The Brain Does Not Store Experience in One Place

This week

This week’s papers redraw a familiar but overly tidy picture of the brain. Memory is not simply filed in the hippocampus; sleep is not one uniform maintenance mode; learning is not merely the accumulation of accurate evidence; and disease is not always a matter of a broken gene or a missing cell. Again and again, researchers found systems that preserve experience by redistributing it—across circuits, cell states, bodily rhythms, and time. That makes the brain more resilient than our textbook diagrams suggest. It also makes it more vulnerable to subtle changes in mood, development, inflammation and sleep.

Memory has more than one overnight route

For decades, the hippocampus has been cast as the night-shift manager of memory: during sleep, it replays recent events and helps train the cortex. A striking mouse study now shows that this is not the only route to consolidation. When mice learned a procedural task, the dorsal striatum replayed neural sequences associated with the skill during offline periods, and the particular sequences replayed predicted later improvement. Crucially, this replay survived complete bilateral hippocampal lesions. The striatum was not waiting for hippocampal instructions; it could run its own consolidation program. [1]

That does not diminish the hippocampus so much as give it a more specific role. In people undergoing intracranial recordings for epilepsy, hippocampal ripple events—brief, high-frequency bursts—preceded a striking expansion in the dimensionality of cortical activity during successful episodic retrieval. The idea is geometrical but intuitive: a compressed hippocampal cue may unfold into a richer, more detailed cortical reconstruction of an experience. When this expansion occurred, participants retrieved associations faster and more accurately. [2]

Together, these findings suggest that “replay” is not a single operation with a single source. The hippocampus may be especially important for reconstructing the relational richness of an episode; the striatum may consolidate skills through its own local replay machinery. Sleep, on this view, is less a central broadcast and more a period in which several memory systems conduct parallel conversations.

Those conversations are shaped by emotion. Across several human experiments, inducing a happy or sad mood after people had encountered emotional material changed what they remembered at least a day later. A later mood biased both the affective tone assigned to past events and recognition toward mood-congruent material—especially for experiences initially judged arousing or personally relevant. [3] The implication is unsettling and important: consolidation is not a neutral delay between experience and recall. What happens while a memory settles can reach backward into how the experience is ultimately represented.

Sleep rhythms may also determine whose memories benefit. In autistic mice, disruptions in thalamic reticular sleep spindles were associated with impaired social memory. Boosting a specific thalamic circuit restored both spindles and social-memory performance in a Neuroligin 2 model; spindle features also helped classify autism in a group of children. [4] This does not make spindles a diagnostic test, nor establish a treatment for autistic people. But it offers a concrete bridge between a sleep rhythm, circuit physiology and a socially meaningful behavior.

Learning means knowing what to ignore—and whom to follow

The brain’s challenge is not only storing information. It must decide what counts.

In infants from the United Kingdom and Singapore, an adult’s eye contact increased the likelihood that infants would learn an artificial language. More remarkably, this social selectivity was mediated by a directional coupling signal from the adult speaker’s brain activity to that of the infant listener. The adult-to-infant measure predicted learning better than the infant’s neural activity alone. [5] “Brain-to-brain coupling” can sound mystical; this result is more grounded. It suggests that live social cues help organize an infant’s attention so that some information is treated as worth learning.

In a very different species, flies appear to make a similarly active judgment about their sensory world. In a virtual odor landscape, Drosophila did not merely reflexively chase an attractive odor plume. They tracked its edge, periodically turning out of the plume and then steering back toward the remembered direction of its boundary. Neurons in the central complex encoded that return vector. [6] A companion study found neural activity that both accumulated evidence across successive odor encounters and persisted after odor disappeared—working-memory-like dynamics that allowed flies to maintain an upwind heading. [7]

These studies make navigation look less like stimulus-response behavior and more like inference. The fly keeps a running internal estimate of where it ought to be; the infant uses a social partner’s gaze to prioritize what should enter memory. Neither animal merely receives information. Each filters it through a model of relevance.

That filtering can systematically distort decisions. Humans and mice gave rare outcomes disproportionate and persistent weight, even when doing so moved them away from optimal choices. In mice, the posterior parietal cortex drove this rarity bias early in learning, but later became necessary for using sensory evidence efficiently. [8] Learning did not simply suppress a bias: it repurposed the circuit that had produced it.

The brain’s vulnerabilities begin before symptoms

Several papers this week argue that neurological and psychiatric conditions are not only failures of neurons. They are failures of the tissue environments—and developmental histories—that make neural computation possible.

A sweeping multi-omic study of 40 human hippocampi across adulthood finds that aging alters gene regulation at a deeper architectural level. Astrocytes, oligodendrocyte precursor cells and endothelial cells declined with age, while microglia shifted toward a primed inflammatory state. Across cell types, the three-dimensional organization of chromatin eroded. [9] Gene expression is often treated as the headline in aging research. This study suggests that the regulatory folding of the genome itself may be part of what comes apart, changing which genes can effectively communicate with their control elements.

Early stress can leave a comparable regulatory residue. In mice, experimentally increasing an enhancer-associated chromatin mark in the juvenile nucleus accumbens opened regulatory regions, altered D2 neuron physiology and made later stress provoke stronger behavioral effects. The same manipulation in adulthood did not have this consequence. [10] Developmental timing, in other words, can convert a transient molecular perturbation into a long-lived sensitivity.

SCN2A-related disorders reveal why this developmental lens matters for treatment. Human forebrain organoids carrying loss-of-function SCN2A variants showed disrupted production of both excitatory and inhibitory cortical neurons, with prematurely generated inhibitory neurons linked to elevated Sonic hedgehog signaling. [11] Yet another study points toward a highly individualized clinical response for a different class of SCN2A mutations. Two boys with severe developmental epileptic encephalopathy received bespoke antisense oligonucleotides designed to reduce the mutant copy of the gene while sparing the healthy one. Seizures fell by 26% and 90%, respectively, alongside improvements in medication burden and developmental skills. [12]

These are two-person, open-label studies, not proof of a general cure. Only an estimated 16% of rapidly diagnosed infants in one cohort carried the necessary genetic configuration for this allele-selective strategy. Still, the work offers a template for precision neurology: not merely identifying a disease gene, but designing a therapy around the exact mutation, phase and molecular context.

Looking ahead

The week’s most consequential lesson may be that the brain is a collection of partially independent systems held together by regulation. Memories can be replayed in parallel circuits; moods can revise them after the fact; infants and flies use internal models to decide what matters; and genome architecture can quietly set the conditions for resilience or decline.

That complexity complicates intervention—but it also creates opportunities. Sleep rhythms, sensory cortex, deep circuits, chromatin regulators and mutant transcripts are all potential points of leverage. The task ahead is to learn which lever matters for which person, at which developmental moment, and without mistaking a promising signal for a finished treatment.