Neuroscientists remain steadfastly uncertain about how the brain encodes memory

Researchers from Monash University, in collaboration with the European Biostasis Foundation and Apex Neuroscience, have revealed that although most neuroscientists agree that long-term memories depend primarily on neuronal connectivity patterns, significant uncertainties persist regarding precisely how these memories are structurally encoded.

Brains can retain memories for days, months and even across a lifetime of decades, through mechanisms that remain elusive to those at the cutting edge of neuroscience. Long-term memory enables animals to shape behaviors by linking past experiences with present contexts.

There are fragile memories, like recalling the name of someone you just met, or the location of where the keys were set down, that can seemingly escape the brain’s data capture. And there are durable memories that can survive periods of global neuronal inactivity and disruption, indicating that ongoing neural activity is not required to maintain stored information.

Distinctions between memory formation and recall also suggest that stable structural changes, rather than transient biochemical processes, underpin long-term retention.

Over the last century, numerous candidates have been proposed as the physical basis of memory storage. Structural alterations span a broad spectrum, from modifications of individual proteins and ion channels to large-scale changes in synaptic connectivity.

Proposed mechanisms include synaptic strength adjustments, synaptogenesis, intracellular molecular modifications, changes in neuronal excitability, and alterations to myelination or extracellular matrix components.

Some researchers assert that ensembles of synaptic connections form a definitive memory trace. Evidence from perturbations such as hypothermia, where fine neural structures transiently disappear without memory loss, raises doubts about the exclusivity of synaptic ensembles as memory substrates.

Many of the proposed mechanisms may coexist, collaborate or compensate for one another, complicating efforts to isolate a singular physical structure for how memory is stored.

After 100 years of study, neuroscientists lack consensus on which neurophysiological features encode long-term memories: whether a structural scale exists between molecular details and macroscopic brain features, and whether memory depends on precise molecular states or coarser patterns of connectivity, leaving the field in a persistent state of uncertainty.

Even in a field of competing hypotheses and conflicting certainties, there must be some level of consensus. So, what exactly are the consensus points in theoretical neuroscience?

In the study, “What are memories made of? A survey of neuroscientists on the structural basis of long-term memory,” published in PLOS One, researchers designed a survey to gauge expert consensus regarding memory’s physical underpinnings with questions about memory preservation and extraction.

Researchers recruited 312 neuroscientists from two distinct groups. One cohort comprised 33 “Engram Experts” who had published research directly related to memory neurophysiology. A second group included 279 attendees of the Computational and Systems Neuroscience (COSYNE) conferences from 2022 to 2024, representing a broader range of neuroscience specializations.

Email invitations reached 305 Engram Experts, yielding 33 responses and a 12.1% response rate. Contact was made with 4,125 COSYNE attendees, resulting in 279 responses and a 7.3% response rate. Completion rates for both cohorts hovered near 75%. Monetary incentives of $75 for Engram Experts and $20 for COSYNE participants aimed to improve response rates.

Participants answered 28 questions organized into six sections: demographics; beliefs about the structural basis of memory; theoretical implications of memory storage; brain preservation; whole brain emulation feasibility; and familiarity with relevant topics. Most questions required responses, with optional fields allowing additional commentary.

To address fundamental yet provocative concepts in neuroscience, there were survey questions asked around the concept of structure. Could detailed maps of neuronal structure alone contain specific memories—such as a learned route through a maze or a memorized password? Around 46% of respondents agreed this was theoretically possible, provided sufficiently sophisticated methods existed. Roughly a third strongly disagreed.

Respondents identified additional information potentially necessary for memory readout, most frequently citing dynamically changing neuronal activity patterns (47%), contextual data on experiences and mental states (43%), and sensory-motor information (37%). Less frequently selected were chemical and electrical gradients (15%) and quantum-level properties (5%).

A majority of 70% agreed that lasting changes in neuronal connectivity and synaptic strength primarily constitute the structural basis of long-term memories, as opposed to molecular or subcellular details.

When asked which physical details would have to be measured in order to decode a specific long-term memory from a static, preserved brain, agreement rose with scale and resolution.

A majority of respondents judged that the types and precise locations of individual biomolecules must be captured. Nearly everyone agreed that sub-cellular structures of ~500 nm and larger are indispensable, while atomic or quantum-level particulars were generally viewed as unnecessary for decoding memory.

Turning to practical implications, the survey probed whether current brain preservation methods might someday permit memory decoding.

The survey found a median 41% belief (as a probability) that brains preserved using aldehyde-stabilized cryopreservation (ASC) retain sufficient information to decode some long-term memories. Responses exhibited a bimodal distribution with peaks near 10% and 75%, suggesting respondents held dramatically split views.

Participants estimated a 40% median probability that a whole brain emulation could (eventually) be created from an ASC-preserved brain without prior electrophysiological recordings, increasing to 62% if such recordings were available beforehand.

Interestingly, neither research background, experimental versus computational, nor expertise level significantly influenced respondents’ perspectives. Expert memory researchers were somewhat more skeptical of decoding memories from preserved brains, but the difference was not statistically significant.

Participants were asked to predict when whole brain emulations might become achievable, estimating median timelines for various species. Predictions converging on whole brain emulations of the worm Caenorhabditis elegans suggest it would likely be achieved by 2045, mice by 2065, and humans by 2125.

Most neuroscientists endorse the notion that long-term memories reside in stable structural features of the brain, primarily involving lasting changes in neuronal connectivity and synaptic strength. Yet, substantial disagreement persists about which specific neurophysiological features or at which spatial scale critical physical substrates of long-term memory storage are contained.

Estimates for the theoretical feasibility of decoding memories from preserved brains or creating whole brain emulations varied widely, indicating unresolved fundamental questions.

According to the researchers, the absence of consensus within neuroscience is not itself a problem, but the result of research setting off in different directions in attempts to fill the profound gaps in neuroscientific knowledge about memory’s physical basis.

They suggest that advances in observational technologies and computational neuroscience may ultimately clarify these disparate ambiguities.

© 2025 Science X Network