Running modulates primate and rodent visual cortex differently
- Departments of Neuroscience and Psychology, Center for Perceptual Systems, Institute for Neuroscience, The University of Texas at Austin, United States
- Departments of Ophthalmology and Psychiatry & Biobehavioral Sciences, Fuster Laboratory for Cognitive Neuroscience, UCLA, United States
- Department of Biology and Program in Neuroscience and Cognitive Science, University of Maryland, United States
- Herbert Wertheim School of Optometry and Vision Science, University of California, Berkeley, United States
Figures
Figure 1
Recording from marmoset V1 during active locomotion.
(a) Apparatus for recording from marmoset V1 while presenting visual stimuli on a high-resolution display, monitoring gaze using an eye tracker, on a toroidal treadmill that allowed the marmoset to run or not run. (b) Schematic example of variables of interest. Visual stimuli were presented (top row). Rasters show activity from a V1 array (second row). Gaze was monitored (third row, x and y time series plotted in black and gray), saccades were detected (red), and pupil size was also measured (fourth row). Running speed was measured using a rotary encoder attached to the treadmill (fifth row). (c) Before the main experiments, receptive fields (RFs) were mapped using sparse noise (Yates et al., 2021). The array of pseudocolor images shows three examples of V1 RFs (two foveal and one peripheral neuron). (d) Main experiment involved presenting full-field sinusoidal gratings that drifted in one of 12 directions (top row), at a variety of spatial frequencies (vertical axis at left). Rasters show example V1 activity during stimulus presentations when running (red) or stationary (black). (e) Summary of RF locations in the mouse dataset (orange, top), and (f) our data from marmosets (blue and green, bottom). In both marmosets, we recorded from a portion of V1 accessible at the dorsal surface of the brain using chronically implanted arrays, which yielded neurons with foveal RFs (green RFs). We also recorded from one marmoset using Neuropixels arrays, allowing us to simultaneously access both peripheral and foveal V1 (blue RFs; peripheral units are analyzed later/separately, see text). (g) Examples of mouse V1 orientation tuning curves, for cells with weak, moderate, and strong orientation tuning. (h) Same, for marmoset V1. (i, j) Histograms of orientation-selectivity indices (OSIs) for mice (i) and marmosets (j). Marmoset OSIs, likely lower than previously reported because we used full-field stimuli not optimized to the spatial frequency tuning of each neuron, and which likely recruited surround suppression. Regardless, the marmoset V1 neurons had strong visual responses and qualitatively conventional tuning. (k, l) Running speeds in mice (k) and marmosets (l). Marmosets were acclimated to the treadmill and motivated to run with fluid rewards yoked to traveling a criterion distance.
Figure 2
Mice and marmosets exhibit different correlations between V1 activity and running speed.
(a) Mice show visually compelling correlations between V1 trial spike counts and running speed. Example session with the highest correlation between running and V1 activity. Raster at top shows spiking activity of all mouse V1 neurons recorded. Population activity is summarized below the raster as the first principal component of the V1 array activity (‘First Neural PC’, orange trace); running speed is plotted underneath it on the same time axis (gray trace). Clearly, the two curves are highly similar. (b) Same, for an example mouse session chosen to have the median correlation between running and V1 activity. In this example, the modulations of running speed and neural activity rise and fall together on a faster time scale than in the example in (a). (c, d) Marmosets show smaller, and typically negative, correlations between V1 spiking activity and running speed. Format same as the mouse data in (a, b), with example sessions chosen to show the maximal and the median correlations between V1 activity and running speed. The (anti-correlated) similarity between V1 activity (First Neural PC) and Running Speed curves is harder to discern in the marmoset. (e, f) Correlations between V1 activity and running in the mouse (e) had a median >0 (median = 0.407, p=, stat = 308, n = 25, Mann–Whitney U test), and many individual sessions had significant correlations with running (filled bars), and all such significant sessions had positive correlations (with significance determined via permutation to remove effects of autocorrelation; Harris, 2021). In the marmosets (f), the distributions of correlations were slightly but reliably negative (median = −0.033, p=0.034, stat = 101, n = 27, Mann–Whitney U test), and all significantly modulated individual sessions exhibited negative correlations (5/27).
Figure 3
Running strongly increases mouse V1 activity and subtly decreases marmoset V1 activity, evidenced at the level of individual units.
Mouse data points are plotted in orange and marmoset data in blue. (a) Scatterplot (log-log) shows firing rate to preferred stimulus for tuned units (orientation-selectivity indices [OSI] > 0.2), during running (y-axis) and stationary (x-axis). Histogram summarizes the projections onto the line of unity and shows a clear shift indicating increases in response during running (geometric mean ratio [running/stationary] = 1.523 [1.469, 1.579], n = 743). Dark-shaded symbols indicate individually significant units. Dashed lines indicate doubling (2×) and halving (0.5×) of response. (b) Same format, but now showing the response aggregated over all stimuli, for all units (geometric mean ratio [running/stationary] = 1.402 [1.365, 1.440], n = 1168). A similar pattern reflecting primarily large increases is evident. (c, d) V1 units in marmoset show a very different pattern. Responses of tuned units to preferred stimuli (c) cluster more closely to the line of unity, with a small but significant shift indicating a subtle decrease in response (geometric mean ratio [running/stationary] = 0.899 [0.851, 0.949], n = 228). Responses to all stimuli for all units (d) show even less running-related modulation (geometric mean ratio [running/stationary] = 1.011 [0.995, 1.027], n = 786).
Figure 4
Shared gain model accounts for fluctuations in both mouse and marmoset V1, and explains species differences.
(a) Structure of shared modulator model. In addition to the effects of the stimulus (and slow drift in responsiveness, not rendered), the model allows for a shared gain/multiplicative term (green). Each simultaneously recorded neuron is fitted with a weight to the latent gain term. (b) The resulting model provides a better account of both mouse and marmoset V1 responses compared to a simple model that only fits stimulus and slow drift terms. Points show variance explained () on test data for each session under each of the two models, plotted against one another. (c) Variance explained for individual units was significantly improved in both species (marmoset: gain model [median = 0.2504] significantly higher than stim + drift [median = 0.1220], p=, stat = 27174, Wilcoxon signed-rank test; mouse: gain model [median = 0.4420] significantly better than stim + drift [median = 0.1697], p=, stat = 25966, Wilcoxon signed-rank test). (d) Example of relationship between neural responses (top raster, blue), the shared gain (green), and running speed (black trace). Visual inspection similar to that in Figure 2 can be performed. (e) Gain modulations span a larger range in mice than in marmosets. Orange, gain term from each mouse session; blue, gain term from each marmoset session. Triangles indicate medians (mouse = 2.17 [2.11, 2.25], marmoset = 1.19 [1.07, 1.27]). (f) Shared gain term is larger during running for mouse data, but is slightly smaller during running for marmoset data (difference is plotted on y-axis; mouse = 0.970 [0.761, 1.225], p=, stat 8.017, one-sample t-test; marmoset = −0.125 [−0.203, −0.059], p=0.002, stat = −3.360, one-sample t-test).
Figure 5
Eye movements and pupil size are modestly different during running.
(a) Each panel shows overlapping histograms of a measurement made on trials when the animal was running (blue) or stationary (red). (a) Saccade rate (in Hz) is slightly higher during running. (b) Saccade magnitude (in degrees of visual angle) is also slightly higher during running. The slight differences in saccade frequency and size did not quantitatively explain the differences in neural activity during running versus stationary periods; see main text for analysis. (c) Pupil size (expressed as % of mean size during stationary) is 8% larger during running (see main text for additional quantification).
Additional files
Download links
A two-part list of links to download the article, or parts of the article, in various formats.
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Running modulates primate and rodent visual cortex differently
eLife 12:RP87736.
https://doi.org/10.7554/eLife.87736.3
