Sensory neurons are often tuned to particular stimulus features, but their responses to repeated presentation of the same stimulus can vary over subsequent trials. This presents a problem for understanding the functioning of the brain, because downstream neuronal populations ought to construct accurate stimulus representations, even upon singular exposure. To study how trial-by-trial fluctuations (i.e., noise) in activity influence cortical representations of sensory input, we performed chronic calcium imaging of GCaMP6-expressing populations in mouse V1. We observed that high-dimensional response correlations, i.e., dependencies in activation strength among multiple neurons, can be used to predict single-trial, single-neuron noise. These multidimensional correlations are structured such that variability in the response of single neurons is relatively harmless to population representations of visual stimuli. We propose that multidimensional coding may represent a canonical principle of cortical circuits, explaining why the apparent noisiness of neuronal responses is compatible with accurate neural representations of stimulus features.

Progress in science requires standardized assays whose results can be readily shared, compared, and reproduced across laboratories. Reproducibility, however, has been a concern in neuroscience, particularly for measurements of mouse behavior. Here, we show that a standardized task to probe decision-making in mice produces reproducible results across multiple laboratories. We adopted a task for head-fixed mice that assays perceptual and value-based decision making, and we standardized training protocol and experimental hardware, software, and procedures. We trained 140 mice across seven laboratories in three countries, and we collected 5 million mouse choices into a publicly available database. Learning speed was variable across mice and laboratories, but once training was complete there were no significant differences in behavior across laboratories. Mice in different laboratories adopted similar reliance on visual stimuli, on past successes and failures, and on estimates of stimulus prior probability to guide their choices. These results reveal that a complex mouse behavior can be reproduced across multiple laboratories. They establish a standard for reproducible rodent behavior, and provide an unprecedented dataset and open-access tools to study decision-making in mice. More generally, they indicate a path toward achieving reproducibility in neuroscience through collaborative open-science approaches.

The sensory neocortex is a highly connected associative network that integrates information from multiple senses, even at the level of the primary sensory areas. Although a growing body of empirical evidence supports this view, the neural mechanisms of cross-modal integration in primary sensory areas, such as the primary visual cortex (V1), are still largely unknown. Using two-photon calcium imaging in awake mice, we show that the encoding of audiovisual stimuli in V1 neuronal populations is highly dependent on the features of the stimulus constituents. When the visual and auditory stimulus features were modulated at the same rate (i.e., temporally congruent), neurons responded with either an enhancement or suppression compared with unisensory visual stimuli, and their prevalence was balanced. Temporally incongruent tones or white-noise bursts included in audiovisual stimulus pairs resulted in predominant response suppression across the neuronal population. Visual contrast did not influence multisensory processing when the audiovisual stimulus pairs were congruent; however, when white-noise bursts were used, neurons generally showed response suppression when the visual stimulus contrast was high whereas this effect was absent when the visual contrast was low. Furthermore, a small fraction of V1 neurons, predominantly those located near the lateral border of V1, responded to sound alone. These results show that V1 is involved in the encoding of cross-modal interactions in a more versatile way than previously thought. The neural substrate of cross-modal integration is not limited to specialized cortical association areas but extends to primary sensory areas. Using two-photon imaging of large groups of neurons, we show that multisensory modulation of V1 populations is strongly determined by the individual and shared features of cross-modal stimulus constituents, such as contrast, frequency, congruency, and temporal structure. Congruent audiovisual stimulation resulted in a balanced pattern of response enhancement and suppression compared with unisensory visual stimuli, whereas incongruent or dissimilar stimuli at full contrast gave rise to a population dominated by response-suppressing neurons. Our results indicate that V1 dynamically integrates nonvisual sources of information while still attributing most of its resources to coding visual information.

Noninvasive behavioral tracking of animals is crucial for many scientific investigations. Recent transfer learning approaches for behavioral tracking have considerably advanced the state of the art. Typically these methods treat each video frame and each object to be tracked independently. In this work, we improve on these methods (particularly in the regime of few training labels) by leveraging the rich spatiotemporal structures pervasive in behavioral video --- specifically, the spatial statistics imposed by physical constraints (e.g., paw to elbow distance), and the temporal statistics imposed by smoothness from frame to frame. We propose a probabilistic graphical model built on top of deep neural networks, Deep Graph Pose (DGP), to leverage these useful spatial and temporal constraints, and develop an efficient structured variational approach to perform inference in this model. The resulting semi-supervised model exploits both labeled and unlabeled frames to achieve significantly more accurate and robust tracking while requiring users to label fewer training frames. In turn, these tracking improvements enhance performance on downstream applications, including robust unsupervised segmentation of behavioral "syllables," and estimation of interpretable "disentangled" low-dimensional representations of the full behavioral video.