Reply to Quintana et al.: Behavior is an unlikely mediator of fluoxetine effects on ovulation in rabbits

In a Letter in PNAS, Quintana et al. (1) raise 2 issues with respect to our paper on the ovulatory homolog model of female orgasm (2). One is about the statistical analysis, and the second is about biological interpretation. We address both issues below.

Statistical Issues

In their Letter, Quintana et al. accuse us of P-hacking, an approach in which the same data are used to discover a pattern and then reused in a statistical test on that pattern, thus conflating hypothesis formation and hypothesis testing (3). In our study, we a priori predict both that fluoxetine has an effect on ovulation, and the direction of the effect, i.e., that it should decrease ovulation rate. The definitional criteria for P-hacking are thus not fulfilled for our study.

The pooling of data from experiments done in different weeks is legitimate if the samples are from the same treatments and if they are statistically indistinguishable. That is the case in our control animals. We used 2 control groups to guard against confounding effects of dates, as for logistical reasons, it was impossible to do all of the experiments in the same week. Since the control groups were indistinguishable, we concluded that there were no confounding effects and pooled the samples, yet reported P values for the treatment–control group comparisons for each cohort separately (ref. 2, figure 1B). The 2 cohorts have not only statistically indistinguishable effects but also similar P values. As an alternative, one could combine groups using Fisher’s method, $X_{2k}^2\sim -2{\displaystyle {\sum }_{i=1}^k\text{ln}\left(p_i\right)}$, and obtain a combined P < 0.025, which is at the same order of magnitude as the pooled data, P = 0.01. We believe that our results are robust.

Biological Interpretation

Quintana et al. raise the interesting possibility that the observed effect of fluoxetine on ovulation rate may be mediated by female behavior. This hypothesis is based on the pioneering work of Pfaus et al. (4–6) on the effect of fluoxetine on copulatory behavior of rats. In rats, repeated copulation is necessary to support the corpus luteum, an effect unknown from rabbits or humans. Copulation in rabbits is easily recognizable and very rapid. Immediately upon mating, the rabbits were separated, i.e., all females only encountered a single round of mating.

Nevertheless, we analyzed 2 traits that could reveal behavioral effects: the time to copulation and the number of attempted mounts by the male. We did find an effect of fluoxetine on both parameters. The time to copulation is shorter in control females (average, 14.3 s; median, 11.28 s) compared to the fluoxetine-treated females (average, 25.6 s; median, 19.01 s; Mann–Whitney U test, P = 0.0129; t test, P = 0.065, two-tailed; Fig. 1A). Also, the number of mounts is lower (control average, 1.11; fluoxetine average, 1.75). Hence there are measurable effects of fluoxetine on the female behavior, as suspected by Quintana et al. We then asked whether these effects can explain the difference in ovulation. Fig. 1 B and C shows that the results do not support the idea that these behavioral traits are related to ovulation rates. We conclude that fluoxetine affects both female behavior and ovulation, but female behavior does not influence ovulation in rabbits.

Fig. 1.

Effect of fluoxetine on behavior and ovulation in the female rabbits. (A) Effect of fluoxetine on time to copulation in seconds. The horizontal lines indicate the median (see text). B and C show the relationship between behavioral traits and ovulation. (B) Time to copulation in relation to number of ovulations; Spearman correlation ρ = 0.0038, P = 0.9897. (C) Number of mounts versus number of ovulations; Spearman correlation ρ = 0.199, P = 0.3867. Note that for neither behavioral trait is there any evidence that behavior is systematically related to the number of ovulations.