Rejuvenating neurons found to restore learning and memory

Age can make memory feel like something that only moves in one direction. A name slips away. A route you know well turns fuzzy. In Alzheimer’s disease, that slide can look even steeper. Yet the brain does not stop adapting. Neurons keep tuning the strength of their connections, and that constant adjustment, called synaptic plasticity, is what lets learning stick.

The problem is that aging and Alzheimer’s disrupt the cellular support systems that plasticity depends on. When those systems wobble, the brain’s “memory trace” can wobble with them.

A new mouse study asks a very pointed question: what if you do not try to “fix the whole brain,” but instead try to revive just the small set of neurons that carry a specific memory?

A team led by Johannes Gräff at EPFL’s Brain Mind Institute reports in Neuron that a short, tightly controlled pulse of three genes, Oct4, Sox2, and Klf4, delivered only to learning-activated “engram” neurons, can restore memory performance after decline has already begun. The trio is referred to as OSK.

Memories are thought to be stored in sparse groups of neurons called engrams. These are the cells that switch on during learning and can later switch on again during recall. In physiological aging and in mouse models of Alzheimer’s disease, engrams can malfunction, and recall suffers.

The EPFL team’s twist was to aim their intervention at those specific memory-related cells instead of treating brain tissue broadly. They used “partial reprogramming,” an approach that has been explored as a way to reset certain age-related cellular features without wiping out a cell’s identity.

The general idea is familiar in regenerative medicine. Cyclic expression of the four Yamanaka factors, Oct4, Sox2, Klf4, and cMyc, or a partial combination such as OSK, has been linked to rejuvenation effects while avoiding de-differentiation and tumorigenesis when controlled carefully. The open question here was narrower and more clinical in spirit: can memory be rescued after decline has started?

In mice, the researchers used adeno-associated viruses delivered by stereotaxic brain injections. Two systems worked together.

One system labeled neurons activated during learning using a c-Fos-based tagging approach (c-Fos:ttA + TRE::GFP). The other system placed OSK under the same tetracycline-responsive control (+TRE::OSK). With this tet-off design, doxycycline (Dox) withdrawal opened a time window: learning-related neurons could be tagged, and OSK could be turned on briefly, then shut back down when Dox returned.

They targeted two memory-linked regions with different jobs.

The dentate gyrus (DG) of the hippocampus is tied to learning and recent recall. The medial prefrontal cortex (mPFC) becomes increasingly important for remote memory expression, including recall two weeks later.

The first behavioral test was contextual fear conditioning, an associative memory task. Aged mice in this paper were defined as 9 to 10 months old, with young controls at 2 to 3 months.

Aged control mice injected with GFP showed the expected drop in freezing during recall compared with young mice, a sign of weakened fear memory. The authors also saw reduced engram reactivation in the DG, consistent with earlier work.

When OSK was induced specifically in DG learning-activated ensembles, memory in aged mice was “rescued to that of young mice.” Reprogrammed Klf4+ engrams were preferentially reactivated compared with non-reprogrammed engrams, pointing to a cell-specific effect rather than a vague boost in arousal or movement.

The team then shifted to the mPFC to ask about remote memory. Using the same tagging and brief OSK induction during learning, they tested recall two weeks later. Aged GFP-injected mice showed reduced freezing at that remote time point. Aged OSK-injected mice did not show that impairment.

Any time you talk about reprogramming factors in the brain, the obvious worry is identity drift. You do not want a mature neuron to start becoming something else.

The study addresses that concern by looking at markers tied to neuronal identity and aging-related nuclear architecture.

In DG granule neurons, the team measured Prox1, a determinant of cell identity. Rather than dropping, Prox1 expression increased with partial reprogramming, largely because OSK reduced the proportion of low-expressing Prox1+ cells in the Klf4+ population. The authors interpret this as strengthening identity rather than eroding it.

They also examined aging-linked markers related to heterochromatin and nuclear structure, including H3K9me3, LaminB1, and nuclear circularity. In reprogrammed engram cells, H3K9me3 and LaminB1 increased, and nuclei became more circular, moving away from aged-like features.

In the mPFC, OSK-expressing viruses mainly infected deeper cortical layers, so the team looked at Ctip2, a marker of deeper layer neurons. Ctip2 expression rose with reprogramming. H3K9me3 and LaminB1 increased there as well, and low-circularity neurons became less common.

The authors then asked whether the same targeted approach could help in Alzheimer’s disease models, where memory problems are tied to pathology rather than physiological aging alone.

They used APP/PS1 mice, which express human mutations in App and Psen1 and show early plaque deposition and cognitive impairments from middle age onward. Fear memory was not impaired in this strain in their hands, so they used a water maze navigational task, which can reveal how mice learn, not just whether they eventually find a platform.

When dentate gyrus engrams were tagged and partially reprogrammed during a five-day water maze training protocol, GFP-injected APP/PS1 mice relied less on hippocampal strategies and took longer paths than GFP-injected wild-type controls. OSK-injected APP/PS1 mice showed a higher contribution of hippocampal strategies and no longer displayed the same navigation impairment during learning.

To test memory retention, they targeted the mPFC in APP/PS1 mice using a shorter, three-day learning paradigm and labeled putative engrams on the last training day. GFP-injected APP/PS1 mice lacked spatial memory in probe tests at both recent and remote times. OSK-injected APP/PS1 mice still showed impaired learning and recent memory, but two weeks later they showed a significant preference for the target quadrant comparable to wild-type mice. That pattern suggests a specific recovery of remote spatial memory.

A key control matters here. When OSK was targeted to cells activated during exploration of a novel environment before water maze training, it did not restore spatial memory. The benefit depended on hitting the memory-relevant ensembles.

To connect behavior to mechanism, the team performed single-nucleus multiome sequencing (snRNA-seq plus snATAC-seq) on mPFC nuclei from GFP-injected wild-type mice, GFP-injected APP/PS1 mice, and OSK-injected APP/PS1 mice. Samples were collected one day after the last recall session in the water maze paradigm.

They mapped reads to viral sequences to identify engram nuclei, finding them mostly among excitatory neurons. In those excitatory engrams, APP/PS1 mice showed decreased expression of identity genes. The authors report that this was fully rescued with OSK.

They also scored cells using gene signatures related to engram formation and AD-dependent shifts. APP/PS1 engrams had a reduced engram score, which OSK partially restored. From a gene ontology perspective, the AD engram signature was enriched for synaptic and neuronal function terms and was largely reversed in OSK engrams. Gene set enrichment analysis pointed to reversals in pathways including cell-death, immune, translation, and synaptic categories.

At the level of individual genes, the authors report that many differentially expressed genes in APP/PS1 were shifted back toward wild-type levels with OSK. Among downregulated genes rescued by OSK were categories tied to synapse organization and potassium ion transport, including several voltage-gated ion channels and neuregulin 1 (Nrg1). Among upregulated genes that were corrected were categories such as amyloid beta response and negative regulation of long-term potentiation, including App and prion protein (Prnp).

Chromatin accessibility changes were more nuanced. A significant proportion of differentially accessible regions in APP/PS1 engrams were reversed by OSK, yet promoter accessibility did not track gene expression changes in a simple way after OSK. The authors suggest a decoupling between chromatin openness at transcription start sites and transcriptional output, with one possibility being recruitment of repressors to open chromatin. They highlight Pou5f1b, a homolog of Oct4, as a top candidate from enrichment analysis of downregulated genes.

They also report that many OSK-induced accessibility gains occurred in distal regions rather than promoters, and that distal peaks gained after OSK were enriched for Klf4 motifs. One example they point to is Kcnj3, which encodes the potassium inwardly rectifying channel Kir3.1.

That gene connects neatly to their electrophysiology.

In ex vivo patch-clamp recordings of mPFC pyramidal engram neurons, APP/PS1 engram cells were hyperexcitable. They fired more action potentials for the same current injections and had reduced rheobase compared with wild-type engrams. OSK treatment normalized firing frequency and rheobase toward wild-type levels. Within the same OSK-injected mice, Klf4+ engram cells showed the rescue more strongly than Klf4− engram cells.

They then applied ML-297, a Kir3.1 activator. It reliably hyperpolarized wild-type engrams but failed to do so in APP/PS1 engram cells, consistent with Kir3.1 disruption. In OSK-reprogrammed engrams, ML-297 again produced hyperpolarization, supporting the idea that restored Kcnj3 expression contributed to normalized excitability.

The team also tried to quantify whether OSK restored young-like learning patterns, not just a few test scores. Using water maze metrics from GFP-injected wild-type mice across ages 16 to 90 weeks (N = 39), they built a regression model that predicted chronological age from behavior, calling it a “cognitive clock.” The model had a correlation of R = 0.57 (p < 0.001) and a median prediction error of 10 weeks.

When they applied the model to OSK-injected wild-type mice at 40 or 90 weeks, predicted age dropped relative to chronological age, particularly in advanced age. In Alzheimer’s models, predicted age ran higher than chronological age in GFP-injected APP/PS1 and 5xFAD mice, which the authors interpret as accelerated cognitive aging. Predicted age returned closer to chronological levels in OSK-injected APP/PS1 and 5xFAD mice.

They also note that non-spatial parameters such as swimming speed were not changed in OSK-injected mice, arguing against a simple fitness or motor explanation.

The paper spells out several limitations.

It used amyloid-based AD models, not tau models, which can differ in progression and pathophysiology. The multi-omic and electrophysiological work focused on a single time point, two weeks after OSK induction, and centered on AD-dependent changes. The authors also note that their engram-targeting approach may have included a small fraction of non-reprogrammed cells in sequencing, which could dilute effects.

They call for future work on earlier molecular changes after OSK induction, inclusion of young animals for mechanistic clarity, and broader cell-type coverage beyond mPFC excitatory engrams. They also point out that while the study rescued remote memories out to two weeks, how long benefits persist beyond that remains unknown.

Research findings are available online in the journal Neuron.

The original story "Rejuvenating neurons found to restore learning and memory" is published in The Brighter Side of News.

Like these kind of feel good stories? Get The Brighter Side of News' newsletter.