While increasing μL therefore increased the dependence of gain on contrast, this trend saturated above μL ≈35 dB SPL ( Figure 5A). At higher mean levels, gain was decoupled from the mean sound level and varied with contrast
alone. Interestingly, although changing mean level had no systematic effect on x-offset in our data ( Figure 5B), reducing the mean level typically increased y-offset, i.e., raised the minimum firing rate ( Figure 5C; examples in Figures S4A and S4B). Given the success of Equation 2 in modeling the relationship between σL and gain, we extended this model to include mean level, μL . The most explanatory model ( Equation 8) was a simple extension of the contrast-dependent model where b could vary with ATM/ATR inhibitor review μL . This allows μL to directly modulate the dependence of gain on contrast. Fitted values for b(μL)b(μL) are presented in Figure 5D, showing that at low μL, b is modulated by μL, whereas b saturates with Selleck Sirolimus high μL. For simplicity, we modeled this with an exponential function
( Equation 8; see also Model 6 in Table S2). This model explained 97% of the total variance in the data set ( Figure 5E). We did not estimate the parameters for individual units, and therefore did not cross-validate this model. All of the above results remained unchanged when gain was expressed as a function of σP/μP rather than σL ( Figure S4C). The above results suggest
that the recent spectrotemporal statistics of the stimulus modulate neural responses to a sound. We predicted that if a particular sound was presented in a low-contrast context, it would generate stronger responses than if presented in high-contrast context. To test this prediction, we embedded a fixed “test sound” into DRC segments of differing ADP ribosylation factor contrasts. This sound was designed to drive all units within an electrode penetration, by having stimulus energy within the receptive fields of the units recorded there (Figure 6A). The different contexts were provided by a DRC sequence that alternated between high (σL = 8.7 dB, c = 92%) and low contrast (σL = 2.9 dB, c = 33%) every 1 s. The same test sound was presented once per 1 s block at a random time relative to the onset of that block, i.e., the last switch in context. Among 63 units that responded reliably to the test sound, all but two responded more vigorously when this sound was presented in a low-contrast context than in a high-contrast context; the firing rate was a median 2.6 times greater in low-contrast context (p ≪ 0.001, sign-rank; Figures 6B–6D). This confirmed our prediction. This experiment also allowed a finer-grained comparison of the time course of responses in high and low context. Similar to the STRF analysis, we found no systematic difference between these (Figure S5).