Genomic environment predicts expression patterns on the human inactive X chromosome

Abstract

What genomic landmarks render most genes silent while leaving others expressed on the inactive X chromosome in mammalian females? To date, signals determining expression status of genes on the inactive X remain enigmatic despite the availability of complete genomic sequences. Long interspersed repeats (L1s), particularly abundant on the X, are hypothesized to spread the inactivation signal and are enriched in the vicinity of inactive genes. However, both L1s and inactive genes are also more prevalent in ancient evolutionary strata. Did L1s accumulate there because of their role in inactivation or simply because they spent more time on the rarely recombining X? Here we utilize an experimentally derived inactivation profile of the entire human X chromosome to uncover sequences important for its inactivation, and to predict expression status of individual genes. Focusing on Xp22, where both inactive and active genes reside within evolutionarily young strata, we compare neighborhoods of genes with different inactivation states to identify enriched oligomers. Occurrences of such oligomers are then used as features to train a linear discriminant analysis classifier. Remarkably, expression status is correctly predicted for 84% and 91% of active and inactive genes, respectively, on the entire X, suggesting that oligomers enriched in Xp22 capture most of the genomic signal determining inactivation. To our surprise, the majority of oligomers associated with inactivated genes fall within L1 elements, even though L1 frequency in Xp22 is low. Moreover, these oligomers are enriched in parts of L1 sequences that are usually underrepresented in the genome. Thus, our results strongly support the role of L1s in X inactivation, yet indicate that a chromatin microenvironment composed of multiple genomic sequence elements determines expression status of X chromosome genes.

Figure 1. Procedure Used to Obtain Overrepresented Oligomers Starting from the Overrepresented 12-Mers

The overrepresented 12-mers were defined with the initial criteria: at least ten occurrences and at least 5-fold enrichment. See Results and Methods for a detailed description.

Figure 2. The Distributions over the Length of L1 Element of Overrepresented Oligomers Found in I Subgenome (Yellow Bars) and of All L1 Sequences within Xp22 (Red Bars)

Only overrepresented oligomers mapping frequently to L1s (>50% of their genomic occurrences in the I subgenome) are shown. Although the full-length L1 is approximately 7 kb long, the alignment of L1 subfamilies was approximately 9 kb long. ORF, open reading frame.

Figure 3. LDA Classification Success Rates for Different Values of the Tuning Parameter τ

(A) Training set derived largely, but not exclusively, from Xp22 (See Table S3). (B) Test set of Xp22 genes, with training performed on genes in (A). (C) Test set of X genes outside of Xp22, with training performed on genes in (A). (D) Training set of all X genes, including genes in Xp22. Dots indicate optimal values of τ (see Table 4 and Methods).

Figure 4. The Distribution of Correctly and Incorrectly Classified Genes along the X Chromosome

Dark green indicates correctly classified genes; light green indicates misclassified genes. X inactivation expression patterns [6] for genes included in this study: yellow indicates inactivated genes, and blue indicates escape genes. Not all genes were analyzed at all distances because sequences that included adjacent genes with different inactivation patterns were excluded from analysis (see Methods). These gene distances remain uncolored.