PRDM9 drives evolutionary erosion of hotspots in Mus musculus through haplotype-specific initiation of meiotic recombination

Abstract

Meiotic recombination generates new genetic variation and assures the proper segregation of chromosomes in gametes. PRDM9, a zinc finger protein with histone methyltransferase activity, initiates meiotic recombination by binding DNA at recombination hotspots and directing the position of DNA double-strand breaks (DSB). The DSB repair mechanism suggests that hotspots should eventually self-destruct, yet genome-wide recombination levels remain constant, a conundrum known as the hotspot paradox. To test if PRDM9 drives this evolutionary erosion, we measured activity of the Prdm9Cst allele in two Mus musculus subspecies, M.m. castaneus, in which Prdm9Cst arose, and M.m. domesticus, into which Prdm9Cst was introduced experimentally. Comparing these two strains, we find that haplotype differences at hotspots lead to qualitative and quantitative changes in PRDM9 binding and activity. Using Mus spretus as an outlier, we found most variants affecting PRDM9Cst binding arose and were fixed in M.m. castaneus, suppressing hotspot activity. Furthermore, M.m. castaneus×M.m. domesticus F1 hybrids exhibit novel hotspots, with large haplotype biases in both PRDM9 binding and chromatin modification. These novel hotspots represent sites of historic evolutionary erosion that become activated in hybrids due to crosstalk between one parent's Prdm9 allele and the opposite parent's chromosome. Together these data support a model where haplotype-specific PRDM9 binding directs biased gene conversion at hotspots, ultimately leading to hotspot erosion.

Figure 1. PRDM9 activity is dependent on genetic background.

(A) H3K4me3 coverage profile from a representative 150 kb window of chromosome 1. H3K4me3 peaks can be allele specific (red boxes - Prdm9^Cst, blue box - Prdm9^Dom2) or novel (black box - Prdm9^Cst hotspot active on the B6 background). (B) Heat map for shared KI and CAST hotspots with significant differences (n = 2,740, FDR<0.01). (C) Top – H3K4me3 profile for representative PRDM9^Cst hotspots with quantitative differences (left) or novel (right). Below - DNA sequence showing SNPs (red) within the hotspot NDR (highlight - PRDM9^Cst motif). (D) SNP frequency between the B6 and CAST haplotypes at PRDM9^Cst hotspots (10 bp running average normalized to background, yellow highlight - PRDM9^Cst motif). Hotspots with differences in H3K4me3 have increased SNP frequency (black –novel hotspots in KI but not CAST, n = 2,035; green – quantitative hotspots from 1B, n = 2,399), while hotspots with similar H3K4me3 level have low SNP frequency (blue – log fold change <0.5 and >−0.5, n = 5,645).

Figure 2. Novel hotspots have biased H3K4me3 modification.

(A) F1 hybrid hotspots were classified as B6 or CAST depending on parental origin, or labeled novel. (B) H3K4me3 haplotype ratio for BxC TSS (n = 12,903, orange - autosomes, black - chromosome X). (C) H3K4me3 haplotype ratio for BxC hotspots (grey - all BxC hotspots, n = 12,271; green - novel BxC hotspots, n = 2,298). (D) Scatterplot of haplotype-specific H3K4me3 for hotspots shared between progeny from reciprocal B6 and CAST crosses (n = 10,977, r = 0.978 without X chromosome). 163 hotspots in lower right (circle) are all on the chromosome X. Black shading reflects signal density. (E) Left - Haplotype-specific PCR showing from genomic DNA samples for hotspots chr1 185.33 Mb and chr1 171.37 Mb. Right - Haplotype-specific PCR from spermatocytes before or after enrichment for H3K4me3 (n = 4, error bars - S.D.). (F) The parental identity and fraction of hotpots in WxP F1 hybrids. (G) H3K4me3 haplotype ratio for WxP TSS (n = 15,856, orange - autosomes, black - chromosome X). (H) H3K4me3 haplotype ratio for WxP hotspots (grey - all WxP hotspots, n = 8,360; green - novel WxP hotspots, n = 2,325).

Figure 3. Novel hotspots are the sites of historic hotspot erosion.

(A) SNP frequency between B6 and CAST haplotypes for PRDM9^Dom2-dependent novel hotspots with haplotype ratio <0.2 (n = 810, points - value at each nucleotide, line - 10 bp running average, highlight - PRDM9^Dom2 motif). (B) Top – CAST haplotype-derived PRDM9^Dom2 motif for hotspots in A. Middle – Amino acid sequence for PRDM9^Dom2 zinc-finger array positions predicted to contact DNA. Bottom – Number of SNPs identified at each base pair position (background subtracted). (C) SNP frequency between B6 and SPRET (blue) or CAST and SPRET (red) for PRDM9^Dom2 novel hotspots in A. (D) SNP frequency between B6 and CAST haplotypes at hotspots unique to the KI strain (n = 2,035, hotspot represented by black line in Fig. 1D). Top – PRDM9^Cst motif derived from the B6 haplotype. Middle – Amino acid sequence for PRDM9^Cst zinc-finger array. Bottom – Number of SNPs identified at each base pair position (background subtracted). (E) Similar to C except comparing SNP frequency between B6 and SPRET (blue) or CAST and SPRET (red) for hotspots used in D. (F) Similar to E for PRDM9^Cst hotspots quantitatively higher in the KI strain compared to CAST (hotspots from Fig. 1D green line, n = 2,187).

Figure 4. Hotspot haplotype influences PRDM9 binding.

(A) H3K4me3 coverage profile from BxC F1 hybrid showing DNA sequences for both B6 and CAST haplotypes and PRDM9^Dom2 motif (red bars – sequences bound by PRDM9^Dom2, black bars - nonbinding sequences, yellow - PRDM9^Dom2 motif). (B) EMSA assay shows specific binding of PRDM9^Dom2 at the novel hotspot chr1 185.33 Mb (lanes 2 vs. 3). Unlabeled DNA oligos 1 and 2, both containing the PRDM9^Dom2 consensus motif, can compete for binding (Lanes 4 and 5). All lanes contain labeled DNA of the CAST haplotype otherwise the composition of the binding reactions is shown above the blot (B – PRDM9^Dom2, C – PRDM9^Cst). (C) EMSA assay shows PRDM9^Dom2 can bind to both B6 and CAST haplotypes (Lanes 2 and 6) although the CAST haplotype is preferentially bound under conditions of competition (Lanes 7 and 8, B - B6 haplotype, C - CAST haplotype).

Figure 5. PRDM9 binding shows large haplotype bias in vivo.

(A) Coverage profile for several representative hotspots for both PRDM9 ChIP-seq and H3K4me3 ChIP-seq in BxC F1 hybrids. (B) 96% of BxC F1 PRDM9 peaks overlap with BxC F1 H3K4me3. (C) Parental origin of BxC PRDM9 peaks. (D) Haplotype-specific heat map of ChIP-seq signal for PRDM9 and H3K4me3 from BxC F1 hybrids. Read densities from a 2 kb window centered on the summit of PRDM9 peaks were clustered. SNPs density between B6 and CAST strains across the same 2 kb were arranged in the order of hotspots. The activating PRDM9 protein variant, and whether the hotspot was identified as novel, is indicated by color bar on the right. (E) Aggregation plots of PRDM9 and H3K4me3 signal. (F) Scatterplot of haplotype-specific PRDM9 binding and haplotype-specific H3K4me3.

Figure 6. The novel hotspot at chr1 40.6 Mb, activated by PRDM9^Cst on the B6 haplotype, shows transmission distortion to the CAST haplotype.

(A) PRDM9^Cst motif aligned to the DNA sequence for both B6 and CAST haplotypes at the peak of PRDM9 ChIP-seq signal (B) PRDM9 and H3K4me3 coverage profile at chr1 40.6 Mb (black lines – SNP positions between B6 and CAST, black box – PRDM9^Cst motif). (C) Proportion of mapped reads to each haplotype (unassigned reads lack informative SNPs). (D) Cumulative frequency of recombination crossovers at hotspot chr1 40.6 Mb (n = 19 mice). (E) Transmission distortion at hotspot chr1 40.6 Mb favors the CAST haplotype at the PRDM9 binding site.

Figure 7. Model for the evolution of Prdm9 alleles and hotspot erosion.

(A) Predicted timeline for the origin of Prdm9 alleles based on SNP frequency found at hotspots comparing B6 and CAST to SPRET. PRDM9^Cst hotspots show an increase in SNPs in the B6 background, suggesting this allele was active in a shared lineage between M.m. castaneus and M.m. domesticus. PRDM9^Dom2 hotspots do not have increased SNPs in the CAST background suggesting this allele was never active in M.m. castaneus. (B) The PRDM9/hotspot lifecycle. Evolutionary erosion driven by biased PRDM9 initiation of recombination decreases hotspot activity over time at many hotspots in parallel. Mutation of Prdm9 creates a new binding domain subsequently shifting the genome-wide position of hotspots.