Figure 4. Simulated neurons suggest a role for adaptation in cortical dereverberation.

To confirm that STRF differences between rooms were genuinely a result of adaptation, we simulated the recorded neurons using a non-adaptive linear-nonlinear-Poisson model and compared STRF measures of the simulated responses with those of the real neuronal STRFs in the different room conditions. (A) The simulated neurons were made in the following way: (1) We fitted a single STRF for each neuron using the combined data from the small and large rooms; (2) We used this STRF along with a fitted non-linearity and a Poisson noise model to generate the simulated firing rate for the small and large rooms separately; (3) Using the small and large room cochleagrams and simulated firing rates, we fitted separate STRFs for the two conditions; (4) We computed the center of mass and peak time metrics as before. (B) Difference in center of mass between the large and small room conditions (large - small room) for the simulated neurons. The $CO{M}^{-}$ values (blue) were larger in the large room (median difference = 4.0ms, mean difference = 5.1ms), and the $CO{M}^{+}$ values (red) were slightly elevated too (median difference = 3.1ms, mean difference = 3.1ms). (C) Reproduction of Figure 3B showing the difference in center of mass of neuronal STRF components between the large and small room conditions (large - small room). The $CO{M}^{-}$ values increased in the larger room (median difference = 9.3ms, mean difference = 12.0ms), whereas $CO{M}^{+}$ did not differ significantly (median difference = 0.32ms, mean difference = 0.59ms). (D) For each unit, the center of mass differences shown in B were subtracted from those in C and plotted as the resulting difference of $COM$ differences (real cortical unit - simulated neuron). The $CO{M}^{-}$ differences between rooms were consistently larger in the neuronal data (median difference = 5.7ms, mean difference = 6.9ms), while the $CO{M}^{+}$ effect was larger in the simulations (median difference = –2.0ms, mean difference = –2.5ms). (E) Difference in peak time between the large and small rooms (large - small) for the simulated neurons. The $P{T}^{-}$ median difference = 6.4ms (mean difference = 13ms) and the $P{T}^{+}$ median difference = –0.50ms (mean difference = –0.43ms). (F) Reproduction of Figure 3D showing the difference in peak time between the large and small rooms (large - small), calculated from neuronal STRFs. The $P{T}^{-}$ values were larger in the large room (median difference = 9.4ms, mean difference = 20.0ms). $P{T}^{+}$ did not differ significantly between the rooms (median difference = 0.0ms, mean difference = 3.0ms). (G) Histogram of the difference in peak time room differences between the cortical units and corresponding simulated neurons (cortical unit - simulated neuron), plotted as in D above. The $P{T}^{-}$ shifts were consistently larger in the neuronal data than in the simulated neurons (median difference = 1.1ms, mean difference = 7.4ms). $P{T}^{+}$, on the other hand, showed larger effects of room size in the simulated data (median difference = 0.95ms, mean difference = 3.5ms). Asterisks indicate the significance of Wilcoxon signed-rank tests: ${\displaystyle {}^{\ast \ast \ast \ast }p<0.0001,{}^{\ast \ast \ast }p<0.001}$.

Figure 4—figure supplement 1. Comparison of real neurons and non-adapting network receptive field-Poisson (NRFP) simulated neurons.

To confirm that STRF differences between rooms were genuinely a result of adaptation, we simulated the recorded neurons using a non-adapting NRFP model and compared STRF measures of the simulated responses with those of the real neuronal STRFs in the different room conditions. The simulated neurons were made using a similar process to that described in Figure 4, with the difference that the linear-nonlinear part (Figure 4A) was substituted with the NRF model, which has more complex non-linearity. (A) Difference in center of mass between the large and small room conditions (large - small room) for the simulated neurons. The $C\mathrm{}O\mathrm{}{M}^{\mathrm{-}}$ values (blue) were larger in the large room (mean difference = 5.5ms, median difference = 5.6ms), and the $C\mathrm{}O\mathrm{}{M}^{\mathrm{+}}$ values (red) were slightly elevated too (median difference = 3.5ms, mean difference = 3.2ms). (B) The center of mass differences between the neuronal data and the simulated NRFP model data were subtracted for each unit and plotted as the resulting difference of differences (real cortical unit - simulated neuron). The $C\mathrm{}O\mathrm{}{M}^{\mathrm{-}}$ differences between rooms were consistently larger in the neuronal data (median difference = 4.7ms, mean difference = 6.3ms), while the $C\mathrm{}O\mathrm{}{M}^{\mathrm{+}}$ effects were modestly larger in the NRFP simulations (median difference = –1.9ms, mean difference = –2.8ms). (C) Difference in peak time between the large and small rooms (large - small) for the simulated neurons. The $P\mathrm{}{T}^{\mathrm{-}}$ increased from the small to the large room (median difference = 2.1ms, mean difference = 9.7ms) and the $P\mathrm{}{T}^{\mathrm{+}}$ showed a more subtle change (median difference = –0.4ms, mean difference = 1.3ms). (D) Histogram of the difference in peak time room differences between cortical units and corresponding simulated neurons (cortical unit - simulated neuron), plotted as in B. The $P\mathrm{}{T}^{\mathrm{-}}$ room effects were consistently larger in the neuronal data than in the simulated neurons (median difference = 1.7ms, mean difference = 10.0ms). $P\mathrm{}{T}^{\mathrm{+}}$ differences between rooms were small overall, but significantly larger in the simulations (median difference = 0.4ms, mean difference = 1.8ms). Asterisks indicate the significance of Wilcoxon signed-rank tests:${\displaystyle {}^{\ast \ast \ast \ast }p<0.0001}$.