Tetrasomic recombination is surprisingly frequent in allotetraploid Arachis.

Arachis hypogaea L. (cultivated peanut) is an allotetraploid (2n = 4x = 40) with an AABB genome type. Based on cytogenetic studies it has been assumed that peanut and wild-derived induced AABB allotetraploids have classic allotetraploid genetic behavior with diploid-like disomic recombination only between homologous chromosomes, at the exclusion of recombination between homeologous chromosomes. Using this assumption, numerous linkage map and quantitative trait loci studies have been carried out. Here, with a systematic analysis of genotyping and gene expression data, we show that this assumption is not entirely valid. In fact, autotetraploid-like tetrasomic recombination is surprisingly frequent in recombinant inbred lines generated from a cross of cultivated peanut and an induced allotetraploid derived from peanut's most probable ancestral species. We suggest that a better, more predictive genetic model for peanut is that of a "segmental allotetraploid" with partly disomic, partly tetrasomic genetic behavior. This intermediate genetic behavior has probably had a previously overseen, but significant, impact on the genome and genetics of cultivated peanut.

WHOLE-genome duplications have occurred in virtually all angiosperm lineages at some time and are believed to be a major driving force in their evolution (Cui ). They generally result in polyploids that are new biological species with increased heterosis and robustness and whose genomes have new possibilities for gene and genome evolution (e.g., Adams and Wendel 2005). There is believed to be a correlation between polyploidy and the ability to adapt to new environments (Te Beest ), and many crop plants are polyploids, perhaps reflecting an increased adaptability to, and increased yield under, cultivation (Renny-Byfield and Wendel 2014).

Perhaps the most frequent class of polyploidy are tetraploids, the result of the duplication of a diploid genome. These are categorized as autotetraploids that result from chromosome doubling of a nonhybrid genome or as allotetraploids that result from hybridization between two species and subsequent chromosome doubling. In autotetraploids, all chromosome sets can pair in meiosis leading to tetrasomic inheritance. Allotetraploids generally undergo disomic recombination because only homologous chromosomes pair at meiosis, but homeologous chromosomes do not. However, some allotetraploids have meiotic behavior that is intermediate between these two extremes: while most chromosomes pairings in meiosis are between homologous chromosomes, some pairing and genetic recombination also occurs between homeologous chromosome pairs (Wu, ).

Arachis hypogaea L. (cultivated peanut), an allotetraploid (2n = 4x = 40) with an AABB genome type, is believed to be derived from the wild diploid species A. duranensis Krapov. & W.C.Greg. and A. ipaënsis Krapov. & W.C.Greg. that contributed the A and B genomes, respectively (Kochert ; Seijo ; Moretzsohn ). These two component genomes diverged an estimated 2.9 MYA (Moretzsohn ). Although relatively recent, this divergence should be put in the biological context of the unusual reproductive biology of Arachis species. Their high rates of self-pollination, the peculiar habit of underground pod and seed development, short-distance seed dispersal, and consequent frequent population bottlenecks are likely linked to the accumulation of Bateson–Dobzhansky–Muller incompatibilities and the high estimated rates of speciation (0.95 speciation events per million years, compared to 0.15 for legumes in general (Magallon and Sanderson 2001; Moretzsohn ). Accordingly, diploid hybrids derived from crosses between A-genome and B-genome species are highly infertile (Krapovickas and Gregory 1994; Krapovickas and Gregory 2007; Mallikarjuna ).

Since most wild species of Arachis are diploid, synthetic allotetraploids have been developed to enable their use in peanut breeding programs (Simpson ; Fávero ; Mallikarjuna ; Leal-Bertioli ). These plants are usually produced by crossing diploid species with A and B genomes, followed by a treatment of the sterile F1 hybrid with colchicine to induce chromosome doubling. The resultant tetraploid plant can be crossed with cultivated peanut. Based on cytogenetics and molecular mapping, A. hypogaea and induced allotetraploids derived from wild diploid species have been considered to have classic allotetraploid genetic behavior. Using this assumption, numerous linkage map construction and QTL identification studies have been successfully carried out (Burow ; Varshney ; Hong ; Fonceka ; Gautami ; Qin ; Shirasawa , 2013; Sujay ; Wang ; Zhou ).

However, in three data sets we observed patterns among unexpected data points. The expression of numerous A. ipaënsis transcribed regions “disappeared” in a derived artificially induced allotetraploid, and, unexpectedly, these regions were genetically clustered. Likewise, in recombinant inbred lines (RILs) derived from a cross of cultivated peanut with an induced allotetraploid, “missing” PCR marker data showed unexpected genetic clustering. Finally, rare unexpected single nucleotide polymorphism (SNP) genotypes, unexplainable with disomic recombination, were observed in both the induced allotetraploid itself and RILs. We realized that these data could be explained by a degree of tetrasomic genetic recombination. Here we report a systematic examination of the data sets to test this hypothesis.

Materials and Methods

Plant material

Plants were maintained in pollinator-free greenhouses, and seeds were saved from year to year. Accessions of the diploid species A. ipaënsis KGBSPSc30076 (hereafter referred to in the abbreviated form K30076), A. duranensis VNvEv 14167 (hereafter referred to in the abbreviated form V14167), and their derived colchicine-induced allotetraploid [A. ipaënsis × A. duranensis]4x (termed an amphidiploid in the original publication) (Fávero ) were obtained from the Active Germplasm Bank of Embrapa Genetic Resources and Biotechnology (Brasília, Brazil). The tetraploid cultivated peanut, A. hypogaea cv. Runner IAC-886, was from the Instituto Agronômico de Campinas (Campinas, Brazil).

Recombinant inbred lines in the F6 generation were derived from a cross between A. hypogaea cv. Runner IAC-886 and the colchicine-induced allotetraploid. These RIL lines are the same population described in Shirasawa and Bertioli . Lines were produced from a single F1 plant, cloned by cuttings to produce enough F2 seeds and single-seed descent to the F6 generation.

The colchicine-induced allotetraploid was derived from chromosome doubling and so is completely homozygous. Our assumption, shown here to be false, was that it would be genetically stable over generations (see Results). Therefore, here we will indicate the number of generations the induced allotetraploid plants were removed from the original polyploidy event for each experimental procedure. For the production of the RIL population, the parental colchicine-induced allotetraploid was only one or two generations removed from the original polyploidy event. For the DNA control for genotyping, the induced allotetraploid was about seven generations removed, and for expression analysis, the induced allotetraploid was about nine generations removed.

Comparison of gene expression of the two-diploid species and their derived allotetraploid

The colchicine-induced allotetraploid [A. ipaënsis K30076 × A. duranensis V14167]4x (Fávero ) and its diploid parents were grown in the greenhouse at the same time and under the same conditions. The induced allotetraploid was an approximately ninth generation progeny derived by single-seed descent from the original polyploidy event. Total RNA was extracted from the first expanded leaf of the main axis using the Qiagen Plant RNeasy kit (Qiagen) with on-column DNAse treatment. Complementary DNA (cDNA) libraries were constructed using equal amounts of RNA from five individuals of each genotype using the TruSeq v2 library construction kit (Illumina). To obtain long reads to improve transcriptome assemblies, size-selected libraries were sequenced using MiSEQ v3.0. For gene expression analyses, libraries were sequenced using HiSEQ 2000 2x150 paired-end sequencing (Illumina).

Adapter and quality trimming was performed using Trim_galore! v0.3.5. (http://www.bioinformatics.babraham.ac.uk/projects/trim_galore/). Transcripts were assembled using Trinity with default parameters (Haas ). Assembled transcripts were filtered to include only the longest isoform from each read cluster. The longest isoforms were then aligned to each other using NCBI blastn v2.2.29. Alignments with 100% sequence identity and ≥90% sequence length were considered redundant and removed from the final assembly.

Mapping of reads from A. duranensis, A. ipaënsis, and the ninth generation induced allotetraploid [A. ipaënsis × A. duranensis]4x onto a joint database of assembled A. ipaënsis and A. duranensis transcripts was performed using bowtie2 v2.2.2, tophat v2.0.11, and cufflinks v2.2.1 (Trapnell ). The densities of mapping (FPKM) were used as measures of expression of the diploid species transcribed regions in the diploid and tetraploid contexts. The use of a joint database helped to avoid erroneous cross-comparison of A- and B-genome transcripts, which have high similarity, because each read from the induced allotetraploid could map only once onto either an A. ipaënsis or an A. duranensis transcript. Furthermore, we considered counts for only A. duranensis assembled transcripts to which no A. ipaënsis reads mapped and A. ipaënsis transcripts to which no A. duranensis reads mapped (“cleanly mapping reads”).

Transcripts that were expressed in the diploid species, but not in the induced allotetraploid, were identified by processing data in Excel with regard to statistical significance of the comparisons.

Transcripts corresponding to sequences that had been previously genetically mapped onto maps from two different crosses, both having A. duranensis as one of the parents (Nagy ; Shirasawa ), were identified using NCBI blastn v2.2.29.

Genetic markers and maps

Simple sequence repeat (SSR) and transposable element PCR-based markers genotyping and map data were from Shirasawa , and SNP marker data were from a 1536 Illumina GoldenGate SNP assay (Nagy ; Bertioli ). The SNPs for this assay were called between two A-genome accessions: A. duranensis PI 475887 (KSSc 36036) and A. duranensis Grif 15036 (PI 666084; WiSVg 1510-B).

In the search for tetrasomic recombinants, we used both standard polymorphic SNP markers (here referred to as “intragenomic SNPs”; see Figure 1) and SNP markers that were not polymorphic for this population, but which harbored different bases on the A and B genomes (here referred to as “intergenomic SNPs”; see Figure 2). Intergenomic SNPs provided unexpected key information for the study. Being nonpolymorphic, they could not be genetically mapped on the tetraploid mapping population used here. Therefore, the diploid SNP A-genome map generated by Nagy using an F2 population derived from two A. duranensis accessions was used to position markers. Linkage groups were reoriented where necessary to ensure forward compatibility with pseudomolecules produced by the Peanut Genome Sequencing Consortium (http://www.peanutbase.org).

Figure 1

Examples of plots of signal intensities generated using GenomeStudio from “intragenomic” SNP assays on the RIL population derived from A. hypogaea and [A. ipaensis × A. duranensis]4x; parentals are indicated by orange and yellow arrows, respectively. Diploid controls, A. duranensis and A. ipaensis, are indicated with green and red arrows, respectively. For A, tetraploid parental genotypes are CaCaCbCb and AaAaCbCb (A and C refer to the DNA bases; the superscript letter refers to the subcomponent genomes; superscript “x” is used when it is not possible to assign the subgenome). For the classic allotetraploid model, progeny RILs are expected to form two frequent disomally homozygous clusters (CaCaCbCb and AaAaCbCb) with an additional cluster with some residual heterozygotes (AaCaCbCb). For the autotetraploid model, five clusters are expected. Note that this example broadly conforms to the expectations of the classic allotetraploid model, but three progenies are tetraplex for the base A, and one progeny may be triplex for A (in parentheses). These genotypes can be formed only by tetrasomic recombination. Note that genotypes that are duplex, monoplex, or nulliplex for the base A may also be the result of tetrasomic recombination, but they are indistinguishable from genotypes derived from disomic recombination. Therefore, this assay can reliably score only one (AaAaAaAa) of the four most frequent genotypes resulting from tetrasomic recombination (the others being AaAaCaCa, CaCaCaCa, CbCbCbCb). Example in B conforms to the classic allotetraploid model only approximately. Although the two expected clusters are the most populated, five clusters of genotypes are clearly visible. This SNP assay is an extreme example in this respect.

Figure 2

Example of a plot of signal intensities generated using GenomeStudio from an “intergenomic” SNP assay on the RIL population derived from A. hypogaea × [A. ipaënsis × A. duranensis]4x. These parentals are indicated by orange and yellow arrows, respectively. Diploid controls for A. duranensis and A. ipaënsis are shown as green and red arrows, respectively. For the classic allotetraploid genetic model, no recombinants are expected, but here four progenies have tetraplex/nulliplex genotypes. These genotypes can arise only from tetrasomic recombination. These intergenomic assays can score all genotypes resultant from tetrasomic recombination.

Analysis of genotyping data

Genotypes resultant from tetrasomic genetic behavior were identified by structured inspection of the results of genotyping assays (Shirasawa ; Bertioli ). This inspection was different from marker scoring for creating standard genotyping data for linkage mapping: it was focused on data points that did not fit the expected disomic model of recombination and assays that were monomorphic between tetraploid parents. Tetrasomic and null alleles were identified in genomic regions from SNP markers and from PCR-based SSR and transposable element markers (the full rationale for the scoring is explained in the Results).

Single-marker association and quantitative trait loci analyses

To investigate if tetrasomic recombination was influenced by genetic factors, we carried out quantitative trait loci analyses using, as a phenotype, the number of null PCR marker data points for each RIL from the PCR marker linkage map of Shirasawa with only codominant markers (supporting information, File S1, worksheet “PCR and SMA,” line 1205).

Single-marker analysis was done by comparing the association between a marker genotype and the trait value using simple linear regression. Composite interval mapping (CIM) (Zeng 1993, 1994) was executed in WinQTL Cartographer, version 2.5 (Wang ). CIM analysis was performed using the Standard Model 6, scanning the genetic map and estimating the likelihood of a QTL and its corresponding effects at every 1 cM, while using eight significant marker cofactors to adjust the phenotypic effects associated with other positions in the genetic map. A window size of 10 cM was used, and therefore cofactors within 10 cM on either side of the QTL test site were not included in the QTL model. Threshold was determined by permutation tests (Churchill and Doerge 1994; Doerge and Churchill 1996), using 1000 permutations and a significance level of 0.05.

Results

Phenotypic observation of parentals and RILs

Over almost a decade of observation of the diploid and tetraploid parental accessions, the induced allotetraploid lines seemed phenotypically stable.

Through the advancement of over six generations, the RILs derived from cultivated × induced allotetraploid were vigorous and fertile. This observation is consistent with the current consensus that A. ipaënsis and A. duranensis are the genome donors to A. hypogaea, and so an induced allotetraploid derived from a hybrid between them effectively “re-creates” the species A. hypogaea (or more strictly speaking, A. monticola Krapov. & Rigoni, which is the wild form of A. hypogaea) (Grabiele ).

HiSeq sequencing of cDNA libraries produced 20,843,737 reads of A. ipaensis, 18,393,012 reads of A. duranensis, and 48,644,287 reads of the induced allotetraploid. The software Trinity produced 79,828 assembled transcripts with a contig N50 of 1330 bp for A. ipaënsis and 82,089 assembled transcripts with a contig N50 of 1625 bp for A. duranensis.

The software pipeline of bowtie2, tophat, and cufflinks mapped 90.7, 91.4, and 89.9% of the reads from A. ipaënsis, A. duranensis, and the induced allotetraploid, respectively, to the combined database of A. ipaënsis and A. duranensis assembled transcripts. A total of 15,099 assembled A. ipaënsis transcripts were cleanly mapping; that is, only A. ipaënsis reads, and no A. duranensis reads, mapped to them. A total of 12,685 assembled A. duranensis transcripts were cleanly mapping; that is, only A. duranensis reads, and no A. ipaënsis reads, mapped to them.

Of these, 909 A. ipaënsis and 35 A. duranensis transcripts were expressed in the diploid state but not in the induced allotetraploid (null expression transcripts). Searches against genetically mapped sequences (Nagy ; Shirasawa ; Shirasawa ) showed that the null expression A. ipaënsis transcripts were disproportionally located and broadly distributed on linkage group 04 (LG04) (Table 1; Figure 3). (Here, when referring to both A- and B-genome homeologous linkage groups, we will use the generic “LG.” When referring specifically to A or B genomes, we will use “LGA” or “LGB.”) Furthermore, on LG04, there were disproportionally few A. ipaënsis transcripts for which there was robust support for expression in the tetraploid (FPKM >10) (Table 1, Figure 3). This genetic grouping of B-genome null expression transcripts suggests a loss of alleles rather than silencing that would tend to be scattered over the genome. Furthermore, inspection of SNP and PCR-based marker data also indicated loss of B-genome alleles on LG04 (see below).

Table 1

A. ipaënsis transcripts positioned on Nagy and Shirasawa maps that are expressed in diploid leaf and have null expression or positive expression in the ninth generation induced allotetraploid leaf

	Expression of A. ipaënsis transcripts in the ninth generation induced allotetraploid
Mapped sequences	Null	Positive (>10 FPKM)
LG01	6	49
LG02	0	24
LG03	1	62
LG04	46	6
LG05	0	26
LG06	2	35
LG07	1	16
LG08	1	33
LG09	0	32
LG10	4	25

Transcripts that have “disappeared” disproportionally from LG04 are in boldface.

Figure 3

Distribution of A. ipaënsis transcripts positioned on LG04 of the Nagy map that are expressed in diploid leaf and have null expression (in green) or positive expression (FPKM >10; in red) in the induced allotetraploid [A. ipaënsis × A. duranensis]4x leaf. Distances are in centimorgans. Other LGs have very few null expressed transcripts and many more expressed transcripts (see Table 1). The A. ipaënsis null expressing transcripts are widely distributed on LG04. This disappearance of A. ipaënsis transcripts is consistent with marker data that show that alleles on LG04 have become nulliplex for B alleles and tetraplex for A alleles (see first column in Figure 5).

Figure 5

A color-map representation of the 47 RILs derived from A. hypogaea × [A. ipaënsis × A. duranensis]4x with >2% of SNP markers indicating genotypes derived from tetrasomic recombination plus one control colchicine-induced allotetraploid ([A. ipaënsis × A. duranensis]4x) that was about seven generations removed the original polyploidy event (far left genotype). The genotypes, inferred from inter- and intragenomic SNPs (Figure 1 and Figure 2), are represented in columns and markers on the horizontal (complete data are in File S1). Markers are ordered with respect to the A-genome linkage map of Nagy . Linkage groups are delimited by horizontal gray lines. Only tetrasomic regions supported by at least two adjacent informative markers are shown. Green regions are tetrasomic for A-genome alleles and nullisomic for B-genome alleles. Red regions are nullisomic for A-genome alleles and tetrasomic for B-genome alleles. Tetrasomic recombination events often cover large portions of linkage groups. Tetrasomic/nullisomic calls were almost absent from LG01 (none shown here). In LG03, they are most frequent: 20 lines represented here have large tetrasomic/nullisomic regions of this linkage group. Note that the allotetraploid [A. ipaënsis × A. duranensis]4x has undergone tetrasomic recombination that covers all informative markers on LG04; this is consistent with PCR marker (File S2) and expression data that are also presented in this article (see Figure 3).

Identifying tetrasomic/nullisomic genomic regions from SNP markers:

For SNP assays of the type used for genetic mapping, the contrasting bases are harbored on homologous bases of the same component A or B genome; in other words, the SNP assay is “intragenomic.” For this type of assay, data points generally clustered into two main groups on GoldenGate signal intensity plots in accordance with the expected genotypes for an allotetraploid RIL population. These correspond to the genotypes of the parentals: e.g., AaAaCbCb and CaCaCbCb (here bases A and C are used as exemplars of SNP polymorphism although any other combination of two different bases could have been used). The superscript letters “a” and “b” indicate the component genome of origin, and we use “x” where it cannot be assigned (see Figure 1, A and B, and Figure 2). Here, the base C is the ancestral base and A is the variant base. In this case, the expected genotypes considering disomic recombination are AaAaCbCb, AaCaCbCb, and CaCaCbCb. Genotypes that are tetraplex and triplex for the base A are not expected and can arise only through tetrasomic recombination (AaAaAaCb, AaAaAaAa; see Figure 1, A and B). Note that genotypes that are duplex, monoplex, or nulliplex for the base A may also be the result of tetrasomic recombination (e.g., AaAaCaCa, AaCaCaCa, CaCaCaCa, CbCbCbCb), but they are indistinguishable from genotypes derived from disomic recombination. Therefore, this assay can distinguish only two of the products of tetrasomic recombination: those that are tetraplex and triplex for the variant base. Since SNPs for the GoldenGate assay were identified using two A. duranensis (A genome) accessions, almost all the intragenomic SNP assays can distinguish genotypes that are tetraplex for A. duranensis (A genome), but not genotypes that are tetraplex for A. ipaënsis (B genome) or A. hypogaea SNPs. (Clearly, this type of assay cannot distinguish CbCbCbCb from CaCaCbCb.) Signal intensity plots for SNP assays were inspected, and SNP genotypes indicative of tetrasomic recombination were recorded manually.

An initially unexpected source of data were also collected from SNP assays that are not polymorphic for classic allotetraploid mapping “intergenomic SNPs” (Figure 2). These are assays where both parental genotypes are the same and are of the type GaGaCbCb; that is, the contrasting bases reside on homeologous genomes (here G and C are used as exemplars of DNA bases although any other combination of two different bases could have been used). With strictly disomic recombination, genotypes of all progeny RILs should be the same as the parents, and recombinants would not be found. However, for many assays, some genotypes of the type GaGaGaGa, GaGaGaCb, CbCbCbGa, and CbCbCbCb were evident in a proportion of the progeny (Figure 2). These genotypes arise through tetrasomic recombination, and data points corresponding to them were recorded for all intergenomic SNP assays. Data collection was performed using GenomeStudio’s software, and subsequent data transformations were done in Excel. Note that, for quantifying the products of tetrasomic recombination, the intergenomic SNP is more sensitive than the intragenomic SNP because, with the former, all products of tetrasomic recombination can be distinguished, but with the latter only two products can.

All 1536 SNP genotyping assays were inspected, and 222 were informative for tetrasomic recombination. Of these, 97 were intragenomic polymorphic SNP markers (almost all informative only for tetraplex and triplex alleles of A. duranensis), and 125 were intergenomic (A vs. B genome) SNPs (File S1, worksheets “IntERgenomic SNPs” and “IntrAgenomic SNPs”). Ninety-one RIL lines were analyzed, but 2 of these were eliminated because they showed contaminant or very aberrant genotypes. This makes 19,758 data points in total, which, after removing 529 with low quality, gives 19,229 effective data points. A total of 604 data points (3.1%) were informative of tetrasomic recombination. Of these, 403 were informative of A-genome tetrasomic recombinants, 193 of B-genome recombinants, and 8 tetrasomic events could not be assigned.

The 89 RIL lines had an average of 3% of marker calls indicative of tetrasomic recombination. Of these, only 9 lines had 0 tetraplex/nulliplex calls, 32 had 0–2% tetraplex/nulliplex calls, and 48 had >2% of tetraplex/nulliplex calls. The line most affected by tetrasomic recombination (T141) had 14.5% tetraplex/nulliplex calls (see Figure 4 for a frequency distribution).

Figure 4

Frequency distribution of the number of RILs derived from A. hypogaea × [A. ipaensis × A. duranensis]4x vs. the percentage of markers with tetra/nullisomic genotyping calls.

When the lines and tetrasomic/nullisomic genotyping calls were ordered using the Nagy ) linkage map, it was obvious that the tetraplex/nullplex genotype calls formed genetically meaningful blocks (Figure 5; File S1, worksheets “IntERgenomic SNPs,” “IntrAgenomic SNPs,” and “Both SNP types”). The spatial distribution of the tetraplex/nullplex regions was highly structured. Most lines harbor only one or two well-supported tetra/nullisomic segments, and these regions cover an entire or a substantial part of a linkage group. Several points of tetrasomic recombination were shared between different lines. This is particularly notable on LG03 where 20 lines have large tetra/nullisomic regions (Figure 5 and Figure 6). Overall, although the total number of A-genome tetraplex calls is larger than that of B-genome tetraplex calls, this mostly reflects the greater density of A-genome informative markers, and the genetic distances covered by A- and B-genome tetraplex regions are similar.

Figure 6

Distribution of percentage of markers with tetra/nullisomic genotyping calls for the linkage groups of RILs derived from A. hypogaea × [A. ipaensis × A. duranensis]4x and the induced allotetraploid parent, as inferred from SNP genotyping (Figure 1 and Figure 2). Numbers above columns are the number of plants with tetra/nulliplex segments supported by at least three contiguous informative markers on the respective linkage group.

The degree of tetrasomic behavior of the linkage groups varies considerably. LG03 displays the most autotetraploid behavior (6.4% of calls are tetra/nullisomic, and 20 RIL lines have large tetra/nullisomic regions). In contrast, we found no robustly supported autotetraploid behavior in LG01 (0.2% of calls tetra/nullisomic, all from isolated points and no lines with well-supported large tetra/nullisomic regions) (Figure 6).

Overall, although the percentage of tetra/nulliplex calls is low, the influence on the genotypes of the RILs is very considerable. Of 89 lines, at least 35 have at least one large genome segment derived from tetrasomic recombination. The RIL line most affected by tetrasomic recombination (T141) has a substantial lower region of the LG03 tetraplex for the B genome and a substantial lower region of LG09 tetraplex for the A genome (far right column in Figure 5). For this line, these regions cover ∼82 cM of a total of 1077 cM. One line (T38) is nulliplex for the B genome at the top of LG04 and nulliplex for the A genome at the bottom (File S1, worksheet “Both SNP types”).

Identifying tetrasomic/null alleles using PCR-based markers:

While inspecting a “colormap” generated from the PCR-based markers for the construction of the tetraploid linkage maps published elsewhere (File S1, worksheet “PCR and SMA”), we noted blocks of “missing” SSR and transposable element PCR-based marker data. Initially, these data were considered failed assays. However, under closer inspection, and because of the block structure, we suspected that these data points may not be missing, but in fact indicate null genotypes derived from tetrasomic recombination. For the RILs with the most prominent blocks of “missing” data, the original genotyping data in the form of gel images and fluorescent traces generated by GeneMapper (Applied Biosystems) was inspected to check if the marker assays had, in fact, failed or if the alleles were actually null. This can be done because many PCR-based markers amplify two pairs of loci from the allotetraploid genomes: one pair from the A genome and one pair from the B genome (Figure 7). Usually only the A or the B alleles are polymorphic, and the nonpolymorphic loci are ignored. However, in this context, the nonpolymorphic loci provided positive controls for the functioning of the genotyping assay. This close observation of PCR markers was also repeated over the whole of the homeologous linkage groups of the RILs in question (for selected traces, see File S2 and File S3).

Figure 7

An example of a trace from capillary electrophoresis (ABI 3730 XL) of a microsatellite marker (AHGS1677), generated by the GeneMapper software and revealing an A-genome null allele of a RIL, T138. The marker amplifies alleles in both A and B subgenomes. Therefore, assuming allotetraploid genetics, all lines should show peaks in both A and B genomes, but T138 has only a B-genome allele. At this locus the line is tetraplex for B alleles (in this case A. ipaënsis alleles) and nulliplex for A-genome alleles.

Observations of a colormap of the PCR-based marker map (File S1, worksheet “PCR and SMA”) indicated that the RILs T138, T101, T125, and T141 displayed prominent blocks of “missing” PCR marker data in LGA02, LGB06, LGB09, and LGB09, respectively. These were further investigated by manual inspection of PCR-based marker traces. All markers in the linkage groups of interest were observed, although some markers were more informative than others. Some markers were dominant and some codominant, some amplified A- and B-genome alleles and some did not; furthermore, some marker electrophoretic profiles gave more distinct and well-separated peaks than others. In addition, lines identified by SNP markers to harbor nulli/tetraplex regions on LG03 and LG04 were chosen for inspection of the PCR marker data. A selection of marker gels and traces is available as File S2. Results of marker inspections are described below. All genetic distances in this section are with respect to Shirasawa .

LGA02 and line T138 (for annotated traces, see File S2):

Genotyping for 32 PCR-based markers on this linkage group was inspected for the line T138 and, as controls, the parentals and lines T101, T125, and T141. Eight microsatellite markers (AHGS1677, AHS1351, AHG2733, AHGS2026, AHS0946, AHGS3738, AHS1949, AHS1008) were particularly suitable for detection of null alleles; they are all codominant, amplify a product from both the A and B genomes, and have easy-to-score, well-separated electrophoretic peaks. All lines presented A- and B-genome alleles except T138, which presented only B-genome alleles. This is consistent with SNP data that scores LG02 in T138 to be tetrasomic for the B genome and nullisomic for the A genome. (See Figure 7 for marker profiles of AHGS1677; for more marker traces see File S2.)

LG03 and lines T040, T074, and T095:

For annotated traces, see File S2. Of the 13 particularly suitable markers between 76.5 and 101.9 cM, 7 and 8 showed null B alleles in T040 and T074, respectively. All 13 markers were null for the A genome in T095. This is consistent will the SNP data except that the genetic distances involved are shorter than those indicated by the SNP markers that are positioned according to the Nagy map. This difference could easily be accounted for by different recombination characteristics of the different populations used to calculate the genetic distances.

LG04 and the induced allotetraploid and line T038:

For annotated traces, see File S2. SNP and expression analysis indicated that the seventh/ninth generation induced allotetraploid was nulli/tetraplex over much of the genetic space of LG04 (Figure 3). SNP data indicated that T038 is nulliplex for B alleles at the top of LG04 and nulliplex for A alleles at the bottom. Inspection of 19 markers that are particularly suitable for the detection of null alleles confirmed the expected genotypes. The induced allotetraploid was nulliplex for B alleles at all marker positions except for marker AHGS2048, which mapped at 56.7 cM on A04 and marker AHGS2390, which mapped at 71.1 cM on A04. Line T038 is nulliplex for B-genome alleles from 0 to 31 cM, diplex from 56.4 to 67.45 cM, and nulliplex for A alleles from 71.1 to 91.0 cM. This shows that line T038 had at least two tetrasomic recombination events involving the same linkage groups.

LGB06 and line T101 (for annotated traces, see File S3):

Data for 25 PCR-based markers distributed along the linkage group B06 were inspected for the line T101. Parentals and lines T125, T138, and T141 were compared as controls. Six markers (AHGS1943, AHGS1464, AHS1510, AHGS1756, AHGS2243, AHGS2195) were particularly suitable for detection of null alleles. These markers detected A- and B-genome alleles in all lines except T101, which presented only A-genome alleles for all the linkage group except three markers at 25.1, 19.9, and 15.6 cM (AHGS2058, AHGS2324, AHGS2317; traces not shown in File S3). This is consistent with SNP data that detects A-genome tetrasomics/B-genome nulls on LG06.

LGB09 and lines T125 and T141 (for annotated traces, see File S3):

Data for 14 PCR-based markers distributed along the linkage group were inspected for lines T125 and T141 and controls. Five markers (AHGS1500, AHG2205, AHGS2324, AHGS2696, AHS1950) were particularly suitable for detection of null alleles. B-genome alleles were detected at the top marker only (AhTE0222, 0 cM; trace not included in File S3); no other marker displayed any B alleles. This matches with the SNP data that scores LG09 in T125 and T141 to be disomic at the top of the linkage group and tetrasomic for the A genome and nullisomic for the B genome at the bottom.

When the dominant markers are removed from the map of Shirasawa and it is incorporated into a colormap coded to enhance missing or null data (File S1, worksheet “PCR and SMA”), then null regions are apparent as blocks within genotypes and linkage groups. The concordance with the SNP calls of nulli/tetraplex genomic regions is good. In summary, the inspection of PCR-based marker data indicated tetraplex/nulliplex regions that were consistent with those identified using SNPs.

Single-marker association and quantitative trait loci analyses:

Single-marker analysis for tendency to undergo tetrasomic recombination (number of null PCR markers per genotype) indicated significant marker associations on LGA02 and LGA07. For LGA02, marker Seq16C03 positioned at 39.6 cM was significantly associated (P ≤ 0.1: R2 14.2%). For LGA07, almost all markers between 35.7 and 54.6 cM (P ≤ 0.1) were significantly associated with the most significant markers positioned at 36.1, 52.2, 52.6, and 54.6 cM (P ≤ 0.01: R2 4.7, 15.1, 15.1, and 11.2%, respectively) (File S1, worksheet “PCR and SMA,” columns CP to CV).

Composite interval mapping identified three QTL with peaks: LGA02 at 37.5 cM, LGA07 at 51.3 cM, and LGB03 at 86 cM (Table 2; File S1, worksheet “CIM QTLs”).

Table 2

Quantitative trait loci for tetrasomic recombination identified by CIM concurring with single-marker association

LG	Position (cM)	Nearest markers	LOD	Additive effect	R2 (%)
LGA02	37.5	AHS0519/Seq16C03	4.6	−8.9027	9.8
LGA07	51.3	AHGS1357_t1/AHS2066	4.7	9.3873	10.9

The table presents the LGs; map position in accordance with the Shirasawa PCR-marker linkage map with the markers linked to the QTL; maximum LOD score obtained for the QTL; additive effects; and the phenotypic variance explained by each QTL. Note that the additive effects of the QTL are opposite, indicating that both wild and cultivated alleles favor tetrasomic recombination. A QTL was also identified on LGB03 by CIM, but not by single-marker association and is not shown here.

Therefore, both CIM and single-marker analysis indicated LGA02 at ∼38 cM and LGA07 at ∼51 cM (Table 2). Although mapping of genes that influence genotype must be undertaken with caution because of the possible difficulties of distinguishing cause from effect, neither of these genetic positions is much affected by tetrasomic recombination (having two and zero null data points, respectively). Therefore, the QTL are very likely valid. Note that the additive effects of the QTL are opposite, indicating that both wild and cultivated alleles favor tetrasomic recombination.

Discussion

The morphology of Arachis chromosomes and their mechanisms of recombination have long been of interest to peanut researchers. Pioneering work showed that A. hypogaea is an allotetraploid species (2n = 4x = 40) with an AABB-type genome (Husted 1936; Smartt ). Meiotic chromosomes were shown to consist of 20 chromosome bivalents in 88–98% of cells, exceptions being rare univalents, trivalents, and quadrivalents, suggesting limited homeologous pairing between A and B genomes (Singh and Moss 1982; Wynne and Halward 1989). Molecular mapping in populations derived from intraspecific crosses of A. hypogaea was for some time impossible because of very low DNA polymorphism. However, as the wild diploids related to the component genomes of A. hypogaea became better defined, it was possible to make inferences using diploids in hybridization schemes. Hybrids between A- and B-genome species are typically very infertile, and bivalent formation is significantly reduced compared to hybrids between species of the same genome type (Krapovickas and Gregory 1994; Krapovickas and Gregory 2007; Robledo and Seijo 2010). The use of wild species also allowed the introduction of more allelic diversity into the cultivated species, facilitating molecular mapping. The first molecular study of alien introgression into cultivated peanut (Garcia ) used a map previously constructed from a diploid cross to order markers (Halward ). A small but significant amount of recombination between the A-genome alien species (A. cardenasii Krapov. & W.C. Gregory) and the B genome of cultivated peanut was detected. However, since the study used the complex hexaploid route for introgression, its relevance to normal allotetraploid Arachis genetics was uncertain. The first map construction in allotetraploid Arachis used a BC1 population derived from a cross of A. hypogaea with an artificially induced allotetraploid [A. batizocoi × (A. cardenasii × A. diogoi)]4x (termed an amphidiploid in the original paper; Burow ). In spite of the complexity of the cross, and the distinctness of A. batizocoi Krapov. & W.C.Greg. from the B genome of A. hypogaea (Robledo and Seijo 2010; Moretzsohn ; Leal-Bertioli ), the material of this study more closely mimicked the natural genetics of A. hypogaea. Although there were some anomalies, as expected, inheritance was reported to be disomic. Since this Burow study to our knowledge, all the molecular mapping studies in allotetraploid Arachis have assumed disomic inheritance for map construction and QTL identification (Varshney ; Hong ; Fonceka ; Gautami ; Qin ; Shirasawa , 2013; Sujay ; Wang ; Zhou ).

To investigate the genetic behavior and genome structure of the peanut, we have over some years built up data sets centered around the two most probable diploid ancestral species of cultivated peanut, an induced allotetraploid derived from them, cultivated peanut, and RILs derived from a cross between them (Shirasawa ; Bertioli ; unpublished data). As our data sets became more populated, infrequent data points that could easily have been dismissed as noise began to form regularities that seemed to imply the involvement of some biological phenomena for which we had not accounted. A systematic review prompted us to interpret the data in a fresh light.

Indications of tetrasomic genetic behavior came together from several lines of evidence. Gene expression and genotyping data showed that a line of induced allotetraploid studied at nine and seven generations removed from the original induced polyploidy had become nulliplex for B-genome alleles over a very large proportion of the genetic space on LG04. Genotyping data showed us that most RILs harbor the expected B-genome alleles on LGB04. Therefore, this genetic recombination must have happened after the second generation when the allotetraploid was used as male parent to generate the F1 from which the RILs were derived.

This in itself may have been dismissed as a rare unexpected genetic event. However, both SNP and PCR-based marker data indicated that many of the RILs had also undergone tetrasomic recombination (we recognize that some of the null marker calls may be due to deletions, but overall the patterns that are observed are clearly best explained by tetrasomic recombination). Although the number of data points indicating that tetrasomic recombination was only ∼3% of the whole data set, the impact of this recombination on the genomic structure of the RILs was surprisingly large. Most lines show some evidence for tetraplex/nulliplex genomic regions, and many of these regions cover substantial proportions or even whole linkage group arms (Figure 5). Their frequency was different for different linkage groups, being most frequent in LGs 02, 03, and 06, of moderate frequency in LGs 04, 09, and 10, infrequent in LGs 05, 07, and 08, and possibly absent from LG01 (Figure 6).

Because the unexpected data points were infrequent, for some time we simply did not notice them, dismissed them as noise, or assumed that they were failed marker assays. Because of their relative infrequency, map construction and QTL identification were possible not taking them into account (Shirasawa ; Bertioli ; Table 2; unpublished results). Furthermore, we believe that the linkage maps and QTL identified are broadly valid. However, without taking tetrasomic recombination into account, the inferred genotypes of many lines would be substantially incorrect.

We recognize that the induced allotetraploid is of recent origin and so may have an enhanced instability (Soltis and Soltis 1995). We also recognize that different genetic backgrounds are likely to influence the tendency to tetrasomic recombination. However, because of the very high similarity of A and B genomes in genic regions (Bertioli ; Moretzsohn ), the fact that the induced allotetraploid in this study was derived from the most likely ancestral species to cultivated peanut, the indication that both wild and cultivated alleles positively influence tetrasomic recombination (Table 2), and the previous studies that indicate a small amount of tetravalent pairing in the meiosis of cultivated peanut (Singh and Moss 1982; Wynne and Halward 1989), we think it probable that tetrasomic recombination has gone unnoticed in previous studies. The Burow study used a backcross population, where null alleles would not be apparent. Furthermore, the linkages created by low-frequency tetrasomic recombination would be too weak to significantly influence the linkage maps. Interestingly, one linkage group consisting of markers originating from different component genomes was reported and the authors do mention the possibility that this may represent a genome region with tetrasomic genetic behavior. The mapping studies based on intraspecific crosses of A. hypogaea have generated mostly lower-density data where the patterns in data points derived from tetrasomic recombination would be difficult to detect. The higher density maps produced by Shirasawa are based on F2 populations, so again null data would be much more difficult to detect.

Occam’s razor, a central principle of scientific inquiry, proposes that, among competing hypotheses, the one with the fewest assumptions should be preferred. Frequently, the most parsimonious hypotheses can be made when aberrant data are discarded or ignored. Sydney Brenner (with tongue firmly in cheek) dubbed this methodology “Occam’s broom,” because inconvenient facts are “swept under the carpet.” Plant genetics has a long and distinguished history of using Occam’s broom [see the “process of sophistication” probably used in the collection of Mendel’s data (Fisher 1936)], and it is routinely used in mapping algorithms even today. The working assumption of disomic inheritance is parsimonious and has served peanut genetics well. However, to build better, more predictive genetic models for peanut, and perhaps for other plants where disomic inheritance has been assumed, “in due course, the edge of the carpet must be lifted and the untidy reality confronted” (Robertson 2009, p. 79). Artificially induced Arachis allotetraploids and the spontaneous allotetraploid A. hypogaea are probably “segmental allotetraploids” (sensu Stebbins 1947, 1950) with predominantly disomic, but partly tetrasomic, genetic behavior. As shown here, even infrequent tetrasomic recombination could have had a significant impact on the genome of A. hypogaea and have an unexpected importance in genetics and breeding programs today.