Vision-dependent specification of cell types and function in the developing cortex

Abstract

The role of postnatal experience in sculpting cortical circuitry, while long appreciated, is poorly understood at the level of cell types. We explore this in the mouse primary visual cortex (V1) using single-nucleus RNA sequencing, visual deprivation, genetics, and functional imaging. We find that vision selectively drives the specification of glutamatergic cell types in upper layers (L) (L2/3/4), while deeper-layer glutamatergic, GABAergic, and non-neuronal cell types are established prior to eye opening. L2/3 cell types form an experience-dependent spatial continuum defined by the graded expression of ∼200 genes, including regulators of cell adhesion and synapse formation. One of these genes, Igsf9b, a vision-dependent gene encoding an inhibitory synaptic cell adhesion molecule, is required for the normal development of binocular responses in L2/3. In summary, vision preferentially regulates the development of upper-layer glutamatergic cell types through the regulation of cell-type-specific gene expression programs.

Keywords: binocular vision; cell types; critical period; inhibitory synapses; layer 2/3; single-nucleus RNA-seq; visual cortex.

Figure 1.. snRNA-seq profiling of V1 during postnatal development

A. Schematic of the mouse visual system. Primary visual cortex (V1). Surrounding higher visual areas: A, anterior; AL, anterolateral; AM, anteromedial; LI, laterointermediate; LM, lateromedial; P, posterior; PM, posteromedial; POR, postrhinal; RL, rostrolateral; TEA, temporal anterior areas. B. Experimental workflow of snRNA-seq profiling of V1 at six postnatal ages. C. Cellular taxonomy of V1. D. UMAP visualization of V1 transcriptomic diversity during postnatal development. Dots correspond to cells and distances between them reflect degrees of transcriptomic similarity. Central panel shows cells from all six ages colored by subclass identity (Table S1). Peripheral panels show cells from different ages, colored by type identity determined via clustering. Data from each age and class were analyzed separately and then merged for visualization purposes.

Figure 2.. Transcriptomic diversity of V1 glutamatergic neurons during postnatal development

A. Schematic of glutamatergic neurons in V1 arranged in layers L1-L6. B. Tracks plot showing subclass-specific markers (rows) in glutamatergic neurons (columns), grouped by subclass (e.g., L2/3). 1000 randomly selected cells from each subclass were used for plotting. Scale on the y-axis (right), normalized, log-transformed transcript counts in each cell. Ccbe1, a L2/3 marker, and Cux2, a L2/3/4 marker, are highlighted. C. The proportions of glutamatergic subclasses are stable with age despite significant variation in the number of cells profiled (Table S2). D. Coronal section through V1 analyzed by fluorescent in situ hybridization (FISH) at P21. Ccbe1 is selective for L2/3 glutamatergic neurons. Cux2 is expressed in L2/3 and L4 glutamatergic neurons and in inhibitory neurons and non-neuronal cells (see Figure S2B for other ages). Scale, 50 μm. E. Transcriptomic similarity identifies temporal associations among V1 glutamatergic neuron types across ages. Sankey diagram computed using a supervised classification approach. Nodes, individual V1 glutamatergic neuron types at each age (as in Figure 1D); edges, colored based on transcriptomic correspondence. F. Adjusted Rand Index (ARI) values quantifying temporal correspondence of glutamatergic types between each pair of consecutive ages based on transcriptomic similarity. Individual bars denote layers. ARI ranges from 0 (no correspondence) to 1 (perfect correspondence). Bar heights, mean ARI computed across pairs of consecutive ages; error bars, standard deviation; ***, P<0.0001 (one-way ANOVA) for L2/3 and L4 against L5 and L6. G. Types in L2/3 and L4, but not L5 and L6, are sensitive to changes in clustering resolution. Glutamatergic neurons at each age are re-clustered at different values of the resolution parameter (x-axis), and the results are compared with the base case corresponding to resolution = 1 (STAR Methods). Line plots, mean ARI values for each layer (colors); error bars, standard deviation across ages.

Figure 3.. Anatomical and transcriptomic maturation of L2/3 glutamatergic neuron types

A. UMAP plots of L2/3 glutamatergic neuron types across ages. B. Dot plot showing expression patterns of L2/3 type-specific genes (rows and colors) across L2/3 neuron types arranged by age (columns). C. FISH images showing type markers Cdh13, Trpc6, and Chrm2 within L2/3 at P8. Vertical colored bars, sublayers expressing the indicated markers; arrows, large cells expressing Cdh13 are not excitatory neurons; they are a subset of inhibitory and non-neuronal cells. Scale, 50 μm. D. Same as C, at P38. E. Pseudo-colored representation of Cdh13, Trpc6, and Chrm2 expression in L2/3 cells at six ages. Cells are colored based on expression levels of one or more of these markers. Each panel is an overlay of five or six images of V1 from three mice. Pial to ventricular axis is oriented horizontally from left to right within each panel. Total number of cells analyzed: P8, 2324; P14, 1142; P17, 1036; P21, 1038; P28, 653; and P38, 1034. Scale bar, 100 μm. Panels E and F are rotated relative to Panels C and D. “Top” and “Bottom” are indicated. F. Line tracings quantifying the number of cells per bin at each position along the pial to ventricular axis corresponding to panel E. 0 on the x-axis, region of L2/3 closest to pia. 14 bins were used over the depth of L2/3. G. Relative proportions of cells within each expression group defined in panel E quantified using FISH data. H. Same as G using snRNA-seq data.

Figure 4.. Visual experience is required to maintain L2/3 glutamatergic neuron types

A. Schematic of experiments. Data collected from three rearing conditions: Dark-reared between P21-P28 (P28DR) and P21-P38 (P38DR), and dark-reared between P21-P28 followed by 8 hrs (P28DL). B. UMAP plots of transcriptomic diversity in P28DR, P38DR, and P28DL. Clusters that match 1:1 to normally-reared (NR) types in Figure 1D are labeled. This was not possible for all L2/3 and two L4 clusters, which correspond poorly to NR types. We therefore provisionally labeled these clusters L2/3_1, L2/3_2, L2/3_3, L4_1, and L4_2. C. Adjusted Rand Index (ARI) quantifying transcriptomic similarity within each layer (x-axis) between glutamatergic clusters observed in dark-reared mice and types observed in normally-reared (NR) mice. Colors correspond to comparisons as indicated. D. Expression of L2/3 type markers (columns) in NR, DR, and DL types and clusters (rows) at P28 and P38. E. Same as panel D for L5. DR and DL clusters are labeled based on their tight transcriptomic correspondence with NR types (Figure S5F, G). F. FISH images showing expression of L2/3 markers in NR, DR, and DL at P28. Arrows, inhibitory neurons expressing Cdh13. The level of Cdh13 is modestly repressed by vision. Scale, 50 μm. G. Pseudo-colored representation of Cdh13, Trpc6, and Chrm2 expression in NR, DR, and DL mice at P28 and P38. Each plot is an overlay of 5–6 images of V1 from three mice. Pial to ventricular axis is oriented horizontally from left to right within each panel. Total number of cells analyzed: P28NR, 653; P28DR, 989; P28DL, 1732; P38NR, 1034; and P38DR, 1177). H. Cells per bin at each position along the pial to ventricular axis corresponding to panel G. 0 on the x-axis, region of L2/3 closest to pia. 14 bins were used over the depth of L2/3. I. Proportions of L2/3 cells within each expression group defined in panel G quantified using snRNA-seq data J. Same as I using FISH data.

Figure 5.. Vision is required to establish L2/3 glutamatergic neuron types

A. Schematic of experiments. B. FISH images of L2/3 markers in normally-reared (NR) and dark reared (DR) mice at P17. Arrows, inhibitory neurons expressing Cdh13. Scale, 50 μm. C. Pseudo-colored representation of Cdh13, Trpc6, and Chrm2 expression in L2/3 cells. Each plot is an overlay of 6 images of V1 from three mice. Cells quantified: P17NR, 1036; P17DR, 1411. D. Line tracings quantifying cells per bin at each position along the pial to ventricular axis corresponding to panel C. 0 on the x-axis, L2/3 region closest to pia. 14 bins were used over the depth of L2/3. E. Proportions of cells in each expression group defined in panel C quantified using FISH data.

Figure 6.. Continuous variation of L2/3 neuron types and vision-dependent gene gradients implicated in wiring

A. Heatmap of L2/3 type-specific genes with graded expression in normally-reared mice (NR). This is disrupted in dark-reared mice (DR) and partially recovered by exposing DR mice to light for 8 hrs (DL). For the full set of L2/3 type-specific genes grouped by expression pattern, see Figure S6A. Genes satisfying criteria in panels B and C (see text) are indicated in red lettering. Two of the three L4 cell types also exhibit graded expression differences (see Figure S6L–M). B. Temporal regulation of cell surface molecules (CSMs) in panel A. Red print, genes upregulated during the classical critical period (P21-P38), downregulated in DR, and upregulated in DL. C. Same illustration as panel B across the conditions P28NR, P28DR, and P28DL. D. Schematic of MDGA1 and IGSF9B interactions with NLGN2 at synapses. MDGA1 prevents NLGN2 interaction with NRXN presynaptically. IGSF9B binds homophilically and interacts with S-SCAM postsynaptically as does NLGN2. E. FISH images of Igsf9b mRNA over time in V1. Three animals per time point, six images per animal. Scale, 20 μm. (Right) Box plot quantifying expression. Wilcoxon Rank Sum Test, **** P <0.0001. Cells quantified: P8,1191; P14,1011; P17, 1389; P21, 1729; P28, 1277; and P38, 1588. F. FISH images showing that dark rearing decreases Igsf9b expression in L2/3, and eight hours of light restores expression. Scale, 50 μm. (Right) Box plot quantifying expression. Three animals imaged per age and condition combination. Cells quantified: P28NR, 1290 cells; P28DL, 1506 cells; P28DR, 1521 cells; P38NR, 1629 cells; and P38DR, 1885. Quantified at 40X. Wilcoxon Rank Sum Test, *** P <0.001. G. FISH quantification of average Mdga1 and Igsf9b expression (y-axis) in glutamatergic cells as a function of distance from the top of L2/3 (x-axis). Shaded ribbons represent standard error of the mean. Cells quantified: P8, 2204; P14, 928; P17, 1037; P21, 1183; P28, 719; and P38, 942. Data from three or four animals at each age. H. Reconstruction of Mdga1 and Igsf9b expression levels averaged across cells based on their inferred L2/3 pseudo-spatial locations in gene expression space (see STAR Methods). Shaded ribbons, standard deviation. I. Same as panel G for P38DR, P28DR, and P28DL. Cell numbers: P38DR, 719; P28DR, 1061; and P28DL, 1053 cells. Data collected from three animals at each time point J. Same as panel H for P38DR, P28DR, and P28DL. Note difference in scale for P28DL to capture the extent of increase in Igfs9b expression.

Figure 7.. Igsf9B is required for vision-dependent maturation of binocular neurons in V1B L2/3

A. Experimental setup for functional analysis. (Top) Schematic of 2-photon (2P) Ca²⁺ imaging using different sinusoidal gratings sequentially presented at 4 Hz. Visual stimuli were presented to each eye separately. The head-fixed mouse was awake on a running wheel. Mice used in this study are WT (Igsf9B^+/+) and KO (Igsf9B^−/−) expressing AAV encoded jGCaMP7f. Panels G, J, K include our unpublished results from NR and DR (dark reared from P22-P36) transgenic mice carrying GCaMP6s expressed in excitatory neurons (from Tan et al., 2020). (Bottom) WT and KO mice were imaged at P21 and P36, the onset and closure of the classical critical period, respectively. Orange, Igsf9b mRNA levels in L2/3 as a function of time. B. Tuning kernel showing response of a single neuron (see Figure S7L) to the contralateral eye. The colors represent response strength (color bar, right) as a function of stimulus orientations (Y-axis) and spatial frequency (log scale; X-axis). C. Response to contralateral (C) and ipsilateral eye (I) of monocular cells. Kernels for each neuron were normalized to the peak inferred spiking. D. As in C, but for matched (top) and unmatched (bottom) binocular neurons. ΔOrientation, the difference in orientation preference between the two eyes. E. (Left) proportions of binocular neurons in WT and KO mice at P21. Each point is from a single imaging plane. Mean and standard deviation, black dots and lines. Mann-Whitney U test. (Right) ΔOrientation of binocular neurons in WT (4 mice, 761 cells) and KO (3 mice, 619 cells) mice at P21. Black horizontal line, median; box, quartiles with whiskers extending to 2.698σ. Mann-Whitney U test. Note the absence of phenotype in binocular neurons at P21. F. As in E but for binocular neurons at P36. WT, 5 mice, 602 cells; KO, 5 mice, 269 cells. G. As in F but for binocular neurons in NR (4 mice, 339 cells) and DR (3 mice, 78 cells) mice. P36 phenotypes in KO and DR mice were similar. The difference in proportion between the WT (panel F) and NR (panel G) likely reflects differences in genetic background or experimental design (i.e., viral versus transgenic expression of GCaMP or differences between GCaMP6s and jGCaMP7f). H. Example of a tuned cell from a WT mouse at P21. Inferred spiking as a function of imaging frames for a neuron with a tuned response. Numbers at the top left indicate imaging frames relative to stimulus onset. For this neuron, the SNR is 3.1, and peak response occurred 5 imaging frames or 323 ms after onset of its optimal stimuli, consistent with the kinetics of jGCaMP7f. I. As in H but for an untuned cell in the same mouse at P21. J. Proportions of tuned neurons in WT and KO mice at P21 and P36, and in NR and DR mice at P36. Each point is from a single imaging plane. Mean and standard deviation, black dots and lines. Mann-Whitney U test. K. (Left) Cumulative distribution of SNR to either eye of all imaged neurons at P21 in WT (4 mice, 3436 neurons) or KO (3 mice, 3457 neurons) mice. Dashed vertical line marks the SNR threshold for visually evoked responses (see STAR Methods). P-value from two-sample Kolmogorov-Smirnov test is shown in the plot. (Middle) As in the left, but for mice at P36 in WT (5 mice, 2698 neurons) and KO (5 mice, 2699 neurons). (Right) As in the middle, but for neurons in NR (4 mice, 1905 neurons) and DR (3 mice, 1188 neurons) mice. Neuronal responses to each eye were measured separately. L. Proportions of tuned neurons as a function of depth in V1B L2/3 in WT and KO mice at P36. Top, middle, and bottom indicate the three imaging planes covering the corresponding sub-laminae within L2/3 in each mouse. Each gray line represents a mouse. Mean and standard error of the mean were shown as black dots and vertical lines. Mann-Whitney U test with Bonferroni correction.