The human inactive X chromosome modulates expression of the active X chromosome

Abstract

The "inactive" X chromosome (Xi) has been assumed to have little impact, in trans, on the "active" X (Xa). To test this, we quantified Xi and Xa gene expression in individuals with one Xa and zero to three Xis. Our linear modeling revealed modular Xi and Xa transcriptomes and significant Xi-driven expression changes for 38% (162/423) of expressed X chromosome genes. By integrating allele-specific analyses, we found that modulation of Xa transcript levels by Xi contributes to many of these Xi-driven changes (≥121 genes). By incorporating metrics of evolutionary constraint, we identified 10 X chromosome genes most likely to drive sex differences in common disease and sex chromosome aneuploidy syndromes. We conclude that human X chromosomes are regulated both in cis, through Xi-wide transcriptional attenuation, and in trans, through positive or negative modulation of individual Xa genes by Xi. The sum of these cis and trans effects differs widely among genes.

Keywords: Klinefelter syndrome; X chromosome inactivation; aneuploidy; dosage sensitivity; gene expression; sex chromosomes; sex differences; turner syndrome.

Graphical abstract

Figure 1

Gene expression analysis of cells from across the spectrum of sex chromosome constitution (A) Collection and processing of samples from individuals with variation in sex chromosome constitution. (B) Schematic of the sex chromosomes featuring the X-Y-shared pseudoautosomal regions, PAR1 and PAR2, and the diverged regions, NPX and NPY. (C) Linear modeling strategy for analyzing RNA-seq data from individuals with one to four copies of Chr X (zero to three copies of Xi). See also Table S1.

Figure 2

Quantitative assessment of Xi contributions to X chromosome gene expression (A) Schematic scatterplot, linear regression line, and ΔE_X calculation for a hypothetical Chr X gene. Each point represents the expression level for an individual sample with the indicated number of copies of Chr X. The calculated coefficients from the linear model in Figure 1C are used to derive ΔE_X. (B–E) Actual scatterplots and regression lines with confidence intervals for selected Chr X genes in LCLs, representing a range of ΔE_X values. Adjusted p values (FDR) < 0.05 indicate that ΔE_X values are significantly different from 0. (F and G) Scatterplots of ΔE_X versus significance for all Chr X genes expressed in LCLs (F) and fibroblasts (G) illustrate variation in Xi contributions to Chr X gene expression; genes with FDR < 0.05 and |ΔE_X| ≥ 0.2 are labeled; genes depicted in (B)–(E) are underlined. (H) Scatterplot comparing ΔE_X in LCLs and fibroblasts for 327 Chr X genes expressed in both cell types. Colors as in (F) and (G). Deming regression line and Pearson correlation are indicated. See also Table S2.

Figure 3

Contributions of Chr Y or 21 copy number to gene expression (A) Chr Y copy number series with zero to four copies. (B) Each point shows the expression of NPY gene KDM5D in one LCL sample across the Chr Y copy-number series, with the regression line and its confidence interval plotted. The formula for calculating ΔE_Y from the regression coefficients is indicated. (C and D) Scatterplot of ΔE_Y versus significance for all Chr Y genes expressed in LCLs (C) or fibroblasts (D); all NPY genes are labeled; KDM5D, depicted in (B), is shown in black. (E) Chr 21 copy-number series with two to three copies. (F) Each point shows the expression of CCT8 in one LCL sample across the Chr 21 copy-number series, with the regression line and its confidence interval plotted. The formula for calculating ΔE₂₁ from the regression coefficients is indicated. (G) Scatterplot of ΔE₂₁ versus significance for all Chr 21 genes expressed in LCLs. CCT8, depicted in (F), is shown in black. (H) Violin plots with median and interquartile range for ΔE values of NPX (without or with an NPY homolog), PAR, NPY, and Chr 21 genes. p values are listed for comparisons referenced in the text. ΔE_X values for NPX genes with and without a Y homolog were compared using Wilcoxon rank-sum test. ΔE_X and ΔE_Y values for PAR1 genes were compared using paired t test. See also Tables S4 and S5.

Figure 4

Comparison of ΔE_X values with allelic ratios (ARs) reveals that Xi modulates Xa expression (A) Stacked barplots for genes with ΔE_X values greater than, less than, or approximately equal to zero, apportioned by their annotated XCI status from published studies (see STAR Methods and Table S6 for newly compiled XCI status calls). (B and C) Venn diagrams comparing LCLs and fibroblasts for genes with ΔE_X values that are either (B) explained or (C) not explained by published XCI status. Genes expressed in both cell types were included in the Venn diagrams, and genes with cell-type-specific expression are noted below. (D and E) Each point shows the mean adjusted AR for an informative gene (with heterozygous SNPs in at least 3 samples with skewed XCI) and whether AR is significantly greater than zero in (D) LCLs or (E) fibroblasts. (F and G) Each point denotes AR and ΔE_X values for an AR-informative gene in (F) LCLs or (G) fibroblasts. The color of the point indicates whether the gene’s AR value is significantly greater than zero (blue) or not (gray); the shape indicates whether the gene’s ΔE_X value is significantly different from zero (circles) or not (squares); and an orange outline indicates that ΔE_X differs significantly from AR. Black diagonal line, AR = ΔE_X. Pearson correlation coefficients (r) and p values are indicated. (H) Venn diagram comparing LCLs and fibroblasts for genes with ΔE_X values not equal to their AR values. Genes expressed and informative in both cell types are depicted in the Venn diagram, with genes that are cell-type specific or informative in only one cell type indicated below. (I) Venn diagram comparing all modulated genes in LCLs and fibroblasts (the union of figures, C and H). All Venn diagram p values, hypergeometric test. See also Table S6.

Figure 5

Combining ΔE_X with metrics of constraint on expression levels identifies genes likely to contribute to phenotypes associated with Xi copy number Scatterplots of ΔE_X versus gene constraint percentile ranking for PAR1 (A and B) or NPX (C and D) genes. Each point represents an expressed gene with scores for at least two of the four expression constraint metrics evaluated, excluding ampliconic genes. Dashed lines indicate |ΔE_X| thresholds of 0.1 for genes to be considered likely contributors to phenotypes driven by Xi copy number; labeled genes include (A and B) SLC25A6, the only PAR1 gene to score above the 50^th percentile for autosomal and PAR genes, and (C and D) among NPX genes with |ΔE_X| > 0.1, the 10 genes with the highest constraint percentile rankings in LCLs or fibroblasts. See also Table S7.