Reproducible stability of verbal and spatial functions along the menstrual cycle

Recent studies have reported brain changes in response to ovarian hormonal fluctuations along the menstrual cycle. However, it remains unclear, whether these brain changes are of an adaptive nature or whether they are linked to changes in behavior along the menstrual cycle, particularly with respect to cognitive performance. To address this knowledge gap, we report results from 3 well-powered behavioral studies with different task designs, leveraging the advantages of each design type. In all three studies we assessed whether verbal or spatial performance (i) differed between cycle phases, (ii) were related to estradiol and / or progesterone levels and (iii) were moderated by individual hormone sensitivity as estimated by premenstrual symptoms. Overall, results of all three studies point towards a null effect of menstrual cycle phase and – to a lesser extent – ovarian hormones on verbal and spatial performance and provided no evidence for a moderation of this effect by individual hormone sensitivity. We conclude that there is substantial consistency in verbal and spatial performance across the menstrual cycle, and that future studies of intra-individual variation are needed.

leading to a final target and were instructed to reach the target as quickly as possible by moving around in the virtual environment using the arrow keys on the keyboard.In order to control for navigation strategy, directions were formulated either from an egocentric (left turn/right turn) or allocentric perspective (north, south, east, west) and referred to either Euclidian (number of blocks) or landmark information (until the tree) to describe distances.Participants could only move on to the next level if they reached the final target.Upon reaching the final target in each level participants were asked to orient themselves by identifying north.Navigation time (time to reach the final target) and orientation accuracy were recorded.

Power analyses
Power estimations were carried out in GPower 3.1.9.7.The sample size required for each study was determined a priori to provide 80% power for small to moderate effect sizes (Study 1: f = 0.20, Study 2: f = 0.15, Study 3: f = 0.25) in repeated measures ANOVA analyses assuming a correlation of 0.5 among repeated measures.For Study 1 we set the alpha error probability to 0.016 since 3 performance measures were assessed in this study and the number of measurements was set to 6, since we followed participants over approximately 2 cycles, selecting 3 phases per cycle.For Studies 2 and 3 we set the alpha error probability to 0.008, since 6 performance measures were assessed in this study.In Study 2 the number of measurements was 6 given that 2 task conditions were assessed over 3 cycle phases.In Study 3, the number of measurements included only the 2 task conditions.Using these input parameters GPower suggested a required sample size of n = 36 for Study 1, n = 71 for Study 2 and n = 180 for Study 3).An additional 4-6 participants was recruited for each study to account for drop-outs, though the final number of exclusions necessary was higher in Studies 2 and 3. Given that the analyses performed were linear mixed models rather than ANOVAs, these numbers should be considered as estimates.

Outlier correction
Conditions with accuracy measures below chance level were excluded from analyses.For outlier correction, items with reaction time (RT) measures of more than 3 standard deviations above the mean were excluded from reaction time analyses (mental rotation task, navigation task).Thus, in Study 1, 11 data points were excluded for MRT accuracy and 9 data points for MRT RT.In Study 2, 12 data points were excluded for MRT accuracy and 23 individual trials were excluded before calculating the average MRT RT.In Study 3, 20 data points were excluded for MRT accuracy and no trials were excluded before calculating the average MRT RT.

Statistical models
All linear mixed effects models included participant number (PNr) as a random factor and menstrual cycle phase (dummy coded) as a fixed effect.For longitudinal studies, session was included as a fixed effect (continuous, first session set to 0 as a reference point) to control for training effects.For Studies 2 and 3, task condition (dummy coded) and its interaction with menstrual cycle phase were included as fixed effects in order to assess menstrual cycle dependent changes in cognitive strategies.Accordingly, the following models were fitted for Study 1 (Performance ~ CyclePhase + Session + 1│PNr), Study 2 (Performance ~ CyclePhase*TaskCondition + Session + 1│PNr) and Study 3 (Performance ~ CyclePhase*TaskCondition + 1│PNr).In order to evaluate hormonal associations with performance measures, estradiol and progesterone, as well as their interaction were included as fixed effects in the models (Performance ~ Estradiol*Progesterone + Session + 1│PNr).In order to evaluate whether the menstrual cycle effect was moderated by individual differences via PMS symptoms, PMS symptoms and their interaction with cycle phase were included in the models (Performance ~ CyclePhase*PMS + Session + 1│PNr).In Study 1, hormone levels were entered once centred around the group mean and once individually centred around menses values in order to assess not only the association to absolute hormone values, but take into account individual variabilities in baseline values and assess the increase relative to the individual baseline.
In case of null effect of menstrual cycle phase, we calculated Bayes factors of a model including the cycle phase relative to the model not including cycle phase using 10.000 iterations for Monte Carlo sampling.In case of non-significant interactions, we calculated Bayes factors of a model including the interaction relative to a model including only main effects.Bayes factors (BF) quantify the relative likelihood of the observed data under two competing models.As per the function default, the prior distribution was a zero-centered Cauchy distribution with scale of 0.707.The participant number was included as random factor in the models.3: Results of hormonal association analyses.For the majority of associations frequentist statistics were backed up by Bayesian analyses indicating about 3 times higher likelihood for the models without estradiol and/or progesterone or their interaction.However, it was unclear, whether progesterone related to verbal fluency performance and whether estradiol related to navigation performance in Studies 2 and 3.

Table 1 : Descriptive statistics of verbal and spatial performance measures in 3 studies
SupplementaryTable 2: Results of menstrual cycle analyses.*p ˂ 0.05, ***p ˂ 0.001