The role of BoFLC2 in cauliflower (Brassica oleracea var. botrytis L.) reproductive development

Genetic, association, and expression studies indicate that a mutation in BoFLC2 explains much of the variability in cauliflower curd formation, curd growth rate, and flowering time.


Introduction
The transition from vegetative to reproductive growth is a critical phase in the life cycle of flowering plants, especially in monocarpic species, whose life cycle only allows a single opportunity to sexually reproduce. Timing of the transition to flowering depends on integration of endogenous factors such as leaf number and gibberellin (GA) biosynthesis (Bernier and Perilleux, 2005;Mutasa-Göttgens and Hedden, 2009) and environmental cues such as photoperiod, light quality, temperature, and stress (Cerdán and Chory, 2003;Yaish et al., 2011). The biological function of endogenous requirements is to prevent the plant from flowering before it has accumulated the necessary reserves required for energy-intensive reproductive development (Matsoukas et al., 2012). Environmental cues serve to synchronize the flowering of conspecific, cross-pollinating plants, and also ensure that flowering occurs when conditions are conducive to fruit set and seed production (Salomé et al., 2011). Although these flowering time regulatory mechanisms have developed to ensure reproductive success in nature, it is often necessary to overcome or otherwise manage them in agricultural systems. For example, cultural practices are often focussed on preventing or delaying flowering in crops grown for vegetative organs. Management of flowering time is also essential where crops are produced for reproductive structures to ensure development and maturity of these organs occurs during favourable environmental conditions. Ensuring synchronous flowering between parent lines with markedly different flowering phenology in the production of hybrid vegetable seed is a third classic example of the need to understand and modify the timing of natural flowering behaviour in agricultural crops.
Molecular technology presents an opportunity to understand and manage reproductive development in such production environments. Prospects for doing so have benefited greatly from recent work with the model species Arabidopsis thaliana. Over the last decade, genetic analysis in Arabidopsis has identified a large number of flowering genes, enabling the development of a model describing the genetic and biochemical pathways that regulate floral induction through both endogenous signals and environmental cues such as vernalization, temperature, and photoperiod (Andrés and Coupland, 2012;Kumar et al., 2012). Many of the genes and genetic pathways involved in Arabidopsis flowering have been conserved to some degree in other plant species. Consequently, there exists great potential to extend the molecular model proposed for the control of flowering in Arabidopsis to species of economic importance such as the closely related B. oleracea, which includes crops such as cauliflower (var. botrytis). Temperature is the main environmental factor influencing reproductive behaviour in this species, with vernalization requirements ranging from near-obligate (e.g. cool-temperate spring cultivars) to virtually none (e.g. tropical cultivars) (Friend, 1985). In vernalization-responsive cultivars, temperature affects important processes such as induction of the edible "curd" and subsequent elongation of inflorescence stems. A better understanding of the molecular and genetic factors regulating vernalization-induced flowering would potentially enable the development of strategies for predicting and managing flowering in a production context, and would also assist breeders seeking to extend the geographical range of B. oleracea cultivation.
An early attempt to extend knowledge of Arabidopsis regulatory networks to Brassica crop species involved the identification of flowering time QTLs in the B. napus and B. rapa chromosomal regions collinear to the chromosomal position of Arabidopsis FLOWERING LOCUS C (FLC) (Osborn et al., 1997). In Arabidopsis, this gene plays a key regulatory role in the vernalization pathway, encoding a MADS-box protein that binds directly to a region of DNA in the first intron and promoter of FLOWERING TIME (FT) and SUPPRESSOR OF OVEREXPRESSION OF CO 1 (SOC1), respectively (Michaels and Amasino, 1999;Sheldon et al., 1999;Helliwell et al., 2006). This prevents transcription of the FT and SOC1 integrator genes, thereby inhibiting flowering. At vernalizing temperatures, genes including VERNALIZATION INSENSITIVE 3 (VIN3), VERNALIZATION1 (VRN1), and VRN2 act together to physically repress FLC mRNA transcription and translation, thereby allowing expression of FT and SOC1 (Sung and Amasino, 2004a, b;He and Amasino, 2005).
Subsequent comparative studies in Brassica crop species have enabled the isolation of multiple copies of Arabidopsis FLC homologues in several Brassica crop species including five in B. napus and four in B. rapa and B. oleracea (Tadege et al., 2001;Schranz et al., 2002;Okazaki et al., 2007). In B. napus, all five copies have been characterized, with both expression studies and complementation studies in transgenic Arabidopsis supporting their conserved function. Genetic and transcriptional studies in B. rapa have provided evidence that, of the FLC genes, BrFLC2 plays the most important role in the regulation of flowering time (Li et al., 2005;Kim et al., 2007;Zhao et al., 2010;Xiao et al., 2013).
There has been comparatively little characterization of FLC genes in B. oleracea. A study by Okazaki et al. (2007) involving biennial, annual, and rapid-cycling forms of B. oleracea found that a frameshift mutation in exon 4 resulting in an in-frame stop codon of BoFLC2 was associated with early flowering behaviour. An earlier study by Lin et al. (2005) showed that cabbage (B. oleracea var. capitata) BoFLC2 transcription (referred to in that paper as BoFLC4-1) was regulated by vernalization in a conserved manner, and also isolated a cabbage homologue of the important flowering time integrator gene FT and found that expression of this transcript in cabbage apices was inversely related to BoFLC2 expression.
These findings suggest that BoFLC2 functionality may underpin important differences between annual and biennial brassica cultivars, particularly with respect to flowering time. However, no research has been conducted to demonstrate the contribution of this gene to flowering time within either annual or biennial brassica types, both of which have broad flowering time ranges across cultivars. In this study, we evaluated the contribution of BoFLC2 allelic variation to curd initiation and flowering time variation in both segregating populations and inbred populations of annual cauliflower (B. oleracea var. botrytis) parent lines. Differences in vernalization response and flowering time were investigated at the molecular level by monitoring the transcriptional dynamics of key genes in response to a range of artificial and natural vernalization treatments. Finally, seasonal changes in gene expression patterns were examined in cauliflowers grown outdoors under natural vernalizing conditions.

Plant material
Fifty four homozygous inbred cauliflower proprietary parent lines were provided by vegetable seed breeding company Rijk Zwaan. These lines had been grouped into five flowering classes by Rijk Zwaan breeders, based on time from planting to full flower under field conditions in the southern Netherlands: early (50-70 days); medium early (70-80 days); medium late (80-120 days); late (120-180 days); very late (180-250 days). Supplementary Table S1 summarises the breeders' information available for these lines.
Two segregating F 2 populations were established by crossing lateflowering cauliflower proprietary parent lines A and C with earlyflowering proprietary parent lines B and D, respectively. Under field conditions present at the Rijk Zwaan Research and Development facility (Fijnaart, the Netherlands), both late-flowering lines had a flowering time of approximately 120 days, and both early-flowering lines flowered after approximately 70 days. Line A had been inbred for two generations (S 2 ), line B had been inbred for four generations (S 4 ), and lines C and D had been inbred for six generations (S 6 ). The F 1 plants arising from these two crosses were manually self-pollinated to generate F 2 seed from the A×B and C×D crosses.

Growth conditions and experimental design
Three of the outdoor experiments were field-based and one was potbased. The field-based flowering time trials were located at Fijnaart (the Netherlands), Oyster Cove (southern Tasmania), and Forthside (northern Tasmania). At each field site, glasshouse-raised seedlings were transplanted into the soil at a spacing of 45-50 cm between plants and 75-80 cm between rows. At Fijnaart, 30 seedlings from each of the A, B, C, and D parent lines were planted into the field in April 2008 along with approximately 380 seedlings from each of the A×B and C×D F 2 populations. At Forthside, a selection of 15 cauliflower parent lines with flowering behaviour ranging from early to very late was sown in 2007. Time of sowing and transplanting was varied according to flowering behaviour in an attempt to achieve uniform curd initiation and flowering between lines; later-flowering lines were sown in mid-summer (late January-early February), and earlier-flowering lines were sown at the end of summer (late February). At Oyster Cove, 50 out of the 54 cauliflower parent lines were transplanted (omitting four lines that failed to germinate) in April 2009. In both Forthside and Oyster Cove trials, three replicates of each parent line were used with approximately ten plants per replicate and both sites were set out in a randomized complete block design.
The pot-based outdoor flowering time trial was conducted at the University of Tasmania's Horticultural Research Centre (Sandy Bay, Tasmania). Fifty seedlings of a late-flowering (BoFLC2) parent line and 50 seedlings of an early-flowering parent line (boflc2) were raised in the glasshouse before being transplanted into 4.5 l pots. To facilitate normal plant growth and development and to synchronise flowering between the two lines, the BoFLC2 line was sown and transplanted earlier than the boflc2. Seeds of the BoFLC2 line were sown in late summer before transplanting outside approximately six weeks later in early autumn. Seeds of the boflc2 line were sown in early autumn, two weeks later than the BoFLC2 line, before transplanting outside approximately seven weeks later in mid-autumn.
In the pot-based, glasshouse flowering time trial, 50 of the 54 cauliflower parent lines were transplanted (omitting four lines that failed to germinate) in April 2009. Seven replicates of each line were transplanted 4.5 l pots. Plants were grown under natural light and photoperiod conditions. Three controlled environment vernalization experiments were conducted in growth cabinets at the School of Plant Science, University of Tasmania. In each case, plants were grown in 2-l pots under a mixture of 36 W cool white fluorescent lamps and 60 W incandescent lamps with an output of 100 µM m -2 s -1 as part of a 16 h photoperiod regime. In the first experiment, seedlings of the romanesco BoFLC2 cauliflower parent line 1 classed as 'late' flowering were grown at 22 °C. At ten weeks of age, a proportion of these plants were transferred to 5 °C growth cabinets. Plants from both groups were sampled for measurement of gene expression every two weeks until 16 weeks of age. In the second experiment, plants from four BoFLC2 parent lines (2, 12, 24, and 51) and four boflc2 parent lines (17, 40, 48, and 50) were grown at 22 °C. At ten weeks of age, a proportion of these plants were transferred to 5 °C, 10 °C, and 15 °C growth cabinets. Gene expression was measured at 3, 6, 10, and 15 weeks of age. In addition, plants vernalized at 5 °C were sampled for gene expression 2, 7, and 21 days after vernalization.
In the third experiment, seeds of the same romanesco variety as above were grown at 22 °C. Plants aged 1-12 weeks were exposed to one-and three-week vernalization treatments at 5 °C.
Phenotyping A semi-quantitative scale of 1-6 was constructed to allow consistent assessment of curd developmental stage across experiments. Stage 1 corresponded to a curd that is visible without damaging excavation amongst apical leaves, and that was compact, with no discernible elongation of peduncles. Stage 2 corresponded to a 'loose' curd that was beginning to 'break'. Stage 3 corresponded to a curd whose early-order inflorescence branches were elongating, but where individual flower buds were not formed, or were rudimentary. Stage 4 curds had fully formed flower buds (similar to the individual buds of a broccoli head). Stage 5 corresponded to a curd that had one or more flowers fully open, but less than twelve flowers fully open. Stage 6 was the highest developmental category, with curds that had more than 12 flowers open, and the possible presence of seed pods. Images of these six stages are provided in Supplementary Fig. S1.

DNA extraction and genotyping
In the segregating F 2 population, DNA was isolated from 8 mm young leaf discs using a mag Plant DNA Isolation Kit (AGOWA) in conjunction with a KingFisher magnetic particle processor (Thermo Scientific) according to manufacturer's specifications. For all other experiments, DNA was extracted using a modified CTAB protocol. CAPS and size markers were developed to genotype BoFLC2 and BoFLC4-1. Details of these markers are shown in Supplementary  Table S2.

Gene expression
Harvested tissue consisted of young, non-fully expanded leaves and apices. Plants at a physiological age of one week did not possess any true leaves, so pooled tissue samples from multiple comprising cotyledon/apex tissue were harvested instead, as with Lin et al. (2005). Samples were frozen in liquid nitrogen and total RNA extracted using the SV total RNA Isolation System (Promega) according to manufacturer's instructions. Reverse transcription was performed in 20 µl with 1 µg of total RNA using the MMLV High Performance Reverse Transcriptase System (Epicentre Biotechnologies) according to manufacturer's instructions. RT negative (no enzyme) controls were included to monitor for gDNA contamination. First-strand cDNA was diluted five times and 2 µl was used in each real-time PCR reaction. Real-time PCR reactions using SYBR green chemistry (Sensimix) were run for 50 cycles in a Rotor-Gene RG3000 (Corbett Research). Two technical replicates and three biological replicates were performed for each sample unless otherwise stated. Transcript levels for experimental genes were normalised against the constitutively-expressed reference gene ACTIN (ACT) using nonequal efficiencies (Pfaffl, 2001). Primer details are included in Supplementary Table S2.

Results and discussion
BoFLC2 is strongly associated with flowering behaviour in annual and biennial Brassica parent lines Cauliflower inbred parent lines representing a diverse range of flowering types were screened for the presence of the single base deletion in exon 4 of BoFLC2 first described by Okazaki et al (2007). Of the 54 parent lines, 33 carried the non-functional allele (boflc2) and 21 carried BoFLC2. The BoFLC2 genotype was strongly associated with delayed flowering as categorized by Rijk Zwaan breeders (Supplementary  Table S1), with a higher proportion of lines carrying functional alleles found in classes defined by later flowering time ( Supplementary Fig. S2), consistent with the role of BoFLC2 as a flowering inhibitor. These findings constitute the first record of functional BoFLC2 alleles in cauliflower varieties, and support the idea that the previously described boflc2 mutation (which is predicted to result in a loss of function owing to a premature termination codon) is a critical determinant of flowering time within the annual brassica subgroup.
Identification of a strong association between BoFLC2 allelic variation and floral induction in 'natural' populations of cauliflower inbred parent lines is potentially of great practical and commercial value. The boflc2 marker could be used for germplasm screening and reliable selection of appropriate cauliflower genotypes in breeding programmes, thereby reducing the need for lengthy field assessments. The number of crosses necessary in doubled haploid (DH) breeding programmes could be reduced by using this marker to help select parent lines with suitable combinations of alleles. In some cases, BoFLC2 could be modified or transformed to alter timing of floral induction without significant negative pleiotropic effects on yield or general fitness. For example, Kim et al. (2007) reported a substantial delay in Chinese cabbage (B. rapa ssp. pekinensis) floral induction by overexpression of FLC. In B. oleracea cultivars where delayed curd induction or flowering is sought, such an approach may find a similar application.

Glasshouse and field trials validate the association of BoFLC2 with reproductive behaviour
In the pot-based glasshouse flowering time trial, curd initiation data for the 50 different parent lines sown was broadly consistent with the classifications provided by Rijk Zwaan pre-breeders (Fig. 1A). Despite considerable differences in genetic background between these parent lines, a clear advance in curd initiation time was observed in the 33 boflc2 lines relative to 17 BoFLC2 lines (108.6 ± 2.1 vs 213.9 ± 9.0 days; P<0.0001). In the initial screen of cauliflower parent lines, the presence of functional BoFLC2 alleles in lines classed as medium early or medium late, and the relatively large number of boflc2 genotypes in the late class seemed somewhat incongruous ( Supplementary Fig. S2). This trial revealed that in most cases, the apparent incongruity in flowering time could be explained by the distinct phenotypes of the lines in question. Four of the earliest curding lines carrying BoFLC2 were of the 'romanesco' variety, with fractally arranged pyramidal curds (Kieffer et al., 1998). Three of the latest boflc2 lines formed small, tightly bunched 'heads' of leaves before detection of a visible curd. This apical morphology made it difficult to routinely distinguish the point at which curds were formed, and complicated the relatively simple BoFLC2 effect observed in field trials.
As with curd initiation, there was a clear separation of BoFLC2 and boflc2 groups based on flowering time with the exception of romanesco and bunchy parent lines (Fig. 1B). BoFLC2 (including romanesco) lines took an average of 14 weeks longer than boflc2 (including 'bunchy') lines to reach full anthesis (flowering stage 6; 286.1 ± 8.0 vs 188.8 ± 1.1 days; P<0.0001). Curd initiation time was usually strongly indicative of flowering time, but some lines that initiated curds at a young age remained in the arrested phase and did not proceed to flowering for a considerable length of time (e.g. line 47). Other parent lines that initiated curds very early were slow to flower (parent line 30) or failed to flower altogether (parent line 37) because of poor growth. Therefore, the distinction in both curd initiation and flowering time between BoFLC2 and boflc2 parent lines is sharp, allowing for the deceptive 'transition' which is primarily due to parent lines with unusual curd initiation behaviour that complicates the BoFLC2 gene effect.
In the Oyster Cove field trial, plant growth was compromised by exceptionally wet and cold conditions leading to poor plant growth and high attrition, particularly among the boflc2 lines that are adapted to warmer climates and lower latitudes. Nevertheless, the average number of days required for curd formation was significantly different (P<0.0001) between the surviving BoFLC2 and boflc2 genotypes (271.8 ± 1.9 and 254.0 ± 1.8 days, respectively), but the difference between the two genotypes was much less clear than in the glasshouse trial. (Fig. 1C). Allowing for the vagaries associated with the production of cauliflowers in environments to which they were unsuited, these trials not only validated the Rijk Zwaan categories, but also enabled identification of the physiological basis for inconsistencies in flowering behaviour between the two genotypes and demonstrated that variation at the BoFLC2 locus is a fundamental determinant of flowering time.
Variation at the BoFLC2 locus contributes to flowering time and curd size in two segregating F 2 populations The correlation of BoFLC2 with flowering time across a diverse range of parent lines suggests that this gene is an important contributor to reproductive behaviour within annual brassicas. To investigate this contribution more precisely, flowering time and curd development were assessed under field conditions in Fijnaart, the Netherlands in populations segregating for functional and mutant BoFLC2 alleles derived from crosses between two members of the same group (cauliflower × cauliflower). To do this, two different late (lines A and C) and early (lines B and D) proprietary parent lines with suitable commercial qualities were selected for crossing. The BoFLC2 genotype of these lines was confirmed, showing that the late and early parents had the expected homozygous BoFLC2 (+/+) and homozygous boflc2 (-/-) genotypes, respectively.
Genotyping of the A×B F 2 population revealed a segregation of 102:175:97 (BoFLC2:het:boflc2) at the BoFLC2 locus. This conforms to the expected theoretical 1:2:1 distribution ratio of 93.5:187:93.5 (df=2; χ 2 crit with α at 0.01=9.21; χ 2 obs =1.67). In the C×D F 2 population, a segregation of 120:186:71 (BoFLC2:het:boflc2) was observed, representing a significant departure from the theoretical expected distribution ratio of 94.25:188.5:94.25 (df=2; χ 2 crit with α at 0.01=9.21; χ 2 obs =12.80), possibly owing to gametic selection for the functional allele. Figure 2A shows that in both the A×B and the C×D segregating F 2 populations, the average length of time from transplanting until curd visibility was significantly different between parent lines and across the three BoFLC2 genotypes (P<0.0001). In the A×B F 2 population, the average length of time from transplanting until curd visibility in +/+, +/-, and -/-plants was 88.2, 84.6, and 80 days, respectively. The differences in curding time observed in the A×B F 2 population are relatively small compared with the differences between parents A and B. This may be partly due to the fact that parent lines A and B were not isogenic, meaning that in addition to BoFLC2, other genes that affect flowering time would also be segregating and influencing the phenotype of the F 2 population, resulting in less extreme phenotypes than in the parent lines. In the C×D F 2 population, the average length of time from transplanting until curd visibility in +/+, +/-, and -/plants was 94.5, 88, and 82.1 days, respectively. In the C×D F 2 population, comparison of developmental stage between genotypic classes was conducted at a fixed point of 117 days after transplanting, with Fig. 2B showing significant (P<0.0001) differences between the three genotypes. Developmental stage was not able to be measured in the A×B population owing to its susceptibility to attack by saprophytic pathogens following curd initiation.
The importance of BoFLC2 to the flowering phenotype in both F2 populations is clear. The gene seems to function in a dosage-dependent manner with curd formation delayed in an additive manner by additional copies of the functional allele. Similarly, as the number of functional alleles increased, the developmental stage at a fixed time point was observed to decline. The incomplete dominance exposed in this study contrasts with other studies that suggested that the annual habit is dominant over the biennial habit (Okazaki et al., 2007). Applying the formula σ 2 phenotype =σ 2 genotype +σ 2 environment (c.f. Supplementary Table S3) to the observed variance under a model of normal distribution would roughly equate to the BoFLC2 gene explaining 40% of total phenotypic variance in flowering time of the C×D F 2 population, and 65% of the genetic variance in flowering time, where flowering time is based on the developmental stage of development 117 days after transplanting, and assuming environmental variance in the F 2 population is the same as the average of the variance in the homozygous parent lines (parental lines were used as environmental controls). It seems likely that the remaining variation may be explained by other flowering time genes. For example, Irwin et al. (2012) suggested that allelic variation at the locus of the brassica orthologue FRIGIDA (BoFRIa) may be responsible for variation in B. oleracea flowering time. A recent QTL study by Uptmoor et al. (2012) using a B. oleracea population derived from a cross between late/early flowering types indicated that flowering time variability was primarily due to differences in vernalization response. However, neither parent used to generate that population carried functional alleles of BoFLC2 and no other BoFLC genes were found to co-segregate with flowering time. The authors raise the possibility that FLC-independent pathways are responsible for the vernalization response seen in their population, and this may also be the case in the populations generated in the study presented here. An attempt was made to map other B. oleracea flowering time orthologues including, BoFLC1, BoFLC3, BoFLC5, BoFT, BoVIN3, SOC1, FRI, BoREM1, CCE1, APETALA1 (AP1), CAULIFLOWER (CAL), and LEAFY (LFY or BoFH), but no polymorphisms suitable for marker development were identified.
Curd size was recorded throughout the observation period in C×D F 2 plants. In the first week of measurement (78 days after planting), the average curd size of -/-plants was 11.1 cm, greater than that of the +/-plants (5.2 cm), which in turn was greater than the average curd size of +/+ plants (2.0 cm; see Fig. 2C). In the -/-genotype, curd size increased steadily before reaching a plateau. The other two genotypes seem as though they would also follow this general trend, but recording ceased before a clear plateau was reached. Despite significant differences (P<0.0001) in average maximum observed curd size (-/-=26.7 cm; +/-=25.0 cm; +/+=22.1cm) it seems that all three genotypes were tracking to reach a roughly similar maximum size, and by fitting quadratic trend lines to the data, similar maximum average curd diameters are predicted for each genotype (24.9 cm, 23.1 cm, and 25.9 cm for -/-, +/-, +/+, respectively). This, along with significant variability in curd size at the point of bolting within each genotype, suggests that BoFLC2 has little bearing on final curd size. However, whereas the homozygous mutant boflc2 lines would be predicted to reach this maximum size after 108 days, the heterozygotes would require a further 6 days to reach this predicted size, and the homozygous functional lines would require yet another 26 days. Differences in curd initiation date were taken into account through the use of a separate slopes model. A comparison of the slope between average Y values at the first two time points in the -/-line and the average Y values at the first three points in the +/-and +/+ lines (Fig. 2C) was made. This revealed that the slopes (indicating rate of curd growth) of the three genotypes were significantly different (P=0.0427). The rate of curd growth was slower in lines with more functional copies of BoFLC2.

Vernalization regulates BoFLC2 expression in parallel with curd induction
In view of the important role of BoFLC2 in vernalizationmediated flowering, we next examined how BoFLC2 is regulated by chilling and how this might be related to curd induction behaviour. Three sets of controlled-environment experiments were conducted to examine physiological and molecular responses to vernalization. In the first, a late-flowering BoFLC2 parent line (line 1) was subjected to vernalization when plants were ten weeks old with gene expression measured two, four, and six weeks after vernalization (Fig. 3A). In the second, ten-week-old BoFLC2 and boflc2 parent lines were exposed to vernalization at a range of temperatures (5 °C, 10 °C, and 15°C), and expression was measured 2, 7, 21, and 35 days after vernalization (Fig. 3B). In the third, parent line 1 plants aged 1-12 weeks were exposed to 1-and 3-week 5 °C vernalization treatments (Fig. 3C).
In the first experiment, a direct relationship between vernalization duration and curd initiation/development was observed, with longer periods of vernalization associated with more rapid curding and curd development upon return to glasshouse conditions. In the second experiment using BoFLC2 and boflc2 parents, numerous instances of reversion from floral to vegetative growth were recorded when vernalized plants were subsequently returned to 22 °C cabinets. Reproductive behaviour and curd size in segregating F 2 populations. The average number of days from transplanting until curds were visible in parent lines and F 2 populations is shown in (A). Plants that had not produced visible curds after 117 days were given a maximum score of 127 days to differentiate them from those plants for which curds were first recorded on the 117th day. In (B), the average developmental stage (c.f. Supplementary Fig. S1) of curds at a fixed point (117 days after transplanting) is shown for C and D parents and F 2 segregants. Lower developmental stage values represent less-developed curds. In (C) average curd sizes in the C×D segregating F 2 population are displayed. In all graphs, +/+ denotes plants homozygous for BoFLC2, -/-denotes plants homozygous for boflc2, and +/-denotes heterozygotes. Error bars represent the standard error of the mean.
Given the capacity of these post-vernalization conditions to induce such floral reversion, it is also possible that devernalization may have significantly delayed curd induction in some cases. Despite this complication, clear differences in reproductive behaviour were observed between BoFLC2 and boflc2 genotypes ( Supplementary Fig. S3). Plants carrying the boflc2 allele were capable of producing curds without any chilling; and yet, a vernalization response was evident, with all vernalization durations increasing the percentage of plants with reproductive structures. Curd induction in BoFLC2 lines was unaffected by 5 °C vernalization of 2, 7, and 21 days, but vernalization for 35 days promoted curd initiation, with decreasing temperatures amplifying this effect.
Many of the phenotypic differences between vernalization treatments and between flowering types were strongly linked to BoFLC2 expression behaviour. It was shown for the first time that the basic down-regulation of BoFLC2 is conserved in cauliflower, with transcript levels markedly reduced by as little as one week of vernalization (Fig. 3B). Colder temperatures and longer periods of vernalization amplified this effect. This transcriptional vernalization response occurred independently of plant age (Fig. 3C). In this regard, BoFLC2 vernalization responsiveness is more closely aligned to that of seed-vernalization-responsive members of the Brassicaceae, such as A. thaliana, B. rapa, and B. napus, than the even more closely related cabbage, whose BoFLC expression is unaffected by vernalization at young ages (Lin et al., 2005).
Although expression responses of these two genes were very similar in both BoFLC2 and boflc2 varieties, the overall level of BoFLC2 transcript was approximately five-times higher than that of the premature termination codon (PTC)carrying boflc2, probably owing to nonsense-mediated decay (NMD) of the boflc2 mRNA (Fig. 3B).

The transcriptional regulation of three additional Arabidopsis vernalization pathway homologues is conserved in B. oleracea
Having shown that the molecular regulation of BoFLC2 by vernalization is consistent with its role as an important regulator of flowering time, we next isolated and characterized the transcriptional behaviour of BoVIN3, BoFT, and BoFLC3 to gain a greater understanding of the brassica vernalization pathway.
A homologue of the upstream epigenetic repressor of FLC, VIN3, was isolated for the first time in a brassica crop species (Supplementary Fig. S4). Our results support the idea that the function of this gene is conserved in cauliflower, with BoVIN3 consistently up-regulated in response to vernalization in all plant tissues, and with no significant differences observed between early-and late-flowering parent lines (Fig. 3). This suggests that the expression response of this BoVIN3 allele to vernalization is not a major contributing factor to variability in vernalization responsiveness or flowering time in the cauliflower lines investigated. In contrast to the Arabidopsis model, in which prolonged cold temperatures (approx. 20 days at 2-4 °C) are required to induce expression (or more accurately, relieve the repression; Kim et al., 2010) of VIN3, results from this study show that BoVIN3 is significantly upregulated by as little as 2 days of 5 °C vernalization and that there is little difference between expression levels in plants vernalized for 2, 7, and 21 days at 5 °C (Fig. 3B). BoVIN3 expression in leaves increased substantially after 35 days of vernalization. It is possible that the up-regulation of BoVIN3 observed following shorter vernalization treatments (i.e. those less than 35 days) is insufficient to elicit stable downregulation of BoFLC (i.e. down-regulation that is maintained even in the absence of the cold stimulus), accounting for the lack of curd initiation in BoFLC2 plants. Lin et al. (2005) raised the possibility that BoVIN3 may be involved in the differential response of seed as opposed to plant vernalization in cabbage, suggesting that BoVIN3 may only be upregulated during vernalization of mature cabbages, but not in vernalized seedlings. However, Fig. 3C shows that in this study, BoVIN3 transcription was up-regulated by vernalization regardless of plant age.
BoFT, the putative downstream target of BoFLC2 was characterized in cauliflower for the first time, and found to be up-regulated in response to vernalization (Fig. 3). Cauliflower BoFT was up-regulated in both leaves and apices (Fig. 3A), unlike cabbage BoFT, which was unaffected by vernalization in leaf tissue (Lin et al., 2005). The overall level of BoFT expression was approximately tenfold lower in BoFLC2 parent lines, reflecting strong suppression by BoFLC2 (Fig. 3B). In lateflowering parent lines, the up-regulation of BoFT in response to vernalization increased with plant age, with younger plants showing little vernalization response (Fig. 3C). This suggests that some factor other than BoFLC is contributing to cauliflower juvenility given that BoFLC2 is significantly (P=0.05) down-regulated by three weeks of vernalization independently of plant age (except for P=0.083 at week 10 where one replicate was excluded owing to qPCR failure). It is possible that delayed BoFT responsiveness is mediated by a transcriptional activator, as opposed to putative negative regulators such as BoFLC and BoVIN3.
BoFLC3 expression patterns were similar to those of BoFLC2. Whereas Lin et al. (2005) found that BoFLC3 transcript levels were too low to detect using a probe derived from the 3'-UTR of BoFLC3, this study showed that BoFLC3 is strongly expressed in cauliflower, at a level substantially higher than BoFLC2, and is also regulated by vernalization in a similar manner to BoFLC2 (Fig. 3). Therefore, it seems probable that BoFLC3 could make at least a minor contribution to the differences in timing of curd formation in boflc2 parent lines following vernalization. Furthermore, recent studies have shown that in species with expanded flowering gene families, transcriptional cross-regulation between different members of the same family may occur (e.g. Hecht et al., 2011); these results suggest that there is no such crossregulation between BoFLC3 and BoFLC2, with BoFLC3 regulation entirely independent of BoFLC2 expression and functionality.

Correlations between gene expression and reproductive state are preserved under field conditions
Although the molecular basis of vernalization-dependent flowering time regulation in Arabidopsis has been intensively studied under controlled conditions, comparatively little is known about flowering gene expression control in more complex, natural environments. Indeed, it has recently been shown that gene regulation in a fluctuating environment is substantially different to that observed in the homogenous and controlled environment of the laboratory (Richards et al., 2009). In 2010, results from a two-year census of the seasonal expression of FLC in an perennial population of outdoor-grown Arabidopsis halleri were reported, showing that seasonal trends of temperature were detected at the level of FLC regulation (Aikawa et al., 2010). However, apart from this study, there are very few reports of the measurement of seasonal/temporal changes in gene expression under natural conditions.
Having examined gene expression in response to artificial vernalization treatments, we next investigated the transcriptional dynamics of flowering genes in pot-grown cauliflower plants exposed to natural, fluctuating temperature conditions University of Tasmania's Horticultural Research Centre (Sandy Bay, Tasmania). To avoid the problems associated with poor genotype-environment matches encountered in earlier outdoor trials, sowings of late-flowering parent line 51 and early-flowering line 48 were staggered to synchronise the transition from vegetative to reproductive growth in the two lines.
Levels of BoFLC2 mRNA in apices of the late-flowering line 51 were initially high during early vegetative growth in mid-autumn (Fig. 4). Expression was rapidly down-regulated throughout late autumn and early winter, reaching minimum values approximately 120 days after sowing, nearly a month before a swollen apical bud was detected. Expression remained low throughout early curd development, eventually rising slightly in curds with distinct floral buds. Fluctuations in temperature throughout the seasons had little effect on expression, suggesting the involvement of a 'buffering' mechanism in FLC regulation that allows seasonal cues to be extracted from the background noise of fluctuating temperatures. Consistent with earlier findings, expression of BoFLC2 in lines carrying the boflc2 mutation was greatly diminished in the non-functional line 48, but the same basic profile was recorded, albeit with more pronounced up-regulation as curds elongated and flowered during the spring months. BoFLC3 patterns were similar. Figure 4 shows that initial BoFT expression levels in line 51 were very low throughout autumn and early winter. A consistent, but slight up-regulation representing a 13-fold increase in mRNA levels from 57 days after sowing to 149 days after sowing may be seen when the data is viewed at a smaller scale, but it is not until 163 days after sowing, when visible curds were first detected, that up-regulation of BoFT accelerates substantially. However, these changes in BoFT expression are minor compared with the marked up-regulation that occurs during curd expansion, and the subsequent up-regulation associated with the development of floral buds throughout spring. The main BoFT up-regulation occurred nearly a month after the BoFT-inhibitor BoFLC2 reached its lowest point. In the younger, early-flowering parent line 48, a similar profile was generated, with a marked increase in BoFT expression also observed immediately following curd formation, before mRNA levels taper off during later curd development.
BoVIN3 expression in line 51 generally increased throughout autumn/winter before a peak 149 days after sowing, when 'fat' apices were first noted (Fig. 4). After this point, BoVIN3 expression trended downwards. In the boflc2 line, BoVIN3 expression was highly erratic from week to week, and there does not seem to be any connection between the stage of reproductive development and BoVIN3 expression.
This long-term gene expression study provides the first realistic profile of gene expression under natural conditions in cauliflower, and is an essential link between controlled-environment experiments with artificial vernalization treatments and 'real-world' application. Importantly, it also indicates robust regulation of the long-term control of BoFLC and BoFT in both lines, with episodic warm temperatures throughout the autumn and winter months seeming to have little influence on the overall direction of gene regulation. This suggests that the epigenetic memory of winter in these two lines is sufficiently long to buffer against short-term temperature fluctuations (c.f. Aikawa et al., 2010).
In the Forthside field trial, two distinct groups were identified, in terms of reproductive behaviour. The "very late" parent lines 2, 3, and 7 formed curds and subsequently went on to flower much later than the other group of BoFLC2 lines which comprised romanesco lines 1, 21 and 44 and "Late" line 25 (Fig. 5A). As seen in Fig. 5B, there was a strong positive correlation between BoFLC2 expression at a fixed time point before curd initiation and both the average number of days to curding (r=0.892, P=0.003) and the average number of days to flowering (r=0.880, P=0.004). These results provide strong evidence in support of the idea that BoFLC2 expression in vegetative plants is a reliable indicator of curding and flowering time in BoFLC2 parent lines. BoFT expression levels in apices at the sampling date were also strongly correlated with curding time (r=-0.760, P=0.024), but not with flowering time. Fig. 4. Gene expression throughout autumn, winter, and spring in apical tissue of cauliflower parent lines 51 (BoFLC2) and 48 (boflc2). Inset: detail of BoFT expression during early growth. Values are based on one replicate only. The developmental stage of both parent lines is shown above each set of graphs, where V=vegetative; F=fat, expanded apex without any discernible curd structure; and 1-5=developmental stages according to Supplementary Fig. S1. Parent line details are provided in Supplementary Table S1.  Supplementary Table S1.
Although interest in the development of genetically informed models of progression towards flowering has increased in recent years (Uptmoor et al., 2008;Wilczek et al., 2009), the concept of using gene expression data to monitor a plant's developmental status has not been explored to any significant extent. These results indicate that the correlations described between gene expression and reproductive state under controlled conditions are also valid under field conditions and illustrate the potential for such expression assays to form the basis of molecular tools for understanding and predicting flowering in dynamic and complex production environments.

Conclusion
Using an integrated physiological, genetic, and molecular approach, we have demonstrated the role of BoFLC2 as a key determinant of flowering time within the annual B. oleracea var. botrytis group. An association approach provided evidence that variation at the BoFLC2 locus is an important determinant of reproductive behaviour, and this hypothesis was strengthened with a more specific study involving segregating populations. Regulation of BoFLC2 expression in response to vernalization is consistent with the FLC function observed in the canonical Arabidopsis vernalization pathway, and transcriptional profiles of both upstream and downstream genes serve to support this view; however, a considerable amount of work remains to fully elucidate the cauliflower flowering pathway. In particular, the ancestral polyploid nature of this species means that there are likely to be multiple additional undiscovered genes even within the families of those genes investigated here that also interact with one another and make significant contributions to the floral transition. The validation of gene expression behaviour under outdoor conditions raises the potential for measurements of relative transcript level to form the basis of assays that could be used to predict curd induction or flowering in production environments.

Supplementary data
Supplementary data are available at JXB online. Figure S1. Curd development key. Figure S2. Distribution of BoFLC2 genotypes across the five cauliflower flowering time classes. Figure S3: Percentage of curds initiated following vernalization of BoFLC2 and boflc2 parent lines. Figure S4: Alignment of BoVIN3 amino acid sequence with B. rapa and A. thaliana homologues. Table S1. Brassica variety/flowering classifications. Table S2. Primer combinations used for gene isolation and RT-qPCR. Table S3. Calculation of the contribution of BoFLC2 to phenotypic and genetic variance.