The Brain’s Talent for Changing Its Mind

This week

This week’s papers converge on an old question with unusually practical stakes: how does a brain revise its model of the world without losing what it already knows? The answer appears at every scale. A fly’s navigation circuit can rapidly turn a sensory measurement into a durable goal. Human neurons in the claustrum register uncertainty while we learn to avoid danger. Safety memories become more portable when extinction happens across varied settings. And at the clinical edge, optogenetics is restoring fragments of sight to people whose photoreceptors have died. The brain is not simply a storehouse of information. It is a system for deciding what to hold onto, what to update, and when to let go.

Learning what matters—and what is safe

The claustrum has long invited grand theories. It is a thin, widely connected structure, sometimes cast as a coordinator of consciousness, but with remarkably little direct evidence for what its neurons actually compute. Recordings from human claustrum neurons during aversive learning now offer a more grounded possibility: the structure tracks hidden features of a situation, including uncertainty and prediction errors—the mismatch between what was expected and what occurred. Its signals overlapped with those in anterior cingulate cortex, but their timing differed; during pauses between trials, only the cingulate carried uncertainty information. The claustrum may therefore be less a singular command centre than a participant in the moment-to-moment work of updating a model when action is required. [1]

That work has consequences for fear. In a three-day human conditioning study, the way brain activity reorganized after fear learning predicted how well an individual would later express extinction memory after transcranial magnetic stimulation. The result did not merely identify “more” or “less” connectivity: it identified a dynamic brain state, including how uncertain its transitions were, that forecast whose safety learning would be most responsive to stimulation. It is an early but intriguing step toward matching neuromodulation to a person’s learning dynamics rather than applying it uniformly. [2]

A behavioral route to sturdier safety memories may be simpler: vary the setting. In mice, extinction conducted across several distinct contexts produced fear suppression that generalized better to a new context than extinction in one setting. The effect depended on the dorsal hippocampus and its dopamine D2 receptors. A few weeks back we looked at disrupted safety updating in PTSD; this study suggests one mechanism by which exposure-based treatment might make safety learning less tied to the therapy room. It did not eliminate all forms of fear return, a useful reminder that “extinction” is new learning rather than deletion of the old threat memory. [3]

Memory is not a fixed thing

Navigation reveals how a nervous system can preserve a memory while remaining open to the present. In the fly’s navigation centre, a recurrent circuit maintains a persistent bump of activity representing a heading goal during odor-guided movement. Yet persistence would be a liability if it could not be overwritten. Slow recurrent excitation holds the memory in place; fast inhibition and a disinhibitory gate temporarily uncouple the circuit during turns, allowing it to follow fresh compass input. The resulting “split attractor” design explains how a circuit can write and erase a goal quickly without destabilizing every stored state. [4]

The sensory anchors for such maps are flexible. In mice, darkness disrupted entorhinal spatial codes—especially grid-like representations—in visually dominated environments. But in darkness with prominent tactile landmarks, spatial cells could form cue-anchored maps, until whisker trimming took away the relevant evidence. Internal maps are not detached from the senses; they are negotiated among the senses available in a particular world. [5]

Even radical metabolic interruption need not wipe the slate clean. Columbian ground squirrels that had learned a spatial maze outperformed naïve animals when they emerged from nine months of wild hibernation, their initial performance approaching their pre-hibernation level. Eight weeks ago, we considered how mouse memories could survive hibernation-like synapse loss. These squirrels extend that puzzle from induced synaptic change to prolonged natural metabolic shutdown: apparent neural remodeling during dormancy may be compatible with a resilient memory architecture rather than evidence of forgetting. [6]

That resilience has limits—and sometimes our technologies encourage the brain to conserve its effort. People told that an image would be externally available as a reminder showed neural patterns resembling deliberate forgetting; the item’s representation faded relative to information they expected to recall unaided. Cognitive offloading is useful precisely because it lets the brain allocate limited storage elsewhere, but its convenience is not memory-neutral. [7]

Seeing, sensing, and deciding

Perception begins before a stimulus arrives. High-resolution human imaging found modality-specific predictive activity in the superior colliculus before visual or tactile stimulation, in patterns resembling its eventual sensory responses. The midbrain is not merely reacting to incoming data; it appears to prepare differently for what it expects to encounter. [8]

And it does not decide alone. In mice performing tactile detection, the two somatosensory cortices shifted choice bias in opposite directions, while the superior colliculus proved necessary for those effects. Removing cortical input altered the colliculus’s decision-related signals more than it erased sensory representation itself. This shifts the familiar cortical hierarchy: a “lower” structure can be where sensory evidence is converted into a choice. [9]

A randomized study of 57 adults offers a striking reminder that what enters the sensory system is not just a stimulus. Nutritionally matched ultraprocessed meals produced larger insulin and energy responses, lower carbohydrate oxidation, and altered striatal responses to food cues compared with less processed meals. Participants did not report valuing the foods differently. But subjective preference and post-ingestive brain-body signals are not the same thing—and food processing may influence the latter beyond calories and macronutrients. [10]

Repairing brains, and revising our assumptions about aging

The week’s most concrete translational result comes from a small open-label trial in advanced retinitis pigmentosa. Ten participants received a viral vector that made retinal ganglion cells responsive to red-shifted light, used alongside stimulation goggles. Seven showed improved light sensitivity and six reached a predefined clinically meaningful threshold, though ocular adverse events were common and one serious injection-related event resolved quickly. This is not restored normal vision, nor yet definitive evidence of efficacy. But optogenetic therapy has moved decisively from a dazzling idea to a clinical strategy capable of restoring aspects of visual function. [11]

Neurodegeneration also looks less cell-autonomous than it once did. In human stem-cell-derived Parkinson’s models, microglia physically nibbled aggregate-containing compartments from dopamine neurons, clearing intraneuronal alpha-synuclein through a GPNMB-dependent process. Activated microglia can be destructive in established disease, but this work identifies an early protective mode—and cautions against treating immune activation as one uniform villain. [12] In Parkinson’s cohorts, meanwhile, greater ongoing physical activity was associated with slower motor and cognitive decline. The observational design cannot show that exercise caused the difference, but the link was strongest for total activity time rather than vigorous intensity alone: consistency may matter more than heroic exertion. [13]

Finally, brain aging may be unusually human. Across adult chimpanzees, bonobos, gorillas, and orangutans, brain weight did not significantly decline even into advanced ages comparable to periods when human brain loss becomes conspicuous. That does not mean other great apes escape every neural change of aging. It does sharpen the mystery: pronounced late-life brain shrinkage may not be an inevitable primate fate, but a vulnerability of our lineage—and perhaps a clue to why human dementia is so prevalent. [14]

Looking ahead

These studies resist a simple picture of the brain as either stable or plastic. Good cognition requires both: a fly circuit that locks onto a goal but can release it, a safety memory that travels beyond one room, an immune cell that clears damage without escalating it, and an aging brain that maintains its structure over decades. Even the medial prefrontal cortex may fit this lesson. People with focal damage there could still reason about others’ beliefs in a strategic game, though they took longer and required more evidence before choosing—suggesting the region shapes the decision process more than it singularly contains the capacity to understand minds. [15] The task now is to turn these distributed, dynamic accounts of brain function into treatments that are equally adaptive: targeted enough to help, humble enough to preserve what the brain already does well.