Slow adaptation in pyramidal cells during rest and locomotion.

A. Top: Mice running on a treadmill were shown a visual stimulus (drifting grating, 20° in size) for 10 seconds. Bottom: example field of view (FOV) showing PCs in V1 expressing GCaMP6f (grey) and corresponding regions-of-interest (ROIs) for individual neurons (rainbow). Scale bar represents 200 µm and stimulus lasts 10 s. B. Top: raster plots showing average calcium responses from individual PCs that were responsive to the stimulus during locomotion and rest (Same cells recordings between rest and locomotion, n = 522 neurons, from 9 mice, cells nested within animals). Responses are averaged over 10 stimulus trials then sorted by time of maximum amplitude in rest and locomotion independently. Bottom: average calcium response of all PCs to the visual stimulus during locomotion (left) and rest (right). C. Distributions of adaptive indices (AIs) in the PC population during locomotion (top) and rest (bottom) in sensitizers (blue), intermediate cells (grey) and depressors (green). Both distributions can be described by Gaussians with similar x0 and different width (Rest, x0 =0.001±0.9, width = 0.93±0.4; Loco, x0 = -0.006±0.3, width = 0.33±0.02). D, E, F. Splitting the distribution of PCs according to their AI reveals that the tertile with lower AI showed slow sensitization, the tertile with intermediate values showed little adaptation, and the tertile with higher AI showed slow depression. G. Differences in gain of sensitizers (blue), intermediate cells (grey) and depressors (green) measured as the integral during the stimulus show that locomotion caused a stronger gain increase in sensitizers than in depressors (p < 0.001, Linear Mixed Model).

Differential effects of locomotion on fast and slow components of adaptation in PCs and interneurons.

A. Estimated spike rate averaged across all sensitizing PCs measured during locomotion (Data from all neurons responsive to the stimulus during locomotion including non-responsive cells during rest, n = 479 neurons from 14 mice). The GCaMP signal averaged across the same set of neurons is shown in grey. The fast depressing component of adaptation is highlighted by the red arrow. Stimulus duration 10 s. B. As A, but for the second (pre-dominantly non-adapting) tertile of PCs. C. As A, but for the depressing tertile. The trace from A is superimposed (light blue) showing the almost identical initial amplitude of the response in sensitizing and depressing PCs. D. Comparison of average spiking response at rest and during locomotion in the sensitizing tertile of PCs in which paired measurements were made (158 neurons in 9 mice). The relative change in the response during locomotion was 4.7 ± 1. E. Comparison of average spiking response at rest and during locomotion in the depressing tertile of PCs. The relative change in the response during locomotion was 1.4 ± 0.3 which was significantly smaller than the sensitizing tertile (T-test, p < 0.001). F. The time-varying difference in spike rate of sensitizing and depressing PCs. G. Average response of VIPs to the visual stimulus during locomotion and rest. The gain during locomotion increased by factor of 5.1 ± 1.0, measured from the integral of the responses. H. Average response of PVs to the visual stimulus during locomotion and rest. The gain during locomotion increased by factor of 1.7 ± 0.2. I. Average response of SSTs to the visual stimulus during locomotion and rest. SSTs showed a much smaller, nonsignificant, gain change 1.2 ± 0.2, p = 0.28. J. Distribution of adaptive indices in VIP interneurons (n = 122 neurons in 2 mice). Note that AI was measured from GCaMP signals, as for PCs in Fig. 1. VIPs were predominantly sensitizing. Psens = 0.56 at rest and 0.90 during locomotion. K. Distribution of adaptive indices in PVs (n = 97 neurons in 3 mice). PVs were predominantly sensitizing. Psens = 0.64 at rest and 0.86 during locomotion. L. Distribution of adaptive indices in SSTs (n = 82 neurons in 4 mice). SSTs were predominantly depressing at rest and during locomotion. Psens = 0.37 at rest and 0.21 during locomotion. Results from D to L are from cell-paired recordings in the two states.

A model of signal flow in layer 2/3.

A. Schematic showing the targets of three external inputs into layer 2/3 of V1. Feedforward excitation targets PVs and PCs; feedback input targets all populations 50,51 and the slower sensitizing input enters through VIPs as well as directly activating PCs and PVs52. (Vertical bar: 0.5 Hz, horizontal bar: 5 s) B. Schematic showing the main excitatory (arrow tip) and inhibitory (round tip) connections between cell types within L2/3 of V1. PCs form excitatory connections with all other neuron types including themselves. PVs mainly target PCs and other PVs while SSTs inhibit all other cell types but avoid inhibiting each other. VIPs almost exclusively inhibit SST interneurons. (Vertical bar: 0.05 ΔF/F, horizontal bar: 5 s). C. Left: schematic showing the experimental paradigm to test if VIPs were a target for a slow sensitizing signal driving adaptation in layer 2/3: VIPs were silenced optogenetically while simultaneously recording activity in SSTs. Right: example response of an SST neuron during stimulus presentation alone and stimulus paired with optogenetic silencing of VIPs. D. Average response of SSTs with and without optogenetic silencing of VIPs (n = 83, from 4 mice). Silencing VIPs almost completely blocked slow adaptation. Light grey shade shows the stimulus time (10 s). E. The slow sensitizing input to the model (SS) was based on the average response of the VIP population during locomotion and consists of a step and sigmoid function with a time-constant = 1.71 s.

Testing the model using optogenetics

A. Top: Average PC response with and without optogenetic activation of SSTs. Separate experiments indicated that this intensity of photoactivating light increased SST activity by a factor of ∼1.79. The red trace shows the model prediction when SST activity within the model was increased by the same factor. Activation of SSTs decreased PC gain and increased sensitization (top, RMSE = 0.27). The grey control trace (No Opto) is averaged over all four optogenetic conditions in A and B. Bottom: Average PC response with and without optogenetic inhibition of SSTs. Separate experiments indicated that this intensity of photoactivating light decreased SST activity by a factor of ∼1.99. The red trace shows the model prediction when SST activity within the model was decreased by the same factor. Silencing of SSTs resulted in an increase in gain and a shift to more depressing dynamics (bottom, RMSE = 0.31). B. Same as in C but when PVs were optogenetically activated (top, by a factor of ∼2.5) or silenced (bottom, by a factor ∼2.0) during stimulus presentations. Activation of PVs decreased the gain of PCs (top, RMSE = 0.32). Silencing of PVs resulted in an increase in gain and shift to more sensitizing dynamics (bottom, RMSE = 0.28). C. Proportion of the initial “good” solutions to the model (chi-square < 10) that also provided good solutions incorporating the results of the optogenetic manipulations in A and B. Filtering of solutions on the basis of adding optogenetic conditions was carried out in series in the order 1-4. Applying filter 4 (PV inhibition) had little further effect on the number of good solutions i.e almost all good solutions incorporating manipulations 1-3 also predicted the effects of PV inhibition. D. Comparison of connection weights (mean ± SD) of all good solutions after applying optogenetic filters 1-3 and 1-4. All weights remain similar after the fourth filter and it is these averages that are presented in Figs. 5 and 6.

Locomotion modulates both electrical activity and the strength of synaptic connections.

A-D. Top: Average firing rates of PCs (A, black), VIPs (B, green), SSTs (C, blue) and PVs (D, dark red) during stimulus presentation at rest and their corresponding fits from the average of solutions calculated by the model (light red, RMSE = 0.19). Note that these traces show the average activity of only partially paired recordings, where only a subset of cells are matched between rest and locomotion (Rest n=158 from 9 mice, Locomotion n = 1437 neurons, from 14 mice). Bottom: average activity during locomotion in the same neuronal populations (RMSE = 0.11). E. Average resting connection weights from all solutions estimated by fitting the model to the activity of responsive neurons (Equation 2). F. The relative change in connection weights caused by locomotion (Equation 4). G. Relative change in synaptic weights based on estimated changes in connection weights (F) and measured changes in number of responsive cells (Supplementary Table 1; equation 3). Note that all excitatory and inhibitory synapses weaken except for synapses from SST to VIP which strengthen. H. The change in the total synaptic input activated by the stimulus, calculated from the average activity (Supplementary Table 2) and the change in connection weight in F (equation 5). The last column is the net change taking into account the polarity of the synapse.

Differences in PV:SST input ratio determine the direction of PC adaptation and the change in gain associated with locomotion.

A. Activity of PCs from the tertile with lower AIs (sensitizers) at rest and during locomotion together with fits from the average of all good solutions calculated by the model (rest, RMSE = 0.22; loco, RMSE = 0.17). These are paired measurements from 158 neurons in 9 mice. During locomotion 7.5 ±1.2 more spikes were fired during the stimulus. B-C. As A, but for the intermediate and depressing tertiles, RMSEs were similar to sensitizers. During locomotion the stimulus generated 6.3 ± 0.9 more spikes in the intermediate group and 2.8 ± 0.4 in depressors. D-E. Heatmaps of the synaptic weights for each PC tertile at rest (D) and locomotion (E); sensitizers (Sen), intermediate (Int) and depressing (Dep). F. Amplitude change induced by optogenetic manipulation of SSTs and PVs in sensitizer (blue) and depressing (green) PCs. Note that sensitizers are preferentially modulated by SSTs and depressors by PVs as predicted by the model. (Inh: silencing with ArchT, Act: activation with ChrimsonR). G. Relative change in connection weights caused by locomotion for each subpopulation of PCs (calculated as equation 4). The key difference is the relative strength of PV and SST inputs. H. Relative change in synaptic weights caused by locomotion for each subpopulation of PCs. I. Change in total input during presentation of the stimulus received by each subpopulation of PCs. Sensitizers receive less SST inhibition than the intermediate and sensitizing tertiles. The last column shows the net change in total input, taking into account the polarity of the synapse.

Summary of effect of locomotion in V1.

From rest (left) to locomotion (right) the connections from external inputs increase while the strength of PC connections to other targets is reduced. Sensitizers (top) receive stronger SST inhibition, being more sensitive to their reduction during locomotion, while depressors (bottom) receive stronger PV inhibition.

Locomotion-dependent changes in adaptive properties of PCs

A. The distribution of AIs at rest and during locomotion, reproduced from Fig. 1C. B. Scatter plot showing the change in AI as a function of resting AI (each point a single cell; n = 522 neurons, from 9 mice). Cells that depress at rest (AI > 0) rest showed AIs closer to sensitization during locomotion (blue box in lower right quadrant) while cells that sensitize (AI < 0) showed AI closer to depression (green box in upper left quadrant). These shifts in measured AI help explaining the narrowing of the AI distribution in A. The red dashed line is fitted to the data collected from the subset of PCs with AI between -0.5 and 0.5 and reflects a correlation coefficient of r = −0.69.

Identification of an external signal determining the time-course of slow adaptation.

A. Similar dynamics of slow depression in SSTs and the depressing tertile of PCs during locomotion. B. Similar dynamics of sensitization in VIPs and the sensitizing tertile of PCs (left axis, blue). C. The slow sensitizing input to the model (SS) was based on the average response of the VIP population during locomotion and consists of a step and sigmoid function with a time-constant = 1.71 s (see Fig. 4).

A brute force approach delimited a parameter subspace of valid solutions.

A. Average connection weights of all the solutions found in the parameter space to give a good fit for the experimental data during rest . B. Average connection weights of all the solutions found in the parameter space to give a good fit for the experimental data during locomotion

Removing external inputs impairs model performance.

A. Average firing rates of PCs (black), VIPs (green), SSTs (blue) and PVs (dark red) during stimulus presentation during locomotion and their corresponding fits from the solution with lower (RMSE = 0.14) calculated by a model without slow-sensitizing (SS) input. Note how VIP and PV dynamics are not properly captured. B. Same as in A for missing feedback input (FB) (RMSE = 0.14). Note how SST and PV dynamics are not properly captured.

Locomotion-associated changes in the number of neurons responding to the visual stimulus

Average response amplitudes in resting and active states

List of parameters used to fit the model