Developmental fine-tuning of medial superior olive neurons mitigates their predisposition to contralateral sound sources
Fig 4
Spike timing-dependent plasticity can gradually and partially compensate for a latency bias.
(A) The model neuron receives 2 populations of inputs with different latencies. The input latency is determined by a frequency-dependent latency and a side-specific latency. Each input starts with the same synaptic weight. The synaptic weights are adjusted based on synaptic activation relative to postsynaptic firing. The activation pattern depends on ITDs. After a number of training rounds, synaptic weights are different from the starting values, leading to a change in the ITD tuning response of the model neuron. (B) Development of bITDs of model neurons with a best frequency of 1.8 or 0.6 kHz. For each best frequency, the left panel shows the simulated membrane potential (normalized scaling) at ITD = 0 ms (green-to-black traces) and the input latencies (bottom, raster, orange = contralateral, blue = ipsilateral). During the training process, the synaptic weights are adjusted, and therefore the membrane potential changes (green: start, black: learning outcome). The right panel shows the ITD-rate curve during the learning process (from green to black: after 1, 200, 400, 600, 800, and 1,000 updates). Best ITDs are indicated by a plus. On the top of the graph, the initial (green) versus outcome bITDs (black) are indicated. Each example was trained with ITDs of 0 ± 0.13 ms, indicated by the yellow area. (C) Weights of ipsilateral (left) and contralateral (right) inputs after the learning process for model neurons with a best frequency of 1.6, 1.0 or 0.4 kHz, and 500 neurons per best frequency were trained for 1,000 updates. Vertical dotted line indicates the weight at the start, which was equal for all inputs. Zero weights (or eliminated inputs) are shown separately as a bar. (D) Average bITD against the number of updates for different best frequencies. Neurons with lower best frequencies need more updates to adjust their bITDs to the training ITDs (0 ± 0.13 ms, yellow area). We simulated 500 neurons for each frequency group. We used 1,000 updates (indicated by dashed line) for the other simulations. (E) Histogram of bITDs before (left) and after training with ITDs in the range of 0 ± 0.13 ms (middle; yellow area), similar to the ecological range of adult gerbils, and after training with ITDs in the range of 0.6 ± 0.13 ms (right). With ITDs of 0.6 ± 0.13 ms contralateral instead of ipsilateral inputs are leading by 0.3 ms. Total number of neurons is 2,400 neurons per graph. (F) Best ITD against best frequency before (left) and after training with ITDs in the range of 0 ± 0.13 ms (middle) or ITDs in the range of 0.6 ± 0.13 ms (right). The mean bITD is indicated by the red, solid lines. The training ITDs are indicated by the yellow area. The initial difference in latencies is indicated by the red, dashed line. The code to generate this figure is available at https://doi.org/10.5281/zenodo.10729468. bITD, best interaural time difference; ITD, interaural time difference.