Behavioral state and stimulus strength regulate the role of somatostatin interneurons in stabilizing network activity

Abstract

Inhibition stabilization enables cortical circuits to encode sensory signals across diverse contexts. Somatostatin-expressing (SST) interneurons are well suited for this role through their strong recurrent connectivity with excitatory pyramidal cells. We tested the necessity of SST cells for inhibition stabilization in mouse primary visual cortex by selectively blocking excitatory glutamatergic receptors on SST cells. Antagonizing this key input for the recruitment of SST cells drives a paradoxical facilitation of their activity-the hallmark of inhibition stabilization-with increasing stimulus contrast, and even more so with high arousal. In a computational model of the visual cortex circuit, increasing sensory input and arousal both move the network toward a regime where other classes of interneurons are no longer sufficient for maintaining network stability. Thus, we reveal that the role of SST cells in cortical processing gradually switches as a function of both input strength and behavioral state.

Keywords: AMPA receptors; CP: Neuroscience; calcium imaging; chemogenetics; inhibition; microcircuit; normalization; pharmacology; visual cortex.

Figure 1.. Cell-type-specific in vivo pharmacology of AMPARs with YM90K^DART

(A) Schematic of cell-type-specific pharmacology with YM90K^DART. HTP: Halo-tag protein. (B) Schematic of circuit manipulation. (C) Average EPSCs in an example simultaneously recorded pyramidal cell (left) and SST cell (right) before (black) and after (blue) infusion of 300 nM YM90K^DART. (D) Summary of average (filled circles) and individual (light lines) EPSC amplitudes normalized to control (Ctrl) for pyramidal (black) and SST (red) cells in YM90K^DART and 10 μM NBQX (n = 6 pairs). Error is SEM across cells. (E) Schematic of cranial window and infusion cannula (left), and wide-field imaging of the calcium indicator GCaMP8s (middle) and flex-dTomato-HTP (right). Scale bar, 1 mm. (F) Alexa 647^DART (1:10 with YM90K^DART) capture before (left), immediately after (middle), and 19 h after (right) infusion for mouse in (E). (G) Left, schematic of experimental setup. Example two-photon imaging field of view of GCaMP (green) and HTP (red) expression in control (middle) and after YM90K^DART infusion (right) for mouse in (E). White triangles highlight example cells identifiable across sessions. Scale bar, 200 μm. See also Figure S1.

Figure 2.. The effect of blocking AMPARs on SST cells depends on stimulus strength

(A) Grand average time courses for ⁺HTP SST (left, solid lines) and ⁻HTP putative pyramidal cells (right, dotted lines) before (black) and after (blue) YM90K^DART infusion, in response to preferred-direction gratings (horizontal black bar) at 25% (top), 50% (middle), and 100% (bottom) stimulus contrast, during stationary epochs. Shaded error is SEM across cells. (B) Mean response during the stimulus period in control and after YM90K^DART for all SST (left) and pyramidal (right) cells. Histogram is the probability distribution of ratios of responses in control and after YM90K^DART. Dashed line is unity line. p values are from paired t test. (C) Mean response during stimulus period, for SST cells (left) and pyramidal cells (right) before (black) and after (blue) YM90K^DART infusion, at each contrast. Error is SEM across cells. p values are from two-way ANOVA: main effect of contrast (black), drug (blue), and interaction (gray). n.s., not significant; **p < 0.01; ***p < 0.001. See also Figure S2.

Figure 3.. Decreased suppression of SST cells by YM90K^DART is consistent with an increase in the role of SST cells for stabilization

(A) Normalized difference ($\frac{{\mathsf{mean}}_{\mathsf{DART}}-{\mathsf{mean}}_{\mathsf{control}}}{{\mathsf{STD}}_{\mathsf{control}}}$) of stimulus response for SST (left) and pyramidal cells (right) as a function of contrast in YM90K^DART. Gray circles are individual cells; boxplots illustrate median, 25% and 75% quartiles. Significance refers to pairwise F tests for variance. (B) Fraction of SST (left) and pyramidal (right) cells that are suppressed (top, cyan) or facilitated (bottom, magenta) by more than 1 std of their control response at each contrast. (C and D) Same as (A) and (B), for YM90K^PEG. n.s., not significant; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001. See also Figure S2.

Figure 4.. SST cells weakly correlated with the local network are more strongly suppressed by YM90K^DART

(A) Mean-subtracted trial-by-trial responses for two SST cells and concurrently recorded pyramidal cells. Each data point represents a single trial. Fit line is from a linear regression; R is the Pearson’s correlation. (B) Grand average time courses for SST cells before (black) and after (blue) YM90K^DART separated into those weakly (left) and strongly (right) correlated to pyramidal activity, during stationary epochs in response to preferred-direction gratings at 25% contrast. Shaded error is SEM across cells. (C) Mean response during stimulus period, for SST cells weakly (left) or strongly (right) correlated to pyramidal activity, at each contrast in control (black) and after YM90K^DART (blue). Error is SEM across cells. p values are from two-way ANOVA: main effect of contrast (black), drug (blue), and interaction (gray). (D) Scatter of noise correlation between each SST cell and the neighboring pyramidal cell population and the normalized difference at 25% contrast. Blue line and p value are from linear regression. (E) Slope of the correlation in (D) for each contrast. Error is confidence interval. n.s., not significant; *p < 0.05; **p < 0.01. See also Figure S3.

Figure 5.. The effect of blocking AMPARs on SST cells depends on the behavioral state

(A) Grand average time courses for SST cells before (black) and after (blue) YM90K^DART during stationary (left) or running (right) epochs, at each contrast. All cells are matched across behavioral states and contrasts. Shaded error represents SEM across cells. (B) Mean response during the stimulus period in control and after YM90K^DART for matched SST cells during running trials. Histogram is the probability distribution of ratios of responses in control and after YM90K^DART. Dashed line is unity line. p values are from paired t test. (C) Mean response during stimulus period, for SST cells during stationary (left) or running (right) epochs, at each contrast. Error is SEM across cells. p values are from two-way ANOVA: main effect of contrast (black), drug (blue), and interaction (gray). (D–F) Same as (A)–(C), for pyramidal cells. (G) Fraction of SST cells suppressed (left, cyan) or facilitated (right, magenta) by more than 1 std of their control response during stationary (light) or running (dark) epochs. p values are from chi-squared test. (H) Same as (G), for pyramidal cells. n.s., not significant; *p < 0.05; **p < 0.01; ***p < 0.001. See also Figure S4.

Figure 6.. A theoretical framework for network stabilization by SST cells

(A) Schematic of the four-cell (left) and reduced two-cell (right) model. (B_i–iii) Schematic of ${\mathsf{r}}_{\mathsf{E}}$ nullcline (dashed black), ${\mathsf{r}}_{\mathsf{S}}$ nullcline in control (solid black), and ${\mathsf{r}}_{\mathsf{S}}$ nullcline after a 50% reduction in $W_{SE}$ (blue) when the slope of the ${\mathsf{r}}_{\mathsf{E}}$ nullcline is negative (B_i), positive (B_ii), and steeply positive (B_iii). Arrows illustrate the shift in stability points (gray dots), and therefore the change in ${\mathsf{r}}_{\mathsf{E}}$ and ${\mathsf{r}}_{\mathsf{S}}$ after decrease in $W_{SE}$. (C) Network stability in the space defined by ${\tilde{W}}_{EE}$ (effective recurrent excitation among $\mathsf{E}$ cells) and ${\tilde{W}}_{ES}$ (effective inhibition of $\mathsf{S}$ to $\mathsf{E}$). Gray arrows illustrate how effective weights in ${\tilde{W}}_{EE}\times {\tilde{W}}_{ES}$ space change when stimulus intensity is increased. (D_i–iii) Simulated activity of pyramidal (dashed lines) and SST cells (solid lines) in response to a visual stimulus (thick black line) in each region of the space defined in (C) and corresponding to the nullclines illustrated in (B_i–iii). See also Figure S5.

Figure 7.. Paradoxical effects indicate the necessity of SST cells for network stabilization

(A) Cost of the best fit for each model. (B) Akaike information criterion (AIC) values for each model. (C) Empirical (dark data points, mean ± SEM from Figures 5B–5D) and simulated (light lines) responses of SST (left) and pyramidal (right) cells to increasing contrast, in stationary (top) or locomotion (bottom) states in control (gray) and after YM90K^DART (light blue). (D) Schematic of changes to weights to fit changes from stationary to running. Weights in red can change across state in the VIP model. Line thickness is proportional to weight change. (E) Position of model best fit parameters at each contrast (shading) and behavioral state (circles, stationary; triangles, running) in the phase space from Figure 5. Instability line (red) corresponds to the high-contrast, running condition. See also Figure S6.