Rapid large-scale genomic introgression in Arabidopsis suecica via an autoallohexaploid bridge

Abstract

Gene flow between species in the genus Arabidopsis occurs in significant amounts, but how exactly gene flow is achieved is not well understood. Polyploidization may be one avenue to explain gene flow between species. One problem, however, with polyploidization as a satisfying explanation is the occurrence of lethal genomic instabilities in neopolyploids as a result of genomic exchange, erratic meiotic behavior, and genomic shock. We have created an autoallohexaploid by pollinating naturally co-occurring diploid Arabidopsis thaliana with allotetraploid Arabidopsis suecica (an allotetraploid composed of A. thaliana and Arabidopsis arenosa). Its triploid offspring underwent spontaneous genome duplication and was used to generate a multigenerational pedigree. Using genome resequencing, we show that 2 major mechanisms promote stable genomic exchange in this population. Legitimate meiotic recombination and chromosome segregation between the autopolyploid chromosomes of the 2 A. thaliana genomes occur without any obvious bias for the parental origin and combine the A. thaliana haplotypes from the A. thaliana parent with the A. thaliana haplotypes from A. suecica similar to purely autopolyploid plants. In addition, we repeatedly observed that occasional exchanges between regions of the homoeologous chromosomes are tolerated. The combination of these mechanisms may result in gene flow leading to stable introgression in natural populations. Unlike the previously reported resynthesized neoallotetraploid A. suecica, this population of autoallohexaploids contains mostly vigorous, and genetically, cytotypically, and phenotypically variable individuals. We propose that naturally formed autoallohexaploid populations might serve as an intermediate bridge between diploid and polyploid species, which can facilitate gene flow rapidly and efficiently.

Introduction

Reproductive isolation is central to the theory of speciation. Hybridization transcends this theory and while often hybrids are evolutionary dead ends. Hybridization transcends this theory and while often hybrids are evolutionary dead ends, in some cases, especially in plants, hybridization events can lead to speciation (Mallet 2007). A special case of hybrid speciation is that of allopolyploidy, where genome duplication simultaneous with hybridization allows the newly formed genome to overcome the challenges of imperfect homolog pairing of diploid hybrids during meiosis (Comai 2005). Because allopolyploids have paired homologous chromosomes and in principle are therefore able to undergo normal meiosis, allopolyploidy has been described as a mechanism for instant speciation (Linder and Rieseberg 2004; Pearse et al. 2006). Although genomic stability and vigor are commonly found in established allopolyploids, early generation allopolyploids (neoallopolyploids), frequently display phenotypic and cytological instabilities and are often characterized by low fitness (Comai et al. 2000; Xiong et al. 2011; Chester et al. 2012; Zhang et al. 2013).

Allopolyploids may benefit from various genomic and epigenetic changes following allopolyploidization (Chen 2007; Madlung and Wendel 2013) but genomic change also brings dangers and genomic conflicts. The similarity between homoeologous chromosomes can, for example, lead to exchanges of chromosome arms and rearrangements between nonhomologous chromosomes. In addition, multivalent chromosome pairing can result in abnormalities during cell division and the formation of aneuploids, a danger that is particularly frequent in autopolyploids, which arise from whole-genome duplication. How the resolution of these conflicts between 2 genomes in a neopolyploid is achieved and how genomic change in response to polyploidy affects the fitness of the new species is still controversial and largely unknown (Mayrose et al. 2011; Soltis et al. 2014). Experiments with synthetic neoallopolyploids have suggested that from an interspecies polyploid cross only few individuals stabilize their genomes throughout the subsequent generations, whereas the majority of neopolyploids is either inviable or infertile (Comai et al. 2000; Gaeta et al. 2007; Gaeta and Pires 2010). Unless genomic stability is restored after allopolyploidization, genomic instability, such as in the form of aneuploidy, can accumulate from generation to generation. This spiraling process has been termed the “polyploid ratchet” (Gaeta and Pires 2010), where genomic errors accumulate over generations until fitness is compromised to the point of extinction.

While genomic rearrangements and aneuploidy constitute genetic errors that can lead to detrimental mutations, legitimate meiotic recombination may be viewed as a genetic mechanism that could facilitate intergenomic gene flow in neoallopolyploids. Whether or not, and if so to what extent, this occurs is not well understood. Strict pairing of homologous chromosomes during meiosis reduces the risk of monovalents during cell division and thus of aneuploid gamete formation. Genetic loci that reduce the threat of intergenomic (homoeologous) pairing have been described in allohexaploid wheat (Griffiths et al. 2006), allotetraploid Brassica napus (Jenczewski et al. 2003), and in allotetraploid Arabidopsis suecica (Henry et al. 2014). It is possible that loss or gain of these (and maybe other) loci in allopolyploids play a role in stabilizing neoallopolyploids. A study of the genus Arabidopsis showed that gene flow can occur between species inconsistent with the notion that speciation gives rise to new species exclusively in a linear manner or via bifurcating phylogenetic trees (Novikova et al. 2016). In their study, Novikova and colleagues (2016) showed that admixture is frequent in the genus Arabidopsis and suggested that evolution in this genus cannot be explained by mutation and divergence alone.

To shed more light on the question what mechanisms might explain how gene flow and introgression in the genus Arabidopsis occurs, we worked with a pedigree that was created by pollinating a diploid (Arabidopsis thaliana, Col-0, 2n = 2x = 10) with an allotetraploid (A. suecica, Sue-1, 2n = 4x = 26), whose genome itself is composed of those of A. thaliana and Arabidopsis arenosa. Here, we show that in this autoallohexaploid controlled meiotic recombination can promote rapid integration of genomic material from A. thaliana into A. suecica (introgression), while out-of-control mechanisms that result in aneuploidy, instability, and the eventual removal of the line from the gene pool effectively act against evolutionarily stable introgression and gene flow. Our data further suggest that an autoallohexaploid might efficiently serve as a bridge between different, related species and might thus provide a mechanism for rapid admixture between genomes.

Materials and methods

Material

The auto-allohexaploid pedigree was created as previously described (Matsushita et al. 2012). Briefly, an individual of the allotetraploid A. suecica Sue-1 (ABRC accession numbers CS22505) was used as the pollen donor in a cross with diploid A. thaliana (Col-0). The resulting F1 offspring was ascertained to be triploid using flow cytometry. Two branches of the otherwise infertile plant became fertile, probably after spontaneous somatic doubling. F₂ individuals were shown to be mostly hexaploid using flow cytometry. F₂ lines were propagated by selfing and single-seed descent (Fig. 1) down to the F₁₄ to F₁₇ (depending on line). For this study, seeds saved from early generations were planted alongside mid- and late generations. To do so, multiple seeds saved from individuals that had been used for propagating the line were planted to produce same-generation sisters. These sisters thus represent short side branches to the descending line. During the production of the pedigree, lines that readily produced seed were preferred, resulting in the selection of fertility traits.

Fig. 1.

Pedigree of the experimental population. Diploid A. thaliana (Col-0) was used as maternal parent in a cross with allotetraploid A. suecica (Sue-1). The triploid F1 individual was sterile with the exception of 2 sectors that arose from cells that had undergone somatic doubling. A total of 37 seeds were recovered and grown (F2). These seeds founded each of the numbered lines 1–37, of which 6 (numbered 2, 5, 6, 12, 14, and 19) were propagated further by single-seed descent. These plants were ascertained by flowcytometry to be generally hexaploid (Matsushita et al. 2012). The lines were propagated through the current 17th generation. Seeds from each individual used for line propagation were set aside. At the time of analysis for this study, 3 seeds from selected lines and generations were grown at the same time in greenhouse-controlled conditions. This figure only shows a small subset of the complete pedigree. The entire list of participating individuals can be found in Supplementary Tables 1 and 2. The pedigree figure is reproduced from Matsushita et al. (2012).

Autoallohexaploids as well as A. suecica plants were grown on soil in 10-cm diameter plastic pots in a greenhouse with supplemental lighting under long-day conditions (16 h light/8h dark) or in an incubator with the same light conditions and a temperature of 20°C. For phenotypic analysis, same-age plants were photographed and their rosette diameter measured using ImageJ. Seeds of same-age plants were later harvested, dried, and weighed to estimate fitness. For short-read sequencing, leaf material was harvested when plants were 7 weeks of age, snap frozen in liquid nitrogen, and stored at −80°C until used. DNA was extracted using a Qiagen Plant DNeasy kit and sequenced at the sequencing center of the Max Planck Institute for Plant Breeding Research.

Assembly and annotation of A. suecica Sue-1

For the genome assembly, DNA was extracted from a single A. suecica Sue-1 individual provided to the Max Planck Genome Centre Cologne, which extracted the DNA that was used for PacBio sequencing on a Sequel II instrument. Sequencing reads were generated at 50× coverage comprising a total length of ∼23.8 Gb. Using Canu software v1.5 (Koren et al. 2017), the A. suecica long reads were corrected, trimmed, and assembled into unitigs using the parameters corOutCoverage = 200, batOptions=-dg 3 -db 3 -dr 1 -ca 500 -cp 50. The assembly was corrected with Pilon v1.22 (Walker et al. 2014) using A. suecica Illumina short paired-end reads. For this, the short reads were first aligned against the assembly and the subsequent BAM file was sorted and indexed using SAMtools (Li et al. 2009), and subsequently used as input for the Pilon tool. Optical mapping was performed using the nicking enzyme Nb.BssSI. The high molecular weight DNA was labeled and stained with the Bionano PrepTM DNA Labeling Kit and analyzed on the Irys platform (Bionano Genomics). Molecules greater than 100 kb were used for the de novo optical map assembly run in the Irys Solve software version 2.1.1. To align the long-read assembly with the optical consensus maps, the PacBio assembly was in silico digested with Nb.BssSI to generate a reference consensus map. Hybrid scaffolding was executed by using the set of the perl scripts integrated in the automated scaffolding pipeline in the Bionano Solve software package. The resulting contigs were aligned against the pseudoreference genome of the concatenated A. thaliana Col-0 and Arabidopsis lyrata (https://plants.ensembl.org; accessed 2022 September 8, v1.0; Hu et al. 2011) genomes using NUCmer from the MUMmer package v3.5 (Kurtz et al. 2004). The parameters for NUCmer were set to—maxmatch and a minimum cluster size of 100. When A. suecica short reads were aligned against the pseudoreference using Bowtie2 v2.2.8 (Langmead and Salzberg 2012), a significant coverage increase was observed at the end of At2 which was not reciprocated by the syntenic region at the end of A. lyrata chromosome 4 and thus was considered to be a duplication. In order to rectify this, the end of the At2 was copied onto the end of the Aa4 in the assembly. The centromeres of the A. arenosa subgenome could not be assembled due to the lack of centromeres in the A. lyrata v1.0 sequence.

The syntenic relationships between the 2 subgenomes in the final assembly were established using SyRI v1.2 (Goel et al. 2019). The subgenome chromosome sequences were first combined in blocks according to the syntenic relationship between A. thaliana Col-0 and A. lyrata (Hu et al. 2011). Arabidopsis arenosa chromosomes 1 and 2 were then combined into 1 sequence and aligned using NUCmer of the MUMmer package with similar settings as before (–maxmatch and minimum cluster size of 100) against A. thaliana subgenome chromosome 1. Chromosomes 3, 4, and 5 of A. arenosa were combined and aligned against chromosomes 2 and 3 of A. thaliana. Similarly, chromosomes 6, 7, and 8 of A. arenosa were aligned against chromosomes 4 and 5 of A. thaliana. These alignments were then used as input for the SyRI software and the resulting synteny between the subgenomes was plotted with SyRI’s plotsr. An inversion was reported between the end of A. thaliana subgenome chromosome 4 and the whole of the A. arenosa subgenome chromosome 6 as well as part of A. arenosa subgenome chromosome 7. As this inversion conflicted with a duplication between the A. thaliana and A. arenosa subgenomes reported in part of the same region, the inversion was shortened to be only between the end of A. thaliana subgenome chromosome 4 and the beginning part of A. arenosa subgenome chromosome 7. This observation is similar to what has been reported in the synteny analysis between A. thaliana Col-0 and A. lyrata (Hu et al. 2011).

The annotation of the assembly was carried out as follows: Protein hints were first generated using prior peptide information from A. thaliana Col-0 and A. lyrata (v1.0). The program Exonerate v2.2.0 (Slater and Birney 2005) was used to align the protein sequences to the assembly. RNA sequence data from previous work (Carlson et al. 2017) was used to generate RNA hints and aid in the annotation. For this, the RNA sequence accessions SRR3676009, SRR3676011, and SRR3676012 (BioProject PRJNA321210) were downloaded from the NCBI. The programs Stringtie v2.1.5 (Pertea et al. 2015) and HiSat2 v2.1.0 (Kim et al. 2019) were used to align the RNA sequences against the assembly and assemble the alignments into potential transcripts. Ab initio runs of the annotation software Augustus 3.4.0 (Stanke et al. 2006), Semi-HMM-based Nucleic Acid Parser (SNAP) version 2006-07-28 (Johnson et al. 2008), and GlimmerHMM (Majoros et al. 2004) were carried out. The EVidenceModeler (Haas et al. 2008) software combined with the Program to Assemble Spliced Alignments (Haas et al. 2003) was then used to combine the annotations predicted based on user-defined weights provided in a configuration file.

After annotating the protein-coding genes, the program RepeatMasker version open-4.0.9 (http://www.repeatmasker.org) was used to identify transposable elements (TEs) in the assembly sequence. The tool intersectBed from the BEDtools suite v2.21.0 (Quinlan and Hall 2010) was then used to find overlaps between the coordinates of the predicted TEs and those of the protein-coding genes. If TEs overlapped 20% or more of the protein-coding gene, the region was annotated as TE-related. The program LiftOff v1.6.1 (Shumate and Salzberg 2021) was used to correct the strand specificity in the annotation file for a contig inversion in chromosome 3 of the assembly. This inversion was also corrected in the actual assembly sequence.

Here, the 13 chromosomes of the A. suecica genome are indicated by the labels “Chr1” to “Chr13,” where the first 5 chromosomes refer to the A. thaliana subgenome and the last 8 chromosomes to the A. arenosa subgenome.

Analysis of meiotic recombination in the autoallohexaploid genomes

Short paired-end reads generated using Illumina technology from individual progeny genomes were aligned against the Sue-1 assembly using Bowtie2 with the default parameters (in end-to-end mode). The BAM files were then sorted and indexed using SAMtools sort and SAMtools index, respectively. The per-base depth of coverage for each alignment was obtained using genomecov from BEDtools, which gave an output file of the cumulative read count per position of each chromosome in each sample. The length of each chromosome was divided into bins of 100 to 500 kb (with stepsizes of 10 kb for 100 kb windows) and the average coverage per base calculated per window normalized by the chromosome length. R ggplot2 was then used to plot the sliding window coverages normalized on a scale between 0 and 100. The only noise was observed in the centromeric regions which stems exclusively from the A. thaliana chromosomes as the centromeres are not sequenced in the A. arenosa chromosomes.

To generate a reference marker list of SNP alleles, short reads of A. thaliana Col-0 obtained from the public Arabidopsis database TAIR were aligned against the A. suecica assembly as well as to the other samples, and variant calling and further filtering of the VCF file were performed. A marker list of Col-0 parental SNP alleles (that differ from the A. suecica SNP alleles) was generated to ensure that the variants not matching the SNP alleles in the A. suecica parent matched the SNP alleles derived from the Col-0 parent. Only biallelic sites in the samples were considered. Variant calling was performed for all the samples using SAMtools mpileup and BCFtools call from SAMtools v1.3.1 (Li et al. 2009). Using SAMtools mpileup anomalous read pairs were retained with the parameter -A and the minimum base quality threshold -Q was set to 0. BCFtools call was run with the option -A to report all possible alternate alleles at variant sites. The variants were then filtered to generate high-quality SNPs by selecting only those whose quality scores were at least 10, and using BCFtools to filter out indels. A sliding window variant analysis was carried out for each individual with windows ranging from 100 to 300 kb and stepsizes of 10 to 30 kb. For each window, the proportion of SNP alleles deriving from the A. suecica parent vs the A. thaliana Col-0 parent was calculated by dividing the number of A. suecica SNP alleles over the total number of SNP alleles. As the genome is comprised of an A. thaliana subgenome and an A. arenosa subgenome SNP alleles derived from the Col-0 parent were only seen in the sliding window analysis of the A. thaliana subgenome. R ggplot2 was used to plot the SNP allele frequencies on a scale of 0 to 1 and overlayed onto the coverage plots. Those windows for which the total number of SNP alleles did not exceed 300 were excluded as these may not give an accurate representation of the ratio of the parental alleles. A frequency or proportion of 1.0 indicates that all the SNP alleles are derived from A. suecica while 0 indicates all of them are derived from the Col-0 parent.

The sliding window analyses of the SNP allele frequencies were used to estimate the extent of homologous recombination across the 4 chromosome copies within the A. thaliana subgenome as follows: For each chromosome sudden, sustained shifts in SNP allele frequencies along the length of the chromosome were counted using SNP frequency graphs with a 300-kb window size rendered in R with ggplot2. The total number of shifts indicates the frequency with which cross-overs occur along a given chromosome, not the accumulated number of recombination events since the original cross.

Identification and validation of aneuploidies

In order to investigate whole chromosome loss or gain within the A. thaliana subgenome, a coverage-based optimization method was used. In this method, a reference with a known karyotype (here A. suecica) was used as a base for the prediction of copy numbers in the sample. Based on the aligned A. suecica reads, we established a reference coverage per copy of each chromosome (i), r_i, which was subsequently used to calculate an expected coverage for each chromosome if the copy number (x_i) is varied. As the search space of different combinations of copy numbers is traversed, the proportion of the expected coverage over the updated total is updated. Due to the condition of the total number of reads needing to stay the same, the updated proportion is normalized by the original total to get the actual expected coverage per chromosome. Each time the copy combination is varied and expected coverage updated, the deviation from the observed proportion of reads in the sample over the total observed reads for that particular chromosome is calculated. These differences are then summed up over the total number of chromosomes, and the optimal copy combination x that minimizes the sum of the differences is chosen as the predicted karyotype of the sample.

The general feasible region for x is [0, inf] can give rise to invariant or multiple optimal solutions. In order to prevent this, in the naïve method for this particular dataset, the feasible region was shrunk to [2, 6] as the most likely number of copies is no than 6 in this case. To further narrow down the optimal solutions, an upper limit of 2 copies was assumed for the A. arenosa chromosomes. Using the coverage information for each sample, a method selecting the optimal copy combination which minimizes the sum of the differences between the expected (for a known karyotype) and observed coverages was used. Placing an upper limit of 2 copies on the A. arenosa chromosome copies reduced the number of invariant or multiple optimal solutions and thus focused us on the aneuploidy predictions in the A. thaliana subgenome. This assumption allowed the following approach: Let there be n chromosomes in a particular sample and let the observed number of reads be y_i for each chromosome i in the sample. The sum of the observed reads over the n chromosomes is given by Y. For a reference genome with a known karyotype, let the coverage per copy for each chromosome i be given by r_i and let the total number of reads for the reference over all the chromosomes be T. For a given number of copies for a particular chromosome i, x_i, the expected number of reads for chromosome i is: (r_i · x_i/N) · T where N refers to the updated total amount of expected reads after changing the copy number of chromosome i. However, as the total number of reads for the reference should stay the same, the coverage computed for chromosome i was normalized against the original total for the reference (T) to give the actual expected number of reads. The copy combination that minimized the sum of the differences between the proportion of total observed reads and the proportion of total expected reads belonging to chromosome i was computed as follows: $\underset{x}{\mathit{ar}\mathit{gmin}}\sum _{i=1}^n\left|\frac{y_i}{Y}-\frac{r_i\cdot x_i}{N}\right|$. The method was implemented in Python.

To verify the results from this method, we also used the allele frequency we had generated for the analysis of meiotic recombination frequencies. As, for example, a partental allele frequency of 0.25 or 0.75 unambiguously indicates a parental ratio of 1:3 or 3:1 copies (and thus the expected copy number of 4), if we ignore extremely unlikely chromosome counts like 6:2 or 12:4, we used histograms of the sliding windows to determine the frequency of windows occurring at each allele frequency from 0.0 to 1.0. Peaks in the number of sliding windows corresponding to certain allele frequency ratio were used to validate the aneuploidy prediction for that particular sample. In cases where the allele frequency was 0.0, 0.5, or 1.0, these could not be unambiguously validated as these frequencies could result in more than one possible copy number.

Identification of homoeologous exchanges between A. thaliana and A. arenosa subgenomes

Significant coverage shifts occurring in the syntenic regions between A. thaliana and A. arenosa were determined using sliding window analysis of the coverage. The mean coverage of all the windows in a particular sample along a particular chromosome was computed. Coverage shifts of one copy or more at the chromosome arm ends were deemed as significant. The read density was plotted to observe the changes in this region and identify those regions where shifts and possible exchanges occurred. The approximate regions of the breakpoints were visualized with the Integrative Genome Viewer (Robinson et al. 2011).

Results and discussion

A multigenerational hexaploid pedigree of a single autoallopolyploidization event shows phenotypic diversity

To find mechanisms that might explain gene flow between congeners resulting in stable introgressions in just a few generations, we created a multigenerational pedigree of autoallohexaploids (Fig. 1). The mother plant resulted from a single cross of diploid A. thaliana (Col-0) whose genome we designated AA, and the natural allotetraploid A. suecica (Sue-1) whose genome we designated A’A’BB. Seeds from the original cross were mostly inviable, except for 1 viable plant. This resulting triploid F₁ plant was sterile except for 2 late-developing side-branches that had undergone spontaneous mitotic whole-genome duplication. These branches bore several siliques with a total of 37 viable seeds. Flow cytometry was used to verify ploidy levels and ascertained additive genome sizes of all analyzed F₂ individuals with the genomic constitution AAA’A’BB (Matsushita et al. 2012). Individuals within and between the lines showed a high degree of phenotypic variability with respect to flowering time, leaf serration, rosette diameter, and fecundity, but we observed no obvious phenotypic patterns that changed with increasing generation number (Fig. 2, Supplementary Fig. 1a). Fitness in this pedigree was assessed by measuring seed mass in a subset of the plants. Only 1 plant (4B in Fig. 2) was completely sterile, all other plants produced plump seeds. Aside from a few particularly fertile plants in F₁₆, individuals in all other generations produced seeds that were nonsignificantly different in weight from each other (Supplementary Fig. 1b).

Fig. 2.

Phenotypes of plant material. Individuals shown in the grid are sorted in rows (1–9) and columns (a–h) corresponding to the entries in Supplementary Tables 1 and 2.

For the current study, we advanced 6 of these lines via single seed descent over a total of 17 generations. We then grew multiple sisters from several generations of each line in the greenhouse to observe their phenotypes and to extract DNA for whole-genome sequencing. Each sample was sequenced with Illumina short-read sequencing with an average of 20× to 40× genome coverage per sample.

A genome assembly of A. suecica Sue-1 reveals a high degree of synteny between its subgenomic chromosomes

To ask questions about the genomic constitution of individual hexaploid plants, we first created a de novo chromosome-level assembly of the paternal parent, A. suecica (Sue-1). Initially, PacBio long reads were assembled into 482 contigs with a total length of 271.6 Mb, which were then polished with Illumina short reads. We also generated 453 optical consensus maps with a total length of 278.4 Mb to correct errors in the assembly and combine the contigs to scaffolds. Upon hybrid scaffolding using the optical consensus maps, the N50 of the assembly improved from 3.2 to 5.7 Mb and the number of PacBio contigs was reduced to 78 scaffolds with a maximum length of more than 15 Mb. The total length of the final assembly sequence including the nonscaffolded contigs was 275.3 Mb, closely resembling in length 2 recently reported assemblies of A. suecica measuring 272.4 Mb for accession As9502 (aka CS22509 or Sue-5; Wang et al. 2006; Jiang et al. 2021), and 262 Mb for accession ASS3 (Burns et al. 2021). As has been done before (Henry et al. 2014; Novikova et al. 2016), we generated a pseudoreference allopolyploid genome by concatenating the A. thaliana (Col-0) and A. lyrata reference sequences and used it as a guide for the generation of a final chromosome-level assembly. The final assembly (N50 = 19.06 Mb) included the sequences for the 13 chromosomes as well as nonscaffolded sequences and scaffolds that were not placed on chromosomes. The annotation of the A. suecica genome resulted in 60,688 protein-coding genes while the number of annotated TEs was 115,872.

Some inversions and other structural variations remained between the final assembly and the pseudoreference, especially between the A. arenosa subgenome and the A. lyrata reference, which we assumed to represent genuine differences between the 2 closely related species. In the alignment of the new assembly with the Col-0 accession, an inversion in chromosome 4 could be seen adjacent to the centromere. This inversion was already described (Fransz 2000) and has previously been reported to be present in other A. thaliana accessions (Zapata et al. 2016). Some of the structural variations between the A. thaliana and A. arenosa subgenomes of A. suecica (shown in Fig. 3) were similar to those observed between A. thaliana Col-0 and A. lyrata (Hu et al. 2011). These included an inversion between A. thaliana subgenome chromosome 1 (At1) and A. arenosa subgenome chromosome 2 (Aa2), and the inversion between At5 and Aa8. The translocation identified between the beginning of At3 and the beginning of Aa3 as well as the duplications between the beginning of At5 and the beginning of Aa6 are also found between A. thaliana Col-0 and A. lyrata (Hu et al. 2011).

Fig. 3.

The A. suecica genome shows a high degree of synteny within its subgenomes. The chromosomes of the A. thaliana (Ath) and A. arenosa (Aa) subgenomes of A. suecica were compared to each other. Translocations, inversions, and duplications between subgenomes are indicated. The syntenic blocks are similar to those previously reported between A. thaliana and A. lyrata (Hu et al. 2011). Blue lines represent the A. thaliana subgenome, orange lines the A. arenosa subgenome. The figure was generated with SyRI (Goel et al. 2019) and plotsr (Goel and Schneeberger 2022).

In summary, comparing our assembly of Sue-1 with the recently published A. suecica assemblies (Burns et al. 2021; Jiang et al. 2021) and genomic comparisons within the Arabidopsis genus (Hu et al. 2011) gave us confidence that this assembly, representing the paternal parent in our pedigree, would serve as a high-quality reference for the analysis of the autoallohexaploid population.

Meiotic recombination occurs between all 4 A. thaliana haplotypes

The autoallohexaploid used in this study contained the allotetraploid genome of A. suecica and the diploid genome of A. thaliana. The allotetraploid A. suecica was itself formed by a hybridization of A. thaliana and A. arenosa and it thus contains both homologous and homoeologous chromosomes (i.e. 4 haplotypes of A. thaliana and 2 of A. arenosa). Since the A. thaliana haplotypes (AA and A’A’) in the hexaploid progeny were derived from 2 different parental A. thaliana genotypes, the A. thaliana subgenome was heterozygous in itself (A vs A’) including the Col-0 accession (from the A. thaliana parent) and an accession similar to Borsk-2 from the Sue-1 parent (Carlson et al., 2017). The 2 A. thaliana-derived genomes were both homozygous implying that the 2 chromosomes from the same parent could not be distinguished. However, the haplotypes from different parents could easily be distinguished using SNP markers. Such genomic state suggests that both disomic and tetrasomic segregation could occur (Tate et al. 2005).

To determine the allelic makeup of the A. thaliana subgenomes in each of the hexaploid individuals, we aligned the Illumina short reads of the autoallohexaploids against the pseudoreference and the Sue-1 assembly. To determine if recombination had occurred between different A. thaliana haplotypes, we assessed the allele frequencies of Col-0 (AA) and Sue-1 (A’A’) along the chromosomes using read counts supporting either of the alleles at a chromosome-wide set of SNP markers within sliding windows (Fig. 4).

Fig. 4.

Evidence for frequent meiotic recombination between the Col-0 and Sue-1 A. thaliana haplotypes in the autoallohexaploid. a) An example of the sliding window analysis estimating the local sequencing coverage (black) and Sue-1 allele frequencies (red). Here, the individual shows 5 major shifts in the allele frequencies on chromosome At1. The chromosome features 3 regions with 2 haplotypes of Col-0 and Sue-1 (beginning, middle, and end) as well as 2 regions where the genome consists of 3 haplotypes of Sue-1 and only one of Col-0 (∼9–12 and ∼21–27 Mb), as well as a short region (∼19–21 Mb), where the genome fixed for the Sue-1 haplotypes. b) In this example, chromosome At1 shows fixation of the Sue-1 alleles at the beginning and at end of the chromosome (i.e. all 4 haplotypes are Sue-1 derived), and zero copies of the Col-0 haplotypes. In the middle of the chromosome, there are no Sue-1 alleles, thus indicating fixation of the Col-0 allele in that area. In both a) and b) the centromeric region (at around 15 Mb shows artifactual noise due to unreliable short-read alignments. c) Cartoon illustrating how homologous (meiotic) exchange can lead to fixed introgressions. Chromosomes on the left correspond to allele frequency plots in the middle and possible gametes arising from the respective generations on the right. d) Number of recombination events for each A. thaliana chromosome for each chromosome based on number of allele frequency shifts as exemplified in Fig. 4, a and b.

If both subgenomes, A and A’, contribute exactly half of the hexaploid’s A. thaliana complement we would expect an allele frequency of 50% for both of the alleles. Any deviation suggests that entire chromosomes or some chromosomal regions were more frequently derived from Col-0 or from Sue-1, for example allele frequencies of 0.25 or 0.75 would indicate that one of the parental haplotypes was represented in 1 or 3 copies, respectively. Chromosome segregation during meiosis, for example, can increase or decrease the number of one or the other parental haplotypes. Moreover, shifts in the allele frequency along a single chromosome are an indication of meiotic recombination (i.e. a crossover) between A and A’, examples of which (along with a cartoon explaining how the analysis was performed) are shown in Fig. 4. Our analysis showed that within the F₂ generation, all individuals displayed shifts of allele frequencies along most of their A. thaliana chromosomes (Fig. 4c) indicating frequent meiotic recombination between the interparental chromosomes already in the early generations. Even though tetraploid genomes need more generations for allelic fixation as compared to diploids, allelic fixation was common in the later generations, and thereby outlines the possibility of stable introgression of one of the alleles into all 4 A. thaliana-derived chromosomes (Fig. 4).

As the 4 A. thaliana chromosomal copies are structurally identical this can lead to the formation of chiasmata with more than 1 homolog, creating multivalents, which may, or may not segregate the 4 chromosomes properly during meiosis (Bomblies et al. 2016) and can lead to aneuploidy. Comparisons of natural diploid and autotetraploid populations of A. arenosa and Cardamine amara showed that chromosomes in some diploids are not inherently adapted to successfully undergo both diploid and tetraploid meiosis (Yant et al. 2013). Instead, a number of genes appeared to undergo selection during autopolyploidization, thereby stabilizing functions such as chromosome cohesion and segregation in the neo-autotetraploid (Hollister et al. 2012; Yant et al. 2013; Bohutínská et al. 2021).

Our data showed that in the hexaploid meiotic recombination occurred between the 2 A. thaliana genomes immediately after genome merger and continued for at least 17 generations (Fig. 4). Analogous to the observations in established autotetraploids, not restricting meiotic recombination across the 2 A. thaliana genomes might serve as one mechanism to allow for introgressions between haplotypes and, in addition, the selection of alleles best suited for survival in the new genomic environment.

Aneuploidy accumulates in the A. thaliana subgenomes

Cytological analysis of early generation (F₂–F₇) individuals within the hexaploid pedigree was previously performed and showed that chromosome number in autoallohexaploid individuals varied from 27 to 39 (Matsushita et al., 2012), while the expected euploid number in this population is 36 (Col 2x, N = 10, Sue-1 4x, N = 26). Most of the variation in these early generations occurred in the A. thaliana subgenomes (Matsushita et al. 2012). In the current study, we developed a read coverage-based algorithm to calculate the number of chromosomes in each of the sequenced samples (see Methods for details), which we used in lieu of conventional cell-based karyotyping. Using this algorithm, we analyzed a total of 210 A. thaliana chromosomes (42 lines times 5 At chromosomes) in the hexaploid individuals (Fig. 5, Supplementary Table 3). Of these, the copy number of 28 chromosomes (∼13%) deviated from the expected number of 4 copies. Within this group, 20 chromosomes showed aneuploidy only in one of the 2 or 3 sister individuals of the same generation. In the other 8 cases, multiple same-generation sisters were found to have similar copy numbers, suggesting possible karyotype destabilization through the earlier occurrence of aneuploidy. Out of the total 42 lines analyzed, only 20 individuals (∼48%) had a euploid A. thaliana chromosome complement. Fourteen of the individuals (∼33%) had a total chromosome count that was less than the expected 20 A. thaliana chromosomes, and 8 (∼19%) had more than 20 A. thaliana copies. To validate the copy number predictions from our count-based algorithm (Fig. 5, Supplementary Table 3), we again used the allele frequencies of the parental alleles along the chromosomes (Fig. 4). A frequency of 0.75, for example, indicates that the ratio of chromosome copies from the Sue-1 parent to those from the Col-0 parent is 3:1, suggesting a total number of 4 copies. An allele frequency of, for example, 0.6 or 0.8, on the other hand, would indicate a ratio of 3:2 or 4:1, respectively, and hence a total copy number of 5. In all cases, where validation was unambiguously possible (in 150 out of 210), the algorithm-based chromosome copy calculations were validated by the allele frequencies. An example of this validation is shown in Supplementary Fig. 2.

Fig. 5.

Read count analysis predicts no obvious pattern of aneuploidy or diploidization in the allohexaploid pedigree. Sample numbers are ordered from earliest to latest generation (y-axis). The number of haplotype copies for each chromosome of the A. thaliana subgenome is shown on the x-axes. The euploid haplotype number of each chromosome of the A. thaliana subgenome is 4.

In some lines seemingly stable chromosome numbers in an earlier generation became unstable again in later generations (Fig. 5, Supplementary Table 3). For example, in line 12 in the F₆ generation both sisters (F₆ 12-1 and F₆ 12-2) appear to have 5 copies of chromosome 5 while all 3 sisters of that same line in the F₁₄ generation show the expected count of 4 copies for chromosomes 5. Similarly, we observed only 3 copies of chromosome 4 in 2 sisters in the F₂ generation of line 19, yet in the F₁₂ and F₁₇ generations of this line, all individuals had the full complement of chromosomes 4. While this might at first sight seem surprising, it is important to point out that our pedigree analysis did not sample direct descendants in subsequent generations but those descending from a separate, unsampled sister of any sampled individual. We therefore only see snapshots in time of the generation sampled. In some cases, genome fractionation (Wendel et al. 2018) appeared to occur in later generations. For example, in line 14 our analysis found only 2 copies of chromosome 2 in both sisters in the F₁₁ generation (Fig. 5). Overall, the chromosome copy numbers did not appear to follow any obvious pattern or trend toward chromosome stabilization or diploidization.

Our analysis suggested that aneuploidy is frequent in all generations of this autoallohexaploid population and is reminiscent of findings in the resynthesized allotetraploids B. napus (Xiong et al. 2011), and Tragopogon miscellus (Chester et al. 2012). Similarly, in a population of resynthesized allohexaploid wheat, aneuploidy was rampant in the first 10 generations (Zhang et al. 2013). Even in resynthesized lines that were specifically bred by euploid selection and selfing to propagate the line to the 10th generation, these researchers found equally as many aneuploids in euploidy-selected lines as in lines that had descended from aneuploids (Zhang et al. 2013). In our study, as in those in Brassica, Tragopogon, and wheat, the analyzed samples do not directly reflect the karyotype of the gametes that the analyzed samples used to produce their offspring. It is thus possible that selection for euploidy occurs at the stage of zygote formation and that postzygotic accumulation of aneuploidy in somatic tissues or in unsuccessful gametes simply does not interfere with evolution to present a severe enough problem to be selected against on the individual level. In fact, the observation described above that later generations “regain” chromosomes toward the full euploid complement, seems to suggest that aneuploid individuals sampled by us in earlier generations were unlikely to stably pass on a new karyotype to subsequent generations. Likewise, it is possible that by selecting fertile individuals during the construction of the pedigree we inadvertently selected for euploid individuals, but when we regrew plants from our seed stocks for sequencing leaf material when fertility was not required, we were more likely to grow aneuploid individuals. Thus, rather than introducing stable variation into the population as could be the case during recombination, aneuploidy might be responsible for selection against certain individuals, thus preventing any previous introgression events (e.g. by recombination) from remaining in the population. Given that aneuploids in this population were most likely to either produce (still unstable) aneuploid offspring or to become a dead-end in the gene pool, aneuploidy could be regarded as a counter-mechanism to stable introgression that essentially prevents individuals with excessive introgression or detrimental loss of whole chromosomes to take a foothold in the evolving population.

Diploidization is a commonly found process in which polyploids slowly return to a diploid state. During this process, repetitive DNA is eliminated via unequal and illegitimate recombination, and entire chromosomes can be lost due to aneuploidy (Soltis et al. 2015). In this context, it is of interest to note that we initially had also observed several lines within our pedigree that had reverted back to a karyotype resembling allotetraploidy much more so than autoallohexaploidy (Supplementary Fig. 3). Upon closer examination, these lines contained cryptic SNPs that did not stem from the parental genomes of Col-0 or Sue-1, but instead likely indicated cross-pollination with different natural A. suecica lines that were growing in the greenhouse at the same time as our advanced autoallohexaploids. We removed these plants from subsequent analyses in this study. It seems likely that the backcross of the hexaploids to natural tetraploid A. suecica produced pentaploid offspring, which then naturally decayed to tetraploids, however, we cannot exclude the possibility that the reverted lines had acquired favorable, stabilizing alleles from the natural A. suecica populations (Henry et al. 2014). But whatever the reason for the reversion to allotetraploidy might have been, the effect is a fixation of the introgressed Col-0 DNA into A. suecica, presenting a possible mechanism for the occurrence of new haplotype DNA in the parental species. Interestingly, some studies have reported natural chromosomal variation in A. suecica (Harmaja and Pellinen 1990), which potentially could indicate dysploidy resulting from naturally occurring hybridization events of A. suecica with A. thaliana and thereby could indicate that such introgressions are also occurring in nature. In this context, it may be of note that introgressions of foreign A. thaliana alleles could result in genome variation in A. suecica that would look like genomic patterns supporting multiple origins (despite resulting from introgressions; Novikova et al. 2017).

Interestingly, while genome evolution in autopolyploids is often the result of problems associated with multivalent formation (Lloyd and Bomblies 2016; Bohutínská et al. 2021), genomic change in allopolyploids is often caused by chromosomal translocations, transcriptome shock, and epigenetic changes (Hegarty et al. 2013; Wendel et al. 2016). This population of autoallohexaploids, although riddled with aneuploidy, appeared cytogenetically relatively stable. While over the 17 generations some sub-lines would die out occasionally (see for example individual F₅ 19-1, B4 in Fig. 2) chromosome number fluctuations did not appear to commonly create a ratchet effect of successively more aneuploidy as was previously reported in resynthesized allopolyploid B. napus (Gaeta and Pires 2010; Xiong et al. 2011). It is possible that the larger hexaploid auto- and allopolyploid genome constitution in our population exerted some sort of genomic buffering function maintaining enough fitness despite repeated chromosome gains or losses to allow for continued intergenomic exchange in subsequent generations before a possible collapse back to allotetraploidy.

Homoeologous exchanges occur between the A. thaliana and A. arenosa subgenomes

Given the large amount of genetic exchange within the autopolyploid subgenome and the wide stretches of syntenic regions between the A. thaliana and A. arenosa subgenomes in the autoallohexaploid (Fig. 3, Hu et al. 2011), we asked if, and to what degree, homoeologous exchange could occur. We hypothesized that homoeologous exchanges would be visible in coverage plots under the following conditions: If homoeologs pair, chromosome arms could be swapped between the A. thaliana and A. arenosa subgenomes and the intrasubgenomic recombinant chromosomes would then segregate according to their centromere identity like any other chromosome within the subgenome. In the next generation, the exchanged chromosome arm would then be subject to segregation. This could lead to an overall gain or loss of copies of the exchanged material, which subsequently could be observed as change in the coverage in both subgenomes. We therefore inspected the coverage plots of each individual in regions of high intrasubgenomic sequence homology. If read coverage changed on one chromosome in a highly homologous area of one subgenome then one would expect to see evidence of a reciprocal change also in the homologous area of the other subgenome (Fig. 6). Over time, segregation could then lead to some of the homeologous exchanges to become fixed, and could lead to genomic loss or gain during the evolution of the population.

Fig. 6.

Identification of homoeologous exchanges. a) Chromosomes from the A. thaliana (blue) and A. arenosa (red) subgenomes share syntenic homologous regions. The coverage plot next to the chromosomes shows the expected relative coverage for both chromosomes during short-read sequencing. Four possible groups of chromosomes (karyotypes) are shown illustrating (1) the normal karyotype, (2) chromosomes after a homoeologous exchange occurred, (3) the chromosomes of the next generation when a homoeologous exchange got fixed due to chromosome segregation, and (4) a reciprocal exchange, which—when balanced—cannot be found in the coverage plots of the short-read resequencing. b) Example of a homoeologous exchange between chromosomes At1 and Aa1 in line 14 and how it is propagated through generations. Sliding window coverage plots of generation 2 (3 sisters corresponding to the 3 lines in the graph), generation 7 (2 sisters), and generation 11 (2 sisters). In the F₂ generation, one of the 3 sisters shows a homoeologous exchange between the beginning of A. thaliana subgenome chromosome 1 (top) and the beginning of A. arenosa subgenome chromosome 1 (bottom, highlighted by the orange box). The same exchange (loss of region in At1 and gain of region in Aa1) is observed in all samples of all later generations (F₇ and F₁₁).

Overall, we observed a total of 19 individuals where such homoeologous exchanges occurred (Table 1). In several of these individuals more than one exchange, usually on different chromosomes, was seen. For example, in line 14, the genetic exchange was initially observed in the F₂ generation in only one of the sisters (Fig. 6). The pattern of variable copy numbers in this genomic region is maintained throughout the subsequent generations. In lines 6 and 14, coverage shifts of at least 1 copy were seen in the syntenic regions at the ends of At1 and Aa2, as well as at the beginning of At1 and Aa1. In line 6 (Supplementary Fig. 4), we observed a pattern of homoeologous exchanges at the ends of At1 and Aa2, where an increase of 1 copy is observed at the end of the A. thaliana subgenome for both tested sisters in generations F₂ and F₇ with a reciprocal loss of DNA in the corresponding region of the A. arenosa subgenome (Supplementary Fig. 4). In the F13 generation of this line, a copy number increase of 2 copies is seen on chromosome At1 while complete loss (i.e. a coverage of zero) of the reciprocal area on Aa2 suggests fixation of this exchange in that line as early as in the F₇. The coverage increase on At1 in one of the F13 sister lines suggested fixation of the genomic material from A. arenosa on the chromosome of A. thaliana (Supplementary Fig. 4).

Table 1.

Summary of homoeologous exchanges between subgenomes.

Line	Generation	Regions	Approximated breakpoint locations (Mb)
5	F2 (1)	Start At5 (↓)/Start Aa6 (↑)	1.5/1.7
6	F2 (2), F7 (2), F13 (2)	End At1 (↑)/End Aa2 (↓)	27.0/10.8
14	F2 (1), F7 (2), F11 (2)	Start At1 (↓)/Start Aa1 (↑)	3.8/4.4
19	F12 (1), F17 (1)	Start At1 (↑)/Start Aa1 (↓)	4.4/5.0
	F12 (1)	Start At1 (↓)/Start Aa1 (↑)	1.9/2.3
	F17 (1)	Start At1 (↓)/Start Aa1 (↑)	4.4/5.0
	F17 (2)	End At5 (↓)/End Aa8 (↑)	24.9/13.5
	F17 (1)	End At5 (↑)/End Aa8 (↓)	24.9/13.5

Arrows indicate whether the coverage (i.e. copy number) increased or decreased in these regions. Fixation of the subgenome exchange between the end of At1 and the end of Ae7 can be observed in all individuals sequenced in line 6 from F₂ to F₁₃. Line 14 shows propagation of the subgenome exchange pattern found in one of the sisters of the F₂ generations to the later generations. In line 19, diverging patterns of coverage shifts are observed in the same regions between same-generation sisters in F₁₂ and F₁₇ generations.

Line: line number; Generation: generation within the line in which the homoeologous exchange was observed (parentheses indicate the number of individual sisters in the given generation where the exchanges occurred); Regions: regions where exchanges occurred; Breakpoint Locations: indicates the approximated genomic location where homoeologous exchanges occurred.

Another example was seen in line 19 (Supplementary Fig. 5) where homoeologous exchange between At1 and Aa1 occurred at the top of the chromosomes in 2 regions. In this case, both sisters showed exchange in opposite directions which is reciprocated between the 2 subgenomes. The 3 sisters in the F₁₇ of this line exemplified a case where each of the 3 individuals varies in copy number within this region of chromosome Aa1 (Supplementary Fig. 5). A similar example in the same line showed homoeologous exchange between chromosomes At5 and Aa8 (Supplementary Fig. 6). With the exception of line 19, the general genomic regions where the breakpoints were located are consistent between sisters and within each line. Although genetic exchanges were found in similar syntenic regions in both lines 14 and 19, the breakpoints are in slightly different locations (∼500 kb apart).

Overall, our data suggested that homoeologous exchanges appeared at a comparatively high rate in the autoallohexaploid offspring (Table 1), in stark contrast to the established allotetraploid A. suecica Sue-1 where we only found one instance of a short homoeologous exchange at the end of chromosome 2. Burns and colleagues (2021) recently reported homoeologous exchanges in 3 of 15 analyzed A. suecica lines, in which a total of 2 different types of homoeologous exchanges were documented (Burns et al. 2021). Homoeologous exchange is thus much more frequent in autoallopolyploids than in established allotetraploids although these types of exchanges apparently can either still occur in the established polyploid or can become fixed in natural populations.

In summary, our data showed how segregation can lead to the fixation of homoeologous exchanges in the autoallohexaploid, and suggests, in addition to intergenomic meiotic recombination discussed earlier, a second mechanism of how genetic material from one species can introgress into the genome of another species while still leading to viable offspring without any apparent effects on fitness.

Conclusions

As in several other genera, there is incomplete reproductive isolation between the species in the Arabidopsis genus, and pervasive gene flow exists between species in this genus (Schmickl and Koch 2011; Novikova et al. 2016). Polyploids have been implicated as one way to facilitate interspecific gene flow, specifically due to the reduced barriers to gene exchange in their compound genome, and for that reason have been referred to “allelic sponges” (Schmickl and Yant 2021) that can increase genetic diversity. Furthermore, hybridization of a reduced and an unreduced gamete via a triploid intermediate, also known as a triploid bridge, has been described as a frequent way to produce tetraploids (Ramsey and Schemske 1998; Schatlowski and Köhler 2012). As our analysis was based only a single cross, it is difficult to extrapolate relevance or frequency of this pathway in nature. However, our data did demonstrate that an intermediate autoallohexaploid arising via an F₁ triploid bridge can provide a mechanism for rapid gene flow between species. Such gene flow can occur using intergenomic legitimate meiotic homologous recombination as a tool for positive or neutral change. Aneuploidy, on the other hand, appears to act as a mechanism preventing evolutionarily stable gene flow and introgression. Homoeologous recombination, as a third mechanism in the response to interspecific genome merger, seems to be able to act both as a mechanism preventing gene flow when it results in instability, or as an agent of facilitating introgression if it occurs against the background of a big enough “genomic buffer” where loss or gain of chromosomal regions does not lead to detrimental change.

Diploidization and genome fractionation are major forces in the constant cycle of polyploidy and diploidy (Soltis et al. 2015). It is thus likely only a matter of time when this autoallohexaploid population will fully revert back to an allotetraploid or diploid state (Baduel et al. 2018). While it is unknown at this point if or how quickly this reversion might occur naturally, the structural differences and vast introgressions seen at this stage would likely remain as a reminder of its former intermediate polyploid state.

Data Availability

The read data for both the allohexaploid samples as well as A. suecica Sue-1 were submitted to the National Center for Biotechnology Information (NCBI) under the BioProject accession number PRJNA753993, where also the assembly of Sue-1 can be found. The names for the hexaploid samples submitted begin with “Hex” followed by the generation, line number, and sister. For example, HexF2-2-1 refers to sister 1 in the F2 generation of line 2. The scripts for the chromosome count predictions are found on the GitHub website (https://github.com/schneebergerlab/Hexaploid_MA_lines).

Supplemental material is available at GENETICS online.