Recombination landscape dimorphism and sex chromosome evolution in the dioecious plant Rumex hastatulus

Abstract

There is growing evidence from diverse taxa for sex differences in the genomic landscape of recombination, but the causes and consequences of these differences remain poorly understood. Strong recombination landscape dimorphism between the sexes could have important implications for the dynamics of sex chromosome evolution because low recombination in the heterogametic sex can favour the spread of sexually antagonistic alleles. Here, we present a sex-specific linkage map and revised genome assembly of Rumex hastatulus and provide the first evidence and characterization of sex differences in recombination landscape in a dioecious plant. We present data on significant sex differences in recombination, with regions of very low recombination in males covering over half of the genome. This pattern is evident on both sex chromosomes and autosomes, suggesting that pre-existing differences in recombination may have contributed to sex chromosome formation and divergence. Our analysis of segregation distortion suggests that haploid selection due to pollen competition occurs disproportionately in regions with low male recombination. We hypothesize that sex differences in the recombination landscape have contributed to the formation of a large heteromorphic pair of sex chromosomes in R. hastatulus, but more comparative analyses of recombination will be important to investigate this hypothesis further.

This article is part of the theme issue ‘Sex determination and sex chromosome evolution in land plants’.

1. Introduction

The distribution of rates of recombination along chromosomes (recombination landscape [1]) influences many aspects of evolutionary genetics, including the efficacy of natural selection [2], genome structure [3] and the dynamics of reproductive isolation [4]. Rates of recombination can vary between species, between and within chromosomes, and between male and female meiosis in both dioecious/gonochoristic and hermaphroditic species [5–7]. We refer to this latter phenomenon as ‘sex differences in the recombination landscape’ [1]. Although sex differences in the rate and distribution of recombination appear to be widespread and variable, the causes and consequences of this variation have only recently been investigated in detail [1,6,8].

Many components of evolutionary processes depend on the sex-averaged rate of recombination. Nevertheless, sex differences in recombination (heterochiasmy) in dioecious populations can have important consequences for the evolution of sex chromosomes. This is because on the sex chromosome restricted to the heterogametic sex (i.e. the Y or W chromosome), sex-specific recombination landscapes will entirely control the rate of recombination, thereby influencing the size of the sex-determining region (SDR) [1]. We define the SDR as the non-recombining region of the sex chromosome including the causal sex-determining gene or genes as well as the entire region tightly linked to them. Recently, heterochiasmy has been proposed as an important factor in maintaining sexually antagonistic (SA) variants on the sex chromosomes, even in the absence of derived recombination modifiers [9]. In particular, SA alleles can spread more easily through populations with XY systems because of pre-existing male-specific reduced recombination, rather than recombination suppression evolving as a secondary consequence of the segregation of SA alleles [9]. Variation among species in patterns of heterochiasmy could thus be an important general determinant of the evolution of sex chromosomes and their turnover [1], potentially contributing to differences among lineages in the likelihood of the formation of heteromorphic sex chromosomes, the maintenance of SA polymorphisms and the size of the SDR.

Several observations are evident in the characteristics of sexual dimorphism in the recombination landscape. First, although data are limited, many eukaryotes have recombination rates biased towards the tips of chromosomes in male meiosis, whereas female recombination rates are more likely to be either elevated towards the centromeres or more uniform across the chromosome [1]. In hermaphroditic plants, three of five taxa studied show this pattern [10–12], although in maize there was limited evidence for large-scale differences in recombination between male and female meiosis [13], and in an interspecific cross between Solanum lycopersicon and S. pennellii (formerly Lycopersicum esculentum and L. pennellii) recombination in male meiosis was reduced genome-wide compared with female meiosis [14]. Preliminary analysis of genetic maps in the dioecious Mercurialis annua do not suggest major sex differences in recombination rates [15], although the genomic context of these maps is still being investigated (J. R. Pannell 2021, personal communication). In general, however, recombination rate landscape dimorphism has not yet been investigated in dioecious plants, which limits our understanding of its potential role in the evolutionary dynamics of plant sex chromosomes.

Although many species show tip-biased recombination in male meiosis [1], the reasons for this pattern are unclear. Both non-adaptive and adaptive explanations have been proposed. If recombination landscape differences are not adaptive, they may simply result from mechanistic differences between the processes of female and male meiosis [1]. For example, it has been hypothesized that recombination tends to be initiated at the tips of chromosomes, leading to lower probabilities of crossover at chromosome centres, particularly for large chromosomes [16]. Sex differences in the degree of chromosome condensation during male and female meiosis may cause a greater periphery-bias in recombination in males [16]. Adaptive hypotheses have also been proposed for sex differences in recombination landscape. For example, if the strength of haploid selection is more intense in males, positive epistatic selection will favour reduced recombination during male meiosis and maintain coadapted allele combinations during haploid selection [6]. Alternatively, selection may favour higher female recombination rates near centromeres because recombination uncouples meiotic drive alleles from the centromere [7]. This process could select for a shorter region of recombination suppression surrounding the centromeres in females [7]. More recently, it has been proposed that SA selection can favour tighter linkage between sex-specific genes and their regulatory elements [1], which could also select for sex differences in recombination landscape. In plants, evidence that female recombination rates are elevated relative to male recombination rates in outcrossing species compared with selfing species [6] is consistent with models of both female meiotic drive and male gametophytic selection, as both are expected to be more intense with higher rates of outcrossing [6,7]. Disentangling these alternatives is challenging and will require more comparative information on sex-specific recombination in both hermaphroditic and dioecious taxa.

Rumex hastatulus (Polygonaceae) is a dioecious, wind-pollinated plant with heteromorphic sex chromosomes. Populations west of the Mississippi River have an XY sex chromosome system, whereas those east of the Mississippi have an XYY system [17,18]. In this species, we have found some evidence for gene loss on the Y chromosome, but the majority of genes retain a Y copy and show low genetic divergence between the X and Y chromosomes [18]. Recent genome sequencing combined with high marker-density linkage mapping has revealed that the SDR is embedded within a very large genomic region of extremely low recombination [19]. Evidence for similarly large non-recombining regions in the likely pericentromeric regions of autosomes suggested that pre-existing recombination suppression may have contributed to the formation of large heteromorphic sex chromosomes in R. hastatulus [19]. However, this study measured sex-averaged recombination rates, thus limiting our ability to investigate the potential roles of heterochiasmy in sex chromosome formation and maintenance. With earlier evidence for an important role for gametophytic selection on the sex ratio in this species [20,21], the influence of pollen competition in the evolution of the sex chromosomes [22], and indications of frequent male and female transmission distortion in related dioecious Rumex taxa [23], there is a strong possibility that haploid selection in males and/or females may contribute to sex-specific selection favouring sexual dimorphism in the recombination landscape of R. hastatulus.

Here, we explore the potential importance of heterochiasmy for the evolution of sex chromosomes and investigate genomic correlates of sex-specific recombination rate differences in R. hastatulus. Using a sex-specific linkage map and an updated draft genome assembly, we determine whether R. hastatulus shows evidence for heterochiasmy and other sex differences in recombination landscape on both the sex chromosome and the autosomes. We then examine the genome-wide correlates of male and female recombination rates and use this information to evaluate the potential role of sexual antagonism, haploid selection and meiotic drive as evolutionary drivers of sexual dimorphism in the recombination landscape.

2. Methods

(a) . Linkage mapping and genome assembly

We generated a mapping population from a single F₁ cross between an unrelated male and female derived from a single population collected in Rosebud, TX [21]. Seeds from the field collection were grown in the glasshouse and at the onset of flowering one male and one female individual from independent maternal families were randomly paired for a controlled cross to develop the F₁ generation. Leaf tissue was collected from the parents, which were immediately moved into a miniature crossing chamber [24] to avoid pollen contamination from other plants growing in the same glasshouse, and F₁ seeds were harvested after maturation. To obtain tissue from F₁ plants, we sterilized seeds using 5% (V/V) bleach and germinated them on filter paper in refrigerated Petri dishes. After germination, we transplanted seedlings into six-inch plastic pots containing a 3 : 1 ratio of Promix soil and sand and a slow-release fertilizer (Nutricote, 14 : 13 : 13, 300 ml per 60 lbs) and placed them in a glasshouse at the University of Toronto, St. George campus. We watered plants every other day, and their positions on benches were randomized weekly. On day 43 or 44 after transplant, between 10 : 00 and 12 : 00, we collected and flash-froze 30 mg of leaf tissue for RNA extraction using liquid nitrogen. We have previously been successful in using RNA sequencing as a strategy for reduced representation genotyping of the large genome of R. hastatulus, including for association and linkage mapping [18,19]. When plants flowered, we phenotyped for sex. We used Spectrum Plant Total RNA Kits (Sigma-Aldrich) for RNA extraction. The sequenced library included 188 individual plants: one sample each of 102 female and 84 male offspring, as well as three replicate samples of each of the two parents of the mapping population.

For library preparation and sequencing, we sent our RNA samples to the Centre d'expertise et de services Génome Québec (CES, McGill University, Montréal, QC, Canada). CES prepared libraries using NEBNext library prep kits and sequenced them on a NovaSeq6000 S4 PE100. Sequencing resulted in a total of 3.1 billion paired-end reads (3 060 570 370), with between 10 and 49 million reads per sample (mean 15 940 471, median 14 548 490). We have deposited raw sequences on the Sequence Read Archive (SRA) under the accession number PRJNA692236 (embargoed until 1 July 2022 or publication).

We aligned our raw sequencing reads to the R. hastatulus Dovetail draft genome assembly [19] using STAR 2-pass v. 2.7.6 [25,26]. We processed the aligned files to sort and mark optical duplicates and to separate reads spanning splice junctions using PicardTools (http://broadinstitute.github.io/picard/) and the Genome Analysis Tool Kit [27]. We initially generated a linkage map using Lep-MAP3 [28] from reads aligned to the original R. hastatulus draft assembly; because Lep-MAP3 relies on haplotype reconstruction and our parental plants were heterozygous for distinct haplotypes, linkage maps can be generated from F₁ families without fixed marker differences at all variable sites. The markers could be split into five linkage groups using a logarithm of the odds (LOD) score limit of 30 for the initial split followed by a LOD score limit of 34 for the resulting largest linkage group (SeparateChromosomes2). Most of the remaining single markers were put into these groups using a LOD score limit of 25 (JoinSingles2All), totalling about 120 000 markers. We then calculated the marker order for each linkage group with OrderMarkers2 (with default settings).

To improve our linkage map, we reduced redundancy in our genome assembly and constructed a new pseudo-chromosome assembly using the Lep-Anchor [29] software. To obtain reliable linkage maps, we removed the 13 most-recombining individuals from the maps and constructed three independent linkage maps (Lep-MAP3: OrderMarkers2) using only male informative markers (parameter informativeMask = 1), only female informative markers (informativeMask = 2) and all markers. These maps were used by the Lep-Anchor software to connect the assembled sequence to the linkage map.

To reduce redundancy in the genome assembly, we first split the existing Dovetail assembly into contigs based on assembly gaps, using contig-joining data included with Dovetail's assembly output. Due to the high number of contigs (more than 44 000), we removed all contigs of fewer than 500 bp, full length haplotypes and joined partial haplotypes in windows of five adjacent contigs (link strength was 6−|distance|−|difference in orientations|, where distance between contig i and j is |i − j| and difference is 0–2 based on how the orientations differ: same = 0, one different = 1, both different = 2). This was done in Lep-Anchor by giving it only the alignment chain computed by HaploMerger2 ([30]; see https://github.com/mapleforest/HaploMerger2) on the WindowMasker [31] soft-masked (contig-split) genome. This allowed us to reduce the number of contigs to about 33 000. All data were transferred to the new contig assembly coordinates using the liftover script and LiftoverHaplotypes module in Lep-Anchor.

We aligned the raw PacBio sequence originally used to generate the Dovetail assembly back to the Dovetail assembly using minimap2 [32] to identify which contig joins were and were not supported by long-read data. We then ran Lep-Anchor (lepanchor_wrapper2.sh) on the final contig assembly using the three linkage maps, new alignment chains (HaploMerger2) and alignments of raw PacBio sequence [32]. The resulting pseudo-chromosome assembly was used to calculate physical order of linkage map markers, and the maps were evaluated (OrderMarkers2 parameter evaluateOrder) in this order to obtain the final linkage maps. After this assembly improvement, we compared contig orders between our previous [19] and new maps using custom R scripts incorporating Plotly [33] interactive plotting.

(b) . Recombination rates and transmission distortion

We quantified recombination rates in two ways. First, we described recombination using map lengths in centimorgans (cM) from the maps produced by Lep-MAP3. Based on the scale of recombination observed in previous work [19] and the current map, we performed all downstream analyses using 1 Mb windows. We also calculated recombination rates as the sum of crossover events per 1 Mb window. To do this, we first calculated the number of crossovers per site from cM differences using the inverse of the Haldane mapping function [34], then summed crossovers in 1 Mb windows. To describe the extent of recombination suppression, we identified the total number of consecutive windows with zero crossovers. We estimated transmission ratio distortion as a likelihood ratio of the deviation from 1 : 1 transmission of haplotypes from the male and female parent using custom scripts by PR.

(c) . Gene and transposable element (TE) content

We also developed a new annotation using MAKER v. 3.01.03 [35]. For our MAKER annotation, we used the soft-masked [31] genome integrated with previously published floral transcriptomes from six individuals [22] and leaf transcriptomes from six populations [18]. Transcripts were assembled with IDBA-tran v. 1.2.0 [36] and annotated in four rounds, using the transcripts and the tartary buckwheat annotation version FtChromosomeV2.IGDBv2 [37] as the evidence for MAKER. We functionally annotated the final gene annotation based on homology using BLAST v. 2.2.28+ [38] and InterProScan 5.52–86.0 [39]. This annotation resulted in 59 121 genes. We also annotated the locations of rDNA repeats using RNAmmer-1.2 [40]. The parameters used, ‘-S euk’ and ‘-m tsu,ssu,lsu’, indicate that the input reference is a eukaryote, and that we are annotating 5/8 s, 16/18 s and 23/28 s rDNA.

We produced the TE annotation using the EDTA (Extensive de-novo TE Annotator) v. 1.9.7 pipeline [41]. This pipeline combines the best-performing structure- and homology-based TE finding programs (LTR_FINDER_parallel [42], LTR_harvest_parallel [43], LTR_retriever [44], TIR-Learner2.5 [45], HelitronScanner v.1.1 [46], Repeatmodeler2.0.1 [47] and RepeatMasker-4.1.1 [48] and filters their results to produce a comprehensive and non-redundant TE library [41]). The optional parameters ‘–sensitive 1’ and ‘–anno 1’ were used to identify remaining unidentified TEs with RepeatModeler and to produce an annotation. The ‘EDTA.TEanno.split.gff3’ output file was used as our non-overlapping TE annotation. This file is produced by EDTA by removing overlaps according to the following priorities: structure-based annotation > homology-based annotation, longer TE > shorter TE > nested inner TE > nested outer TE [41].

For all gene content analyses, we used a stringently filtered set of genes to remove gene annotations associated with transposable elements. We first used BEDTools [49] to remove any exons that overlapped a TE, although genes containing both exons that overlapped TEs and exons that did not overlap TEs were retained. We then removed any gene functionally annotated with ‘transpos*’ (transposon, transposase, etc.), ‘ribonuclease H’, ‘pol poly’, ‘mitochondri*’, ‘chloroplast’ or ‘retrovirus’. This filtered annotation contained 30 641 genes.

(d) . Differential expression and SNP calling

We performed differential expression analyses using DESeq2 (v. 1.28.1) [50] and our new annotation. For DESeq2 analyses, we aligned reads to the new genome pseudomolecules using STAR v. 2.7.6a [50] and generated readcounts using featureCounts (2.02) [51]. Cutoffs for differential expression were as follows: FDR-adjusted p-value less than 0.1, absolute log₂fold change greater than 1. Note that because we are primarily interested in having a large set of differentially expressed genes for analysis of the correlates of recombination rather than a set of candidate genes, we used a relatively permissive adjusted p-value. We identified genes that were differentially expressed between male and female leaf tissues using published leaf RNA sequence data from population samples of the XY cytotype [18], and between male and female floral tissue using published RNA sequence data from the XY cytotype [22]. Genes with fewer than 20 reads across all samples were removed from these analyses. We also identified sequences that were differentially expressed in pollen tissue compared to male leaf tissue, using published sequence data [22]. Finally, we identified sequences differentially expressed in pollen tubes compared to pollen, using pollen from the individuals in the mapping population described in [19]. We collected pollen using a kief box (Wacky Willy's, Victoria, BC, Canada), germinated and grew it in 100 µL of media [52] for 24 h, and flash froze the resulting pollen tubes in LN2. After removing media, we lysed cells and extracted total RNA using Spectrum Plant Total RNA Kits (Sigma-Aldrich) for RNA extraction. To identify sex-linked SNPs and fixed differences between the X and Y chromosomes (all females homozygous reference or non-reference, all males heterozygous) for our new assembly, we used FreeBayes v. 0.9.10-3-g47a713e [53] to call SNPs from the population transcriptome data from the XY cytotype (six males and six females) [18] and the crossing transcriptome data from the same cytotype (six male and six female offspring, plus parents) [18]. We filtered the SNPs to exclude any with a SNP quality score of lower than 60, any sites with missing data, and SNPs that exhibited fixed heterozygosity across all samples and therefore likely reflecting paralogous mapping.

(e) . Linear modelling predictors of recombination rate

We combined our linkage map data with our annotation, TE annotation, differential expression data and summed and averaged variables in 1 Mb windows to perform analyses of recombination landscape, gene content and differential expression in R v. 4.1.0 [54] in RStudio v. 1.4.1717 [55] using the packages dplyr v. 1.0.7 [56] and stringr v. 1.4 [57]. We performed correlations using R's built-in cor function and estimated partial correlations using the package ppcor v. 1.1 [58].

To identify factors associated with genome structure that predicted recombination rates and recombination rate differences [59], we created linear models with the following response variables: female crossovers per window, male crossovers per window, sex-averaged crossovers per window, crossover number sex difference per window and female- versus male-biased recombination across window. We fitted all responses using generalized linear models with either negative binomial or Tweedie distributions except for female- versus male-biased recombination, for which we used logistic regression. We performed linear models in glmmTMB v. 1.12 [60], evaluated model fit using DHARMa v. 0.4.3 [61] and compared models using ANOVAs. We performed separate models for each response variable on each chromosome. Scripts are available at https://github.com/joannarifkin/Rumex-sex-specific.

3. Results

(a) . Linkage mapping and genome assembly improvement

We identified five linkage groups, consistent with both the karyotype of the XY cytotype of R. hastatulus [17,62] and our previous sex-averaged linkage mapping results [19] (table 1). We confirmed two apparently metacentric linkage groups (A1 and A2) and three apparently submetacentric linkage groups (A3, A4, XY) based on the patterns of recombination across the chromosomes (figure 1a), cytogenetic evidence for the relative positioning of centromeres [62], and the identities of the scaffolds constituting the linkage groups. We have retained the same autosomal labels across both maps and they continue to reflect chromosome sizes from largest (A1) to smallest (A4).

Table 1.

The linkage groups (LG) of the revised Rumex hastatulus genome assembly, including sex-averaged, male, and female map lengths (length in Mb), gene content, and the extent of the non-recombining region.

LG	sex-averaged map length (cM)	male map length (cM)	female map length (cM)	Mba	number of genes	largest size of markers with 0 crossover events (Mb), in males/females
A1	104.57	94.624	114.516	388.3386 (344.5)	7512	238.6/94.3
A2	91.67	79.032	104.301	278.2766 (260.43)	7301	33.4/22.9
A3	69.09	48.387	89.785	171.7219 (175.01)	3923	91.5/44.3
A4	61.29	52.688	69.892	135.2128 (158.24)	3502	62.4/30.2
XY	79.03	64.516	93.548	239.0056 (150.39)	4752	212.1/55.3

^aValues in brackets indicate the lengths from the previous assembly [19].

Figure 1.

Distribution of recombination, segregation distortion, and gene content in Rumex hastatulus. (a) Male (purple/lighter) and female (green/darker) Marey maps of the chromosomes. (b) Difference in crossover number for 1 Mb windows along the chromosome (male crossovers per window – female crossovers per window). Positive: male crossover excess. Negative: female crossover excess. (c) Segregation distortion for male (purple/lighter) and female (green/darker) haplotypes. Dashed red lines indicate significance at 0.05 and 0.01 levels according to chi-square tests. (d) Genes per 1 Mb window along the genome. (e) TEs per 1 Mb window along the genome. Yellow/lighter: DNA TEs. Blue/darker: RNA TEs. (Online version in colour.)

Our larger genetic mapping population and improved genome assembly led to considerable improvement in higher-order chromosome-scale scaffolding of the genome of R. hastatulus. Our improved genome assembly contained 1.45 Gb, a reduction of 0.2 Gb from our previous 1.65 Gb assembly [19] due to the collapsing of redundant haplotypes (see §2). For this assembly, 1.212 Gb (84%) is now grouped in the five linkage groups (previously 1.08 GB, 65% of the previous primary assembly), with the remaining 0.23 Gb in smaller contigs (ranging in length from 502 to 446 089 bp, mean 10 736). These corrections have substantially increased the size of the assembled sex chromosome, with an additional 88.6 Mb of sequence assembled on the sex chromosome, the largest increase of any of the chromosomal scaffolds (table 1). Consistent with this increase, in our past genome assembly, only 52% of sex-linked SNPs mapped to the assembled sex chromosome; our new assembly, integrated with transcriptomes from independent crossing data [18], now positions 94% of SNPs showing X–Y segregation patterns on the sex chromosome.

(b) . Recombination rates

As in our previous study [19], we found that recombination was unevenly distributed across the genome, with very large non-recombining regions on all chromosomes (figure 1). We identified clear evidence of heterochiasmy (table 1 and figure 1a). Male map lengths were shorter than female map lengths: across chromosomes, female map length averaged 1.4× male map length (table 1 and figure 1a) and the sex chromosome was not an obvious outlier for this metric. Furthermore, although regions of low male recombination tended to also be regions of low female recombination, males had longer continuous blocks of non-recombining windows across all chromosomes. To summarize this pattern, we identified the longest stretches of markers on each chromosome with zero crossovers. On the autosomes, males had runs of non-recombining windows approximately twice as long as those of females, with male-specific non-recombining regions as large as 238 Mb (table 1). By this measure, the sex chromosomes were an exception: the largest run of male-specific non-recombining windows on the sex chromosome was four times the length of the longest run of female-specific non-recombining sequence (table 1). Thus, although all chromosomes exhibited reduced male recombination rates, the XY chromosome showed the largest size difference between the sexes even though it was not the largest chromosome. Overall, despite broad similarities, male and female recombination rates in R. hastatulus were only weakly correlated (r = 0.333, correlation of male and female crossover number among 1 Mb windows across all chromosomes; figure 2 and electronic supplementary material, table S1).

Figure 2.

Correlations among genome window characteristics across the whole genome of Rumex hastatulus. Colours and sizes correspond to the strength and direction of the correlations. (Online version in colour.)

The extent of sex differences in recombination varied both within and between chromosomes (figure 1b). Chromosomes A1, A2 and the sex chromosome conformed to the common pattern of more tip-biased recombination in males, but the submetacentric chromosome A3 showed female-biased recombination in the more highly recombining end, and A4 appeared to have low-recombination regions at both ends. This differs from the previous linkage map, likely because of the difficulty of positioning low-recombination regions. In general, regions of low recombination were similar in both sexes, but female map lengths were larger in these regions and showed apparent hotspots of recombination with large jumps in centimorgan position (figure 1). All five chromosomes had regions of female-biased and male-biased recombination, as well as shared recombining and non-recombining regions (figure 1b). This complex pattern creates a highly heterogeneous recombination landscape.

(c) . Gene and TE content

As in our previous study [19], we found that genes were generally concentrated in high-recombination regions (figure 1d). However, A1 contained one gene-dense yet low-recombination region (approx. 100–200 Mb). RNA (class 1) TEs were concentrated in low-recombination regions, whereas DNA (class 2) TEs were concentrated in high-recombination regions (figure 1e). Despite the reduced gene density in low-recombination regions, the large size of non-recombining regions means that a large fraction of genes in the genome (approximately 37%) are in these large regions with no male recombination, many of which include at least limited recombination in females. Ribosomal genes were concentrated mostly on A3 (62 rDNA features annotated) and A4 (28 rDNA features annotated). rDNA genes occurred in the first 50 Mb of A3 (5S subunit sequence), with additional rDNA sequence located around 130 Mb of A3 (18S and 28S subunit sequence) and in the first 3 Mb of A4 (18S and 28S subunit sequence). These rDNA locations are consistent with past cytological findings [62], further confirming our identification of the A3 and A4 chromosomes.

(d) . Characterization of the sex-determining and pseudoautosomal regions

With female and male recombination separated, it is clear that the male-specific non-recombining region of the sex chromosome in R. hastatulus, i.e. the SDR, is extensive. In particular, our linkage mapping suggests that the SDR is as large as 209 Mb (14% of the total assembly: table 1 and figure 1), including as many as 3595 genes (12% of the total filtered annotated genes). The gene-dense recombining pseudoautosomal region of the sex chromosome is similarly the smallest male recombining segment of any chromosome, representing only approximately 13% of the physical size of the chromosome. Note that the ‘true’ SDR may be narrower and the pseudoautosomal region larger as rare male recombination may have gone unobserved in a single cross. However, evidence suggests that most of this region is non-recombining and linked to the SDR. Specifically, fixed differences between the X and Y chromosome from our population samples [18] mapped onto the assembly are common across the first 210 Mb (electronic supplementary material, figure S1).

(e) . Transmission ratio distortion

We identified loci with biased haplotype transmission through either paternal or maternal inheritance based on the haplotype reconstructions from Lep-Map 3. Transmission ratio distortion varied between chromosomes (figure 1c and electronic supplementary material, figure S2 and table S2). More sites distributed across a larger fraction of the genome experienced biased transmission through maternal (276) than paternal (48) inheritance. On A1, 18 sites were significantly distorted in transmission from males using a 0.05 cutoff in a chi-squared distribution (LOD greater than 3.841), and on A4 30 sites were significantly distorted in transmission from males with a 0.05 cutoff. Although deviations from 1 : 1 male haplotype transmission occurred on other chromosomes, there were no significant male-distorted sites on A2, A3 or the sex chromosome. In contrast, female haplotype distortion was extensive on A2 (91 sites in females at 0.05, 38 at 0.01 cutoff of LOD greater than 6.635), A3 (108 sites at 0.05 cutoff, 88 at 0.01) and the sex chromosome (76 sites at a 0.05 cutoff, 26 at 0.01), but negligible on A1 (0 sites) and A4 (1 site at 0.01).

(f) . Correlates of recombination rate differences

Across the genome, the number of genes, leaf- and flower-expressed genes, and DNA (class 2) TE density were all positively correlated with recombination rate, whereas RNA (class 1) TE density was negatively correlated with recombination rate (figure 2 and electronic supplementary material, table S1 and figures S3–S6). Male crossover number and female crossover number were both positively correlated with gene density, but this correlation was stronger for male crossovers (male Pearson's r = 0.390, female Pearson's r = 0.278). However, the correlations between transmission ratio distortion and crossover number were in opposite directions: in females more distorted regions were also more recombining (Pearson's r = 0.111), whereas in males distortion was negatively correlated with crossover number (Pearson's r = −0.197), reflecting the fact that male transmission distortion signals were enriched in the large non-recombining regions of male meiosis (figure 1). Correlations varied in strength between chromosomes, with notable differences in correlates of transmission ratio distortion and recombination rate difference, both of which varied in magnitude, position and direction within and between chromosomes (figure 2 and electronic supplementary material, figure S7 and table S3). The difference in crossover number between the sexes was most strongly correlated with signals of male distortion (figure 2), where regions of particularly low male crossover number represented regions with larger signals of male distortion. We also estimated partial correlation coefficients controlling for gene density, which was consistently correlated with many genomic variables (electronic supplementary material, figures S8 and S9 and tables S4 and S5). Correlations between distortion and crossover number in both males and females persisted after controlling for gene density and varied in direction across chromosomes (electronic supplementary material, figure S9). However, all chromosomes except A2 showed a consistent negative correlation between male-biased recombination and male transmission distortion, even when gene density was controlled.

(g) . Linear models of recombination rate differences

We used generalized linear models to identify predictors of sex-specific and sex-averaged recombination rates, whether recombination was male- or female-biased, and the magnitude of the recombination rate difference in R. hastatulus. Full modelling results are available in electronic supplementary material, tables S6 and S7.

In our linear models, the variables that significantly predicted male recombination rate varied between chromosomes (table 2a). The number of genes predicted increased male recombination rate on three chromosomes (as well as a fourth in some models), and position along the genome emerged as an important predictor on two submetacentric chromosomes, suggesting that distance from chromosome peripheries predicts male recombination (figure 1). RNA TE count also appeared to play a role, but in inconsistent directions, predicting increased male recombination on two chromosomes and decreased male recombination on a third; this pattern also appears in the partial correlations with gene density removed (electronic supplementary material, figure S9). Transmission ratio distortion predicted reduced male recombination rate on three chromosomes (A2, A3 and A4) and predicted increased male recombination on A1. Either pollen bias or pollen-tube bias predicted male recombination rate on three chromosomes (negatively on A4 and the sex chromosome, positively on A3).

Table 2.

Variables identified by linear models as significant predictors of recombination in windows of the R. hastatulus genome. (a) Male recombination rate predictors. Each cell contains the estimate for the effect of the term in the model. In A3, two models fit equally well; we report both, although the more complex model may be overfitted. Interaction effects in best model—A1, number of RNA TEs : male distortion (–0.003). A2, number of genes : male distortion (0.026*). A3 (more complex model), position window : number of genes (–0.003*), male distortion : position window (0.080), number of genes : male distortion (–0.085*), number of genes : position window : male distortion (less than 0.001*). A3 (simpler model), male distortion : position window (0.079). A4, none. XY, position window : number of genes (–0.004). * indicates that t​he term was not significant alone, typically because the term is significant because of an interaction effect. (b) Female recombination rate predictors. Each cell contains the estimate for the effect of the term in the model. On the XY, two models fit equally well; we report both, although the more complex model may be overfitted. Interaction effects in best model—A1, number of RNA TEs : pollen-tube bias (0.005). A2, number of genes : position window (>0.001*). A3, number of genes : position window (–0.009), leaf expression : position window (0.012). A4, number of genes : position window (>0.001*). XY (simpler model), position window : female flower bias (–0.007). XY (more complex model), position window : female flower bias (–0.009), position window : pollen-tube bias (–0.013*), flower female bias : pollen-tube bias (–1.893), position window : pollen-tube bias : flower female bias (0.008). * indicates that t​he term was not significant alone, typically because the term is significant because of an interaction effect.

	# genes	position	# RNA TEs	male distortion	pollen-tube bias	pollen bias
(a) male recombination rate predictors:
A1	0.242		0.003	0.980
A2	0.045		0.005*	−0.215*
A3 (neo)	0.204*	−0.026*		−2.606*	0.441
	NA	0.078		−4.454	0.434
A4			−0.009	−0.523	−0.558
XY	1.035	0.243				−0.084

	# genes	position	# RNA TEs	flower female bias	pollen-tube bias	leaf expression
(b) female recombination rate predictors:
A1	0.0563		−0.004*		−2.607
A2	0.0232	−0.004*
A3 (neo)	0.280	−0.010*				−0.331
A4	0.094	0.047
XY	0.055	0.0207		0.844	0.440*
	0.049	0.0290		1.464	3.665

In contrast, the predictors of female recombination are more consistent across chromosomes (table 2b). The number of genes per window positively predicted female recombination rate on all five chromosomes, and position affected female recombination rate on four chromosomes. Only one other variable, pollen-tube-biased expression, predicted female recombination rate on more than one other chromosome. Female transmission ratio distortion did not emerge as a significant predictor of recombination rate on any chromosome.

Predictors of both sex-averaged recombination and of the magnitude of the recombination rate difference between males and females reflected predictors of male and female recombination rate independently. The number of genes, position along the chromosome, the number of RNA TEs, female-biased floral expression and pollen-tube-biased expression were all significant predictors, but their importance and direction varied between chromosomes (electronic supplementary material, tables S6A and S6B).

Finally, we used logistic regression to identify variables that predicted whether windows exhibited female or male recombination bias (electronic supplementary material, table S6C). Our logistic regressions also suggested considerable variation in the factors that predicted sex differences in recombination. The number of genes per window was an important predictor for four out of five chromosomes, but in variable directions, predicting both male bias (A1, XY) and female bias (A2, A3) in recombination.

4. Discussion

The main findings of this study are consistent with the general pattern of extensive low recombination that we previously inferred based on sex-averaged recombination in R. hastatulus [19]. However, our sex-specific maps indicate that recombination suppression is not evenly distributed between males and females, and that the very large observed regions of low sex-averaged recombination are particularly influenced by male recombination landscapes. Across all chromosomes, females recombine more frequently than males, and males have much larger continuous non-recombining blocks than females. Although general patterns of low and high recombination in males and females have similarities, a key difference is in the length of non-recombining segments (table 1). Our results are in line with several studies of hermaphroditic plants, as well as other eukaryotes [1], with the very large male-specific non-recombining regions that we report making this an extreme case.

Rumex hastatulus does not, however, strictly follow the common eukaryotic pattern of tip-biased male recombination [1] across all chromosomes. The recombination landscapes of both the metacentric and submetacentric chromosomes suggest greater variation in the distribution of recombination than simply less male recombination in the centres and more at the chromosome ends, with highly recombining regions scattered along chromosomes (particularly A2, A3 and A4). Thus, although male recombination is more concentrated, it is not always concentrated at the tips of the chromosome. These differences among chromosomes may reflect an ongoing history of chromosomal rearrangements in Rumex and additional patterns of chromosome structure such as the locations of centromeres and rDNA clusters [62].

The pattern of larger non-recombining regions in males is consistent with an evolutionary bias toward the evolution of male heterogamety and XY sex chromosomes in Rumex [1]. In particular, very low recombination can facilitate the invasion and maintenance of SA variants linked to SDRs [63], and SDRs that evolve in large non-recombining regions, especially sex-specific ones, may contribute to the subsequent evolution of sex chromosomes [64]. The existence of male-specific non-recombining regions may thus facilitate the evolution of XY rather than ZW sex chromosomes, although similar patterns of recombination dimorphism can still occur in ZW systems, albeit less frequently [1].

Our findings of no crossovers in male meiosis over very large fractions of each chromosome suggest that recombination suppression may have been ancestral to the evolution of dioecy in Rumex even if subsequent/additional recombination modifiers evolved following the origin of the SDR. The observed size of the SDR, which is over 200 Mb using population-validated sex-linked genes and includes over 14% of the assembled genome and over 3500 genes, is much larger than those recently reviewed in plants [65], although it is possible that the sex-linked region of Silene latifolia may be larger [66]. However, larger population samples of R. hastatulus should be used to test for very rare recombination events between the X and Y. High-coverage phased assemblies of X and Y chromosomes will also be useful to assess the possible role of chromosome rearrangements, which to date have not been detected in the pairing region of the sex chromosomes of R. hastatulus [17,62]. As with transmission distortion below, it is important to note that our conclusions are currently based on a single cross, and these patterns may vary across genotypes and populations. Furthermore, while large-scale sequence divergence can also contribute to recombination suppression between the sex chromosomes [67], this explanation seems unlikely in R. hastatulus, as average sequence divergence is very low between X and Y gametologues (synonymous site divergence (Ks) average of approximately 0.02 [18]).

Our results suggest that sex differences in the recombination landscape could have played an important role in determining the large size of the SDR and may have facilitated the evolution of large heteromorphic sex chromosomes in this system. However, we cannot rule out a role for subsequent recombination modifiers, including chromosome rearrangements. Further research on R. hastatulus and comparative studies of heterochiasmy in both hermaphroditic and other dioecious species in Rumex will be important to further assess the extent to which ancestral heterochiasmy and derived changes in recombination rates have contributed to sex chromosome evolution in this lineage.

Models for the evolution of heterochiasmy due to male haploid selection and female meiotic drive both predict lower overall male recombination rates and higher female recombination near centromeres [1,6,7], as we observed in R. hastatulus. Our mapping population provided evidence for both male and female transmission ratio distortion, which varied within and among chromosomes and between the sexes. Overall, more sites displayed significant distortion in female than in male transmission, but significant regions of both types of distortion were observed across the genome. Transmission ratio distortion through female inheritance is generally thought to be consistent with female meiotic drive. In contrast, transmission ratio distortion through male inheritance may result from haploid (pollen) competition [68].

Zygotic distortion (i.e. differential seed germination or seedling survival) could also lead to transmission ratio distortion, and can result from alleles inherited from either parent [68]. In our study, we genotyped reproductive adults rather than pollen or seeds, which conflates several opportunities for natural and sexual selection to cause biased transmission. Zygotic selection may be particularly likely to explain our observed female distortion, since regions showing female distortion were not focused on low-recombination, likely pericentromeric regions, where meiotic drive is expected to act [7]. In contrast, signals of male transmission distortion are particularly enriched in regions of low male recombination and high sex bias in recombination (figures 1 and 2 and table 2), suggesting that haploid selection in males could be an important selective pressure for sex differences in recombination. However, distorted regions can vary widely even between populations of the same species [69], so distortion in a single cross should be interpreted with some caution. Furthermore, patterns of biased pollen or pollen-tube expression do not show similarly consistent enrichment in regions of low male recombination except for the sex chromosome (table 2 and electronic supplementary material, figure S6), although direct observation of transmission distortion likely provides a stronger indicator of loci involved in pollen competition. Taking these caveats into account, our results provide some evidence for the hypothesis that pollen competition may play a role in sex differences in recombination.

We used both pairwise correlations and regression models to investigate various genomic correlates of male and female recombination and sex bias in recombination, in order to further explore other possible factors favouring sex differences in recombination. It is important to note that because numerous factors can contribute to recombination rate heterogeneity and may show collinearity with each other and because recombination rate can in turn influence genome structure [3,16], any inference of causal factors should be treated with caution. On a genome-wide scale, both male and female recombination rates in R. hastatulus follow the widely observed patterns that genes and class 2 DNA TEs concentrate in high-recombination regions and class 1 RNA TEs concentrate in low-recombination regions (figure 2; [3,16,59]). In addition, these patterns are consistent with recombination in plants occurring preferentially upstream of genes, and with epigenetic silencing of retrotransposable elements causing a reduction of recombination, although this pattern could also be explained if transposable elements preferentially accumulate in regions of low sex-averaged recombination [3].

Because the recombination landscape varied widely between the chromosomes of R. hastatulus, we also sought to identify chromosome-specific correlates of recombination rates and recombination rate bias using linear models and partial correlations controlling for gene density. At this more granular scale, a different and more complex picture emerged (table 2 and electronic supplementary material, figures S8 and S9). In particular, our logistic regression models of recombination bias direction support the possibility that different mechanisms may be contributing to variation in recombination rates on different chromosomes. Both the number of genes per window and their position along the genome predicted female-biased recombination on some chromosomes and male-biased recombination on others. Aside from physical position and gene density, diverse factors predicted both male and female bias on different chromosomes, including RNA TEs, number of genes expressed in different tissues, sex-biased floral expression, pollen-biased expression and transmission ratio distortion (figure 1c). These findings suggest a general picture in which positional effects, the chromosomal positions of gene density and possibly centromeres may shape recombination on a ‘global scale’, but variation in gene content and haploid selection could lead to region-specific selective forces acting on both male and female recombination rates. Large-scale comparative data on sex differences in recombination rate landscapes remain rare, but a recent comparative study in fish [70] found that sex differences in recombination are labile at the species level yet do not exhibit clear trends consistent with adaptive hypotheses across species. This could be explained by multiple selective forces shaping sex differences in recombination rate over shorter microevolutionary timescales in ways that vary between chromosomes and populations and do not result in consistent patterns over longer macroevolutionary timescales.

5. Conclusion

Our study has provided some of the first evidence of sex differences in recombination and identified one of the largest known SDRs in a dioecious plant, or in fact in any eukaryote [66]. We identified both genome-wide and chromosome-specific factors correlated with sex differences in recombination, and also found evidence consistent with a possible role for male gametophytic selection in driving these differences. Future work in this system should allow more precise dissection of the genetic and evolutionary mechanisms favouring sex differences in recombination landscapes. In particular, exploring both sex-specific expression quantitative trait locus (eQTL) positions and further study of transmission ratios in pollen and seeds will allow us to further differentiate between SA cis epistasis in diploids and epistasis in haploids [1]. Finally, characterizing sex-specific recombination landscapes of hermaphroditic Rumex species should make it possible to determine whether these sex differences in recombination landscape did indeed precede and promote the evolution of a heterogametic XY sex-determining system with a very large SDR, as we have hypothesized.

Data accessibility

Raw sequence has been deposited on the Sequence Read Archive (SRA) under the accession number PRJNA692236 (embargoed until 1 July 2022 or publication). Our new genome assembly, transcriptome annotation, rDNA annotation and TE annotation have been deposited in the CoGe comparative genomics platform at https://genomevolution.org/coge/GenomeInfo.pl?gid=62326. Scripts used in the analyses have been deposited on Github at https://github.com/joannarifkin/Rumex-sex-specific. The data are provided in electronic supplementary material [71].

Authors' contributions

J.L.R.: conceptualization, data curation, formal analysis, investigation, methodology, project administration, writing— original draft, writing—review and editing; S.H.: data curation, formal analysis, investigation, writing—review and editing; M.Y.: data curation, formal analysis, investigation, writing—review and editing; B.M.S.: data curation, formal analysis, investigation, writing— review and editing; B.I.C.: formal analysis, investigation, methodology, writing—review and editing; Y.G.: formal analysis, investigation, writing—review and editing; P.R.: formal analysis, investigation, methodology, writing—review and editing; S.C.H.B.: conceptualization, funding acquisition, investigation, project administration, writing—original draft, writing—review and editing; S.I.W.: conceptualization, formal analysis, investigation, project administration, writing—original draft, writing—review and editing. All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Competing interests

We declare we have no competing interests.

Funding

This research was supported by Discovery grants from the Natural Sciences and Engineering Research Council of Canada to S.C.H.B. and S.I.W. J.L.R. was supported by an EEB post-doctoral fellows. P.R. was supported by the Academy of Finland (grant 343656).