Pervasive hybridizations in the history of wheat relatives

Abstract

Cultivated wheats are derived from an intricate history of three genomes, A, B, and D, present in both diploid and polyploid species. It was recently proposed that the D genome originated from an ancient hybridization between the A and B lineages. However, this result has been questioned, and a robust phylogeny of wheat relatives is still lacking. Using transcriptome data from all diploid species and a new methodological approach, our comprehensive phylogenomic analysis revealed that more than half of the species descend from an ancient hybridization event but with a more complex scenario involving a different parent than previously thought-Aegilops mutica, an overlooked wild species-instead of the B genome. We also detected other extensive gene flow events that could explain long-standing controversies in the classification of wheat relatives.

Fig. 1. Reconstructed phylogeny of the Aegilops/Triticum genus.

(A) Phylogenetic tree of the Aegilops/Triticum genus. This same topology was obtained by both the ML analysis of 8739 gene alignments concatenation (supermatrix) and the supertree combination of 11,033 individual gene trees. All bootstrap values of the supermatrix analysis are 100 except those designated by an asterisk (* = 98). Support values for the supertree analysis are given for each interspecies node [percentage of triplets supporting a given node (13)]. Time scale was obtained by making the ML tree ultrametric and assuming a divergence of 15 Ma with Hordeum (7). (B) “Cloudogram” of 248 trees (in gray) inferred from non-overlapping 10-Mb genomic window concatenations. The global phylogeny is superposed in black.

Fig. 2. Rationale of the quartet method.

The dataset is composed of the counts of the 10 informative site patterns associated with four taxa (so nine degrees of freedom): 0 and 1 are the ancestral and derived states, respectively (polarization with an outgroup). A scenario corresponds to a network with four taxa and up to two hybridizations. It can be decomposed into components (i) with probabilities given by the hybridization proportions (γ_i). The model also includes the times of hybridization (T_i) and the coalescent rates on each branch (α_i) (eight parameters in total). For each component, (ii) the probabilities of embedded coalescent trees and (iii) the probabilities of site patterns given a coalescent tree are computed. They are function of T_i and α_i. Together, they give the expected frequencies of each site pattern for a given scenario. The likelihood of a scenario is given by the multinomial distribution of observing the count vector {v₁,…, v₁₀} given the expected frequencies {p₁,…, p₁₀}. Likelihood comparison was used to choose the best scenario.

Fig. 3. Distribution of the hybridization index for the origin of the D clade.

(A) Violin plots of the hybridization index for the nine species of the D lineage as a function of the A (T. urartu or T. boeoticum) and B (Ae. speltoides or Ae. mutica) parents. The dotted lines correspond to a perfect 50/50 hybridization. All indices are significantly different from 0 (P < 10⁻⁶ after Bonferroni correction). (B) Distribution of the mean hybridization index [and 95% confidence interval (CI)] calculated on 10-Mb windows, along chromosome 3. Red dashed line, chromosome mean; blue line, loess regression with 95% CI in dark gray. The Sitopsis section and Ae. speltoides were excluded because of additional introgression (event 3 on Fig. 4).

Fig. 4. The best scenario for the origin of the D clade determined by the quartet method.

(A) Schematic representation of the two-hybridization tested scenarios (A, species from the A clade; D, species from the D clade; M, Ae. mutica; S, Ae. speltoides). (B) Akaike Information Criterion (AIC) of the saturated model and the four tested scenarios. Models were run with the 10 different combinations of species from the A and D clades. The best AIC are in bold. In two cases, two models have close AIC (the second one is in italics). Scenario 4 is the best model in nine combinations and the second one (with close AIC) in one combination. Point estimates of γ₁ and γ₂ are given for scenario 4: D is the result of two successive hybridizations A + S → M then A + M → D. For the three first combinations, there is a second best model with a very close AIC with a much lower γ₁, in agreement with other values. Scenarios with no or only one reticulation were also tested, and all have much higher AIC (text S4).

Fig. 5. The proposed scenario for the history of diploid Aegilops/Triticum species.

The proposed events obtained from the analysis of hybridization indices and from the quartet method have been added to the global phylogeny. Well-supported clades have been collapsed. The length of triangles corresponds to the divergence age as in Fig. 1. For event 4, the question mark indicates the uncertainties of this complex event. Solid arrows correspond to the most likely detected events, and dashed arrows correspond to possible additional ones: We could not exclude hybridization from Sitopsis as we could not formally test this hypothesis (text S3).