Fig 2: Insertion of an ERV element in place of A-repeat restores Spen binding and local Xist silencing.(a) Diagram showing the experimental rationale for replacing the A-Repeat with a 9x tandem insert of an ERV-derived Spen binding site. (b) Xist-ERV restores Spen recruitment. RIP-qPCR for Xist and ActB in Xist-WT, Xist-?A, and Xist-ERV clones 1 and 2 mESCs. qRT-PCR values are normalized to input samples. Error bars represent standard error for four independent biological replicates. p values are derived from two-tailed t-test. (c) Xist-ERV causes epigenetic silencing of the Xist locus. Xist expression level in +Dox compared to –Dox conditions for Xist-?A mESCs and Xist-ERV mESCs with and without HDAC1/2 inhibitor treatment. Xist levels are first normalized to ActB and then to –Dox levels. The red line is drawn at y = 0, representing no increase in expression of Xist with the addition of doxycycline. For each condition, four independent experiments were conducted. p values are derived from two-tailed T test. (d) Xist-ERV silencing requires Spen and HDAC3. Xist expression level in +Dox compared to –Dox conditions for Xist-?A mESCs and Xist-ERV mESCs with and without siRNA-mediated knockdown of Spen or siRNA-mediated knockdown of Spen + HDAC3 inhibition. For each condition, four independent experiments were conducted. (e) Model for how Spen evolved to bind Xist through its ERV binding mechanism in eutherian mammals.

Fig 3: Spen’s RNA binding domains bind specifically to ERV RNAs in vivo.(a) Diagram of irCLIP-seq rationale. irCLIP-seq was performed for the Spen RNA binding domains to identify direct RNA binding partners of Spen genome-wide. (b) Pie chart showing the number of irCLIP-seq clusteres called on TE RNAs and mRNAs and lncRNAs. (c) Pie chart showing the number of RNA transcripts containing at least one irCLIP-seq cluster in TEs and mRNAs and lncRNAs. (d) Binding strength of RRM2-4 for the top 50 Spen-bound RNAs. Xist (#1) and Spen (#204) are highlighted in red. (e) Tracks showing the irCLIP-seq signal across the A-repeats of Xist. Tracks are shown for Spen KO +RRM2-4 (+ / - Xist), Spen full-length, and V6.5 male mESCs. irCLIP-seq clusters are shown in black at the top. (f) Tracks showing the irCLIP-seq signal across exon 2 of Spen. Tracks are shown for Spen KO +RRM2-4 (+ / - Xist), Spen full-length, and an IgG control. irCLIP-seq clusters are shown in black at the top. (g) Graph showing the number of bound TE loci for each TE subfamily with significant Spen RRM2-4 binding. (h) Graph showing the percentage of TE loci bound for each TE subfamily with significant Spen RRM2-4 binding. (i) Expression of RRM-bound LTRs in WT mESCs (green) and Spen KO mESCs (orange). Shown is the log2 transformed normalized RNA-seq read counts. For all boxplots, the thick line represents the median, while the box gives the IQR. P values are calculated using a two-tailed T test. (j) Same as in i for RRM-bound ERVKs. (k) Same as in i for RRM-bound MMETns.

Fig 4: Spen knockout blocks XCI chromosome-wide and leads to derepression of autosomal ERVs.(a) Diagram of experimental set up. Spen KO mESCs are cultured +/- Doxycycline for 48 hr to induce Xist expression and then ATAC-seq is performed. We ask which sites are de-repressed in Spen KO compared to WT cells both during XCI and on the autosomes. (b) Log2 ratio of ATAC-seq signal in +Dox and –Dox samples at all ATAC-seq peaks on the X chromosome in WT and Spen KO mESCs. Two independent Spen KO mESC clones are shown. (c) Same as in b for all autosomal peaks. (d) Heatmap showing ATAC-seq signal for all differential peaks between WT and Spen KO mESCs. Values depict the z-scored value for normalized ATAC-seq reads within each peak for WT, Spen KO Cl2, and Spen KO Cl3 cells (with two replicates each). (e) Pie charts depicting the proportion of ATAC-seq peaks that cover each genomic region annotation. Annotations are derived from HOMER. (f) Genome Ontology enrichment (by HOMER) for all sites gaining accessibility in Spen KO mESCs compared to WT (n = 288). Plotted is the –log10 p-value for enrichment. P values (one-sided binomial) are corrected for multiple hypothesis testing using the Benjamini-Hochberg correction. (g) Two example ERVK regions on chromosome 16 (left) and chromosome 1 (right) showing a gain in chromatin accessibility specifically in two clones of Spen KO mESCs. Shown are tracks of uniquely mapping reads (top) and tracks where multimapping reads are randomly assigned to one location (bottom).

Fig 5: Spen RRMs recognize a common structural feature in A-Repeat and ERV RNA.(a) irCLIP-seq signal from full-length Spen (Kahles et al., 2016) across the 8 Xist A-repeats. Signal for each repeat is stacked across the conserved repeat region. (b) Same as in a for RRM2-4 irCLIP-seq. (c) icSHAPE reactivity scores for each of the Xist A-repeats performed in vitro. Signal for each repeat is stacked across the conserved repeat region. (d) Phylo-P scores across the A-repeats, showing conservation of repeat sequence at the RRM2-4 binding sites. (e) Structural model for Xist A-repeats 4 and 5 which form a hairpin structure. Structural prediction is supported by icSHAPE data. At each nucleotide, the RRM2-4 irCLIP-seq signal strength is plotted as a bar for each repeat (in color). (f) irCLIP-seq signal (top, black) and icSHAPE reactivity scores (bottom, gray) for the consensus sequence of RLTR13G elements. Included are elements for which we have both confident RRM2-4 binding and icSHAPE data. (g) Same as in f for ETnERV. (h) Same as in f for MMETn-int. (i) Structural model for ETnERV. Structural prediction is supported by icSHAPE data. At each nucleotide, the RRM2-4 irCLIP-seq signal strength is plotted as a bar.