Historical Contingencies Modulate the Adaptability of Rice Yellow Mottle Virus

The rymv1-2 and rymv1-3 alleles of the RYMV1 resistance to Rice yellow mottle virus (RYMV), coded by an eIF(iso)4G1 gene, occur in a few cultivars of the Asiatic (Oryza sativa) and African (O. glaberrima) rice species, respectively. The most salient feature of the resistance breaking (RB) process is the converse genetic barrier to rymv1-2 and rymv1-3 resistance breakdown. This specificity is modulated by the amino acid (glutamic acid vs. threonine) at codon 49 of the Viral Protein genome-linked (VPg), a position which is adjacent to the virulence codons 48 and 52. Isolates with a glutamic acid (E) do not overcome rymv1-3 whereas those with a threonine (T) rarely overcome rymv1-2. We found that isolates with T49 had a strong selective advantage over isolates with E49 in O. glaberrima susceptible cultivars. This explains the fixation of the mutation T49 during RYMV evolution and accounts for the diversifying selection estimated at codon 49. Better adapted to O. glaberrima, isolates with T49 are also more prone than isolates with E49 to fix rymv1-3 RB mutations at codon 52 in resistant O. glaberrima cultivars. However, subsequent genetic constraints impaired the ability of isolates with T49 to fix rymv1-2 RB mutations at codons 48 and 52 in resistant O. sativa cultivars. The origin and role of the amino acid at codon 49 of the VPg exemplifies the importance of historical contingencies in the ability of RYMV to overcome RYMV1 resistance.


Introduction
Rice yellow mottle virus, of the genus Sobemovirus, causes a major disease in Africa [1]. The virus has a narrow host range restricted to the two cultivated rice species, Oryza sativa and O. glaberrima, and a few wild grasses [2]. The Asiatic rice O. sativa, which was introduced to East Africa and Madagascar in the 11 th century and to West Africa in the 15 th century, is now grown throughout Africa. In contrast, the African rice O. glaberrima, which was domesticated in the Niger Interior Delta ca. 3000 years ago, has been grown in West Africa only [3][4][5][6]. A few rice cultivars exhibit a high resistance to RYMV characterized by an absence of symptoms and no viral detection by ELISA [7]. The inheritance of this high resistance is recessive. The resistance gene RYMV1 was identified as an eIF(iso)4G gene [8]. Four rymv1 resistance alleles have been characterized, one in O. sativa (rymv1-2) and three in O. glaberrima cultivars (rymv1-3, rymv1-4, rymv1-5) [9]. The genetic determinants of the ability to break resistance alleles rymv1-2, rymv1-3 and rymv1-4 were investigated in this study.
The genetic determinants of RYMV1 resistance are located in the central domain of eIF(iso)4G1 (named MIF4G domain) [8,10]. The resistance allele rymv1-2 of the two O. sativa cultivars Gigante and Bekarosaka is due to a point mutation leading to the replacement of a glutamic acid by a lysine at codon 309 ( Figure 1a).
The resistance allele rymv1-3 of the cultivar Tog5681 and of a few other O. glaberrima cultivars [9,11] is caused by a deletion of codons 322-324 [8]. The rymv1-4 resistance allele of the cultivar Tog5672 and of a few other O. glaberrima cultivars is due to a point mutation resulting in the replacement of a glutamic acid by a lysine at codon 321 [8]. Interestingly, there is a single amino acid-specific difference between O. glaberrima and O. sativa at position 303 of the MIF4G domain, close to the genetic determinants of RYMV1 resistance. An alanine occurs in O. sativa whereas an aspartic acid exists in O. glaberrima (Figure 1a). This result was first established on a few sequences [8,9] and recently confirmed on a larger data set (C. Lirette and E. Hébrard, unpublished results).
Earlier studies established the features of the RYMV1 resistance breakdown (RB) process [12][13][14][15]. In most instances, the inoculation of resistant plants does not cause infection. Occasionally however, fully systemic symptoms are expressed, the virus content reaches that of susceptible plants, and the virus is readily transmitted from plant-to-plant: i.e. the resistance is overcome. Subsequently, wild-type (WT) isolates are displaced in the resistant plants by newly-adapted mutants with higher multiplication rates. Resistance-breaking process involves the Viral Protein genome linked (VPg), which is an intrinsically disordered protein [16]. The RB mutations occur in codons -referred to as virulence codons -in the central domain of the VPg between positions 37 and 52. The rymv1-2 resistance is most frequently overcome by the substitution of arginine (R) by glycine (G) at codon 48 and then of glycine by glutamic acid (E) at the same codon [13] (Figure 1b). Minor RB mutations occur at codons 48 (R48I and R48W), 42 (N42Y), 43 (T43A) and 52 (H52Y). The rymv1-3 resistance breakdown of the O. glaberrima cv. Tog5681 involves the replacement of serine (S) by proline (P) or alanine (A) at codon 41, and/or the replacement of histidine (H) by tyrosine (Y) at codon 52 [15]. Interestingly, H52Y is the only RB mutation that overcomes both rymv1-2 and rymv1-3 resistance alleles. Recently, the direct interaction between the MIF4G domain and the VPg was shown by co-purification and two-hybrid assays [17]. Mutations mimicking the resistance alleles in MIF4G strongly diminished the interaction with the VPg. Virulence mutations that become fixed in the VPg partially or fully restored the interaction.
One of the most salient feature of RYMV1 RB process is the converse genetic barrier to rymv1-2 and rymv1-3 resistance breakdown. Isolates that can break rymv1-2 never overcome rymv1-3 whereas isolates that can break rymv1-3 rarely overcome rymv1-2. This specificity is associated with the amino acid at codon 49 of the VPg, a position which is adjacent to the main virulence codons 48 and 52. Isolates with a glutamic acid (E) do not overcome rymv1-3 [15] whereas those with a threonine (T) rarely overcome rymv1-2 [13]. The changes at codon 49 are coordinated with those at the adjacent codons 48 and 50 (Figure 1c). Mutations R(AGA)48R(AGG), E(GAG)49T(ACG) and R(AGG)50K(AAG) are generally associated, R(AGA)48E49R50 being the ancestral motif and R(AGG)48T49K50 being the derived motif [13]. There are, however, a few isolates with the REK motif [13]. This stretch of four coordinated changes (one synonymous and three nonsynonymous) at codons 48-49-50 is a notable case of epistasis [18].
In this study, the identity, role and origin of the genetic determinants of the ability to break RYMV1 resistance alleles were investigated. First, we tested whether the amino acid at codon 49 was a key genetic determinant and not just a molecular 'signature' of the ability to overcome the RYMV1 resistance alleles. Second, we examined the effects of the amino acid at codon 49 on the fixation of RB mutations at the adjacent virulence codons 48 and 52. Third, we studied whether the threonine and glutamic acid at codon 49 resulted from past adaptation to O. glaberrima and O. sativa, respectively. This hypothesis arose from the matching geographical distribution of the amino acid polymorphism at codon 49 of the VPg of the virus and that of the two cultivated rice species in Africa. O. glaberrima cultivars and RYMV isolates with T49 occur in West Africa only, whereas O. sativa cultivars and RYMV isolates with E49 occur everywhere in Africa [3,15]. Moreover, codon 49 is under strong diversifying selection [13], which may reflect past adaptation to rice species. In conclusion, we  [9]. (B) RYMV1 resistance-breaking mutations in the VPg. Alignment of the 48-52 region of the central part of the VPg from the wild-type (WT) isolates with E49 and T49 and from the main rymv1-2 and rymv1-3 resistance breaking (RB) mutants; adapted from [13,15]

Author Summary
A fundamental question in virus evolution is whether emergence into a new host species/cultivar requires adaptation of the virus during the early stages of infection or whether it is largely via a chance transmission of a viral strain with the requisite characteristics. Our studies of the breakdown of RYMV1 resistance to Rice yellow mottle virus (RYMV) in rice show that emergence of a virus into a new host is a two-step process, where adaptive mutations can be fixed only by isolates with the necessary genetic determinants. We found the identity, effects and origin of these genetic determinants. Past host adaptation and subsequent genetic constraints, epitomized in a single amino acid, modulate the current ability of RYMV to overcome RYMV1 resistance through epistatic relationships. The importance of epistatic interactions in virus adaptation is increasingly acknowledged, but has seldom been studied within an historical context. The origin and role of the genetic determinants in RYMV1 resistance breakdown exemplify the importance of historical contingencies in the adaptability of a virus to a new host, a scenario at variance with the widely accepted co-evolution arm race model between pathogens and their hosts.
found that RYMV-rice relationships do not fit a simple coevolutionary arms race model. An alternative evolutionary scenario is proposed that emphasizes the role of historical contingencies in virus adaptation to a new host.

Results
The amino acid at codon 49 of the VPg is a key genetic determinant of the ability to break RYMV1 resistance The mutant CIa*49E was obtained from the infectious clone of the isolate CIa (T49) to establish the causal role of the amino acid at codon 49 in the ability to break RYMV1 resistance alleles. The RB ability of the isolate CIa (T49) and of the mutant CIa*49E were then compared. The isolate CIa and the mutant CIa*49E were inoculated at the same concentration to the rymv1-2 resistant O. sativa indica cv. Gigante and to the rymv1-3 resistant O. glaberrima cv. Tog5681. Forty-two days after the inoculation of the isolate CIa (T49), ca. 96% of the rymv1-3 resistant plants but only ca. 5% of the rymv1-2 resistant plants showed severe symptoms and reached a high virus content ( Table 1). The high ability of the isolate CIa (T49) to overcome rymv1-3 and its low ability to overcome rymv1-2 are typical of isolates with T49 [13,15]. The substitution of threonine by a glutamic acid at codon 49 of the mutant CIa*49E reversed these trends. The mutant CIa*49E fully lost its ability to overcome rymv1-3 whereas its ability to overcome rymv1-2 increased dramatically from 5 to 40%. Consequently, the resistance breaking pattern of the mutant CIa*49E became similar to that of isolates with E49. The link between the amino acid at codon 49 and the ability to break the rymv1-2 and rymv1-3 resistance alleles was highly significant (x 2 = 17.5, P,0.001 and x 2 = 95.3, P,0.001 for rymv1-2 and rymv1-3, respectively).
Not only did the substitution of threonine by glutamic acid at codon 49 in the mutant CIa*49E reverse the RYMV1 RB ability of isolate CIa (T49), but it also modified the mutational pathways. Sequencing the viral population after rymv1-2 resistance breakdown by the mutant CIa*49E revealed the presence of the *48G mutation (i.e. the first step of the major mutational pathway toward rymv1-2 resistance breakdown). Furthermore, minor rymv1-2 RB mutations *42Y and *43A were observed. These mutations are sometimes detected after rymv1-2 resistance breakdown by isolates with E49 but never by isolates with T49 [13,15]. Moreover, the rymv1-2 RB mutation *48W which sometimes emerged after the inoculation of the isolates with T49 [13,15] was not detected after the inoculation of the mutant CIa*49E. Considering both its resistance breaking ability and the kind of RB mutations fixed, the mutant CIa*49E responded like the isolates with E49, and no longer like the isolate CIa (T49) or any other isolates with T49. Overall, these results show that the amino acid at codon 49 is a key genetic determinant of the ability to overcome rymv1-2 and rymv1-3 resistance alleles.
Effect of a threonine at codon 49 on the fixation of rymv1-2 RB mutations at codon 48 It has been established that directed mutagenesis resulting in the replacement of arginine at codon 48 of the VPg by glycine within the infectious clone CIa (T49) was lethal to the virus [13]. It was then postulated that it was the cause of the inability of isolates with T49 to follow the main rymv1-2 RB mutational pathway R.G.E at codon 48. To test this hypothesis, the double mutant CIa*48G*49E was constructed by directed mutagenesis, and its infectivity was compared with that of the single mutant CIa*48G (T49). The viral RNAs obtained by in vitro transcription were inoculated to the susceptible O. sativa indica cv. IR64 and to the rymv1-2 resistant cv. Bekarosaka. Thirty days after inoculation, the single mutant CIa*48G (T49) was not detected by DAS-ELISA either in the susceptible or the resistant cultivar ( Figure 2), confirming that the combination of amino acids G48-T49 was unfit. By contrast, all plants -either resistant or susceptibleinoculated with the double mutant CIa*48G*49E expressed symptoms in non-inoculated leaves and reached a high virus content.
The infection of the rymv1-2 resistant plants by the mutant CIa*49E (R48) was followed. As mentioned above, the *48G mutation was fixed after the inoculation of the mutant CIa*49E (R48) to the rymv1-2 resistant plants. At 90 days post-inoculation (dpi), the mutant CIa*48G*49E was itself displaced by CIa*48E*49E ( Figure 3). Interestingly, synonymous differences at codon 48 R(AGG or AGA), G(GGG or GGA) and E(GAG or GAA) did not interfere with the ability of the mutant CIa*49E to follow the mutational pathway R.G.E. Reversion from CIa*48E*49E to CIa*48G*49E in susceptible rice cultivars has been reported earlier [14]. In the present experiment, reversion from CIa*48G*49E to CIa*48R*49E was observed in susceptible cultivars. Altogether, the substitution of threonine by glutamic acid  at codon 49 of the VPg of isolate CIa restored the viability of the mutant CIa*48G*49E and also its ability to adapt to rymv1-2 resistant cultivars or to revert to the avirulent form in susceptible hosts. The mutant CIa*49E responded like the isolates with E49 and no longer like the isolate CIa (T49) or any other isolates with T49. These results validated the claim that lack of fitness of genotypes with the combination G48-T49 is the cause of the inability of isolates with T49 to overcome rymv1-2 by the main mutational pathway R.G.E at codon 48. Collectively, these results demonstrate the critical effect of the amino acid at codon 49 on the ability to fix rymv1-2 RB mutations at the virulence codon 48.
Effect of a glutamic acid at codon 49 on the fixation of rymv1-3 RB mutations at codon 52 Mutation H52Y is unique in that it breaks both rymv1-2 and rymv1-3 resistance alleles. However, the fixation of mutation *52Y depends upon the amino acid at codon 49. In rymv1-2 resistancebreakdown, the mutation *52Y is only fixed after inoculation of isolates with E49 [13] whereas for the rymv1-3 resistant cv. Tog5681, the mutation *52Y is only fixed after the inoculation of isolates with T49 [15]. Several isolates with *52Y and T49 (CIa, Ma105 and Ma203) or E49 (Mg16) obtained earlier during passage experiments [13,15] were inoculated to rymv1-2 resistant O. sativa indica cvs. Gigante and Bekarosaka and to rymv1-3 resistant O. glaberrima cv. Tog5681. The results depended on the amino acid at codon 49. The isolates with T49 and *52Y were infectious in both rymv1-2 and rymv1-3 resistant plants, whereas the isolates with E49 and *52Y were infectious in rymv1-2 but not in rymv1-3 resistant plants ( Table 2).
The single mutant CIa*52Y (T49) and the double mutant CIa*49E*52Y were constructed to establish the role of the amino acid at codon 49 in this contrasted infectivity. The mutants were then inoculated at the same concentration to the rymv1-2 resistant cv. Bekarosaka and to the rymv1-3 resistant cv. Tog5681. The single mutant CIa*52Y (T49) was detected at a high concentration in both resistant cultivars. In contrast, the double mutant CIa*49E*52Y was infectious in cv. Bekarosaka but not in cv. Tog5681 (Table 2). Two nearly isogenic lines (NILs) were tested to understand the role of the rice genetic background on the relationship between codons 49 and 52 during the RYMV1 RB process. The viral accumulation of the single mutant CIa*52Y (T49) and that of the double mutant CIa*49E*52Y in each NIL and in the parental O. glaberrima resistant cv. Tog5681 was compared. At 30 dpi, the single mutant CIa*52Y (T49) genotype was detected at high levels in both rymv1-2 and rymv1-3 resistant plants (Figure 4a). In contrast, the double mutant CIa*49E*52Y was infectious in the rymv1-2 NIL, but not in the resistant O. glaberrima cv. Tog5681 or in the rymv1-3 NIL. Similarly, the isolate Mg16*52Y, an RB isolate with E49 of a different strain, was not infectious in rymv1-3 resistant cultivar cv. Tog5681 and NILs ( Figure 4b). Thus, the presence of a threonine at codon 49 is a prerequisite for fixation of RB mutation *52Y in rymv1-3 resistant plants. Conversely, isolates with E49 are unable to fix the rymv1-3 RB mutation *52Y.
It was earlier established that rymv1-2 RB mutants with E48 or Y52 were not able to overcome rymv1-4 although a similar substitution -glutamic acid into a lysine -causes the rymv1-2 and rymv1-4 resistance at positions 309 and 321, respectively [12]. In contrast, the response to RYMV inoculation of rymv1-3 and rymv1-4 resistance alleles, which both occur in O. glaberrima cultivars, showed marked similarities: (i) the single mutant CIa*52Y (T49) was infectious in cv. Tog5672, although at a lower rate than in the rymv1-3 resistant cv. Tog5681; (ii) only isolates with T49 were able to break the rymv1-4 resistance; (iii) after inoculation of isolate Ng25 (T49), the RB mutation S41P was fixed to rymv1-3 resistant cv. Tog5681 and rymv1-4 resistant cv. Tog5672 (data not shown). These similarities in the rymv1-4 and rymv1-3 in the RB process suggest that the breakdown of RYMV1 resistance is host-species dependent. Comparison of the response of the parental lines and NILs indicates that the origin of the contrasted responses between rymv1-2 and rymv1-3 occurs in the MIF4G domain rather than elsewhere in the genome. Position 303 is the most likely candidate as it is the only position that distinguishes O. sativa from O. glaberrima and it is located close to the genetic determinants of resistance.
Effect of the amino acid at codon 49 on the VPg/MIF4G interaction Yeast double-hybrid assays were used to investigate the effect of the amino acid at codon 49 on the VPg/MIF4G interaction. The Figure 3. Effect of the amino acid at codon 49 on the rymv1-2 mutational pathway. Wild-type (WT) isolates with E49 follow the two-step major mutational pathway R.G.E at codon 48, whereas WT isolates with T49, including isolate CIa, are unable to do so (adapted from [13]. Mutation *49E in the isolate CIa restored the ability of the mutant to follow the rymv1-2 major mutational pathway at codon 48, subsequently responding like isolates with E49. The dotted arrow indicates a reversion to the wild type. doi:10.1371/journal.ppat.1002482.g003 strength of the interaction between the MIF4G domain with the VPg of wild-type isolates (R48-H52), and of RB mutants *48I, *48G, *48E and *52Y with T49 or E49 was assessed. The mutated MIF4G domain mimicking the rymv1-2 resistance substantially reduced the interaction with the wild-type VPg, whereas the VPg mimicking RB mutations partly or fully restored the interaction ( Figure 5). These findings extended previous results with genotypes with T49 and RB mutations at codon 48 [17] to genotypes with E49 and RB mutations at codon 52. Overall, there was a marked effect of the amino acid at codon 49 on the strength of the interaction between the mutated MIF4G domain and the VPg of WT isolates and RB mutants. The interaction with the rymv1-2 mutated MIF4G domain was significantly lower when VPg was carrying T49 than that observed in VPg with *49E. This was true both for the WT isolate (R48) and for RB mutants *48I, *48G and *52Y (P,0.001, P,0.001, P,0.025, respectively). The interaction was also lower in VPg with *48E-T49 than with *48E-*49E, although the difference was not statistically significant. Lower affinity to the rymv1-2-MIF4G domain may contribute to the low propensity of isolates with T49 to overcome rymv1-2 resistance. In particular, it may explain the inability of isolates with T49 to fix the rymv1-2 RB mutation *52Y. Interaction between the VPg and the mutated MIF4G domain mimicking the rymv1-3 resistance was different. There was no significant influence of the amino-acid at codon 49 on the strength of the interaction between the WT or *52Y mutated VPg and the rymv1-3 mutated MIF4G (data not shown).

Role of the amino acid at codon 49 in the adaptation to African and Asiatic rice species
We tested whether the amino acid at codon 49 of the VPg in wild-type isolates conferred a selective advantage in the infection of susceptible O. glaberrima, O. sativa indica and O. sativa japonica cultivars. Isolates representative of the diversity of RYMV were selected (cf. Materials and Methods). Preliminary experiments showed that single accumulation of each isolate, evaluated by qRT-PCR, was not significantly different between susceptible O. sativa and O. glaberrima cultivars (data not shown). Several pairs of isolates with E49 and T49 were then co-inoculated at the same concentration to O. glaberrima, O. sativa indica and O. sativa japonica cultivars. The outcome of the competition was assessed by sequencing the VPg of the viral population at successive dates after inoculation. Each co-inoculation was replicated at least twice. Altogether, the results of 92 competitions were sequenced during this study.
The outcome of the co-inoculation of T49 and E49 isolates in susceptible cultivars depended upon the rice species (Table 3). In O. glaberrima cultivars, the results were always clear-cut. Whatever the cultivar tested or the pair of isolates inoculated, the isolates with E49 were rapidly excluded (i.e. within 28 days) and only the isolates with T49 were fixed. The relationship between the amino acid at codon 49 and the result of the competition was highly significant in O. glaberrima (x 2 = 36, P,0.001). In O. sativa indica cultivars, the exclusion of one of the two genotypes also occurred but more slowly (i.e. within 60 days). Either isolates with E49 or T49 were fixed and there was no significant relationship between the amino acid at codon 49 and the outcome of the competitions. In O. sativa japonica cultivars, isolates with E49 or T49 were fixed, but there was a preferential fixation of isolates with T49 (x 2 = 9.8, P = 0.02). However, coexistence of the two genotypes was sometimes observed until 100 dpi.
The isolate CIa (T49) and the mutant CIa*49E were inoculated at the same concentration to susceptible O. glaberrima, O. sativa indica and O. sativa japonica cultivars to further assess the effect of the amino acid at codon 49 on the outcome of the competitions. Preliminary experiments showed there was no significant difference in virus accumulation after single inoculations (evaluated by DAS-ELISA and qRT-PCR) between the O. sativa indica cv. IR64 and the O. glaberrima cv. Tog5673 ( Figure S1). The results of the competitions were consistent among the six replicates. In the O. glaberrima susceptible cv. Tog5673, the isolate CIa (T49) always displaced the single mutant CIa*49E. In the O. sativa indica and the O. sativa japonica susceptible cultivars, the single mutant CIa*49E displaced the isolate CIa (T49).

Discussion
The first objective of this study was to establish that the amino acid at codon 49 of the VPg was not only a molecular 'signature' but was a key genetic determinant of the ability to overcome rymv1-2 and rymv1-3 resistance alleles. This was shown by directed mutagenesis of an infectious clone. The replacement of threonine by glutamic acid at codon 49 reversed the ability of isolate CIa to overcome RYMV1 resistance alleles and modified the mutational pathways which were followed. Considering both its resistance breaking ability and the kind of RB mutations fixed, the mutant CIa*49E responded like the isolates with E49, and no longer like the isolate CIa (T49) or any other isolates with T49. This result validates the claim that the amino acid at codon 49 modulates the ability of RYMV isolates to overcome rymv1-2 and rymv1-3 resistances either by promoting or by restraining the fixation of RB mutations at virulence codons. Consistently, the virulence  [19]. Indeed, there is increasing phylogenetic and experimental evidence that epistatic relationships play a critical role in the adaptation of viruses (reviewed in [20]). This justifies our attempts to investigate the modes of action of the amino acid at codon 49 on the ability to break RYMV1 resistance.
The second objective of this study was to determine the effects of the amino acid at codon 49 in the main rymv1-2 and rymv1-3 virulence pathways. The lack of fixation of the rymv1-2 RB mutation *48G in genotypes with T49, and of the rymv1-3 RB mutation *52Y in genotypes with E49 involved different processes. Genotypes with E49-*52Y were not infectious in rymv1-3 resistant cultivars but were fit in susceptible and rymv1-2 resistant cultivars. In contrast, genotypes with *48G-T49 were unfit in both susceptible and resistant cultivars. Combination *48W-T49 [15] in the clone CIa was also unfit in both susceptible and resistant cultivars. Actually, the range of possible changes within this motif is restricted, as indicated by the coordinated changes associated with the E49T substitution at codons 48-49-50 during RYMV evolution. Double-hybrid tests revealed that a threonine at codon 49 reduced the affinity of the RB mutants with the MIF4G domain Figure 5. Two-hybrid assays of the interaction between rymv1-2 mutated MIF4G and VPg with T49 or E49. Cells were plated as serial dilutions (optical density at 600 nm of 5610-2, 5610-3 and 5610-4) on a control plate with (left panel) or without (middle panel) the histidine amino acid and the adenine base. Comparisons of yeast growth were estimated as percentages after quantifying spot intensities presented in the middle panel using ImageJ (fourth panel), with 100% control efficiency being assigned to the susceptible MIF4G (rymv1-1)/VPg (T49) binary interaction. Results were highly reproducible and are the means of four independent experiments. Significant differences in efficiency between isolates and mutants with T49 (grey bars) and E49 (white bars) at P = 0.05 after Student's test are indicated by a star. doi:10.1371/journal.ppat.1002482.g005 of rymv1-2. The lower affinity to the MIF4G domain of rymv1-2 possibly explain why mutation *52Y was not fixed after inoculation of isolates with T49. Altogether, the effects of the amino acid at codon 49 on RYMV1 resistance breakdown found in this study are complex, critical, and numerous, and yet unlikely to be exhaustive. Similarities in the response of rymv1-3 and rymv1-4 O. glaberrima resistant cultivars to RYMV despite the differences of their genetic determinants (deletion vs. point mutation) suggested that RYMV1 resistance breakdown was host-species dependent. Codon 49 of the VPg and codon 303 of MIF4G epitomized on the virus and plant sides, respectively, the effect of the genetic background on the outcome of the infection of RYMV1 resistant cultivars. This was exemplified by mutation *52Y which was fixed in isolates with E49 in rymv1-2 resistant O. sativa cultivars with A303. In constrast, mutation *52Y was fixed in isolates with T49 in rymv1-3 resistant O. glaberrima cultivars with D303. Consistently, mutation *52Y was also fixed after inoculation of isolate with T49 to the rymv1-4 resistant O. glaberrima cv. Tog5672 (A. Nicaise and A. Pinel-Galzi, unpublished results).
It was earlier stressed that the strategy of a RNA plant virus such as RYMV to gain virulence against host resistance showed striking parallels with HIV resistance against antiviral treatments [13]. The present finding supports this parallel. Epistatis within the 48-49-50 motif of the VPg is a major constraint to RYMV evolution by preventing key adaptive mutational pathways. Recently, it was also established that a major consequence of epistatis for HIV evolution is that many of the minimum-length mutational trajectories between the wild type and the mutant with highest fitness are selectively inaccessible [21].
The third objective of this study was to determine the origin of the amino acid polymorphism at codon 49. Altogether, a threonine at codon 49 conferred a strong selective advantage over glutamic acid in the O. glaberrima cultivars. Independent fixation events of T49 account for the diversifying selection estimated at codon 49 [13,15]. The selective advantage of isolates with T49 in O. glaberrima over isolates with E49 is a likely explanation of these fixations. This is one of the first experimental validation of the adaptive role of a codon predicted to be under positive selection in the host range expansion of a pathogen in a natural environment [22]. The selective advantage of isolates with T49 in O. glaberrima also provides an explanation of the link between the threonine at codon 49 and the ability to overcome rymv1-3 resistance in O. glaberrima. Isolates with T49, which are better adapted to O. glaberrima than isolates with E49, would be more prone to break O. glaberrima rymv1-3 resistant cultivars. In contrast, isolates with E49, which are less fit in O. glaberrima, would not be able to overcome rymv1-3 resistance. Consistently, the frequency of breakdown of an eIF4E resistance in pepper by mutations in the VPg of PVY was high in a susceptible genetic background and low when the same gene was introgressed in a partially resistant genetic background [23]. Results of the competitions between isolates with E49 and T49 in O. sativa were more variable than in O. glaberrima. The narrow genetic diversity of O. glaberrima [24][25][26] may explain the homogenous outcome of the competitions whereas the wide genetic diversity in O. sativa may account for the diversity in responses.
It has been proposed that a direct interaction between the eIF4E of pepper and the VPg of PVY drives the co-evolution between resistance and pathogenicity, leading to the diversifying selection of both genes [27,28]. This is considered to be the best example of plant-virus co-evolution [29]. It was of interest to see whether this evolutionary scenario was relevant to RYMV in rice. Superficially, the evolution of eIF(iso)4G in rice and of the VPg in RYMV presents features similar to such an ''arms race'' coevolution model. Under a co-evolution model similar to that between pepper and PVY, the resistance mutations in rice should have been fixed under the selection pressure imposed by RYMV. However, this hypothesis does not match what is known about the epidemiology of the virus. There is evidence that RYMV diversified only two centuries ago and that epidemics developed only in recent decades [30,31]. This short time-scale is inconsistent with the hypothesis of rice adaptation to RYMV. More likely, the RYMV1 resistance alleles resulted from past selection pressure imposed under a wider time-frame [30], possibly that of RYMVrelated sobemoviruses [9]. RYMV variation does not fit a simple rice-RYMV co-evolution scenario either. With PVY, the codons of the VPg under diversifying selection are the virulence codons involved in eIF4E resistance breakdown [32,33]. According to a co-evolution model, the virulence mutations were fixed under the selection pressure imposed by the resistance genes. With RYMV however, virulence codons of the VPg of wild type isolates are not variable, possibly because resistant cultivars have not yet been deployed widely in the fields. Most importantly, codon 49 -which is a under strong diversifying selection [13] -is not a virulence codon itself but is adjacent to virulence codons 48 and 52. Its indirect mode of action on resistance breakdown suggests an evolutionary scenario different from co-evolution.
In conclusion, our results -although superficially consistent with a co-evolution scenario -suggest a different historical relationship between RYMV and rice. On the host plant side, the RYMV1 resistance alleles were presumably not fixed under the selection pressure of RYMV. On the virus side, the current ability to overcome resistances in the two rice species is modulated by past host adaptations and subsequent genetics constraints which are epitomized by a single amino acid of the VPg. This genetic determinant of RYMV1 resistance breakdown exemplifies the importance of historical contingencies in the adaptability of a virus to a new host. glaberrima genes potentially involved in the viral cycle such as translation initiation factor genes were not introgressed in the susceptible O. sativa indica cultivar (data not shown). Then, differences between the rymv1-2 and rymv1-3 NILs were mostly restricted to the RYMV1 gene. The plants were kept in a growth chamber under 12 hours illumination at 120mE-2s-1 at 28uC and 90% humidity.

Viral isolates, mutagenesis of infectious clone and inoculation
Our experiments followed a two-step approach: (i) directed mutagenesis of the VPg of the infectious clone CIa (T49), (ii) comparison of a range of field isolates with E49 and T49. The mutants CIa*49E, CIa*48G, CIa*48G*49E, CIa*52Y (T49) and CIa*49E*52Y were constructed by directed mutagenesis of the infectious clone CIa (T49) with QuikChange Site-Directed Mutagenesis Kit (Stratagene) [13,15]. Transcription of mutated clones and inoculation in planta of viral RNAs was done as described elsewhere [34]. None of the results obtained suggested a loss of competitive fitness of the mutated infectious clone CIa*49E. Nine wild-type isolates, including isolates of the three clades with T49, were selected as representative of the diversity of the virus to be used in the experiments (Table S1).
Inoculum was prepared by grinding infected frozen leaves in 0.1 M phosphate buffer (0.1 g/ml, pH 7.2) and was rubbed on leaves of 14-day-old rice seedlings. Viral quantification was performed by q-RT-PCR as described previously [14]. For competition experiments, ca. 10 12 copies of each isolate were mixed and inoculated to the different cultivars. In parallel, the same amount of inoculum of each isolate was inoculated singly to each cultivar. To test the RB ability of CIa and CIa*49E genotypes, ca. 10 13 copies of each genotype were inoculated to resistant hosts. The accumulation of mutants CIa*48G, CIa*48G*49E, CIa*52Y and CIa*49E*52Y was compared after the inoculation of ca. 10 12 copies of each genotype per plant. All genotype sequences were checked before and after inoculation (see below).

Viral quantification, sequence analysis and two-hybrid assays
For each competition experiment in susceptible plants, expanded leaves of different plants were collected at 14, 21, 28, 56 and 100 dpi. The VPg was then sequenced by analyzing RT-PCR fragments amplified from total RNA purified by RNeasy Plant Mini Kit (QIagen) as previously described [35]. Dual peaks in the sequencing electropherograms at the same position were interpreted as reflecting the co-existence of the two genotypes in the viral population [13,36]. In constrast, a single peak at a site polymorphic between the two competing genotypes suggested that one genotype predominated. The ELISA tests were performed on the leaf samples as described previously [37].
Two hybrid assays were conducted with the Matchmaker GAL4 Two-Hybrid System 3 (Clontech). The yeast strain AH109 was cotransformed with either pGAD::MIF4G (rymv1-1) or pGAD:: MIF4G* (mimicking the rymv1-2 allele) and either pGBK::VPg (wild-type with R48-H52) or pGBK::VPg* (mimicking RB mutants *48I, *48G, *48E and *52Y) constructs. Doubletransformation of yeast cells and spotting on control or selective medium plates was performed as previously reported [17]. Yeast cells were grown at 30uC for 4 days. Growth intensity was monitored with ImageJ software [38]. The data were normalized to positive and negative controls and expressed as a percentage after quantifying spot intensities, assigning 100% efficiency to the susceptible rymv1-1 MIF4G/wild-type VPg interaction. At least four independent experiments were conducted. The results of the virus accumulation estimated by DAS-ELISA, and of the interaction efficiency evaluated in yeast two hybrid systems were analysed by ANOVA (Statistica software version 6.0). Table S1 Name, strain and isolation host of the isolates and mutants used, and accession numbers of their VPg. (XLS)