Introduction

Temperature has a strong effect on many life history traits, including growth, development and reproduction (Angilletta et al., 2006; de Jong and van der Have, 2009). This is especially true for ectotherms, such as fungi, whose internal temperature directly reflects the surrounding environment (Angilletta et al., 2002). Two general strategies for adapting to different thermal environments are phenotypic plasticity and genetic differentiation (Dybdahl and Kane, 2005; Knies et al., 2006; Yamori et al., 2010). Phenotypic plasticity is an individual-based response that allows the same genotype to produce different phenotypes in different environments (de Jong and van der Have, 2009). Genetic differentiation is a population-based response where individuals carrying different alleles are selected according to the local environment. Over generations, selection pressure for different temperature optima can lead to changes in the genetic architecture of temperature adaptation. For example, nonsynonymous substitutions can generate protein isoforms that differ for thermal stability and/or optimized function at different temperatures (Schoville et al., 2012). In a population, genetic diversity for thermal adaptation can be maintained by antagonistic pleiotropy (genetic trade-offs), in which an allele that has optimal function at one temperature exhibits maladaptation at a different temperature (Fields, 2001). A better understanding of the genetic basis of thermal adaptation is likely to provide insight into the ability of species to adapt to global climate change (Austin, 2007).

Thermal adaptation is important for many plant pathogenic fungi. Before 2000, the wheat stripe rust pathogen Puccinia striiformis f. sp. tritici caused epidemics only in cooler wheat-growing regions. After 2000, two new clonal lineages of the pathogen able to cause epidemics in warmer climates emerged, allowing the pathogen to expand into previously unfavorable environments (Milus et al., 2009; Mboup et al., 2012). In the globally distributed barley pathogen Rhynchosporium commune, there was evidence for local adaptation to grow at higher temperatures (22 °C), a key adaptation that may have enabled the global emergence of a pathogen that originated in the cold climate of Scandinavia (Zaffarano et al., 2006; Linde et al., 2009; Stefansson et al., 2013).

In some cases, pathogenic fungi grow saprophytically as filamentous hyphae outside their host, and then undergo a morphological transformation to yeast-like growth after infecting their host. These morphological switches enhance pathogenicity by facilitating dispersal within the host and enabling evasion of host immune responses (Nadal et al., 2008; Gauthier, 2015). In other cases, the switch is in the opposite direction, which is mainly the case for plant pathogens (e.g. Ustilago maydis), where yeast-like growth is observed outside the host and filamentous growth is observed within the host. For some fungal pathogens (e.g. Talaromyces marneffei; Boyce and Andrianopoulos, 2013), a change in temperature triggers a switch in growth morphology (Gauthier, 2015).

Fungi use several mechanisms to cope with changing temperatures, including the induction or upregulation of proteins related to thermal stress (e.g. heat-shock proteins (HSPs); Feder and Hofmann, 1999), changes in cellular composition (e.g. activation of the high-osmolarity glycerol (HOG) pathway resulting in glycerol accumulation in cells; Panadero et al., 2006) and adjustments in membrane fluidity (e.g. lipid saturation in cell membranes; Leach and Cowen, 2014). Little is known regarding the genetic basis of how fungi tolerate different temperatures. The majority of research on this topic has been conducted in yeasts and is oriented around the regulation of HSPs (Feder and Hofmann, 1999; Leach and Cowen, 2014). HSPs are regulated through the heat-shock transcription factor 1, which has been extensively studied in Saccharomyces cerevisiae (Sorger and Pelham, 1988; Nicholls et al., 2009). Quantitative trait locus (QTL) mapping of high-temperature growth in the progeny of a cross between a heat-tolerant clinical strain of S. cerevisiae and a heat-sensitive strain isolated from a rotting fig identified four linked genes involved in high-temperature growth, but none of these genes had a function known to affect thermal adaptation (Steinmetz et al., 2002; Sinha et al., 2008). A bulk segregant analysis applied to yeast populations (Ehrenreich et al., 2010) identified 21 QTLs associated with high-temperature growth in a cross of a heat-tolerant North American oak tree strain and a heat-sensitive West African palm wine strain (Parts et al., 2011). A transcriptome analysis of 48 Neurospora crassa isolates from subtropical and tropical environments identified two genes involved in thermal adaptation, including an MRH4-like RNA helicase (Ellison et al., 2011). Despite the progress illustrated by these studies, many gaps remain in our understanding of the genetic mechanisms underlying thermal adaptation in fungi (Robert et al., 2015) and other organisms.

Zymoseptoria tritici (syn Mycosphaerella graminicola) is the fungal pathogen causing Septoria tritici blotch, currently the most damaging wheat disease in Europe (Jorgensen et al., 2014; O’Driscoll et al., 2014). The 39.7 Mb reference genome is among the best-assembled fungal genomes available. It contains ~11 000 predicted genes distributed across 21 chromosomes sequenced from telomere to telomere, with an average of ~21% repetitive DNA (Goodwin et al., 2011). Z. tritici is a dimorphic fungus than can grow either as filamentous hyphae or yeast-like budding cells when grown in vitro (Nadal et al., 2008). The pathogen is found in wheat-growing areas worldwide covering a wide range of temperature regimens (Eyal et al., 1987; Zhan and McDonald, 2011; Jorgensen et al., 2014; O’Driscoll et al., 2014). An earlier study that included 138 Z. tritici strains sampled from diverse thermal environments on three continents found evidence for thermal adaptation and concluded that most of the thermal adaptation was due to genetic differentiation rather than phenotypic plasticity (Zhan and McDonald, 2011). Here, we sought to determine the genetic architecture of thermal adaptation and identify candidate genes based on QTL mapping of temperature sensitivity (TS) in wild-type strains sampled from the same thermal environment. This approach allowed us to identify and characterize genetic polymorphisms segregating within natural populations (Franks and Hoffmann, 2012; Robert et al., 2015). Knowledge about co-occurring genetic variants conferring different levels of thermal adaptation allows us to predict how a population will be able to adapt to climate change. To determine the genetic architecture of thermal adaptation in Z. tritici, we used restriction site-associated DNA sequencing (RADseq) for genotyping, coupled with high-throughput digital image analysis to measure growth rates under two temperatures (15 and 22 °C). We used QTL mapping to determine the genetic architecture of natural variation for TS and identify QTLs and their associated chromosomal segments. Comparisons of complete genome sequences of the parental isolates allowed us to identify polymorphisms in candidate genes for thermal adaptation.

Materials and methods

Mapping population, genotyping, genetic mapping and gene expression

Methods used for genetic mapping and gene expression were described earlier (Brunner et al., 2013; Lendenmann et al., 2014, 2015). Briefly, four Swiss wild-type strains were used to create two mapping populations, following an established protocol (Kema et al., 1996). The mapping populations were composed of 263 (ST99CH3D1 (3D1: SRS383146) × ST99CH3D7 (3D7: SRS383147)) and 261 (ST99CH1A5 (1A5: SRS383142) × ST99CH1E4 (1E4: SRS383143)) retained progeny. BioProject numbers PRJNA256988 and PRJNA256991 contain the NCBI Short Read Archive accession numbers for the retained progeny of crosses 3D1 × 3D7 and 1A5 × 1E4, respectively. RADseq (Baird et al., 2008) generated more than 7000 single-nucleotide polymorphism (SNP) markers for each cross. Resequencing of all four parents (Croll et al., 2013) allowed identification of the retained RADseq SNPs in the parents and candidate genes responsible for the observed QTLs. R/qtl version 1.27-10 (Arends et al., 2010) was used to construct two dense genetic maps (Lendenmann et al., 2014). Candidate genes within QTL confidence intervals were characterized for their transcriptional profile, using RNA-Seq data from an in planta virulence time-course assay (Brunner et al., 2013).

Phenotyping

Radial growth rates (mm per day) (Trinci, 1971; Lendenmann et al., 2015) of single spore colonies grown at two temperatures (15 and 22 °C) and observed over three time points (8, 11 and 14 days post inoculation (dpi)) were used to calculate TS. Plate inoculation procedures and digital image analyses in ImageJ (Schneider et al., 2012) were described earlier (Lendenmann et al., 2014). Isolates were grown in Petri plates on potato dextrose agar (4 g l−1 potato starch, 20 g l−1 dextrose, 15 g l−1 agar) and placed at a constant temperature of 15 or 22 °C with 70% humidity in the dark. TS for each isolate was calculated as the radial growth rate at 22 °C divided by the radial growth rate at 15 °C. Two absolute value traits (radial growth rate at 15 and 22 °C) and one relative value trait (relative growth rate coefficient=TS) were calculated for each isolate and used for QTL analysis. The 22 °C growth rate trait of this study was called growth rate (fungicide absent) in an earlier investigation mapping QTLs involved in fungicide sensitivity (Lendenmann et al., 2015).

In addition to growth rate, two growth morphologies were scored for each strain. Although Z. tritici is considered a filamentous fungus that grows mainly as hyphae, a yeast-like morphology is often observed in culture on rich media, suggesting that Z. tritici is dimorphic (Mehrabi and Kema, 2006; Mehrabi et al., 2006b; Nadal et al., 2008). To determine if the yeast/hyphae dimorphism can be mapped, we conducted a visual inspection of the digital images to classify the colonies of each isolate as either hyphal- or yeast-like. Three people (ML, JP and ES) independently classified each isolate using all five technical replicates at 11 dpi for both temperatures. The scoring day of 11 dpi was chosen because the colonies were too small to make an accurate visual scoring at 8 dpi, while classifications at 14 dpi gave the same result. Progeny that exhibited only yeast-like colonies were scored as ‘0’, whereas progeny exhibiting at least two hyphal colonies on a Petri dish were scored as ‘1’ (Supplementary Figure S1). Only progeny with identical classifications from all three scorers were included in this analysis. At 15 °C, >98% of the progeny with a consensus score in cross 1A5 × 1E4 grew with a yeast-like morphology (binary value ‘0’: 247 progeny/binary value ‘1’: 3 progeny), hence this treatment was not included in the analysis. The binary counts were as follows: cross 3D1 × 3D7 at 22 °C binary value ‘0’: 111 progeny/binary value ‘1’: 90 progeny; at 15 °C binary value ‘0’: 186 progeny/binary value ‘1’: 52 progeny; cross 1A5 × 1E4 at 22 °C binary value ‘0’: 104 progeny/binary value ‘1’: 33 progeny.

Absolute and relative phenotypes as well as the yeast/hyphae scores for the retained progeny of each cross are available from the Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.cj432.

QTL mapping

Simple interval mapping (SIM) was used for QTL analysis performed in R/qtl version 1.27-10 (Arends et al., 2010) using R (R_Core_Team, 2012). A binary model was applied for the yeast/hyphae dimorphism phenotype. A total of 1000 genome-wide permutations were used to calculate the significant logarithm of odds (LOD) threshold. Further analyses considered only QTLs that showed P-values <0.05. Bayesian credible intervals (Manichaikul et al., 2006) were used to calculate 95% confidence intervals for each QTL. Genetic positions (centiMorgans, cM) of markers were converted into physical positions (base pair, bp) by using the reference IPO323 genome (Goodwin et al., 2011). QTLs were assumed to be different if their confidence intervals did not overlap. More detailed information on methods used for QTL mapping were described earlier (Lendenmann et al., 2014).

Single spore colonies were also scored for their degree of melanization (Lendenmann et al., 2014). Because melanin has been associated with protection against extreme temperatures as well as fungicide resistance (Butler and Day, 1998; Nosanchuk and Casadevall, 2006; Taborda et al., 2008), we investigated pleiotropy among the growth rate traits, TS and melanization (11 dpi), and also looked for evidence of pleiotropic effects among temperature and fungicide sensitivity. This analysis was conducted by overlapping full SIM genome scans and using linear regression among the traits based on a general linear model in R with Pearson’s correlation coefficient. Regression analysis between the yeast/hyphae dimorphism and growth rate at each temperature was conducted using a one-way analysis of variance to calculate the phenotypic variance explained by the yeast/hyphae dimorphism phenotypes upon growth rate. A t-test was used to compare mean differences between the two morphologies in R. Marker allele effects were calculated using R/qtl to confirm the linear regressions.

Identification of candidate genes within QTL confidence intervals

Resequenced parents of each cross were aligned against the IPO323 reference genome, and sequence variants were called and annotated using the open-source tools SnpEff and SnpSift version 3.3 h (Cingolani et al., 2012). Synonymous SNPs were omitted from any further investigation. A gene was considered a candidate for explaining the QTL if it contained at least one sequence variant within the boundaries of the 95% confidence interval. More details regarding methods used to identify candidate genes within QTL confidence intervals were provided earlier (Lendenmann et al., 2014). Because heat-shock proteins have an important role in maintaining cellular functions under thermal stress (Feder and Hofmann, 1999), a hypergeometric test was used to determine whether there was a significant enrichment of proteins with an InterPro domain classified as HSPs within the QTL 95% confidence intervals (Tables 1 and 2). The QTL confidence intervals (Tables 1 and 2) were also investigated for the presence of HSPs as candidate genes.

Table 1 Positions and effects of the five temperature sensitivity-associated QTLs identified in cross 3D1 × 3D7
Full size table
Table 2 Positions and effects of the five temperature sensitivity-associated QTLs identified in cross 1A5 × 1E4
Full size table

Results

Genetic architecture of TS

The general linear model of growth rate based on radial development of colonies over time provided an average fit (r2) >98% across both crosses and treatments (Supplementary Figure S2), justifying the method used to measure growth rate for each isolate. The growth rate and TS phenotypes showed a continuous distribution consistent with a quantitative trait in both crosses. Transgressive segregation was found for all traits, with many progeny showing more extreme phenotypes than their parents (Figure 1). The high broad-sense heritability (H2>80%) found for average colony area indicates a low environmental variance among the five technical repeats (Supplementary Figure S2).

Figure 1
figure 1

The distributions of growth rate and TS in both crosses. Positions of the parental phenotypes within each distribution are indicated by white dashed lines with corresponding parental names above their phenotypes.

Full size image

For the TS phenotype, one significant QTL was found in cross 3D1 × 3D7 and three were found in cross 1A5 × 1E4. These four QTLs were on chromosomes 1, 2, 4 and 10 with LOD values ranging from 4.3 to 11.8 (Figures 2 and 3 and Tables 1 and 2). For the absolute value phenotypes, cross 3D1 × 3D7 had four growth rate QTLs on chromosomes 2, 3, 7 and 11 and cross 1A5 × 1E4 had two growth rate QTLs on chromosomes 5 and 8 (Tables 1 and 2). There was a confidence interval overlap between the two crosses for the QTLs on chromosome 2, suggesting a shared QTL. This QTL overlap was already described in a previous study (Lendenmann et al., 2015). On average, ~30% of total variance was explained by the significant QTLs for each trait in both crosses (Table 3). No significant QTLs were mapped on the accessory chromosomes 14–21. Detailed information for every mapped QTL can be found in Supplementary Tables S1 and S2.

Figure 2
figure 2

LOD plots from SIM analysis over all chromosomes for growth rate at 15 and 22 °C, as well as TS for cross 3D1 × 3D7. The dashed horizontal red line represents the significance threshold (P=0.05) obtained using 1000 genome-wide permutations. A full color version of this figure is available at the Heredity journal online.

Full size image
Figure 3
figure 3

LOD plots from SIM analysis over all chromosomes for growth rate at 15 and 22 °C, as well as TS for cross 1A5 × 1E4. The dashed horizontal red line represents the significance threshold (P=0.05) obtained using 1000 genome-wide permutations. A full color version of this figure is available at the Heredity journal online.

Full size image
Table 3 Genetic architecture of each trait for both crosses
Full size table

QTLs associated with the yeast/hyphae dimorphism

For the yeast/hyphae dimorphism phenotype, we considered QTLs to be different if their confidence intervals did not overlap. In cross 3D1 × 3D7, we mapped five significant QTLs distributed on chromosomes 1, 3, 7 and 11 (Supplementary Figure S3 and Supplementary Tables S3 and S4), with two QTLs on chromosome 3 (Supplementary Table S3). In cross 1A5 × 1E4, two significant QTLs were identified on chromosomes 7 and 11 (Supplementary Figure S4 and Supplementary Tables S5 and S6). Both of these QTLs overlapped with QTLs of cross 3D1 × 3D7, suggesting these QTLs were shared in the two crosses (Supplementary Tables S3 and S5).

Identification and characterization of candidate genes within QTL confidence intervals

All annotated genes within each QTL 95% confidence interval were investigated to identify candidate genes affecting each trait. We excluded from consideration all genes that lacked sequence variation or that possessed only synonymous SNPs. For cross 3D1 × 3D7, a summary of the candidate genes for each trait is shown in Supplementary Tables S4 and S7. Supplementary Tables S6 and S8 summarize the candidate genes for cross 1A5 × 1E4.

On average, we found ~200 candidate genes per confidence interval, with a minimum of six candidate genes (chromosome 10, TS, cross 3D1 × 3D7) and a maximum of 1382 candidate genes (chromosome 1, growth rate (15 °C), cross 1A5 × 1E4). On average, ~40% of the candidate genes had no known function (Supplementary Tables S4 and S6–S8). There was no evidence for enrichment of HSPs in the QTL confidence intervals relative to the rest of the genome. However, some HSPs were among the candidate genes found within the QTL confidence intervals (Tables 1 and 2).

A more detailed investigation was conducted on the three major QTLs (LOD ⩾4.5) containing ⩽30 candidate genes in narrow confidence intervals (<85 kb). The chromosome 10 QTL associated with TS in cross 3D1 × 3D7 had an LOD score of 11.8 and contained only six candidate genes (Supplementary Table S7). The chromosome 1 and 3 QTLs associated with the yeast/hyphae dimorphism at 15 °C in cross 3D1 × 3D7 contained 15 candidate genes each (Supplementary Table S4) with LOD scores of 4.5 and 10.2, respectively. The lists of candidate genes associated with each QTL are summarized in Table 4 and Supplementary Tables S9 and S10.

Table 4 Genes within the chromosome 10 temperature sensitivity QTL confidence interval containing ⩽30 candidate genes for cross 3D1 × 3D7a
Full size table

Evidence for pleiotropy

In a previous study (Lendenmann et al., 2015), we described a shared QTL on chromosome 11 for melanization at 22 °C, growth rate at 22 °C and fungicide sensitivity in cross 3D1 × 3D7. We hypothesized that a rare mutation in PKS1 (protein ID 96592) found in the 3D7 parent was responsible for this pleiotropy (Lendenmann et al., 2014). In this study, we found evidence for pleiotropy between melanization and growth at 15 °C following the same pattern described earlier (Lendenmann et al., 2015), namely the PKS1 allele that increases melanization also slows growth at 15 °C. However, there was no evidence for pleiotropy between melanization and TS associated with the chromosome 11 QTL in cross 3D1 × 3D7 (Figure 4). On the other hand, we found strong evidence for pleiotropy among melanization at 15 °C, growth at 15 °C and TS based on a shared chromosome 10 QTL in cross 3D1 × 3D7. This QTL displayed the following pattern: the 3D7 allele that increases melanization also accelerates growth at 15 °C and decreases TS. There was some evidence for pleiotropy among melanization and TS in cross 1A5 × 1E4 due to a shared chromosome 2 QTL and there were significant correlations between melanization and growth at 15 °C for QTLs on chromosomes 5 and 8 (Figure 4). In both crosses, we found no evidence for pleiotropy among TS and fungicide sensitivity.

Figure 4
figure 4

Evidence for pleiotropy among melanization, growth rate at 15 °C and TS for the chromosome 10 QTL in cross 3D1 × 3D7. Melanization is represented by gray values (11 dpi), with lower values indicating higher melanization (Lendenmann et al., 2014). The 3D7 allele that increases melanization also accelerates growth at 15 °C (a) and decreases TS (c). Panels b and d, which show corresponding full genome scan overlaps for cross 1A5 × 1E4, indicate no pleiotropy among the same three traits. Each panel shows a full genome scan LOD overlap plot based on SIM analysis, with traits shown in different colors (red=melanization; green=growth rate at 15 °C or TS). Horizontal lines in the LOD plots represent the color-coded significance thresholds (P=0.05) obtained with 1000 genome-wide permutations. Each panel includes the corresponding linear correlation among the two traits. The name above each QTL peak indicates which parental allele provided the higher phenotypic mean. Arrows point to the QTL peaks that could explain the observed positive or negative correlation among the two traits. A full color version of this figure is available at the Heredity journal online.

Full size image

The QTL peaks for the yeast/hyphae dimorphism overlapped with growth rate QTLs on chromosomes 3, 7, 10 and 11, providing evidence for pleiotropy in both crosses. There also was a strong correlation between growth rate and the yeast/hyphae dimorphism. In both crosses, the allele associated with faster growth was also associated with filamentous growth (Supplementary Tables S1, S3, S5 and S11 and Figures 5 and 6), with >17% of the total growth rate variance explained by the yeast/hyphae dimorphism and significant (P<0.001) mean differences between the growth rates of the two morphologies (Figures 5 and 6). On the other hand, QTLs on chromosomes 1, 2, 4 and 8 did not overlap between the yeast/hyphae dimorphism and growth rate, indicating that growth morphology is not the sole determinant of growth rate (Figures 5 and 6). For example, in cross 3D1 × 3D7 there is a significant QTL for the yeast/hyphae dimorphism at 15 °C on chromosome 1, but there is no growth rate QTL at this position, whereas there is a significant QTL for growth rate at 15 °C on chromosome 2, but no corresponding QTL for the yeast/hyphae dimorphism (Figure 5b).

Figure 5
figure 5

Evidence for pleiotropy among the yeast/hyphae dimorphism and growth rate phenotypes for the chromosome 3, 7, 10 and 11 QTLs in cross 3D1 × 3D7 (a=22 °C/b=15 °C). Each panel shows a full genome scan LOD overlap plot based on SIM analysis, with traits separated by colors (green=growth rate at 15 °C/22 °C, purple=yeast/hyphae dimorphism trait). Horizontal lines in the LOD plots represent the color-coded significance thresholds (P=0.05) obtained with 1000 genome-wide permutations. Each panel includes a corresponding box plot for the two traits. Mean growth rate values are presented either above or below the median line. The name above each QTL peak indicates which parental allele provided the higher phenotypic mean. Allele effects for nonsignificant, but visible pleiotropic peaks for either of the phenotypes were obtained by modeling the peaking marker of the significant peak for either of the phenotypes upon the corresponding opposite phenotype. For each pleiotropic peak, the allele providing faster growth also contributes to a more hyphal phenotype (a and b). A full color version of this figure is available at the Heredity journal online.

Full size image
Figure 6
figure 6

Evidence for pleiotropy among the yeast/hyphae dimorphism and growth rate phenotypes at 22 °C for the chromosome 7 and 11 QTL in cross 1A5 × 1E4. The figure shows a full genome scan LOD overlap plot based on SIM analysis, with traits separated by colors (green=growth rate at 15 °C/22 °C, purple=yeast/hyphae dimorphism trait). Horizontal lines in the LOD plots represent the color-coded significance thresholds (P=0.05) obtained with 1000 genome-wide permutations. In the box plot for the two traits, mean growth rate values are presented either above or below the median line. The name above each QTL peak indicates which parental allele provided the higher phenotypic mean. Allele effects for nonsignificant, but visible pleiotropic peaks for either of the phenotypes were obtained by modeling the peaking marker of the significant peak for either of the phenotypes upon the corresponding opposite phenotype. For each pleiotropic peak, the allele providing faster growth also contributes to a more hyphal phenotype. A full color version of this figure is available at the Heredity journal online.

Full size image

Discussion

Genetic architecture of TS

The TS phenotype we used to measure thermal adaptation reflects differences in growth rates at two temperatures. Strains with a TS=1 are ‘generalists’ that grow at the same rate at both temperatures. Strains with TS>1 grow faster at the higher temperature and strains with TS<1 grow faster at the lower temperature. Strains that grow quickly at both temperatures (i.e. high growth rates and TS=~1) exhibit the greatest degree of thermal tolerance. Although the TS phenotypes of the parental strains used in both crosses were similar, the progeny from these parents exhibited a variance in TS comparable to the variance measured in natural field populations from around the world (Zhan and McDonald, 2011), indicating that the parents carried a high degree of genetic diversity for this trait. This finding also illustrates the potential for recombination to generate novel combinations of alleles that can affect thermal adaptation, providing a mechanism for a rapid evolutionary response to change in temperature. The presence of several visible but nonsignificant peaks on the genome scans, coupled with the finding of a maximum of 48.2% of the total phenotypic variance explained by the significant QTLs leads us to predict that additional QTLs will be revealed in future studies that include a greater number of markers and a larger number of offspring.

All four TS QTLs had allele means >1, indicating faster growth at 22 °C than at 15 °C. An earlier analysis of Swiss populations of Z. tritici found many isolates with a TS<1, consistent with adaptation to growth at lower temperatures (Zhan and McDonald, 2011). These findings suggest that genes involved in adaptation to cooler temperatures were not variable in the four Swiss parents. We found no evidence for antagonistic pleiotropy (genetic trade-offs), but we postulate that testing our mapping populations under more extreme temperatures (>22 °C and <15 °C) could reveal such effects.

The finding of different TS QTLs in the two crosses likely reflects differences in the alleles found in the parents and may reflect different selective processes underlying thermal adaptation. For example, in cross 1A5 × 1E4 the significant TS QTLs on chromosomes 2 and 4 were due mainly to growth QTLs exhibited in the 22 °C environment. Pathogen strains that grow faster at warmer temperatures are likely to be favored during the asexual phase of reproduction that occurs during the summer (Ponomarenko et al., 2011). The finding of different temperature-specific QTLs within the same cross suggests that different protein isoforms encoded by different alleles of the same gene could be better adapted to one temperature and maladapted to another temperature or that the corresponding genes are differentially regulated at the two temperatures, a known mechanism for thermal tolerance (Schoville et al., 2012).

All identified QTLs except the chromosome 10 TS QTL in cross 3D1 × 3D7 had large confidence intervals containing more than 10 genes. These QTLs could contain more then one segregating explanatory variant. Future studies using targeted backcross mapping (Sinha et al., 2008) could be used to determine if multiple linked QTLs exist within any of these QTLs.

QTLs associated with the yeast/hyphae dimorphism

We identified seven QTLs associated with the yeast/hyphae dimorphism, with two of these QTLs shared between the crosses. On average, ~35% of the overall variation in growth morphology was explained by these QTLs. These findings indicate that the yeast/hyphae dimorphism is inherited as a quantitative trait in Z. tritici. The yeast/hyphae dimorphism QTLs correlate with the growth QTLs, indicating pleiotropic effects. QTL alleles associated with faster growth were also associated with filamentous growth in both crosses. We consider it likely that the filamentous morphology is responsible for the faster growth, as the yeast morphology results in a more localized growth of Z. tritici colonies (Mehrabi and Kema, 2006). However, it is also possible that morphology is a threshold trait that is triggered through faster growth. Our results also indicate that growth rate is not determined solely by the colony morphology, as we found three growth QTLs that were not associated with the dimorphism phenotype. In Z. tritici the switch between morphologies can be induced in vitro by changing from a nutrient-rich medium, which favors yeast-like growth, to a poor medium that favors hyphal growth (Mehrabi and Kema, 2006; Mehrabi et al., 2006b). We found that temperature significantly affected dimorphism in our mapping populations, with more progeny exhibiting hyphal growth at 22 °C compared with 15 °C. While diverse environmental signals can affect dimorphism, the main signaling pathways appear well conserved, with the cAMP-dependent protein kinase A and mitogen-activated protein kinase pathways (Nadal et al., 2008) having major roles. In Z. tritici both of these pathways were shown to affect the dimorphism phenotype (Cousin et al., 2006; Mehrabi and Kema, 2006; Mehrabi et al., 2006a, b; Gohari et al., 2014). It is not known if the yeast-like morphology plays any role in the natural history of Z. tritici. Filamentous growth is obvious during all phases of its life cycle, including biotrophic, necrotrophic and saprotrophic phases of growth and development, but a yeast-like morphology has never been reported in planta. Regardless of the purpose of the yeast phase in Z. tritici, our findings illustrate that Z. tritici offers a new model for elucidating the mechanisms responsible for the yeast/hyphal dimorphism in fungi.

Identification and characterization of candidate genes within QTL confidence intervals

Earlier gene disruption studies in the IPO323 reference strain identified 10 genes (MgTpk2, MgGpa1, MgGpa3, MgGpb1, MCC1, MVE1, MgSlt2, MgHog1, MgFus3 and ZtWor1) that affected the yeast/hyphae growth morphology (Cousin et al., 2006; Mehrabi and Kema, 2006; Mehrabi et al., 2006a, b, 2009; Choi and Goodwin, 2011a, b; Gohari et al., 2014). We did not find any of these genes in the 95% confidence intervals of the dimorphism QTLs identified in our study, indicating that our QTL mapping approach identified novel gene candidates associated with the yeast/hyphae dimorphism in Z. tritici.

Out of the 15 candidate genes found in the 3D7 × 3D1 chromosome 1 QTL for the yeast/hyphae dimorphism at 15 °C, gene 88644 encodes a Ca+2/calmodulin-dependent protein kinase, which is part of a kinase family involved in various signaling cascades. In the dimorphic ascomycete Sporothrix schenckii, calmodulin-dependent protein kinases were shown to be involved in the dimorphism (Valle-Aviles et al., 2007). Hence, we consider gene 88644 as the most likely candidate to explain this chromosome 1 QTL.

The 3D7 × 3D1 chromosome 3 QTL for the yeast/hyphae (15 °C) dimorphism contains 15 candidate genes. Among these, gene 56742 encodes a guanine nucleotide exchange factor, a family of proteins involved in various intracellular signaling pathways. In a previous study in Ustilago maydis, a guanine nucleotide exchange factor was shown to be involved in dimorphism (Muller et al., 2003), hence we consider this the most likely candidate gene explaining this QTL.

Evidence for pleiotropy

The mitogen-activated protein kinase kinase candidate gene (JGI ID: 76249 (MgPbs2)) in the chromosome 10 TS QTL of cross 3D1 × 3D7 is the ortholog of the S. cerevisiae gene PBS2. Pbs2 phosphorylates the Hog1 protein (a mitogen-activated protein kinase) in S. cerevisiae as well as in Aspergillus nidulans (Panadero et al., 2006; Duran et al., 2010). HOG1 is part of the HOG pathway that is induced upon osmotic stress. Recent studies indicate that other stress responses, including thermal stress, are also governed by the HOG pathway. For example, the HOG pathway provides a mechanism for thermal tolerance in S. cerevisiae, which accumulates glycerol under cold stress to protect against freezing (Panadero et al., 2006). In Aspergillus fumigatus, the HOG pathway is thought to affect the fluidity of plasma membranes, providing protection under thermal stress (Ji et al., 2012). Changes in cell membrane fluidity are thought to be a common mechanism for microorganisms to adapt to changing temperatures (Leach and Cowen, 2014). A mutant study (Mehrabi et al., 2006b) of the Z. tritici ortholog HOG1, MgHog1 (JGI ID: 76502), found higher melanization associated with greater hyphal growth as well as larger colony sizes, the same pattern of phenotypes exhibited by the chromosome 10 TS QTL. Many of the chromosome 10 QTL pleiotropic effects observed in this study were found at 15 °C but not at 22 °C. This suggests that PBS2, and the HOG pathway, may be induced at lower temperatures in Z. tritici, providing a plausible mechanism to explain phenotypic plasticity.

Pleiotropy also explains the shared QTL peak on chromosome 11 in cross 3D1 × 3D7 for melanization (11 dpi), growth rate and the yeast/hyphae dimorphism for both temperatures (Lendenmann et al., 2014). In this case, we hypothesize that PKS1, in particular the postulated quantitative trait nucleotide at amino-acid position 1783 (Lendenmann et al., 2014), is the source of the observed trait variance. This hypothesis is supported by analyses showing that the RADseq marker closest to the PKS1 gene explains >7% of the variance for each of these traits.

Conclusions

We elucidated the genetic architecture of thermal adaptation in a pathogenic fungus and identified multiple QTLs as well as candidate genes underlying this trait. To our knowledge, this is the first study to use QTL mapping to identify genes involved in thermal adaptation in a plant pathogen. Our findings suggest that thermal adaptation in Z. tritici involves not only changes in growth rate but also morphological adaptations, including degree of melanization and changes in growth morphology. This suggests that Z. tritici has evolved many strategies for thermal adaptation, in part explaining its global distribution across a wide temperature range and leading us to predict that it will adapt rapidly to global warming. Future investigations aiming to functionally validate the candidate genes responsible for the mapped QTLs are underway. Additional population genetic diversity studies are needed to analyze the variation for these genes found in global populations and may further elucidate the potential of Z. tritici to adapt to changes in temperature. These studies point the way towards understanding the potential response of fungal pathogens to global warming.

Data archiving

Sequence data of parental strains and progeny have been submitted to NCBI: NCBI BioSample accession numbers of parental strains: SRS383146 (ST99CH3D1), SRS383147 (ST99CH3D7), SRS383142 (ST99CH1A5) and SRS383143 (ST99CH1E4). NCBI BioProject accession numbers of progeny strains: PRJNA256988 (cross: ST99CH3D1 × ST99CH3D7) and PRJNA256991 (cross: ST99CH1A5 × ST99CH1E4).

Progeny phenotypes have been submitted to Dryad: DOI:10.5061/dryad.cj432.