1 | INTRODUCTION
The adaptation of a population to a new environment can involve traits controlled by only a few genes that have a major effect, but such oligogenic adaption is relatively rare (Bell, 2009; van’t Hof et al., 2011; Bastide et al., 2016). Indeed, many adaptive traits are genetically complex and involve large numbers of loci, each of which contributes little to the phenotype (Pritchard et al., 2010; Sella & Barton, 2019). With the large amount of genomic data now available, many authors have been able to identify the genetic basis of complex adaptive traits in different organisms (Daborn, 2002; Cook et al., 2012; Linnen et al., 2013) but identifying the genetic basis of a polygenic trait is not sufficient to understand adaptive potential of a species. In addition, the effect size of the genes (i.e. their contribution to the genetic variance of a trait, (Park et al., 2010)), interactions between genes (i.e. additivity, dominance, epistasis and pleiotropy, (Hansen, 2006)) and redundancy (i.e. when several genotypes share the same phenotype by accumulating different combinations of mutations (Barghi et al., 2020)) need to be evaluated.
Identifying the genetic architecture of adaptive traits has been the main focus of two fields of evolutionary biology (Höllinger et al., 2019; Barghi et al., 2020). The first approach is based on molecular population genetics and assumes that adaptive traits result in the directional selection of a limited number of beneficial mutations that have major effects on the traits concerned. A hitchhiking effect on other linked loci leads to loss of diversity in the surrounding genomic regions; this footprint is called a “selective sweep” (Maynard-Smith & Haigh, 1974; Messer & Petrov, 2013). Genome scan methods have been developed to detect this footprint across the genome by measuring differentiation between populations, by detecting variations in the site frequency spectrum (SFS) and/or identifying haplotypes under strong linkage disequilibrium (reviewed by (Vitti et al., 2013, Vatsiou et al., 2016 and Pavlidis & Alachiotis, 2017). The second approach is based on quantitative genetics and focuses on the phenotype to identify the genes responsible for phenotypic variation (Bazakos et al., 2017). Evolution of a polygenic trait is supposed to be the result of a collective effect of a large number of loci with infinitesimally small variations, leading to more subtle footprints called “shifts” (Barton et al., 2017; Boyle et al., 2017). Analyses of quantitative trait loci (QTL) or genome wide association studies (GWAS) are used to decipher the genetic architecture of a phenotypic trait by identifying correlations between loci and the phenotype (Barton & Keightley, 2002; Visscher et al., 2017). Molecular population genetics and quantitative genetics views are not incompatible. Pritchard and Di Rienzo in 2010 proposed a unifying view of polygenic adaptation as the result of sweeps and shifts acting simultaneously. Thus, combining the two approaches could be a good way to decipher the genetic architecture underlying polygenic adaptation (Gagnaire & Gaggiotti, 2016).
Genetic architecture of traits can be viewed as the genetic potential for phenotype variation through mutation. However, this concept is not sufficient to fully understand adaptation in natural populations, and Barghi et al. 2020 recently proposed the notion of adaptive architecture to better describe the adaptive potential of species. This notion extends the genetic architecture concept by including other factors involved in population adaptation such as the frequency of contributing alleles, pleiotropy fitness constraints, and genetic forces other than mutation, including selection, drift, and recombination. All these factors play a role in shaping the relative contribution of genes to the adaptation of a population and also in the degree of parallelism when different populations are compared that evolve in the same environment, and could consequently be considered as replicates. Experimental evolution is one possible approach to investigate the genomic responses related to adaptation and to measure the degree of parallelism between populations faced with a controlled environmental constraint and has been successfully applied in Drosophila(Graves et al., 2017; Griffin et al., 2017) and Escherichia coli(Tenaillon et al., 2012). Alternatively, in biological situations (like epidemics) that are difficult to reproduce in the laboratory, adaptive architecture can be investigated in natural systems comprising multiple populations that evolve independently in similar environments (Barghi et al., 2020).
The adaptive architecture concept proposed by Barghi et al. 2020 provides a unified framework to understand how pathogens adapt to plant genetic resistance which is more and more used in agriculture to control diseases as an alternative to applying chemicals. Two categories of resistance have been described in the literature: qualitative and quantitative resistance. Qualitative resistance is often based on ‘effector-triggered immunity’ (ETI), in which major genes confer almost complete protection after recognition of effectors produced by certain pathogen genotypes referred to as avirulent genotypes (Cowger & Brown, 2019). Qualitative resistance is usually not durable because the high specificity of the host-pathogen interactions exerts strong selective pressure on pathogen populations and can lead to rapid selection and fixation of a beneficial mutation (Parlevliet, 2002; Zhong et al., 2017), a process corresponding to the selective sweep concept described above. The genetic basis of quantitative resistance may rely on only a small number of QTLs but can be also polygenic, i.e. involve a large number of QTLs (Cowger & Brown, 2019). Diverse mechanisms can be involved and quantitative resistance is generally considered as the most durable (Pilet-Nayel et al., 2017). However, following changes in quantitative traits of pathogenicity (also referred to as aggressiveness (Lannou, 2012), many examples of erosion of quantitative resistance have recently been reported (reviewed in Pilet-Nayel et al. 2017, Cowger & Brown, 2019). In contrast to quantitative resistance of plants, only a few studies have provided information on the genetic basis of quantitative pathogenicity in pathogens. A complex genetic architecture of fungal quantitative pathogenicity was found in a comprehensive QTL mapping analysis of the wheat pathogen Zymoseptoria tritici supported by genome wide association studies (GWAS) of a global sample of isolates (Hartmann et al., 2017; Dutta et al., 2021). However, description of the adaptive architecture on one particular host requires comparison of several fungal populations which can have notable differences on standing genetic variation and population size (McDonald & Linde, 2002).
The ascomycete fungus Pseudocercospora fijiensis , which is responsible for black streak disease (BLSD) of banana, is an interesting pathogen model to describe adaptive architecture to quantitative plant resistance. BLSD is the most damaging foliar pathogens of banana worldwide (Guzmán et al., 2019). The BLSD pandemic started around 1960 in South-East Asia/Oceania. In 1972, the disease was detected for the first time in Latin America, in Honduras, and spread rapidly throughout the region (Carlier et al., 2021a). The Fundación Hondureña de Investigación Agrícola (FHIA) produced several quantitatively resistant hybrids that were used in Cuba in the 1990s and 2000s and have been used in the Dominican Republic since 2005. However, after five to 10 years of cultivation, in both countries, erosion of resistance was reported in FHIA 18 and FHIA 21 cultivars in the field (Pérez Miranda et al., 2006; Guzmán et al., 2019). Local adaptation of P. fijiensispopulations explaining the erosion of resistance of FHIA hybrids in the two countries was demonstrated in cross-inoculation experiments (Dumartinet et al., 2019). An even more recent study based on pool sequencing (Pool-Seq) supported the existence of convergent adaptation in both resistant and susceptible cultivars in less than 10 genomic regions, suggesting oligogenic architecture underlies this adaptation (Carlier et al., 2021b). However, other genomic regions that did not converge were detected across the populations analyzed and neither redundancy nor phenotype-genotype relationship was tackled in that study.
The aim of the present work was thus to characterize the adaptive architecture underlying the quantitative resistance adaptation ofP. fijiensis . To this end, we analyzed a large number of P. fijiensis samples from susceptible and resistant cultivars in Cuba using a paired population sampling design. We first used a genome scan based on pool-sequencing data to detect host selection footprints in key genomic regions. Isolates from one location characterized for one trait of pathogenicity (the diseased leaf area) were individually sequenced to perform GWAS and to investigate correlations between the phenotype and the genotype in candidate genomic regions. We then combined all these data to compare adaptive architecture between populations.