A response was deemed a failure if the integrated response area after the flash was less than 64% greater than the integral of the absolute value of the baseline-corrected trace before the flash; the time window for integration was 400 and 600 ms in wild-type and GCAPs−/− rods, respectively. This 64% difference criterion was selected for its empirical robustness: in one test rod, the difference in failure count between using a 50% criterion and a 90% criterion corresponded

to a difference in 38 versus 42 failures out of 155 total responses. The 64% criterion counted 41. The failure count provided an estimate of the number of singletons (ns), using Poisson statistics. We then assumed our singleton population to be the set of ns Nintedanib order nonfailure responses having the smallest amplitudes. The mean singleton and the amplitude c.v. did not vary greatly when the singleton population was altered by the

inclusion of 1–2 additional responses or the exclusion of 1–2 of the largest responses. In addition, the average singleton determined from the set of ns was indistinguishable from that average single click here photon response calculated from variance-to-mean analysis. Likewise, the average of the responses that were deemed failures showed no time-dependent changes in current ( Figures 6A–6B). We elected not to use matched-filter analysis to identify singletons, as this approach assumes that all SPRs have an effectively identical shape combined with broadband noise—an

assumption that begs the question of how the SPR shape would change with variation in R∗ lifetime. Nonetheless, we compared the method used here to the traditional matched filter analysis-histogram method (e.g., Field and Rieke, 2002). We found that the matched filter method had a tendency to produce lower coefficients of variation for both WT and GCAPs−/− singletons, though this difference was not significant (GCAPs−/− c.v.: 0.40 ± 0.02 using matched filter analysis versus 0.42 ± 0.02 using smallest oxyclozanide nonfailures; WT: 0.31 ± 0.03 using matched filter analysis versus 0.34 ± 0.01 using smallest nonfailures). The matched filter analysis identified 246 total wild-type singletons from 5 rods, while the number of singletons expected from the number of identified failures is 263. The corresponding numbers for 4 GCAPs−/− rods are 142 and 152. In sum, matched filter analysis tended to identify fewer responses as singletons, in particular excluding more of the large amplitude responses on the fringe that would escape the boundaries imposed on the singles peak. When these responses are included in the statistical expectation method employed in the paper, they tend to increase the standard deviation. For each rod used for measuring reproducibility, the identified singletons were used to generate the time-dependent average.