The histone modification H3K27me3 is retained after gene duplication and correlates with conserved noncoding sequences in Arabidopsis

Abstract

The histone modification H3K27me3 is involved in repression of transcription and plays a crucial role in developmental transitions in both animals and plants. It is deposited by PRC2 (Polycomb repressive complex 2), a conserved protein complex. In Arabidopsis thaliana, H3K27me3 is found at 15% of all genes. These tend to encode transcription factors and other regulators important for development. However, it is not known how PRC2 is recruited to target loci nor how this set of target genes arose during Arabidopsis evolution. To resolve the latter, we integrated A. thaliana gene families with five independent genome-wide H3K27me3 data sets. Gene families were either significantly enriched or depleted of H3K27me3, showing a strong impact of shared ancestry to H3K27me3 distribution. To quantify this, we performed ancestral state reconstruction of H3K27me3 on phylogenetic trees of gene families. The set of H3K27me3-marked genes changed less than expected by chance, suggesting that H3K27me3 was retained after gene duplication. This retention suggests that the PRC2-recruiting signal could be encoded in the DNA and also conserved among certain duplicated genes. Indeed, H3K27me3-marked genes were overrepresented among paralogs sharing conserved noncoding sequences (CNSs) that are enriched with transcription factor binding sites. The association of upstream CNSs with H3K27me3-marked genes represents the first genome-wide connection between H3K27me3 and potential regulatory elements in plants. Thus, we propose that CNSs likely function as part of the PRC2 recruitment in plants.

Keywords: CNS; PRC2; PRC2 recruitment; ancestral state reconstruction.

Fig. 1.—

Gene families are enriched in either marked or nonmarked genes. (A) Distribution of gene families according to the proportion of H3K27me3-marked genes. (B) Bar heights show an average log₂ ratio of observed/expected number of gene families in each bin for 1,000 permutations. Stars indicate bars where observed values were more extreme than the expected values for all 1,000 permutations.

Fig. 2.—

H3K27me3 distribution in gene families reflects their phylogeny. (A) An example phylogenetic tree with ancestral state reconstruction shows AP2-domain transcription factors. H3K27me3 data are represented as squares at the tips. Reconstructed ancestral states are shown as circles. (B) Histogram of average number of H3K27me3 changes per branch when gain penalty is 1.1. Observed number of changes is lower than number of changes in any of the randomized cases.

Fig. 3.—

Paralogous CNSs are associated with H3K27me3. (A) Overlap between bigfoot gene pairs and α WGD gene pairs with H3K27me3 where both genes (two red squares), one gene (a red and a black square), or none (two black squares) are marked. (B) Overlap between paralogous pairs with CNSs (as determined by Baxter et al. 2012) with H3K27me3-marked pairs. Background gene set consists of all α paralogous pairs (3,817 pairs).

Fig. 4.—

Orthologous CNSs are associated with H3K27me3. (A) Overlap between genes with H3K27me3 and genes from the Haudry CNS data set with >10 CNS. (B) Overlap between marked genes and genes with Baxter data set CNSs. (C) Overlap between genes with >5 CNSs and H3K27me3-marked genes for upstream and downstream CNSs. The background data set for all Venn diagrams contains 32,678 genes. (D) Two examples of CNS inheritance and H3K27me3. Left panel: only one of Arabidopsis homologs is H3K27me3-marked and has a CNS conserved with poplar and grapevine orthologs. Right panel: both Arabidopsis paralogs have H3K27me3 and a shared CNS. There is no H3K27me3 data for grapevine and poplar.