Synaptic engram underlies memory

Can We Target the Specific Connections That Hold Memories?

Researchers have discovered that specific connections between memory-encoding cells in different brain regions strengthen during learning. By using light to selectively weaken only these specific "engram" connections, they were able to erase a memory without affecting other nearby brain cells. This precision offers a new way to view how the brain stores and retrieves our experiences.

The search for the physical seat of memory

When we learn something new, our brains undergo physical changes. Scientists believe these changes occur within "engram cells"—small, sparse populations of neurons that represent a specific memory. While these cells are necessary for memory, a fundamental question remains: where exactly is the memory stored? Is it held within the internal electrical properties of the individual cells, or is it hidden in the strength of the connections between them?

Specifically, the authors investigate whether "engram synapses"—the connections between a sending engram cell and a receiving engram cell—drive memory. Because these cells are so sparsely distributed, these specific synapses represent a tiny fraction of total brain connections. The researchers sought to determine if manipulating this minuscule population would disrupt memory. They also tested if the brain relies on broader changes, such as shifting the overall excitability (the ease with which a cell fires) of the entire cell population.

Cracks in the single-cell model

For years, research has focused heavily on the activity of engram cells. The prevailing logic suggested that memory might be sustained by making these cells easier to trigger. This is known as intrinsic plasticity (changes in a cell's own excitability). Another possibility was homeostatic plasticity (adjustments to non-engram synapses to maintain balance).

However, observing these connections is difficult. Because engram cells are spatially intermingled with non-engram cells, traditional stimulation hits everything at once. It is like trying to turn down the volume on one singer in a crowded stadium. Usually, you just end up muffling the entire crowd. Previous studies showed that strengthening these interregional connections correlates with learning .

But correlation is not causation. The field lacked a way to selectively "unlearn" a specific connection without affecting the surrounding neural architecture.

Implementing optical spike-timing-dependent plasticity

To solve this, the authors developed a tool called optically-implemented spike-timing-dependent plasticity (oSTDP). This method uses a biological principle called STDP. In STDP, the precise millisecond-level timing between a sending neuron's fire and a receiving neuron's fire dictates the connection's strength. If the receiving cell fires just before the sending cell, the connection weakens. This is called synaptic depression.

The researchers controlled this timing using light. They used two different "opsins" (light-sensitive proteins that act as biological switches) with distinct colors. They used ChR2 to stimulate the axons of entorhinal cortex (EC) cells. They used ChrimsonR to target the dentate gyrus (DG) engram cells .

By alternating 473 nm and 593 nm light pulses with millisecond precision, they could instruct the synapses to weaken.

The team validated this in vitro (in a controlled slice of brain tissue) before moving in vivo (inside a living animal). They used Neuropixels 2.0 probes—high-density recording arrays—to track individual neurons in the DG. They delivered light pulses simultaneously .

This confirmed that the oSTDP protocol specifically reduced the probability of a DG engram cell firing in response to EC engram inputs. Meanwhile, neighboring non-engram cells remained untouched .

Decoupling connectivity from memory

The most striking result is the direct link between these specific synapses and behavior. After applying the oSTDP protocol to mice that underwent fear conditioning, researchers observed a significant drop in "freezing" behavior. Freezing is a standard measure of fear memory in rodents .

The authors report that this memory impairment was highly specific. In the group receiving targeted oSTDP-LTD (long-term depression) stimulation, the mice showed significantly decreased freezing levels during their second exposure to the fear context . Crucially, this effect was absent in control groups. These groups received "pseudo" stimulation, which used light pulses at timings that did not trigger the plasticity rule .

Furthermore, the study shows this wasn't a side effect of making cells less excitable. The researchers found that the reduction in memory was driven by the removal of AMPA receptors (proteins that facilitate fast synaptic transmission) from the engram synapses. This occurred without changing the cells' intrinsic ability to fire . This proves the memory was lost because the "bridge" between the two brain regions was dismantled.

Implications for the architecture of thought

This work shifts the focus of memory research from the cell to the connection. If these findings generalize, it implies that many episodic memories are maintained by a web of interregional synaptic strengths. This happens rather than relying on the mere presence of active neurons.

There are two major implications here. First, it suggests the brain achieves high storage capacity by using specific, sparse "wiring diagrams" between regions. Second, it provides a powerful new toolkit. Instead of broadly suppressing brain activity, researchers can now theoretically target the specific circuits responsible for particular memories.

The paper does not explore if this specific STDP-like mechanism is the primary way the brain learns in natural settings. It also does not specify exactly what information is carried across these EC-DG synapses. A logical next step would be to use this oSTDP tool to attempt the selective erasure of different types of memories. Researchers could see if the same synaptic rules apply across different cognitive domains.