Abstract
In wheat, the timing and dynamics of stem elongation are tightly linked to temperature. It is yet unclear if and how these processes are genetically controlled. We aimed to identify quantitative trait loci (QTL) controlling temperature-response during stem elongation and to evaluate their relationship to phenology and height. Canopy height of the GABI wheat panel was measured between 2015 and 2017 in bi-weekly intervals in the field phenotyping platform (FIP) using a LIDAR. Temperature-response was modelled using a linear regression between stem elongation and the mean interval temperature.
The temperature-response was highly heritable (H2 = 0.81) and positively related to a later start and end of stem elongation as well as an increased final height (FH). Genome-wide association mapping revealed three temperature-responsive and four temperature-irresponsive QTL. Furthermore, putative candidate genes for temperature-response QTL were frequently related to the flowering pathway in A. thaliana while temperature–irresponsive QTLs corresponded with growth and reduced height genes. These loci, together with the loci for start and end of stem elongation accounted for 49% of the variability in height.
This demonstrates how high throughput field phenotyping in combination with environmental covariates can contribute to a smarter selection of climate-resilient crops.
Introduction
Temperature is a major abiotic factor affecting plant growth and development. As a consequence of Global warming, wheat production could decrease by 6% per °C global temperature increase (Asseng et al., 2015). While heat stress during critical stages can drastically reduce yield (Gibson and Paulsen, 1999; Farooq et al., 2011), warm temperatures can decrease yield by accelerating development and thereby shortening critical periods for yield formation (Fischer, 1985; Slafer and Rawson, 1994). However little is known about how temperature affects development and growth, and how this is genetically controlled.
The critical phase for yield formation in wheat is stem elongation (SE); happening between the phenological stages of terminal spikelet initiation and anthesis (Slafer et al., 2015). The start of SE coincides with the transition from vegetative to reproductive development, when the apex meristem differentiates from producing leaf primordia to producing spikelet primordia (Trevaskis et al., 2007; Kamran et al., 2014). During SE, florets are initiated at the spikelets until booting (Kirby, 1988; Slafer et al., 2015). An increased duration of stem elongation increases the number of fertile florets due to longer spike growth and higher dry matter partitioning to the spike (González et al., 2003). This in turn increases the number of grains per spike and therefore yield (Fischer, 1985). Modifying the timing of the critical phenological stages (transition to early reproductive phase and flowering) and SE duration has been proposed as way to increase wheat yield or at least mitigate adverse climate change effects on yield (Slafer et al., 1996; Miralles and Slafer, 2007; Whitechurch et al., 2007). The recent warming trend causes a faster advancement in phenology. For example over the past decade flowering time occurred earlier in Germany, which is attributable to both, increased temperature and selection for early flowering (Rezaei et al., 2018).
Final height is also an important yield determinant. During the “green revolution” wheat yields increased by the introduction of reduced height genes (Rht). The resulting dwarf and semi dwarf varieties benefit from improved resource allocation from the stem to the spike and reduced lodging, allowing more intensive nitrogen application (Hedden, 2003). Gibberellin insensitive Rht genes (Rht-A1, Rht-B1, and Rht-D1) were shown limit cell wall extensibility which decreases growth rates (Keyes et al., 1989) without affecting development (Youssefian et al., 1992). Whilst the allele Rht-B1c (Wu et al., 2011) and the GA sensitive Rht12 dwarfing gene (Chen et al., 2013) delay heading.
The main abiotic factors affecting the timing of floral initiation and flowering are temperature and photoperiod; with temperature affecting both vernalisation and general rate of development (Slafer et al., 2015). These developmental transitions are controlled by major genes involved in the flowering pathway, namely; vernalisation (VRN), photoperiod (PPD) and earliness per se (EPS) genes (Slafer et al., 2015). The PPD and VRN genes define photoperiod and vernalisation requirements which jointly enable the transition to generative development and define time to flowering. Whereas EPS genes fine tune the timing of floral transition and flowering, after vernalisation and photoperiod requirements are fulfilled (Zikhali and Griffiths, 2015). While vernalisation and photoperiod response are well known, the role of temperature per se remains less clear. Temperature affects all developmental phases and warmer ambient temperatures generally accelerate growth and development in crops (Slafer and Rawson, 1994, 1995a,c; Atkinson and Porter, 1996; Fischer, 2011; Slafer et al., 2015). But it is unclear, if temperature-response governs growth rate and development independently. If so, the question remains whether there is enough genetic variability in temperature-response to be used in a breeding context (Parent and Tardieu, 2012).
Genotypic variation for growth response to temperature was reported for wheat leaf elongation rate (Nagelmüller et al., 2016), as well as for canopy cover growth (Grieder et al., 2015). Kiss et al. (2017) reported significant genotype by temperature interactions in the timing of stem elongation as well as temperature dependent differences in the expression of VRN and PPD genes under controlled conditions. Under field conditions, the response of stem elongation to temperature has not yet been investigated in high temporal resolution.
In recent years, new high throughput phenotyping technologies have enabled monitoring plant height with high accuracy and frequency in the field (Bendig et al., 2013; Friedli et al., 2016; Holman et al., 2016; Aasen and Bareth, 2018; Hund et al., 2019). We have previously demonstrated that the ETH field phenotyping platform (FIP, Kirchgessner et al., 2016) can be used to accurately track the development of canopy height in a large set of wheat genotypes using terrestrial laser scanning (Kronenberg et al., 2017). Considerable genotypic variation was detected for the start and end of SE which correlated positively with final canopy height (Kronenberg et al., 2017).
While many temperature-independent factors affecting plant height are known, the influences of temperature-dependent elongation and timing of the elongation phase is less clear. To address this, we aimed to dissect final height into the following components: i) temperature-independent elongation, ii) temperature-dependent elongation and iii) the duration of the elongation phase determining by the start and end of the process. To achieve this we present a method to assess and measure these three processes under field conditions by means of high-frequency, high-throughput phenotyping of canopy height development. The resulting data were combined with genetic markers to identify quantitative trait loci controlling the aforementioned processes.
Material and Methods
Experimental setup, phenotyping procedures and extracted traits
Field experiments were conducted in the field phenotyping platform FIP at the ETH research station in Lindau-Eschikon, Switzerland (47.449°N, 8.682°E, 520 m a.s.l.; soil type: eutric cambisol). We used a set of approximately 330 winter wheat genotypes (335 – 352 depending on the experiment) comprising current European elite cultivars (GABI Wheat; Kollers et al., 2013), supplemented with thirty Swiss varieties. These were monitored over three growing seasons in 2015, 2016 and 2017. Details about the experimental setup for the growing seasons 2015 and 2016 are described in Kronenberg et al. (2017). Briefly, the field experiments were conducted in an augmented design with two replications per genotype using micro plots with a size of 1.4 by 1.1 m. In the growing season 2017, the experiment was repeated again, with minor changes in genotypic composition. This resulted in 328 genotypes present across all three experiments.
Canopy height was measured twice weekly from the beginning of shooting (BBCH 31) until final height using a light detection and ranging (LIDAR) scanner (FARO R Focus3D S 120; Faro Technologies Inc., Lake Mary USA) mounted on the FIP (Kirchgessner et al., 2016). Canopy height data was extracted from the LIDAR data as described in Kronenberg et al. (2017). Spatial heterogeneity at each measuring date was corrected by applying two-dimensional P-splines to the raw canopy height data within each year using the R-package SpATS (Rodríguez-Álvarez et al., 2018). The start, end, and duration of stem elongation with final canopy height (FH) were extracted from the height data as described by Kronenberg et al. (2017): Normalized canopy height was calculated as percent of final height at each day of measurement for every plot and then linearly interpolated between measurement points. Growing degree-days until 15% final height (GDD15) and 95% final height (GDD95) were used as proxy traits for start and end of stem elongation, respectively. SE duration was recorded in thermal time (GDDSE) as well as in calendar days (timeSE), as the difference between GDD95 and GDD15 (Kronenberg et al., 2017).
In order to investigate short-term growth response to temperature, average daily stem elongation rates (SER) were calculated for each plot as the difference (∆) in canopy height (CH) between consecutive timepoints (t):
Extracting growth response to temperature
Temperature response was modelled by regressing average daily stem elongation rates (SER) against average temperature of the respective interval for each plot within the respective year following where T is the ambient temperature, a is the coefficient of the linear regression (i.e. growth response to ambient temperature; slpSER~T) and ε denotes the residual error. bTcrit is the model intercept, estimated at the temperature, at which the correlation between intercept and slope is zero (intSER~T). Tcrit was determined empirically for each year by sequentially estimating the intercept between 1°C and 22°C Fig. 1A). Per definition, the intercept would be estimated at T = 0 °C, i.e. far outside the range of observed temperatures. In the observed data, an intercept at T = 0°C correlated strongly negative with the slope (Fig. 1A) and, thus, did not add much additional information concerning the performance of the evaluated genotypes. Grieder et al. (2015) performed a similar analysis for the canopy cover development during winter and found a similar, strongly negative correlation between temperature-response (slope) and growth at 0°C (intercept). Likewise an intercept at 20°C at the upper range of the observed data was correlated strongly positive with the slope. Hence, Tcrit is the turning point from negative to positive correlation as the position of the intercept increases, which is the point where intercept and slope are independent. Therefore, two genotypes can show the same growth at Tcrit but differ markedly in temperature-response (Fig. 1B), have the same temperature-response but differ in growth at Tcrit (Fig. 1C), or differ for both, intercept and slope (Fig. 1D). Following this, intSER~T would be interpreted as intrinsic, temperature-independent growth, hereinafter referred to as “vigour”.
Statistical Analysis
All statistical analysis were performed in the R environment (R Core Team, 2018). Best linear unbiased estimations (BLUEs), predictors (BLUPs) and broad sense heritabilities (H2) were determined for all traits using the R-package asreml (Butler, 2009). In a first step, BLUEs were calculated within each year using:
Where Y is the respective trait (FH, GDD15, GDD95, GDDSE, intSER~T or slpSER~T), μ is the overall mean, g the fixed genotype effect and ε is the residual error.
In a second step, 3-year BLUPs were calculated using where Y are the single-year BLUEs for the respective traits derived from eq. 3, μ is the overall mean, g is the genotype effect, y is the year effect and ε is the residual error. Broad sense heritabilities were calculated following Falconer and Mackay (1996) as where and are genotypic and residual variance, respectively, from eq. 4.
The 3-year BLUPs of GDD15, GDD95, GDDSE, FH, intSER~T, and slpSER~T were used for correlations and genome wide association study (GWAS).
Association study
The genetic basis of temperature-response was investigated by GWAS. GWAS was performed on the different traits to compare the phenotypic correlations with the underlying genetic architecture of the traits. As a positive control final height data made in Germany and France by Zanke et al. (2014b) was also compared and analysed.
Genotyping data was made previously by the GABI wheat consortium represented by the Leibniz Institute of Plant Genetics and Crop Plant Research (IPK; Zanke et al., 2014a) using the 90K illumina SNP-chip (Cavanagh et al., 2013; Wang et al., 2014). Monomorphic SNPs were discarded. The remaining markers were mapped to the IWGSC reference genome (Consortium (IWGSC) et al., 2018) by BLASTN search using an E-value threshold < 1e−30. The genome position with the lowest E-value was assigned as the respective marker location. Markers that could not be unequivocally positioned were dropped. After filtering SNPs with a minor allele frequency and missing genotype rate < 0.05, a total of 13,450 SNP markers and 315 genotypes remained in the set. The reference genome position of RHT, PPD, VRN and putative EPS genes was determined with BLASTN search as described above using published GenBank sequences (Table S1).
To mitigate against multiple testing, relatedness and population structure; three different methods were used to calculate marker trait associations (MTA) between phenotypic BLUPs and SNP markers:
We used a mixed linear model (MLM) including principal components among marker alleles as fixed effects and kinship as random effect to account for population structure (Zhang et al., 2010). This approach was chosen to stringently prevent type I errors. The MLM GWAS was performed using the R Package GAPIT (v.2, Tang et al., 2016). Kinship was estimated according to VanRaden (2008).
In a generalised linear model (GLM) framework implemented in PLINK (Purcell et al., 2007), association analysis was performed using SNP haplotype blocks consisting of adjacent SNP triplets. Using haplotype blocks takes the surrounding region of a given SNP into account, thus increasing the power to detect rare variants (Purcell et al., 2007)
Finally, the FarmCPU method (Liu et al., 2016) was used, which is also implemented in GAPIT. FarmCPU tests individual markers with multiple associated markers as covariates in a fixed effect model. Associated markers are iteratively used in a random effect model to estimate kinship. Confounding between testing markers and kinship is thus removed while controlling type I error, leading to increased power (Liu et al., 2016).
For all methods, a Bonferroni correction was applied to the pointwise significance threshold of α = 0.05, to avoid false-positives. Hence, only markers above −log10(P-value) >= 5.43 considered significant.
Linkage disequilibrium (LD) among markers was estimated using the squared correlation coefficient (r2) calculated with the R package SNPrelate (Zheng et al., 2012). A threshold of r2 = 0.2 (Gaut and Long, 2003) was applied to calculate the chromosome specific distance threshold of LD decay. Putative candidate genes were identified by searching the IWGSC annotation of the reference genome (Consortium (IWGSC) et al., 2018) for genes associated with growth and development within the LD distance threshold around the respective MTA.
Results
Phenotypic results
We measured the canopy height of 710 – 756 plots per year, containing 335 – 352 wheat genotypes, for three consecutive years. In each season measurements were made between 17 and 22 times during stem elongation. Thus resulting in an average of 122 canopy height measurement points per genotype. From these data we extracted growth rates and the timing of critical stages. Plot based growth rates within single years indicate a clear relation between growth and temperature for the period of stem elongation, as depicted in Fig. 2. Towards the end of the measurement period in June, there was a larger deviation, which was also reflected in the quality of plot based linear model fits of SER versus temperature (see eq. 2), summarized in Fig. S1. For the 2015 and especially the 2016 experiment, R2 values were low and except for the 2017 experiment, the parameter estimates were not statistically significant (Fig. S1A). Inspection of the best and worst model fits however shows, that failure of fitting the model for single plots was levelled out by the replications within genotypes (Fig. S1B), therefore allowing for confident estimates of genotypic means of the model parameters (see below). Analysis of variance revealed significant (P < 0.001) genotypic effects for both slpSER~T and intSER~T within single years as well as across three years. Both traits showed high heritabilities across years (H2 = 0.81 for slpSER~T and H2 = 0.77 for intSER~T) and very high heritabilities within single years (Table 1). Using the BLUPs of slpSER~T, intSER~T and temperature sum for stem elongation (GDDSE), final height could be predicted with high accuracy across different years (0.82 <= R2 <= 0.85) by training a linear model on the BLUPs of one year and predicting it on the BLUPs of another independent year. Training the model on the 3-year BLUPs resulted in a prediction accuracy of single years between R2 = 0.93 and R2= 0.95 (Fig. 3). High heritabilities within years (0.75 <= H2 <= 0.99) as well as across three years (0.54 <= H2 <= 0.98; Table 1), were also found for other traits; final height, start of SE, end of SE and SE duration.
Phenology, temperature-response and final height were positively correlated
To evaluate the relationships between the traits measured, Pearson correlation coefficients were calculated for each trait pair. If not indicated otherwise, the reported correlations were highly significant (P < 0.001)
Positive correlations were found among GDD15, GDD95 and FH (0.36 <= r <= 0.64, Fig. 4), indicating that taller varieties were generally later. Temperature response (slpSER~T) and vigour (intSER~T) also showed a strong, positive relationship with final height (r = 0.85 and r = 0.65, respectively). However, only temperature-response correlated with GDD15 and GDD95 (r = 0.63 and r = 59, respectively), whereas vigour did not (r < 0.26, Fig. S2).
As expected, stem elongation duration in thermal time (GDDSE) was negatively correlated with GDD15 (r = −0.44) and positively correlated with GDD95 (r = 0.4). But, GDDSE did not correlate with final height (r = −0.01, P = 0.878) or temperature-response (r = 0.006, P = 0.289). Although GDDSE negatively correlated with vigour (r = −0.32). In contrast, SE duration in calendar days (timeSE) was negatively correlated with temperature-response (r = −0.35) and GDD15 (r = −0.82), indicating a longer SE phase for earlier genotypes. Other weak correlations (r < 0.3), that are not discussed, are shown in Fig. S2.
Linkage disequilibrium and population structure
Prior to MTA analysis we evaluated population structure and LD. Principal component analysis of the marker genotypes revealed no distinct substructure in the investigated population. The biplot of the first two principal components showed no apparent clusters, with the first component explaining 8% and the second component explaining 3.3% of the variation in the population (Fig. S5). This is consistent with prior work using the same population (Kollers et al., 2013; Yates et al., 2018). On average across all chromosomes, LD decayed below an r2 of 0.2 at a distance of 9 MB. There was however considerable variation in this threshold among the single chromosomes (Table S2).
Association study
Genome-wide association results differed markedly depending on the applied model. Using a MLM with kinship matrix and PCA as covariates resulted in no significant MTA for any trait (Fig. S3). In contrast, the GLM using the haplotype method yielded 2949 significant MTA for α < 0.05 and 1846 MTA for α < 0.001 respectively. However, investigation of the respective QQ-plots showed large P-value inflation in the haplotype method whereas the P-values were slightly deflated when using the MLM approach (Fig. S3, Fig. S4). In contrast, with FamCPU the QQ-plots (Fig. 5) showed no P-value inflation, except for some markers. This pattern is expected, if population structure is appropriately controlled. Therefore, FarmCPU was chosen to be the most appropriate method for the given data, despite identifying less significant MTA.
As a positive control we compared our final height data and associated markers with data made by Zanke et al. (2014b). Final canopy height correlated strongly between the two studies (r = 0.95), which is in accordance with the high heritability of the trait. In this study, we found 11 significant MTA for final height (Table 2, Fig. 5). Zanke et al. (2014b) reported 280 significant MTA for final height across several environments. Of these, only marker RAC875_rep_c105718_585 on chromosome 4D overlapped with the MTA found in this study. However, by considering flanking markers, we found that of the remaining ten significant MTA for final height, six were in LD with MTA found by Zanke et al. (2014b; Table S3). The significant MTA found for FH in this study are near known genes controlling FH. For example, Tdurum_contig64772_417, is 4 MB upstream of Rht-B1 and RAC875_rep_c105718_585, is 7 MB downstream of Rht-D1 on their respective group 4 chromosomes.
Temperature-response loci are independent of vigour loci
For slpSER~T we detected one significant (LOD = 5.77) MTA on chromosome 1B (wsnp_Ex_c1597_3045682) and two almost significant (LOD = 5.39 / LOD = 5.02) MTA on chromosomes 4B (CAP7_c10839_300) and 5D (IAAV7104), respectively (Fig. 5). All associated markers for slpSER~T yielded small but significant allelic effects ranging from −0.049 mm °C−1d−1 to −0.041 mm °C−1d−1 (Table 2). The GWAS for intSER~T yielded four significant MTA on chromosomes 2B, 4B, 4D and 5D respectively (Table 2, Fig. 5). Start and end of SE yielded four MTA each (Table 2, Fig. 5).
Comparing the GWAS results for temperature-response, vigour, final height, GDD15 and GDD95 revealed no common quantitative trait loci (QTL) between slpSER~T and any other trait. Only one marker (Excalibur_c74858_243) was significantly associated with both GDD15 as well as GDD95. The lack of overlap, of MTA, between temperature-response, vigour and timing of critical stages indicate they are genetically independent. However, there is a genetic connection between vigour and FH on the one hand and between the start and end of stem elongation on the other.
To identify potential causative genes underlying the QTL, we searched the reference genome annotation around the respective QTL intervals. For temperature-response we found an increased presence of genes or gene homologues involved in the flowering pathway, i.e. EARLY FLOWERING 3, FRIGIDA and CONSTANS (Table 3). Around the QTL associated with vigour the annotation showed genes associated with growth (i.e. GRAS, CLAVATA, BSU1, Argonaute) as well as developmental progress (i.e. Tesmin/TSO1-like CXC domain, BEL1, AGAMOUS (Table 4). Importantly, we found GAI-like protein 1 6MB upstream of marker Kukri_rep_c68594_530, which we identified as RHT-D1 by blasting the RHT-D1 sequence (GeneBank ID AJ242531.1) against the annotated reference genome.
Vigour, temperature-response and the timing of SE affect final height
The phenotypic correlations show a strong connection between temperature-response, vigour and FH as well as weaker connections between GDD15, GDD95 and FH. In order to examine this interdependency on a genetic level, we used a linear model to predict FH with the SNP alleles of the QTL for slpSER~T, intSER~T, GDD15 and GDD95 as predictors. The model was able to predict FH with an accuracy R2 = 0.49, with significant contributions by QTL of all three traits (Fig. 6, Table 5).
Discussion
In this study we present a method to measure temperature response during stem elongation of wheat using high throughput phenotyping of canopy height in the field. The results show a highly heritable genotype-specific ambient temperature response of wheat which affects both growth and timing of the developmental key stages. We modelled temperature-response in a simple linear framework with the intercept estimated at the temperature of zero correlation to the slope. This allowed for the decomposition of growth dynamics into a genotype-specific vigour component and temperature-response component. Thereby we could assess interdependence between vigour and temperature-response to plant height and the timing of developmental key stages.
Linear models were used before to describe wheat growth response to temperature for leaf elongation (Nagelmüller et al., 2016), canopy cover (Grieder et al., 2015) as well as stem elongation rate (Slafer and Rawson, 1995a). Others proposed the use of a more complex, Arrhenius type of function to account for decreasing growth rates at supra optimal temperatures (Parent and Tardieu, 2012). Wheat has its temperature-optimum at around 27°C (Parent and Tardieu, 2012). As temperatures in the measured growth intervals during stem elongation did not exceed 25°C and given the temporal resolution of the data, a simple linear model is justified (Parent et al., 2018).
The results of the correlation analysis show a clear connection between FH and temperature-response (slpSER~T) as well as between FH and vigour (intSER~T). This is consistent with part i) and ii) of our hypothesis: Final height can be described as a function of temperature-independent growth processes and as a function of temperature-response during SE. Importantly, among all components, the temperature-response was a major driver of final height and also had a strong influence on the timing. Temperature-response delayed the beginning of stem elongation leading to a later start and end of the whole phase. This finding might appear counter intuitive: given the assumption that plants develop faster under higher ambient temperatures a more responsive genotype should develop faster compared to a less responsive one. Slafer and Rawson (1995b) reported an accelerated development towards floral transition with increasing temperatures up to 19°C whereas higher temperatures slowed development. In that respect, a more responsive genotype would experience a stronger delay of floral transition under warm temperatures.
In terms of their correlation to FH, the effects of the timing of start and end of stem elongation (part iii) of the initial hypothesis) are less distinct. Final height was more a function of faster growth than duration of growth, especially since genotypes with a strong temperature-response have a shorter duration of SE. However, the timing of start and end of stem elongation was linked with temperature-response. Based on this result and the according correlations, it would appear that temperature-response influences FH directly as well as indirectly by mediating start and end of stem elongation.
The question, whether these trait correlations are due to pleiotropic effects will substantially impact the breeding strategy (Chen and Lübberstedt, 2010). If these effects are pleotropic, they have a huge impact on breeding as they indicate that temperature-response, timing and height are to a large degree determined by the same set of genes. Alternative explanations are linkage and population structure. As the examined traits are major drivers of adaptation to the different regions of Europe we anticipate a very strong selection for both, temperature response as well as timing of critical stages. The GABI wheat panel is made of wheat varieties from different regions of Europe. Even if there is no apparent population structure at neutral markers, there may be a strong population structure at selected loci with strong effect on local adaptation. However, pleiotropy between height and flowering time is known for maize and rice, supporting the hypothesis of pleiotropy here. The DWARF8 gene of maize encoding a DELLA protein is associated with height and flowering time (Lawit et al., 2010) and strongly associated with climate adaptation (Camus-Kulandaivelu et al., 2006, page). The rice GHD7 locus has a strong effect on number of days to heading, number of grains per panicle, plant height and stem growth (Xue et al., 2008). To further examine the relationship among the different traits we consider the following GWAS analysis using stringent correction of population structure.
The GWAS results indicate an independent genetic control of final height, temperature response and the timing of critical stages. Whereas vigour and FH as well as start and end of SE appear to be partly linked. Yet, final height could be predicted with surprising accuracy using the QTL for temperature response, vigour, start and end of SE which reflects the correlations found in the phenotypic data.
Previous studies investigating the control of developmental key stages in wheat with respect to temperature generally adopted the concept, that after fulfilment of photoperiod and vernalisation, EPS genes act as fine tuning factors independent of environmental stimuli (Kamran et al., 2014; Zikhali and Griffiths, 2015). Temperature, apart from vernalisation is thought to generally quicken growth and development independent of the cultivar (Slafer and Rawson, 1995b; Porter and Gawith, 1999; Slafer et al., 2015). A genotype-specific temperature effect on the duration of different phases was not considered (Takahashi &Yasuda 1971, Slafer & Rawson 1995c). It was however reported, that photoperiod effects vary depending on temperature (Slafer and Rawson, 1995c). Under long days, Hemming et al. (2012) reported faster development and fewer fertile florets under high compared to low temperatures. Temperature-dependent effects were also found for different EPS QTL (Slafer and Rawson, 1995c; Gororo et al., 2001). It has previously been suggested, that EPS effects could be associated with interaction effects between genotype and temperature fluctuations (Slafer and Rawson, 1995c; van Beem et al., 2005).
The mechanisms of ambient temperature sensing and its effects on growth and development are not yet well understood (Sanchez-Bermejo and Balasubramanian, 2016). However, important findings regarding ambient temperature effects on flowering time as well as on hypocotyl elongation have come from the model species Arabidopsis thaliana (Wigge, 2013). With respect to these two traits, Sanchez-Bermejo and Balasubramanian (2016) reported distinct genotypic differences in temperature-sensitivity. According to their results, the flowering pathway genes FRIGIDA (FRI), FLOWERING LOCUS C (FLC) and FLOWERING LOCUS T (FT) are major candidate genes for ambient temperature mediated differences in flowering time (Sanchez-Bermejo and Balasubramanian, 2016). In the present study, we found FRI homologues near two of the three QTL for temperature-response. FRI and FLC acts as main vernalisation genes in A. thaliana (Johanson et al., 2000; Amasino and Michaels, 2010). In wheat, these genes are not yet well described. However, FLC orthologues were found to act as flowering repressors regulated by vernalisation in monocots (Sharma et al., 2017).
Another promising candidate gene for temperature response found near the QTL on chromosome 1B is EARLY FLOWERING 3 (ELF3). In A. thaliana, ELF3 was found to be a core part of circadian clock involved in ambient temperature response (Thines and Harmon, 2010). In Barley, ELF3 was shown to be involved in the control of temperature dependent expression of flowering time genes (Ejaz and von Korff, 2017). A mutant ELF3 accelerated floral development under high ambient temperatures while maintaining the number of seeds (Ejaz and von Korff, 2017). Furthermore, ELF3 has been reported as a candidate gene for EPS1 in Triticum monococcum (Alvarez et al., 2016).
One important aspect we could not address in the current study is the interaction of genotype specific temperature response with vernalisation and photoperiod (Slafer and Rawson, 1995c; Gol et al., 2017; Kiss et al., 2017). It also remains unclear if and to which extent temperature response varies across different developmental phases and how temperature-response relates to other environmental stimuli such as vapour pressure deficit or radiation. Nevertheless, the results of this study present valuable information towards a better understanding of temperature response in wheat and may be of great importance for breeding. Temperature-response could provide a breeding avenue for local adaptation as well as the control of plant height.
With the recent advancements in UAV-based phenotyping techniques, the growth of canopy cover and canopy height can be measured using image segmentation and structure from motion approaches (Bareth et al., 2016; Aasen and Bareth, 2018; Roth et al., 2018). Thus, temperature-response can be investigated during the vegetative canopy cover development (Grieder et al., 2015) and during the generative height development as demonstrated here. It can also be assessed in indoor platforms (e.g. Parent and Tardieu, 2012) and the field using leaf length tracker (Nagelmüller et al., 2016) measuring short-term responses of leaf growth to diurnal changes in temperature. Combining this information may greatly improve our understanding about the genetic variation in growth response to temperature.
Conclusion
Modern phenotyping platforms hold great promise to map the genetic factors driving the response of developmental processes to environmental stimuli. To the best of our knowledge, this is the first experiment dissecting the stem elongation process into its underlying components: temperature-dependent elongation, temperature-independent vigour and elongation duration. The independent loci detected for these traits, suggest that it is possible to select them independently. The detected loci may be used to fine tune height and the beginning and end of stem elongation as they explain a substantial part of the overall genotypic variation. With increases in automation, growth processes may be monitored in the field on a daily basis or even multiple times per day. This will increasing the precision in assessing genotype responses to the fluctuation in meteorological conditions and quantifying the relationship of these responses to yield. Remote sensing by means of unmanned aerial vehicles in combination with photogrammetric algorithms will allow to measure these traits in breeding nurseries. We believe that this is paving the road for a more informed selection to climate adaptation within individual growing seasons.
Supplementary Data
Fig. S1: Summary of plot based linear model fits of stem elongation rate vs. temperature.
Fig. S2: Pearson correlation coefficients among 3-year BLUPS of all investigated traits.
Fig. S3: Manhattan plots and quantile-quantile plots depicting the GWAS results using the MLM approach.
Fig. S4: Manhattan plots and quantile-quantile plots depicting the GWAS results using the GLM approach.
Fig. S5: Principal component analysis among marker genotypes.
Table S1: Genes of interest related to floral transition and flowering.
Table S2: Chromosome wise distance thresholds for LD-decay < r2 = 0.2.
Table S3: Corresponding marker-trait associations for final canopy height with respect to Zanke et al. 2016.
Table S4: 3-year BLUPs of the investigated traits FH, GDD15, GDD95, GDDSE, timeSE, slpSER~T, intSER~T.
Author contribution
LK conducted the laser scans, did all statistical analyses and drafted the manuscript; SY assisted with the GWAS and candidate gene evaluation; MB assisted with the spatial correction; NK developed the analysis pipeline for the laser scans; AW drafted the grant application and supervised the overall concept; AH made the experimental design, developed the phenotyping models and assisted with the statistical analysis. All authors contributed to the drafting of the manuscript.
Acknowledgements
We sincerely thank Hansueli Zellweger for managing and nursing our field experiments. We further thank the members of the ETH crop science and the ETH molecular plant breeding groups, especially Michelle Nay and Beat Keller, for many fruitful discussions. We also thank Martina Binder for doing the correlation analysis between the final height data of this study and the data made by Zanke et al. (2014b) in the framework of her MSc-Thesis. We would like to thank Marion Röder (IPK Gatersleben) for supply of the GABI wheat panel including genetic information. Finally, we thank the anonymous reviewers for their helpful comments and suggestions. This work was supported by the Swiss National Foundation (SNF) in the framework of the project PhenoCOOL (project no. 169542).