A novel allele of ASY3 promotes meiotic stability in autotetraploid Arabidopsis lyrata

In this study we performed a genotype-phenotype association analysis of meiotic stability in ten autotetraploid Arabidopsis lyrata and A. lyrata/A. arenosa hybrid populations collected from the Wachau region and East Austrian Forealps. The aim was to determine the effect of eight meiosis genes under extreme selection upon adaptation to whole genome duplication. Individual plants were genotyped by high-throughput sequencing of the eight meiosis genes (ASY1, ASY3, PDS5b, PRD3, REC8, SMC3, ZYP1a/b) implicated in synaptonemal complex formation and phenotyped by assessing meiotic metaphase I chromosome configurations. Our results reveal that meiotic stability varied greatly (20-100%) between individual tetraploid plants and was associated with segregation of a novel allele orthologous to the budding yeast RED1 chromosome axis protein, Asynapsis3 (ASY3), derived from A. lyrata. The adaptive ASY3 protein possesses a putative in-frame tandem duplication (TD) of a serine-rich region upstream of the coiled-coil domain that has arisen at sites of DNA microhomology. The frequency of multivalents observed in plants homozygous for the ASY3 TD haplotype was significantly lower than plants heterozygous for TD/ND (non-duplicated) ASY3 haplotypes. Chiasma distribution was significantly altered in the stable plants compared to the unstable plants with a shift from proximal and interstitial to predominantly distal locations. The number of HEI10 foci at pachtyene that mark class I crossovers was significantly reduced in meiotic nuclei from ASY3 TD homozygous plants compared to ASY3 ND/TD heterozygotes, indicating an adaptive consequence of the ASY3 TD allele. From the ten populations, fifty-eight alleles of these 8 meiosis genes were identified, demonstrating dynamic population variability at these loci which nevertheless exhibit signatures of strong hard selective sweeps. Widespread chimerism between alleles originating from A. lyrata/A. arenosa and diploid/tetraploids indicates that this group of rapidly evolving genes provide precise adaptive control over meiotic recombination in the tetraploids, the very process that gave rise to them. Author summary Whole genome duplication can promote adaptability, but is a dramatic mutation usually resulting in meiotic catastrophe and genome instability. Here we focus on a case of coordinated stabilization of meiotic recombination in ten autotetraploid Arabidopsis lyrata and A. lyrata/A. arenosa hybrid populations from the Wachau region and East Austrian Forealps. We fuse population genomic data with a genotype-phenotype association study, concentrating on the effects of eight meiosis genes (ASY1, ASY3, PDS5b, PRD3, REC8, SMC3, ZYP1a/b) implicated in synaptonemal complex formation in the tetraploids under extreme selection. Our analysis demonstrates that a novel allele of the meiotic chromosome axis protein Asynapsis3 that contains an in-frame duplication of a serine-rich region is the major determinant of male meiotic stability. This adaptive restabilisation appears to be achieved by a reduction in the number of meiotic crossovers as well as a shift in their positioning towards the chromosome ends. Of the eight genes, fifty-eight alleles were identified, indicating dynamic population variability at these loci under extreme selection. In addition, widespread allelic chimerism between alleles originating from A. lyrata/A. arenosa and diploid/tetraploids indicates that this group of rapidly evolving genes provide precise adaptive control over meiotic recombination in the tetraploids, the very process that gave rise to them.


Introduction
Whole genome duplication (WGD) occurs in all eukaryotic kingdoms, and is associated with adaptability, speciation and evolvability [1,2]. At the same time, it is also one of the most dramatic mutations observed, usually resulting in catastrophic problems during meiosis, when ensuring stable chromosome segregation and genome integrity is paramount [3]. Because efficient meiosis is required for the formation of euploid gametes during sexual reproduction, selection acts strongly on standing variation from the progenitor diploids in newly arisen polyploids.
In allopolyploids (formed by both genome duplication and interspecies hybridization), loci required for correct chromosome pairing and recombination have been identified in wheat [4], oil seed rape [5,6] and Arabidopsis suecica [7]. However, in autopolyploids (which form within-species, without hybridization), there has been no functional confirmation of any gene controlling correct chromosome pairing, synapsis and crossing over (CO), although we have detected clear signatures of extreme selection in eight meiosis genes associated with the synaptonemal complex (SC) (ASY1, ASY3, PDS5b, PRD3, REC8, SMC3, ZYP1a, ZYP1b) in the young autotetraploid Arabidopsis arenosa [8].
The SC is a tripartite protein structure consisting of two lateral elements and a central element, specific to meiotic prophase I that is required for normal levels of COs in the majority of sexually reproducing eukaryotes [9]. In Arabidopsis, the chromosome axes (which come to form the SC lateral elements) consist of a scaffold of cohesin proteins (SMC1, SMC3, PDS5, REC8 and SCC3) [10][11][12][13][14] that organise sister chromatids into a loop/base conformation [15]. PRD3, the budding yeast MER2 homolog, is required for double-strand break (DSB) formation and is not an SC protein per se but may juxtapose the potential DSB site with the chromosome axis to promote inter-homolog recombination [16,17]. In Sordaria, MER2 also transfers and releases recombination complexes to and from the SC central region [16]. The meiosis specific proteins ASY1, ASY3 and ASY4 load onto the chromosome axis defined by the cohesin scaffold, to promote inter-homolog recombination [18][19][20]. ASY1 and ASY3 are the functional homologs of budding yeast HOP1 and RED1 and, HORMAD1/2 and SCP2 in mammals, respectively, that facilitate correct chromosome pairing and synapsis, required for wild-type COs [19, 5 | P a g e 21,22]. In Arabidopsis, synapsis is initiated by installation of the transverse filament proteins ZYP1a/b between homologous chromosomes, thus ensuring appropriate levels of COs [23].
Arabidopsis lyrata and A. arenosa represent powerful models for investigating adaptation to autopolyploidy, particularly their populations from the eastern Austrian Forealps, where interspecific hybridization and introgression is frequent [8] [24][25][26]. As a consequence, these populations represent 'natural mapping experiments' that can be studied to understand the relative contributions of the suite of alleles known to exhibit strong signatures of selection. These eight meiosis loci displaying highly differentiated alleles in A. arenosa were also reported in A. lyrata autotetraploids, along with signatures of extensive bidirectional gene flow [25]. At meiotic metaphase I in A. arenosa, chiasma frequency was reduced in autotetraploids carrying the derived alleles compared to the diploids, indicating an ongoing adaptive consequence of their evolution [8]. However, thus far, no formal confirmation of a direct effect on meiotic stabilisation in tetraploid A. arenosa or A. lyrata has been attributed to these evolved alleles.
In this study we fuse genomic, genetic and cytological approaches to investigate the effects of rapidly evolved adaptive haplotypes in these meiosis genes under strong selection. We measure the consequences of alternative evolved haplotypes at these loci in autotetraploid A. lyrata, A. arenosa, and natural introgressants of these species across a hybrid zone. Our analysis reveals functional evidence of a novel ASY3 haplotype that modulates meiotic recombination in both A. lyrata and A. arenosa autotetraploids, thus stabilising chromosome bivalent formation and genomic integrity.

A metaphase 1 analysis to determine meiotic stability in A. lyrata/A. arenosa populations
A thorough examination of meiotic stability was performed on fifty-two plants obtained from individual maternal seed lines sampled from tetraploid populations covering a range of known 6 | P a g e genomic backgrounds and demographic histories from the Wachau region and east Austrian Forealps.
Relative genomic contributions (proportions of admixture) from A. lyrata and A. arenosa from sampled populations had previously been determined [25]. Sampled populations were: LIC, MOD, PIL, SCB, KAG, ROK (A. lyrata and A. lyrata-like hybrids, that contain >50% genomic contribution from A. lyrata) and, TBG, SEN and WEK (A. arenosa dominant/A. lyrata introgressants, that contain >50% genomic contribution from A. arenosa) (Fig 1). Meiotic stability was assessed in individual plants by performing cytological analyses on metaphase I (MI) chromosome spreads of pollen mother cells (PMCs). A rod bivalent forms when only one chiasma (the cytological manifestation of a CO) connects a homologous chromosome pair (Fig 2A, D1-3). A ring bivalent forms when chiasmata occur in both chromosome arms of homologous pairs (Fig 2A, D4). Quadrivalents are structures formed of four chromosomes, usually two pairs of homologous chromosomes and, multivalents form between multiple chromosomes either by chiasmata or interlocks ( Fig 2C). As a control, chiasma frequency and distribution was scored at MI in diploid A. lyrata PMCs (Fig 2A). For the tetraploids, MI nuclei were scored as stable when 16 individual bivalents could be observed aligned on the MI plate and unstable if quadrivalents or multivalents were observed (Fig 2B -2C and S1 Fig). For each maternal line we scored blind the percentage of stable versus unstable nuclei, revealing a range from 20-100% ( Fig 2E).
Furthermore, we scored chiasmata as distal, interstitial or proximal to the centromere based on chromosome bivalent shapes (Fig 2D1-4). In unstable nuclei, only a proportion of individual bivalents could be scored per nucleus (ranging from 1-11 per cell), that were not associated with other chromosomes (Fig 2C, dashed ellipse). A FISH analysis utilizing the 5S and 45S rDNA probes revealed that in the samples of unstable nuclei all chromosomes that could be scored were observed both associated with multivalents and unassociated with multivalents. From the total sixteen pairs of chromosomes per nucleus, seven did not hybridize with the 5S and 45S rDNA probes, five with the 5S only, two with 45S only, and two with both 5S and 45S. However, the FISH analysis revealed that there was a chromosome bias for those that could be scored in the unstable nuclei with an underrepresentation of chromosomes containing the 45S rDNA nucleolar organizing regions (NOR) 7 | P a g e and an overrepresentation of chromosomes without the 45S rDNA. The expected frequency for bivalents that could be scored from the seventy seven chromosome counts based on a random expectation of occurrence from the proportion of labelled/unlabelled chromosomes was 34, 24, 9.5 and 9.5 for no probes, 5S only, 45S only, and both 5S and 45S, respectively, and the observed values were 46, 21, 6 and 4. A Chi-squared test revealed that these values were significantly different ([3] 2 =9.54, P < 0.05), indicating that chromosomes with the 45S rDNA NOR were more likely to form multivalents ( Fig 2C). As all chromosomes that could be scored were present in the sample from unstable nuclei, these were randomly grouped into pseudo-nuclei containing sixteen bivalents to determine chiasmata frequency and distribution, although with the caveat that these were less likely to form multivalents. Overall, significantly more chiasmata were observed in diploid A. lyrata MI bivalents than those from tetraploid stable or unstable nuclei (1.52±0.3, n=312; versus 1.12±0.2, n=960 and 1.26±0.3, n=590, respectively, Mann Whitney Test, P < 0.001). The frequency of distal chiasmata was not significantly different between bivalents in diploid and stable tetraploid nuclei, but was reduced in bivalents from unstable tetraploid nuclei ( Fig 2F). Bivalents from stable tetraploid nuclei had significantly fewer interstitial and proximal chiasmata compared to bivalents from diploids and those from unstable nuclei, whereas interstitial and proximal chiasmata were not significantly different between bivalents from diploids and unstable tetraploid nuclei ( Fig 2F).
A HEI10 immunocytological analysis was performed at late pachytene to confirm whether there was a difference in CO frequency between stable and unstable nuclei (Fig 2G-I). MAU8.11 (98% stability) and SEN2.2 (21% stability) were selected as extreme examples of meiotic stability (S1 Table).
During pachytene HEI10 marks class I CO sites [27] and in our analysis HEI10 also marked heterochromatic DNA, which was not scored as designated class I CO sites. In stable MAU8.11 nuclei, an average of 20.4 HEI10 foci per pachytene (n=30) were scored and in the unstable SEN2.2 nuclei, an average of 22.5 HEI10 (n=30) were scored, revealing that the unstable nuclei had significantly greater numbers of HEI10 foci (Wilcoxon rank sum test, P<0.05) (Fig 2I). Stable nuclei contained an average of 1.28 HEI10 foci/bivalent, whereas the unstable nuclei contained an average of 1.4 HEI10 foci/bivalent.

Association of haplotypes with meiotic stability in A. lyrata/A. arenosa tetraploids
The fifty-two tetraploid plants phenotyped for male meiotic stability were then genotyped for the proportion of each meiosis gene haplotype by high-throughput sequencing. Accurate genotyping required obtaining precise population reference sequences from published genomic data from these populations [25]. Degenerate primers were used to amplify full length gene amplicons of the eight meiosis genes from the fifty-two tetraploid plants (including exons and introns) for construction of Nextera LITE libraries. Libraries were barcoded per plant and sequenced by MiSeq, generating an average sequence depth across all loci of >2000x, from which we determined the proportion of each haplotype per plant by SNP frequency. Because the coding regions of all eight meiosis genes from representative populations were cloned and Sanger sequenced (described below in 'Adaptive polymorphisms in meiosis genes'), it was possible to resolve individual haplotypes. As these plants were drawn from a diversity of wild populations it was not surprising that an average of 7.25 alleles were identified for each of the eight meiosis genes. Given the limited sample size of fifty-two individuals it was not possible to statistically interrogate associations between all individual haplotypes for the eight genes. Consequently, similar haplotypes were collapsed together and classified into two groups: haplotypes with derived tetraploid alleles, and those with ancestral diploid alleles. Derived tetraploid haplotypes were those possessing conserved polymorphisms compared to the diploid reference sequences. For each gene (except ZYP1b, which was homozygous in all populations tested) a large proportion of the individuals carried four derived tetraploid haplotypes, whilst the others carried a mixture of derived tetraploid and diploid haplotypes. We therefore tested whether the presence of diploid haplotypes influenced meiotic stability. To do this, we classified the allele state at each of the eight meiosis genes in each individual tetraploid as either homozygous (i.e. exclusively either derived tetraploid haplotypes, or alternatively ancestral diploid haplotypes) or heterozygous (individuals harbouring both ancestral and derived alleles together at a given locus), and 9 | P a g e tested for any associations between these genotypes and meiotic stabilty by cytological analysis (Bonferroni corrected pairwise Mann-Whitney-Wilcoxon; S1 Table). This revealed that only the meiotic chromosome axis gene Asynapsis3 (ASY3) had a significant effect on meiotic stability (Fig 3).
Plants that were heterozygous for the ASY3 ancestral diploid haplotype and the derived tetraploid haplotype had significantly more unstable male metaphase I nuclei than plants homozygous for the derived ASY3 haplotype (4n Hom ~ = 88.9, IQR = 15.1, n= 41, ~ = 66, IQR =41.7, n=11, p=0.008). There was a large range of meiotic stability within the ASY3 heterozygotes, so smaller effects from the other seven meiosis genes cannot be excluded. For example SMC3 showed a trend, whereby the ancestral diploid allele may be associated with lower meiotic stability than the tetraploid homozygotes or diploid/tetraploid allele heterozygotes, although sampling sizes were not great enough to statistically confirm this trend.

Adaptive polymorphisms in meiosis genes
Previous studies have inferred polymorphic amino acids in meiosis genes between diploids and tetraploids in A. arenosa and A. lyrata by aligning short read sequences to the A. lyrata reference, but could not infer contiguous autotetraploid alleles [8,25,28]. To overcome this and resolve individual haplotypes, we amplified, cloned and Sanger sequenced the coding regions of the eight meiosis genes from diploid A. arenosa SNO and A. lyrata PER populations, tetraploid A. arenosa, SEN, TBG and WEK populations, as well as A. lyrata KAG and MAU populations. This approach provided high-resolution sequence polymorphism data for a total of three hundred and twenty cDNA transcripts, consisting of fifty-eight alleles with an average of seven alleles for each of the eight meiosis genes, thus allowing us to detect both structural variation including indels and divergent SNP variation ( Fig S2, S2-4 Tables). The eight encoded proteins associated with the synaptonemal complex are conserved at the secondary and tertiary amino acid level, rather than at the primary sequence [29]. It is therefore difficult to infer functional or non-functional amino acid polymorphisms with complete confidence, but results from KinasePhos2.0 and NetPhos3.1 suggest that overall 45% of those we detect were either loss or gain of putative serine/threonine phosphosites and 55% were nonphosphosites. The ASY3 derived adaptive allele and ZYP1b exhibited the greatest quantity of residue changes compared to the diploids that were conserved in tetraploid populations (45 and 44, respectively) (SFig 4), whilst SMC3 had none (although there were population specific SMC3 SNPs). Of the 45 polymorphic residues between the ancestral A. lyrata diploid ASY3 allele and the derived tetraploid ASY3 allele, 27 were due to a tandem duplication (TD allele) in a serine-rich region of the protein. The serine-rich region is upstream of the coiled-coil domain, possessing putative ATM/CKII phosphosites and a predicted SUMO site (K556, GPS-SUMO), and the TD allele contains 14 serines in this region, compared to 7 in the ancestral A. lyrata non-duplicated allele (ND) ( Fig 4A). As the ASY3 TD allele had highest sequence similarity to the diploid A. lyrata ASY3 ND allele, to investigate the provenance of the adaptive tandem duplication, we exhaustively screened the local diploid A. lyrata population (PER) geographically adjacent to the A. lyrata autotetraploid LIC and MOD populations (Fig   1). One hundred and twenty-eight plants from the diploid A. lyrata PER population were screened, but the ASY3 TD allele was not identified, indicating the current absence (or vanishingly low frequency) of the TD allele, although it did reveal the presence of a deletion (DEL) allele at 7% frequency, where the entire serine-rich region is absent. We therefore cloned and sequenced genomic DNA from ASY3 ND, TD and DEL and aligned them. This revealed a 78bp region of exon 2 is duplicated in-frame in the ASY3 TD allele that is missing in the ASY3 DEL allele between two AGAGA sites that possess DNA microhomology. We speculate that this DNA microhomology may have been instrumental in the formation of the ASY3 TD allele by homologous DNA repair through a replication error or during meiotic recombination (Fig 4B). 11   populations tested and 95% in tetraploid A. lyrata populations (Fig 5). The analysis revealed a small number of ancestral diploid A. arenosa and A. lyrata alleles in these populations, except ZYP1b that was completely homozygous. In contrast, the ASY1 allele derived from diploid A. arenosa had a frequency of 94% in A. arenosa tetraploid and 93% in A. lyrata tetraploid, indicating bidirectional gene flow (Fig 5), as well as bidirectional gene flow of ancestral SMC3 and REC8 diploid alleles in the tetraploids (Fig 5).

Widespread chimerism in meiosis gene alleles
Our analysis of Sanger and MiSeq data identified a novel chimeric allele of ZYP1b in all tetraploid populations; chimeric PRD3 in SCB, SEN and WEK; and chimeric PDS5b in ROK. While rare chimeras could potentially result from PCR artefacts [30], these alleles have the same breakpoints in multiple individuals which suggests alterations in the genomic DNA through homologous DNA repair in planta [31]. At the 3' end of all ZYP1b tetraploid alleles we detect evidence for a 474bp gene conversion (GC) to ZYP1a (S7A Fig). There is also evidence of GC in PRD3 between A. arenosa and A. lyrata ancestral diploid alleles (S7B Fig). In eleven plants from A. arenosa and A. lyrata populations, the first 740bp of the diploid PRD3 allele is more similar to the diploid A. arenosa than to A. lyrata (7 vs 21 SNPs), while the remaining 625bp of coding sequence has a higher similarity to diploid A. lyrata than to A. arenosa (5 vs 24 SNPs). In PDS5b, at the 5' end of tetraploid alleles from five A. arenosa plants we observe evidence for a GC of the first two exons to the diploid allele, providing evidence of GC (or CO) between ploidy levels (S8A Fig). In addition, analysis of genome resequencing data [25] revealed a 3' GC from a diploid ASY1 A. lyrata allele to a tetraploid A. arenosa allele in the KAG population (S8B Fig). The widespread presence of such evidence of gene conversion products in these loci exhibiting the most dramatic signatures of selection suggests a mechanism by which the peaks of differentiation generally found in this system are so narrow [3,8,25,32,33], despite a recent origin, and possible bottleneck, of the tetraploids [25].

Discussion
Here we aimed to determine the impact of strongly selected meiosis alleles that underwent recent selective sweeps on the rapid evolution of autotetraploid meiotic stability in A. lyrata/A. arenosa hybrids and introgressants and to trace their evolutionary origin. By associating genotypic and cytological phenotypic data we provide evidence that ASY3 is the major locus currently stabilising autotetraploid male meiosis in these populations. We identified structural variation of meiosis alleles including a novel derived, ASY3 allele with a tandem duplication (TD) in a serine-rich region that underlies the stable chromosome meiotic phenotype in the tetraploids, as well as novel ASY1, PDS5b, PRD3 and ZYP1b chimeric alleles between diploids and tetraploids and A. arenosa and A. lyrata origins.
A cytological metaphase I (MI) analysis revealed that chiasmata in the autotetraploids were significantly reduced in both stable and unstable nuclei compared to diploid A. lyrata. Moreover, chiasma frequencies in meiotically stable nuclei were significantly reduced in regions proximal and interstitial to the centromere. A shift in chiasma distribution may reflect a fundamental mechanism 13 | P a g e for meiotic adaptation to autopolyploidy [34]. The unstable nuclei occur due to unregulated meiotic recombination between multiple chromosomes, either homologous or non-homologous. All chromosomes appeared to associate with multivalents, although there was a bias for the 45S rDNA containing chromosomes, which may be due to the NORs clustering during prophase I, thus bringing non-homologous chromosomes into close proximity. A reduction in numbers of chiasmata in stable nuclei was supported by an immunocytological approach counting HEI10 foci that mark designated class I COs. HEI10 foci numbers were significantly lower in the stable pachytene nuclei (20.4 HEI10 foci per cell) compared to the unstable nuclei (22.5 HEI10 foci per cell)(P<0.01).
Our genotype-phenotype association study revealed that among these eight meiosis genes, the allele state of the structurally variable meiotic chromosome axis protein ASY3 was the major factor governing whether nuclei were stable. We hypothesise that the ancestral diploid ASY3 ND allele promotes high levels of proximal and interstitial chiasmata, but in the tetraploid it acts dominantly over the evolved ASY3 TD allele, promoting interstitial and proximal chiasmata as well as complex chromosome structures including multivalents. Such multivalents have previously been observed in diploid A. thaliana ZYP1 RNAi lines where the authors postulated that chiasmata may have formed between extensive duplications with high sequence similarity on non-homologous chromosomes [23].
The tandemly duplicated serine-rich region in the ASY3 TD allele may function in a manner similar to the budding yeast N-terminus MSH4 degron that destabilizes the protein until it is phosphorylated [35]. Further studies are required to determine if the serine-rich region destabilizes the ASY3 protein thus creating a hypomorphic variant or whether these sites are phosphorylated. The analysed brassica SC phosphoproteome [36] did not recover peptides for ASY3 in the serine-rich region, although similar (serine-aspartic acid) residues recovered from ASY1 were phosphorylated. The ASY3 TD allele also contains 19 derived residues outside the serine-rich region, of which 10 are predicted phosphosite gains or losses, that cannot be ruled out as functionally important along with unknown trans effects.
Chiasmata are distalized in A. thaliana chromosome axis mutants asy1, asy3 and asy4, presumably due to telomere proximity enabling sufficient inter-homolog pairing, whereas a complete meiotic axis 14 | P a g e is required to promote high levels of recombination between spatially separated regions in nuclei along the arms of the chromosomes [19,20,37].
From our data and extensive literature we can speculate a model wherein the serine-rich duplicated region destabilizes the ASY3 TD protein and may target the protein for degradation by the proteasome [35, [38][39][40]. A destabilized protein may be hypomorphic to the ancestral ASY3 ND allele as there less protein may be available to bind to the chromosome axes and/or it could be less effective at promoting interhomolog recombination when bound at the axis. A hypomorphic protein may act similarly to ASY1, ASY3 and ASY4 mutants in distalizing chiasmata [19,20,37]. However, when heterozygous with the dominant ancestral ASY3 ND allele, rates of inter-homolog and nonhomologous recombination increase. We hypothesise that axis components that favour interstitial and proximal recombination in diploids promote associations with non-homologous chromosomes in the tetraploids, especially in regions with high sequence homology. However, for stable bivalents, once a distal CO site is designated, CO interference may prevent further COs forming. In budding yeast, the TopoII interference pathway requires SUMOylation of TopoII and Red1, the ASY3 orthologue [22,41]. The serine-rich duplicated region in ASY3 possesses a predicted SUMO site, which could play a role in protein function, although this hypothesis needs functional testing.
Phylogenetic analysis revealed that the ASY3 TD allele most likely originated from diploid A. lyrata, consistent with [25]. A screen of the diploid A. lyrata population geographically adjacent to the tetraploids for the ASY3 TD did not detect the allele, but instead identified one with the same region deleted (DEL). The ASY3 DEL allele coding region is in-frame and contains a second deletion in the coiled-coil. It may be a coincidence that the same region is lost and gained in the ASY3 alleles, or that this sequence is more susceptible to DNA replicative errors. The region of interest is flanked by DNA microhomology (AGAGA) that is positioned at the putative exchange points and may have led to DNA polymerase slippage or aberrant replication fork repair. It is unlikely to have occurred during meiotic recombination due to the low level of sequence homology. However, we observe numerous examples of apparent gene conversion that appear to have arisen during meiotic recombination in ASY1, PRD3, PDS5b and ZYP1a/b between both species and ploidies. The bidirectional gene flow between A. arenosa and A. lyrata tetraploids has enlarged the gene pool for beneficial alleles to be borrowed and selected upon and the novel gene converted chimeric alleles may precisely coalesce advantageous sequences from differing origins for adaptation, although this would require further testing with an even larger sample size.
Evidence suggests that the origin of the adaptive ASY3 TD allele in the tetraploid populations is relatively recent, but as it is under extreme selection, it has spread extensively and introgressed within tight boundaries in the A. lyrata/A. arenosa hybrid genomes tested [25]. Our current analysis has provided haplotype-level sequence data that supports this hypothesis. The orthologous diploid ASY3 alleles from A. arenosa and A. lyrata are highly divergent, and yet the tetraploid ASY3 TD allele is very similar among tetraploid populations. We speculate that is therefore possible that preceding the origin of the ASY3 TD allele, gene flow of adaptive alleles from A. arenosa (ASY1, PRD3, REC8, SMC3 and ZYP1a/b) was necessary to establish meiotic stability in the nascent A. lyrata tetraploids, but has since been relaxed due to the presence of the predominant ASY3 TD allele. thaliana female compared to male [44]; d) in the case of SMC3, diploid alleles may be beneficial for male meiosis; e) inability to purge genetic load in autotetraploids [45] and; f) limited effect on overall male pollen fecundity, due to an excess of grains transmitted, despite variable quality.
Taken together, our data indicate multiple mechanisms for rapid meiotic evolution in autotetraploid A. lyrata. They reveal a predominant association of a duplication of the serine-rich region in the ASY3 TD allele with MI stability. Furthermore, tetraploid A. lyrata has introgressed ASY1 PRD3, REC8, SMC3 and ZYP1a/b alleles from A. arenosa by gene flow. Finally, novel chimeric genes of ASY1, PDS5b, PRD3, and ZYP1a/b have arisen evidently through gene conversion, suggesting highly dynamic mechanisms to generate variation that may be selected upon by evolution to ensure meiotic success in these populations.

Cloning and sequencing of meiosis gene transcripts
Plants were grown from seed to obtain fresh flower buds from diploid A. arenosa SNO and A. lyrata PER populations, and tetraploid A. arenosa, SEN, TBG and WEK and A. lyrata KAG and MAU populations [25]. These buds were collected, flash frozen in liquid nitrogen and stored at -80C until RNA extraction. Total RNA was extracted using a Bioline ISOLATE II RNA Plant Kit (Bioline Ltd, London, UK), following manufacturer's instructions, eluting into a final volume of 60 μl nuclease free water.
Concentration and purity were determined using a NanoDrop spectrophotometer (LabTech International, Lewes, UK) and one microgram of total RNA was electrophoresed on a non-denaturing 1% (w/v) agarose gel to check for degradation. First strand cDNA was reverse transcribed from 0.5 μg of total RNA using a Quantitect Reverse Transcription Kit (Qiagen, Hilden, Germany) that incorporates genomic DNA removal prior to reverse transcription. The coding regions of 8 meiosis genes (ASY1, ASY3, PRD3, PDS5b, REC8, SMC3, ZYP1a and ZYP1b) were amplified by PCR using 0.2 μM primers (S Table 5) designed using HiSeq data [33] and Platinum ™ Taq DNA Polymerase High Fidelity (ThermoFisher Scientific, MA, USA). PCR conditions were as follows: 94 °C for 2 min, followed by 35 cycles of 94 °C for 15 s, 60-65 °C for 30 s and 68 °C for 2-5 min (see S5 Table), with a final extension of 68 °C for 5-10 min. PCR products were electrophoresed on a 2% (w/v) agarose gel, and single bands of the expected size were excised and purified with a Monarch ® DNA Gel Extraction kit (New England Biolabs, MA, USA). Purified PCR products were cloned into pCR-XL-TOPO ™ vector using a TOPO ™ XL PCR Cloning Kit following the manufacturer's instructions. For each gene a total of 8 clones from each plant were isolated from overnight LB cultures using an ISOLATE II Plasmid Mini Kit (Bioline) prior to sequencing with universal M13F and M13R primers by GATC Biotech (Konstanz, Germany). Nucleotide sequences of the cDNAs were processed in Geneious 11.1.2 (https://www.geneious.com) to remove vector and low-quality sequence before using BLASTN to search the June 2010 (v.1.0/INSDC) assembly of the North American A. lyrata reference genome [46] and NCBI nonredundant (nr) database for confirmation that the obtained cDNAs were the expected gene transcripts. Primer walking was then used to sequence the entire length of the transcript. For each meiosis gene, cDNAs from each population were aligned with the respective Ensembl gene predictions from the A. lyrata reference genome (S6 Table), and to act as outgroups, A. thaliana transcripts obtained from The Arabidopsis Information Resource (TAIR) using the MUSCLE 3.8.425 plugin in Geneious 11.1.2 with default settings [47]. Phylogenetic trees were constructed using a maximum likelihood (ML) method with PhyML v3.

Meiotic haplotype genotyping
Genomic DNA was extracted from leaf material of each of the 52 plants in the study using a DNeasy Plant Mini Kit (Qiagen) and eluting into 100 μl nuclease free water. Full length coding regions of each of the 8 meiosis genes (including introns) were amplified from this genomic DNA by PCR using Platinum ™ SuperFi ™ Green PCR Master Mix (ThermoFisher Scientific) and 0.5 μM primers (S5 Table) designed against the Sanger sequenced cDNA of the 8 meiosis genes (see above). PCR conditions were as follows: 98 °C for 30 s, followed by 35 cycles of 98 °C for 30 s, 60-63 °C for 10 s and 72 °C for 2.5-10 min (see S5 Table), with a final extension of 72 °C for 5-10 min. PCR products were electrophoresed on a 1% (w/v) agarose gel, and single bands of the expected size were excised and purified with a  Table) from the North American A. lyrata reference genome [46] using Geneious 11.1.2 Read Mapper (Medium sensitivity; 5 iterations and default settings). SNPs relative to the reference genome genes were then called (Minimum Variant Frequency 0.25; Maximum Variant P-value 6x10 -6 ) and used to identify and determine the proportion of 2n and 4n alleles for each gene per plant using a set of allele specific SNPs (as revealed from the Sanger sequencing of meiosis gene transcripts described above; S7 Table). Allele specific indels (eg ASY3 TD) were identified by associated SNPs.

Cloning and sequencing of ASY3 alleles
Genomic DNA was extracted from diploid A. lyrata PER plants as above and a short section of ASY3 was PCR amplified using 0.5 μM primers (S5 Table) and MyTaq ™ Red Mix (Bioline). PCR conditions were as follows: 95 °C for 2 min, followed by 35 cycles of 95 °C for 30 s, 69 °C for 30 s and 72 °C for 30 s, with a final extension of 72 °C for 5 mins. PCR products were electrophoresed on a 2% (w/v) agarose gel and the lower band (~125 bp) corresponding to a partial ASY3 DEL allele was excised and purified with a Monarch ® DNA Gel Extraction kit (New England Biolabs). Purified PCR products were cloned into pCR ™ 4-TOPO ® vector using a TOPO ™ TA Cloning ™ for Sequencing Kit (ThermoFisher Scientific) following manufacturer's instructions. A total of 12 clones were isolated, purified and Sanger sequenced as above. Nucleotide sequences were processed in Geneious 11.1.2 (https://www.geneious.com) to remove vector and low-quality sequence before aligning with ASY3 ND/TD transcript alleles as described above. Primers designed against the partial DEL sequence were used to obtain the 3' end of the transcript from 3 µg ASY3 ND/DEL heterozygous 2n A. lyrata (PER) floral bud total RNA using a GeneRacer ™ Kit (ThermoFisher Scientific) following manufacturer's instructions. Purified PCR products were cloned into pCR ™ 4-TOPO ® vector, sequenced and processed as above.
ASY3 ND, TD and DEL alleles were PCR amplified from genomic DNA extracted from diploid and tetraploid A. lyrata respectively using 0.2 µM primers designed against ASY3 cDNA sequences obtained above (S5 Table) and Q5 ® High-Fidelity DNA Polymerase (New England Biolabs). PCR conditions were as follows: 98 °C for 2 min, followed by 35 cycles of 98 °C for 10 s, 63°C for 30 s and 72 °C for 4 min, with a final extension of 72 °C for 10 min. PCR products were electrophoresed on a 1% (w/v) agarose gel, and single bands of the expected size were excised and purified as above.

Cytology
Chromosome spreads were performed [49] [25] on all populations used in this study. The HEI10 immunocytological analysis was performed using the protocol [50] with anti-AtSMC3 rat and anti-AtHEI10 rabbit antibodies described in [51]. Nikon Eclipse Ci and Ni-E microscopes installed with NIS Elements software were used to capture images of chromosomes.

Statistical analyses and Map drawing
Statistical analysis was performed using the R Stats package. Mann-Whitney Wilcoxon tests were performed with function wilcox.test. Bonferoni adjusted p values were calculated using the function p.adjust. The map was drawn using ggmap [55].

Data availability
All sequences in this study including cDNA transcripts and genomic DNA sequences have been        Table. Genotype and phenotype data 28 | P a g e S2