Description of statistical models and model outputs.

Genome-wide statistical models were of the form lmer(Y ~ AZS + (1|indID) + (1|chrID/xxxxID)), with Y being (A) log2(FPKM)-values, (B) log2(Fold change)peakheight and (C) asin(sqrt(5mCG)). AZS is a factor with four levels: Female Autosomes (AAf), Male Autosomes (AAm), Female Z (Zf) and Male Z (Zm). We included the identity of the sample (indID) as a random effect. Depending on the unit of measurement (i.e. genes, ATAC peaks, 5mC sites), we included either gene (geneID), peak (peakID) or 5mC site (siteID) nested in chromosome (chrID) as additional random effects. In (A) we included chromosome and gene lengths as covariates and in (B) and (C) we included chromosome length as a covariate. Models were fitted for genes (including transcription start site, exons, introns and transcription termination sites) and regulatory regions (transcription start and transcription termination sites) and for liver and spleen separately. For the gene-centered analyses, we included log(gene length) as an offset term and in the analyses of the regulatory regions we included Z-transformed gene length. We did not observe overdispersion. We also fitted models focusing on chrZ, separating the PAR from the rest of the chromosome as lmer(Y ~ AZP + (1|indID) + (1|xxxxID)) or glmer(N Peaks ~ AZP + (1|peakID)). Here, AZP is a factor with four levels: Female PAR (PARZf), Male PAR (PARZZm), Female rest of chromosome Z (Zf) and Male rest of chromosome Z (ZZm). The random effect structure was similar to the above models, except that we left out chrID as a random effect. In (A), (B) and (C) we provide sample sizes and the variance explained by each random effect (Vxxx), parameter estimates with their standard errors, z- and p-values and also the ratios between parameter estimates with bootstrapped 95% confidence intervals.
(XLSX)