In this experiment, we utilized a subset of previously described A. thaliana (L.) Heynh accessions for this work (Lewandowska-Sabat et al., 2010, 2012a,b, 2017). First-generation descendants from individuals collected from 10 locations with diverse climates throughout Norway were used in this study (Supplementary Figure 1 and Supplementary Table 1). These accessions displayed both facultative and obligate vernalization requirements, varying flowering times, and varying responses to long photoperiods, all of which correlated with location and climate variables from the accessions’ homesites (Lewandowska-Sabat et al., 2012a, 2017). Individuals were subjected to 8- and 19-h photoperiods at 16°C after 13 days in 8-h photoperiods at 23°C, and then 14 weeks of vernalization at 4°C and 8-h photoperiods. The long vernalization period was used to saturate the vernalization response. Hydrargyrum quartz iodide (HQI) lighting systems (Osram, Hungary) were used for the 8-h photoperiod as a source of photosynthetically active radiation (PAR), while the 19-h photoperiod was created by adding 1.5 h of light from incandescent bulbs (LU400/XO/T/40 Philips General Electric, Munich, Germany), which have a lower red:far-red ratio, prior to and after a 16-h period using the HQI system to simulate dawn and dusk. Plants were grown in 6.5 cm-diameter soil-filled (Hasselfors Garden AB, Örebro, Sweden) pots and bottomed watered two times per week or once per week during vernalization. For five individuals from each accession, the number of days post-vernalization to the production of a visible bolt (days to bolt; DTB), days to first visible flower (days to flower; DTF), rosette leaf number at flowering, and bolt height at flowering were used for this study and previously described as part of a larger assessment of response to photoperiod (Lewandowska-Sabat et al., 2017). Leaf petiole length, recorded as the average petiole length per plant, and rosette diameter at flowering were also recorded, but have not been previously described. Bolt height data for one accession, and petiole length and rosette diameter for four accessions were not recorded on the same plants used for flowering, and therefore replaced by corresponding data from other descendants from the same parent as the other flowering traits were relatively uniform across descendants from a single parent. Three individuals per treatment of two common lab accessions, Col-0 and Landsberg erecta-0 (Ler-0), were included in this planting for comparison. Data from Col-0 and Ler-0 has not been previously reported.

Leaf tissue from individuals grown in the 19-h photoperiod and harvested at 1, 9, 13, 17, 20, and 24 h after onset of the PAR, 5 days after being moved from vernalization to the photoperiod treatment, were used for comparison of mRNA accumulation. Tissue from 17 h after PAR onset (zeitgeber 17; ZT17) from days 3, 5, and 8 after vernalization was used to assess gene expression change over time. Tissue was stored at –80°C until processed, moved to 2-ml Safe Seal Microtubes (Sarstedt, Medline, Netherlands) containing three 3.2 mm stainless steel beads (Biospec Products, Fisher Scientific, Norway), cooled in liquid N, then ground using a TissueLyser (Qiagen, Norway). mRNA was isolated using the Illustra RNAspin Mini kit including on-column DNAse treatment (GE Healthcare, Fisher Scientific, Norway), concentrations were determined using a NanoDrop^TM 8000 Spectrophotometer (Thermo Fisher Scientific, Norway), mRNA quality was determined using an Automated Electrophoresis Bioanalyzer System (Agilent, Denmark), cDNA was synthesized using Superscript VILO cDNA synthesis kit (Applied Biosystems, Fisher Scientific, Norway), and qPCR was conducted using SYBR Select Master Mix (Applied Biosystems, Fisher Scientific, Norway) on a QuantStudio^TM Real-Time PCR system (Thermo Fisher Scientific, Norway). Expression of CONSTANS (CO), FLOWERING LOCUS T (FT), and FLOWERING LOCUS C (FLC) was assessed relative to ACTIN and IPP2 (Sawa et al., 2007) using the ΔCT method. Primers are listed in Supplementary Table 2.

Leaf tissue from the 8-h photoperiod treatment at the 9-h time point was isolated for comparison against the leaf tissue harvested at the 17-h time point in the 19-h photoperiod on day 5 after vernalization. These time points were selected because they were 1 h post, and did not directly coincide with a rapid switch from light to dark (8-h treatment) or from the HQI to incandescent lighting systems (19-h treatment) in our experimental conditions, as well as because they were close to the end of the light period when FT should be strongly expressed as its upstream inducer CO is degraded in the dark (Kobayashi and Weigel, 2007). In this case, RNA from both photoperiod treatments was isolated using TRIzol^TM (Thermo Fisher Scientific, Norway) followed by DNase treatment in solution using the RNase-Free DNase Set (Qiagen, Norway) followed by RNA cleanup using isopropanol. The remaining steps were as above except that FT cycle threshold (CT) values were relativized to IPP2.

In all cases, three individuals from each accession, separate from those used to record flowering phenotypes, were sampled at each time point for biological replication. Technical duplicates from randomly selected time points were processed for gene expression analysis to ensure uniformity of processing and pipetting. No-template controls were included with all qPCR runs to ensure sample contamination did not occur. For FT, some timepoints within replicates, primarily in Lod-1 and Tje-1, yielded undetectable values. As the same samples yielded consistent expression for the four other genes measured, wherein FT was overall low in these accessions, we interpreted the expression of FT in these samples to be very low and therefore undetectable.

Correlations among DTB, DTF, leaf number, bolt height, petiole length, and rosette diameter were assessed using the cor function in R (method = pearson, version 3.6.3). Differences in DTB across the two photoperiod treatments were assessed using ANOVA (aov function in R, version 3.6.3). In the 8-h treatment, the line from Kvi-1 was missing one individual as it had died early in the experiment. These missing values are imputed by averaging the remaining four to ensure a balanced design (Janssen et al., 2010). Flowering times of Col-0 and Ler-0 were included in plots for visual comparison, but not included in the statistical analyses of flowering times as their sample size differed (three vs. five). However, they were included in any analyses that included gene expression.

To explore relationships between gene expression and flowering phenotypes, we first approximated the total amount of each gene expressed each day by calculating the area under the curve (AUC) for each time course using the trapz function in R (pracma package, version 3.0.1) (Borchers, 2021). Expression for Kvi-1 was excluded from further analyses as its levels for CO were outside the range of all other lines (Supplementary Figure 2), outside those shown for CO for that accession in a pilot experiment with other lines (Supplementary Figure 3), and because its leaves were purple and appeared stressed. Then, we assessed the relationship between the flowering phenotypes and FT, CO, and FLC through ANOVA using the aov function in base R and the anova function to compare different models for final model selection. These tests were done using accession means as the individuals used for gene expression were not the same as those used to assess flowering, although they were from the same grow-up. Assumptions of normality were assessed on the final models using the boxcox power transformation (boxcox function, MASS package in R) (Box and Cox, 1964; Venables and Ripley, 2002), and the dependent variable was transformed if necessary. As transformations were determined using the fitted values relative to the independent variables in each model, transformations could differ for the same dependent variable across models, and sometimes no transformations were required. Relationships among genes were analyzed in the same manner, except those values for individual plants were used as all five genes were measured in each individual, allowing one-to-one correspondence. In addition to transformations to meet normality assumptions, we also log-linearized FT and FLC to assess behavioral trends across accessions grouped into different flowering types.

To compare gene expression over time, we grouped the lines by rapid and slow flowering in terms of their DTB and included the lab accessions as a separate group. We then normalized the expression in each group by the maximum value within each accession or strain so that we could compare patterns of change rather than level. Finally, we used ANOVA to compare the effects of time, flowering type (group), and the time by type interaction using the aov function in R.

We first assessed correlations across the phenotypic characteristics in the winter-annual accessions to assess which characteristics we would pursue for further study and to better understand how the plant structure might vary across these wild accessions. We focused on the long photoperiod as the original seeds were collected well after photoperiods exceeded 16 h and plants from these accessions likely flowered in the wild when days were long (Supplementary Table 1). We noted that petiole length and rosette size were correlated, while petiole length and bolt height negatively correlated with the number of DTB (Figure 1A), which meant that later flowering plants were potentially flowering at a smaller size and perhaps had a more compact growth habit than early flowering plants as their petioles were shorter. Leaf number at bolt did not correlate with any other plant trait, even though leaf number is frequently used as a proxy for flowering time in Arabidopsis (Pouteau and Albertini, 2009). Plants growing in the short photoperiod had patterns like those in the long photoperiod, except that petiole length positively correlated with DTB. Thus, it is possible that growth form, specifically the compact growth form of the leaves, changes seasonally with daylength in natural settings, but that was not tested here.

Since DTB, DTF, and leaf number are common measures of flowering time (Pouteau and Albertini, 2009), and DTB and DTF were strongly correlated, we moved forward with the comparison of DTB and leaf number, using DTB at it was closer in time to the physiological transition from production of vegetative to reproductive structures. In these winter-annual accessions, we noted two visible groupings in timing to flowering in DTB, especially in the 19-h photoperiod, which were those that flowered more rapidly and those that flowered more slowly post vernalization (Figure 1B). Hereafter, we referred to these groups as rapid flowering type and slow flowering type. This was not dependent on vernalization requirement as Nfro-1 and Lod-1, the two lines facultative for vernalization, fell into different groups (Lewandowska-Sabat et al., 2012a). DTB showed a clear effect due to photoperiod across accessions, while leaf number did not (Figure 1C). We noted that some winter-annual accessions showed little difference across the short and long photoperiod treatments. However, this did not seem to correlate with the flowering type (rapid vs. slow flowering) (Figure 1D).

To assess whether flowering times correlated with gene expression and which measure of flowering time-correlated most strongly, we compared the expression levels of FT, FLC, and CO from leaves harvested at six timepoints on day 5 after vernalization ended and photoperiod treatments began using tissue collected from the long photoperiod treatment. We included two accessions commonly used for lab studies, namely Col-0 and Ler-0, for which the relationship between FT and flowering had previously been shown across treatments (Kobayashi et al., 1999; Corbesier et al., 2007; Kinmonth-Schultz et al., 2016). These accessions were summer annuals, allowing us to assess whether the behaviors in gene expression differed across summer- and winter-annual variants of Arabidopsis. FT tended to show a two-peak profile across the ten winter-annual Norwegian and two summer-annual lab accessions, showing the highest level at 17 h after onset of photosynthetically active radiation (PAR, zeitgeber 17, ZT17) which occurred at the end of the light period, and a second peak at ZT9 which occurred midway through the 16-h period of PAR light (Supplementary Figure 2). The latter peak is consistent with the morning peak observed in outdoor-grown Arabidopsis that was due to a lower red:far-red ratio, which we mimic at dawn and dusk in our conditions (Song et al., 2018). FLC showed its highest expression at ZT1, while CO tended to peak at the end of the light period. Expression levels of CO for accession Kvi-1 were outside the range of all other accessions, showing an atypical mid-day peak, and different from those shown for CO for that accession in a pilot experiment (Supplementary Figure 3). We also noted that its leaves were purple and appeared stressed. Thus, we excluded Kvi-1 from further analysis.

Afterward, we assessed whether the expression of FT, FLC, or CO could be predictive of flowering time and whether the two upstream regulators FLC and CO could be predictive of FT expression. To assess flowering time, we used estimates of total expression from day 5 after vernalization, as calculated by determining the AUC for each time course and gene, as independent variables, comparing them against the dependent variables DTB and final rosette leaf number. We used average values across replicates as the plants for which gene expression was measured were destructively harvested and therefore not assessed for flowering, although they were grown with those used for flowering. Since CO protein is degraded in the dark (Valverde et al., 2004), CO mRNA produced during the day should have a greater impact on FT levels and flowering time. Therefore, we compared eight different models using ANOVA for each measure of flowering time (leaf number and DTB), specifically the three genes with and without their interactions, the three genes including only daytime CO expression with and without their interactions, and each of these models including flowering type as an additional variable. The three flowering types were rapid and slow flowering for the winter-annual accessions, as determined from DTB, and summer-annual accessions (hereafter, referred to as “lab” accessions). We sequentially compared nested variants in both groups of models, those including total CO and those including only daytime CO, to determine which terms to include in the final models.

First, we described the selection process and results for models including DTB as the dependent variable. For the group of models containing total CO, no model performed significantly different than the model containing the three genes without interactions, and so neither flowering type nor the interaction terms were included in the final model from that group (Supplementary Table 3). For the group of models containing daytime CO, a model containing the three genes without interactions but including flowering type was significantly different from the simplest model form, so it was used for further analysis. Since total CO and daytime CO describe slightly different biological processes, not different subsets of the same set of independent variables, we report the results of the final models in both groups here (Table 1). For the model containing total CO, FT, and FLC, only FLC showed a significant effect on flowering time. While for the model including daytime CO, FT, FLC, and flowering type, CO also had a significant effect. FT did not have a significant effect on flowering at this time point 5 days after vernalization. Thus, two upstream regulators of FT and flowering time, CO and FLC, likely drive differences in flowering times in these accessions of Arabidopsis.

For the models including rosette leaf number as the dependent variable, no model including interaction terms or flowering type differed from the simplest models in both model groups (Supplementary Table 3). Therefore, we continued only with the simplest models in each group, and across both models, no term in FT, FLC, total CO, or daytime CO showed a significant effect on rosette leaf number (Table 1). These results indicated that the expression of FT, FLC, and CO are not predictive of the total number of leaves produced before flowering in these winter-annual accessions of Arabidopsis.

Since FT acts downstream of FLC and CO (Kobayashi and Weigel, 2007), we next assessed the degree to which FLC and CO were predictive of FT levels across these accessions and time points using ANOVA. In this instance, we were able to compare individuals as the values across genes came from the same plants. We compared models including CO and FLC with and without their interaction as well as with a model including the three flowering types [rapid-flowering winter annuals, slow-flowering winter annuals, and summer annuals (lab)] as a covariate. As the model including the interaction between CO and FLC and the model including flowering type differed from the model including only CO and FLC, we included both their interaction and flowering type in the final model (Supplementary Table 4). We found strongly significant effects of FLC and a strong interaction between CO and FLC (Table 2). There was also a significant effect of CO which was strongly significant when flowering type was included in the model. To understand how the relationships among these genes varied across the flowering types, we plotted FT relative to both CO and FLC individually across the three groups and assessed the individual of effects CO and FLC incorporating their interaction with the flowering type (Table 2). The interactions between FLC and type and between CO and type were both significant as were the individual effects, indicating that each gene affected FT as expected and that the effect of each gene on FT differed across flowering types. Each flowering type showed very different relationships among the three genes. FT showed very low levels across a range of CO values in the slow-flowering winter-annual accessions, while FT positively correlated with CO in both the rapid-flowering winter-annual accessions and summer-annual lab accessions (Figure 2A). The lab accessions showed overall higher levels of FT across the same values of CO than either the rapid- or slow-flowering winter-annual accessions, but a similar rate of increase to that observed across the rapid-flowering winter-annual accessions although the latter trend appeared driven by a few points. These patterns were the same when replicates were averaged within each accession (Supplementary Figure 4). Together, the data in the rapid-flowering winter-annual accessions and lab accessions were consistent with CO as an activator of FT and potentially with a scenario in which CO activates FT at similar rates regardless of flowering type. However, some factor other than CO was influencing the overall level of FT. Furthermore, CO was not predictive of FT in the slow-flowering winter-annual accessions at the time point tested, 5 days post-vernalization.

The relationship between FT and FLC showed a different pattern from that of FT and CO. FT was primarily expressed when FLC was very low, while FT ranged near zero at moderate to high levels of FLC (Figure 2B). To better understand the relationship among these genes we log-linearized them both. Doing so revealed different patterns across flowering types. The lab accessions showed no clear relationship between FT and FLC, likely due to the basal levels of FLC expression observed in those accessions as Columbia and Landsberg erecta both retain functional variants of FLC (Koornneef et al., 1994; Michaels and Amasino, 2001; Gazzani et al., 2003; Schmalenbach et al., 2014). FT correlated negatively with FLC in the two winter-annual groups. However, the rate at which FT decreased with an FLC increase was steeper in the slow-flowering winter-annual accessions (Figure 2C). Again, these patterns were the same when replicates were averaged within each accession (Supplementary Figure 4). These data suggested that FLC is repressing FT with the strongest effect being in the slow-flowering winter-annual accessions. It is likely that the presence of FLC is inhibiting FT transcriptional activation by CO in the slow-flowering winter-annual accessions and influencing overall FT levels in the rapid-flowering winter-annual accessions relative to the summer-annual lab accessions. Possibly, there is a threshold level of FLC over which FT expression is strongly inhibited since the relationship between FT and FLC across all summer and winter-annual accessions tested was strongly non-linear with FT only being expressed above residual levels when FLC was low.

While the levels of CO mRNA are influenced primarily by the circadian clock and should be relatively stable through time in the constant temperature and daylength conditions used here (Song et al., 2015), FT appeared to accumulate over time to influence flowering (Krzymuski et al., 2015; Kinmonth-Schultz et al., 2019), while FLC varied temporally post vernalization in accessions of Arabidopsis that differ in their strength of vernalization-induced FLC repression (Li et al., 2014). Therefore, we hypothesized that the levels of FLC and FT would change through time post vernalization in these accessions. We compared single time points across days 3, 5, and 8 post vernalization, and to select the time point to use as FT and FLC peaked at different times of the day, we first determined whether ZT1 or ZT17 from day 5 showed relationships with DTB similar to the full time course. Since ZT17 showed the most similar correlations for both genes (Supplementary Table 5), we selected ZT17 for analysis across the 3 days. We compared values normalized by the maximum value within the three flowering types to allow us to assess change across time rather than the different relative levels among flowering types. We found that FT had a strong effect on time and type, as well as a significant interaction between the two terms (Figure 2D). FT increased over time across all groups and was highest on day 8. However, the pattern was less clear for the lab accessions, likely contributing to the significant interaction. FLC showed a different pattern. There was a slight effect of time and a strong time-by-type interaction. Rapid-flowering winter-annual accessions showed no change over time while slow flowering accessions decreased over time (Figure 2D). The latter was consistent with the FLC declines observed post vernalization in some Arabidopsis accessions previously (Li et al., 2014). We also assessed the relationship between FT, FLC, and DTB on days 3, 5, and 8 after vernalization. We noted a strong correlation between FT and FLC on all 3 days that was strongest on day 3 after vernalization (Supplementary Table 6 and Supplementary Figure 5). Conversely, the relationship between FT and DTB was strongest on day 8, while the relationship between FLC and DTB was similarly strong across the 3 days (Supplementary Table 6 and Figure 2E). Therefore, the influence of FT on flowering likely increased over time, while FLC might have retained some influence on FT early after vernalization, especially in the slow-flowering winter-annual accessions. A factor other than FLC might influence FT later.

Since we saw little difference in DTB between the short and long photoperiods in some accessions, we hypothesized that FT may be more similarly expressed across photoperiods in these accessions. We compared FT levels from timepoints near the end of the light periods between plants grown in 8- and 19-h photoperiods as the upstream inducer of FT, namely CO protein, was degraded in the dark (Valverde et al., 2004). We found very low levels of expression across all accessions in the 8-h photoperiod (Figure 2F). Within the winter-annual accessions classified as slow, there was little difference between long and short-day treated plants with both treatments having similarly low levels of FT. However, rapidly flowering accessions had much higher FT levels in long days (Figure 2F). We assessed whether short-day FT levels might explain either DTB in short days or the relative change in DTB between the 8 and 19-h photoperiods. However, linear models between FT in 8-h photoperiods and DTB in 8-h photoperiods or between FT in 8-h photoperiods and the relative change in flowering between 8- and 19-h photoperiods were not significant. Thus, a factor other than FT might be contributing to flowering in short days in these accessions.