Functional specificity of recurrent inhibition in visual cortex

Abstract

In the neocortex, neural activity is shaped by the interaction of excitatory and inhibitory neurons, defined by the organization of their synaptic connections. Although connections among excitatory pyramidal neurons are sparse and functionally tuned, inhibitory connectivity is thought to be dense and largely unstructured. By measuring in vivo visual responses and synaptic connectivity of parvalbumin-expressing (PV+) inhibitory cells in mouse primary visual cortex, we show that the synaptic weights of their connections to nearby pyramidal neurons are specifically tuned according to the similarity of the cells' responses. Individual PV+ cells strongly inhibit those pyramidal cells that provide them with strong excitation and share their visual selectivity. This structured organization of inhibitory synaptic weights provides a circuit mechanism for tuned inhibition onto pyramidal cells despite dense connectivity, stabilizing activity within feature-specific excitatory ensembles while supporting competition between them.

Keywords: cortical circuits; inhibitory neurons; neuronal connectivity; visual cortex.

Figure 1. Heterogeneity of synaptic strength of PV+ neuron connections

(A) Three possible motifs of synaptic strength of PV+ neurons. (B) Rates of pyramidal/PV+ neuron connectivity. (C) Distribution of the strength of excitatory connections from pyramidal cells onto PV+ neurons. (D) Distribution of the strength of inhibitory connections from PV+ neurons onto pyramidal cells. (E) Example recording of a PV+ neuron reciprocally connected to two pyramidal cells. The PV+ neuron provides stronger inhibition to the pyramidal cell that provides it with stronger excitatory input. (F) EPSP and IPSP strengths are correlated for reciprocally connected PV+/pyramidal neuron pairs. Black line: best fit regression line of log-IPSP magnitude against log-EPSP; gray shading: 95% confidence interval for the regression line estimated from bootstrap resampling; R and p are Pearson correlation and its p value, respectively. Cell pairs in (E) are highlighted. (G) In recordings with multiple pyramidal neurons reciprocally connected to a PV+ cell, the correlation between EPSP and IPSP magnitude persists after controlling for slice quality by normalizing each by the geometric mean of EPSP/IPSP strength in the recording. Notation as in (F). p value was estimated using a shuffling procedure (see STAR Methods). See also Figures S1 and S2.

Figure 2. Response properties of PV+ neurons

(A and B) Visual responses of example PV+ (A) and PV− (B) neurons to drifting sinusoidal gratings. Traces show responses at the preferred direction (gray, single trials; black, mean response). Color maps show mean responses across spatial and temporal frequencies and directions during the moving phase of the grating. (C) PV+ cells are less selective than PV− cells (p = 1.2 × 10⁻⁶², rank-sum test), quantified by skewness of their response distributions across stimulus types. Triangles indicate medians. (D) PV+ cells are broadly tuned to orientation; FWHM, full width at half maximum response (p = 2.3 × 10⁻⁵⁹, rank-sum test). N.T., cells untuned for orientation. (E) PV+ cells are less direction selective than PV− cells (p = 9.2 × 10⁻²⁸, rank-sum test). (F and G) PV+ cells are more broadly tuned to spatial (C, p = 7.7 × 10⁻³⁶, rank-sum test) and temporal (D, p = 3.0 × 10⁻⁸, rank-sum test) frequency than PV− cells.

Figure 3. Response similarity of PV+/PV− neuron pairs is heterogeneous

(A) Tuning properties and response similarity of example simultaneously recorded PV+ and PV− neurons. Response similarities < 0 are not shown. (B) Mean fluorescence image and spatial location of example neurons shown in (A). Note some neurons are located in other imaging planes. Scale bars, 100 μm. (C) Distribution of response similarity of PV+/PV− neuron pairs (234,994 pairs between 255 PV+ and 8,427 PV− neurons). (D–G) Two-dimensional histograms showing the relationship between response similarity and differences in preferred spatial frequency (D), temporal frequency (E), orientation (F), or distance between PV+/PV− cell pairs. (H) Correlation coefficients of response similarity and differences in preferred spatial frequency, temporal frequency, orientation, or distance.

Figure 4. Response similarity predicts the strength of PV+ neuron connections

(A) Example PV-pyramidal neuron pair shown in vivo (left), in the brain slice (center), and during whole-cell recording (right). The in vivo image was generated by resampling the in vivo z stack to match to the coordinate frame of the in vitro volume. (B) Example in vivo calcium traces (left) of the cells in (A) and in vitro current-clamp recordings of evoked action potentials in the PV+ cell (top right) and IPSPs in the pyramidal cell (bottom right). (C) Left: scatterplots of dF/F trial-average responses on individual imaging frames across the stimulus ensemble for 3 reciprocally connected PV/pyramidal cell pairs. Center: visual tuning (blue, PV+ cell; red, pyramidal cell). Polar plots show normalized responses at spatial and temporal frequency that evoked the highest mean response. Colormaps show the normalized mean spatial and temporal frequency responses across directions. Right: post-synaptic potentials. Cell pair 1 is depicted in (A) and (B). (D and E) Frequency of inhibitory connections from PV+ onto pyramidal cells (D) and excitatory connections from pyramidal cell onto PV+ cells (E) does not depend on response similarity. p value corresponds to the slope coefficient of logistic regression of connection probability against response similarity. (F) Relationship of the strength of inhibitory connections from PV+ onto pyramidal cells and response similarity. Cell pairs shown in (C) are highlighted. Black line: best fit regression line of log-IPSP magnitude against response similarity; gray shading: 95% confidence interval for the regression line estimated by bootstrap resampling; R and p are Pearson correlation and its p value, respectively. (G) Relationship of the strength of excitatory connections from pyramidal onto PV+ cells and response similarity. Notation as in (F). (H) Similarity of trial-average responses but not selectivity for individual visual features predicts the strength of inhibitory connections from PV+ cells onto pyramidal cells (blue) and excitatory connections from pyramidal cells onto PV+ cells (red). Error bars are 95% confidence intervals. **p < 0.01, *p < 0.05. See also Figures S3–S7.

Figure 5. Simulations of networks with specific inhibitory connectivity demonstrate their role in network stability and recapitulate in vivo observations

(A–E) Networks with different distributions of excitatory and inhibitory synaptic strength. Top: connectivity matrices for E-to-E, E-to-I, and I-to-E connections; bottom: responses of example neurons to stimuli spanning feature 1 at the preferred value for the other stimulus dimension. (A) Network with uniform excitatory and inhibitory connectivity. Responses of example neurons (B) Specific excitatory connectivity between neurons co-tuned for both stimulus features (s_E = 0.48, see STAR Methods) enables amplification of neuronal responses. (C) Further increase in the specificity of excitatory connections (s_E = 0.55) leads to network instability. (D) Specific inhibitory connectivity (s_I = 0.8) between neurons co-tuned for both stimulus features balanced excitatory specificity, restoring network stability. (E) On the other hand, lateral inhibition, whereby inhibitory neurons strongly target excitatory neurons with different stimulus preferences (s_I = − 0.8), promotes instability. (F) In the network in (D), measures of overall response similarity are better predictors of connection strength of inhibitory neurons than individual feature preferences. Error bars, 95% confidence interval computed by subsampling connections using the sample sizes as in the experiments in Figure 4H. (G and H) Excitatory and inhibitory inputs received by excitatory cells are correlated in networks with specific excitatory connections, independent of the presence of specific inhibitory connectivity. (I) Inhibition received by excitatory cells in the network in (D) is co-tuned with excitation for individual stimulus features.

Figure 6. Specific but broad inhibitory connectivity supports neuronal competition

(A and B) Stimulation of a small cohort of excitatory neurons (11 of 4,000) in the network shown in Figure 5C leads to facilitation of the majority of inhibitory neurons and suppression of the majority of excitatory neurons (B). (C–F) In the subset of excitatory neurons with similar tuning to the stimulated population, stimulation boosts activity (C and E) driven by excitatory inputs exceeding inhibition (D). In other cells, stimulation suppresses firing (C and F) driven by strong multi-synaptic inhibitory inputs in the absence of excitatory inputs (D).