Spontaneous activity in RL sparsifies over development, with more mature activity in S1 compared to V1 and RL.

A. Primary visual cortex (V1), somatosensory cortex (S1), and rostro-lateral area (RL) imaged in sequence over a 100 min recording session. B. Sample field of views of the individual cortices. Scale bar corresponds to 50 µm. C. Sample z-profile traces from the three different cortices (columns) in three mice at different postnatal ages (rows). Vertical scale bar corresponds to 20% ΔF/F0. Horizontal scale bar corresponds to 50 seconds. D-G. Amplitude (D), duration (E), participation rate (F), and event rate (G) as a function of postnatal age. Each dot represents the average value for one animal for one cortical area at the indicated postnatal age, computed across all detected events in that area. N = 9 was the total number of animals included across all ages. Because animals were distributed across postnatal ages and some points overlap visually, fewer than 9 points are visible at any individual age. Regression lines derived from the linear mixed model in H. H. Linear mixed-model estimates for each activity feature, where age and region are included as fixed effects, and mouse ID and sample ID are included as random effects. Values are shown relative to V1 at PN8. The V1 term corresponds to the model intercept, the Age term corresponds to the age-dependent slope in V1, and the S1, RL, Age:S1, and Age:RL terms indicate differences from this reference (see Methods). Stars indicate statistical significance (* p<0.05, ** p<0.01, *** p<0.001). Because intercepts and slopes have different units and scales, full model estimates, confidence intervals, and comparisons using S1 or RL as reference areas are provided in Supplementary Tables S1-S3.

Spontaneous activity in V1, S1, and RL is near-synchronous and spatially organized.

A. Schematic of the wide-field experiments. Cortical surface area encompassing the primary visual (V1) and somatosensory (S1) cortices as well as RL in a PN9 mouse. Higher-order areas of the visual cortex are drawn according to [9]. LM: lateromedial area, AL: anterolateral area, A: anterior area. B. ΔF/F0 traces from 3 ROIs in V1, S1 and RL (shown in A). C. Expanded view of the traces shown in the area marked by a light grey box in B. D. Four ΔF/F0 frames over a period of 900 ms indicated by a dark grey vertical bar in C. E. Functional correlation map of the recording shown in B-D. Pixel values represent Pearson correlation coefficients between each pixel and the areas marked by blue, green and red squares in RL across the recording’s duration. The color of each pixel is the RGB composite of the coefficients of the areas marked by the three colored squares in RL. F. Same as E except that each pixel is assigned to the color channel (red, green or blue) with the highest coefficient of correlation value across the three channels. G. Pearson correlation between activity in somatosensory and visual cortex as a function of temporal lag at PN9. Individual animals are shown in gray and the mean in black. The vertical black line indicates zero lag. Arrows indicate the temporal order of activity at different lags. H. Temporal correlations between V1 and S1. Violin plots where each dot indicates the Pearson correlation coefficient for one 2-minute bin. PN9 N = 6; PN10 N = 5; PN11 N = 4; PN12 N = 2.

Model set-up and assumptions.

A. Schematic of a three-population feedforward model of the developing V1, S1, and RL. B. Example of random initial connectivity matrix between V1 and RL (connectivity matrix for S1 and RL is chosen identically) with a topographic bias denoted by stronger connections along the diagonal inspired by activity-independent mechanisms (see Methods). C. Sample of spontaneous events in V1 (top) and S1 (bottom) across cells as a function of time. Boxes highlight two spatiotemporally correlated events in V1 and S1 (see Methods). D. Example of refinement of connectivity matrix between V1 and RL (top) and S1 and RL (bottom), from t0 to tN. Same axes as in Figure 2B. E. Example of connectivity matrices in steady state from simulations with low (correlation=0.1, left), medium (correlation=0.5, center) to highly (correlation=0.9, right) temporally correlated activity between V1 and S1. F. Schematic to illustrate topography and alignment measures (see Methods). G-I. Topography (G), alignment (H), and percentage of bimodal cells (I) as a function of V1 to S1 correlation. Red dots correspond to the examples from E. Each point represents an independent simulation run in which both the initial connectivity matrix and the stochastic sequence of spontaneous events were resampled to mimic different developmental realizations. J. Excerpt of activity in RL at the beginning (left) and end (right) of the simulation (corresponds to panel D, left and right).

More mature spontaneous events in S1 effectively instruct map alignment between V1 and RL.

A. Three pairs of connectivity matrices (V1 left, S1 right) with increasing (top to bottom) S1 bias. B. Corresponding connectivity matrices in steady-state after plasticity. C-E. For all simulations, V1-S1 correlation of 0.5 was used. C S1 and V1 to RL topography. Different S1-to-RL connectivity biases indicated by color. D. Alignment of V1 and S1 to RL topography as a function of S1-to-RL connectivity bias. Linear fit with R = 0.44, p value< 0.001. E. Percentage of bimodal neurons in RL as a function of S1-to-RL connectivity bias. Linear fit with R = 0.42, p value< 0.001.

A mixture of bimodal and unimodal neurons can optimally decode activity.

A. Schematic of the regression procedure to calculate R2 for reconstructing V1 or S1 activity from RL activity with a non-linear transfer function (a sigmoid) (see Methods). A linear regression is performed using the non-linearly transformed RL activity (dependent variable) and both V1 and S1 activity (independent variables). R2 is calculated from V1 or S1 activity separately using the fitted regression line. B. Reconstruction of V1 or S1 activity (R2) as a function of the percentage of connected cells from the respective cortex over all values of correlation chosen uniformly between 0 and 1 (see Methods). Color indicates topography. C. Reconstruction of V1 and S1 activity (R2) as a function of the percentage of bimodal cells (denoted by color). For all simulations, V1-S1 correlation of 0.5 was used.

Parameters used for simulations.

Linear Mixed Model with V1 as reference.

Values are estimates ± 95% CI width. Significance: * p < 0.05, ** p < 0.01, *** p < 0.001.

Linear Mixed Model with S1 as reference.

Values are estimates ± 95% CI width. Significance: * p < 0.05, ** p < 0.01, *** p < 0.001.

Linear Mixed Model with RL as reference.

Values are estimates ± 95% CI width. Significance: * p < 0.05, ** p < 0.01, *** p < 0.001.

Functional correlation maps of 3 recordings at P9, P10 and P13.

Pixel values represent Pearson correlation coefficients between each pixel and the three seed areas marked by blue, green, and red squares across the recording duration. Seed areas were placed in V1, S1, or RL, as indicated in each example. A-C. the color of each pixel is the RGB composite of the coefficients of the areas marked by the three colored squares. A’-C’. Same correlation values as in the top rows, but each pixel is assigned to the RGB channel with the highest correlation coefficient.

Effects of S1 activity statistics across different S1-to-RL connectivity biases.

S1-to-RL connectivity bias was varied across simulations from 0.05 to 0.5. For all simulations, V1-S1 correlation was fixed at 0.5. A. V1-to-RL and S1-to-RL topography for different S1 inter-event intervals (IEI, in ms; indicated by color). B. V1-to-RL and S1-to-RL topography for different S1 event amplitudes (indicated by color).

Mapping sensory cortex.

A. Imaged cortical surface with sensory areas labeled (bottom). B. Functional correlation map, showing how correlated each pixel is with each of the seed areas shown as red, green and blue squares in V1 (for details see Supplementary Figure S1). Since topographic organization within V1 is monotonous, reversal of correlation distributions indicate areas outside V1, e.g., the red pixels in the top-left corner of the field of view. C. ΔF/F0 representations of individual network events in the imaged area. The extent of these events give a clear indication of the boundaries of V1.