1. Introduction
Floral traits, such as flowering time and flower size, are determinants of reproductive success and consequently fitness. In outcrossing species, variation in floral traits typically co-evolves with pollinators, with resources invested in producing larger or more attractive flowers (Goodwillie et al., Reference Goodwillie, Sargent, Eckert, Elle, Geber, Johnston, Kalisz, Moeller, Ree, Vallejo-Marin and Winn2010). In contrast, selfing species, such as Arabidopsis thaliana, in which reproduction is largely independent of pollinator availability, have typically evolved smaller and less attractive flowers that promote self-fertilization (Sicard & Lenhard, Reference Sicard and Lenhard2011). This strategy likely reflects a reallocation of resources toward other physiological processes (Goodwillie et al., Reference Goodwillie, Sargent, Eckert, Elle, Geber, Johnston, Kalisz, Moeller, Ree, Vallejo-Marin and Winn2010). The importance of floral traits acting as reproductive and ecological niche barriers (Krizek & Anderson, Reference Krizek and Anderson2013) makes it crucial to understand how the rising global temperatures (FAO, 2023) will affect these traits and overall plant fitness.
Temperature plays a critical role in regulating plant growth and development in a process known as thermomorphogenesis (Casal & Balasubramanian, Reference Casal and Balasubramanian2019; Delker et al., Reference Delker, Quint and Wigge2022; Quint et al., Reference Quint, Delker, Franklin, Wigge, Halliday and van Zanten2016). In nature, temperature is known to vary across latitudes, which is often used as a proxy when studying temperature-mediated changes in plant populations (de Villemereuil et al., Reference de Villemereuil, Mouterde, Gaggiotti and Till-Bottraud2018; Ren et al., Reference Ren, Guo, Liu, Yu, Guo, Wang, Ye, Lambertini, Brix and Eller2020; Stotz et al., Reference Stotz, Salgado-Luarte, Escobedo, Valladares and Gianoli2021). The effects of ambient temperature on vegetative traits both in plants grown at controlled and in nature are well known. For example, warmer conditions often lead to hypocotyl elongation, and higher temperatures also tend to reduce overall biomass production, indicating constraints on vegetative growth (Casal & Balasubramanian, Reference Casal and Balasubramanian2019; Delker et al., Reference Delker, Quint and Wigge2022; Quint et al., Reference Quint, Delker, Franklin, Wigge, Halliday and van Zanten2016). It has also been shown that temperature effects on plant vegetative growth can be predicted from temperature-induced changes in pathways of plant central metabolism (Wendering et al., Reference Wendering, Andreou, Laitinen and Nikoloski2025). Much less is understood about how temperature influences floral and reproductive traits.
Among floral traits, flowering time has been found to be highly sensitive to temperature (Song et al., Reference Song, Ito and Imaizumi2013; Wei et al., Reference Wei, Guo, Mu, Alladassi, Mural, Boyles, Hoffmann, Hayes, Sigmon, Thompson, Salas-Fernandez, Rooney, Kresovich, Schnable, Li and Yu2025), enabling plants to avoid flowering under unfavourable conditions. Moreover, elevated temperatures can reduce petal number in outcrossing Cardamine hirsuta (Mckim et al., Reference Mckim, Routier-Kierzkowska, Monniaux, Kierzkowski, Pieper, Smith, Tsiantis and Hay2017), and decrease flower size in several outcrossing species, including strawberry and rose (Liang et al., Reference Liang, Wu and Byrne2017; Shin et al., Reference Shin, Lieth and Kim2001), petunia (Sood et al., Reference Sood, Duchin, Adamov, Carmeli-Weissberg, Shaya and Spitzer-Rimon2022), salvia, marigold (Moccaldi & Runkle, Reference Moccaldi and Runkle2007), blue clips (Niu et al., Reference Niu, Heins, Cameron and Carlson2001) and sweet cherry (Mahmood et al., Reference Mahmood, Carew, Hadley and Battey2000). Even in selfing species such as A. thaliana, natural genetic variation exists for flower size and for temperature-mediated flower size plasticity (Andreou et al., Reference Andreou, Messer, Tong, Nikoloski and Laitinen2023; Wiszniewski et al., Reference Wiszniewski, Uberegui, Messer, Sultanova, Borghi, Duarte, Vicente, Sageman-Furnas, Fernie, Nikoloski and Laitinen2022).
Floral trait plasticity refers to the degree of change in a trait of an individual in response to temperature. Genetic variation in floral trait plasticity has been observed for flower size in A. thaliana (Wiszniewski et al., Reference Wiszniewski, Uberegui, Messer, Sultanova, Borghi, Duarte, Vicente, Sageman-Furnas, Fernie, Nikoloski and Laitinen2022) and for many other floral and fitness traits (Arnold et al., Reference Arnold, Wang, Notarnicola, Nicotra and Kruuk2024; Ren et al., Reference Ren, Guo, Liu, Yu, Guo, Wang, Ye, Lambertini, Brix and Eller2020; Stotz et al., Reference Stotz, Salgado-Luarte, Escobedo, Valladares and Gianoli2021). This variation implies genotype-by-environment interaction (GxE) for these traits (Napier et al., Reference Napier, Heckman and Juenger2023). The mean value of a trait and its plasticity in response to environment (GxE) are often genetically unlinked (Laitinen & Nikoloski, Reference Laitinen and Nikoloski2018; Wiszniewski et al., Reference Wiszniewski, Uberegui, Messer, Sultanova, Borghi, Duarte, Vicente, Sageman-Furnas, Fernie, Nikoloski and Laitinen2022) and as a result, may be independently shaped by natural selection (Josephs, Reference Josephs2018). Despite growing evidence that floral traits are temperature-sensitive, we lack a comprehensive assessment of how floral traits and their plasticity respond to different temperatures and whether these responses translate into fitness differences.
We hypothesize that only a few degrees change in ambient temperature alters floral traits and their plasticity, and that this plasticity contributes to fitness differences among A. thaliana accessions. The broad native distribution of A. thaliana across the northern hemisphere and its presumed adaptation to diverse environments make it an ideal model for investigating these questions (Hoffmann, Reference Hoffmann2002). To test our hypothesis, we examined how four ambient temperatures, from 17 °C to 27 °C, affect seven life-history traits, divided into floral and fitness traits. Our findings offer new insights into how rising temperatures may shape the adaptive potential of selfing plants. Moreover, this study provides evidence for the role of phenotypic plasticity in enabling plants to cope with rapid climatic change, particularly in species with limited genetic recombination, such as selfing A. thaliana.
2. Results
2.1. Influence of temperature on floral and fitness traits
We investigated the effect of four different temperatures on seven life-history traits, namely: flowering time (FT), early silique development time (ESDT), flower size (FD), seeds per silique (SS), single seed weight (SSW), seed number (SN) and flower number (FN). From these, SN was determined by the ratio of the total weight of the seeds to SSW, while FN (corresponding to the number of siliques) was estimated by dividing the total number of seeds by SS. These traits were further divided into three floral traits (i.e., FT, ESDT and FD) and four fitness traits (i.e., SS, SSW, SN and FN). A set of 34 A. thaliana accessions was selected based on our previous finding to cover the range of temperature-mediated flower size plasticity (Wiszniewski et al., Reference Wiszniewski, Uberegui, Messer, Sultanova, Borghi, Duarte, Vicente, Sageman-Furnas, Fernie, Nikoloski and Laitinen2022). To ensure flowering, all accessions were vernalized for six weeks, whereafter the plants were grown in four replicates to constant 17 °C, 20 °C, 24 °C and 27 °C until they had produced ten siliques (see Methods).
After removing outliers, we examined the contribution of the quantified traits to the overall trait variance of all measurements using principal component analysis (PCA). PCA indicated a separation of measurements by temperature, but not by accession (Figure 1a). The first two principal components explain 55.5% of the total variance in the dataset (Figure 1a, PC1: 31%, PC2: 24.1%). Moreover, PC2 was primarily associated with floral traits, where FD had the strongest loading with 0.59, followed by FT (0.53) and ESDT (0.23). In contrast, PC1 was mainly characterised by variation in fitness traits, with SN showing the largest loading (0.63), followed by FN (0.60) and SSW (-0.42). Thus, our PCA highlights that the major sources of variation in floral traits and fitness traits are captured separately.
Seven phenotypic traits across four temperatures. A panel of 34 Arabidopsis thaliana accessions grown at 17 °C, 20 °C, 24 °C and 27 °C were analysed for flowering time (FT), early silique development time (ESDT), flower diameter (FD), seeds per silique (SS), single seed weight (SSW), seed number (SN) and flower number (FN). (a). Principal component analysis (PCA) ellipses outline temperature groups in blue, green, orange and red, respectively, based on interquartile range (IQR) filtering. Each accession is represented by a glyph, with glyph size increasing proportionally to temperature. Trait contributions to the PCA are shown as loading vectors with arrows indicating direction and magnitude. (b). Boxplot for each trait and temperature (n ≥ 98), with each point representing an individual measurement. Boxes represent the IQR, with horizontal lines indicating the median. Letters indicate statistical difference (p-value ≤ 0.05, Kruskal–Wallis test, followed by pairwise comparisons using Dunn’s test with Benjamini–Hochberg correction). (c). Partial correlations for all traits at the four temperatures. Correlations were adjusted for all remaining traits other than the ones compared, (* p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001, permutations = 10000, adjusted with Benjamini–Hochberg).
On average, we observed that the plants flowered earlier, the siliques developed faster and flowers were significantly smaller with increasing temperature (Figure 1b). In addition, the number of seeds per silique and the seed number were reduced (Figure 1b). In all traits, apart from FN and SSW, the decrease of the means from 17 °C to 27 °C was between 1.3 and 1.38-fold. In contrast, SSW showed a slight increase (1.17-fold) across temperatures. FN exhibited the most stable trait means across temperatures, with only a 1.03-fold decrease from 17 °C to 27 °C.
We then used the coefficient of variation (CV) to quantify the trait variation across the different accessions in each of the temperatures (Supplementary Figure 1). Apart from SS, the highest amount of variation among the accessions for each trait was observed at 27, indicating a cryptic genetic variation in these traits that is revealed at 27 °C (Supplementary Figure 1). Partial correlation analysis (see Methods) for the traits at the different temperatures revealed a strong positive correlation between FN (estimated from the SN and SS) and SN (Figure 1c). This, together with the strong positive correlation between SS and SN at each temperature (r > 0.79, p < 0.001), indicates that the more easily measured SS could serve as a proxy for fitness in this set of accessions (Figure 1c).
2.2. Temperature-mediated plasticity in floral and fitness traits
Next, we investigated whether the floral and fitness traits exhibited temperature-mediated plasticity and whether any of the trait plasticities showed strong associations. To reveal the different plasticity patterns, we clustered accessions according to their reaction norms of the seven traits (Figure 2a). The highest number of clusters was observed in FD reaction norms, in which accessions were divided into five groups (k-means clustering, k determined by silhouette index analysis, Figure 2a, Supplementary Table 1). In all clusters, the FD reaction norms declined from 17 °C to 27 °C. The reaction norm of cluster 5 for FD contained half of all tested accessions and showed a pattern of largely linear reduction in FD from 17 °C to 27 °C. In smaller clusters, the FD reaction norm slope showed the strongest decline only at 24 °C (cluster 2) or at 20 °C (cluster 3). For the reaction norm of cluster 1 of FD, the strongest reduction was observed from 20 °C to 24 °C.
Temperature-mediated trait-plasticities. (a). Reaction norm clusters showing different responses to temperature (k-means clustering, using Euclidean distances, k was determined with silhouette scores) (b). Stacked bar plots showing the proportion of total phenotypic variance explained by genotype (G), environment (E), genotype-by-environment interaction (G×E) and residual variation (R) estimated using linear mixed-effects models.
For FT, ESDT and SN, the accessions could be grouped into four distinct clusters. In all clusters of reaction norms for FT, the strongest decrease was observed between 20 °C and 24 °C. In ESDT, all reaction norm clusters decreased until 24 °C, followed by a slight increase at 27 °C. The clusters of reaction norms for SN showed different temperature optima, yet, apart from two accessions in cluster 4, they all displayed the smallest number of seeds at 27 °C. These results further support the different temperature optima for the accessions. The reaction norms of SSW could be grouped into three clusters, and those of SS and FN into two clusters, indicating reduced natural variation among the accessions in plasticities of SSW, SS and FN. Furthermore, FN cluster one, which contained most of the tested accessions, did not seem to depend on temperature (Figure 2a).
To find out whether the similarity in reaction norms is associated with the geographic origin of the accessions, we examined the Mantel correlation between the distances of reaction norms and geographic distances between the points of origin for the studied accessions. We did not observe any correlation between the geographic distances of the accessions and their reaction norm distances for any of the analysed traits (Supplementary Figure 2). This suggests that local environmental factors rather than geographic origin are driving the temperature-mediated plasticity response on traits.
2.3. Variance partitioning reveals strongest GxE effects in SN and FN
The observed natural variation in the reaction norms suggests that the temperature-mediated plasticity in these traits is due to genotype-by-environment interaction (GxE). To further validate this claim, we investigated the contribution of genotype (G), environment (E) and genotype-by-environment interaction (GxE) on the observed variation across the temperatures. For all traits, GxE contributed to some degree to the total phenotypic variance (Figure 2b). FD showed the smallest proportion of GxE interaction (3.7%). SN and FN showed the least variation due to genotype, 10% and 3%, respectively. However, these traits exhibited the largest proportion of variance explained by GxE with 28% and 34%, respectively (Figure 2b, Supplementary Table 1). The sizeable GxE variance findings point out that there is a genetic basis underpinning the observed differences in temperature-mediated plasticity in FN and SN.
2.4. Partial correlations between trait plasticities
Here, due to the possible confounding effects of life-history traits, we used partial correlation to analyse trait and trait plasticity relationships. We first asked if the trait plasticities are associated with the mean value of the respective traits. We did not observe any significant partial correlations between the trait means and the overall trait plasticities (Supplementary figure 3). Next, we investigated whether the different plasticities exhibited strong correlations. For this, we quantified three consecutive plasticities (17→20 °C, 20→24 °C, 24→27 °C) for each trait. Partial correlation analysis revealed that for all plasticities, SN and FN showed significant positive correlation (Figure 3b, S upplementary Figure 4). Furthermore, all plasticities of SS showed positive correlation with the plasticities of SN and negative correlation with the plasticities of FN. This is in line with the observed partial correlation between the mean values of these traits (Figure 1c) and further confirms the strong effect of FN on seed yield. In addition, the plasticity of FT between 20 °C and 17 °C showed significant negative correlation with FN and ESDT plasticity but resulted in positive correlation with the SN plasticity. These findings indicate that at lower temperatures, later flowering increased the overall SN but reduced the FN and the ESDT. Furthermore, FT plasticity showed a significant positive correlation with FD plasticity between 17 °C and 27°C, indicating that there is a link between the flowering time and flower size, but this only becomes significant with larger temperature changes.
Correlations between trait plasticities. (a). Partial correlations for plasticities of the three consecutive temperature changes and the overall temperature change. permutations = 10000, adjusted p-values with Benjamini–Hochberg. (b). Linear model showing predictive power of all floral traits combined as one predictor at each temperature (indicated by number on the bar) for seed number plasticity. R2 value specified above bars for predictions with R2 ≥ 0.3. (c). Spearman correlation of traits and latitude at the four temperatures, as well as overall plasticity. P-values calculated by permutation (n = 10,000, BH adjusted). * p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001; Rho shown in cells for significant correlations.
In addition, we asked if the combination of the three floral traits (i.e., FT, ESDT, FD) can be used to predict the plasticity in each of the four fitness traits (i.e., SSW, SS, SN, FN). To this end, we used linear multivariate regression analysis (Supplementary Figure 5) that allows us to investigate whether variation in floral traits is predictive of plasticity differences in fitness-related traits. Our results showed that SN plasticity between consecutive temperature intervals could be predicted from floral traits measured at either temperature. For the 17→20 °C interval, R 2 was 0.47 when using traits at 17 °C and 0.48 when using traits at 20 °C. For 20→24 °C, R 2 was 0.30 (20 °C traits) and 0.45 (24 °C traits), and for 24→27 °C, R 2 was 0.33 (24 °C traits) and 0.35 (27 °C traits). Unlike the plasticity between consecutive temperature intervals, the overall plasticity in fitness traits across the full temperature range (17→27 °C) could not be predicted from the floral traits (R 2 < 0.3; Figure 3b), suggesting that the link between floral traits and the plasticity of fitness traits is subtle and condition-dependent.
2.5. The correlation between traits and latitude depends on temperature
Our previous analysis suggested that local environmental factors, rather than geographic proximity of accessions, underlie the different trait plasticity responses in plants (Supplementary Figure 2). Correlation analysis with all traits, their temperature-mediated plasticities, and latitude of accession origin revealed that SN and FN at 17 °C showed a significant negative correlation with latitude and a significant positive correlation at 27 °C. At 20 °C, correlations remained negative, while at 24 °C they were positive, although weaker than at the extremes, reflecting a clear temperature-mediated shift in trait–latitude relationships (Figure 3c, Supplementary figure 6). In addition to fitness traits, flowering time at 17 °C and at 20 °C showed significant positive correlation, and early silique development at 17 °C and at 20 °C showed significant negative correlation with latitude. Moreover, FD showed a positive correlation with latitude at 17 °C and 20 °C (Figure 3c). This finding, together with the significant negative correlation between FT and ESDT specifically at 17 °C and 20 °C (Figure 1c), suggests that accessions tend to flower later with larger flowers but initiate early silique development more rapidly with higher latitudes, likely reflecting adaptation to shorter growing seasons. Correlation analysis of trait plasticity between 17 °C and 27 °C revealed a significant positive relationship with latitude for ESDT, SN and FN. This finding indicates that northern accessions tend to exhibit greater plasticity in these traits across the temperature range, consistent with the temperature-dependent trait–latitude patterns observed at individual temperatures (Figure 3c, Supplementary figure 6). These results, together with the lack of correlation among ESDT, FD, SN and FN and longitude (Supplementary Figure 7), support our hypothesis that temperature-mediated traits, specifically ESDT, FD, SN and FN, are connected to the latitudinal origin of the accession.
3. Discussion
Reproduction through selfing in A. thaliana is associated with a reduction in flower size in comparison to its sister species, A. lyrata (Sicard & Lenhard, Reference Sicard and Lenhard2011). However, A. thaliana remains capable of outcrossing, which may provide a fitness advantage in highly variable environments where genetic diversity is an advantage. This is supported by reports of natural A. thaliana populations with general outcrossing rates ranging from <0.3% to 2.5% (Abbott & Gomes, Reference Abbott and Gomes1989; Bomblies et al., Reference Bomblies, Yant, Laitinen, Kim, Hollister, Warthmann, Fitz and Weigel2010; Picó et al., Reference Picó, Méndez-Vigo, Martínez-Zapater and Alonso-Blanco2008). However, under certain circumstances these can be as high as 20% (Bomblies et al., Reference Bomblies, Yant, Laitinen, Kim, Hollister, Warthmann, Fitz and Weigel2010). Outcrossing ability does further correlate with floral display (Goodwillie et al., Reference Goodwillie, Sargent, Eckert, Elle, Geber, Johnston, Kalisz, Moeller, Ree, Vallejo-Marin and Winn2010), yet whether there is a connection between flower size and fitness in selfing A. thaliana has not been studied so far. This study provides novel information on how floral traits and their plasticities are connected to fitness traits across different ambient temperatures of A. thaliana.
Previous studies have examined the effects of temperature changes on various traits across the full life cycle of A. thaliana but included only a limited number of accessions and did not consider flower size (Huang et al., Reference Huang, Footitt and Finch-Savage2014; Ibañez et al., Reference Ibañez, Poeschl, Peterson, Bellstädt, Denk, Gogol-Döring, Quint and Delker2017). In our study, flower diameter in A. thaliana did not show a strong correlation with fitness. Only seeds per silique showed a significant negative correlation with flower size at 17 °C that could reflect a trade-off between flower size and the number of ovules per flower. This suggests that floral morphology plays a limited role in reproductive success, consistent with A. thaliana’s predominantly selfing reproductive strategy. In a survey of angiosperms, it was shown that in selfing species, resource allocation to investment in traits such as flower number and flower size is generally reduced compared to outcrossing species (Goodwillie et al., Reference Goodwillie, Sargent, Eckert, Elle, Geber, Johnston, Kalisz, Moeller, Ree, Vallejo-Marin and Winn2010). Based on our data, reaction norms for flower number (FN) as well as the seeds per silique (SS) were key estimates for the reproductive success of selfing A. thaliana.
To capture non-linear plasticity patterns, we examined temperature-mediated trait plasticity using reaction norms across 17 °C, 20 °C, 24 °C and 27 °C. All traits showed variation among genotypes in reaction norms across the temperatures, also indicative of the presence of GxE to temperature. Among all traits analysed, flower number exhibited the lowest plasticity. This was indicated by the shallowest reaction norm slopes, the smallest fold changes and the fewest distinct clusters. This buffering against temperature variation further supports the essential role of flower number in fitness. Due to the non-linearity observed in some trait responses, we calculated correlations not only using fold changes across the entire temperature range, but also between consecutive temperature steps. As expected, due to positive trait correlation, FN and seed number (SN) showed significant correlation also for each pairwise plasticity. These findings are consistent with the idea that fitness traits tend to be canalized, while underlying morphological or developmental traits display greater plasticity (Stearns & Kawecki, Reference Stearns and Kawecki1994). The observed GxE in the traits further indicates that plasticity to temperature in these traits could be under selection and play a role in plant evolution.
Our analysis of the relationship of phenotypic plasticity with latitude revealed that FN and SN plasticity were strongly and positively correlated with latitude, indicating that high latitude accessions increase their reproductive output in response to increasing temperature. Even though little is known about latitudinal clines in A. thaliana, especially under warming, a global meta-analysis comprising 126 studies with several species has shown that warming has an increased positive effect at higher latitudes on traits such as flower number, fruit number and fruit weight (Dobson & Zarnetske, Reference Dobson and Zarnetske2025). We also observed a strong positive correlation of flowering time (FT) with latitude at 17 °C and 20 °C, suggesting a strategy in the northern accessions that prevents flowering too early in the spring. A latitudinal cline in flowering time has previously been shown for accessions carrying functional FRIGIDA (FRI) alleles grown in a common garden (Debieu et al., Reference Debieu, Tang, Stich, Sikosek, Effgen, Josephs, Schmitt, Nordborg, Koornneef and de Meaux2013). In addition, flowering time has been shown to increase with the latitudinal origin of accessions with and without functional FRI alleles (Lempe et al., Reference Lempe, Balasubramanian, Sureshkumar, Singh, Schmid and Weigel2005). Interestingly, as in our results, this correlation was stronger at 16 °C in comparison to 23 °C. Conversely to flowering time, our data showed that ESDT exhibited a strong negative correlation with latitude when plants were grown at 17 °C and 20 °C. We also found that, in general, late flowering plants produce siliques faster at lower temperatures, a trend lost with rising temperature.
This study indicated that latitude-dependent adaptation has shaped the balance between reproductive success and temperature response. Yet, correlations between traits and latitude of origin do not by themselves provide conclusive evidence of adaptation, but this requires further experiments. The contrasting responses of seed number and seed weight further suggest a trade-off, where northern accessions shift towards seed number at high temperatures, whereas southern accessions may prioritize investment in fewer, larger seeds.
4. Materials and methods
4.1. Plant material and growth conditions
To assess fitness and trait associations with respect to temperature, we ordered 34 A. thaliana accessions from the European Arabidopsis Stock Centre (NASC). We chose the accessions based on their temperature-mediated plasticity reported in Wiszniewski et al. (Reference Wiszniewski, Uberegui, Messer, Sultanova, Borghi, Duarte, Vicente, Sageman-Furnas, Fernie, Nikoloski and Laitinen2022) and are listed in Table S1. Prior to sowing, seeds were stratified in 0.1% agarose in the dark at 4 °C for 3 days, whereafter we sowed them in a 2:1 (v/v) peat and vermiculite mixture and germinated them at 23 °C/19 °C and 12h light (180 μmol/m2/s) /12h dark for ten days. After this, we vernalised the plants for six weeks in 12h light/ 12h dark at +4 °C. After vernalisation, we transplanted 16 plants, of each accession, four for each of the temperatures, to new pots (1:2 (v/v) peat and vermiculite mixture) and grew the plants in growth chambers (Percival Plant Research Chamber SE-41AR2CLT assembled by CLF Plant Climatics) at four different temperatures, 17 °C, 20 °C, 24 °C and 27 °C, with 16h/8h day/night cycle and 60% of relative humidity under 180 μmol/m2/s light. We randomised the pots in trays of 40 plants and rotated the trays every 3 days to minimize edge effects on phenotypes. We applied Fertiliser (NPK 7-2-2) after 50% of plants in a tray had bolted (inflorescence ~ 1 cm). Due to space limitations and the fact that some of the accessions did not fit in the growth chamber, we kept the plants in the growth chamber until they had produced ten siliques. Silique number was monitored independently for each plant, and after reaching ten siliques, each plant was moved to the greenhouse (average conditions during trial: 23 °C, 61% rH, with a total radiation average of 364 μmol/m2/s measured outside the greenhouse). At this point, they were fertilised once more. We stopped watering all accessions after 10 days in the greenhouse. We harvested seeds and siliques from fully dry plants.
4.2. Phenotyping and trait quantification
In all four temperatures, we scored the plants for flowering time (FT), flower diameter (FD), the early silique development time (ESDT), single seed weight (SSW), seeds per silique (SS), seed number (SN) and flower number (FN). Summary data can be found in Table S1 and raw data is in Table S2. Summary statistics are in Table S3. FT was quantified as the number of days it took the seedlings to bolt (main inflorescence ~ 1cm). For FD measurements, the flowers were placed in 96-well plates containing 0.7% agar mixed with 0.5% char coal and photographs were taken from above. FD was measured using ImageJ as the average distance between the far ends of each pair of opposing petals from 4 open flowers picked from each replicate between the 3rd and 10th flower of the main inflorescence. ESDT was scored as the number of days to produce ten siliques from the time of bolting. To score SSW and SS, we weighed and counted seeds from ten siliques from the main stem of each replicate for each accession along the height of the plant. SSW was then calculated for each replicate per accession separately as the weight of seeds per 10 siliques/ number of seeds per ten siliques. SS was calculated for each plant as number of seeds per 10 siliques /10. To calculate SN, we additionally scored the total seed weight, which we divided by SSW. FN, reflecting the number of siliques, was estimated as SN/SS. All traits were averaged across biological replicates and pairwise plasticities were calculated as fold change by dividing the warmer temperature by the colder temperature relative to the temperatures compared. The coefficient of variation (CV) across the average trait values among accessions per trait for each temperature was measured as the standard deviation of the mean trait values of all accessions divided by the average of the mean trait values of all accessions.
4.3. Analysis of the phenotype data
We performed all data analysis in R (R version 4.5.0). For the overall quality of the data, we performed PCA using the stats package and the prcomp() function. Outliers were removed for each trait within each temperature treatment using the 1.5× Interquartile Range (IQR) rule (Tukey, Reference Tukey1977), according to which 1.5 IQR points below the first quartile of data or above the third quartile is an outlier. After outlier removal, we partitioned the phenotypic variance of each trait into components attributable to genotype, environment and genotype-by-environment interaction using a linear mixed-effect model implemented with the lmer() function from the lme4 package (Bates et al., Reference Bates, Mächler, Bolker and Walker2015). Accessions, temperature treatments and their interaction were treated as random effects in the model to estimate the proportion of variance explained by each source. After determining the sources of variance, the different plasticity patterns of the traits were investigated from the reaction norms across traits and temperatures.
4.4. Analysis of the reaction norms
To ensure the reliability of the data, only accessions with≥ 3 replicates were clustered based on their response for each of the traits across the four temperatures (17 °C, 20 °C, 24 °C and 27 °C). To do this, we first constructed a custom matrix encoding plastic responses for each trait between consecutive temperature pairs. Each accession was assigned a value of -1, 0 or 1 for every comparison, representing a significant negative change, no significant change or a significant positive change, respectively, based on Wilcoxon rank-sum tests (α = 0.05). Accession similarity was then defined using Euclidean distance, and K-means clustering was applied to group accessions based on their overall plasticity patterns. To ensure stability and accuracy of the results, clustering was repeated 10 times with different random seeds. The number of clusters (k) was varied from 2 to half the number of accessions included, and the optimal cluster number was determined using silhouette scores (Maechler et al., Reference Maechler, Rousseeuw, Struyf, Hubert and Hornik2021; Rousseeuw, Reference Rousseeuw1987). If the improvement in average silhouette width from k to k+1 was <0.05, the lower-k solution was preferred. Final cluster assignments were taken from the best-performing k-means model, ensuring a robust representation of plasticity patterns across temperature treatments.
4.5. Correlations among traits, traits plasticity and geography
To investigate whether accessions that are different in reaction norm are also spatially separated, we performed a Mantel test using mantel() from the vegan package (Mantel, Reference Mantel1967; Oksanen, Reference Oksanen2025) https://cran.r-project.org/web/packages/vegan and evaluated the correlation between the distance in reaction norms and geographical distance. Trait distance was represented by the Euclidean distances calculated from our custom matrix (as described in the previous section) and geographic distance was represented by the haversian distance between the origin of accessions. The Mantel correlation coefficients (r M) were calculated as Spearman’s rank correlation between the two matrices. A permutation test (n = 10000) was used to generate a null distribution for significance testing (Manly, Reference Manly and Manly2007). P-values were calculated as the proportion of permuted Mantel statistics that were equal to or more extreme than the observed r M value. To account for multiple testing across traits, p-values were adjusted using Benjamini–Hochberg false discovery rate implemented via p.adjust() (Benjamini & Hochberg, Reference Benjamini and Hochberg1995).
To assess whether trait values and trait plasticities were associated with latitude or longitude, we calculated Spearman rank correlation coefficients (ρ, rho) between each at every temperature treatment. A permutation test (n = 10000) was used to generate a null distribution for significance testing. P-values were computed as the proportion of permuted rho that were at least as extreme as the observed rho, using a directional test based on the sign of the observed coefficient. To account for multiple testing across all comparisons, p-values were adjusted using Benjamini–Hochberg false discovery rate implemented via p.adjust().
To investigate how trait–latitude relationships varied across growth temperatures, we fitted linear mixed-effects models (LMMs) (Bates et al., Reference Bates, Mächler, Bolker and Walker2015; Kuznetsova et al., Reference Kuznetsova, Brockhoff and Christensen2017) with trait values as the response, latitude and temperature as fixed effects, their interaction and accession identity as a random effect to account for repeated measurements. Model fit was assessed using marginal and conditional R 2 values, extracted using MuMIn (Bartoń, Reference Bates, Mächler, Bolker and Walker2022) and quantifying the variance explained by fixed effects alone versus the full model including random effects. The significance of the latitude × temperature interaction was evaluated using type III ANOVA with Satterthwaite’s approximation for degrees of freedom (Kuznetsova et al., Reference Kuznetsova, Brockhoff and Christensen2017). To identify which temperature contrasts exhibited significantly different slopes of trait vs latitude, we performed pairwise comparisons of estimated marginal slopes (Lenth, Reference Lenth2026) (emtrends from the emmeans package; https://CRAN.R-project.org/package=emmeans). For visualization, simple linear regressions were fitted separately for each temperature to illustrate trait–latitude relationships, with R 2 values reported.
4.6. Analysis of correlations between traits and trait plasticity
Next, to examine trait–plasticity correlations independently of shared influences, we performed partial correlation as we expected the traits to influence each other. Trait relationships were assessed using Spearman partial correlation coefficients (ρ, rho) calculated with pcor (method = spearman) from the ppcor package. A permutation test (n = 10000) was used to generate a null distribution for significance testing. P-values were computed as the proportion of permuted partial-correlation coefficients that were at least as extreme as the observed rho, using a directional test based on the sign of the observed coefficient. To account for multiple testing across all comparisons, p-values were adjusted using Benjamini–Hochberg false discovery rate implemented via p.adjust().
First, we tested partial correlation between our traits as well as plasticities at each temperature or temperature change. Additionally, we assessed the correlation between traits and plasticity across the entire temperature range for which we calculated plasticity as a fold change between 17 °C and 27 °C and the mean for each trait across all four temperatures.
Finally, we used linear regression (lm()) to test whether all floral traits (FT, FD, ESDT) combined could predict fitness trait plasticity. For this, we considered two sets of predictors: (i) floral traits measured at either of the temperatures of the target plasticity interval were used separately as one predictor, (ii) floral traits averaged across all temperatures, combined to predict fitness trait plasticities across the overall temperature range from 17 °C to 27 °C. The R 2 values were extracted to assess the strength of these associations.
Open peer review
To view the open peer review materials for this article, please visit http://doi.org/10.1017/qpb.2026.10045.
Acknowledgements
The authors thank Airi Lamminmäki, Jesse Pikkarainen and Leena Grönholm for their help with experiments and growing plants.
Competing interest
The authors declare no conflict of interests.
Data availability statement
The summary and raw data that support the findings of this study are available in supplemental information. The codes are available at GitHub. https://github.com/PlantAdaptationGroup/Temperature-mediated-plasticity-of-floral-and-fitness-traits-in-Arabidopsis-thaliana
Author contributions
J.H., G.M.A-H. and R.A.E.L. designed research; J.H. and G.M.A.-H. performed the experiments; J.H. analysed data; Z.N. and R.A.E.L. supervised the research; and J.H. and R.A.E.L. wrote the paper with input from G.M.A.-H. and Z.N.
Funding statement
R.A.E.L. is supported by grants from the Emil Aaltonen foundation and Research Council of Finland (grant number: 356746) and from University of Helsinki. Z.N. was supported by funding from the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) - SFB 1644/1 - 512328399.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/qpb.2026.10045.
Use of AI tools
ChatGPT was used to correct grammatical errors in the language of the manuscript and for assistance in writing R scripts.
Open access
Comments
Dear Dr. Hamant,
We would like to submit for the first time our manuscript entitled “Temperature-mediated plasticity of floral and fitness traits in Arabidopsis thaliana” for consideration for publication in Quantitative Plant Biology.
Temperature is a major environmental cue projected to increase in future, with potentially dis-tressing effects on many organisms, including plants. In plants, elevated temperatures are known to influence the growth and development of various traits, a phenomenon known as thermomorphogenesis. Recent studies have also shown that temperature can induce plasticity in floral traits; for instance, petal size and number are known to vary in response to temperature. Floral traits are directly linked to reproductive strategies, but it is still unclear whether temperature-mediated changes in floral traits and their plasticity are associated with fitness.
In this manuscript, we investigate seven life-history traits across four ambient temperatures in a panel of 34 A. thaliana accessions. Based on reaction norms for trait means across temperatures, we found that all traits exhibited temperature-mediated plasticity. The clustering of the reaction norms and GxE analysis revealed that the different traits had independent underlying genetic mechanism. From the traits and their plasticities, the strongest positive correlation was found between flower and seed numbers, indicating that flower number is strongly linked with fitness. In addition, we found a temperature-dependent significant posi-tive correlation between flower size and latitude, with correlation only at 17 °C and 20 °C, but not at 24 °C or 27 °C. This result suggests local adaptation for flower size. Altogether, our results provide new insights into the impact of temperature on trait variation and fitness in a predominantly selfing species.
We suggest the following experts as potential referees for our manuscript:
• Fabrice Roux, LIPME, CNRS, Toulouse, fabrice.roux@inrae.fr
• Xavier Picó, Estación Biológica de Doñana (EBD-CSIC), Sevilla, Spain, xpico@ebd.csic.es
• Johanna Schmitt, UC Davis, California, USA, jschmitt@ucdavis.edu
• Pieter Arnold, The Australian National University, pieter.arnold@anu.edu.au
Thank you for considering our manuscript for publication in Quantitative Plant Biology.
Sincerely,
Roosa Laitinen, corresponding author, Roosa.Laitinen@Helsinki.fi
On behalf of Jan Hoffman, Gregory Andreou-Huotari and Zoran Nikoloski (co-authors)