Do Pregnancy-Induced Brain Changes Reverse? The Brain of a Mother Six Years after Parturition

Neuroimaging researchers commonly assume that the brain of a mother is comparable to that of a nulliparous woman. However, pregnancy leads to pronounced gray matter volume reductions in the mother’s brain, which have been associated with maternal attachment towards the baby. Beyond two years postpartum, no study has explored whether these brain changes are maintained or instead return to pre-pregnancy levels. The present study tested whether gray matter volume reductions detected in primiparous women are still present six years after parturition. Using data from a unique, prospective neuroimaging study, we compared the gray matter volume of 25 primiparous and 22 nulliparous women across three sessions: before conception (n = 25/22), during the first months of postpartum (n = 25/21), and at six years after parturition (n = 7/5). We found that most of the pregnancy-induced gray matter volume reductions persist six years after parturition (classifying women as having been pregnant or not with 91.67% of total accuracy). We also found that brain changes at six years postpartum are associated with measures of mother-to-infant attachment. These findings open the possibility that pregnancy-induced brain changes are permanent and encourage neuroimaging studies to routinely include pregnancy-related information as a relevant demographic variable.

. Image processing and statistical analysis. This figure describes the longitudinal symmetric diffeomorphic pipeline used in the study.

Supplementary Methods and Results 1
During the acquisition of the POST6y session, there was an MRI software update on the Philips scanner (from version 5.1.7 to version 5.3.1). Six participants out of twelve were scanned with a different software version. Despite maintaining constant the image reconstruction parameters, when inspecting in detail the images, we noticed that the MRI software update slightly affected the image reconstruction. In one of the software versions, there were incompatibilities between acquisition and reconstruction parameters. These incompatibilities affected the relation between the voxel size and the acquisition field of view. As a result, images with the newest software version suffered from a subtle enlargement on the phase encoding direction. The magnitude of this enlargement was estimated based on affine nine degrees of freedom registrations between the postpartum session (POST) and the six years postpartum session (POST6y). Affine registrations were performed with FSL-FLIRT between the brain-extracted images using weighting volumes to give more importance to the inner structures. The average of the 3x3 matrix determinants resulting from the affine nine of transformation was calculated. To perform the correction, we applied the diagonal transformation matrix with the scaling factor in the phase encoding direction to the images. Importantly, the same scaling factor was applied to all the subjects with the newest MRI software, thus ensuring that we maintain intrasubject variability.
The reconstruction differences between the software versions 5.1.7 and 5.3.1 were communicated to Philips so they can inform other users whose longitudinal data could also be affected.
To demonstrate that the inclusion of this correction step did not significantly change the interpretation of our findings, below we show the results --group differences--obtained without applying the software correction.
As observed in Supplementary Figure 1, the tendencies in the mothers' group are maintained. However, in the control subjects, which were all scanned after the software update, gray matter (GM) volume reductions are closer to zero, thus accentuating and biasing both group differences (all POST6y-PRE P-values< 0.0152). Mean values (circle) with their respective standard error of the mean (vertical lines) and slopes (lines joining the circles) are represented. Black and gray lines represent mothers and nulliparous women, respectively. The blue shadow indicates the approximated period of pregnancy. Abbreviations are as follows: GM= gray matter, L.= Left hemisphere, R.= Right hemisphere, PRE= pre-pregnancy session, POST= early postpartum session, POST6y= six years after parturition session. Asterisks indicate group differences at q< 0.05 FDR-corrected for multiple comparisons.

Methods:
Given our limited sample size, we calculated the Positive Predictive Value (PPV) of our main results as a post hoc analysis. The PPV is the probability that a "positive" or significant finding reflects a true effect. This probability was calculated for the between group differences in GM volume change (POST6y-PRE) in the "All ROIs" mask. This probability depends on three parameters: 1) the statistical power, which in turn depends on the effect size and sample size; 2) the threshold for statistical significance, and 3) the odds that a tested effect is a truly non-null effect among the effects being tested, also known as pre-study odds. The PPV can be estimated with the following formula: Where β is the probability of obtaining a false negative, (1−β) is the statistical power, α is the probability of obtaining a false positive, and R is the pre-study odds.
The study of Hoekzema et al., 2017 is the only publication available from which to estimate our pre-study odds of the long-term effects of pregnancy. We are taking a confirmatory approach in an a priori selected ROI. Thus, the pre-study odd would be larger than 1. We have calculated the PPV as a function of different pre-study odds (from 0 to 2). To be prudent, we considered pre-study odd equals to 1, meaning that, based on previous literature, we expect the same probability for non-null effects and null effects.

Results:
The estimated Cohen's d effect size of the between-group differences in the GM volume change (POST6-PRE) of the "All ROIs" mask was 1.635, and the statistical power of the test was 0.768. The large effect size counteracted the reduced PPV commonly associated with a reduced sample size. As indicated in Supplementary Figure 2, for a pre-study odds equal to 1, the probability that our group differences in the "All ROIs" mask reflect a true effect is 0.939. (GM) volume changes between the pre-pregnancy and six years after parturition sessions in mothers and nulliparous women for the "All ROIs" mask. b) Graphical representation of the estimated Positive Predictive Value as a function of different pre-study odds. The considered significance level is P-value< 0.05, and the statistical power of the test is 0.768. The blue shadow indicates the values of R larger than 1. Abbreviations are as follows: PRE= pre-pregnancy session, POST6y= six years after parturition session.

Supplementary Methods and Results 3
Due to a technical problem, the radiofrequency head coil (RFHC) was replaced in three participants out of twelve. To make sure that our findings do not depend on this variable, we repeated the main analysis excluding the three participants scanned with a different RFHC.
As observed in the figure below, the patterns of GM volume changes and group differences are very similar.  et al., 2017]. Mean values (circle) with their respective standard error of the mean (vertical lines) and slopes (lines joining the circles) are represented. Black and gray lines represent mothers and nulliparous women, respectively. The blue shadow indicates the approximated period of pregnancy. Abbreviations are as follows: GM= gray matter, L.= Left hemisphere, R.= Right hemisphere, PRE= pre-pregnancy session, POST= early postpartum session, POST6y= six years after parturition session. Asterisks indicate group differences at q< 0.05 FDR-corrected for multiple comparisons. Table S1. Clinical and parturition data.   ('PRE' and 'POST' sessions) between the primiparous mothers and nulliparous control group. Shapiro-Wilk tests indicated that some of these variables did not follow a normal distribution, and therefore non-parametric two-tailed Mann-Whitney U tests were applied. Equal variances were confirmed using the non-parametric Levene's test. There were no significant changes in performance on these measures, although it should be noted that larger sample sizes would be required to more reliably examine this type of data and reveal subtle effects. For the TAVEC, complete PRE/POST datasets are available of 23 primiparous women and 16 nulliparous control women (not all subjects participated in these tests in both sessions). For the N-back and RT test this is 22/15 and for the Digits test: 25/16. TAVEC = Test de Aprendizaje Verbal España-Complutense, based on the California Verbal Learning Test, Digits = the Digits subtest of the Wechsler Adult Intelligence Scale III, N-back = a visual 2-back test, RT = a simple visual reaction time task. Abbreviations: Mdn= Median, IQR = Interquartile range. Table S3. Group differences in gray matter changes.