Blocking a Single Receptor Subunit Restores Memory in an Alzheimer’s Model

Memory loss in Alzheimer’s disease has long been viewed as an irreversible slide, but a new computational study suggests that a targeted molecular intervention could restore memory function at the level of entire neural networks—not just individual cells. Researchers from Italy, Finland, and Lithuania built a biophysically detailed model of the hippocampal CA1–CA3 network and demonstrated that the memory impairments caused by amyloid-β accumulation, a hallmark of early Alzheimer’s disease, can be substantially reversed by partially blocking a specific subunit of the N-methyl-D-aspartate receptor (NMDAR) called GluN2B. The findings, published in Frontiers in Computational Neuroscience, represent the first time this rescue mechanism has been demonstrated at a network level, bridging the gap between cellular synaptic dysfunction and system-level cognitive impairment.

The hippocampus is the brain’s primary memory-encoding structure, and its CA1 and CA3 regions are critical for both storing new information and retrieving stored patterns. Long-term potentiation (LTP), the strengthening of synaptic connections that underlies memory formation, depends heavily on NMDA receptors—ion channels that open in response to glutamate binding and postsynaptic depolarization, allowing calcium to enter the cell and trigger the molecular cascades that consolidate synaptic change. In early Alzheimer’s disease, pathological levels of amyloid-β peptide (Aβ) bind to NMDA receptors and drive excessive activation, ultimately disrupting the delicate balance between potentiation and depression that memory storage requires.

Previous experimental work had shown that pharmacologically blocking GluN2B-containing NMDA receptors with the antagonist ifenprodil could rescue impaired LTP in CA1 pyramidal neurons exposed to Aβ. Computational models at the single-neuron level had further elucidated the biophysical mechanisms underlying this rescue. But whether these cellular-level mechanisms remain effective when synaptic impairments are distributed across a population of neurons—and how they ultimately affect memory storage and recall at a network level—could not be readily addressed experimentally. The new study by Fabio Librizzi, Saana Seppälä, Marja-Leena Linne, Michele Migliore, and Ausra Saudargiene set out to answer exactly that question.

The researchers built their model on a well-established CA1 network architecture comprising 100 hippocampal CA1 pyramidal neurons with four types of inhibitory interneurons: oriens-lacunosum moleculare, basket, bistratified, and axo-axonic cells. Each CA1 neuron was modeled with 15 compartments reflecting the soma, axon, and dendritic regions. The network receives excitatory inputs from the entorhinal cortex (EC) and CA3 Schaffer collaterals, activated by gamma bursts (30–80 Hz) coupled with a theta rhythm (4–8 Hz), mirroring the oscillatory patterns that govern memory encoding in the living brain. Each CA3 input creates two synaptic connections on the proximal dendrites of each CA1 neuron—one AMPA-mediated and one NMDA-mediated—while EC inputs arrive 9 milliseconds earlier and form AMPA-mediated connections on distal dendrites.

To simulate early Alzheimer’s pathology, the team modeled the effects of accumulated Aβ by reducing synaptic density to 80 percent of its original value, increasing free glutamate concentration to 120 percent, and modifying the shape of the GluN2B-dependent LTP function so that potentiation was abolished even when the subunit was activated. These adjustments, grounded in prior electrophysiological observations, produced a strong inhibition of LTP and a complete lack of synaptic weight increase during the storage protocol. The researchers then tested whether partially blocking GluN2B function could restore the impaired plasticity.

The results were striking. Reducing GluN2B conductivity to 40 percent of its original value produced a recovery of roughly 90 percent of the original synaptic weight increase during the 25-second storage protocol. The recovery fraction depended strongly on the degree of GluN2B attenuation, with 0.4 microsiemens providing the maximum rescue for the network. Importantly, control simulations in which AMPA receptor conductance was reduced across a range of values produced no recovery whatsoever, confirming that the rescue is specific to the GluN2B-dependent mechanism rather than reflecting a nonspecific reduction of excitatory drive. This specificity aligns with the known role of GluN2B–NMDAR-dependent calcium/calmodulin kinase II (CaMKII) signaling in Aβ-induced LTP disruption.

The team then evaluated network performance in storing and retrieving multiple patterns, both orthogonal (non-overlapping) and non-orthogonal (overlapping). With 100 CA1 cells and 20-input patterns, a maximum of five orthogonal patterns could be presented. Recall quality was nearly perfect for orthogonal patterns, as expected, since they activate entirely separate sets of synapses. For non-orthogonal patterns, quality decreased even with only five stored patterns, mainly due to spurious spiking of cells not involved in the recalled pattern. When 50 non-orthogonal patterns were stored, recall quality approached the threshold corresponding to the spiking of all 100 CA1 neurons—effectively rendering the output uninformative about which memory was retrieved.

When the researchers introduced different fractions of AD-affected synapses—0, 20, 50, 80, and 100 percent—they found that recall quality decreased significantly as the fraction of impaired cells increased. However, reducing GluN2B activity to 40 percent of its original value strongly counteracted this effect, to the point that in rescue conditions the quality factor almost approached its maximum value even when 100 percent of synapses were AD-affected. For non-orthogonal patterns, the partial GluN2B inhibition appeared to reduce spurious spiking of cells outside the recalled pattern, improving recall specificity, although it simultaneously reduced the average number of spikes, weakening quality in some cases. The interplay between these two effects produced a nuanced network-level outcome that could not have been predicted from single-synapse mechanisms alone.

The authors emphasize that the degree of recovery depends on population-level variables such as the fraction of affected CA1 neurons and the structure of the stored patterns. The model’s robust behavior across different levels of synaptic impairment and pattern separation supports the presented mechanisms for synaptic recovery in Alzheimer’s-like conditions. The results indicate that accumulated Aβ disrupts GluN2B function in hippocampal microcircuits, impairing memory storage and recall especially by affecting theta-gamma coupling, which can be partly restored by suppressing GluN2B conductivity.

While no specific GluN2B antagonist has yet been approved as a therapeutic agent for Alzheimer’s disease, unspecific NMDA receptor antagonists such as memantine have received FDA approval for their modest symptomatic effects. The new findings provide a proof-of-concept that selective GluN2B blockade could counteract early AD impairments not only at the single-neuron level but also at the network level in memory storage and recall. The authors stress that the 40 percent reduction identified here should not be considered a universal optimum, but rather that an intermediate level of GluN2B activity exists that optimally restores LTP under the conditions studied. Future work incorporating astrocytic dynamics and other molecular mechanisms into large-scale network models could further elucidate how early synaptic dysfunction propagates to network-level deficits and how targeted interventions can restore function in Alzheimer’s disease.

Subject of Research: Network-level restoration of impaired memory function in early Alzheimer’s disease via GluN2B modulation: insights from a hippocampal CA1–CA3 model

Article Title: Network-level restoration of impaired memory function in early Alzheimer’s disease via GluN2B modulation: insights from a hippocampal CA1–CA3 model

Article References: Librizzi, F., Seppälä, S., Linne, M.-L., Migliore, M., & Saudargiene, A. (2026). Network-level restoration of impaired memory function in early Alzheimer’s disease via GluN2B modulation: insights from a hippocampal CA1–CA3 model. Frontiers in Computational Neuroscience, 20, Article 1953352. https://doi.org/10.3389/fncom.2026.1953352

Image Credits: AI Generated

DOI: 10.3389/fncom.2026.1953352

Keywords: Network-level, restoration, impaired, memory, function, early, Alzheimer, disease, GluN2B, modulation, insights, hippocampal