Carlos D. Brody

Decision making is associated with frontal brain circuits ^1–5 and spatial navigation with the hippocampus ^6–8 . In addition, recent work in spatial decision making tasks ^9–11 found single neurons in both areas encoding space conjunctively with other task-relevant variables. However, circuit function is not determined by tuning alone, but also by representational geometry, i.e. the representation of task-relevant variables in neural state space. Here, using Neuropixel ^12,13 recordings in a complex spatial decision making task combined with nonlinear dimensionality reduction ^8,14 , we show an intrinsically low-dimensional neural manifold in medial prefrontal cortex (mPFC) on which key task variables were represented as smooth gradients. This geometry resembled the hippocampal (HPC) map. The mPFC and HPC manifolds from one mouse can predict the behavior across other mice and brain areas. A non-linear representational map between the mPFC and HPC manifolds demonstrates alignment in time. Our work suggests that the representational geometry in HPC and mPFC is distributed and time-aligned using low-dimensional neural codes.

How does cortical connectivity support decision-making? Behavioral tasks often involve multiple sequential phases that implement different computations. For perceptual decision-making during navigation these can include evidence accumulation, decision commitment, and motor program read out. How are these different phases implemented in circuits with fixed anatomical synaptic connectivity? One potential contribution is that the connectivity of neurons is modulated in the different phases, but this has never been tested. Here we used an all-optical method to probe the causal connectivity of excitatory neurons in layer 2/3 of mouse retrosplenial cortex during different behavioral epochs of a navigation-based decision-making task, as well as in the absence of the task. In-task connectivity was different from no-task connectivity: furthermore, these differences were selective to the cue / decision phase, tapering off in later stages of the task. We propose that fast modulation of connectivity is a prevalent mechanism in neural circuit function.

Decision making is traditionally thought to be mediated by neurons that accumulate evidence through persistent activity. However, recent decision-making experiments in rodents have observed neurons across the brain that fire sequentially, rather than persistently, with the subset of neurons in the sequence depending on the animal’s choice. We developed two candidate circuit models in which neurons are active sequentially and transfer evidence faithfully to the next active population. One model encodes evidence in the relative firing of two competing chains of neurons, and the other in the network location of a stereotyped, bump-like pattern of neural activity. Neural recordings from four brain regions during an evidence accumulation task revealed that different regions displayed evidence tuning consistent with different candidate models. This work provides a mechanistic explanation for how graded information may be precisely accumulated within and transferred between neural populations, and suggests that different brain regions may accumulate evidence through different circuit mechanisms.

Cognition is thought to emerge from the flexible organization of neural activity, yet how this organization reconfigures across behaviors varying in cognitive load remains unclear. We investigated how the structure of working-memory representations in the cortex compares to task representations that do not involve working memory. We used a task-switching paradigm in virtual reality, where mice alternated between a navigation-based working-memory task and a simpler task with matched sensorimotor demands. During behavior, we simultaneously imaged three cortical areas: higher visual area AM, and two association areas-premotor (M2) and retrosplenial cortex. At the single-neuron level, trial-averaged activity appeared similar across tasks. However, pairwise correlations decreased during the working-memory task, particularly in association areas. In addition, the corresponding linear task subspace explained the variance of both tasks equally well, whereas the simpler task subspace failed to do so, suggesting an asymmetric relationship between them. Nonlinear dimensionality reduction revealed a shared low-dimensional structure across tasks. Yet, the organization of neuronal firing fields along this shared structure accounted for the difference in pairwise correlations: in the working-memory task, firing fields were more disjoint, especially among neurons in association areas that formed sequences along the memory dimension. Moreover, the degree of overlap between these firing fields predicted the mice's behavioral reliance on working memory. We conclude that behaviors varying in cognitive demands are supported by a single low-dimensional neural structure, which can expand or contract depending on cognitive load. We thus provide a framework for how task representations across the cortex reconfigure to support cognitive processes.

Center-port fixation is a common prerequisite for many freely-moving rodent tasks in neuroscience and psychology. However, typical protocols for shaping this behavior are non-standardized and inefficient. Moreover, motor errors in fixation termed ‘violations’ often account for a significant fraction of experimental trials, leading to notable data loss during experiments. In light of this, we developed FixGrower, a standardized protocol for center-port fixation training. FixGrower (1) requires a longer initial fixation requirement, (2) increases the required fixation duration at session boundaries customized to each animals’ performance, and (3) delays the introduction of violation penalties until the end of training. We demonstrate FixGrower decreases training time by 61%, yields low violation rates, and generalizes across rodent species and task difficulty. Moreover, the success of this curriculum is well supported by theories of operant conditioning and reinforcement learning. Our findings establish FixGrower as an efficient and broadly applicable curriculum for training fixation behavior in rodents, thereby accelerating training of many tasks in the field.

Perceptual decision-making is thought to be mediated by neuronal networks with attractor dynamics ^1,2 . However, the dynamics underlying the complex neuronal responses during decision-making remain unclear. Here we use simultaneous recordings of hundreds of neurons, combined with an unsupervised, deep-learning-based method, to discover decision-related neural dynamics in the rat frontal cortex and striatum as animals accumulate pulsatile auditory evidence. We found that trajectories evolved along two sequential regimes: an initial phase dominated by sensory inputs, followed by a phase dominated by autonomous dynamics, with the flow direction (that is, neural mode) largely orthogonal to that in the first regime. We propose that this transition marks the moment of decision commitment, that is, the time when the animal makes up its mind. To test this, we developed a simplified model of the dynamics to estimate a putative neurally inferred time of commitment (nTc) for each trial. This model captures diverse single-neuron temporal profiles, such as ramping and stepping ^3,4 . The estimated nTc values were not time locked to stimulus or response timing but instead varied broadly across trials. If nTc marks commitment, evidence before this point should affect the decision, whereas evidence afterwards should not. Behavioural analysis aligned to nTc confirmed this prediction. Our findings show that decision commitment involves a rapid, coordinated transition in dynamical regime and neural mode and suggest that nTc offers a useful neural marker for studying rapid changes in internal brain state.

Different brain systems have been hypothesized to subserve multiple “experts” that compete to generate behavior. In reinforcement learning, two general processes, one model-free (MF) and one model-based (MB), are often modeled as a mixture of agents (MoA) and hypothesized to capture differences between automaticity vs. deliberation. However, shifts in strategy cannot be captured by a static MoA. To investigate such dynamics, we present the mixture-of-agents hidden Markov model (MoA-HMM), which simultaneously learns inferred action values from a set of agents and the temporal dynamics of underlying “hidden” states that capture shifts in agent contributions over time. Applying this model to a multi-step, reward-guided task in rats reveals a progression of within-session strategies: a shift from initial MB exploration to MB exploitation, and finally to reduced engagement. The inferred states predict changes in both response time and OFC neural encoding during the task, suggesting that these states are capturing real shifts in dynamics.

The ability to flexibly switch our responses to external stimuli according to contextual information is critical for successful interactions with a complex world. Context-dependent computations are necessary across many domains^1–3, yet their neural implementations remain poorly understood. Here we developed a novel behavioural task in rats to study context-dependent selection and accumulation of evidence for decision-making^4–6. Under assumptions supported by both monkey and rat data, we first show mathematically that this computation can be supported by three dynamical solutions and that all networks performing the task implement a combination of these solutions. These solutions can be identified and tested directly with experimental data. We further show that existing electrophysiological and modelling data are compatible with the full variety of possible combinations of these solutions, suggesting that different individuals could use different combinations. To study variability across individual subjects, we developed automated, high-throughput methods to train rats on our task and trained many subjects using these methods. Consistent with theoretical predictions, neural and behavioural analyses revealed substantial heterogeneity across rats, despite uniformly good task performance. Our theory further predicts a specific link between behavioural and neural signatures, which was robustly supported in the data. In summary, our results provide an experimentally supported theoretical framework to analyse individual variability in biological and artificial systems that perform flexible decision-making tasks, open the door to cellular-resolution studies of individual variability in higher cognition, and provide insights into neural mechanisms of context-dependent computation more generally.