Relating sex-bias in human cortical and hippocampal microstructure to sex hormones

Abstract

Determining sex-bias in brain structure is of great societal interest to improve diagnostics and treatment of brain-related disorders. So far, studies on sex-bias in brain structure predominantly focus on macro-scale measures, and often ignore factors determining this bias. Here we study sex-bias in cortical and hippocampal microstructure in relation to sex hormones. Investigating quantitative intracortical profiling in-vivo using the T1w/T2w ratio in 1093 healthy females and males of the cross-sectional Human Connectome Project young adult sample, we find that regional cortical and hippocampal microstructure differs between males and females and that the effect size of this sex-bias varies depending on self-reported hormonal status in females. Microstructural sex-bias and expression of sex hormone genes, based on an independent post-mortem sample, are spatially coupled. Lastly, sex-bias is most pronounced in paralimbic areas, with low laminar complexity, which are predicted to be most plastic based on their cytoarchitectural properties. Albeit correlative, our study underscores the importance of incorporating sex hormone variables into the investigation of brain structure and plasticity.

Fig. 1. Intracortical T1w/T2w signal intensity profiling.

A Parcellation scheme. B Intracortical sampling to build microstructural profiles. Twelve equivolumetric surfaces are put between cortical surface and white matter boundary of a single subject, yielding 12 sample points at different intracortical depths. C Left: Group averages (N = 1093) of microstructural measures (i-iii), plotted on the cortical surface. Right: examples for parcels with a high and low profile mean (i) and skewness (ii), per intracortical sample point, respectively. Microstructural profile covariance matrix (MPC) (iii) based on correlations of microstructural profiles between pairs of parcels. In the right MPC, parcels are ordered according to their microstructural differentiation, using the principal component derived from diffusion embedding. D Map of the hippocampal subfields after extraction and unfolding of the hippocampus (i), and the group-average T1w/T2w signal intensity (ii) for the left and right hippocampus. T1w/T2w the ratio of T1- over T2 weighted magnetic resonance imaging (MRI), MPC microstructural profile covariance matrix, CA Cornu ammonis, Sub Subiculum, A anterior, P posterior, M medial, L lateral. Source data are provided as a Source Data file.

Fig. 2. Sex differences in the brain’s microstructural organization.

A FDR-thresholded Cohen’s d maps showing significant sex differences (females (n = 594) > males (n = 499)) in intracortical microstructure: T1w/T2w based intracortical profile mean, profile skewness, and the microstructural gradient. Red colors represent microstructural values higher for females, blue represent values higher for males. B 2D and 3D FDR thresholded Cohen’s d maps of the unfolded hippocampus showing significant differences in T1w/T2w mean between females (n = 500) and males (n = 367 male). C Boxplots represent Pearson’s r-values between unthresholded t-statistics resulting from two respective split-halves of the sample (n = 1000 permutations) comparing profile mean, profile skewness, and microstructural gradient between females and males, indicating the reliability of each measure. The median is shown as the central mark, the box indicates the 25th and 75th percentile; whiskers include all values not considered outliers (1.5*IQR from the quartile). Source data are provided as a Source Data file.

Fig. 3. Comparing males to different female sub-samples, grouped by menstrual cycle phase.

A Schematic of normative trajectories of estrogen and progesterone fluctuations during the menstrual cycle, based on^,,. Horizontal lines under the x-axis indicate grouping in this work: purple reflects progesterone (dotted = low; solid = high); turquoise reflects estrogen (dotted = low; solid = high). B Distribution of sex-difference effect per parcel, by group comparison for each microstructural measure. Brackets indicate significant differences in the cortex-wide sex-difference effect distribution between respective groups, based on post-hoc contrasts of ANOVA results using Tukey’s honest significant difference procedure (n = 400 parcels, two-sided, *** black brackets p = 0.0000; ** dashed black brackets = p < 0.01, * dashed grey brackets p < 0.05). Test details for tests with 0.05 > p > 0.0000 (all skewness): cortex-wide average $d_{highprogfemales-males}=0.1995$ vs cortex-wide average $d_{allfemales-males}$ = 0.1681: CI [0.0057–0.0569] p = 0.0066; cortex-wide average $d_{highprogfemales-males}$ = 0.1995 vs cortex-wide average $d_{lowprogfemales-males}$ = 0.1680: [0.0058–0.0571]p = .0062; cortex-wide average $d_{highprogfemales-males}$ = 0.1995 vs cortex-wide average $d_{lowestrfemales-males}$ = 0.1731: CI[0.0007–0.0520] p = 0.0396. C FDR-thresholded Cohen’s d maps projected on the cortical surface and the hippocampus of T1w/T2w profile mean (i), T1w/T2w profile skewness (ii), and microstructural gradient (iii) between males and subsamples of females divided by OC use and menstrual cycle phase. For completeness, all other FDR-thresholded Cohen’s d maps (all group-comparisons, for each of the three measures) are plotted in the supplement. OC oral contraceptives, A anterior, P posterior, M medial, L lateral. Source data are provided as a Source Data file.

Fig. 4. Differences between NC females and OC females for each intracortical measure.

FDR-Thresholded Cohen’s d for microstructural differences between female groups, comparing OC females with all NC females (n = 284), as well as OC females (n = 170) with specific NC subgroups, divided by hormone estimations according to self-reported days after menstruation (n_{low estrogen} = 100; n_{low progesterone} = 171; n_{high estrogen} = 184; n_{high progesterone} = 113). Columns are the three microstructural measures T1w/T2w mean (A), T1w/T2w skewness (B), and the microstructural gradient (C). Purple areas are parcels which had significantly higher values in OC females, orange shows significantly higher values for NC females after FDR-thresholding (all Cohen’s d). OC oral contraceptives, NC naturally cycling, prog progesterone, estr estrogen. Source data are provided as a Source Data file.

Fig. 5. Spatial overlap between effect maps of sex differences for the microstructural gradient, profile mean, and profile skewness.

Transcriptomic maps of genes are sorted by categories: sex hormone synthesis-related genes, and androgen, estrogen, and progesterone receptor-related genes. We test for spatial specificity by comparing against the principal component of all genes (baseline). Shades of red represent positive r-values, shades of blue represent negative correlations; circle size and shading indicate size of correlation. Values with significant p-values (p < 0.05) after permutation spin-testing are marked with a black outline (one-sided). Note that no correlation is significant when accounting for multiple testing at an FDR-threshold. Source data are provided as a Source Data file.

Fig. 6. Contextualization of effects by histological decoding.

A Schematic of cortical types according to von Economo and Koskinas and Garcia-Cabezas^,,. B Results were put into context by spatial correlations with a hierarchy of laminar differentiation (cortical types). Figures show links between cortical type hierarchy and effect values (Cohen’s d for each of the 400 parcels) for each of the T1w/T2w profile-based intracortical measures. Raincloud plots show distribution of sex-difference effects per parcel, binned by cortical type. Also binned by cortical type, boxes show the median and interquartile (25–75%) range of the respective distribution of sex-difference effects, whiskers depict the 1.5*IQR from the quartile. C Zero-distributions between random hierarchies and effect maps in comparison to the statistical r-value (one-sided). Profile skewness and gradient correlate significantly with histological hierarchy according to spatial autocorrelation significance level; profile mean does not. Kon koniocortex, EU-III Eulaminate III, EU-II Eulaminate II, Eu-I Eulaminate I, Dys Dysgranular, Ag Agranular. Source data are provided as a Source Data file.