Sympatric speciation of wild emmer wheat driven by ecology and chromosomal rearrangements

Significance

In plants, which are stressed constantly by diverse abiotic and biotic stresses, ecological divergence is substantial in driving speciation, yet its demonstration under sympatry is still limited in plants and needs elaboration. By analyzing wild emmer wheat at the Evolution Canyon microsite, a hot spot of sympatric speciation, we highlighted how disruptive ecological selection drives primary sympatric speciation in situ. Natural selection, overruling the homogenizing effects of gene flow, split the metapopulation of wild emmer wheat into three reproductively isolated populations. Each population evolved unique evolutionary trajectory with divergent adaptive strategies, including fungal disease resistance, irradiance tolerance, and divergent flowering time. Our work provides a model for plant adaptive ecological sympatric speciation under biotic and abiotic stressors.

Abstract

In plants, the mechanism for ecological sympatric speciation (SS) is little known. Here, after ruling out the possibility of secondary contact, we show that wild emmer wheat, at the microclimatically divergent microsite of “Evolution Canyon” (EC), Mt. Carmel, Israel, underwent triple SS. Initially, it split following a bottleneck of an ancestral population, and further diversified to three isolated populations driven by disruptive ecological selection. Remarkably, two postzygotically isolated populations (SFS1 and SFS2) sympatrically branched within an area less than 30 m at the tropical hot and dry savannoid south-facing slope (SFS). A series of homozygous chromosomal rearrangements in the SFS1 population caused hybrid sterility with the SFS2 population. We demonstrate that these two populations developed divergent adaptive mechanisms against severe abiotic stresses on the tropical SFS. The SFS2 population evolved very early flowering, while the SFS1 population alternatively evolved a direct tolerance to irradiance by improved ROS scavenging activity that potentially accounts for its evolutionary fate with unstable chromosome status. Moreover, a third prezygotically isolated sympatric population adapted on the abutting temperate, humid, cool, and forested north-facing slope (NFS), separated by 250 m from the SFS wild emmer wheat populations. The NFS population evolved multiple resistant loci to fungal diseases, including powdery mildew and stripe rust. Our study illustrates how plants sympatrically adapt and speciate under disruptive ecological selection of abiotic and biotic stresses.

Understanding the mechanisms underlying speciation, the splitting of a group of interbreeding populations into two reproductively isolated groups, has been a major focus of evolutionary biology (1). In the widespread process known as “allopatric speciation,” populations that become geographically isolated experience selective pressures and/or genetic drift that ultimately give rise to two reproductively isolated new species (2). In contrast, there are dramatically fewer known instances of speciation occurring without geographical isolation, a concept known as sympatric speciation (SS) proposed by Darwin (2–4). It is generally agreed that an instance of SS must be substantiated with evidence showing that the relevant species occur in the same geographic location, are in reproductive isolation, and display a sister relationship (3). Claims of SS are bolstered by evidence that any earlier periods of allopatric speciation are unlikely (5). Major questions about SS typically center on how reproductive isolation has arisen in sympatry and on the nature of the selection pressures and genetic drift that has driven the speciation (6).

An increasing number of mathematical models have been established to sustain the plausibility of SS, often including the influence of disruptive selection that is particularly validated in nature by the examples in animals (e.g., resource competition, assortative mating, etc.) (5, 7, 8). In plants, which are stressed constantly by a diversity of biotic and abiotic stresses, ecological divergence was presumed essential in driving speciation. However, accepted examples of ecological SS are still limited in plants, and evolutionary mechanisms for such speciation have not been well established. One compelling case is that of sister species of palm (Howea) that seem to have split on a tiny isolated island in the Tasman Sea, possibly driven by divergent soil preference, yet any underlying cellular and genetic mechanisms associated with this split have not been demonstrated (9).

Guided by the widely held speculation that divergent selection via microcosmic ecological differences are likely essential in facilitating SS, a long-term research project since 1990 in Israel has enabled tracking of evolutionary histories at a microscale by designating four natural laboratories: the so-called “Evolution Canyon” microsites (EC-I to EC-IV). At each of these EC sites, there are sharp ecological contrasts within minuscule interslope distances (6). The EC-I microsite comprises twin slopes with extremely sharp ecological contrasts: the south-facing slope (SFS) and north-facing slope (NFS) are only separated by an average of 250 m, but the SFS receives 200 to 800% higher solar radiation than the NFS, resulting in a substantially hotter and drier slope (Fig. 1A). In contrast, the higher relative humidity and reduced solar radiation of the NFS results in cooler and wetter conditions (10). These dramatic differences in microclimates in extremely close proximity have enabled very high spatial resolution studies unfolding the influence of environmental gradients on fitness-related phenotypic traits, and offers one of the best available opportunities to study the interplay between natural selection, adaptation, and further speciation. In past decades, a variety of population-based analyses with five model organisms [Bacillus simplex (11), Drosophila melanogaster (12), spiny mice Acomys cahirinus (13), wild barley Hordeum spontaneum (14), and sawtoothed grain beetles Oryzaephilus surinamensis (15)] have marshaled evidence to strongly suggest that incipient SS is under way at EC-I between populations present on the SFS or NFS.

Fig. 1.

Population history of wild emmer accessions in EC-I. (A) The geographic distribution of the emmer wheat accessions used in this study. The green circles represent the collection location outside EC-I. The red circle represents the location of EC-I. The arrows represent the collection stations on each slope. (B) Phylogenetic tree of 168 wild emmer wheat accessions by neighbor-joining method with 1,000 bootstraps. The arabic notation following the collection location in Israel is shown in the map. (C) Inferred historical population sizes by pairwise sequential Markovian coalescent (PSMC) analysis. Three typical samples from each population were shown. The lower x axis gives time measured by pairwise sequence divergence, and the y axis gives the effective population size measured by the scaled mutation rate. (D) Distribution of Ks values for 12,352 one-to-one orthologous gene sets between each population. The full-length transcripts from one representative sample from each population (SFS1-7, SFS2-14, and NFS-8) was employed for Ks evaluation.

Here, seeking to apply modern genome-scale technologies to examine potential speciation events in tetraploid wheat populations at EC-I, we collected 47 accessions of wild emmer wheat (Triticum dicoccoides Koern), the progenitor of bread wheat, from EC-I. Using genomics, cytogenetics, analysis of hybrid sterility patterns, and both observational and hypothesis-driven confirmatory studies addressing observed divergence for ecologically important traits, we assessed the possibility of SS of wild emmer wheat at EC-I.

Results

Three Distinct Populations of Wild Emmer Wheat Have Diverged within EC-I following a Recent Population Bottleneck.

To apply modern high-resolution genetic analysis methods to the exceptionally diverse populations of species known to exist at the EC-I site, we collected 47 accessions from the tropical hot and dry savannoid south-facing slope (SFS) and abutting temperate, humid, and cool forested north-facing slope (NFS) locations of this site, within a transect of only 300 m. Notably, most of these accessions were collected from altitudes above the midlevel of each slope (stations 1, 2 on SFS; stations 6, 7 on NFS). We also obtained materials from the close vicinity around EC-I (e.g., ∼3.5 km from Beit Oren, ∼13 km from Bat Shlomo, and ∼30 km from Afula) and from surrounding countries to help determine the population structure of the materials we collected from EC-I (Fig. 1A and Datasets S1). We first determined which of the SNPs present on the 55K Affymetrix Wheat GeneChip would be suitable for analysis of these wild emmer wheat accessions using BLAST analysis of the chip probe sequences against the emmer reference genome; a total of 30.6K SNPs were suitable for use (e.g., SNPs having no more than two hits in the A and B subgenomes). These SNPs were used to construct a neighbor-joining phylogenetic tree (Fig. 1B).

Most of the wild emmer accessions collected from Israel were clustered together in the tree, and the accessions collected from Turkey, Syria, and other locations likewise clustered together (Fig. 1 A and B). The terminal taxa of the phylogenetic tree include three distinct subpopulations among the accessions collected at EC-I: two populations from the SFS locations (“SFS1” and “SFS2”) and one from the NFS location (“NFS”). These three populations are all monophyletic, and were resolved as sister relationships that are more similar to each other than to the other emmer accessions from surrounding areas (Fig. 1B). All of the accessions collected from NFS and SFS clustered only their own populations in the phylogeny, without any migration or radiation, thus clearly indicating apparent reproductive isolation and adaptive divergence of these populations primarily based on the local environment (Fig. 1 A and B), which is important to assess sympatric speciation.

To estimate the levels of shared ancestry of each population in EC-I, we subjected the SNP data to an “admixture” model based on a maximum-likelihood method. When the K values were increased from 2 to 10 (optimal K value), the NFS, SFS2, and SFS1 populations were sequentially separated from other populations, apparently suggesting the lineage splits of the SFS1, SFS2, and NFS populations (SI Appendix, Fig. S1). Similar results were also obtained from analysis using the “fastStructure” model (SI Appendix, Fig. S1).

To further define the three populations within EC-I deriving from ancestral introgression or incomplete lineage sorting, we performed multiple iterations of ABBA-BABA D-statistic tests using the GeneChip data (16). When using the SFS1 population as Pop. 3 and the populations from Bat Shlomo or other areas in Israel as outgroups (SI Appendix, Table S1), no significant gene flow or introgression was detected between the SFS1 population and the NFS or SFS2 populations. Similarly, we did not detect any gene flow between the SFS1 population and the populations from surrounding areas (SI Appendix, Table S1). These results are in agreement with the phylogenetic relationship and the admixture analyses (Fig. 1B and SI Appendix, Fig. S1), supporting independent adaptive branching of each population within EC-I rather than admixture or secondary introduction.

To further assess the genetic diversity and evolutionary history of the wild emmer population represented by the accessions collected at EC-I, 16 accessions from the NFS population, 9 accessions from the SFS1 population, and 10 accessions from the SFS2 population were subjected to whole-genome sequencing, with an average sequencing depth of 9.45 to 18.49×. The mapping rate for the different accessions was over 98%, thus covering 89.29 to 92.89% of the 10.1-gigabase (Gb) assembly of the emmer reference genome “Zavitan” (Datasets S2). After validation by BQSR (base quality score recalibration) and VQSR (variant quality score recalibration), a total of 46,301,637 high-quality SNPs were identified and used in subsequent analyses (Materials and Methods provides characteristics of the genomic distribution of these SNPs; SI Appendix, Fig. S2). The average SNP density along the genomes of these accessions was 4.36 SNP/Kb, with 4.02 SNP/Kb for the A subgenome and 5.05 SNP/Kb for the B subgenome (SI Appendix, Figs. S2 and S3).

These SNPs were used to generate a phylogenetic tree and to conduct a population structure admixture and PCA analysis. This revealed similar results to those generated by the corresponding analyses based on the data from the 55K Affymetrix Wheat GeneChip: e.g., the SFS1, SFS2, and NFS populations sequentially diverged and then further diversified (SI Appendix, Figs. S4 and S5). In order to reconstruct the population history of these three lineages, haploid pairwise sequential Markovian coalescent (PMSC) analysis was conducted to infer the branching times and inbreeding effective population size (Ne) of the three populations of accessions collected from the EC-I microsite (17). This analysis revealed that, consistent with the phylogeny, these three populations have branched from the same ancestral population (Fig. 1C).

Furthermore, each of the three populations apparently experienced the same bottleneck recently (∼3,000 to 6,000 y ago if using substitution rate for rice) (18). The pronounced contemporaneity of the ancestral population bottleneck and the population splits suggests that genetic drift induced by a population bottleneck or founder effect was likely a strong influence on the initial divergence of these three populations at EC-I (19). After these splits in the wild emmer populations at EC-I, all three populations gradually recovered Ne, albeit to differing extents, with the SFS1 population eventually having the largest population size and the SFS2 population having the smallest (Fig. 1C and SI Appendix, Table S2).

To validate the divergence times inferred from the PMSC results, we used an alternative method of comparing the synonymous substitution values (Ks) of orthologous genes. Total RNA from three accessions (SFS1-7, SFS2-14, and NFS-8) representative of each population was extracted and analyzed using both the PacBio Sequel platform for single-molecule sequencing and the Illumina HiSeq XTen platform for short reads. After identifying the full-length reads from the PacBio data, the high-depth coverage (>100×) short read data were used for SNP correction (Materials and Methods). After mapping to the emmer reference genome, a total of 12,352 one-to-one orthologous gene pairs with high specificity were employed in the Ks and K_a analyses using model averaging (MA) (20) (SI Appendix, Fig. S6). The SFS1 population and SFS2 population were estimated to have diverged first (Ks = 4.94E-05), around 4,800 y ago based on the nucleotide substitution rate of rice (18); the divergence of the SFS2 population and the NFS population was estimated as slightly later (Ks = 4.16E-05; Fig. 1D). These results support the estimates from the phylogenetic and PMSC analyses indicating that the branching of these populations of wild emmer wheat in EC-I occurred recently.

The SFS2 and NFS Populations Display Prezygotic Reproductive Isolation, whereas the SFS1 Population Displays Postzygotic Barriers.

To evaluate the level of genetic differentiation among populations, the population fixation statistics (Fst) were examined using the same whole-genome SNP data set (Fig. 2A and SI Appendix, Fig. S7). Very high Fst values (>0.46) were distributed across the whole genome, especially between the SFS2 population and NFS populations, reflecting a large differentiation among the three populations in EC-I (Fig. 2A and SI Appendix, Fig. S7 and Table S3). Lower divergence was found around regions of heterochromatin in several chromosomes such as 3A, 6A, and 7B (Fig. 2A), but the average Fsts between each population still reached 0.468 for SFS1/SFS2, 0.479 for SFS1/NFS, and 0.634 for SFS2/NFS (SI Appendix, Fig. S7 and Table S3), values much higher than in previously studied emmer wheat populations (21). Note that lower genetic diversity was found for the SFS2 population (π = 0.36 × 10⁻³) and NFS population (π = 0.52 × 10⁻³) as compared to the SFS1 population (π = 0.98 × 10⁻³), findings consistent with the aforementioned Ne estimates (Fig. 1C and SI Appendix, Fig. S7 and Table S2).

Fig. 2.

Genetic differentiation of wild emmer populations in EC-I. (A) Genetic differentiation revealed by Fst across genome window between each population. (B) Progeny ratio of F₁ hybrids between accessions of SFS1 population and the other two populations. A 100% ratio of progeny was supposed as the seed setting rate of three progeny seeds per spikelet, and more than 10 spikes per sample were employed for seed setting rate calculation.

We did not detect any apparent gene flow between each population using a TreeMix analysis based on a maximum-likelihood tree (SI Appendix, Fig. S8), similar to results from D-statistic analyses using the GeneChip data (SI Appendix, Table S1). The genetic drift parameter for the SFS2 population and the NFS population is higher (>0.2). This is consistent with their relatively small Ne, indicating that genetic drift has contributed to the fast diversification of these two populations (SI Appendix, Fig. S8 and Fig. 1C). Consistent with the phylogeographic and phylogenetic relationships, the results of Fst and TreeMix indicate that prezygotic reproductive isolation has developed between each population, blocking gene flow among populations.

To test whether any postzygotic reproductive isolation has developed, we selected several individuals from each population, conducted crosses with interpopulation materials, and monitored the fertility of progeny. Hybrids between individuals from the SFS2 and NFS populations produced normal fertile progeny (Datasets S3). However, large-scale hybrid sterility was found when crossing SFS1 individuals with SFS2 individuals and with NFS individuals. For self-pollination, all nine individuals from the SFS1 population produced a high ratio of progeny (here defining “high” as 70%, as emmer wheat can generally produce two seeds per spikelet, presuming three seeds per spikelet is 100%; Fig. 2B). However, interpopulation hybrids from the SFS1 individuals (SFS1-1, SFS1-2, and SFS1-3) only produced a ∼30 to 50% ratio of progeny when crossed with SFS2 and NFS individuals (Fig. 2B). Further, the hybrids from SFS1-4 and SFS1-5 produced about a 20% ratio of progeny, and the hybrids from SFS1-6 to SFS1-9 only produced a ∼2 to 15% ratio of progeny in crosses with SFS2 and NFS individuals (Fig. 2B). The much reduced fertility of these hybrids shows that postzygotic isolation has formed between the SFS1 population and the other populations, supporting the inference of SS of the SFS1 population.

Chromosomal Rearrangements in the SFS1 Population Caused the Postzygotic Reproductive Isolation.

Motivated by the unanticipated finding that the SFS1 and SFS2 populations have in situ diversified as postzygotically isolated populations on SFS, a small area located within 30 m, we subsequently attempted to unravel the underlying mechanism using a cytogenetics approach. The karyotypic variation via fluorescence in situ hybridization (FISH) was examined for the emmer wheat accessions in this study using four probes [Oligo-pSc119.2, Oligo-pTa535, Oligo-(GAA)7, and Oligo-3A1; Materials and Methods]. The banding pattern generated by these four probes allows identification of all individual chromosomes of hexaploid wheat (22), and here the emmer wheat showed similar banding patterns with the hexaploid wheat cultivar “Chinese Spring” (SI Appendix, Fig. S9 A and B), e.g., the Oligo-pSc119.2 shows a single band on the short arm of Chr. 5A (SI Appendix, Fig. S9B, green). Thus, the diagram of signal pattern was generated for identifying chromosomal polymorphisms of the wild emmer accessions (SI Appendix, Fig. S9C).

Three atypical banding patterns were observed in the SFS1 population, findings which suggested the possible occurrence of Robertsonian translocations that were not present in the SFS2 or NFS populations (Fig. 3A and SI Appendix, Figs. S10 and S11). Based on these FISH signals, we determined that some SFS1 individuals have experienced chromosomal breakage and refusion of chromosomes 2A and 5A, thereby forming two new linkage groups: 2AS.5AS/2AL.5AL (Fig. 3A). We also observed other translocations in the SFS1 population, including1BS.4BS/1BL.4BL and 2BS.3BL/2BL.3BS (Fig. 3A and SI Appendix, Fig. S11).

Fig. 3.

Chromosome rearrangements in SFS1 population. (A) The representative for three typical Robertsonian translocations by FISH. (B) The pattern genome rearrangements for different individuals in SFS1 population. The paring and division events during meiosis of F₁ are shown in the table as percentages. (C) The representative of paring and division events during meiosis of F₁. The abnormal paring or division is shown by arrows.

All of the translocation events were homozygotes (SI Appendix, Fig. S10), indicating they have become stable during evolution. A pattern of various genotypes with/without translocations exists in the SFS1 population (Fig. 3B). First, none of the Robertsonian translocations were identified for the SFS1-1, SFS1-2, and SFS1-3 individuals, which show relatively ancestral positions on the phylogeny. Other structural changes of the genome could not be ruled out, as some pollen mother cells (8 to 12%) of the hybrid (crosses with NFS or SFS2 individuals) showed disordered bridges during meiosis anaphase I (Fig. 3 B, c₁). Second, three genotypes with one or two translocation events were found in independent phylogenetic lineages. The lineages of SFS1-4 and SFS1-5 only involve the 1BS.4BS/1BL.4BL translocation. In addition to this translocation, the lineages of SFS1-6 and SFS1-7 also harbor the 2BS.3BL/2BL.3BS translocation, and the lineages of SFS1-8 and SFS1-9 harbor 2AS.5AS/2AL.5AL (Fig. 3B). These results suggest that the 1BS.4BS/1BL.4BL translocation was the initial translocation type in the SFS1 population, after which the other two translocations evolved.

In crosses with the normal chromosome accessions of the NFS or SFS2 populations, we indeed found severe meiotic disorders in the hybrids, mainly tetravalent status at metaphase I, reflecting a typical Robertsonian translocation (23). Consistent with the FISH results, single tetravalent status (26 to 31%) in pollen mother cells was observed in the SFS1-4 and SFS1-5 individuals that harbor the 1BS.4BS/1BL.4BL translocation, while double tetravalent status (9 to 16%) was observed for the individuals bearing two translocation events (1BS.4BS/1BL.4BL and 2BS.3BL/2BL.3BS; Fig. 3 B and C). We did not find abnormal pairing or division during meiosis in hybrids from crosses between the SFS2 and NFS individuals, which all showed equal metaphase I division status (Fig. 3, c₂).

Thus, we have clear cytogenetic evidence that three Robertsonian translocation events have evolved sequentially in the SFS1 population. Moreover, we demonstrate that these homologous chromosome rearrangements gained in the course of evolution cause meiotic disorders in hybrids from crosses between SFS1 individuals with SFS2 and NFS individuals. We also showed that the extent of such meiotic disorders is elevated for hybrids from crosses with SFS1 individuals harboring multiple chromosome rearrangement events. Collectively, these results establish a plausible cellular mechanism to explain the postzygotic isolation of hybrid sterility we observed (Fig. 2B and SI Appendix, Table S3).

Disruptive Ecological Selection Has Driven the Discrete Adaptations of the SFS1, SFS2, and NFS Populations.

To unravel the evolutionary mechanism(s) underlying the three diverged populations in EC-I, especially for the sympatric speciation of SFS1 at the site based on chromosome rearrangements, we further examined the adaptation strategies of these populations of emmer wheat. Relating directly to the ecology of the humid NFS, all accessions in the NFS population displayed high resistance to most of the tested powdery mildew (Blumeria graminis f. sp. tritici) isolates and to three leaf rust (Puccinia triticina) isolates, while most accessions from the SFS1 and SFS2 populations were highly susceptible to the tested pathogen isolates (Fig. 4A and SI Appendix, Fig. S12A). There were two patterns of powdery mildew resistance within the NFS population (SI Appendix, Fig. S12A), indicating polymorphism for resistant genes within the NFS population.

Fig. 4.

Distinct adaptation mechanisms of wild emmer populations in EC-I. (A) Representative photograph of phenotypic variance of accessions from each population. (B) The distribution of the XP-CLR score along 14 chromosomes with 20-Kb sliding window. The genome-wide threshold was defined both by the top 5% and 1%, as marked in black dotted lines. (C) GO terms enriched by candidate regions selected in the SFS1 population. The top 20 terms with the most significant enrichment are shown. The GO items involved with antioxidant pathway are noted with asterisk. (D) Phenotype of transgenic wheat overexpressing TaPRX1 under irradiance condition. The seedlings were subjected to high radiation with 30,000 Lux light for 1 wk and then photographed.

We constructed F_2:3 populations by crossing the susceptible parent of SFS2-12 with resistant parents of NFS-10 and NFS-2 independently. For NFS-10, representing resistance pattern 1 (SI Appendix, Fig. S12A), the F₂ populations displayed a segregation ratio of observed number, 3:1, showing qualitative regulation by a single resistance gene. Bulked segregant RNA sequencing (BSR-seq) revealed a single strong peak on the terminus of the Chr. 4A long arm (SI Appendix, Fig. S12B), a newly evolved resistance locus to our knowledge. However, the resistance signals found for the NFS-2 individual (representing pattern 2) had multiple weaker peaks in BSR-seq analysis (SI Appendix, Fig. S12C), indicating multiple quantitative resistance loci.

Ecological studies of EC-I have reported early flowering phenotypes for many plants on the SFS, with some SFS emmer wheat plants showing flowering 1 to 2 mo earlier than those from the NFS (6, 24). We grew all of the collected accessions from EC-I in SFS-mimic conditions (with relatively mild abiotic stress to avoid excessive pressure on the NFS population; Materials and Methods). The SFS2 population exhibited much earlier flowering times than the SFS1 population; the SFS2 plants also flowered earlier than NFS plants (Fig. 4A and SI Appendix, Fig. S13). These findings suggest that the SFS2 population has evolved an early flowering strategy to cope with the latter harsh abiotic stress of its environment on the south-facing slope.

The relatively late flowering status of the SFS1 population is puzzling, and is apparently inconsistent with the SFS2-type adaptation to its local environment. We thus checked the SNP variation of FT1 loci, 34 genes of the CONSTANTS-like family, and 10 PRR genes that have been reported to function in plant flowering regulation. An SNP in the CCT domain of the CONSTANS-like gene CO6 (TRIDC7BG063900) had an amino acid change from Lys to Glu at position 321 in the SFS1 population; this mutation was not detected in any non-SFS1 emmer accessions in our study (SI Appendix, Fig. S14 A and B). CONSTANS can activate the transcription of the FT1 gene by binding to the CORE element in its promoter (25), and we confirmed a flowering-promoting function for CO6 in vivo: three independent transgenic Brachypodium lines overexpressing emmer CO6 flowered significantly earlier than wild type plants (SI Appendix, Fig. S14 F and G).

Biochemically, both yeast one-hybrid and gel-shift assays revealed that CO6 (321Lys) from the SFS2 and NFS populations can strongly bind to this CORE element (SI Appendix, Fig. S14 C–E), but that CO6 (321Glu) from the SFS1 population had substantially reduced binding activity to the CORE element (SI Appendix, Fig. S14D). Moreover, the binding affinity was recovered when position 321 of the SFS1 CO6 protein was mutated back to Lys (SI Appendix, Fig. S14D). Thus, we hypothesized that the 321Glu mutation of the CO6 gene can at least partially account for the differences in flowering times between the SFS2 and SFS1 populations that encounter the similar abiotic stress conditions of the SFS of EC-I.

The SFS1 populations’ flowering phenotypes are apparently inconsistent with efficient adaptation to the local environment of the SFS, but we found that the SFS1 population has evolved an alternative strategy for mitigating the harsh abiotic conditions: higher tolerance to irradiance. We did not find significant differences in drought tolerance among the three EC-I emmer populations when growing plants at 20 °C with a 5,000-Lux light at the seedling stage (data not shown). The SFS1 plants had significantly higher tolerance when the light was elevated to 30,000 Lux, exhibiting reduced oxidative injury on leaves and consistently reduced levels of superoxide radicals (O₂⁻) than the accessions from other populations (Fig. 4A and SI Appendix, Fig. S15), indicating an example of unique adaptive evolution for increased antioxidant activity. This result is in agreement with the fact that there is an average of 200 to 800% higher radiation on the SFS than that on the NFS (10).

To identify potential selective signals underlying this change, we scanned genomic regions by comparing the SFS1 population and the other two populations using a likelihood method (XP-CLR) (26). To lessen the problem of fixation due to genetic drift, we pooled the populations of SFS2 and NFS and conducted selection analyses in comparison with the SFS1 population. With the cutoff value of 5%, we subjected the genes in selective sweeps to Gene Ontology (GO) enrichment analyses and found that the selective genomic regions were extremely enriched for genes with annotations relating to oxidative stress tolerance (Fig. 4 B and C and Datasets S4), such as the members of cytochrome P450 and peroxidases (PRX) (27) (Datasets S5), further supporting the involvement of antioxidant pathways in the evolution of the SFS1 population.

We detected that a total of 30 general ROS scavengers of PRX encoding loci in the SFS1 population exhibited elevated selection signals (Datasets S5), in which nine PRX genes showed fixed nonsynonymous mutations in the SFS1 population (Datasets S6). We then cloned the alleles for these genes from the SFS1 and SFS2 populations and expressed them in fission yeast (Materials and Methods). In vitro assays using the same concentrations of the recombinant proteins indicated that three SFS1 PRX proteins, encoded by TRIDC2AG012890, TRIDC7AG049310, and especially TRIDC5AG058230 (here named PRX-SFS1), showed significantly improved oxidative activity for ABTS substrates, suggesting that the SFS1 PRX-SFS1 allele has evolved improved ROS scavenging capacity compared to the SFS2 allele (SI Appendix, Fig. S16). To test whether this gene is involved in oxidative stress tolerance in vivo, we overexpressed PRX-SFS1 in wheat using Agrobacterium-mediated transformation and analyzed three independent T₃ lines in light stress conditions. Under light treatment at 30,000 Lux, the transgenic lines displayed apparently reduced leaf injury compared to wild type plants. We also detected significantly higher peroxidase activity and reduced superoxide radical (O₂⁻) content in the transgenic plants.

Collectively, these results clearly illustrate that the three populations in EC-I have undergone distinct evolutionary trajectories in their respective adaptations to the local environment (disease resistance for the NFS population, early flowering for the SFS2 population, high irradiance tolerance for the SFS1 population). These adaptive differences support the driving roles of disruptive ecological selection in the branching and diversification of each population (Fig. 5). For the sympatric diversification of the SFS1 and the SFS2 populations, the difference in flowering time should have been the major prezygotic barrier against gene flow. However, the accumulation of mutations in each population, especially the series of chromosome rearrangements that have occurred among SFS1 plants, have resulted in the final postzygotic reproductive isolation of SFS1 from the other EC-I populations. It is noted that the independent clustering in phylogeny, large genomic differentiation, lack of gene flow, and distinct adaptive strategies for the NFS and SFS2 populations all suggest that they are prezygotically isolated in the local environment at EC-I, indicating incipient sympatric speciation.

Fig. 5.

Sympatric ecological speciation model of wild emmer wheat at EC-I. The blue dotted line indicates the extent of genetic drift. The red full lines indicate postzygotic reproductive isolation, and the red dotted line indicates prezygotic reproductive isolation. The yellow five-pointed stars indicate powdery mildew-resistant genes evolved, and the orange one indicates strip rust-resistant gene.

Discussion

Sympatric Speciation of Wild Emmer Wheat at EC-I.

In plants, clear evidence supporting sympatrically split species with reproductive isolation and identifying the underlying mechanisms is still limited. Here, we present evidence that wild emmer wheat, the progenitor of tetraploid and hexaploid wheats, is incipiently sympatrically speciating at EC-I, a hot spot microsite where SS from bacteria to mammals has been demonstrated (28). Beyond simply identifying interslope diversification, our data clearly support also the split of two postzygotically isolated populations (SFS1 and SFS2) on the same tropical hot south-facing slope (SFS), within a transect of less than 30 m, beside a third incipient speciation on the opposite NSF. We successfully ruled out the possibility of secondary contact or introgression using genetic diversity data from wild emmer wheat accessions collected from the EC-I microsite, its surrounding areas across Israel, and surrounding countries, which represent the origin (North Israel) and distribution region of wild emmer wheat in the Near East Fertile Crescent (Fig. 1 A and B). All our phylogeographic, phylogenetic, admixture, D-statistic, and PMSC analyses (Figs. 1 and 2 and SI Appendix, Fig. S1 and Table S1) provide evidence that the SFS1, SFS2, and NFS populations at the EC-I microsite in Mount Carmel have diverged sequentially from a single ancestral population, following a population bottleneck that occurred less than 10,000 y ago (Fig. 1 C and D).

In addition to prezygotic isolation detected between any of the three diverged populations at EC-I, postzygotic barriers by chromosomal rearrangements of Robertsonian translocations were unique to the SFS1 population (Fig. 3 A and B and SI Appendix, Figs. S9–S11). This resulted in a large extent of hybrid sterility when crossing was conducted with the individuals from other populations (Fig. 3 B and C). We thus conclusively demonstrate a case of primary SS and characterize molecular events explaining the reproductive isolation that has enabled the origin of separate emmer wheat species at the same EC-I microsite.

Disruptive Ecological Selection Drives Sympatric Speciation of Wild Emmer Wheat at EC-I.

Theoretical models for SS often include disruptive speciation mechanisms such as divergent habitat preference, resource partitioning, and/or assortative mating (3–6, 12, 13, 29–31). Several animal cases, like African indigobirds, apple maggots, and cichlid fish, indicate that such diversifying selection-causing sympatric speciation models are plausible (5–9). However, in sessile plants, which are stressed constantly by a diversity of biotic and abiotic stresses and which cannot escape to refugia like animals, mechanistic evidence is lacking for ecological SS (9, 24). In the present study, our haploid pairwise sequential Markovian coalescent (PMSC) analysis clearly indicated that the branching of the SFS1, SFS2, and NFS populations occurred soon after the bottleneck of an ancestral population less than 10,000 y ago (Fig. 1C), suggesting that genetic drift induced by a founder effect might have been involved in the initial split of these three populations at EC-I (19). With relatively small population sizes (SI Appendix, Fig. S7), genetic drift was involved in the further diversification of these populations (SI Appendix, Fig. S8), and may partially account for the rapid differentiation of genetic pool frequencies and genomic components of each population at EC-I (SI Appendix, Fig. S1) (32).

However, even with genetic drift, our data indicate that disruptive ecological selection has driven wild emmer wheat to diversify as reproductively isolated populations, adaptive to the contrasting abiotic (SFS) and biotic (NFS) stresses at the EC-I microsite. Remarkably, on the same slope, the SFS2 population has adapted to the strong irradiance, heat, and drought by temporally advancing its reproductive development to occur during a time window with relatively weaker abiotic stress (Fig. 4A and SI Appendix, Fig. S13). The SFS1 population has, in contrast, developed much stronger irradiance tolerance over the past several thousand years (Fig. 4A and SI Appendix, Fig. S15), with relatively late flowering (Fig. 4A and SI Appendix, Fig. S13). On the opposite slope, the NFS population evolved pronounced resistance to plant fungal pathogens that thrive in humid conditions (Fig. 4A and SI Appendix, Fig. S12). Thus, both the data in this study and the previously identified cases of incipient SS at EC sites repeatedly confirm that divergent selection can overrule gene flow and genetic drift to drive a metapopulation of organisms into distinct isolated populations undergoing SS (6, 11–14, 33, 34).

The Evolution of Reproductive Isolation in Wild Emmer Wheat.

One major question about sympatric speciation is understanding how reproductive isolation has arisen in sympatry. In this study, the fact that a series of sequentially formed chromosome rearrangements in the SFS1 population, which resulted in postzygotic barriers, helps rule out the possibility of secondary contact (Fig. 3B). It is highly unlikely that a genetically identical but different Robertsonian translocation could simultaneously have migrated to EC-I from surrounding areas. Indeed, similar patterns of these chromosome rearrangements were not found for the accessions outside EC-I in the present study (data not shown) or in previous studies that screened karyotypes of wild emmer populations from the areas or countries surrounding EC-I (35, 36).

For the chromosome rearrangements, an intriguing possibility is that the particular adaptive strategy of the SFS1 population may have facilitated the evolution of postzygotic reproductive isolation. UV radiation is widely recognized as a major cause of chromosome disruption (27). This is supported by the evidence that the genes in the SFS1 accessions exhibiting selection signals are particularly enriched for antioxidant activity (Fig. 4 B–D), e.g., modified activity of peroxidases (Fig. 4D and SI Appendix, Fig. S15), highlighting that the SFS1 population has been subjected to strong selection pressure due to the high irradiance stress characterizing its site positioned below a cliff (station 1), which accepts the highest irradiance at EC-I (SI Appendix, Fig. S17) (10).

Extensive mathematical modeling (4, 29, 30, 37) and experimental studies, including those from the Evolution Canyon model (28), have provided evidence that evolutionary branching in situ followed by sympatric speciation is not only feasible, but may be a common model of the origin of species, substantiating Darwin’s insight of SS. This is plausible since geologically, edaphically, climatically, and biotically divergent canyons abound across our planet (28). Nevertheless, more evidence for completed SS with reproductive isolation is desirable, since the frequency and potential universality of the SS model is still debatable (3). Our present study presents long-sought and unambiguously clear mechanistic evidence that in situ sympatric speciation has occurred among the wild emmer plants—the progenitor for cultivated wheat—not only interslope but also intraslope in the SFS of Evolution Canyon.

Materials and Methods

Phylogenetic Tree and Population Structure Analyses.

A total of 1,364,793 SNPs in coding regions (exonic and intronic) were used for population genetics analysis. To clarify the phylogenetic relationships from a genome-wide perspective, an individual-based neighbor-joining tree was constructed using the p-distance as implemented in TreeBeST (v1.9.2), with bootstrap values determined from 1,000 replicates (38). The population genetic structure was examined using the program ADMIXTURE (v1.23) (39). A total of 35 accessions were used to estimate the genetic ancestry, specifying a K ranging from 2 to 3. A principal component analysis was also conducted to evaluate the genetic structure of the populations using GCTA (40). First, a genetic relationship matrix was obtained using the parameter “–make-grm”; then, the top three principal components were estimated with “–pca3”.

Detection of Gene Flow Using TreeMix.

Population relatedness and migration events were inferred using TreeMix (41). The 1,364,793 coding-region SNPs were used to build a maximum-likelihood tree, with a window size of 2,000 SNPs. This was repeated 10 times. The tree with the lowest SE for the residuals was selected as the base tree topology. The population pairs with an above-zero standard residual error were identified as candidates for admixture events, representing populations that the data indicate are more closely related to each other than is demonstrated in the best-fit tree (41). TreeMix was then run using between one and six introduced migration events. When three migration events were added, the residuals were much lower than for the trees generated using other numbers of migration events.

PSMC Analysis.

PSMC was used to analyze the evolutionary demographic changes of wild emmer wheat in “Evolution Canyon” (Mt. Carmel, Israel). We selected three samples from each population with the highest depth to represent each population. Genotype calls for each position were made using the SAMtools (v1.19) mpileup command, filtering for reads with minimum base and mapping quality scores of 30. Occasional heterozygous sites were dealt with by randomly sampling one allele. Population divergence time and effective population sizes were estimated by PSMC with a mutation rate of 6.5 × 10⁻⁹ substitutions per site per year and a generation time of one generation per year.

Genome-Wide Selective Sweep Analysis.

In order to identify selective sweeps in the SFS1 population, we combined irradiance-intolerant populations (SFS1 and NFS) in a single group to exclude the potential effect of genetic drift. We used a likelihood method (the cross-population composite likelihood ratio, XP-CLR) with the following parameter: –w1 0.02 200 10000 –p0 0.95. To run XP-CLR, we assigned all SNPs to genetic positions based on 0.18 cM/Mb for the A and B subgenomes (26). Windows with the top 5% XP-CLR values were identified. For candidate genes, Gene Ontology (GO) enrichment analysis was performed using the R program with the default parameters (P < 0.05 and FDR < 0.05).

Cytogenetic Analysis.

Seeds were germinated on moist filter paper in the dark at room temperature until the roots grew to 1.5 to 2.0 cm in length. The excised root tips were then treated as described before (42). The synthesized oligonucleotide probes Oligo-pSc119.2, Oligo-pTa535, and Oligo-(GAA)7 (Invitrogen Biotechnology) were 5′ end-labeled with 6-carboxyfluorescein (6-FAM) for green or 6-carboxytetramethylrhodamine (Tamra) for red signals. Probe Oligo-3A1 (FAM-5′-AATAATTTTACACTAGAGTTGAACTAGCTCTATAAGCTAGTTCA-3′) was used to distinguish signal size and location differences in chromosomes 3A and 7A. The probe solution (20 ng/μL in 2× SSC and 1× TE buffer, pH 7.0) was denatured for 5 min in boiling water and then placed on ice. A 6-μL probe solution was used for each slide. The hybridization was performed at 37 °C overnight in a humid chamber. After hybridization, the slides were washed using 2× SSC and then mounted with a Vectashield mounting medium containing 1.5 μg/mL 4,6-diamidino-2-phenylindole (DAPI; Vector Laboratories). Images were captured with an Olympus BX-51 microscope equipped with a DP-70 CCD camera. Particular chromosomes were identified by comparing the signal pattern of probes hybridized to hexaploid wheat according to ref. 22.

Data and Material Availability.

The resequencing and SNP data generated in this work are deposited in the Genome Sequence Archive in the BIG Data Center, Beijing Institute of Genomics, Chinese Academy of Sciences, under the accession no. CRA001873.