For decades, neuroscientists treated astrocytes as the brain’s custodial staff. These star-shaped cells mop up chemical spills at synapses, ferry nutrients to neurons, and help maintain the blood-brain barrier. Important work, but not the kind that earns a role in the story of how you remember your first kiss or where you parked your car. A series of experiments published between late 2023 and early 2025, now drawing wider attention as of June 2026, has upended that assumption. Astrocytes do not just support memory. They carry molecular tags tied to specific experiences, and when those tags are disrupted, recall falls apart, even when every neuron in the circuit is firing normally.
Catching astrocytes in the act
The clearest evidence comes from two studies published in Nature that used a gene called c-Fos as a kind of cellular timestamp. c-Fos switches on when a cell becomes active, so by engineering mice whose astrocytes would glow under a fluorescent reporter only when c-Fos fired during a learning task, researchers could pinpoint exactly which glial cells participated in forming a memory.
One team found that these learning-associated astrocyte ensembles clustered near the neuronal engram cells already known to store memories in the hippocampus. Silencing those astrocytes degraded recall. “We were stunned that removing astrocytes from the equation was enough to break the memory, even though the neurons were still there,” one of the study’s lead authors told reporters at the time of publication. A second group discovered that emotionally significant experiences prime astrocytes through noradrenaline, the brain’s alertness signal, creating a trace that persists for days. When the animal later retrieves the memory, those primed astrocytes fire again, this time releasing a protein called IGFBP2 that appears to stabilize the memory after recall.
A third Nature study attacked the question from a different angle entirely. Using spatially resolved transcriptomics, a technique that maps gene activity across thin slices of brain tissue, researchers found that memory-linked gene-expression signatures persisted for weeks in both neurons and neighboring astrocytes. That durability is significant. It means astrocytes are not just helping encode a memory and then resetting to baseline. They carry a lasting molecular record of the experience, one that outlasts the initial learning session by a wide margin. And because this finding came from a method that does not depend on c-Fos at all, it reduces the chance that the tagging results are an artifact of a single labeling technique.
Shaping circuits, not just synapses
The story extends beyond local tagging. Experiments published in Cell Reports showed that manipulating astrocytic pathways during memory acquisition changed the strength of both recent and remote memories by altering how the hippocampal CA1 region communicates with the anterior cingulate cortex through a specific long-range projection. In a separate line of work published in Nature Neuroscience, chemogenetic suppression of CA1 astrocytes during learning selectively impaired recall weeks later without touching short-term memory.
Taken together, these results suggest that astrocytes do not just stamp a local molecular tag on a memory. They help shape the long-distance brain circuits that determine whether that memory survives over time or fades.
What the data cannot tell us yet
Every tagging and manipulation experiment so far has been conducted in mouse hippocampi. As of mid-2026, no study has recorded or sampled human astrocyte ensembles to confirm the same mechanism operates in people. The standard behavioral task in most of these papers is contextual fear conditioning, a protocol in which mice learn to associate a chamber with a mild foot shock. That tests a narrow slice of memory. Whether astrocyte ensembles behave differently for neutral versus emotionally charged experiences, or for declarative versus procedural learning, remains untested.
The two-step model — noradrenaline primes astrocytes during learning, IGFBP2 stabilizes the trace during recall — is compelling but built on inference. No group has published real-time recordings of noradrenaline release in awake, behaving animals during both the priming and recall phases. The causal chain rests on pharmacological and genetic manipulations rather than direct observation. Independent replication by labs outside the original research teams has not yet appeared in the literature.
Longitudinal tracking in the available studies extends only a few weeks. Whether tagged astrocytes maintain their molecular signatures over months, or whether those signatures degrade in ways that mirror natural forgetting, is an open question. The field also lacks data on how astrocyte ensembles interact with sleep-dependent memory consolidation, a process known to reshape neuronal engrams overnight.
Why this changes the search for memory-related therapies
If astrocyte dysfunction contributes to recall problems in aging, traumatic brain injury, or neurodegenerative disease, the implications for drug development are substantial. Nearly all current memory-related therapies target synaptic receptors on neurons. Astrocytes open a parallel set of targets that the field has barely explored.
IGFBP2 is a secreted protein, meaning it could in principle be measured in cerebrospinal fluid. That raises the speculative but testable possibility of a biomarker for memory stabilization, something clinicians currently lack. Alzheimer’s research, which has increasingly focused on the role of neuroinflammation and glial dysfunction alongside amyloid and tau pathology, may find particular relevance in these findings. Astrocytes are already known to become reactive in Alzheimer’s brains; the question now is whether that reactivity disrupts the tagging process these studies describe.
None of this is proven in humans yet. But the direction of the evidence is clear, and it carries a simple message: any serious account of how the brain remembers now has to include the cells that textbooks once dismissed as glue.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
