Understanding how pathogens evolve according to pressures exerted by their plant hosts is essential for the derivation of strategies aimed at the durable management of resistant cultivars. The spectrum of action of the resistance factors in the partially resistant cultivars is thought to be an important determinant of resistance durability. However, it has not yet been demonstrated whether the pressures exerted by quantitative resistance are different according to their spectrum of action.
To investigate selection pressures exerted by apple genotypes harbouring various resistance quantitative trait loci (QTLs) on a mixed inoculum of the scab disease agent, Venturia inaequalis, we monitored V. inaequalis isolate proportions on diseased apple leaves of an F1 progeny using quantitative pyrosequencing technology and QTL mapping.
Broad-spectrum resistances did not exert any differential selection pressures on the mixed inoculum, whereas narrow-spectrum resistances decreased the frequencies of some isolates in the mixture relative to the susceptible host genotypes.
Our results suggest that the management of resistant cultivars should be different according to the spectrum of action of their resistance factors. The pyramiding of broad-spectrum factors or the use of a mixture of apple genotypes that carry narrow-spectrum resistance factors are two possible strategies for the minimization of resistance erosion.
Since Dr Roy Johnson defined durable resistance 30 years ago (1981), breeding for durable plant resistance has remained a long-sought goal despite some progress in the understanding of factors that influence resistance durability in terms of both the pathogen and the host. McDonald & Linde (2002) suggested that the evolutionary potential of the pathogen is the main factor that determines resistance durability. Pathogen species with a mixed reproduction system, a high potential for gene flow, large effective population sizes and high mutation rates pose the greatest risk to overcome genetic resistance (McDonald & Linde, 2002). Other authors have investigated the potential for host resistance variation to determine evolutionary trajectories of pathogen populations. In natural systems, Thrall & Burdon (2003) demonstrated that the mean virulence, that is the average number of resistance genes overcome, of a pathogen population increases directly with the mean resistance of plant populations. Life history traits of the host, including host resistance, are essential factors for the determination of the evolution of the pathogen population, especially for obligate pathogens (Barrett et al., 2008). Two extreme categories of resistance are generally recognized: qualitative resistance conditioned by a single gene and quantitative resistance conditioned by multiple genes of partial effect. However, the distinction between these two categories is not straightforward and there is a ‘great deal of gray area’ between these extremes (Poland et al., 2009). In agroecosystems, genetically homogeneous crops with qualitative resistance facilitate strong directional selection on pathogen populations when they are planted over a large area, and can lead to resistance breakdown. The breakdown of major genes has been demonstrated in many pathosystems (Parisi et al., 1993; Bayles et al., 2000; Rouxel et al., 2003; Caffier & Laurens, 2005; Stokstad, 2007; Peressotti et al., 2010; Gladieux et al., 2011). Polygenic quantitative resistances based on several genes of partial effect (i.e. quantitative trait loci, QTLs) have been frequently considered to be broad-spectrum and have been empirically shown to be more durable (Parlevliet, 2002). Two hypotheses have been proposed to explain this extended durability. First, a pathogen would require the combination of a larger number of mutations in its genome to overcome polygenic resistance than to overcome monogenic resistance. Second, selection pressures exerted on the pathogen by quantitative resistance would be lower and distributed among several genes, which would reduce the risk of emergence of virulent variants from the pathogen population (Lindhout, 2002; Poland et al., 2009). The pyramiding of quantitative resistance factors with major resistance genes has also been shown recently to increase the durability of the latter resistance factors (Palloix et al., 2009; Brun et al., 2010).
Nevertheless, erosion of quantitative resistance over time following pathogen adaptation has been observed in some pathosystems (Abang et al., 2006; Montarry et al., 2006; Andrivon et al., 2007; Le Guen et al., 2007; Lehman & Shaner, 2007; Antonovics et al., 2011). Other authors have also shown that partially resistant cultivars exert directional selection on pathogen populations, but at a slower rate than the susceptible cultivar (Zhan et al., 2002; Sommerhalder et al., 2011). Such a weak selection could be the result of ‘minor-gene-for-minor-gene interaction’ proposed by Parlevliet & Zadoks (1977) to explain specific, despite partial, interactions observed between quantitative resistance and different pathogen isolates. Partially resistant hosts could therefore exert differential selection pressures on pathogens. Following a more theoretical approach, Gandon & Michalakis (2002) used a simulation model to demonstrate that quantitative resistance selects for more aggressive pathogens, leading to erosion of the resistance. It was assumed that all partially resistant host genotypes were infected by all pathogens, allowing within-host competition that selects for higher aggressiveness.
Many genetic mapping studies of disease resistance QTLs have been performed by independent inoculation of several pathogen isolates on a given mapping population. Such experiments have frequently shown strong isolate × QTL interactions, indicating isolate-specific (i.e. narrow-spectrum) resistance QTLs, as well as QTLs detected with all or most of the isolates, suggesting broad-spectrum resistance QTLs (Talukder et al., 2004; Jorge et al., 2005; Marcel et al., 2008). The spectrum of action of the resistance factors in the partially resistant cultivars is thought to be an important determinant of resistance durability (Kou & Wang, 2010). One of the putative mechanisms underlying quantitative resistance is the mutation of genes involved in basal resistance. Basal resistance is a broad-spectrum resistance and would be more difficult for a pathogen to evade and thus more durable (Poland et al., 2009). In the rice blast pathosystem, constitutive expression of defence is a ‘hallmark’ of partial resistance (Vergne et al., 2010). On the contrary, a narrow spectrum of action would imply a specific recognition system that could be more easily overcome by a pathogen following differential selection. It has not yet been demonstrated whether the pressures exerted by quantitative resistance are different according to the spectrum of action. More precisely, the identification of selection pressures exerted by individual resistance factors could make it possible to control virulence evolution by the management of host resistance through space and time (Palumbi, 2001; Sapoukhina et al., 2009).
There is a real lack of data at this time on the impact of quantitative resistance genes on the adaptation of pathogen populations, preventing plant breeders and landscape managers from proposing strategies for the choice of quantitative resistance combinations and to optimize their deployment. In this article, we have investigated the impact of apple quantitative resistances on the genetic composition of a mixed inoculum of Venturia inaequalis isolates. This pathosystem is particularly suited to the investigation of such questions. First, many quantitative resistance factors have been identified, including both specific and broad-spectrum QTLs of resistance (Calenge et al., 2004; Soufflet-Freslon et al., 2008). Second, this fungus has a great evolutionary potential, especially because of its mixed reproduction system (McDonald & Linde, 2002). Indeed, phylogeographic studies on V. inaequalis have demonstrated its great capacity to adapt to new environments (Gladieux et al., 2008, 2010; Lê Van et al., 2012). Another convincing example of adaptation was given by the breakdown in multiple orchards of the major resistance gene Rvi6 (Vf), widely used in apple breeding programmes (Gladieux et al., 2011).
Advances in molecular technology, such as the development of pyrosequencing, have permitted new insights into microorganism community composition (Benson et al., 2010). Quantitative measurement of the allele proportions of pathogen populations should make it possible to evaluate whether different plant genotypes exert differential selection pressures on a diverse pathogen population according to their resistance QTLs. Several isolates are generally inoculated independently to define the spectrum of action of the QTL. However, such practices are not adapted to the field situation, where isolates are in competition. The use of co-inoculated isolates is an alternative, but could lead to the identification of false broad-spectrum factors that are, in fact, resistance factors specific to the most competitive isolate in the mixture. Thus, we first investigated whether the definition of specific or broad-spectrum resistance QTLs is congruent when isolates are co-inoculated or inoculated independently. Second, using quantitative pyrosequencing technology, we evaluated whether host genotypes exert differential selection pressures on co-inoculated V. inaequalis isolates according to the spectrum of action of the resistance QTLs they carry. We hypothesized that broad-spectrum resistance does not exert differential selection, whereas narrow-spectrum resistance does. Evidence for this would be that broad-spectrum resistance factors do not select for particular isolates, whereas narrow-spectrum resistances decrease the frequencies of some isolates in the mixture relative to the susceptible host genotypes.
Materials and Methods
An apple progeny of 149 F1 individuals (Malus × domestica Borkh.) was derived from the cross ‘Discovery’ × ‘TN10-8’. ‘Discovery’ is an English cultivar derived from the ‘Worcester Pearman’ × ‘Beauty of Bath’ cross. ‘TN10-8’ is a French hybrid derived from a cross between cv. ‘Reinette Clochard’ and a hybrid derived from ‘Schmidt's Antonovka P.I. 172632’. Both parents are partially resistant to apple scab (Laurens et al., 2004). This progeny was used previously for genetic map construction and the detection of QTLs conferring resistance to apple scab employing seven single monoconidial isolates (Calenge, 2004; Calenge et al., 2004). Seven resistance QTLs were detected in Calenge's studies (2004) that could be classified into three classes according to their spectrum of action: a first class, including highly specific QTLs of resistance, that is detected with only one or two isolates, on linkage groups (LGs) LG5, LG12, LG13 and LG15; a second class, including moderately specific QTLs of resistance on LG1 and LG2, that is detected with four and five isolates on seven, respectively; and a third class, including one broad-spectrum QTL of resistance on LG17. The resistance QTL on LG17 was detected with all monoconidial isolates, except for one (isolate 1066) that possessed an avirulence gene towards the major gene Vg that also segregated in this progeny. The absence of resistance QTL detection with this isolate may be a result of the strong effect of the corresponding resistance gene that prevented the detection of resistance QTLs with a smaller effect. This isolate was discarded from our study. Six replications of dormant shoots were grafted onto ‘MM106’ apple rootstock for each individual of the F1 progeny.
Inoculation procedure and phenotyping
The environmental conditions, experimental design, inoculation procedure and disease assessment were identical to those in the study of Calenge et al. (2004). One genotype replicate was disposed per block in a randomized complete block design with six blocks. A mixture of six monoconidial isolates of Venturia inaequalis (Cooke) Wint. used by Calenge et al. (2004) and Calenge (2004) was prepared (Table 1). The method for producing inoculum has been described by Parisi et al. (1993). Monoconidial suspensions were first independently adjusted at a final concentration of 3 × 105 conidia ml−1 and then mixed in equal proportions to prepare the mixed inoculum. Consequently, the final concentration of each isolate in the mixture was 5 × 104 conidia ml−1. To verify conidia viability, the germination rate was assessed for each monoconidial suspension on agar plates before mixing. The germination rates were 68%, 90%, 80%, 70%, 91% and 84% for the isolates 104, 163, 302, EU-B04, EU-NL24 and EU-D42, respectively. The mixed inoculum was sprayed onto grafted trees. The humidity was maintained at 90% for the first 48 h after inoculation to allow conidial germination and fungus penetration. The humidity was then reduced to 70% and the temperature was maintained at c. 17°C. Disease severity (DS) was measured by sporulation severity and scored at 14, 21 and 28 d after inoculation (dai) using an ordinal scale ranging from 0 (no sporulation) to 7 (75–100% of sporulating leaf area; Parisi et al., 1993). At 28 dai, all sporulating leaves were harvested, air dried and conserved at −20°C.
Table 1. Origin of Venturia inaequalis monoconidial isolates
Cultivar of origin
‘Prima’ × ‘A143/24’
Venturia inaequalis DNA extraction
DNA was extracted from leaf fragments that revealed sporulating symptoms using a cetyltrimethylammonium bromide (CTAB) extraction procedure (Aldrich & Cullis, 1993). The four most highly sporulating replicates were chosen among the six replicates of each genotype. Four extractions per genotype replicate were performed using a square fragment of c. 1 cm2 of the most highly sporulating area. For genotypes without any sporulating symptoms, only a 1-cm2 fragment was extracted. The four DNA extracts from fragments of the same genotype replicate were pooled. DNA samples were diluted 1 : 50 before PCR amplification.
Venturia inaequalis marker selection and primer design
Sequences of a candidate effector gene, Vice16 (Kucheryava et al., 2008), and of an elongation factor, EF1α, were aligned for the six isolates after DNA extraction from in vitro cultures, amplification and sequencing according to the protocols described previously (Guerin & Le Cam, 2004). Five isolate-specific single nucleotide polymorphisms (SNPs) and one SNP with one allele shared by two isolates (104 and 302) and the other allele shared by the four other isolates were identified. PCR and sequencing primers (Tables 2, 3) were designed using the allele quantification option in PSQ Assay Design Software (Version 1; Biotage AB, Uppsala, Sweden).
Table 2. List of PCR primer pairs used in this study
Table 3. List of sequencing primers used in this study
Seq104 + 302
PCR amplification and allele proportion determination using pyrosequencing
Amplification of the target sequences was performed in a total reaction volume of 10 μl containing 1 × PCR 10X buffer, 1.5 mM MgSO4 for primer EF1 or 2 mM for primer Vice16, 0.2 mM deoxynucleoside triphosphate (dNTP) mix, 0.4 μM of each PCR primer, 0.03 U Taq DNA polymerase High Fidelity (Invitrogen) and 0.2 ng μl−1 of DNA. The reaction conditions were 94°C for 5 min, 45 cycles with denaturation at 94°C for 30 s, annealing at 60°C for 30 s and elongation at 68°C for 15 s, with a final elongation at 68°C for 5 min. To verify amplicon size and specificity of amplification, the PCR products were analysed by electrophoresis through a 1% agarose gel (E-gel; Invitrogen), and visualized under UV light (300 nm). Four microlitres of deionized water were added to the PCR before pyrosequencing. Amplicons were prepared for pyrosequencing analysis using the Vacuum Prep Tool (Qiagen) according to the manufacturer's instructions. Briefly, biotinylated PCR products (10 μl) were bound to streptavidin sepharose beads, captured on the Vacuum Prep Tool filter probes, washed with 70% ethanol, denatured with 0.2 M NaOH, and washed with 10 mM Tris-acetate, pH 7.6. After purification, single-stranded DNA was annealed with sequencing primer (0.4 μM). Pyrosequencing reactions were performed with a PyroMark Q96 MD instrument according to the manufacturer's instructions, using PSQ 96 Pyro Gold Reagents (Biotage AB) that contained enzymes, substrates and nucleotides. The quantification of the relative nucleotide proportion at each target location was performed using PSQ 96MD SNP Software, version 2.02 (Biotage AB), and the results were calculated using the AQ mode. In addition to the pyrosequencing of the four biological replicates, a technical replication was performed for one of the four biological replicates. Two PCRs followed by two allele quantifications by pyrosequencing were thus performed for the technical replicates of all the F1 progeny to determine the accuracy of the measurement.
Statistical analyses of the phenotypic data
All statistical analyses were performed with Statistical Analysis System (SAS) software (SAS Institute, Cary, NC, USA) and R 2.10.1 (R-Development-Core-team, 2008). Analysis of variance (ANOVA) was carried out separately for DS and the proportion of each isolate in the mixture, using the general linear model in SAS (PROC GLM). The proportion of each isolate was provided directly by the corresponding nucleotide proportion measured by pyrosequencing, except for the proportion of isolate 302. The latter was equal to the proportion of the shared nucleotide between isolates 302 and 104, minus the proportion of the nucleotide specific to isolate 104. Individual mean DS and individual mean proportions were calculated after an adjustment to block effect when it was significant. Normality of the individual mean DS distribution was assessed using the Shapiro test. Pearson's correlation coefficients were calculated between traits used in this study and sporulation severities obtained in a previous study (Calenge et al., 2004) using single isolate inocula. Broad sense heritabilities were estimated from ANOVA according to the following formula:
(, genetic variance; , residual variance; n, mean number of replicates per genotype).
QTL mapping was carried out using the individual genotypic data and the female and male integrated linkage map developed by Calenge et al. (2004). Two types of mapping software were considered: MapQTL 5.0 (Van Ooijen, 2004) and MCQTL (Jourjon et al., 2005). The former is based on a maximum likelihood approach, whereas the latter uses a linear regression model. For MapQTL analyses, simple interval mapping (IM) was first performed for each trait. Markers close to significant QTLs were then used as cofactors for multiple-QTL model mapping (MQM). This procedure was repeated until QTL detection was stabilized. For MCQTL analyses, the iterative QTL mapping (iQTLm) technique of Charcosset et al. (2000) was used. This is a scan method used for multiple QTL models at the genome level, with an exclusive window of 5 cM around the putative QTL and a forward stepwise method to select genetic cofactors from the whole genome. Depending on the software used, a genome-wide logarithm of odds ratio (LOD) score (MapQTL) or Fisher test (MCQTL) significance threshold was estimated trait by trait with the resampling method and permutation of the trait data (1000 iterations), according to Churchill & Doerge (1994). On the basis of the Fisher test value, an approximate LOD test value was obtained using the following formula:
where ddl is the numerator degree of freedom of the Fisher test. For each significant QTL, a 95% confidence interval was determined using the ‘2-LOD drop-off method’.
For each trait, a multiway ANOVA without interaction was performed with molecular markers near the QTL peaks to estimate the total percentage of phenotypic variation (R2) jointly explained by the significant QTLs. Digenic epistatic interactions between pairs of markers were tested using a two-way ANOVA with interaction. The terms ‘QTLDS’ and ‘QTLPROP’ are used here to refer to QTLs associated with DS variation and the variation in the relative proportion of isolates, respectively. Two categories of QTLDS are distinguishable: ‘QTLDS mono’ identified by Calenge et al. (2004) using the monoconidial isolates inoculated separately; and ‘QTLDS mix’ identified in this study using the mixture of six monoconidial isolates.
For each QTL mapped, the mean proportions, corrected by the germination rate, of each isolate in genotypes with and without the favourable allele, without considering alleles at the other QTLs, were compared using Student's t-test. The favourable allele was the allele that conferred a reduction of DS following infection with the mixed inoculum. In susceptible genotypes (i.e. genotypes without favourable alleles at any QTLDS), the expected mean proportion of each isolate corrected by the germination rate was compared with the observed proportion using Student's t-test, where the null hypothesis was a mean equal to 0.14, 0.19, 0.17, 0.14, 0.19 or 0.17 for isolates 104, 163, 302, EU-B04, EU-NL24 and EU-D42, respectively.
The ANOVA on the DS of the F1 progeny showed highly significant effects of the genotypes (P < 0.001), regardless of the scoring date. Distributions of individual mean DSs (Fig. 1) were not significantly skewed from normality (P = 0.28). Broad-sense heritability for DS at 21 dai was 0.76.
DS QTL analyses (QTLDS)
QTLDS mix detected at 14, 21 and 28 dai were the same, but the LOD scores were higher at 21 dai. As a result, we focused on DSs at 21 dai. Only QTLs detected with MQM and confirmed with MCQTL analyses are presented (Table 4). Four QTLDSmix were detected on LG1, LG2, LG14 and LG17, with DS measured at 21 dai using the six-isolate mixed inoculum. All detected QTLs, except QTLDSmix on LG14, co-localized with some QTLDS mono identified by Calenge et al. (2004; Fig. 2), but no co-localization was observed with highly specific QTLDS mono. The total percentage of phenotypic variation (R2) explained by the significant QTLDS mix was 0.36. A significant digenic epistatic interaction was found between markers on LG2 and LG17 (P = 0.037).
Table 4. Quantitative trait loci (QTLs) detected for two traits, disease severity scored after inoculation with the mixed inoculum of Venturia inaequalis isolates (DSmix) and relative isolate proportion (Prop), in the F1 apple progeny (‘Discovery’ × ‘TN10-8’) on the different linkage groups (LG)
For DS traits, the parameters associated with the QTLs were obtained using the multiple-QTL model mapping (MQM) method, whereas ‘Prop’ traits were obtained with the iterative QTL mapping (iQTLm) method.
Maximum logarithm of odds ratio (LOD) score.
Map position of the maximum LOD score.
Confidence interval in cM corresponding to a two-LOD score drop-off on either side of the likelihood peak.
Proportion of the explained phenotypic variance.
Molecular marker closest to the likelihood peak of each QTL.
The accuracy of allele quantification in V. inaequalis isolates was high with correlation coefficients between two technical replicates ranging from 0.72 to 0.94, depending on the isolate (Table 5). Broad-sense heritability of each isolate proportion was high and ranged from 0.67 to 0.95, depending on the isolate (Table 5). Here again, all detected QTLPROP (Fig. 2) co-localized with some QTLDS mono (Calenge et al., 2004), but a strict consistency was not observed isolate per isolate. QTLPROP were detected on LG1 for proportions of isolates EU-B04, EU-NL24, 163 and 302. For the first three isolates, a consistent co-localization was observed with QTLDS mono. The QTLDS mono for the isolate EU-NL24 was a suggestive QTL in the study of Calenge et al. (2004) (i.e. present but not significant). Conversely, no QTLDS mono was observed for isolate 302 (Calenge et al., 2004). No QTLPROP were detected for isolates 104 and EU-D42 despite the fact that QTLDS mono were observed for these isolates (Calenge et al., 2004). Moreover, the same favourable (i.e. decreasing DS) allele was always involved for QTLDS mono (Calenge et al., 2004), whereas this allele either decreased (for EU-B04, 163 and 302) or increased (for EU-NL24) the isolate proportion (Fig. 3). On LG2, QTLPROP were detected for isolates 302 and EU-D42, but not for isolates EU-B04, 104 and EU-NL24, although corresponding QTLDS mono were detected previously at that position (Calenge et al., 2004). Here again, the favourable alleles previously identified for QTLDS mono were not always in accordance with the increase or decrease of the corresponding isolate proportions on the leaves that sporulated after mixed inoculation. QTLPROP were detected for isolates EU-B04 and 302 on LG5. These QTLPROP co-localized with the isolate-specific QTLDS mono previously detected with isolate EU-B04 (Calenge et al., 2004). Reduction of DS for isolate EU-B04 was consistently associated with the reduction in its proportion in the isolate mixture, whereas an increased proportion was observed for isolate 302 (Fig. 3). The total percentages of phenotypic variation (R2) jointly explained by the different QTLPROP detected for isolates EU-B04 and 302 were 0.49 and 0.55, respectively (Table 4). In contrast with QTL detection with the mixed inoculum, no QTLPROP were detected that co-localized with the broad-spectrum QTLDS on LG17 and LG14. Overall, only QTLPROP that co-localized with highly and moderately specific QTLDS mono were detected. Indeed, the mean proportions of isolates in genotypes with the favourable allele of highly or moderately specific QTLDS mono (i.e. QTLs on LG1, LG2 and LG5) were, in some cases, significantly different from the mean proportions of isolates in genotypes without the favourable allele (Fig. 3). However, for the broad-spectrum QTLDS mono, the mean proportions of isolates were not significantly different between genotypes with or without the favourable allele (Fig. 3). These results underlined the selection pressure exerted by moderately and highly specific QTLDS mono.
Table 5. Accuracy of allele quantification and broad-sense heritability (h2) for each Venturia inaequalis isolate proportion
Correlation coefficient between the two independent allele quantifications performed on two independent PCR amplifications of one of the four genotype replicates for all the F1 progeny.
Prop104 + 302
Isolates were inoculated in the mixture in equal proportions. The expected relative proportion corrected by the germination rate for each isolate was between 0.14 and 0.19. However, for susceptible genotypes (i.e. genotypes without favourable alleles at any QTLDS mono), the observed isolate proportion means (0.11, 0.11, 0.08, 0.43, 0.31 and 0.13 for isolates 104, 163, 302, EU-B04, EU-NL24 and EU-D42, respectively) were significantly skewed from the expected relative proportion, except for isolates 104 and EU-D42 (t =−1.25, P =0.24 and t =−2.01, P =0.07, respectively). Thus, in the absence of selection pressure of resistance QTLs, isolate proportions varied as a result of a competition effect.
Correlation between DS and relative isolate proportion
The mean DSs evaluated for each genotype with the isolates EU-B04, 104, 302 and 163 inoculated independently in Calenge et al. (2004) and in Calenge (2004) were significantly correlated (P <0.001) with the relative proportions of the respective isolates using Pearson's correlation coefficients of 0.66, 0.37, 0.42 and 0.37, respectively. Conversely, the proportions of isolates EU-D42 and EU-NL24 were not significantly correlated with the corresponding mono-isolate DSs (P >0.05). The observed proportions of each isolate in the mixture were therefore not uniquely dependent on the individual isolate aggressiveness.
Host populations have a strong impact on the evolutionary dynamics of pathogen populations (Thompson & Burdon, 1992; Regoes et al., 2000). Adaptive modification of pathogen populations following the introduction of major gene resistance is a well-known phenomenon. However, whether the introduction of quantitative resistances can lead to a shift in pathogen populations is much less well understood. Notably, the impact of the spectrum of action of quantitative resistance on pathogen populations has not been investigated previously. To our knowledge, this is the first study to demonstrate directly the differential selection pressures exerted by quantitative resistance according to their spectrum of action on a mixed inoculum in plants.
Mixed inoculum allowed the detection of broad-spectrum and moderately specific QTLs
QTL analysis for DS, scored after inoculation with a mixed inoculum, made it possible to confirm the QTL on LG17 and the QTLs on LG1 and LG2, previously characterized as broad-spectrum and moderately specific, respectively (Calenge et al., 2004). The highly specific QTLs, that is QTLs detected with only one or two isolates (Calenge et al., 2004), were not detected using the mixed inoculum, regardless of their efficacy. This result suggests that the reduction in DS as a result of the presence of the resistance allele of a highly specific QTL was compensated for by the increase in sporulation of the other isolates not controlled by this QTL. This finding has important consequences for screening for new resistances. Breeding programmes using mixed inoculum may fail to identify isolate-specific QTLs. However, the use of mixed inoculum did not restrict QTL detection to broad-spectrum QTLs, as moderately specific QTLs were detected. The mixed inoculum also permitted the detection of a new apple scab resistance QTL on LG14, which can be considered as a broad-spectrum QTL, as it was not detected using isolate proportions. However, apple scab resistance factors have never been identified on this LG. This QTL thus needs to be confirmed in a new experiment or by the use of other apple mapping populations and/or other isolates. A similar result was observed in wheat, where a QTL was detected using a mixture of three isolates, whereas it was not found to significantly reduce DS for the three individual isolates of Puccinia striiformis f. sp. tritici when inoculated independently (Christiansen et al., 2006).
Contrary to broad-spectrum QTLs, specific QTLs exerted a differential selection on co-inoculated isolates
In this article, we demonstrated a strong differential selection among co-inoculated isolates, mainly driven by the spectrum of action of highly or moderately specific QTLs. This is the first time that such an impact of the host genotype on the genetic composition of a mixed inoculum has been monitored within a mapping population in a plant species. A somewhat similar approach was published by Benson et al. (2010), who identified QTLs that controlled relative abundance of specific microbial taxa in mouse gut using quantitative pyrosequencing. In our study, a major insight was obtained by the comparison between QTL analyses performed either on single monoconidial inoculations or on mixed inoculations (QTLDS) coupled with the monitoring of each isolate proportion (QTLPROP). QTL detection at a finer scale using the relative isolate proportion allowed us to detect a QTLPROP that co-localized with the first of three highly specific QTLDS previously identified on LG5, LG13 and LG15 (Calenge et al., 2004). This result outlines the differential selection pressure exerted by specific QTLDS. The highly specific QTLDS on LG5 had a rather strong effect on one of the isolates (EU-B04) and explained > 20% of the phenotypic variation. The two other highly specific QTLDS were minor QTLs (R2 < 8%). The minor effect of these QTLDS could explain why we did not detect any QTLPROP on LG13 and LG15. The mechanisms underlying the three highly specific QTLDS may also vary. Conversely, it was not possible to identify any QTLPROP that co-localized with the broad-spectrum QTLDS on LG17, suggesting the absence of differential selection pressure exerted by such a QTL. The results obtained here concerning the differential selection of isolates could vary greatly according to the mixture of isolates chosen. Differences in aggressiveness or competitive ability could lead to different selection pressures. Although only a restricted range of isolates could be co-inoculated in this study, it represents an attempt to model the interactions among isolates that infect the same leaf of a tree in an orchard.
Selection pressures exerted by specific QTLs remodulated within-host competition
The detection of QTLPROP was not fully consistent with the previous detection of QTLDS for the corresponding single isolates. This observation could be partly explained by within-host competition between isolates. This competition was suggested by the significant distortion of isolate proportions with regard to the expected proportion in the susceptible genotypes. Highly and moderately specific QTLDS may have interfered with isolate competition in two connected ways: first, these QTLs may have changed the overall competition situation by filtering certain isolates; and second, the same QTL, especially when pyramided in individual genotypes, may have reduced the mycelium density within the invaded leaves, thus attenuating the competition intensity by widening sporulating spots. A signature of such a complex interplay between differential selection pressures exerted by specific QTLDS and the reorganization of isolate competition patterns and intensities was illustrated by the fact that the reduction in one isolate proportion caused by a specific QTL effect was followed by different distributions of other isolates according to the genotype. For example, the reduction in the proportion of isolate EU-B04, caused by the effect of the QTL on LG1 or LG5, is followed by either a decrease or an increase in the isolate 302. Such contrasting patterns cannot be explained simply by considering the individual contribution of QTL regions according to their spectrum of action. Indeed, isolate 302 is not controlled by the QTLs on LG1 and LG5. The implication of differential competition situations can thus be invoked. Competition between isolates is also supported by the absence of correlation between disease scored for some isolates inoculated separately and their relative proportion in the mixture. This result is in agreement with other studies, and illustrates the difficulty of predicting competitive outcome based on aggressiveness in single inoculations (Miedaner et al., 2004; Von der Ohe & Miedaner, 2011). However, direct quantification of each isolate during a single infection should be compared with isolate quantification in multiple infections to demonstrate inter-isolate competition. Indeed, a high-frequency isolate does not necessarily compete with other isolates in the case of extensive spatial distribution. Similarly, a low-frequency isolate is not necessarily a poor competitor. It could be an isolate that poorly exploits its host, even during a single infection.
It is tempting to advocate the use of broad-spectrum QTLs rather than specific QTLs. Our results allowed us to better understand the selection pressures exerted by resistance QTLs according to their spectrum of action, but did not make it possible to reject any of the resistance factor type. However, the QTL spectrum must be taken into account for the management of resistance durability, which will be different according to the nature of the quantitative resistance. The pyramiding of broad-spectrum factors or the use of a mixture of apple genotypes that carry narrow-spectrum resistance factors are two possible strategies aimed at the minimization of resistance erosion.
We could expect that QTLs exerting differential selection pressure on the mixed inoculum will tend to select specific isolates, whereas a more stable relative proportion over time is expected with QTLs, such as the QTL on LG17, for which no differential selection pressure was found. It would be possible to determine whether the differential selections of isolates exerted by resistance QTLs are stable after several asexual cycles, as even small differences in competitive ability can be exacerbated. Differences in selective pressure exerted by QTLs need to be validated in field populations using a high-throughput genotyping approach that makes it possible to compare pathogen genetic diversity in relation to QTLs carried by apple genotypes. If validated, the strategy of pyramiding several broad-spectrum QTLs rather than specific QTLs is expected to be the most durable. In contrast, pyramiding only specific QTLs may lead to a particular differential selection pattern that accelerates the selection of a subpopulation, as in a gene-for-gene relationship.
As demonstrated by Gandon & Michalakis (2002), broad-spectrum resistance that allows multiple infections can lead to the selection of isolates with higher aggressiveness, and therefore to resistance erosion. As genotypes with specific QTLs also allow multiple infections, a selection for a particular subpopulation is expected. If orchards were to be set up with multiple apple genotypes that carry different specific QTLs, we might hypothesize that their complementary effects could impede the selection of one particular subpopulation (Lannou, 2001). Studies of pathogen epidemic modelling have made it possible to evaluate the best spatial pattern in a given landscape, but most of these studies have focused on gene-for-gene interactions (Sapoukhina et al., 2009; Fabre et al., 2012). There is a real lack of theoretical models at this time that provide information on the best way to manage QTLs, both for the construction of genotypes and their subsequent spatial deployment. We therefore assume that modelling approaches that use knowledge on the selection pressures exerted by different quantitative resistance factors are the most promising way to achieve disease resistance durability.
We would like to thank Caroline Denancé, Pascale Expert, Marie-Noëlle Bellanger and Marie De Gracia for their excellent technical assistance, and Frédérique Didelot for advice on statistical analyses. We are grateful to all of the technicians of the INRA Experimental Unit (UE Horticole, directed by Arnaud Lemarquand) and of the glasshouse facilities (Installations Experimentales, directed by Nicolas Dousset), especially Lysiane Leclout for taking care of the plant material. We would also like to thank Anne-Charlotte Bernard and the Plateforme de Biotechnologie Moléculaire of Angers for their help in primer design. We acknowledge Laurence Hibrand-Saint Oyant and the ANAN platform of SFR QUASAV. Amandine Lê Van was supported by a fellowship from INRA (SPE and GAP departments), and the Region Pays de La Loire. This work was partially funded by the European Network of Excellence, ENDURE, and by the COSAVE programme (Région Pays de La Loire).