Natural Variation of the RICE FLOWERING LOCUS T 1 Contributes to Flowering Time Divergence in Rice

In rice (Oryza sativa L.), there is a diversity in flowering time that is strictly genetically regulated. Some indica cultivars show extremely late flowering under long-day conditions, but little is known about the gene(s) involved. Here, we demonstrate that functional defects in the florigen gene RFT1 are the main cause of late flowering in an indica cultivar, Nona Bokra. Mapping and complementation studies revealed that sequence polymorphisms in the RFT1 regulatory and coding regions are likely to cause late flowering under long-day conditions. We detected polymorphisms in the promoter region that lead to reduced expression levels of RFT1. We also identified an amino acid substitution (E105K) that leads to a functional defect in Nona Bokra RFT1. Sequencing of the RFT1 region in rice accessions from a global collection showed that the E105K mutation is found only in indica, and indicated a strong association between the RFT1 haplotype and extremely late flowering in a functional Hd1 background. Furthermore, SNPs in the regulatory region of RFT1 and the E105K substitution in 1,397 accessions show strong linkage disequilibrium with a flowering time–associated SNP. Although the defective E105K allele of RFT1 (but not of another florigen gene, Hd3a) is found in many cultivars, relative rate tests revealed no evidence for differential rate of evolution of these genes. The ratios of nonsynonymous to synonymous substitutions suggest that the E105K mutation resulting in the defect in RFT1 occurred relatively recently. These findings indicate that natural mutations in RFT1 provide flowering time divergence under long-day conditions.


Introduction
The appropriate flowering time is important for reproductive success in plants.Flowering time is controlled by developmental regulation and environmental conditions, such as daylength (photoperiod) and temperature [1,2].Rice (Oryza sativa L.), a facultative short-day (SD) plant, flowers when the days become shorter than a critical daylength [3].Rice is grown in tropical, subtropical and temperate regions, and variation in flowering time (called "heading date" in rice) allows it to adapt to different climate conditions.The high level of flowering time variation is also one of the most important factors in rice breeding.A number of flowering time QTLs and genes have been identified and characterized by QTL analysis using natural variation in rice.Hd1/Se1, Ehd1, Ghd7/Lhd4, DTH8/ Ghd8/Lhd1/Hd5, Hd6, Hd16, Hd17 and Hd3a have been identified as rice flowering time genes [4][5][6][7][8][9][10][11][12][13][14][15][16][17][18][19][20].Hd1, which has several alleles, promote flowering under SD and represses it under long-day (LD) conditions [4].The loss of Hd1 function, which results in a decrease in photoperiodic response, has enabled expansion of the cultivation area of rice [21][22][23][24][25][26].Hd1 is the main source of flowering time diversity in cultivated rice [23].Ehd1 encodes a B-type response regulator.Ehd1 is a key promoter of flowering under both SD and LD conditions [6].Cultivars that have a defective Ehd1 allele (ef1) have only been identified in Taiwan so far [6,27,28].The ef1 allele is useful for long vegetative growth period in low and mid-low latitude areas [29].Ghd7 encodes a CCT domain protein, which is a strong repressor of Ehd1 expression [7,30].Ghd7 acts as a repressor of flowering under LD conditions, and the loss-of-function mutations in this gene cause early flowering under LD conditions [7,8].Such defective Ghd7 alleles are found in high-latitude areas in China and Japan [7,31].DTH8 encodes a putative HAP3 subunit of the CCAAT box-binding transcription factor, and is an ortholog of AtHAP3b in Arabidopsis [9][10][11][12].DTH8 promotes flowering under SD and represses it under LD conditions [10].Many frame-shift mutations in DTH8 cause a weak photoperiod response and early flowering in Asian cultivated rice [9][10][11][12].Cultivars with double defects in Lhd4 and Hd5, grown at the northern limit of the rice cultivation area in Japan, show extremely early flowering [32].Hd6 encodes the alpha subunit of casein kinase II, and represses flowering indirectly via Hd1 under LD conditions [13,14].Hd6 causes a strong photoperiod response and late flowering [15,16].Recently, variation in Hd16 and Hd17 was found among japonica cultivars [17][18][19].Hd16 controls flowering by regulating Ghd7 activity [18].Since Hd16 loss-of-function plants show moderately early flowering, the non-functional allele has been used to breed cultivars able to grow in a wide range of areas in Japan [18].Hd17 encodes the ELF3-like protein [19], and is allelic to Ef7 and Hd3b [33,34].A single-nucleotide polymorphism (SNP) in Hd17, found as a natural mutation in japonica cv.Koshihikari, contributes to flowering variation via regulation of Ghd7 [19].The flowering time genes mentioned above act in one or more pathway(s) that regulate flowering time by controlling the florigen genes Hd3a and RICE FLOWERING LOCUS T 1 (RFT1) [21,22,35,36].
Florigen was originally described as a product of FLOWERING LOCUS T (FT) in Arabidopsis [37].FT and its orthologs in other plants are long-distance mobile floral stimuli that move from leaves to the shoot apex [37][38][39][40].They are members of the phosphatidylethanolamine-binding protein (PEBP) family.Duplications and divergence of PEBP genes have produced three subfamilies in angiosperms: FT, TFL1 and MFT1 [41].Variation of FT-like genes contributes to flowering time variation in Arabidopsis and several crops [42][43][44][45][46][47].In rice, 19 PEBP genes have been described, and 13 of them are FT-like genes.Hd3a, RFT1 and FTL have the ability to promote flowering [35], and Hd3a and RFT1 are considered to be rice florigen genes because double RFT1-Hd3a RNAi plants do not flower [36].The expression of Hd3a and RFT1 is detected in leaf blades, and respective GFP fusion proteins have been detected in the shoot apical meristem and vascular tissue [39,40].Hd3a RNAi plants delay flowering under SD, but not under LD conditions.In contrast, RFT1 RNAi plants delay flowering under LD, but not under SD conditions.Thus, Hd3a and RFT1 function as florigens under SD and LD conditions, respectively [48].Although Hd3a and RFT1 are tandemly duplicated, highly homologous FT-like genes, they are regulated differently.Both are regulated by Ehd1, whereas Hd3a is also regulated by Hd1 [40,48].RFT1 is also regulated by the SDG724 histone methyltransferase [49].Thus, Hd3a and RFT1 gene functions and regulation are relatively well understood.However, there is little experimental evidence regarding the contributions of natural mutations in FT-like genes in rice flowering time variation [20,23].
A region close to Hd3a and RFT1 has been detected as a flowering time QTL among various cultivars and wild rice accessions [20,[50][51][52][53]. Kojima et al. [20] provided evidence for natural mutations in Hd3a in cultivars Nipponbare (Nip) (O.sativa ssp.japonica) and Kasalath (Kasa) (O.sativa ssp.indica) using near-isogenic lines (NILs) and transgenic plants carrying the NipHd3a genomic region.Although both KasaHd3a and NipHd3a are functional and accelerate flowering, the effect of KasaHd3a is stronger, and its expression levels are slightly higher than those of NipHd3a, probably because of the differences in the regions encoding the C-termini (P179N) or polymorphism(s) in the 3′ untranslated regions (UTRs) [20].Kojima et al. [20] also noted that RFT1 promoted flowering.Dung et al. [53] reported the flowering time QTL se-pat in the RFT1/Hd3a region from indica cv.Patpaku.Sano et al. [54] reported the flowering time QTL enhancer of Se1 (En-Se1) in this region from Oryza rufipogon (W593) and showed that En-se1 strongly represses flowering in the presence of Se1 (Hd1).Hagiwara et al. [50] resolved the se-pat and en-se1 regions into three QTLs (Hd3b, RFT1 and Hd3a), and showed that the nucleotide diversity in RFT1 is higher than in Hd3a.Hagiwara et al. [50] and Ebana et al. [51] predicted two causal polymorphisms in the Hd3a and RFT1 translated regions by comparing their sequences.Uga et al. [52] detected a flowering time QTL around the Hd3a/RFT1 genomic regions by using an F 2 population and backcross progenies from a cross between extremely late-flowering cv.Nona Bokra (ssp.indica) and early-flowering cv.Koshihikari (ssp.japonica), grown under natural-daylength field (ND) conditions.Although the Hd3a and RFT1 genomic regions have been frequently detected by QTL analysis, no loss-of-function alleles have been reported in rice florigen genes.Takahashi et al. [23] found six types of Hd3a alleles in 64 cultivars from a rice core collection, some of them carrying three nonsynonymous substitutions.They also showed that the Hd3a promoter activity is similar in all alleles, and there is no clear relationship between Hd3a nucleotide sequence variation and flowering time variation.An SNP close to Hd3a and RFT1, strongly associated with the flowering time variation, was reported by a genome-wide association study that used a global collection of 950 rice varieties [55].This suggests that the Hd3a and RFT1 genomic region is involved in rice flowering time variation.
The purpose of this study was to identify the causal gene of extremely late flowering.We performed fine mapping by using progenies derived from a cross between Koshihikari and SL520 [56], a line containing an introgressed segment from Nona Bokra on chromosome 6 (Chr.6).By using transformation and expression analysis, we also achieved molecular identification of a QTL for extremely late flowering that mapped to the RFT1 and Hd3a region.Here we show that Nona Bokra RFT1 is a defective gene and the main causal gene for extremely late flowering and no flowering under ND and LD conditions, respectively.Furthermore, by comparing the flowering times of different RFT1 haplotypes, we show that the defective RFT1 plays an important role in 'extremely' late flowering in rice.Therefore, we conclude that RFT1 can have a considerable effect on the diversity of flowering time.

Delimitation of a candidate genomic region for a lateflowering time QTL on the short arm of chromosome 6
To identify the causal gene(s) on the short arm of Chr. 6 that are involved in a delay in flowering time, we performed highresolution mapping of a QTL associated with flowering time.We selected the chromosome segment substitution line SL520 [56], which contains an introgressed segment on Chr. 6 (including the target QTL region) from cv.Nona Bokra in the genetic background of cv.Koshihikari (Figure 1A).SL520 showed late flowering under LD and ND conditions, but not under SD conditions (Figure 1B), which suggested that this line can be used to develop a population for QTL mapping on the short arm of Chr. 6.We used a candidate gene approach with the F 2 progeny from a cross between Koshihikari and SL520.On the basis of the previous studies [52], we assumed that RFT1 and Hd3a could be candidates for the QTL.Out of 2,750 F 2 plants, we selected those with recombination in or near RFT1 and Hd3a, and scored their flowering time (Figure S1).As we expected, some plants showed association between the genotype of the RFT1/Hd3a region and flowering time.But despite carrying the Nona Bokra segment in the short arm of Chr. 6, some plants showed early flowering under ND conditions (Figure S2A).This indicated that a Nona Bokra segment at a position other than the short arm of Chr. 6 affected flowering time in the F 2 population.Therefore, we surveyed the whole genome genotype by using 1001 SSR markers (Table S1) to define the genetic background of SL520 in detail, and found two additional small Nona Bokra segments in Chr. 2 and 3 (Figure S2B).We investigated the genotypes of these two regions using F 2 populations.Genetic interaction between the RFT1/Hd3a region and the small segment in Chr. 3 was found under ND and LD conditions (Figure S2C, D).The position of the small segment in Chr. 3 was consistent with Hd16 [18].The presence of the functional Hd16 allele from Nona Bokra made it possible to observe the effect of RFT1/ Hd3a (Figure S2C, D).Thus, Hd16 had to be fixed for the Nona Bokra allele to enable fine mapping of the RFT1/Hd3a region.Ten F 2 plants with homozygous Hd16 and the recombinant chromosome within RFT1 and Hd3a region were selected (Figure S1).In ten F 3 lines, we selected homozygous plants and measured their flowering time under ND conditions (Figure 1C).The flowering time was categorized into four classes: similar to the control line (3095#1), and flowering 10, 15 and 20 days later than the control line (Figure 1C).Similar flowering time segregation was also observed under LD conditions (Figure S3).Therefore, we suggested that at least three genomic regions in RFT1 and Hd3a are likely to be involved in late flowering in Nona Bokra.Region I (2.4 kb) was defined by two DNA markers, SNP1 and InDel3, and included the promoter region (2,283 bp from the transcription start site) and the 1 st exon of RFT1.Region II (11.1 kb) was defined by InDel3 and SNP3, and included the 1 st intron of RFT1 and the entire Hd3a promoter region (−30 bp from ATG).Region III (1.3 Mb) was defined by SNP3 and RM7488, and included the entire Hd3a coding region (Figure 1D).

Nona Bokra RFT1 is mainly responsible for late flowering or lack of flowering under LD conditions
Among the three predicted regions within the RFT1/Hd3a region, Regions I and II of the Nona Bokra allele delayed flowering compared to the Koshihikari allele, but Region III caused earlier flowering (Figure 1C, S3).As RFT1 and Hd3a encode florigens, we assumed that delayed flowering was due to a malfunction of one of them.We assessed whether Hd3a was functional by measuring flowering time in a nonfunctional Hd1 background.To develop the necessary line (SL520-hd1), SL520 was crossed with a nonfunctional Hd1 near-isogenic line (Figure S1).Whereas late flowering of SL520 could be observed in a functional Hd1 background, the flowering time of SL520-hd1 was almost the same as that of Koshihikari (Figure 2).This indicated that the Nona Bokra Hd3a protein had normal function.Furthermore, sequence analysis of the entire coding and regulatory Hd3a regions (8 kb) showed that they were identical to those in Kasalath (data not shown).The Hd3a allele of Kasalath is functional, and causes earlier flowering than the Nipponbare allele (which is identical to that of Koshihikari) [20].Thus, we concluded that the causal gene(s) for late flowering of SL520 are in the RFT1 region.
To examine the function of RFT1, we carried out a complementation analysis by introducing the RFT1 genomic region (5 kb) of Koshihikari into SL520 and Nona Bokra backgrounds.T 2 plants homozygous for a transgene containing the RFT1 genomic region showed early flowering (similar to Koshihikari) compared to the vector-transformed control plants (Figure 3A).Plants carrying the Koshihikari RFT1 genomic region in the Nona Bokra background could flower (after 120 days) under LD conditions (Figure 3B).These results indicated that the causal gene for late flowering was located in Regions I and II, and thus could be RFT1 (Figure 1D).In order to test the genetic effect of Region III, we transformed two progeny lines, 3098#2 and 3095#3 (selected from the cross between SL520 and Koshihikari, Figure S1) with Koshihikari RFT1.Both RFT1 and Hd3a regions were of Nona Bokra origin in 3098#2, whereas in 3095#3 only the RFT1 region was from Nona Bokra (Figure 3C).Both 3098#2 and 3095#3 lines transformed with Koshihikari RFT1 showed earlier flowering than respective vector-transformed controls, but 3095#3 showed later flowering than 3098#2.These results demonstrate that Koshihikari RFT1 functions to promote flowering (Figure 3).Furthermore, the flowering time difference between 3098#2 and 3095#3 indirectly suggests that genomic Region III from Koshihikari is associated with late flowering, that RFT1 (Regions I and II) from Nona Bokra is involved in late flowering, and that Region III from Nona Bokra is involved in early flowering.

A single amino acid substitution (E105K) in RFT1 causes a defect in promoting flowering
To further define the functional nucleotide polymorphisms (FNPs) in Regions I and II, we compared the deduced amino acid sequences of Koshihikari, Kasalath and Nona Bokra RFT1 with those of FT-like proteins of Arabidopsis, and found that six amino acids are different between Koshihikari and Nona Bokra.Among them, one substitution, E105K, was observed only in Nona Bokra RFT1, but not in any other proteins, including Arabidopsis FT-like proteins (Figure 4A).To verify this candidate FNP, we developed Nipponbare (nipRFT1ox), Kasalath (kasaRFT1ox) and Nona Bokra (nbRFT1ox) lines overexpressing RFT1 (Figure 4B).As RFT1 nucleotide sequences were identical in Koshihikari and Nipponbare (Figure S4), we used Nipponbare cDNA to make wild-type and artificially mutated (E105K) RFT1 genes under control of the constitutive 35S promoter (Figure 4B).T 0 transgenic nipRFT1ox and kasaRFT1ox plants showed extremely earlyflowering phenotypes in a growth chamber even under continuous light (LL) conditions (Figure 4C-F), whereas nbRFT1ox, nipRFT1(E105K)ox and control plants did not (Figure 4F).Some nbRFT1ox and nipRFT1(E105K)ox lines flowered earlier than the control lines under LD conditions (Figure 4F).In addition, nipRFT1ox plants had increased expression of other FT-like genes, Hd3a and FTL, despite the absence of Ehd1 expression (data not shown) and no changes in Hd1 expression under LL conditions (Figure 4G).This effect was less pronounced in the case of nbRFT1ox compared to nipRFT1ox plants.These results demonstrate that nbRFT1 and nipRFT1(E105K) are almost all functionally defective, and suggest that the E to K substitution is the main reason for the functional defect and for the causal variation in region II.
We also evaluated causal variation in Region I. Region I contained the 1 st exon of RFT1 (Figures 1D, 4A), and one nonsynonymous substitution leading to the V33A variation between Koshihikari (V) and Nona Bokra (A) (Figure 4A, B).However, Kasalath also had A in this position (Figure 4A, B), and KasaRFT1 could promote flowering (Figure 4C, F); therefore, causal variation in Region I was unlikely to be due to this substitution.This suggests that the causal variation in Region I is in the regulatory region of RFT1.

Expression patterns of RFT1 and Hd3a under SD, LD and ND conditions
The flowering time correlates with the expression levels of Hd3a under SD, but not RFT1 [23].Therefore, we re-examined the relationship between flowering time and the expression levels of RFT1 and Hd3a not only under SD, but also under LD and ND conditions using 24 cultivars (Figure S5A-D and Table S2).The expression of both RFT1 and Hd3a correlated with flowering time under SD (R 2 = -0.46 and -0.58, respectively), LD (R 2 = -0.69 and -0.44) and ND (R 2 = -0.64 and -0.51) conditions (Figure S5D).Nona Bokra showed low expression levels of RFT1 and Hd3a under LD and ND conditions, and the correlation coefficient increased when the data for Nona Bokra under ND conditions were removed (Figure S5C, D).
To investigate how RFT1 and Hd3a are regulated in Nona Bokra, we grew Nona Bokra, Koshihikari and Nipponbare (as a control) in a paddy field (ND conditions), and sampled their leaves from June to flowering time: late July (Koshihikari), early August (Nipponbare) and mid-November (Nona Bokra) (Figure 5).In Koshihikari and Nipponbare, RFT1 and Hd3a mRNA levels increased gradually from mid-June despite relatively long days under ND conditions (Figure 5).These expression patterns were supported by microarray data in the RiceXPro database [58] (Figure S6).In Nona Bokra, RFT1 levels increased very slowly from August to November (the cultivar's flowering time).Although high levels of RFT1 transcript were detected in Nona Bokra in early October, flowering was not observed until the end of November (Figure 5).Hd3a expression in Nona Bokra was strongly repressed from June to early September, and was induced when the daylength became less than 13 h.Hd3a expression patterns clearly showed photoperiodic response under ND conditions in Nona Bokra (Figure 5).These results revealed that expression patterns of RFT1 and Hd3a in Nona Bokra were very different from those of Koshihikari and Nipponbare under ND conditions.

A region in the short arm of chromosome 6 affects RFT1 and Hd3a expression patterns
We investigated the effect of the short arm of Chr. 6 on RFT1 and Hd3a expression under LD conditions.As RFT1 and Hd3a expression is regulated by several flowering time genes, detecting expression differences by regulatory variation was very difficult.Therefore, the SL line SL520-#3098 (with Nona Bokra allele from the distal short arm of Chr. 6 to the 1 st intron  of RFT1) derived from the cross between SL520 and Koshihikari was used for the expression analysis (Figure 6, S1).In spite of similar expression levels of Ehd1, the induction of RFT1 and Hd3a expression in SL520-#3098 was delayed in comparison with the control line (Figure 6), suggesting that the short arm of Chr. 6 affects the induction of RFT1 and Hd3a expression.Furthermore, in Nona Bokra, Ehd1 expression was not detected under LD conditions (Figure 6).This result indicated that Ehd1 was repressed by other flowering time genes in Nona Bokra, but not in Koshihikari.As a consequence, RFT1 and Hd3a were not induced, and no flowering occurred in Nona Bokra under LD conditions (Figure 1B).

Nona Bokra and similar haplotypes of RFT1 from a rice core collection are strongly associated with extremely late flowering under ND and LD conditions
The available data suggested that RFT1 is involved in extremely late flowering [50][51][52].To evaluate this hypothesis, we investigated the nucleotide polymorphism in the regulatory and coding regions of RFT1 (3.5 kb) among 66 cultivars from the rice core collection [59,60].We identified 20 haplotypes and classified them into four groups (I to IV) according to their similarity (Figure 7).Group I contained indica and japonica (including Koshihikari) cultivars (Figure S7).Most japonica cultivars had the group I allele.Group II contained only indica cultivars (including Nona Bokra).Groups III and IV were mixtures of indica and japonica cultivars (Figure S7).We found eight nonsynonymous substitutions in the RFT1 coding sequence in the rice core collection.Nona Bokra RFT1 (group II-4) had six nonsynonymous substitutions, including the E105K FNP site.Two other substitutions were found in groups I and III.Group I-5 RFT1 (cv.Khau Mac Kho [KMK]) had a P160S substitution (Figure 7).A flowering time QTL was not detected in RFT1/Hd3a regions between Koshihikari and cv.KMK [51].Group III RFT1 had a V12I substitution.Cv.Badari Dhan (BAD) in group III had functional Hd1, but its flowering time was moderately increased under LD (80 days) and ND (107 days) conditions (Table S3).In addition, no flowering time QTL was detected in the RFT1/Hd3a regions between Koshihikari and cv.Jarjan (in group III-3) (Figure 7, S8).These data suggest that the RFT1 allele encoding the P160S and V12I substitutions has normal function.
In the RFT1 regulatory region (1,756 bp upstream of the transcription start site), we found three ARR1 binding sites (Figure 7, S5).The ARR1 binding element is a candidate site for GARP, a DNA binding motif of Ehd1 [6], but there were no polymorphisms in the ARR1 binding sites in the rice core collection.We sequenced an additional regulatory region (−3,254 to −1,755 bp) of RFT1 in Nona Bokra, and found three additional ARR1 binding sites and three CCAAT boxes (Figure S4).DTH8 can potentially bind the CCAAT boxes in the RFT1 regulatory region [9][10][11][12], but there were no polymorphisms in ARR1 binding and CCAAT box sites in Nona Bokra.We found 43 SNPs and one insertion in region I in Nona Bokra, but no polymorphism was found in recognizable potential cis-element sites (Figure S4).
We investigated the relationships between RFT1 haplotype groups and flowering time.As the RFT1 phenotype appeared in a functional Hd1 background, cultivars were divided into two categories: hd1 (nonfunctional) or Hd1 (functional) (Figure 8).Most cultivars with extremely late flowering under LD and ND conditions had group II RFT1 and functional Hd1.This clearly indicated that group II RFT1 has an important role in delaying flowering in the presence of functional Hd1.However, there were two exceptions: cv.Bei Khei (BKH) and cv.Bingala (BIN).BKH had functional Hd1, but showed early flowering under LD and ND conditions (Figure 8, Table S3).This suggested that BKH had a defect in a flowering time gene other than Hd1.DTH8 could be responsible for this effect [51].BIN had functional Hd1 (the same type as BKH) and presumably functional RFT1 (group IV), but it showed extremely late flowering under ND conditions (Figure 8).The Hd3a allele in BIN was of the Nona Bokra type.This indicated that BIN should have strong flowering suppressor gene(s) other than Hd1.Cv.Radin Goi Sesat (RGS) had functional Hd1 (the same type as BKH and BIN), and showed extremely late flowering under ND conditions (Figure 8, Table S3).RGS did not have the E105K substitution, although the RFT1 regulatory region was of the Nona Bokra type (II-8 on Figure 7), indicating that this regulatory region is involved in late flowering in RGS.Taken together, these results indicated that the group II haplotype of RFT1 has an important role in 'extremely' late flowering under LD and ND conditions.

Diversification of RFT1 and Hd3a in rice
Nucleotide diversity is used to measure the degree of polymorphism within a population [61].Sequence analysis of fragments from six cultivars and four wild rice species revealed that the nucleotide diversity was higher in RFT1 than in Hd3a [50].To expand this analysis, we additionally sequenced the coding regions of RFT1 and Hd3a in cultivated and wild rice accessions (204 in total, including the accessions from the core collection, see above), and found 16 amino acid changes in RFT1 and 10 changes in Hd3a (Figure S9 and Table S3).No frameshifts or premature stop codons were found.The nucleotide diversity () of RFT1 in the coding region and in the entire gene region was higher than that of Hd3a (Table 1), as found previously [50].In addition, since defective RFT1 was found in indica, we also analyzed and Watterson's estimator () [62] within the two subspecies (indica and japonica).Although the value was similar among the two subspecies, the value of RFT1 in indica was almost four times that in japonica, indicating the higher diversity in RFT1.The haplotype diversity of RFT1 was similar to that of Hd3a in cultivated rice.The RFT1 haplotype number was larger than that of Hd3a in the entire gene region, but smaller in the coding region.Spontaneous mutations may have occurred at particular sites in the Hd3a coding region, increasing the number of haplotypes.Mutations in RFT1 were similar among the accessions, and the haplotype number appeared to decrease despite high nucleotide diversity (Table 1).Furthermore, we sequenced the entire coding regions of RFT1 (7 species) and Hd3a  ).We found a frameshift mutation in RFT1 in O. glaberrima and O. barthii, and a premature stop codon in O. meridionalis; no frameshift mutations or premature stop codons were found in Hd3a.Nonsynonymous substitutions were found in both RFT1 (42 sites) and Hd3a (27 sites) (Figure S9).If RFT1 and Hd3a were important as florigens, their amino acid sequences would be highly conserved in Oryza species.Do the differences in the extent of nucleotide diversity and in the number of nonsynonymous substitutions indicate pseudogenization of RFT1?
To evaluate the effect of natural selection, we compared the evolutionary rates [61] of RFT1 and Hd3a.Since the nucleotide diversity was higher in RFT1 than in Hd3a, this implies that functional constraint was relaxed in RFT1 after gene duplication.If so, the evolutionary rate in RFT1 would have increased.Phylogenetic analysis and colinearity of genes around RFT1 and Hd3a revealed that BradHd3a (Brachypodium distachyon Bradi1g48830) and SbHd3a (Sorghum bicolor Sb10g003940) are orthologs of RFT1 and Hd3a [63][64][65].We compared the evolutionary rates of RFT1 and Hd3a by two methods.First, the Tajima's relative rate test [66] showed no significant differences between RFT1 and Hd3a in each lineage (χ 2 test, P>0.05) (Table S4, S5).Even in Nona Bokra, no significant difference was observed.Second, the ratio of nonsynonymous substitutions per synonymous substitutions (dN/dS) was low (below 0.25) in both BradHd3a and SbHd3a.A scatter plot of dN/dS for RFT1 and Hd3a in each lineage did not show any significant differences, even though the ratio was smaller in RFT1 in some cultivars (Figure S10).Therefore, there was no clear difference in the evolutionary rates of RFT1 and Hd3a, and indicating that they are under the same degree of functional constraint.Both genes are highly conserved in Oryza species.
Haplotype group II RFT1 alleles with E105K were found in indica cultivars and wild rice, but not in japonica cultivars (indica n = 40, japonica n = 0, O. rufipogon n = 1) (Table S3 and Figure S7).We investigated the relationship between RFT1 haplotypes and the distribution of their areas of cultivation.The cultivars with group II RFT1 are found in the whole of Asia, and accessions with group II RFT1 in a functional Hd1 background are found at lower latitudes (<23°60′N) (Figure 9).

SNPs in E105K and the RFT1 regulatory region have strong linkage disequilibrium with a flowering timeassociated SNP
A genome-wide association study using 950 rice accessions from a global collection showed that the SNP most strongly associated with flowering time is located on chr.6 (SNP chr6_2912415) [55].We found that the physical distances from this SNP to ATG of RFT1 and Hd3a were about 15 and 27 kb,  respectively (Figure S11).Using Rice HapMap data [67], we investigated the linkage disequilibrium of SNPs in RFT1 and Hd3a with the peak SNP (chr6_2912415) in global rice accessions (O.sativa n = 950).We found 25 SNPs in the RFT1 region, and 21 SNPs in the Hd3a region in 217 cultivated rice accessions (Table S6).Most SNPs in the regulatory and coding regions of RFT1 and Hd3a showed strong linkage disequilibrium with the peak SNP (D' > 0.90).The D' value for RFT1 FNP(E105K) was 1.00 and that for Hd3a candidate FNP(P179N) [20] (chr6_2927179) was 0.93 (Figure S11).This indicates that RFT1 and Hd3a regions are coinherited with the peak SNP.RFT1 SNP(V33A) showed very high r 2 values (0.97), but this nonsynonymous substitution had no effect on the flowering time in the overexpression experiment (Figure 4).The r 2 value for RFT1 FNP(E105K) was 0.38 and that for Hd3a candidate FNP(P179N) was 0.56 (Figure S11).

The mechanism of extremely late flowering or lack of flowering in Nona Bokra under ND and LD conditions
In this study, we found that three regions (I to III), which include RFT1 and Hd3a, are detected as flowering time QTLs, and we revealed the roles of each region.We found that Nona Bokra RFT1 has a functional defect because of the E105K mutation.We further showed that sequence variations in the regulatory region may also reduce RFT1 function at the transcriptional level.We demonstrated that RFT1 and Hd3a were never expressed in Nona Bokra, which never flowered under LD conditions (Figure 6), but transformation with kosRFT1 restored flowering (Figure 3).This indicated that Koshihikari RFT1 with its own regulatory region was induced normally in Nona Bokra under LD conditions.This suggested that the abnormal regulation of RFT1 expression (Region I) is the major cause of the lack of flowering under LD conditions.In addition, the region from the distal end of Chr. 6 to the 1 st exon of RFT1 (the Nona Bokra segment) delayed the induction of RFT1 and Hd3a expression (Figure 6).Because the Nona Bokra segment on the distal end of Chr. 6 above the InDel2 marker did not affect flowering time (Figure 1C, D), the delay in RFT1 and Hd3a expression might be caused by Region I (Figure 6).It is known that RFT1 expression is regulated by chromatin modification [36].Region I contains the H3K9 acetylation locus (5′-UTR), and the levels of H3K36 methylation are high at the RFT1 locus [49].In Arabidopsis, quantitative modulation of chromatin silencing through cis variation (nucleotide changes) in the FLC locus was reported [68].These reports imply that the regulatory region of RFT1 (Region I) is affected by chromatin silencing, which is in turn affected by variations in this region.
Under ND conditions, RFT1 and Hd3a showed unique expression patterns.RFT1 expression in Nona Bokra was induced in summer, but Hd3a was induced after the daylength became relatively short, which preceded flowering (Figures 5,  10).This indicates that Hd3a expression is entirely controlled by photoperiod in Nona Bokra.In contrast, in Koshihikari and Nipponbare, Hd3a was induced even under relatively long-day conditions (Figure 5).This induction can be explained by defective Hd16 in Koshihikari [18], but not in Nipponbare.We found that high expression of functional RFT1 can induce Hd3a expression without inducing Ehd1 expression in Nipponbare (Figure 4G).Therefore, we suggest that RFT1 plays an important role in regulation of Hd3a expression in photoperiodsensitive varieties under LD conditions and during summer under ND conditions.
We found that functional Hd1 and Hd16 are necessary for late flowering in plants with defective RFT1 (Figure 2, S2).
These results indicate that when Hd3a expression is repressed, the RFT1 effect appears.Hd1, Hd2, Hd6 and RFT1 regions were detected by QTL analysis of the difference in flowering time between Koshihikari and Nona Bokra [52].DTH8 and Ghd7 were not detected in crosses between Koshihikari and Nona Bokra or between Koshihikari and Nipponbare [52,17].Koshihikari has functional Hd1, DTH8 and Ghd7 (Nipponbare type), nonfunctional Hd6 (Kasalath type), and a dominant Hd2 allele (because the Hd2 locus was not detected in QTL analysis in a cross between Koshihikari and Nipponbare, but was detected in a cross between Nipponbare and Kasalath [17,69]).Thus, Nona Bokra has strong flowering repression genes Hd1, Hd2, Hd6, DTH8, Ghd7 [52] and Hd16 (located near Hd6) [17,18].Its non-flowering phenotype under LD may be caused by the defect in RFT1 and by strong repression of Hd3a (Figure 10).This model can explain the following phenomenon: under LD conditions, cv.Norin 8 with RFT1 suppressed by RNAi flowers extremely late, but does flower eventually [36].Because Norin 8 has nonfunctional Hd6 (e3) [15,16], the level of Hd3a repression in Norin 8 is weaker, which allows flowering under LD conditions.
We observed the effect of the Hd3a-containing locus (Region III) on flowering in the nbRFT1 background under ND and LD conditions (Figures 1C, 3C, S3).The Nona Bokra Region III caused earlier flowering.The Hd3a nucleotide sequence (including promoter) is identical between Koshihikari and Nipponbare, and between Nona Bokra and Kasalath.The Koshihikari/Nipponbare and Nona Bokra/Kasalath types of Hd3a are functional, but the Kasalath allele (KasaHd3a) promotes earlier flowering than the Nipponbare allele [20], although the reported difference between Nipponbare and a near-isogenic line carrying kasaHd3a was small (5 days) under LD conditions [20].In this study, in a defective RFT1 background, the difference between line III (nbRFT1/kosHd3a) and SL520 (nbRFT1/nbHd3a) under LD conditions was larger (>20 days) (Figure S3).Therefore, we speculate that if the RFT1 pathway is not functional, a small functional difference in the Hd3a (pathway) can be detected (Figure 10).Our data show that RFT1 can regulate the expression of Hd3a and FTL (Figure 4G).The RFT1 and Hd3a pathways are coordinated and can precisely control flowering time [48].The first induced florigen initiates the flowering process (Figure 10).In Nona Bokra, Hd3a appears to be the only florigen gene: its repression under LD conditions causes extremely late flowering under ND conditions and no flowering under LD conditions.

Functional divergence and pseudogenization of RFT1 and Hd3a
We found that the duplication of the RFT1 and Hd3a ancestor gene occurred after the divergence between rice and Brachypodium (about 40 Mya [70,71]).Since RFT1 and Hd3a genes are present in all Oryza species (Figure S9), we assume that a single duplication event occurred before their divergence.The evolutionary rates were not significantly different between functional and defective RFT1 (Table S4 and Figure S10), and the defective alleles accumulated very few mutations, suggesting a functional constraint on both genes.This is consistent with earlier reports that both RFT1 and Hd3a are important for flowering time regulation in cultivated rice [36].The E105K allele is still under functional constraint for some reason.Our model implies that RFT1 can be expressed under ND conditions, and can regulate Hd3a.We have revealed that variation in two loci, RFT1 regulatory (I) and coding (II) regions, is involved in the regulation of flowering.Most group II RFT1 cultivars had the Nona Bokra-like regulatory region and the defective protein (E105K).Only two cultivars (II-8 type, RGS and Puluik Arang) had Nona Bokralike regulatory regions, but no E105K substitution (Figure 7).The II-8 allele might be the ancestral allele of group II defective RFT1.Furthermore, the haplotype network indicated that the group IV allele was the original allele of group II and III RFT1.These data suggest that the loss-of-function event in the regulatory region occurred first, and that the G to A (E105K) substitution was the second mutation.Could RFT1 be an example of pseudogenization?The presence of two florigens in rice, encoded by Hd3a and RFT1, might generate functional redundancy.However, previous studies and our current studies have shown that RFT1 is regulated by unique mechanisms [40,49] and promotes flowering in the absence of Hd3a function [36].The RFT1 pathway is important in the late-flowering varieties (e.g.Hd1 functional cultivars) under LD conditions or in high-latitude areas (Figure 10).Thus, the mechanisms may have been established by the gain of function after gene duplication.In general, indica rice is grown in tropical and subtropical regions at low latitudes, i.e., in warmer climate and shorter days, and these factors accelerate flowering.Under SD conditions, flowering time is mainly regulated by the Hd3a pathway (Figure 10), because Hd3a expression is induced earlier than RFT1.This situation may lead to a relaxation of functional constraint on RFT1.However, sometimes a mutation in a florigen protein leads to an opposite function: for example, paralogous genes TFL1 and FT are key controllers of flowering but have opposite effects in Arabidopsis [38].It was reported that only a single amino acid change is sufficient to switch the flowering promoter or repressor function of TFL1 and FT [72,73].Therefore, limited mutations may be permissible in florigen genes.
We reconfirmed in a large scale test that RFT1 and Hd3a diversity in O. sativa is lower than in O. rufipogon (Table 1) [50].The number of haplotypes for the coding region of RFT1 was lower than that of Hd3a in both species, and the same loss-of-function RFT1 allele was found in both indica and O. rufipogon (Table 1, Figure S7).If the loss of function occurred in the common ancestor of indica and rufipogon, some mutations would have accumulated in the RFT1 allele.However, the sequence of the defective RFT1 allele of O. sativa indica (group II-1) is identical to that of O. rufipogon (W1723).This implies that the defective RFT1 in W1723 was introgressed from an indica cultivar.
Perennial wild rice typically shows extremely late flowering; after germination, it may take two years or more to reach flowering even under SD conditions (at low latitudes).Vegetative growth and reproduction is the main propagation strategy of the wild rice species.Wild rice has strong photoperiod sensitivity, and some QTLs were reported around the RFT1/Hd3a region in perennial wild rice [50,54,74].The nonsynonymous substitutions and a frameshift mutation we found in the RFT1 and Hd3a coding regions in several wild Oryza species may underlie the variability of flowering time.Further studies are required to elucidate the role of RFT1 and Hd3a in Oryza wild species and cultivars.

Flowering time control for rice breeding
In this study, we found four major haplotypes of RFT1.The relationship between flowering time and RFT1 haplotype suggested that group I, III and IV RFT1 are functional alleles, whereas group II RFT1 is nonfunctional (Table S3).No flowering time QTLs were previously detected around RFT1 in crosses between Koshihikari (allele group I) and the following cultivars: Nipponbare (group I) [17], JAR (III) (Figure S8), NAB (IV) [51] or KASA (IV) [69] (Figure 7), consistent with group I, III and IV RFT1 being functional haplotypes.Thus, these haplotype data would be useful for flowering time prediction.Group II RFT1 haplotypes were found in the whole of Asia, whereas at lower latitudes they were found only in the presence of functional Hd1 (Figure 9).This distribution was consistent with flowering responses to daylength (Figure 8).Furthermore, the varieties carrying group II and IV RFT1 have mainly group I and IV Hd1 alleles, respectively (Table S3) [24].The indica subspecies contains two genetically distinct subgroups, indica and aus.Because independent events of domestication of O. rufipogon carrying the ancestral types of Hd1 alleles I and IV of indica and aus group have occurred, respectively [24].This suggests that the defective mutation occurred in the indica lineage after the divergence of indica/ aus.Indica rice is grown mainly in tropical and subtropical regions.Therefore, the wide distribution of group II RFT1 haplotypes might indicate that the defective RFT1 allele first appeared at lower latitudes, and subsequently expanded to northern areas due to loss of function of flowering repressor gene(s) (e.g.Hd1) and artificial breeding.
We sequenced RFT1 and Hd3a in 204 accessions, including parental lines for rice breeding.We found some elite cultivars and accessions that have the defective RFT1 allele (Table S3), for example IR36 (a "Green Revolution" rice cultivar from the International Rice Research Institute [IRRI]) [75] (Table S3).Although IR36 is an early-flowering cultivar, in breeding programs some of its progeny showed late flowering [76].One of the parents of IR36, cv.TKM6, has defective RFT1 (E105K) and is photoperiod-insensitive, but its progenies show various (and even high) photoperiod sensitivity [77].The defective RFT1 may be an unfavorable allele for breeding programs in middle-to high-latitude areas.On the other hand, we detected the defective RFT1 in deep-water cultivars (Table S3).All deep-water rice accessions are photoperiod-sensitive [78]; flowering time is critical for propagation of their seeds after seasonal flooding.Our data suggested that Hd3a pathway acts as a single flowering activator in deep-water rice.To continue vegetative growth in water during the rainy season (in summer), Hd3a has to be repressed.According to our model (Figure 10), Hd3a is repressed by flowering time genes under LD conditions (in summer), but is induced after being subjected to SD conditions (in autumn), except in equatorial areas.Therefore, under particular environmental conditions (e.g.seasonal rapid flooding), defective RFT1 might be a favorable allele.
The effect of defective RFT1 on flowering is strong in functional Hd1 and other flowering repressor gene backgrounds (e.g.Hd16) (Figure 2, S2).The tropical and subtropical indica cultivars often have not only defective RFT1, but also strong flowering repressor genes (e.g.Hd1, Hd2 and Hd6) [50][51][52][53][54]. Therefore, when parents with defective RFT1 are used in middle-or high-latitude areas, removing the defective RFT1 region is advantageous.The ability to select for or against defective RFT1 should make it easier to breed for cultivars with the desired flowering time.Our experimental evidence suggests that RFT1 is the primary gene responsible for flowering time variation (see also 55).Most SNPs in the RFT1 region showed strong linkage disequilibrium with the flowering-associated SNP [chr6_2912415] (Figure 7, S11), although the r 2 value for RFT1 FNP(E105K) was relatively low (0.38).This may be explained by allele frequency differences and by the likely role of the RFT1 regulatory regions (as opposed to the E105K substitution) as the primary determinant of late flowering under ND conditions.Therefore, it is possible to remove or select a defective RFT1 haplotype by using group II RFT1pro-specific SNPs and FNP(E105K).

Plant materials and growth conditions
We crossed Koshihikari and SL520 [56], and used the progeny from self-pollinated individuals (Figure S1) for highresolution mapping of the RFT1 and Hd3a regions.Only Nona Bokra-type Hd16 lines were selected for mapping in F 3 (Figure S1).Within the F 3 population, three regions were mapped around the RFT1 and Hd3a regions in a paddy field in Tsukuba, Japan: 36°03′N, 140°10′E (Figure 1).The progeny from self pollination of these plants (F 4 ) was used for the flowering time test and for the expression analysis under LD conditions.To verify the interaction between RFT1 and Hd1, we crossed Koshihikari and KantoIL1, which contains the Hd1 chromosomal region (Hd1 nonfunctional allele) from the Kasalath genome [79] (Figure S1).We measured the flowering time in the self-pollinated progeny of these plants (F 2 ) in a functional Hd16 background.The molecular markers used for high-resolution mapping are listed in Tables S1 and S7.Genomic regions I and II, which include the RFT1 promoter, coding region and recombination region are shown in Figure 1D.For fine mapping and flowering time evaluation under ND conditions, plants were grown from the middle of April.The daylengths during growth are shown in Figure 5B.For flowering evaluation and gene expression analysis, plants were grown in a growth cabinet under SD (10 h light/14 h dark) or LD (14.5 h light/ 9.5 h dark) conditions.Days to flowering under each condition were scored as the number of days required from germination to the emergence of the first panicle.

RFT1 cloning and generation of transgenic plants for complementation test
The Koshihikari RFT1 genomic region (4,914 bp; Figure S4) was amplified by PCR and cloned into the SalI site of pCR-Blunt II-TOPO (Invitrogen), resulting in TOPO-RFT1.TOPO-RFT1 was digested with ApaI/KpnI and KpnI/XhoI, and subcloned into pPZP-2Hlac, digested with ApaI and XhoI [80], yielding pPZP-gRFT1.This plasmid was introduced into Agrobacterium strain EHA101 and subsequently into SL520, progeny from crossing between Koshihikari and SL520, and Nona Bokra by using the Agrobacterium-mediated rapid method [81].

RFT1 overexpression
Fragments containing the 5′-UTR of RFT1 from Nipponbare (nipRFT1), Nona Bokra (nbRFT1) and Kasalath (KasaRFT1) were amplified by PCR with primers designed for Nipponbare cDNA (Table S7), and cloned into pCR-Blunt II-TOPO.The E105K mutation was introduced by PCR with RFT1-F/ RFT1(E105K)-R and RFT1(E105K)-F/RFT1-R primer sets (Table S7) containing appropriate restriction sites, and the fragment obtained (RFT1[E105K]) was inserted into pPZP-Ha3(+) [80], resulting in pPZP-35S:RFT1.These plasmids were introduced into Nipponbare as mentioned above [81].Regenerating transformed plants (T 0 ) were incubated on Murashige and Skoog medium to induce roots under LL conditions for 20 days, and then transplanted to soil under LD conditions.We defined the flowering time (for each individual plant) as the time from transplanting onto the regeneration medium to the appearance of the first panicle.

Expression analysis
Nipponbare (reference genome cultivar), Koshihikari and Nona Bokra were used for expression analysis under ND conditions.They were cultivated from April to October.Nona Bokra was transplanted to a greenhouse from October to avoid low temperatures.Leaf samples were collected every 7 days from 37 to 120 days after sowing, and every 14 days from 135 to 218 days (at 9:00 am).
The control line, #3098-2-1-1 and Nona Bokra were used for expression analysis under LD conditions.Leaf samples were collected every 10 days from 10 to 90 days after sowing.Additionally, Nona Bokra samples were collected 120 and 180 days after sowing.
Total RNA extraction and TaqMan quantitative real-time PCR (qRT-PCR) analysis were performed as described previously [14].Expression data were normalized against UBQ expression.

Sequencing of RFT1 and Hd3a in O. sativa accessions and wild rice species
The upstream regions and the ORFs of RFT1 and Hd3a were amplified by PCR (94°C for 30 s, 60°C for 60 s, and 72°C for 60 s; 35 cycles) with primers listed in Table S7 and AmpliTaq DNA (Applied Biosystems).The nucleotide sequences of the PCR products were analyzed with an ABI3700 capillary sequencer (Applied Biosystems).Accessions were provided by NIAS Gene Bank [60]

Sequence analysis
Multiple alignment and evolutionary analyses of the FT-like proteins were conducted in MEGA5 [83].To consider the codon structure, multiple alignment of 363 sequences was performed in MUSCLE [84].For further analyses, positions with gaps were excluded.To analyze the equality of evolutionary rates of RFT1 and Hd3a in each accession, Tajima's relative rate test was performed with orthologs from either B. distachyon (Bradi1g48830) or S. bicolor (Sb10g003940) as an outgroup [63][64][65].The numbers of nonsynonymous substitutions per nonsynonymous site and of synonymous substitutions per synonymous site were calculated by using the modified Nei-Gojobori model (assumed transition/transversion bias = 2) [86].

Figure 1 .
Figure 1.Comparison of SL520 with its parent lines Koshihikari and Nona Bokra, and delimitation of the candidate genomic regions for RFT1 and Hd3a.(A) Graphical genotype of SL520.The white and black regions denote the regions homozygous for Koshihikari and Nona Bokra alleles, respectively.(B) Flowering time of SL520, Koshihikari and Nona Bokra under short-day (SD:10L/14D), long-day (LD:14.5L/9.5D)and natural field (ND) conditions in Tsukuba, Japan in 2009.Each bar represents mean ± SD (n = 10).(C) Delimitation of the candidate genomic regions for RFT1 and Hd3a.Left: Graphical genotypes of the RFT1 and Hd3a regions in 10 lines, in which recombination occurred between InDel1 and RM7488.Right: Days to flowering under ND conditions.Each bar represents mean ± SD.The number of plants is indicated in parentheses.(D) Division of the RFT1/ Hd3a region into three subregions.The 2.4-kb, 11.1-kb and 1.3-Mb candidate regions were defined by linkage analysis.The effect of the Nona Bokra allele on flowering time is shown below each region.The underlined 5-kb (chr.6: 2,923,569-2,928,437; see Figure S4) fragment was used for complementation analysis.

Figure 4 .
Figure 4. Amino acid sequences of RFT1 and Hd3a and phenotypes of RFT1 overexpressors.(A) Comparison of amino acid sequences of rice and Arabidopsis FT-like proteins.Rice RFT1 and Hd3a (from three cultivars as indicated), and Arabidopsis FT, TSF and TFL1 are shown.Conserved amino acids are shaded in black, dark gray or light gray depending on the level of conservation.The blue and red arrows indicate the V33A and E105K sites, respectively.The red boxes indicate the binding interface with 14-3-3 protein [57].(B) 35S:RFT1 constructs.RFT1 alleles used: Nip, Nipponbare; Kasa, Kasalath; NB, Nona Bokra.E105K, Nipponbare allele with the introduced point mutation.Black and gray boxes denote ORF and UTR, respectively.Amino acids differing from those in Nipponbare are shown as white letters.(C) Regenerated plants (cv.Nipponbare) transformed with 35S:RFT1 constructs or with the empty vector (Vec).Bar = 1 cm.(D, E) 35S:nipRFT1 (D) and 35S:kasaRFT1 (E) plants at higher magnifications.Bar = 2 mm.(F) Flowering time of 20 T 0 plants overexpressing each of the 35S:RFT1 constructs or vector control under LL (continuous light) and LD conditions.Day 0 corresponds to the date of transplanting onto regeneration media under LL conditions.On day 20, counting of the flowering plants was started, and plates were transferred to LD.Each open ellipse indicates an individual plant.(G) Expression levels of RFT1, Hd3a, FTL and Hd1 expression in T 0 plants transformed with the 35S:RFT1 constructs or vector under LL conditions.N.D. indicates that transcripts were not detected.Samples were taken 20 days after transplanting onto regeneration media.Samples from five plants were mixed together.Each bar represents the mean ± SD (technical replicates n=3).Expression values were plotted on a log 10 scale.Ehd1 was not detected (data not shown).doi: 10.1371/journal.pone.0075959.g004

Figure 5 .
Figure 5.The mRNA levels of RFT1 and Hd3a in Koshihikari, Nipponbare and Nona Bokra during the growing period in a natural field.(A) Expression levels of RFT1 and Hd3a in top leaves.Expression values were plotted on a log 10 scale.Black triangles indicate the flowering date for each cultivar.(B) Changes in daylength at the growing location.Open triangle indicates the sowing date for each cultivar.Yellow boxes represent the period of long-day conditions (>13 h daylength).

Figure 7 .
Figure 7. RFT1 allele haplotypes.RFT1 gene regions of plants from the rice core collection were compared with those of Nipponbare (NIP).BKH, Bhei Khei; DAV, Davao; NB, Nona Bokra; JAR, Jarjan; NAB, Naba; KASA, Kasalath.Site numbering and physical positions are based on the Nipponbare sequence (RAP-DB build 5.0 [85]).Nucleotide substitutions are highlighted in orange.Red vertical bars indicate ARR1 binding elements.Gray and purple boxes represent UTRs and ORF of RFT1, respectively.Amino acid changes are indicated below the alignment.O or X on the right indicate whether a QTL was detected or not, respectively, in this region in the population derived from the cross between Koshihikari and particular cultivars.

Figure 8 .
Figure 8. Flowering time of plants from the core collection under SD, LD and ND conditions.Accessions were divided into four groups by RFT1 haplotypes and Hd1 function (functional Hd1 is shown in bold).RFT1 haplotypes were divided into two groups, candidate functional (groups I, III and IV) or defective (group II) haplotypes.Accessions were grouped according to flowering time (in 10-day steps).The diameter of the black circles reflects the number of accessions.BIN, Bingala; BKH, Bei Khei; RGS, Radin Goi Sesat.The distributions of flowering time of group II with functional Hd1 deviated from other distributions under LD and ND conditions (P<0.01,Kolmogorov-Smirnov test), but not under SD conditions (P = 0.135).doi: 10.1371/journal.pone.0075959.g008

Figure 9 .
Figure 9. Geographic distribution of rice accessions and RFT1 haplotype groups.(A) Distribution and RFT1 haplotypes of 143 O. sativa and 16 O.rufipogon accessions.(B) The subset containing only accessions with functional Hd1.Black line indicates 23.6° north latitude.To the south of this line, maximum daylength is less than 13 h.doi: 10.1371/journal.pone.0075959.g009

Figure 10 .
Figure 10.Schematic models of the flowering pathway in Nona Bokra and Koshihikari under ND and LD conditions.RFT1 in Nona Bokra is defective, but the pathway that represses Hd3a expression is functional.Koshihikari has functional RFT1, but has defects (in Hd6 and Hd16) in the pathway that represses Hd3a.Under ND conditions, RFT1 begin to express from summer, but it cannot induce Hd3a expression in Nona Bokra.In autumn (short day conditions), Hd3a expression is induced and Nona Bokra can flower.Under LD conditions, RFT1 expression does not occur, and Hd3a is repressed by flowering repressor genes, and Nona Bokra cannot flower.In Koshihikari, Hd3a and RFT1 express under LD and ND condition, and it shows early flowering.Geographic location of the original cultivation regions of Koshihikari and Nona Bokra is shown on the right.doi: 10.1371/journal.pone.0075959.g010

Table 1 .
Nucleotide diversity and divergence in each region of RFT1 and Hd3a gene.