Feedback generates a second receptive field in neurons of the visual cortex

Abstract

Animals sense the environment through pathways that link sensory organs to the brain. In the visual system, these feedforward pathways define the classical feedforward receptive field (ffRF), the area in space in which visual stimuli excite a neuron¹. The visual system also uses visual context-the visual scene surrounding a stimulus-to predict the content of the stimulus², and accordingly, neurons have been identified that are excited by stimuli outside their ffRF^3-8. However, the mechanisms that generate excitation to stimuli outside the ffRF are unclear. Here we show that feedback projections onto excitatory neurons in the mouse primary visual cortex generate a second receptive field that is driven by stimuli outside the ffRF. The stimulation of this feedback receptive field (fbRF) elicits responses that are slower and are delayed in comparison with those resulting from the stimulation of the ffRF. These responses are preferentially reduced by anaesthesia and by silencing higher visual areas. Feedback inputs from higher visual areas have scattered receptive fields relative to their putative targets in the primary visual cortex, which enables the generation of the fbRF. Neurons with fbRFs are located in cortical layers that receive strong feedback projections and are absent in the main input layer, which is consistent with a laminar processing hierarchy. The observation that large, uniform stimuli-which cover both the fbRF and the ffRF-suppress these responses indicates that the fbRF and the ffRF are mutually antagonistic. Whereas somatostatin-expressing inhibitory neurons are driven by these large stimuli, inhibitory neurons that express parvalbumin and vasoactive intestinal peptide have mutually antagonistic fbRF and ffRF, similar to excitatory neurons. Feedback projections may therefore enable neurons to use context to estimate information that is missing from the ffRF and to report differences in stimulus features across visual space, regardless of whether excitation occurs inside or outside the ffRF. By complementing the ffRF, the fbRF that we identify here could contribute to predictive processing.

Extended Data Figure 1 |. Robust responses to inverse stimuli with blurred edges.

a, Experimental configuration. b, Top: Schematic of stimuli used for size-tuning functions. Bottom: Population-averaged size tuning of classical and inverse stimuli with sharp edges (left) and stimuli with blurred edges (right; see Methods). Here and in all other figures, black and red traces are responses to classical and inverse stimuli, respectively, and shaded areas are periods of stimulus presentation. Solid lines are fits to the data (see Methods). Triangles above size-tuning functions indicate median preferred size for each condition. Insets: Maximum responses. Horizontal lines, medians. Two-sided Wilcoxon signed-rank test; sharp edge, ***: p = 2.0 × 10⁻⁹; blurred edge, ***: p = 4.5 × 10⁻¹⁰; 773 neurons in 4 mice. Traces and data points represent mean ± SEM (shading or error bars, respectively). Here and in all other figures, error bars are present but sometimes smaller than symbols.

Extended Data Figure 2 |. Illustration of classical and inverse-tuned neurons.

a, Classical(-only) neuron. Left: The response of a neuron probed with classical stimuli (black) increases with the size of the stimulus until it peaks at the neuron’s preferred size (top horizontal dotted line). The response then decreases due to surround suppression (maximum suppressed level indicated by the lower dotted horizontal line). The response of the same neuron probed with inverse stimuli (red) starts at the maximally surround suppressed activity level (an inverse stimulus with a size of 0° is a full-field grating) and then decreases as the diameter of the gray patch increases, consistent with visual stimulation being progressively removed from the classical feedforward receptive field (ffRF). Right: Schematic of a neuron’s ffRF surrounded by its classical suppressive zone. b, Inverse-tuned neuron. Left: The response of the neuron probed with inverse stimuli (red) starts, as for the classical-only neuron, at the maximally surround suppressed activity level but then it increases until reaching the neuron’s preferred inverse stimulus size and decreases with larger diameters of the gray patch consistent with visual stimulation being progressively removed from the feedback receptive field (fbRF). Right: Schematic of a neuron’s ffRF surrounded by its fbRF. c, Four example stimuli: Two classical stimuli (1 and 2 of sizes x and y, respectively) and two inverse stimuli (3 and 4, also of sizes x and y, respectively). The inner dotted circle represents the outer border of the classical ffRF. The outer dotted circle represents the outer border of the suppressive region and, for inverse-tuned neurons, also the outer border of the fbRF. The response amplitudes to the four example stimuli (1 to 4) in a classical-only neuron and in an inverse-tuned neuron, are marked in (a) and (b), respectively, at the intersection between the green vertical lines (stimulus size) and the size-tuning functions.

Extended Data Figure 3 |. Classical and inverse tuning properties in layer 2/3 excitatory neurons.

a, Scatter plot of L2/3 excitatory neuron peak responses (maximum responses to size tuning curves in Fig. 1c) of inverse-tuned neurons to classical and inverse stimuli. Classical and inverse median, 0.25 and 0.30 ΔF/F, respectively. Two-sided Wilcoxon signed-rank test; p = 7.7 × 10⁻⁵; same excitatory layer 2/3 (L2/3) neurons as in Fig. 1c; 1190 neurons in 9 mice. b, Top: Schematic of stimuli presented at different orientations to map the classical and inverse orientation preferences. We tested 8 orientations at intervals of 45° at the neuron’s preferred stimulus size and location using either a classical or an inverse stimulus. Bottom: Calcium responses of two example neurons in primary visual cortex (V1) for each orientation using classical and inverse stimuli. c, Population-averaged tuning curve for classical and inverse stimuli. Each neuron’s preferred orientations (independently for classical and inverse stimuli) were aligned to 0° and its activity normalized to its maximum response (367 neurons in 4 mice). Solid lines are fits to the data (see Methods). d, Tuning widths of orientation tuning curves obtained with classical stimuli compared to those obtained with inverse stimuli. For each neuron, tuning width was defined as the full width at half maximum (FWHM) of the fitted tuning curve. Two-sided Wilcoxon signed-rank test; p = 1.8 × 10⁻²¹; 367 neurons in 4 mice. Green symbols represent the example neurons shown in (b). e, Same for orientation selectivity indexes. The horizontal and vertical lines at 0.3 delimit the orientation-selective population. Two-sided Wilcoxon signed-rank test; p = 7.0 × 10⁻¹⁶; 367 neurons in 4 mice. f, Same for direction selectivity indexes. Two-sided Wilcoxon signed-rank test; p = 0.46; 367 neurons in 4 mice. g, Distribution of orientation offsets. For orientation-selective neurons only (see (e), with both OSIs ≥ 0.3), an orientation offset was computed, defined as the absolute difference in orientation between a neuron’s preferred orientation for a classical and an inverse stimulus. h, Contrast response map. Classical and inverse stimuli were presented simultaneously, and different combinations of contrasts were tested. The contrast heat map was obtained by averaging normalized activity (86 neurons in 4 mice). Traces and data points represent mean ± SEM (shading or error bars, respectively). i, Scatter plot of L4 neuron peak responses (maximum responses to size tuning curves in Fig. 1f) of inverse-tuned neurons to classical and inverse stimuli. Classical and inverse median, 2.2 and 0.41 ΔF/F, respectively. Two-sided Wilcoxon signed-rank test; p = 2.5 × 10⁻⁷; same layer 4 (L4) neurons as in Fig. 1f; 35 neurons in 6 mice. j, Scatter plot of PV neuron peak responses of inverse-tuned neurons to classical and inverse stimuli. Classical and inverse median, 0.40 and 0.48 ΔF/F, respectively. Two-sided Wilcoxon signed-rank test; p = 0.021; same PV neurons as in Fig. 2b, bottom; 60 neurons in 7 mice. k, Scatter plot of VIP neuron peak responses of inverse-tuned neurons to classical and inverse stimuli. Classical and inverse median, 0.98 and 0.54 ΔF/F, respectively. Two-sided Wilcoxon signed-rank test; p = 3.6 × 10⁻⁴; same VIP neurons as in Fig. 2c, bottom; 74 neurons in 8 mice. l, Scatter plot of SOM neuron peak responses of inverse-tuned neurons to classical and inverse stimuli. Classical and inverse median, 2.9 and 1.5 ΔF/F, respectively. Two-sided Wilcoxon signed-rank test; p = 1.3 × 10⁻²³; same SOM neurons as in Fig. 2d, bottom; 179 neurons in 5 mice.

Extended Data Figure 4 |. Inverse tuning is not due to low mapping resolution.

a, Top left: Schematic of regular receptive field mapping. Stimulus diameter of 20° with a grid spacing of 15°. Center left: Trial-averaged calcium responses from an example neuron for each stimulus location. Bottom left: Population-averaged receptive field for responses to classical or inverse stimuli aligned to the center of the ffRF (489 neurons in 4 mice). Right: Same but with fine receptive field mapping. Stimulus diameter of 10° with a grid spacing of 5° (only for part of the visual space covered with the regular mapping, see dotted rectangle on the left). b, Top: Spatial offset of regular ffRF mapping compared to fine ffRF mapping (same 489 neurons in 4 mice). For each neuron, its ffRF center estimated by the fine grid mapping is aligned at [0,0] and the localization of its estimated ffRF center estimated by the regular grid is plotted with respect to the fine grid estimated center. Bottom: Distribution of distances between the center of ffRF estimated by fine grid mapping and the center estimated by regular grid mapping (approximately 90% of neurons have a distance between the two centers below 10°). Green symbol represents the example neuron shown in (a). c, Population-averaged receptive field for responses to classical or inverse stimuli aligned to the center of the ffRF and only for L2/3 neurons that had a preferred ffRF size of more than 15° (319 neurons in 9 mice). d, Population-averaged size-tuning functions for classical (black: L2/3 neurons with ffRF > 15°, 335 neurons in 9 mice; gray: L4 neurons, 35 neurons in 6 mice) and inverse (red: L2/3 neurons with ffRF > 15°, 335 neurons in 9 mice; orange: L4 neurons, 35 neurons in 6 mice) stimuli. Solid lines are fits to the data (see Methods). Triangles above size-tuning functions indicate median preferred size for each condition. Inset: Maximum responses. Horizontal lines, medians. L2/3 neurons with ffRF size > 15°; Two-sided Wilcoxon signed-rank test; *: p = 6.7 × 10⁻³; 335 neurons in 9 mice. Traces and data points represent mean ± SEM (shading or error bars, respectively).

Extended Data Figure 5 |. Responses to inverse stimuli in layer 5/6.

a, Receptive field mapping of layer 5/6 (L5/6) neurons using classical and inverse stimuli. Top: Experimental configuration. Electrophysiological recordings were obtained in awake mice. The silicon probe spanned all layers, including deep layers (see Methods for layer definition). Center: Receptive fields were mapped using classical and inverse stimuli. Bottom left: Population-averaged ffRFs for L5/6 neurons. Bottom right: Same for inverse stimuli, aligned relative to the center of the ffRF (248 units in 20 mice). b, Size tuning of L5/6 neurons using classical and inverse stimuli. Top: Schematic of stimuli used for size-tuning functions. The classical and inverse stimuli were presented at the same location (within 10° of the estimated center of the ffRF). Bottom, normalized size-tuning functions for classical and inverse stimuli. Solid lines are fits to the data (see Methods). Triangles above size-tuning functions indicate median preferred size for each condition. Inset: Maximum responses. Horizontal lines, medians. Two-sided Wilcoxon signed-rank test; ***: p = 1.1 × 10⁻⁴; 119 units in 20 mice. c and d, same as (a and b) but for subset of L5/6 units defined both as surround suppressed and inverse tuned (as compared to (b) where all L5/6 units that responded to at least one classical stimulus size were included; see Methods) (c), 22 units in 12 mice; (d), Two-sided Wilcoxon signed-rank test; *: p = 0.016; 24 units in 12 mice. Data points represent mean ± SEM (error bars).

Extended Data Figure 6 |. Impact of anesthesia is more pronounced in higher visual areas.

a, Experimental configuration for intrinsic imaging of V1 and higher visual areas (HVAs). To estimate the visual area locations and their retinotopic maps using intrinsic imaging, we presented a narrow white bar (5°) on a black background, slowly drifting (10°/s) in one of the cardinal directions (“Fourier”). We calculated the temporal phase of the Fourier component at the frequency of the bar presentation. This gave us the complete extent of V1. For locating HVAs, we cross checked the Fourier maps with those obtained from the responses to 25° patches of gratings at different retinotopic locations (“episodic”). b, Left: Blood vessel pattern visible through thinned skull. Center: Fourier map of same field of view obtained with a vertical var moving from nasal to temporal. Right: Episodic map of same field of view. Scale bar, 1 mm. c, Other example episodic maps. Scale bar, 1 mm. d, Experimental design to assess impact of anesthesia on V1 and HVAs. The responses to classical stimuli of neurons in a HVA, LM or PM, and V1 were recorded using two-photon calcium imaging. The experiment started in awake mice by imaging either a HVA or V1. After induction of anesthesia, the same neurons were imaged again. To reduce the influence of variability in anesthesia levels, the first imaged area under anesthesia was imaged again at the end of the experiment. e, Peak responses in visual areas. Top left: Example calcium response of a neuron located in PM and another neuron located in V1 in an awake mouse (black) and responses of the same neurons in the anesthetized mouse (gray). Top right: Trial-averaged peak response for the same neurons shown on the left for the awake (black) and anesthetized mouse (gray). Bottom, same for a different mouse but recorded in V1 and LM. f, Population-averaged peak responses in awake and anesthetized mice. Top: Population-averaged peak responses in V1, LM and PM for awake (black) and anesthetized mice (gray). Two-sided Wilcoxon signed-rank test; V1, p = 6.2 × 10⁻⁴⁰, 431 neurons in 5 mice; LM, p = 9.9 × 10⁻¹⁹, 106 neurons in 3 mice; PM, p = 1.1 × 10⁻¹⁰; 55 neurons in 2 mice. Bottom: Population-averaged difference between normalized neuronal activity for awake and anesthetized state. For each neuron, all responses were normalized by the peak activity in the awake state before computing the differences. Two-sided Wilcoxon rank-sum test; V1, 431 neurons in 5 mice; LM, 106 neurons in 3 mice; PM, 55 neurons in 2 mice. V1-LM, ***: p = 1.2 × 10⁻²⁵; V1-PM, ***: p = 9.0 × 10⁻¹³; LM-PM, ns: p = 0.48. Traces and bars represent mean ± SEM (shading or error bars, respectively).

Extended Data Figure 7 |. Strong silencing by spatially restricted excitation of local inhibitory units.

a, Experimental configuration. A silicon probe was inserted in V1, spanning all cortical layers, in mice expressing Channelrhodopsin-2 in inhibitory neurons (VGAT-ChR2). To assess the strength of inhibition of excitatory units when using the laser scanning technique (Fig. 5; see Methods), the V1 recording site as well as seven other locations were scanned at 125 Hz. b, Raster plot of example excitatory unit in L5/6 in response to classical and inverse stimuli of 15° in diameter under control conditions (30 trials each) and during silencing of V1 (blue; V1 sil.). Black and blue horizontal lines are periods of stimulus presentation and V1 silencing, respectively. Classical and inverse stimuli were presented in random order; trials with V1 silencing were randomized as well but are separated here for clarity. c, Reduction in firing of excitatory units. The reduction in firing was measured as one minus the ratio between the optogenetic condition and the control condition. Note that silencing reached nearly 100% for both responses to classical and inverse stimuli, and for the baseline activity (26 units in 10 mice). d, Experimental configuration. To assess the effect of distance on the optogenetic stimulation of inhibitory units at the recording site, two medial and two lateral locations at 400 μm and 800 μm from the V1 recording site were targeted for laser stimulation while recording in V1. e, Modulation of the baseline of inhibitory units. The modulation index was defined as the difference between the activity during the optogenetic and the control condition divided by the sum of the two. The modulation index was high at the recording site (at 0 μm) and quickly dropped with distance (gray bars; Two-sided Student’s t-test; 0 μm, ***: p = 2.0 × 10⁻⁷; 400 μm, ns: p = 0.26; 800 μm, ns: p = 0.51; 16 units in 8 mice). As a comparison, the distance of the HVAs from the recording site is plotted on the same axis (black dots, right y-axis; 21 recording sites, 12 mice), suggesting that when pointing the laser at HVAs, direct activation of inhibitory neurons at the V1 recording site is unlikely. f, Experimental configurations. To assess the effect of the laser stimulation of HVAs on inhibitory units at the recording site, all 8 (top) or individual HVAs (bottom) were targeted for laser stimulation while recording in V1 (same configurations as during the experiments in Fig. 5 and Ext. Data Figs. 8 and 9). g, Modulation of the baseline of inhibitory units. The modulation indices were either negative or not significantly different from zero, indicating that the laser stimulation was unlikely to directly activate inhibitory neurons at the V1 recording site. Two-sided Student’s t-test; HVA, *: p = 0.045; 16 units in 8 mice; M, ns: p = 0.16; 5 units in 4 mice; PM, ns: p = 0.24; 16 units in 8 mice; AM, ns: p = 0.11; 16 units in 8 mice; RL, ns: p = 0.46; 5 units in 4 mice; AL, ns: p = 0.051; 16 units in 8 mice; LM, ns: p = 0.064; 16 units in 8 mice; LI, *: p = 0.015; 5 units in 4 mice; P, *: p = 0.010; 5 units in 4 mice. h, Are there many inhibitory neurons projecting from HVAs to V1? i, Methodology. A retrograde virus, AAVretro.CAG.Flex.tdTomato, was injected in V1 of GADcre mice to label GAD-positive neurons projecting to the site of injection. j, Left: Outlines of the cortical section where the confocal images shown on the right were acquired. The location of the imaged area is further indicated by the dotted square depicted on the outline. Rostro-caudal distance to bregma is indicated below the outline. Right: Average intensity projection. Top right: DAPI staining highlights the higher density of neurons in layer 4 in V1 used to define V1 borders (white lines). Bottom: The tdTomato fluorescence reveals numerous cell bodies in V1 around the site of injection and even more distal in L1. k, Same as in (j) but only for the tdTomato fluorescence and for all HVAs targeted for laser stimulation in Fig. 5 and Ext. Data Figs. 8 and 9. White lines delimit area’s boundaries. l, Quantification of tdTomato-positive neurons at the area’s center. The number of tdTomato-positive neurons were counted in the section containing the center of the investigated area. “HVAs” represents the sum of tdTomato-positive neurons in all HVAs. Note the sparse inhibitory projections from HVAs to V1 but the abundance of local inhibitory projections within V1 (3 mice). Bars represent mean ± SEM (error bars).

Extended Data Figure 8 |. Silencing higher visual areas reduces surround suppression in V1.

a, Experimental configuration. A laser beam is scanned over HVAs around V1 for optogenetic silencing while recording in V1. b, Size-tuning function of an example unit (baseline subtracted firing rates) to classical stimuli with (blue) or without (black) HVA silencing. Note the relief of surround suppression at larger stimulus sizes upon silencing HVAs. c, Scatter plot of the classical suppression index with or without silencing of HVAs (see Methods). Two-sided Wilcoxon signed-rank test; p = 0.033; 34 units in 12 mice. Closed and open symbols are units from L2/3 and L5/6, respectively. Green symbol represents the example neuron shown in (b). Data points represent mean ± SEM.

Extended Data Figure 9 |. Silencing individual higher visual areas differentially affects responses to classical and inverse stimuli.

a, Schematic of results and experimental configuration. Individual HVAs are targeted for optogenetic silencing while recording in V1. b, Difference in firing rates (baseline subtracted and normalized) between control conditions and individual HVA silencing for classical and inverse stimuli. Two-sided Wilcoxon signed-rank test; M, ns: p = 0.18; 22 units in 5 mice; PM, ns: p = 0.46; 42 units in 12 mice; AM, ns: p = 0.88; 42 units in 12 mice; RL, ns: p = 0.81; 22 units in 5 mice; AL, ns: p = 0.20; 42 units in 12 mice; LM, *: p = 0.013; 42 units in 12 mice; LI, ns: p = 0.51; 22 units in 5 mice; P, *: p = 0.020; 22 units in 5 mice. c, Scatter plot of the modulation indexes of individual HVA silencing for responses to classical and inverse stimuli (see Methods). Closed and open symbols are units from layer 2/3 and 5/6, respectively. Two-sided Wilcoxon signed-rank test; M, p = 0.033; 22 units in 5 mice; PM, p = 0.50; 42 units in 12 mice; AM, p = 0.47; 42 units in 12 mice; RL, p = 0.14; 22 units in 5 mice; AL, p = 0.19; 42 units in 12 mice; LM, p = 0.017; 42 units in 12 mice; LI, p = 0.067; 22 units in 5 mice; P, p = 0.039; 22 units in 5 mice. Note that for the visual stimulus parameters used here, LM showed the strongest effect in preferentially reducing responses to inverse stimuli. Bars represent mean ± SEM (error bars).

Extended Data Figure 10 |. Dual-color imaging of LM boutons and their putative V1 targets.

a, Left: Experimental configuration. To localize V1 and LM, we used intrinsic optical imaging (see Methods). Right: Response map to a nasal (magenta) and temporal patch of gratings (green). White lines represent area borders. b, Left: Blood vessel pattern overlaid with area borders defined by the intrinsic map (black lines). The red-shifted calcium indicator RGECO1a was injected in V1 and GCaMP6f was injected in LM. Right: Fluorescence of calcium indicators in V1 and LM. The black square delimits the example imaging site shown in (c). Same scale bar as in (a). c, Left: The responses of LM boutons and of V1 cell bodies were recorded within the same cortical location. Center: Example imaging site of V1 cell bodies recorded 190 μm below surface. The white square delimits the example imaging site shown on the right. Right: Example imaging site of LM boutons in V1 recorded 110 μm below surface. d, Top: Schematic of receptive field mapping. Left: Trial-averaged calcium responses from an example LM bouton aligned to its putative V1 target. Right: same but from an example bouton that is retinotopically offset with respect to its putative V1 target. e, Top: Schematic of stimuli used for size-tuning functions. Left and right: Trial-averaged calcium responses from the same example neurons as in (d). f, Left: Distance of population-averaged receptive field center of V1 neurons from center of size tuning stimuli (20 sites in 5 mice). Right: same for LM boutons. Note that all average V1 receptive field centers are located within 10° and that average LM receptive field centers are more spread with larger standard deviations. g, Retinotopic spread measured as cumulative distance from population-averaged receptive field center. The ffRF centers of LM boutons (solid green line) were more retinotopically spread than V1 neurons measured over the same cortical surface (solid black line) or measured over approximately 6 times the surface of the LM bouton site (dotted black line). Two-sided Wilcoxon rank-sum test; LM-V1 same surface, ***: p = 1.2 × 10⁻⁵; LM-V1 6× surface, ***: p = 3.1 × 10⁻⁴; LM, 311 boutons in 5 mice; V1 same surface, 530 neurons in 5 mice; V1 6× surface, 2352 neurons in 5 mice. h, Population-averaged size-tuning function of LM boutons (711 boutons in 5 mice) that are NOT retinotopically aligned with their V1 target. Note that both classical and inverse stimuli were presented at the ffRF location of their putative V1 targets (see Methods) and NOT at the ffRF location of the imaged LM boutons. Solid lines are fits to the data (see Methods). Triangles are median preferred size. Insets: Maximum responses. Horizontal lines, medians. Two-sided Wilcoxon signed-rank test; ***: p = 1.4 × 10⁻¹¹; 711 neurons in 5 mice. Data points represent mean ± SEM (error bars, respectively). i, Experimental configuration. j, Population-averaged size-tuning functions for classical and inverse stimuli. Solid lines are fits to the data (see Methods). Triangles are median preferred size. Insets: Maximum responses. Horizontal lines, medians. Two-sided Wilcoxon signed-rank test; ***: p = 4.7 × 10⁻¹⁰; 115 neurons in 3 mice. Data points represent mean ± SEM (error bars, respectively). k, Distribution of inverse tuning indices (ITIs) of LM (black) and V1 neurons (gray; same neurons as in Fig. 1c; 0: classical only; 0.5 equal peak response to classical and inverse stimuli; 1: inverse only; see Methods). Triangles above the distribution indicate medians. Two-sided Wilcoxon rank-sum test; ***: p = 2.9 × 10⁻¹⁵; 115 neurons in 3 mice and 1190 neurons in 9 mice for LM and V1, respectively.

Figure 1 |. Layer-specific responses to inverse stimuli.

a, Experimental configuration. b, Top: Example trial-averaged calcium responses of an excitatory layer 2/3 (L2/3) neuron for each stimulus location. In all figures: shaded areas, stimulus presentation periods. Bottom: Population-averaged RFs aligned to the center of the classical feedforward RF (ffRF; 2601 excitatory L2/3 neurons in 9 mice). c, Top: Example trial-averaged calcium responses. Stimuli are centered on the ffRF. Bottom: Size-tuning functions, normalized to maximum response to classical stimuli. In all figures: solid lines, fits to the data; triangles, median preferred size. Inset: Inverse tuning index (ITI). Median: 0.54 (1190 excitatory L2/3 neurons in 9 mice). d, Schematic of results and experimental configuration. e and f, As above but for layer 4 (L4) excitatory neurons. e, 24 neurons in 4 mice. f, Inset: ITI. Median: 0.053. Two-sided Wilcoxon rank-sum test; ***: p = 1.5 × 10⁻¹⁸; black: L4, 35 neurons in 6 mice; gray: L2/3 (Fig. 1c). In all figures, traces and data points represent mean ± SEM (shading or error bars, respectively).

Figure 2 |. Neuron-type specific response to inverse stimuli.

a, Top: Population-averaged RF for PV neurons aligned to the center of ffRF (82 neurons in 6 mice). Bottom: Size-tuning functions, normalized to maximum response to classical stimuli. Inset: ITI. Median: 0.53. Two-sided Wilcoxon rank-sum test; ns: p = 0.79; black: PV, 60 neurons in 7 mice; gray: excitatory L2/3 (Fig. 1c). b, Same as in (a) but for VIP neurons. Top: 126 neurons in 4 mice. Bottom: Inset: ITI. Median: 0.42. Two-sided Wilcoxon rank-sum test; ***: p = 1.1 × 10⁻⁵; black: VIP, 74 neurons in 8 mice; gray: excitatory L2/3 (Fig. 1c). c, Same as in (a) but for SOM neurons. Top: 315 neurons in 5 mice. Bottom: Inset: ITI. Median: 0.12. Two-sided Wilcoxon rank-sum test; ***: p = 1.1 × 10⁻⁴²; black: SOM, 179 neurons in 9 mice; gray: excitatory L2/3 (Fig. 1c).

Figure 3 |. Slow and delayed responses to inverse stimuli.

a, Schematic of results and experimental configuration for extracellular recordings in V1. b, Responses of a layer 5/6 (L5/6) unit to classical and inverse stimuli centered on its ffRF. Top: Raster plot (1000 trials each stimulus). Bottom: Peristimulus time histogram (PSTH; 10 ms bins). c, Population-averaged PSTHs normalized to average response to classical stimuli (15 units in 4 mice). d, Onset slope of the response to classical and inverse stimuli. Green symbol, example unit from (b). Inset: PSTH from (b) to illustrate differences in slopes. Dotted lines, lower and upper thresholds to compute slopes. Two-sided Wilcoxon signed-rank test; p = 1.8 × 10⁻⁴; 15 units 4 mice. e, Same as (d) but for rise time. Two-sided Wilcoxon signed-rank test; p = 7.8 × 10⁻³; 15 units in 4 mice. f, Same as (d) but for delay. Two-sided Wilcoxon signed-rank test; p = 9.8 × 10⁻⁴; 15 units in 4 mice. g, Mean onset slopes, rise times, and delays of independently tuned units. Two-sided Wilcoxon rank-sum test; onset slope, **: p = 1.1 × 10⁻³; rise time, *: p = 0.017; delay, *: p = 0.031; classical: 51 units in 8 mice; inverse: 29 units in 8 mice.

Figure 4 |. Anesthesia preferentially reduces responses to inverse stimuli.

a, Top: Example calcium responses to classical and inverse stimuli in an awake mouse. Bottom: Same example neuron but under isoflurane anesthesia. b, Size-tuning functions in awake (top) and anesthetized (bottom) mice, normalized to maximum awake response to classical stimuli. Insets: ITI. Top: Median: 0.50. Bottom: Median: 0.32. Two-sided Wilcoxon signed-rank test; ***: p = 1.5 × 10⁻⁵; black: anesthetized; gray: awake; same 49 excitatory L2/3 neurons in 5 mice. c, Peak responses of inverse-tuned neurons in awake (top) and anesthetized (bottom) mice. Green symbol, example neuron from (a). Two-sided Wilcoxon signed-rank test; top: p = 0.96; bottom: p = 2.0 × 10⁻⁵; same neurons as above.

Figure 5 |. Higher visual areas contribute to inverse tuning in V1

a, Schematic of results and experimental configuration to optogenetically silence higher visual areas (HVAs). b, Top: Raster plot of example L5/6 unit (30 trials each). Black lines, stimulus presentation; Blue lines, HVA silencing. Bottom: Example size-tuning function with or without HVA silencing c, Size-tuning function with or without HVAs silencing, normalized to maximum control response to classical stimuli. Two-sided Wilcoxon signed-rank test; classical, ***: p = 3.2 × 10⁻⁴; inverse, **: p = 3.0 × 10⁻³. Inset: Mean baseline firing rate with or without HVA silencing. Two-sided Wilcoxon rank-sum test; ***: p = 5.0 × 10⁻¹⁰; 44 units in 12 mice. d, Difference in firing rates between control conditions and HVA silencing. Two-sided Wilcoxon signed-rank test; 5°, *: p = 8.0 × 10⁻³; 15°, **: p = 1.4 × 10⁻³; 25°, ***: p = 4.0 × 10⁻⁵; 35°, **: p = 3.5 × 10⁻³; 45°, ns: p = 0.70; 44 units in 12 mice. e, Optogenetic modulation indexes (see Methods). Green symbol, example neuron from (b). Two-sided Wilcoxon signed-rank test; p = 7.4 × 10⁻⁴; 44 units in 12 mice. f, Schematic of results and experimental configuration. g, Size-tuning function for LM boutons retinotopically aligned to their putative V1 targets. Insets: Left: Maximum responses. Horizontal lines, medians. Two-sided Wilcoxon signed-rank test; ***: p = 6.1 × 10⁻¹⁶. Right: ITI. Median: 0.11. Two-sided Wilcoxon rank-sum test; ***: p = 2.8 × 10⁻³⁸; black: LM, 87 boutons in 5 mice; gray: L2/3 (Fig. 1c). h, Top: Schematic of results. Bottom: Retinotopic spread of the ffRF of V1 neurons and LM boutons (2352 neurons and 311 boutons in the same 5 mice). i, Size-tuning functions of LM boutons not retinotopically aligned with their putative V1 targets. Inset: Maximum responses. Horizontal lines, medians. Two-sided Wilcoxon signed-rank test; ***: p = 4.8 × 10⁻³⁶; 362 boutons in 5 mice.