Abstract
We assessed temperature influence on flowering by incorporating temperature-responsive flowering mechanisms across developmental age into an existing model. Temperature influences both the leaf production rate and expression of FLOWERING LOCUS T (FT), a photoperiodic flowering regulator, in leaves. The Arabidopsis Framework Model incorporated temperature influence on leaf growth but ignored the consequences of leaf growth on and direct temperature influence of FT expression. We measured FT production in differently aged leaves and modified the model, adding the mechanistic temperature influence on FT transcription, and linking FT to leaf growth. Our simulations suggest that in long days, the developmental timing (leaf number) at which the reproductive transition occurs is influenced by day length and temperature through FT, while temperature influences the rate of leaf production and the time (in days) the transition occurs. Further, we demonstrated that FT is mainly produced in the first 10 leaves in the Columbia ecotype, and that FT accumulation alone cannot explain flowering in conditions in which flowering is delayed. Our simulations supported our hypotheses that: 1) temperature regulation of FT, accumulated with leaf growth, is a component of thermal time, and 2) incorporating mechanistic temperature regulation of FT can improve model predictions in fluctuating temperatures.
Introduction
Ambient temperature during the growing season correlates with the timing of plants’ transition from vegetative to reproductive growth. Germination, organ emergence, leaf expansion, photosynthesis, and respiration display similar relationships (Parent et al., 2010). These findings have led to the concept of “thermal time” (Lehenbauer, 1914), a metric that asserts that temperature-driven metabolic rates govern development (Zavalloni et al., 2006), and to models that use the empirical relationship between temperature and development to predict plant response (e.g., Chuine, 2000; Jones et al., 2003).
Thermal time accumulation describes an aggregate of underlying plant responses. Thermal units accumulate more quickly, and reach a predetermined threshold sooner to predict flowering, during warm growing seasons than cool ones. Thermal time implies 1) that all plant physiological rates increase in tandem with temperature increases and 2) that fluctuating and constant temperatures have the same influence on most physiological rates if the mean temperature remains stable. However, processes do not always slow under cool temperatures. The up-regulation of cryo-protective genes (Jaglo-Ottosen et al., 1998) and the circadian clock’s buffering to temperature changes (Rensing & Ruoff, 2002) are just two examples.
Furthermore, predicting plant response to future climates remains imprecise when considering temperature alone or in conjunction with CO2 (Asseng et al., 2013; Makowski et al., 2015). The effect of non-stressing temperatures varies among cultivars (Karsai et al., 2013), and plants may respond differently to temperature fluctuations than predicted from constant temperatures (Yin & Kropff, 1996; Kim et al., 2007; Karsai et al., 2008). As most plant models incorporate some variant of thermal time (Ritchie & Otter, 1985; Jamieson et al., 1998a,b; Wilczek et al., 2009; He et al., 2012; Kumudini et al., 2014), they may fail to capture aspects of temperature response. Differing day-length or climate responses may also confound model prediction, with the same cultivar showing different thermal-time requirements, depending on planting date, location, or growth conditions (Piper et al., 1996; Kumudini et al., 2014; Carter et al., 2017). Incorporating the molecular mechanisms of cultivar response in different environments should improve models’ predictive capacity.
A more mechanistic approach would decompose environmental influences into separate model processes (Welch et al., 2003; Kim et al., 2012; Zheng et al., 2013; Brown et al., 2013). One such approach, in wheat, noted that the number of leaves produced before the reproductive transition decreased as the environmental signal’s strength increased (Jamieson et al., 1998b). Prolonged cold, vernalizing temperatures followed by longer days reduced the leaf number at which the transition occurred, while ambient temperature influenced the rate the leaves were produced (Brown et al., 2013). Modeling accumulation of VRN3, a key flowering gene, in response to vernalization and day length cues, and as a function of thermal time, accurately predicted final leaf number and timing of flowering (Brown et al., 2013).
VRN3 is an orthologue of FLOWERING LOCUS T (FT) in Arabidopsis thaliana (Yan et al., 2006), an integrator of environmental cues in the photoperiodic flowering pathway (Song et al., 2015). FT levels correlate strongly with the leaf number present when flowering occurs (Krzymuski et al., 2015; Seaton et al., 2015; Kinmonth-Schultz et al., 2016). In turn, day length, vernalization, and ambient temperature changes regulate FT expression (Blazquez et al., 2003; Amasino, 2010; Song et al., 2015). FT simulated as a function of day length and accumulated as a function of thermal time can accurately predict flowering in some conditions (Chew et al., 2012). Under constant and fluctuating temperature conditions, cool temperatures suppress FT through the interaction of SHORT VETIGATIVE PHASE (SVP) and the FLOWERING LOCUS M (FLM)-β splice variant on the FT regulatory regions (Blazquez et al., 2003; Posé et al., 2013; Lee et al., 2013; Lutz et al., 2015; Sureshkumar et al., 2016; Kinmonth-Schultz et al., 2016). However, temperature fluctuations from warm to cool induce FT through induction of CONSTANS (CO), a chief transcriptional activator of FT (Schwartz et al., 2009; Kinmonth-Schultz et al., 2016). As there is no simple correlation between temperature decrease and FT level reduction, the linear accumulation of flowering gene products with thermal time may not adequately capture the influence of temperature on final leaf number, especially in fluctuating temperatures.
Further, FT protein is expressed in the leaves and moves to the shoot apex where it complexes with FLOWERING LOCUS D (FD) protein to induce the transition from leaf to floral production (Abe et al., 2005; Corbesier et al., 2007). The amount of FT protein perceived at the shoot apex likely depends on the amount of leaf tissue present. Leaf production and growth are strongly temperature dependent (Parent et al., 2010). We proposed that a key mechanism underlying thermal time accumulation could be either the accumulation of gene product (e.g., FT protein) or the increasing capacity for transcript production as a plant grows. In either case, the rate of FT accumulation would be further adjusted by day length and by direct temperature influence on FT gene expression.
To predict whole-plant FT accumulation we must consider changes in FT expression with developmental age. Likely, FT expression is not consistent in all leaves or developmental stages. The transcriptional reporter, pFT:GUS was expressed in the tips of the two true leaves in six-day-old seedlings, but ranged across the leaf in 12-day-old plants having five to seven true leaves (Takada & Goto, 2003). Further, whole-plant transcript levels increase from age five to 15 days relative to an internal control, indicating changing capacity for FT expression with age (Mathieu et al., 2009). FT transcript levels have neither been measured in leaves of different ages, nor has this been considered in flowering models, but it could improve our understanding of how day length and temperature impact FT to control flowering across developmental age.
In earlier work we found FT levels correlate with flowering across a range of temperature conditions ((Kinmonth-Schultz et al., 2016). We also observed that FT can be both induced and suppressed by cool temperatures depending on whether constant or fluctuating temperatures are applied (Kinmonth-Schultz et al., 2016). This provided us with an opportunity to determine the relative influences of FT transcriptional control versus whole-plant FT accumulation via leaf production. One possibility is that despite FT induction by a temperature drop, flowering would be delayed because whole-plant leaf production is slowed. Alternatively, FT induction could result in flowering times that are earlier than predicted. To address these questions we utilized an existing model (The Arabidopsis Framework Model; FM-v1.0) capable of simulating plant growth and flowering times in response to temperature (Chew et al., 2014). We assessed FM-v1.0’s capacity to simulate growth in fluctuating temperature conditions. We then quantified the level of FT produced in leaves of different ages and built new models describing the behavior of FT across leaves and the influence of fluctuating temperatures on FT. We integrated these models into FM-v1.0, linking FT accumulation to leaf tissue production. Using this altered model, referred to as FM-v1.5, we explored the sensitivity of FT accumulation to both gene expression and leaf growth, and demonstrated how each component may influence flowering times. FM-v1.0 used a more traditional thermal-time approach to determine flowering times, whereas FM-v1.5 uses a more mechanistic approach based on FT levels, hence we also compared mechanistic and thermal-time methods of simulating temperature influence on flowering.
Material and Methods
Description of Arabidopsis Framework Model and Modifications
The Arabidopsis Framework Model (FM-v1.0, Figure S1, Chew et al., 2014) combines plant growth and mechanistic flowering regulation for Arabidopsis. FM-v1.0 is run in two phases. In phase one, the timing of flowering is determined by thermal time accumulation (T(t) − Tbase, calculated hourly) in the Phenology module, with daytime temperature given more weight (Wilczek et al., 2009; Chew et al., 2012). Thermal time is modified by day length, to produce Modified Photothermal Units (MPTUs), through mechanistic circadian- and day-length FT transcriptional regulation in the Photoperiodism module (Salazar et al., 2009). The number of days required to reach the MPTU threshold determines the stopping point of vegetative growth and onset of flowering, and is used as an input in phase two. In phase two, the climate parameters affect vegetative growth. Growth is determined by the rate of photosynthesis and carbon partitioning between roots and shoots (Carbon Dynamic module, Rasse & Tocquin, 2006), and includes the rate of organ production as a function of thermal time, including production of individual leaves (Functional Structural Plant module, Christophe et al., 2008). To modify FM-v1.0, we removed the thermal time accumulation used in phase one of FM-v1.0 and instead incorporated mechanistic temperature influence on FT into the Photoperiodism module. We maintained thermal time control over leaf tissue production in phase two, but modified the SLA and respiration components to improve the response of leaf growth to fluctuating temperatures. Then, rather than running the model in two phases, we called the Phenology and Photoperiodism modules at each time step, considering their outputs FT gene expression per unit of leaf tissue. We used the leaf number, age, and area outputs at each time step to determine the relative FT produced by each leaf, and summed the value of FT across all leaves to get a whole-plant FT value. Our modifications (FM-v1.5, Figure 1) are described in detail below.
1. FT transcript accumulation in fluctuating temperatures simulated through SVP and CO influence
Under long days (LD), in 22 °C day, 12 °C night temperature-cycle conditions (22°C /12 °C-night), FT was suppressed at dusk compared to 22 °C constant temperatures (22°C-constant) (Kinmonth-Schultz et al. 2016) likely through the action of SVP and the FLM-β splice variant, consistent with prior observations under constant temperatures (Blazquez et al., 2003; Lee et al., 2007, 2013; Posé et al., 2013). SVP protein levels increased shortly after exposure to cool temperatures (Kinmonth-Schultz et al. 2016), as did the ratio of FLM-β to FLMS-δ splice variants (Posé et al., 2013). FLM-β facilitates SVP binding, and SVP and FLM-β protein levels increase with decreasing temperatures (Lee et al., 2013). Both SVP and FLM-β are present at 23 °C; a transfer from 23 °C to 27 °C resulted in SVP decay that occurred within 12 h (Lee et al., 2013). We used a single term to simulate the combined SVP and FLM-β behavior termed “SVP activity”. Consistent with the observed behavior of these proteins, we modeled SVP activity to increase in response to a decrease in temperature, as shown below.
SVPnew is the newly synthesized protein (nmol/h), VTSVP describes the degree SVP synthesis decreases in response to a temperature increase, the intercept (a) is used to adjust the overall amount of SVP synthesized, T is temperature (°C), and t is time (t = 0 at sowing). The influence of SVP may decline over time, as cool-temperature suppression of FT disappeared over a two-week period (Figure S2a-b). To simulate this, SVPmx declines relative to days post emergence of the first true leaves (dFTL, eq. [1.2], Figure S2c). SVPnew is synthesized every hour, and is input into a differential equation calculated continuously [1.3]. Values and units of each coefficient are in Table S1. To capture the suppression of FT at dusk, we set the SVP decay rate to be slightly lower than its production. This caused SVP to remain higher at 22 °C after a 12 °C night than in 22°C-constant conditions, even after several hours (Figure S2c). The decay rate (vSVP) is proportional to the present SVP concentration.
In LD 22°C /12°C-night, FT levels are higher at dawn coinciding with higher CO mRNA and protein in cool nights (Kinmonth-Schultz et al., 2016). While SVP activity may respond to absolute changes in temperature (Lee et al., 2007, 2013; Posé et al., 2013), CO accumulation is induced by rapid changes from warm to cool (Kinmonth-Schultz et al., 2016). The degree of temperature change is likely a factor, as a drop of 10 °C (22°C /12°C-night) yielded more CO transcript accumulation than did a drop of 5 °C (22°C/17°C-night) relative to 22 °C constant temperatures (Kinmonth-Schultz et al., 2016). This relationship was linear across the three treatments (Figure S3a). We correlated CO mRNA induction (KT) linearly with the difference (dT) between the maximum and current temperatures (eq. [1.4]). To determine dT, the model queries the temperature at each time step, and compares the current temperature against the previous maximum temperature. If higher, the current temperature is set as the new maximum temperature. dT may be zero if there has been no decrease in temperature, and KT cannot fall below zero.
Coefficient c describes the rate at which CO induction changes with dT. The influence of a temperature change fades over several days if the temperature remains cool over that timeframe (Figure S3b). To account for this, ddT is the time (days) since the change in temperature occurred. KT is used to modify the CO mRNA (COm) amount produced (eq. [1.4]), as temperature seems to influence CO through transcription (Kinmonth-Schultz et al., 2016). COm is an input for the CO protein (COp) equation as in Chew et al., 2014, as shown below (eq. [1.5]). Decay occurs only at night (L1 = Light period).
The SVP/FLM-β complex and CO may act competitively at the FT promoter (Bratzel & Turck, 2015), with CO overcoming suppression by SVP/FLM-β at night when its levels are high. The Photoperiod module in FM-v1.0 (Chew et al., 2014) describes the relationship between FT transcription and CO protein (eq. [S2.1]). We incorporated the interaction between CO and SVP/FLM-β using a modified Michaelis-Menten function for competitive inhibition (Segal, 1976), such that the k of FT induction by is influenced by SVP activity as below.
The lower-case v and k are Michaelis-Menten constants either describing the FT synthesis rate as influenced by CO protein (COp) or SVP activity, or FT degradation. CO and FT induction were observed when the temperature dropped both at dawn and dusk (Kinmonth-Schultz et al., 2016), like previous observations (Thines et al., 2014). However, daytime CO induction was lower than nighttime induction while FT induction was higher. The higher daytime CO protein production captured in equation [1.6] was not enough to capture this behavior. While dusk regulation of FT is well understood (Song et al., 2015), the mechanisms governing the morning FT induction sometimes observed (Corbesier et al., 2007) are not known. To capture the observed behavior, we increased FT transcriptional sensitivity in the morning (L2) using a switch function that relied on a model component that peaks around dawn, specifically the circadian clock component, LHY, from the Photoperiodism module, (Figure S4). This enabled us to approximate the observed behavior of FT.
To entrain the diurnal FT and CO patterns, we incorporated data from three different treatment types all in 16-h photoperiods grown at ~60 umol m2 s−1 photon flux density: warm-day (22 °C), cool-night (12 or 17 °C) temperature cycles, in which the temperature change occurred at dusk (24 wild-type replicates, six including 17 °C, and five including the svp mutant line); constant warm (22 °C) temperatures shifting to constant cool (12 or 17 °C) temperatures at dawn (eight and three replicates respectively); and growth at 12 and 17 °C from seed (three replicates each) (Kinmonth-Schultz et al., 2016). In all instances, growth from seed at 22 °C was used as the control. The temperature-cycle harvests including 17 °C spanned two days. An ANOVA comparison of models including and excluding day as a factor, showed no difference. The days were counted as separate replicates for model training. FT and CO gene expression were pooled across all replicates within a treatment and normalized across treatments to the mean peak FT expression (ZT 16) and mean peak CO expression (ZT 16 and 20 mean) in the 22 °C control. Parameter values for change in CO induction and SVP activity over a period of days were determined using experiments with four replicates each (Kinmonth-Schultz et al., 2016). As we were interested in the cumulative influence of FT, we assessed model fit and performance in three ways: (1) minimizing RMSE between observed and predicted gene expression profiles over the 24-h harvest period (14 d after sowing), (2) comparing observed and predicted amounts of CO and FT as calculated as the area under the curve (AUC) 14 d after sowing, and (3) maintenance of gene expression patterns through time.
2. Incorporating FT as a function of leaf and plant age
We found that FT expression declined as leaves aged. Newer leaves in older plants seemed to lose capacity to express FT (Figure 2, S5). To simulate the proportion of FT per unit tissue (FT, nmol cm−2) of each leaf, we used a beta function (eq. [3.1], Yin et al., 1995) based on relative leaf age (r), beginning with the youngest emerged leaf as leaf one. βFT yields a value between zero and one. βFTmx describes the maximum value that can be attained by a leaf of a single plant, Ropt is the relative age at which that maximum value is attained, Rcrit is the oldest leaf that can express FT, and e describes the steepness of the curvature. This function causes the dependent variable to oscillate if the independent variable spans a broad range. To avoid this behavior, we set βFT to be zero below and above the relative ages where βFT first attains a minimum. βFTmx and Ropt are dependent on the total number of leaves on a plant (l), as described below, avoiding the need to reparameterize for plants of different ages. f and g are coefficients.
3. Determining whole-plant FT levels and accumulating FT to a threshold
To link FT transcript accumulation to leaf tissue production, the Phenology module is called at each time step. We consider the output of the Phenology module to be the amount of FT produced per unit leaf area (FT, nmol cm−2). This value is adjusted by leaf area (LA, cm2) and capacity of each leaf to express FT (βFT, unitless modifier), as FT induction is dependent on light intercepted by the leaf.
At each time step, FTleaf (nmol leaf−1) is determined for each leaf, summed across all leaves, and added to the value from the previous time step to determine whole-plant FT levels. Such FT accumulation is consistent with the observation that several days of FT induction are needed to induce flowering (Corbesier et al., 2007; Krzymuski et al., 2015; Kinmonth-Schultz et al., 2016). To predict flowering, the model runs until a threshold level of FT is reached. This threshold is determined by simulating whole-plant FT, at constant 22 °C in LDs, accumulated until a target leaf number is reached. All other treatments are run to this threshold under the assumption that it remains conserved under different growing temperatures.
In FM-v1.0, the development rate towards flowering, as influenced by FT amount and photoperiod, is limited below and above two critical daylengths (10 and 14 h) using a different parameter set for each photoperiod (Chew et al., 2014).
Here, we removed this function and considered direct FT accumulation. Determining the absolute amount of FT required to induce flowering and whether there are threshold levels of transcription, below and above which flowering time is unaffected, will be a useful future study. We maintained the vernalization component from FM-v1.0 to maintain model flexibility, as vernalization should modify overall levels of FT (Helliwell et al., 2006; Searle et al., 2006). This value falls between zero and one and now modifies the levels of FT produced within the Phenology model rather than modifying the thermal unit accumulation rate.
4. Adjusting FM-v1.0 leaf-area response to fluctuating temperature
FM-v1.0 was parameterized for constant temperatures. It captured the leaf areas of plants exposed to different constant temperatures, but simulated larger areas for plants grown in fluctuating temperatures than the constant-temperature control (Figure S6a). Observed plants accumulated similar biomass, but a lower Specific Leaf Area (SLA, m2 g−1) under fluctuating temperatures relative to a constant-temperature control (Pyl et al., 2012). In FM-v1.5, we adjusted the SLA and respiration components to improve the relationship among leaf areas across fluctuating temperature conditions (described below).
The larger leaf area under fluctuating temperatures in FM-v1.0 occurred for two reasons. First, SLA decreases with increasing thermal time (i.e. developmental time, Christophe et al. 2008). In FM-v1.0, this causes simulated SLA to be lower in warmer conditions because development is faster (Figure S6b-c), while all treatments begin at a similar biomass. Second, FM-v1.0 relates maintenance respiration to temperature using the Arrhenius function, causing respiration to be lower under cooler temperatures. Under warm daytime temperatures, plants simulated in fluctuating temperatures accumulate the same amount of stored carbon as the control (Figure S6d). Once shifted to cooler temperatures, the lower maintenance respiration rate (Figure S6e) leaves a larger stored carbon pool that can be used for growth, causing larger leaves.
Respiration, carbon storage, or growth may be altered by temperature in ways not captured in the model. In cold-tolerant woody species, respiration of stem cuttings increased near freezing, rather than following the trend predicted by the Arrhenius function, as did the pool of non-structural carbohydrates (NSC) (Sperling et al., 2015). Respiration may also increase at more moderate temperatures in cases where freezing tolerance is induced, as in Arabidopsis at 16 °C in light with a low red/far-red ratio (Franklin & Whitelam, 2007). In chrysanthemum, cool nighttime temperatures decreased leaf area while increasing dry weight, by increasing stored starch (Heinsvig Kjær et al., 2007). FM-v1.0 does not incorporate these complexities nor consider sinks for carbon other than growth, such as NSCs. Therefore, to simulate the relative relationships in leaf area across temperature conditions needed for our study (Figure S6f), we removed the temperature sensitivity of maintenance respiration and adjusted the Specific Leaf Area (SLA, m2 g−1) to decline with decreasing temperature using observations from Pyl et al. 2012 (Figure S7). A more accurate representation of respiration and carbon pools should be incorporated into future models to improve plant growth predictions in a range of temperature conditions.
Plant growth conditions, RNA expression, GUS tissue analysis, and statistical analysis and experimental controls can be found in the supplemental material.
Results
Behavior of CO and FT transcript accumulation in fluctuating temperatures in FM-v1.5
The FT induction by fluctuating temperatures was incorporated through CO transcript, which was induced in response to a change to cool temperatures like that observed (Figure 3a-b). There was a strong relationship between the amount of simulated and observed CO transcript across treatments, as calculated as the area under the curve (AUC, Figure 3c); although, FM-v1.5 does not incorporate the CO suppression observed when plants are grown at constant 12 °C from seed (12°C-constant) (Kinmonth-Schultz et al., 2016). These model modifications, coupled with increased FT transcriptional sensitivity near dawn, resulted in induction of FT after a temperature drop at dawn or dusk like that observed (Figure 3d-e). Suppression of FT through SVP activity, mimicked the observed FT suppression at dusk. When the SVP influence is removed in FM-v1.5 to mimic an svp mutant, dusk FT suppression in warm-day, cool-night conditions (22°C /12°C-night) disappears as is observed; however, simulated morning induction of FT is higher, perhaps because SVP activity accounts for both SVP and FLM-β (Figure S8). This strong induction through CO was necessary in FM-v1.5 to simulate FT induction by cool temperatures in wildtype.
For flowering to occur, favorable conditions must occur over several days (Kinmonth-Schultz et al., 2016; Krzymuski et al., 2015; Corbesier et al., 2007). Our aim was to approximate FT behavior within a day and through time. Observed FT suppression at dusk in 22/12°C-night conditions occurs by day two of the temperature-cycle treatment (Figure S2a). This is true with FM-v1.5 as well, although FT levels continue to decline until day four relative to the constant-temperature control (Figure S2b). Over two weeks, the increase in dusk FT levels in 22/12°C-night conditions relative to the 22°C-constant control is similar between observed and simulated data (Figure S2a-b). Together, FM-v1.5 can accommodate the wide range in FT transcribed across treatments (Figure 2f), and FT behaves similarly over time to that observed, allowing us to explore the temperature influence on FT expression and flowering in LDs.
Assessment of FT accumulation in FM-v1.5 across temperatures
FM-v1.5 allows us to assess the relative temperature influence on FT accumulation through both gene expression and leaf development. We compared the total FT accumulated 9 days post emergence, equivalent to 1 week in fluctuating temperature treatments, considering 1) influence of temperature on gene expression only (GE), 2) FT accumulated with leaf tissue production as influenced by thermal time, temperature influence on gene expression excluded (LTP), and 3) gene expression changes incorporated with leaf tissue production (LTP+GE, full FM-v1.5 model). The influence of age on a leaf’s capacity to express FT is incorporated into both the LTP and LTP+GE model variants.
When considering LTP+GE, total FT declined, relative to the 22°C-constant control, with increasing exposure times to cool temperature as would be expected from leaf area changes (Figure 4a). When only transcriptional changes were considered (GE), FT accumulated at a faster rate than the control for some treatments (i.e. a drop in daytime temperature, Figure 4b). For treatments in which FT accumulated more slowly than the control, as in 12°C-constant, the relative difference from the control was less extreme than in LTP+GE. For comparison, we explored the relative difference in accumulated MPTUs, which control flowering time in FM-v1.0, over this timeframe. MPTUs across treatments differed to a lesser degree than accumulated FT transcript in LTP+GE, even when nighttime temperatures carried the same weight as daytime temperatures (Figure 4a).
To assess the influence FT transcriptional changes due to temperature have on whole-plant FT levels, we used the LTP model variant, meaning that temperature influenced FT only through leaf production modulated by thermal time. LTP did differ in whole-plant accumulation. Total FT accumulation in the warm-day, cool-night temperature cycle treatments moved closer to that of the control compared to LTP+GE (Figure 4a). When the daytime temperature dropped from 22 °C to 12 °C (22/12°C-day) FT accumulated more quickly in LTP+GE than in LTP.
Assessing capacity of FM-v1.5 to predict flowering
How well can FT accumulation predict flowering? What impacts do transcriptional changes have compared to that of whole-pant FT accumulation? To assess this, we simulated experiments for plants grown in warm-day, cool-night temperature cycles (Kinmonth-Schultz et al., 2016) as plants often experience such temperature fluctuations in nature. We first assumed that FT accumulates to a threshold in a manner like thermal time accumulation. This assumption is consistent with observations that FT induction must occur over a period of days before flowering is induced (Corbesier et al., 2007; Krzymuski et al., 2015; Kinmonth-Schultz et al., 2016). We set the threshold as the value of FT accumulated when plants reached 15 and 8 leaves, which was the nearest whole number to the average leaf number at bolt for Columbia-0 (Col-0) and Landsburg erecta (Ler), respectively, grown in LD 22°C-constant conditions (Kinmonth-Schultz et al., 2016). We maintained the strain-specific parameters for rate of emergence and leaf initiation from FM-v1.0, as they were comparable to our results (Figure 5a), but added a 7-d delay after initiation of the final leaf to improve the fit across strains at 22 °C. This was to account for the time between initiation of the leaf primordia as modeled (Christophe et al., 2008) and growth of a visible bolt, counted when the stem below the bolt head was 2 mm in length (Kinmonth-Schultz et al., 2016).
We then compared the predicted final leaf number and days to bolt for warm-day, cool-night temperature-cycle treatments in the LTP and LTP+GE model variants in FM-v1.5. Cool temperatures delay bolting and increase leaf number (Blazquez et al., 2003, Kinmonth-Schultz et al., 2016). In LTP, we expected that cool-nighttime temperatures would cause flowering to occur at a similar leaf number to the 22°C-constant control because temperature was not influencing gene expression; however, plants would still flower later due to slower whole-plant FT accumulation through slower leaf growth. LTP predicted a trend opposite that observed, with a lower leaf number for both 22/17°C-night and 22/12°C-night treatments (Table 1, Figure 5b), because leaves that are present continue to produce FT such that it accumulates over time as well as with leaf growth. This caused FT to reach the threshold at a lower leaf number. As expected, both 22/17°C-night and 22/12°C-night treatments bolted later than the 22°C-constant control (Table 1). The full LTP+GE variant followed a trend close to that observed, increasing the final leaf number for both cool-night temperature treatments and causing a stronger delay in days to bolt than LTP (Table 1, Figure 5a-c).
We compared this behavior to flowering predicted using MPTU accumulation by FM-v1.0, adjusting the MPTU threshold to our LD 22°C-constant conditions, as recommended (Chew et al., 2014). If FM-v1.0 adequately captured temperature influence, then the MPTU threshold should be similar across treatments, with negligible differences between predicted and observed results for all three temperature regimes. FM-v1.0 predicted fewer leaves in both 22/12°C-night and 22/17°C-night conditions than in the 22°C-constant control, because it reached the MPTU target before reaching the observed final leaf number (Table 1, Figure 5b). FM-v1.0 accurately captured days to bolt for Col-0 and Ler grown in 22°C-constant conditions, and showed an expected delay in days to bolt for both 22/12°C-night and 22/17°C-night. However, the days to bolt were lower than observed (Table 1, Figure 5d). Recalibrating to equalize the influence of nighttime and daytime temperatures (daytime temperatures are given more weight in FM-v1.0 (Chew et al., 2012)) reduced but did not eliminate these trends (Table 1-2, Figure 5e-f). Therefore, incorporating mechanistic FT accumulation can improve model predictions in fluctuating ambient temperature conditions (Table 2).
Influence of FT accumulation in conditions causing later flowering
As later produced leaves may lose the capacity to express FT (Figure 2), we wondered how this would impact FT accumulation and flowering over longer developmental time periods, such as in cool constant temperatures when FT is suppressed and Arabidopsis flowers at a higher leaf number (Blazquez et al., 2003). We grew Col-0 at 12 °C-constant or 22/12°C-day conditions (in the latter treatment, plants then remained at 12°C). We observed flowering at 24 and 28 leaves, respectively, and at 60 and 61 days after sowing. In the full FM-v1.5 LTP+GE variant, FT failed to accumulate to the threshold set in 22 °C conditions (Figure 6). Simulated FT in the LTP variant (temperature influence on FT gene expression removed), did reach the threshold in 22°C/12°C-day conditions (data not shown). FT attained the threshold in 12°C-constant, only after influence of leaf age was removed from the LTP model as well. Therefore, whole-plant FT accumulation, as influenced by leaf age, leaf tissue production, and transcriptional regulation of FT by temperature may not be sufficient to predict flowering in conditions in which FT is strongly suppressed under the assumption of a constant FT threshold.
Influence of short-term temperature fluctuations on FT and flowering
Although long-term exposure to cool temperatures suppressed whole-plant FT and delayed flowering, temperature changes at dawn in LDs (22/12°C-day or 22/17°C-day) caused short-term FT induction (Kinmonth-Schultz et al. 2016). As FT transcript must accumulate over several days before flowering can occur (Krzymuski et al., 2015), we wondered whether a short-term temperature drop, causing FT induction, could complement FT produced in subsequent warm temperatures to accelerate flowering, or if slower whole-plant FT accumulation with slower leaf growth would delay flowering. To compare the predicted influence of FT induction by temperature fluctuations, we used the FM-v1.5 LTP+GE and LTP variants to simulate two-week-old plants moved to 12 °C in LDs for two, four, or six days (12°C-2d, -4d, or -6d), then moved to warm, LD conditions. We also grew plants in these conditions. Control plants were moved directly to warm, LD conditions at two weeks.
Simulating these conditions in the full LTP+GE variant of FM-v1.5, we found little difference in days to bolt between 12°C-2d and the control and a three-day difference between 12°C-6d and the control. There was a decline in leaf number from 15 to 14 leaves in plants exposed to 12°C-2d and 12°C-4d, indicating flowering at a slightly younger developmental age that translated to little difference in days to bolt between the control and 12°C-2d. In 12°C-6d, the leaf number increased again to be like the control. In the LTP variant, the leaf number of all three treatments was the same as the control, whereas there was an increase in days to bolt for each consecutive two-days at 12 °C, consistent with slowed accumulation of FT due to slower leaf growth.
We observed slowed growth (relative to the control) in the cool-temperature treatments. Visible leaf number was significantly lower after four and six days in 12 °C (Figure 7a). On day seven, after completion of all cool-temperature treatments, there was a gradient in leaf area across treatments, with plants from 12°C-6d being the smallest (Figure 7b, S8). We observed a statistically significant delay in the number of days to visible bolt in both 12°C-4d and12°C-6d, like both simulations (P<0.001, Table 3, Figure 7c). While we did not observe a significant difference in leaf number in either 12°C-2d or 12°C-4d relative to the control, plants in 12°C-6d produced approximately 1.5 more leaves before flowering than the other three treatments (P<0.001), more like the predicted increase in leaf number from 12°C-2d and 12°C-4d to 12°C-6d in the LTP+GE model variant (Table 3).
Observed treatments counted significantly different from the control when P<0.05 and the confidence interval of the difference from the control does not contain zero.
Discussion
Incorporating underlying mechanisms could improve model utility for a range of conditions without requiring recalibration (White, 2009; Boote et al., 2013). Here, we found that thermal time (MPTUs) did predict delays in days to bolt under fluctuating temperature conditions in LDs relative to the constant-temperature control, but the delays were less than observed and more like FM-v1.5 LTP, in which FT accumulated only with leaf growth, a function of thermal time (Table 1, Figure 5b & d). Adding direct temperature regulation of FT improved model predictions by increasing the degree of predicted difference between the warm-day, cool-night treatments and the control.
FT was reduced in later-produced leaves (Figure 2). This change in FT expression with developmental age was incorporated into FM-v1.5 using leaf age as a proxy, and caused FT to fail to accumulate to a preset threshold to predict flowering in constant cool temperatures. This finding enables integration of qualitative (presence/absence) and quantitative (dosage response) aspects of FT effects on flowering, and has implications for other conditions in which FT is suppressed, such as in short daylengths. It can help us quantify when FT plays a role during development, when FT alone is a poor predictor of flowering, and when it may act synergistically or competitively with other flowering factors.
For instance, the FT threshold requirement should be influenced by shoot-apex genes; their sensitivity likely changes with climate and developmental age. For example, in short-days, high temperatures may reduce SVP activity at the shoot apex to initiate flowering despite lower FT levels (Fernández et al., 2016). At the shoot apex, SVP suppresses SUPPRESSOR OF OVEREXPRESSION OF CONSTANS (SOC1), which is positively regulated by FT, and which activates LEAFY (LFY), a key player in the floral transition (Schmid et al., 2003; Lee et al., 2008; Jang et al., 2009). FT protein also activates APETALA1 (AP1) at the shoot apex (Lee & Lee, 2010). AP1, in turn, is involved in the down regulation of TERMINAL FLOWERING1 (TFL1), a FT homolog. TFL1 is thought to compete with FT for binding with FD to suppress LFY, as well as AP1, forming a negative feedback loop (Kaufmann et al., 2010; Wickland & Hanzawa, 2015). Both the decrease in SVP and TFL1 would likely decrease the FT threshold needed to induce flowering. Like SVP, TFL1 may be temperature sensitive (Kim et al., 2013).
A changing threshold, due to different LATE FLOWERING alleles in Pea, a homologue of TFL1 in Arabidopsis (Foucher et al., 2003), aids flowering time predictions (Wenden et al., 2009). Incorporating such a mechanism – influenced by climate and developmental age – may aid understanding of how climate influences flowering. As proof of concept, we caused the FT threshold level to change with developmental age (thermal time) (Figure 6). Doing so improved the predictive capacity of FM-v1.5 in constant, cool temperatures.
SVP, in conjunction with FLM, suppresses FT in response to cool temperatures (Blazquez et al., 2003; Lee et al., 2007, 2013). We demonstrated that residual SVP and FLM activity after shortterm cold exposures could be important for FT regulation. For instance, to mimic observed dusk suppression of FT in warm-day, cool-night temperature cycles, simulated SVP activity decayed slowly after at 12 °C night, such that it was higher after 16 hs at 22 °C, than it was in constant 22 °C conditions. Our model also highlights the need to clarify the degree of temperature influence in FT activation and suppression at a range of temperatures. For example, in FM-v1.5, FT is not induced to observed levels, and induction is not maintained as long, after dawn exposure to 17 °C (Figure 3f). It is possible that SVP activation is lower in 17 °C, than predicted from our model. However, the relative difference in transcript levels across treatments is similar to the relative difference in daytime FT expression, which correlates most strongly with flowering (Krzymuski et al., 2015; Kinmonth-Schultz et al., 2016).
Our simulations, while requiring validation in other temperature conditions, are consistent with approaches that use day length and vernalization to influence the leaf number at which the reproductive transition occurs (Brown et al., 2013). However, our work demonstrates that ambient temperature should be incorporated to influence leaf number as well, not only developmental rate. For instance, we altered FT accumulation, either by removing temperature influence on FT transcription (FM-v1.5 LTP, Table 2) or by short-term, cool-temperature exposure (Figure 3d-e, Table 3), affecting final leaf number. In each instance, FT still accumulated with leaf production as influenced by temperature, demonstrating that temperature influences when (in days) the reproductive transition occurs by influencing the developmental rate and whole-plant FT accumulation. We further suggest that tissue accumulation through growth is an underlying factor in the accumulation of thermal time as it causes gene products to accumulate. Together, this work demonstrates that decomposing the influences of climate and development can improve our understanding of plant responses in a range of conditions.
Supplementary Data
Section S1: Materials and Methods for plant growth conditions, RNA expression, GUS tissue analysis, and statistical analysis and experimental controls.
Section S2: Equation used in FM-v1.0 to describe FT transcription as a function of CO protein.
Table S1: Coefficients values for equations used in FM-v1.5.
Figure S1. Graphic representation of FM-V1.
Figure S2. SVP/FLM activity declines over time.
Figure S3. Behavior of CO mRNA in response to different temperature regimes.
Figure S4. Simulated expression profile of LHY, plotted over time used to increase morning transcriptional sensitivity of FT.
Figure S5. The spatial expression profile of FT changes with leaf age.
Figure S6: Behavior of morphological and physiological parameters in FM-v1.0 and v1.5.
Figure S7. Original photograph used for Figure 7 showing Specific Leaf Area (SLA) declines after growth in cool constant temperatures or in warm-day, cool-night temperature cycles relative to a constant, warm-temperature control.
Figure S8. Simulated FT expression profile in FM-v1.5 in the svp mutant mimics the pattern but not relative amplitude of that observed.
Acknowledgements
Many thanks the Jennifer Nemhauser lab for use of their dissecting scope and to Paul Panipinto for help on this project. This work was supported in part by a Cooperative Research Program for Agricultural Science and Technology Development (PJ0127872017), Rural Development Administration, Republic of Korea and a Specific Cooperative Agreement (58-8042-6-097) between University of Washington and USDA-ARS to S.-H.K; by National Institute of Health grant (GM079712), Next-Generation BioGreen 21 Program (SSAC, PJ011175, Rural Development Administration, Republic of Korea) to T.I.; by a Biotechnology and Biological Sciences Research Council awards (BB/L026996, BB/N012348) to A.J.M.; and by the University of Washington Biology Department Frye-Hotson-Rigg Fellowship to H.K.S.
Footnotes
Highlight We examined temperature influence on transcript regulation, organ-specific whole-plant FT accumulation, and flowering time using the Arabidopsis Framework Model. We also quantified FT’s changing systemic interactions throughout development.