Identification of the VERNALIZATION 4 gene reveals the origin of spring growth habit in ancient wheats from South Asia

Significance

A precise regulation of flowering time is critical for plant reproductive success and for cereal crops to maximize grain production. In wheat, barley, and other temperate cereals, vernalization genes play an important role in the acceleration of reproductive development after long periods of low temperatures during the winter (vernalization). In this study, we identified VERNALIZATION 4 (VRN-D4), a vernalization gene that was critical for the development of spring growth habit in the ancient wheats from South Asia. We show that mutations in regulatory regions of VRN-D4 are shared with other VRN-A1 alleles and can be used to modulate the vernalization response. These previously unknown alleles provide breeders new tools to engineer wheat varieties better adapted to different or changing environments.

Abstract

Wheat varieties with a winter growth habit require long exposures to low temperatures (vernalization) to accelerate flowering. Natural variation in four vernalization genes regulating this requirement has favored wheat adaptation to different environments. The first three genes (VRN1–VRN3) have been cloned and characterized before. Here we show that the fourth gene, VRN-D4, originated by the insertion of a ∼290-kb region from chromosome arm 5AL into the proximal region of chromosome arm 5DS. The inserted 5AL region includes a copy of VRN-A1 that carries distinctive mutations in its coding and regulatory regions. Three lines of evidence confirmed that this gene is VRN-D4: it cosegregated with VRN-D4 in a high-density mapping population; it was expressed earlier than other VRN1 genes in the absence of vernalization; and induced mutations in this gene resulted in delayed flowering. VRN-D4 was found in most accessions of the ancient subspecies Triticum aestivum ssp. sphaerococcum from South Asia. This subspecies showed a significant reduction of genetic diversity and increased genetic differentiation in the centromeric region of chromosome 5D, suggesting that VRN-D4 likely contributed to local adaptation and was favored by positive selection. Three adjacent SNPs in a regulatory region of the VRN-D4 first intron disrupt the binding of GLYCINE-RICH RNA-BINDING PROTEIN 2 (TaGRP2), a known repressor of VRN1 expression. The same SNPs were identified in VRN-A1 alleles previously associated with reduced vernalization requirement. These alleles can be used to modulate vernalization requirements and to develop wheat varieties better adapted to different or changing environments.

Hexaploid wheat (Triticum aestivum L., genomes AABBDD) originated in the coastal area of the Caspian Sea ∼8,000 y ago from the hybridization of a tetraploid wheat (Triticum turgidum L., genomes AABB) and diploid Aegilops tauschii (genome DD) (1, 2). Its polyploid nature and rapidly changing genome facilitated wheat adaptation to a wide range of environments as it spread westward into Europe and eastward into Asia (3). Since then, wheat has become a major global crop contributing approximately one-fifth of the calories and proteins consumed by humans. Although wheat production in 2013 exceeded 700 million tons (Food and Agriculture Organization, faostat.fao.org), further increases are required to satisfy the food demands of a fast-growing human population. We hypothesize that a better understanding of the adaptative changes that occurred in wheat during the early expansion of agriculture can contribute to the development of new strategies to increase wheat productivity in changing environments.

Wheat adaptation to different latitudes and planting dates is mainly associated with natural variation in the photoperiod gene PPD1 that promotes flowering under long days (4–6) and in the vernalization gene VRN1 that modulates the requirement of long exposures to cold temperatures (vernalization) to induce flowering (7–9). The photoperiod and vernalization pathways converge at the regulation of FT1 (= VRN3 in wheat) (10), which encodes a mobile protein that travels from leaves to the shoot apical meristem (SAM) (11, 12). Once in the SAM, FT1 becomes part of a florigen activation complex that binds to the VRN1 promoter, inducing its transcription (13–15). In common wheat, most genes exist in three copies (homeologs), which are designated using their respective genomes (e.g., VRN-A1, VRN-B1, and VRN-D1).

VRN1 is a MADS-box (MCM1/AGAMOUS/DEFICIENS/SRF) transcription factor homologous to the Arabidopsis meristem identity gene APETALA 1 (7–9), and its activation by vernalization (16) or by FT1 (14) accelerates the transition of the SAM from the vegetative to the reproductive phase. Wheat varieties carrying the ancestral VRN1 allele have a strong vernalization requirement and are classified as “winter wheats.” Mutation in the VRN1 promoter (17–19) or large deletions in the first intron (20) greatly reduce or eliminate the vernalization requirement, and wheat varieties carrying these alleles are classified as “spring wheats.” A critical regulatory region in the VRN1 first intron is transcribed and contains binding sites for the RNA-binding protein TaGRP2 (21). TaGRP2 is homologous to the Arabidopsis GLYCINE-RICH RNA-BINDING PROTEIN7 (GRP7), which is a single-stranded RNA-binding protein involved in the regulation of flowering (22). In the absence of cold, TaGRP2 binds to VRN1 pre-mRNA and inhibits VRN1 expression. During vernalization, TaGRP2 is gradually modified (O-GlcNAcylation), allowing the interaction with a carbohydrate-binding protein called VER2, which reduces TaGRP2-mediated repression of VRN1 expression (21).

In wheat and other temperate grasses, VRN1 is also expressed in the leaves where it acts as a repressor of VRN2 (23). VRN2 encodes a protein containing a putative zinc finger and a CCT protein–protein interaction domain (24) that acts as a long-day repressor of FT1 to prevent flowering during the fall (25, 26). The induction of VRN1 during winter prevents the up-regulation of VRN2 when daylength increases during spring (23). In the absence of VRN2, FT1 is up-regulated by long days, further increasing VRN1 expression, and closing a positive feedback loop that leads to an irreversible acceleration of flowering (27).

A fourth vernalization locus, VRN-D4, has been mapped on the proximal region of the short arm of chromosome 5D (henceforth, 5DS) (28, 29), but the gene underlying this locus has not been identified yet. VRN-D4 was first described in the Australian wheat variety Gabo (30), which likely inherited it from the Indian cultivar Muzaffarnagar (31). VRN-D4 is found at higher frequency among wheat accessions from South Asia than from other parts of the world (32–34). VRN-D4 was transferred from Gabo into Triple Dirk to develop Triple Dirk “F” (TDF), which is part of a set of isogenic lines carrying different vernalization genes (35, 36). Segregating populations generated from these isogenic lines revealed that VRN-D4 has strong epistatic interactions with the other vernalization genes and is part of the same vernalization pathway (29).

The VRN-D4 gene is located in a proximal region of 5DS that shows no recombination (29), limiting the use of map-based cloning to identify this gene. In this study we used alternative approaches to show that the VRN-D4 locus originated from the insertion of a large 5AL chromosome segment including the VRN-A1 gene into the proximal region of chromosome arm 5DS. We also show that this insertion is almost fixed in the ancient T. aestivum ssp. sphaerococcum (Perc.) Mac Key from South Asia and is associated with a reduction in diversity in the neighboring chromosome regions. Finally, we characterize natural variation in the regulatory regions of the first intron of VRN-D4 and related VRN-A1 alleles and demonstrate that they affect the binding of a repressor of VRN1 expression.

Results

VRN-D4 Is Linked to Early Expression of VRN1.

In a previous study, we showed that the presence of the Vrn-D4 allele was associated with the up-regulation of VRN1 and FT1 and the down-regulation of VRN2 transcript levels in the leaves (28). Based on this result and on the epistatic interactions among these genes, we suggested that VRN-D4 likely operates upstream or as a part of the positive regulatory feedback loop connecting these three genes. To determine which of these genes is affected first by VRN-D4, we conducted a developmental time course analysis of gene expression under nonvernalizing conditions in isogenic lines with and without the dominant Vrn-D4 allele.

Triple Dirk “C” plants (TDC, winter growth habit, vrn-D4) exhibited low transcript levels of VRN1 and FT1 and high levels of VRN2 at the three tested sampling points (Fig. 1 A–C) confirming previous results (28). By contrast, Triple Dirk F plants (TDF, spring growth habit, Vrn-D4) showed increased expression of VRN1 and FT1 and decreasing levels of VRN2 at week 5 (Fig. 1 A–C). At the earliest sampling point (week 1), the differences in VRN1 between TDF and TDC were already highly significant (P = 0.002, 38-fold change, Fig. 1A), whereas the differences in VRN2 were only marginally significant (P = 0.05, 2-fold change, Fig. 1B). Differences in FT1 transcript levels between TDC and TDF were detected only in the latest sampling point (week 5, Fig. 1C). These results suggest that VRN1 is the earliest target (direct or indirect) of VRN-D4 among these three genes.

Fig. 1.

Characterization of the TDF (Vrn-D4) line. Transcript levels of (A) VRN-1 (all homeologs combined), (B) VRN2, and (C) FT1. (D) VRN-D4 effect on heading time. The experiment was stopped 114 d after sowing. (E) Transcript levels of the three different VRN1 homeologs. (F) VRN-A1 haploid copy number. Jagger, Munstertaler, and Moldova were included as controls for VRN-A1 copy number equal to 1, 2, and 3, respectively (37). (A–E) Means were calculated from eight biological replications. (F) Means were calculated from four biological replications. Error bars in all graphs represent SEM. TDC and CS5402: winter, no allele for spring growth habit. TDF: spring, Vrn-D4. TDD: spring, Vrn-A1a. BW (Bobwhite): spring, Vrn-A1, Vrn-B1, Vrn-D1. NS, nonsignificant; *P < 0.05, **P < 0.01, and ***P < 0.001.

In the absence of vernalization, TDF flowered much earlier than TDC but still 12.6 d later than Triple Dirk “D” (TDD, strong Vrn-A1a allele) (17) and 18.3 d later than Bobwhite (BW, alleles for spring growth habit at the three VRN1 homeologs) (Fig. 1D). The later flowering of TDF relative to TDD and Bobwhite was associated with lower transcript levels of VRN1 and FT1 and higher transcript levels of VRN2 (Fig. 1 A–C, week 3).

Surprisingly, we found that VRN-A1 was the only VRN1 homeolog significantly up-regulated in TDF relative to TDC. VRN-A1 transcript levels were higher from week 1, and the levels were similar in TDF and TDD (Fig. 1E). By contrast, neither VRN-B1 nor VRN-D1 transcripts were detected at the first (Fig. 1E) or third week (SI Appendix, Fig. S1A) in TDF or TDD. VRN-B1 and VRN-D1 were detected in Bobwhite (dominant Vrn-A1 Vrn-B1 Vrn-D1) after the first week (SI Appendix, Fig. S1). These results suggested that either the VRN-A1 homeolog is a specific target of VRN-D4 or that an additional copy of VRN-A1 is present in TDF.

TDF Carries a Second Copy of VRN-A1 Completely Linked to VRN-D4.

Using a VRN1 copy number Taqman assay described in previous studies (6, 37), we observed one copy of VRN-A1 in winter lines TDC and CS5402 (recessive vrn-D4 allele), but two copies of VRN-A1 in TDF (Fig. 1F). When we sequenced the coding region of VRN-A1 in TDF and Gabo (dominant Vrn-D4 allele), we discovered a fixed A/C polymorphism at position 367 (henceforth, A367C, SI Appendix, Fig. S2A) that was not present in any of the VRN1 sequences deposited in GenBank. Lines carrying the recessive vrn-D4 allele showed only the canonical “A” at position 367, present in all available VRN-A1 sequences (SI Appendix, Fig. S2A). The A367C SNP was completely linked with spring growth habit in a high-density mapping population (3,182 gametes) segregating for VRN-D4 on chromosome arm 5DS (29). The six plants with the VRN-D4 closest recombination events are shown in SI Appendix, Fig. S2B. Based on this result, we hypothesized that the duplicated copy of VRN-A1 on chromosome arm 5DS was a good candidate for VRN-D4.

This hypothesis was further supported by the earlier expression of the VRN-A1 gene inserted on chromosome arm 5DS transcript. In the leaf samples collected from 1- and 3-wk-old TDF plants, all of the transcripts carry the C variant at position 367 (SI Appendix, Fig. S3). Only by the fifth week, low expression of the A variant (vrn-A1 allele) was detected in TDF together with the stronger expression of the C variant. The control lines showed the expected results: the winter TDC showed no VRN1 expression, whereas the spring TDD and BW plants showed early expression of the A variant (Vrn-A1a allele from 5AL, SI Appendix, Fig. S3). Taken together, these results support the hypothesis that the VRN-A1 copy in chromosome 5DS is VRN-D4.

Validation of the VRN-D4 Candidate Gene Using Induced Mutations.

To test the previous hypothesis, we mutagenized a population of 1,153 TDF lines with ethyl methane sulphonate (EMS) and screened them using a CelI assay as described before (38). We identified four nonsynonymous mutations and one splicing site mutant in the VRN1 copy in 5DS (SI Appendix, Method S1). Based on their predicted effect on protein function, we selected two mutations for phenotypic characterization (SI Appendix, Table S1). The first mutation generated a change from glutamic acid to lysine at position 158 (E158K, Fig. 2A). Although this is usually a tolerated amino acid change (BLOSUM 62 score = 1), it is located in a conserved position that suggests some evolutionary constraints (SI Appendix, Figs. S4 and S5). In a population segregating for the E158K mutation, plants heterozygous and homozygous for the mutation flowered 5.1 d (P = 0.0004) and 9.1 d (P = 2.15e⁻⁵) later than plants carrying the wild-type allele, respectively. Plants carrying the wild-type allele flowered at the same time as the TDF control (P = 0.76, Fig. 2B).

Fig. 2.

Effect of VRN-D4 TILLING mutants on flowering time. (A) VRN-D4 gene structure; arrows indicate the location of natural SNP A367C, and TILLING mutants E158K and splicing site mutant (SI Appendix, Table S1). (B) Effect of the E158K mutation on heading time. WT, homozygous wild-type allele; het, heterozygous; E158K, homozygous mutant allele. (C) Effect of the splicing site mutation on heading time. Spl, homozygous splicing site mutation. Bars indicate means of at least six biological replications. Error bars indicate SEM. NS, nonsignificant; *P < 0.05, **P < 0.01, and ***P < 0.001.

The second mutation, found in homozygous state, eliminated the donor splicing site at the fourth exon (Fig. 2A). Translation of the adjacent intron sequence predicts a premature stop codon 8 bp downstream of the induced mutation that eliminates 40% of the VRN-D4 protein. The presence of this premature stop codon was confirmed in the transcripts from the TDF mutant (SI Appendix, Method S1). TDF control plants flowered on average 48 d after sowing, whereas plants homozygous for the splicing site mutation flowered on average 146 d later than TDF (P < 5.04e⁻¹⁴) and at the same time of the winter TDC control (P = 0.910, Fig. 2C). These two independent mutations confirmed that the duplicated VRN-A1 copy on the proximal region of 5DS is VRN-D4.

Physical Map of the VRN-D4 Region.

A BLAST search of the International Wheat Genome Sequencing Consortium (IWGSC) draft sequence of Chinese Spring (CS) chromosome arms (39) using VRN-A1 as query revealed the presence of VRN-D4 (carrying the diagnostic A367C SNP) in the 5DS database. Although CS does not carry VRN-D4, the CS ditelosomic 5DS line (henceforth, CS-dt5DS) used to sequence the 5DS arm was generated using the variety Gabo (40), which does carry VRN-D4 (36, 41).

The physical map of the 5DS arm from CS-dt5DS, developed as part of the IWGSC, was used to determine the size of the inserted 5AL segment. A screen of the CS-dt5DS BAC library minimum tiling path (MTP) clones (SI Appendix, Method S2) (42) yielded BAC TaeCsp5DShA_0038_M14, which is part of overlapping contigs CTG87-CTG61 (SI Appendix, Table S2). We sequenced 12 BACs from these two contigs (SI Appendix, Method S2) and deposited the assembled ∼680-kb sequence in GenBank (AC270085). The comparison of the assembled sequence with databases for CS-5AL (39), Triticum urartu (diploid donor of the A genome) (43), and three databases for Ae. tauschii (D genome) (SI Appendix, Method S2) identified a ∼290-kb region with high identity (>99%) to the A genome databases, flanked by sequences with high identity (>99%) to Ae. tauschii (Fig. 3A). This result indicates that the VRN-D4 locus originated by the insertion of a large segment from chromosome arm 5AL into chromosome arm 5DS.

Fig. 3.

Graphical representation of the VRN-D4 region in chromosome 5DS. (A) CS-dt-5DS sequence assembly (∼680 kb, AC270085) compared with 5DS and 5AL chromosome arm databases (IWGSC), T. urartu, and three different Ae. tauschii databases (SI Appendix, Method S2). Two genes [VRN-D4 and Cyclin A2-1 (TRIUR3_24995)] and one hypothetical gene TRIUR3_08792 were annotated. (B) Schematic representation of CS-dt-5DS sequence. Dark blue, CS-dt-5DS sequence; red, >99% identical to CS 5AL; black, Ae. tauschii contigs aegilops.wheat.ucdavis.edu/ATGSP/; gray rectangles, CS-dt-5DS BACs; yellow, regions >99% identical; and triangles, 5DS/5AL insertion point upstream and downstream of VRN-D4. (C and D) Molecular marker for the 5DS/5AL insertion sites upstream and downstream of VRN-D4, respectively. The 1,440-bp product indicates the presence of the 5DS/5AL upstream insertion and the 687-bp band is a PCR amplification control from BJ315664 (SI Appendix, Method S3). The 1,283-bp product indicates the presence of the 5DS/5AL downstream insertion and the 534-bp band is a PCR amplification control from BE606654 (SI Appendix, Method S3).

Annotation of the assembled region of CS-dt-5DS revealed the presence of three genes (AC270085). VRN-D4 was the only gene in the 290-kb 5AL region. None of the genes known to flank VRN-A1 in chromosome arm 5AL (7) were detected in the sequence of the inserted 5AL segment or by PCR in the DNA extracted from sorted 5DS chromosome arms from CS-dt-5DS (SI Appendix, Fig. S6). The two other genes were located in the D-genome region downstream of VRN-D4 (CYCLIN-A2-1 and hypothetical gene TRIUR3_08792). No genes were detected in the 225-kb sequence upstream of the 5AL insertion.

The first border of the 5DS/5AL insertion is located 88.3 kb upstream of the start codon of VRN-D4 (henceforth “upstream insertion site,” Fig. 3B) and the second border is located between 201 and 203 kb downstream of VRN-D4 (henceforth “downstream insertion site,” Fig. 3B). We developed diagnostic PCR markers for each border (Fig. 3C and SI Appendix, Method S3), that were perfectly correlated with the A367C SNP diagnostic for VRN-D4 in the critical recombinants of the TDF × CS5402 fine mapping population (SI Appendix, Fig. S2C) and in all of the accessions described in SI Appendix, Tables S4 and S5. These results suggest that VRN-D4 originated from the same 5DS/5AL insertion event.

We could not find a single Ae. tauschii contig matching both D-genome borders of the CS-dt-5DS sequence. Ae. tauschii contig 5265.1 showed >99% identity with a 104-kb region of CS-dt-5DS upstream of the 5AL insertion site (Fig. 3B). However, the additional 200 kb of this Ae. tauschii contig showed no similarity with the CS-dt-5DS D-genome sequence on the other side of the 5AL insertion. This result suggests that one of the original borders of the 5DS/5AL insertion was likely modified during or after the insertion event. This hypothesis was further supported by high levels of identity between different Ae. tauschii contigs and the CS-dt-5DS region downstream of the 290-kb insertion (Fig. 3B).

To identify the closest source of the 5AL segment inserted on chromosome arm 5DS, we compared the genomic sequence of VRN-D4 with seven VRN-A1 alleles from polyploid wheat (SI Appendix, Fig. S7). VRN1 sequences form Triticum monococcum and T. urartu were included as outgroups. A neighbor joining tree showed that the closest sequence to VRN-D4 was the VRN-A1 allele from CS. The two sequences differed only in one SNP (A367C) in 14,912 bp. The VRN-A1 sequences from Jagger and Claire were also very similar, showing only one and three additional SNPs in the analyzed region, respectively (SI Appendix, Fig. S7).

Characterization of the VRN-D4 Allele for Spring Growth Habit.

Although VRN-D4 is expressed early in development in the absence of vernalization, its sequence lacks previously known mutations associated with spring growth habit (7, 17–20). Instead, VRN-D4 has three close SNPs in the RIP-3 region of the first intron (SI Appendix, Fig. S8), which acts as a binding site for TaGRP2, a known negative regulator of VRN1 (21) (Fig. 4A).

Fig. 4.

Binding of GRP2 to RNA probes including the RIP-3 site. (A) GRP2 binding region in VRN1 first intron. (B) Predicted RNA structure of intron one region including 50 nt or 200 nt at each side of the RIP-3 binding site (highlighted with a blue line). (C) Predicted RNA structure of the three 34-nt probes (L). RIP-3L, canonical sequence; RIP-3L^VRN-A1, SNP T2783C; and RIP-3L^VRN-D4, SNPs G2780C, T2783C, and C2784T. Arrows indicate the position of the SNPs. Nucleotide colors represent base pair probability for the predicted RNA secondary structure, with red representing the highest probability. (D) RNA-electrophoretic mobility shift assay (REMSA) showing the interaction of TaGRP2 with the three RNA probes. Interactions were tested with two different protein amounts (1: 1 µM and 2: 2 µM). Nonbiotinylated RIP-3L was used as competitor. (E) Same probes as in D were tested at two different binding temperatures (lines 1–3: 4 °C and lines 4–7: 23 °C).

We found the canonical RIP-3 sequence in VRN-B1 and VRN-D1 in polyploid wheat and in VRN-A^m1 in diploid wheat T. monococcum (SI Appendix, Fig. S8). By contrast, we detected two different RIP-3 haplotypes in the VRN-A1 sequence (SI Appendix, Fig. S8). The first one includes a T-to-C mutation at position 2,783 from the start codon (henceforth, T2783C or RIP-3^VRN-A1) and was found in all 33 T. urartu, 13 T. turgidum ssp. dicoccoides, 13 T. turgidum ssp. dicoccum, and 12 T. aestivum sequenced accessions (SI Appendix, Table S3). The second haplotype, found in VRN-D4 and the related VRN-A1 sequences in CS, Jagger, and Claire (henceforth RIP-3^VRN-D4), included the previous T2783C SNP and two additional SNPs at positions 2,780 (G2780C) and 2,784 (C2784T). Interestingly, the VRN-A1 alleles from Jagger and Claire have been previously linked to a weak vernalization requirement (6, 44).

To test if these SNPs at the RIP-3 region affect the binding of the TaGRP2 protein (SI Appendix, Fig. S9A), we performed RNA-electrophoretic mobility shift assays (REMSA, SI Appendix, Method S4). First, we tested a 27-nt RNA probe (SI Appendix, Fig. S9B) identical to the previously published RIP-3 fragment (21) and compared it to two 27-nt RNA probes containing the SNPs present in RIP-3^VRN-A1 and RIP-3^VRN-D4 (SI Appendix, Fig. S9C). TaGRP2 bound to the RIP-3 fragment in a dose-dependent manner (SI Appendix, Fig. S9B). The strength of the binding decreased in the presence of the T2783C mutation in RIP-3^VRN-A1 and was almost eliminated in the presence of the three mutations characteristic of the RIP-3^VRN-D4 allele (SI Appendix, Fig. S9C).

The predicted secondary structure of the 27-nt RIP-3 sequence was different from the one generated using longer RIP-3 flanking sequences (200 nt) and was affected by the SNPs in RIP-3^VRN-A1 and RIP-3^VRN-D4 (Fig. 4B and SI Appendix, Fig. S9D). To test if these differences have an effect on the differential binding of TaGRP2, we developed longer RIP-3 RNA probes (34 nt, henceforth RIP-3L) that have similar structures to that predicted for RIP-3 with longer flanking regions (Fig. 4B). The three polymorphic SNPs are all located within a single-stranded RNA loop of the predicted RIP-3L structure and their presence does not affect the predicted structure of the probes (Fig. 4C). TaGRP2 showed a strong binding to the RIP-3L probe with the canonical sequence, a weaker binding to the RIP-3L^VRN-A1 probe with one SNP, and almost no binding to the RIP-3L^VRN-D4 probe with 3 SNPs (Fig. 4 D and E). A similar result was found when the experiment was repeated under low binding temperature (4 °C, Fig. 4E). In summary, the two probe lengths and two binding temperatures tested in the in vitro assays showed similar relative binding intensities for RIP-3, RIP-3^VRN-A1, and RIP-3^VRN-D4.

Geographical Distribution of VRN-D4.

The two markers for the 5DS/5AL insertion and for the A367C SNP are diagnostic for the presence of VRN-D4 and were used to study the distribution of VRN-D4 in hexaploid wheat. The RIP-3^VRN-D4 haplotype is also informative, but its presence should be interpreted with caution because this haplotype is also present in the VRN-A1 locus in some accessions.

These four markers were all present in 22 of the 26 T. aestivum ssp. aestivum accessions (SI Appendix, Table S4) previously proposed to carry VRN-D4 based on genetic analyses (32–34). However, markers for the 5DS/5AL insertion and for the A367C SNP were not detected in accessions IL193, CL035, IL346, and CL030 (SI Appendix, Table S4). Copy number assays showed that these four accessions have no additional copies of VRN-A1 outside the ones in chromosome arm 5AL. Accessions IL193 and CL035 have only one copy of VRN-A1 carrying the RIP-3^VRN-D4 haplotype (same as Jagger, Claire, and CS), whereas accessions IL346 and CL030 have two linked copies of VRN-A1. The last two lines have a C349T fixed SNP (SI Appendix, Table S4) that is diagnostic for a tandem duplication of VRN-A1 on chromosome 5AL (6). These results confirmed the absence of VRN-D4 in these four accessions, suggesting that the original crosses and genetic analyses should be repeated.

The collection sites of the T. aestivum ssp. aestivum accessions carrying VRN-D4 (SI Appendix, Table S4) partially overlap with the distribution of T. aestivum ssp. sphaerococcum (Punjab region, current Northeast Pakistan and Northwest of India, SI Appendix, Fig. S10). This last subspecies has been previously proposed as a putative source of VRN-D4 (41), a suggestion that is supported by the high frequency of VRN-D4 (94%) among the 33 T. aestivum ssp. sphaerococcum accessions characterized in this study (SI Appendix, Table S5). The only two accessions classified as T. aestivum ssp. sphaerococcum lacking VRN-D4 were collected in China (CItr 8610 and CItr 10911). These two lines also differed from the other T. aestivum ssp. sphaerococcum accessions by the presence of the Vrn-D1a allele (SI Appendix, Table S5), which carries a large intron deletion that is associated with a spring growth habit and that is present at high frequency in China (45).

Analyses of additional molecular markers for six VRN1 and VRN3 alleles showed no additional alleles for spring growth habit among the 31 T. aestivum ssp. sphaerococcum accessions that carry VRN-D4 (SI Appendix, Table S5). For each accession, we also extracted RNA from leaves of 1-wk-old unvernalized plants and performed RT-PCR with primers that amplify both VRN-D4 and VRN-A1. Sequencing of the amplification products showed only the VRN-D4 allele (A367C SNP), confirming that this gene is expressed early in development in these T. aestivum ssp. sphaerococcum accessions (SI Appendix, Table S5). To determine if the absence of VRN-D4 from these two lines from China was the result of admixture with other T. aestivum subspecies, we compared them with accessions from five different T. aestivum subspecies (SI Appendix, Method S5 and Table S6) using the 90K iSelect SNP chip (46). The five subspecies were clustered into six groups, including two different groups for T. aestivum ssp. sphaerococcum and one group combining the spring accessions of T. aestivum ssp. aestivum with the accessions from T. aestivum ssp. compactum (Fig. 5B). Chinese accessions CItr 8610 and CItr 10911 exhibited evidence of extensive admixture, which may explain the absence of VRN-D4 in these two lines (asterisks in Fig. 5C and SI Appendix, Fig. S11).

Fig. 5.

Genetic relationships among T. aestivum subspecies. (A) Dark gray indicate presence and light gray absence of VRN-D4 based on the markers developed in this study. (B) Population structure based on the analysis of 100 lines with 16,371 SNP markers (six groups). Each bar represents one genotype, and the identification numbers below correspond to the same numbers in SI Appendix, Table S6 describing each accession. (C) Classification of T. aestivum subspecies based on GRIN. Green, aestivum (spring growth habit); orange, compactum; blue, sphaerococcum; yellow, macha; pink, aestivum (winter growth habit); and cyan, spelta. Asterisks indicate two Chinese lines classified as ssp. sphaerococcum that show extensive admixture and lack VRN-D4. (D) Level of genetic differentiation between all pairs of subspecies determined using the fixation index (F_ST). All comparisons are significant at P < 0.001. (E) Standardized genetic diversity values for the distal and proximal regions in chromosomes from homeologous group 5. Note the high reduction in diversity in chromosome 5D proximal region of T. aestivum ssp. sphaerococcum relative to the other subspecies.

The VRN-D4 gene was found in all of the T. aestivum ssp. sphaerococcum accessions from the Punjab region, suggesting that it likely contributed to local adaptation and was favored by positive selection. To test this possibility, we estimated the average genetic diversity (47) across chromosomes and chromosome regions in the different T. aestivum subspecies using 14,236 SNPs with known chromosome locations. To reduce the possible effect of ascertainment bias, diversity values per chromosome were divided by the average diversity of their respective genome and subspecies (SI Appendix, Method S6 and Table S7). These values, referred to hereafter as “standardized genetic diversity,” are summarized in SI Appendix, Fig. S12. For chromosome 5D, the standardized genetic diversity of the T. aestivum ssp. sphaerococcum accessions carrying VRN-D4 is less than half the average of the values from the other four T. aestivum subspecies. None of the other 20 chromosomes showed such a large difference in standardized genetic diversity among subspecies (SI Appendix, Fig. S12A).

To determine which region of chromosome 5D was responsible for the reduced diversity, we divided each chromosome arm into one proximal (1/3 of the genetic length including VRN-D4) and one distal region (2/3 of the genetic length). For comparison, a similar analysis was performed for homeologous chromosomes 5A and 5B (Fig. 5E) and for all of the chromosomes of the D genome (SI Appendix, Fig. S12B). Chromosome 5D of T. aestivum ssp. sphaerococcum showed a 1.8-fold reduction in standardized genetic diversity in the distal region and a 4.7-fold reduction in the proximal region compared with the average of the other four subspecies (Fig. 5E). Such a drastic relative reduction in diversity in the proximal region was not observed in chromosomes 5A and 5B (Fig. 5E) or in any of the other chromosomes of the D genome in T. aestivum ssp. sphaerococcum (SI Appendix, Fig. S12B). The centromeric region of chromosome 5D, where VRN-D4 is located, also showed the highest level of genetic differentiation (F_ST) between T. aestivum ssp. sphaerococcum and the pooled data from the other four subspecies (SI Appendix, Fig. S13). In summary, these results indicate that the proximal region of chromosome 5D in T. aestivum ssp. sphaerococcum accounts for most of the observed reduction in genetic diversity in chromosome 5D in this subspecies and contributes to its genetic differentiation of T. aestivum ssp. sphaerococcum from the other subspecies of T. aestivum.

Discussion

The Dynamic Wheat Genome.

This study shows that the VRN-D4 gene originated from the insertion of a large region of chromosome arm 5AL (including VRN-A1) into chromosome arm 5DS. The presence of genes and gene fragments in noncolinear locations is relatively frequent in the large genomes of the temperate grasses, which originated from massive amplifications of repetitive elements (39, 48–50). Some DNA transposons in wheat (e.g., CACTA) are able to move genes and gene fragments to noncolinear locations (51), as reported for rice ‘Pack-MULEs’ (52) and maize helitrons (53). However, this mechanism is associated with mobilization of small DNA fragments (<3 kb) (54) and is unlikely to explain the 290-kb insertion that originated VRN-D4.

An alternative mechanism proposed to explain noncolinear DNA sequences involves the insertion of nonhomologous DNA segments to repair double-strand breaks generated by the insertion of repetitive elements (54). A hallmark of this mechanism is the presence of a retrotransposon insertion in one of the borders of the acceptor site (54). This mechanism is associated with a wide range of insertion sizes and therefore can explain the 290-kb 5AL insertion in chromosome arm 5DS. Unfortunately, one of the original downstream borders of the 5AL insertion has been altered in the CS-dt-5DS sequence, limiting our ability to determine the molecular mechanism that originated this large 5AL insertion.

The insertion of the 5AL chromosome segment into the 5DS chromosome arm provides a well-documented example of the ability of large wheat chromosome regions to move to different chromosome locations within historical times. This result suggests that contigs in a chromosome arm database that match sequences in different chromosomes may represent real insertion events and should not be automatically discarded as DNA contaminations. In addition, our results demonstrate that some of the sequences deposited in the wheat survey sequences of CS (39) actually belong to different wheat varieties. Differences between CS and CS-derived cytogenetic stocks have been reported before (40) and should be considered when analyzing the CS survey sequences.

The Origin of VRN-D4.

Previous genetic studies identified the presence of VRN-D4 in a few accessions of T. aestivum ssp. sphaerococcum and suggested that this subspecies could be the source of VRN-D4 (32, 41, 55). This subspecies, first described in 1921 (56), has been found mainly in the Punjab region, including vast areas of eastern Pakistan and northern India. The characteristic round grains are abundant in archeological sites from the Indus Valley Civilization between 4,600 and 3,900 y ago (57).

The identification and characterization of VRN-D4 allowed us to study the distribution of this gene in a larger germplasm collection (SI Appendix, Tables S4 and S5). We found VRN-D4 in 31 of the 33 accessions of T. aestivum ssp. sphaerococcum (SI Appendix, Table S5) but in none of the accessions from the other subspecies described in SI Appendix, Table S6. These results, together with the higher frequency of VRN-D4 in T. aestivum ssp. aestivum accessions from South Asia (SI Appendix, Table S4), support the hypothesis that VRN-D4 originated in T. aestivum ssp. sphaerococcum. The only two accessions classified as T. aestivum ssp. sphaerococcum that lacked the VRN-D4 were collected in China (outside the central area of distribution for this subspecies) and carry the alternative VRN-D1a allele for spring growth habit (SI Appendix, Table S5). These two accessions exhibit high levels of admixture (Fig. 5), which may explain the absence of VRN-D4 and the presence of a different allele for spring growth habit.

Except for the previous two accessions from China, none of the T. aestivum ssp. sphaerococcum accessions described in SI Appendix, Table S5 carry known VRN1 or VRN3 alleles for spring growth habit (SI Appendix, Table S5). This result suggests that VRN-D4 may be the only allele for spring growth habit in this subspecies. This was confirmed for accessions K-5498 and K-23822 (SI Appendix, Table S5) in a previously published genetic study (32). The importance of VRN-D4 in the induction of spring growth habit in this subspecies was also supported by the early expression of this gene during development in the absence of vernalization. Early expression was confirmed for all of the accessions of T. aestivum ssp. sphaerococcum carrying this gene (SI Appendix, Table S5).

Patterns of Genetic Diversity in the VRN-D4 Region.

The proximal region of chromosome 5D in T. aestivum ssp. sphaerococcum showed a four- to fivefold reduction in standardized genetic diversity and a local increase in genetic differentiation relative to the other four subspecies. We did not observe reductions in genetic diversity of this magnitude in the proximal regions of homeologous chromosomes 5A and 5B (Fig. 5E) or in the proximal regions of the other six chromosomes from the D genome (SI Appendix, Fig. S12B). These patterns of genetic diversity are likely associated with adaptive selection for Vrn-D4 or a tightly linked gene in the centromeric region of chromosome 5D.

Indirect support for Vrn-D4’s role in local adaptation is provided by previous studies showing evidence of selection for different VRN1 loci (58). Among modern wheat varieties, the strong Vrn-A1a allele for early flowering is predominant in regions of very cold winters, where spring wheats are planted in the spring. By contrast, the weaker Vrn-B1a and VrnD1a alleles are present at higher frequencies in the Mediterranean climates where spring wheats are planted during the fall (45, 59, 60). These results provide an example of the selective forces that could have also operated on VRN-D4. Adaptive selection for VRN-D4 might have been favored by the absence of other alleles for spring growth habit in the T. aestivum ssp. sphaerococcum accessions from South Asia (SI Appendix, Table S5) and by the partial dominant nature of the Vrn-D4 allele for early flowering (29).

We currently do not know when the VRN-D4 gene originated in T. aestivum ssp. sphaerococcum and how fast it was adopted. However, if DNAs can be obtained from ancient archeological samples of T. aestivum ssp. sphaerococcum, the diagnostic markers developed in this study for VRN-D4 can be used to answer these questions.

Genetic Differentiation of T. aestivum ssp. sphaerococcum.

In addition to the characterization of the VRN-D4 region, this study provides an overall view of T. aestivum ssp. sphaerococcum genetic diversity that is very different from the one suggested in earlier studies. Early studies based on few morphological traits proposed that T. aestivum ssp. sphaerococcum and ssp. aestivum differed only in a few loci (61, 62). However, our study based on a genome-wide sample of SNPs, suggests a very different picture. Both the F_ST and cluster analyses showed that T. aestivum ssp. sphaerococcum is one of the most differentiated subspecies of T. aestivum (Fig. 5D and SI Appendix, Fig. S11). A genome-wide scan of F_ST values across the 21 chromosomes revealed multiple genomic regions with strong differentiation, including VRN-D4 (SI Appendix, Fig. S14). It would be interesting to investigate if these differentiated regions are also targets of adaptive selection.

We hypothesize that as common wheat moved eastward and westward from its original area on the south coast of the Caspian Sea, it differentiated and adapted to new environments. Limited exchanges between the western and Asian human populations during the early stages of agriculture expansion likely favored the simultaneous differentiation of crops, cultures, and languages. The divergence of T. aestivum ssp. sphaerococcum from the western T. aestivum subspecies parallels the differentiation of the Indo-European languages, where the Indo-Iranian subfamilies are more distinct from the western Celtic, Germanic, Italic, and Balto-Slavic families (63).

VRN-D4 Provides Insights into the Regulation of Vernalization Requirement.

None of the polymorphisms previously known to be associated with spring growth habit in VRN1 regulatory regions (17–20) were found in VRN-D4. Therefore, alternative mechanisms are required to explain the dominant spring growth habit associated with the Vrn-D4 allele. We discuss below three nonmutually exclusive hypotheses.

Hypothesis 1: “Chromosome location.”

It is possible that the new chromosome location of the VRN-D4 gene affects its regulation. The original VRN-A1 gene is located in the middle of the 5AL arm in a region of normal recombination, whereas VRN-D4 is located in a more proximal location where no recombination has been detected so far (29). It is well established that chromosome translocations and rearrangements can have a deep impact on gene expression. Neighboring genes can have highly correlated expression profiles (64, 65) and can impact the expression of relocated genes. However, no genes were identified close to VRN-D4 limiting their potential effect on VRN-D4 expression.

Position effects on gene expression can also be associated with changes in the proximity to euchromatic or heterochromatic regions. In Drosophila, these effects can spread over considerable distances, but the genes located closest to the breakpoint are commonly the most severely affected (66). However, in the VRN-D4 locus, the duplicated VRN1 gene is still flanked by large segments of the original 5AL chromatin (∼80 kb upstream and ∼200 kb downstream of VRN-D4), which may buffer potential effects of the different chromosome location on VRN-D4 expression.

Hypothesis 2: “Effect of the K123Q polymorphism.”

The A367C SNP in VRN-D4 results in an amino acid change (K123Q) that has not been detected so far in any published VRN1 allele (or in related MADS box proteins, SI Appendix, Figs. S4 and S5). Therefore, we currently have no information to determine if this polymorphism can contribute to the differences in flowering time. Any proposed mechanism for this polymorphism will have to explain why the transcript levels of VRN-D4 are higher than those from VRN-A1 during early developmental stages (Fig. 1E and SI Appendix, Fig. S3), when both genes have identical promoters. Hypothesis 3 provides a potential explanation for these differences.

Hypothesis 3: “SNP mutations in the RIP-3 region.”

The three linked SNPs in the transcribed RIP-3 region in the VRN-D4 first intron provide an attractive explanation for the early expression of VRN-D4. These three linked SNPs are located within a predicted single-stranded RNA loop targeted by TaGRP2 (Fig. 4C). The RIP-3 region in wheat is similar to the binding site of the Arabidopsis GRP7 homolog (SI Appendix, Fig. S8A), and one of the three SNPs found in RIP-3^VRN-D4 is in a position that has been shown to be critical for the binding of GRP7 in Arabidopsis (67). Variation in AtGRP7 affects Arabidopsis flowering time by regulating the expression of the MADS-box FLOWERING LOCUS C (FLC) (22).

Previous studies in transgenic wheats have shown that overexpression of TaGRP2 delays flowering, whereas its down-regulation by RNAi accelerates flowering relative to the wild-type control. These results confirmed that TaGRP2 functions as a flowering repressor (21). Our REMSA results show that the presence of three SNPs in the RIP-3^VRN-D4 region is sufficient to disrupt the binding between TaGRP2 and its target site (Fig. 4). TaGRP2 binding to the VRN1 pre-mRNA is required for the inhibition of VRN1 expression (21), so the disruption of this interaction can explain the earlier expression of VRN-D4. This mechanism is also consistent with the spring growth habit associated with large deletions in the first intron of VRN1, including the RIP-3 region (20, 68).

We found that the same three SNPs present in the RIP-3 region in VRN-D4 are also present in the RIP-3 regions of the related VRN-A1 alleles in common wheat varieties Jagger, Claire, and CS (SI Appendix, Fig. S8). The VRN-A1 allele from CS has not been characterized in detail, but the winter wheat varieties Jagger and Claire have a reduced vernalization requirement (6, 44) that maps to the VRN-A1 locus. The cause of the reduced vernalization requirement is still controversial and has been attributed to copy number variation in Claire (6) and to a polymorphism (A180V) in the VRN-A1 protein in Jagger (44). The discovery that both Claire and Jagger carry the RIP-3^VRN-D4 allele provides an alternative explanation for their reduced vernalization requirement. We propose that the limited binding of the TaGRP2 repressor to the RIP-3^VRN-D4 allele in these two varieties results in higher VRN-A1 transcript levels and reduced vernalization requirement. This last interpretation is consistent with the expression reported for these two varieties. VRN-A1 transcript levels increased more rapidly in Jagger than in the winter line 2174 (RIP-3^VRN-A1) (44) and also more rapidly in Claire than in the winter varieties Malacca and Hereward (6).

Despite carrying the same RIP-3^VRN-D4 allele, the effects of VRN-D4 and the VRN-A1 allele present in Jagger and Claire are not identical. A short vernalization treatment (3 wk) accelerates flowering in Claire and Jagger more than 1 mo (6, 44), but has no significant effects in TDF (VRN-D4) (28). This result suggests that additional mechanisms contribute to the earlier flowering associated with the VRN-D4 allele. The different chromosome location or the additional K123Q polymorphism in VRN-D4 can contribute to these differences, but it is also possible that the different genetic backgrounds in these varieties can also modulate the effect of the RIP-3^VRN-D4 allele.

An interesting observation was that the VRN-A1 alleles characterized so far in diploid T. urartu and in wild and cultivated polyploid wheat accessions (SI Appendix, Table S3) all have the T2783C SNP in the RIP-3 region (RIP-3^VRN-A1 allele). Our REMSA experiments showed that this mutation is sufficient to weaken the interaction between TaGRP2 and its target RNA sequence, although the disruption is not as strong as the one observed when the three SNPs are present in the RIP-3^VRN-D4 allele (Fig. 4). It is tempting to speculate that the T2783C SNP in RIP-3^VRN-A1 can contribute to the higher transcript levels in VRN-A1 relative to VRN-B1 and VRN-D1 observed in winter wheat plants after 6 wk of vernalization (69). We are currently generating the genetic stocks required to compare the expression levels of the different RIP-3 alleles at the VRN-A1 locus.

In summary, the cloning of a vernalization gene originated in the ancient Indus Valley Civilization has improved our understanding of the natural variation of the main wheat vernalization gene in modern winter wheat varieties. Breeders can use these alleles to engineer wheat varieties better adapted to different or changing environments. One more time, “What’s past is prologue.”

Materials and Methods

Plant Materials and Growth Conditions.

Triple Dirk F (TDF, spring, Vrn-D4) and Triple Dirk C (TDC, winter, vrn-D4), and Triple Dirk D (TDD, Vrn-A1a) are part of a series of isogenic lines carrying different alleles for spring growth habit in the genetic background of Triple Dirk (35, 36). The Australian variety Gabo is the source of the VRN-D4 allele in TDF. Lines with the closest recombination events to VRN-D4 were used to validate the VRN-D4 candidate gene. These lines were identified in a high density map of VRN-D4 generated from the cross between TDF and a substitution line of chromosome 5D from synthetic wheat 5402 in CS (CS5402) (28, 29).

We characterized two common wheat varieties with reduced vernalization requirement linked to the VRN-A1 locus (6, 44). Jagger (KS-82-W-418/STEPHENS), a hard red winter wheat, was the most widely planted variety in Kansas and Oklahoma for 12 y. Its early maturity helps Jagger to partially avoid wheat rust and heat stress. Claire (WASP/FLAME) is a variety from the United Kingdom released in 1999 that remains popular for its suitability for early sowing.

To determine the geographic distribution of VRN-D4, we characterized a collection of 26 T. aestivum accessions with known vernalization genes (32–34) (SI Appendix, Table S4) and a collection of 33 T. aestivum ssp. sphaerococcum accessions obtained from the National Small Grains Collection (NSGC) and the Vavilov Institute of Plant Industry (SI Appendix, Table S5). For progeny tests and expression experiments, plants were grown in PGR15 growth chambers (Conviron) adjusted to 16 h of light (22 °C) and 8 h of darkness (18 °C).

Expression Profiles and Copy Number Variation.

RNA samples were extracted from leaves using the Spectrum Plant Total RNA Kit (Sigma-Aldrich). VRN1, VRN2, FT and ACTIN primers for SYBR GREEN quantitative PCR were developed in previous studies (10, 70, 71). VRN-A1, VRN-B1 and VRN-D1 expression was tested using genome specific qRT-PCR primers described before (69). Quantitative RT-PCR was performed using SYBR Green and a 7500 Fast Real-Time PCR system (Applied Biosystems). Amplified products were confirmed by sequencing. VRN-A1 and VRN-D4 transcripts were differentiated by the presence of an additional BstNI restriction site in VRN-D4 (A367C polymorphism, SI Appendix, Fig. S3). VRN-A1 copy number estimation was performed using the Taqman assay described before (6).

Physical Map.

The MTP clones of the CS-dt-5DS BAC library (42) were screened at Sabanci University (Turkey) with three pairs of primers for VRN-D4 listed in SI Appendix, Table S8. The selected BAC and 11 overlapping BACs belonging to two overlapping contigs (SI Appendix, Table S2) were sequenced using a combination of Illumina and 454 sequencing (SI Appendix, Method S2). Annotation was conducted using “Triannot” (72). The assembled sequence was compared with the 5AL arm database of the IWGSC (39), T. urartu (v1.26 plants.ensembl.org) (43), and three Ae. tauschii (D genome) databases (plants.ensembl.org/Aegilops_tauschii/, GCA_000347335.1 (73), aegilops.wheat.ucdavis.edu/ATGSP/ (74), and https://urgi.versailles.inra.fr/download/iwgsc/TGAC_WGS_assemblies_of_other_wheat_species, TGAC_WGS_tauschii_v1). We used BLAT v. 35 filtered for matches that were longer than 500 bp and >99% identical (SI Appendix, Method S2).

“Targeting Local Lesions in Genomes” Mutants.

A targeting local lesions in genomes (TILLING) population of 1,153 lines of hexaploid wheat line TDF (VRN-D4) was mutagenized with 0.9% ethyl methane sulphonate (EMS) and screened using a CelI assay as described before (38). DNA pools (four lines per pool) were screened with three sets of primers (SI Appendix, Method S1). The potential effects of the different mutations on protein function were predicted using programs PROVEAN, SIFT, and PolyPhen-2 (SI Appendix, Method S1).

T. aestivum ssp. sphaerococcum Population Structure.

A total of 100 lines including 33 T. aestivum ssp. sphaerococcum, 15 T. aestivum ssp. macha, 17 T. aestivum ssp. compactum, 15 T. aestivum ssp. spelta, and 20 T. aestivum ssp. aestivum (10 spring and 10 winter) accessions obtained from the NSGC (SI Appendix, Table S6) were genotyped with the iSelect 90k SNP array (46). The population structure analysis was conducted using the program STRUCTURE version 2.3.4 (SI Appendix, Method S5). The fixation index (F_ST) between all possible pairwise combinations of the five subspecies was calculated using the program Arlequin (75). Mean F_ST values were calculated using a sliding window of seven mapped SNPs (46) with steps of four SNPs to generate smoother curves. F_ST values for each SNP were calculated by comparing T. aestivum ssp. sphaerococcum with pooled data from the other four subspecies.