Sequence features accurately predict genome-wide MeCP2 binding in vivo

Abstract

Methyl-CpG binding protein 2 (MeCP2) is critical for proper brain development and expressed at near-histone levels in neurons, but the mechanism of its genomic localization remains poorly understood. Using high-resolution MeCP2-binding data, we show that DNA sequence features alone can predict binding with 88% accuracy. Integrating MeCP2 binding and DNA methylation in a probabilistic graphical model, we demonstrate that previously reported genome-wide association with methylation is in part due to MeCP2's affinity to GC-rich chromatin, a result replicated using published data. Furthermore, MeCP2 co-localizes with nucleosomes. Finally, MeCP2 binding downstream of promoters correlates with increased expression in Mecp2-deficient neurons.

Figure 1. MeCP2 binding is accurately predicted by GC%.

(a) Wiggle plots showing sequence GC%, CpG methylation level, mCpG% and MeCP2 ChIP-seq and Input profiles for each replicate around the Bdnf locus. ChIP-qPCR bars indicate MeCP2 ChIP/Input (%). See Supplementary Table 1 for primer sequences. (b) ROC curve for predicting MeCP2 ChIP-seq peaks using Random Forest regressor based on GC% in 200 bp windows. (c) Relative importance of different sequence features in predicting MeCP2 binding using the Random Forrest regressor algorithm (Methods). Trees with maximal depth 8 were used, but the dominant importance of GC% did not depend on this choice. (d) GC% and CpG% dependence of the mean MeCP2 enrichment (colours) calculated using 150 bp windows. The contours indicate the genome-wide joint distribution p(GC%, CpG%) and contour labels indicate the enclosed genome fraction. The inset shows mean MeCP2 enrichment versus GC%.

Figure 2. MeCP2 binding follows GC% in regions with extreme GC% and CpG%.

(a–c) Alignment plots showing MeCP2 enrichment, GC%, and CpG% (colour) around regions with no CpG and GC≥60%. These regions were identified genome wide using sliding windows, and overlapping regions were joined (the resulting regions are outlined by dashed curves). (d) Comparison of the aligned MeCP2 ChIP-seq signal in this and previously published studies using the regions in a–c. For Baubec et al., the figure shows the ESC data. The y axis represents ChIP/Input for all data sets, except for Skene et al. which does not have Input. (e–g) Same as a–c but showing regions with CpG≥3% and GC≤35%. (h) Same as d but using the regions from e–g.

Figure 3. MeCP2 enrichment depends weakly on mCpG%.

(a) GC% (green), mean MeCP2 enrichment (purple) and mCpG% (red) around proximal (that is, located within 500 bp of a TSS, solid) and distal (dashed) CGIs. (b) Gaussian graphical model (Methods) describing the full partial correlations between the MeCP2 ChIP, Input, GC%, CpG%, mCpG% and mCpH%, all evaluated using 150 bp binning. (c) Same as b but using MeCP2 ChIP-seq data from Skene et al. (no Input was available) and methylation data from Lister et al. (d) Same as b but using ES cell MeCP2 ChIP and Input from Baubec et al. and methylation data from Stadler et al. (no mCpH% was available). (e) Same as b but using MeCP2 ChIP and Input from Chen et al. and methylation data from Lister et al. (f) Same as b but using MeCP2 ChIP and Input from Gabel et al. and methylation data from Lister et al.

Figure 4. MeCP2 binding coincides with nucleosomes.

(a) Enrichment plot of WT MNase-seq profile (colour) around MeCP2 ChIP-seq peaks (sorted by length and outlined by dashed curves). (b) GC% and CpG% dependence of the mean WT MNase-seq coverage (colours). The contours indicate the genome-wide joint distribution p(GC%, CpG%). (c) MeCP2 enrichment as a function of GC% and MNase-seq fragment coverage. The contours indicate the genome-wide joint distribution p(GC%,MNase-seq).

Figure 5. MeCP2 binding at the 5′ end of genes is associated with increased transcription in Mecp2 KO.

(a) Top plot shows the alignment of MeCP2 enrichment profiles (colour) in 8 kb regions surrounding the significantly up- and down-regulated genes after Mecp2 KO (separated by thick horizontal black line). The TSSs are ordered by fold change in expression between KO and WT (indicated by curved grey line). The bottom plot shows the mean MeCP2 enrichment in up- and down-regulated genes. (b) Same as a but showing GC%. (c) Same as a but showing mCpG%. (d) Dependence of the differential expression on MeCP2 enrichment, GC%, mCpG% and gene length. Each subplot shows how the mean log₂-fold expression change (colour) depends on a pair of covariates (averaged across the first 500 bps downstream of TSS) after grouping by quintile (columns and rows in subplots). Only genes with significant differential expression are included. Grey indicates combinations without data. The direction of the colour gradient indicates the strongest dependence of the differential expression. (e) Olfactory sensory neurons were identified by the expression of olfactory marker protein (OMP, green) and the nuclei were identified by DAPI staining (blue). No significant changes in cell body and nuclei sizes were observed between wild-type (identified by MeCP2 expression (Red), left panel) and KO (shown by lack of MeCP2 expression, right panel). Scale bar, 20 μm. Quantification is shown in Supplementary Fig. 10.