Vision and Locomotion Shape the Interactions between Neuron Types in Mouse Visual Cortex

Abstract

Cortical computation arises from the interaction of multiple neuronal types, including pyramidal (Pyr) cells and interneurons expressing Sst, Vip, or Pvalb. To study the circuit underlying such interactions, we imaged these four types of cells in mouse primary visual cortex (V1). Our recordings in darkness were consistent with a "disinhibitory" model in which locomotion activates Vip cells, thus inhibiting Sst cells and disinhibiting Pyr cells. However, the disinhibitory model failed when visual stimuli were present: locomotion increased Sst cell responses to large stimuli and Vip cell responses to small stimuli. A recurrent network model successfully predicted each cell type's activity from the measured activity of other types. Capturing the effects of locomotion, however, required allowing it to increase feedforward synaptic weights and modulate recurrent weights. This network model summarizes interneuron interactions and suggests that locomotion may alter cortical computation by changing effective synaptic connectivity.

Keywords: circuit; disinhibition; inhibition; inhibition stabilized network; interneurons; locomotion; neural field model; primary visual cortex; recurrence; surround suppression.

Figure 1

Genetic Targeting and Activity Statistics Identify Pyr, Pvalb, Vip, and Sst Cells in the Awake Cortex (A₁) Experimental setup showing the air-suspended ball surrounded by the three screens for stimulus presentation. (A₂) Green fluorescence from an Emx1-Cre mouse expressing GCaMP6m via virus injections. (A₃) Normalized fluorescent trace from a representative Pyr neuron. Blue shading above axes represents periods of locomotion (>1 cm/s). (A₄) Histogram of fluorescence values for the example neuron in (A₃). The number indicates the skewness of the distribution. (A₅) Distribution of skewness values over all Pyr neurons. (B₁) Green fluorescence from a mouse expressing GCaMP6 following virus injection. Scale bars, 100 μm. (B₂) Red fluorescence from the recordings in (B₁), indicating tdTomato expression in Pvalb neurons. (B₃ and B₄) Same as (A₃) and (A₄) for a representative Pvalb neuron. (B₅) Same as (A₅) for all Pvalb neurons. (C) Similar analysis for Vip cells. (D) Similar analysis for Sst cells. (E₃ and E₄) Normalized fluorescent traces from an unlabeled neuron recorded simultaneously with the Sst example in (D₃) and (D₄). (E₅) Distribution of skewness values over all unlabeled neurons. Unlabeled cells above a skewness threshold of 2.7 (dashed vertical line) are classified as putative Pyr (E₅).

Figure 2

The Effects of Locomotion on Baseline Activity Depend on Cell Type and Depth (A₁) Fluorescence of representative L2/3 pyramidal neuron (top) and simultaneous running speed trace (bottom). (A₂) Correlation coefficient of recorded pyramidal cells with running speed plotted versus cell depth. Circles represent cells with significant correlations at p < 0.05 (shuffle test); dots represent cells with insignificant correlations. For clearer visualization, we plotted only a random subsample of 800 of genetically identified Pyr cells. Dashed line represents fitted dependence of correlation versus depth. Black circle indicates example cell shown in (A₁). (A₃) Histogram of correlation coefficients of all pyramidal cells. Solid bars indicate significant correlations at p < 0.05 (shuffle test). Values left and right of the histogram represent the percentage of cells with a negative or positive correlation, respectively. (B) Similar analysis for Pvalb neurons. The two traces in (B₁) (top) show fluorescence traces of representative Pvalb cells of upper and lower L2/3 (blue and cyan, respectively). The average correlation with speed was slightly negative (${\rho }_{gray}=-0.05\pm 0.03$, SE) among Pvalb cells in superficial L2/3 (depth < 300 μm, n = 843) and weakly positive (${\rho }_{gray}=0.11\pm 0.04,\text{ SE},\text{ n }=831$) in deeper L2/3. (C) Similar analysis for Vip cells. (D) Similar analysis for Sst cells.

Figure 3

Locomotion Modulates Sensory Responses Differently from Baseline Activity (A₁) Visual responses of representative L2/3 pyramidal neuron (top) and simultaneous running speed trace (bottom) for the same cell as in Figure 2A1. In the top plot, dots represent the size of visual responses relative to the mean response to the presented stimulus, while the continuous line represents a smoothed interpolation of these points. (A₂) Average increased visual responses of recorded pyramidal cells by locomotion (modulation of visual responses by locomotion) plotted versus cell depth. Continuous line represents fitted dependence of correlation versus depth. Black circle indicates example cell shown in (A₁). (A₃) Histogram of modulation of visual responses by locomotion for all pyramidal cells. Solid bars indicate significant correlations at p < 0.05 (t test). Values left and right of the histogram represent the percentage of cells with a negative or positive modulation, respectively. (A₄) Modulation of spontaneous activity by locomotion as a function of modulation of visual responses by locomotion. Each point corresponds to a different neuron. Black circle indicates example cell shown in (A₁) and Figure 2A₁. (B) Similar analysis for Pvalb neurons. The two traces in (B₁) (top) show fluorescence traces of the same representative Pvalb cells of upper and lower L2/3 (blue and cyan, respectively) as in Figure 2B₁. (C) Similar analysis for Vip cells with the example cell in (C₁) being the same as in Figure 2C₁. (D) Similar analysis for Sst cells with the example cell in (D₁) being the same as in Figure 2D₁.

Figure 4

Locomotion Increases Sst Cell Responses to Large Stimuli and Vip Cell Responses to Small Stimuli (A₁) Responses of a representative Pyr neuron. Black curves show trial-averaged response in the stationary (dashed line) and locomotion (continuous line) conditions. Panels show responses to stimuli of diameter 5° (left) and 60° (right). Gray shaded regions indicate the 1 s stimulus presentation period. (A₂) Size-tuning curve for this example cell. Solid line, locomotion; dashed line, stationary. Error bars correspond to standard error. (A₃) Size-tuning curve averaged over Pyr cells after normalization. Solid line, locomotion; dashed line, stationary. Error bars correspond to standard error. (A₄) Scatterplot showing change with locomotion of normalized responses to large stimuli (diameter 60°; y axis) and to small stimuli (diameter 5°; x axis). Circles represent cells whose responses have a significant interaction between size and locomotion (multi-way ANOVA over stimuli of diameter 5° and 60° only), squares represent cells that did not have a significant interaction but did have a significant effect of locomotion, and dots represent cells with no significant effect of locomotion. Arrow marks example cell shown in (A₁) and (A₂); square marks mean response. Numbers above and to the right of the dashed diagonal represent the percentage of cells with a significant positive or negative interaction between stimulus size and running speed. (B–D) Similar analysis for Pvalb (B), Vip (C), and Sst (D) neurons.

Figure 5

The Relationship of Interneurons with Pyr Cells Is Linear in Pvalb Cells and Nonlinear in Vip and Sst Cells (A₁ and A₂) Summed activity of Pvalb population versus Pyr population in the gray screen condition during stationary periods (A₁) and during locomotion (A₂). Each circle represents the simultaneous normalized value of the excitatory and the inhibitory populations at one time point. Dashed line indicates linear regression estimate of signal correlation. (A₃ and A₄) Average stimulus response of Pvalb population versus average response of Pyr population in a typical experiment during stationary periods (A₃) and during locomotion (A₄). Each point represents a response to a stimulus, with larger circles representing larger stimuli. Dashed line represents nonlinear interpolation of the Pvalb and Pyr size tuning obtained from their size-tuning curves. (B and C) Same as (A), but for Vip (B) and Sst (C) interneurons. Note the nonlinear signal correlations. (D₁ and D₂) Summary plots of spontaneous correlations during stationary periods (D₁) and during locomotion (D₂). Error bars correspond to standard error. (D₃ and D₄) Linear signal correlation (Pearson correlation coefficient) between the Pyr population and the three classes of interneurons averaged across all experiments during stationary periods (D₃) and during locomotion (D₄). Error bars correspond to standard error. (E₁) Same plot represented in (C₄) showing the characteristic angles used to illustrate the nonlinear relationship between each interneuron class and Pyr mean visual responses. Circles with black outline indicate the minimum size (diameter 5°, black filled), maximum sizes (diameter 60°, red filled), and the Pyr cells’ preferred size (white filled). θ₁ is the angle relative to the horizontal axis of the line joining the response to stimuli of diameter 5° and the preferred stimulus for Pyr neurons. θ₂ is the angle between the latter line and the line joining the response at the Pyr cells’ preferred size to the response to stimuli of diameter 60°. (E₂ and E₃) Angle θ₂ versus θ₁ as defined in (E₁) for each experiment during stationary periods (E₂) and during locomotion (E₃). In E₃, arrows indicate examples in (A₃)–(C₃); in E₄, arrows indicate examples in (A₄)–(C₄).

Figure 6

A Recurrent Network Model Accurately Predicts the Visual Responses of Each Cell Type (A₁) Input synapses received by Pyr cells and their modulation by locomotion. Pyr cells receive feedforward visual excitatory inputs (e.g., from thalamus or other cortical layers) and subtractive inhibition from Pvalb and Sst cells. Locomotion increases the synaptic weights from external visual inputs to Pyr cells (filled circles) while decreasing recurrent Pyr-to-Pyr connections (empty circles). Pyr cells integrate from a broad pool of Sst cells (large Gaussian curve). Gray indicates connections that could be mathematically integrated out in the model when computing effective connections. (A₂–A₅) Model fit (black curves) of the size-tuning data during stationary periods (A₂ and A₃) and locomotion (A₄ and A₅), visualized for centered cells (offset = 0, A₂ and A₄) and off-center cells (offset = 20, A₃ and A₅). Error bars correspond to standard error. (B₁) Input received by Pvalb cells. Locomotion boosts feedforward synaptic weights to Pvalb neurons while decreasing recurrent Pvalb-to-Pvalb synaptic weights. (B₂–B₅) Same as (A₂)–(A₅), but for Pvalb cells. (C₁) Input received by Vip cells. Vip cells receive divisive inhibition from Sst cells; no modulation of these synapses by locomotion is required to obtain a good fit. (C₂–C₅) Same as (A₂)–(A₅), but for Vip cells. (D₁) Input received by Sst cells. In addition to inputs from Pyr and Vip cells, Sst cells receive a feedforward input that we propose to originate from thalamus or other cortical layers. As with Pyr and Pvalb cells, the strengths of these synapses are boosted by locomotion. (D₂–D₅) Same as (A₂)–(A₅), but for Sst cells.