Cortico-cortical feedback engages active dendrites in visual cortex

Abstract

Sensory processing in the neocortex requires both feedforward and feedback information flow between cortical areas¹. In feedback processing, higher-level representations provide contextual information to lower levels, and facilitate perceptual functions such as contour integration and figure-ground segmentation^2,3. However, we have limited understanding of the circuit and cellular mechanisms that mediate feedback influence. Here we use long-range all-optical connectivity mapping in mice to show that feedback influence from the lateromedial higher visual area (LM) to the primary visual cortex (V1) is spatially organized. When the source and target of feedback represent the same area of visual space, feedback is relatively suppressive. By contrast, when the source is offset from the target in visual space, feedback is relatively facilitating. Two-photon calcium imaging data show that this facilitating feedback is nonlinearly integrated in the apical tuft dendrites of V1 pyramidal neurons: retinotopically offset (surround) visual stimuli drive local dendritic calcium signals indicative of regenerative events, and two-photon optogenetic activation of LM neurons projecting to identified feedback-recipient spines in V1 can drive similar branch-specific local calcium signals. Our results show how neocortical feedback connectivity and nonlinear dendritic integration can together form a substrate to support both predictive and cooperative contextual interactions.

Fig. 1. Mesoscale mapping of interareal functional connectivity in mouse visual cortex using simultaneous two-photon optogenetic stimulation and two-photon calcium imaging.

a, Top, illustration of the experimental set-up to examine interareal functional connectivity between V1 and LM. Bottom, schematic of pyramidal neurons with layer-specific projection preferences. A, anterior higher visual area; AL, anterolateral higher visual area; AM, anteromedial higher visual area; PM, posteromedial higher visual area; RL, rostrolateral higher visual area. b, Mean two-photon images for one example FOV, showing the large spatial extent of expression. Top, GCaMP6s expression driven transgenically. Bottom, C1V1 expression, driven using adeno-associated viruses (AAVs). Insets: representative cell bodies. Scale bars, 250 μm. c, Top, cellular-resolution retinotopic mapping across V1 and LM with a large FOV, obtained using sparse noise stimulation and two-photon population imaging. The grey bar delineates a 150-μm-wide border zone, which was excluded from stimulation and responder detection. Bottom, the mean photostimulation response of example stimulation groups (pixelwise stimulus-triggered averages). d, Trial structure. Full field sinusoidal gratings were presented either alone, or paired with a photostimulus. Responses to the two trial types were compared to detect responders. Vis., visual. e, Example local responders from one session. Data are mean ± s.e.m. across trials. Black lines represent presentation of the visual stimulus. Grey bars with lightning symbols represent the photostimulus. Both examples are from LM. V, visual stimulus only; V+P, visual stimulus and photostimulus. f, Example across-border (in the area opposite the stimulated one) responders from one session. Facilitated example from LM, suppressed from V1. For e and f, scale bars, 1 s (horizontal) and 0.1 ΔF/F (vertical).

Fig. 2. Cortico-cortical feedback is relatively suppressive of topographically matched centre locations and relatively facilitating of mismatched surround locations.

a, Left, retinotopic map in azimuth used to assign recorded neurons to a cortical area was obtained by smoothing cellular-resolution maps constructed using sparse noise stimuli and two-photon population imaging. Right, smoothed retinotopic map in elevation. b, Left, example photostimulation group in LM (probing functional connectivity in the feedback direction) and the corresponding retinotopic location in V1. Right, measurement of the absolute retinotopic (rather than physical) distance of an example across-border responder to a photostimulated cluster. c, The probability distribution of responder retinotopic distances divided by the probability distribution of distances for all available neurons. Data are mean ± s.e.m. across stimulation groups. n = 129 (LM) and n = 180 (V1) clusters from 42 sessions in 11 animals. The horizontal grey lines mark y = 1 and represent uniform spatial sampling. Suppressed responders are plotted downwards by convention. Red arrow indicates the location of stimulation, the green arrow indicates the location of responders measured. Top, stimulation and readout in V1. Bottom, stimulation and readout in LM. d, The same as in c, but for interareal stimulation and readout, and also including the average retinotopic location of responders for each stimulation group (red and blue dots), which were used to make the comparison. Top, stimulation in V1 and readout in LM (feedforward). Bottom, stimulation in LM and readout in V1 (feedback). Feedforward-facilitated and feedforward-suppressed responders do not differ in spatial distribution (centroids over stimulation groups, rank-sum test, P = 0.61). Feedback-facilitated and feedback-suppressed responders are displaced relative to each other (rank-sum test, P = 9.2 × 10⁻⁵). NS, not significant. e, Schematic illustrating the suppressive feedback from retinotopically aligned and facilitating feedback from retinotopically offset projections.

Fig. 3. Visual stimuli that recruit facilitating feedback drive local calcium events in apical tuft dendrites.

a, Experimental design. Ultrasparse expression within layer 5 IT neurons defined by Cre expression was achieved by combining AAV-mediated Cre-dependent FLP expression with AAV-mediated FLP-dependent GCaMP6s expression. Isolated somata were identified using two-photon imaging, their receptive fields were mapped and apical dendrites were traced. Visual stimuli were then positioned relative to the receptive field and the visual responses of the identified dendrites were imaged. Scale bars, 50 μm (left), 8° (middle) and 25 μm (right). b, Example local dendritic events from two different neurons. For each neuron, top left, mean dendritic segment fluorescence during global events. Middle left, mean fluorescence during local dendritic event. Bottom left, local dendritic event magnified. Note that at least two spines were simultaneously active along with a limited extent of the adjacent branch. Top right, manually drawn ROIs, and illustration of their location on an idealized pyramidal cell morphology. Inset: magnification of an independent event. Scale bars, 10 μm (neuron 1 and 2, top and middle) and 2 μm (neuron 1 and 2, bottom). c, Stimulus-dependence of independent events showing that the inverse stimulus, which provides the most effective stimulation of the surround, drives the most independent events. Results from n = 13 branches belonging to 9 neurons. The lines connect responses obtained from a single branch (17 trials per minute). Kruskal–Wallis test across all stimuli (P = 0.01) followed by post hoc Dunn’s test: 8° versus inverse (P = 0.041), 16° versus inverse (P = 0.047), inverse versus annulus (P = 0.034) and inverse versus full field (P = 0.48). Stimuli are schematized as disks to illustrate size and shape, presented experimentally as sinusoidal gratings with Gaussian masks and no sharp edges. RF, receptive field. d, Schematic illustrating topographically offset feedback facilitating events in apical tuft dendrites.

Fig. 4. LM feedback drives branch-specific local dendritic calcium signals in V1.

a, Two-photon optogenetic stimulation of LM neurons during simultaneous two-photon imaging of apical dendrites in V1. b, Left, opsin expression and target groups in LM for clustered stimulation. Right, sparse GCaMP expression in V1. c, Top, example dendrite. Bottom, fluorescence from the blue ROI. Photostimulation is indicated by the red lines. d, Average photostimulation response (blue) and average blank trial (black). n = 71 dendrites in 14 animals. Statistical analysis was performed using signed-rank tests; P = 1.1 × 10⁻⁷ (before versus after stimulation), P = 0.74 (before versus after blank), P = 3.2 × 10⁻⁶ (stimulation versus blank). e, Grid of targets for random stimulation; one group is shown in yellow. f, Top, example dendrite and spine ROIs. Bottom, fluorescence from three ROIs. The arrowhead indicates an independent spine event. g, Testing a putative connection (target 1095 to spine 1) after it was identified through analysis of independent events. Stim., stimulation. h, Isolated stimulation confirms connection. Top, photostimulus-triggered average image showing responsive spine. Bottom, blank and photostimulus responses from spine. i, Two examples of photostimulation-triggered local events. Left, global event showing stimulated and reference branch ROIs. Middle, responsive spines. Right, a single trial showing a branch-specific event. j, Fluorescence from ROIs in i. k, Example FOV for the boosting analysis showing an ROI on a reference branch; the box indicates a branch carrying a feedback-recipient spine. l, Top, higher-magnification image showing a feedback-recipient spine (arrowhead) and an ROI excluding it. Bottom, schematic showing layout of ROIs for the boosting analysis. m, Top, example photostimulation trial showing fluorescence from the synaptically stimulated branch and reference branch. Bottom, the same for a blank trial. n, Distributions of boosting indices for stimulation and blank trials from one recording with significant boosting. Statistical analysis was performed using a rank-sum test; P = 1.16 × 10⁻⁸. The dashed lines indicate the mean values. o, Mean boosting indices (BI) for photostimulation trials are higher than for blank trials across recordings. Statistical analysis was performed using a paired t-test; P = 0.0002. The filled circles indicate individually significantly modulated neurons (P < 0.05). Scale bars, 400 μm (b,e and g), 20 μm (c (top), f (top), h (top), i and k), 10 μm (l).

Extended Data Fig. 1. Photoactivation parameters, response magnitude and reliability.

a, Spatial extent of two-photon optogenetic photoactivation. Top, response probability by distance from targeted locations. Curves obtained from the experiments that produced data in Figs. 1 & 2. Bottom, pixelwise photostimulus-triggered response averages of one example photostimulation group in three dimensions. b, Response reliability of facilitated responders in the stimulated area (top, “local”) and in the other area (bottom, “across”) as a function of false discovery rate (n = 310 stimulation groups, two-sided Wilcoxon rank sum test). Boxes indicate median, 75^th and 25^th percentile, whiskers indicate upper and lower adjacent values. c, Most locally facilitated responders are weakly activated, with a small number being strongly activated. d, The number of facilitated and suppressed responders as a function of false discovery rate. Bar plots represent mean ± s.e.m, dots indicate individual stimulation groups (n = 130 in LM and 180 in V1). Statistical test used was a Wilcoxon signed rank test between facilitated and suppressed responder counts. Note that the y-axis was set to the 95^th percentile of the data for FDR 0.1 to facilitate visualization.

Extended Data Fig. 2. Topographic map weighting and dependence on FDR.

a, Un-weighted retinotopic distance distributions of responders. These distributions are averages across all experiments. Data shown in Fig. 2c,d are weighted, where the responder retinotopic distribution obtained for each stimulation group is weighted by the corresponding “all neurons” distribution for that stimulation group. b, Un-weighted distance distributions of responders (as in a) for one example stimulation group (across-area facilitated and suppressed followers after LM photostimulation). c, Un-weighted distance distributions of locally facilitated responders and target locations. Dashed lines s.e.m. across stimulation groups. d, Topographic biases do not depend on FDR. Top left, the difference between un-weighted retinotopic distance distributions for facilitated feedback responders and their corresponding “all neurons” distributions. Bias across distance changes gradually as a function of FDR. Right, the integral of the difference plotted on the left as a function of FDR (mean ± s.d., n = 129 stimulation groups). e, Each dot is the centroid of the retinotopic locations of responders recruited by one stimulation group (only stimulation groups with ≥ 1 across border responder, Feedback: n = 51/74/103 groups (FDR 2.5%, 5%, 10%) facilitated, n = 64/90/106 suppressed; Feedforward: n = 42/63/112 facilitated, n = 61/83/116 suppressed). Responder distributions are displaced between facilitated and suppressed responders in the feedback direction but not in the feedforward direction (see also Fig. 2; Wilcoxon rank sum test). This comparison remains significant across different FDR thresholds for feedback.

Extended Data Fig. 3. Differences in stimulation do not account for across-border response topography.

a, Procedure for matching local facilitated responder numbers (top), retinotopic spread (middle) and across-area responder physical distances from target location (bottom). Left, distributions for all stimulation groups before matching, which was performed by randomly sampling the overlapping segment of the distributions. Right, distributions after matching (see Methods). b, Across-border response topography in the feedback (left) and feedforward (right) direction computed from resamples (mean ± s.d. across samples). Relative patterns of facilitation and suppression are maintained (Fig. 2). c, Proportion of resamples with a negative difference between centroids of weighted probability distributions for facilitated and suppressed neurons as a function of FDR. Negative difference indicates suppressed neurons displaced relative to facilitated. d, Left, retinotopic distance of responder to target location as a function of physical distance of responder to target location. Right, density plot showing the availability of all neurons. e, Retinotopic distance distributions of facilitated and suppressed responders more than 500 μm away from the target location. f, Number of across-border facilitated (left) and suppressed (right) neurons as a function of the number of local responders. g, Photostimulation response magnitude as a function of retinotopic distance for all responders at FDR = 2.5%. Response magnitude is the difference between mean responses to the V+P trials and V trials. There is no effect of distance on response magnitude (ANOVA, p = 0.34 facilitated, p = 0.35 suppressed). Solid lines indicate binned and averaged response magnitude. h, Probability of a feedback stimulation group producing at least one suppressed, at least one facilitated and at least one of each type of responder. Probability of finding both is not significantly less than would be expected based on the individual rates (permutation test, p = 0.22).

Extended Data Fig. 4. Apical tuft dendrite imaging experiments.

a, Example illustrating experimental workflow, which consisted of finding a sparsely labelled group of neurons, mapping visual receptive fields, tracing from the soma to the apical tuft while imaging at 800 nm and then finding a branch that was relatively uninterrupted by other processes before displaying visual stimuli relative to the known receptive field. b, Visual stimuli used in this experiment. c, Two additional examples of dendrites that exhibited independent events. Bottom example illustrates challenges to automated detection. Event #1 was included in our data as an independent event that clearly involved at least two distinct spines and the branch. Event #2 was similar to #1, but only one spine and the branch were visibly activated during this smaller magnitude event. Event #3 was a global event that happened to be larger magnitude in the distal ROI in comparison to the proximal ROI.

Extended Data Fig. 5. Spatial scale of local events in apical tuft dendrites.

a, Two separate ROIs were drawn over each dendritic segment where events were identified. One “mask” ROI encapsulated the entire branch with its spines, and the other “line” ROI traced a single pixel wide line along the branch itself. b, Fluorescence measured in each pixel in the mask ROI was averaged into the nearest pixel of the line ROI, along with all other mask ROI pixels for which that line ROI pixel was the nearest. The geodesic distance between each line ROI pixel and the most proximal line ROI pixel was measured. c, Fluorescence over time and geodesic distance along an example dendrite. The mean fluorescence in each resulting geodesic distance pixel was smoothed with a 4 s moving average and then converted to (F-F₀)/F₀, where F₀ was assigned as the 10^th percentile of the fluorescence in a running window 45 s wide. We then normalized this ΔF/F by the standard deviation of the whole trace to facilitate comparisons across experiments. d, The spatial profile of fluorescence over geodesic distance averaged over the frames identified with red band in c. e, As in c for another example neuron. f, As in d for another example neuron. g, The spatial profile of all identified events, aligned by their peak location and normalized to that value in grey, their average in blue. h, Same data as in g, average ± s.d. i, Average spatial profile is well fit by a 3-term Gaussian with full-width at half-maximum of 11.2 μm.

Extended Data Fig. 6. Visual stimuli that recruit enhanced facilitating feedback preferentially drive dendritic activity.

a, Semi-sparse expression allows tracing of apical trunks to their parent somata. b, Left, soma and dendrite regions-of-interest. Right, fluorescence from a connected soma and apical dendrite. c, Population average dendritic fluorescence as a function of somatic fluorescence, for six stimulus sizes. Trials sorted by somatic activity level, binned in 5% increments, averaged across cells (n = 57 neurons with receptive fields <20° from stimulus centre, and size tuning (p < 0.01) with preference for 20° gratings). Two-way ANOVA on population data to determine effect of somatic activity, stimulus size, and their interaction on dendritic fluorescence showed a significant effect of stimulus size, F(5,131) = 5.96, p = 5 x 10⁻⁵, and a significant interaction, F(5,131) = 4.74, p = 5 x 10⁻⁴. d, Population average somatic and dendritic calcium fluorescence traces for different stimuli from one somatic activity bin, indicated in panel c. e, Dendritic fluorescence as a function of stimulus size, averaged across neurons after effect of somatic fluorescence is removed on a cell-by-cell basis (residual after subtraction of fit to somatic fluorescence; Methods). One way repeated measures ANOVA on n = 57 neurons shows significant effect of size, F(5,280) = 2.681, p = 0.02. Post-hoc tests show residual at 5° is significantly smaller than residuals at larger stimulus sizes (paired t-tests, p < 0.05). f, Dendritic and somatic size preference. Dendrites prefer larger stimuli as a population (paired t: p = 1 x 10⁻⁶). Data from n = 131 neurons with receptive fields <20° from stimulus centre, significant size tuning (p < 0.01), and preferring 10°,20°, 40° or 60° gratings. Data from 26 sessions in 9 mice. Dendritic preference computed from residuals as in e. g, Schematic illustrating topographically offset feedback facilitating dendritic excitation during global events.

Extended Data Fig. 7. Identifying connected soma-apical dendrite pairs with semi-sparse expression.

a, Correlation between the fluorescence traces (X) and between the deconvolved event traces (Y) for all possible pairs of somata and dendrites from one session. Dashed line: thresholds used to identify connected pairs. b, Soma plane from this example session. c, Two dendrite planes from same example session. d, Calcium signals in the soma and dendrites highlighted by red circles. Both dendrites belong to circled soma, but only one pair makes it past the conservative thresholds. e, Another example that illustrates variability in expression density. Denser end of the range we obtain still allows identification of apical trunk and nexus below layer 1. f, Dendrite fluorescence as a function of soma fluorescence, n = 76 neurons (receptive field <10 degrees of stimulus centre, preferring 10, 20, 40 or 60 degrees). Fluorescence was baseline-subtracted (F₀ = 10^th percentile in a 1 min running window) normalized by its standard deviation over the whole recording duration and decimated to 1 Hz for this plot. Note that high fluorescence in the dendrite in the absence of high fluorescence in the soma is not a feature of the data.

Extended Data Fig. 8. Modulation of global events in apical dendrites relative to basal dendrites by visual stimuli.

a, Receptive field mapping with sparse noise stimuli and example receptive field. b, Left, schematic with volume imaging planes. Right, example neuron showing ultra-sparse expression. Isolated somata were identified, their receptive fields mapped, and apical dendrites traced. Visual stimuli (same as in Fig. 3) were then positioned relative to the receptive field. c, Traces from different dendritic locations. Arrows highlight events with variable ratio of apical to basal fluorescence. d, Top left, apical dendritic responses to inverse and full field stimuli. Bottom left, basal dendritic responses to same stimuli. Top right, apical dendritic responses to Gabor stimuli. Bottom right, basal dendritic responses to same stimuli. Each trace shows one trial, with fluorescence from all apical and all basal ROIs averaged to get one trace per compartment. e, Same data as in d, averaged across trials (mean ± s.e.m.) with responses to Gabor stimuli in blue and responses to inverse and full field stimuli in red, separately for apical (top) and basal dendrites (bottom). f, Top, distribution of apical-to-basal fluorescence ratios across trials for the two stimulus classes for one cell. Bottom, same data, comparing the real difference between the apical-to-basal fluorescence ratios (difference of means of histograms above) to the shuffling stimulus ID for the same trials. g, Top, apical-to-basal ratio differences for 26 recordings. Significantly modulated neurons shown in dark grey. Dashed line shows mean of all cells. Bottom, for the five significantly positively modulated neurons recorded, the apical-to-basal ratio was computed individually for each apical ROI against the average of all basal ROIs. The correlation between the apical-to-basal ratios across apical ROIs, with correlation of trunk and tuft ROIs in red and the correlation of tuft ROIs in blue.

Extended Data Fig. 9. Two-photon photostimulation-triggered spine mapping.

a, Example dendrite and ROIs used for spine fluorescence extraction. Same experiment as shown in Fig. 4, where panel f shows a section of the FOV shown here at higher magnification. b, Fluorescence from six example ROIs. c, Photostimulation responses of example ROI #1 plotted against the mean photostimulation response across all spine ROIs. Trials that produced independent activity were identified as passing a threshold applied to signals from that spine and not passing a threshold applied to the average spine. Targets active on those trials, like the example target #1095 shown, were identified. d, Four example spines from four recordings. Top, stimulation-triggered average image showing independently activated spines. Bottom, the stimulation triggered fluorescence traces from those example spines. e, An example spine and stimulation target pair that were identified in two separate experimental sessions, 3 days apart. f, Distribution of retinotopic and physical distances between effective target in LM and the parent soma of the responsive spine. Retinotopic distances were measured using widefield intrinsic imaging under anaesthesia and temporal retinotopic mapping (Kalatsky et al., 2003). g, Retinotopic position of effective targets in LM plotted relative to the retinotopic position of the parent soma of the responsive spine, positioned at the origin.

Extended Data Fig. 10. Branch-specific boosting analysis.

a, Two additional examples of branch-specific local events in response to photostimulation of identified presynaptic targets in LM. Left, average image of global events, showing stimulated (incl. spine) and reference branch ROIs. Middle, spines responsive to feedback. Right, average image of single trial showing branch-specific event. b, Fluorescence extracted from those ROIs, showing the variability of stimulated vs reference branch signal strength and the branch-specific events driven by photostimulation, indicated by red bars. c, Top, schematic illustrating ideal layout of ROIs used for boosting analysis. Bottom, distribution of minimum distances between stimulated branch ROIs and the identified feedback-recipient spine. d, Mean boosting indices for blank vs photostimulation trials. Each recording appears once for “near” ROIs and once for “far”. “Near” ROIs same as reported in Fig. 4. Filled circles are individually significant neurons (t-test, p < 0.05). Right, same data plotted as a histogram of differences between boosting indices measured during stimulation and during blank trials. e, Boosting index (difference between photostimulation and blank, same data as in Fig. 4) as a function of retinotopic distance between target location and parent soma location. f, Schematic illustrating visual stimulation delivered simultaneously with feedback stimulation. g, Boosting indices measured from Near ROIs while the visual stimulus was a grey screen. Left, scatter plot of indices measured on blank vs on stimulation trials. Right, same data shown as a histogram of differences between trial types. h, Boosting indices measured from Near ROIs while the visual stimulus was an inverse grating positioned around the receptive field of the V1 neuron. Left, scatter plot of indices measured on blank vs on stimulation trials. Right, same data shown as a histogram of differences between trial types.

Extended Data Fig. 11. Locomotion regulates feedback and dendritic excitability.

a, FOV for measuring modulation of visual responses in V1 and LM. b, Locomotion facilitates visual responses more in LM than V1 (signed-rank test, 20°: p = 0.017, full-field: p = 0.042). c, Increased feedback predicts increased suppression for smaller stimuli, and increased suppression and facilitation for larger stimuli. d, Dendritic fluorescence as a function of somatic fluorescence, during locomotion and stationarity (mean ± sem, n = 131 neurons). A 2-way ANOVA to determine effect of somatic fluorescence and locomotion on dendritic fluorescence shows significant interaction: F(1,48) = 8.86, p = 0.0046. e, Change in dendritic response with locomotion (mean ± sem). For each neuron we removed the effect of somatic activity and categorized stimuli as smaller than preferred (paired t-test, p = 0.003,) preferred (p = 0.112), larger than preferred (p = 0.067). f, Apical dendrites are suppressed by locomotion when small stimuli, but not larger stimuli, are presented. g, Somatic size tuning during locomotion and stationarity (n = 131). h, Somatic response modulation. (mean ± s.e.m., smaller than preferred: paired t-test, p = 5.45x10⁻⁷, preferred: p = 0.32, larger than preferred: p = 0.002). i, Locomotion reduces somatic response to small stimuli but enhances response to large stimuli. j, Top, strategy for dual-colour input-output imaging. Bottom, stimulus-triggered average glutamate and calcium fluorescence during stationarity and locomotion (n = 12 sessions, 4 mice). k, Interaction coefficients between locomotion and apical/basal glutamate from linear model (Methods). Basal coefficients are significantly above (t-test, p = 0.01), apical coefficients below zero (p = 0.02). Filled circles are sessions with individually significant effect of omitting interactions. l, Locomotion-induced enhancement of feedback, suppression of apical dendrites and somata during presentation of smaller stimuli, and enhancement of basal inputs and somata for larger stimuli.

Extended Data Fig. 12. Locomotion-induced changes in glutamate, calcium signals and their relationships.

a, Schematic illustrating dual-colour expression strategy. b, Basal glutamate fluorescence over entire FOV during stationary and locomotion periods, averaged over all trials in one session. c, Apical glutamate fluorescence over entire FOV during stationary and locomotion periods, averaged over all trials in one session. d, Population activity quantified as deconvolved events across all ROIs over entire FOV during stationary and locomotion periods, averaged over all trials in one session. e, Raw traces of iGluSnFR fluorescence and deconvolved calcium signals summed across the population. Grey bars are visual stimulus. f, iGlu-SnFR fluorescence in basal dendrites during the baseline period did not change with locomotion (n = 12 sessions in n = 4 mice, t-test, p = 0.12). g, iGlu-SnFR fluorescence in apical dendrites during the baseline period did not change with locomotion (t-test, p = 0.18). h, Deconvolved events during the baseline period did not change with locomotion (t-test, p = 0.25). i, Stimulus evoked change in mean iGlu-SnFR fluorescence in the basal dendrite planes increased with locomotion (t-test, p = 1.3 x 10⁻⁶). j, Stimulus evoked change in mean iGlu-SnFR fluorescence in the apical dendrite planes increased with locomotion (t-test, p = 1.8 x 10⁻⁵). k, Stimulus evoked change in deconvolved events increased with locomotion (t-test, p = 0.02). l, Three dimensional population level input-output function for one session. m, Linear model fit to each session in order to measure the interaction between locomotion and basal and apical glutamate to population output gain. n, Coefficients of the interaction show that locomotion and basal glutamate interact positively (t-test, p = 0.01), while locomotion and apical gain interact negatively (t-test, p = 0.02). Solid data points are sessions with individually significant interactions between running and glutamate (Wald test, p < 0.05, Methods).

Extended Data Fig. 13. Locomotion suppresses spontaneous GABAergic signalling in layer 1 neurons but enhances sensory-evoked GABAergic signals.

a, Expression and excitation strategy for dual-colour imaging. b, Average iGABA-SnFR fluorescence (top) and jRGECO1a fluorescence (bottom) in a typical field-of-view (300 μm). Visual stimuli were 20° or full field sinusoidal gratings and they produced qualitatively similar results. c, Example raw trace of iGABA-SnFR fluorescence recorded over the entire field of view and averaged over two imaging planes in layer 1. d, Mean GABA fluorescence over entire field-of-view during stationary and locomotion periods, averaged over trials in one session. e, Mean GABA fluorescence over entire field-of-view during stationary and locomotion periods, averaged across n = 33 sessions. f, Mean fluorescence during stationarity vs locomotion. Top, GABA fluorescence during baseline periods significantly decreased with locomotion (t-test, p = 5.5 x 10⁻⁵). Bottom, stimulus evoked (baseline-subtracted) GABA fluorescence significantly increased with locomotion (t-test, p = 3 x 10⁻⁷).