Experimental setup, visual stimulation, and tissue alignment protocol for calcium imaging in V1 of Chrm2tdT x Emx 1-Cre mice.

(a) In vivo image of four injection sites of AAV2/1.hSyn.Flex.GCaMP6f in V1 after head plate implantation, visualized by overlaying brightfield image of surface blood vessels with GFP expression. Inset shows ex-vivo tangential section with GCaMP6f injection sites (green) overlaid with Chrm2tdT expression in L1 (magenta). AL (anterolateral area), LI (laterointermediate area), LM (lateromedial area), P (posterior area), PM (posteromedial area), POR (postrhinal area), V1 (primary visual area). (b) Schematic of two photon calcium recording in V1 of awake headfixed mice running on a wheel, facing a drifting square wave grating stimulus on a monitor (left). Mapping of receptive fields by stimulating with 20°-wide patches of a drifting gratings at one of 24 locations (center). Stimuli used for recording tuning curves to square wave gratings of varying spatial frequencies, temporal frequencies, and orientations (right). The orientation and length of red arrows indicate the direction and speed of stimulus motion. (c-h) Alignment of recorded cells with M2+ patches. (c) Time-averaged in vivo GCaMP6f signals from pyramidal cells showing somas of active V1 L2/3 cells contained in the white box outlined in panel a. Examples of responsive cells are labeled by arrows. (d) Cellular GCaMP6f expression (arrows) in ex vivo paraformaldehyde-fixed 40 µm tangential section at the same location as shown in panel c. (e) Overlay of ex vivo GCaMP6f expression shown in panel d aligned with ex vivo M2+ patches in L1. (f) Image of M2+ patches shown in panel e after high-pass filtering and blurring. (g) Image from panel f with contours outlining 6 intensity quantiles. (h) In vivo GCaMP6f image from panel c overlaid with M2 quantiles from panel g, allowing cells (white) from in vivo recordings to be assigned to M2 quantiles. M2+ patches are colored magenta.

Locomotion modulation and orientation tuning in L2/3 pyramidal cells of V1, aligned with M2+ patches and M2− interpatches of L1.

(a) Tuning curves from example cells for spatiotemporal stimulus parameters. Red traces show tuning curves generated by trials with locomotion (≥ 1 cm/sec). Black curves indicate trials during which mice were immobile (< 1 cm/sec). The leftmost column contains M2− interpatch cells (quantiles 1-2), the rightmost contains M2+ patch cells (quantiles 5-6), the middle column represents cells at the border between M2+ patches and M2− interpatches (quantiles 3-4). Fitted tuning curves are shown with dots indicating responses from individual trials. Inset: trial-averaged time courses of a sample M2− interpatch neuron to its preferred spatial frequency stimulus during locomotion (red; n = 4 trials) and rest (black; n = 6 trials). Error bars show SEM in each time bin. (b) Locomotion Modulation Index (LMI) of visual responses in M2+ patch and M2− interpatch cells shows preferential modulation of M2− interpatch cells. (c) Distribution of Orientation Selectivity Index (OSI) computed from combined trials during locomotion and immobility shows increased selectivity of M2+ patch cells. Error bars represent SEM across cells.

Pairwise correlated activity of L2/3 M2+ patch and M2− interpatch cells in V1.

Pairs were divided into those containing two M2+ patch cells (M2+/M2+, magenta), two M2− interpatch cells (M2−/M2−, green), and one of each type (M2+/M2−, black). (a) Pearson correlation coefficient of ΔF/F signals (mean ± SEM) of cell pairs in darkness at different cell-to-cell distances. Asterisks indicate significant differences between correlation coefficients in M2+/M2+ of M2−/M2− and M2+/M2− cell pairs at different cell-to-cell distances (t-test; ⁎ = p < 0.05, ⁎⁎ = p < 10-2, ⁎⁎⁎ = p < 10-3, ⁎⁎⁎⁎ = p < 10-4). (b) Coefficient of response noise correlation (mean ± SEM) between cell pairs during visual stimulation at different cell-to-cell distances. Asterisks same as in panel a. (c) Comparison of response noise correlations (mean ± SEM) between cell pairs in which both cells were locomotion-responsive and the cell-to-cell distance was > 275 µm. Asterisks indicate significance level (t-test with Bonferroni correction; ns = not significant, ⁎ = p < 10-2, ⁎⁎ = p < 10-3, ⁎⁎⁎ = p < 10-4). (d) Comparison of response noise correlations (mean ± SEM) between cell pairs in which neither cell was locomotion-responsive and the cell-to-cell distance was > 275µm. Note, that the number of pairs in panel d is larger than in c, which includes pairs with one locomotion-tuned and one non-locomotion-tuned cell. Asterisks same as in panel c.

Looped modular like-to-like of connections between V1 and secondary motor area (MOs).

(a) Diagram of flatmap of left mouse cerebral cortex. Boxed area indicates location of image shown in panel b. A (Anterior area), ACC (Anterior cingulate cortex), AL (Anterolateral area), AM (Anteromedial area), LI (Lateral intermediate area), LM (Lateromedial area), MM (Mediomedial area), MOp (Primary motor area), MOs, P (Posterior area), PM (Posteromedial area), POR (Postrhinal area), RL (Rostrolateral area), RSP (Posterior retrosplenial area), SSp (Primary somatosensory area), V1 (Primary visual area). (b) Tangential section through L4 of flatmounted cortex of Ai9 mouse, stained with an antibody against M2AChR (M2) (white). Injection site in MOs (asterisk) of cocktail of viral tracers (AAV2/1-hSyn-EGFP.WPRE.bGH [green], rAAV2-Retro-CAG.Cre [magenta]). (c) Tangential section through L1 of flatmounted V1. White regions are M2AChR immunostained M2+ patches, outlined by white lines. Dark regions that lack M2, represent M2−interpatches, and are outlined by blue lines. (d) Anterogradely AAV-labeled MOs→V1 projections (green) in L1. (e) Overlay of panels c and d, showing preferential input of MOs→V1 axons to M2− interpatches. (f) Retrogradely labeled spine-covered apical dendrites of V1→MOs projecting cells arborizing in L1 of V1. (g) Apical dendrites of retrogradely AAV-labeled V1→MOs projecting neurons in L1 (magenta) preferentially branch in M2− interpatches outlined by white lines. M2+ patches are surrounded by blue lines. (h) Overlay of apical dendrites (magenta) and M2+ patches (white). (i) Overlay of MOs→V1 axon projections (green), apical dendrites of V1→MOs projection neurons (magenta) and M2+ patches (white) in L1 of V1. (j) Labeling density of MOs→V1 axons in different M2 quantiles. Pearson correlation (r), error bars ±SEM. (k) Labeling density of dendrites of V1→MOs projecting cells different M2 quantiles. Pearson correlation (r), error bars ±SEM. (l) Laminar distribution of retrogradely labeled cells bodies shows that most cells are located in L2/3 and exhibit no preferential alignment with M2+ patches (magenta) and M2− interpatches (green). KS test. Shading ±SEM.

Looped clustered like-to-like connections between V1 and LM.

(a) Tangential section through L1 of V1 of Ai9 mouse stained with an antibody against M2AChR (M2). M2+ patches (white) are outlined by white lines, M2− interpatches are indicated by blue lines. (b) Anterogradely AAV-labeled (for tracer, see Figure 4) LM→V1 axons (green) in L1 of V1. (c) Overlay of panels a and b. Axons preferentially terminate in M2+ patches. (d) Apical dendrites of retrogradely AAV-labeled V1→LM-projecting neurons (magenta) in L1 outlined by white lines. M2− interpatches outlined by blue lines. Note that the section is not perfectly parallel to the pial surface. It runs from right to left through L1 and the top of L2, which shows cell bodies (arrows) in M2+ patches and M2− interpatches. (e) Apical dendrites of V1→LM-projecting neurons (magenta) preferentially branch in M2+ patches (white). Retrogradely AAV-labeled cell bodies (arrows). (f) Overlay of panels d and e. Apical dendrites of V1→LM-projecting neurons (magenta) terminate in M2+ patches (white). Retrogradely AAV-labeled cell bodies (arrows). (g) LM→V1 axon projection density in different M2 quantiles of L1. Pearson correlation (r), error bars ±SEM. (h) Labeling density of retrogradely AAV-labeled V1→LM dendrites in different M2 quantiles of L1. Pearson correlation (r), error bars ±SEM. (i) Laminar distribution of retrogradely labeled V1→LM-projecting cells shows that most cell bodies are located in L2/3 and are randomly distributed relative to M2+ patches (magenta) and M2− interpatches (green) in L1. KS test. Shading ±SEM.

Looped modular like-to-like of connections between V1 and PM.

(a) Tangential section through L1 of V1 of Ai9 mouse stained with an antibody against M2AChR (M2). M2+ patches (white) are outlined by white lines, M2− interpatches by blue lines. (b) Anterogradely AAV-labeled PM→V1 axon projections (green) in L1 of V1 terminating in M2− interpatches outlined by blue lines. M2+ patches outlined by white lines. (c) Interdigitation of PM→V1 axon projections (green) and M2+ patches (white). (d) Apical dendrites of retrogradely AAV-labeled V1→PM-projecting cell bodies (magenta) at the L1/2 border of V1, preferentially arborize in M2− interpatches (blue outlines) of L1. M2+ patches are outlined by white lines. (e) Interdigitation of dendrites of retrogradely labeled V1→PM-projecting cells (magenta) and M2+ patches (white). (f) Overlay of panel e with anterogradely labeled axons shown in panel c. (g) Density of PM→V1 axon projections in different M2 quantiles of L1. Pearson correlation (r), error bars ±SEM. (h) Labeling density of retrogradely labeled dendrites of V1→PM-projecting neurons in L2/3 aligned with different M2 quantiles of L1. Pearson correlation (r), error bars ±SEM. (i) Laminar distribution of retrogradely labeled cells bodies shows that most cells are preferentially aligned with M2− interpatches (green) in L1. Magenta indicates cells aligned with M2+ patches. KS test. Shading ±SEM.

Preferential expression of SSTtdT-expressing axons and dendrites in M2− interpatches of L1 of V1 of Sst-IRES-cre x Ai9 mouse.

(a) Tangential section through V1 shows M2+ patches (magenta, white outlines) in L1 of V1 labeled with an antibody against M2AChR. (b) tdT-expressing SST axons and dendrites in M2− interpatches (false colored green, blue outlines) of L1 of V1, (c) Overlay of panels a and b. (d) Intensity of SSTtdT expressing axons in different M2 quantiles, shows a strong preference for M2− interpatches. Error bars ±SEM. (e) SSTtdT intensity in each M2 quantile after shuffling images. (f) Distribution of labeled cell bodies shows that the majority of SST cells is preferentially aligned with M2− interpatches (green). (g) Cell counts shows that in all layers, SST cells are enriched in M2− interpatches. Shading in f and g indicates ±SEM.

Connectivity rules of visual processing streams.

(a) Flatmap of processing streams between V1 and higher cortical areas. The scheme is referenced to the clustering of M2+ patches (red) and M2− interpatches (grey) in L1 of V1. Each L1 module contains apical dendrites of different sets of output neurons in L2-5, and receives overlapping feedback input from corresponding higher areas to L1. The connections, represented by double arrowed lines, indicate module-selective, like-to-like reciprocal loops. M2+ patches and M2− interpatches split the dorsal and ventral intracortical processing streams into distinct sub-streams. The dorsal stream, represented by loops between V1 and AL, RL, PM, RSP and MOs, is divided into two subnetworks: the M2+ dorsolateral subnetwork (light red) composed of V1↔AL (Ji et al., 2015; D’Souza et al., 2019) and V1↔RL, and the M2− dorsomedial subnetwork (grey) represented by V1↔PM, V1↔RSP, and V1↔MOs. The ventral stream (dark red) is represented by the M2+ subnetwork and forms a loop between V1 and LM. For abbreviations, see Figure 4a. (b) Block of V1, constructed from 60-80-µm wide bundles (microcolumns) of apical dendrites (bold lines) distributed preferentially across multiple M2+ patches (red) and M2− interpatches (grey), contained within ∼300 x 300 µm slab of L1. M2 expression in layers below L1 is spatially uniform, but varies in density (grades of red shading) in layer-specific fashion. Intracortical L2/3 output PCs (triangles) project to LM, AL and RL (red) with apical dendrites (bold red lines) in M2+ patches receive overlapping feedback projections from areas LM, AL and RL, with which they form looped, like to-like connections. Output PCs (triangles) projecting to MOs (green), RSP (grey), PM (grey) and LP (black) with apical dendrites in M2− interpatches receive overlapping feedback input from areas MOs, PM, RSP and LP, with which they form looped, like to-like connections. In each pathway the dendrites are aligned with the M2+ patches or M2− interpatches in L1. In contrast, the cell bodies of L2/3 LM-, AL-, RL- and MOs-projecting neurons are randomly distributed across M2+ patches and M2− interpatches (notice red triangles aligned with M2+ and M2− domains), while the cell bodies of PM-, RSP, and LP-projecting neurons are vertically aligned with M2− interpatches (notice green triangles underneath M2+ patches and M2− interpatches). The pathway-dependent spectrum, random to clustered, of cell body distributions, sharply contrasts with the tight spatial alignment of dendrites with M2+ patches or M2− interpatches. This shows that the random distribution of LM-, AL-, RL- and MOs-projecting L2/3 neurons contrasts with the tight non-random spatial arrangement of apical dendrites in L1. The spatial organizations of L2/3 PM, RSP- and L5 LP-projecting neurons (grey, black triangles) is tighter, showing a strict alignment of apical dendrites and cells bodies with M2− interpatches. The common principle of all pathways depicted here, is that multiple types of projection neurons whose distribution may not be strictly columnar are grouped by clustered apical dendrites in M2+ patches or M2− interpatches, into distinct target-specific “output units” (Innocenti and Vercelli, 2010).

Demonstration of image registration of in vivo recording planes with ex vivo sections.

(a) Non-registered image of time-averaged GCaMP6f fluorescence across a session in the recording plane. Cyan arrows point to blood vessels and red arrows point to GCaMP6f-labeled cells which were clearly identifiable in both this in vivo image and the corresponding ex vivo section (panel c). These points were manually chosen in both images and used as fiducial points for registration. (b) Registered version of the in vivo image. Registration consisted almost entirely of rotation, with very little distortion or change in relative distances between points in the in vivo image. The Matlab function ‘fitgeotrans’ was used by inputting point pairs, including those labeled with red arrows, from the non-registered in vivo image (panel a) and the ex vivo image (panel c). In vivo landmarks were used as “moving points” and ex vivo landmarks were used as “fixed points.” (c) GCaMP6f fluorescence in ex vivo section at same location as in vivo recording. Cyan and red arrows are in the same locations and point to the same anatomical structures in this section and in the registered in vivo image (panel b), demonstrating that the registration process was successful in precisely aligning locations across the two conditions.

Effect of recording session number on locomotion modulation.

(a) Mean LMI across cells in all mice according to recording session number. Cells were divided into 5 groups depending on how many experimental sessions the mouse had experienced prior to that cell being recorded. Experience in prior recording sessions neither systematically increased nor decreased LMI of cells (r = 0.006, p = 0.882, Pearson correlation). (b) Relative goodness of fit of linear models for LMI which either included or excluded session number as an independent variable. Akaike Information Criterion (AIC) was used to compare model suitability. Using session number as the only predictor (3rd model) was inferior to a model using only a fixed intercept (1st model). Similarly, adding session number as a predictor (4th model) to a model using M2 module as a predictor (2nd model) reduced the model suitability.

Effects of locomotion and M2 module on visual tuning parameters of neural responses.

(a-c) Half-width at half maximum (HWHM) of spatial frequency (SF), temporal frequency (TF), or orientation tuning computed from locomotion (mean forward velocity > 0.1cm/s) and stationary trials (mean forward velocity < 0.1cm/s). For each cell, a tuning curve while stationary or during running were computed. Locomotion state did not change the HWHM of any stimulus parameter (p > 0.05, paired t-test). (d-f) HWHM of SF, TF and orientation tuning of cells in M2+ patches and M2− interpatches. M2+ patch and M2− interpatch cells did not show differences in HWHM for any tuning parameter (p > 0.05, t-test). (g-h) Peak SF and TF of cells in M2+ patches and M2− interpatches. M2+ patch and M2− interpatch cells did not show differences in peak SF or TF (p > 0.05, Pearson correlation).

Modular organization of connections between V1 and RL.

(a) Tangential section through L1 of V1 of Ai9 mouse stained with antibody against M2. M2+ patches (white) are outlined by white contours, M2− interpatches by cyan lines. (b). Anterogradely AAV-labeled RL→V1 axons (green) in L1 of V1 shows preferential termination in M2+ patches. (c) Overlay of panel a and b. (d-f) Apical dendrites of retrogradely AAV-labeled V1→RL-projecting neurons in L1 preferentially branch in M2+ patches. (g) Labeling density of axons in different M2 quantiles. Pearson correlation (r), error bars ±SEM. (h) Labeling density of dendrites in different M2 quantiles. Pearson correlation (r), error bars ±SEM. (i) Percent of retrogradely AAV-labeled V1→RL-projection neurons in M2+ patches (magenta) and M2− interpatches (green) in different layers of V1. Shading ±SEM. KS test.

Modular organization of connections between V1 and RSP.

(a) Tangential section through L1 of V1 of Ai9 mouse stained with antibody against M2. M2+ patches (white) are outlined by white contours, M2− interpatches by cyan lines. (b, c). Anterogradely AAV-labeled RSP→V1 axons (green) in L1 of V1 shows preferential termination in M2− interpatches. (d-f) Apical dendrites of retrogradely AAV-labeled V1→RSP-projecting neurons in L1 preferentially branch in M2− interpatches. (g) Labeling density of axons in different M2 quantiles. Pearson correlation (r), error bars ±SEM. (h) Labeling density of dendrites in different M2 quantiles. Pearson correlation (r), error bars ±SEM. (i) Percent of retrogradely AAV-labeled V1→RSP-projection neurons in M2+ patches (magenta) and M2− interpatches (green) in different layers of V1. Shading ±SEM. KS test.

Modular organization of connections between LP and V1.

(a) Tangential section through L1 of V1 of Ai9 mouse stained with antibody against M2. M2+ patches (magenta) are outlined by white contours, M2− interpatches by cyan lines. (b, c) Apical dendrites (green) of retrogradely AAV-labeled V1→LP-projecting L5 neurons preferentially branch in M2− interpatches of L1. (d) Labeling density of dendrites in different M2 quantiles shows that dendritic branches are denser in M2− interpatches. Pearson correlation (r), error bars ±SEM. (e) Retrogradely labeled V1→LP-projecting cell bodies show that cell bodies in L5 are preferentially (p = 0.02, KS) aligned with M2− interpatches (green). Shading ±SEM. KS test.