Spatially coordinated heterochromatinization of long synaptic genes in fragile X syndrome

Abstract

Short tandem repeat (STR) instability causes transcriptional silencing in several repeat expansion disorders. In fragile X syndrome (FXS), mutation-length expansion of a CGG STR represses FMR1 via local DNA methylation. Here, we find megabase-scale H3K9me3 domains on autosomes and encompassing FMR1 on the X chromosome in FXS patient-derived iPSCs, iPSC-derived neural progenitors, EBV-transformed lymphoblasts, and brain tissue with mutation-length CGG expansion. H3K9me3 domains connect via inter-chromosomal interactions and demarcate severe misfolding of TADs and loops. They harbor long synaptic genes replicating at the end of S phase, replication-stress-induced double-strand breaks, and STRs prone to stepwise somatic instability. CRISPR engineering of the mutation-length CGG to premutation length reverses H3K9me3 on the X chromosome and multiple autosomes, refolds TADs, and restores gene expression. H3K9me3 domains can also arise in normal-length iPSCs created with perturbations linked to genome instability, suggesting their relevance beyond FXS. Our results reveal Mb-scale heterochromatinization and trans interactions among loci susceptible to instability.

Keywords: CRISPR; DNA FISH; Hi-C; chromatin; epigenetics; fragile X syndrome; heterochromatin; higher-order genome folding; repeat expansion disorders; short tandem repeats; topologically associating domains.

Figure 1.. A Megabase-sized H3K9me3 domain spreads upstream of the FMR1 locus in iPSC-derived NPCs and post-mortem caudate nucleus brain tissue from FXS patients.

(A) Schematic of iPSC lines used to model FMR1 CGG expansion in FXS, including normal-length (NL), premutation-length (PM), and mutation-length (FXS). (B) Representative Nanopore long-reads across the FMR1 5’UTR. Colors reflect nucleotides (orange: A, blue: T, green: C, red: G, dark green: CGG). (C) Number of CGG triplets in the FMR1 5’UTR from Nanopore long-reads. (D) FMR1 mRNA levels normalized to GAPDH via qRT-PCR. Horizontal line, mean n=2 biological replicates. (E) Proportion of 19 CpG dinucleotides methylated in the 500 bp FMR1 promoter computed from Nanopore long-reads. (F) Proportion of CGG triplets methylated within the 5’ UTR STR using STRique. Each dot, one allele. (G) Hi-C and ChIP-seq in iPSC-NPCs across a 5Mb region around FMR1. (H) Hi-C fold-change interaction frequency maps. Gained and lost contacts compared to NL_18 highlighted in red and blue, respectively. (I) SLITRK2 and SLITRK4 mRNA levels via RNA-seq. Horizontal lines, mean n=2 biological replicates. (J) H3K9me3 CUT&RUN in brain tissue from N=2 FXS patients with sex- and age-matched N=2 normal-length individuals.

Figure 2.. Heterochromatin domains and synaptic gene silencing on autosomes in FXS patient-derived iPSC-NPCs and brain tissue.

(A) Two classes of autosomal H3K9me3 domains (i) FXS-recurrent: consistently gained in all three FXS iPSC-NPCs and not in NL/PM iPSC-NPCs or (ii) Genotype-invariant: present in NL/PM/FXS iPSC-NPCs. (B) Hi-C and ChIP-seq for a 3.5 Mb region around a H3K9me3 domain encompassing DPP6. (C) DPP6 mRNA levels via RNA-seq. Horizontal lines, mean n=2 biological replicates. (D) Average H3K9me3 and CTCF ChIP-seq signal across autosomal FXS-recurrent H3K9me3 domains. (E) Boundary strength in NL_18 and FXS_426 iPSC-NPCs for one TAD boundary per autosomal FXS-recurrent H3K9me3 domain. (F) mRNA levels via RNA-seq for N=25 expressed protein-coding genes in autosomal and chrX FXS-recurrent H3K9me3 domains. Each point, mean per gene n=2 biological replicates. P-values, one-tailed MWU, where * P-value <0.05 versus NL_18. (G) Gene ontology for all N=36 protein-coding genes in autosomal and chrX FXS-recurrent H3K9me3 domains. (H) Expression of N=54 coding/noncoding genes in FXS-recurrent H3K9me3 domains across GTEX tissues. (I) Number of autosomal H3K9me3 domains arising in FXS patient-derived brain tissue compared to sex- and age-matched normal-length control tissue. (J) H3K9me3 CUT&RUN in brain tissue from N=2 FXS patients with sex- and age-matched N=2 normal-length individuals at DPP6, RBFOX1, and CSMD1.

Figure 3.. Engineering the mutation-length FMR1 CGG STR to premutation-length attenuates a subset of H3K9me3 domains and de-represses gene expression.

(A) Schematic of N=7 mutation-length and premutation-length single-cell-derived CGG CRISPR cut-back iPSC clones generated from the FXS_421 parent iPSC line. (B) FMR1 mRNA levels normalized to GAPDH and shown relative to FXS_421 using qRT-PCR. Error bars, standard deviation n=2 biological replicates. (C) Number of CGG triplets in the FMR1 5’UTR computed from Nanopore long-reads. (D) Average input normalized H3K9me3 signal for the chrX FXS-recurrent H3K9me3 domain. Dots represent equal sized bins (N=5) across the domain. (E) FXS-recurrent H3K9me3 domains amenable (red) and refractory (black) to reprogramming. For each domain, we measured the fraction of iPSC clones with persistent, lowered, or removed H3K9me3 signal for all mutation-length (N=7) and premutation-length (N=7) clones. (F) Hi-C and ChIP-seq for a 5 Mb region around FMR1 in FXS_421 and PMcut_scClone1 iPSCs. (G) Log2 fold change of gene expression in FXS_421 vs. PMcut_scClone1 with respect to NL_18. Each dot, one gene. P-values, one-tailed MWU.

Figure 4.. Autosomal heterochromatin domains spatially connect with FMR1 via inter-chromosomal interactions in FXS.

(A) Trans interactions between each of the N=10 FXS-recurrent H3K9me3 domain on autosomes and FMR1 on chrX. (B) Hi-C inter-chromosomal interaction heatmaps binned at 1 Mb resolution. Green arrows, trans interactions. (C-D) Hi-C inter-chromosomal interactions among FXS-recurrent H3K9me3 domains (C) FXS_426 (upper triangle) versus NL_18 (lower triangle) iPSC-NPCs and (D) FXS_421 (upper triangle) versus PMcut_scClone1 (lower triangle) iPSCs. H3K9me3 ChIP-seq signal plotted above Hi-C heatmaps. Blue boxes, FXS-gained trans interactions. Green boxes, attenuated trans interactions after premutation-length cutback. (E+H) DNA FISH images for the H3K9me3 domain on chrX interacting with (E) the chr12 domain or (H) all domains in NL_18, FXS_421, and PMcut_scClone1 iPSC nuclei. Scale bars, 10 μm. (F-G) Distances between chrX and chr12 H3K9me3 domains in iPSCs, including (F) proportion of measurements stratified by distance and (G) measurements directly compared with a two-tailed MWU, where * P-value <1E-6. (I) Average distance per cell between the chrX and all other FXS-recurrent H3K9me3 domains. (J) Kernel density estimation of the number of foci per nucleus. (I-J) Two-tailed MWU, where * P-value <1E-12.

Figure 5.. Autosomal H3K9me3 domains are enriched for late replicating long synaptic genes and replication stress-induced double strand breaks.

(A-H) Empirical randomization test assessing the enrichment of (A+E) gene density, (B+F) gene length, (C+G) replication timing, and (D+H) replication stress-induced double stranded breaks in (A-D) FXS-recurrent H3K9me3 domains or (E-H) genotype-invariant H3K9me3 domains compared to N=1000 draws of random genomic intervals matched by size. (I) FXS-recurrent H3K9me3 domains encompassing CSMD1 (gene length: ~2.10 Mb), DPP6 (gene length: ~1.15 Mb), PTPRT (gene length: ~1.16 Mb), and RBFOX1 (gene length: ~2.47 Mb). Replication stress-induced double strand breaks, dark green. Replication timing, yellow (early S phase) and black (late S phase). (J) Empirical randomization test assessing the enrichment of CGG tracts (>=CGGx3) in TSSs + 2kb within FXS-recurrent H3K9me3 domains compared to N=1000 draws of random genomic intervals matched by size. (K) Examples of CGG tracts in FXS-recurrent H3K9me3 domains encompassing DPP6 and TCERG1L.

Figure 6.. Autosomal FXS-recurrent H3K9me3 domains can harbor STR tracts prone to stepwise somatic instability.

(A) Schematic depicting the pipeline for identifying candidate long STRs with potential for somatic instability using GangSTR and ExpansionHunter. (B) Venn diagram depicting “FXS long STRs” identified in FXS iPSCs as significantly longer than expected in N=120 ancestry-, sex-, sequencing depth-, and cell type-matched normal-length individuals. (C) Stratification of “FXS long STRs” into those exhibiting patterns potentially consistent with somatic instability (green: >=3 alleles per FXS iPSC line per STR) and those that do not (orange: somatically stable). (D) Empirical randomization test assessing the enrichment of FXS-reproducible stepwise somatically unstable STRs in FXS-recurrent H3K9me3 domains compared to N=1000 draws of random genomic intervals matched by size. (E) Distribution of STR tract length (bp) across N=240 alleles of ancestry-, sex-, sequencing depth-, and cell type-matched normal-length HipSci iPSC lines. Overlayed blue dashed lines indicate the maximum STR length in each of the three FXS iPSC lines. Empirical one-tailed P-value. Distributions shown for “FXS long STRs” in RBFOX1 (left) and an intergenic region on chr5 (right). (F) Representative reads for direct visualization of stepwise STR expansion events in short-reads across all 3 FXS iPSC lines as well as verified in FXS_421 with Nanopore long-reads (top). STR lengths computed directly from reads via the CIGAR string (bottom).

Figure 7.. Specific normal-length iPSC lines made with p53 perturbation exhibit an intermediate level of H3K9me3 signal at BREACHes.

(A) Venn diagram showing the overlap between the genes localized with BREACHes from this study and down-regulated genes in Fmr1 knock-out mouse cortical neurons. (B-C) RNA-seq comparing expression of BREACH-localized genes in normal-length versus Fmr1 knock-out neurons. (D) Venn diagram showing reproducibly down-regulated genes (n=38) in mutation-length FXS compared to normal-length and premutation iPSC-NPCs. Red genes localize with BREACHes. Blue genes are linked to the DNA damage response. (E) Gene ontology for reproducibly down-regulated genes (n=34) not present in BREACHes. (F-H) Genomic features at BREACHes in normal-length iPSCs (red) and FXS iPSCs from this study derived without p53 shRNA (blue), as well as two prototypic iPSC lines derived with p53 shRNA (grey). (F) H3K9me3, (G) trans interaction frequency, and (H) summed burden of STR instability. (I) STR length computed directly from reads via the CIGAR string for an AAAT tract on chr5. (J) Schematic model of BREACHEs – Beacons of Repeat Expansion Anchored by Contacting Heterochromatin.