Abstract
Pathogens play an important role in the evolution of plant populations, but genetic mechanisms underlying disease resistance may differ greatly between geographic areas as well as over time. Local adaptation is thought to be an important step in plant evolution, and may be impacted by differential pathogen pressures in concert with abiotic factors. This study uses locally adapted ecotypes of the native perennial switchgrass (Panicum virgatum) to examine the temporal and spatial variation in the genetic architecture of resistance to fungal pathogens, namely switchgrass leaf rust (Puccinia novopanici). To identify loci underlying variation in pathogen resistance in switchgrass, we scored rust damage across an outcrossed mapping population at eight locations across the central United States from southern Texas to Michigan. We followed rust progression at these sites for three years and mapped quantitative trait loci (QTLs) using function-valued transformations of rust progression curves. Overall, we mapped 51 QTLs that varied in presence and strength over the three-year period. Two large-effect QTLs were consistently associated with variation in rust progression in multiple sites and years, and are therefore potentially the result of the same underlying resistance genes. Interestingly, these two large-effect QTLs were almost exclusively detected in northern sites. This pattern could be caused by geographic difference in genetic architecture. The distribution of rust strains or variation in climatic conditions across the field sites could result in genotype-by-environment interactions in efficacy of rust resistance loci. Beyond reducing rust damage by 34%, the beneficial alleles at the two loci also increased biomass by 44%, suggesting a direct benefit by pleiotropy or indirect benefit through genetic linkage. Our results suggest an important role for fungal pathogens in the local adaptation of switchgrass and illustrate an influential geographic component of the genetic architecture of plant disease resistance.
Introduction
Understanding the factors that determine how well a particular population is adapted to its environment is a major goal of evolutionary biology. Plant populations often exhibit local adaptation, in which populations are more successful in their local environments than foreign genotypes in that environment (Leimu & Fisher 2008; Kawecki & Ebert 2004). Traditional studies of local adaptation have focused on the abiotic factors that contribute to differential population success (e.g. Clausen et al. 1940). Recent advances in molecular biology have allowed researchers to directly link abiotic stresses to genetic differences between populations, showing variation in the genetic basis of local adaptation (McKay et al. 2003; Lowry & Willis 2010; Fournier-Level et al. 2011; Price et al. 2018; He et al. 2018). However, biotic interactions have been less-studied in this context, but may be essential for explaining variation in local adaptation (Macel et al. 2007; Grøndahl & Ehlers 2008; Crémieux et al. 2008). Pathogens can impose strong disruptive selection on plant populations when they are constrained in the areas they infect, and therefore shape local adaptation (Giraud et al. 2017; Mursinoff & Tack 2017). Therefore, a full understanding of plant adaptation requires characterization of the genetic basis of local adaptation to pathogens in conjunction with the abiotic environment.
Molecular coevolution between microbe and host has been documented throughout the tree of life (Hooper & Gordon 2001; Gagneux et al. 2006; Alfano & Collmer 2004; Dodds et al. 2006) and is an important tool to understand large-scale evolution (Moran et al. 2008; Woolhouse et al. 2002). The most basic genetic mechanism underlying this dynamic is a gene-for-gene interaction, whereby infection success is determined by variation at a single locus in both the pathogen and host (Agrios 1997; Kniskern & Rausher 2006). In the host, resistance (R) genes usually code for proteins that recognize either the direct product of a single pathogen gene or some downstream protein in its signaling pathway (Jones & Dangl 2006). Pathogen genes that have a corresponding R-gene in their host are termed ‘avirulence’ genes because their presence means that plant resistance will be effective. Plant R-gene products trigger a cascade of changes in response to the avirulence gene that may result in the hypersensitive response and systemic acquired resistance (Kniskern & Rausher 2006). The hypersensitive response (HR) is a local upregulation of various protective mechanisms, including production of reactive oxygen species (ROS) and free radicals that may kill a pathogen, but typically at the cost of cell death and possible loss of fitness and growth (Tian et al. 2003; Jones & Dangl 2006). Systemic acquired resistance (SAR) is more akin to an innate immune response in vertebrates, whereby pathogen recognition molecules are increased in expression throughout the plant, improving resistance to later infections (Jones & Dangl 2006). SAR also shows a fitness cost, but has been demonstrated to maintain fitness under disease pressure (Traw et al. 2007).
The gene-for-gene coevolutionary model is appealingly simple and can sometimes explain the pattern of race-specific resistance in many plant species (e.g. van Leur et al. 1989). However, quantitative resistance may also evolve, whereby resistance is governed by allelic composition at several to many loci (Geiger & Heun 1989; Agrios 1997; Young 1996). These mechanisms are typically more complex, involving polygenic adaptations for morphological or phenological changes, production of antimicrobial compounds, and modification of effector targets (Niks et al. 2015). Studies have found several examples of both gene-for gene resistance (Kniskern & Rausher 2006; Bourras et al. 2016), and quantitative resistance (Quesada et al. 2010), but little consensus on the factors that determine which will evolve in a particular system. Rather, it appears that the expression of host resistance as well as parasite virulence are environmentally-dependent traits, not stable phenotypes as has been traditionally assumed (Penczykowski et al. 2016). Since resistance mechanisms are generally investigated under either stable laboratory conditions or at a single field site, it is challenging to breed for durable resistance, resistance that prevents disease in many locations and for more than a few years (Mundt 2014). Addressing the environmental variation in the genetic architecture of plant disease resistance requires an experiment replicated over both time and space.
The prairie grass switchgrass (Panicum virgatum) and its obligate fungal pathogen, switchgrass leaf rust (Puccinia novopanici), are an ideal system to study how loci contributing to pathogen resistance vary across space. Panicum virgatum L. is a long-lived, polyploid, C4, perennial grass native to North America east of the Rocky Mountains from northern Mexico to southern Canada (Gleason and Cronquist 1991). It is a common prairie and pasture grass and is grown as both a forage crop and as a bioenergy feedstock (Casler 2012; Parrish et al. 2012), and has become an important study system for ecological specialization. P. virgatum is split into two locally adapted ecotypes, upland and lowland (Morris et al. 2011; Lowry et al. 2014; Milano et al. 2016). The upland ecotype is more common in northern North America, and exhibits small stature (up to 190 cm) and low pathogen resistance (Casler 2012; Uppalapati et al. 2013; Milano et al. 2016; Lovell et al. 2016). In contrast, the more southerly lowland ecotype is large (up to 285 cm) and is more resistant to fungal pathogens (Casler 2012; Uppalapati et al. 2013; Milano et al. 2016; Lovell et al. 2016). While the lowland ecotype produces more biomass, it also has lower freezing tolerance (Lee et al. 2014; Peixoto & Sage 2016), possibly explaining the rarity of lowland ecotypes in more northern climates.
Since P. virgatum ecotypes differ in their susceptibility to rust infection, this host-pathogen system is useful for testing the role of local variation in the evolution of resistance. Switchgrass is infected with at least five species of rust (Puccinia spp.; Demers et al. 2017), but Puccinia novopanici is thought to be dominant in the central US (Gary Bergstrom, pers. obs.). P. novopanici is a basidiomycete fungi that infects only living leaf tissue of P. virgatum. As such, these fungi are thought to be extirpated from northern populations every winter. In closely-related and well-studied wheat rust (P. triticana), wind-borne spores blow from warm refugia in southern Texas across the Great Plains every summer in what is known as the “Puccinia pathway” (Eversmeyer & Kramer 2000). While this has not been directly examined for switchgrass rust, it is the dominant explanation for epidemiological patterns, and fits with our observations. Rust infection is virtually inevitable in switchgrass stands in North America by the late summer, though damage is less severe in varieties from the lowland ecotype.
Over the range of switchgrass in North America, plant-pathogen interactions occur in a wide range of abiotic conditions. The environmental dependence of resistance can have important consequences for disease prediction and breeding for durable resistance (McDonald & Linde 2002; Michelmore et al. 2013). In Nicotiana, for instance, elevated temperature inhibits plant defense responses, prompting a need for development of heat-resistant R-genes (Zhu et al. 2010). The reasons for altered pathogenicity in different climates can be due to failure of plant defenses as a result of multiple causes, including stress (Zhu et al. 2010), promotion of beneficial pathogen conditions (Doke 1983), and variation in pathogen ecology (Weller et al. 2002). While there have been few studies of the environmental niche of switchgrass rusts, closely related wheat rusts can offer some clues. In wheat stripe rust (P. striiformis), resistance genes are temperature-dependent (Fu et al. 2009). Further, it is well-established that variation in humidity can greatly impact the infection by foliar fungal pathogens (Magarey et al. 2005), and may therefore play a role in plant resistance. Thus, over the geographic range of the switchgrass-rust interactions, variation in the abiotic environment may play an important role in influencing both plant defenses and pathogen virulence.
The primary goal of our study was to characterize the genetic architecture of switchgrass resistance to rust pathogens to better understand the causes of pathogen-mediated local adaptation. To overcome past limitations due to environmental and temporal variation in host-pathogen relationships, we measured fungal resistance over three years at a continental scale. We planted clonally replicated quantitative trait locus (QTL) mapping populations at eight sites across more than 7000 km of latitude to map QTLs for rust resistance. We first tested whether rust resistance is controlled by large-effect loci corresponding to specific pathogens in a gene-for-gene model or by small-effect loci in a polygenic resistance model. The geographic patterns of QTL presence and strength allowed us to fully summarize the spatial distribution of pathogen resistance. Therefore we were able to test the degree to which resistance exhibits a genotype-by-environment interaction, or is homogeneously expressed. In addition, we assessed the quantitative impact of those QTLs on resistance and other morphological and fitness traits to determine the evolutionary impacts of pathogen resistance. We hypothesized that resistance alleles would have evolved in lowland ecotypes and would be correlated with morphological differences between ecotypes, such as biomass and tiller count.
Methods
Development of mapping populations
To identify loci controlling variation in rust progression, we used a previously developed a four-way phase-known (pseudo-testcross) population (for full cross details, see Milano et al. 2016). We clonally divided the outbred populations by manually splitting rhizomes at the Brackenridge Field Laboratory in Austin, TX. In May-July of 2015, the F0, F1, and F2 clones were potted, moved by truck, and transplanted at eight sites (Figure 1) in Kingsville, TX; Austin, TX; Temple, TX; Overton, TX; Columbia, MO; Manhattan, KS; Mead, NE; and Hickory Corners, MI. We assigned plants randomly to a honeycomb design, with 1.56 m between each plant. To reduce edge effects, we planted a border of lowland plants around the plot that were not measured experimentally. We watered plants by hand in 2015 to facilitate establishment. To develop a linkage map for QTL mapping we genotyped 431 second-generation genotypes by whole genome resequencing (for full details, see Lowry et al. 2019).
Phenotyping
At each site we scored the presence of leaf rust in 2016, 2017, and 2018. We used a method developed for rust on wheat (Triticum aestivum; McNeal et al. 1971; Roelfs et al. 1992), which translates well to switchgrass and has been used in previous studies (Uppalapati et al. 2013). At each site, we scored rust on a 0-10 scale based on the total proportion of the canopy covered in rust spores, a score which we have defined as ‘rust damage’ for this study (Figure 2).
In conjuction, we used a simple resistance definition as 1 – rust damage (Simms & Triplett 1994). Other fungal pathogens such as anthracnose and Bipolaris were present in plots, but were generally less common than rust and were not reflected in our ratings.
The effort extended to phenotyping rust damage varied among sites and years due to logistical challenges. However, for the most part, sampling began three weeks after green-up (the point at which ~50% of plants had emerged from the soil) and continued weekly until damage stopped increasing (Figure 3). Over three years, this resulted in over 149,000 rust ratings, which were used for the QTL analyses. In addition, we measured other morphological and physiological traits at all sites, including the number of tillers, weekly plant height, date of first flowering, and end-of-season aboveground biomass (see Milano et al. 2016 and Lowry et al. 2019 in review for details of this phenotyping effort).
Rust abundance and composition
Visual pathogen scoring regimes are well known to be subject to statistical artifacts (Lesaffre et al. 2012) and our data were no exception. Pathogen scores followed a tail-inflated (“U”-shaped) distribution. Therefore, we used nonparametric Wilcoxon signed-rank and Kruskal-Wallis tests using the functions wilcox.test and kruskal.test in the stats package of R to test the differences in damage between lineages, years, and sites (R core team 2018). To test for cytoplasmic effects on rust prevalence, we compared rust scores between F2 individuals with maternal cytoplasm from upland and lowland F1s, also with the aforementioned nonparametric tests.
Genomic architecture
We first mapped QTLs using traditional QTL mapping on pathogen ratings for each individual time point. Raw pathogen ratings were processed in R (v3.4; R core team 2018) using both packages qtl and funqtl (Broman et al. 2003; Kwak et al. 2016). To examine QTL effects over time, we scanned for QTLs using Haley-Knott regression for each site by year combination, with each time point as a separate trait using the functions scanone in qtl and geteffects in funqtl (Broman et al. 2003; Kwak et al. 2016).
We additionally examined QTLs controlling the overall progression of rust by modeling damage as a function-valued trait (Kwak et al. 2014; 2016). For each individual, we fit a curve to rust damage ratings using the R package funqtl (Kwak et al. 2016). This method has the advantage of decreasing bias introduced by differences among raters at different sites by using the parameters of an damage curve as traits, rather than the absolute rating values. In addition, it overcomes several statistical challenges of QTL mapping of time-valued traits by replacing trait data with a smoothed approximation, then applying functional principal component analysis (FPCA) as a dimension-reduction technique (Kwak et al. 2016). QTLs are then mapped for a small number of principal components (PCs; Kwak et al. 2016). These PCs represent the shape of the pathogen damage curve of rust, and therefore include both the timing and rate of infection spread. Previous studies have used area under disease progress curve (AUDPC) measurements to quantify resistance (Jeger & Viljanen-Rollinson 2001). AUDPC is robust and useful, but may show bias when infection timing differs among sites (Jeger & Viljanen-Rollinson 2001). FPCA generally shows the same results as AUDPC, but is less impacted by phenology differences. We used the first four PCs to map QTLs, and combined their effects by taking the mean LOD (logarithm of the odds) score, the SLOD score, at each genomic position (Kwak et al. 2016). Then we conducted 1000 permutations to calculate a penalty for the SLOD score that reduces the rate of inclusion of extra loci to 5% (Broman & Sen 2009). We examined geographic and temporal variation by mapping QTLs separately for each site and year, but we also generated a combined test that summarizes variation in this experiment. For this test, we summed SLOD scores across multiple sites and years, and concatenated permutations to generate a critical SLOD cutoff. We produced all plots using funqtl and ggplot2 (Kwak et al. 2016; Wickham & Wickham 2007).
Finally, we estimated the allele-specific effects of the significant QTLs we discovered in our study. In the four-way cross design, second-generation offspring can express four possible allele combinations, L1_L2, U2_L1, U1_L2, or U1_U2. These represent alleles from each of the four parents (lowland AP13: L1; lowland WBC: L2; upland DAC: U1; and upland VS16: U2). For each locus, we compared the pathogen scores for the individuals with the “resistant” QTL alleles to those with the “susceptible” alleles. We additionally made this comparison for morphological traits including biomass, flowering time, tiller count, and green-up date. We tested for difference in means using nonparametric Wilcoxon signed-rank tests for pathogen ratings, and a two-sample t-test for all morphological traits.
Results
Rust abundance and composition
Though infection timing varied, rust was present at all sites throughout the study period (Figure 3). The largest genetic source of variation was between upland and lowland F0 (grandparental) plants, with upland plants experiencing 39.15% more rust damage (W = 1.15e7, p < 0.0001). Rust damage differed between generations in the cross (χ2 = 656.98, p < 0.0001), with F0 plants showing the least amount of rust, and F1 plants the greatest (Figure S1). Rust damage was negatively correlated with green-up date, biomass, flowering time, height, and tiller count (Figure S2). Field sites differed significantly in mean rust damage across years (χ 2 = 1.99e5, p < 0.0001), though this was confounded by differing sampling periods across sites. Mean rust damage declined in almost every site each year (Figure S3), decreasing by 19.75% in 2017 and 30.74% in 2018. This change correlates with biomass increases of 85.64% in 2017 and 46.89% in 2018. Our cross design allowed us to test the phenotypic effect of maternal cytoplasm, the difference between second-generation plants with an upland dam and those with a lowland dam. We compared rust scores between second-generation individuals with maternal cytoplasm from upland and lowland F1s. There was a cytoplasmic effect (W = 1.72e9, P = 0.0059), but rust scores were only 1.33% higher in plants with lowland cytoplasm (Figure S1).
Genomic architecture
Since we collected data on pathogens over several weeks at all sites, we were able to examine how the genomic architecture of resistance changed over a single season at each site by scanning for QTLs as if each time point were a distinct phenotype. We found extensive variation between sites, but some patterns were shared across several sites. We identified QTLs for resistance at multiple sites on chromosomes 9N and 3N, though they differed in effect direction (Milano et al. 2016). That is, the allele from a lowland grandparent decreased rust damage on chromosome 9N, but increased damage on chromosome 3N. These QTLs had significant effects for ~40-50 days, with the 9N QTL becoming detectable about one week after the 3N QTL (Figure 4 and Supp.). Overall patterns were similar across years, although the collection of fewer time points in 2017 resulted in lower-resolution data.
We mapped 51 total QTLs using function-valued traits (Figure 5). Overall, we found the highest number of QTLs at the most northen site (KBSM), though there was not a clear geographic pattern in QTL number. Additionally, there was variation between years, with the greatest number of QTLs in 2016 (24 QTLs), and fewer in 2017 and 2018 (15 and 12, respectively). We found the same large-effect QTLs using function-valued traits to map overall QTLs (Figure 5). Across several sites and all years, QTLs on chromosomes 3N and 9N had the highest LOD scores (Figure 5), indicating close associations between loci and pathogen resistance, and each explained ~14% of the variation in rust damage.
Given their strength and consistency across sites and years, we considered these the most important QTLs. These large-effect QTLs differed greatly between northern sites and southern sites. LOD scores were significant at the large-effect sites in northern sites 17 times over the three years, but only four times in southern sites.
Allelic effects
We calculated allele-specific effects for each QTL in our dataset to quantify the direction and strength of each QTL. To understand the alleles underlying each QTL, we categorized them into either “Ecotype-specific” or “Genotype-specific” based on the resistance of each grandparental allele (Figure 6bc; Milano et al. 2016). For instance, if both upland alleles showed higher rust damage, and both lowland alleles low rust damage, this QTL would be considered ecotype-specific. If only one of the grandparental genotypes was resistant or susceptible, we scored that as genotype-specific. We added an additional category, “Cross-specific,” for effects that were shown in particular upland/lowland combinations (Figure 6a). A majority of loci showed genotype-specific effects (31 out of 51). Relatively fewer showed ecotype-specific effects (16 out of 51) and cross-specific effects were rarest (4 out of 51). In addition, we counted the number of resistance QTLs for each grandparental allele (Figure 6d). Many loci were in the same direction as the parental divergence, with both lowland alleles causing a reduction in rust prevalence (23 out of 51), but the same number showed the lowest rust with an upland/lowland allele combination. A few (5 out of 51) showed opposite pattern, with the least rust with both upland alleles (Figure 6e).
We then calculated the combined effects of the two large-effect QTLs we identified by examining only individuals containing either resistant or susceptible combinations of alleles at these loci. For instance, the allele from the upland VS16 grandparent was most resistant at the 9N locus, but the allele from the lowland WBC3 grandparent was most resistant at the 3N, so the individuals with both of these alleles had the “resistant” combination of alleles. We also chose the individuals with alleles that increased rust, that had the “susceptible” combination of alleles. By comparing these individuals, we estimated the phenotypic impacts of our large-effect QTLs. The effects of QTLs were most apparent early in the season, when the “resistant” combination resulted in 34.08% lower rust scores (W = 56156, P = 0.0003). Later in the summer, the difference between resistant and susceptible decreased to 6.22%, but variation also decreased (W = 41666, P < 0.0001). We repeated this comparison for several other morphological variables. The resistance alleles increased biomass by 44.3% (t = 5.44, P < 0.0001, Figure 5), tiller count by 7.59% (t = 2.58, P = 0.010), and height by 5.52% (t = 6.12, P < 0.0001), and contributed to a 2.07% later flowering time (t = 2.04, P = 0.041; Table 1).
Discussion
Our results show that there is a marked difference between the genetic architecture of rust resistance between populations planted in northern and southern regions. In the north, resistance appears qualitative, and driven by two large-effect loci that explain ~14% of variation in rust damage, and clearly influence plant morphological traits. These QTLs were largely stable across all three years of the study, indicating that they confer durable resistance. In the south, resistance is more quantitative, driven by many small-effect loci that vary year-to-year.
We found support for a gene-for-gene model in northern populations, but southern populations were representative of polygenic resistance. This geographic difference leads us to conclude that there is a strong genotype-by-environment interaction for rust resistance in switchgrass. Our hypothesis that the majority of resistance QTLs evolved in the lowland ecotypes was supported, but at many loci, an upland-lowland allele combination produced the greatest rust resistance. We also found support for the hypothesis that resistance alleles are correlated with morphological differences, indicating that the two large-effect loci we mapped in the north may have contributed to ecotypic differentiation between upland and lowland P. virgatum. Overall, our results suggest an important role for two large-effect loci in northern populations, but primarily minor effect loci underlying variation in resistance at southern field sites.
Genomic architecture
Locally adapted northern upland and southern lowland ecotypes of P. virgatum are divergent for many traits (Casler et al. 2004, 2007; Lowry et al. 2014; Milano et al. 2016; Lowry et al. 2018 in review), including cold tolerance, biomass, and leaf architecture (Casler 2012 p.30). Ecotypic differences in fungal pathogen resistance have been well documented for P. virgatum and will play a key role in the success of switchgrass as a forage and biofuel crop (Cornelius & Johnston 1941; Uppalapati et al. 2012; Sykes et al. 2016). Both of the large-effect resistance QTLs colocalize with previously-identified large-effect QTLs for biomass, height, and tiller number using the same mapping populations (Lowry et al. 2019 in review).
The most striking pattern in QTLs over time and space was between northern and southern sites. In the north, the two large-effect QTLs are consistently present over time and across four sites. This pattern would be expected for qualitative gene-for-gene resistance, in which the two QTL peaks are caused by resistance genes on chromosomes 3N and 9N. However, in the south, there are many more small-effect QTLs. This pattern is more indicative of quantitative resistance (Corwin & Kliebenstein 2017). It is possible that the large-effect QTLs in the northern sites obscured several small-effect QTLs that would have otherwise been above the critical LOD score, a phenomenon known as the Beavis effect (Beavis 1998; Xu 2003). The Beavis effect obscures small-effect QTLs and causes effect size overestimation of detected QTLs (Xu 2003). However, this bias is greatly decreased in sample sizes near to 500, so we expect that the impact on our study was minimal.
Typically, resistance to a particular pathogen is either quantitative or qualitative, but this pattern is not generally thought of as being geographically dependent. Studies in wheat have found that rust resistance can be geographically constrained because resistance is typically strain-specific (Kolmer 2005). Since rust populations in the central US are dominated by a single species, P. novopanici (Gary Bergstrom pers. obs.), resistance differences are inconsistent with species differences. Rather, we expect that north-south resistance differences may either be driven by differences in rust strain diversity or a GxE interaction on the resistance loci. Broadly, we expect higher strain diversity in the south, since this pattern has been documented for wheat rust in Asia (Ali et al. 2014) and anther smut in Silene (Bueker et al. 2016). Southern populations may have responded to higher stain diversity with more resistance loci. Alternatively, resistance may be environment-dependent, as a study found for temperature-dependent resistance to wheat stripe rust (Puccinia striiformis f. sp. tritici; Fu et al. 2009). In wheat stripe rust and other systems, immunity-related proteins often exhibit temperature-dependent activity (Franklin & Wigge 2014), suggesting a potential mechanism for GxE in disease resistance. Further work in this system should focus on quantifying the population genetics of P. novopanici to better understand whether resistance is strain-specific.
Previous efforts to map QTLs for rust resistance in switchgrass yielded limited results. Milano et al. (2016) mapped one QTL on chromosome 8K for a rust prevalence in a population planted in Austin, Texas. We found small-effect QTLs on this chromosome at three field sites, but these loci were not consistently detected across years. The differences in QTLs may be due to temporal differences, but are more likely traceable to inconsistency in phenotyping by different field technicians. The previous study scored only five tillers per plant and used principal-component transformation of a 1-4 rating scale (Milano et al. 2016). Our study improved upon this method as we evaluated rust damage on a whole-plant basis and mapped QTLs to function-valued transformations of rust progression, allowing for less bias due to tiller selection and evaluation of the full rust progression curve.
Allelic effects
For the two large-effect QTLs, one showed higher resistance with the upland allele, and the other with the lowland allele (Figure 5). This is surprising given that the lowland ecotypes typically display the highest rust resistance (Uppalaptati et al. 2013), so one might expect that all resistance alleles would come from lowland plants. The presence of upland resistance alleles may provide further evidence that pathogens differ between the north and south, since upland plants are much more common in the north.
Previous studies of switchgrass found many fixed differences between ecotypes (Milano et al. 2016), and considerable variability in effects of loci across geographic locations (Lowry et al. 2019, in press). Our pathogen resistance results were similar, although a surprising number of loci had genotype-specific effects. This may indicate divergent selection within ecotypes, and that mechanisms for pathogen resistance differ between populations of the same ecotype, as would be expected if pathogen strain differences are substantial. A similar pattern was found in melon (Cucumis melo), in which resistance QTLs vary between cultivars due to differences in Fusarium strains (Perchepied et al. 2005). Since the lowland grandparental genotypes were much more rust-resistant than the upland, it was surprising that the most rust-resistant plants expressed resulted from a combination of an upland and a lowland alleles across QTLs.
The two large-effect QTLs decreased rust substantially throughout the season, though the effect decreased over time. Leaf rust may show differing mechanisms of seedling and adult plant resistance, so this pattern is not unexpected (German & Kolmer 1992). Similarly, resistance alleles were associated with higher biomass and other overall morphological traits. This result runs contrary to the expectation of a growth-defense tradeoff, whereby resistant genotypes should be slower-growing due to limited resources (Herms & Mattson 1992). However, there may be many reasons why a negative correlation between growth and defense does not occur (Kliebenstein 2016; Hahn & Maron 2016). In switchgrass, we find that resistance QTLs colocalize with biomass and tiller count QTLs (Lowry et al. 2019 in review), suggesting either close linkage or a pleiotropic effect. Importantly for the emerging perennial bioenergy industry, the correlation between resistance and biomass indicates that breeding switchgrass will be able to combine positive traits without major trade-offs in biomass across most environmental conditions. However, both biomass and resistance may instead trade off with freezing tolerance, which is an important survival trait for switchgrass in high latitudes (Lowry et al. 2019 in review). The major barrier to high-yield lowland traits in northern climates is winter temperatures, so identifying the molecular basis of upland freezing tolerance in switchgrass is an important goal. If freezing tolerance is not linked to rust resistance, there is great potential for improvement of switchgrass for biofuel, especially in northern marginal areas.
Conclusion
Our results show the temporal and geographic variation in the genetic architecture of rust resistance in locally adapted switchgrass, including two large-effect loci that explain both pathogen defense and morphological differences between ecotypes, but show a limited effectiveness in the south. This pattern raises important questions about the drivers of genetic architecture of pathogen resistance and underscores the importance of assaying pathogen resistance across both time and space to capture the inherent variability in the interplay of biotic and abiotic drivers of genetic change. These loci may allow for more effective breeding strategies for rust resistance, if there are not trade-offs with other traits, such as cold tolerance. The role of rust in differentiating these ecotypes illustrates the synergistic role pathogens play in the evolution of different ecotypes and ultimately contributing to genetic variation within species.