Co-evolution in a landrace meta-population: two closely related pathogens interacting with the same host can lead to different adaptive outcomes

We examined the local adaptation patterns in a system comprising several interconnected heterogeneous plant populations from which populations of two phylogenetically closely related pathogens were also sampled. The host is Hordeum vulgare (cultivated barley); the pathogens are Pyrenophora teres f. teres (net form) and Pyrenophora teres f. maculata (spot form), the causal agents of barley net blotch. We integrated two approaches, the comparison between the population structures of the host and the pathogens, and a cross-inoculation test. We demonstrated that two closely related pathogens with very similar niche specialisation and life-styles can give rise to different co-evolutionary outcomes on the same host. Indeed, we detected local adaptation for the net form of the pathogen but not for the spot form. We also provided evidence that an a-priori well-known resistance quantitative-trait-locus on barley chromosome 6H is involved in the co-evolutionary ‘arms race’ between the plant and the net-form pathogen. Moreover, data suggested latitudinal clines of host resistance and that different ecological conditions can result in differential selective pressures at different sites. Our data are of interest for on-farm conservation of plant genetic resources, as also in establishing efficient breeding programs and strategies for deployment of resistance genes of P. teres.

Scientific RepoRts | 5:12834 | DOi: 10.1038/srep12834 forms, which indicates that there are two closely related, but reproductively isolated, pathogen groups in Sardinia 28,29 . Moreover, the low level of linkage disequilibrium 28 and the co-existence in each field of both mating types at similar frequencies 30 strongly suggest that as well as clonal reproduction, sexual reproduction is also likely to be very important in the generation of the observed diversity in both of these Ptt and Ptm pathogens. Finally, it has also been suggested that drift (or founder effects) and low gene flow might have a more prominent role in the Ptt than the Ptm pathogen 28 .
Different host genes confer resistance to the Ptt and Ptm pathogens (e.g. table 1 of 31 ) which suggests that there have been separate host-pathogen evolutionary process for these two pathogen forms. Thus, in the present study, we investigated this 'host-two pathogens' plant-pathosystem for the local adaptation of the host and/or pathogen. Specifically, we ask the question whether the patterns of variations observed for the Ptt and Ptm pathogens reflect the same or at least partially independent underlying evolutionary processes.
With these aims, we compared the population genetics structure of the host H. vulgare with those of the pathogens P. teres f. teres and P. teres f. maculata, and we used two cross-inoculation designs to directly test for the local adaptation, separately for both of the two formae of P. teres. This allowed us to determine the consistency of different approaches for the detection of local adaptation, and we discuss further the differences between the Ptt and Ptm pathogens in their co-evolutionary interactions with their host.

Results
Neutrality tests. Among the 134 SSAP markers used to characterise the barley populations, six (4.48% of the total) were under divergent selection, with F ST varying from 0.499 to 0.777 (average, 0.601) and with P values between 0.05 and < 0.001. Among the AFLPs used to characterise the pathogens, eight markers out of 112 (7.14% of the total) were under selection among the Ptt populations of the pathogen, while only two out of 114 (1.75% of the total) were under selection among the Ptm populations. The average F ST values of the markers under selection in the Ptt and Ptm were 0.716 and 0.292, respectively. The P values were between 0.05 and < 0.001 for the Ptt, and < 0.05 for the Ptm. From hereafter, when necessary, we present calculations performed on the entire dataset or only using putatively neutral or putatively selected loci.

Comparison between plant and pathogen population structure for molecular markers.
Within population diversity. Considering all of the markers, the genetic diversity of neither the Ptt populations nor the Ptm populations were significantly correlated with the genetic diversity of the barley host, albeit a stronger effect was noted for the Ptt than for the Ptm (r = 0.674, P = 0.110 vs r = − 0.215, P = 0.728; Fig. 1). Interestingly, when only putatively neutral loci were considered, a difference between the Ptt and Ptm forms emerged: while the diversity of the Ptt populations and host populations were strongly correlated (r = 0.677, P = 0.0012), there was no apparent association between the Ptm and the barley diversity (r = − 0.201, P = 0.745; Fig. 1). When loci putatively under divergent selection were used, the correlations did not reach significance, although a difference between the Ptt and Ptm forms of the pathogen was still suggested (Ptt: r = 0.682, P = 0.110 vs Ptm: r = 0.545, P = 0.842; Fig. 1).
Between population diversity. Considering all of the markers, simple correlation analysis indicated that the genetic distances between the Ptt populations were significantly associated to those between the barley host populations (r = 0.586; P = 0.0107), while this was not the case for the Ptm populations (r = − 0.079; P = 0.577) ( Table 1). There was a similar pattern also for the correlations between the genetic distances and the geographic distances, albeit these did not reach statistical significance (r = 0.700, P = 0.0725 and r = 0.200, P = 0.306 for the Ptt and Ptm, respectively; Table 1).
The correlation between host population divergence and geographic distance never reached statistical significance; however, positive trends were observed (r = 0.246 and 0.431, respectively; Table 1). Thus, to disentangle the effects of geographic distance from those due to genetic distances between host populations on the pathogen population genetic divergence, a three-way Mantel test was performed to calculate the partial correlations (r) between the pathogen (Ptt or Ptm) and the host genetic distance, and between the pathogen genetic distance and the geographic distance. This confirmed that the genetic divergence between the barley populations explained a significant (r = 0.597, P = 0.0084) and non-trivial portion of the variance of the pair-wise population divergence for the Ptt (25.8%). However, this correlation was not significant for the Ptm (r = − 0.186, P = 0.6381), with only 1.59% of the variance explained (Table 1). Similarly, the geographic distances tended to have a stronger impact on the Ptt population divergence (41.4%) than on the Ptm population divergence (5.75%) ( Table 1).
The analysis conducted with the loci under selection confirmed that the genetic divergence between the barley populations explained a significant (r = 0.443, P = 0.0028) proportion of the variance of the pair-wise population divergence for the Ptt (14.7%). This was not the case for the Ptm (r = − 0.021, P = 0.433), with only 0.10% of the variance explained ( Table 1). The calculations with the neutral dataset also suggested differences, albeit less sharp, between the Ptt and Ptm forms (25.3%, P = 0.0570; 9.31%, P = 0.2082; respectively). Similarly, the geographic distance tended to have a stronger impact on the Ptt population divergence (33.1%-34.7% of the variance explained, depending on the dataset considered) than on the Ptm population divergence (0.05%-8.90%) ( Table 1).
In Table 2, the mean F ST values are presented for both the host and the pathogens. The genetic divergence among populations was statistically highly significant (AMOVA: P < 0.001 in all cases). Considering the entire marker datasets, the Ptt populations (F ST = 0.276) were about four-fold more differentiated than the Ptm populations (F ST = 0.071). The F ST obtained for the barley landrace populations (F ST = 0.189) was in between those obtained for the Ptt and Ptm forms of the pathogen. Thus, the host populations were about 30% less differentiated than those of the Ptt pathogen (F ST = 0.276), and 2.0-fold to 2.5-fold more divergent than those of the Ptm pathogen (F ST = 0.071). The Wilcoxon non-parametric-tests indicated significant differences in the mean F ST among the three organisms (P < 0.05), which further resulted in these all being separated based on the Tukey-Kramer tests (Table 2). Moreover, the Ptt and Ptm forms of the pathogens showed disjoint F ST bootstrap 95% confidence intervals. When the host and the pathogens were compared, the 95% confidence intervals marginally overlapped for the barley and the Ptt, and were disjoint for the barley and the Ptm (Table 2).
Assuming an island model of population structure, differences in the F ST estimated can be translated into differences in migration rates, mN isl ( Table 2). The Ptm showed the highest mN isl , followed by the The significance of each correlation is based on the Spearman-rank non-parametric method. Calculations were performed for the Ptt (red) and Ptm (green) separately, and using all of the markers, only the putatively neutral markers, or the markers putatively under divergent selection. barley host and then the Ptt. This pattern among these three organisms held also when the 'neutral' marker datasets were considered, even if, as expected, the F ST decreased and the estimates of the migration rates consequently increased ( Resistance structure and local adaptation. The barley fields differed significantly in terms of the distribution of the two formae speciales of the pathogen (χ 2 test, P < 0.001) ( Table S1). The strongest difference was observed for CUM, in the north of the island, which showed highly significant prevalence of the Ptm (P < 10 −6 ), and then for STU, in the south of the island, where in contrast, there was significant prevalence of the Ptt (P = 2 × 10 −9 ). Thus, we first tested for the significance of the pathogen forma specialis × barley populations interaction. Interestingly, ANOVA showed that the barley populations responded differently towards the Ptt and Ptm forms of the pathogen (Table 3). In particular, four populations (i.e., CUM, NXM, STU and COR) were more susceptible to the Ptm than the Ptt, while no significant differences where observed for the remaining two populations (PIR and TER) (Fig. 2). Overall, the Ptm produced bigger necrotic lesions than the Ptt ( + 16.84%; P < 10 −4 ). Moreover there was a significant, but not strong, correlation between the host-line susceptibility to Ptt and Ptm pathogens (r = 0.530, P = 0.0009; Fig. 3).    All of these findings suggested differences among the populations and single individuals in the resistance genes against Ptt and Ptm. Thus, from here on we considered the results from the cross-inoculation experiments for the Ptt and Ptm pathogens separately.
Here, a significant variation in the mean resistance among the barley populations was observed, which was true for both the Ptt and Ptm (Table 4). Moreover, the resistance of the host populations increased when moving from the south to the north of the island; i.e., there was a latitudinal cline in the host genetic resistance (Fig. 4). This was true for both the Ptt and Ptm pathogens, albeit there was a stronger effect for the Ptm (Fig. 4). Although we detected a statistically significant difference in the mean aggressivity among the Ptt and Ptm pathogen populations (Table 4), there was no associated latitudinal pattern ( Figure S1). Moreover, there was no significant correlation between the mean resistance of the host populations and the mean aggressivity of the associated pathogen samples (P > 0.05).
Interestingly, the ANOVA showed that the Sympatric versus Allopatric term was significant for the Ptt but not for the Ptt form (Table 4). This was seen by analysing the entire dataset (Table 4) or the data from the populations from which the pathogen samples were available (Table S2). In particular, we observed that the Ptt isolates produced 8.87% greater lesions in the sympatric interactions than in the allopatric interactions (Fig. 5); this provided evidence that on average and at the meta-population level the Ptt pathogen is locally adapted. Conversely, for the Ptm, no local adaptation or maladaptation (i.e. lower ability to infect local host than allopatric host populations) can be inferred.
Jack-knifing over the populations indicated that the significance of the Sympatric versus Allopatric term observed for the Ptt was mainly due to the CUM and STU populations (Table 5). Indeed, elimination of CUM or STU from the dataset resulted in non-significant P values (Table 5), while elimination of one of the population among the PIR, TER and NXM do not modify the inference of local adaptation. The marked effects of the CUM and STU populations was also confirmed when we analysed the data   from the populations from which the pathogen samples were sampled (Table S3). Interestingly, these populations were quite far apart (about 150 km), and as stated above, they were also characterised by very different Ptt/Ptm ratios (Table S1).
Comparison between population structure and cross-inoculation studies. As we showed local adaptation for the Ptt pathogen, we further investigated the relationships between the population structure and the local adaptation patterns. Interestingly, the strength of the pathogen local adaptation was positively and very strongly significantly correlated with the pathogen population divergence as  measured by AFLPs (r = 0.920, P = 0.0005; Fig. 6A). On the contrary, local adaptation did not significantly depend on the host population divergence when this was measured by all of the SSAPs (r = 0.628; P = 0.0816; Fig. 6B). Furthermore, using the putatively 'neutral' markers, the local adaptation was significantly correlated with the population divergence of both the pathogen (r = 0.906; P = 0.0002) and the host (r = 0.679; P = 0.025) (Fig. 6A1,B1), which was not the case when using markers that carried the signature of the divergent selection among the populations, albeit also here there was a slightly stronger effect for the pathogen compared to the host (AFLP: r = 0.541, P = 0.060; SSAP: r = 0.503, P = 0.726; Fig. 6A2,B2). Moreover, the correlation between the local adaptation and geographic distance (r = 0.666, P = 0.033; Fig. 6C) was less strong than the correlations between the local adaptation and the neutral AFLP ( Fig. 6A1) or neutral SSAP (Fig. 6B1), which suggested a fundamental role for the population structure of the two players in the co-evolution process.
The ratios between pathogen migration and host migration were not significantly predictive of the Sympatric versus Allopatric differences for the pairs of populations (r = 0.476, P = 0.108; Fig. 6D). However, after removing two population pairs for which the population divergence was not significant (one for the pathogen, one for the host; Table S4), the correlation become negative and significant (r = 0.908, P = 0.021; Fig. 6D1), which suggested that in this system there was higher pathogen local adaptation when the pathogen migrated less than the host.
Finally, we tested the hypothesis that a major QTL on chromosome 6H that confers the resistance of barley against the Ptt of barley net blotch in several genetic backgrounds 32,33 is responsible (at least partially) for the observed patterns of variation. With this aim, we genotyped the individual barley plants at a microsatellite locus (Bmag0009) that is associated (at a distance < 0.7 cM) to this major resistance QTL 34 . Here, the Bmag0009 locus do not show any signature of divergent selection (P > 0.05). However, the populations showed moderate significant allelic divergence at this locus (AMOVA = average F ST = 0.132; P < 0.0001; Fig. 7. In this regard, it is of note that the CUM and STU populations that are the most divergent at the Bmag0009 locus (F ST = 0.277, P < 0.0001; Fig. 7; Table S4) were also those with the most biased and contrasting Ptt/Ptm ratios (Table S1) and that produced the strongest signals of local adaptation ( Table 5). As expected, the different alleles at Bmag0009 locus are associated to different levels of resistance (Fig. 7). Notably, Fig. 7 also evidenced that moving from North to South the allele associated with the lowest susceptibility decreases in frequency while tend to be more frequent that associated with the highest susceptibility. Additionally, the ANOVA showed that the Bmag0009 genotype × Ptt pathogen population interaction was significant (F 20,3391 = 2.268, P = 0.001), which indicated that there were different effects of the Ptt pathogen populations on necrosis size in barley with different Bmag0009 genotypes, or alternatively, that there were different effects of the Bmag0009 genotype on necrosis size in barley infected by the different Ptt pathogen populations. In contrast, the interactions for Bmag0009 × pathogen population were not significant for the Ptm (F 20,3305 = 0.601, P = 0.915).
Interestingly, the divergence between the Ptt populations was more correlated with Bmag0009 (r = 0.809, P = 0.0158; Fig. 8A) than with SSAP (r = 0.614, P = 0.0677; Fig. 8B), which indicated that the Ptt population structure better-mirrored the spatial distribution of the locus-specific QTL resistance alleles (linked to Bmag0009) than the multilocus population structure depicted by the anonymous SSAP markers. Furthermore, the local adaptation of the Ptt pathogen was best explained by the host population divergence at Bmag0009 (r = 0.722, P = 0.0257; Fig. 8C) than by the host SSAP divergence (r = 0.614, P = 0.0677; Fig. 8B) or the geographic distances (r = 0.666; P = 0.0330; Fig. 6C). Thus, the pathogen populations collected from the host populations that diverged at the Bmag0009 locus (i.e., at the Ptt resistance QTLs on chromosome 6H) were also genetically more differentiated and produced stronger signals of pathogen local adaptation.
Collectively, this suggests that the pathogen variation was probably tracking the Ptt resistance QTLs on chromosome 6H.  Table 5. Results of the jack-knifing over the populations. The same model as in Table 5 was applied while sequentially eliminating each host population, to determine the P value for the Sympatric versus Allopatric term for the Ptt pathogen. This analysis was performed considering all six of the host populations (i.e., including COR, for which no pathogen sample was available). The populations are in descending order based on their 'weight' on the P value for the Sympatric versus Allopatric term.

Discussion
To study the co-evolutionary relationships between barley and P. teres, we conducted analyses at the meta-population level, through integration of the comparisons of the population structures of both host and pathogens with the cross-inoculation design.
In this study, the AFLP analyses do not have the main aim of estimating the variability within populations. Moreover our study does not have the aim of comparing differences in allele frequencies among populations at single-locus level, a case for which large samples are needed. The main aim was to investigate the multilocus genetic distances among the populations (F ST ), overall and between pairs   Table 4. Genetic distances between the pathogen populations are based on AFLPs (A), and those among the host populations are based on SSAPs (B). For each correlation, we report the coefficient of correlation (r), while the P values were determined by the Spearman-rank non-parametric method.
Scientific RepoRts | 5:12834 | DOi: 10.1038/srep12834 of populations, and to put this in relation to the migration rates, with the genetic distance among the host populations, and with the geographic distance among the sampling sites. In this regard, it should be noted that lower sample size are needed and that the reliability of the F ST estimates not only depends on the number of individuals per population, but also on the number of sampled populations and the number of markers.
The comparison of the host and pathogen population structures suggests that the organisation of the genetic diversity in the Ptt pathogen mirrors that observed for the barley host; this is the case for both the within and between population components of the genetic diversity. In contrast, no match was apparent between the barley and the Ptm pathogen. Furthermore, as the Ptt population structure is more strongly associated with the barley population structure than with the geographic distances, this suggests that host-mediated gene flow is more important than gene flow along the direct-line distance. This suggests that the barley and the Ptt pathogen have similar population dynamics (e.g., drift or bottleneck) and/ or patterns of gene flow. Moreover, the observed pattern might at least in part also reflect the reciprocal selection pressures. Indeed, the Ptt population structure better correlates with the host population structure at a single SSR locus that is linked to a Ptt resistance major QTL, than with the multilocus population structure depicted by the anonymous SSAP markers. Taken together, this suggests that the barley and the Ptt pathogen are involved in a co-evolutionary relationship.
If co-evolution is indeed taking place, with the evolutionary rate of a pathogen usually faster than that of its host , the usual expectation is that the pathogen is locally adapted. These arguments apply also to our system here, where the host is strictly selfing 19 and has only one generation per year; in contrast, the pathogen can undergo both sexual reproduction and clonal reproduction, with many generations during the host cropping cycle 35 . However, a large body of literature has shown that the population structure is also crucial in the determination of co-evolutionary trajectories. Single experimental studies, and simulation studies and meta-analyses have all emphasised the link between evolutionary potential, gene flow, and adaptation [5][6][7] . In particular, models and simulations show that in the meta-population context, the distribution of resistance and virulence and the detection of adaptive changes will depend on the ratio of the host and pathogen migration rates 5 . When pathogens migrate more than the host (m p > m h ), it is expected that the pathogen will be locally adapted. When the pathogen migrates at a level similar to its host (m p ≈ m h ), it will be expected that the relative evolutionary potential is not influenced by migration 5 . When a pathogen migrates less than their host (m p < m h ), it is possible that gene flow can equalise, or even reverse, the asymmetry in the evolutionary rates between the host and the pathogen; i.e., pathogens might have a co-evolutionary disadvantage and be less able to infect sympatric hosts than allopatric hosts 5 . In our context, local adapted (maladapted) parasites are more (less) able to infect host individuals from their local population. Local adapted (maladapted) host are more (less) resistant to pathogen individuals from their local population 1 .
In our system, the host migrates mainly by seed exchange among farmers 15,22 , whereas the pathogen migrates by infected seeds, and by wind and water splashing 35 . Albeit we did not see sharp differences, the three organisms investigated here showed different levels of average population subdivision (F ST ), which translate into different average rates of migration. We have here inferred more restricted gene flow of the Ptt pathogen than the Ptm pathogen (m Ptt < m Ptm ). The moderately high population structure observed for Ptt in this system is in line with other estimates [36][37][38] , as also the difference in the population structure between Ptt and Ptm 38,39 . We also observed a higher level of gene flow of the Ptm pathogen compared to the barley host (m Ptm > m h ), and a lower or similar level of gene flow of the Ptt pathogen compared to the host (m Ptt ≤ m h ). Thus, based on host/pathogen co-evolutionary models 5 , we would expect that in our system the Ptm will be locally adapted, and/or that the Ptt will be locally maladapted (m Ptt ≤ m h ) or not involved in a co-evolving interaction (m Ptt = m h ). Inferring migration from population structures relies on a series of assumption that are unlikely to be met in real populations 40,41 . However, this has been applied in several host/pathogen systems and it has allowed comprehensive and very meaningful meta-analyses that have been aimed at the deciphering of the connections between migration and local adaptation 7 .
In concluding here, the correlation between the host and pathogen population structures suggests that local adaption is more likely for the barley-Ptt than for the barley-Ptm interactions. Moreover, the differences in the migration rates among these organisms further suggest that the interaction of the barley with these two pathogens will have followed different co-evolutionary trajectories.
Detached leaf assays conducted on the first leaf have allowed the identification of international standard set of barley differential genotypes for P. teres 42 , as also the screening of a large germplasm collection 43 , the investigation of a gene-for-gene model in a P. teres/barley pathosystem 44 , and the mapping of resistance QTLs using multiple host populations and multiple isolates 45,46 . We also used this method to investigate local adaptation in our system.
When analysing data from the cross inoculation test and testing for effects at the pathogen-form level, we observed that the Ptm induced necrotic lesions that are significantly larger (+ 16.84%) than those of the Ptt. The rapid spread of the Ptm in several parts of the world appears to be a more recent phenomenon, and the reasons for this are not clear 31 . However, if the Ptm arrived in Sardinia more recently than the Ptt, i.e., if the Sardinian barley evolved without (or with limited) selective pressure exerted by the Ptm, it would be expected that the Ptm pathogen is more virulent on the Sardinian barley than the Ptt pathogen, which is indeed what we have observed. This equal testing of whether the metapopulation of Sardinian barley landraces is 'better-infected' by a resident (Ptt) than a non-resident (Ptm) pathogen thus inferring host adaptation to the Ptt. However, such a conclusion might be hampered by differences in the pathogen-form quality and not by host fitness variations.
Furthermore, we observed that the Ptt and Ptm forms of the pathogen were non-randomly distributed across the six barley fields, and that there was a pathogen form × host population interaction. This all suggests that different demes have experienced different selective histories against the Ptt and the Ptm.
When testing for effects at within-form local adaptation, a latitudinal cline of host resistance was observed for both the Ptt and the Ptm, with higher levels of resistance in the north than in the south of the island. Pyrenophora teres prefers moist and cold conditions 31 , and these are more likely in the north than in the south of the island (http://www.sar.sardegna.it/). Thus, the pathogen might have exerted differential selective pressures on the host populations at different sites. As the resistance genes in barley against the Ptt and Ptm are genetically independent (see table 1 of 31 ), this suggests that the observed pattern is due at least in part to parallel selection, rather than to a migration/drift process. Interestingly, selection gradients that are oriented along the north-to-south axis in Sardinia were also suggested for the barley host 15,22,23 .
The main result from the cross-inoculation design is that two closely related pathogens with very similar niches of specialisation and life-style 31 can give rise to different co-evolutionary outcomes on the same host. This is demonstrated by the ANOVA based on the infection data, where the Ptt of the pathogen was shown to be locally adapted, while for the Ptm there were no significant plant × pathogen population interactions. Moreover, we also provide evidences that a major resistance QTL on barley chromosome 6H is likely to be involved in the barley-Ptt co-evolutionary relationship.
Our cross inoculation experiment was conducted only once. While several informative studies have been conducted as only a single entire cross-inoculation experiment e.g. (12,47), we acknowledge that the repetition of the entire experiment would be of some benefit to reinforce our conclusion. However, it is relevant to note that the detached leaf assay used in this experiment has allowed the mapping of the same QTLs in different genetic backgrounds by different authors and in different laboratories even by conducting the overall experiment only once 45,46 . This points towards a good repeatability of our experiment.
Our experiment comprises a relatively low number of isolates per population. However, as shown by Blanquart et al. 13 by simulations, when studying local adaptation, it is better to have a small number of individuals from several populations, instead of sampling many individuals from only a small number of populations, and that at the extreme, it is better to have a lot of populations and only one isolate per population 13 . Overall, we considered six fields where the barley and pathogens were sampled. This is an appropriate number of fields 13 , and most importantly, these represent a high fraction of the existing fields where Sardinian barley landraces and the relative infecting P. teres pathogens can be sampled. This is relevant, as the estimated local adaptation will also be closer to the true local adaptation of the whole metapopulation 13 .
Moreover, as we adopted the same experimental design for the two forms, if our experiment has low power due to the low number of isolates per population, we would expect that local adaptation will not be detected either for Ptt or Ptm. However, we observed local adaptation for Ptt and not for Ptm. Thus, we can assume that local adaptation is indeed present in the Ptt form and not in the Ptm form, or alternatively, that local adaptation is present in both forms, but the magnitudes of the effects are very different between the two forms: strong in Ptt and weak in Ptm, such that only in the first case was this detectable in our experiments. Moreover, even if we cannot completely exclude that with more isolate per population the observed scenario would change for some of the details, when the population structures of the two forms were compared, numerous pieces of evidence combined to suggest that the hypothesis that the signal of local adaption found for the Ptt form and the contrasting results obtained for Ptt and Ptm with the cross-inoculation tests might be due to insufficient isolates sampling, is not parsimonious.
Interesting considerations emerged with the comparison of the results of the population structure and the cross-inoculation study. Indeed, the parallelism between the host and pathogen population structures correctly suggested a co-evolving interaction between the barley and the Ptt form. This has also been observed in contexts different from that of a landrace metapopulation 8 . Moreover, based on differential migration rates, we would expect local adaptation for the Ptm and/or maladaptation (or no signal of adaptation) for the Ptt, while the cross-inoculation revealed an almost opposite pattern; i.e., local adaptation for the Ptt, and no signal of co-evolving interactions for the Ptm.
The absence of a signal for local adaptation for the Ptm is in counter tendency compared to the results of a meta-analysis on pathogen local adaptation 7 that supports the concept that pathogen populations with high relative migration rates are more likely to be adapted to infection of their local host. A possible explanation for this is that in the present study, the geographical scale is not as needed (e.g., it is too small or too large) to detect local adaptation for the Ptm, but it is adequate for such for the Ptt 4 . Indeed, local adaptation can be detected at the scale of the individual, population or region, depending on the properties and dynamics of any given system 42 . If the pathogen migration is high between the populations (as appears to be the case for the Ptm form in our system) and the host populations are characterised by varying frequencies of different resistance types, then it would possible for the pathogen to adapt to the overall metapopulation frequency of the particular host genotypes 4 . In this case, local adaptation might be apparent when testing parasite performance at the local vs. foreign local metapopulation (i.e., at a higher geographic scale). Our study does not specifically show this point. However, as also noted above, the differences in necrosis size between the two forms at the metapopulation levels, as bigger for Ptm than for Ptt, suggests that at the whole metapopulation level, the Ptm might be better adapted than the Ptm, although this remains to be investigated in a specific study.
From another side, the same meta-analysis showed that when pathogens have a migration rate lower than that of their hosts, there is not a higher probability that pathogens perform worse on their local host than on allopatric hosts 7 .
Nonetheless, Kaltz et al. 48 suggested that low pathogen migration relative to host migration might account for local maladaptation in a smut fungus-plant system 11 . In this case, the pathogen populations were 12-fold more differentiated that those of its host 11 (i.e., the difference in the population structure between the host and pathogen were about six-fold that observed in the case of the barley-Ptt interaction). Moreover, in their system, the host was an obligate outcrosser, while the pathogen self-fertilises. In this regard, the present system shows a reversed situation, as the barley is a strictly selfing species while the pathogen can undergo both sexual and clonal reproduction.
Scientific RepoRts | 5:12834 | DOi: 10.1038/srep12834 In our system, the host and pathogens also have very different generation times, with barley undergoing one generation per year, during which time P. teres can undergo several cycles 31 . Moreover, the pathogen here can survive on the barley staples where sexual structures (asci) can also be formed 31 ; thus, infection will usually be reinitiated by the sexual products of a resident population. Overall, this suggests that factors other than migration must be taken into account here; i.e., that the higher migration rates of the host compared to the Ptt pathogen might not be sufficient to reverse the evolutionary advantage due to differences in lifestyle among the two antagonists.
More in particular, based on the correlation analysis, we observed that for explaining the pathogen local adaptation, the host population structure for a locus linked to a QTL for Ptt resistance is more important than the overall SSAP structure and the geographic distances. Thus, locus-specific effects are stronger than the 'neutral' baseline (represented by the neutral SSAP structure) and the isolation by distance. All of this suggests a role for selection. If in a deme the host resistance genes are selecting pathogens and the pathogen population is highly variable for virulence, it can be expected that the selection will be more efficient when it only acts on the individual of the population; i.e., without the migration that incorporates new genotypes from the surrounding populations. Indeed, in this case, successful pathogens will be diluted together with other (perhaps also less virulent) foreign pathogens, which will reduce the gain of selection. This might be important for P. teres, as a typical staples-born pathogen where infections are usually reinitiated by the sexual products of a resident population, and where secondary clonal reproduction can 'amplify' a successful pathogen genotype in the population. This would explain why the local adaptation of the pathogen (i.e., barley maladaptation) is high when the pathogen population divergence is high (i.e., when migration is low) also relative to the host. Moreover, this might be even more important when the pathogen has a relatively high virulence, a case where it is possible to observe local adaptation of the pathogen even in the face of a low relative migration rate 6 .
To correlate the differences in the local adaptation patterns observed between the Ptt and Ptm with differences in lifestyle is not easy. Indeed, both the Ptt and Ptm forms have very similar lifestyles, and both can undergo sexual reproduction and are staple-borne pathogens. Interestingly, Lightfoot and Able 49 differentiated the Ptt and Ptm on the basis of infection characteristics. They observed that while the Ptt mainly grows only intercellularly and can affect cells that are not immediately associated with the mycelia, in contrast, the Ptm initially forms haustorial-like intracellular vesicles. On this basis, the Ptt would be considered as a necrotroph, while the Ptm would be a hemibiotroph. However, it is not clear how (and if) this difference might lead to different co-evolutionary outcomes. Moreover, Lajenesse and Forbes 50 showed that studies based on broad host-range pathogens are less likely to demonstrate local pathogen adaptation. They argued that "this may relate to evolutionary lags during diffuse co-evolution of broad host-range pathogens with their hosts". This might suggest that the Ptm of P. teres would have a broader host range than the Ptt. However, while comprehensive studies on the host range of Ptt have been performed, little information is available on the host range of Ptm 31,35 . Moreover, the Sardinian populations of barley landraces are cultivated often in sympatry with bred barley cultivars. Moreover, as pointed out by Brown 51 , "crucial in the co-evolutionary dynamic is whether the islands of landrace amid a sea of the same species act as an alternate host with a particular resistance structure or in rotation with bred cultivars of the same species in the same fields" as "the pathogen populations would be subjected to diversifying selection on the alternative populations of the host". This appears to be important also for P. teres, and in particular for the Ptm, where epidemics in several parts of the world have been correlated to the large diffusion of susceptible commercial varieties 31 .
The populations used in this study were collected almost 15 years ago; however, the conclusions of our study remain valid: i.e., that coevolution in a landrace metapopulation at this geographical scale is possible, that the evolutionary potential of P. teres f. teres is high, and that at a given time point the outcomes of the co-evolutionary dynamics can be different for two closely related pathogens occupying the same niche of specialisation. We cannot affirm that repeating the same analyses today with new plant and pathogen populations will lead to the same results. However, this would not indicate that our results are not repeatable, but instead that the populations of both players have changed in number and/or quality during this period of time. However, while we acknowledge that this might be very interesting, our study was not designed to accomplish an over-time analysis of the plant and pathogen relationships.
Concluding, our results have relevant implications for the on-farm conservation of crop resources, plant breeding and deployment of resistance genes.
Most of the interest in on-farm conservation of crop biodiversity is due to the claim that, contrary to the ex situ conservation strategy, this provide the opportunity for plants and pathogens to coevolve. Our data suggest that this is indeed the case. However, our results also support the view that the products of coevolutionary dynamics not necessarily encounter the breeders needs and support the argument that the "presence of resistance genes in landraces 'unfrozen' on farm will inevitably evoke changes in the pathogen populations that could render the resistances obsolete" 51 . In parallel, our experiment outlines the high evolutionary potential of P. teres and suggests that an effective breeding strategy against P. teres should focus on quantitative resistance (instead of major resistance) and that, if quantitative resistance are not available, the major resistance genes can better be deployed in rotation trough time or space 52 .

Materials and Methods
All of the observations in the present study are from six sites in different areas on the island of Sardinia (Italy). Each of these sites corresponded to a field of a highly diverse multigenotypic population of the S' orgiu sardu barley landrace 24 , where both the plant (H. vulgare) and pathogen (P. teres) were sampled ( Table 6).

Host and pathogen sampling.
The host: Hordeum vulgare. The six host populations of 'S' orgiu sardu' were sampled in 1999 with the harvesting of 100 spikes along a diagonal in each field 15 . The fields were of similar areas (2-3 hectares). Based on information from the respective farmers, the fields that were chosen for the sampling had been cultivated with each population on the same site for at least 30 years. About 30 lines per population were used for the SSAP and SSR molecular marker characterisation, and six lines per population were used in the cross-inoculation design ( Table 6).
The pathogen: Pyrenophora teres. Leaves infected with P. teres were collected from the same six H. vulgare fields between January and April 2000 (Table 6). Infected leaves were collected from randomly selected, non-contiguous plants (at about 10 m apart) along the diagonals of the fields. Single monoconidial isolates were obtained from single lesions from each leaf. Each isolate was designated as P. teres f. teres (net form, Ptt) or P. teres f. maculata (spot form, Ptm) based on the morphology of the original lesion, on the AFLP analysis and the controlled re-inoculation tests 28 , and on the DNA sequence phylogeny 29 . A total of 150 isolates (68 Ptt, 82 Ptm) were considered here for the population structure analysis, and 34 (17 Ptt,17 Ptm) for the cross-inoculation study (Table 6).
For clarity and brevity the plant and the pathogen samples collected from the same field are labelled with the same name (that of the collection site). The distribution of the collecting sites across the regional territory is shown in Fig. 9.

Molecular analysis. Hordeum vulgare.
To characterise the host plants, we used the SSAP markers also used for a study of the population structure and linkage disequilibrium of a larger set of Sardinian barley landrace populations 15 . Each barley line was analysed using six SSAP primer combinations 15 . The SSAP method 53 exploits the combination of a primer designed on the long terminal repeat sequence of a barley retrotransposon (e.g., Sukkula, Nikita, BAGY-2, BARE-1) and an Mse primer, which usually generates high levels of polymorphism. For further details, the reader is recommended to refer to 15 .
To further characterize host plants, we considered the genotype data obtained by Bellucci et al. 23 at a microsatellite locus (Bmag0009), which is associated (at a distance < 0.7 cM) to a major quantitative trait locus (QTL) on chromosome 6H that confers resistance against Ptt [31][32][33][34] . DNA extraction and SSR genotyping were performed as described by Bellucci et al. 23 .
Pyrenophora teres. The field isolates were genotyped using AFLPs 54 , as described elsewhere 28,30 . Rau et al. 28 reported the data for two primer combinations, using two EcoRI (E) primers with two selective nucleotides (E-AC, E-GC) and 1 MseI (M) primer with a selective nucleotide (M-C). Rau et al. 30 extended the analysis, with data from two other primer combinations (E-AC/M-A, E-GC/M-A), to determine whether the isolates studied by Rau et al. 30 that shared the same multilocus genotype are true clones, and to better infer phylogenetic relationships among the isolates 29 . Here, for all of the calculations, we used data from all four primer combinations.
Host population sample (1) Corresponding pathogen sample (2) Table 4 of 28 ; and Table 6). Thus, the Ptt and the Ptm pathogens were fully comparable for four barley populations (CUM, PIR, SOR, NXM). Cross-inoculation tests were performed using a detached leaf assay 42,44 with some modifications Barley seeds were put in plateaux that were kept in a growth chamber under controlled conditions (temperature, 16 °C; relative humidity, 70%). Ten-day-old seedlings were then used. The first leaf of each seedling was cut and segments of 2.0 cm to 2.5 cm in length were placed in Petri dishes on a 0.5-mm layer of 1.2% w/v water agar (Agar Technical N° 3, Oxoid). A single drop (0.1 ml) of a suspension of each monoconidial isolate was put in the centre of each leaf segment. The suspensions were adjusted to 1,000 infective unit/ml 28,55 .
The experimental layout was a completely randomised design with six replicates (i.e., six leaf segments for each line × isolate combination were put in six different Petri dishes and inoculated). The experiments were conducted only once. The Petri dishes were then incubated under 12-h day and 12-h night, at a relative humidity of 80%, for 7 days. To score the infection responses, a semi-automated method was adopted. First, each Petri dish was scanned, to produce an image at a resolution of 600 dpi, which was then imported into a Kodak 1D software scientific image system v. 3.6.1. For each leaf segment, the necrotic area was delimited using the drawing instruments and 'excised' using the 'cut' options of the software (Fig. 10A,B). The necrotic area was then measured directly by counting the number of pixels in the excised image using the ROI function of the software.
The size of the necrotic lesion was adopted as the measure of fitness for both the host and the pathogen, as microscopic studies have shown that the growth of the pathogen after penetration is significantly slower in resistant host tissue 56 . This trait has also been successfully used to select barley differential lines and to map quantitative resistance loci 43,45,46 .

Statistical analysis.
Molecular analysis. To disentangle the effects of stochastic and demographic processes (mutation, drift, migration) from those putatively due to selection, we searched for loci that showed signatures of selection in the Sardinian landrace SSAP datasets of 189 individuals and 134 polymorphic loci. We used the 'detection of loci under selection' from the F-statistic procedure implemented in Arlequin, version 3.5 57 . This is essentially the FDIST approach 58 . We set 10 6 simulations to build the neutral expectations. Based on the results of this analysis, the full dataset was split into a putatively 'neutral' dataset (discarding loci identified as putatively under divergent selection; P < 0.05) and a putatively 'non-neutral' dataset (retaining only the loci for which P < 0.05). The statistical analyses were repeated using the three datasets (full, neutral and non-neutral).
Within-population genetic diversity. The within-population genetic diversity was calculated as the average gene diversity (H E ) across loci using the Arlequin version 3.5 software. The significance of the Pearson's r coefficients (r) between the within-population genetic diversity statistics of the host and the pathogen were calculated according to the Spearman's 'Rho' non-parametric method, using the JMP software version 7 (Sas Insitute, 2007).
Between-populations genetic differentiation. Population divergence estimates were obtained through the calculation of the Wright's F-statistics, or F ST 59 . For both the plant and pathogen populations, the total variance was partitioned into two levels, as among individuals within populations and among populations, using the hierarchical analysis of molecular variance (AMOVA) implemented in the Arlequin version 3. 5 software. The F ST was determined as the ratio between the variance among the populations and the total variance. The F ST were calculated separately for the Ptt and the Ptm populations of the pathogen. The significance of all of the F ST estimates was tested by permutations (10 5 replicates), and the F ST 95% confidence intervals were calculated by bootstrapping over loci with 10 5 replicates.
To determine whether the different organisms were characterised by populations with different average levels of population subdivision (F ST ), we also performed analysis of variance according to the Wilcoxon non-parametric test, using the JMP version 7. We estimated the average level of gene flow among populations using based on Wright 59 , which showed that for a haploid organism and for neutral alleles, F ST = 1/(1 + 2 Nm), from which Nm isl = (1-F ST )/2 Nm, where N is the local population size and m is the average rate of immigration in an 'island' model of population structure.
Geographic distances (km) were calculated on the assumption that the dispersal between populations occurred via a straight line 10 . Partial correlation analysis was performed to examine the relationships for the pathogen genetic distance (as F ST ) versus 1. geography (as km) without the confounding influence of the host, or 2. genetic distance between host populations (F ST ) without the confounding influence of the geography 10,60 . The significance of the standardised correlation coefficients (r) calculated was tested by adopting three-way Mantel tests with permutations (100,000 replicates) using the Arlequin version 3. 5 software.
The differences across the different barley fields in the ratio between the two formae speciales of the pathogen were tested using the Fisher exact test, as implemented in the JMP ver. 7 software (Sas Institute, 2007).
Cross-inoculation design. As suggested by Blanquart et al. 13 , to test for local adaptation we used the following linear model with three factors: host population, pathogen population, and Sympatric versus Allopatric plant and pathogen population combinations. Thus, we tested the null-hypothesis that there is no fitness advantage of being in sympatry relative to allopatry. This test is more powerful and easier to interpret than those based on Home versus Away and Local versus Foreign comparison because it takes into account the variance explained by pathogen and host populations 13 .
Specifically, we adopted the model proposed by Blanquart et al. 13 : where W i→j , k is the performance of the individual host plant k of population i inoculated with pathogens from population j; γ i is the effect of barley population i, ψ j is the effect of pathogen population j, δ i,j is a binary variable that is 1.0 if i = j and 0.0 if i ≠ j, α measures the Sympatric versus Allopatric effect, ε i,j represents the Genotype × Environment (GxE) interaction at the level of the population (with variance σ 2 ε,pop ), and ε i,j,k that accounts for individual-level error (with variance σ 2 ε,ind ) 13 . We tested α against the remainder of the interaction (the ε i,j ), and not against the individual error (the ε i,j,k ). Indeed, as also pointed out by Blanquart et al. 13 when testing for local adaptation the appropriate unit of replication is the population not the individual. We determined the level of significance of the F statistics for the Sympatric versus Allopatric test i with [1, P 2 -2(P-1)-2] degrees of freedom, where P is the number of populations considered in the experiment 13 .
To normalise the distributions of the residuals, the pathogen aggressivity scores were square-root transformed 8 . To determine the 'weight' of each population in the meta-population, the same ANOVA model described above was repeated, by jack-knifing over populations; i.e., by sequentially eliminating one population from the dataset.
All of the ANOVAs were performed using JMP version 7.