4 | DISCUSSION
The aim of this study was to investigate the adaptive architecture
involved in the response of the fungal pathogen Pseudocercospora
fijiensis to selection exerted by banana quantitative resistance. We
combined genome scan and quantitative genetic approaches to compare
paired population samples of P. fijiensis collected on banana
cultivars with different levels of quantitative resistance in Cuba.
The results provide first insights
into the adaptive architecture behind the response to quantitative
resistance in a fungal plant pathogen which was revealed to be complex.
Thirty-two putatively selected genomic regions were detected using the
genome scan approach, suggesting a polygenic basis of host adaptation inP. fijiensis (Table 2, Figure 2). Genome sequencing of pools of
isolates (Pool-Seq), a cost-effective method that enables estimation of
allele frequency using large samples
(Schlötterer et al.,
2014), was used in this first approach. The main problem using genome
scan methods is the detection of false positives
(Vatsiou et al.,
2016). Considering signals that are convergent between several
replicates is one way to limit the number of false positives, which is
what we did in this study
(Lotterhos &
Whitlock, 2015; Hoban et al., 2016). As also suggested in Dalongeville
et al. 2018 to limit the number of false positives, we combined
different genome scan methods and used the local score method that
considers linkage disequilibrium between SNPs showing a significant
signal to delimit genomic regions
(Bonhomme et al.,
2019; Fariello et al., 2017). It is worth noting that the sizes of the
genomic regions delimited in the present study using the local score
method were in line with the results of linkage disequilibrium analysis.
The average size of the candidate regions was around 14 kb, a distance
for which there still was some linkage disequilibrium between two
adjacent SNP markers (Figure S2). We used the same paired design as in
Carlier et al. 2021b,
but we more than doubled the number of population pairs analyzed (Table
1, Figure 1). The 8/14 new populations were sampled two years later in
the same locations and in the same cultivars. Based on this larger
number of samples, a convergent selection footprint was detected in at
least two locations in 90% (27/30) of the candidate genomic regions. As
suggested in Carlier et al.
2021b, increasing the
number of locations analyzed allowed us to detect more candidate
regions. The genetic basis of host adaptation in P. fijiensisconsequently appears to be more polygenic than we previously thought.
The genome scan based on Pool-Seq was an efficient and low-cost way to
obtain insights into the genetic basis of adaptation of P.
fijiensis to quantitative resistance. However, to further characterize
the genetic architecture of this adaptation, other approaches were
required to identify adaptive variation and estimate the contribution of
the candidate genomic regions in adaptive traits.
The genetic architecture of adaptation of P. fijiensis to
quantitative resistance was further investigated by searching for
correlations between a trait (the diseased leaf area, DLA) involved in
quantitative pathogenicity and haplotypes identified from individual
sequencing of isolates originating from one location (Table 3). This
trait, previously used to detect local adaptation in the populations
sampled in 2011
(Dumartinet et al.,
2019), appeared to be a good proxy of parasite fitness (see Material and
Method). A correlation was found between 61% (17/28) of the candidate
regions and DLA measured in both cultivars. These results first
confirmed that more than half the candidate regions detected by genome
scan were associated with a phenotypic variation related to quantitative
pathogenicity in at least one cultivar. In 14 regions out of 28, we were
able to detect a haplotype that would confer a fitness advantage only on
the resistant cultivar (9 regions) or the susceptible one (5 regions).
These results support the hypothesis that adaptation to quantitative
resistance can involve specific host-pathogen (or genotype x genotype)
interactions that may result in a local adaptation pattern already
described in the same cultivars
(Dumartinet et al.,
2019). However, the advantageous haplotypes identified in one cultivar
did not result in a disadvantage in the other cultivar, thus supporting
the absence of a fitness-cost, as previously observed in Dumartinet et
al. 2019. Host
specificity in genes involved in quantitative pathogenicity was also
suggested in the fungus Z. tritici (which has similar biology toP. fijiensis ;
Ohm et al., 2012) by comparing two wheat cultivars
(Hartmann et al.,
2017). Three regions (S2R1-Cu, S4R4-Cu and S9R3-Cu) were found to be
significantly correlated with the trait measured in the two cultivars.
In the two first regions, the same haplotype conferred an advantage on
both cultivars, meaning that the same genes in these regions could be
selected in both cultivars or alternatively, different linked genes,
since the regions contained several genes. In the third region, two
different haplotypes were correlated with increased pathogenicity in the
two cultivars, again suggesting the existence of specific host-pathogen
interactions. No correlation was detected between the DLA and 39%
(11/28) of the candidate genomic regions. This could be due to
insufficient statistical power or, alternatively, the genes in these
regions may play a role in other traits related to the pathogen’s life
cycle not measured in this study, such as the latent period, spore
production rate, or the latent infectious period
(Guzmán et al., 2019).
The level of disease in the field depends on the value taken by all
these quantitative traits
(Lannou, 2012).
Currently, no method is available to measure all these traits inP. fijiensis in laboratory conditions but methods that test
associations at the population level like BayPass may make it possible
to study some of them directly in the field.
In the case of polygenic adaptation, variants can have different effect
sizes on a given phenotypic trait and measuring these effects provides
insight into the genes that contribute the most to adaptation
(Park et al., 2010;
Shabana et al., 2018). The effect sizes can be estimated using GWAS
(Korte & Farlow,
2013). However, in the present study, no association between loci and
phenotypic traits was detected using this analysis. GWAS is not always
appropriate to study the genetic basis of highly variable traits and/or
traits involving a large number of loci with minor effects, because
associations can only be detected with using a large number of
individuals (Barton &
Keightley, 2002; Korte & Farlow, 2013; Visscher et al., 2017).
Concerning quantitative pathogenicity in plant pathogenic fungi, SNPs
associated with a trait related to reproduction in two cultivars were
detected using the GWAS approach in Z. tritici (Hartman et al.
2017). However, in a recent GWAS analysis with the same pathogen, SNPs
associated with a trait related to reproduction and to leaf disease area
(like the trait used in the present study) were detected in only 3/12
and 2/12 inoculated wheat cultivars, respectively, although the authors
used a not too conservative statistical threshold (false discovery rate,
FDR=10%, (Dutta et
al., 2021)). The size effect of the candidate regions detected in the
present study was rather tackled by estimating their contribution to the
two traits using redundancy analysis. In the present study, the
redundancy analysis (RDA, Figure 4) indicated unequal contribution of
the regions associated with both study cultivars, thus suggesting that
these regions may contain variation in genes with different effect
sizes. Moreover, different multilocus genotypes across the candidate
regions leading to an increase in quantitative pathogenicity were found
among the study populations, suggesting genetic redundancy among the
loci involved in adaptation of a fungus like P. fijiensis to its
host (Table 4). Genetic redundancy has been proven to create
heterogeneous signatures of adaptation and therefore to influence the
adaptive architecture of polygenic traits, as already observed in plants
or animals facing environmental variations
(Yeaman et al., 2016;
Barghi et al., 2019).
Although convergent adaptation was revealed between some P.
fijiensis populations in most candidate regions (Table 2), overall, a
low level of convergence was found across all the populations and the 32
candidate regions analyzed. In some evolve and resequence experiments,
parallel evolution was found to be relatively rare
(Graves et al., 2017;
Griffin et al., 2017). Barghi et al. 2020 suggested that non-parallel
evolution should be considered as the most likely scenario because the
majority of adaptive traits are complex and polygenic. Non-parallelism
is expected for 1) polygenic traits controlled by multiple genes with
small effects, 2) when there is some redundancy between the genes
involved in the adaptive trait, and 3) when populations are
differentially affected by evolutionary forces
(Barghi et al., 2020).
The results obtained in this study suggest a polygenic basis for host
adaptation with genetic redundancy. In addition, the 14 natural
populations studied could have been affected by demographic events.
As already discussed in Carlier
et al. 2021b, the
change in the allele frequency spectrum observed in all the 14
populations, with an increase in intermediate-frequency alleles (Figure
3), may reflect a concomitant effect in candidate genomic regions of
population contraction and selection on P. fijiensis Cuban
populations. Constraints such as crop management, pesticide applications
or introduced resistance create bottlenecks and genetic drift will
randomly maintain some mutations that are putatively beneficial but not
others (McDonald &
Linde, 2002). Thus, the different history of the populations studied
could play a role in the non-parallelism we observed. Furthermore, theP. fijiensis populations are panmictic
(Carlier et al.,
2021a) and recombination should also favor non-parallelism
(Barghi et al., 2020).
Finally, other factors such as pleiotropy and epistasis not taken into
consideration in the present study can influence parallelism
(Bailey et al., 2017)
and further investigations are needed to better understand the relative
role of all the potential factors shaping adaptive architecture in plant
pathogen populations.
The results of this study suggest a polygenic basis for adaptation to
quantitative resistance and specific host-pathogen interactions inP. fijiensis . Specific interactions are not always detected in
erosion of quantitative resistance in other plant pathogens
(Cowger & Brown,
2019) and general adaptation to quantitative resistance could emerge is
such situations even through selective sweeps in a few genes or through
polygenic adaptation. General adaptation could lead to an impasse in the
use of quantitative resistance since greater pathogen aggressiveness
could be selected
(Zhan et al., 2015).
On the other hand, specific interactions in different cultivars, as
observed in P. fijiensis and other plant pathogens
(Montarry et al.,
2012; Delmas et al., 2016; Frézal et al., 2018), can lead to local
adaptation patterns
(Dumartinet et al.,
2019). Specific interactions in different cultivars associated with
fitness cost could lead to antagonist selection pressures on the
pathogen populations. Assuming that cultivars are the main habitat of
pathogen populations, such a situation would resemble the so-called
antagonistic pleiotropy process. In this process, alleles have opposite
effects on fitness in different habitats, and this is the most important
form of genotype x environment interaction involved in local adaptation
(Kawecki & Ebert,
2004; Mitchell-Olds et al., 2007; Anderson et al., 2013). Antagonistic
adaption to different quantitative-resistant cultivars could thus be
exploited to define durable resistance constraining the evolution of
pathogen populations. To this end, fitness cost and adaptive
architecture of pathogen populations need to be first analyzed in a wide
range of resistant cultivars using similar approaches to the ones
applied in this study. Finally, from the without a priori approach used
in this study, we were able to
highlight major candidate genes which accumulated several
characteristics and could now, using functional analysis, be further
investigated to better understand the mechanisms involved in the
quantitative pathogenicity of fungi such as P. fijiensis.