Arabidopsis thaliana ecotype Columbia (Col) was used throughout this study. T-DNA insertion lines phyB (SALK_022035, Mayfield et al., 2007; Ruckle et al., 2007), pft1-2 (SALK_129555, Kidd et al., 2009), flc-6 (SALK_41126, Schönrock et al., 2006a) and msi1-5 (WiscDs Lox302B08) were obtained from NASC and confirmed by PCR. Seeds of FRI-Col, esd1-10, ft-10 and soc1-2 have been described (Lee and Amasino, 1995; Lee et al., 2000; Yoo et al., 2005; Martin-Trillo et al., 2006) and were kindly provided by J. Jarillo (FRI-Col, esd1-10), D. Weigel (ft-10), I. Lee (soc1-2), B. Ayres (co-1). The mutant co-1 (accession La-0, Redei, 1962) was backcrossed into Col. The line msi1-tap1 (accession Col) has been described before (Bouveret et al., 2006). Double mutants were identified among progeny of appropriate crosses by PCR with gene-specific primers (Supplementary Table 1).

To construct plants that ectopically overexpress FT (35S::FT), the full-length coding sequence was inserted into the binary destination vector pK7WG2 (Karimi et al., 2002) downstream of the cauliflower mosaic virus (CaMV) 35S promoter and transformed into msi1-tap1 plants. Transformants were selected on kanamycin plates and genotyped by PCR. Hemizygous T2-generation plants of three independent T1 lines were analyzed for flowering time.

Seeds were sterilized and plants were grown on Murashige and Skoog (MS) basal salt medium (Duchefa, Brussels, Belgium) after stratification at 4°C for 2–3 days. Plants were analyzed on plates or transferred to soil (“Einheitserde,” H. Gilgen optima-Werke, Arlesheim, Switzerland) 10 days after germination. Plants were kept in Conviron growth chambers with mixed cold fluorescent and incandescent light (130 μmol m⁻² s⁻¹, 21 ± 2°C) under (LD, 16 h light) or short-day (SD, 8 h light) photoperiods or were raised in green houses [LD: 14 h light, 19°C/10 h dark, 14°C; SD: 8 h light, 20°C/16 h dark, 20°C; supplemented with mercury vapor lamps (Sylvania Lighting S.A., Meyrin, Switzerland) to a maximum of 150 μmol m⁻² s⁻¹]. Flowering time was scored as described (Möller-Steinbach et al., 2010).

Total RNA was extracted as previously described (Hennig et al., 2003; Leroy et al., 2007; Alexandre et al., 2009). 1 μ g RNA treated with DNase I (Promega, Dübendorf, Switzerland) was transcribed into cDNA using a RevertAid First Strand cDNA Synthesis Kit (Fermentas, Nunningen, Switzerland) according to manufacturer's instructions. qRT-PCR with gene-specific primers (Supplementary Table 2) was performed on three technical replicates with the Fast Start Universal Probe Master (Rox) reagent and the Universal Probe Library set (UPL) (Roche Diagnostics, Rotkreuz, Switzerland) according to the manufacturer's instructions and results were normalized to PP2A as described (Exner et al., 2009). Shown is one of at least two independent biological experiments with similar results.

Previously, we had reported that MSI1 antisense lines and msi1 mutants partially complemented with untagged pMSI1::MSI1 or tagged pMSI1::MSI1:TAP constructs were late flowering (Bouveret et al., 2006). This suggests that undisturbed MSI1 levels are needed for normal flowering promotion. Homozygous msi1 null mutants are lethal (Köhler et al., 2003; Guitton et al., 2004). Here, we analyzed heterozygous plants of the original msi1-1 and a novel msi1-5 allele and found that both flowered later than wild type under LD but not SD (Supplementary Figure 1). Similarly, msi1-1^−/− plants partially complemented with a pMSI1::MSI1:HA construct were late flowering (Supplementary Figure 1). Therefore the dose of MSI1 is important for flowering time. The observation of a late flowering phenotype for msi1-mutants and transgenic lines motivated us to investigate the genetic pathway(s) in which MSI1 acts to affect flowering. Because the flowering delay was considerably more severe for the msi1-1^−/− pMSI1::MSI1:TAP (msi1-tap) line than for heterozygous msi1 mutants, we used the msi1-tap1 line in subsequent experiments. Unlike heterozygous msi1 mutants, msi1-tap1 flowered much later than wild type in SD (Bouveret et al., 2006). This is consistent with the generally milder late flowering phenotype of heterozygous msi1 mutants. The normal flowering in SD may suggest that a single wild-type MSI1 allele can largely suffice for normal MSI1 functionin SD. It remains to be tested whether MSI1 requirements are lower in SD or whether a potential compensatory mechanism can more efficiently up-regulate the remaining MSI1 allele in SD than in LD.

We then tested a potential function of MSI1 in the light quality pathway, which functions through phytochromes and independently of the circadian system. PHYB, PHYD, and PHYE repress FT expression and therefore flowering, with PHYB having the major role in this process (Kim et al., 2008). The PHYB target PFT1 was proposed to directly activate both, CO and FT expression, while it simultaneously acts as negative regulator of phytochrome signaling by inactivation of PHYB protein (Cerdan and Chory, 2003; Wollenberg et al., 2008). Null alleles of phyB and pft, which are early and late flowering, respectively, were introduced into the msi1-tap1 background and flowering time was analyzed in LD. The loss of PHYB in the phyB msi1-tap1 double mutant led to flowering with 15 rosette leaves (RL) in LD, which was intermediate between the phyB single mutant (4 RL) and msi1-tap1 (19 RL, Table 1). This result suggests an additive interaction between MSI1 and PHYB. The loss of PFT1 in the pft1 msi1-tap1 double mutant resulted in a synergistic delay in flowering (41 RL) compared to the pft and msi1-tap1 single mutants (14 and 17 RL respectively, Table 1), suggesting likewise independent effects of MSI1 and PFT1 in flowering promotion but likely on the same common targets. Together, these results propose a function of MSI1 independent of the light quality pathway in floral induction.

Next, we investigated the potential role of the flowering repressor FLC in MSI1 effects on flowering time. FLC is one of the main regulators of flowering time in Arabidopsis, and altered flowering time is often caused by altered FLC expression. Consistent with earlier observations (Bouveret et al., 2006), analysis of flc msi1-tap1 double mutants suggested that late flowering of msi1-tap1 was independent of FLC, as evident from the largely unaffected late flowering of msi1-tap1 flc plants (Table 1). Previously, we observed reduced SOC1 expression in msi1-tap1 (Bouveret et al., 2006). Here, we tested whether the reduced expression of SOC1 was independent of FLC. SOC1 expression in the double mutant was as low as in msi1-tap1, suggesting that the flc mutation could not lift SOC1 repression in msi1-tap1 plants (Figure 1A). Because MSI1 was recently shown to be involved in FLC control as part of plant PRC2 complexes (De Lucia et al., 2008; Derkacheva et al., 2013), we performed additional genetic tests of a potential role of FLC in msi1-tap1 late flowering. The active FRI-allele of the late flowering Arabidopsis accession San Feliu (Sf2) crossed into Columbia (Col FRI), was introgressed into msi1-tap1. As previously reported (Lee et al., 1993; Clarke and Dean, 1994), Col FRI flowered very late, possibly due to high FLC expression (Table 1). Col FRI msi1-tap1 plants flowered much later (85 RL) than either parent (73 and 20 RL for Col FRI and msi1-tap1, respectively). Some of the Col FRI msi1-tap1 plants were not able to flower at all and died without completing their life cycle. This additive delay in flowering suggests an independent role of MSI1 and FRI in flowering. Previously, we found a strongly synergistic interaction between MSI1 and FVE (Bouveret et al., 2006). FVE is part of the autonomous pathway, which represses FLC, and genes in this pathway were grouped in two epistasis groups. While FVE represents one of the two groups, FCA is a gene from the second group. Here, we tested the genetic interaction between MSI1 and FCA. The fca msi1-tap1 double mutants were extremely delayed in flowering. They ceased to produce leaves without starting to bolt or flower leading to a smaller rosette leave number than for fca. After an extended period of developmental inactivity they eventually died (Table 1). The strongly synergistic interaction suggests that MSI1 and FCA do not function in the same genetic pathway to control flowering time.

Another activator of FLC is EARLY IN SHORT DAYS1 (ESD1, also known as SUPPRESSOR OF FRIGIDA 3 and ACTIN RELATED PROTEIN 6). Mutations in ESD1 hasten flowering through reduced FLC expression in LD and SD (Martin-Trillo et al., 2006; Choi et al., 2007; Lazaro et al., 2008). An early flowering esd1 mutant allele was crossed into msi1-tap1. In LD, the esd1 msi1-tap1 double mutant flowered intermediate (12 RL) to both parents esd1 and msi1-tap1 (5 and 20 RL, respectively) disclosing an additive effect between ESD1 and MSI1 on flowering (Table 1). These data suggest that ESD1 and MSI1 function in separate genetic pathways. Together, these results firmly established that MSI1 can function independently of FLC to affect flowering time in LD.

FT and its homolog TSF are activators of SOC1 (Yamaguchi et al., 2005; Yoo et al., 2005). Increased FT expression was found in msi1-tap1 suppressor mutants, which rescued the msi1-tap1 late flowering phenotype (Exner et al., 2009, 2010). In LD-grown wild-type Arabidopsis, CO activates FT at the end of the day (Suarez-Lopez et al., 2001; Yanovsky and Kay, 2002). To test whether the diurnal rhythm of FT was affected in msi1-tap1, FT expression was profiled throughout the light-dark cycle in seedlings (Figure 1B). In wild type, FT had its expression peak toward the end of the light and beginning of the dark period as previously reported. The FT expression in msi1-tap1 followed the same pattern as in wild type, but expression values were lower, especially at the end of the day (EOD), when expression of FT was reduced by up to 50%. Additionally, we tested whether MSI1 affected the temporal activation of FT or its homolog TSF (Figures 1C,D). Under LD conditions, FT and TSF levels increased steadily in wild type between 3 and 17 days. In msi1-tap1, FT transcripts started to accumulate similarly to wild type but the increase was much slower leading to considerably reduced FT levels. The accumulation of TSF transcripts was even stronger reduced in msi1-tap1 leading to 70% lower levels than in wild type at 17 days after germination. These results demonstrate that normal MSI1-function is needed for typical activation of FT and its homolog TSF in LD.

To test whether higher FT expression can be sufficient to suppress the late flowering phenotype of msi1-tap1, a 35S::FT transgene was introduced into msi1-tap1. The FT over-expression caused extremely early flowering (Figure 2A), which is consistent with the notion that reduced FT expression contributed to the late flowering of msi1-tap1.

To substantiate that delayed activation of FT and therefore of SOC1 was responsible for the late flowering of msi1-tap1, a ft mutant allele was crossed into msi1-tap1 for flower time measurements. The double mutant soc1 msi1-tap1, which was already described in LD before (Bouveret et al., 2006), was included into the analysis (Figures 2B–D; Supplementary Figure 2). Under SD conditions, ft flowered similar to wild type, and the ft msi1-tap1 line flowered similar to msi1-tap1, confirming that FT does not play a major role under these conditions (Figure 2B) (Yanovsky and Kay, 2002; Corbesier et al., 2007). In contrast to FT, SOC1 functions in induction of flowering in SD (Borner et al., 2000) and the soc1 single mutant flowered later than wild type (Figure 2C). While the soc1 msi1-tap1 line needed longer until flowering than either parent, it produced a similar number of RL as the msi1-tap1 parent suggesting that delayed activation of SOC1 contributes at least partially to the late flowering of msi1-tap1 in SD. Thus, during flowering induction in SD, MSI1 and SOC1 appear to function partially in the same genetic pathway.

Under LD conditions, both ft and soc1 flowered later than wild type consistent with their roles in photoperiodic flowering (Borner et al., 2000). The double mutant soc1 msi1-tap1 exhibited an additive late flowering phenotype (42 RL) compared to the msi1-tap1 and soc1 parents (23 and 24 RL, respectively, Figure 2D) confirming earlier results (Bouveret et al., 2006). The ft msi1-tap1 line flowered with 42 RL similar to the ft parent (45 RL) supporting the notion that reduced FT expression is the main reason for late flowering of msi1-tap1 in LD (Figure 2D). In summary, MSI1 affects full activation of FT, TSF and SOC1 expression to promote timely flowering.

Because CO is a main activator of FT, SOC1, and TSF (Suarez-Lopez et al., 2001; Hepworth et al., 2002; Yamaguchi et al., 2005), we asked whether reduced expression of FT, SOC1, and TSF in msi1-tap1 was caused by defects in CO regulation. CO is under strong circadian and diurnal control (for review see Searle and Coupland, 2004), and CO expression in msi1-tap1 was tested throughout an entire light-dark cycle. This experiment revealed that CO expression was considerably lower in msi1-tap1 than in wild type (Figure 3A). The CO expression in wild type showed the previously reported peak toward the end of the day and beginning of the dark. Similarly, this expression pattern was observed for msi1-tap1 suggesting that diurnal regulation was not grossly altered. This conclusion was supported by normal diurnal cycling of CCA1 and TOC1, two components of the central circadian oscillator. However, under the tested conditions, CCA1 and TOC1 showed lower amplitudes of peak expression values in msi1-tap (Figure 3B). Further, we analyzed the CO transcript levels at different developmental time points until 17 days after germination (Figure 3C). Under our conditions, CO increased steadily in wild type during 10 days after germination. In msi1-tap1, CO transcripts started to accumulate similarly to wild type but the increase was slower leading to considerably reduced CO levels. Together, the expression data suggest the hypothesis that MSI1 affects expression of FT, TSF, and SOC1 and flowering time in LD via CO.

To test genetically whether reduced CO expression was responsible for delayed flowering of msi1-tap1, a co mutant allele was introduced into the msi1-tap1 line. Consistent with earlier findings (Koornneef et al., 1991; Robson et al., 2001), the co mutant was late flowering in LD. While msi1-tap1 delayed flowering substantially in the CO wild-type background, it only slightly delayed flowering of a co mutant (Figure 4A) suggesting that late flowering in msi1-tap1 is caused mainly by effects on CO. The similar flowering time of ft msi1-tap1 and the ft co msi1-tap1 triple mutant (Figure 4A) further supported the notion of an epistatic genetic interaction between MSI1 and CO.

Because GIGANTEA (GI) is a major activator of CO expression (Imaizumi et al., 2005; Sawa et al., 2007), we tested whether GI expression was altered in msi1-tap1. At 5 d after germination, when CO levels did not differ between WT and msi1-tap1, GI was also not affected (Figure 3D). In contrast, at 10 d and 17 d, not only CO but also GI expression was substantially reduced in msi1-tap1.

Together, these data suggest that MSI1 acts on flowering in response to the photoperiod through GI and CO on FT.

To further test the importance of MSI1 in the photoperiodic pathway, we performed a SD-LD-SD shift experiment. In SD, FT is only very weakly activated due to immediate destabilization of CO protein after synthesis, and flowering is very much delayed. A brief LD experience can be sufficient to activate FT and induce flowering if the photoperiodic pathway functions normally (Corbesier et al., 1996; King et al., 2008). We cultivated plants under SD conditions interrupted by 1 or 3 days of LD at 45 days after germination and measured flowering time (Figure 4B). Under these conditions, Col was highly sensitive to the additional LD exposures, and flowering was accelerated by about 3 weeks. In contrast, the effect on flowering time of msi1-tap1 was minor and not statistically significant. Heterozygous msi1-1 mutants reacted like wild type to three additional LDs but showed a reduced response to a single additional LD (Figure 4B). Together, these results demonstrate that MSI1 is needed for normal sensitivity of the photoperiodic pathway.