Natural variation in meiotic recombination rate shapes introgression patterns in intraspecific hybrids between wild and domesticated barley.

Meiotic recombination rates vary considerably between species, populations, and individuals. The genetic exchange between homologous chromosomes plays a major role in evolution by breaking linkage between advantageous and deleterious alleles in case of introgressions. Identifying recombination rate modifiers is thus of both fundamental and practical interest to understand and leverage variation in meiotic recombination rates. We investigated recombination rate variation in a large intraspecific hybrid population (named HEB-25) derived from a cross between domesticated barley and 25 wild barley accessions. We observed quantitative variation in total crossover number with a maximum of a 1.4-fold difference between subpopulations and increased recombination rates across pericentromeric regions. The meiosis-specific ɑ-kleisin cohesin subunit REC8 was identified as a candidate gene influencing crossover number and patterning. Furthermore, we quantified wild barley introgression patterns and revealed how local and genome-wide recombination rate variation shapes patterns of introgression. The identification of allelic variation in REC8 in combination with the observed changes in crossover patterning suggest a difference in how chromatin loops are tethered to the chromosome axis resulting in reduced crossover suppression across pericentromeric regions. Local and genome-wide recombination rate variation is shaping patterns of introgressions and thereby directly influences the consequences of linkage drag.


Introduction
Introgression is the transfer of genetic material between or within species through hybridization followed by backcrossing. Both interspecific and intraspecific introgressions are widespread in nature, as revealed by recent genome-scale sequence data (Mallet et al., 2016). As such, adaptive introgressions have been found in plants, humans, butterflies and birds (Heliconius Genome Consortium, 2012;Huerta-S anchez et al., 2014;Lamichhaney et al., 2015;Arnold et al., 2016). Hybridization between wild and domesticated species is commonly used in plant breeding to develop new cultivars possessing desirable traits (Hufford et al., 2012;Wendler et al., 2014;Maurer et al., 2015;He et al., 2019). The genomic landscape of introgressions is influenced by selection, drift and recombination (Martin & Jiggins, 2017). Introgressions may simultaneously introduce beneficial and deleterious alleles. In these cases, the efficacy of selection depends partially on the local recombination rate in order to break linkage between beneficial and deleterious alleles. Linkage drag is of hindrance in plant breeding, where deleterious alleles are introgressed along with a desired beneficial allele (Young & Tanksley, 1989). Therefore, both local and genome-wide recombination rate variation play an important role in shaping the evolution of naturally occurring hybrid genomes and the genetic gain achieved in plant breeding (Schumer et al., 2018;Tourrette et al., 2019).
In the present study, we explore the association of genotypes with variation in recombination rates of the nested-association mapping population HEB-25 (Maurer et al., 2015), which consists of 25 intraspecific subpopulations derived from hybridization between domesticated and wild barley (Hordeum vulgare (cv. Barke) 9 Hordeum vulgare subsp. spontaneum). We investigate how recombination rate variation directly affects genomic patterns of introgression. By comparing genome-wide recombination landscapes among the 25 subpopulations, we observe quantitative variation in the total number of crossovers as well as crossover patterning, with an increase across pericentromeric regions in high-vs low-recombining subpopulations. We identify the meiosis-specific a-kleising cohesin subunit REC8 as a putative genome-wide recombination rate modifier. Utilizing genome-wide recombination rate variation among the 25 subpopulations, we provide empirical evidence regarding how the size and distribution of introgressions is shaped by recombination. Together, our study links variation in recombination rate to allelic variation in a meiotic gene involved in tethering chromatin loops to the chromosome axis and provides experimental evidence for the relationship between recombination rate variation and patterns of introgression.

Plant material and SNP array genotyping
In this study, we used a large intraspecific hybrid population (HEB-25) derived from a cross between the domesticated barley cultivar 'Barke' (Hordeum vulgare L.) and 25 wild barley (Hordeum vulgare subsp. spontaneum (K. Koch) Thell) accessions previously developed by Maurer et al. (2015). This population is composed of 25 subpopulations with between 21 and 72 BC 1 S 3:8 lines per subpopulation (i.e. one backcross to the domesticated barley parent and eight generations of selfing), resulting in a total of 1340 lines. Plants were grown under field conditions from 2011 until 2018 in central Germany (Halle (Saale), 51°29 0 45.6″ N, 11°59 0 40.2″E). DNA of 12 pooled individual plants of each BC 1 S 3:8 line was extracted using the Qiagen BioSprint 96 DNA Plant Kit and a BioSprint workstation following the manufacturer's instructions. Single nucleotide polymorphism (SNP) genotyping was performed by TraitGenetics (Gatersleben, Germany) for all 1340 lines on the barley 50k iSelect SNP array . SNP markers that did not meet the quality criteria (polymorphic in at least one HEB family, < 10% failure rate, < 12.5% heterozygous calls) were removed from the dataset. Furthermore, 256 redundant SNPs were removed as they showed the exact same segregation among all HEB lines, indicating that they were in complete linkage disequilibrium (LD). Only one of these markers was kept. SNP raw data were deposited on e!DAL under the following https://doi.org/10.5447/ipk/2019/20 (Arend et al., 2014;Maurer & Pillen, 2019). Physical map positions of all SNPs according to the most recent reference sequence of the barley genome (TRITEX Morex V2) (Monat et al., 2019) were downloaded from https://ics.hutton.ac.uk/50k/.

Crossover counting, recombination landscape and introgression analysis
All the following analyses were done using basic functions of the R statistical environment if not stated otherwise (R Core Team, https://www.rproject.org/contributors.html).
To count crossovers, the entire SNP matrix containing raw data was first converted into numerical format, where the 'Barke' allele is coded as 0, heterozygous alleles are coded as 1 and wild barley alleles are coded as 2. The entire SNP matrix was then split into subpopulation matrices from which subpopulation-specific monomorphic markers were removed, as well as markers with missing values. Subpopulation SNP matrices were further filtered against a minor allele frequency of <0.1. Finally, to avoid false positive crossover counts caused by small structural variations or falsely assigned physical positions, we aggregated SNPs by counting the most common allele in sliding windows of a size of 20 consecutive SNPs using the SlidingWindow() function of the EVOBIR package (Blackmon et al., 2015) of the R statistical environment. Thereby, single markers giving rise to suspicious double crossovers were removed. Using the aggregated SNP matrix of each subpopulation, crossovers were then counted as changes in the allelic state along the physical length of the chromosome. The New Phytologist (2020) Ó 2020 The Authors New Phytologist Ó 2020 New Phytologist Trust www.newphytologist.com total crossover count (TCO) of each line was used to calculate the mean crossover number per subpopulation, and lines showing more than 60 crossovers (37 out of 1340 lines) were removed as outliers. Outliers were removed based on strong deviations from the population mean, that is outside the three-fold interquartile range. The 37 outliers showed a mean TCO of 103.19 compared to a population mean of 24.38, and an outlier-pruned population mean of 22.17 (Supporting Information Fig. S2). ANOVA was performed to test for differences among the entire population and Tukey's honest significant difference (HSD) test was performed to test for multiple pairwise differences between subpopulations. Crossover data are summarized in Dataset S1. To analyse the recombination landscape of each subpopulation, we summarized recombination frequencies between pairs of SNPs in 1 Mb intervals multiplied by a factor of 100 to obtain cM Mb À1 values along the physical length of the chromosome. Our ability to detect recombination events decreased with each generation due to a decrease in heterozygosity of 50% per generation caused by self-pollination. Using this approach, we systematically underestimated the true per-generation recombination rate, because individuals of our HEB-25 population were genotyped after nine rounds of meiosis. Mitotic recombination events in the tissue which gave rise to meiocytes could potentially have contributed to the total number of crossovers detected as well. Therefore, recombination rates reported here should not be interpreted as a per-generation recombination rate. Recombination landscapes of each chromosome of each subpopulation were visualized by plotting recombination rates in 1 Mb intervals (cM Mb À1 ) against their physical position. To compare the genome-wide recombination landscapes of selected subpopulations, we calculated the mean recombination rate across all seven chromosomes in relative chromosomal intervals of 0.5%, where the physical length of each chromosome was normalized to range from 0 to 100%. We then split the relative length of the chromosome into three intervals (e.g. 0-25%, 25-75%, 75-100%) and used cM Mb À1 values in 0.5% intervals to perform Welch's two-sample t-tests to infer significant differences between subpopulations. The same approach was used to compare SNP densities between subpopulations, for which SNPs were counted in 0.5% intervals (Fig. S1).
To analyse introgression size and distribution, we used the same SNP matrix as for counting crossovers. Introgressions were detected as contiguous homozygous wild barley alleles or heterozygous positions along the chromosome. The mean distance between the first or last marker position of each contiguous region and the previous or next marker, respectively, defined the size of an introgression, and the mean of the first and last marker defined its central position. We calculated Spearman's rank correlation coefficient (⍴) between the size of an introgression (Mb) and the local recombination rate (LRR, cM Mb À1 ) in a given 0.5% chromosomal interval to analyse the correlation between introgression size and LRR. Similarly, we calculated Spearman's rank correlation coefficient (⍴) between genome-wide mean introgression size (Mb) and GWRR (cM Mb À1 ). GWRR was calculated by summarizing recombination frequencies between marker pairs along each chromosome (M, cM = M 9 100) and dividing by a total genome size of 5300 Mb.

Genome-wide association analysis
Genome-wide association analysis was performed using the total number of crossovers accumulated over nine meiotic divisions as a quantitative phenotypic trait. SNPs which were monomorphic in a specific subpopulation were coded as 0 (i.e. domesticated 'Barke' allele). Missing SNPs (0.77% of all SNPs) were imputed using the mean score of polymorphic flanking markers (matrix E (Maurer & Pillen, 2019), https://doi.org/10.5447/ipk/2019/20).
The scan for genome-wide marker trait associations followed a robust three-step procedure performed in SAS 9.4 (SAS Institute Inc., Cary, NC, USA). (1) The whole set of SNPs was included in a multiple linear regression model to find potentially associated SNPs based on stepwise selection with SAS PROC GLMSE-LECT. For this, 80% of phenotype data were randomly assigned to the prediction set and 20% were assigned to the validation set. SNPs were consecutively included in the final model as long as they were able to decrease the average square error (ASE) of phenotype prediction in the validation set. This procedure was repeated 100 times with independent random assignments and the number of times an SNP was included in the final model was recorded. Only those SNPs that were included in more than one out of 100 models were treated as robust enough to be included in the second step.
(2) A second model fitting was performed with the SNPs selected in step 1. This model included the whole phenotype dataset and stepwise selection of SNPs based on minimizing the Schwarz Bayesian Criterion (SBC (Schwarz, 1978)) with SAS PROC GLMSELECT. The list of selected SNPs, hereafter called cofactors, was then included in the third step. (3) Each single SNP was finally screened for significance by multiple linear regression with the cofactors modelled in the background with SAS PROC REG.
Additionally, we applied a second layer of cross-validation to remove the effects of local recombination rate modifiers (Jordan et al., 2018). Briefly, we screened only markers of a given chromosome for associations with the number of crossovers on the remaining chromosomes based on the procedure mentioned above. Using this approach, the effects of local recombination rate variation are reduced and trans-acting recombination rate modifiers are identified.

Candidate gene analysis
Candidate gene analysis was performed by manually screening all annotated high-confidence genes in a AE 0.5 cM window surrounding the quantitative trait locus (QTL). Genetic map positions were derived from Maurer et al. (2015). Furthermore, orthologues of known meiotic genes described by Mercier et al. (2015) were identified using PLAZA (https://bioinformatics.psb. ugent.be/plaza/versions/plaza_v4_5_monocots/) (Vandepoele et al., 2013;Van Bel et al., 2018). The physical positions of putative orthologues of known meiotic genes were used to assist candidate gene analysis. A candidate gene located within the 0.5 cM window surrounding the QTL was tested for allelic variation within the entire population using SNPs identified by exome capture. SNPs within the candidate gene were then tested for their association with the crossover phenotype, assuming that the wild barley allele is dominant in the case of rare heterozygous genotypes. First, a one-way ANOVA was performed followed by Tukey's honest significant difference test to analyse the effects of each SNP. Gene models were retrieved from the EnsemblPlants database (Release 46) (Cunningham et al., 2019). Gene expression data were retrieved from BARLEX (The Barley Genome Explorer; Colmsee et al., 2015;Mascher et al., 2017).

Recombination rate variation in 25 intraspecific hybrid populations between wild and domesticated barley
To explore natural variation in meiotic recombination rates, we took advantage of the barley population HEB-25 derived from a cross between one domesticated barley cultivar ('Barke') and 25 wild barley accessions (Badr et al., 2000;Maurer et al., 2015). Each of the 25 subpopulations was propagated for eight generations (BC1 1 S 3:8 ) via self-pollination resulting in a total of nine meiotic divisions over the course of which recombination events were accumulated. A total of 1340 lines (21-72 lines per subpopulation) were subjected to high-resolution SNP genotyping on the barley 50k iSelect array, yielding a total of 32 120 physically mapped SNPs Monat et al., 2019). After removing subpopulation-specific monomorphic SNPs and SNPs with a minor allele frequency of <0.1, a total set of 7174-11 246 SNPs with a mean inter-SNP distance of 386-669 kb was used to measure meiotic recombination events.
We measured an average of 22.17 crossovers (range 18.7-25.6) across all subpopulations, indicating quantitative variation in total crossover number almost resembling a normal distribution (Fig. 1b, Shapiro-Wilk normality test, P = 0.043). The average number of crossovers detected in population HEB-25 seemed lower than would be expected over nine meiotic divisions assuming an average of 10 crossovers per generation accumulated over nine meiotic divisions (Phillips et al., 2015;Dreissig et al., 2017). However, as heterozygosity decreased by 50% per generation in inbreeding barley (Morrell et al., 2003;Abdel-Ghani et al., 2004), so did our ability to detect crossovers.
Given the difference of 6.9 crossovers between the two subpopulations at the ends of the scale, we looked at each subpopulation's recombination landscape to understand the nature of this increase in crossover number (Tukey's HSD test, P = 0.0068, Fig. S3). Interestingly, comparing recombination rates (cM Mb À1 ) between the two most extreme subpopulations 23 and 24 along each chromosome showed a tendency towards an increase across low-recombining pericentromeric regions as well as distal subtelomeric regions in subpopulation 23 (Fig. 2a,c). This difference was not explained by a higher number of SNPs in one population, nor did we observe a correlation between differences in SNP number and the number of crossovers measured across the entire population (Fig. 2e,f; Pearson's r = À0.04, P = 0.08). Our observation was confirmed by summarizing mean recombination rates (cM Mb À1 ) across the low-recombining pericentromeric interval (25-75%) of all seven barley chromosomes, showing a difference of 0.13 cM Mb À1 (Welch's two-sample t-test, P = 0.042). Again, this increase in recombination rate did not stem from a difference in SNP number (Welch's twosample t-test, P = 0.513, difference in mean SNP number = 0.3 across the 25-75% interval). No significant increase was observed in the 75-100% distal region (+ 0.319 cM Mb À1 , Welch's twosample t-test, P = 0.089), and in the 0-25% region (+ 0.27 cM Mb À1 , P = 0.28).
Regarding the underlying causes, a multitude of genetic, epigenetic and structural factors influencing recombination rates have been described in various organisms. Allelic variation in procrossover factors, that is the ubiquitin E3 ligase HEI10 (Ziolkowski et al., 2017), or simultaneous mutation of anticrossover factors RECQ4a, RECQ4b and combination with increased HEI10 copy number , was shown to result in a massive increase in recombination rate in distal chromosomal regions in Arabidopsis, but not across the pericentromere. Between closely related Drosophila species, allelic variation in a single gene (MEI17/MEI18) explained large differences in both the rate and the patterning of recombination events (Brand et al., 2018). In plants, a shift in the patterning of recombination events was shown to be linked to DNA methylation in non-CG sequence contexts and histone 3 lysine 9 dimethylation (H3K9me) . A possibly confounding factor in population HEB-25 may be structural variations, as these were shown to affect crossover patterning at multiple scales, that is SNPs inhibiting crossovers at base-pair resolution, juxtapositioning of homozygous and heterozygous regions shifting crossovers into heterozygous regions, and inversions inhibiting crossovers at the megabase scale (Borts & Haber, 1987

Research
New Phytologist Schneeberger, 2020). As our SNP array data did not allow us to identify structural variations such as inversions, we instead focused on the possibly confounding effects of SNP density and total number of SNPs. A higher number of SNPs may simply increase the probability of detecting crossovers. Between the two populations showing the highest and lowest number of crossovers, the total number of SNPs was almost identical (11 125 and 11 294, respectively). Across the entire population, we did not observe a correlation between differences in SNP number and crossovers (Pearson's r = À 0.04, P = 0.85, Fig. 2f). There was, however, a strong correlation between SNP density and recombination frequency at the chromosome level (Fig. S4). This is in strong agreement with previous studies reporting positive correlations between SNP density, recombination frequency and gene density in other small-or large-genome plants Darrier et al., 2017;Jordan et al., 2018;Kianian et al., 2018;Gardiner et al., 2019;Rowan et al., 2019). Given the suppressive effects of chromosomal inversions, their rather low frequency and small size (few Mb) previously shown in barley make them unlikely to contribute to genome-wide differences in recombination patterning (Konishi & Linde-Laursen, 1988;Keilwagen et al., 2019). Another possibly confounding factor is the environment. Because population HEB-25 was propagated under field conditions, differences in flowering time may result in meiosis taking place at different temperatures, which would affect recombination rates (Phillips et al., 2015;Lloyd et al., 2018). To test this, we looked at the correlation between total crossover number and flowering time data reported by (Maurer et al., 2015). The absence of a correlation between flowering time and crossover number indicated that differences in flowering time probably did not result in meiosis taking place at extremely different temperatures (Pearson's r = À 0.048, P = 0.084, Fig. S5).
Although the effects of structural variations and environmental conditions could not be fully excluded in our population, we did

Identification of trans-acting recombination rate modifiers
To identify genetically determined recombination rate modifiers in our intraspecific hybrid population, we performed a genomewide association study (GWAS). We utilized quantitative variation in the total number of crossovers per line and associated these with the genotype of each line. Using this approach, we were associating the number of crossovers accumulated over nine meiotic divisions with the genotypic constitution of the final generation (BC 1 S 3:8 ), which did not allow us to determine whether a recombination rate modifier was present/absent or homozygous/ heterozygous throughout previous meiotic divisions. Therefore, as already argued by others, we are systematically underestimating the effects of recombination rate modifiers and are only detecting dominant modifiers with strong effects (Esch et al., 2007;Jordan et al., 2018;Gardiner et al., 2019). Furthermore, to identify trans-acting recombination rate modifiers rather than identifying common recombination breakpoints, we applied a cross-validation step in which SNPs of a given chromosome were only screened for associations with the number of crossovers on the remaining six chromosomes (Jordan et al., 2018). Finally, only those QTLs retained after cross-validation were considered transacting recombination rate modifiers. Throughout both types of cross-validation (Maurer et al., 2015;Jordan et al., 2018) we consistently detected one significant QTL on chromosome 1H (Fig. 3, Àlog 10 P = 9.1, Àlog 10 Bonferroni threshold = 6.5) with an effect size of the wild barley allele of + 42% (i.e. + 7.84 total crossovers, minor allele frequency of 0.03) that explained 3% of the total variation observed across the entire population (hereafter named 1H.recQTL1, Dataset S2).
By screening a total of 37 annotated high-confidence genes Monat et al., 2019) in a AE 0.5 cM (2 243 219 bp) window surrounding 1H.recQTL1, we identified the meiosis-specific kleisin cohesin subunit REC8 as a candidate gene genetically linked within 0.3 cM and physically separated by 1.8 Mb (Dataset S3). Hordeum vulgare REC8 (HvREC8) exists in two copies, HORVU.MOREX.r2.1HG0050790 and HORVU.MOREX.r2.1HG0075790, both located on chromosome 1H at 422 and 513 Mb, respectively. However, only the latter gene is closely linked to the significantly associated SNP within a distance of 1.8 Mb. Both REC8 genes show expression peaks in developing inflorescences (1-1.5 cm) and high similarity to REC8 orthologues in related Triticeae species (Figs S6, S7; Colmsee et al., 2015;Beier et al., 2017;Mascher et al., 2017). An essential prerequisite for a putative candidate gene is to show allelic variation within the population. Otherwise, if all individuals harboured the same variant of the gene, it would be unlikely to be the cause of recombination rate variation. Therefore, we used exome capture data of the entire population to search for SNPs within the presumably causative HvREC8 paralogue (i.e. HORVU.MOREX.r2.1HG0075790). We identified a total of five high-confidence SNPs located within two exons and two introns, which differentiate the HvREC8 variant of the domesticated cultivar from those of the 25 wild barley accessions (Fig. 3b,c). By looking at the effects of each SNP on total crossover number, we identified three out of five SNPs showing a significant effect on total crossover number (Fig. 3d, Tukey's HSD P < 0.05). Interestingly, both intron variants (SNP_2 and SNP_5) were associated with an increase in crossover number (+ 11.8 and + 6.8%, respectively) and occurred exclusively in different subpopulations (Fig. 3c). These intron variants may affect mRNA splicing and could be of regulatory function (Pagani & Baralle, 2004). In cattle, REC8 intron variants were also identified as trans-acting recombination rate modifiers, suggesting a regulatory mode of action potentially conserved across kingdoms (Sandor et al., 2012). Both intron variants seem to be located in conserved intron regions in related Triticeae species (Fig. S7). By contrast, a synonymous SNP (SNP_3) located in exon 9 of HvREC8 was associated with a reduction in crossover number (À24.3 %) and occurred in subpopulation 15 only (Fig. 3c). This SNP is located in a region where the barley genomic sequence differs from those of related Triticeae species. Whether these SNPs have a causative role in recombination rate variation or whether other not yet identified polymorphisms play a role cannot be answered here and remains to be addressed in future studies. However, our data support the idea of different HvREC8 variants being present in wild barley.
Regarding the function of REC8, it was recently shown that REC8 occupancy is enriched in gene bodies, exons, GC-rich sequences and H3K27me3-modified genes in Arabidopsis (Lambing et al., 2020). REC8 is involved in tethering chromatin loops to the chromosome axis and its abundance is highest in pericentromeric regions and correlates with crossover suppression (Kugou et al., 2009;Ito et al., 2014;Sun et al., 2015;Folco et al., 2017;Patel et al., 2019;Schalbetter et al., 2019;K€ ohler et al., 2020;Lambing et al., 2020). REC8 was also identified in sheep, cattle and red deer as a factor mediating quantitative differences in genome-wide recombination rates (Sandor et al., 2012;Johnston et al., 2016Johnston et al., , 2018. Interestingly, Sandor et al. (2012) also identified REC8 intron variants associated with recombination rate variation, suggesting alternative regulation of REC8 alleles as the causative agent. A different variant of REC8, altered gene expression or protein structure may have an effect on chromatin loop structure and therefore result in a pericentromeric region that is more permissive for crossovers to occur, as is evident in population HEB-25.

Patterns of introgression are shaped by local and genomewide recombination rate variation
In intra-or interspecific hybrids, the minor parent genome (i.e. the parental genome being introgressed) may introduce beneficial or deleterious alleles into the major parent genome (Brown et al., 1989;Martin & Jiggins, 2017;Schumer et al., 2018). Linkage drag (i.e. the simultaneous introgression of deleterious alleles along with beneficial alleles) is a result of alleles being genetically linked by low local recombination rates, a universal genetic phenomenon observed across kingdoms (Harris & Nielsen, 2016;Juric et al., 2016). It is of great hindrance in plant breeding and  . Asterisks indicate significant effects determined by ANOVA followed by Tukey's honest significant difference test (P < 0.05).
Ó 2020 The Authors New Phytologist Ó 2020 New Phytologist Trust New Phytologist (2020) www.newphytologist.com plays an important role in the evolution of hybrid genomes (Brown et al., 1989;Schumer et al., 2018;Martin et al., 2019). However, there is little empirical data exploring the relationship between recombination rates and patterns of introgression in intraspecific hybrid populations.
Given the amount of local and genome-wide recombination rate variation we observed in population HEB-25, we investigated how the size and position of introgressions were shaped over relatively few generations. Wild barley (minor parent) introgressions into the domesticated parent (major parent) genome were detected as contiguous wild barley alleles along the chromosome. We detected a total of 19 635 wild barley introgressions of varying size, ranging from < 1 to 670 Mb (Fig. S8). By plotting the size of each introgression against its position on the chromosome, we observed a striking relationship between local recombination rate and introgression size (Figs 4, 5a,b; Spearman's rank correlation ⍴ = À 0.9, P < 2.2 9 10 À6 ).
As one would intuitively assume, introgressions were smaller in regions of high recombination and larger in regions of low recombination. Heterozygous introgressions were generally smaller than homozygous introgressions, demonstrating that meiotic recombination events were continuously shrinking heterozygous regions until fixed as homozygous regions. Along the chromosome, introgressions in distal high-recombining regions ranged from < 1 to 200 Mb in size, whereas most introgressions in pericentromeric low-recombining regions ranged from 400 to 600 Mb. This shows how patterns of introgression were directly shaped by local recombination rate in the absence of selection. As a consequence, naturally occurring introgressions between wild and domesticated species would exhibit varying deleterious burdens on the major parent genome depending on local recombination rate. In strictly self-fertilizing species, such as barley, introgressions are rapidly fixed in a homozygous state and thereby permanently fixed in size as well. This is directly relevant to mapping genes or QTLs via LD (LD-mapping), as large introgressions in low-recombining regions do not permit accurate mapping.
In addition to local recombination rate variation, we also observed genome-wide differences. A genome-wide increase in recombination rate in a given population would predict smaller introgressions and thereby reduced linkage drag and increase the efficiency of selection (Martin & Jiggins, 2017). We summarized the genome-wide recombination rate (cM Mb À1 ) of each subpopulation and compared it to its average introgression size (Fig. 5c,d).
However, we did not detect strong correlations between genome-wide recombination rate and mean introgression size, and only heterozygous introgressions were significantly correlated with the genome-wide recombination rate (Spearman's rank correlation ⍴ = À0.41, P = 0.042). This probably reflects that local recombination rate variation along the chromosome is usually greater than genome-wide recombination rate variation between New Phytologist (2020) Ó 2020 The Authors New Phytologist Ó 2020 New Phytologist Trust www.newphytologist.com

Research
New Phytologist subpopulations (i.e. 0-8 cM Mb À1 along the chromosome and 0.35-0.55 cM Mb À1 between subpopulations). As such, rather small differences in genome-wide recombination rate do not strongly explain differences in genome-wide mean introgression size.
Taken together, we conclude that the effects of local recombination rate variation on patterns of introgression are stronger than those exerted by genome-wide differences, at least in this experimental population. This is expected due to stronger recombination rate variation along the chromosome than between individuals or populations. However, as we did observe similar tendencies for the effects of local and genome-wide recombination rate variation on patterns of introgression, we have been able to provide empirical evidence for a genomic phenomenon that was previously largely supported by intuition, rather than experimental data. Genome-wide recombination rate (cM Mb -1 ) Genome-wide homozygous introgression size (Mb) (d) Fig. 5 Correlation between recombination rate and introgression size. (a, b) Correlation between size of heterozygous (a) or homozygous (b) wild barley introgressions (Mb, y-axis) and local recombination rate (cM Mb À1 , x-axis) determined in relative chromosomal intervals of 0.5%. (c, d) Correlation between genome-wide recombination rate (cM Mb À1 , x-axis) and genome-wide mean size (Mb, y-axis) of heterozygous (c) or homozygous (d) introgressions for each subpopulation.
Ó 2020 The Authors New Phytologist Ó 2020 New Phytologist Trust New Phytologist (2020) www.newphytologist.com Project DEAL. The authors declare that they have no competing interests.

Author contributions
SD analysed the data regarding recombination, candidate genes and introgression, and conceptualized and wrote the manuscript. AM analysed raw SNP data, performed the GWAS, and wrote and edited the manuscript. RS, LM and AJF collected the exome capture data and edited the manuscript. TS analysed the exome capture data, and wrote and edited the manuscript. KP coordinated the project and edited the manuscript. All authors read and approved the final manuscript.

Supporting Information
Additional Supporting Information may be found online in the Supporting Information section at the end of the article.

Dataset S2 Summary of GWAS for recombination rate modifiers
Dataset S3 List of genes surrounding 1H.recQTL1 and respective transcription profiles across 16 tissues        Please note: Wiley Blackwell are not responsible for the content or functionality of any Supporting Information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office.