When an animal learns to navigate a new environment, certain brain cells begin responding at specific locations. These neurons have long been seen as key to how spatial memories form. Once a memory is established, it’s often assumed to be stable.
But a new study in Nature Neuroscience has challenged that view. Researchers tracked more than 2,500 neurons in mice learning to run a virtual track for a reward. They found that even the most stable memory signals were reconstructed daily through plasticity, a kind of neuronal rewiring.
“It was indeed surprising,” said the study’s first author Sachin Vaidya, associate professor at Baylor College of Medicine, Texas. “Memory may not be fixed by lasting synaptic strength, but instead may rely on a few stable synapses that re-trigger plasticity across the network each day.”
In other words, the mice weren’t just retrieving a memory: they were actively reconstructing it.
“It’s not relearning,” Vaidya said. “It’s a mechanism that balances stability and flexibility, preserving the memory’s core while letting the network adapt.”
This echoes what some scientists have long suspected.
“Even though the location may remain the same, time is always moving forward,” said behavioural neuroscientist Tracey Shors of Rutgers University. “Each time you reflect on an experience, time has advanced, and therefore this memory is, at least in part, also new.”
See how they run
To understand how memories persist over time, researchers trained mice to run on a virtual track. The mice stood on a small platform and viewed a moving visual scene to simulate forward motion. Reaching the end earned them a water reward. Over several days, the mice learned to anticipate the reward and ran more reliably.
As they ran, scientists monitored activity in hundreds of individual neurons in a hippocampal region called CA1, which helps animals track their location in space. Using two-photon calcium imaging, the team recorded the same set of cells across multiple days.
They focused on a type of signal called a place field, where a cell becomes active at a specific location. These “GPS tags” in the brain are considered key indicators of spatial memory. In the study, the team identified place fields that persisted over many days — the kind typically taken as signs of memory stability.
Tracking the same cells over time, they asked: did these spatial signals remain stable or shift with time?
Stability that isn’t static
As training progressed, more neurons formed place fields, firing reliably at specific locations on the virtual track. These patterns suggested that a spatial memory had formed.
Some place fields appeared briefly and then disappeared while others stayed active in consistent locations. These longer-lasting patterns were typically interpreted as evidence of a stable memory trace. The team focused on these more persistent patterns.
They looked for signs of a plasticity mechanism called behavioural timescale synaptic plasticity. It’s a bit like stamping a memory into place: when a neuron receives a strong input, it fires a burst of activity that leaves a lasting mark. This appeared as a sudden jump in activity at a new location, followed by continued firing there.
Strikingly, even neurons with stable place fields showed new behavioural timescale synaptic plasticity signatures at the same spot the next day. Stability, in other words, wasn’t static: it was rebuilt through new plasticity events.
These reactivations often occurred at the same location and were more likely in cells active the day before. Over time, cells with prior activity became increasingly likely to reactivate, suggesting that a memory’s stability emerged from repeated recruitment.
But what marked certain synapses as the ‘stable’ ones?
Sourav Banerjee, a neuroscientist and professor at the National Brain Research Centre in Manesar, pointed to long non-coding RNAs (lncRNAs), molecules that don’t make proteins but help regulate gene activity at specific sites.
“Our lab found a lncRNA at CA1 synapses that appears to do exactly that,” Banerjee explained. “When we knocked it out using synapse-targeted CRISPR [the gene-editing tool], those synapses lost activity and the animal showed clear memory deficits.”
In the mice, reactivation occurred even though the memory remained the same. The finding suggested that some brain cells may need to be re-engaged repeatedly to keep the memory active.
Each time we recall a memory, the brain may be reconstructing it anew.
“Our working idea is that stable synapses make reactivation of place cells more likely,” Vaidya said. “But plasticity is probabilistic. A cell might go silent for a time and still reappear later. This may be how long-term memory endures despite temporary lapses in activity.”
To test whether this kind of probabilistic reactivation — i.e. how likely a neuron is to fire again — could explain what they were seeing, the researchers turned to modelling.
Memory as probability, not permanence
The researchers simulated three models. One assumed that a neuron stayed active forever once it became active, like an engraved trace. Another treated neuron activity as random, switching on or off daily with no memory of past states. A third, called the cascade model, made reactivation more likely each time it occurred, letting stability build gradually over time.
Only the cascade model matched the real brain data, according to the researchers capturing both the rise of stable place cells and their consistent firing across days. This echoed a concept called metaplasticity, where a neuron’s past activity makes it more likely to change again.
“I have always wondered whether stable forms of plasticity, like lasting synaptic strengthening, could account for the dynamic nature of memories,” said Shors, of Rutgers University. “A more dynamic form, this metaplasticity, is seemingly necessary.”
That is, memory may not be fixed or random but be shaped by experience and sustained by practice.
When reactivation doesn’t occur
The mouse study did raise another question: what happens when reactivation fails or when the brain actively dismantles a memory?
Another new study, this one published in Current Biology, turned to fruit flies to explore exactly that, revealing how a short-term memory trace could shift and then disappear if it wasn’t maintained.
In the fruit flies, researchers from Tsinghua University in Beijing examined an associative memory formed after a sugar-reward task. Immediately after training, a signal appeared at a synapse between neurons that process reward, but faded within an hour. Meanwhile, a second signal emerged in a different brain region involving a new set of connections.
The researchers called this a trace shift because the memory moved from one site in the body to another. In the latter, neurons began forming fresh active zones, the structures where neurotransmitters are released. But these new zones didn’t last. It was as if the brain had opened a second memory warehouse and marked it for demolition. Molecules like Rac1 and Ephrin acted as foremen, instructing the removal of the new active zones and dismantling the second trace. When these molecular demolition crews were blocked, the memory lingered for far longer.
Do similar forgetting mechanisms exist in mammals? Banerjee said yes. In one study that he was part of, researchers found that deleting a specific lncRNA in the infralimbic cortex disrupted the extinction of a fear even after repeated exposure.
His team also uncovered a metabolic link: an lncRNA that regulated ATP production at hippocampal synapses.
“Disrupt sleep, lose energy, and the trace collapses,” Banerjee said. “It shows how molecular and metabolic factors can directly influence whether a memory fades or persists.”
Rather than passive decay, the fly study described forgetting as an orchestrated, multi-step process triggered once a memory moves to a site marked for removal.
Banerjee’s findings suggested mammalian parallels: that forgetting, too, is an active and regulated process, albeit more nuanced and complex.
“Rac and Ephrin do affect the shape of synapses, and that can make them less stable,” Banerjee said. “But I don’t think these molecules alone explain the kind of memory shift we saw in flies. That level of change likely depends more on how groups of neurons behave together than just on what’s happening at individual connections. We’re not quite there yet in linking those big-picture patterns to molecular details — but that’s where we need to go.”
When memory persists, or not
We often think of memories as static: something laid down and ready to be recalled. But what happens after a memory forms may be just as important as how it forms.
“There’s a strong analogy to spaced versus massed learning,” Banerjee said. “Think of two students: one studies steadily for months, the other crams the night before. The first usually does better because repeated exposure helps lock in the memory. In the brain, we do something similar. Spaced learning re-engages the same synapses repeatedly.”
A widely supported theory for how this works is called circuit remodelling.
“The idea is that repeated activation doesn’t just strengthen a connection, it triggers molecular changes that help stabilise it,” Banerjee said. “New proteins are made at active synapses, which signal back to the nucleus and activate protein synthesis. Later, other synapses can ‘capture’ those proteins and grow stronger. This two-stage process is how fleeting activity becomes a durable memory.”
In mice, memory traces returned to the same location across many days of training — but only by being actively rebuilt. In flies, short-term memories shifted and were dismantled unless preserved. Together, the studies show that memory is dynamic: rebuilt when needed, let go when it’s not.
This kind of flexibility hints at a deeper function.
“The purpose of memory is not to reminisce about the past,” said Shors. “We use memories to learn what we should do now and in the future.”
In this view, memory is less a record than a rehearsal that tunes the mind to act.
Despite differences in species, brain regions, and memory types, both studies point to a provocative idea: memory isn’t a static imprint the brain stores by default. It’s a living process: rebuilt, reinforced, and actively protected against decay, from the molecular level to the scale of entire neural circuits.
Anirban Mukhopadhyay is a geneticist by training and science communicator from Delhi.
