Summary
Plant pathogen populations inhabit patchy environments with contrasting, variable thermal conditions. We investigated the diversity of thermal responses in populations sampled over contrasting spatiotemporal scales, to improve our understanding of their dynamics of adaptation to local conditions.
Samples of natural populations of the wheat pathogen Zymoseptoria tritici were collected from sites within the Euro-Mediterranean region subject to a broad range of environmental conditions. We tested for local adaptation, by accounting for the diversity of responses at the individual and population levels on the basis of key thermal performance curve parameters and ‘thermotype’ (groups of individuals with similar thermal responses) composition.
The characterisation of phenotypic responses and genotypic structure revealed: (i) a high degree of individual plasticity and variation in sensitivity to temperature conditions across spatiotemporal scales and populations; (ii) geographic adaptation to local mean temperature conditions, with major alterations due to seasonal patterns over the wheat-growing season.
The seasonal shifts in functional composition suggest that populations are locally structured by selection, contributing to shape adaptation patterns. Further studies combining selection experiments and modelling are required to determine how functional group selection drives population dynamics and adaptive potential in response to thermal heterogeneity.
Introduction
Populations in natural settings often experience environmental heterogeneity (Li & Reynolds, 1995), which dictates physiological responses (Cavieres & Sabat, 2008) and can drive the emergence of local adaptation patterns (Thompson, 2005; Nuismer & Gandon, 2008). As such, heterogeneity is regarded as one of the most important elements driving the emergence and maintenance of genetic variation within populations (Levins, 1974; Hedrick, 1986; Ravigné et al., 2009), thereby shaping population dynamics (Hughes et al., 2008). Gathering information about the way a given community, species or population copes with this environmental heterogeneity is crucial for the understanding and prediction of its distribution and responses to current and future environmental changes (Austin, 2007).
The adequate capture of eco-evolutionary responses requires an integration of physiological variation across biological (individual, group, population, species) and spatiotemporal (seasonal, geographic) scales, given the significant implications of this variation for dynamics (Saloniemi, 1993; Vindenes et al., 2008; Schreiber et al., 2011). It is therefore important to go beyond summarising diversity through average trait values (Bolnick et al., 2011; Violle et al., 2012), and to account for the individual specialisation of phenotypic responses by taking into account both phenotypic plasticity (within-individual differences; Pigliucci, 2001) and interindividual variation (between-individual differences; Dall et al., 2012). This paradigm shift has been made possible by progress in the measurement and analysis of this specialisation at the individual and population level.
The ecological concept of “reaction norm”, describing the set of phenotypes generated by a given genotype in response to environmental cues (Schlichting & Pigliucci, 1998), is particularly effective for accounting for individual specialisation (Bolnick et al., 2002; Araújo et al., 2008). Most of the comparisons on the continuous variation of a given phenotypic trait among individuals under different environmental conditions have been conducted to date on reaction norm descriptors (e.g. comparisons of phenotypic mean and range values at the maximum; Gibert et al., 1998) or degree of plasticity (e.g. breakdown of the shifting and stretching of non-linear reaction norms into non-exhaustive biological modes, i.e. lower-higher, faster-slower, specialist-generalist directions; Izem & Kingsolver, 2005; Martin et al., 2011; van de Pol, 2012). Such approaches have proved highly valuable, but may not be suitable for decomposing the overall variation or distinguishing differential responses between populations (Bulté & Blouin-Demers, 2006) or including intra- and interindividual sources of error (Angilletta, 2006) in ANOVA and random regression approaches (Lynch & Gabriel, 1987; Gilchrist, 1995).
One possible complementary approach to the description of variation between reaction norms involves the use of functional ecology to describe significant variations in the intensity of individual specialisation within populations and species (Garnier & Navas, 2012). The idea is to translate reaction norms into functional traits (Violle et al., 2007) by grouping individual reaction norms into ‘functional groups’ (Gitay & Noble, 1997). Each of these functional groups responds to the environment in its own way (e.g. low- or high-performance specialists), according to a classification system that is not predetermined (i.e. constrained modes of variation). This approach accounts more effectively for diversity and adaptation patterns, through the characterisation of three functional components: richness, evenness, and divergence (Mason et al., 2005).
This approach is particularly useful for deciphering variation in continuous reaction norms describing performance as a function of temperature (thermal performance curves or TPC; Huey & Stevenson, 1979), and for documenting patterns of thermal adaptation to prevailing local conditions (Kawecki & Ebert, 2004) across a range of environments (e.g. Mitchell & Lampert, 2000). These patterns plays an important role in the case of microorganisms impacting ecosystems, human health, and food security (Fisher et al., 2012) as local adaptation to temperature conditions governs their geographic distribution, phenology, and abundance (Kraemer & Boynton, 2017). This results in impacting the expansion ranges of plant pathogens (e.g. Milus et al., 2009; Robin et al., 2017), as well as the onset and severity of disease epidemics (e.g. Ferrandino, 2012).
The studies of thermal responses in plant pathogenic microorganisms performed to date have focused mostly on either summarising the individual variance of aggressiveness traits as population-scale averages (problematic use of single mean species values; Suffert & Thompson, 2018) or phenotyping individuals under a limited set of temperatures when considering variances (generally about three temperatures in thermal biology studies; Dell et al., 2013; Low-Décarie et al., 2017). These strategies have provided useful information about species distribution, making it possible to detect signatures of interindividual variation and adaptation within species and populations (Milus et al., 2006). However, they cannot be used to infer group selection driving population dynamics (Lavergne et al., 2010) or to assess the relevant scales of functional diversity (Woodcock et al., 2006; Martiny et al., 2011). Such analyses go well beyond simple comparisons of mean trait values and would require the characterisation of entire TPCs and their variation across different scales.
This study develops a functional approach ascertaining patterns of diversity across space (geographic range) and time (local seasonal dynamics) in a way that explores how functional diversity in thermal responses varies with environmental conditions and uncovers the role that adaptation plays in generating this diversity. The analysis of the plasticity and variation of thermal sensitivity across individuals, populations, and scales was conducted in the case of the fungal wheat pathogen Zymoseptoria tritici (formerly Mycosphaerella graminicola; Steinberg, 2015). This pathogen, the causal agent of one of the most economically important wheat diseases (Septoria tritici blotch or STB; Dean et al., 2012; Fones & Gurr, 2015), is now considered a model species for both basic and applied research. Besides its agronomic relevance, we chose to study Z. tritici as its aggressiveness traits are known to be temperature-sensitive (Shaw, 1990; Lovell et al., 2004) and to display interindividual variation (Bernard et al., 2013; Boixel et al., 2019). Furthermore, its populations present signatures of adaptation to a wide range of contrasted environments over space (wheat-growing areas worldwide; Zhan & McDonald, 2011) and time (covering seasonal changes e.g. from late autumn to early summer in Europe; Suffert et al., 2015). Drawing on previous local adaptation studies conducted by Zhan & McDonald (2011) and Suffert et al. (2015), we designed a sampling scheme to grasp the levels of functional diversity shaping responses of its populations to contrasted environments based on a finer-grained sampling and phenotyping resolution.
Materials and Methods
Tailored sampling survey design (Fig. 1 – step 1)
Samples were collected from 12 Z. tritici populations for the exploration of spatial and temporal components of thermal adaptation. Spatial variation was investigated within the Euro-Mediterranean region, with samples from eight sites in wheat-growing areas representative of the highly contrasting climatic conditions over this large geographic area (see the 8 populations for the geographic scale, covering three out of seven Köppen-Geiger climate zones in which Z. tritici is reported as a notable pathogen, in Table 1 and Fig. S1). One of these sites (Grignon, France) was selected for a comparison of the thermal responses of two pairs of winter and spring subpopulations sampled from neighbouring fields, to capture seasonal dynamics over a wheat growing season (i.e. over the course of an annual epidemic; see the 4 populations of the seasonal scale in Table 1 and Fig. S2). These pairs of subpopulations were subject to seasonal variation from November to February and from March to June, respectively. For each of the 12 populations, we collected from 25 to 30 isolates at random from wheat leaves with STB symptoms, from which single-spore isolates were prepared (Methods S1) and which were later confirmed to be genetically unique strains with the microsatellite analysis. We chose to consider 25 or 30 strains (i.e. individuals) per population instead of the minimum level of 15 identified on the basis of a rarefaction analysis (Fig. S3) for estimating the diversity of thermal responses in Z. tritici with sufficient power, accuracy and precision (Dale & Fortin, 2014).
Phenotypic variations in thermal responses (Fig. 1 – step 2)
Thermal responses were phenotyped by determining the in vitro growth rates of the strains in liquid glucose peptone medium (14.3 g.L−1 dextrose, 7.1 g.L−1 bactopeptone and 1.4 g.L−1 yeast extract) over a four-day period at 12 temperatures ranging from 6.5 to 33.5°C (6.5, 9.5, 11.5, 14.5, 17.5, 20.0, 22.5, 24.5, 26.5, 28.5, 30.5 and 33.5°C). The growth rate μ of each strain at each temperature was calculated according to the standardised specific experimental framework developed by Boixel et al. (2019) which have been validated to be representative of in planta responses with respect to discrimination between cold- and warm-adapted individuals. TPCs describing in vitro growth rate as a function of temperature were established by fitting a quadratic function to the temperature–growth rate (or performance P) estimates for each strain: P(T) = Pmax + Curv(T − Topt)2 where Curv is a shape parameter (Table S1 for more information on the model selection process). The key properties of TPCs were estimated through thermal traits commonly used to compare thermal sensitivities (Kingsolver, 2004; Angilletta, 2006). We have retained three parameters to describe the shape of these TPCs and quantify their characteristics: first, maximum performance (Pmax) which informs on TPC height (‘vertical shift’ modes of variation); second, thermal optimum (Topt) which informs on TPC position at the peak performance (‘horizontal shift’ modes of variation); third, thermal performance breadth (TPB80) which informs on the sensitivity of the response to temperature change around Topt (‘width shift’ modes of variation). The estimates of Tmin and Tmax were not retained for further analysis as they fell outside the range of temperatures tested. TPC variation was assessed in two ways: (i) differences in the range and mean values of Pmax, Topt, and TPB80, assessed with parametric or non-parametric (depending on whether the assumptions of normality and homoscedasticity were verified) statistical tests for comparing variances and means; (ii) typological comparisons grouping together TPCs with similar thermal characteristics (functional thermal groups, referred to hereafter as ‘thermotypes’) based on a K-means clustering procedure applied to the covariation of Pmax, Topt, and TPB80 for all TPCs (Methods S2). The magnitude and distribution of diversity for thermal traits and thermotypes were analysed across individuals, populations, and scales, to detect differentiation in phenotypic patterns, in Chi-squared tests on the observed frequency distribution of thermotypes.
Neutral genetic variation and population differentiation (Fig. 1 – step 3)
The 350 individuals composing the 12 Z. tritici populations were genotyped for 12 neutral genetic markers on DNA extracted from 50 mg of fresh fungal material from five-day cultures, following SSR amplification and sequencing in one multiplex PCR sample, and allele size annotation (Gautier et al., 2014; Methods S3a). Population structure was inferred with a Bayesian clustering approach under an admixture and correlated allele frequencies model implemented in STRUCTURE (Pritchard et al., 2000). The degree and significance of genetic variability within a population (genetic diversity and allele richness) and differentiation between populations (pairwise estimates of Weir and Cockerham’s F-statistic – FST – and hierarchical analyses of molecular variance – AMOVA) were evaluated with random allelic permutation procedures in GENETIX (Belkhir, 2004) and Arlequin (Excoffier & Lischer, 2010) software (Methods S3b-d).
Dimensionality of mesoclimatic environmental variation (Fig. 1 – step 4)
Air temperature data for the closest weather stations within a mean 30-km radius of the eight sampling sites were retrieved from archives of global historical weather and climate data, to obtain: (i) monthly-averaged values of 1961-1990 climate normals (Norwegian Meteorological Institute, 2019); (ii) daily data over the sampling year (US National Climatic Data Center NCDC-CDO, 2019). Local variation in mesoclimatic temperatures were summarised for four climatic time windows (1961-1990 climate normals and thermal conditions for the year of sampling averaged over the calendar year, the wheat growing season, the winter/spring periods) and for five metrics (thermal mean, range, maximum, minimum and variance), giving a total of 20 thermal variables at each site. We then established a thermal niche (i.e. temperature conditions of a given sampling site) classification, by assessing the importance of each of these variables for discriminating between the three contrasting Köppen-Geiger climatic zones prospected, with a nonlinear and nonparametric random forest algorithm (RF; Breiman, 2001) in the ‘randomForest’ package of R (Liaw & Wiener, 2002). The importance of variables was compared on the basis of two metrics assessing the inaccuracy of RF zone classification if the variable concerned is not accounted for (RF mean decrease in prediction accuracy and node impurity, i.e. Gini coefficient).
Testing for signatures of local adaptation (Fig. 1 – step 5)
Two steps were taken to detect genetic and phenotypic signatures of local adaptation underlying the observed differentiation between populations. First, the degree of genetic differentiation for the set of neutral markers (FST index; Weir & Cockerham, 1984) was compared with that for phenotypic traits (PST index; Leinonen et al., 2006). This made it possible to infer departures from neutral expectations (Merilä & Crnokrak, 2001), to determine whether thermal traits were under selection rather than subject to genetic drift (Brommer, 2011). FST-PST comparisons were conducted separately for seasonal (on Topt) and geographic populations (on Topt and TPB80), on the basis of sensitivity analyses assessing the robustness of the conclusions to variations in the approximation of QST by PST (Methods S3e). Second, correlations between local climate conditions and Z. tritici thermal sensitivity were evaluated, to detect signatures of adaptation. Pearson correlation coefficients and their statistical significance were established for all possible combinations of thermal traits or thermotypic compositions and for the 20 spatiotemporal thermal variables defining the thermal niche of a climatic site.
Results
Marked interindividual variation in thermal traits at all scales
We observed a very high level of interindividual variation for the three thermal traits chosen to describe TPCs, within a range of 0.17 to 0.46 h-1 for Pmax (in vitro growth rate), 9.6 to 25.1°C for Topt, and 2.8 to 30.9°C for TPB80, across all 350 strains. Individual thermal phenotypes are summarised in Fig. 2 and available in Dataset S1. The average metapopulation-level responses in the seasonal and geographic data sets were remarkably similar in terms of their quadratic parameters (Welch’s two-sample t-test, P > 0.05): P(T)seasonal = 0.30 - 0.00077 × (T - 18.3)² vs. P(T)geographic = 0.30 – 0.00088 × (T - 18.2)². Interindividual variation around this average TPC was greater for the seasonal than for the geographic scale, as demonstrated by the standard shift in TPC position along the x- and y-axes (Fig. 2a) and the distinctly larger density distributions of the three thermal traits at the seasonal scale (Fig. 2b-d; Levene’s test for homogeneity of variance: P = 0.01 for Pmax; P < 0.01 for Topt; P = 0.02 for TPB80). Interindividual variation within populations was similar at both the geographic and the seasonal scales, with equivalent variances for Pmax (x̅ ± 0.06 h-1 [SD] on average), Topt (x̅ ± 2.59 °C [SD] on average) and TPB80 (x̅ ± 5.72 °C [SD] on average) within the 12 populations (Levene’s test for homogeneity of variance: P = 0.07; 0.51; 0.13, respectively). The populations may therefore be considered similar in terms of their individual variances for thermal traits. By contrast, they were not similar in terms of the corresponding population means, as significant differences were detected for Topt and TPB80 (P < 0.05) but not for Pmax (Pgeographic = 0.09; Pseasonal = 0.75; contrary to what would be expected under the ‘warmer is better’ hypothesis in thermal biology but which is not surprising given the fact that this is a fungus with limited growth at high temperatures; Bennett, 1987).
A functional reading grid for diversity in individual thermal responses
TPCs were classified into thermotypes with similar thermal responses (Hopkins’ statistic of 0.71, indicating clustered data and justifying the establishment of such a typology; Methods S2a). The diversity of TPCs encountered in the data set was optimally partitioned into 13 thermotypes (Th1 to Th13; Fig. S4), for which relative degrees of temperature specialisation were described in terms of the Topt (cold- vs. warm-adapted), TPB80 (specialist vs. generalist), and Pmax (low vs. high performer) dimensions (Fig. 3a). These thermotypes illustrated two commonly documented non-exclusive shifts in TPC along thermal gradients: a horizontal shift (low-temperature vs. high-temperature generalists or low-temperature vs. high-temperature specialists; e.g. Th1 vs. Th13 in Fig. 3b) and a generalist-specialist shift without (Th8 vs. Th9 in Fig. 3c) or with (Th1 vs. Th3 or Th11 vs. Th13 in Fig. 3d) trade-offs between Pmax and TPB80 (i.e. when one cannot increase without a decrease in the other). Indeed, regression analysis revealed a significant negative correlation between Pmax and TPB80 across all individuals (Pearson’s correlation coefficient: R = -0.44; P < 0.01). About 10% of individuals did not follow this pattern, with high values of both Pmax and TPB80. These individuals (i.e. the strains of Th8), are both ‘jack-of-all-temperatures’ and ‘masters of all’, as they perform well at all temperatures (Huey & Hertz, 1984; Fig. 3e). Each cluster included strains from both geographic and seasonal populations (Fig. S4), but with an uneven distribution (difference in Jaccard distance, with a highest pairwise difference of 0.62 between WIN1 and SPR1) and an uneven relative abundance of the 13 thermotypes over the two scales. This relative abundance varied by a factor of up to two for the slightly adapted thermotypes Th5 and Th7. The various thermotypes were not equally distributed across the 12 populations either (Chi-squared test for given probabilities, P < 0.01). This heterogeneous distribution was particularly pronounced for high-temperature generalists (see the contributions of Th12 and Th13 to the total Chi-squared score for the comparison of distributions across seasonal and geographic populations in Fig. S5d and S6c).
Four thermotypes together accounted for almost half the entire data set (Th5, Th6, Th7 and Th10). The distinguishing features of these four thermotypes were their average behaviour with respect to Topt (Th5, Th6, Th7), TPB80 (Th5, Th10) and Pmax (Th7, Th10).
Thermal phenotypic differentiation of Euro-Mediterranean populations
For population-level TPCs, significant variation was observed for thermal trait means for Topt (Kruskal-Wallis, P < 0.01) and TPB80 (Kruskal-Wallis, P < 0.01), but not for Pmax (Kruskal-Wallis, P = 0.09), for which no population differentiation was detected (Table 2). Pmax values may have been constrained by the detection thresholds for optical density (smoothing of data for individuals with ‘extreme performance phenotypes’). A significant horizontal shift in Topt, by about 2°C compared to the 7 other populations, was observed for the IS population (Table 2), which consisted of individuals performing best at higher temperatures (Fig. 4a). Indeed, the proportion of high-temperature generalists (Th12 and Th13) was higher in the IS population (1:3 vs. 1:15 on average for the other geographic populations), accounting for 20.7 % of the imbalance in the distribution of thermotypes between populations (see contributions to the total Chi-squared score in Fig. S5d). The thermotypes best adapted to colder conditions (CA+, Th1-Th2-Th3) were particularly abundant in the Dfb populations (RU-KZ-LV), as shown by their long-tailed distributions skewed towards lower temperatures (with 6 highlighted individuals in Fig. S5a presenting a Topt of about 10.4 ± 0.7°C, i.e. about 7°C below the mean value). The IS population was characterised by a higher TPB80 for its average population response than the other populations, particularly DK (19.5 vs. 12.7°C; Table 2). These two populations had opposite patterns in terms of their respective proportions of thermal specialists and generalists (Fig. 4b and Fig. S5b). More broadly, the individuals with the greatest thermal breadth (G+, Th1-Th13) were less abundant in Cfb populations (DK-FR-IR), which were characterised by a higher proportion of more highly specialist individuals (S+, Th4 and Th9) than the average (accounting for 10% of the total Chi-squared score; Fig. S5d).
Seasonal adaptive shifts within local populations
Spring subpopulations (SPR1 and SPR2) had a higher thermal optimum than winter subpopulations (ANOVA, P < 0.01), with a horizontal shift of Topt towards higher temperature of the order of 5°C (SPR1) and 2.3°C (SPR2) on average (Table 2 and Fig. 5a). In terms of thermotype composition, these two pairs of subpopulations differed principally in their relative proportions in strains highly adapted to warm conditions (WA+). WA+ strains were significantly more abundant in SPR populations (Fig. 5b) than in WIN populations, accounting for 33.4% of the total Chi-squared score for difference in thermotype distributions between WIN and SPR populations. Conversely, WIN populations had a higher proportion of individuals highly adapted to cold conditions (CA+; Fig. S6).
Signatures of local adaptation to mean annual temperature conditions
Neutral molecular markers revealed that all strains were genetically different. We observed no difference in the genetic structure of the 12 populations, with similar allele frequencies at each locus (Fig. S7 and Table S2), suggesting a constant mixing of populations through substantial continental gene flow as underlined in previous studies for Z. tritici (Boeger et al., 1993). Strong evidence of local adaptation was detected, with the occurrence of strong phenotypic divergence (Fig. 6) and a robust PST - FST difference for the Topt of both geographic and seasonal populations and for the TPB80 of geographic populations (Fig. S8). An analysis of possible correlations between these thermal traits and the temperature conditions of the eight sampling sites (monthly averaged values of 1961-1990 climate normals) indicated that the mean thermal optimum of geographic populations increased with mean annual temperature (Fig. 7a). The level of cold adaptation of these populations (measured as the ratio of highly cold-adapted to highly warm-adapted strains) was negatively and significantly correlated with the same environmental variable (Fig. 7b). The mean annual temperature over the 1961-1990 period to which populations seemed to be adapted was one of the three thermal variables most strongly differentiating between the three Köppen-Geiger climatic zones considered in this study (highest random forest mean decrease in accuracy; Fig. S9), together with mean annual temperature over the year of sampling and seasonal contrasts. This difference between mean spring and mean winter temperatures gave the highest mean decrease in Gini index (0.27 vs. 0.25 for mean annual temperature over the 1961-1990 period). This finding highlights the potential importance of seasonal conditions in structuring the thermal responses of these geographic populations.
Discussion
Striking spatiotemporal diversity and distribution of Z. tritici thermal responses
By characterizing the TPCs of Z. tritici strains collected over different spatiotemporal scales, we were able to develop a fine description of the extensive interindividual variation in thermal plasticity: maximum performance (Pmax), thermal optimum (Topt), and thermal performance breadth (TPB80). This detailed characterisation was made possible by the large range of temperatures and the high resolution of this experimental study (12 temperatures, ranging from 6.5 to 33.5 °C), the extensive sampling strategy (350 strains from 12 populations collected within the Euro-Mediterranean region) and the use of a dedicated and previously validated experimental framework based on turbidity measurements (Boixel et al., 2019). It is important to bear in mind that these turbidity measurements may not reflect the sole growth multiplication rate via yeast-like budding but more precisely quantify the total fungal biomass that could be affected by the pleomorphic nature of some strains of Z. tritici under some environmental stimuli (e.g. partial transition to pseudohyphae or induction of a few chlamydospores, a very recently highlighted form at high temperatures; Francisco et al., 2019). Precautions were taken to work under culture conditions limiting morphological transitions in the four-day time window of the experiments: very few hyphae were observed at 96 h when validating the method (see ESM1-3 in Boixel et al., 2019), and no chlamydospore has so far been described in the literature before 96 h in liquid medium (Francisco et al., 2019). Furthermore, effects of these potential – although undetected here – morphological transitions on the estimation of thermal parameters can be neglected as it results from a double integration based on kinetics, which mathematically limits the impact of the latest time point measurements that are the most likely to be affected by morphological transitions. As such, this framework enables to detect differences in thermal sensitivity between isolates (whatever the physiological bases that underpin these differences) and to go beyond the usual tests of ‘thermal sensitivity’ based on two temperatures, which can be misleading due to the non-linearity of reaction norms (Angiletta, 2009). The interindividual variation of thermal traits was conserved across populations (similar variance within populations) but was generally more marked over the seasonal scale (for a similar average metapopulation-level response between seasonal and geographic scales). These findings are particularly striking because the choice of geographic populations made it possible to cover three contrasting Köppen-Geiger climatic zones (Fig. S1).
Singular geographic patterns of Z. tritici population adaptation to local conditions
The geographic variation of TPCs provides evidence of thermal adaptation to local conditions in Z. tritici, with: (i) an increase in the mean thermal optimum of a given population with the annual mean temperature of its location of origin; (ii) a particularly marked adaptation to high temperatures of the population sampled in Israel, consistent with the results obtained for another Israeli population investigated by Zhan & McDonald (2011); (iii) differences in the level of specialisation of individuals between populations with higher proportions of specialist individuals in the Cfb (climatic zone with lower annual temperature range) than in the Dfb (climatic zone with higher annual temperature range) populations, consistent with the assumption that thermal generalists are favoured in more variable environments. By contrast, over a smaller geographic scale (France), using the same experimental method, we detected: (i) high levels of local diversity but no structuring of thermal responses between spring populations sampled along a gradient of increasing mean annual temperature; (ii) a marked difference between post-winter populations sampled along a gradient of increasing annual temperature range: presence of thermal generalists in the population exposed to the largest annual temperature range (19.9°C) vs. the complete absence of such generalists in the population exposed to the smallest annual temperature range (11.9°C; Boixel et al., 2019). The phenotypic differentiation of thermal responses at the population level probably results from local short-term selection of the fittest strains over the course of an annual epidemic. We investigated the adaptation to the location of origin of populations with respect to mesoclimatic temperature conditions. The patterns of adaptation detected may have have been blurred by a non-optimal descriptive resolution of the thermal niche. Indeed, the microenvironment actually perceived by organisms can diverge from the surrounding macroenvironment due to complex biophysical filters across scales (here phylloclimate vs. mesoclimate; Chelle, 2005). In a second approach scaling the actual climate perceived by Z. tritici populations down to the phylloclimate would help refining the definition of a thermal niche for each population (Pincebourde & Woods, 2012; Pincebourde & Casas, 2019). Such an approach might provide deeper insight into the maintenance of high levels of diversity and some degree of maladaptation in individual thermal responses within each population.
Seasonal dynamics of thermal responses in two local Z. tritici populations
Sampling over the geographic scale occurred during spring, between the two time points investigated at the seasonal scale (i.e. post-winter and post-spring conditions). These seasonal samplings highlighted a marked seasonal shift of TPCs towards higher temperatures and changes in the thermotype composition of two local Z. tritici populations. This result is consistent with previous observations of seasonal short-term selection on aggressiveness traits (Suffert et al., 2015). This study thus reveals a two-tier thermal adaptation, with seasonal dynamics nested within and potentially occurring in each geographic local adaptation over annual epidemics. This key finding shows that adaptive patterns are ‘eco-evolutionary snapshots’ that should be interpreted with caution, to such an extent that certain evolutionary dynamics of microbial populations can be of one type over a very short time scale and another type over longer time scales. Indeed, adaptive dynamics may differ with the time scale investigated (annual or pluriannual), particularly for annual crop pathogens with both sexual and asexual reproduction cycles, such as Z. tritici (Suffert et al., 2018). Our findings could be summarised by the counterintuive statement ‘local seasonal adaptation is stronger but more fleeting than geographic adaptation’ although we would expect that regions with lower seasonal contrasts in temperature (e.g. with mild winters) will exert weaker selective pressure.
The use of sequential temporal sampling would make it possible to capture shifts in thermal adaptation over and between wheat-growing seasons and to detect potential trade-offs between aggressiveness and survival over winter (e.g. Montarry et al., 2007).
From adaptation patterns to eco-evolutionary processes
Consistent with previous studies, our findings highlight the existence of high levels of genetic diversity and an absence of its structuration across Z. tritici populations collected from local wheat fields (Zhan et al., 2001) up to the regional and continental scales (Schnieder et al., 2001; Linde et al., 2002) or over the course of an epidemic cycle (Chen et al., 1994; Morais et al., 2019). The high level of gene flow suggested by this low level of genetic differentiation between populations may partly explain the maintenance of some degree of maladaptation to local conditions (e.g. the detection of three CA+ individuals in the IS population). More generally, we observed almost all the Topt-adapted thermotypes (CA+, CA, WA, WA+) in each phenotyped population (except that CA individuals were absent from the IS population and CA+ individuals were absent from the SPR1 population), despite the clear patterns of adaptation observed for Topt and the large differences in environmental temperatures. This maintenance of diversity suggests that Z. tritici is highly tolerant to thermal variations (high probability that environmental conditions are favourable to the development of at least some individuals in a given local population). One possible explanation for this finding is that the substantial adaptation of populations to their environments (e.g. only warm-adapted individuals under a warm environment) is hindered by a balance between gene flow and local selection (Ronce & Kirkpatrick, 2001). It also raises the issues of the occurrence of counter-selection during the interepidemic period that might explains how local populations shift in thermotype structure to reestablish similar structures between years through heritability and genetic reassortment during sexual reproduction which is driven by antagonistic density-dependent mechanisms (Lendenmann et al., 2016; Suffert et al., 2019). Further studies are required to determine the extent to which the detected pattern of geographic adaptation is driven by the thermal conditions of the environment. For this, the potential counteracting effects of selection, gene flow, random genetic drift, mutation and recombination on the increase or decrease in genetic variation would need to be assessed (Hanson et al., 2012). In particular, the combination of the high diversity of thermal responses in Z. tritici highlighted here, their heritability (Lendenmann et al., 2016) and the high level of local heterogeneity within wheat canopies (Chelle, 2005) suggest that local thermal conditions probably exert strong selection pressure on thermal sensitivity (for which TPCs are probably the best proxy as they may themselves be direct targets of selection; Scheiner, 1993; Via, 1993), even in the presence of high gene flow.
Functional group composition: a browsing pattern for investigating population dynamics
Our study illustrates how the functional classification of TPCs into thermotypes with multivariate statistical procedures can provide a complementary means of deciphering diversity patterns in the biological responses quantified in reaction norms. In particular, it constitutes an operational tool for assessing functional similarity at the individual level (i.e. whether the apparent variation observed in thermal parameters is functionally significant; Fig. 8a and 8b) and at the population level (i.e. whether the thermotypes constituting a population are more or less well-differentiated within the whole functional space; Fig. 8c). However, caution will be required in the extension of this approach to comparisons over multiple data sets, through either: (i) the development of comparable classification systems, taking into account the variation of the classification with the populations sampled by explicitly stating which ranges of trait values are hidden behind a given group description (e.g. affixing levels of adaptation: very low, low, high, very high); or (ii) the validation of group delineations between experiments, by combining a priori and a posteriori methodologies. This description of populations in terms of functional groups makes it possible to move from a description of phenotypic patterns and shifts in population composition to an inference process. This process may, for example, be based on comparisons of the competitive advantage of thermotypes under given thermal scenarios: e.g. ‘do more variable environments favour thermal generalists?’ or ‘is there a shift in the optimal range of thermal responses with mean temperature conditions?’ (Fig. 8d). This classification into thermotypes enabled here to go beyond a purely descriptive framework and future investigations will need to be undertaken to tackle the physiological bases of these differentiations in thermal responses. The thought-provoking results of Francisco et al. (2019) could be used to test whether several strains belonging to those thermotypes also correspond to specific or main morphotypes that would increase their tolerance under some environmental conditions (e.g. if warm-adapted individuals exhibit higher proportions of stress tolerant growth forms such as chlamydospores under warmer temperatures). All in all, this functional approach lays the foundations for future studies of the potential of pathogen populations to adapt to changes in their environment, from seasonal changes in the short term, to global warming in the long term. In particular, it will prove useful in gaining a fuller understanding of how new aggressive fungal strains may emerge and expand into previously unfavourable environments (Milus et al., 2009; Mboup et al., 2012; Stefansson et al., 2013). This is a crucial area of investigation that is all too often overlooked in models for predicting plant disease epidemics in conditions of climate change (West et al., 2012).
Concluding remarks
We detected a high level of functional divergence in the plasticity and variation of individual thermal responses over geographic and seasonal scales, highlighting the occurrence of a two-tier dynamics in thermal adaptation. These findings raise intriguing questions regarding the mode of selection operating on these functional groups of individuals with similar competitive advantages in given thermal conditions. Deciphering the mechanisms underlying this maintenance of diversity in population phenotypic composition will prove useful for expanding our understanding of eco-evolutionary responses and the potential of populations, species and communities to adapt to environmental change.
Author contributions
ALB, MC and FS designed the study and wrote the manuscript. ALB performed the experiments and analysed the data.
Supporting information
Additional Supporting Information may be found online in the Supporting Information section at the end of the article.
Supporting Information
FIGURES
Fig. S1 Selection of the sampling sites in the Euro-Mediterranean wheat-growing area
Fig. S2 Selection of the seasonal subpopulations
Fig. S3 Appropriate sample size for estimating diversity in TPCs within Z. tritici populations
Fig. S4 Clustering of Z. tritici strains into 13 thermotypes
Fig. S5 Distribution of thermotypes across the eight Euro-Mediterranean Z. tritici populations
Fig. S6 Distribution of thermotypes across the four French seasonal Z. tritici subpopulations
Fig. S7 Genetic diversity and population structure of the 12 Z. tritici populations
Fig. S8 Sensitivity analyses for the robustness of PST–FST comparisons
Fig. S9 Characterisation of the thermal niche at each sampling site
TABLES
Table S1 Selection of a candidate mathematical model for establishing TPCs
Table S2 Population-pairwise genetic distance (matrix of FST-values)
METHODS
Methods S1 Procedure for the sampling, collection and recovery of Z. tritici strains
Methods S2 Definition of Z. tritici ‘thermotypes’ (functional thermal groups)
Methods S3 Procedure for acquiring and analysing multilocus genotypic data
Acknowledgements
We would like to thank Aigul Akhmetova (CIMMYT, Kazakhstan), Yerlan Dutbayev (National Agrarian University, Kazakhstan), Andrea Ficke (Bioforsk Plant Health and Plant Protection, Norway), Inga Gaile (Integrētās Audzēšanas Skola, Latvia), Lise Jørgensen (Aarhus University, Denmark), Steven Kildea (Teagasc, Ireland), Elena Pakholkova (Research Institute of Phytopathology, Russia), Antonio Prodi (University of Bologna, Italy), Hanan Sela (Tel-Aviv University, Israel), Sandrine Gélisse, Thierry Marcel and Anne-Sophie Walker (INRA BIOGER, France) for their crucial help in collecting wheat leaves with STB symptoms, which was required for the establishment of the Euro-Mediterranean Z. tritici collection upon which this study is based. We would also like to thank Sylvain Pincebourde (IRBI-CNRS, France) for many fruitful discussions. This research was supported by a grant from the French National Research Agency (ANR) as part of the ‘Investissements d’Avenir’ programme (SEPTOVAR project; LabEx BASC; ANR-11-LABX-0034) and by a PhD fellowship from the French Ministry of Education and Research (MESR) awarded to ALB.