Alexis O. Kaya - Theoretical Neuroscience Perspective
Abstract
A memory can remain retrievable for decades even though the molecules, synapses, cellular states, and population activity supporting cognition undergo continual change. The apparent contradiction between biological turnover and mnemonic endurance has encouraged two incomplete pictures: the memory as a fixed object stored somewhere in the brain, and memory as an unrestrictedly distributed process without identifiable physical constraints. This perspective argues for a more discriminating alternative: memory persistence is a multiscale achievement of neural organization, in which some relations are stabilized, others are reconstructed, and still others are replaced without necessarily destroying the capacity to remember. The argument integrates engram allocation, synaptic remodeling, systems consolidation, representational drift, non-neuronal regulation, and the distinction between preservation and accessibility. Recent work on the synaptic architecture of hippocampal engrams, the reorganization of hippocampal circuits during systems consolidation, and astrocytic contributions to memory stabilization sharpens the claim that an enduring memory need not be an unchanging anatomical trace. Building on these findings, Neurotenacity is retained as a proposed conceptual label for functionally consequential continuity across biological change—not as a newly discovered mechanism or proof that neuronal longevity causes lifelong memory. An experimentally oriented framework distinguishes trace persistence, retrieval accessibility, representational fidelity, and autobiographical integration and identifies predictions capable of separating organizational continuity from mere cell survival. The central thesis is that memory endures not by escaping biological change, but by maintaining usable relations through it.
Keywords: memory engrams; neural ensembles; systems consolidation; representational drift; synaptic remodeling; memory accessibility; Neurotenacity; autobiographical continuity
1. The Endurance Problem
The persistence of a remembered experience is familiar enough to seem unremarkable. A voice heard decades earlier can become recognizable in an instant. A place not visited since childhood can acquire unexpected familiarity. A skill may be performed fluently despite years without practice. Yet the biology responsible for these achievements is anything but static. Proteins are synthesized and degraded; receptors are inserted and removed; dendritic spines form, stabilize, and disappear; intrinsic excitability changes; cellular metabolism continuously renovates the conditions of signaling. At larger scales, hippocampal and cortical representations evolve during learning, sleep, aging, and retrieval. The organism can remember even while the material implementation of remembering changes.
This apparent paradox is often answered too quickly. To say that memory is stored in a stable network begs the question of what stability means. Must the same neurons respond to a cue across years? Must the same synapses remain present and hold equivalent strengths? Must the representation preserve all its original details? Or can a memory count as persistent if a different neural configuration supports an appropriately related recollection or behavior? These are different propositions and demand different measurements.
The argument developed here is that enduring memory requires constrained continuity, not frozen structure. At least some causally relevant organization must survive or be reconstructed with adequate functional correspondence; yet successful remembering may remain compatible with cellular turnover, shifting ensemble participation, changes in synaptic weight, and partial transformation of the remembered content. This is not an argument that every memory survives. The brain forgets, and some traces are degraded or destroyed. It is instead an attempt to specify why the existence of change does not by itself establish the loss of information.
This article therefore differs from previous reflections on neuronal persistence, the costs of replacement, and retrieval failure. Its concern is the integrative problem of how molecular, synaptic, cellular, circuit, and systems-level mechanisms jointly establish the conditions under which a remembered relation can remain recognizable after the substrate has changed.
2. From a Storage Metaphor to a Causal Trace
The idea that experience leaves a physical trace predates modern circuit neuroscience. Richard Semon’s concept of the engram treated memory as the consequence of a lasting biological modification; Hebb later articulated how coordinated neuronal activity might modify future interactions within cell assemblies (Semon, 1921; Hebb, 1949). Both ideas were valuable because they made remembering a question about mechanisms rather than an exclusively introspective phenomenon. But a trace is not necessarily a discrete object, and biological localization does not entail a simple correspondence between one cell and one remembered episode.
Contemporary engram research offers a more experimentally constrained vocabulary. Neurons recruited during learning can be labeled, manipulated, and examined at recall; in some rodent paradigms, selective reactivation of these populations is sufficient to elicit memory-related behavior. Allocation to an engram is partly influenced by neuronal excitability, and overlap among ensembles can contribute to the linking or separation of experiences (Josselyn and Frankland, 2018; Josselyn and Tonegawa, 2020). These findings justify speaking of identifiable memory-related cell populations while resisting the notion of an invariant neural filing cabinet.
A memory trace should therefore be understood as a causally relevant, distributed, and temporally evolving configuration. Its biological realization may include synaptic weights, local microcircuit structure, intrinsic excitability, coordinated ensemble dynamics, and interactions among brain regions. Its retrieval is an event involving cues, neural states, reconstruction, and behavior—not the passive display of a preserved recording.
Two qualifications follow. First, not all kinds of memory share the same implementation. Episodic recollection, semantic knowledge, procedural expertise, affective learning, and working memory recruit partially distinct systems. Second, the phrase ‘distributed memory’ does not imply that every part of the brain contributes equally to every memory. Causal necessity is task- and time-dependent, and particular circuits can be indispensable under defined experimental conditions. An architectural account must explain differentiated organization, rather than dissolve it into the vague assertion that everything is connected.
3. What May Persist Across Biological Turnover?
The central difficulty is a problem of scale. A stable psychological capacity cannot be mapped straightforwardly onto a molecule, a spine, a neuron, or an anatomical edge. Each level can change on a different timescale, and one level may compensate for changes at another. Long-lived neurons offer an enduring cellular scaffold in many brain regions, but the synapses they support remain active sites of molecular renovation. At the population level, the same behavior can sometimes be sustained while the neurons most strongly expressing a representation vary across sessions.
Longitudinal recordings illustrate this distinction. In mice, Ziv and colleagues (2013) observed that representations of a familiar environment in hippocampal CA1 were assembled from changing subsets of neurons over weeks, while a smaller overlapping component maintained consistent spatial tuning. This is evidence of representational dynamics, not proof that the animals’ episodic memories were unchanged. The broader literature on ensemble fluidity likewise treats drift as potentially compatible with, and sometimes useful for, updating learned representations (Mau, Hasselmo, and Cai, 2020).
The important distinction is between microscopic identity and functional correspondence. Microscopic identity requires exactly the same local configuration. Functional correspondence requires preservation of relations adequate to guide a particular output: discrimination between contexts, recovery of a learned association, recognition of a location, or reconstruction of an episode. One does not imply the other. A neural configuration can change without a measured loss in performance; conversely, an anatomically intact circuit can become functionally inaccessible under altered physiological conditions.
The form of preservation that matters must consequently be stated relative to an observable function. If an organism remembers a previously learned discrimination, which aspects of the underlying representation predict that outcome better than baseline firing rates, cell counts, or general network integrity? If a remote autobiographical recollection changes in detail, what exactly is being preserved: the original episode, its semantic gist, an emotional valuation, or merely the ability to give a coherent narrative? ‘Memory persistence’ is too broad a label unless these possibilities are separated.
4. The Engram Is Not an Immutable Wiring Diagram
Recent cellular and ultrastructural findings make the anatomical interpretation of engrams both more powerful and more complicated. Using three-dimensional electron microscopy of labeled hippocampal pathways in mice, Uytiepo and colleagues (2025) identified experience-associated modifications involving multisynaptic boutons, synaptic scaling, organelles, and interactions with astrocytes. Their findings did not establish a simple rule in which neurons co-active during learning preferentially wired only to one another. Instead, the study showed structural features compatible with a changing, selectively modified substrate of memory.
A consequential conclusion follows: memory-related connectivity is richer than a static list of neuronal pairs. Synaptic architecture includes the number, strength, spatial arrangement, and compartmental context of contacts, as well as postsynaptic and presynaptic properties. Two networks with superficially similar node-and-edge diagrams might have substantially different dynamics. Conversely, some effective computational relations might be sustained by different physical implementations. Neither proposition warrants assuming that a connectome alone contains a complete, decodable autobiography.
Engram experiments also distinguish cell allocation from enduring cell exclusivity. A neuron recruited to one experience can later participate in additional representations; altered excitability can shift its recruitment; reconsolidation can reorganize the neural support of a remembered event. In a study of remote memory updating in mice, Lei and colleagues (2025) reported the recruitment of new hippocampal engram populations during systems reconsolidation. The resulting lesson is not that memories are endlessly rewritten in every circumstance, but that retrieval and subsequent reorganization can be causally important under specified conditions.
For an architectural account, an engram is therefore a dynamically maintainable set of causal dispositions, not necessarily a frozen snapshot of synaptic anatomy. The hypothesis must remain open to the possibility that some memories rely more heavily on stable microstructure than others. Detecting which relations must remain intact is precisely an empirical question.
5. Systems Consolidation: Persistence Through Reorganization
Episodic memory persistence cannot be understood entirely at the level of local synapses. Experiences are encoded and later expressed through coordinated hippocampal, cortical, and other circuits, while the contributions of those circuits evolve with time and experience. Systems consolidation does not simply involve moving a complete memory from one storage container into another. Different theoretical accounts emphasize transformation, redistribution, continued hippocampal involvement, and the interaction of detailed episodic representations with more generalized knowledge.
In influential mouse experiments, Kitamura and colleagues (2017) identified prefrontal engram populations initiated during learning that underwent later functional maturation. Their findings supported a temporally reorganized network in which hippocampal and cortical engram contributions differed between recent and remote retrieval. Such a result provides an experimentally grounded meaning for continuity through change: an initially recruited circuit can help establish later functional dependence on additional circuitry.
Newer work has complicated the assumption that these transformations are exclusively cortical. Ko and colleagues (2025) reported that systems consolidation reorganizes hippocampal engram circuitry itself and that changes in hippocampal neurogenesis can alter the balance between precise event memory and generalized gist-like responding in mice. The study is especially pertinent because it demonstrates that persistence need not imply fixed precision: a memory may be retained in an altered representational form with different generalization properties. Changes in performance should not automatically be interpreted as complete erasure or perfect conservation.
Sleep-associated reactivation, neuromodulation, and ongoing activity can participate in the stabilization and modification of memory. These processes make the term ‘consolidation’ less like an instantaneous sealing operation and more like repeated coordination among changing network states. The critical theoretical task is to characterize the invariants or reliably recoverable relations that bridge these states. Without such specification, saying that a memory ‘migrates’ or that its ‘architecture persists’ supplies a metaphor rather than an explanation.
6. Beyond Neurons: The Regulatory Architecture of Memory
An exclusively neuron-centered theory also risks overlooking the physiological environments that support the trace. Astrocytes, microglia, vascular supply, extracellular ion regulation, neuromodulators, and metabolic availability can change the probability that a circuit remains plastic, stabilizes after experience, or is reactivated during a later test. These components need not each encode a specific autobiographical detail to be causally important for memory stability.
Recent experiments provide unusually direct evidence. Dewa and colleagues (2025) identified astrocytic ensembles in mice that underwent experience-dependent state changes and influenced memory stability and precision during later recall. A 2026 perspective by Sánchez Romero and Navarrete argues that astrocytic ensembles deserve consideration in accounts of the cellular substrate of memory. Such findings broaden the relevant architecture, although they do not establish that glial signals alone instantiate a complete memory or permit generalizations from rodent fear conditioning to the whole of human autobiographical experience.
The distinction between bearing information and regulating its availability is important. Some non-neuronal processes may shape engram formation or recall without constituting the same representational content as the neuronal ensemble. Other experiments may reveal a more integrated division of labor. A theory of memory persistence should not decide the issue through terminology; it should compare causal perturbations, information measures, and longitudinal changes in neuronal and non-neuronal components.
For this reason, ‘architecture’ will be used here in an explicitly expanded sense. It encompasses not only structural connectivity but the regulatory processes required for a network to remain capable of expressing or reconstructing learned relations. The architecture is living and metabolically maintained. A preserved diagram does not establish a functioning memory system.
7. Memory Preservation, Retrieval Accessibility, and Fidelity
The existence of a neural trace and the ability to retrieve it are related but distinguishable. In rodent experiments, Ryan and colleagues (2015) showed that direct optogenetic activation of tagged engram cells could elicit memory-related behavior despite amnesia under conventional testing conditions. The findings support a separation between expression of a memory through ordinary retrieval cues and the continued existence of some memory-related circuitry. They do not show that all apparently forgotten experiences remain intact, or that external stimulation restores a complete and veridical autobiographical record.
An architectural model benefits from separating four dimensions. Trace persistence concerns whether biologically consequential traces of prior learning remain detectable. Retrieval accessibility concerns whether a cue and a given network state can recruit those traces. Representational fidelity concerns how closely the retrieved response corresponds to the original event or learned discrimination. Autobiographical integration concerns whether the outcome can be situated within an individual’s broader understanding of their life. These dimensions can dissociate, and no single behavioral test measures them all.
For example, an animal might retain a conditioned response while losing the ability to discriminate between a training context and a similar one. The trace may support generalized behavior, while contextual fidelity has changed. A person may recognize a childhood place without recollecting the original event; familiarity need not demonstrate accurate episodic reinstatement. A detailed recollection can also be mistaken. The subjective certainty or vividness of remembering is not equivalent to external verification.
Forgetting is consequently heterogeneous. It can reflect interference, failed encoding, compromised retrieval, adaptive inhibition, representational transformation, or genuine loss of the supporting substrate. Studies of retrieval-dependent updating and reconsolidation offer evidence for modification after reactivation, but the exact boundary conditions and reproducibility of particular reconsolidation effects remain contested (Jardine et al., 2022; Brunswick et al., 2025). An architectural theory should resist two symmetrical errors: assuming non-recall proves erasure, and assuming any apparent recovery proves a perfectly preserved hidden memory.
8. Neurotenacity: A Proposed Principle, Not a Demonstrated Mechanism
Neurotenacity has been introduced in this series as a conceptual framework connecting neuronal endurance to the maintenance of functionally consequential organization. In its most defensible form, the term does not mean that neurons are unchanging, that every stored relation remains fixed, or that one particular cell must survive for a memory to persist. It denotes the ability of a living neural system to sustain or reconstruct selected learned relations across time despite component-level turnover.
This formulation retains the value of neuronal longevity while restricting its explanatory reach. Long-lived neurons may reduce the amount of cellular replacement required in regions carrying established representations, but longevity by itself cannot explain why some memories remain accessible, why others generalize, or why others disappear. Redundant representations, synaptic remodeling, intrinsic excitability, systems consolidation, glial regulation, sleep-dependent activity, and contextual retrieval may all contribute. Neurotenacity should be useful only if it helps organize predictions beyond the phenomena already described by these mechanisms separately.
There is also a genuine alternative to the persistence-centered interpretation. A system could maintain memory through continuous replacement if a sufficiently faithful process transferred task-relevant relations to new components. Regenerating nervous systems in other species, and experience-dependent neuronal recruitment in adult mammalian circuits, make a universal prohibition against replacement untenable. The question is not whether a neuron was born recently, but whether replacement disrupts, preserves, or transforms measurable learned relations.
An additional hazard is teleology. The relative longevity of many human neurons does not prove that evolution selected their lifespan for the specific purpose of autobiographical preservation. Multiple developmental, metabolic, anatomical, and phylogenetic constraints may be involved. The proposed framework makes no claim of a demonstrated evolutionary design. Its scientific value depends on observable relations among cellular persistence, network maintenance, information recovery, and behavior—not on the appeal of an architectural analogy.
9. Experience, Autobiographical Memory, and the Self
Long-term memory also contributes to the organization of a person’s history. Episodic recollection, semantic knowledge, learned habits, affective dispositions, and relationships provide partially overlapping forms of continuity. Autobiographical remembering integrates information about one’s past with current goals, beliefs, and circumstances; it therefore involves both preserved constraints and present reconstruction.
The implication is not that identity is simply an inventory of remembered episodes. A person may lose episodic details while retaining dispositions, attachments, procedural abilities, semantic knowledge, and social relationships. In neurodegenerative illness, these dimensions can change at different rates. Moreover, psychological continuity and numerical personal identity are not equivalent philosophical questions, and neither should be used to assign moral worth. Cognitive impairment does not abolish the dignity or status of a person.
The architectural perspective offers a restrained proposition: continuity of neural organization contributes to some of the capacities through which a person recognizes, interprets, and narrates their history. It does not establish that a connectome is sufficient for consciousness, that personal identity can be reduced to network similarity, or that memories must remain unchanged for identity to endure. The very capacity to reinterpret past experience is part of a continuing life.
In this sense, memory is neither a perfectly faithful archive nor an unrestricted invention. It is a present reconstruction constrained by biological and experiential history. Such a description allows the past to remain causally consequential without requiring the mind to preserve every past state in its original form.
10. An Empirical Program for Organized Continuity
A scientific account must specify observations that would favor or challenge it. The following research program translates the architectural proposal into testable comparisons. Its central aim is not to assign a single numerical index to ‘memory’ or ‘identity’, but to isolate which relations predict which forms of persistence.
Prediction 1 — Functional persistence despite ensemble drift. During longitudinal population recording, behaviorally stable memories should sometimes remain decodable across sessions even when individual cell participation shifts. The relevant comparison is between models trained on fixed cell identities and models tracking relational or low-dimensional population structure, controlling for task performance, neural sampling, and movement. A failure of relational measures to predict delayed performance beyond simpler activity measures would weaken this proposal.
Prediction 2 — Dissociation of trace persistence and accessibility. In paradigms with validated engram tagging, experimental conditions that reduce natural-cue recall should sometimes leave experimentally measurable trace-related responses available. Rescue by artificial activation would support an accessibility component, but it would not by itself prove unchanged content. Appropriate controls would test new learning during stimulation, nonspecific arousal, and the specificity of the recovered response. If no trace-dependent response survives, the preservation explanation should not be presumed.
Prediction 3 — Selective preservation during consolidation. Across recent and remote tests, the organization that best predicts discrimination, gist, and contextual detail should differ. Longitudinal manipulation of hippocampal and cortical ensembles would test whether systems-level reorganization preserves task-relevant relations even as mnemonic precision shifts. The possibility that the representation is transformed, rather than fully conserved, must be an explicit outcome.
Prediction 4 — Regulatory contributions independent of cell count. Perturbations to astrocytic signaling, sleep-dependent reactivation, or neuromodulatory conditions should alter memory maintenance or access even when gross neuronal survival is unchanged. This prediction is a check against reducing Neurotenacity to longevity. Null or contradictory perturbation findings would narrow the role attributed to particular regulatory pathways.
Prediction 5 — Replacement cost depends on preserved relations. Where neuronal addition or experimental replacement can be manipulated without unacceptable injury confounds, memory outcomes should track the preservation or re-establishment of task-relevant connectivity and dynamics more closely than the sheer number of new neurons. The same intervention could improve flexibility while reducing precise retrieval, and such mixed outcomes would not be treated as anomalous. If replacement preserves function through successful reorganization, the hypothesis must accommodate it rather than declare it impossible.
Across these experiments, a core methodological issue is identifiability. Similar behavior can be produced by different network configurations, and similar recordings can reflect different memories. Longitudinal cellular measurements, causal intervention, detailed behavioral probes, and cross-scale modeling must therefore be interpreted jointly. Neural similarity alone is not memory identity, and successful behavior alone is not evidence that a particular original trace was preserved.
11. Limits, Competing Accounts, and Research Boundaries
The architectural view is consistent with much contemporary memory science, but compatibility is not evidence of novelty. Its strongest claims overlap with engram theory, multiple-trace and transformation accounts of systems consolidation, theories of representational drift, and network models of robust computation. To become a distinct scientific contribution, Neurotenacity must specify measurable effects that existing constructs do not already predict, and it must accept failure if those effects are not observed.
Several limits are especially important. Most causal engram-manipulation experiments discussed here were performed in rodents, often using comparatively restricted forms of conditioning or spatial learning. Human autobiographical memory involves language, social context, semantic interpretation, and phenomenology that cannot be inferred directly from a conditioned freezing response. Anatomical and molecular findings can reveal candidate mechanisms but do not by themselves validate a theory of identity. Long-term memory also varies by type: a procedural skill and a richly contextualized episode may have different vulnerabilities and neural requirements.
Finally, continuity need not entail constancy of content. Some forgetting enables interference reduction, adaptive generalization, and abstraction; some updating corrects outdated knowledge. A theory that equates every representational change with damage would misunderstand a major function of memory. Conversely, treating all change as adaptive would deny the reality of deterioration, false recollection, and irreversible loss. The problem is to identify which changes preserve a useful relation, which transform it, and which destroy it.
Conclusion — The Living Form of Memory
Why can an experience continue to matter after the material substrate through which it was learned has changed? No single mechanism offers a complete answer. Neuronal longevity provides one form of continuity; synaptic and intrinsic plasticity adjust the participating circuits; systems consolidation redistributes and transforms their contributions; neuromodulatory and glial processes regulate their stability and accessibility. Together these processes can make learned relations usable across time. None guarantees permanent retention, complete fidelity, or unrestricted recall.
The most productive shift is therefore not from storage to architecture as though one metaphor simply replaced another. It is from the image of an inert record to a set of testable causal questions: which relations persist, at what scale, under which conditions, with what behavioral and experiential consequences? A memory can be physically instantiated without being a fixed physical object. It can remain meaningful without being unchanged. It can become inaccessible without necessarily being erased. It can also be transformed or lost.
Neurotenacity, understood provisionally as organized continuity amid biological turnover, offers a name for this problem rather than its solution. Its merit will depend upon whether it helps reveal the mechanisms by which nervous systems preserve, reconstruct, and sometimes relinquish their histories. The architecture of memory is not the architecture that escapes change. It is the living organization through which some of the past remains capable of shaping the present.
Declarations
Article type: Theoretical perspective. No original experiments, human-subject data, or animal data are reported.
Funding: None declared. Competing interests: None declared.
Data availability: No new dataset was generated or analyzed in preparing this conceptual perspective.
References
Brunswick, C. A., Carpenter, C. M., Dennis, N. A., and Kwapis, J. L. (2025). Not the same as it ever was: A review of memory modification, updating, and distortion in humans and rodents. Neuroscience & Biobehavioral Reviews, 106195. https://doi.org/10.1016/j.neubiorev.2025.106195
Dewa, K.-I., Kaseda, K., Kuwahara, A., et al. (2025). The astrocytic ensemble acts as a multiday trace to stabilize memory. Nature, 648, 146–156. https://doi.org/10.1038/s41586-025-09619-2
Hebb, D. O. (1949). The Organization of Behavior: A Neuropsychological Theory. Wiley.
Jardine, K. H., Huff, A. E., Wideman, C. E., McGraw, S. D., and Winters, B. D. (2022). The evidence for and against reactivation-induced memory updating in humans and nonhuman animals. Neuroscience & Biobehavioral Reviews, 136, 104598. https://doi.org/10.1016/j.neubiorev.2022.104598
Josselyn, S. A., and Frankland, P. W. (2018). Memory allocation: Mechanisms and function. Annual Review of Neuroscience, 41, 389–413. https://doi.org/10.1146/annurev-neuro-080317-061956
Josselyn, S. A., and Tonegawa, S. (2020). Memory engrams: Recalling the past and imagining the future. Science, 367(6473), eaaw4325. https://doi.org/10.1126/science.aaw4325
Kitamura, T., Ogawa, S. K., Roy, D. S., et al. (2017). Engrams and circuits crucial for systems consolidation of a memory. Science, 356(6333), 73–78. https://doi.org/10.1126/science.aam6808
Ko, S. Y., Rong, Y., Ramsaran, A. I., et al. (2025). Systems consolidation reorganizes hippocampal engram circuitry. Nature, 643, 735–743. https://doi.org/10.1038/s41586-025-08993-1
Lei, B., Kang, B., Hao, Y., Yang, H., Zhong, Z., Zhai, Z., and Zhong, Y. (2025). Reconstructing a new hippocampal engram for systems reconsolidation and remote memory updating. Neuron, 113(3), 471–485.e6. https://doi.org/10.1016/j.neuron.2024.11.010
Mau, W., Hasselmo, M. E., and Cai, D. J. (2020). The brain in motion: How ensemble fluidity drives memory-updating and flexibility. eLife, 9, e63550. https://doi.org/10.7554/eLife.63550
Ryan, T. J., Roy, D. S., Pignatelli, M., Arons, A., and Tonegawa, S. (2015). Engram cells retain memory under retrograde amnesia. Science, 348(6238), 1007–1013. https://doi.org/10.1126/science.aaa5542
Sánchez Romero, J., and Navarrete, M. (2026). Astroengrams: Rethinking the cellular substrate for memory. Nature Reviews Neuroscience, 27, 289–300. https://doi.org/10.1038/s41583-025-01012-2
Semon, R. (1921). The Mneme (L. Simon, Trans.). George Allen & Unwin.
Uytiepo, M., Zhu, Y., Bushong, E., et al. (2025). Synaptic architecture of a memory engram in the mouse hippocampus. Science, 387(6740), eado8316. https://doi.org/10.1126/science.ado8316
Ziv, Y., Burns, L. D., Cocker, E. D., et al. (2013). Long-term dynamics of CA1 hippocampal place codes. Nature Neuroscience, 16, 264–266. https://doi.org/10.1038/nn.3329



