 2 [ChR2(H134R)-EYFP] in all GABAergic neurons (Zhao et al., 2011), and immunohistochemical confirmation of transgene expression in cortex revealed >93% of ChR2(H134R)-EYFP positive neurons were labeled with antibodies against glutamate decarboxylase (GAD67; Zhao et al., 2011), or GABA (King et al., 2016). PvAi32 mice express ChR2(H134R)-EYFP only in Pvalb+ interneurons (Madisen et al., 2012). In these mice, immunohistochemical confirmation of transgene expression in cortex revealed 99% of Pvalb-Cre expressing cells were labeled with antibodies against parvalbumin (Pfeffer et al., 2013), and strong Cre-induced expression of transgene mRNA was found in Pvalb+ neurons (Madisen et al., 2012).

The tip of the fiberoptic cannula was positioned ~0.2–0.5 mm above the surface of V1, and a 470 nm fiber-coupled LED was used for optogenetic photostimulation (0.4 mm diameter; 0.39 NA; Thor Labs). LED activation was coordinated with visual stimuli by the CED 1401 (Figures 1B,C). ChR2(H134R)-EYFP expressing neurons pass measurable photocurrent at a light intensity of ~0.02 mW/mm², which saturates at ~1 mW/mm² (Asrican et al., 2013). Our fiber power output of 0.089–1.16 mW (median: 0.092 mW) was estimated to yield 0.14–1.8 mW/mm² (median: 0.15 mW/mm²) at 0.8 mm cortical depth (Stujenske et al., 2015), which is sufficient light intensity to induce photocurrents in ChR2(H134R)-EYFP expressing interneurons even in layer 6. However, since photostimulation intensity was always strongest near the cortical surface it was important to consider the layer distribution of interneurons in V1. No GABAergic interneurons in layer 1 express Pvalb, and few express SOM (~2%) or VIP (~5%). In layers 2/3 a similar proportion of interneurons express VIP (~20%) and Pvalb (~20%), with fewer expressing SOM (8%). In deeper layers ~50% of interneurons express Pvalb, 20–30% express SOM, and ~7% express VIP (Xu et al., 2010; Pfeffer et al., 2013). Thus, our photostimulation was sufficient to activate a large population of Pvalb+ interneurons in PvAi32 mice, and most if not all interneuron subtypes in VGAT mice. We ensured identical photostimulation parameters had no effect on neural firing in wildtype C57BL6J mice that did not express any optogenetic proteins, illumination from the visual stimulus was too dim to inadvertently activate ChR2(H134R)-EYFP expressed in the retina, and that transgenic mice had normal visually guided behavior (King et al., 2016).

Like previous work (Atallah et al., 2012; Lee et al., 2012), we used low to moderate photostimulation intensities to modulate responses over a range where tuning curve shape was retained, which was checked online. Stronger photoactivation of all GABAergic neurons or Pvalb+ cells alone can silence pyramidal cells completely (Atallah et al., 2012; King et al., 2016). Our photostimulation was continuous (rather than eliciting time locked spikes with high intensity flashed photostimulation) to maintain potentially important temporal features of pyramidal and interneuron visual responses such as onset transients, firing rate decay over time, or phase preference for the visual stimulus (Figures 1B,C). Neurons that were activated by photostimulation at low latencies (< 20 ms) irrespective of the visual stimulus were excluded from further analyses because it was likely these cells expressed ChR2(H134R)-EYFP themselves (e.g., Figure 1C; Atallah et al., 2012).

The receptive fields of isolated visually responsive units were mapped using hand-driven light bars and spots. Quantitative stimuli programmed in MATLAB using the Psychophysics Toolbox extension (Brainard, 1997; Pelli, 1997) were then presented within the classical receptive field on a calibrated CRT monitor (LG Flatron 915FT Plus 19 inch display, 100 Hz refresh, 1024 × 768 pixels, mean luminance = 30 cd/m²) at a viewing distance of 15–30 cm. Orientation and direction tuning was measured with full contrast square-wave gratings that drifted in 16 randomly interleaved directions for 2s each trail (fundamental spatial frequency = 0.03 cycles per degree; temporal frequency = 2 Hz). During photostimulated trials LED illumination occurred for the full 2s trial (Figure 1B). Grating stimuli were presented in a circular aperture surrounded by a gray field of mean luminance for 8–12 repetitions. A gray of mean luminance was presented during each 3s inter-trial interval.

Spike arrival times were exported to MATLAB (Math Works, Natick, MA) and neuronal responses were represented as spike density functions with 1 kHz resolution, generated by convolving a delta function at each spike arrival time with a Gaussian window. Tuning curves for each neuron were fit to double von Mises functions using the least squares method:

where R is the response in spikes/second; θ is the drift angle of the visual stimulus in degrees; A₁and A₂ are the amplitudes of each peak; σ₁and σ₂ are the width constants; ϕ₁and ϕ₂ are the centers of each peak; and B is the baseline firing rate. Goodness of fit to the curves was measured using r². An inclusion criterion of r² > 0.5 ensured only tuning curves that were reasonably smooth and orientation tuned were analyzed further. Width constants were converted to half-width at half-height (HWHH; Chang et al., 2012) for more intuitive interpretation:

Orientation selectivity index (OSI) was calculated for each neuron as 1—circular variance (Atallah et al., 2012), where circular variance was calculated as (Ringach et al., 1997):

where R_k is the neuron's response to orientation k. Direction selectivity index (DSI) was calculated for each neuron using the amplitudes from equation 1 (Atallah et al., 2012):

We modeled the effect of photostimulation on orientation tuning as either divisive or subtractive inhibition. To this end we fit each neuron's response during photostimulation (LEDon) to a double von Mises function where the parameters described for equation 1 were held constant using the estimated parameters from control responses (LEDoff), but with one added term to model either divisive scaling:

where g is the scaling term, or subtractive shifting with rectification:

where h is the shifting term. Each model was fit using the least squares method. Fitted models were compared by calculating an F statistic where the number of parameters are equal (Motulsky and Ransnas, 1987):

where SS_DIV and SS_SUB are the sum of squared residuals between the observed LEDon data and the divisive or subtractive model fits, respectively. For a given neuron a positive F value smaller than 1 indicated a superior fit for the divisive model, and an F value > 1 indicated a superior fit for the subtractive model.

We used parametric and nonparametic analyses where appropriate (specific tests noted in Results), and applied the Benjamini-Hochberg procedure for controlling false discovery rate (20 total comparisons). Adjusted p-values are reported (Benjamini and Hochberg, 1995).

The effects of GABAergic inhibition on V1 tuning curves were characterized across the population as changes in several parameters extracted from control and photostimulated double von Mises curve fits. Changes in pertinent parameters were analyzed with mixed repeated-measures ANOVAs to examine the main effect of photostimulation and the interaction effect between photostimulation and genotype. None of the measures discussed below showed significant interactions (p > 0.05), indicating that there was no evidence that the magnitude of any main effects differed in data obtained from the two mouse genotypes. Both types of mice showed a significant decrease in peak amplitude (A₁) with photostimulation [Figures 3A,C; F_{(1, 119)} = 95, p < 6.8 × 10⁻¹⁶]. Baseline firing (B) also decreased significantly in both types of mice [Figures 3B,D; F_{(1, 119)} = 61, p < 1.4 × 10⁻¹¹]. Most of the sample neurons in Figure 2 showed a greater decrease in peak firing than baseline activity, and this difference was significant in the population data from both types of mice [Figures 3E,F; F_{(1, 119)} = 195, p < 1.6 × 10⁻²⁵]. We found no evidence that photostimulation consistently changed tuning width, as measured with HWHH, for either the primary [F_{(1, 119)} = 3.1, p = 0.14] or secondary peaks [F_{(1, 119)} = 0.14, p = 0.94]. Direction selectivity (DSI) decreased significantly with photostimulation [Figures 3G,H; F_{(1, 119)} = 19, p < 9.6 × 10⁻⁵]. This effect appeared to arise from the substantial number of negative DSI values in the photostimulation condition, which indicated a reversal in preferred direction. However, these reversals in direction preference appeared to be quite subtle on the tuning curves themselves (e.g., Figures 2B,E,G) because most of these cells were directionally biased (0 < DSI < 0.5) rather than strongly directional (DSI > 0.5). There was no evidence that preferred orientation was changed during photostimulation [F_{(1, 119)} = 0.34, p = 0.8], and both the main (filled circles) and secondary peak locations (empty circles) clustered tightly along the line of equality in the scatter plots in Figures 3I,J. Orientation tuned or directionally biased V1 neurons are expected to have peaks separated by ~180°, but we did not constrain the peak locations in our double von Mises curves (ϕ₁, ϕ₂) to determine whether photostimulation could affect the distance between peaks. As expected, the median peak separation pooled across genotypes was 178° in the control condition. There was no evidence that photostimulation consistently altered the peak-to-peak distance [Figures 3I,J insets; F_{(1, 119)} = 0.045, p = 0.96].

We quantified the magnitude of suppression as the proportional decrease in firing to the preferred direction induced by photostimulation. We aimed to suppress firing by about a quarter with our photostimulation, which equated to a decreased rate of only 1–3 spikes/s for most neurons. Nonetheless, there was enough variability that a minority of neurons from both VGAT (21%) and PvAi32 (16%) mice had more than −0.5 suppression, which allowed the effects of stronger inhibition to be examined as well. There was no evidence of a difference between the median proportional decrease in firing for VGAT (−0.28) and PvAi32 (−0.25) mice (Figure 4A; Wilcoxon Rank Sum, p = 0.68). However, the proportional decrease in firing per mW/mm² of photostimulation irradiance was significantly greater for VGAT mice (Figure 4B; Wilcoxon Rank Sum, p < 1.5 × 10⁻⁴), indicating dimmer photostimulation was able to produce the desired level of suppression when all interneuron types were simultaneously activated.

It has been reported that the magnitude of optogenetically induced inhibition can affect the pattern of changes observed in V1 orientation tuning curves (El-Boustani and Sur, 2014; El-Boustani et al., 2014; Lee et al., 2014), so we correlated tuning curve changes with our measure of suppression. The change in tuning breadth (HWHH) with photostimulation was poorly correlated with the magnitude of suppression in both kinds of mice (Figures 4C,D; r-values inset). Conversely, the change in DSI with photostimulation showed a weak to moderate correlation with the magnitude of suppression in both kinds of mice (Figures 4E,F; r-values inset). These correlations appear to be driven by two distinct effects. First, many of the aforementioned reversals in direction preference occurred under moderate suppression. Second, there were a few neurons where stronger suppression flattened responses in the anti-preferred direction (A₂) so that DSI increased substantially. The change in orientation selectivity with photostimulation, as measured with OSI, was poorly correlated with the magnitude of suppression in both kinds of mice (Figures 4G,H; r-values inset). We included this measure to compare with previous work (e.g., Atallah et al., 2012), however in our hands changes in OSI were noisy because suppression in neurons with little spontaneous firing caused OSI to decrease (Figure 2C) or remain unchanged (Figure 2A), whereas suppression in neurons with even moderate untuned responses in their tuning curves caused OSI to increase (Figures 2B,F). Overall, photostimulation appeared to produce a reliable constellation of changes to orientation tuning, regardless of whether inhibition arose from optogenetically driving just Pvalb+ cells or all interneuron types together.

Previous work has indicated that different types of interneurons in V1 can either produce a divisive scaling or subtractive shift in pyramidal cell orientation tuning, with the majority of studies suggesting that Pvalb+ neurons induce divisive scaling (Atallah et al., 2012, 2014; Lee et al., 2012, 2014; Wilson et al., 2012; El-Boustani and Sur, 2014; El-Boustani et al., 2014). Several of the photostimulation effects we describe above in PvAi32 mice are consistent with Pvalb+ mediated inhibition taking the form of divisive scaling, but importantly we also observed similar effects in VGAT mice. First, divisive scaling should produce smaller decreases in baseline (B) compared to peak firing (A₁), and this is exactly what we observed (Figures 3E,F). Second, divisive scaling is not expected to substantially narrow tuning breadth (HWHH), and this is what we observed as well (Figures 4C,D).

To compare divisive and subtractive inhibition most directly in our data sets, we used each neuron's control double von Mises curve to generate a divisively scaled model and a subtractively shifted model to fit its photostimulated data (see Methods). The subtractive model was rectified to avoid producing negative firing rates. For the example orientation tuning curves shown in Figures 5A–F, the data points as well as the control double von Mises curve fits have an identical format to Figure 2. However, in Figure 5 models of divisive and subtractive inhibition are shown for each cell as cyan and magenta curves, respectively. The divisive model fit the photostimulated data better than the subtractive model for most example neurons because the subtractive model tended to overshoot peak responses and undershoot baseline responses (e.g., Figures 5A–C,F). We normalized all neurons by their maximal control firing rate and then calculated the sum-of-squared residuals for each model. Larger sum-of-squared residuals indicated poorer curve fits, and SS_SUB was larger than SS_DIV for 61/64 (95%) neurons recorded in VGAT (Figure 5G), and 53/57 (93%) neurons recorded in PvAi32 mice (Figure 5H). A mixed repeated-measures ANOVA examining the main effect of photostimulation and the interaction between photostimulation and genotype indicated the divisive model provided significantly better fits to the photostimulated data than the subtractive model [F_{(1, 119)} = 92, p < 1.1 × 10⁻¹⁵], and there was no evidence of an interaction [F_{(1, 119)} = 0.03, p = 0.96]. Furthermore, the subtractive model did worse with greater levels of suppression: the F-statistic comparing the two model fits (see equation 7 in Methods) was strongly correlated with the optogenetically induced proportional decrease in firing to the preferred direction in both VGAT (r = 0.82; p < 1 × 10⁻⁵) and PvAi32 mice (r = 0.66; p < 1 × 10⁻⁵). Overall, the model fitting supported previously published reports that Pvalb+ interneurons provide divisive inhibition to pyramidal neurons (Atallah et al., 2012; Wilson et al., 2012; El-Boustani and Sur, 2014; El-Boustani et al., 2014), but also indicated that this divisive inhibition prevails when all interneurons are activated simultaneously.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.