Summary
A bioRxiv preprint combining mathematical modelling with intracellular recordings in behaving mice reports that CA3 pyramidal neurons have an activation exponent slightly above 1, as predicted for stable memory recall. The analysis included 49 new recordings and 133 cells from two earlier studies.
A mathematical analysis and recordings from behaving mice point to a specific cellular property that may help the hippocampus retrieve memories reliably. In a bioRxiv preprint posted on September 20, 2026, researchers report that pyramidal neurons in hippocampal area CA3 have an activation exponent slightly above 1 on average, matching a prediction from a model of stable memory recall.
The study combined a theoretical analysis of a recurrent neural-network model with in vivo intracellular recordings. The researchers recorded 49 CA3 pyramidal neurons and compared the resulting voltage traces with a range of mathematical models. They also analysed 133 additional cells from two earlier studies, obtaining activation-exponent estimates highly consistent with the new recordings.
From a recurrent network model to a neuron
Area CA3 is widely believed to contribute to memory by retrieving distributed patterns of neural activity stored in recurrent connections between neurons. In this kind of circuit, activity can spread through feedback connections and recreate a previously stored pattern.
The researchers examined how the intrinsic properties of individual neurons influence that process. A neuron's activation function describes how its output changes as its input changes. In the model studied here, the activation exponent determines the degree of nonlinearity in that relationship. An exponent slightly above 1 means that output increases somewhat more than proportionally as input rises within the model's operating range.
The mathematical analysis produced three experimentally testable predictions for recurrent circuits capable of stable memory recall. CA3 pyramidal cells were predicted to have elevated intrinsic excitability, while recurrent connections were predicted to be dominated by inhibition. The model also predicted that neuronal activation functions would have an exponent slightly above 1.
The first two predictions were consistent with previous experimental findings, according to the authors. The new work focused on testing the prediction about the activation function.
Recordings matched the predicted exponent
The researchers performed intracellular recordings from CA3 pyramidal neurons in behaving mice. Intracellular recording measures changes in a neuron's electrical voltage, allowing the relationship between incoming activity and the cell's response to be estimated more directly than by recording spikes alone.
After fitting several parametric models to the voltage traces, the researchers used statistical model comparison to estimate the activation exponent. Across the 49 newly recorded neurons, the average exponent was slightly above 1, in line with the theoretical prediction. The cells showed substantial variation, however, and that heterogeneity is not yet accounted for by the theory.
The analysis of 133 cells drawn from two other studies produced estimates that were highly consistent with the new data. This cross-study agreement supports the idea that the activation pattern is a recurring feature of CA3 pyramidal neurons rather than an isolated result from one recording set.
The evidence is a combination of mathematical modelling and mouse systems-neuroscience data. It connects a property measured at the level of individual neurons with a proposed circuit mechanism for recalling stored activity patterns. The authors suggest that CA3 may be tuned to support reliable memory recall through the combined properties of its neurons and recurrent connections.
The work is a bioRxiv preprint, and its physiological evidence comes from behaving mice. A central open issue within the study is how the substantial cell-to-cell variation in activation exponents should be incorporated into the model of memory recall.