Genome analyses reveal the hybrid origin of the staple crop white Guinea yam (Dioscorea rotundata)

Genetic Diversity of Guinea Yam

We obtained DNA samples from 336 accessions of D. rotundata maintained at the International Institute of Tropical Agriculture (IITA) in Nigeria, representing the genetic diversity of Guinea yam landraces and improved lines from West Africa. We subjected these samples to whole-genome resequencing on the Illumina sequencing platform. We aligned the resulting short reads to the newly assembled reference genome (SI Appendix, sections S1 and S2) and extracted single nucleotide polymorphism (SNP) information for use in genetic diversity studies (SI Appendix, Table S1 and section S3). Based on admixture analysis with the sNMF program (12), we defined five major clusters (Fig. 1A). When K = 2, cluster 1 was clearly separated from the other accessions. Principal component analysis (PCA) also separated cluster 1 from the rest of the clusters (Fig. 1B). Accessions in cluster 1 had significantly higher heterozygosity and ∼10-fold more unique alleles than those in the four remaining clusters (SI Appendix, Figs. S2 and S3 and Table S2). Because flow cytometry analysis confirmed that all 10 accessions analyzed in cluster 1 were triploids (Dataset S1), we hypothesized that cluster 1 represents triploid D. rotundata, a hybrid of D. rotundata and D. togoensis (4). After removing the cluster 1 accessions, the nucleotide diversity of D. rotundata was estimated as 14.83 × 10⁻⁴ (SI Appendix, Table S3), which is ∼1.5-fold larger than that reported previously (10), presumably because we used a larger number of samples with diverse genetic backgrounds in our study. Linkage disequilibrium of diploid D. rotundata showed a decay of r² = 0.13 in a 200-kb genomic region (SI Appendix, Fig. S4), which is slower than that of cassava, another clonally propagated crop (13).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Genetic diversity and phylogenomics of Guinea yam and its wild relatives. (A) Ancestry proportions of each Guinea yam accession with 6,124,093 SNPs. TDr96_F1 is the sample used as the reference genome. (B) PCA result of the 336 Guinea yam accessions. (C) NJ tree of four African yam lineages reconstructed using D. alata as an outgroup based on 463,293 SNPs. The numbers indicate bootstrap values after 100 replications. The sequences of D. rotundata in the previous study (10) were included in the tree. (D) Evolutionary relationship of three African wild yam lineages (D. abyssinica, Western D. praehensilis, and Cameroonian D. praehensilis) as inferred by ∂a∂i (15) using 17,532 SNPs. N, M, and T represent the relative population size from N_anc, migration rate, and divergence time, respectively.

Phylogenomic Analysis of African Yam

Using the SNP information, we constructed a rooted neighbor-joining (NJ) tree (14) based on 308 Guinea yam accessions sequenced in the present study (excluding cluster 1 triploid accessions), as well as 80 D. rotundata, 29 D. abyssinica, 21 Western D. praehensilis, and 18 Cameroonian D. praehensilis accessions that were sequenced in a previous study (10) using two accessions of Asian species D. alata as an outgroup (Fig. 1C). Throughout the analyses described below, we used 388 D. rotundata accessions by combining our samples and those used previously (10). According to this NJ tree, the D. rotundata accessions sequenced in this study are genetically close to the D. rotundata accessions reported previously (10) (Fig. 1C). However, the NJ tree showed that D. rotundata is more closely related to D. abyssinica than to Western D. praehensilis (Fig. 1C), which is inconsistent with a previous finding (10) that D. rotundata is most closely related to Western D. praehensilis.

To elucidate the evolutionary relationships of the three wild Dioscorea species that are closely related to D. rotundata—D. abyssinica (designated as A), Western D. praehensilis (P), and Cameroonian D. praehensilis (C)—we performed diffusion approximations for demographic inference (∂a∂i) analysis (15), which allows for estimation of demographic parameters based on an unfolded site frequency spectrum. First, we tested three phylogenetic models—{{A, P}, C}, {{P, C}, A}, and {{C, A}, P}—using 17,532 SNPs that were polarized using D. alata as an outgroup without considering migration among the species. Of the three models, {{A, P}, C} had the highest likelihood (SI Appendix, Table S4).

This result is not consistent with the previous finding that {{P, C}, A} had the highest likelihood (10), as determined using a different method with fastsimcoal2 software (16). To exactly repeat the previous analysis, we tested these three models with fastsimcoal2 (16) using the previous reference genome (11), which indicated that {{A, P}, C} had the highest likelihood (SI Appendix, Table S5). Taken together, our results are inconsistent with the previous report (10) but are consistent with the PCA result from the same report, which separated Cameroonian D. praehensilis from the other African yams in PC1 (figure 2A of ref. 10).

Based on the assumption that {{A, P}, C} describes the true evolutionary relationship among the three wild Dioscorea species, we reestimated the evolutionary parameters with ∂a∂i, allowing symmetric migration (gene flow) among the species (Fig. 1D). Since the results indicated that Cameroonian D. praehensilis is distantly related to D. rotundata and was not likely involved in genetic exchange with D. rotundata (Fig. 1C), we focused on Western D. praehensilis, which we refer to as D. praehensilis hereinafter for brevity.

Hybrid Origin of Guinea Yam

We propose three hypotheses for the origin of Guinea yam (D. rotundata) based on the NJ tree (Fig. 1C) and ∂a∂i (15) (Fig. 1D). The first hypothesis is that D. rotundata was derived from D. abyssinica (hypothesis 1 in Fig. 2A); the second is that D. rotundata was derived from D. praehensilis (hypothesis 2 in Fig. 2A). However, in hypotheses 1 and 2, the divergence time of D. rotundata from the wild species might not be sufficient to separate the three lineages, and there may be incomplete lineage sorting among the species. The third hypothesis is that D. rotundata originated as an admixture between D. abyssinica and D. praehensilis (hypothesis 3 in Fig. 2A).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Evidence for the hybrid origin of Guinea yam. (A) Hypotheses for the domestication of Guinea yam (D. rotundata). Hypothesis 1 assumes that D. rotundata diverged from D. abyssinica. Hypothesis 2 assumes that D. rotundata diverged from D. praehensilis. Hypothesis 3 assumes that D. rotundata was derived from a hybrid between D. abyssinica and D. praehensilis. D. alata served as an outgroup. (B) Frequencies of individuals homozygous for D. abyssinica allele (A, in yellow), homozygous for D. praehensilis allele (P, in blue), and heterozygous for A and P (in white) among the 388 D. rotundata sequences as studied for 144 SNPs. (C) Evolutionary parameters related to the hybrid origin of Guinea yam as inferred by ∂a∂i (15) using 15,461 SNPs. N, M, and T represent the relative population size from N_AP, migration rate, and divergence time, respectively. f_A and f_p indicate the genomic contributions from D. abyssinica and D. praehensilis when the hybridization occurred, respectively. (D) F_ST between the wild and cultivated yams. This was conducted with a 100-kb window and a 20-kb step. Chromosome 11 of D. rotundata containing the sex-determining locus shows a shorter distance to that of D. abyssinica and a longer distance to that of D. praehensilis.

Before estimating the evolutionary parameters for the three hypotheses, we studied the allele frequencies of the 388 D. rotundata sequences, focusing on 144 SNPs that are positioned over the entire genome and are oppositely fixed in the two candidate progenitors (Fig. 2B and SI Appendix, Fig. S5). If hypothesis 1 or 2 is correct, then the allele frequencies in these 144 SNPs should be highly skewed to either of the progenitors. The patterns of allele contributions from the two candidate species to D. rotundata were nearly identical, however. This result suggests that hypothesis 3, the admixture origin of Guinea yam, is most likely correct.

We tested the three hypotheses by ∂a∂i (15) with symmetric migration (gene flow) rates using 15,461 SNPs polarized by D. alata (Fig. 2A), which showed that hypothesis 3 had the highest likelihood and the lowest Akaike information criterion value (Fig. 2C and SI Appendix, Table S4). This result supports the admixture hypothesis, that D. rotundata was derived from crosses between D. abyssinica and D. praehensilis. The parameters estimated by ∂a∂i indicate that the hybridization between D. abyssinica and D. praehensilis was relatively recent in relation to the divergence between the two wild species. This analysis also indicated that the genomic contributions from D. abyssinica and D. praehensilis during the hybridization period were ∼68% and 32%, respectively. Introgression generally results in highly asymmetric genomic contributions from the parental species, whereas hybridization shows symmetric genomic contributions (17). The intermediate genomic contributions revealed by this analysis support the hybridization hypothesis rather than the introgression hypothesis. Our findings are in line with the hybrid origin of the Guinea yam proposed by Coursey in 1976 based on morphology (8) and supports his speculation that spontaneous hybridization between wild yams could have occurred at the artifactual “dump heaps” created by people living in the savannah between the forest and the Sahara (9).

To evaluate the genetic distances of D. rotundata from the two parental species for each chromosome, we calculated fixation index (F_ST) values (18) (Fig. 2D and SI Appendix, Table S6). The genetic distances from the two parents varied across the different chromosomes, and the overall genetic distance of D. rotundata from D. abyssinica was shorter than that from D. praehensilis (SI Appendix, Table S6). Intriguingly, chromosome 11, to which we previously mapped the candidate locus for sex determination (11), had the shortest genetic distance from D. abyssinica and the longest genetic distance from D. praehensilis among all chromosomes, indicating that chromosome 11 of D. rotundata is highly skewed to D. abyssinica (Fig. 2D and SI Appendix, Table S6). Similarly, interspecies divergence is different between the autosomes and sex chromosome of the dioecious plant species Silene (19).

Evolutionary History of Guinea Yam

In angiosperms, plastid genomes are predominantly inherited maternally (20), making them useful for studying maternal lineages. To infer the maternal history of Guinea yam, we constructed a haplotype network of the whole plastid genome with all samples used in the NJ tree (Fig. 1C), as well as the triploid accessions in cluster 1 (Fig. 3A and SI Appendix, section S6). According to this haplotype network, Cameroonian D. praehensilis has the longest genetic distance from D. rotundata. This result is in line with the phylogenomic trees of African yam (Fig. 1 C and D). Strikingly, the plastid genomes of diploid and triploid D. rotundata are uniform and very similar to those of Nigerian or Beninese D. abyssinica, although the latter has another plastid genome lineage distant from that of D. rotundata. The plastid genomes of D. praehensilis from Nigeria, Benin, and Ghana appear to be derived from Nigerian or Beninese D. abyssinica. These results indicate that D. abyssinica is an older lineage than D. praehensilis, and that the places of origin of D. rotundata and D. praehensilis are probably around Nigeria or Benin. Based on the whole-genome diversity of D. rotundata, a recent study (10) hypothesized that the origin of D. rotundata was around north Benin, as supported by the current results. The plastid genomes of some wild species are identical to those of cultivated Guinea yams. Hybridization between cultivated yams and wild yams may account for this observation (7).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Evolutionary scenario of African yam origins. (A) Haplotype network of the whole plastid genomes of 416 D. rotundata (including the triploid accessions), 68 wild relatives, and two D. alata accessions as the outgroup. The number of vertical dashes represents the number of mutations. Western (Nigerian, Beninese, and Ghanaian) D. praehensilis and D. rotundata seem to have diverged from Nigerian and Beninese D. abyssinica. (B) Possible scenario of domestication of Guinea yam. The blue line represents paternal origin, and the red line represents maternal origin. (C) Changes in population sizes of D. rotundata and its wild relatives as inferred by ∂a∂i (15). The parameters except population size were identical to those used in Fig. 2C. After the domestication of D. rotundata, the population size of D. rotundata has increased with migration from the wild progenitors.

The results of nuclear genome admixture (Fig. 2) and plastid haplotype network (Fig. 3A) analyses indicate that the maternal origin of diploid D. rotundata is D. abyssinica and its paternal origin is D. praehensilis (Fig. 3B). Hybridization between D. abyssinica and D. praehensilis is rare (10), but such rare hybrids appear to have been domesticated by humans. The triploid D. rotundata shares its plastid haplotype with diploid D. rotundata, indicating that diploid D. rotundata served as the maternal parent and D. togoensis was the paternal parent. D. cayenensis is reported to have D. rotundata as the maternal parent and D. burkilliana as the paternal parent (3, 4). All cultivated Guinea yams are hybrids containing D. abyssinica plastid genomes.

To explore the changes in population size, we reinferred the demographic history of African yam by ∂a∂i (15), allowing migration (Fig. 3C and SI Appendix, section S7). We used the same dataset as in Fig. 2C. By fixing the parameters predicted in Fig. 2C except population size, we reestimated each population size at the start and end points after the emergence of these species, assuming an exponential increase/decrease in population size. According to this analysis, since the emergence of the wild progenitors of Guinea yam, the population size of D. abyssinica has been decreasing, while that of D. praehensilis has been increasing (Fig. 3C). This finding suggests that the D. praehensilis population was derived from D. abyssinica, which is consistent with the results of haplotype network analysis (Fig. 3A).

Extensive Introgression at the SWEETIE Locus

To explore multiple introgression to D. rotundata from the two wild species, we analyzed the f₄ statistic (21) using four groups: D. rotundata clusters 2 and 5, D. rotundata cluster 4, D. abyssinica, and D. praehensilis (SI Appendix, section S8). The f₄ statistic reveals the representation of two alternative discordant genealogies (Fig. 4A). The f₄ value is close to zero if the first two groups of D. rotundata show a concordant genealogy in relation to D. abyssinica and D. praehensilis. In contrast, the f₄ value diverges from zero if the two groups of D. rotundata exhibit discordant genealogy and a large genetic distance to each other. We obtained the f₄ statistic f₄ (P₂₅, P₄, P_P, P_A) for each SNP and performed sliding window analysis (Fig. 4B). The f₄ value was close to zero across the genome, indicating that overall, we cannot decide between topology 1 and topology 2. However, the genomic regions around the SWEETIE gene showed the lowest f₄ (P₂₅, P₄, P_P, P_A) [Z(f₄) = −5.66], with overrepresentation of topology 2 in the SWEETIE gene (DRNTG_01731) (SI Appendix, Table S7).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Signature of extensive introgression around the SWEETIE gene. (A) Topology of f₄ (P₂₅, P₄, P_P, P_A) in clusters 2, 4, and 5 and wild yams. Positive f₄ values represent the long internal branch of the upper tree (topology 1), and negative f₄ values represent the long internal branch of the bottom tree (topology 2). (B) f₄ values across the genome. This was conducted with a 250-kb window and a 25-kb step. Red dots indicate outliers of the sliding window, which have |Z(f₄)| > 5. The locus around the SWEETIE gene shows extraordinarily negative f₄ values. (C) Neighbor-Net around the SWEETIE gene (4∼4.15 Mb on chromosome 17). This was constructed by SplitsTree (22) using a total of 458 SNPs.

To explore the genealogical relationships around the SWEETIE gene, we constructed a Neighbor-Net (22) around this locus (4.00 to 4.15 Mb on chromosome 17) (Fig. 4C). The Neighbor-Net showed that the locus of cluster 4 was close to that of D. praehensilis, while the loci of clusters 2 and 5 and some other accessions were close to those of D. abyssinica. These results indicate that the SWEETIE gene was introgressed from the wild species more than once. The SWEETIE gene encodes a membrane protein involved in the general control of sugar flux (23). The Arabidopsis thaliana sweetie mutant shows pronounced changes in the accumulation of sugar, starch, and ethylene along with significant changes in growth and development (24). We still do not know the effect of this introgression on the phenotype of Guinea yam, but this locus appears to be a target of selection.

Homoploid Hybrid Formation as the Trigger of Domestication

The importance of hybridization and polyploidization for crop domestication is well documented (25, 26), including in bread wheat (27) and banana (28). Compared with allopolyploidy, only a limited number of homoploid hybridizations have been reported in plants (29), and homoploid hybridizations have rarely contributed to the origin of crops (30). Homoploid hybridization can increase genetic variation via recombination between distantly related species, and it often allows the hybrid to adapt to unexploited niches (31). In the case of Guinea yam, the savannah-adapted wild species D. abyssinica and the rainforest-adapted wild species D. praehensilis are not suitable for agriculture; however, their hybrid, D. rotundata, could have been adopted for cultivation by humans. Gene combinations from different wild yams might have contributed to the domestication of Guinea yam. The present study provides an example of the origin of a crop through homoploid hybridization.

Use of Wild Species to Improve Guinea Yam

A project for the improvement of Guinea yam by crossbreeding has been initiated (AfricaYam: https://africayam.org). However, the current breeding projects depend solely on D. rotundata genetic resources. Systematic efforts are needed to introgress beneficial alleles from wild species into crops; these alleles will increase disease resistance and abiotic stress tolerance to improve crop resiliency and productivity (32). Our study revealed that the two wild progenitor species (D. abyssinica and D. praehensilis) of Guinea yam contain much greater genetic diversity than D. rotundata (Fig. 2C), suggesting that these wild species could be useful sources for alleles of agricultural importance. However, the D. abyssinica and D. praehensilis accessions in IITA GenBank account for only 1.6% of the total Dioscorea accessions maintained as of 2018 (33). Therefore, it will be important to collect and preserve wild Dioscorea species as genetic resources for improving Guinea yam. Our findings suggest that new alleles of loci, such as the SWEETIE gene, were introgressed from wild yams into cultivated Guinea yams multiple times, which likely conferred the plants with phenotypes preferred by humans. Many more alleles from wild species remain to be exploited for systematic breeding. Our findings highlight the need to consider how to effectively leverage the gene pools of wild species from different habitats for the rapid breeding of Guinea yam using genomic information.