Abstract
Natural selection driven by water availability has resulted in considerable variation for traits associated with drought tolerance and leaf level water-use efficiency (WUE). In Arabidopsis, little is known about the variation of whole-plant water use (PWU) and whole-plant WUE (TE). To investigate the genetic basis of PWU, we developed a novel proxy trait by combining flowering time and rosette water use to estimate lifetime PWU. We validated its usefulness for large scale screening of mapping populations in a subset of ecotypes. This parameter subsequently facilitated the screening of water-use but also drought tolerance traits in a recombinant inbred line population derived from two Arabidopsis accessions with distinct water use strategies, namely C24 (low PWU) and Col-0 (high PWU). Subsequent quantitative trait loci (QTL) mapping and validation through near-isogenic lines identified two causal QTLs, which showed that a combination of weak and non-functional alleles of the FRIGIDA (FRI) and FLOWERING LOCUS C (FLC) genes substantially reduced plant water-use without penalising reproductive performance. Drought tolerance traits, stomatal conductance, intrinsic water use efficiency (δ13C) and rosette water-use were independent of allelic variation at FRI and FLC, suggesting that flowering is critical in determining life-time plant water use, but not leaf-level traits.
Introduction
Water availability is essential for the optimal allocation of resources to achieve maximal growth and reproductive fitness (Anderson 2016). Consequently, a water deficit may force survival trade-off costs resulting in reduced reproductive fitness (Von Euler, Ågren & Ehrlén 2014; Sletvold & Ågren 2015). In natural populations, adaptations to water deficits encompass a number of different ecological strategies that include drought escape and avoidance leading to drought resistance. While drought escape is characterised by rapid growth and early flowering to reproduce before the onset of terminal drought, avoidance limits growth during periods of dehydration through lowering stomatal conductance and transpiration (Ludlow 1989; Kooyers 2015). Drought resistance traits, characterised by the ability to survive a water deficit, have traditionally been used to assess plant performance under reduced water availability. However, the usefulness of drought resistance as a trait to optimize plant productivity has been questioned, as the improvement of various drought resistance related traits has been demonstrated to reduce productivity under some circumstances, regardless of the ability of plants to survive the period of drought stress (Blum 2005, 2009; Passioura 2007). It is widely accepted that drought resistance facilitates plant survival, but it does not contribute towards the maintenance of yield following drought stress or in water replete conditions (Blum 2005, 2009; Passioura 2007). The identification of plant varieties that are able to produce stabilized or improved yields with reduced water inputs is therefore an important goal for plant breeders, physiologists and molecular biologists alike (Parry, Flexas & Medrano 2005; Morison, Baker, Mullineaux & Davies 2008).
Water use efficiency at the leaf level is the net amount of CO2 fixed per unit of transpired water, hereafter referred to as instantaneous water-use-efficiency (WUEi, A/E) (Condon et al. 2004; Table 1). It relates equally to water loss by transpiration and net carbon gain achieved via gas exchange (Long, Marshall-Colon & Zhu 2015). Alternatively, carbon isotope composition (δ13C; Table 1), as an estimator of intrinsic water use efficiency, that is the ratio of net CO2 assimilation to stomatal conductance for water vapour (A/gs; Farquhar & Von Caemmerer 1982; Farquhar et al. 1989), are regularly used to describe integrated leaf level intrinsic water use efficiency and have been targeted in several studies as a primary trait to achieve “more crop per drop” as well as enhancing drought resistance (Morison et al. 2008; Blum 2009).
The value of leaf level water use efficiency estimates for improving crop yield has previously been questioned. For example, it has been shown that despite the association between δ13C and water use efficiency in many species (Farquhar et al. 1989), its relation to yield across multiple environments and genotypes is much often variable (Condon et al. 2004). This suggests that both additional intrinsic plant factors, as well as environmental conditions, impact the relationship between intrinsic water use efficiency and agronomic water use efficiency, i.e. the amount of yield produced per unit of water transpired. Therefore, leaf level intrinsic water use efficiency estimates may not be a useful proxy to select for yield under water limited conditions. This lack of consistent upscaling from leaf-to whole-plant water use efficiencies may be a product of the heterogeneity of net CO2 assimilation rates within and across individual photosynthetic organs or it may also be due in part to the lack of integration of night-time transpiration and plant respiration rates in leaf-level WUE measurements (reviewed in Cernusak, Winter & Turner 2009; Cernusak et al. 2013). Furthermore, this inconsistency may be related to changes in environmental conditions leading to variations in other processes that affect CO2 supply and demand (Seibt, Rajabi, Griffiths & Berry 2008; Medrano et al. 2015). In addition, discrepancies may occur due to genotypic variation in carbon isotope signatures of crop plants being often driven by variation in stomatal conductance (Blum 2005; Monclus et al. 2006; Monneveux, Sánchez, Beck & Edmeades 2006; Marguerit et al. 2014), thereby limiting carbon assimilation and productivity. It should be noted, however, that in some species variation in δ13C has also been attributed to variation in carbon fixation as well as stomatal conductance (Masle, Gilmore & Farquhar 2005; Donovan, Dudley, Rosenthal & Ludwig 2007; Brendel et al. 2008).
Investigating the natural variation in whole-plant water-use efficiency and the mechanisms of drought resistance in natural populations is challenging, due to difficulties in re-creating realistic drought conditions in an experimental setting. For example, in short-dehydration experiments (Bechtold et al. 2010, 2016; Ferguson et al 2018), water loss is greater in larger plants creating substantial heterogeneity in the timing of water deficits (Kooyers 2015). While plant size greatly contributes to water use in Arabidopsis, drought response traits are independent of the transpiring leaf surface (Ferguson et al. 2018). This suggests that above ground biomass impacts water use and consequently whole-plant water-use-efficiency, but not necessarily drought tolerance. Central to the determination of whole-plant water use efficiencies, such as transpiration efficiency (TE, here ratio between aboveground biomass and transpired water; Table 1) or water productivity (WP, here ratio between seed biomass and transpired water; Table 1), is the quantification of water lost by the plant. We have previously shown that leaf-level WUE is not representative of absolute vegetative (rosette) water use (VWU), or biomass production (Ferguson et al. 2018), as the transpiring leaf surface is a major upscaling factor. We also demonstrated in a few selected ecotypes and mutants that differences in life-time plant water use (PWU, Table 1) and plant-level water-use efficiency (TE and WP) exist (Bechtold et al. 2010, 2013). This is an important consideration since water use is maintained for an extended period following floral transitioning (Bechtold et al. 2013), however little is known about the underlying molecular mechanisms of the variation in PWU and TE/WP. In Arabidopsis, the measurement of life-time PWU has received little attention, mainly due to the difficult and time consuming nature of manually phenotyping PWU on a daily basis for the majority of the lifetime of the plant (Bechtold et al. 2010, 2013). As plants begin to develop stalks and flowers, automated watering systems (Granier & Tardieu 2009; Tisné et al. 2013) would cause considerable disturbance of the tall structures. Conversely, non-conveyor belt platforms (Halperin, Gebremedhin, Wallach & Moshelion 2017), or a manual approach involving careful handling of flowering plants limits the potential for harmful effects occurring due to movement and touch induced changes (Van Aken et al. 2016). From limited studies of this nature, the C24 ecotype has emerged as drought tolerant and highly water use efficient (Bechtold et al. 2010), additionally it demonstrates resistance to numerous abiotic and biotic perturbations (Brosché et al. 2010; Lapin et al. 2012; Xu et al. 2015; Bechtold et al. 2018).
Our recent study of 35 Arabidopsis ecotypes confirmed the above-described uniqueness of C24 in uniting several desirable water use and drought response traits (Ferguson et al. 2018). To build upon these findings, we set out to ascertain whether PWU of C24 was reduced compared to other ecotypes and whether this had a heritable and genetically discernible basis. We therefore employed a C24 x Col-0 recombinant inbred line (RIL) population (Törjék et al. 2006) to identify QTLs that underlie the natural variation of these traits. However, due to the difficulties of manually phenotyping PWU, development of a suitable proxy trait was required to phenotype the mapping population in a high-throughput manner. Arabidopsis represents an ideal system through which to develop and evaluate the usefulness of proxy traits, such as WUEi, δ13C, flowering time, VWU and biomass parameters for predicting PWU and whole-plant water-use efficiencies. To this end, we assessed the usefulness of this suite of traits for acting as proxies to predict whole-plant water-use efficiencies (TE and WP, see Table 1) in a set of 12 summer annual ecotypes. A highly accurate proxy trait was subsequently identified and employed in a forward genetic screen for whole-plant water use traits.
Materials and methods
Plant material and plant growth
A selection of 12 facultative summer annual Arabidopsis thaliana (Arabidopsis) ecotypes (Table S1) and 164 RILs derived from a cross between ecotypes Col-0 and C24 (Törjék et al., 2006) were employed to assess the natural variation of long-term plant water use. The genetic map and genotype information for the RIL population are as described in Törjék et al., 2006 (Table S2). The Col-0 x C24 RIL mapping population was used to identify QTL relating to key traits associated with water use. Detected QTL regions of interest were further investigated using near-isogenic lines (NILs) that captured Col-0 alleles in a homogenous C24 genomic background and vice versa (Törjék et al. 2008). The ecotypes, RILs, and NILs were phenotyped for water-use (VWU and PWU), flowering time, and above ground biomass parameters. Additionally, the 12 ecotypes and NILs were phenotyped for δ13C (Fig. 1). Plants were sown in peat-based compost (Levington F2+S, The Scotts Company, Ipswich, UK.) and stratified at 4°C in darkness for 4 days. After stratification plants were grown in a growth chamber at 23°C under short-day (8h:16h; light:dark) conditions, under a photosynthetically active photon flux density (PPFD) of 150 ± 20 μmol m-2 s-1, and at 65% RH (VPD of 1kPa, Fig. 1). Plants were transferred to the glasshouse at distinct stages depending on the applied watering regime (see below and Fig. 1). Within the glasshouse, the environmental conditions were variable, as temperature and external light cycles fluctuated during the experimental periods. Supplemental lighting was maintained at a minimum PPFD threshold of ~200 μmol m-2s-1 at plant level for a 12h day (long day conditions). Plants were watered according to the different watering regimes (see Fig. 1), and their positions within the two growth environments (short- and long day) were changed daily. In this study, we deliberately opted for transitions between short- and long-day conditions (growth chamber to glasshouse) without a vernalization period, which resulted in delayed flowering compared to some studies. This decision was taken as physiological measurements (snapshot measurements for WUEi) required a minimal rosette size that would normally not be achieved in vernalized plants.
Watering regimes
(i) Short term dehydration experiment for the determination of vegetative water use (VWU)
All lines undergoing a short-dehydration experiment were grown in the growth chamber in 6cm diameter (0.11L) pots for the determination of vegetative rosette water-use (VWU) as described in (Ferguson et al. 2018). Briefly, at 50 days post sowing, plants were left to progressively dry to 20% relative soil water content (rSWC), at which point they were re-watered and transferred from the controlled environment room to the glasshouse for flowering time determination and seed production. VWU was calculated as the slope of the linear regression of the rate of drying from 95 – 20% rSWC (lasting between 10 – 12 days; Fig. 1a, Table 1). Plants were transferred to the glasshouse after re-watering and maintained well-watered to determine flowering time and the number of rosette leaves at bud initiation. Plant biomass components were separated and measured as rosette biomass (vegetative biomass), chaff biomass (stalks and pods; reproductive biomass), and seed yield (reproductive biomass), and the sum of all biomass components produced the total above ground biomass value. PWU was calculated as VWU multiplied by the time it took from germination to flowering to generate calculated lifetime plant water use (cPWU, Table 1). WP was calculated as seed biomass divided by either calculated or measured life-time water use (cWP or mWP, Table 1). This watering regime is designated as short day (SD), as plants spend most of their life time under short day conditions (~65 days)
(ii) Continuous maintenance of moderate drought for determination of life-term plant water use (PWU)
For the determination of PWU, 8-cm diameter (0.3L) pots were filled with the same volume of soil following the experimental setup as described in Bechtold et al. (2010). The soil surface was covered with 0.4 cm diameter polypropylene granules to limit soil evapotranspiration. Plants were germinated in the previously described growth chamber before being transplanted into individual pots 12 days after sowing at the initiation of the rosette growth stage (Boyes et al. 2001). Four days after being transferred into individual pots, plants were moved into the glasshouse, where pots were weighed daily (Kern PCB, 350-3 balance) to determine and maintain the pots at a moderate drought level of 40% rSWC (Bechtold et al. 2010). Daily water use was recorded after plants were transferred into the glass house. Control pots without plants were also measured daily to estimate evaporation from the soil surface. Flowering time and number of leaves at bud initiation were recorded, and once the final flower had opened, watering ceased, and plants were bagged for harvesting. During harvest the vegetative (rosette) and reproductive (stalks, pods, and seeds) biomass components were separated. mPWU was determined as the sum of water added every day until bagging, minus the water lost through evaporation from control pots. This parameter is also termed measured plant water use (mPWU) in order to distinguish it from cPWU (Table 1). This watering regime is denoted as long-day (LD), as plants only spend 16 days from germination under short day conditions, the remaining time plants were grown under long day conditions (Fig. 1b).
Estimating drought sensitivity (DS)
For analysing in more detail the data used for calculating VWU, we applied the Davies test (Davies, 2002) and segmented regression analysis as part of the segmented package in R (Muggeo 2017) in order to test (i) for a significant difference in slope parameter and (ii) for the breakpoint in the regression. This analysis produced the breakpoint in the drying period and the slopes before (stage 1) and after (stage 2) the breakpoint. VWU plasticity was calculated as the slope before the breakpoint (stage1; supposed to represent transpiration under control conditions) - slope after breakpoint (stage2; supposed to represent transpiration under drought conditions) / slope before breakpoint (stage1). Both breakpoint (in terms of rSWC) and VWU plasticity were used to estimate the drought sensitivity as per Ferguson et al. (2018).
Physiological measurements
(i) Photosynthetic rate (snapshot measurements) in the short dehydration experiment Instantaneous measurements of net CO2 assimilation rate (A), and stomatal conductance to water vapour (gs) and transpiration rate (E) were taken on leaf 7, using an open gas exchange system (PP Systems, Amesbury, MA, USA). Leaves were placed in the cuvette at ambient CO2 concentration (Ca) of 400 μmol mol-1, leaf temperature was maintained at 22 ± 2 °C and vapour pressure deficit was ca. 1 kPa and irradiance was set to growth conditions (150 μmol m-2 s-1). A reading was recorded after the IRGA conditions had stabilized (ca. 1.5 min), but before the leaf responded to the new environment (Parsons, Weyers, Lawson & Godber 1997). Instantaneous water use efficiency WUEi was estimated as A/E.
(i) Delta carbon 13 analysis
The carbon isotope composition (δ13C) of bulk leaf material was assessed for the 12 ecotypes comprising the SD experiment (well-watered samples), and the NILs and parental lines from the continuous moderate drought experiment. The harvested leaves had developed during moderate drought stress (40% rSWC). δ13C was measured as described in Roussel et al. (2009) and Ferguson et al. (2018). δ13C was calculated as: (Rs – Rb)/Rb x 1000, where Rs and Rb represent the 13C/12C ratio in the samples and in the Vienna Pee Dee Belemnite standard respectively (Craig, 1957).
Statistical Analysis
All statistical analyses were performed within the R software environment for statistical computing and graphics (R Core Team 2015). Experiments using the RIL population were performed across several blocks over a period of two years. Each temporally divided block contained the two parental ecotypes and between 20-40 RILs. One-way analysis of variance (ANOVA) comparison of means tests were performed across all lines and all blocks to determine the existence of experimental block effects that could potentially confound further analysis and the QTL mapping. Best linear unbiased predictors (BLUPs) were extracted using the following general linear mixed model (GLMM): Y = E + B + Residual (Error) variance, where Y represents the phenotypic trait parameter of interest, and both E (Ecotype) and B (Experimental block) are treated as random effects, while controlling for fixed effects, i.e. temporal block effects (Lynch & Walsh 1998). Predicted means were obtained for each trait and for each RIL by adding the appropriate BLUP value to the population mean. Predicted means were employed for all subsequent analyses involving the RILs and for QTL mapping. The GLMMs allowed for the determination of phenotypic (VP) and genotypic (VG) variation for all trait parameters. These parameters were used to obtain estimates of broad sense heritability (H2) as VG/VP.
QTL Mapping
We mapped for QTLs underlying all assessed parameters using the qtl R package (Broman et al. 2003; Broman 2009). The Lander-Green algorithm (Lander & Green 1991) i.e. the hidden Markov model technology, was used to re-estimate the genetic map using the Kosambi map function to convert genetic distance into recombination fractions with an assumed genotyping error rate of 0.0001. The re-estimated genetic map, based on the lines incorporated in this study, was preferred to the original genetic map, which was based on over 400 RILs. The hidden Markov model technology and Kosambi map function were further employed to calculate the probabilities of true underlying genotypes at pseudo-marker points between actual markers based on observed multipoint marker data, whilst allowing for the same rate of genotyping errors. Genotypes were calculated at a maximum distance of 2 cM between positions.
Multiple QTL mapping was performed using the predicted means derived from BLUPs. The best multiple QTL models were identified via the Haley-Knott regression method (Haley & Knott 1992) using genotype probabilities at both genetic markers and calculated pseudo-markers. We used the Haley-Knott regression method because the genetic map marker density was relatively high (Average inter-marker distance: 3.87 cM). We additionally tested the expectation-maximization and multiple-imputation methods for QTL mapping but did not notice a perceptible difference in QTL effect size or position between these three methods (Fig. S4).
10,000 permutations were used to determine LOD significant thresholds for incorporating both additive QTL and epistatic interactions at an experiment-wise α = 0.05. Automated stepwise model selection was performed, where additive and epistatic QTL were scanned for at each step (Manichaikul, Moon, Sen, Yandell & Broman 2009). The penalties for the stepwise model selection were derived from a two-dimensional genome scan. Finally, the positions of detected QTLs were refined, and the model was fitted with ANOVA to calculate the effect size, percentage variance explained, LOD score for each QTL, and penalized LOD score for each model. Interval estimates of all detected QTLs were obtained as 95% Bayesian credible intervals.
Following MQM, the log10 ratio comparing the full model and the single QTL model from the two-dimensional genome scan was directly assessed to test for the presence of an epistatic interaction between the two main effect QTL for cPWU. To assess the impact of flowering time, vegetative biomass, and VWU on the QTL mapping for cPWU, we performed single-QTL mapping for cPWU using these traits as individual covariates.
Genotyping using InDel markers
Insertion-deletion markers (InDels) polymorphic between Col-0 and C24 alleles of FRI and FLC were obtained to address the hypothesis that these genes underlie the two major QTLs detected. A 16-bp deletion in the Col-0 allele of FRI was scored using primers developed by Johanson et al. (2000). A 30-bp deletion in the Col-0 allele of FLC was scored using primers developed by Gazzani et al. (2003). InDel markers with a single PCR band for both InDels (Fig. S1a; Table S3) were assayed by qPCR and high-resolution melting (HRM) genotyping using the CFX96TM Touch Real-Time PCR Detection System (BIO-RAD). This information for 138 individuals of the RIL population and both parents was subsequently integrated into the re-estimated genetic map (Fig. S1b, Table S4).
Analysis of publicly available RNAseq and microarray datasets
Publicly available RNAseq (Xu et al. 2015; GSE61542) and microarray datasets of C24 and Col-0 (Bechtold et al. 2010, E-MEXP-2732) were analysed for differentially expressed genes. These datasets were compared with the protein coding genes within mapping intervals using VENNY (Oliveros 2007).
RNA extraction and gene expression analysis by qPCR
Leaves of a minimum of four biological replicates were harvested from the NILs and both parental lines at 26 and 43 days post germination, and frozen in liquid nitrogen. Total RNA was extracted using Tri-reagent (SIGMA, Aldrich, UK) according to the manufacturer’s instructions. For cDNA synthesis, 1 μg of total RNA was treated with RNase-free DNase (Ambion) according to manufacturer’s instructions and reverse transcribed as previously described (Bechtold et al. 2008). Quantitative real-time PCR (qRT-PCR) was performed using a cybergreen fluorescence based assay as described previously (Bechtold et al. 2008). Gene-specific cDNA amounts were calculated from threshold cycle (Ct) values and expressed relative to controls and normalized with respect to Actin and Cyclophilin cDNA according to Gruber, Falkner, Dorner & Hämmerle (2001). To calculate the standard error of the calculated ratios of fold differences for gene expression data, the errors of individual means were combined "in quadrature", and the final ratio was a combination of the error of the two-different means of the NILs and Col-0 samples. The primers used for RT-qPCR can be found in Table S3.
Results
We used a selection of 12 facultative summer annual ecotypes of Arabidopsis that previously demonstrated variation for drought sensitivity and water use associated traits (Table S1, Ferguson et al. 2018), as well as a RIL mapping population and associated NILs (BC4F3-4) to examine natural variation of PWU and above ground biomass allocation (Table S2, Table S5). The assessment of natural variation for VWU, PWU, biomass accumulation and drought sensitivity was followed by QTL mapping to establish the genetic basis of these traits. Two experimental setups were used as part of this study: (i) 12 ecotypes and RILs - a short dehydration experiment under predominantly short-day conditions to measure a range of leaf level WUE parameters (WUEi, δ13C), VWU, flowering time, biomass parameters and drought sensitivity (Fig. 1a, Ferguson et al. 2018), and (ii) 12 ecotypes and NILs - a continuous moderate drought experiment under predominantly long-day conditions, during which rSWC was maintained at moderate drought levels (~40% rSWC) to measure leaf level WUE parameters (δ13C), VWU, PWU, flowering time and biomass parameters (Bechtold et al. 2010; Fig. 1b).
Identification of a proxy trait for lifetime (plant) water use (PWU)
We analysed a range of parameters associated with plant water status by performing a short dehydration as well as a continuous maintenance of moderate drought experiment on 12 selected Arabidopsis ecotypes (Fig. 1, Table 1). We determined VWU (Ferguson et al. 2018; Fig. 1a, Table 1), lifetime PWU (Fig. 1b, Table 1), flowering time, above ground biomass parameters, δ13C, and calculated whole-plant water-use efficiency parameters, namely TE and WP (Table1; Fig. 1a, b; Bechtold et al. 2010, 2013, 2016; Ferguson et al. 2017). Both δ13C and WUEi measurements were taken to determine the influence of leaf-level processes on whole plant traits (i.e. transpiring leaf surface area), however we did not observe a significant relationship with whole-plant water-use efficiency parameters such as TE and WP (Fig. S2). We continued to focus on the determination of lifetime PWU and the genetic dissection of PWU and productivity traits, instead of the leaf-level WUE parameters, δ13C and WUEi.
Our usual approach of a manual determination of PWU (Fig. 1b) requires the weighing and watering of individual pots until the terminal flower has opened (Bechtold et al. 2010). The manual determination of PWU is challenging and time consuming (see Introduction), thus to facilitate large-scale manual screening of PWU of the mapping population, we first set out to identify an adequate proxy. We compared biomass production, flowering time, VWU and PWU between the short-dehydration and continuous moderate drought experiment carried out on the 12 Arabidopsis ecotypes (Fig. 1). The continuous moderate drought experiment revealed that measured PWU (mPWU) was significantly correlated with both flowering (Fig. 2a) and vegetative (rosette) biomass (Fig. 2b, Table S6). Based on these relationships we developed the proxy parameter “calculated life time (plant) water-use (cPWU)”, as a product of VWU and flowering time:
The continuous moderate drought experiment allowed us to directly relate mPWU with cPWU, which showed a highly significant positive correlation within the experiment (Fig. 2c). In addition, the correlation between mPWU with cPWU was tighter than the correlations with rosette biomass and flowering time (Fig. 2a, b). Importantly, a significant correlation between calculated and measured PWU was also observed when comparing mPWU from the continuous moderate drought experiment under long-day conditions, with cPWU of a short-dehydration experiment under short-day conditions (Fig. 2d). Therefore, we reasoned that PWU calculated from flowering time in a short-dehydration experiment would provide a robust estimate of mPWU.
Furthermore, the short-dehydration approach allowed us to quantify the drought responses of individual ecotypes by calculating the threshold at which plants enter drought stress (breakpoint) and the plasticity of the drought response (VWU plasticity; Ferguson et al. 2018). The breakpoint negatively correlated with the VWU plasticity, indicating that lines responding to drought stress at higher rSWC, showed less absolute change in transpiration throughout the dehydration period, and therefore exhibited reduced VWU plasticity (Fig. 2e). Therefore, a short dehydration experiment allowed us to not only screen and dissect the genetic basis for the natural variation of cPWU and biomass, but also assess drought response parameters at the same time.
The genetic dissection of cPWU, drought response and biomass parameters
Short dehydration experiments (Fig. 1a) were subsequently performed on 163 individuals of the Col-0 x C24 RIL population (Table S2) including both parents. To control for experimental block effects, BLUPs were extracted and predicted means were calculated for all traits. The variation in predicted means for all traits was not significantly different from what would be expected of a normal distribution (P > 0.05; Kolmogorov-Smirnov normality test) and all traits demonstrated transgressive segregation (Fig. S3). We calculated genetic variance (VG), total phenotypic variance (VP) and broad sense heritability (H2), where all 13 traits assessed demonstrated variation that had a significant heritable basis within the RIL population (Table 2).
Adjusted linkage maps were constructed based on the individuals used for mapping. Analyses indicated that 97.5% of the markers had been genotyped for all the RILs, and we observed a virtually even split in the allelic form of these markers, with 50.3% coming from the Col-0 parental line and 49.7% from the C24 parental line. To identify the genetic variation that causes the observed phenotypic variation in VWU, cPWU, flowering time, productivity and drought sensitivity traits, multiple QTL mapping was performed (MQM; see Material and methods) on a minimum of 163 selected individuals. No significant QTL models were identified for seed biomass (Fig. S5a), dehydration response (VWU plasticity; Fig. S5b), and the breakpoint (Fig. S5c). For VWU, FT, cPWU and slope 1, a total of 9 main effect QTLs were detected (Fig. 3; Table 3; p <0.001). The percentage of phenotypic variance explained for the cPWU QTLs ranged from 4.8 to 25.5%, for flowering time from 3.6 to 18.2 %, and for VWU from 3 and 5% (Table 3). Consistent with the previously observed significant correlation between flowering time and mPWU (Fig. 2a), there was co-localization between the two main effect QTLs on chromosomes 4 and 5 for flowering time and cPWU (Fig. 3a, b; Table 3). The strong positive correlation observed between flowering time and PWU suggests that the co-localizing QTLs for these traits were likely to represent the same genes or linkage between causal genes. On the other hand, QTLs detected for VWU did not co-localise with flowering time QTLs (Table 3, Fig. 3).
When performing single QTL-mapping for cPWU whilst incorporating flowering time as a covariate in the analyses, the main effect QTL on chromosomes 4 and 5 are not detected, however the QTL on chromosome 3 that is also detected when mapping for VWU becomes more significant (Fig S6c). Similarly, when incorporating vegetative biomass as a covariate, the effect of these QTL is reduced, however they are still significant (Fig S6b). Incorporating VWU as covariate, removes the importance of the QTL on chromosomes 3 and heightens the significance of the QTLs on chromosomes 4 and 5 (Fig. S6d).
The two significant cPWU and flowering time QTLs on chromosomes four and five (Fig. 3a, b) contained two well characterised flowering time genes, FRIGIDA (FRI, Chromosome 4; AT4G00650) and FLOWERING LOCUS C (FLC, Chromosome 5; AT5G10140). The ecotype Col-0 possesses a non-functional allele of FRI (fri) and a functional allele of FLC (FLC), while the ecotype C24 contains a functional allele of FRI (FRI) and a weak allele of FLC (flc; (Johanson et al. 2000; Michaels, He, Scortecci & Amasino 2003). A significant epistatic interaction was detected between these QTLs when comparing the full model to the model which just accounts for one of these two QTL (Fig. S7). Transcriptional levels of FLC are positively regulated by FRI (Deng et al. 2011), thus the epistatic interaction between these QTL further suggests that FRI and FLC are the causal genes. InDel markers were designed for both candidate genes and the RIL population was scored for the allelic variant of both genes (see Materials and Methods). This information was incorporated into the genotypic data and the genetic map was re-estimated, which demonstrated that FRI and FLC were present between the markers that flanked the main effect QTLs on chromosomes four and five respectively (Fig. S1b). The RIL population was sub-divided according to the different allelic combination of FRI and FLC of each individual line (Table S4) to confirm the importance of the functionality of these genes on the traits of interest here.
The genetic action of non-functional and weak alleles of FRI and FLC reduces water-use
We determined the allelic state of FRI and FLC in all RILs and divided the population into four groups: i) fri: FLC (Col-0), ii) FRI: FLC, iii) fri: flc, and iv) FRI: flc (C24). One-way ANOVA comparisons of means and post-hoc Tukey tests were performed to determine the effect of different allelic combinations on water use and plant development (Fig. 4). There were significant and parallel differences in cPWU and flowering time between the four groups (Fig. 4a, b). Possessing non-functional and weak alleles of FRI and FLC, respectively, significantly reduced flowering time and cPWU (Fig. 4a, b).
To further test the hypothesis that cPWU is a suitable proxy of mPWU and to confirm that increased life-span through a combination of FRI and FLC is the main factor underlying PWU, we subsequently obtained NILs that harboured the Col-0 allele of FRI and FLC separately in a homogenous C24 genomic background and vice versa (Table S5). Seven NILs and two parental lines were subjected to a continuous moderate drought experiment, where flowering time, mPWU, VWU, cPWU, productivity parameters, mean daily water use as well as δ13C and stomatal conductance were determined (Fig. 1b). The hypotheses regarding cPWU that emerged from the RIL population were essentially confirmed. The combination of both non-functional and weak alleles of fri (Col-0) and flc (C24) led to significantly reduced mPWU (Fig S8a), but had no impact on VWU (Fig. S8c,d).
Due to the significant relationship between flowering time and mPWU (Fig. 2a), we assessed whether the different allelic combinations of FRI and FLC had pleiotropic effects on VWU. There was no significant difference in VWU in both the NILs and RILs under either short-(RILs) or long-day (NILs) conditions (Fig. S8c,d).
Interestingly we observed a significant relationship between mean daily water-use, days to flowering and rosette biomass in the moderate drought experiments for the 12 ecotypes and the NILs (Fig. S9a,b; Fig 5a,b), leading to high mPWU (Fig. S9c; Fig. 5c). Therefore, late flowering ecotypes and NILs appear to sustain increased daily water use over a longer period, which was independent of the allelic combinations of FRI and FLC (Fig. 5d).
δ13C, while significantly different between Col-0 and C24, did not show a significant difference among the remaining allelic combinations of FRI and FLC (Fig. S10a), which suggests that δ13C was independent of FRI and FLC. A significant negative correlation between δ13C and stomatal conductance indicated that low gs leads to increased instantaneous WUE (A/gs) (gs; Fig. S10b; R2 = 0.781 p < 0.01), which also coincided with the distinct rosette growth phenotype of C24 (Fig. S10b,d). In addition, the lack of significant QTLs for VWU, VWU plasticity and the breakpoint (Fig. S5) suggests that leaf-level drought responses were not genetically controlled in this mapping population, and therefore independent of the detected genetic control of flowering time. This was confirmed by the non-significant differences in VWU, VWU plasticity and breakpoint for the four allelic FRI/FLC groups (Fig. S8c,d; Fig. S11a,b).
Importantly, the observation that a combination of fri (Col-0) and flc (C24) in the NILs led to significantly reduced mPWU (Fig S8a), and significant variation in δ13C (Fig S10a) that did not match the variation for mPWU, support our observations from the diverse suite of ecotypes. Taken together, this suggests that cPWU is a reliable proxy for mPWU.
Biomass variation and distribution is independent of the genetic action of FRI and FLC, and growth conditions
We also assessed whether the different allelic combinations of FRI and FLC resulting in significantly different PWU, had pleiotropic impacts on biomass parameters. For example, the decrease in cPWU in the fri: flc group did not result in a significant reduction in above ground-, seed- or vegetative biomass in the RILs (Fig. 6a-c) or the NILs (Fig. S12a-c), yet the combination of FRI:FLC significantly decreased seed- and increased vegetative biomass (Fig. 6b,c; Fig. S12c). This suggests that the additionally acquired photosynthates acquired by later flowering plants are translocated primarily to vegetative as opposed to reproductive sinks.
Biomass allocation (HI) showed substantial variation amongst the NIL and the RIL populations (Fig. S13a, b), due to different experimental conditions (short day vs long day, well-watered vs moderate drought). Despite these experimental differences, relative proportions were highly correlated between the well-watered and moderate drought experiments (Fig. 7), suggesting allelic combinations with low HI in the short-dehydration experiments (RILs) also showed low HI in the continuous moderate drought experiment (NILs; Fig. 7a,). Equally, cPWU significantly correlated across the distinct experiments for the different allelic groups (Fig. 7b). A similar relationship for PWU and HI across different experiments was also observed in the 12 natural ecotypes (Fig. 2d, Fig. 7c). This suggests that the distribution of biomass and PWU was independent of environmental growth conditions including watering status and day length in both the mapping population and the natural ecotypes.
Gene expression
The detected QTL regions contained many genes, as such we explored gene expression differences between the two parents within the mapping intervals for all three mapped traits. This was achieved using a publicly available microarray experiment comparing C24 and Col-0 (Bechtold et al. 2010) and RNAseq data of both parental accessions (Brosché et al. 2014). In total 9906 protein coding genes were identified within the 95% Bayesian credible intervals on chromosomes 4 and 5 (Table 3), of which 304 showed differential expressions between Col-0 and C24 (Tables S8, S9). We randomly selected three to four differentially expressed genes (up and down) for each interval, whilst also including FRI, FLC and FLOWERING LOCUS T (FT; Chromosome One) for analysis of gene expression in the NILs and both parental lines (Table S10) at 26- and 48-days post germination.
Early studies have shown that FRI up-regulates FLC expression in ecotypes that have the active allele of FRI (Michaels & Amasino 1999; Sheldon et al. 1999). NILs carrying the C24 FRI allele (Table S5) showed elevated FLC expression at 26- and 43-days post germination in plants grown under short-day controlled environment conditions (Fig. 1, 8a). Variation in FLC and FRI expression at 43 days post germination showed a significant association with flowering time and mPWU (Table S11), which was independent of FT expression (Table S11). This is in line with QTL mapping results where a significant association of the allelic state of FRI and FLC with flowering time and PWU was observed under short-day controlled environment conditions (Fig. 3a,b; Fig. 4; Fig. S8a,b). Other highly differentially expressed genes in the mapping intervals on chromosome 4 and 5 showed no specific pattern that significantly correlated with the flowering time phenotype or mPWU observed in the NILs across the two developmental stages (Table S11).
Discussion
The ecotype C24 has an unusually rare combination of traits resulting in increased drought resistance, reduced VWU and increased WP (Bechtold et al. 2010; Ferguson et al. 2018), as well as resistance to a number of other abiotic and biotic stresses (Brosché et al. 2010; Lapin et al. 2012; Xu et al. 2015; Bechtold et al. 2018).
WUEi is considered to play a key role in plant water use (Steduto et al. 2007) as it relates equally to water loss by transpiration and net carbon gain, thus impacting on biomass production (Steduto et al., 2007; Long et al., 2015). Because of the relationship between leaf and plant-level WUE parameters, high leaf-level WUE is seen as an important trait for minimising water loss in many different plants species (Blum 2009; Sinclair & Rufty 2012; Vadez, Kholova, Medina, Kakkera & Anderberg 2014). In addition, WUE is often referred to as a drought adaptation trait (Condon et al. 2004; Comstock et al. 2005; McKay et al. 2008) because of the A/gs correlation, where WUE can increase during drought stress when stomata close, especially when A is not yet proportionally affected (Meinzer, Goldstein & Jaimes 1984; Gilbert, Holbrook, Zwieniecki, Sadok & Sinclair 2011; Easlon et al. 2014). However, WUE only evaluates how much water a plant needs to fix carbon, and in Arabidopsis, where within species variation in WUE is predominantly driven by variation in stomatal conductance (Easlon et al. 2014; Ferguson et al. 2018), overall plant water use will therefore be the main driver of TE.
The importance of flowering time for plant water-use strategies
In natural populations, such as Arabidopsis, few studies have compared leaf-level measurements with whole-plant estimates of WUE (i.e. TE or WP; Bechtold et al. 2010, 2013; Easlon et al. 2014), and often leaf-level WUE measurements have been exploited as a screening tool to identify genes that could optimise water requirements and yield (McKay, Richards & Mitchell-Olds 2003; McKay et al. 2008; Hausmann et al. 2005; Juenger, Mckay, Hausmann, Keurentjes & Sen 2005; Masle et al. 2005). Natural genetic variation for δ13C has been demonstrated in Arabidopsis (Bouchabke-Coussa et al. 2008; Easlon et al. 2014; Kenney et al. 2014; Verslues & Juenger 2011), and QTL mapping has successfully elucidated the genetic basis of δ13C (Ghandilyan et al. 2009; Lovell et al. 2015; McKay et al. 2003; Hausmann et al. 2005; Juenger et al. 2005; Masle et al. 2005; McKay et al. 2008). Interestingly, a positive genetic correlation between flowering time and δ13C has been reported (McKay et al., 2003; Easlon et al., 2014), while other studies found a negative genetic correlation between flowering time and water content (Loudet et al., 2002, 2003). Despite these differences, the link between flowering time and plant water status is undeniable. Furthermore, natural polymorphisms of FRI and FLC have been identified as key determinants of the natural variation in δ13C (McKay et al. 2003, 2008; Kenney et al. 2014; Lovell et al. 2015), and FLC is also known to control the circadian rhythm of leaf movement (Edwards et al., 2006). It was therefore suggested that FLC may also regulate stomatal transpiration (Edwards et al., 2006), because accessions with a non-functional allele of FLC, showed reduced flowering time and increased water content (Loudet et al., 2002, 2003). Similarly, C24 possess a non-functional allele of FLC, exhibits a high relative water content (RWC) and low stomatal conductance (Bechtold et al., 2010; Fig. S10a,b). Our data suggests that although flowering time achieved through different combinations of weak or non-functional alleles of FRI and FLC explained most of the variation in plant water use (Fig. 4; Fig S8a,b), leaf-level traits associated with the lowered stomatal conductance phenotype were independent of variation at these genes (Fig S10). In addition, VWU, average daily water-use or the dehydration response were also not affected by the allelic combinations of FRI and FLC (Fig. S5c,d, Fig. 5d, Fig. S8c). Accordingly, QTLs identified for VWU did not overlap with the two major intervals containing FRI and FLC (Fig. 3c, Table 2). Importantly, plants with high mPWU also used more water on a daily basis, which suggests that life-time PWU is not only driven by flowering time but also by short-term water use strategies (Fig. 5c, Fig. S9).
In this study, cPWU and mPWU was clearly associated with increased flowering time (Fig. 2a). Mapping identified three QTLs for cPWU located on chromosomes 3, 4 and 5, and given the observed relationships between lifespan and water use (Fig. 2a), two also overlapped with flowering time QTLs (Fig. 3, Table 2). FRI and FLC were determined to be the causal genes underlying the overlapping QTLs on chromosomes 4 and 5 respectively (Fig. S1), which reinforced the role of flowering time in determining lifetime PWU. This is perhaps unsurprising, since a plant that lives for a longer period is likely to use more water, however this occurred without apparent gain of reproductive biomass (Fig. 6b; Fig. S12b). Interestingly, other development associated genes such as ERECTA (Masle et al., 2005; Villagarcia et al., 2012; Shen et al., 2015), SHORT VEGETATIVE PROTEIN (SVP or AGL22; Bechtold et al., 2016) and HEAT SHOCK TRANSCRIPTION FACTOR A1b (Bechtold et al., 2013; Albihlal et al., 2018), have been shown to affect stomatal function, stress tolerance and plant development in Arabidopsis and other plant species.
Similarly, the lack of a significant positive correlation between δ13C and flowering time in the NILs suggested that the variation in δ13C was independent of FRI and FLC in this mapping population (Fig S10c). However, increased δ13C coincided with reduced stomatal conductance and the distinctive growth phenotype of the C24 rosette (Fig. S10b,d). In Arabidopsis, δ13C is regulated by variation in stomatal conductance (Masle et al. 2005), which clearly corroborates the observed link between gs and δ13C in the NILs and the independence from FRI and FLC. C24 is also more drought tolerant compared to Col-0 based on rosette wilting phenotypes after dehydration (Bechtold et al. 2010), and the drought response parameters were also independent of FRI and FLC in the RIL population (Fig. S11).
The impact of day length on flowering time and water use
Col-0 is a rapid cycling ecotype (Shindo et al. 2005) and the higher FLC expression levels in C24 would suggest a late-flowering phenotype compared to Col-0 (Fig. 8a). However, early genetic studies have shown that C24 contains an allele of FLC that suppresses the late flowering phenotype caused by dominant alleles of FRI, whereas Col-0 contains an allele of FLC that does not suppress the late-flowering caused by dominant FRI alleles (Koornneef et al. 1994; Lee et al. 1994; Sanda & Amasino 1995). Therefore, we do not see a significant difference in flowering time between Col-0 and C24 in un-vernalised plants (Fig. 4b). The transition from short- to long-day conditions as part of our growing regimes (Fig. 1) mimics the natural progression in day length from spring to summer, which is commonly experienced by spring/summer annuals. Despite the difference in day length and watering regimes between the short dehydration and moderate drought treatments (Fig. 1), PWU and biomass allocation were significantly correlated between experiments (Fig. 7). This suggested that even though absolute values for HI and PWU were different the relative difference between lines remained the same (Fig. 7), indicating that day length does not alter overall water use and developmental strategies in a genotype-by-environment specific manner.
With respect to the above, it is worth noting that subjecting summer or winter annual ecotypes to long photoperiods may result in outcomes that could be problematic especially when assessing mechanisms related to leaf level WUEi drought resistance strategies, since these are often closely linked to flowering time. For example, Riboni et al. (2013, 2014) demonstrated that the induced drought escape mechanisms in Arabidopsis are promoted by the drought mediated up-regulation of florigens in an ABA- and photoperiod-dependent manner, so that early flowering (drought escape) can only occur under long days, independent of FT and CONSTANS. This is in line with our observation that flowering time and mPWU are associated with FRI and FLC expression, but seemingly independent of FT expression (Fig. 8., Tables S10, S11).
The role of FRI and FLC in determining water use and biomass allocation
FRI and FLC respond to seasonal variation in temperature, thus play a crucial role in floral transitioning (Koornneef et al. 1994; Lee et al. 1994; Michaels & Amasino 2001). FLC is a MADS box transcription factor that inhibits the transition to flowering by repressing the expression of floral integrators, such as FT and SUPPRESSOR OF OVEREXPRESSION OF CONSTANS1 (SOC1, Hepworth et al. 2002; Helliwell et al. 2006; Deng et al. 2011). Most rapid-cycling accessions of Arabidopsis contain naturally occurring loss-of-function mutations in FRI and therefore have low levels of FLC expression and are early flowering even in the absence of vernalization (Johanson et al. 2000).
Despite variation in cPWU mapping to FLC and FRI, we cannot explicitly rule out an indirect effect of flowering time differences on water use (Fig. S6c). Especially since FLC expression remained high in C24 and two NILs throughout the experiment (Fig. 8a), independent of the FLC allele present (Table S5). However, the reduction in mPWU attained via introgression of the non-functional Col-0 allele of FLC or the functional C24 FRI allele into the C24 and Col-0 genomic background respectively, demonstrates that although flowering time ultimately impacts PWU it does not confound the importance of these genes in determining PWU.
Interestingly, two major FLC haplogroups were associated with flowering time variation in Arabidopsis under field-like conditions, but only in the presence of functional FRI alleles (Caicedo, Stinchcombe, Olsen, Schmitt & Purugganan 2004). This is in line with our finding that the functional C24 allele of FRI (FRI) was required for increased FLC expression, even though FRI expression was not significantly altered (Fig. 8b, Table S10, S11). Furthermore, a study of ~150 accessions showed that the role of FLC in regulating flowering time is less important under short day conditions (Lempe et al. 2005), which suggests that the impact of FLC on PWU in our experiments may have been influenced by the environmental growth conditions such as photoperiod and potentially watering status (Fig. 1).
However, since FLC also acts in conjunction with other MADS-box proteins to regulate various aspects of plant development through a large variety of target genes (Deng et al. 2011), and rapid-cycling accessions contain a number of other genes regulating FLC expression, collectively known as the autonomous floral-promotion pathway (Michaels & Amasino 1999; Sheldon et al. 1999), we cannot rule out that other genetic factors affecting flowering time may indirectly contribute to the variation in whole plant water use.
The analysis of such putative relationships was beyond the scope of this study. Yet, the considerable number of FLC targets and their involvement in different developmental pathways may reflect an important strategy to integrate environmental signals and plant development to ensure reproductive success under many different conditions.
Short-term stress-mediated initiation of flowering pathways also involves the repression of FLC expression. Cold or saline stress-dependent activation of miR169b, was shown to repress the expression of the NF-YA2 transcription factor, which in turn reduces FLC expression promoting early flowering (Xu et al. 2014). Here, stress treatments were shown to accelerate flowering (escape response) involving the above-described signalling cascade. We have previously demonstrated that the experimental watering regimes employed in this study (Figure 1), do not initiate a similar escape response in the progenitors of the mapping population and a number of other rapid cycling ecotypes (Ferguson et al. 2018; Bechtold et al. 2010; 2013). Heat sensitivity has been associated with late flowering haplotypes in vernalised plants, and FLC haplotypes resulting in late flowering showed reduced silique length, suggesting a negative correlation between flowering time and seed productivity (Bac-Molenaar et al. 2015). This negative correlation corroborates our findings, where late flowering RILs and NILs, produced less seed biomass and vice versa independent of photoperiod and watering conditions (Fig. 6b, S13b).
However, well-known work from the previous decade has demonstrated a pleiotropic link between flowering time and δ13C (WUE; McKay et al. 2003, Juenger et al. 2005). Similarly, phenotypic positive associations between flowering time and δ13C have been reported (Easlon et al. 2014, Kenney et al. 2014). It has therefore been suggested that functional alleles of FRI and FLC indirectly increase δ13C, suggesting that late flowering genotypes have greater WUE (McKay et al. 2003). The other referenced studies here support this notion in terms of flowering time and WUE, but not with respect to the allelic state of FRI and FLC. The identification that non-functional and weak alleles of FRI and FLC facilitate reduced water use and improved whole plant water use efficiency (Fig. 4, 6) challenges this previous work and illuminates the necessity to assess WUE at the whole plant and life-time level.
The relationship between leaf-level and whole-plant measures of water use
Leaf level measures of WUE, taken during vegetative growth are not representative of whole plant measures such as TE or WP (Fig. S2a, b). This suggests that plants with improved δ13C and/or WUEi are not necessarily diverting additionally acquired photosynthates toward reproductive growth. In addition, our estimation of TE is clearly biased towards the final above ground biomass, neglecting root architecture. It is well established that both root depth and density play a major role in optimising water uptake depending on the hydrological conditions (Falik, Reides, Gersani & Novoplansky 2005; Czyz & Dexter 2012), but variation here may have been limited due to their likely pot bound nature. However, the relative performance of NILs and ecotypes was highly correlated between different experiments (Fig 7), suggesting that the variation observed for TE even though biased may reflect actual genotypic differences.
Different drought resistance mechanisms, such as avoidance by maintaining high plant water status and/or drought escape through early flowering (Levitt 1985) are critical from an ecological standpoint, facilitating population persistence in regions characterised by frequent and/or extended periods of reduced water availability (Araus, Slafer, Reynolds & Royo 2002; Gechev, Dinakar, Benina, Toneva & Bartels 2012; Kooyers 2015; Kooyers, Greenlee, Colicchio, Oh & Blackman 2015). However, leaf-level traits such as high WUEi/δ13C, aimed at preserving water may not always ensure high productivity, while lifespan also determines water use but not necessarily biomass production (Fig. 6; Fig. S12b), or allocation (Fig. S12a-c; Ferguson et al. 2018). In late flowering plants, photosynthates are not translocated to reproductive sinks, but instead to vegetative biomass (Figs. S2d), which either suggests poor resource allocation in late flowering ecotypes, or a diversion of resources toward abiotic stress defence mechanisms associated with reduced water availability (Claeys, Inze & Inzé 2013). Recent studies on the perennial species Arabidopsis lyrata and 35 Arabidopsis thaliana accessions highlighted that populations increased their reproductive output while reducing vegetative growth (Remington et al. 2013; Ferguson et al. 2018), which may be even more prevalent in annual plants that only have one opportunity at reproduction. While recent reports have clearly shown that there is a selection on early flowering in Arabidopsis due to increased plant fitness (Ågren, Oakley, Lundemo & Schemske 2017; Austen, Rowe, Stinchcombe & Forrest 2017; Gnan, Marsh & Kover 2017), still little is known about the genotype-to-phenotype basis of this resource allocation trade-off.
Conclusion
We conclude that flowering is the predominant determinant of lifetime PWU strategies, a critical life history trait that is important for seed production. Absolute water use at the vegetative growth stage contributes to overall PWU, albeit to a much-reduced degree. The causal genes that underlie these QTLs are ambiguous and will require further fine-mapping. We have demonstrated that Arabidopsis plant water use strategies are independent of traditional leaf-level measures of drought tolerance, WUE and biomass traits, and consequently genes identified based on these traditional performance traits may not lead to improved productivity under water limiting or water-replete conditions.
Acknowledgements
We thank Susan Corbett and Philip M Mullineaux for help with the continuous moderate drought experiments. J.N.F was funded by a BBSRC CASE award (BB/J012564/1). U.B. is supported by the University of Essex. J.N.F. and U.B. performed and analysed all experiments. O.B and R.M provided and analysed plant material, including the RIL and NIL populations. U.B., J.N.F., and M.H. planned and designed the experiments. U.B. and J.N.F wrote the manuscript with the input from all authors. Isotopic measurements were performed by C. Hossann at the Plateforme Technique d’Ecologie Fonctionnelle (PTEF) (OC 081, INRA Nancy, France).