Representational drift: Emerging theories for continual learning and experimental future directions

Metaplasticity and continual learning: mechanisms subserving brain computer interface proficiency

Objective.Brain computer interfaces (BCIs) require substantial cognitive flexibility to optimize control performance. Remarkably, learning this control is rapid, suggesting it might be mediated by neuroplasticity mechanisms operating on very short time scales. Here, we propose a meta plasticity model of BCI learning and skill consolidation at the single cell and population levels comprised of three elements: (a) behavioral time scale synaptic plasticity (BTSP), (b) intrinsic plasticity (IP) and (c) synaptic scaling (SS) operating at time scales from seconds to minutes to hours and days. Notably, the model is able to explainrepresentational drift-a frequent and widespread phenomenon that adversely affects BCI control and continued use.Approach.We developed an all-optical approach to characterize IP, BTSP and SS with single cell resolution in awake mice using fluorescent two photon (2P) GCaMP7s imaging and optogenetic stimulation of the soma targeted ChRmineKv2.1. We further trained mice on a one-dimensional BCI control task and systematically characterized within session (seconds to minutes) learning as well as across sessions (days and weeks) with different neural ensembles.Main results.On the time scale of seconds, substantial BTSP could be induced and was followed by significant IP over minutes. Over the time scale of days and weeks, these changes could predict BCI control proficiency, suggesting that BTSP and IP might be complemented by SS to stabilize and consolidate BCI control.Significance.Our results provide early experimental support for a meta plasticity model of continual BCI learning and skill consolidation. The model predictions may be used to design and calibrate neural decoders with complete autonomy while considering the temporal and spatial scales of plasticity mechanisms. With the power of modern-day machine learning and artificial Intelligence, fully autonomous neural decoding and adaptation in BCIs might be achieved with minimal to no human intervention.

Evolving Engrams Demand Changes in Effective Cues

A longstanding principle in episodic memory research, known as the encoding specificity hypothesis, holds that an effective retrieval cue should closely match the original encoding conditions. This principle assumes that a successful retrieval cue remains static over time. Despite the broad acceptance of this idea, it conflicts with one of the most well-established findings in memory research: The dynamic and ever-changing nature of episodic memories. In this article, we propose that the most effective retrieval cue should engage with the current state of the memory, which may have shifted significantly since encoding. By redefining the criteria for successful recall, we challenge a core principle of the field and open new avenues for exploring memory accessibility, offering fresh insights into both theoretical, and applied domains.

A comprehensive suite for extracting neuron signals across multiple sessions in one-photon calcium imaging

We developed CaliAli, a comprehensive suite designed to extract neuronal signals from one-photon calcium imaging data collected across multiple sessions in free-moving conditions in mice. CaliAli incorporates information from blood vessels and neurons to correct inter-session misalignments, making it robust against non-rigid brain deformations even after substantial changes in the field of view across sessions. This also makes CaliAli robust against high neuron overlap and changes in active neuron population across sessions. CaliAli performs computationally efficient signal extraction from concatenated video sessions that enhances the detectability of weak calcium signals. Notably, CaliAli enhanced the spatial coding accuracy of extracted hippocampal CA1 neuron activity across sessions. An optogenetic tagging experiment showed that CaliAli enhanced neuronal trackability in the dentate gyrus across a time scale of weeks. Finally, dentate gyrus neurons tracked using CaliAli exhibited stable population activity for 99 days. Overall, CaliAli advances our capacity to understand the activity dynamics of neuronal ensembles over time, which is crucial for deciphering the complex neuronal substrates of natural animal behaviors.

Assistive sensory-motor perturbations influence learned neural representations

Task errors are used to learn and refine motor skills. We investigated how task assistance influences learned neural representations using Brain-Computer Interfaces (BCIs), which map neural activity into movement via a decoder. We analyzed motor cortex activity as monkeys practiced BCI with a decoder that adapted to improve or maintain performance over days. Over time, task-relevant information became concentrated in fewer neurons, unlike with fixed decoders. At the population level, task information also became largely confined to a few neural modes that accounted for an unexpectedly small fraction of the population variance. A neural network model suggests the adaptive decoders directly contribute to forming these more compact neural representations. Our findings show that assistive decoders manipulate error information used for long-term learning computations like credit assignment, which informs our understanding of motor learning and has implications for designing real-world BCIs.

Behavioral timescale synaptic plasticity in the hippocampus creates non-spatial representations during learning and is modulated by entorhinal inputs

Behavioral timescale synaptic plasticity (BTSP) is a form of synaptic potentiation where a single plateau potential in hippocampal neurons forms a place field during spatial learning. We asked whether BTSP can also form non-spatial responses in the hippocampus and what roles the medial and lateral entorhinal cortex (MEC and LEC) play in driving non-spatial BTSP. Two-photon calcium imaging of dorsal CA1 neurons while mice performed an odor-cued working memory task revealed plateau-like events which formed stable odor-specific responses. These BTSP-like events were much more frequent during the first day of task learning, suggesting that BTSP may be important for early learning. Strong single-neuron stimulation through holographic optogenetics induced plateau-like events and subsequent odor-fields, causally linking BTSP with non-spatial representations. MEC chemogenetic inhibition reduced the frequency of plateau-like events, whereas LEC inhibition reduced potentiation and field-induction probability. Calcium imaging of LEC and MEC temporammonic CA1 projections revealed that MEC axons were more strongly activated by odor presentations, while LEC axons were more odor-selective, further confirming the role of MEC in driving plateau-like events and LEC in relaying odor-specific information. Altogether, odor-specific information from LEC and strong odor-timed activity from MEC are crucial for driving BTSP in CA1, which is a synaptic plasticity mechanism for generation of both spatial and non-spatial responses in the hippocampus.

Synaptic architecture of a memory engram in the mouse hippocampus

Memory engrams are formed through experience-dependent plasticity of neural circuits, but their detailed architectures remain unresolved. Using three-dimensional electron microscopy, we performed nanoscale reconstructions of the hippocampal CA3-CA1 pathway after chemogenetic labeling of cellular ensembles recruited during associative learning. Neurons with a remote history of activity coinciding with memory acquisition showed no strong preference for wiring with each other. Instead, their connectomes expanded through multisynaptic boutons independently of the coactivation state of postsynaptic partners. The rewiring of ensembles representing an initial engram was accompanied by input-specific, spatially restricted upscaling of individual synapses, as well as remodeling of mitochondria, smooth endoplasmic reticulum, and interactions with astrocytes. Our findings elucidate the physical hallmarks of long-term memory and offer a structural basis for the cellular flexibility of information coding.

Sampling representational plasticity of simple imagined movements across days enables long-term neuroprosthetic control

The nervous system needs to balance the stability of neural representations with plasticity. It is unclear what the representational stability of simple well-rehearsed actions is, particularly in humans, and their adaptability to new contexts. Using an electrocorticography brain-computer interface (BCI) in tetraplegic participants, we found that the low-dimensional manifold and relative representational distances for a repertoire of simple imagined movements were remarkably stable. The manifold's absolute location, however, demonstrated constrained day-to-day drift. Strikingly, neural statistics, especially variance, could be flexibly regulated to increase representational distances during BCI control without somatotopic changes. Discernability strengthened with practice and was BCI-specific, demonstrating contextual specificity. Sampling representational plasticity and drift across days subsequently uncovered a meta-representational structure with generalizable decision boundaries for the repertoire; this allowed long-term neuroprosthetic control of a robotic arm and hand for reaching and grasping. Our study offers insights into mesoscale representational statistics that also enable long-term complex neuroprosthetic control.

Beyond Mechanism-Extending Our Concepts of Causation in Neuroscience

In neuroscience, the search for the causes of behaviour is often just taken to be the search for neural mechanisms. This view typically involves three forms of causal reduction: first, from the ontological level of cognitive processes to that of neural mechanisms; second, from the activity of the whole brain to that of isolated parts; and third, from a consideration of temporally extended, historical processes to a focus on synchronic states. While modern neuroscience has made impressive progress in identifying synchronic neural mechanisms, providing unprecedented real-time control of behaviour, we contend that this does not amount to a full causal explanation. In particular, there is an attendant danger of eliminating the cognitive from our explanatory framework, and even eliminating the organism itself. To fully understand the causes of behaviour, we need to understand not just what happens when different neurons are activated, but why those things happen. In this paper, we introduce a range of well-developed, non-reductive, and temporally extended notions of causality from philosophy, which neuroscientists may be able to draw on in order to build more complete causal explanations of behaviour. These include concepts of criterial causation, triggering versus structuring causes, constraints, macroscopic causation, historicity, and semantic causation-all of which, we argue, can be used to undergird a naturalistic understanding of mental causation and agent causation. These concepts can, collectively, help bring cognition and the organism itself back into the picture, as a causal agent unto itself, while still grounding causation in respectable scientific terms.

Complementary benefits of multivariate and hierarchical models for identifying individual differences in cognitive control

Understanding individual differences in cognitive control is a central goal in psychology and neuroscience. Reliably measuring these differences, however, has proven extremely challenging, at least when using standard measures in cognitive neuroscience such as response times or task-based fMRI activity. While prior work has pinpointed the source of the issue-the vast amount of cross-trial variability within these measures-solutions remain elusive. Here, we propose one potential way forward: an analytic framework that combines hierarchical Bayesian modeling with multivariate decoding of trial-level fMRI data. Using this framework and longitudinal data from the Dual Mechanisms of Cognitive Control project, we estimated individuals' neural responses associated with cognitive control within a color-word Stroop task, then assessed the reliability of these individuals' responses across a time interval of several months. We show that in many prefrontal and parietal brain regions, test-retest reliability was near maximal, and that only hierarchical models were able to reveal this state of affairs. Further, when compared to traditional univariate contrasts, multivariate decoding enabled individual-level correlations to be estimated with significantly greater precision. We specifically link these improvements in precision to the optimized suppression of cross-trial variability in decoding. Together, these findings not only indicate that cognitive control-related neural responses individuate people in a highly stable manner across time, but also suggest that integrating hierarchical and multivariate models provides a powerful approach for investigating individual differences in cognitive control, one that can effectively address the issue of high-variability measures.

Transcriptional determinants of goal-directed learning and representational drift in the parahippocampal cortex

Task learning involves learning associations between stimuli and outcomes and storing these relationships in memory. While this information can be reliably decoded from population activity, individual neurons encoding this representation can drift over time. The circuit or molecular mechanisms underlying this drift and its role in learning are unclear. We performed two-photon calcium imaging in the perirhinal cortex during task training. Using post hoc spatial transcriptomics, we measured immediate-early gene (IEG) expression and assigned monitored neurons to excitatory or inhibitory subtypes. We discovered an IEG-defined network spanning multiple subtypes that form stimulus-outcome associations. Targeted deletion of brain-derived neurotrophic factor in the perirhinal cortex disrupted IEG expression and impaired task learning. Representational drift slowed with prolonged training. Pre-existing representations were strengthened while stimulus-reward associations failed to form. Our findings reveal the cell types and molecules regulating long-term network stability that is permissive for task learning and memory allocation.

Experience directs the instability of neuronal tuning for critical period plasticity in mouse visual cortex

Brief monocular deprivation during a developmental critical period, but not thereafter, alters the receptive field properties (tuning) of neurons in visual cortex, but the characteristics of neural circuitry that permit this experience-dependent plasticity are largely unknown. We performed repeated calcium imaging at neuronal resolution to track the tuning properties of populations of excitatory layer 2/3 neurons in mouse visual cortex during or after the critical period, as well as in nogo-66 receptor (ngr1) mutant mice that sustain critical-period plasticity as adults. The instability of tuning for populations of neurons was greater in juvenile mice and adult ngr1 mutant mice. We propose instability of neuronal tuning gates plasticity and is directed by experience to alter the tuning of neurons during the critical period.

A combinatorial neural code for long-term motor memory

Motor skill repertoire can be stably retained over long periods, but the neural mechanism that underlies stable memory storage remains poorly understood1-8. Moreover, it is unknown how existing motor memories are maintained as new motor skills are continuously acquired. Here we tracked neural representation of learned actions throughout a significant portion of the lifespan of a mouse and show that learned actions are stably retained in combination with context, which protects existing memories from erasure during new motor learning. We established a continual learning paradigm in which mice learned to perform directional licking in different task contexts while we tracked motor cortex activity for up to six months using two-photon imaging. Within the same task context, activity driving directional licking was stable over time with little representational drift. When learning new task contexts, new preparatory activity emerged to drive the same licking actions. Learning created parallel new motor memories instead of modifying existing representations. Re-learning to make the same actions in the previous task context re-activated the previous preparatory activity, even months later. Continual learning of new task contexts kept creating new preparatory activity patterns. Context-specific memories, as we observed in the motor system, may provide a solution for stable memory storage throughout continual learning.

Representational drift in barrel cortex is receptive field dependent

Cortical populations often exhibit changes in activity even when behavior is stable. How behavioral stability is maintained in the face of such "representational drift" remains unclear. One possibility is that some neurons are more stable than others. We examined whisker touch responses in layers 2-4 of the primary vibrissal somatosensory cortex (vS1) over several weeks in mice stably performing an object detection task with two whiskers. Although the number of touch neurons remained constant, individual neurons changed with time. Touch-responsive neurons with broad receptive fields were more stable than narrowly tuned neurons. Transitions between functional types were non-random: before becoming broadly tuned, unresponsive neurons first passed through a period of narrower tuning. Broadly tuned neurons in layers 2 and 3 with higher pairwise correlations to other touch neurons were more stable than neurons with lower correlations. Thus, a small population of broadly tuned and synchronously active touch neurons exhibits elevated stability and may be particularly important for behavior.

An optical brain-machine interface reveals a causal role of posterior parietal cortex in goal-directed navigation

Cortical circuits contain diverse sensory, motor, and cognitive signals, and form densely recurrent networks. This creates challenges for identifying causal relationships between neural populations and behavior. We developed a calcium imaging-based brain-machine interface (BMI) to study the role of posterior parietal cortex (PPC) in controlling navigation in virtual reality. By training a decoder to estimate navigational heading and velocity from PPC activity during virtual navigation, we discovered that mice could immediately navigate toward goal locations when control was switched to BMI. No learning or adaptation was observed during BMI, indicating that naturally occurring PPC activity patterns are sufficient to drive navigational trajectories in real time. During successful BMI trials, decoded trajectories decoupled from the mouse's physical movements, suggesting that PPC activity relates to intended trajectories. Our work demonstrates a role for PPC in navigation and offers a BMI approach for investigating causal links between neural activity and behavior.

Analysis methods for large-scale neuronal recordings

Simultaneous recordings from hundreds or thousands of neurons are becoming routine because of innovations in instrumentation, molecular tools, and data processing software. Such recordings can be analyzed with data science methods, but it is not immediately clear what methods to use or how to adapt them for neuroscience applications. We review, categorize, and illustrate diverse analysis methods for neural population recordings and describe how these methods have been used to make progress on longstanding questions in neuroscience. We review a variety of approaches, ranging from the mathematically simple to the complex, from exploratory to hypothesis-driven, and from recently developed to more established methods. We also illustrate some of the common statistical pitfalls in analyzing large-scale neural data.

Memory engram stability and flexibility

Many studies have shown that memories are encoded in sparse neural ensembles distributed across the brain. During the post-encoding period, often during sleep, many of the cells that were active during encoding are reactivated, supporting consolidation of this memory. During memory recall, many of the same cells that were active during encoding and reactivated during consolidation are reactivated during recall. These ensembles of cells have been referred to as the memory engram cells, stably representing a specific memory. However, recent studies question the rigidity of the "stable memory engram." Here we review the past literature of how episodic-like memories are encoded, consolidated, and recalled. We also highlight more recent studies (as well as some older literature) that suggest that these stable memories and their representations are much more dynamic and flexible than previously thought. We highlight some of these processes, including memory updating, reconsolidation, forgetting, schema learning, memory-linking, and representational drift.

Representational dissimilarity component analysis (ReDisCA)

The principle of Representational Similarity Analysis (RSA) posits that neural representations reflect the structure of encoded information, allowing exploration of spatial and temporal organization of brain information processing. Traditional RSA when applied to EEG or MEG data faces challenges in accessing activation time series at the brain source level due to modeling complexities and insufficient geometric/anatomical data. To overcome this, we introduce Representational Dissimilarity Component Analysis (ReDisCA), a method for estimating spatial-temporal components in EEG or MEG responses aligned with a target representational dissimilarity matrix (RDM). ReDisCA yields informative spatial filters and associated topographies, offering insights into the location of "representationally relevant" sources. Applied to evoked response time series, ReDisCA produces temporal source activation profiles with the desired RDM. Importantly, while ReDisCA does not require inverse modeling its output is consistent with EEG and MEG observation equation and can be used as an input to rigorous source localization procedures. Demonstrating ReDisCA's efficacy through simulations and comparison with conventional methods, we show superior source localization accuracy and apply the method to real EEG and MEG datasets, revealing physiologically plausible representational structures without inverse modeling. ReDisCA adds to the family of inverse modeling free methods such as independent component analysis (Makeig, 1995), Spatial spectral decomposition (Nikulin, 2011), and Source power comodulation (Dähne, 2014) designed for extraction sources with desired properties from EEG or MEG data. Extending its utility beyond EEG and MEG analysis, ReDisCA is likely to find application in fMRI data analysis and exploration of representational structures emerging in multilayered artificial neural networks.

The representation of decision variables in orbitofrontal cortex is longitudinally stable

The computation and comparison of subjective values underlying economic choices rely on the orbitofrontal cortex (OFC). In this area, distinct groups of neurons encode the value of individual options, the binary choice outcome, and the chosen value. These variables capture both the choice input and the choice output, suggesting that the cell groups found in the OFC constitute the building blocks of a decision circuit. Here, we show that this neural circuit is longitudinally stable. Using two-photon calcium imaging, we record from the OFC of mice engaged in a juice-choice task. Imaging of individual cells continues for up to 40 weeks. For each cell and each session pair, we compare activity profiles using cosine similarity, and we assess whether the neuron encodes the same variable in both sessions. We find a high degree of stability and a modest representational drift. Quantitative estimates indicate that this drift would not randomize the circuit within the animal's lifetime.

Measuring instability in chronic human intracortical neural recordings towards stable, long-term brain-computer interfaces

Intracortical brain-computer interfaces (iBCIs) enable people with tetraplegia to gain intuitive cursor control from movement intentions. To translate to practical use, iBCIs should provide reliable performance for extended periods of time. However, performance begins to degrade as the relationship between kinematic intention and recorded neural activity shifts compared to when the decoder was initially trained. In addition to developing decoders to better handle long-term instability, identifying when to recalibrate will also optimize performance. We propose a method, "MINDFUL", to measure instabilities in neural data for useful long-term iBCI, without needing labels of user intentions. Longitudinal data were analyzed from two BrainGate2 participants with tetraplegia as they used fixed decoders to control a computer cursor spanning 142 days and 28 days, respectively. We demonstrate a measure of instability that correlates with changes in closed-loop cursor performance solely based on the recorded neural activity (Pearson r = 0.93 and 0.72, respectively). This result suggests a strategy to infer online iBCI performance from neural data alone and to determine when recalibration should take place for practical long-term use.

2P-NucTag: on-demand phototagging for molecular analysis of functionally identified cortical neurons

Neural circuits are characterized by genetically and functionally diverse cell types. A mechanistic understanding of circuit function is predicated on linking the genetic and physiological properties of individual neurons. However, it remains highly challenging to map the transcriptional properties to functionally heterogeneous neuronal subtypes in mammalian cortical circuits in vivo. Here, we introduce a high-throughput two-photon nuclear phototagging (2P-NucTag) approach optimized for on-demand and indelible labeling of single neurons via a photoactivatable red fluorescent protein following in vivo functional characterization in behaving mice. We demonstrate the utility of this function-forward pipeline by selectively labeling and transcriptionally profiling previously inaccessible 'place' and 'silent' cells in the mouse hippocampus. Our results reveal unexpected differences in gene expression between these hippocampal pyramidal neurons with distinct spatial coding properties. Thus, 2P-NucTag opens a new way to uncover the molecular principles that govern the functional organization of neural circuits.

Centering cognitive neuroscience on task demands and generalization

Cognitive neuroscience seeks generalizable theories explaining the relationship between behavioral, physiological and mental states. In pursuit of such theories, we propose a theoretical and empirical framework that centers on understanding task demands and the mutual constraints they impose on behavior and neural activity. Task demands emerge from the interaction between an agent's sensory impressions, goals and behavior, which jointly shape the activity and structure of the nervous system on multiple spatiotemporal scales. Understanding this interaction requires multitask studies that vary more than one experimental component (for example, stimuli and instructions) combined with dense behavioral and neural sampling and explicit testing for generalization across tasks and data modalities. By centering task demands rather than mental processes that tasks are assumed to engage, this framework paves the way for the discovery of new generalizable concepts unconstrained by existing taxonomies, and moves cognitive neuroscience toward an action-oriented, dynamic and integrated view of the brain.

Degraded tactile coding in the Cntnap2 mouse model of autism

Atypical sensory processing is common in autism, but how neural coding is disrupted in sensory cortex is unclear. We evaluate whisker touch coding in L2/3 of somatosensory cortex (S1) in Cntnap2-/- mice, which have reduced inhibition. This classically predicts excess pyramidal cell spiking, but this remains controversial, and other deficits may dominate. We find that c-fos expression is elevated in S1 of Cntnap2-/- mice under spontaneous activity conditions but is comparable to that of control mice after whisker stimulation, suggesting normal sensory-evoked spike rates. GCaMP8m imaging from L2/3 pyramidal cells shows no excess whisker responsiveness, but it does show multiple signs of degraded somatotopic coding. This includes broadened whisker-tuning curves, a blurred whisker map, and blunted whisker point representations. These disruptions are greater in noisy than in sparse sensory conditions. Tuning instability across days is also substantially elevated in Cntnap2-/-. Thus, Cntnap2-/- mice show no excess sensory-evoked activity, but a degraded and unstable tactile code in S1.

Temporal information in the anterior cingulate cortex relates to accumulated experiences

Anterior cingulate cortex (ACC) activity is important for operations that require the ability to integrate multiple experiences over time, such as rule learning, cognitive flexibility, working memory, and long-term memory recall. To shed light on this, we analyzed neuronal activity while rats repeated the same behaviors during hour-long sessions to investigate how activity changed over time. We recorded neuronal ensembles as rats performed a decision-free operant task with varying reward likelihoods at three different response ports (n = 5). Neuronal state space analysis revealed that each repetition of a behavior was distinct, with more recent behaviors more similar than those further apart in time. ACC activity was dominated by a slow, gradual change in low-dimensional representations of neural state space aligning with the pace of behavior. Temporal progression, or drift, was apparent on the top principal component for every session and was driven by the accumulation of experiences and not an internal clock. Notably, these signals were consistent across subjects, allowing us to accurately predict trial numbers based on a model trained on data from a different animal. We observed that non-continuous ramping firing rates over extended durations (tens of minutes) drove the low-dimensional ensemble representations. 40% of ACC neurons' firing ramped over a range of trial lengths and combinations of shorter duration ramping neurons created ensembles that tracked longer durations. These findings provide valuable insights into how the ACC, at an ensemble level, conveys temporal information by reflecting the accumulation of experiences over extended periods.

Prefrontal cortical ripples mediate top-down suppression of hippocampal reactivation during sleep memory consolidation

Consolidation of initially encoded hippocampal representations in the neocortex through reactivation is crucial for long-term memory formation and is facilitated by the coordination of hippocampal sharp-wave ripples (SWRs) with cortical slow and spindle oscillations during non-REM sleep. Recent evidence suggests that high-frequency cortical ripples can also coordinate with hippocampal SWRs in support of consolidation; however, the contribution of cortical ripples to reactivation remains unclear. We used high-density, continuous recordings in the hippocampus (area CA1) and prefrontal cortex (PFC) over the course of spatial learning and show that independent PFC ripples dissociated from SWRs are prevalent in NREM sleep and predominantly suppress hippocampal activity. PFC ripples paradoxically mediate top-down suppression of hippocampal reactivation rather than coordination, and this suppression is stronger for assemblies that are reactivated during coordinated CA1-PFC ripples for consolidation of recent experiences. Further, we show non-canonical, serial coordination of independent cortical ripples with slow and spindle oscillations, which are known signatures of memory consolidation. These results establish a role for prefrontal cortical ripples in top-down regulation of behaviorally relevant hippocampal representations during consolidation.

A combinatorial neural code for long-term motor memory

Motor skill repertoire can be stably retained over long periods, but the neural mechanism underlying stable memory storage remains poorly understood. Moreover, it is unknown how existing motor memories are maintained as new motor skills are continuously acquired. Here we tracked neural representation of learned actions throughout a significant portion of a mouse's lifespan, and we show that learned actions are stably retained in motor memory in combination with context, which protects existing memories from erasure during new motor learning. We used automated home-cage training to establish a continual learning paradigm in which mice learned to perform directional licking in different task contexts. We combined this paradigm with chronic two-photon imaging of motor cortex activity for up to 6 months. Within the same task context, activity driving directional licking was stable over time with little representational drift. When learning new task contexts, new preparatory activity emerged to drive the same licking actions. Learning created parallel new motor memories while retaining the previous memories. Re-learning to make the same actions in the previous task context re-activated the previous preparatory activity, even months later. At the same time, continual learning of new task contexts kept creating new preparatory activity patterns. Context-specific memories, as we observed in the motor system, may provide a solution for stable memory storage throughout continual learning. Learning in new contexts produces parallel new representations instead of modifying existing representations, thus protecting existing motor repertoire from erasure.

Magnetic voluntary head-fixation in transgenic rats enables lifespan imaging of hippocampal neurons

The precise neural mechanisms within the brain that contribute to the remarkable lifetime persistence of memory are not fully understood. Two-photon calcium imaging allows the activity of individual cells to be followed across long periods, but conventional approaches require head-fixation, which limits the type of behavior that can be studied. We present a magnetic voluntary head-fixation system that provides stable optical access to the brain during complex behavior. Compared to previous systems that used mechanical restraint, there are no moving parts and animals can engage and disengage entirely at will. This system is failsafe, easy for animals to use and reliable enough to allow long-term experiments to be routinely performed. Animals completed hundreds of trials per session of an odor discrimination task that required 2-4 s fixations. Together with a reflectance fluorescence collection scheme that increases two-photon signal and a transgenic Thy1-GCaMP6f rat line, we are able to reliably image the cellular activity in the hippocampus during behavior over long periods (median 6 months), allowing us track the same neurons over a large fraction of animals' lives (up to 19 months).

Paradoxical replay can protect contextual task representations from destructive interference when experience is unbalanced

Experience replay is a powerful mechanism to learn efficiently from limited experience. Despite several decades of compelling experimental results, the factors that determine which experiences are selected for replay remain unclear. A particular challenge for current theories is that on tasks that feature unbalanced experience, rats paradoxically replay the less-experienced trajectory. To understand why, we simulated a feedforward neural network with two regimes: rich learning (structured representations tailored to task demands) and lazy learning (unstructured, task-agnostic representations). Rich, but not lazy, representations degraded following unbalanced experience, an effect that could be reversed with paradoxical replay. To test if this computational principle can account for the experimental data, we examined the relationship between paradoxical replay and learned task representations in the rat hippocampus. Strikingly, we found a strong association between the richness of learned task representations and the paradoxicality of replay. Taken together, these results suggest that paradoxical replay specifically serves to protect rich representations from the destructive effects of unbalanced experience, and more generally demonstrate a novel interaction between the nature of task representations and the function of replay in artificial and biological systems.

Drift of neural ensembles driven by slow fluctuations of intrinsic excitability

Representational drift refers to the dynamic nature of neural representations in the brain despite the behavior being seemingly stable. Although drift has been observed in many different brain regions, the mechanisms underlying it are not known. Since intrinsic neural excitability is suggested to play a key role in regulating memory allocation, fluctuations of excitability could bias the reactivation of previously stored memory ensembles and therefore act as a motor for drift. Here, we propose a rate-based plastic recurrent neural network with slow fluctuations of intrinsic excitability. We first show that subsequent reactivations of a neural ensemble can lead to drift of this ensemble. The model predicts that drift is induced by co-activation of previously active neurons along with neurons with high excitability which leads to remodeling of the recurrent weights. Consistent with previous experimental works, the drifting ensemble is informative about its temporal history. Crucially, we show that the gradual nature of the drift is necessary for decoding temporal information from the activity of the ensemble. Finally, we show that the memory is preserved and can be decoded by an output neuron having plastic synapses with the main region.

Representational drift as a result of implicit regularization

Recent studies show that, even in constant environments, the tuning of single neurons changes over time in a variety of brain regions. This representational drift has been suggested to be a consequence of continuous learning under noise, but its properties are still not fully understood. To investigate the underlying mechanism, we trained an artificial network on a simplified navigational task. The network quickly reached a state of high performance, and many units exhibited spatial tuning. We then continued training the network and noticed that the activity became sparser with time. Initial learning was orders of magnitude faster than ensuing sparsification. This sparsification is consistent with recent results in machine learning, in which networks slowly move within their solution space until they reach a flat area of the loss function. We analyzed four datasets from different labs, all demonstrating that CA1 neurons become sparser and more spatially informative with exposure to the same environment. We conclude that learning is divided into three overlapping phases: (i) Fast familiarity with the environment; (ii) slow implicit regularization; and (iii) a steady state of null drift. The variability in drift dynamics opens the possibility of inferring learning algorithms from observations of drift statistics.

Synaptic architecture of a memory engram in the mouse hippocampus

Memory engrams are formed through experience-dependent remodeling of neural circuits, but their detailed architectures have remained unresolved. Using 3D electron microscopy, we performed nanoscale reconstructions of the hippocampal CA3-CA1 pathway following chemogenetic labeling of cellular ensembles with a remote history of correlated excitation during associative learning. Projection neurons involved in memory acquisition expanded their connectomes via multi-synaptic boutons without altering the numbers and spatial arrangements of individual axonal terminals and dendritic spines. This expansion was driven by presynaptic activity elicited by specific negative valence stimuli, regardless of the co-activation state of postsynaptic partners. The rewiring of initial ensembles representing an engram coincided with local, input-specific changes in the shapes and organelle composition of glutamatergic synapses, reflecting their weights and potential for further modifications. Our findings challenge the view that the connectivity among neuronal substrates of memory traces is governed by Hebbian mechanisms, and offer a structural basis for representational drifts.

What does the mean mean? A simple test for neuroscience

Trial-averaged metrics, e.g. tuning curves or population response vectors, are a ubiquitous way of characterizing neuronal activity. But how relevant are such trial-averaged responses to neuronal computation itself? Here we present a simple test to estimate whether average responses reflect aspects of neuronal activity that contribute to neuronal processing. The test probes two assumptions implicitly made whenever average metrics are treated as meaningful representations of neuronal activity: Reliability: Neuronal responses repeat consistently enough across trials that they convey a recognizable reflection of the average response to downstream regions.Behavioural relevance: If a single-trial response is more similar to the average template, it is more likely to evoke correct behavioural responses. We apply this test to two data sets: (1) Two-photon recordings in primary somatosensory cortices (S1 and S2) of mice trained to detect optogenetic stimulation in S1; and (2) Electrophysiological recordings from 71 brain areas in mice performing a contrast discrimination task. Under the highly controlled settings of Data set 1, both assumptions were largely fulfilled. In contrast, the less restrictive paradigm of Data set 2 met neither assumption. Simulations predict that the larger diversity of neuronal response preferences, rather than higher cross-trial reliability, drives the better performance of Data set 1. We conclude that when behaviour is less tightly restricted, average responses do not seem particularly relevant to neuronal computation, potentially because information is encoded more dynamically. Most importantly, we encourage researchers to apply this simple test of computational relevance whenever using trial-averaged neuronal metrics, in order to gauge how representative cross-trial averages are in a given context.

Spontaneous Dynamics of Hippocampal Place Fields in a Model of Combinatorial Competition among Stable Inputs

We present computer simulations illustrating how the plastic integration of spatially stable inputs could contribute to the dynamic character of hippocampal spatial representations. In novel environments of slightly larger size than typical apparatus, the emergence of well-defined place fields in real place cells seems to rely on inputs from normally functioning grid cells. Theoretically, the grid-to-place transformation is possible if a place cell is able to respond selectively to a combination of suitably aligned grids. We previously identified the functional characteristics that allow a synaptic plasticity rule to accomplish this selection by synaptic competition during rat foraging behavior. Here, we show that the synaptic competition can outlast the formation of place fields, contributing to their spatial reorganization over time, when the model is run in larger environments and the topographical/modular organization of grid inputs is taken into account. Co-simulated cells that differ only by their randomly assigned grid inputs display different degrees and kinds of spatial reorganization-ranging from place-field remapping to more subtle in-field changes or lapses in firing. The model predicts a greater number of place fields and propensity for remapping in place cells recorded from more septal regions of the hippocampus and/or in larger environments, motivating future experimental standardization across studies and animal models. In sum, spontaneous remapping could arise from rapid synaptic learning involving inputs that are functionally homogeneous, spatially stable, and minimally stochastic.

Local prediction-learning in high-dimensional spaces enables neural networks to plan

Planning and problem solving are cornerstones of higher brain function. But we do not know how the brain does that. We show that learning of a suitable cognitive map of the problem space suffices. Furthermore, this can be reduced to learning to predict the next observation through local synaptic plasticity. Importantly, the resulting cognitive map encodes relations between actions and observations, and its emergent high-dimensional geometry provides a sense of direction for reaching distant goals. This quasi-Euclidean sense of direction provides a simple heuristic for online planning that works almost as well as the best offline planning algorithms from AI. If the problem space is a physical space, this method automatically extracts structural regularities from the sequence of observations that it receives so that it can generalize to unseen parts. This speeds up learning of navigation in 2D mazes and the locomotion with complex actuator systems, such as legged bodies. The cognitive map learner that we propose does not require a teacher, similar to self-attention networks (Transformers). But in contrast to Transformers, it does not require backpropagation of errors or very large datasets for learning. Hence it provides a blue-print for future energy-efficient neuromorphic hardware that acquires advanced cognitive capabilities through autonomous on-chip learning.

Measuring instability in chronic human intracortical neural recordings towards stable, long-term brain-computer interfaces

Intracortical brain-computer interfaces (iBCIs) enable people with tetraplegia to gain intuitive cursor control from movement intentions. To translate to practical use, iBCIs should provide reliable performance for extended periods of time. However, performance begins to degrade as the relationship between kinematic intention and recorded neural activity shifts compared to when the decoder was initially trained. In addition to developing decoders to better handle long-term instability, identifying when to recalibrate will also optimize performance. We propose a method to measure instability in neural data without needing to label user intentions. Longitudinal data were analyzed from two BrainGate2 participants with tetraplegia as they used fixed decoders to control a computer cursor spanning 142 days and 28 days, respectively. We demonstrate a measure of instability that correlates with changes in closed-loop cursor performance solely based on the recorded neural activity (Pearson r = 0.93 and 0.72, respectively). This result suggests a strategy to infer online iBCI performance from neural data alone and to determine when recalibration should take place for practical long-term use.

The Representation of Decision Variables in Orbitofrontal Cortex is Longitudinally Stable

The computation and comparison of subjective values underlying economic choices rely on the orbitofrontal cortex (OFC). In this area, distinct groups of neurons encode the value of individual options, the binary choice outcome, and the chosen value. These variables capture both the input and the output of the choice process, suggesting that the cell groups found in OFC constitute the building blocks of a decision circuit. Here we show that this neural circuit is longitudinally stable. Using two-photon calcium imaging, we recorded from mice choosing between different juice flavors. Recordings of individual cells continued for up to 20 weeks. For each cell and each pair of sessions, we compared the activity profiles using cosine similarity, and we assessed whether the cell encoded the same variable in both sessions. These analyses revealed a high degree of stability and a modest representational drift. A quantitative estimate indicated this drift would not randomize the circuit within the animal's lifetime.

Visual word processing engages a hierarchical, distributed, and bilateral cortical network

Although the Visual Word Form Area (VWFA) in left temporal cortex is considered the pre-eminent region in visual word processing, other regions are also implicated. We examined the entire text-selective circuit, using functional MRI. Ten regions of interest (ROIs) per hemisphere were defined, which, based on clustering, grouped into early vision, high-level vision, and language clusters. We analyzed the responses of the ROIs and clusters to words, inverted words, and consonant strings using univariate, multivariate, and functional connectivity measures. Bilateral modulation by stimulus condition was evident, with a stronger effect in left hemisphere regions. Last, using graph theory, we observed that the VWFA was equivalently connected with early visual and language clusters in both hemispheres, reflecting its role as a mediator in the circuit. Although the individual ROIs and clusters bilaterally were flexibly altered by the nature of the input, stability held at the level of global circuit connectivity, reflecting the complex hierarchical distributed system serving visual text perception.

Representational drift as a result of implicit regularization

Recent studies show that, even in constant environments, the tuning of single neurons changes over time in a variety of brain regions. This representational drift has been suggested to be a consequence of continuous learning under noise, but its properties are still not fully understood. To investigate the underlying mechanism, we trained an artificial network on a simplified navigational task. The network quickly reached a state of high performance, and many units exhibited spatial tuning. We then continued training the network and noticed that the activity became sparser with time. Initial learning was orders of magnitude faster than ensuing sparsification. This sparsification is consistent with recent results in machine learning, in which networks slowly move within their solution space until they reach a flat area of the loss function. We analyzed four datasets from different labs, all demonstrating that CA1 neurons become sparser and more spatially informative with exposure to the same environment. We conclude that learning is divided into three overlapping phases: (i) Fast familiarity with the environment; (ii) slow implicit regularization; (iii) a steady state of null drift. The variability in drift dynamics opens the possibility of inferring learning algorithms from observations of drift statistics.

Hippocampal sequences span experience relative to rewards

Hippocampal place cells fire in sequences that span spatial environments and non-spatial modalities, suggesting that hippocampal activity can anchor to the most behaviorally salient aspects of experience. As reward is a highly salient event, we hypothesized that sequences of hippocampal activity can anchor to rewards. To test this, we performed two-photon imaging of hippocampal CA1 neurons as mice navigated virtual environments with changing hidden reward locations. When the reward moved, the firing fields of a subpopulation of cells moved to the same relative position with respect to reward, constructing a sequence of reward-relative cells that spanned the entire task structure. The density of these reward-relative sequences increased with task experience as additional neurons were recruited to the reward-relative population. Conversely, a largely separate subpopulation maintained a spatially-based place code. These findings thus reveal separate hippocampal ensembles can flexibly encode multiple behaviorally salient reference frames, reflecting the structure of the experience.

Unconscious Manipulation of Conceptual Representations with Decoded Neurofeedback Impacts Search Behavior

The necessity of conscious awareness in human learning has been a long-standing topic in psychology and neuroscience. Previous research on non-conscious associative learning is limited by the low signal-to-noise ratio of the subliminal stimulus, and the evidence remains controversial, including failures to replicate. Using functional MRI decoded neurofeedback, we guided participants from both sexes to generate neural patterns akin to those observed when visually perceiving real-world entities (e.g., dogs). Importantly, participants remained unaware of the actual content represented by these patterns. We utilized an associative DecNef approach to imbue perceptual meaning (e.g., dogs) into Japanese hiragana characters that held no inherent meaning for our participants, bypassing a conscious link between the characters and the dogs concept. Despite their lack of awareness regarding the neurofeedback objective, participants successfully learned to activate the target perceptual representations in the bilateral fusiform. The behavioral significance of our training was evaluated in a visual search task. DecNef and control participants searched for dogs or scissors targets that were pre-cued by the hiragana used during DecNef training or by a control hiragana. The DecNef hiragana did not prime search for its associated target but, strikingly, participants were impaired at searching for the targeted perceptual category. Hence, conscious awareness may function to support higher-order associative learning. Meanwhile, lower-level forms of re-learning, modification, or plasticity in existing neural representations can occur unconsciously, with behavioral consequences outside the original training context. The work also provides an account of DecNef effects in terms of neural representational drift.

Reward Expectation Reduces Representational Drift in the Hippocampus

Spatial memory in the hippocampus involves dynamic neural patterns that change over days, termed representational drift. While drift may aid memory updating, excessive drift could impede retrieval. Memory retrieval is influenced by reward expectation during encoding, so we hypothesized that diminished reward expectation would exacerbate representational drift. We found that high reward expectation limited drift, with CA1 representations on one day gradually re-emerging over successive trials the following day. Conversely, the absence of reward expectation resulted in increased drift, as the gradual re-emergence of the previous day's representation did not occur. At the single cell level, lowering reward expectation caused an immediate increase in the proportion of place-fields with low trial-to-trial reliability. These place fields were less likely to be reinstated the following day, underlying increased drift in this condition. In conclusion, heightened reward expectation improves memory encoding and retrieval by maintaining reliable place fields that are gradually reinstated across days, thereby minimizing representational drift.

Cortical ripples mediate top-down suppression of hippocampal reactivation during sleep memory consolidation

Consolidation of initially encoded hippocampal representations in the neocortex through reactivation is crucial for long-term memory formation, and is facilitated by the coordination of hippocampal sharp-wave ripples (SWRs) with cortical oscillations during non-REM sleep. However, the contribution of high-frequency cortical ripples to consolidation is still unclear. We used continuous recordings in the hippocampus and prefrontal cortex (PFC) over the course of spatial learning and show that independent PFC ripples, when dissociated from SWRs, predominantly suppress hippocampal activity in non-REM sleep. PFC ripples paradoxically mediate top-down suppression of hippocampal reactivation, which is inversely related to reactivation strength during coordinated CA1-PFC ripples. Further, we show non-canonical, serial coordination of ripples with cortical slow and spindle oscillations. These results establish a role for cortical ripples in regulating consolidation.

The Precision of Place Fields Governs Their Fate across Epochs of Experience

Spatial memories are represented by hippocampal place cells during navigation. This spatial code is dynamic, undergoing changes across time, known as representational drift, and across changes in internal state, even while navigating the same spatial environment with consistent behavior. A dynamic code may provide the hippocampus a means to track distinct epochs of experience that occur at different times or during different internal states and update spatial memories. Changes to the spatial code include place fields (PFs) that remap to new locations and place fields that vanish, while others are stable. However, what determines place field fate across epochs remains unclear. We measured the lap-by-lap properties of place cells in mice during navigation for a block of trials in a rewarded virtual environment. We then determined the position of the place fields in another block of trials in the same spatial environment either separated by a day (a distinct temporal epoch) or during the same session but with reward removed to change reward expectation (a distinct internal state epoch). We found that place cells with remapped place fields across epochs tended to have lower spatial precision during navigation in the initial epoch. Place cells with stable or vanished place fields tended to have higher spatial precision. We conclude that place cells with less precise place fields have greater spatial flexibility, allowing them to respond to, and track, distinct epochs of experience in the same spatial environment, while place cells with precise place fields generally preserve spatial information when their fields reappear.

Circular and unified analysis in network neuroscience

Genuinely new discovery transcends existing knowledge. Despite this, many analyses in systems neuroscience neglect to test new speculative hypotheses against benchmark empirical facts. Some of these analyses inadvertently use circular reasoning to present existing knowledge as new discovery. Here, I discuss that this problem can confound key results and estimate that it has affected more than three thousand studies in network neuroscience over the last decade. I suggest that future studies can reduce this problem by limiting the use of speculative evidence, integrating existing knowledge into benchmark models, and rigorously testing proposed discoveries against these models. I conclude with a summary of practical challenges and recommendations.

Representational drift in barrel cortex is receptive field dependent

Cortical populations often exhibit changes in activity even when behavior is stable. How behavioral stability is maintained in the face of such 'representational drift' remains unclear. One possibility is that some neurons are stable despite broader instability. We examine whisker touch responses in superficial layers of primary vibrissal somatosensory cortex (vS1) over several weeks in mice stably performing an object detection task with two whiskers. While the number of touch neurons remained constant, individual neurons changed with time. Touch-responsive neurons with broad receptive fields were more stable than narrowly tuned neurons. Transitions between functional types were non-random: before becoming broadly tuned neurons, unresponsive neurons first pass through a period of narrower tuning. Broadly tuned neurons with higher pairwise correlations to other touch neurons were more stable than neurons with lower correlations. Thus, a small population of broadly tuned and synchronously active touch neurons exhibit elevated stability and may be particularly important for downstream readout.

Degraded tactile coding in the Cntnap2 mouse model of autism

Atypical sensory processing in autism involves altered neural circuit function and neural coding in sensory cortex, but the nature of coding disruption is poorly understood. We characterized neural coding in L2/3 of whisker somatosensory cortex (S1) of Cntnap2-/- mice, an autism model with pronounced hypofunction of parvalbumin (PV) inhibitory circuits. We tested for both excess spiking, which is often hypothesized in autism models with reduced inhibition, and alterations in somatotopic coding, using c-fos immunostaining and 2-photon calcium imaging in awake mice. In Cntnap2-/- mice, c-fos-(+) neuron density was elevated in L2/3 of S1 under spontaneous activity conditions, but comparable to control mice after whisker stimulation, suggesting that sensory-evoked spiking was relatively normal. 2-photon GCaMP8m imaging in L2/3 pyramidal cells revealed no increase in whisker-evoked response magnitude, but instead showed multiple signs of degraded somatotopic coding. These included broadening of whisker tuning curves, blurring of the whisker map, and blunting of the point representation of each whisker. These altered properties were more pronounced in noisy than sparse sensory conditions. Tuning instability, assessed over 2-3 weeks of longitudinal imaging, was also significantly increased in Cntnap2-/- mice. Thus, Cntnap2-/- mice show no excess spiking, but a degraded and unstable tactile code in S1.

The Wire Is Not the Territory: Understanding Representational Drift in Olfaction With Dynamical Systems Theory

Representational drift is a phenomenon of increasing interest in the cognitive and neural sciences. While investigations are ongoing for other sensory cortices, recent research has demonstrated the pervasiveness in which it occurs in the piriform cortex for olfaction. This gradual weakening and shifting of stimulus-responsive cells has critical implications for sensory stimulus-response models and perceptual decision-making. While representational drift may complicate traditional sensory processing models, it could be seen as an advantage in olfaction, as animals live in environments with constantly changing and unpredictable chemical information. Non-topographical encoding in the olfactory system may aid in contextualizing reactions to promiscuous odor stimuli, facilitating adaptive animal behavior and survival. This article suggests that traditional models of stimulus-(neural) response mapping in olfaction may need to be reevaluated and instead motivates the use of dynamical systems theory as a methodology and conceptual framework.

Flexible regulation of representations on a drifting manifold enables long-term stable complex neuroprosthetic control

The nervous system needs to balance the stability of neural representations with plasticity. It is unclear what is the representational stability of simple actions, particularly those that are well-rehearsed in humans, and how it changes in new contexts. Using an electrocorticography brain-computer interface (BCI), we found that the mesoscale manifold and relative representational distances for a repertoire of simple imagined movements were remarkably stable. Interestingly, however, the manifold's absolute location demonstrated day-to-day drift. Strikingly, representational statistics, especially variance, could be flexibly regulated to increase discernability during BCI control without somatotopic changes. Discernability strengthened with practice and was specific to the BCI, demonstrating remarkable contextual specificity. Accounting for drift, and leveraging the flexibility of representations, allowed neuroprosthetic control of a robotic arm and hand for over 7 months without recalibration. Our study offers insight into how electrocorticography can both track representational statistics across long periods and allow long-term complex neuroprosthetic control.

Time and experience are independent determinants of representational drift in CA1

In this issue of Neuron, Khatib et al.1 and Geva et al.2 present complementary and breakthrough discoveries demonstrating that elapsed time and active experience independently affect unique aspects of representational drift in the hippocampus.

Synaptic weights that correlate with presynaptic selectivity increase decoding performance

The activity of neurons in the visual cortex is often characterized by tuning curves, which are thought to be shaped by Hebbian plasticity during development and sensory experience. This leads to the prediction that neural circuits should be organized such that neurons with similar functional preference are connected with stronger weights. In support of this idea, previous experimental and theoretical work have provided evidence for a model of the visual cortex characterized by such functional subnetworks. A recent experimental study, however, have found that the postsynaptic preferred stimulus was defined by the total number of spines activated by a given stimulus and independent of their individual strength. While this result might seem to contradict previous literature, there are many factors that define how a given synaptic input influences postsynaptic selectivity. Here, we designed a computational model in which postsynaptic functional preference is defined by the number of inputs activated by a given stimulus. Using a plasticity rule where synaptic weights tend to correlate with presynaptic selectivity, and is independent of functional-similarity between pre- and postsynaptic activity, we find that this model can be used to decode presented stimuli in a manner that is comparable to maximum likelihood inference.

Representational drift as a window into neural and behavioural plasticity

Large-scale recordings of neural activity over days and weeks have revealed that neural representations of familiar tasks, precepts and actions continually evolve without obvious changes in behaviour. We hypothesise that this steady drift in neural activity and accompanying physiological changes is due in part to the continuous application of a learning rule at the cellular and population level. Explicit predictions of this drift can be found in neural network models that use iterative learning to optimise weights. Drift therefore provides a measurable signal that can reveal systems-level properties of biological plasticity mechanisms, such as their precision and effective learning rates.