Volatile working memory representations crystallize with practice.

Working memory, the process through which information is transiently maintained and manipulated over a brief period, is essential for most cognitive functions1-4. However, the mechanisms underlying the generation and evolution of working-memory neuronal representations at the population level over long timescales remain unclear. Here, to identify these mechanisms, we trained head-fixed mice to perform an olfactory delayed-association task in which the mice made decisions depending on the sequential identity of two odours separated by a 5 s delay. Optogenetic inhibition of secondary motor neurons during the late-delay and choice epochs strongly impaired the task performance of the mice. Mesoscopic calcium imaging of large neuronal populations of the secondary motor cortex (M2), retrosplenial cortex (RSA) and primary motor cortex (M1) showed that many late-delay-epoch-selective neurons emerged in M2 as the mice learned the task. Working-memory late-delay decoding accuracy substantially improved in the M2, but not in the M1 or RSA, as the mice became experts. During the early expert phase, working-memory representations during the late-delay epoch drifted across days, while the stimulus and choice representations stabilized. In contrast to single-plane layer 2/3 (L2/3) imaging, simultaneous volumetric calcium imaging of up to 73,307 M2 neurons, which included superficial L5 neurons, also revealed stabilization of late-delay working-memory representations with continued practice. Thus, delay- and choice-related activities that are essential for working-memory performance drift during learning and stabilize only after several days of expert performance.

Decoding and perturbing decision states in real time.

In dynamic environments, subjects often integrate multiple samples of a signal and combine them to reach a categorical judgment1. The process of deliberation can be described by a time-varying decision variable (DV), decoded from neural population activity, that predicts a subject's upcoming decision2. Within single trials, however, there are large moment-to-moment fluctuations in the DV, the behavioural significance of which is unclear. Here, using real-time, neural feedback control of stimulus duration, we show that within-trial DV fluctuations, decoded from motor cortex, are tightly linked to decision state in macaques, predicting behavioural choices substantially better than the condition-averaged DV or the visual stimulus alone. Furthermore, robust changes in DV sign have the statistical regularities expected from behavioural studies of changes of mind3. Probing the decision process on single trials with weak stimulus pulses, we find evidence for time-varying absorbing decision bounds, enabling us to distinguish between specific models of decision making.

Shared Dorsal Periaqueductal Gray Activation Patterns during Exposure to Innate and Conditioned Threats.

The brainstem dorsal periaqueductal gray (dPAG) has been widely recognized as being a vital node orchestrating the responses to innate threats. Intriguingly, recent evidence also shows that the dPAG mediates defensive responses to fear conditioned contexts. However, it is unknown whether the dPAG displays independent or shared patterns of activation during exposure to innate and conditioned threats. It is also unclear how dPAG ensembles encode and predict diverse defensive behaviors. To address this question, we used miniaturized microscopes to obtain recordings of the same dPAG ensembles during exposure to a live predator and a fear conditioned context in male mice. dPAG ensembles encoded not only distance to threat, but also relevant features, such as predator speed and angular offset between mouse and threat. Furthermore, dPAG cells accurately encoded numerous defensive behaviors, including freezing, stretch-attend postures, and escape. Encoding of behaviors and of distance to threat occurred independently in dPAG cells. dPAG cells also displayed a shared representation to encode these behaviors and distance to threat across innate and conditioned threats. Last, we also show that escape could be predicted by dPAG activity several seconds in advance. Thus, dPAG activity dynamically tracks key kinematic and behavioral variables during exposure to threats, and exhibits similar patterns of activation during defensive behaviors elicited by innate or conditioned threats. These data indicate that a common pathway may be recruited by the dPAG during exposure to a wide variety of threat modalities.SIGNIFICANCE STATEMENT The dorsal periaqueductal gray (dPAG) is critical to generate defensive behaviors during encounters with threats of multiple modalities. Here we use longitudinal calcium transient recordings of dPAG ensembles in freely moving mice to show that this region uses shared patterns of activity to represent distance to an innate threat (a live predator) and a conditioned threat (a shock grid). We also show that dPAG neural activity can predict diverse defensive behaviors. These data indicate the dPAG uses conserved population-level activity patterns to encode and coordinate defensive behaviors during exposure to both innate and conditioned threats.

Long-Term Characterization of Hippocampal Remapping during Contextual Fear Acquisition and Extinction.

Hippocampal CA1 place cell spatial maps are known to alter their firing properties in response to contextual fear conditioning, a process called "remapping." In the present study, we use chronic calcium imaging to examine remapping during fear retrieval and extinction of an inhibitory avoidance task in mice of both sexes over an extended period of time and with thousands of neurons. We demonstrate that hippocampal ensembles encode space at a finer scale following fear memory acquisition. This effect is strongest near the shock grid. We also characterize the long-term effects of shock on place cell ensemble stability, demonstrating that shock delivery induces several days of high fear and low between-session place field stability, followed by a new, stable spatial representation that appears after fear extinction. Finally, we identify a novel group of CA1 neurons that robustly encode freeze behavior independently from spatial location. Thus, following fear acquisition, hippocampal CA1 place cells sharpen their spatial tuning and dynamically change spatial encoding stability throughout fear learning and extinction.SIGNIFICANCE STATEMENT The hippocampus contains place cells that encode an animal's location. This spatial code updates, or remaps, in response to environmental change. It is known that contextual fear can induce such remapping; in the present study, we use chronic calcium imaging to examine inhibitory avoidance-induced remapping over an extended period of time and with thousands of neurons and demonstrate that hippocampal ensembles encode space at a finer scale following electric shock, an effect which is enhanced by threat proximity. We also identify a novel group of freeze behavior-activated neurons. These results suggest that, more than merely shuffling their spatial code following threat exposure, place cells enhance their spatial coding with the possible benefit of improved threat localization.

Inferring single-trial neural population dynamics using sequential auto-encoders.

Neuroscience is experiencing a revolution in which simultaneous recording of thousands of neurons is revealing population dynamics that are not apparent from single-neuron responses. This structure is typically extracted from data averaged across many trials, but deeper understanding requires studying phenomena detected in single trials, which is challenging due to incomplete sampling of the neural population, trial-to-trial variability, and fluctuations in action potential timing. We introduce latent factor analysis via dynamical systems, a deep learning method to infer latent dynamics from single-trial neural spiking data. When applied to a variety of macaque and human motor cortical datasets, latent factor analysis via dynamical systems accurately predicts observed behavioral variables, extracts precise firing rate estimates of neural dynamics on single trials, infers perturbations to those dynamics that correlate with behavioral choices, and combines data from non-overlapping recording sessions spanning months to improve inference of underlying dynamics.

Redundant Information Neural Estimation.

We introduce the Redundant Information Neural Estimator (RINE), a method that allows efficient estimation for the component of information about a target variable that is common to a set of sources, known as the "redundant information". We show that existing definitions of the redundant information can be recast in terms of an optimization over a family of functions. In contrast to previous information decompositions, which can only be evaluated for discrete variables over small alphabets, we show that optimizing over functions enables the approximation of the redundant information for high-dimensional and continuous predictors. We demonstrate this on high-dimensional image classification and motor-neuroscience tasks.

Latent Factors and Dynamics in Motor Cortex and Their Application to Brain-Machine Interfaces.

In the 1960s, Evarts first recorded the activity of single neurons in motor cortex of behaving monkeys (Evarts, 1968). In the 50 years since, great effort has been devoted to understanding how single neuron activity relates to movement. Yet these single neurons exist within a vast network, the nature of which has been largely inaccessible. With advances in recording technologies, algorithms, and computational power, the ability to study these networks is increasing exponentially. Recent experimental results suggest that the dynamical properties of these networks are critical to movement planning and execution. Here we discuss this dynamical systems perspective and how it is reshaping our understanding of the motor cortices. Following an overview of key studies in motor cortex, we discuss techniques to uncover the "latent factors" underlying observed neural population activity. Finally, we discuss efforts to use these factors to improve the performance of brain-machine interfaces, promising to make these findings broadly relevant to neuroengineering as well as systems neuroscience.

Considerations in using recurrent neural networks to probe neural dynamics.

Recurrent neural networks (RNNs) are increasingly being used to model complex cognitive and motor tasks performed by behaving animals. RNNs are trained to reproduce animal behavior while also capturing key statistics of empirically recorded neural activity. In this manner, the RNN can be viewed as an in silico circuit whose computational elements share similar motifs with the cortical area it is modeling. Furthermore, because the RNN's governing equations and parameters are fully known, they can be analyzed to propose hypotheses for how neural populations compute. In this context, we present important considerations when using RNNs to model motor behavior in a delayed reach task. First, by varying the network's nonlinear activation and rate regularization, we show that RNNs reproducing single-neuron firing rate motifs may not adequately capture important population motifs. Second, we find that even when RNNs reproduce key neurophysiological features on both the single neuron and population levels, they can do so through distinctly different dynamical mechanisms. To distinguish between these mechanisms, we show that an RNN consistent with a previously proposed dynamical mechanism is more robust to input noise. Finally, we show that these dynamics are sufficient for the RNN to generalize to tasks it was not trained on. Together, these results emphasize important considerations when using RNN models to probe neural dynamics.NEW & NOTEWORTHY Artificial neurons in a recurrent neural network (RNN) may resemble empirical single-unit activity but not adequately capture important features on the neural population level. Dynamics of RNNs can be visualized in low-dimensional projections to provide insight into the RNN's dynamical mechanism. RNNs trained in different ways may reproduce neurophysiological motifs but do so with distinctly different mechanisms. RNNs trained to only perform a delayed reach task can generalize to perform tasks where the target is switched or the target location is changed.

Trial-by-Trial Motor Cortical Correlates of a Rapidly Adapting Visuomotor Internal Model.

Accurate motor control is mediated by internal models of how neural activity generates movement. We examined neural correlates of an adapting internal model of visuomotor gain in motor cortex while two macaques performed a reaching task in which the gain scaling between the hand and a presented cursor was varied. Previous studies of cortical changes during visuomotor adaptation focused on preparatory and perimovement epochs and analyzed trial-averaged neural data. Here, we recorded simultaneous neural population activity using multielectrode arrays and focused our analysis on neural differences in the period before the target appeared. We found that we could estimate the monkey's internal model of the gain using the neural population state during this pretarget epoch. This neural correlate depended on the gain experienced during recent trials and it predicted the speed of the subsequent reach. To explore the utility of this internal model estimate for brain-machine interfaces, we performed an offline analysis showing that it can be used to compensate for upcoming reach extent errors. Together, these results demonstrate that pretarget neural activity in motor cortex reflects the monkey's internal model of visuomotor gain on single trials and can potentially be used to improve neural prostheses.SIGNIFICANCE STATEMENT When generating movement commands, the brain is believed to use internal models of the relationship between neural activity and the body's movement. Visuomotor adaptation tasks have revealed neural correlates of these computations in multiple brain areas during movement preparation and execution. Here, we describe motor cortical changes in a visuomotor gain change task even before a specific movement is cued. We were able to estimate the gain internal model from these pretarget neural correlates and relate it to single-trial behavior. This is an important step toward understanding the sensorimotor system's algorithms for updating its internal models after specific movements and errors. Furthermore, the ability to estimate the internal model before movement could improve motor neural prostheses being developed for people with paralysis.

A Non-Human Primate Brain-Computer Typing Interface.

Brain-computer interfaces (BCIs) record brain activity and translate the information into useful control signals. They can be used to restore function to people with paralysis by controlling end effectors such as computer cursors and robotic limbs. Communication neural prostheses are BCIs that control user interfaces on computers or mobile devices. Here we demonstrate a communication prosthesis by simulating a typing task with two rhesus macaques implanted with electrode arrays. The monkeys used two of the highest known performing BCI decoders to type out words and sentences when prompted one symbol/letter at a time. On average, Monkeys J and L achieved typing rates of 10.0 and 7.2 words per minute (wpm), respectively, copying text from a newspaper article using a velocity-only two dimensional BCI decoder with dwell-based symbol selection. With a BCI decoder that also featured a discrete click for key selection, typing rates increased to 12.0 and 7.8 wpm. These represent the highest known achieved communication rates using a BCI. We then quantified the relationship between bitrate and typing rate and found it approximately linear: typing rate in wpm is nearly three times bitrate in bits per second. We also compared the metrics of achieved bitrate and information transfer rate and discuss their applicability to real-world typing scenarios. Although this study cannot model the impact of cognitive load of word and sentence planning, the findings here demonstrate the feasibility of BCIs to serve as communication interfaces and represent an upper bound on the expected achieved typing rate for a given BCI throughput.

Control of feeding by a bottom-up midbrain-subthalamic pathway.

Investigative exploration and foraging leading to food consumption have vital importance, but are not well-understood. Since GABAergic inputs to the lateral and ventrolateral periaqueductal gray (l/vlPAG) control such behaviors, we dissected the role of vgat-expressing GABAergic l/vlPAG cells in exploration, foraging and hunting. Here, we show that in mice vgat l/vlPAG cells encode approach to food and consumption of both live prey and non-prey foods. The activity of these cells is necessary and sufficient for inducing food-seeking leading to subsequent consumption. Activation of vgat l/vlPAG cells produces exploratory foraging and compulsive eating without altering defensive behaviors. Moreover, l/vlPAG vgat cells are bidirectionally interconnected to several feeding, exploration and investigation nodes, including the zona incerta. Remarkably, the vgat l/vlPAG projection to the zona incerta bidirectionally controls approach towards food leading to consumption. These data indicate the PAG is not only a final downstream target of top-down exploration and foraging-related inputs, but that it also influences these behaviors through a bottom-up pathway.

A cortical information bottleneck during decision-making.

Decision-making emerges from distributed computations across multiple brain areas, but it is unclear why the brain distributes the computation. In deep learning, artificial neural networks use multiple areas (or layers) to form optimal representations of task inputs. These optimal representations are sufficient to perform the task well, but minimal so they are invariant to other irrelevant variables. We recorded single neurons and multiunits in dorsolateral prefrontal cortex (DLPFC) and dorsal premotor cortex (PMd) in monkeys during a perceptual decision-making task. We found that while DLPFC represents task-related inputs required to compute the choice, the downstream PMd contains a minimal sufficient, or optimal, representation of the choice. To identify a mechanism for how cortex may form these optimal representations, we trained a multi-area recurrent neural network (RNN) to perform the task. Remarkably, DLPFC and PMd resembling representations emerged in the early and late areas of the multi-area RNN, respectively. The DLPFC-resembling area partially orthogonalized choice information and task inputs and this choice information was preferentially propagated to downstream areas through selective alignment with inter-area connections, while remaining task information was not. Our results suggest that cortex uses multi-area computation to form minimal sufficient representations by preferential propagation of relevant information between areas.

A pilot study of closed-loop neuromodulation for treatment-resistant post-traumatic stress disorder.

The neurophysiological mechanisms in the human amygdala that underlie post-traumatic stress disorder (PTSD) remain poorly understood. In a first-of-its-kind pilot study, we recorded intracranial electroencephalographic data longitudinally (over one year) in two male individuals with amygdala electrodes implanted for the management of treatment-resistant PTSD (TR-PTSD) under clinical trial NCT04152993. To determine electrophysiological signatures related to emotionally aversive and clinically relevant states (trial primary endpoint), we characterized neural activity during unpleasant portions of three separate paradigms (negative emotional image viewing, listening to recordings of participant-specific trauma-related memories, and at-home-periods of symptom exacerbation). We found selective increases in amygdala theta (5-9 Hz) bandpower across all three negative experiences. Subsequent use of elevations in low-frequency amygdala bandpower as a trigger for closed-loop neuromodulation led to significant reductions in TR-PTSD symptoms (trial secondary endpoint) following one year of treatment as well as reductions in aversive-related amygdala theta activity. Altogether, our findings provide early evidence that elevated amygdala theta activity across a range of negative-related behavioral states may be a promising target for future closed-loop neuromodulation therapies in PTSD.

GABAergic CA1 neurons are more stable following context changes than glutamatergic cells.

The CA1 region of the hippocampus contains both glutamatergic pyramidal cells and GABAergic interneurons. Numerous reports have characterized glutamatergic CAMK2A cell activity, showing how these cells respond to environmental changes such as local cue rotation and context re-sizing. Additionally, the long-term stability of spatial encoding and turnover of these cells across days is also well-characterized. In contrast, these classic hippocampal experiments have never been conducted with CA1 GABAergic cells. Here, we use chronic calcium imaging of male and female mice to compare the neural activity of VGAT and CAMK2A cells during exploration of unaltered environments and also during exposure to contexts before and after rotating and changing the length of the context across multiple recording days. Intriguingly, compared to CAMK2A cells, VGAT cells showed decreased remapping induced by environmental changes, such as context rotations and contextual length resizing. However, GABAergic neurons were also less likely than glutamatergic neurons to remain active and exhibit consistent place coding across recording days. Interestingly, despite showing significant spatial remapping across days, GABAergic cells had stable speed encoding between days. Thus, compared to glutamatergic cells, spatial encoding of GABAergic cells is more stable during within-session environmental perturbations, but is less stable across days. These insights may be crucial in accurately modeling the features and constraints of hippocampal dynamics in spatial coding.

An artificial intelligence that increases simulated brain-computer interface performance.

Objective.Brain-computer interfaces (BCIs) translate neural activity into control signals for assistive devices in order to help people with motor disabilities communicate effectively. In this work, we introduce a new BCI architecture that improves control of a BCI computer cursor to type on a virtual keyboard.Approach.Our BCI architecture incorporates an external artificial intelligence (AI) that beneficially augments the movement trajectories of the BCI. This AI-BCI leverages past user actions, at both long (100 s of seconds ago) and short (100 s of milliseconds ago) timescales, to modify the BCI's trajectories.Main results.We tested our AI-BCI in a closed-loop BCI simulator with nine human subjects performing a typing task. We demonstrate that our AI-BCI achieves: (1) categorically higher information communication rates, (2) quicker ballistic movements between targets, (3) improved precision control to 'dial in' on targets, and (4) more efficient movement trajectories. We further show that our AI-BCI increases performance across a wide control quality spectrum from poor to proficient control.Significance.This AI-BCI architecture, by increasing BCI performance across all key metrics evaluated, may increase the clinical viability of BCI systems.

Dorsal premammillary projection to periaqueductal gray controls escape vigor from innate and conditioned threats.

Escape from threats has paramount importance for survival. However, it is unknown if a single circuit controls escape vigor from innate and conditioned threats. Cholecystokinin (cck)-expressing cells in the hypothalamic dorsal premammillary nucleus (PMd) are necessary for initiating escape from innate threats via a projection to the dorsolateral periaqueductal gray (dlPAG). We now show that in mice PMd-cck cells are activated during escape, but not other defensive behaviors. PMd-cck ensemble activity can also predict future escape. Furthermore, PMd inhibition decreases escape speed from both innate and conditioned threats. Inhibition of the PMd-cck projection to the dlPAG also decreased escape speed. Intriguingly, PMd-cck and dlPAG activity in mice showed higher mutual information during exposure to innate and conditioned threats. In parallel, human functional magnetic resonance imaging data show that a posterior hypothalamic-to-dlPAG pathway increased activity during exposure to aversive images, indicating that a similar pathway may possibly have a related role in humans. Our data identify the PMd-dlPAG circuit as a central node, controlling escape vigor elicited by both innate and conditioned threats.

Dorsal periaqueductal gray ensembles represent approach and avoidance states.

Animals must balance needs to approach threats for risk assessment and to avoid danger. The dorsal periaqueductal gray (dPAG) controls defensive behaviors, but it is unknown how it represents states associated with threat approach and avoidance. We identified a dPAG threatavoidance ensemble in mice that showed higher activity farther from threats such as the open arms of the elevated plus maze and a predator. These cells were also more active during threat avoidance behaviors such as escape and freezing, even though these behaviors have antagonistic motor output. Conversely, the threat approach ensemble was more active during risk assessment behaviors and near threats. Furthermore, unsupervised methods showed that avoidance/approach states were encoded with shared activity patterns across threats. Lastly, the relative number of cells in each ensemble predicted threat avoidance across mice. Thus, dPAG ensembles dynamically encode threat approach and avoidance states, providing a flexible mechanism to balance risk assessment and danger avoidance.

Coordination of escape and spatial navigation circuits orchestrates versatile flight from threats.

Naturalistic escape requires versatile context-specific flight with rapid evaluation of local geometry to identify and use efficient escape routes. It is unknown how spatial navigation and escape circuits are recruited to produce context-specific flight. Using mice, we show that activity in cholecystokinin-expressing hypothalamic dorsal premammillary nucleus (PMd-cck) cells is sufficient and necessary for context-specific escape that adapts to each environment's layout. In contrast, numerous other nuclei implicated in flight only induced stereotyped panic-related escape. We reasoned the dorsal premammillary nucleus (PMd) can induce context-specific escape because it projects to escape and spatial navigation nuclei. Indeed, activity in PMd-cck projections to thalamic spatial navigation circuits is necessary for context-specific escape induced by moderate threats but not panic-related stereotyped escape caused by perceived asphyxiation. Conversely, the PMd projection to the escape-inducing dorsal periaqueductal gray projection is necessary for all tested escapes. Thus, PMd-cck cells control versatile flight, engaging spatial navigation and escape circuits.

Deep Learning Neural Encoders for Motor Cortex.

Intracortical brain-machine interfaces (BMIs) transform neural activity into control signals to drive a prosthesis or communication device, such as a robotic arm or computer cursor. To be clinically viable, BMI decoders must achieve high accuracy and robustness. Optimizing these decoders is expensive, traditionally requiring animal or human experiments spanning months to years. This is because BMIs are closed-loop systems, where the user updates his or her motor commands in response to an imperfectly decoded output. Decoder optimization using previously collected "offline" data will therefore not capture this closed-loop response. An alternative approach to significantly accelerate decoder optimization is to use a closed-loop experimental simulator. A key component of this simulator is the neural encoder, which synthetically generates neural population activity from kinematics. Prior neural encoders do not model important features of neural population activity. To overcome these limitations, we use deep learning neural encoders. We find these models significantly outperform prior neural encoders in reproducing peri-stimulus time histograms (PSTHs) and neural population dynamics. We also find that deep learning neural encoders better match neural decoding results in offline data and closed-loop experimental data. We anticipate these deep-learning neural encoders will substantially improve simulators for BMIs, enabling faster evaluation, optimization, and characterization of BMI decoder algorithms.

Structure in Neural Activity during Observed and Executed Movements Is Shared at the Neural Population Level, Not in Single Neurons.

In multiple cortical areas, including the motor cortex, neurons have similar firing rate statistics whether we observe or execute movements. These "congruent" neurons are hypothesized to support action understanding by participating in a neural circuit consistently activated in both observed and executed movements. We examined this hypothesis by analyzing neural population structure and dynamics between observed and executed movements. We find that observed and executed movements exhibit similar neural population covariation in a shared subspace capturing significant neural variance. Further, neural dynamics are more similar between observed and executed movements within the shared subspace than outside it. Finally, we find that this shared subspace has a heterogeneous composition of congruent and incongruent neurons. Together, these results argue that similar neural covariation and dynamics between observed and executed movements do not occur via activation of a subpopulation of congruent single neurons, but through consistent temporal activation of a heterogeneous neural population.