Locomotion Controls Spatial Integration in Mouse Visual Cortex

1We started by characterizing the properties of spatial integration and found that most neurons in area V1 of awake mice showed clear surround suppression (Figure 1). We recorded multiunit activity and single-unit activity from V1 while mice were head-restrained on an air-suspended ball (Figure 1A). The visual stimuli were gratings centered on the receptive fields of the recorded neurons, varying randomly in size. Consistent with previous reports in monkey and mouse, the responses of most neurons showed first an increase with increasing size, followed by suppression as the grating size exceeded an optimal value (e.g., Figure 1B). To quantify the strength of this surround suppression, we defined a suppression index as (Rp – RL)/RL, where RL and RP are the responses to the largest size and to the preferred size. For robustness, we defined the latter as the smallest size that elicited >95% of the maximal response (Figure 1B). Based on suppression index and preferred stimulus size, the neurons in the population could be roughly divided into two groups (Figure 1C). The majority (60 of 89) showed preferred sizes in the range of 16° ± 5° (median ± median absolute deviation) and positive suppression indices (30% ± 13%, Figure 1C, black dots). The rest (29 of 89) showed no evident suppression: their suppression indices were negative, indicating that the size tuning curves saturated or kept on growing with increasing size (Figure 1C, gray dots). These cells might have showed surround suppression if we had been able to show larger stimuli. Indeed, the display spanned only 45° vertically, so our largest stimuli (60° diameter) were partially clipped and larger stimuli were not feasible. In the following, we report results for both groups, but our interest is in the cells where we could measure surround suppression.

2Locomotion strongly decreased the strength of surround suppression (Figure 2). Animals typically spent ∼51% ± 22% (SD, n = 4 mice, 9 sessions) of the time being stationary (speed ≤ 1.2 cm/s). During these periods, cells typically showed strong surround suppression (e.g., Figure 2A, red, suppression index = 47%). During locomotion, however, this surround suppression was typically greatly reduced (e.g., Figure 2A, blue, suppression index = 3%). Similar effects were seen in the other cells in our sample (e.g., Figure 2B, suppression index from 50% to 23%) and indeed in the average of all cells that showed some surround suppression (Figure 2C). On average, the suppression index decreased from 38% ± 3% (SEM, n = 60) when mice were stationary to 23% ± 3% during locomotion. Locomotion increased the responses more for large stimuli than for small stimuli, and this increase in response grew linearly with stimulus size, both for the example neurons (Figures 2D and 2E) and for the population (Figure 2F). Across cells, locomotion increased both the peak firing rate and the spontaneous firing rate (Figures 2G and 2H). Moreover, it markedly changed the properties of spatial integration, increasing the preferred stimulus size and decreasing the strength of size tuning (Figures 2I and 2J). These effects were consistent across cells, and they were similar in cells with broad or thin spikes (see Figure S1 available online), suggesting that they may affect all cells equally. Conversely, locomotion did not affect other properties of the responses, such as orientation tuning (data not shown).

3These effects of locomotion could not be explained by artifacts of eye movements (Figure S2). Eye movements were invariably horizontal and only slightly larger during locomotion (averaging 4.4° ± 3.9°, SD) than when the animal was stationary (2.5° ± 2.6°). These effects were too small to disrupt estimates of surround suppression: locomotion affects most the responses to the largest stimuli, which are one order of magnitude larger than typical eye movements. Moreover, eye movements were rare: the eye was within 5° of the central position in 66% of trials during locomotion and 78% of trials while the animals were stationary. Finally, the effects of locomotion were seen even in trials where the eye did not move. In these trials, locomotion reduced the suppression index from 38% ± 3% (SEM, n = 58) to 26% ± 3% (Figure S2). Similarly, the effects of locomotion were not due to artifactual changes in spike isolation (Figure S3) or to artifactual correlations between the visual preferences of the neurons (preferred orientation or receptive field center) and the properties of the stimulus (Figure S4).

4Spatial integration by neurons in area V1 is known to depend strongly on contrast. To fully describe how spatial integration is affected by locomotion, therefore, we sought to characterize it with stimuli varying in size and contrast. We considered a divisive normalization model (Figure 3A), where the responses are driven by the output of a “driving field” and suppressed by the output of a wider “suppressive field”. We modeled both fields as Gaussians and defined the model’s response to a stimulus with contrast c and diameter d asHere R0 is the baseline (spontaneous) firing rate, m and n are exponents, RD and RS are the strengths of driving field and suppressive field, and D(d) and S(d) are the fractions of the driving field and suppressive field covered by the stimulus (between 0 and 1), which depend on the sizes σD and σS of the driving field and suppressive field. An additional parameter δ allows the model to account for slight miscentering of the stimuli. The model is therefore specified by eight parameters, and we obtained those parameters by fitting responses to 41 stimuli: all combinations of eight diameters and five contrasts, and a blank stimulus.

5The model provided good fits to the responses, accounting for the effects of both size and contrast (Figures 3C–3F). With a single set of parameters, the model accounted for the increase in surround suppression with contrast (Figure 3C) and for the associated increase in contrast saturation with size (Figure 3D). Similar results were seen in the other cells (Figures 3E and 3F). For 99 of 105 neurons that we probed with combinations of stimulus size and contrast, the model accounted for >80% of the explainable variance (Figure 3B). We can therefore use the model to quantify the effects of contrast on spatial integration in awake mouse V1. By doing so, we observed two effects of increasing contrast: an increase in the strength of surround suppression (Figure 3G) and a decrease in the preferred stimulus size (Figure 3H). Similar effects have been observed in primates.

6We then applied the divisive normalization model to characterize the effects of locomotion (Figures 4A and 4B). We considered the 99 of 105 cells whose responses were well fitted by the normalization model (the model accounted for >80% of the explainable variance), and we fitted their responses separately depending on whether the animal was stationary or moving (Figures 4A and 4B). To minimize the number of free parameters, we allowed as few parameters as possible to vary with locomotion. We could achieve good fits as long as we allowed three parameters to vary: the baseline firing rate R0, the strength of the driving field RD, and the strength of the suppressive field RS. The remaining parameters (m, n, σD, σS, and δ) were kept constant across conditions (Figures 4G–4L).

7Locomotion had two main effects: it increased the baseline firing rate, and it reduced the strength of the suppressive field relative to the driving field (Figures 4C–4F). Across cells, locomotion increased the baseline firing rate R0 by a factor of 1.3 ± 0.2 (median ± median absolute deviation, n = 99 neurons, Figure 4C), a significant increase (p < 0.005, sign test). Similarly, locomotion decreased the strength of the suppressive field RS, by a factor of 0.9 ± 0.1 (Figure 4D, p < 0.005). Locomotion also increased the strength of the driving field RD, by a factor of 1.1 ± 0.2 (p < 0.005). However, these changes in RD traded inversely with changes in RS, leading to an apparent variability across cells. Indeed, locomotion increased the ratio RD/RS, by a median factor of 1.7 (Figure 4F, significantly >1, p < 0.005). At any given stimulus size, this ratio is a measure of responsiveness, and responsiveness was indeed increased by locomotion.

8We have shown that locomotion modulates the responses of V1 neurons by changing their properties of spatial integration. In agreement with previous results, we found that locomotion increases both baseline activity and visually driven activity. By varying stimulus size, however, we discovered that this increase is largest for large stimuli and progressively smaller for smaller stimuli. Locomotion, in summary, profoundly reduces the strength of surround suppression experienced by V1 neurons.

9These results indicate that locomotion has more than a simple modulatory effect. Previous data on the effects of locomotion were obtained at a fixed stimulus size and suggested that locomotion affects responses to all stimuli equally. Such a modulatory role could in principle be given a number of simple explanations, including one as simple as an increase in brain temperature. However, our finding that locomotion shapes spatial selectivity makes it unlikely to operate through a global quantity.

10Our findings indicate that the effects of locomotion differ substantially from those of spatial attention. Spatial attention typically results in an increase in effective stimulus contrast and therefore in stimulus competition. In primate V1, increasing stimulus contrast increases the strength of surround suppression. Our measurements indicate that similar effects are present in mouse V1. However, these effects are opposite to those of locomotion, which reduces spatial competition. Therefore, locomotion does not seem to cause an increase in effective contrast. Its effects are likely to rely on mechanisms different from those affected by spatial attention.

11Perhaps the effect of locomotion on mouse V1 does resemble the effect of spatial attention seen in primate V1, but only for attention delivered to the periphery of the visual field. Indeed, whereas near the fovea of primate V1 attention appears to strengthen surround suppression, in the periphery it appears to weaken it. This similarity suggests that locomotion may operate on similar mechanisms as peripheral attention.

12We found that in mouse V1, the spatial integration and its dependence of contrast could be well described by a simple divisive normalization model. This model had previously been used to summarize spatial integration and contrast integration in cat and primate. It is attractive because it can directly predict the interactions of stimulus contrast and size, without the need for additional parameters. Our results suggest that mouse V1 performs similar computations and extend previous results obtained with stimuli of a single size.

13The normalization model allowed us to summarize the effects of locomotion concisely: an increase in baseline activity and a decrease in the relative strength of the divisive suppressive field. This phenomenological description does not by itself indicate what mechanisms are at play, because the mechanisms that underlie normalization are currently unclear. However, having a simple, closed-form equation that summarizes the effects of contrast, spatial extent, and locomotion can help guide the search for the underlying mechanisms.

14One of the possible mechanisms at play is synaptic inhibition. There is debate as to whether inhibition mediates surround suppression. In mouse V1, however, there is convincing evidence that it does, at least for a class of inhibitory neurons that express somatostatin and are particularly active in the awake cortex. Indeed, in awake mouse V1 there is strong synaptic inhibition from broad regions of visual field. Perhaps these interneurons expressing somatostatin are less active during locomotion, thus explaining the decrease in surround suppression. Other classes of interneurons could also play a role. Niell and Stryker found neurons with thin spikes that are suppressed by stimuli when the animal runs. We did not find such cells (in only 3 of 215 neurons were responses decreased by stimuli when the animal was running), but the effect is clear in their measurements and points to a class of interneurons expressing parvalbumin, because these neurons have thin spikes. Perhaps during locomotion, large stimuli suppress a larger number of these interneurons, thereby reducing the overall amount of inhibition. To elucidate these matters, future experiments will need to clarify whether and how each inhibitory cell class is affected by locomotion.

15Conversely, our results seem to rule out the possible involvement of mechanisms relying on acetylcholine. In primate V1, acetylcholine does typically increase visual responses, but it also appears to increase the strength of surround suppression, which is the opposite of what we found for locomotion.

16We do not know whether the effects of locomotion that we have uncovered in mouse also extend to other species such as cats, primates, or humans. If these effects are present also in the other species, then this points to an interaction of early vision and locomotion that has been hitherto unsuspected. If instead they are not, this may point to a fundamental difference between visual systems. Perhaps the visual system of the mouse is devoted to assisting the animal’s navigation more than to pattern vision and shape perception. In that case, it might be advantageous for the mouse visual cortex to have its spatial integration properties depend on locomotion, because this may assist in some spatiotemporal computations that are useful for navigation. However, the relevant computations need to be elucidated. The literature contains compelling theories for how surround suppression could help neural coding, yet nothing in these theories seems to suggest that this suppression should be altered by locomotion.

17All procedures were conducted in accordance with the UK Animals Scientific Procedures Act (1986). Experiments were performed at University College London under personal and project licenses released by the Home Office following appropriate ethics review. For detailed methods, please see the Supplemental Experimental Procedures.