Introduction
Mnemonic Discrimination (MD) refers to the ability to distinguish novel stimuli from similar memories [1]. It is hypothesized to involve dentate gyrus (DG) pattern separation (PS) [1], which is impaired by the hyperactivity of DG granule cells (DGGCs). DGGC hyperactivity has been found in bipolar disorder [2]; such hyperactivity may subsequently impair MD. It is unclear whether DG PS is involved in both the encoding and retrieval phases of MD, or solely during encoding [3]. This distinction is important because it may clarify how changes in DGGC activity affects MD performance.
Methods
To address this gap, we developed a computational model of the hippocampus capable of executing the gold standard MD task for humans, the Mnemonic Similarity Task (MST) [1]. Our model simulates the ventral visual stream and entorhinal cortex via pre-trained ResNet-derived representations. It simulates DG pattern separation via k-winner-take-all dynamics, and a continuous log-sum-exp modern Hopfield network simulates the CA3’s autoassociative behaviour. Subsequently, our model’s MD performance was compared between retrieval conditions where the DG was active vs. inactive. Finally, a mediation analysis was conducted to examine if the relationship between DG excitability and MD is mediated via DG pattern separation.
Results
Our preliminary findings suggest that the DG is active during both encoding and retrieval, as these models exhibited better MD performance than those with the DG only active during encoding. Additionally, the mediation analysis indicated that MD performance is significantly partially mediated by DG PS. The proportion mediated ranged from 0.42 to 0.46. Total effects were also statistically significant, with coefficients between -0.80 and -0.87, indicating that, in our model, DG hyperexcitability impairs MD performance.
Discussion
We present a computational model of the hippocampus capable of simulating the MST. By demonstrating that PS partially mediates the relationship between DG excitability and MD performance, we therefore present a candidate mechanistic explanation for memory impairments seen in people with BD. A potential direction for future research is to explore why PS did not fully mediate MD performance. Alternatively, future studies can fit model parameters to behavioural data at the individual level, deepening our understanding of individual differences in hippocampal functioning.
References
1.\tStark, S. M., Kirwan, C. B., & Stark, C. E. L. (2019). Mnemonic Similarity Task: A Tool for Assessing Hippocampal Integrity. Trends in Cognitive Sciences, 23(11), 938–951.
2.\tBakker, A., Krauss, G. L., Albert, M. S., Speck, C. L., Jones, L. R., Stark, C. E., Yassa, M. A., Bassett, S. S., Shelton, A. L., & Gallagher, M. (2012). Reduction of hippocampal hyperactivity improves cognition in amnestic mild cognitive impairment. Neuron, 74(3), 467–474.
3.\tBernier, B. E., Lacagnina, A. F., Ayoub, A., Shue, F., Zemelman, B. V., Krasne, F. B., & Drew, M. R. (2017). Dentate Gyrus Contributes to Retrieval as well as Encoding: Evidence from Context Fear Conditioning, Recall, and Extinction. The Journal of Neuroscience, 37(26), 6359–6371.
Acknowledgement
