Genomic variability as a driver of plant–pathogen coevolution?

Abstract

Pathogens apply one of the strongest selective pressures in plant populations. Understanding plant–pathogen coevolution has therefore been a major research focus for at least sixty years [1]. Recent comparative genomic studies have revealed that the genes involved in plant defense and pathogen virulence are among the most polymorphic in the respective genomes. Which fraction of this diversity influences the host–pathogen interaction? Do coevolutionary dynamics maintain variation? Here we review recent literature on the evolutionary and molecular processes that shape this variation, focusing primarily on gene-for-gene interactions. In summarizing theoretical and empirical studies of the processes that shape this variation in natural plant and pathogen populations, we find a disconnect between the complexity of ecological interactions involving hosts and their myriad microbes, and the models that describe them.

Introduction

Pathogens proliferate within plant tissues by secreting effectors — virulence-associated proteins, RNAs and metabolites — that down-regulate basal defenses. Plants, in turn, employ myriad defense mechanisms to prevent pathogenic proliferation. Studies of plant– pathogen co-evolution have repeatedly implicated one class of interaction, the gene-for-gene interaction, as a primary interface at which pathogens evolve to evade detection, and plants evolve to improve detection. In this interaction, plants encode resistance (R) products that either directly or indirectly recognize the action of specific effectors. This recognition, which requires the presence of a specific R gene and a specific effector, induces localized cell death (hypersensitive response; HR) and a systemic defense response [2]. In this review we will focus primarily on the well-studied gene-for-gene interaction, relating to other means of plant–pathogen interaction only tangentially.

Plants encode hundreds of R-genes with differing specificities, and pathogens encode dozens to hundreds of effectors [3,4••,5]. These genes vary widely in copy number, single nucleotide polymorphisms (SNP) [4••,6,7], and expression levels [8] among individuals. This variation has the potential to impact disease epidemiology; host resistance polymorphisms can dampen pathogen spread [9] and virulence polymorphisms can determine host range [10]. In this review, we describe genomic variability in plants and microbes, and the mechanisms that generate and maintain this variability. We highlight recent breakthroughs that reveal complexities critical to the evolution of naturally occurring plant–microbe interactions.

Multiple processes maintain genomic variation

Both plant R-genes and pathogen effectors are located in genomic regions that undergo exceptional rates of mutagenesis. As a result, de novo diversity is generated quickly, providing material to allow rapid adaptation on the part of each species, but simultaneously creating variability that is maladaptive or neutral. Determining the impact of this diversity remains a central challenge.

Mutagenesis and selection in pathogens

Infection loci in pathogens are hypermutagenic

Effectors frequently contain high SNP diversity, many small insertions and deletions (indels), and presence-absence polymorphisms [4••,11] among isolates. As an example, whole genome sequencing of 19 phylogenetically diverse isolates of Pseudomonas syringae identified 58 type III secreted effectors (T3SE), only five of which were common to all isolates [4••]. Effectors in the Xanthomonads also vary widely among species and pathovars [12]. How is this variability generated?

In general, effectors are expected to mutate frequently, enabling pathogens to avoid detection within extant hosts or adapt to new hosts [13–15]. These virulence loci are often located in genomic islands or on separate chromosomes [13] that can be reshuffled, transferred between pathogens or lost from the genome [16]. In bacteria, effectors and pathogenicity islands tend to insert into tRNA genes that are targets for integration and excision. For oomycetes and fungi, effectors commonly lie in gene-sparse, repeat-driven regions, which presumably allows duplications, deletions, or gene fusions to occur without major fitness penalties [15].

Indeed, the loss and gain of effector-containing plasmids and islands can be observed after a few in planta passages [17]. This rampant exchange of genetic material can result in chimeric genomes, as exemplified by Xanthomonas species, which contain genes from groups as disparate as the Archaea and Eukarya [18]. It can also result in the immediate transfer of virulence to a previously non-pathogenic strain [13]

Selection on variability in pathogens is different between agricultural and natural ecosystems

Our view of how selection shapes variation at virulence genes is biased by agricultural systems, where most studies have focused. Several agriculturally relevant pathogens have undoubtedly experienced selection. For example, the recent deletion of the avirulence locus AvrLm1 has nearly swept to fixation (~90%) in the fungus that causes stem canker in rapeseed [19]. Similarly, P. syringae pv. tomato is starting to overcome field resistance conferred by the tomato genes Pto and Prf [20]. Indeed, even though naturally occurring effector loci harbor extensive variability [21], most agricultural research suggests that effectors quickly sweep to fixation in a manner consistent with an arms-race (but see [22]). This is inconsistent with the variation that is generated and segregates in nature. What explains the discrepancy?

First, plant–pathogen interactions are expected to differ between agricultural and natural ecosystems. Both theoretical and empirical studies have demonstrated that genetically heterogeneous plant populations, the norm in natural communities, are less susceptible to disease than plants grown in monoculture [9], as is common in agriculture. This lack of host diversity provides a uniform host environment for the pathogen, allowing for the selective sweep of effector loci beneficial in that one host background.

More importantly, as the major determinants of host specificity, effectors contribute to niche expansion [10]. Consistent with this, the effector repertoire of a single pathovar, found on a single host, is less extensive than the effector repertoire of the pathogen species when all hosts are considered [4••]. While selective sweeps may be apparent as a pathovar adapts to a particular host-genotype (i.e. in an agricultural setting), they should be less obvious when considering adaptation in the face of genetically variable hosts and pathogens. What’s more, the presence of effectors can shape interactions within the microbial community [23]. The complexity of species interactions in natural communities complicates analyses of plant–pathogen interactions, a point described in more detail below.

Mutagenesis and selection in plants

Resistance loci in plants are hypermutagenic

R-genes are the most polymorphic loci in plant genomes [6,24], largely because they occur in clusters of paralogous copies [5,25]. Their high sequence similarity and repetitive sequences (e.g. the leucine rich repeat (LRR) domain common to most R-genes) predisposes them to slippage and non-allelic homologous recombination, and leads to their frequent turnover and diversification (fusion, duplication, or deletion) [26,27]. Furthermore, the highest frequency of deleterious mutations, including premature stop codons or indels resulting in frameshifts, fall within members of the NBS-LRR, F-box and Receptor-like kinase (RLK) gene families [8], all of which have been implicated in disease resistance.

R-genes exhibit extreme variability not only at the sequence level, but also at the expression level. In A. thaliana the NBS-LRR gene family shows the highest levels of differentially methylated regions [28••] and gene expression differences [8] among individuals.

Plant resistance loci frequently show a signature of balancing selection

Allelic variation in many R-genes is likely to be ecologically and functionally relevant. Population genetic analyses in several plant species provide insight into the adaptive significance of this variation, identifying dozens of R-genes likely to have undergone diversifying selection [29–32]. For example, elevated Ka/Ks ratios are common in the LRRs of paralogous R-genes and suggest the rapid diversification of LRRs within a species [31]. Indeed, it has recently been shown that a diverse repertoire of functional R alleles, capable of recognizing a quickly evolving pathogen, occurs not only within, but also between, host species [33••]. Finally, elevated Ka/Ks ratios among orthologs suggest that related species may have evolved specificities to different virulence factors.

Interestingly, R-genes and other immune genes frequently exhibit a signature resembling balancing selection (Figure 1, Table 1) [34,35••,36,37,38••,39–46]. A clear example of this signature surrounds the gene RCR3 in the wild tomato species Solanum peruvianum. RCR3 is a cysteine protease cleaved by effectors from several pathogen species. Several complex haplotypes at RCR3 coexist within populations, and the proteins encoded by these different haplotypes are cleaved at similar rates by the pathogen effectors. These protein variants differ, however, in the level of defense response induced by this cleavage. These haplotypes exhibit high values of Tajima’s D, suggesting a role of balancing selection in maintaining quantitative variation in the stimulation of a defense response [35••]. This example is notable in revealing the evolution not of sites involved in pathogen recognition, but instead in the level of the response triggered by this recognition [41]. Huard-Chauveau et al. [38••] similarly show a signature suggestive of balancing selection acting on the expression level of a gene involved in resistance to Xanthamonas campestris.

Figure 1.

Evaluating the allele frequency spectrum to assess evidence of selection on a locus. Balanced polymorphisms can be distinguished by the (excess) intermediate frequency SNPs that arise on the deepest branches (a) of a genealogy. In this example, because of the long branches, the genealogy on the right is expected to have a higher number of intermediate frequency SNPs than the tree on the left. These common SNPs will tilt Tajima’s D (b), a commonly used test of selection, towards positive values, which are consistent with balancing selection (or complex population structure) and can be distinguished from neutrality using permutations or an outlier approach.

Table 1.

Examples of selection on R gene variation. This table summarizes the pattern of variation observed in intraspecies comparisons of R-gene orthologues in several plant species. Ka >Ks is interpreted as evidence of diversifying selection, while an increased time to coalescence and elevated Tajima’s D are signatures consistent with balancing selection. This table does not distinguish between the different processes that generate a signature resembling balancing selection, (see Fig. 1)

Species	Gene	Polymorphism type	Spp. recognized	Recognized effector	Putative selection	Evidence for selection	Reference
Arabidopsis thaliana	27 R-genes	Various	NA	NA	Balancing selection	~25% exhibit elevated Ka/Ks values, elevated Tajima’s D, low Fst	[40]
A. thaliana	RKS1	SNPs	Xanthomonas campestris	NA	Balancing selection	Elevated Tajima’s D 5′ of gene	[38••]
A. thaliana	RPM1	Deletion of locus	Pseudomonas syringae	avrRpm1	Balancing selection	Increased time to coalescence of presence absence polymorphism	[47]
A. thaliana	RPP13	SNPs	Hyaloperonospora arabidopsidis	ATR13, others NA	Balancing selection, diversifying selection	Ka > Ks, increased time to coalescence	[42,45]
A. thaliana	RPS2	SNPs	Pseudomonas syringae	avrRpt2	Balancing selection	Increased time to coalescence at the 5′ end of the LRR	[37]
A. thaliana	RPS4	SNPs	Pseudomonas syringae, Colletotrichum higginsianum, Ralstonia solanacearum	avrRPS4, others NA	Positive selection (selective sweep)	Reduced variation between alleles	[46]
A. thaliana	RPS5	Deletion of locus	Pseudomonas syringae	avrPphB	Balancing selection	Increased time to coalescence of presence/ absence polymorphism	[44]
Capsella rubella, Capsella grandiflora	Nine R-genes	Various	NA	NA	Balancing selection, diversifying selection	~20% exhibit increased time to coalescence, low Fst	[43]
Lactuca sativa (lettuce)	Several genes within RGC2 locus	SNPs, indels	NA	NA	Balancing selection	Trans-species polymorphism	[48]
Lycopersicon	Pto	NA	Pseudomonas syringae	avrPto, avrPtoB	Balancing selection	Elevated Pn/Ps levels, trans- species polymorphism	[34,36]
Solanum peruvianum	RCR3	SNPs	Cladosporium fulvum, Phytophthora infestans	avr2, EPIC1,2B	Balancing selection (possibly local adaptation)	Elevated Tajima’s D 3′ of gene, high Fst	[35••]

Several of the balanced resistance polymorphisms studied to date appear to be millions of years old, even persisting as trans-species polymorphisms [36,43,47–49]. The long-term maintenance of these polymorphisms indicates the ubiquity and stability of the dynamics that generated them.

Why does host–pathogen coevolution result in signatures of balancing selection?

Many processes generate signatures of balancing selection

What does the ubiquitous signature of balancing selection indicate about host–pathogen dynamics? The signature of balancing selection itself does not reveal the dynamics that produced it; very different adaptive processes can generate similar genomic signatures (Figure 2). Balancing selection indicates the maintenance of co-occurring alleles, which may result from heterozygote advantage, frequency-dependent selection, or environmental fluctuations that alternately favor different resistance and virulence alleles [50,51]. Local adaptation [52–54] and metapopulation dynamics [55] can also yield a signature that resembles balancing selection when a global population is considered, further complicating analyses.

Figure 2.

Processes that maintain variation in plant genotypes. This figure presents the hypothetical expected allele frequency, and genomic signatures (outlier statistics) resulting from three distinct selective scenarios capable of maintaining diversity at a resistance locus. Only scenarios (a) and (c) are classically considered scenarios of balancing of selection. Most studies to date have been unable to distinguish between scenarios (a) and (b) in molecular evolution analyses.

How do we determine the dynamics underlying a signature of balancing selection?

Two alternative approaches have been adopted to disentangle the processes behind a signature of balancing selection. First, finer analysis of molecular variation can point to particular dynamics [56]. Stahl et al. [47] suggested that the ancient balanced polymorphism at RPM1 in A. thaliana was maintained by frequency-dependent selection due to a skew in the molecular variation segregating within R and S classes. By embedding an ecological model within a coalescent framework, such a skew was shown to be consistent with the fluctuations that result from frequency dependence. In contrast, high F_st values associated with variation at the RCR3 locus in tomato suggest that its variability has been maintained by local adaptation [35••].

Second, one can ask whether the assumptions of models generating balancing selection are consistent with particular host–pathogen systems. Most models require tradeoffs in resistance and virulence [57,58,59], and evidence for these tradeoffs is mounting. For example, R-genes increase the fitness of infected plants [60] but incur an energetic cost that reduces fitness in the absence of infection [49]. Similar tradeoffs have been observed in pathogens [61].

In one of the few studies that have investigated these dynamics in field conditions, Thrall et al. [62••] tracked the allele frequencies of resistance genes and effectors over a six-year period in the Australian native flax (Linum marginale) and its fungal rust pathogen (Melampsora lini). They found rapid adaptation of the host and cycling polymorphisms in both the host and the pathogen. After challenging host populations from multiple years with pathogen strains from multiple years, they failed to find any systematic increase in pathogen virulence over time, suggesting the lack of an arms race. In another field study, Tack et al. [63] found that the interaction between Plantago lanceolata and its obligate pathogen Podosphaera plantaginis is shaped by frequency dependent selection on resistance traits. Although their study populations spanned hundreds of kilometers and exhibited genetic isolation by distance, most phenotypic variation in defense was maintained within host populations. Both of these studies point to the longterm maintenance of polymorphisms locally.

Plants interact with many pathogen species simultaneously

Pathogens rarely specialize on one host species

Whereas most studies of coevolution focus on the dynamics of tightly interacting pairs, host–pathogens in nature lie on a spectrum of specificities, with tight coupling accounting for fewer than half [64]. Without a tight, obligate association, it is not given that a pathogen will evolve in response to changes in a particular host species. Kniskern et al. [65] illustrate this point in showing that the ubiquitous pathogen Pseudomonas syringae is maladapted to its A. thaliana host in the Midwestern USA. That is, Midwestern A. thaliana are more resistant to the local P. syringae strains than to non-native P. syringae strains. It appears that P. syringae has tailored its infectivity to an alternative host or hosts rather than to the weedy, ephemeral, A. thaliana, which is unlikely to be a major host reservoir for this pathogen.

Hundreds/thousands of microbial species simultaneously interact with a plant

Another level of complexity lies in the enormous microbial diversity that shares a single host plant. A. thaliana, for example, hosts a microbial community consisting of thousands of species of microbiota [66]. The composition of this community is influenced by the surrounding environment, by the host genotype and by other microbes in the community [23,66,67].

A pathogen strain within that community will encounter not only host defenses but also those thousands of other microbial species [66,67], themselves competing and cooperating for the extraction of resources [23,68••]. All of these interactions complicate coevolutionary dynamics and call into question the relevance of models that consider the coevolution of a host with a single pathogen species.

Conclusion

Resistance and virulence genes exhibit exceptional levels of variation, in part due to the genomic processes that generate it and in part due to selection that promotes its maintenance. The persistence of this variation is unexpected on the basis of the dynamics that we see in agriculture, yet in nature there are clear indications of diversifying and balancing selection in action. Given the genetic and ecological complexities that we have touched upon, it is surprising that polymorphisms are maintained for long periods of time. While precious few examples exist to tease apart natural dynamics, we suggest that community level dynamics play a key role in explaining the ubiquity of virulence and resistance polymorphisms. Consideration of systems containing multiple hosts and pathogen species, interacting as assemblages, is an area ripe for future studies.