Epigenetic activation of meiotic recombination near Arabidopsis thaliana centromeres via loss of H3K9me2 and non-CG DNA methylation

1Eukaryotic centromeres are the sites of kinetochore attachment to spindle microtubules that allow chromosome segregation. Centromere identity is governed by nucleosomes containing CENPA/CENH3-related histone variants, which occupy large arrays of tandemly repeated satellite sequences. A conserved feature of centromeres shared across eukaryotes is suppression of meiotic crossover. Crossover suppression is important for fertility, as centromere-proximal recombination events have been associated with chromosome segregation errors and aneuploidy. Plants, animals, and fungi also typically possess repetitive pericentromeric heterochromatin, containing a high density of transposable elements. In crop genomes, including wheat, barley, maize, and tomato, extensive pericentromeric heterochromatin occupies more than half of the chromosome and is also crossover-suppressed. However, the genetic and epigenetic factors that shape meiotic recombination patterns in eukaryote centromeric and pericentromeric regions remain to be fully understood.

2Arabidopsis thaliana centromeres consist of megabase tandem arrays of the 178- to 180-base pair CEN180 satellite repeat. The centromeric regions are also densely DNA methylated and enriched for H3K9me2 and histone variant H2A.W. Within the centromeric satellite arrays, a subset of repeats are occupied by nucleosomes containing the centromeric variant histone H3 CENH3. Surrounding the CEN180 satellite arrays are repetitive, transposon-dense regions of pericentromeric heterochromatin. Plant transposable elements are transcriptionally silenced by H3K9me2 and DNA cytosine methylation in CG and non-CG (CHG and CHH, where H = A, C, or T) sequence contexts. A. thaliana mutants that lose maintenance of CG or non-CG DNA methylation show elevated transposon transcription and mobility at high and moderate levels, respectively.

3In A. thaliana, the chromodomain cytosine methyltransferases CHROMOMETHYLASE2 (CMT2) and CHROMOMETHYLASE3 (CMT3) recognize heterochromatic H3K9me2 via BAH and chromodomains, and methylate-associated DNA in CHH and CHG contexts, respectively. Methylation of histone H3K9 requires the SET domain methyltransferases KRYPTONITE/SUPPRESSOR OF VARIEGATION HOMOLOG4 (KYP/SUVH4), SUPPRESSOR OF VARIEGATION HOMOLOG5 (SUVH5), and SUPPRESSOR OF VARIEGATION HOMOLOG6 (SUVH6), which are recruited to methylated DNA by SRA methyl-cytosine binding domains. The de novo DNA methyltransferase DOMAINS REARRANGED METHYLTRANSFERASE2 (DRM2) is also required for maintenance methylation of non-CG contexts and thus can also affect H3K9me2. The kyp suvh5 suvh6 mutant abolishes almost all H3K9me2 and CHG/CHH methylation, but CG methylation remains intact, while cmt3 mutants lose CHG methylation and H3K9me2 is reduced, but CHH methylation and CG methylation are largely unchanged. In contrast to plants, in fission yeast, which lacks DNA methylation, the SET domain histone lysine methyltransferase CRYPTIC LOCI REGULATOR4 (CLR4) is recruited to methylated H3K9me2 directly via its chromodomain. Thus, by separating chromodomains (CMT2 and CMT3) from SET domains (KYP/SUVH5/SUVH6), plants have introduced non-CG DNA methylation as an additional layer of epigenetic control underlying H3K9me2. Alongside these mechanisms, the METHYLTRANSERFERASE1 (MET1) cytosine methyltransferase, VARIANT IN METHYLATION1 (VIM1) family proteins, and the DECREASE IN DNA METHYLATION1 (DDM1) SWI/SNF chromatin remodeling protein are required for maintenance of CG context DNA methylation. In this work, we use mutations in A. thaliana heterochromatic silencing pathways to investigate epigenetic control of meiotic recombination in the pericentromeric regions.

4Meiotic crossovers form via inter-homolog repair of DNA double-strand breaks (DSBs) that are generated by SPO11 topoisomerase-related complexes. Diverse eukaryotes have evidence for recombination hotspots, which are approximately kilobase-size regions with an elevated frequency of meiotic DSBs or crossovers, compared to the genome average or surrounding regions. Hotspots in different eukaryotic lineages are controlled to varying degrees by genetic and epigenetic information. At the chromosome-scale, A. thaliana crossover frequency is highest in gene-dense euchromatin, whereas the heterochromatic centromeres are crossover-suppressed. At the fine-scale, plant crossover hotspots occur at gene promoters and terminators, and recombination is promoted by euchromatic modifications, including histone variant H2A.Z. Acquisition of DNA methylation and H3K9me2 via the RNA-directed DNA methylation (RdDM) pathway is sufficient to silence A. thaliana euchromatic crossover hotpots. This is consistent with DNA methylation suppressing meiotic DNA double-strand breaks in mouse, and silencing crossovers in Ascobolus. Furthermore, loss of RNAi and Clr4-dependent H3K9me2 elevates centromeric crossovers in fission yeast, and Drosophila position effect variegation (PEV) suppressor mutations (Suppressor of Variegation) can modify centromeric crossover frequency. Interestingly, the A. thaliana met1 and ddm1 CG methylation mutants associate with remodeling of meiotic recombination along chromosomes, with crossover increases in the chromosome arms and decreases across the pericentromeres. However, how non-CG DNA methylation and other epigenetic silencing pathways contribute to recombination landscapes along plant chromosomes has not been fully explored. In this study, we address the roles of H3K9me2 and non-CG DNA methylation in suppression of meiotic DSBs and crossovers within A. thaliana pericentromeric heterochromatin.

5To investigate meiotic recombination frequency in A. thaliana non-CG DNA methylation and H3K9me2 mutants, we used fluorescent crossover reporter lines (fluorescent tagged lines, FTLs). FTLs express different colors of fluorescent protein under seed (NapA)- or pollen (LAT52)-specific promoters, from linked T-DNAs insertions (Fig. 1A). The scoring of fluorescent color inheritance in the progeny seed or pollen (male gametes) of FTL hemizygotes allows the measurement of sex-averaged or male-specific crossover frequency, respectively, in defined chromosomal intervals (Fig. 1A;;; ).

6We first analyzed crossover frequency within the 5.4-megabase (Mb) CEN3 FTL interval, which spans the centromere and pericentromeric heterochromatin of Chromosome 3, in wild type versus non-CG/H3K9me2 pathway mutants (Fig. 1B; ). Genetic ablation of the H3K9 methyltransferases (kyp suvh5 suvh6) or the non-CG DNA methyltransferases (drm1 drm2 cmt2 cmt3) eliminates both H3K9me2 and non-CG DNA methylation, while single and double mutants have intermediate effects. We observed that mutations that disrupt H3K9me2 and non-CG DNA methylation to progressively greater extents resulted in progressively greater increases in CEN3 crossover frequency (drm1 drm2 < kyp < cmt3 < drm1 drm2 cmt3 < kyp suvh5 suvh6; all χ2 P < 2.0 × 10−16) (Fig. 1C; Supplemental Table S1;, ). The suvh5 suvh6 and cmt2 mutants did not show significant differences compared to wild type (Fig. 1C; Supplemental Table S1).

7We used published bisulfite sequencing data to analyze DNA methylation levels within the CEN3 interval in the genotypes analyzed for crossover frequency. Within this series of mutants, levels of CHG DNA methylation showed a strong negative correlation with CEN3 genetic distance (Pearson's r = −0.93, P = 2.36 × 10−3), whereas CG and CHH methylation were not significantly correlated (Fig. 1D,E; Supplemental Table S2; ). In contrast to non-CG/H3K9me2 mutants, heterozygous ddm1/+, met1/+, or homozygous met1 mutants, which inherit chromosomes hypomethylated in the CG context, have reduced CEN3 recombination, as reported previously (Fig. 1C; Supplemental Table S1;;;;, ). We confirmed release of crossover suppression in cmt3 across the Chromosome 5 centromere and pericentromeric heterochromatin, using additional FTLs in Col (CTL5.11, χ2 P = 1.30 × 10−3) and Ler (LTL5.4, χ2 P = 2.09 × 10−9) inbred backgrounds (Supplemental Fig. S1; Supplemental Tables S3, S4). We also crossed cmt3 alleles in Col (cmt3-11) and Ler (cmt3-7) accessions together to generate Col/Ler F1 progeny that were cmt3 mutant and carried the CEN3 FTL (Fig. 2A; ). We observed significantly increased CEN3 genetic distance in cmt3 hybrids (χ2 P = 1.27 × 10−86), similar to the increase observed for inbreds (Fig. 2B; Supplemental Tables S1, S5). In contrast, recombination in the euchromatic 420 FTL interval on Chromosome 3 did not significantly change in cmt3 inbreds and slightly decreased in hybrids (χ2 P = 1.96 × 10−3), compared to wild type (Fig. 2C; Supplemental Table S6). Together, these data indicate that mutations in the H3K9me2/non-CG pathway primarily activate crossover frequency in proximity to the centromeres.

8We next sought to map crossovers genome-wide in wild type compared with a H3K9me2/non-CG mutant background, using segregation of single nucleotide polymorphisms (SNPs). We used cmt3 mutant alleles in both Col (cmt3-11) and Ler (cmt3-7) backgrounds to generate wild type (Col × Ler) and cmt3 (cmt3-11 × cmt3-7) F2 populations of >700 individuals each (Fig. 2A). To assess centromeric recombination levels in these populations, we genotyped Col/Ler simple sequence length polymorphism (SSLP) markers on Chromosomes 1 and 3 (Supplemental Table S7). This confirmed significant increases in pericentromeric recombination in the cmt3 population compared to wild type, consistent with our previous FTL measurements (Supplemental Table S7). To map crossovers at high resolution, we performed genotyping by sequencing (GBS) of 437 wild-type and 384 cmt3 F2 individuals, which identified 3320 and 2803 crossovers, respectively (Fig. 2D–F; Supplemental Table S8;;; ). The crossovers were mapped between Col/Ler SNPs to a mean resolution of 887 bp. The total number of crossovers per wild-type F2 individual (mean = 7.6) was comparable to that observed in similar F2 populations and was not significantly different in cmt3 (mean = 7.3; Mann–Whitney–Wilcoxon test, P = 0.101) (Fig. 2E; Supplemental Table S8).

9To analyze crossover distributions throughout the genome, we defined centromeres as the crossover-suppressed regions that surround gaps in the chromosome assembly, the pericentromeres as the contiguous regions flanking the centromeres with higher than average DNA methylation, and chromosome arms as the remainder of the genome (Supplemental Table S9). Consistent with our FTL analysis, we observed that GBS-mapped crossovers were significantly increased in the cmt3 pericentromeric regions (24.6% versus 27.8% of events were pericentromeric in wild type versus cmt3, χ2 P = 5.60 × 10−3), which are strongly depleted of CHG DNA methylation in cmt3 (Fig 2D,F; Supplemental Table S10; ). We also observed elevated centromeric crossovers (n = 13) in cmt3 (χ2 P = 2.63 × 10−4), which were completely absent in wild type (Supplemental Table S10). The chromosome arms showed a significant decrease of crossovers in cmt3 (χ2 P = 1.50 × 10−3) (Supplemental Table S10). These data confirm that crossovers increase in proximity to cmt3 centromeres, but that the increase is strongest in the flanking pericentromeric regions (Fig. 2D,F). However, we note that some pericentromeric regions showed higher crossover frequency in wild type than cmt3—for example, the right pericentromere of Chromosome 2—and therefore, region-specific effects also likely play an important role, such as structural genetic variation.

10To assess H3K9me2 patterns during meiosis, we performed immunocytological staining using antibodies against this histone modification and the chromosome axis HORMA domain protein ASYNAPTIC1 (ASY1) (Fig. 3A; ). During meiosis, A. thaliana centromeres and pericentromeres undergo progressive clustering during prophase-I (Fig. 3A; ). We observed that the heterochromatic clusters, detected by DAPI staining of DNA, stain strongly for H3K9me2 throughout leptotene, zygotene, and pachytene (Fig. 3A), which are the key meiotic stages during which DSB formation and crossover maturation occur. At the leptotene stage, when meiotic DSBs initiate, the H3K9me2 signal was significantly reduced to background levels in kyp suvh5 suvh6 (Mann–Whitney–Wilcoxon test, P = 7.47 × 10−9). Although the mean H3K9me2 signal was lower in cmt3 compared to wild-type controls, the difference was not statistically significant (Fig. 3B,D; Supplemental Fig. S2; Supplemental Tables S11, S12), consistent with intermediate reductions in H3K9me2 observed previously in cmt3 somatic cells. Thus, H3K9me2 accumulates strongly in A. thaliana heterochromatin during meiosis, and its loss or reduction results in increased pericentromeric crossover.

11Crossovers can be detected by immunostaining for MLH1, which marks class I interfering crossover foci. Therefore, we scored MLH1 foci associated with euchromatin or heterochromatin, based on DAPI staining, in wild type and kyp suvh5 suvh6 mutants (Fig. 3C,E; Supplemental Table S13). We observed a slight but significant increase in MLH1 foci numbers in kyp suvh5 suvh6 (mean = 11.1) compared to wild type (mean = 9.4) at diakinesis (Mann–Whitney–Wilcoxon test, P = 1.64 × 10−7) (Fig. 3C,E; Supplemental Table S13). Importantly, MLH1 foci were also significantly increased in kyp suvh5 suvh6 heterochromatin (mean = 2.9), compared to wild type (mean = 1.7; Mann–Whitney–Wilcoxon test, P = 4.22 × 10−5) (Fig. 3C,E; Supplemental Table S14). This provides cytological support for our crossover mapping data and indicates that MLH1-dependent repair contributes to the increase in pericentromeric crossovers observed in H3K9me2/non-CG DNA methylation mutants.

12As MLH1 foci were increased in kyp suvh5 suvh6, we further investigated the relationship of class I and class II crossover pathways to the observed recombination changes. Approximately 85% of A. thaliana crossovers are dependent on the class I repair pathway and are interference-sensitive. Mutants in the class I pathway, for example, zip4, cause a strong reduction in crossovers and fertility. The fancm mutation restores fertility in fancm zip4 double mutants by increasing class II noninterfering crossovers. Therefore, we constructed cmt3 zip4 and cmt3 fancm double mutants and compared CEN3 crossover frequency and fertility to wild type and single mutants (Fig. 4A–C; Supplemental Tables S14, S15). Unlike fancm, cmt3 was unable to suppress zip4 infertility (Fig. 4A,B; Supplemental Table S15). However, a small but significant CEN3 crossover increase was observed in cmt3 zip4, compared with zip4 alone (χ2 P = 2.99 × 10−9), which indicates that the class II pathway may contribute to increased crossovers in cmt3 mutant centromeres (Fig. 4C; Supplemental Table S14). Additionally, the cmt3 fancm double mutant shows an additive increase in CEN3 crossover frequency, compared with cmt3 (χ2 P = 3.34 × 10−10) and fancm (χ2 P = 2.66 × 10−7) single mutants (Fig. 4C; Supplemental Table S14). From these data, we conclude that increased pericentromeric crossovers in H3K9me2 and non-CG DNA methylation mutants may involve contributions from both interfering and noninterfering repair pathways.

13During A. thaliana meiosis, SPO11-1 acts with SPO11-2 and MTOPVIB to generate DSBs, which can undergo inter-homolog repair to form crossovers. SPO11 enzymes are related to topoisomerase-VI transesterases and become covalently bound to ∼20- to 50-base target site oligonucleotides during DSB formation. We have purified and sequenced A. thaliana SPO11-1-oligos in order to map patterns of meiotic DSBs genome-wide, using a complementing SPO11-1-Myc spo11-1 line. In order to profile meiotic DSBs in H3K9me2/non-CG DNA methylation mutants, we generated a SPO11-1-Myc spo11-1 kyp suvh5 suvh6 line and used this to purify and sequence SPO11-1-oligos (Supplemental Table S16). The coverage of combined unique and multiple mapped SPO11-1-oligo reads were normalized by library size. Replicate libraries showed significant correlation at multiple scales (Supplemental Table S17). For example, Pearson's r values between replicate libraries at the 10-kb scale were between 0.91 and 0.99 (Supplemental Table S17). Further normalization was performed using a single-ended Col genomic DNA library with reads trimmed to 50 base pairs, which were aligned as for SPO11-1-oligos, and then used to calculate log2(SPO11-1-oligo/gDNA) values. Finally, z-score standardization was applied, such that scores represent the signed number of standard deviations from the mean, and these values were used for downstream analysis.

14To analyze chromosome-scale patterns, SPO11-1-oligo levels were calculated in 10-kb adjacent windows and plotted along the chromosomes using a rolling average. We also calculated a SPO11-1-oligo differential (Δ) by subtracting wild-type from kyp suvh5 suvh6 values. We observed that the centromeric and pericentromeric regions showed a striking increase in SPO11-1-oligos in kyp suvh5 suvh6 (Fig. 5A,B). We similarly calculated mean CG, CHG, and CHH methylation levels in 10-kb windows and calculated kyp suvh5 suvh6 differential values in the same way, using published data. The kyp suvh5 suvh6 mutant shows strong reduction of CHG and CHH methylation in the pericentromeric regions, which was significantly correlated with the SPO11-1-oligo Δ differential (Pearson's CHG r = −0.751, P <2.2 × 10−16, CHG r = −0.737, P = <2.2 × 10−16) (Fig. 5A,B). The increase in SPO11-1-oligos in kyp suvh5 suvh6 occurred more strongly within the genetically defined centromeres, compared with the crossover changes in cmt3, which occurred most strongly within the adjacent pericentromeres (Figs. 2D,F, 5A,B). Hence, while both meiotic DSBs and crossovers increase in H3K9me2/non-CG mutant heterochromatin, they were elevated in adjacent centromeric and pericentromeric regions, respectively.

15We next analyzed DSB frequency around copies of the CEN180 satellite repeat, which are found in proximity to the centromeres (Fig. 5C). Each A. thaliana chromosome sequence contains a centromere gap, which contains megabase arrays of CEN180 repeats. Further matches to the CEN180 consensus flank these gaps, and we identified 3397 repeats in the Col reference genome that generally occur in tandemly repeated arrays (Fig. 5C). SPO11-1-oligo density was analyzed in 20-kb windows around these repeat positions and compared to the same number of randomly chosen windows, in wild type and kyp suvh5 suvh6 (Fig. 5D). Consistent with the SPO11-1-oligo increase in centromeric regions at the chromosome scale (Fig. 5A), we observed a pronounced increase in SPO11-1-oligo density within the CEN180 repeats at the fine-scale in kyp suvh5 suvh6 (Fig. 5D). The pattern of SPO11-1-oligos within the CEN180 repeat units is also altered in kyp suvh5 suvh6 (Fig. 5D). Together, this shows that meiotic DSBs increase in the CEN180 repeats in H3K9me2 mutant backgrounds. A relatively small number (141) of transposons are transcriptionally up-regulated in kyp suvh5 suvh6 mutants and associated with decreased non-CG methylation. Some of these elements were associated with elevated SPO11-1-oligonucleotides in kyp suvh5 suvh6 mutants, but others were not (Supplemental Fig. S3), indicating that transcriptional activation is not strictly coupled to up-regulation of meiotic SPO11-1-oligo formation at A. thaliana transposons.

16In a large scale screen for de novo insertions of the nonautonomous maize transposable element Dissociation (Ds) introduced into A. thaliana, a transgene strongly expressing the Activator (Ac) transposase was crossed to plants containing Ds “launch-pads,” triggering Ds transposition in clonal cell lineages within F1 plants. As Ds elements are known to preferentially transpose to linked sites, a positive–negative selection scheme was implemented in F2 progeny to select against the Ds launch-pad and for transposed Ds. In this way, recovery of F2 transpositions was dependent on recombination between the transposed Ds and the launch-pad and should be mostly unlinked.

17Three launch-pads on Chromosome 1, DsE2, DsE3, and DsE6 (Fig. 6), were used to generate 9622 independent transpositions, which were mapped using TAIL-PCR. Unexpectedly, we recovered a dramatic enrichment of homozygous and heterozygous insertions into and nearby the CMT3 locus (214 out of 4044 insertions in total), but only when Ds elements were launched from a closely linked (∼360 kb) centromere-proximal locus on Chromosome 1 (DsE3) and not from a centromere-distal locus (DeE6) located a comparable distance from CMT3 (∼624 kb) (Fig. 6). One possible explanation was that flowers carrying homozygous cmt3 insertions arose by transposition into CMT3, followed by imprecise excision and mitotic recombination. Subsequent meiotic recombination between heterozygous transposed Ds elements and the launch-pad might be enhanced in cmt3 anthers, leading to increased recovery of F2 “transposant” progeny carrying insertions near CMT3. Ac elements in maize and Ds elements in A. thaliana can stimulate mitotic recombination between flanking repeats, but the excision/recombination scenario envisioned here has not been previously observed. Further work will be required to investigate the role of mitotic and meiotic recombination in the enhanced recovery of Ds insertions near CMT3, and the extent that this might be caused by disruption of the non-CG DNA methylation/H3K9me2 pathway is currently unknown.

18Meiotic crossover in proximity to centromeres has been associated with chromosome missegregation and aneuploidy in fungi and animals. Hence, suppression of crossovers within centromeres is thought to play an important role in maintaining fidelity of genome transmission during meiosis. However, abundant evidence for recombination-associated polymorphism in centromeric repeats exists, consistent with the effects of replication slippage, unequal crossover, and gene conversion. For example, maize centromeric CRM retrotransposons have been observed to undergo meiotic gene conversion but not crossover, which indicates initiation of meiotic DSBs but downstream inhibition of recombination leading to crossover. We also observe evidence for SPO11-1-dependent DSBs within the A. thaliana CEN180 repeats, meaning that meiotic recombination may contribute to polymorphism within centromeric satellite repeat arrays in this species.

19As centromeric DSBs increase in kyp suvh5 suvh6 mutants, this demonstrates that epigenetic information, including H3K9me2 and non-CG DNA methylation, plays important roles in suppressing initiation of meiotic recombination in these regions. However, our data also reveal complexity in how chromatin shapes meiotic recombination around plant centromeres. First, while the genetically defined centromeric regions show increased DSBs in kyp suvh5 suvh6, in cmt3 we observed that crossovers were most elevated in adjacent pericentromeric regions. An important distinction between these experiments is that SPO11-1-oligos were mapped in a Col/Col homozygous background, whereas mapping crossover necessitates use of polymorphic Col/Ler hybrids. In a Col/Ler hybrid context, centromeric crossovers are likely to be additionally suppressed by structural polymorphisms. For example, the inhibitory effect of structural polymorphism is evident within the ∼1.17-Mb heterochromatic knob inversion on Chromosome 4, which is suppressed for crossovers in both wild-type and cmt3 populations (Fig. 2D;, ). Following DSB formation, resection occurs to generate 3′ single-stranded DNA that can perform strand invasion of homologous chromosomes. Inter-homolog recombination downstream from strand invasion is sensitive to heterology between the recombining chromosomes, which can have a local inhibitory effect on crossover formation and instead promote noncrossover repair and gene conversion. A. thaliana centromeric regions exhibit extensive structural variation, including within Gypsy retroelements, which may therefore suppress centromeric crossover formation downstream from inter-homolog strand invasion, despite activation of meiotic DSBs in non-CG/H3K9me2 mutants. It is also possible that additional chromatin or epigenetic features enriched within the centromeres suppress crossover repair. For example, the kinetochore, CENH3 nucleosomes, or further heterochromatic marks such as H2A.W may be differentially enriched within the centromeric versus pericentromeric regions and cause inhibition of crossover maturation.

20An important question raised by our study is why pericentromeric crossover frequency increases in the H3K9me2/non-CG pathway mutants reported here, but not in met1 and ddm1 where reduced pericentromeric crossover frequency is observed, despite both kyp suvh5 suvh6 and met1 showing increased SPO11-1-oligos in proximity to centromeres. MET1 and DDM1 play major roles in the maintenance of DNA methylation in the CG context. However, their molecular roles in other respects are distinct: (1) Non-CG DNA methylation is reduced in ddm1 to a greater extent than met1; (2) gene body methylation is eliminated in met1 but not ddm1; and (3) H3K9me2 is reduced more strongly in ddm1 compared with met1. Therefore, we postulate that their common feature, loss of CG methylation within heterochromatin, alters progression and maturation of the meiotic recombination pathway, such that crossovers are favored in the chromosome arms, at the expense of the pericentromeres. In contrast, H3K9me2/non-CG pathway mutants activate recombination in the heterochromatic regions such that maturation of crossovers in the pericentromeres is increased. Indeed, distinctions in recombination phenotype are consistent with the different effects on transcription, transposition, and chromosomal conformation associated with loss of CG versus non-CG DNA methylation maintenance pathways.

21We propose that, while both CG and non-CG DNA methylation inhibit centromeric meiotic DSBs, only non-CG methylation and/or H3K9me2 inhibit crossovers. In agreement with this idea, euchromatic crossover hotspots in A. thaliana can be silenced by RNA-directed DNA methylation associated with gain of H3K9me2 and both CG and non-CG methylation. Furthermore, in fission yeast, arrested recombination intermediates accumulate strongly in wild-type heterochromatin, but not in clr4 and rik1 (Recombination In K) heterochromatin, which lose H3K9me2 and undergo mitotic and meiotic recombination. Mouse dnmt3l mutants also have altered DNA methylation and chromatin signatures and increased DSB initiation within retrotransposons, which is associated with meiotic catastrophe and infertility. In contrast, A. thaliana H3K9me2 and DNA methylation mutants are fully fertile, despite increased recombination initiation in the centromeric regions, suggesting that increased meiotic DSBs in transposons do not, per se, cause infertility. Suppression of heterochromatic recombination is a major barrier to introducing genetic diversity in crop plants like maize and wheat, where the majority of the chromosome is composed of pericentromeric heterochromatin and yet contains important genetic variation in functional genes and traits. Therefore, an exciting prospect will be to modulate H3K9me2 and non-CG DNA methylation to unlock pericentromeric crossovers in crop breeding programs.

22A. thaliana plants were grown under long-day conditions (16 h light/8 h dark, at 150 µmol light intensity) at 20°C. We used the following mutant alleles: kyp-6 (SALK_041474), cmt3-11 (SALK_148381), cmt3-7, kyp suvh5 suvh6 (SALK_041474, GK-263C05, SAIL_1244_F04), drm1-2 drm2-2 (SALK_031705, SALK_150863), drm1-2 drm2-2 cmt3-11 (SALK_031705, SALK_150863, SALK_148381), cmt2-3 (SALK_012874), ddm1 (SALK_000590), zip4-2 (SALK_068052), and fancm-1 (EMS point mutant). The centromeric FTLs CTL5.11 and LTL5.4 were obtained from the Traffic line population.

23Generation of the Ds transposant lines was previously reported. Ds insertion sites were amplified by TAIL-PCR.

24Crossover scoring using the pollen FTL CEN3 was performed by flow cytometry, as previously reported. Crossover scoring using seed FTLs (420, CTL5.11, LTL5.4) was performed by fluorescent imaging, as previously reported. Statistical analysis of fluorescent count data was performed as described.

25Illumina sequencing libraries were constructed in 96-well format, as previously reported, with the following minor modifications. DNA was extracted from three rosette leaves of 5-wk-old plants and 150 ng of DNA used as input for each library. DNA shearing was carried out for 20 min at 37°C with 0.4 units of DNA Shearase (Zymo Research). Each set of 96 libraries was sequenced on an Illumina NextSeq 500 (300-cycle Mid Output run). Sequencing data was analyzed to identify crossovers as previously reported, using the TIGER pipeline. Three hundred eighty-four cmt3-11 × cmt3-7 F2 individuals were sequenced. Wild-type crossovers (CO) were mapped by sequencing 245 Col × Ler F2 individuals, which were combined with data from 192 F2 individuals.

26The coordinates of crossover intervals called by TIGER were used for subsequent analysis. Centromeres were genetically defined as contiguous regions flanking the TAIR10 centromeric assembly gap that show an absence of crossovers in wild type. We define the pericentromeric regions as regions flanking the centromeres with higher than chromosome average levels of DNA methylation. The euchromatic arms constitute the remainder of the chromosomes, from the telomeres to the pericentromeres (Supplemental Table S9). Crossovers (midpoints) were counted in these regions and compared to those expected at random according to physical distance and χ2 tests performed (Supplemental Table S10).

27For chromosomal plots of crossovers and DNA methylation, crossovers were tallied in 10-kb windows along the chromosomes, and crossover/10-kb values were then divided by the number of F2 individuals analyzed for each genotype. For DNA methylation, mean values were calculated in 10-kb windows. Finally, for both crossovers and DNA methylation data, a rolling mean calculation was applied to smooth the data prior to plotting using the R function filter. The crossover and DNA methylation differentials were calculated by subtracting wild-type from cmt3 values.

28For telomere-centromere analysis, chromosome arms were first oriented such that each began at the telomere and ended at the centromere. The arms were then divided into windows representing 1% of their proportional lengths and crossovers assigned to these windows. Crossover values were divided by the number of F2 individuals analyzed for each genotype. These values were averaged across all chromosome arms and then plotted along the telomere-centromere axis with smoothing applied using the R function smooth.spline. For DNA methylation analysis along telomere-centromere axes, values were first calculated in 10-kb windows along the chromosomes. Chromosome arms were then oriented such that each began at the telomere and ended at the centromere and divided into windows representing 0.5% of their proportional lengths. Methylation values were then averaged across all chromosome arms and plotted along the telomere-centromere axis.

29A detailed protocol and analysis methodology are provided in an accompanying manuscript. A complementing SPO11-1-Myc spo11-1 line was crossed with kyp suvh5 suvh6 triple mutants and SPO11-1-Myc spo11-1 kyp suvh5 suvh6 plants identified for analysis. The CEN180 consensus sequence (5′-AACCTTCTTCTTGCTTCTCAAAGCTTTCATGGTGTAGCCAAAGTCCATATGAGTCTTTGGCTTTGTGTCTTCTAACAAGGAAACACTACTTAGGCTTTTAAGATCCGGTTGCGGTTTAAGTTCTTATACTCAATCATACACATGACATCAAGTCATATTCGACTCCAAAACACTAACC-3′) was matched to the TAIR10 reference sequence using the R function matchPattern with max.mismatch set to 90. The coverage value of normalized SPO11-1-oligonucleotides was analyzed in 20-kb windows around CEN180 matches and compared with analysis of the same number of randomly chosen positions.

30Chromosome spreads of A. thaliana pollen mother cells and immunostaining of ASY1 and H3K9me2 were performed using fresh buds, as described. Immunostaining of MLH1 was performed on acetic acid chromosome spreads on fixed buds, as described. The following antibodies were used: α-ASY1 (rabbit, 1/500 dilution, gift from Chris Franklin [University of Birmingham]), α-H3K9me2 (mouse, 1/200 dilution, Abcam, ab1220), and α-MLH1 (rabbit, 1/200 dilution, gift from Mathilde Grelon [INRA, Versailles]). Microscopy was conducted using a DeltaVision Personal DV microscope (Applied Precision/GE Healthcare) equipped with a CDD CoolSNAP HQ2 camera (Photometrics). Image capture was performed using softWoRx software version 5.5 (Applied Precision/GE Healthcare). All slides within an experiment (e.g., Col, cmt3, and kyp suvh5 suvh6) were prepared alongside one another and images captured using the same exposure time. The staining pattern of ASY1 and DNA was used to identify cells at leptotene, zygotene, or pachytene stage. Wild-type (Col) cells were first analyzed and a threshold pixel intensity value identified that removed background signal. This threshold was then applied to all images prior to further processing. Individual cell images were acquired as Z-stacks of 16 optical sections of 0.25 µm each, and the maximum intensity projection of the cell was reconstructed using ImageJ, as described. The boundaries of each cell were manually defined and the total signal intensity within the cell measured. An adjacent region outside of the cell was used to measure mean background intensity and this value used to subtract from the within-cell intensity. The same methods were used for analysis of meiotic and somatic cells.

31All data from this study have been submitted to ArrayExpress (https://www.ebi.ac.uk/arrayexpress/): SPO11-1-oligonucleotide sequencing data in wild type and kyp suvh5 suvh6 under accession number E-MTAB-5041; control libraries for SPO11-1-oligonucleotide sequencing under accession number E-MTAB-6257; and GBS crossover data from wild type and cmt3 populations under accession numbers E-MTAB-4657, E-MTAB-5476, and E-MTAB-6577 (GBS).