Polarized gene conversion at the bz locus of maize

Significance

Meiotic recombination between homologous paternal and maternal chromosomes allows the shuffling of genes that otherwise would be coinherited and originates at nonrandom sites termed hotspots. In maize, an organism with >90% repetitive DNA, recombination hotspots correspond to genes. Gene conversion is a nonreciprocal form of recombination that leads to an aberrant allelic segregation ratio. In most higher organisms, gene conversions can be identified as intragenic recombinants having a parental arrangement of flanking markers. Here, we show that in maize heterozygotes where most recombinants can be attributed to gene conversion, mutant sites at either end of the gene convert to WT more frequently than those in the middle, arguing for a polarized distribution of recombination initiation sites within the gene.

Abstract

Nucleotide diversity is greater in maize than in most organisms studied to date, so allelic pairs in a hybrid tend to be highly polymorphic. Most recombination events between such pairs of maize polymorphic alleles are crossovers. However, intragenic recombination events not associated with flanking marker exchange, corresponding to noncrossover gene conversions, predominate between alleles derived from the same progenitor. In these dimorphic heterozygotes, the two alleles differ only at the two mutant sites between which recombination is being measured. To investigate whether gene conversion at the bz locus is polarized, two large diallel crossing matrices involving mutant sites spread across the bz gene were performed and more than 2,500 intragenic recombinants were scored. In both diallels, around 90% of recombinants could be accounted for by gene conversion. Furthermore, conversion exhibited a striking polarity, with sites located within 150 bp of the start and stop codons converting more frequently than sites located in the middle of the gene. The implications of these findings are discussed with reference to recent data from genome-wide studies in other plants.

Gene conversion in organisms where all four products of a meiotic tetrad can be recovered refers to a departure from the normal 2:2 segregation of alleles and arises from the repair of meiotic double-strand breaks (DSBs) by a homologous recombination mechanism (1). Gene conversion represents the nonreciprocal, but faithful, transfer of information between two homologous DNA sequences, usually located in homologous chromosomes. The stretch of DNA that is transferred during a gene conversion event, called the conversion tract, can vary in yeast from a few hundred bases to more than 12 kb and is composed of sequences found in only one of the parental chromosomes, i.e., it is continuous, not patchy. Conversion polarity within a gene, which is the tendency of markers located near one end of the gene to convert more frequently than those located at the opposite end, has been reported in several Ascomycete fungi: Ascobolus (2), Neurospora (3), yeast (4), and Aspergillus (5). The high conversion end is usually the 5′ end (6), but can also be the 3′ end (7). Conversion frequency gradients are generally accepted to reflect a preferential initiation site for recombination that is located at the high conversion end of the gene (8, 9).

In most organisms that undergo meiosis, only one of the four meiotic products is ordinarily recovered, so it is not possible to identify gene conversion on the basis of aberrant segregation ratios. A notable exception is the Arabidospsis quartet1 (qrt1) mutant that allows pollen tetrad analysis and has been used to demonstrate gene conversion events unambiguously by their classic 3:1 segregation (10) to estimate genome-wide conversion frequencies (11, 12) and to measure the tract lengths of such conversion events (12, 13). Usually, however, gene conversions have been identified by the flanking marker arrangement of intragenic recombinants (IGRs). This convention derives from the observation that, in yeast asci displaying gene conversion of a central marker flanked by two closely linked outside markers, the convertant spore could carry a parental or a recombinant arrangement of flanking markers with about equal frequency, on average (14, 15). Therefore, geneticists studying recombination in organisms where tetrad analysis is not possible have tended to refer to IGRs bearing a parental or noncrossover (NCO) arrangement of flanking markers as convertants, a convention that we also follow in this paper. In these cases, a convenient way of detecting conversion polarity is to compare the relative frequencies with which the two parentally marked IGRs are recovered from heteroallelic combinations (5, 16).

In contrast to observations in fungi, it was noted repeatedly in maize recombination experiments that the vast majority of IGRs were crossovers (COs), i.e., associated with an exchange of flanking markers (17). Most of those studies dealt with polymorphic heterozygotes in which the recombining heteroalleles were derived from progenitor alleles that differed by single nucleotide and indel polymorphisms in as much as 1.6% of their sequences (18) and often included a transposon insertion allele. This experimental setup helped to place recombination junctions, but affected the outcome of the experiment. A different picture emerged from recombination studies with dimorphic heterozygotes at the bz locus, in which the recombining heteroalleles were derived from the same progenitor and differed only at the two sites between which recombination was being measured (19). In dimorphic heterozygotes, one CO class still predominated (that expected from the relative location of the mutations in the gene), but the NCO classes occurred in much higher numbers, so that the CO/NCO ratio was often less than 1. Similar results have been obtained at the r locus (20). Thus, the CO/NCO ratio in maize can be allele dependent. The variability in the CO/NCO ratio for different recombination hotspots observed in a highly polymorphic yeast hybrid (21) may reflect the extent of allelic polymorphisms among different loci.

The sharp difference in the outcome of experiments involving polymorphic and dimorphic heterozygotes was explained in terms of the dual recombination pathway proposed by Allers and Lichten (22) and supported by other work (23–25), whereby repair of the initiating DSB produces COs via a double Holliday junction (DHJ) intermediate and NCOs via a synthesis-dependent strand annealing (SDSA) pathway. In this model, the decision to repair a DSB as a CO (via a DHJ) or a NCO (via SDSA) would happen at or soon after the initial step of strand invasion. The bz and r data suggest that that decision is affected by the extent of mismatches in the heteroduplex DNA formed by the invading strand. In the absence of extensive heterologies, mismatch repair proteins would not bind to the heteroduplex and the newly synthesized strand would be displaced readily leading to the recovery of NCO products.

In the studies with bz dimorphic heterozygotes, the two NCO classes occurred in roughly similar numbers, i.e., the NCO class that carried the flanking markers of the proximal (5′) allele was recovered approximately as frequently as the NCO class that carried the flanking markers of the distal (3′) allele. Thus, there was no indication of preferential conversion of the proximal or distal allele (26, 27). However, the sample of mutant sites used in those studies did not include any at either end of the gene. Because most conversion gradients in yeast show strong 5′ or 3′ polarity, we could have missed a conversion gradient at bz that was steep at one or both ends but hardly detectable thereafter. To examine at greater depth the issue of conversion polarity within a higher plant gene, we now isolated a series of new bz mutations from the Bz-McC progenitor allele and extended our analysis to include sites at both the 5′ and 3′ ends of the bz gene. Mutations covering the length of the gene were tested in all possible pairwise combinations in two large diallels, with remarkably similar results in the two experiments. We find that bz mutants derived from the Bz-McC allele, which is flanked by single-copy DNA sequences on either side (28), show a U-shaped conversion gradient, with higher conversion frequencies at both the 5′ and 3′ ends than at the center. The implications of these findings are discussed with reference to recent data from genome-wide studies in several organisms.

Results

Isolation of bz Mutants from Bz-McC.

New stable bz mutations spanning the length of the gene were isolated from the well-characterized Bz-McC allele by a combination of chemical and transposon-excision mutagenesis (Fig. 1). Three mutations were produced by seed ethylmethane sulfonate (EMS) mutagenesis: they carry single G-C to A-T transitions at positions 453 (E1), 958 (E3), and 1,217 (E2). Transposon footprint mutations were generated by moving Ac in and out of Bz-McC in a series of sequential transposition experiments. They have either a 7- or 8-bp excision footprint (xis) at the previous site of Ac insertion: s32.1 at −1-(+7). s30.1 at 142–149, s23.5 at 678–685, s2.2 at 755–762, s24.1 at 1,409–1,416, and s19.1 at 1,429–1,436. s39.8 is a 1.8-kb proximal deletion that removes the promoters of bz and the adjacent stk1 gene. The stability of the mutations as homozygotes was tested before combining them in heteroallelic pairs (Table S1).

Fig. 1.

Location of bz mutations. The Bz-McC start codon is given position 1. WT and EMS-induced mutant nucleotides are shown above and below the map, respectively. Most s mutants are Ac excision footprints; s39.8 is a 1.8-kb deletion of the bz promoter. The intron is cross-hatched. The proximal marker wx is on the left, the distal marker sh, on the right.

Analysis of Recombination in a Diallel.

To analyze conversion patterns across the length of the Bz-McC allele, we designed a diallel cross in which eight mutations spread throughout the gene were paired in all possible heterozygous combinations. First, we synthesized 26 of the 28 heterozygotes possible in such a diallel (n{n − 1}/2). The sh-wx region exhibits very high chiasma interference (29), so double crossovers in the region are rare. Then, the heterozygotes were grown in a large isolation plot and hand-pollinated with a sh bz-R wx stock (bz-R is an internal deletion in the bz gene). Bz IGRs occurring at meiosis in these heterozygotes were identified as exceptional purple seeds among the bronze testcross progeny, scored visually for flanking markers, and PCR tested for the unique bz-R allele contributed by the pollen parent.

The results of the diallel are reproduced in Table 1. From a population of almost 1.4 million gametes screened, 1,098 Bz IGRs were obtained. Bz IGRs carrying the parental (NCO) markers of the proximal allele are shown in blue (n = 355), those with the parental markers of the distal allele in red (n = 299), those with the CO arrangement of flanking markers expected based on the location of the mutations in green (n = 276), and those with the opposite recombinant arrangement of flanking markers in black (n = 116). The total number of expected COs (n = 276) was much lower than the total number of NCOs we observed (n = 706), confirming that intragenic recombination events in heterozygotes largely devoid of heterologies are often not associated with flanking marker exchange (19). In fact, assuming that conversion does not interfere with crossing over, as in fungi (30), and 27% exchange between the sh and wx outside markers (29), 967 (n = 706/0.73) or 88% of all IGRs can be accounted for by gene conversion, leaving only a residual 12% of the total (n = 131) to be accounted for by crossing over. We classified Bz IGRs as convertants if they carried parental (NCO) arrangements of flanking markers. This classification is a good approximation because the same proportion of proximal and distal convertants gets scored as COs as a consequence of coincidental exchanges in the adjacent intervals. We can examine whether the assumption of no interference between conversion and crossing over is valid in our system by considering the opposite CO class. This class represents mainly distal convertants accompanied by an exchange in the 25-cM bz-wx region, so it should account for roughly one-fourth of all distal convertants. The number of NCO distal convertants in Table 1 is 344; therefore, 115 distal convertants would be expected to also have an exchange in the bz-wx region and 116 do, validating our assumption. For each heterozygote, we then determined the allelic conversion frequencies and the conversion ratios of the pair of alleles.

Table 1.

Bz recombinants from bz-McC diallel 1

Black, nonexpected CO; blue, proximal convertant; green, expected CO; red, distal convertant.

Fig. 2 summarizes the data in two ways. Fig. 2A plots the frequency of conversion of each heteroallele at its specific location along the length of the bz gene (x axis) against each of the alleles that it was paired with in the diallel. Frequencies are expressed as percentages of each class in the population doubled to account for the undetectable reciprocal event. The results with the bz-s39.8 deletion are not included here and will be discussed separately. Thus, there are seven curves in Fig. 2A: one for each mutation (solid lines). As expected, frequencies tend to increase with distance between the mutations, but the conversion frequencies of mutations located centrally (E1, s2.2, E3, and E2) are generally lower than those of mutations located at either end (s32.1, s30.1, and s19.1). The average conversion frequency (dotted line) shows a U-shaped curve, with a steep increase at mutation s19.1, located close to the 3′ end. The mutation s2.2, located in the middle of the gene, has the lowest average conversion frequency and converts at a lower frequency than its heteroallele in most combinations.

Fig. 2.

Plots of CFs from diallel 1. (A) CFs of each of the seven mutant sites in the Bz-McC allele (x axis) vs. every other mutation in the diallel, color-coded as indicated on the right. (B) CFs of the allelic pair in each heterozygote (identified on the right). (C) CFs of each of the seven mutant sites in the Bz-McC allele (x axis) vs. either the bz-s39.8 deletion or the bz-s32.1 indel mutation.

The data show a clear polarity of conversion. This polarity is best illustrated in Fig. 2B, which simply compares the conversion frequencies of the two mutations in each heterozygote. In most heteroallelic pairs (13/19), the 5′ allele converts more frequently than the 3′ allele (declining slopes from left to right). In four pairs (E1/s2.2, s2.2/s19.1, E3/s19.1, and E2/s19.1), the 3′ allele converts more frequently (ascending slopes) and in two pairs (s32.1/s19.1 and s30.1/s19.1), there is no difference. That is, five of the six exceptions involve s19.1. The overall ratio of 5′ to 3′ allele conversions, 355:299, is slightly, although significantly, skewed toward the 5′ end (χ² = 4.8; P < 0.05). However, the difference is greatly amplified if the six heterozygotes involving the 3′ terminal allele bz-s19.1 are considered separately. Then, the ratio becomes very significantly skewed toward the 5′ end (255:145; χ² = 30.3; P < 0.001), and the residual ratio among the six bz-s19.1 heterozygotes becomes significantly skewed toward the 3′ end (100:154; χ² = 11.5; P < 0.001), even though both bz alleles convert with about equal frequencies in the three heterozygotes between bz-s19.1 and 5′ alleles within the first 450 bp of the bz gene (s32.1/s19.1, s30.1/s19.1, and E1/s19.1). The main conclusion from this analysis is that the bz gene, or at least the Bz-McC allele, shows a general 5′ to 3′ conversion gradient that reverses sharply close to the 3′ end, suggesting preferential sites of conversion initiation at either end.

To verify the existence of a second zone of conversion initiation at the 3′ end, we made use of the 1.8-kb bz-s39.8 promoter deletion, which should remove most of the conversion initiation sites at the 5′ end without affecting those at the 3′ end. This deletion begins within the bz 5′ UTR and extends proximally past the bz promoter into the second exon of the adjacent stk1 gene. Compared with the s32.1 8-bp indel mutation located just 32 bp upstream, the 1.8-kb bz-s39.8 deletion caused a major reduction in conversion frequency of heteroalleles on the 5′, but not on the 3′, side of the gene (Table 1 and Fig. 2C), suggesting that conversion of alleles at the 3′ end rarely, if ever, initiates from a DSB at the 5′ end. This finding supports the concept that there are separate preferential sites of conversion initiation at the 5′ and 3′ ends of the bz gene and suggests that conversion tracts rarely extend the length of the bz gene.

To determine whether the observed U-shaped conversion gradient within the Bz-McC allele was reproducible, we synthesized a second diallel. This diallel included six, rather than eight, bz mutations spread across the gene, so that larger gamete populations from fewer heterozygous combinations could be concurrently tested, and replaced the central s2.2 allele with s23.5 (Fig. 1) to achieve a more uniform distribution of the six markers along the length of the bz gene. Again, the 15 possible heterozygotes were grown in a large isolation plot in the same season and hand-pollinated with a sh bz-R wx stock. Exceptional purple seeds were selected, scored visually for flanking markers, and genotyped by PCR for the unique bz-R allele of the pollen parent.

The results of the second diallel are presented in Table S2. Among the 1,443 Bz IGRs scored), the total number of NCOs (n = 999) was more than three times the number of expected COs (n = 301) and, assuming no interference between conversion and crossing over, 1,368 or 95% of them can be accounted for by gene conversion. Thus, the overall outcome was very similar to that of the first diallel (Table 1). The data are summarized in the two graphs of Fig. 3, which closely parallel those shown in Fig. 2, and display a similar conversion polarity. Fig. 3A shows the conversion frequency of each mutation on the x axis against each mutation that it was paired with in the diallel. The average conversion frequency of the bz alleles decreases sharply from the 5′ end to the middle of the gene and reverses sharply close to the 3′ end. Fig. 3B compares the conversion frequencies of the two mutations in each of the 15 heterozygotes. In 11 of them, the 5′ allele converts more frequently than the 3′ allele (declining slopes), and three of the four exceptions involve the s19.1 mutation. Overall, the 5′ allele converts significantly more frequently than the 3′ allele (564:435, χ² = 16.7; P = 0.00004). Again, the difference is greatly amplified if heterozygotes involving the 3′ terminal allele bz-s19.1 are not considered (394:202; χ² = 61.8; P = 3 × 10⁻¹⁵) and reverses among the 5 bz-s19.1 heterozygotes (170:233; χ² = 9.9; P = 0.002), even though 5′ alleles within the first 450 bp of the bz gene convert with about equal frequencies as bz-s19.1.

Fig. 3.

Plots of CFs from diallel 2. (A) CFs of each of the six mutant sites in the Bz-McC allele (x axis) vs. every other mutation in the diallel, color-coded as indicated on the right. (B) CFs of the allelic pair in each heterozygote (identified on the right).

The sharp polarity reversal at the s19.1 site prompted us to examine the conversion behavior of a different 3′ end mutant site, s24.1, an 8-bp indel mutation located just 20 bp upstream of s19.1 (Fig. 1). Rather than performing another complete diallel, we simply compared the recombinational behavior of s19.1 and s24.1 side by side in heterozygotes with the 5′ mutation s30.1 to determine whether their conversion frequencies were similar in the two cases. The results in Table S3 show that the two 3′ mutations convert at similarly high frequencies, which are just slightly lower than those of the 5′ mutation s30.1, as had been seen in the two diallels (Table 1 and Table S2).

Discussion

Polarized NCO Gene Conversion at bz.

We provide here extensive evidence supporting our earlier finding (19) that gene conversion not associated with flanking marker exchange is the predominant form of intragenic recombination in dimorphic heterozygotes. The two alleles in these heterozygotes derive from the same progenitor and thus differ only at the two mutant sites between which recombination is being measured. In two large diallel crosses involving eight and six alleles, respectively, a screen of more than 3 million gametes, and the scoring of more than 2,500 IGRs, >90% of recombinants could be accounted for by gene conversion. Furthermore, we find that conversion exhibits a striking polarity, with sites located within 150 bp of the start and stop codons converting more frequently than sites located in the middle of the gene. In an earlier study (31), we did not see preferential conversion of either the proximal or distal allele. That study included a set of bz mutant sites stretching from 458 bp downstream of the start codon to 201 bp upstream of the stop codon in the Bz-W22 progenitor allele, so we could have missed a steep conversion gradient at either end of the gene. Using a combination of chemical and transposon excision mutagenesis of the Bz-McC progenitor allele, we were able to isolate more terminally located mutant sites that uncovered a steep 5′ to 3′ conversion frequency gradient at the 5′ end, which reverses within the last 150 bp of the bz coding sequence.

Conversion polarity in yeast has been hypothesized to reflect a fixed recombination initiation site from which conversion tracts of variable length extend into the gene (8, 9, 14). This view derives from the analysis of conversion frequencies of single mutant alleles in heterozygotes with WT. In our study, we compare the conversion frequencies of two mutant alleles in a heterozygote and we assume that, as in yeast (1, 15), most conversion tracts are continuous, i.e., composed of sequences found in only one of the parental chromosomes. In this light, we interpret the conversion frequency gradient to reflect the variable probabilities with which the DSB that initiates meiotic recombination, and hence heteroduplex formation, occurs across the gene. If so, the concave conversion gradient that we see at bz would suggest that meiotic DSB formation in bz is highest at the 5′ and 3′ ends but declines toward the center of the gene. The decline appears steeper at the 3′ end, particularly in the second diallel, where the bz mutant sites were more evenly distributed (Fig. 3).

Conversion gradients in yeast that are high at the 5′ end of the gene are abolished, and conversion frequencies for all sites are reduced in strains homozygous for promoter deletions (8, 32), consistent with the presence of a DSB recombination initiation site in the promoter region. In our system, we tested the effect of a bz promoter deletion in heterozygous condition. For a meiotic DSB to function efficiently as a recombination initiator, i.e., for a heteroduplex to form, the allelic sequence must be present in the homolog. Thus, promoter deletions, even if heterozygous, should interfere with the initiation of recombination, as they do in yeast, where they lower both DSB formation and the frequency of conversion of a downstream marker (33). We found that the bz-s39.8 deletion, which includes the bz promoter, caused a major reduction in the conversion frequency of heteroalleles on the 5′, but not the 3′, half of the gene (Fig. 2C). This finding supports the presence of separate preferential sites of conversion initiation at the 5′ and 3′ ends of the bz gene and suggests that the average conversion tract is shorter than half the length of the bz locus or 750 bp.

In yeast, a concave conversion gradient seen at DED81 has been similarly attributed to sites of DSB formation at both ends of the gene (6). However, the elevated conversion at the 3′ end of DED81 corresponds to the site of DSB formation at the 5′ end of the adjacent ARG4 locus, whereas the 3′ ends of the bz gene and of the adjacent stc1 gene face each other (28), so the high conversion at the bz 3′ end cannot be the result of 5′ conversion initiation sites in the adjacent gene. The reversal of polarity in bz is also not sequence or site specific, as the s24.1 mutation, located very close to s19.1 at the 3′ end (Fig. 1), converts at about the same frequency as s19.1 itself and s30.1 at the opposite end (Table S3). A trivial explanation for the observed conversion gradient would be that it arises from the differential repair of indel and point mutations. However, the conversion frequencies of centrally located indel (s2.2 and s23.5) and point (E2 and E3) mutations were uniformly low, and the nadir of the conversion frequency curve occurred at the site of an indel mutation (s2.2) in diallel 1 and of a point mutation (E2) in diallel 2 (Figs. 2A and 3A). These observations argue that the conversion polarity observed at bz is due to the position rather than the nature of the mutations used in the study. A recent genome-wide study also detected positional recombination polarity in a flowering plant. In Mimulus genes with five or more exons, the average historical recombination was highest in the first exon and decayed with distance from the start codon (34).

The chromosomal neighborhood of the bz locus varies greatly among different maize lines (35, 36). In the McC haplotype used in our study, the bz-stc1 intergenic region consists of about 1 kb of mostly single-copy DNA interrupted by three miniature inverted-repeat transposable elements (MITEs). In the B73 haplotype, conversely, that segment is made up mainly of a 25-kb methylated, most likely heterochromatic, retrotransposon cluster that strongly reduces recombination in the adjacent bz and stc1 genes (37). In yeast, chromatin structure has been shown to affect the genome-wide distribution of the DSBs that initiate meiotic recombination (38), and the suggestion has been made that DSB formation is opportunistic, occurring in nucleosome-depleted regions where DNA is sufficiently exposed to allow access by the SPO11 endonuclease (39). It will be interesting to analyze the conversion pattern of the bz locus in the B73 haplotype, which is so structurally dissimilar to McC at the bz 3′ end. If a conversion gradient exists, as is likely, will it also evince a reversal of polarity at the 3′ end? It is conceivable that the conversion pattern of maize loci will be affected by the remarkable variation in local genome structure and thus will also turn out to be a polymorphic trait in this highly variable species.

The bz locus is both a recombination hotspot (29) and a highly GC-rich gene (18), a correlation predicted by the biased gene conversion model proposed to explain the origin of GC-rich isochores in mammals (40) and grasses (41). According to this model, GC-biased gene conversion results from the GC bias of the mismatch repair machinery, so highly recombinogenic genes, which would form heteroduplexes at a higher frequency, would tend to be more GC rich. Recently, it was shown that grasses, in general, display strongly decreasing GC gradients along transcripts (42), and it was suggested that GC-biased gene conversion would explain them if gene conversion was higher at the 5′ end of grass genes. Except for the sharp reversal at the 3′ end, which may not be a universal feature of bz haplotypes, the 5′ to 3′ bz gene conversion gradient described here would support that suggestion.

Sequence Diversity and Recombinational Outcome.

In maize, crossing over is the almost exclusive mode of recombination between alleles that differ in multiple polymorphisms, whereas NCO gene conversion is the predominant mode of recombination between alleles that are essentially identical. At what level of polymorphism does recombination begin to switch from one mode to the other? We can consider our own data on intragenic recombination at bz in an attempt to obtain limits. From the work with dimorphic heterozygotes presented here, we know that sequences varying at only two sites within the gene recombine predominantly by NCO gene conversion and because the average distance between markers in this study is 0.71 kb, that level of polymorphism translates to 2.8 heterologies/kb. In contrast, polymorphic sequences differing in 15.3 heterologies/kb, spread throughout the gene, recombine mostly by CO (19, 31). Therefore, recombination switches from one predominant mode to another somewhere in the range of 2.8–15 heterologies/kb of recombining sequences. A comparison of the more than 20 bz alleles in GenBank reveals that polymorphisms between some allelic pairs are distributed very unevenly, so that stretches of identity measuring up to 500 bp occur adjacent to highly polymorphic stretches. It is possible that in maize hybrids carrying such pairs of alleles, one part of the gene will show predominantly NCO gene conversion, whereas the other part will show predominantly CO recombination and that, by extension, both types of recombination will be found across the genome as a function of the degree of identity of the recombining sequences. In maize inbreds, genes will recombine predominantly by NCO gene conversions. Although considerably less numerous, the crossovers required for the proper segregation of homologs at meiosis also do occur, as is clear from the data in Table 1 and Table S2, where about 10% of the total number of Bz IGRs cannot be accounted for by NCO gene conversions and where the CO class expected based on the position of the markers in the gene is more than twice as frequent as its reciprocal.

Extensive NCO gene conversion in a higher plant was reported in a recent genome-wide sequencing study in Arabidopsis. Yang et al. (43) resequenced 40 F₂ progeny plants from a Columbia/Landsberg F₁ hybrid and detected a large number of gene conversions, which were defined as short double switches (<10 kb) between the two parental sequences (conversion tracts). The authors estimated that small gene conversion tracts represented >90% of all recombination events in the Columbia/Landsberg hybrid. In contrast, two other Arabidopsis genome-wide studies came to a different conclusion. Lu et al. (13) and Wijnker et al. (12) sequenced the four products of, respectively, 2 and 13 pollen tetrads from a Columbia/Landsberg qrt1 F₁ and detected an average of only one to two NCO gene conversion events per meiosis, most of which included just a single marker. A subsequent reanalysis (44) of the Yang et al. high-throughput sequencing data (43) led to the conclusion that many of the NCO gene conversions had arisen from misinterpretations of the short read sequence data and that the true number was much lower and consistent with other findings.

In assessing whether findings in one system can be extrapolated to another, the extent of polymorphism between the recombining sequences should be considered. As discussed above, the degree of NCO gene conversion appears to be a function of the percent identity of the recombining sequences, so we can expect it to vary both at the gene level within a species and at the genome level between species. Another caveat in extrapolating genome-wide results from one species to another is that in some species, like Arabidopsis, most recombination events occur in intergenic regions (13, 43, 45), whereas in others, like maize, most recombination is intragenic (17, 46). The latter observation is not surprising given the high degree of intergenic gross structural polymorphism in maize (36, 47). It will be interesting to examine the genome-wide breakdown of recombination in rice, a grass species with intragenic recombination hotspots (48), like maize, but with much less repetitive DNA (49).

Methods

Genetic Procedures.

All of the alleles used in this study were in the common genetic background of the inbred W22. The mutations sh (shrunken endosperm) and wx (waxy endosperm) were used as markers flanking bz. They map, respectively, 3 cM distal and 25 cM proximal to bz in 9S (29) and were introduced by recombination into the different bz mutants incorporated into the diallel. The bz alleles and the analysis of the diallel are described in detail in SI Methods and Table S4.

Molecular Procedures.

DNA extraction, PCR, and sequencing were performed as previously described (46).