How do chromosomal regions with differing degrees of homology and homeology interact at meiosis? We provide a novel analytical method based on simple genetics principles which can help to answer this important question. This method interrogates high-throughput molecular marker data in order to infer chromosome behavior at meiosis in interspecific hybrids.
We validated this method using high-resolution molecular marker karyotyping in two experimental Brassica populations derived from interspecific crosses among B. juncea, B. napus and B. carinata, using a single nucleotide polymorphism chip.
This method of analysis successfully identified meiotic interactions between chromosomes sharing different degrees of similarity: full-length homologs; full-length homeologs; large sections of primary homeologs; and small sections of secondary homeologs.
This analytical method can be applied to any allopolyploid species or fertile interspecific hybrid in order to detect meiotic associations. This genetic information can then be used to identify which genomic regions share functional homeology (i.e., retain enough similarity to allow pairing and segregation at meiosis). When applied to interspecific hybrids for which reference genome sequences are available, the question of how differing degrees of homology and homeology affect meiotic interactions may finally be resolved.
Successful transfer of genetic information from one generation to the next is essential for all organisms. The mechanism for this transfer in most sexually reproducing eukaryotic species is meiosis. Chromosome behavior during meiosis must be strictly controlled: in order to ensure correct segregation of chromosomes into daughter cells, each chromosome must pair with its homologous partner. However, homolog recognition is one of the least well understood meiotic processes (Tiang et al., 2012). In most species, the broader process of meiosis is the same: each homologous chromosome must find its partner, associate with it and undergo the reciprocal ‘crossing-over’ process of genetic exchange, creating one or more physical ties (chiasma) between the chromosomes to ensure correct first division disjunction (Wilson et al., 2005). During homologous chromosome pairing, homologs are roughly aligned before close-range ‘homology checking’ and elimination of associations based on repetitive DNA sequences is then thought to occur (Bozza & Pawlowski, 2008). Although homologous chromosome pairing relies on DNA sequence homology (Bozza & Pawlowski, 2008), the exact relationship between DNA sequence similarity and homolog recognition is unknown (Tiang et al., 2012): how similar do genomic sequences have to be to initiate chromosome pairing at meiosis? In particular, how are homologs recognized, and how are meiotic interactions regulated between genomic regions with different degrees of sequence homology?
This question can be addressed in allopolyploid species that are formed when two related diploid species hybridize. Polyploidy is common in many, if not most, plant and animal species lineages (Otto & Whitton, 2000; Leggatt & Iwama, 2003; Van de Peer et al., 2009; Jiao et al., 2011). Some angiosperm families such as the Brassicaceae have undergone multiple polyploidy events such that the present-day species contain homeologous (i.e., ancestrally related) chromosomal regions with a spectrum of sequence similarities (Fig. 1). High-resolution molecular karyotyping is now becoming feasible in nonmodel species with the increasing availability of high-density genotyping arrays and genotyping-by-sequencing (Elshire et al., 2011; Poland et al., 2012; Cavanagh et al., 2013; Edwards et al., 2013). Therefore, the stage is set for significant advances in our understanding of just how similar DNA sequences have to be for homolog recognition to occur, and of how hybrids and polyploids regulate meiosis when several genomic regions with different degrees of sequence homology exist. However, the analytical tools for interrogating large molecular genotyping datasets to answer these questions remain underdeveloped.
In this study, we address this issue by developing a novel method that is capable of interrogating large molecular karyotyping data sets to rapidly assess chromosome behavior at meiosis in allopolyploid species or interspecific hybrids. This approach uses a set of simple statistical assumptions (Fig. 2) to infer chromosome pairing behavior. A hybrid between two species/genotypes is crossed to a third species/genotype to make a testcross, and the resulting progeny are then assessed for allele inheritance. Segregation of alleles from the two original parent species is inferred to result from chromosome pairing during meiosis in the hybrid. To validate this method we used the well-characterized Brassica system (Fig. 1), for which a high-density single nucleotide polymorphism (SNP) genotyping array has recently been developed (http://www/illumina.com). We show that this SNP array can be applied to detect homologous and homeologous chromosome pairing in segregating progeny derived from hybrids between the Brassica allotetraploid species (Fig. 3). This analysis is supported by a previous cytogenetic study of the same genotypes (Mason et al., 2010) and confirms the validity of our molecular karyotype analysis pipeline, providing novel, interesting data related to chromosome pairing behavior in Brassica.
Materials and Methods
Plant material and growth conditions
Two experimental populations were developed by intercrossing homozygous lines of three allotetraploid Brassica species as described previously by Mason et al. (2012) and summarized in Fig. 3. Two to five F1 hybrids per hybrid type (B. napus ‘N1’ × B. juncea ‘J1’; B. juncea ‘J1’ × B. napus ‘N1’; B. juncea ‘J1’ × B. carinata ‘C1’; and B. juncea ‘J1’ × B. carinata ‘C2’) were used to generate testcross progeny; generation of the F1 hybrids is described in Mason et al. (2011). No significant differences in allele inheritance were observed between ‘N1J1’ and ‘J1N1’ reciprocal hybrids (B. juncea × B. napus (F1: AjAnBjCn)) or between the ‘J1C1’ and ‘J1C2’ genotypes (B. juncea × B. carinata (F1: BcBjAjCc)): these were subsequently considered as two single populations for the purposes of the analyses. A total of 41 (B. juncea × B. napus (F1: AjAnBjCn)) × B. carinata experimental progeny (out of 42 seeds planted, germination rate 98%) and 62 (B. juncea × B. carinata (F1: BcBjAjCc)) × B. napus experimental progeny (out of 73 seeds planted, germination rate 85%), derived from meiotically reduced gametes of AjAnBjCn and BcBjAjCc hybrids (Mason et al., 2012), comprised the final experimental populations in this study (Fig. 3).
Molecular karyotyping using the Illumina 60K Brassica SNP chip
Recently, an Illumina Infinium 60 000 SNP array was developed and released for B. napus (http://www.illumina.com/). This chip comprised 52 157 SNPs distributed across the A and C genome chromosomes. B-genome inheritance could not be characterized in our populations because of a lack of available high-throughput molecular tools for this genome. Hybridization protocols were run according to manufacturer's instructions for all samples in the population plus controls (parent species and F1 hybrids), and chips were scanned using an Illumina HiScanSQ. Genotyping data were visualized using Genome Studio V2011.1 (Illumina, Inc., San Diego, CA, USA).
Single nucleotide polymorphisms were filtered in Genome Studio to retain genome-specific SNPs that were: polymorphic between the B. juncea and B. napus A genomes (An and Aj) and did not amplify in B. carinata (2n = BcBcCcCc); or polymorphic between the B. napus and B. carinata C genomes (Cn and Cc) and did not amplify in B. juncea (2n = AjAjBjBj) for the genotypes used in the experimental populations. Filtering for polymorphism between the segregating parent alleles yielded 4883 SNPs for the A genome and 7215 SNPs for the C genome. These 12 098 SNPs were then manually screened in Genome Studio: SNPs that did not show three clear genotype clusters (AA, AB, BB) in the population were removed, leaving 3705 A-genome and 6489 C-genome SNPs (10 194 SNPs in total). After filtering and quality control measures were implemented, the total number of SNP markers retained for segregation analysis was 3046 in the B. napus An genome, 4643 in the B. napus Cn genome, 3055 in the B. juncea Aj genome and 4687 in the B. carinata Cc genome (Supporting Information, Table S1). On average, 406 SNP markers were retained per chromosome, with a minimum of 124 (B. napus Cn5) and a maximum of 1279 (B. carinata Cc4) (Table S1).
Marker genotype calls were imported into Microsoft Excel 2010 where SNPs were sorted by chromosome, with reference to the published and available draft reference Brassica diploid genome sequences: Brassica rapa (Wang et al., 2011) and Brassica oleracea (http://brassicadb.org/brad/; Cheng et al., 2011). Further data cleaning was undertaken to remove SNPs that showed allelic segregation patterns inconsistent with the determined genomic locations; appeared to detect two or more loci; had an excess of no-calls (NCs); or were null for one of the parental alleles.
In the AjAnBjCn population, several blocks of markers (A01_015–A01_37; A05_300–A05_305; A10_001–A10_005; C1_041–C1_053; C4_1278–C4_1279; C8_0001–C8_0004) amplified both an A- and a C-genome allele in B. napus, but only an A-genome or a C-genome allele in B. juncea and B. carinata, respectively. These markers were eliminated from the AjAnBjCn (B. juncea × B. napus) population analyses, but retained in the BcBjAjCc (B. juncea × B. carinata) population analyses. For the known B. napus A7–C6 interstitial homeologous reciprocal translocation in the parental genotype ‘Surpass 400’ (Osborn et al., 2003), markers were named according to their physical location, such that ‘A7’ in B. napus comprised what would be part of C6 in B. oleracea and B. carinata, and ‘C6’ in B. napus comprised part of what would be A7 in B. juncea and B. rapa (see Table S1). A putative duplication in B. napus covered 17% of An10, from markers A10_284 to A10_341, but was not obviously linked to any other genomic region.
The SNP data set was highly saturated: many markers were redundant in terms of detecting chromosome segment inheritance and recombination events across each population. Hence, in order to facilitate the segregation analyses and crossing-over analyses, SNP marker data were reduced to one representative marker per haplotype block, representing unique information per chromosome across each population (i.e. marker bins). SNPs providing redundant information across the population were removed using an R script that treated missing values (NCs) as wild cards (either 1 or 0) and retained markers with the fewest missing values (Notes S1). In the BcBjAjCc testcross population, an average of 19 marker bins per Aj-genome chromosome (range 3–33) and 18 marker bins per Cc-genome chromosome (range 3–29) were retained after elimination of redundant SNPs. In the AjAnBjCn testcross population, an average of 58 marker bins (range 36–93) were retained in the B. juncea Aj and B. napus An genomes and an average of six marker bins (range 2–18) were retained in the B. napus Cn genome after elimination of redundant SNPs.
Data analysis, data visualization and statistics
Fisher's exact test for count data with a significance level of P < 0.05 was used to test whether pairs of marker alleles were segregating with one another, as would be expected from, for example, the presence of the two alleles at a single homologous locus (Fig. 2). To test whether any given two marker alleles present anywhere in the genome(s) were segregating with each other, every possible combination of two alleles was concatenated across individuals, such that each individual was scored for presence of both alleles (1/1), presence of only one of the pair of alleles (1/0 or 0/1) or absence of both alleles (0/0). The number of observations in each category was then summed and compared with expected values derived from sums for 2× 2 contingency tables (Fig. 2). Expected values were derived from normal Mendelian expectations for segregation of individuals with both alleles present (1/1), one allele present (1/0 or 0/1) or neither allele present (0/0) for each pair of alleles (1 : 1 : 1 : 1), using contingency tables to adjust for alleles with an excess of 0 or 1 scores. Every marker allele with a known genome location was tested against every other marker allele with a known genome location.
Deviation from this ratio as a result of linkage was expected to result in an excess of individuals with scores of 0/0 and 1/1 as a result of cosegregation of pairs of alleles (Fig. 2). Deviation from this ratio as a result of segregation of the two alleles was expected to result in an excess of individuals with scores of 0/1 or 1/0 (only one allele present of the pair) (Fig. 2). A significant deviation which resulted in an excess of individuals with a score of 0/1 or 1/0 (one allele present for each pair) was deduced to be the result of segregation of the two alleles (Fig. 2). Significant (P < 0.05) results of the pairwise Fisher's exact tests for count data are presented in figures only where the sum of 0/1 and 1/0 scores was greater than the sum of 0/0 and 1/1 scores, as expected from segregation of alleles. P-values are presented uncorrected for multiple testing, and instead with a range of P-values encompassing rigorous correction cutoffs, as the assumption of independence is invalid for tests of linked alleles. However, the stringent Bonferroni corrections for multiple testing values assuming independent tests were P < 0.00014 for the BcBjAjCc hybrid population and P < 0.000041 for the AjAnBjCn hybrid population for the final set of alleles used for analysis of each population.
In the absence of meiotic crossing over in AjAnBjCn and BcBjAjCc hybrids, marker alleles on each haploid chromosome in the experimental progeny were expected to be either all present or all absent. Recombination was manifested as changes in the presence or absence of marker alleles along haploid chromosomes. Recombination events were recorded for all A- and C-genome chromosomes in each population.
R version 3.0.0 (The R Project for Statistical Computing, http://www.R-project.org) was used to carry out the statistical analyses, to reduce duplicate markers to single representatives per block and to generate the associated figures (Notes S1). Heatmap figures showing allele associations were generated using the R software package ‘gplots’, and linkage disequilibrium was calculated using the ‘LD’ function in software package ‘genetics’.
A total of 3 046 SNPs polymorphic between the B. napus and B. juncea A genomes (An and Aj), and a total of 4 643 SNPs polymorphic between the B. napus and B. carinata C genomes (Cn and Cc) were selected for final analysis. Linkage disequilibrium values (r, P-value and D′) were calculated for every pair of alleles in each of the two populations to provide additional information, and linkage disequilibrium was identified between both adjacent alleles and segregating alleles (Tables S4, S5). Using the molecular karyotyping pipeline in R (Notes S1), every SNP allele in each of the two testcross populations (Fig. 3) was tested for significant segregation (Fig. 2) with every other SNP allele in the same population. This approach allowed discrimination between linkage disequilibrium as a result of segregation of alleles, and linkage disequilibrium as a result of cosegregation of alleles (linkage). Increasing significance corresponded to increasing numbers of individuals in the population with either one allele or the other (but not both or neither) for any two alleles.
A-genome allele segregation in AABC hybrids
In testcross population A (Fig. 3), either an Aj or an An allele was present at most homologous loci, as evidenced by the strong diagonal of significant segregation between Aj and An alleles in Fig. 4. In the A genome, which was present as homologous chromosomes from B. juncea (Aj) and B. napus (An), most segregating allele pairs corresponded to homologous loci in the A genome (Fig. 4). Most B. juncea Aj-genome chromosomes segregated with high fidelity with their B. napus An-genome homologs. An exception to this was part of chromosome A7, where strong associations between Aj and An alleles were absent and this region instead showed an Aj7 – Cn6 association.
Segregation in the BBAC hybrids between alleles sharing primary homeology
In testcross population B (Fig. 3), B. juncea A-genome alleles (Aj) segregated with B. carinata C-genome alleles (Cc) with which they shared primary homeology (Fig. 5; see Fig. 1 for detailed primary homeologous relationships). These associations between homeologous alleles were of similar significance to those between two alleles at a single homologous locus (P < 0.00000001, Fig. 5).
Four Aj–Cc chromosome pairs showed significant segregation of homeologous alleles along their entire length: Aj1–Cc1, Aj2–Cc2, Aj3–Cc3 and Aj7–Cc6 (Fig. 5), as indicated by the diagonal red lines across the length of the chromosome blocks. Chromosomes Aj4 and Cc4 showed segregation of alleles along the whole length of Aj4, but with decreasing significance (0.00000001 < P < 0.001) towards the end of Cc4 (from c. 40 to 55 Mbp). Aj9 and Cc8 alleles segregated along the entire length of Cc8, with the top 8 Mbp of Aj9 segregating instead with the top of Cc9. Clear evidence of trivalent formation was apparent among chromosomes Aj9, Cc8 and Cc9, with Aj9 chromosomes showing matching allele segregation patterns and crossovers with Cc8 at one end and Cc9 at the other end of the chromosome. Likewise, strongly significant segregation was observed between alleles on part of Aj10 and Cc9, with a switch in Cc9 preference to Aj9.
Other chromosomes showed more fragmentary allele segregation across regions of A–C homeology (Fig. 5, Table S4). For example, Aj5 alleles preferentially segregated with Cc5 alleles, but segregation with Cc4 and Cc6 alleles was also observed. Aj8 alleles showed little tendency to segregate with Cc alleles, but significant segregation associations were observed with the top of Cc8 and the bottom of Cc9. Similarly, alleles on chromosome Cc7 showed only weak segregation with the alleles in the region of primary homeology at the bottom of chromosome Aj3.
Segregation between alleles sharing secondary homeology
Evidence was also obtained for weakly significant segregation between alleles in regions of secondary homeology (derived from ancient polyploidy events) within each haploid genome in both testcross populations, that is, autosyndesis (Aj–Aj, within the B. juncea A genome; Cc–Cc, within the B. carinata C genome; and Cn–Cn, within the B. napus C genome) (Figs 6, S4, S5).
The largest clusters of segregation between regions of secondary homeology were observed for chromosomes Aj2 and Aj10 and for chromosomes Cc2 and Cc9 (Fig. 6a,b). The Aj2–Aj10 association spanned c. 5–10 Mbp based on putative SNP positions, whereas the Cc2–Cc9 association was significant over 35–45 Mbp. Another association was also present between Cc3 and Cc9 over a range of c. 15–35 Mbp. Only two weak regions of segregation between Cn alleles were detected within the haploid B. napus C genome in the AjAnBjCn hybrid type. One of these associations was between Cn2 and Cn9 (Fig. 6c), as was observed for Cc2 and Cc9.
Recombination events: distribution and frequency
Chromosome segregation as predicted by the molecular karyotyping analysis using R (Notes S1) was independently validated by manual inspection of chromosome segregation and recombination events: Brassica Aj and Cc allele segregation patterns between known primary homeologs were assessed (Table 1). In general, predicted associations from molecular karyotyping could be validated by manual assessment of allele segregation and evidence of crossover formation for each chromosome pair (Fig. 5, Table 1). The location of recombination breakpoints along individual chromosomes was assessed for chromosomes that were neither completely absent nor completely present (i.e. recombinant chromosomes). The distribution of breakpoints along each chromosome is shown in Figs S1–S5.
Table 1. Proportion of chromosomes segregating with primary homeolog(s) based on matching patterns of allelic inheritance (presence of one chromosome and absence of the other, with or without recombination between the two) in a population of n = 62 plants derived from Brassica juncea × Brassica carinata (BjBcAjCc) hybrids segregating for B. juncea A-genome (Aj) and B. carinata C-genome (Cc) alleles
Proportion of nonrecombinant chromosomes (transmitted without recombination breakpoints indicating crossover formation), regardless of segregation with primary homeolog. Indicates maximum possible percentage univalent inheritance for each chromosome.
Percentages that add up to > 100% are strongly suggestive of multivalent formation involving that chromosome.
Homologous Aj and An chromosomes were highly recombined in the AjAnBjCn testcross population, averaging 2.4 crossover breakpoints per chromatid per individual (Fig. 7a). As many as eight breakpoints per chromatid were observed for A3 and A9. A-genome chromatids were transmitted intact without recombination only 8% of the time on average (range 0–15%), except for An7 (unrecombined 22% of the time) which appeared to have the structural An7–Cn6 translocation variant common to many B. napus genotypes. The region on An7 corresponding to the translocation had no recombination breakpoints within it, but many recombination breakpoints flanking this region (Fig. S2). Otherwise, recombination breakpoints were distributed along the lengths of the chromosomes, with a general tendency towards increased frequency approaching distal regions (Figs S1, S2).
Cn recombination frequencies
Recombination involving secondary homeologs in the haploid B. napus C genome in the AjAnBjCn testcross population occurred at much lower frequencies, averaging 0.26 Cn chromosome breakpoints per plant (Fig. 7b). Cn chromosomes were either wholly present or wholly absent (unrecombined, either lost or transmitted intact as univalents) 59–95% of the time, except for 46% of Cn6 chromatids which had undergone recombination with chromosome Aj7 (again as a result of the An7–Cn6 translocation variant). Cn6 had a proximal cluster of recombination breakpoints. However, few recombination breakpoints were observed overall in Cn chromosomes, and these tended to be distally located (Fig. S3).
Aj–Cc recombination frequencies
Recombination was frequent between primary homeologs in the Aj and Cc genomes in the BcBjAjCc hybrid progeny (Fig. 7c) with an average of 0.46 chromosome breakpoints observed per chromosome per plant. On average, 58% of Aj and Cc chromosomes were transmitted without detectable recombination events. This was not significantly different from the 50% expected to result from a single crossover event per homologous chromosome pair per meiosis, whereby two out of four chromatids would be expected to be transmitted without evidence of recombination (P = 0.09, Student's paired t-test against the mean). However, some chromosomes were far less likely to recombine than others (Fig. 7c): chromosomes Cc7 and Aj8 each had only two detectable recombination events across the whole population. Recombination breakpoints were distributed distally on every chromosome, except in chromosomes Aj7 and Cc6, which had proximal clusters of breakpoints (Figs S4, S5).
Detecting nonrandom coinheritance of alleles across genomes
In addition to the segregation and recombination analyses discussed in the preceding sections, we also sought evidence of positive associations between alleles that would indicate significant coinheritance of chromosomes. In addition to the expected coinheritance of alleles physically located on the same chromosomes, we found some parts of the genome were significantly associated in this way (Tables S2, S3). This generally occurred when two chromosomes both associated with a third chromosome (Tables S2, S3). Examples included Cc2 and Cc3 (both paired with Cc9), Cc8 and Cc9 (both paired with Aj9), Cc7 and Cc3 (both paired with Aj3) and Aj2 and Cc9 (both paired with Cc2).
Transmission frequencies of marker alleles from AjAnBjCn and BcBjAjCc hybrids to progeny
In the B. juncea × B. napus (AjAnBjCn) hybrids, most Aj, An and Cn genome alleles were transmitted to the testcross population according to normal Mendelian expectations (50% frequency; P > 0.05, Pearson's χ2 test), (Fig. S6). Some bias towards retention of Aj or An alleles was observed over short regions (Fig. S6). Alleles from the B. napus chromosome Cn8 and half of chromosome Cn4 were retained more often than expected by chance, and no chromosome was lost more often than the expected 50% frequency (Fig. S7). In the B. juncea × B. carinata (BcBjAjCc) testcross population, B. juncea Aj-genome chromosomes were retained more often than B. carinata Cc genome chromosomes for every chromosome except Aj7 and Cc6 (Fig. S8).
In this study, we developed a novel open source analytical pipeline suitable for inferring chromosome segregation in large data sets (see Notes S1). This molecular karyotyping approach uses a set of simple genetic principles (Fig. 2) to infer chromosome pairing behavior. A hybrid between two species or two genotypes is crossed to another species/genotype to make a testcross (Fig. 3), and the resulting progeny are then assessed for allele inheritance using high-throughput SNP molecular markers. Segregation of alleles from the original two species can then be inferred to result from chromosome pairing during meiosis in the hybrid without previous knowledge of chromosome relationships. This method can provide new insight into homeologous and homologous relationships in species complexes by determining homologous and homeologous relationships between chromosomes and estimating stability of chromosome inheritance. By assessing functional chromosome interactions, it would also allow empirically based predictions of transfer of alleles between genomes in interspecific hybrids, a matter of importance for estimating the risk of transgene escape from transgenic crop species to their wild relatives (Chèvre et al., 1997). In future, and with the increasingly availability of genotyping-by-sequencing approaches, this analytical method may provide us with the answer to the question of how similar DNA sequence has to be to pass homology checks and initiate chromosome pairing during meiosis.
Successful detection of homologous, homeologous and autosyndetic chromosome segregation
High-resolution molecular karyotyping of two novel interspecific Brassica populations was carried out to validate this approach. A spectrum of genome interactions was revealed, ranging from almost completely regular pairing and recombination of homologs (A-genome chromosomes from B. napus and B. juncea in the AjAnBjCn hybrids; Fig. 4), to mostly regular pairing and reduced recombination frequencies between primary homeologs (A- and C-genome chromosomes in the BcBjAjCc hybrids; Fig. 5), down to low-level autosyndetic chromosomal interactions between secondary homeologs (within the haploid B. napus Cn genome, Fig. 6c). In other words, this molecular karyotyping analysis revealed exactly which regions of both ancient (secondary) and recent (primary) homeology in the Brassica A and C genomes have retained enough sequence similarity to form pairing associations at meiosis (Figs 4-6), and this was validated by manual inspection of allele segregation patterns (Table 1, Fig. 7).
Molecular karyotyping detects segregation between alleles at homologous loci
Segregation of homologous alleles belonging to the B. juncea and B. napus Aj and An genomes (Fig. 4) supports the status of these genomes as predominantly unrearranged relative to each other (Parkin et al., 1995; Axelsson et al., 2000). However, one major discrepancy was observed: an An7–Cn6 translocation in the B. napus line used in this experiment (Osborn et al., 2003) resulted in associations between Aj7 and part of Cn6 (Figs 4, S1, S3, Table S3), rather than between Aj7 and An7, over the region of the translocation. Several genomic regions showed amplification of two B. napus alleles at each locus rather than one B. napus allele (An1, Cn1, An5, Cn4, An10 and Cn8), but amplified only one allele per locus in B. juncea and B. carinata. This phenomenon may have been the result of nonspecificity of SNP primer binding and/or divergence between the genomic sequences of B. napus, B. juncea and B. carinata. Therefore, these regions were removed from the analysis, except for the largest, which comprised 30% of chromosome An10 (5 Mb). However, it is more probable that this observation is the result of the presence of duplication/deletion events between the An and Cn genomes in the B. napus parent, as similar chromosome rearrangements have been observed in other cultivars of this young allopolyploid species (Udall et al., 2005).
Molecular karyotyping predicts segregation between regions of primary homeology
In the BcBjAjCc hybrids, Aj–Cc allosyndetic (between-genome) recombination between primary homeologous regions approached the high levels observed between homologs in natural Brassica species. Nicolas et al. (2007) observed that only half of the chromosome rearrangements transmitted from B. napus haploids resulted from recombination between regions of primary homeology, with the rest resulting from recombination between other regions of the genome. By contrast, in our BcBjAjCc population, the vast majority of crossover events appeared to derive from regions of primary homeology: for chromosome pairs Aj1–Cc1, Aj2–Cc2 and Aj3–Cc3 we predicted 94–100% pairing between primary homeologs (Table 1), and in general only a few convincing associations between regions of secondary homeology were detected (Fig. 6). The contrasting results of these two studies are surprising, but may be the result of differences in genome structure or genetic factors between the two population types (Leflon et al., 2010; Suay et al., 2014), or of different selection pressures in the generation of the experimental populations.
Our study suggests that the most frequently formed Aj–Cc bivalents are likely to be collinear pairs Aj1–Cc1, Aj2–Cc2, Aj3–Cc3 (Fig. 5), supporting previous analyses of primary A–C homeology (Parkin et al., 1995). However, the chromosome pair Aj7–Cc6, which lacks whole-chromosome collinearity (Parkin et al., 2003), was also very strongly associated across their entire length (Fig. 5). This is surprising, as the B. juncea A genome and the B. carinata C genome lack the interstitial An7–Cn6 reciprocal translocation found in many genotypes of B. napus (Osborn et al., 2003), including the genotype used in this study to generate the AjAnBjCn hybrid population. In fact, several Aj–Cc chromosome regions that would be predicted to show segregation based on primary homeology (Fig. 1) showed little to no association (Fig. 5, Table 1). For instance, Aj7 and Cc6 also share primary homeology with Aj6 and Cc7 (Fig. 1), but no Aj6–Cc6 or Aj7–Cc7 segregation was observed (Fig. 5). Hence, factors other than primary sequence homeology must be influencing the Aj7–Cc6 association, and the lack of association between other known homeologs. Concentration of recombination breakpoints around the 8–10 Mbp region of Aj7 and the 15–25 Mbp region of Cc6 in this population (Figs S4, S5) suggests an unusual mode of bivalent formation between these two chromosomes: other Aj–Cc chromosome pairs appeared to form mostly distal associations (Figs S4, S5).
Genetic factors present on single chromosomes that affect genome-wide recombination frequency have recently been found in Brassica (Suay et al., 2014). However, genetic factors influencing recombination between particular chromosomes have yet to be identified in the Brassica genus, although chromosome-specific genetic factors have been identified in other species (Jackson et al., 2002). Hence, certain chromosomes or chromosome pairs (e.g. Aj7–Cc6) may harbor, or be the target of, specific genetic factors encouraging crossover formation involving that chromosome. The ‘choice’ of pairing partner may also be influenced by a combination of availability of other suitable partners (Nicolas et al., 2008), genetic factors and the location of the homeologous sequence in relation to the centromeres, as well as the degree of sequence similarity (Nicolas et al., 2012).
Chromosome associations detected for Aj8, Cc8, Aj9, Cc9 and Aj10 were strongly indicative of multivalent formation or partner swapping (Fig. 5), consistent with the status of these linkage groups as the most rearranged between the A and C genomes (Parkin et al., 2003). At least one multivalent association per cell involving A- and C-genome chromosomes only was observed by cytogenetic means by Mason et al. (2010) for this BcBjAjCc hybrid type: 0.5 Aj–Aj–Cc (zero to two per cell), 0.3 Cc–Cc–Aj (zero to two per cell) and 0.1 Aj–Aj–Cc–Cc (zero to one per cell). These multivalents are most likely attributable to Cc8, Aj9, Cc9 and Aj10, based on these current findings.
Molecular karyotyping predicts segregation between regions of secondary homeology
Significant autosyndesis (within-genome pairing) in both the A and C genomes was predicted on the basis of segregation between alleles (Fig. 6). These associations are probably the result of shared ancestral karyotype blocks resulting from the ancient genomic triplication before the divergence of the Brassica A and C genomes (Parkin et al., 2005; Schranz et al., 2006; Cheng et al., 2013). The three strongest associations observed were between Aj2 and Aj10, Cc2 and Cc9 and Cc3 and Cc9, all in the BcBjAjCc hybrid population. These chromosomes all share two large conserved ancestral karyotype blocks in the same orientation and terminal to the end of the chromosome (Schranz et al., 2006; Cheng et al., 2013). An association between Cn2 and Cn9 was also observed in the AjAnBjCn hybrid population. These autosyndetic associations suggest that associations may form between regions of ancient homeology even in the presence of recent homeologs.
Recombination frequency and allelic selection pressure
Recombination between the B. juncea and B. napus A genomes was greatly enhanced relative to that observed in natural and resynthesized B. napus (Nicolas et al., 2008). This is probably because of the presence of additional univalent chromosomes during meiosis in this hybrid type (Mason et al., 2010), as previously observed in other Brassica hybrids (Leflon et al., 2010). Hence, this hybrid type may offer a useful bridge in Brassica for breeders aiming to break up regions of undesirable linkage disequilibrium, following a trend of interspecific hybridization for crop improvement in this genus (Rygulla et al., 2007; Navabi et al., 2010; Chen et al., 2011; Zou et al., 2011). In the BcBjAjCc hybrid-derived progeny, A-genome loci were retained in preference to C-genome loci (Fig. S8). As B. juncea was the maternal parent in the initial crosses, this may indicate a cytoplasmic effect on preferential genome inheritance, as previously suggested in crosses between natural and synthetic B. napus (Szadkowski et al., 2010). Several genomic regions also appeared to be under retentive selection pressure in both hybrid populations (Figs S6–S8), and may harbor alleles related to success in interspecific hybridization or other processes conferring a viability advantage.
Selection for particular chromosome complements through viability advantage
Viability advantage for particular karyotypes could affect the success of the molecular karyotyping approach. For instance, Xiong et al. (2011) found strong ‘dosage compensation’ between homeologs, such that presence of four copies of primary homeolog pair A1–C1 (e.g. A1–A1–C1–C1, A1–A1–A1–A1, A1–C1–C1–C1 etc.) was selected for in advanced generations of synthetic B. napus. Similar dosage compensation of homeologous chromosome sets has been observed in neoallopolyploid Tragopogon miscellus (Chester et al., 2012, 2013). Hence, a similar effect could occur in our BcBjAjCc hybrids to eliminate gametes that have not inherited a copy of either chromosome A1 or C1 for example, to give the false appearance of chromosome segregation and hence pairing at meiosis. However, recombination events between primary homeologs were manually validated (Table 1), and putative crossover frequency was not significantly different from predictions for one crossover event per chromosome per gamete. Therefore, we predict that the majority of segregation associations identified through the molecular karyotyping analysis are the result of actual chromosome pairing and segregation. In systems with unknown chromosome homeology, even dosage compensation effects will provide information about chromosome similarity, as only primary homeologs are predicted to provide ‘dosage compensation’ (Xiong et al., 2011).
In future, as reference genome sequences for polyploid species (including Brassica crop species) become available, we will have the opportunity to identify at base-pair resolution the chromosomal regions initiating homeologous pairing at meiosis. We provide a simple analytical pipeline that can, once whole genome sequences become available, be interrogated by whole-chromosome comparative sequence analysis to deduce exactly how similar genomic regions need to be to pair and segregate at meiosis. Similar approaches using our analysis pipeline may also be taken in highly complex polyploid species, helping us to understand how chromosome pairing is controlled in complex polyploids with homeology from ancient and common ancestors.
This work was made possible by prior support from the Australian Research Council Linkage Project LP0667805, with industry partners Council of Grain Grower Organisations Ltd and Norddeutsche Pflanzenzucht Hans-Georg Lembke KG. A.S.M. was supported, and additional work funded, by an Australian Research Council Discovery Early Career Researcher Award (DE120100668).