The Response of COL and FT Homologues to Photoperiodic Regulation in Carrot (Daucus carota L.)

Carrot (Daucus carota L.) is a biennial plant requiring vernalization to induce flowering, but long days can promote its premature bolting and flowering. The basic genetic network controlling the flowering time has been constructed for carrot, but there is limited information on the molecular mechanisms underlying the photoperiodic flowering response. The published carrot genome could provide an effective tool for systematically retrieving the key integrator genes of GIGANTEA (GI), CONSTANS-LIKE (COL), FLOWERING LOCUS T (FT), and SUPPRESSOR OF OVEREXPRESSION OF CONSTANS 1 (SOC1) homologues in the photoperiod pathway. In this study, the bolting time of wild species “Songzi” (Ws) could be regulated by different photoperiods, but the orange cultivar “Amsterdam forcing” (Af) displayed no bolting phenomenon. According to the carrot genome and previous de novo transcriptome, 1 DcGI, 15 DcCOLs, 2 DcFTs, and 3 DcSOC1s were identified in the photoperiod pathway. The circadian rhythm peaks of DcGI, DcCOL2, DcCOL5a, and DcCOL13b could be delayed under long days (LDs). The peak value of DcCOL2 in Af (12.9) was significantly higher than that in Ws (6.8) under short day (SD) conditions, and was reduced under LD conditions (5.0). The peak values of DcCOL5a in Ws were constantly higher than those in Af under the photoperiod treatments. The expression levels of DcFT1 in Ws (463.0) were significantly upregulated under LD conditions compared with those in Af (1.4). These responses of DcCOL2, DcCOL5a, and DcFT1 might be related to the different bolting responses of Ws and Af. This study could provide valuable insights into understanding the key integrator genes in the carrot photoperiod pathway.

In the photoperiod pathway, the functional hierarchy GIGANTEA (GI)-CONSTANS (CO)-FT has been identified, and plays a key role in regulating flowering 7,11,[15][16][17][18] . During the late afternoons of long days (LDs), GI forms a complex with the FLAVIN-BINDING, KELCH REPEAT, F BOX 1 (FKF1) to degrade CYCLING DOF FACTORs (CDFs), which contains the major regulators of CO transcription, and consequently, CO protein accumulates to promote flowering by activating the transcription of FT and TWIN SISTER OF FT (TSF) in the leaf vasculature 5,7,8,[19][20][21] . The degradation complex cannot be formed during short days (SDs) due to GI and FKF1 expression not coinciding 20,22 . In rice, HEADING DATE 1 (Hd1), which is the homologue of CO, promotes flowering by activating the expression of the FT-like gene HEADING DATE 3a (Hd3a) on SDs, and represses flowering on LDs 12,15,23,24 . By integrating light and circadian clock signals to regulate the downstream florigen gene, CO plays a central role in the mechanism of photoperiod flowering in Arabidopsis 3,11,18 . Extensive gene duplication events have occurred in this gene family, resulting in 16 other CO-like (COL) genes with different functions 25,26 . Among them, COL1 and COL2 have little effect on the flowering time, while COL3, COL4, and COL9 represent flowering repressors [27][28][29] . The overexpression of COL8 delays flowering under LDs, whereas COL5 promotes flowering 30,31 . Multiple CO/COL homologues have been identified in different species: 16 in rice, 9 in barley, 13 in sugar beet, and 28 in soybean [32][33][34][35] . GmCOL5 can rescue the late-flowering phenotype of the co mutant 33 . OsCOL13 negatively regulates flowering under LD and SD conditions in rice 36 . There is little evidence to prove that CO homologues are potential regulators of FT-like genes 12 .
The flowering habits of many domesticated crops have changed greatly following human selection, which provides an important route for studying flowering mechanisms by comparing cultivars with their ancestors and wild relatives [37][38][39][40] . Carrot (Daucus carota L.) is a biennial plant requiring vernalization to induce flowering, and the seedling is usually not responsive to a low temperature (between 0 and 10 °C) until it has 8-12 leaves 41 . Moreover, long days can promote premature bolting in carrot 42 . With a change of cultivation system, carrots can be harvested and abundantly supplied all year round in markets. The premature bolting of carrot has been a serious risk during the winter-spring period or spring cultivation. The seedlings grow under a low temperature for a long time at the early stage and grow under long days at the late stage, this is why some varieties which are not tolerant to bolting are prone to bolting. Selection tolerance to premature bolting has been a constant concern in carrot breeding [43][44][45][46] . The wild species D. carota subsp. carota Songzi (Ws) is sensitive to flower induction by vernalization and photoperiod, and orange cultivar D. carota var. sativus Amsterdam forcing (Af) is tolerant 42 . Through a de novo transcriptome comparison of Ws and Af, a basic genetic network controlling the flowering time was constructed, including photoperiod, vernalization, and gibberellin pathways 42 . Furthermore, an FT-like gene was identified in carrot 47 . There is substantial evidence that FT homologues have a conserved role in promoting flowering and reflect variation in the copy number in divergent angiosperms 12 . Despite carrot being one of the 10 most important vegetables, understanding on the molecular mechanisms of COL and FT homologues underlying the photoperiodic flowering response is limited. In this study, new homologues of COL, FT, and SOC1 were retrieved based on the carrot genome 48 . The structure, phylogenetic relationship, and molecular evolutionary rate variation of GI, COL, FT, and SOC1 homologues were investigated using 21 D. carota var. sativus accessions (as DCS), 4 D. carota subsp. gummifer species and 9 D. carota subsp. carota species (as DCC), and 5 Daucus species (as Dau) (Supplementary Table S1). The circadian rhythms of GI, COL, FT, and SOC1 homologues under different photoperiods and their expression trends during growth in different seasons were analyzed in Ws and Af. DcCOL2 and DcCOL5a were transformed into Arabidopsis thaliana plants to analyze their function in regulating flowering. This study provides an improved understanding of the regulatory network of photoperiodic flowering in carrots.

Results
phylogenetic and nucleotide diversity of DcGI, DcCOLs, DcFTs, and DcSOC1s. We previously reported a preliminary study of the photoperiod pathway in carrot 42 . Only one DcGI sequence was retrieved from the carrot genome database and localized at chromosome (Chr.) 1, like Dct293 (Supplementary Table S2). The nucleotide diversity (π) of DcGI in 21 DCS accessions and 13 DCC accessions was significantly lower than that in 5 Dau accessions (Fig. 1, Supplementary Table S1). DcGI endured selective pressures during evolution, according to the neutral tests of Tajima's D (TD) 49 and Fu and Li's F (FF) 50 . Protein sequence alignment was performed to further explore their relationships with 27 GI homologues from other species. The tree was divided into four groups, and DcGI was clustered within group I and closely associated with ChGIL, SiGIL, and EgGIL ( Fig. 2A).
Although 28 CO/COL sequences were retrieved, only 16 sequences had complete zinc finger B-box 1 and CCT domains 25 (Supplementary Table S2). The size of the DcCO sequence was only 297 bp, without complete B-box 1 and CCT domains. DcCOL2, like Dct43207, had two tandem duplications and was localized at Chr. 2. DcCOL4, like Dct43377, was localized at Chr. 1. DcCOL5 had two homologues and was localized at Chrs. 3 and 6, like Dct7859 and Dct20940, respectively. DcCOL13 also had two homologues and was localized at Chrs. 1 and 3, but only DcCOL13b was found in the transcriptome, like Dct3283. DcCOL15, like Dct39974, was localized at Chr. 7. Moreover, 1 DcCOL9, 3 DcCOL10, and 4 DcCOL16 homologues were retrieved, which had not previously been reported in the transcriptome 42 . DcCOL9 was localized at Chr. 3. Three DcCOL10 homologues with 83.05% identity were localized at Chrs. 4, 5, and 6, respectively. Four DcCOL16 homologues with 55.61% identity had only B-box 1 and CCT domains and were localized at Chrs. 3, 4, 7, and 9, respectively. Except for DcCOL13b, the nucleotide diversity (π) of the other DcCOLs in DCS and DCC was lower than that in Dau (Fig. 1). According to the neutrality tests of TD and FF, DcCOL10a/c, DcCOL13a, and DcCOL16a/d had significantly negative values, which indicated that these genes might be selected with a higher pressure during evolution. Additionally, it was interesting that DcCOL5b, DcCOL11, and DcCOL15 in DCS had significantly positive values, which indicated that these genes might be strongly selected during carrot evolution. Based on the phylogenetic analysis, 15 DcCOLs were divided into three groups (Fig. 2B). DcCOL2, DcCOL4, and DcCOL5a/b were clustered into group I, which were grouped with AtCO/AtCOL1/2, AtCOL3/4, and AtCOL5, respectively. DcCOL16a/b/c/d were clustered into group II and were closely associated with AtCOL6 and AtCOL16. DcCOL9, DcCOL10a/b/c, DcCOL13a/b, and DcCOL15 were assigned to group III and were closely associated with AtCOL9/10, AtCOL13, and AtCOL14/15, respectively.
Four DcFT homologues were retrieved and annotated as Hd3a 48 , but only two sequences had the complete phosphatidylethanolamine-binding protein (PEBP) domain and were localized at Chrs. 1 and 7, respectively (Supplementary Table S2). Additionally, DcFT1 had a 23.66 kbp large intron by compared its cDNA sequence with carrot genome 48 (Fig. 3B), and DcFT2 was the same as DcFT KY768910 (GenBank number) 47 . The nucleotide diversity (π) of DcFT2 in DCS was significantly lower than that in DCC and Dau, and it might be selected during evolution, according to the neutrality tests of TD and FF (Fig. 1). Twenty-eight FT/Hd3a protein sequences of other species were aligned to explore these gene relationships. The phylogenetic tree was divided into four groups. DcFT1 was assigned to group I and associated with EuHd3a, LsFT, and DlFT1, while DcFT2 was assigned to group II and was closely associated with VvFT, InFTL, and NtFTL (Fig. 2C).
Six DcSOC1 homologues were retrieved, but only four sequences had the complete MADS-box domain, K-box domain, and SOC1/MOTIF (Supplementary Table S2). DcSOC1a has not previously been reported and was localized at Chr. 2. DcSOC1b, like Dct4069, had two tandem duplications and was localized at Chr. 2. DcSOC1c was localized at Chr. 4, like Dct34200. For DcSOC1a, the nucleotide diversity (π) of DCS was significantly lower than that of DCC and Dau, while similar to that of DcSOC1b/c (Fig. 1). Twenty-two SOC1 protein sequences of other species were aligned to explore the relationships. The phylogenetic tree was divided into four groups. DcSOC1a/b/c were assigned to group III and were associated with HaSOC1L (Fig. 2D). circadian rhythm of DcGI, DcCOLs, DcFT1, and DcSOC1s under different photoperiods. Based on the sequences and phylogenetic analysis, DcGI, DcCOL2, DcCOL4, DcCOL5a/b, DcCOL13b, DcCOL15, DcFT1, and DcSOC1b/c were screened for a further study of the circadian rhythm. When the seedlings of Ws and Af had 3-4 leaves, the plants were subjected to SD and LD treatments for 27 days, respectively, and half of the seedlings under SD conditions were then subjected to the above LD conditions for 7 days (as SD-LD). Under LD conditions, Ws plants began to bolt for 67 days after germinating, and the bolting ratio reached 100% after 108 days. Under SD-LD and SD conditions, the bolting of Ws was delayed for approximately 5 and 22 days, and the bolting was 81% and 40% after 108 days, respectively. No bolting plant was found in Af under the above treatments (Fig. 3A). Except that of DcCOL13b, which was only expressed in Ws, the other gene open reading frame (ORF) regions were separately cloned from both Ws and Af (Fig. 3B).
The circadian patterns of DcGI expression in Ws and Af were similar to those described for AtGI 51 and OsGI 15 (Fig. 3C). Under SD conditions, the expression of DcGI began to increase from zeitgeber time (ZT) 0 to 8 and reached trough levels 4 h later, and its peak value was similar for Ws and Af (323.1/334.8). Under SD-LD conditions, the expression of DcGI peaked at ZT 12 and reached trough levels 8 h later, while the peak value (322.3) of DcGI in Ws was higher than that in Af (237.6). The DcGI expression pattern under LD conditions was similar to that under SD-LD conditions, and its peak value was similar for Ws and Af (256.7/281.8).   3,13,14 . In order to further understand the functions of DcGI, DcCOLs, DcFT1, and DcSOC1s during carrot growth, their expression levels were analyzed in Ws and Af in spring and autumn. The seeds were sown directly in the field under natural photoperiods and temperature conditions on 19 March (spring) and 4 August (autumn), respectively. In spring, Ws plants began to bolt for about 65 days after germinating and the bolting ratio rapidly reached 80.3% after 79 days after germinating, while in autumn, the bolting plants initiated 39 days after germinating and peaked at 93.3% after 74 days. No bolting plants were observed for Af during these two seasons (Fig. 4A).
The expression level of DcGI in Ws constantly increased, but later decreased in Af during spring, and fluctuated during autumn (Fig. 4B). DcCOL2 in Af maintained a higher expression level than that in Ws at the beginning and fluctuated during spring, while it maintained a constantly higher level in Af than that in Ws during autumn. The expression of DcCOL4 in Ws remained at low levels, but was high in Af during spring, while it fluctuated during autumn. The expression of DcCOL5a in Ws maintained a higher level than that in Af during both spring and autumn, but that of DcCOL5b was similar.  www.nature.com/scientificreports www.nature.com/scientificreports/ two seasons. DcSOC1b in Ws was expressed at a significantly higher level than that in Af during the two seasons, but DcSOC1c showed a similar expression level.  www.nature.com/scientificreports www.nature.com/scientificreports/ transformed into wild-type Columbia (WT) Arabidopsis plants ( Supplementary Fig. S1A). In the 14 35 S::DcCOL2 plants (T 1 ), 13 were detected in the DcCOL2 cDNA products, but one was not detected (Supplementary Fig. S1B). All 12 35 S::DcCOL5a plants (T 1 ) were detected in the DcCOL5 cDNA products. The bolting time of 35S::DcCOL2 plants (T 2 ) was significantly delayed by about 18.1 days, but the rosette leaf number was similar to that of the WT (Fig. 5A,B). Conversely, the bolting time of 35 S::DcCOL5a plants (T 2 ) was slightly accelerated by 3.0 days and the rosette leaf number decreased by about 1.8.

Discussion
For cultivated carrot, it takes almost a year for it to complete its lifecycle. Many endogenous and external factors can affect carrot bolting and flowering during growth [41][42][43] . Plants with 8-12 leaves are required to respond to a low temperature in carrot 41 , but some landraces and cultivars can initiate bolting after a short vernalization period 52,53 . During the winter-spring period or spring cultivation, the premature bolting of carrot might be interconnected by vernalization and photoperiod pathways 3,[12][13][14] . A basic and putative flowering time network was suggested in carrot 42 , but knowledge of its photoperiodic regulation is limited. In order to further understand the molecular mechanisms of the photoperiod pathway, the establishment of a set of carrot materials with a sensitivity to different photoperiods is required. In this study, Ws plants began to bolt 67 days after germinating and reached 100% after 108 days under LD conditions (Fig. 3A). In spring, under natural conditions, Ws plants similarly began to bolt about 65 days after germinating, but the bolting ratio increased rapidly (Fig. 4A). Ws plants initiated bolting 39 days after germinating and rapidly reached 93.3% after 74 days in autumn (Fig. 4A), but there was no low temperature recorded during the growth. In addition, the bolting time of Ws could be delayed for 22 days and its bolting ratio was significantly reduced to 40% about 108 days after germinating (Fig. 3A), which was similar to a previous report showing that the floral initiation of carrot can be inhibited by SDs 42 . Meanwhile, the orange cultivar Af showed no bolting phenomenon, with tolerance of flowering characterization under different photoperiods (Figs. 3A and 4A). This result suggests that Ws and Af can be regarded as suitable materials for further studying photoperiodic flowering regulations. www.nature.com/scientificreports www.nature.com/scientificreports/ According to the carrot genome 48 , previously reported de novo transcriptome genes 42 could be localized to specific chromosomes and some new photoperiod genes were identified in this study, including 9 DcCOLs, 1 DcFT, and 1 DcSOC1, but there is no complete structure of CO 6,11,[15][16][17][18] (Supplementary Table S2). GI plays a general role in controlling circadian rhythms for flowering and is highly conserved in seed plants 16,19,54 . One DcGI was identified in Daucus species and was strongly selected during evolution (Fig. 1). The circadian pattern of DcGI is similar to that described for AtGI 19,54 and OsGI 15 , and its peak could be delayed by a long photoperiod (Fig. 3C). However, the response for Af and Ws was different that the peak value was quickly reduced in Af under SD-LD conditions, but was maintained in Ws, which was consistent with the trend of expression in spring (Fig. 4B). According to the bolting ratio and time of Ws (Figs. 3A and 4A), DcGI might promote flowering in the carrot photoperiodic pathway.
CO/COL plays a central role in regulating the downstream florigen gene by integrating light and circadian clock signals 3,11,18,25,26 , and multiple homologues have been identified in many species [32][33][34][35] . Only some CO homologues in other species can be regarded as potential regulators of FT-like genes 12 . In this study, 15 DcCOLs were identified with complete B-box 1 and CCT domains in carrot (Supplementary Table S2, Fig. 2B). According to the nucleotide diversity (π) and neutrality tests, 8 DcCOLs might be selected during evolution, especially DcCOL5b, DcCOL11, and DcCOL15 in cultivars (Fig. 1). DcCOL2 showed a similar morning-phase rhythm in Ws and Af, which was comparable to AtCOL2 28 and ClCOL2 55 . In Ws, the peak of DcCOL2 was advanced and higher under SD conditions than under SD-LD and LD conditions (Fig. 3C), but the bolting time was significantly delayed and the ratio was reduced under SD conditions (Fig. 3A). In Af, the patterns of DcCOL2 were similar and the peak value was higher than that in Ws (Fig. 3C). In the two seasons, the expression of DcCOL2 in Af showed higher levels than in Ws, but no bolting plants were found in Af (Fig. 4A). The bolting time of 35 S::DcCOL2 Arabidopsis plants (T 2 ) was also significantly delayed, but its rosette leaf number was similar to that of the WT (Fig. 5). DcCOL4 also had a morning expression pattern and peaked at dawn under SDs, similar to DcCOL2. It was strongly repressed under a long photoperiod in Ws, but it could maintain a constantly high level in Af (Fig. 3C), even during spring (Fig. 4B). A recent study showed that an increase of COL4 expression causes flowering delay in Arabidopsis 29 , but COL2 has little effect on the flowering time 28 . Maintaining a higher expression of DcCOL2 and DcCOL4 might be related to the bolting tolerance in Af, but this claim needs more supporting evidence.
Two DcCOL5 homologues were identified and their circadian rhythms under SDs were similar to that of ClCOL5 55 , but their responses to a long photoperiod were different (Fig. 3C). It was obvious that the peak of DcCOL5a was postponed and maintained a similar level under a long photoperiod, but that of DcCOL5b was advanced and reduced in Ws. In Af, the expression of DcCOL5a was downregulated, but there was no DcCOL5b response under SD-LD and LD conditions. In the two seasons, the expression level of DcCOL5a in Ws was constantly higher than that in Af, but the expression levels of DcCOL5b fluctuated (Fig. 4B). The bolting time of 35 S::DcCOL5a Arabidopsis plants (T 2 ) was only slightly accelerated (Fig. 5). DcCOL13b was only expressed in Ws and maintained similar expression levels under different photoperiods (Fig. 3C), but its circadian rhythm presented contrasting patterns under SD and SD-LD/LD conditions, similar to that of OsCOL13 36 . Whether DcCOL13b represses flowering or not requires further investigation.
FT is the major primary target of CO in leaves 4,6,7,12 . In the carrot genome, there are two FT homologues localized at Chrs. 1 and 7, respectively 48 (Supplementary Table S2). According to the phylogenetic analysis, DcFT1 was assigned to group I and associated with EuHd3a, LsFT, and DlFT1, while DcFT2 was assigned to group II and closely associated with VvFT, InFTL, and NtFTL (Fig. 2C). It is interesting that DcFT1 had a 23.66 kbp large intron by compared its cDNA sequence with the carrot genome 48 (Fig. 3B). The circadian rhythm of DcFT1 in Ws was similar to that of AtFT and its expression levels were significantly upregulated under SD-LD/LD conditions, but constantly low in Af 4,7 (Fig. 3C). The expression patterns of DcFT1 in Ws and Af were different, but the levels were similar during the two seasons (Fig. 4B). The expanded studies show that FT homologues have a conserved function in promoting flowering 12 . Different responses of DcFT1 in Ws and Af might be related to the different bolting response, which could give a clue for further studying its function in flowering regulation.
SOC1 integrates multiple flowering signals to interact with multiple MADS-box proteins and regulate the expresssion of flowering genes 8,9,21,56,57 . Three DcSOC1s with complete structures were identified and found to be highly conserved during evolution (Fig. 1). The circadian rhythm patterns of DcSOC1b/c were different under the three photoperiod treatments (Fig. 3C). The expression of DcSOC1b in Ws maintained significantly higher levels than those in Af, while DcSOC1c showed similar levels. The same results were observed in the two studied seasons (Fig. 4B). These results suggest that DcSOC1b might play a role in regulating downstream flowering genes, like AtSOC1 9,20 and ZmSOC1 57 .
In general, Ws plants could initiate bolting 39 days after germinating without low-temperature treatment, which suggests that it can be regarded as a suitable material for studying photoperiodic flowering regulation. Based on previous studies, 1 DcGI, 15 DcCOLs, 2 DcFTs, and 3 DcSOC1s were identified in the photoperiod pathway. The circadian rhythm peaks of DcGI, DcCOL2, DcCOL5a, and DcCOL13b could be delayed under LD conditions. The peak value of DcCOL2 in Af was significantly higher than that in Ws under SD conditions, and their values could be reduced under SD-LD/LD conditions. Under the photoperiod treatments, the peak values of DcCOL5a in Ws were constantly higher than those in Af, even during the two seasons. The expression levels of DcFT1 in Ws were significantly upregulated under SD-LD/LD conditions compared with those in Af. These responses of DcCOL2, DcCOL5a, and DcFT1 might be related to the different bolting responses of Ws and Af. This study could provide valuable information that increases our understanding of the key integrator genes in the carrot photoperiod pathway. Future work focusing on characterizing the function of these candidate genes will be helpful for screening those accessions with tolerance to premature bolting.

Methods
Retrieval and phylogenetic analysis of DcGI, DcCOL, DcFT, and DcSOC1 sequences in carrot.
GI, COL, FT, and SOC1 are the key transcript regulators in the photoperiodic pathway 4,6,7,11,16,17 . Based on the carrot genome 48 , the sequences of 1 DcGI, 28 DcCOL, 4 DcFT, and 6 DcSOC1 homologues were retrieved and aligned with transcriptome unigenes 42 using DNAMAN software (Supplementary Table S1). Twenty-seven GI, 22 CO/COL, 28 FT/Hd3a, and 22 SOC1 homologue protein sequences of other species were retrieved from the NCBI database and used to perform a phylogenetic analysis with the abovementioned genes through MEGA5.0 using the neighbor-joining method, respectively. The phylogenetic tree architecture was validated using the probabilistic bootstrap test with 1000 replicates. nucleotide diversity and selection of DcGI, DcCOLs, DcFTs, and DcSOC1s in Daucus species.
Twenty-one D. carota var. sativus accessions (as DCS), 4 D. carota subsp. gummifer species and 9 D. carota subsp. carota species (as DCC), and 5 Daucus species (as Dau) were used to perform the evolutionary analysis of GI, COL, FT, and SOC1 homologues (Supplementary Table S1). The genome of cultivar Af, breeding lines 17166 C, 170P2A, 17P25A, and wild species Ws were re-sequenced with each leaf sample in the Illumina HiSeq. 2500 sequencing platform (Biomarker Technologies Co., Ltd. Beijing, China). The sequence data of 34 accessions were downloaded from NCBI (https://www.ncbi.nlm.nih.gov/genome/?term=carrot) 48 . All genome sequence reads from 39 individual accessions were trimmed and filtered prior to analysis, and aligned to the carrot genome 48 using BWA software 58 . Alignment files were converted into SAM/BAM files and simple nucleotide polymorphism (SNP) calling was performed using SAMtools 58 . Low-quality SNPs with a base quality value of <20 and a read depth of <4× or those with >32× coverage from the sequences were excluded because these SNPs may be false positives. The FASTA sequences of 39 individuals were extracted using BCFtools 59 . Nucleotide diversity (π) for the genomic sequence of each data group and neutral tests of TD 49 and FF 50 were estimated based on the neutral model prediction by DnaSP 6 60 .
Gene cloning and circadian rhythm analysis of DcGI, DcCOLs, DcFT1, and DcSOC1s. For the circadian rhythm analysis, Ws and Af seeds were sown in flats (90 cm × 30 cm × 30 cm) filled with a 1:2:3 mixture of soil, vermiculite, and turf in the greenhouse at a temperature of 13-16/25-28 °C night/day in late February. The average maximum light intensity per day in the greenhouse (55.4 ± 3.7 W • m −2 ) was only 30.6% of that in the field (180.9 ± 14.4 W • m −2 ). When the seedlings had 3-4 leaves and 37 days after germinating, the plants were moved to the field under a natural temperature and subjected to SD (8 h light/16 h dark cycle) and LD (16 h light/8 h dark cycle, supplemented with white fluorescent light of 30 μmol/m 2 /s) treatments for 27 days. Then, half of the seedlings that had 6-7 leaves under SD treatment were subjected to the above LD conditions for 7 days (as SD-LD), and half remained under SD conditions. During the treatment, there were about 12 days with a low temperature (below 10 °C) 41 . The leaves of each treatment were sampled from five plants with three biological replicates at 4 h intervals after dawn at zeitgeber time 0 (ZT 0). All samples were immediately frozen in liquid nitrogen and stored at −80 °C until use. After this, all plants were grown under natural conditions, and the number of bolting plants was counted every 2-4 days until 108 days after germinating.
The candidate genes of DcGI, DcCOL2, DcCOL4, DcCOL5a/b, DcCOL13b, DcCOL15, DcFT1, and DcSOC1b/c were selected to study the circadian rhythm under different photoperiod treatments. The specific primers for the open reading frame (ORF) sequence clone and real-time qPCR were designed using Primer Premier 5.0 based on the abovementioned gene sequences (Supplementary Table S3). The operation procedures for RNA extraction, cDNA synthesis, reverse transcription PCR, and real-time qPCR were carried out as described by Wang et al. 61 . Tublin was amplified along with the target genes as an endogenous control to normalize the expression levels between samples 62 . Three biological repetitions were performed for each expression data point. The comparative C T (2 −ΔΔCt ) method was applied for calculation 63 . The lowest expression level of each gene was used for calibration. Values are reported as the mean ± SE from the replicates.
Expression analysis of DcGI, DcCOLs, DcFT1, and DcSOC1s in spring and autumn. The expression levels of DcGI, DcCOLs, DcFT1, and DcSOC1s were further analyzed to understand their functions during carrot growth under natural conditions. The seeds of Ws and Af were sown with 20 cm row spacing in the field under natural photoperiods and temperature conditions on 19 March (spring) and 4 August (autumn), respectively. The cultivation and management of carrot referred to that of Rubatzky et al. 45 . There were about 15 days with a low temperature (below 10 °C) after the seeds had germinated in spring and no low temperature was recorded before the plants had bolted in autumn. When the plants had about five leaves, the leaves from five plants were sampled with three biological replicates from 08:00 to 09:00 at 55, 63, 71, and 79 days after germinating in spring, and at 32, 39, 46, and 62 days after germinating in autumn. The number of bolting plants was also investigated. Sample collection, total RNA extraction, cDNA synthesis, gene expression, and data analysis were performed as described by Wang et al. 61 .
Flowering time of transgenic Arabidopsis with DcCOL2 and DcCOL5a. To further understand the functions of DcCOL2 and DcCOL5a, their full-length cDNAs were amplified using gene-specific primers and subcloned into the vector PBI121 plasmid and agrobacterium GV3101 (Tiangen, China) to generate 35 S::DcCOL2 and 35 S::DcCOL5a, which were introduced into the wild-type Columbia (WT) Arabidopsis by the floral dip method 64 . Transgenic plants were screened on Murashige and Skoog (MS) agar plates supplemented with 50 mg/L kanamycin. After being left in darkness at 4 °C for 3 days, kanamycin-resistant seedlings were transferred to soil after 10 days and grown in a controlled growth room (~22 °C, 16 h light/8 h dark). The seeds from T 0 plants were