Targets of selection in a disease resistance network in wild tomatoes

SUMMARY

Studies combining comparative genomics and information on biochemical pathways have revealed that protein evolution can be affected by the amount of pleiotropy associated with a particular gene. The amount of pleiotropy, in turn, can be a function of the position at which a gene operates in a pathway and the pathway structure. Genes that serve as convergence points and have several partners (so‐called hubs) often show the greatest constraint and hence the slowest rate of protein evolution. In this article, we have studied five genes (Pto, Fen, Rin4, Prf and Pfi) in a defence signalling network in a wild tomato species, Solanum peruvianum. These proteins operate together and contribute to bacterial resistance in tomato. We predicted that Prf (and possibly Pfi), which serves as a convergence point for upstream signals, should show greater evolutionary constraint. However, we found instead that two of the genes which potentially interact with pathogen ligands, Rin4 and Fen, have evolved under strong evolutionary constraint, whereas Prf and Pfi, which probably function further downstream in the network, show evidence of balancing selection. This counterintuitive observation may be probable in pathogen defence networks, because pathogens may target positions throughout resistance networks to manipulate or nullify host resistance, thereby leaving a molecular signature of host–parasite co‐evolution throughout a single network.

Most proteins do not operate in isolation, but as components of complex pathways or metabolic networks. The ‘connectivity’ of a protein (i.e. the number of interactions of a protein with other components of a pathway) may determine the level of constraint, and hence the rate of molecular evolution. Indeed, in yeast, the connectivity of genes in the network is correlated with their rate of evolution (Costanzo et al., 2010). Similarly, the position in the pathway or network can affect the evolutionary constraint on the protein. For example, downstream proteins that serve as convergence points of a diverse group of signalling upstream molecules may be subject to greater evolutionary constraint than the upstream molecules (Alvarez‐Ponce et al., 2009). This can be viewed in terms of the extent of the pleiotropic effects of amino acid substitutions on proteins which serve as convergence points for different signalling molecules. Mathematically, it has been shown that highly pleiotropic genes ought to exhibit much reduced molecular variation (Waxman and Peck, 1998). Finally, the level of constraint may be affected by the degree of redundancy of genes in a pathway, which, in turn, may depend on whether the proteins are encoded by single‐copy genes or by duplicate genes with overlapping functions (Costanzo et al., 2010; Wagner, 2001).

Some of the first studies of pathways in plants indicated that upstream genes in biochemical pathways showed the greatest protein conservation when compared with downstream genes (Lu and Rausher, 2003; 1999, 2008). Recent studies in Arabidopsis have evaluated a number of defence genes, some of which are known to operate together in specific signalling pathways (2006, 2008; Caldwell and Michelmore, 2009). Although these studies were not explicitly designed to test the effect of pathway position on evolutionary rates, the combined analysis of 27 resistance (R) genes and 27 downstream genes in Arabidopsis by 2006, 2008) revealed that, although some R genes showed histories of transient balancing selection or partial selective sweeps, genes further downstream experienced almost exclusively purifying selection. These results are consistent with expectations that genes downstream in defence networks experience greater evolutionary constraint and upstream genes can be subject to adaptive change.

In this article, we describe sequence variation at genes operating at different points in a network controlling disease resistance in wild tomatoes. Such case studies complement the analyses of large protein databases, because, in case studies, the forces underlying evolutionary constraints can be analysed in much greater detail and can potentially capture a different evolutionary timescale (e.g. Wagner, 2000, 2001). Furthermore, the evolution of components of a signalling network may not conform to expectations based on biochemical pathways. As a first step in our analysis of this resistance network, we have previously characterized the sequence variation of the pathogen resistance R gene, Pto, and the functional consequences of this variation within and between populations of seven Solanum species (Rose et al., 2005, 2007). We found evidence for elevated levels of amino acid polymorphism at this R gene, consistent with balancing selection at this locus. In this article, we describe the sequence variation and level of evolutionary constraint for an additional four genes involved in this signalling network: Fen, Prf, Pfi and Rin4.

This network has multiple signalling molecules, including Pto, Fen and Rin4, as potential inputs (Fig. 1). Pto is a receptor kinase that is activated on binding of pathogen‐derived molecules, AvrPto or AvrPtoB (reviewed in Oh and Martin, 2010). Fen is a paralogue of Pto and confers sensitivity to the insecticide fenthion (Chang et al., 2002; Martin et al., 1994). Fen can also recognize and activate defence responses to versions of the Pst effector AvrPtoB lacking E3 ligase activity (Rosebrock et al., 2007). However, wild‐type forms of AvrPtoB ubiquitinate Fen, which leads to its degradation in plant cells. The Rin4 gene was originally identified in Arabidopsis and plays a role in several R gene signalling pathways (Kim et al., 2005). Identification and functional studies of Rin4 in tomato indicate that it is degraded in the presence of pathogen effectors, AvrPto and AvrPtoB, and this degradation depends on Pto and Prf (Luo et al., 2009). As Rin4 is believed to be a negative regulator of basal defence, its degradation is predicted to activate downstream defences.

Figure 1.

Schematic representation of the Pto signalling pathway in tomato.

Prf interprets and transduces these signals. Prf is located in the same cluster as the Pto gene family, although it is phylogenetically unrelated to Pto and its paralogues. The Prf protein contains regions showing homology to nucleotide‐binding sites (NBS), coiled‐coil (CC) domains and leucine‐rich repeats (LRR). Silencing of Prf prevents signalling by Fen or Pto, indicating that Prf acts epistatically to Fen and Pto. It has been demonstrated that the two kinases, Pto and Fen, physically interact with the same N‐terminal portion of Prf (Mucyn et al., 2006; Ntoukakis et al., 2009). The fifth gene in our study, Pfi (originally named Prf‐interactor 30137), was cloned via yeast two‐hybrid screen with a portion of the CC and NBS region of Prf as bait (Tai, 2004). This gene encodes a protein with homology to basic helix–loop–helix transcription factors. Functional testing of Pfi indicated that over‐expression in tomato suppresses the hypersensitive response (HR), whereas viral‐induced gene silencing of Pfi showed no phenotypic effect (Tai, 2004). As such, this gene appears to be a negative regulator of HR. Our further bioinformatic analyses of Pfi revealed two additional regions of interest: a nuclear localization signal (NLS) and a region with homology to hydrolases.

Based on this network structure, we might expect Prf, which acts epistatically to Pto and Fen, to show a stronger signature of purifying selection, because mutations in this gene affect at least two phenotypes (signalling via Fen and Pto). However, the genes Pto, Fen and Rin4 may show a signature of relaxed constraint because they are less pleiotropic. Alternatively, these three genes may be subject to adaptive changes because they interact directly with pathogen molecules. However, as the precise locations of both Rin4 and Pfi in this network remain to be elucidated, definitive a priori predictions about the presence or absence of selective constraint operating at these genes are not possible.

For our study, we sampled multiple individuals of S. peruvianum from a single, large population in Tarapaca, Chile. S. peruvianum is endemic to the western coast of South America and is closely related to the cultivated tomato. This species is widespread and often occurs in large stands in central and southern Peru and northern Chile (reviewed in Chetelat et al., 2009). Individuals of this species are diploid, obligate outcrossing, short‐lived perennials, and census population sizes range from a few plants to several hundred (tgrc.ucdavis.edu). For this study, plants were grown under standard glasshouse conditions. DNA isolation, polymerase chain reaction (PCR) conditions and sequencing methods are described in Notes S1 (see Supporting Information). The entire coding regions of the five network genes were amplified and sequenced, yielding 20 fully resolved alleles for each gene (for a total of 14 167 base pairs). For comparisons between loci, the sequences of alleles of 14 other loci (Table 1) were obtained from Baudry et al. (2001) and Roselius et al. (2005). These reference genes were amplified from the same individuals of this S. peruvianum population.

Table 1.

Summary statistics across loci.

Gene	π * total (sites)†	π syn (sites)	π non (sites)	π non/π syn	Haplotype diversity	D ‡	R §	ZnS¶
Pto	0.01450 (928)	0.02038 (210.66)	0.01278 (710.34)	0.62	0.912	0.06383	0.0168	0.1698
Fen	0.00871 (966)	0.0156 (214.54)	0.00676 (745.46)	0.43	0.942	−0.55114	0.0406	0.1241
Prf	0.00667 (5541)	0.01386 (1206.59)	0.00448 (4211.41)	0.32	0.947	−0.27218	0.0117	0.1125
Pfi	0.01662 (5556)	0.02233 (494.17)	0.01277 (1707.83)	0.57	0.971	−0.59193	0.0108	0.1243
Rin4	0.00924 (1176)	0.01984 (221.34)	0.00320 (714.66)	0.16	0.947	−0.88121	0.0517	0.1113
CT066	0.00984 (1346)	0.03366 (332.10)	0.00204 (1011.90)	0.06	0.933	−0.30729	0.0306	0.1871
CT093	0.00568 (1388)	0.01763 (248.30)	0.00105 (780.70)	0.06	0.956	−0.14066	0.0591	0.1677
CT099	0.01827 (1198)	0.02138 (238.97)	0.00709 (649.03)	0.33	0.978	−0.81459	0.0408	0.1606
CT114	0.00820 (1165)	0.01605 (162.00)	0.00000 (510.00)	0.00	0.933	0.18940	0.0188	0.2535
CT143	0.01839 (1616)	0.01652 (113.00)	0.00000 (355.00)	0.00	0.978	0.25007	0.0141	0.2173
CT148	0.01433 (1394)	0.02010 (135.98)	0.00524 (407.02)	0.26	1.000	−1.05934	0.1020	0.1348
CT166	0.01429 (2602)	0.00699 (209.50)	0.00078 (732.50)	0.11	0.967	−0.03372	0.0198	0.2316
CT179	0.01069 (958)	0.03457 (153.00)	0.00000 (426.00)	0.00	0.911	−0.16307	0.0524	0.1980
CT189	0.01010 (1353)	0.00726 (82.55)	0.00000 (271.45)	0.00	1.000	−1.57754	0.0047	0.2295
CT198	0.02924 (760)	0.05648 (76.33)	0.00182 (256.67)	0.03	0.911	0.06862	0.0619	0.1958
CT208	0.00746 (1746)	0.00674 (168.10)	0.00000 (566.90)	0.00	0.778	−0.59745	0.0021	0.2866
CT251	0.01400 (1678)	0.03448 (317.78)	0.00721 (1005.22)	0.21	0.933	−0.24984	0.0125	0.2098
CT268	0.00941 (1884)	0.02587 (435.60)	0.00446 (1448.40)	0.17	1.000	−0.73472	0.0089	0.2027
sucr	0.01406 (1434)	0.02692 (267.43)	0.00401 (842.57)	0.15	1.000	−0.66941	0.1519	0.1358

^* Average pairwise differences.

^† Total number of sites analysed excluding gaps.

^‡ Tajima's D (Tajima, 1989); all sites analysed.

^§ Hudson (1987), all sites analysed.

^¶ Kelly (1997), all sites analysed.

Standard population summary statistics and tests of neutrality were conducted using DnaSP v. 4.0 (Rozas et al., 2003). Coalescent simulations were used to examine whether the patterns of substitution at nonsynonymous and synonymous sites at these genes differed from those of the 14 reference genes. We used the arithmetic mean of π (the average pairwise differences) of the reference genes as the estimate of θ (the population mutation parameter). A thousand simulations were executed in DnaSP and, subsequently, we determined whether the value of π observed at the network genes fell within the 95% confidence interval of the simulations based on θ estimated from the 14 reference genes. For these simulations, we assumed no recombination.

In our analysis of 20 alleles from S. peruvianum of these five genes, polymorphism, quantified by average pairwise differences across all sites (π), ranged from 0.6% (Prf) to 1.6% (Pfi) (Table 1). For comparison, the mean across the set of 14 reference genes from these same individuals was 1.3%. Pfi and Pto showed the highest polymorphism at synonymous sites (2.2% and 2.0%, respectively), as well as at nonsynonymous sites (1.3% at both loci). The ratio of π _non to π _syn was 0.57 for Pfi and 0.62 for Pto, whereas this ratio was consistently much lower at the 14 reference loci (mean π _non to π _syn= 0.10, Table 1).

We used neutral coalescent simulations to test whether the values of π observed at nonsynonymous and synonymous sites at the set of resistance genes differed from those of the 14 reference genes (Hudson, 1990). These simulations indicated that both Pfi and Pto show excess variation at nonsynonymous sites (P < 0.001), whereas, at synonymous sites, the observed level of variation is within the 95% confidence interval based on θ across these 14 other genes (Table 2). Previous studies of Pto alleles from this tomato population have revealed that a larger proportion of ancestral variation is maintained at Pto when compared with these reference loci (Rose et al., 2007). As the presence of elevated levels of amino acid polymorphisms is equally consistent with a hypothesis of relaxed constraint, we evaluated the functional consequences of these segregating amino acid variants. We found that the frequency spectrum of amino acid polymorphisms known to negatively affect Pto function is skewed towards low frequency compared with polymorphisms that do not affect function (Rose et al., 2005, 2007). This reveals that deleterious mutations are not at high frequency in this population, and that the evolution of Pto appears to be influenced by a mixture of both purifying and balancing selection.

Table 2.

Results of coalescent simulations.

Locus	π syn	Prob (π exp > π obs)†	π non	Prob (π exp > π obs)†
Reference genes‡	0.023		0.0024
Fen	0.016	0.668	0.0068	0.015
Pfi	0.022	0.413	0.013	<0.001
Prf	0.014	0.787	0.0045	0.085
Pto	0.020	0.390	0.013	<0.001
Rin4	0.020	0.520	0.0032	0.191

^† Probability of observing a value of π greater than that observed at network genes in 1000 coalescent simulations, conditioned on the π values of the reference genes.

^‡ Arithmetic mean of π from 14 genes (CT066, CT093, CT099, CT114, CT143, CT148, CT166, CT179, CT189, CT198, CT208, CT251, CT268 and sucr).

The evaluation of both polymorphism and divergence at Pfi using the McDonald–Kreitman test lent support to the hypothesis of balancing selection operating at Pfi (McDonald and Kreitman, 1991). Here, we detected significantly more variation at nonsynonymous sites (i.e. polymorphisms segregating in this population) than expected under neutrality (Table S1, see Supporting Information). A closer inspection of the distribution of variation across this large gene reveals that the region encoding the putative hydrolase motif and NLS harbours substantial amounts of nonsynonymous variation (π _non= 0.0216; Fig. 2a). Outside of this region, nonsynonymous variation is much lower: 0.00423. What is striking is that the region of elevated polymorphism is centred on the hydrolase region, but does not affect neighbouring coding regions. This could potentially be the target of natural selection driven by co‐evolution with pathogens, as discussed below.

Figure 2.

Sliding window analysis of average pairwise differences among Pfi alleles (a) and Prf alleles (b) in Solanum peruvianum. Silent polymorphisms include both synonymous polymorphisms and polymorphisms in nonprotein coding regions, such as introns. The structure of the genes is shown below the graph. Boxes indicate exons, solid lines indicate introns. The putative functional regions are indicated below the appropriate exons. bHLH, basic helix–loop–helix; CC, coiled‐coil; LRR, leucine‐rich repeat; NB‐ARC, nucleotide binding adaptor shared by APAF‐1, certain R proteins, and CED‐4; NLS, nuclear localization signal.

Of the five network genes, Fen and Prf showed the lowest levels of polymorphism, but intermediate values of π _non/π _syn. Fen, like Pto, is a small gene and encodes a functional protein kinase. Although Fen is known to interact with some pathogen ligands, no resistance function similar to that of Pto has been assigned to this gene. However, as this gene is expressed and retains its kinase activity across multiple wild tomato species, it is probably functionally important for tomatoes (Chang et al., 2002; Riely and Martin, 2001; Rosebrock et al., 2007). Selective constraint at this locus may explain the absence of a strong signature of either directional or balancing selection at Fen.

In contrast with Fen, Prf is a large gene, made up of both well‐defined and poorly defined domains. These domains show different evolutionary histories, as captured in the sliding window analyses (Fig. 2b). In contrast with many other R genes, the LRR region of Prf does not show an excess of amino acid polymorphism. Instead, the N‐terminal portion of the protein, known to bind Pto and Fen, has peaks of replacement site polymorphism. This elevated polymorphism provides hints that balancing selection at Pto may be carrying over to its interacting partner, Prf. These types of correlated selective history open the way to more complex forms of selection, such as epistatic selection between molecules. In comparison with the first half of Prf, the second half shows greater evolutionary constraint, consistent with its presumed role in downstream signalling.

The gene showing the greatest level of evolutionary constraint is Rin4. This gene has the lowest level of nonsynonymous polymorphism and the lowest π _non/π _syn of the five genes (Table 1). Indeed, based on the distribution and levels of polymorphism, Rin4 appears to be indistinguishable from the 14 reference loci. However, the frequency spectrum of mutations and the presence of a young, but divergent, allelic type, carrying several derived mutations, expose additional aspects of the Rin4 history (Fig. S1, see Supporting Information). Alleles 7232.1, 7233.1 and 7240.1 show nine fixed differences relative to the other alleles. All nine nonsingleton polymorphisms distributed throughout the Rin4 coding sequence are in significant linkage disequilibrium (LD). Two of these differences are nonsynonymous, whereas the others do not encode protein differences. These nine sites are derived relative to the alleles from the outgroup species, S. habrochaites and S. pennellii. Seven of these changes are derived relative to the more distant outgroup, S. lycopersicoides (Fig. S2, see Supporting Information). The absence of evidence of recombination between this sequence type and the others, the strong pattern of LD involving derived changes, two of which are nonsynonymous, and the low to moderate frequency of this sequence type are consistent with the presence of a partial or ongoing sweep at Rin4.

In our study of five genes involved in the Pto signalling network, we found two loci with elevated amino acid polymorphism, consistent with balancing selection, namely Pto and Pfi. A third gene, Prf, showed signatures of both balancing selection and purifying selection, whereas two other genes, namely Fen and Rin4, showed predominantly purifying selection. Previous studies have reported that Pto is subject to balancing selection within different wild tomato species and, given the substantial functional information available for Pto, a scenario of balancing selection is not surprising (Rose et al., 2005, 2007). Pto binds and recognizes two different pathogen ligands and triggers a defence response in wild tomato. The maintenance of different host resistance proteins in natural populations is consistent with an ongoing co‐adaptation between host and pathogen.

The second gene that showed elevated amino acid polymorphism relative to neutral expectations was Pfi. The protein product of Pfi physically interacts with Prf, has a putative NLS and is predicted to encode a transcription factor (Tai, 2004). As such, it may respond to an activated form of Prf by moving into the nucleus; however, the precise action of this protein remains to be elucidated. As a probable component of the signalling network, rather than a known pathogen target, it is surprising to uncover a signal of balancing selection at Pfi. The signature of balancing selection is located in a region that encodes a putative hydrolase, although enzymatic assays to confirm hydrolytic activity have yet to be conducted. Provided that this molecule is enzymatically active, it is possible that natural selection operates directly on the enzymatic function, and that protein variation is maintained in this region as a result of selection for different substrate specificities, perhaps involved in pathogen defence. Alternatively, this molecule could serve as a direct target by other tomato pathogens. Recent studies have revealed that all points in immune signalling networks can be vulnerable to pathogen manipulation (reviewed in Brodsky and Medzhitov, 2009). Pathogens may specifically secrete proteins (i.e. effector molecules) to target downstream points in the network to suppress host resistance (reviewed in Zhou and Chai, 2008). Pfi is described to be a negative regulator of defence; therefore, alteration of protein stability could result in suppression of the HR (Tai, 2004). In this case, balancing selection may not be specifically operating on enzymatic function, but rather on pathogen evasion. Alternative forms of Pfi found in these natural populations may vary in their ‘resistance’ to manipulation by pathogen molecules.

Prf, one of the central molecules of this network, showed two distinctive signals of natural selection. The region known to physically interact with Pto and Fen showed elevated amino acid polymorphism, providing the first hint that balancing selection at Pto may be carrying over to its interacting partner Prf. In comparison with the first half of Prf, the second half of this gene shows greater evolutionary constraint, consistent with its presumed role in downstream signalling.

The Fen and Rin4 genes showed the greatest evolutionary constraint of the five genes. Although Fen is known to interact with some pathogen ligands, no resistance function similar to that of Pto has been assigned to this gene. It is possible that Fen does not operate in the same isolate‐specific manner as Pto. If Fen is involved in basal defence and not in isolate‐specific defence, it would be subject to different evolutionary forces. One such molecule that is known to contribute to basal defence and is involved in different resistance pathways (at least in Arabidopsis thaliana) is Rin4. We observed strong protein conservation at Rin4 in our tomato population. However, the frequency spectrum of mutations and pattern of LD among Rin4 alleles expose additional aspects of the history of Rin4, including the presence of a young, but divergent, Rin4 allelic type, carrying several derived mutations. One possible explanation of this pattern is that this divergent Rin4 allele is passaging through the population as an advantageous allele. However, detecting a selective sweep in progress is quite unlikely because the sojourn times of advantageous alleles are generally too fast. The fact that this Rin4 allele with several derived changes is segregating with two other distinct alleles, all at moderate frequency, and none of these three allelic types shows any evidence of recombination, indicates that the frequency spectrum of these alleles has been perturbed in the recent history of this plant population.

Recent studies in Arabidopsis have evaluated a number of defence genes, some of which are known to operate together in specific signalling pathways (Bakker et al., 2006, 2008; Caldwell and Michelmore, 2009). At a broad scale, these results are consistent with expectations that genes downstream in defence pathways experience greater evolutionary constraint, and upstream genes are subject to adaptive change. However, a subset of these same genes was recently evaluated more extensively by another team, and they came to slightly different conclusions (Caldwell and Michelmore, 2009). In a study of 10 downstream defence genes in A. thaliana, three genes (NPR1, EDS1 and PAD4) showed interesting patterns of past adaptive evolution. This signature of balancing selection in these three genes may have been missed by Bakker et al. (2008) because, in the original study, only portions of the coding regions were analysed, rather than the entire genes. Interestingly, a fourth gene in the study by Caldwell and Michelmore (2009) overlapped with one in our study, namely Rin4. In their initial analyses, Rin4 was identified as a potential outlier based on Hudson–Kreitman–Agaude (HKA) tests. The HKA test can be used to evaluate whether the level of intraspecific polymorphism at the locus of interest is elevated relative to the neutral expectation, and therefore serves as a powerful means to identify genes affected by balancing selection (Hudson et al., 1987). Although the results from the HKA test for Rin4 were inconclusive following correction for multiple testing, the authors reported that Rin4 harbours substantial polymorphism within Arabidopsis, displaying more genetic variation than found at 93.5% of the genes in a set of 355 reference loci (Caldwell and Michelmore, 2009). To what degree this elevation in genetic diversity reflects past selective events has not been investigated.

Compared with these other studies, we did not find a strong correlation of selective constraint and network position. In a recent study of the genetic interactions in yeast, the authors reported that genes that have many interactions and serve as hubs are typically evolutionarily conserved (Costanzo et al., 2010). However, not all genes with many interactions evolve slowly. For example, genes that encode proteins with disordered regions have a higher rate of evolution, despite having numerous genetic interactions. Given the observation that the rate of evolution is not only a function of the number of genetic interactions, but also dependent on other features, such as protein structure, our current models of network evolution may be too simplistic and may not apply to defence pathways.

A second explanation for the lack of correlation between network position and selective constraint may reflect the biological reality that genes at several points in defence networks can be targets of adaptive evolution. Consequently, population genetic studies, such as ours, can uncover interesting candidates for future functional studies. For example, the consequences of Rin4 protein polymorphism on Pto‐specific resistance and possibly basal defence responses could be tested using the methods presented recently by Luo et al. (2009). Likewise, a better understanding of the functional consequences of protein polymorphism around the enzymatic core of the Pfi protein will probably reveal novel aspects of the defence repertoire of plants as, although this gene displays a signature of balancing polymorphism similar to other R genes in plants, it does not share the motifs of most other R genes.

Supporting information

Fig. S1 Maximum parsimony tree based nucleotide sequences of Rin4 alleles from S. peruvianum.

Fig. S2 Distribution of non‐singleton polymorphisms at the Rin4 gene among individuals of S. peruvianum.

Table S1 McDonald–Kreitman test on Pfi.

Notes S1 Experimental procedures.

Supporting info item

Supporting info item

Supporting info item