Schematic Overview of our study.

RNA-seq datasets from 54 human tissues (including 13 brain regions) were downloaded from GTEx Portal. GE and AS quantifications were performed by Paean. The individual sex or age and age-by-sex effects on global transcriptomic variation were evaluated by designing a pcSVR value, and the age-differential genes/AS events were subsequently identified in each sex. Sex-biased age-associated AS events were used to find the sex-specific associations with certain diseases. We also constructed AS regulatory networks to find the regulatory age-associated splicing factors (SF) and explored the transcriptional regulation of sex hormones via nuclear receptors. Moreover, we focused on age-associated GE and then evaluate the aging rate and their modulated genes in different sexes.

Individual sex or age effects and combined age-by-sex effects on global transcriptome variation.

(A-B) pcSVR between different sex or age groups calculated by GE (a) and AS (b). The yellow triangles show the significant sex-pcSVR calculated between females and males, while black triangles show the significant age-pcSVR calculated between young (i.e., 20-40 years old) and old (i.e., >60 years old) using the empirical p-value cutoff 0.1. Insignificant data points are labeled in grey.

(C) The age effect on transcriptome variation as calculated by GE vs. AS. The X-axis of the scatter plot show the log2 transformed age-pcSVR between young and old calculated by GE, while the Y-axis shows the values calculated by AS. Blue dots indicate the tissues with significant pcSVR specific in AS, while green dots indicate the tissues with significant pcSVR specific in GE. Tissue names in each class are labeled under the scatter plot.

(D) The sex effect on transcriptome variation as calculated by GE vs. AS. The analyses and labels are similar to panel c except the sex-pcSVR is used in all calculations.

(E-F) Scatter plots of the age effect between females vs. males calculated by GE (E) and AS (F). The X-axis shows log2 transformed age-pcSVR calculated in females, while the Y-axis shows the age effect calculated in males. Dashed lines showed equal age-pcSVR in females and males. Dark red and blue dots indicate the tissues with female-specific and male-specific age effects.

Sex-stratified age-associated genes and AS events across tissues.

(A-B) The numbers of age-associated genes (A, left) and AS events (B, left) in each sex. The red bars indicate the numbers of age-associated genes or AS events in females, while blue bars indicate those in males. The numbers of age-associated genes and AS events common in both sexes are shown in bars with relatively lighter colors. The significance of the overlapped age-associated genes (A, right) or AS events (B, right) between sexes are estimated using hypergeometric test. Tissues with significant overlapped p-values are labeled by dark red dots.

(C) Correlation between the age effects on genes (green) or AS events (blue) in females vs. males. The Y-axis represents the correlations of the effect sizes of age (βF, βM) between two sexes across all genes/AS events (Spearman’s correlation). Grey lines link a pair of tissue in GE and AS. The p-values are estimated using Wilcoxon signed-rank test.

Male-biased associations between the AS changes during aging and Alzheimer’s disease.

(A) GO analysis of the sBASEs in each sex and brain region. Heatmap shows the sex-specific pathways that are significantly enriched in more than 3 brain regions. Clustering is conducted by default parameters in pheatmap functions. The -log10 transformed enrichment p-values are shown in the color scale.

(B) Venn diagram between the sBASEs in females and males with AD-related AS events. The p-values are calculated using hypergeometric test.

(C) Correlation between AD-associated and age-associated AS changes in females and males. The X-axis indicates the PSIOld-PSIYoung, while the Y-axis indicates PSIAD- PSIControl. sBASEs in females are labeled in red, while sBASEs in males are in deep blue. The estimated Rho and p-value by spearman’s correlation test in each sex are labeled on the top.

(D) Model for AD prediction and feature importance evaluation. sBASEs are used for predicting AD in females and males respectively. 90% of samples are randomly selected as training sets for 100 iterations. The recursive feature elimination approach is used for feature selection. Feature importance is evaluated by the averaged MDA across 100 iterations.

(E) AS levels of the skipped exon 10 on SLC43A2 during plaque stages in males and females.

(F) AS levels of the skipped exons 3 and 4 on FAM107A during tangles stages in males and females. Mean ± standard error is shown in the error bar.

(G) Performances of sex-stratified and merge-sexes models predicted by sBASEs or randomly selected AS events for 100 iterations. The control models (i.e., the sex-stratified model trained by randomly selected AS events and the merge-sexes model trained by sBASEs in each sex) are highlighted with dashed lines. The mean ± s.d. (standard deviation) of AUC across 100 iterations is shown in the boxplot.

Sex-dimorphic AS regulation during aging in decision-related brain region.

(A) Schematic diagram for constructing AS regulatory networks (left) and the networks of the decision-related brain region (right). The blue and red lines indicate the regulations in males and females. Hub genes in octagons are SFs. The red ones are female-specific age-associated SFs, while the age-associated SFs common in both sexes are labeled in yellow. The color of the ellipse indicates the significance of sBASEs during aging, and the thickness of the line shows -log10 spearman’s correlation p-values between the TPMs of SFs and PSIs of AS events during aging.

(B) Examples of female-specific age-associated SFs (left) and age-associated SFs common in both sexes (right) during aging. The p-value and foldchange between old and young are labeled at the bottom.

(C) Percentages of the sBASEs regulated by the sex-specific age-associated SFs in four brain regions.

(D) Functional enrichment of sBASEs regulated by female-specific age-associated SFs based on MsigDB.

(E) Schematic diagram for the transcriptional regulation on age-associated SFs via nuclear receptors (left). Boxplot shows the ESR1 binding scores on age-associated SFs (right). The Y-axis indicates the ratio of ESR1 binding scores of each SF divided by the median binding score of age-associated SFs common in both sexes.

(F) Expression levels of ESR1 during aging. The legends of each group are the same as those in B.

(G) Expression levels of SRSF1 and SRSF7 treated by estrogen receptor agonist (PPT) vs. DMSO in MCF-7 cell line (N=3). The error bars indicate the mean ± s.d..

(H) Expression levels of SRSF1 and SRSF7 treated by 1nM estradiol (E2) vs. vehicle control (veh) in ESR1 wildtype, Y537S, and D538G mutant MCF-7 cells (N=4).

(I) Expression levels of SRSF1 and SRSF7 during aging, as plotted by mean ± s.e. (standard error).

Sex-dimorphic aging rate of GE during aging process.

(A) Workflow of time series and breakpoint analysis of gene expression in each sex.

(B) Aging rates and breakpoints in the aorta (left) and transverse colon (right). The Y-axis shows the estimated aging rate (see Materials and Methods). The p-values are calculated using Wilcoxon signed-rank test.

(C) Characteristics of the major breakpoints across tissues in each sex. The left panel shows the distribution of the major breakpoints across multiple tissues. The right panel shows the aging rate at the major breakpoints. The p-values are calculated using Wilcoxon signed-rank test.