Ihe brain is thought to produce decisions by gradual accumulation of sensory evidence through a hierarchically organized feedforward cascade of neuronal activities that transforms early stimulus representations in the primary somatosensory cortex (S1) to a perceptual decision processed in premotor areas. Recently, this prevailing view has been challenged by observation of choice-correlated neural activity as early in the hierarchy as S1. Here, to reconcile these seemingly controversial observations, we employ ethological whisker-guided navigation of mice in a tactile virtual reality paradigm combined with dense electrophysiological recordings in whisker-related wS1. Leaving only a pair of C2 whiskers for mice to navigate with, we effectively designed an information bottleneck for sensory input to decision-making. We show that neural activity during sensory evidence accumulation exhibits dramatic collapse of the high-dimensional spiking activity to just a single latent variable followed by a slower and almost synchronous ramping up across the whole cortical column. We show that this variable is consistent with models of gradual accumulation of noisy sensory evidence to a decision bound. These observations indicate that S1 may directly participate in a categorical coding of all-or-none decision variable via cortico-cortical feedback loops through which sensory information reverberates to be transformed into perception and action.
The prevailing view on perceptual decision-making, mostly adapted from influential research in the visual system of primates, has been that sensory flow is gradually transformed from stimulus representations in the primary somatosensory cortex (S1) (1) to neural activity correlated with decision in the secondary somatosensory cortex (2) (S2) up to a categorical all-or-none perceptual decision computed in premotor areas (3, 4). Here, close to the top of the cortical hierarchy, internal decision variables (DV) are accumulated in a form of ramping neural activity (3–6) that is transformed into a decision when a decision boundary is reached (7).
Serial transformation of sensory inputs along bottom–up hierarchical pathways has been observed in rodents engaged in whisker-discrimination tasks (8, 9) with an increasingly stronger association with perceptual choice in S2 (10, 11) and premotor areas (12–14) rather than whisker-related wS1 (9). The function of wS1 is usually analyzed (15) in the context of extracting complex, high-order features of sensory input (16–18). It has been observed, however, that wS1 may integrate sensory information over time (19) and contribute to goal-directed behavior (20, 21). More recent experiments (22–25) have shown neural activity in subpopulations of layer 2/3 of wS1 neurons is correlated with behavioral choice. In particular, wS1 neurons projecting to S2 were found to increase their spike rate, predicting the choice (20, 22). Correlation with choice was found to be strongest for the feedback S2 to wS1 projecting neurons (23). Hence, these new findings, although seemingly contradicting earlier views, emphasize the role of feedback loops through which sensory information reverberates (26) before transforming into perception and action.
Most decision-making studies in rodents, and in part, primates, rely on stereotyped behavioral tasks requiring intensive weeks-long training to obey artificial decision rules that may affect underlying neural computations. To gain further insights into processes of transformation of sensory inputs into perception and decision, it is crucial (27) to employ behavioral tasks that are closer to the animals’ behavior in their natural habitat, which brain circuits were specialized to produce during evolution. Animal decision-making is often associated with fast dynamics of navigating terrain, outrunning predators, or pursuing prey that echoes fast ms-scale dynamics of “reverberations” of sensory information along feedback loops. While 2-photon microscopy used in previous studies (10, 11, 17, 22–25) can provide access to thousands of neurons simultaneously, it is typically limited to shallow cortical layers and difficulty resolving fast neural dynamics.
Here, addressing these challenges, we harnessed the natural whisker-guided navigation behavior of mice using tactile virtual reality (VR) (17, 28, 29) combined with dense multielectrode array electrophysiological recordings. Head-fixed mice navigate at elevated speeds through a virtual corridor, defined by a pair of motorized walls, using only their whiskers, closely resembling navigation through burrows in their natural habitat (28). As opposed to traditional designs, our task is untrained and unrewarded. Leaving only a pair of C2 whiskers for mice to navigate with, we designed an information bottleneck for sensory input to decision-making (29), strategically placing a microelectrode probe into the C2 principal whisker barrel and aligning it perpendicular to the cortical layers to record neural activity across entire cortical depth.
This design reveals unexpectedly strong correlation of neural activity across the columnar circuits with DVs. In fact, during decision formation, variance of the high-dimensional spiking activity collapses to just a single latent variable, while spiking synchronously ramps up across the cortical column. We show that this variable is consistent with models of gradual accumulation of noisy sensory evidence to a decision bound (5, 7, 26, 30). These observations indicate that the primary sensory cortex not only retains relatively weak and sparse choice-related activity, as it was shown previously (10, 11, 19–25), but also, in our naturalistic behavioral task, directly participates in a categorical coding of all-or-none DV into which the majority of neural populations are recruited.
Loading cortical circuits with difficult yet manageable cognitive tasks and at the same time elicit ethologically relevant behavior, we designed a naturalistic tactile VR (17, 28, 29) paradigm (Materials and Methods) where head-fixed mice run on a 3D treadmill, while navigating using only their whiskers (Fig. 1A). The animal’s movement in lateral (Vlat) and forward (Vfor) directions is used to provide positive feedback to movement of the walls (Vwall) positioned at both sides of the snout, generating a closed-loop (CL) VR corridor for the animal to follow (Fig. 1B). Importantly, all but the C2 whiskers (second longest vibrissae oriented parallel to the ground) are trimmed, restricting sensory information flow into the primary somatosensory cortex.
Typically, animals require only 3 d of habituation sessions with CL straight trials (Fig. 1C) to acclimatize to head-fixed locomotion (Materials and Methods) with water rewards given only to encourage running at fast speeds (21 ± 4 cm/s, mean ± SD, SI Appendix, Fig. S1A). When habituated, animals undergo a single terminal 2-h long behavioral session that is unrewarded (Fig. 1C). To generate an analogous two-alternative forced choice (2AFC) paradigm, we introduce an oddball deviant stimulus (OL turn trials, Fig. 1D). After a baseline CL straight section, the feedback signal is terminated, and walls are briefly (within 12 ms) removed beyond the reach of the C2 whiskers to produce consistent sensory input across trials. Then just one wall approaches the snout (within 0.2 s) and stays at a predefined constant distance for 1.8 s before retracting again (Fig. 1D). Sensing the fast-approaching obstacle with their C2 whiskers, animals are forced to change running direction (OL left or OL right turns, Fig. 1D) to avoid an expected collision. Typical 20 s-long trajectory of the animal’s locomotion is presented in Fig. 1E exhibiting sections of straight runs (CL) and randomized left (OL_L) and right (OL_R) turns performed while the animal is running at elevated speed.
We previously showed (29) that local ablation of layer 4 (L4) of the principal C2 whisker barrel column, but not neighboring barrels, results in the inability of mice to produce turns guided by a whisker contralateral to the lesion, while still performing turns guided by the ipsilateral whisker. This is consistent with previous reports on impairment of whisker-guided gap crossing after subcolumnar microinfarction (31). Thus, we created a bottleneck of sensory information critical for making decisions in our untrained and unrewarded 2AFC task.
Content provided by PNASB (www.pnas.com) Note: Content, including the the headline, may have been edited for style and length.
