Conservation and innovation in the DUX4-family gene network

Abstract

Facioscapulohumeral dystrophy (FSHD; MIM158900, MIM158901) is caused by misexpression of the DUX4 transcription factor in skeletal muscle. Animal models of FSHD are hindered by incomplete knowledge regarding the conservation of the DUX4 transcriptional program in other species. Despite the divergence of their binding motifs, both mouse DUX and human DUX4 in mouse and human muscle cells, respectively, activate genes associated with cleavage-stage embryos, including MERVL and ERVL-MaLR retrotransposons. We found that human DUX4 expressed in mouse cells maintained modest activation of cleavage-stage genes driven by conventional promoters but did not activate MERVL-promoted genes. Thus, the ancestral DUX4-regulated genes are characteristic of cleavage-stage embryos and are driven by conventional promoters, whereas divergence of the DUX4 and DUX homeodomains correlates with retrotransposon specificity. These results provide insight into how species balance conservation of a core transcriptional program with innovation at retrotransposon promoters, and establish a basis for animal models recreating the FSHD transcriptome.

Figure 1. Dux and DUX4 activate an early cleavage-stage embryo gene signature in muscle cells of their respective species

(a) Dux transcriptome in C2C12 mouse muscle cells: red dots are genes affected more than absolute(log2FoldChange)>=2 and adjusted p-value<=0.05. Normalized counts are calculated by DESeq2 (normalized count = read count/size factor, where size-factors are estimated with the median-of-ratios method). Control samples were un-induced cells of the same cell line. (b) Gene set enrichment analysis (GSEA): gene set is 2C-like gene signature, x-axis is log2FoldChange-ranked Dux transcriptome. Enrichment score (ES) increases when a gene in the Dux transcriptome is also in 2C-like gene set and a black vertical bar is drawn in lower panel; ES decreases when a gene isn’t in 2C-like gene set. P-value was empirically determined based on 1,000 permutations of ranked gene lists. (c) Human DUX4, mouse Dux and canine DUXC homeodomain alignments (%=percent amino acid identity, *=four predicted DNA-contacting residues). (d) GSEA: gene set is the top 500 most upregulated genes in DUX4-expressing human cells, x-axis is log2FoldChange-ranked Dux transcriptome in mouse cells. This cross-species comparison required limiting both gene set and transcriptome to 1:1 mouse-to-human orthologs. The converse comparison is in Supplementary Figure 4a. (e) GSEA: gene set is the human orthologs of the mouse 2C-like gene signature, x-axis is log2FoldChange-ranked DUX4 transcriptome in human muscle cells. Both gene set and transcriptome are limited to 1:1 mouse-to-human orthologs. Note: mouse 2C-like gene signature has 469 genes total, 297 of these genes have simple 1:1 mouse-to-human orthology.

Figure 2. Despite binding motif divergence and general transcriptome divergence, DUX4 transcriptome in mouse muscle cells is enriched for the 2C-like gene signature

(a) Dux and DUX4 binding motifs as derived de novo from ChIP-seq peaks using MEME algorithm. DUX4 ChIP-seq data was previously published, but re-analyzed using the methods of this study. Note the divergence in the first half of the motif and the conservation of the second half of the motif. E-values listed reflect an estimate of the expected number of motifs, with the given motif’s log likelihood ratio (or higher) and with the same width and site count, that one would find in a similarly sized set of random sequences (where each position in each sequence is independent and letters are chosen according to the background letter frequencies). Histogram to the right shows that 578 peaks out of the 600 used to generate the Dux motif carry a match to the motif and that the motifs are centrally located within each ChIP-seq peak. DUX4 histogram is also shown. (b) GSEA: gene set is the mouse 2C-like gene signature, x-axis is the log2FoldChange-ranked DUX4 transcriptome in mouse cells. Since the mouse 2C-like gene signature and this DUX4 transcriptome were both identified in mouse cells, neither gene set nor transcriptome was limited to genes with 1:1 mouse-to-human orthology.

Figure 3. Dux, but not DUX4, activates transcription of repetitive elements characteristic of the early embryo in mouse muscle cells

(a) Expression levels of repeats during Dux expression in mouse cells compared to un-induced cells of the same cell line, broken down by repeat class. For LTR elements broken down by family, see Supplementary Figure 6a–c. Each dot is a repeatName as defined by RepeatMasker. Red color indicates differential expression at absolute(log2-Foldchange)>=1 and adjusted p-value<=0.05. Number in parentheses is log2-FoldChange. (b) Same as (a) for DUX4-expressing mouse muscle cells compared to un-induced cells of the same cell line. (c) Same as (a) for DUX4-expressing human muscle cells compared to un-induced cells of the same cell line, data previously published. (d) Example of a Dux ChIP-seq peak in MERV-L (MT2-element in RepBase nomenclature). Track height is 200 reads for all tracks. mm10 genome location is chr15:52,742,953–52,744,319. (e) Luciferase assay comparing the activation of a 2C-active MERV-L element reporter by either Dux, DUX4 or an empty vector. The MERV-L element contains a match to the Dux motif and was mutated as shown in cartoon to the right and the full sequence is in Supplementary Figure 6d. Activation of the mutated MERV-L reporter is also shown. Data shown are mean fold change over empty vector of 3 cell cultures prepared in parallel for each condition. Error bars are s.e.m. The non-mutated MERV-L reporter activation experiment was repeated on three separate occasions with consistent results. The mutated MERV-L reporter experiment was performed on one occasion.

Figure 4. Dux and DUX4 use different types of LTR elements as alternative promoters for protein-coding genes

(a) Histogram where black bars are counts of genes in the 2C-like signature that are MERV-L-promoted and activated by the indicated factor. White bars are genes detected by RNA-seq, but are not upregulated compared to control samples. Gray bars are genes with no reads by RNAseq. MERV-L promoted genes for this plot were determined by presence of an MT2-type element that overlaps the annotated TSS of a gene in the published 2C-like gene signature. (b) Histogram showing the number of genes in the 2C-like signature where the indicated factor bound a MERV-L (MT2-type) element based on ChIP-seq data and there was at least one RNA-seq read that connected the ChIP-seq peak range to an annotated exon in mouse muscle cells, termed “Peak-Associated Genes” (PAGs). Cartoon depiction of PAGs that overlap MERV-Ls is to the right. For two examples of PAGs that start in MERV-L (MT2-type) elements, see Supplementary Figure 7a–b. (c) LTR-family distribution of PAGs that overlap any LTRs (CHIP-seq peak in an LTR with at least one RNA-seq read that connects the element to an annotated exon). Note that although Dux and DUX4 both have PAGs that start in ERVL-MaLRs, they are predominantly different ERVL-MaLRs (only 1/31 DUX4_PAGs in ERVL-MaLRs was also identified as a Dux_PAG). (d) Two examples of DUX4 binding an LTR to induce novel transcription. LTR element = gray box. Track height in reads is given in brackets below each browser shot.

Figure 5. Transcriptional divergence between DUX4 and Dux maps to the two DNA-binding homeodomains

(a) Cartoons of chimeric proteins; MMH is the two Dux homeodomains and the DUX4 C-terminus; MHM is Dux with HD2 from DUX4; HMM is Dux with HD1 from DUX4. (b–d) RT-qPCR data for 2C-like genes in mouse muscle cells of various classes, defined below. Data shown are mean of 3 separate cell cultures for each condition with s.e.m. error bars. The experiments in (b) and (d) were also repeated on three separate days and showed consistent results. The experiments in (c) were completed on one occasion. (b) 2C-like genes with MERV-L promoters (c) 2C-like genes with conventional promoters that are induced by DUX4 and Dux (d) 2C-like genes with conventional promoters that are induced only by Dux (e) Cartoons of reciprocal set of chimeric proteins; HHM is the two DUX4 homeodomains and the Dux C-terminus; HMH is DUX4 with HD2 from Dux; MHH is DUX4 with HD1 from Dux. (f) RT-qPCR data for DUX4-target genes in human rhabdomyosarcoma cells. Data shown are mean of 3 separate cell cultures for each condition with s.e.m. error bars. These experiments were completed on one occasion.