Extended Data Fig. 6: Rate based circuit model of direct hippocampal feedback modulation of cortical output in ECL2/3.
From: Hippocampus shapes entorhinal cortical output through a direct feedback circuit
a. Schematic of a rate-based model: Two multimodal inputs (I1, I2) project onto an ECL2/3 read-out neuron with weights w1 and w2. In the hypothetical behavior task, only I1 is paired with hippocampal feedback, leading to ITDP-like potentiation of w1 (thick red arrow). b. Case 1 (Slice Condition): With intact HC → ECL2/3 but severed ECL2/3→HC connections, pairing I1 with hippocampal feedback increases w1 by ~1.5-fold, simulating slice ITDP results. The output of the unpaired input I2 remains unchanged. Potentiation scales with pairing repetitions and hippocampal feedback strength (blue=50, yellow=100, green=150 iterations). Case 2 (Intact Loop): Case 1 likely underestimates the effect due to severed axons (connections) between EC and HC. To answer whether learning rate increases with reciprocally driven activity between cortex and hippocampus, we simulated the in vivo condition where the whole ECL2/3→HC→ECL2/3 loop is intact. In this condition, the bar graph indicates that upon simulating an intact ECL2/3-HC-ECL2/3 loop, the same number of pairings of I1 with HC-ECL2/3 feedback and the subsequent reciprocal activation of the circuit sped up the learning rate and lead to a higher potentiation of I1 than seen in Case 1. Again, potentiation scales with pairing iterations and HC feedback strength (blue=50, yellow=100, green=150). Thus, our computational model suggests that reverberating activity in the reciprocal connections between hippocampus and ECL2/3 leads to a true feedback circuit, which iteratively improves the learning and output behavior of the system. Case 3: Silencing HC-EC loop – Upon silencing the entirety of the ECL2/3-HC-ECL2/3 loop, both the paired and unpaired inputs fail to potentiate as demonstrated by their weight holding at 1.0. Case 4: Silencing HC feedback – Silencing only the HC feedback to ECL2/3 is sufficient to eliminate the potentiation of the sensory input post-pairing.
