Distinct Viral Populations Differentiate and Evolve Independently in a Single Perennial Host Plant

The complex structure of virus populations has been the objectof intensive study in bacteria, animals, and plants for overa decade. While it is clear that tremendous genetic diversityis rapidly generated during viral replication, the distributionof this diversity within a single host remains an obscure areain this field of science. Among animal viruses, only Human immunodeficiencyvirus and Hepatitis C virus populations have recently been thoroughlyinvestigated at an intrahost level, where they are structuredas metapopulations, demonstrating that the host cannot be consideredsimply as a "bag" containing a homogeneous or unstructured swarmof mutant viral genomes. In plants, a few reports suggesteda possible heterogeneous distribution of virus variants at differentlocations within the host but provided no clues as to how thisheterogeneity is structured. Here, we report the most exhaustivestudy of the structure and evolution of a virus population everreported at the intrahost level through the analysis of a Prunustree infected by Plum pox virus for over 13 years followinga single inoculation event and by using analysis of molecularvariance at different hierarchical levels combined with nestedclade analysis. We demonstrate that, following systemic invasionof the host, the virus population differentiates into severaldistinct populations that are isolated in different branches,where they evolve independently through contiguous range expansionwhile colonizing newly formed organs. Moreover, we present anddiscuss evidence that the tree harbors a huge "bank" of viralclones, each isolated in one of the myriad leaves.

INTRODUCTION

Top

Introduction

Viruses are notable for their large census population size,short generation time, and high replication and mutation rates.In fact, because of the lack of proofreading activity associatedwith RNA-dependent RNA polymerase, RNA viruses have the highestmutation rates among living entities (8), with extremely highgenetic variability being generated rapidly within viral populations(7). Analysis of the genetic structure and evolution of populationsis a crucial area of general biology and, in the case of viruses,highly relevant to the development of strategies for diseaseor epidemic control. Over the last 20 years, interest in thegenetic structure and evolution of virus populations has emergedand developed tremendously, as evidenced by the ever-increasingnumber of papers published on this subject (for recent reviews,see references 18, 21, 28, and 39).

The concept of the viral population has been hotly debated aroundthe term quasispecies (6, 12, 22, 47), but it is undisputedthat an RNA virus population is a swarm of mutant genomes amongwhich complementation occurs to a variable extent for differentvital functions (16). The limit or frontier of virus populationsis difficult to determine and varies widely depending on thescientific questions addressed by various authors. This limitis often logically determined by the physical or geographicalbarriers that separate host populations. Infected pluricellularhosts have been, and are still, often considered as delineatingthe minimal virus population since no clear isolation is perceived(or known) among groups of viral genomes replicating in variouslocations within this host, and mixing can occur through thevascular system, resembling a panmictic situation. Consequently,in animals and most particularly in plants, the structure ofvirus populations is seldom described in detail at an intrahostlevel. When sequence analysis of a genome pool originating froma single host is reported, this pool is usually considered aunique population sample and is rarely regarded as a mix ofseveral possible bulk genomes originating from various organsand tissues (14, 19, 45).

The few data available in the literature at this intrahost levelsuggest a more complex pattern than the genuine panmictic situation.Remarkably, for Human immunodeficiency virus type 1 (HIV-1),it was shown that the assumption that populations are panmicticwithin a host is inconsistent with the observed data. Instead,the metapopulation concept, where a population is made up ofseveral discrete subpopulations that undergo turnover throughfrequent foundation and extinction events, could play a centralrole in the evolution of HIV-1 (17). Differentiation of subpopulationsin various organs of the same host has been further confirmedmore recently for HIV (24, 30, 33, 37) and has also been reportedfor other animal virus species such as Hepatitis C virus (1,9, 25) and TT virus (29).

In plants, available information is scarcer. Analysis of thegenetic composition of virus populations during systemic invasionof tobacco plants by Tobacco mosaic virus (36) and Cucumbermosaic virus (26) or wheat plants by Wheat streak mosaic virus(15, 20) has clearly revealed the existence of severe bottleneckswhen a virus population is colonizing new leaves, possibly inducingstrong genetic drift and differentiation within each leaf, relatedto the founder effect. Perennial plants are particularly wellsuited as models for the study of the structure of virus populations.Indeed, infections commonly persist for many years without killingthe host, and the viral population can potentially be analyzedafter a much longer within-host evolution than in annual plants.It is thus surprising that very few studies have investigatedthe structure of virus populations within perennial hosts. Previoushints, however, indicate that the composition of sequence variants(also designated haplotypes) may differ in various locationsof a single host tree for Apple stem grooving virus (27) andCitrus tristeza virus (10), although other data appear to becontradictory (11).

To further study the differentiation of several subpopulationsof plant viruses at the intrahost level and to elucidate themechanisms of such differentiation, we have decided to use aperennial plant model chronically infected by a potyvirus followinga unique inoculation event. The extensive analysis presentedin this report reveals a series of striking phenomena that havethus far been mostly overlooked in plant virology. Thirteenyears after the initial inoculation, we demonstrate that a highgenetic diversity has built up in the virus population, whichis structured by the architecture of the tree. Beyond the observationthat viral genetic diversity increases when moving up from old(trunk and then limb) to newly formed (branches and then leaves)organs, our data clearly demonstrate that several viral subpopulationsdifferentiate over the years as they become isolated in differentlimbs of the host tree and evolve independently by "contiguousrange expansion." In addition, we present evidence that eachindividual leaf is infected by a single haplotypic variant;thus, the tree harbors, through its myriad leaves, a huge collectionof mutant genomes comprising the complex sum of a virus population(s).

MATERIALS AND METHODS

Top

Materials and Methods

Virus and plant materials. The host-virus pair chosen in this study involves a single peachtree, Prunus persica cv. Alexandra, that has been infected withPlum pox virus (PPV) strain M since 1991. More detailed informationon the organization of the RNA genome of PPV is available inthe supplemental material (see Fig. S1 in the supplemental material).PPV strain M isolate 91-001, used as an inoculum, had previouslybeen maintained in pea plants by sap inoculation transfer. Inthe spring of 1991, aphids, previously starved for 2 h, wereplaced on an infected pea leaf for a 7-min acquisition accessperiod and immediately transferred onto a single leaf on theyoung Prunus tree (1 year old, single-shoot stage) for an overnightinoculation access period. Since then, the infected Prunus treehas been grown and kept in an insect-proof screened house toprevent any possibility of secondary infection.

Sampling protocol. In June 2004, 13 years after PPV inoculation, we carried outa large-scale stratified sampling on the infected Prunus tree.As schematized in Fig. 1, we collected the following sampleson the tree: (i) 3 main roots (bark only) and 18 young terminalroots (pieces of well-separated fibrous root); (ii) 12 barksamples from the trunk, evenly distributed around the peripheryof a virtual transverse section; (iii) 12 bark samples fromthe base of each of the five constitutive limbs, evenly distributedat the periphery of virtual transverse sections; (iv) 1 barksample from the base of each of 23 newly formed branches, witha maximum of 5 branches (when available) being sampled on eachof the five constitutive limbs; and (v) 10 whole-leaf samples(when available) from each of the 23 young branches. In total,230 whole leaves, 23 bark samples of young branches, 60 barksamples of limbs, 12 bark samples of the trunk, 3 bark samplesof main roots, and 18 young terminal roots were collected. Allsamples were stored at –80°C until use for immunocapturereverse transcription (RT)-PCR and single strand conformationpolymorphism (SSCP) analysis.

View larger version (27K):
[in this window]
[in a new window]
FIG. 1. Sample locations within the Prunus persica host. The architecture of the tree is drawn schematically, comprising terminal roots, main roots, trunk, constitutive limbs, yearling branches, and leaves. An extensive sampling was conducted according to the stratified pattern as follows: dots situated at the periphery of each of the upper branches indicate whole-leaf samples; the dot situated at the base of each of the upper branches indicates a branch bark sample; dotted circles illustrate virtual transverse sections, around which each dot represents one bark sample; and dots on the main and young terminal roots indicate bark and whole tissue samples, respectively. Groups of samples Br1 to Br23, Li1 to Li5, Tr, and Ro are hypothesized to contain distinct subpopulations (see text). Samples from the main and young terminal roots were grouped in a single Ro subpopulation.

cDNA synthesis and PCR amplification. Three genomic regions, designated A, B, and C, correspondingto sequences encoding the N terminus of HC-Pro (nucleotide positions1136 to 1371); a region overlapping the C terminus of P3, allof 6K1, and the N terminus of CI (positions 3425 to 3728); anda region overlapping the C terminus of Nib and the N terminusof CP (positions 8466 to 8891), respectively, were analyzed.The genome organization of PPV, summarizing the known functionsof its different genes and the positioning of regions A, B,and C, is given in the supplemental material (see Fig. S1 inthe supplemental material). Three pairs of primers (A, B, andC) were designed for performing RT-PCR for these three regionsof the PPV genome; primer sequences are also available in supplementalmaterial (see Table S2 in the supplemental material).

Plant extracts were prepared by grinding leaves, barks, andfragments of roots 1/20 (wt/vol) in phosphate-buffered saline(PBS)-Tween (PBS containing 0.05% [vol/vol] Tween 20 and 2%[wt/vol] polyvinylpyrrolidone). For each sample, 100 µlof supernatant from a low-speed centrifugation (12,000 x g,5 min) was incubated overnight at 4°C in tubes previouslycoated with PBS-Tween containing 1 µg ml^–1 polyclonalanti-PPV immunoglobulin G; the supernatant was then washed awaywith two rinses with the same buffer. RT was performed directlyin these immunocapture tubes using a reverse transcription kit(Promega, Madison, WI), according to the supplier's recommendations.One microliter of the cDNA-containing mixture was then usedfor further PCR amplification. The yields of the resulting RT-PCRproducts were systematically checked on agarose gels and thenstored at a concentration of 100 ng µl^–1.

SSCP analysis and sequencing. For SSCP analysis, 5 µl (0.5 µg) of each of theRT-PCR products was mixed with 12 µl of denaturing solution(95% formamide, 20 mM EDTA, pH 8.0, 0.05% bromophenol blue,and 0.05% xylene cyanol), heated for 10 min at 95°C, andimmediately cooled on ice. The DNA strands were separated bynondenaturing 8% polyacrylamide gel electrophoresis in Tris-borate-EDTAbuffer (89 mM Tris-borate, pH 8.0, 2 mM EDTA), with a constantcurrent of 28 mA per minigel, and the temperature was maintainedat 4°C. The duration of the electrophoresis run varied withthe size of the fragment analyzed and was 4, 5, and 6 h forregions A, B, and C, respectively. The gels were finally stainedwith a DNA silver staining kit (Amersham Biosciences, LittleChalfont, United Kingdom).

Tissue samples yielding different SSCP patterns in at leastone of the genomic regions analyzed were considered to be containingdifferent haplotypes. On the one hand, sequencing of all differenthaplotypes confirmed that they indeed differ by one or moremutations; on the other hand, sequencing of up to five samples(when available) containing identical haplotypes confirmed thatthey displayed strictly identical sequences.

Fragmentation of the data set for statistical analysis. Samples from each of the 23 newly formed branches (10 leaf samplesand one bark sample per branch, named Br1 to Br23), from eachof the five constitutive limbs (12 bark samples per limb, namedLi1 to Li5), from the trunk (12 bark samples, named Tr), andfrom the whole root system (named Ro) were a priori consideredto be containing discrete population units. Thus, in total,we primarily hypothesized that we were working with 30 putativelydistinct subpopulations positioned as shown in Fig. 1.

Descriptive statistics of genetic structure. Genetic differentiation between the 30 PPV subpopulations definedabove was estimated by the F statistic (46), using the Arlequinversion 2.000 program, and the significance was tested by bootstrapanalysis based on 10,000 replicates.

The potential effect of a limb that supports branches and leaveson the total variation of haplotype distribution was studiedby analysis of molecular variance (AMOVA) (13). This procedurewas performed by evaluating the type and frequency of haplotypesdetected in each limb (including associated branches and leaves)and by estimating the probability that their distribution wasrandom. This test was contrasted using a bootstrap analysiswith 10,000 repetitions. The statistic for differentiation betweengroups of subpopulations on different limbs was F_Ct, the significanceof which is similar to the corresponding percentage of variance.The second level of variation was between subpopulations withineach limb. This source of variation was tested by exchangingthe haplotypes among Br and Li subpopulations in each limb,and the statistic of differentiation was F_SC, the significanceof which is also evaluated as a percentage of the correspondingvariance. The third level of variation was within subpopulations,and the statistic of differentiation was F_ST, the significanceof which is also evaluated as a percentage of the correspondingvariance.

For neutrality testing, the program DnaSP (34) was used to studyTajima's D and Fu and Li's F* statistics.

Nested clade analysis. We used a nested clade analysis (NCA) to evaluate the associationbetween haplotypes and their geographical range. The NCA wasperformed according to the procedure described previously byTempleton et al. (43) and can be divided into a series of successivesteps implemented with the computer programs TCS version 1.13(2) and Geodis (32). First, a haplotype tree or network wasconstructed according to a statistical parsimony procedure (23,40). The same TCS program was used to reconstruct a nonrootedhaplotype genealogy and to identify the most ancestral haplotype.Second, the nesting procedure (3, 4, 40, 42, 44) was appliedto the haplotype network. The resulting nested clades are designatedby C-N, where C is the nesting level of the clade and N is thenumber assigned to a particular clade at a given nesting level(41). Once the cladogram had been converted into a nested seriesof linked clades, the geographical data were quantified viaa distance matrix (in centimeters) based on the tree (host)structure. Finally, the computation of different distance statistics(D_c, the geographical range of a particular clade, and D_n, thedistribution of a particular clade relative to its sister clades)was performed using Geodis (32) according to the proceduresdescribed previously by Templeton et al. (43).

In cases where clades were nonrandomly distributed, the biologicalcauses of the haplotype-geography association were interpretedusing the inference key previously proposed by Templeton etal. in 1995 (43) and updated recently (31). This inference keyaims to discriminate between distinct evolution patterns throughoutthe hierarchical clade levels. These patterns are designatedas restricted gene flow, past fragmentation, and range expansion.

RESULTS

Top

Results

To study the genetic structure of a plant virus population atthe intrahost level, we extensively sampled a Prunus tree infectedwith PPV strain M following a single inoculation event thatoccurred more than 13 years ago. The sampling was carefullydesigned to show the possible relationships between the distributionof virus variants, the nature of the tree organs, and/or thearchitecture of the tree (Fig. 1) (see Materials and Methodsfor details). Three genomic regions (A, B, and C) of the PPVgenome were analyzed by SSCP and DNA sequencing (sequences areavailable upon request), and each identified haplotype was characterizedby a combination of these three regions. Thirty-three distincthaplotypes were identified, indicating a large within-host variationfor PPV. Among the 333 samples analyzed, the frequency of the33 haplotypes was highly variable. Two haplotypes, H1 and H4,were by far the most frequent, being found in 39.9 and 30.9%of the samples, respectively. Thirteen additional haplotypeswere found in more than one sample, ranging from 2 to 20 samples,whereas 18 haplotypes were detected only once. Table 1 liststhe 33 haplotypes, their frequencies, and the origin of thesamples in which they were found. At this point, it should bementioned that no selection for or against any of the 33 haplotypeswas detected. Indeed, Tajima's D and Fu and Li's F statistics(see Materials and Methods) gave nonsignificant negative valuesnot only on the complete data set (–1.14749 and –0.94357,respectively; P < 0.1) but also on the five groups of samplescompared in the AMOVA analysis (data not shown) described below.Consequently, any heterogeneity of the haplotype distributionwithin this tree could not be attributed to local adaptation(selection) but likely was due to stochastic phenomena suchas genetic drift and/or founder effect.

View this table:
[in this window]
[in a new window]
TABLE 1. Distribution of the 33 haplotypes within samples composing the 23 putative subpopulations

In order to intuitively evaluate whether and how the viral geneticdiversity is structured within the host tree, and to elaborateworking hypotheses, we represented each haplotype as a thinrectangle composed of three distinct regions of different colors,illustrating variations in regions A, B, and C, and superimposedthe haplotypic distribution on the architecture of the hosttree (Fig. 2). This representation immediately suggests a numberof interesting hypotheses that can all be tested using the seriesof statistical tools described in Materials and Methods: (i)the haplotypic diversity is distributed heterogeneously andincreases when sampling is done towards the newly formed organsof the tree, going from the trunk to the major limbs and thento branches and leaves; (ii) even in comparable organs, thehigh genetic diversity of the PPV population appears to be structuredaccording to the tree architecture, with each supporting limbcontaining seemingly differentiated subpopulations; (iii) thedistribution of apparent lines of PPV haplotypes seems to matchthe architecture of the tree, suggesting that the virus populationhas expanded in a "one-way" fashion, concomitant with tree growthand the appearance of new organs and tissues, with potentiallyno or little return and mixing. A detailed description and thoroughstatistical testing of these hypotheses are presented below.

View larger version (41K):
[in this window]
[in a new window]
FIG. 2. Haplotype distribution superimposed on the schematic architecture of the tree. Each haplotype is represented as a thin rectangle composed of three distinct compartments illustrating the different sequences of regions A, B, and C. To facilitate the schematic representation of haplotypes, the first detected haplotype, H1, for regions A, B, and C is represented in green. When the sequence of a given genomic region varies, the color of the corresponding compartment changes. Blank haplotypes correspond to samples from which the immunocapture RT-PCR failed repeatedly. (1) Haplotypes were detected in the 18 young terminal root samples. (2) Haplotypes were detected in the three main root samples. (3) Haplotypes were detected in the 12 bark samples from the trunk. (4) Haplotypes were detected in the 60 bark samples from constitutive limbs. (5) Haplotypes were detected in the 23 bark samples from yearling branches. (6) Haplotypes were detected in the 230 leaf samples.

Distribution of haplotype variants is highly heterogeneous within the host tree. The first striking observation is that in all samples analyzedand for all three genomic regions (A, B, and C), the SSCP profilesclearly demonstrated the presence of a single detectable sequencevariant (data not shown). Despite the large genetic diversityfound in our experimental tree, we never detected a mixtureof haplotypes in any given sample. This surprising finding hasvery important implications that will be further discussed below.

To test for possible differentiation of PPV populations withinthis single host tree, the whole sample set was divided into30 putative subpopulations according to the architecture ofthe tree, with 23 representing subpopulations from each yearlingbranch (Br1 to Br23), 5 representing subpopulations from thebark of each major limb (Li1 to Li5), and 2 representing subpopulationsfrom the bark of the trunk and from the root system (Tr andRo, respectively). We estimated the haplotype (H) and nucleotide( ${pi}$ ) diversity within each of the 30 putative subpopulations (Table2). The values were highly variable and ranged from 0.0 to 0.800(±0.114) and from 0.0 to 0.0240 (±0.00054) forH and ${pi}$ , respectively. By comparing H and ${pi}$ obtained from bulksamples from the trunk, the limbs, the branches, and the leaves(Table 2), it appears that the diversity increases significantlywhen sampling is done upward in the host tree from the trunkto the leaves, from older to younger organs. Our sampling ofthe root system was only exploratory and not as extensive asthat of the rest of the tree. For this reason, the variabilityfound in young fibrous roots is not further considered in thedetailed analysis described below and will be discussed later.

View this table:
[in this window]
[in a new window]
TABLE 2. Haplotype and nucleotide diversity within PPV putative subpopulations

PPV subpopulations differentiate in various locations of the host tree. Genetic differentiation between the 30 putative subpopulationswas assessed using the F statistic. In this test, out of 435possible pairwise comparisons, 301 showed a significant difference(see Table S1 in the supplemental material), thus confirmingthat the haplotypic distribution is not uniform, even amongcomparable organs such as the 23 yearling branches (Br1 to Br23).Indeed, out of the 301 comparisons showing a significant differentiation,179 pairwise comparisons were between different yearling branches.

In order to evaluate whether the differentiation detected betweenthe subpopulation units is structured, and how it is structured,we performed an AMOVA test on the haplotype distribution atdifferent hierarchical levels: the major limbs, the yearlingbranches within each limb, and the leaves within each yearlingbranch. While the AMOVA results (Table 3) showed significantgenetic structure at the three hierarchical levels examined,the limb hierarchical level appears to explain the majorityof the variation (55.04%). Differences between yearling branchesof the same limb contributed to only 13.05% of the total variation.The remarkably large residual variation of 31.9% of the totalis due to the high polymorphism within the 23 yearling branches(discussed in detail below).

View this table:
[in this window]
[in a new window]
TABLE 3. AMOVA within the PPV sample set

PPV populations evolve by contiguous range expansion in the infected host tree. To further understand the structure and the evolutionary processof the PPV population within its host tree, we conducted anNCA on our data set (see Materials and Methods). When the haplotypephylogeny network was created, the Hudson estimator (23, 40)did not reject the hypothesis of parsimony for the entire dataset, meaning that all haplotype sequences are tightly relatedand can be included in the same network (Fig. 3).

View larger version (21K):
[in this window]
[in a new window]
FIG. 3. Maximum parsimony network and nested clades of the 33 PPV haplotypes. Haplotypes are named as described in Table 1, and each connecting line represents a single mutational step between two haplotypes. Missing (hypothetical) haplotypes are represented by a "0." Dotted-line rectangles enclose step 1 level clades and are designated by "1-n"; plain-line rectangles enclose step 2 level clades and are designated "2-n"; and the thick line separates step 3 level clades (3-1 and 3-2). Haplotype H1 has the highest outgroup probability, as indicated by the TCS program.

The nesting procedure also illustrated in Fig. 3 organized thehaplotype network into a total of 15 step 1 level clades, nestedin 6 step 2 level clades, which are themselves nested in 2 step3 level clades. While the high number (15) of step 1 level cladesfurther indicates a strong genetic differentiation within thePPV population, the fact that only two nesting levels can satisfactorilyorganize the network agrees with the close relatedness of allhaplotype sequences.

Clade 1-1 contains the majority of the samples (141/333) and,in particular, samples containing the haplotype H1 (133/333),which is proposed to be the root of the network.

We measured the geographical distance (in centimeters) betweenthe positions of all samples in the Prunus tree and confrontedthe haplotype network with the distance matrix. A significantnonrandom geographical association was found for 10 clades outof 20 clades tested. According to Templeton's inference key(31, 43), the inferred evolutionary process (Table 4), whichexplains the geographical association of clades, was contiguousrange expansion for clades 1-1, 1-4, 1-6, and higher clade 2-2and allopatric fragmentation for clade 1-7. However, for reasonsdiscussed below, we believe the case of clade 1-7 is to be consideredwith caution. Finally, no conclusive inference could be determinedfor clade 1-2 and higher clades 2-1, 2-5, 3-1, and 3-2.

View this table:
[in this window]
[in a new window]
TABLE 4. Evolution process of PPV populations within a single host tree

DISCUSSION

Top

Discussion

Our biological system, a Prunus persica tree infected with PPVfor 13 years, is a unique case of a controlled long-term plant/virusassociation. The fact that this tree has been protected fromany possible secondary virus entry since the initial inoculationevent makes it an excellent model for the basic study of thestructure and evolution of a virus population within a singlehost. The use of SSCP as the primary technique to detect geneticvariation in various locations of the host allowed the analysisof a large number of samples, thus providing the most thoroughanalysis of the genetic structure of a virus population everreported at the intrahost level.

The observation that a unique sequence variant is detected ineach of the very numerous samples is remarkably novel and hasconsiderable implications. For main roots, trunk, and constitutivelimbs, both the number of bark samples and the haplotypic diversityare low, and mixes could easily be overlooked under such conditions.In contrast, in leaves, the genetic diversity is very large,and up to 230 samples were analyzed. This situation was alsoconfirmed in unrelated experiments, where single-sequence PPVvariants were found in each of 450 additional whole-leaf samplesfrom experimentally inoculated Prunus trees (C. Jridi, unpublishedresults). Previous reports have demonstrated the existence ofstrong genetic bottlenecking during the invasion of leaves byplant virus populations (15, 20, 26, 36). However, these studiessuggested that more than one infectious unit was most ofteninitiating the within-leaf population. To make sure that thesurprising observation reported here was not an artifact ofour experimental SSCP conditions, we pooled leaf samples bypairs and verified that haplotype mixes were readily detectable(data not shown). Moreover, we conducted SSCP on variable ratiosof two PPV sequence variants in mixed solutions and observedthat a sequence variant present at a frequency of 0.1 is stilldetectable (data not shown), a value also reported previously(35). From the latter test, we concluded that each leaf containsone major sequence with no other variant reaching a frequencyof 0.1. Even in branches where two haplotypes are found witha high frequency among the 10 corresponding leaves (i.e., Br6,Br8, and Br11), no mix was ever detected within a single leaf.The only logical conclusion from these results is that the viruspopulation is submitted to biological cloning at most (if notall) leaf invasions, and the tree thus harbors a tremendousnumber of clones within the complex virus population, each individuallyisolated in a single leaf. In apparent contradiction, a previousstudy on PPV has reported mixes of sequence variants withinsingle leaves (5). In this study, healthy plants were coinoculatedwith two variants that were then colonizing "free-land" cellsand tissues at their own pace, obviously reaching some leavesconcomitantly. Interestingly, in an analysis of coinfectionat the cellular level, those same authors reported that singlecells were rarely (if ever) coinfected, thus suggesting thatPPV does not easily invade "occupied land," making conciliationwith our results possible. Indeed, under our experimental conditions,with the virus being present well before expansion of new leaves,we propose that the cells initiating buds might become rapidlyinfected by very few virus genomes (most often a single one)present in neighboring cells, subsequently precluding secondaryinfection by other variants. In this hypothesis, newly formedtissues are colonized by locally established haplotypes andnot by those circulating in the phloem, as one would expectfrom the general scheme that describes virus primary invasionof a healthy host (38).

H1 was the only haplotype detected in the bark of the trunk.Interestingly, the bark samples of the major limbs contain H1and three additional haplotypes, the bark samples of yearlingbranches contain the four haplotypes detected below plus twoadditional ones, and the leaves contain the 6 detected belowplus 27 new ones. This addition of new haplotypes when goingfrom old to young organs suggests that the tree reflects thechronology of the appearance of virus diversity. Consistently,the phylogeny network containing all haplotypes (Fig. 3) proposesthat H1 is the most probable root from which diversity has builtup slowly over the years. Although indirect, these data partiallycompensate for the fact that the original virus source plant,on which the aphids were fed prior to inoculation of the treeanalyzed here, was lost 13 years ago, preventing any possibilityof directly identifying the ancestral haplotype. H1 was alsofound in old main roots, whereas additional haplotypes appearedin young terminal roots, perhaps suggesting a similar phenomenonof PPV diversification downward with the growth of the rootsystem. The presence (together with H1) of some haplotypes ata low frequency (below 0.1) in the initial inoculum, and theirulterior increase in frequency due to genetic drift and/or foundereffect in samples of limbs, branches, and leaves, cannot beexcluded. However, because the virus has replicated for 13 yearsin the tree and because the haplotypic diversity increases incomparable younger tissues (see H and ${pi}$ in barks of trunk, barksof limbs, and barks of branches in Table 2), it is likely thatmost haplotypes were derived from H1, accumulating new mutationsafter the original inoculation event.

As mentioned in above, fragmentary data had previously hintedat a within-host differentiation of virus populations in perennialplants. AMOVA based on the breakdown of our data set into 30discrete subpopulations confirms the importance of this phenomenonfor PPV in a Prunus host. More innovatively, AMOVA demonstratesthat this differentiation is structured by the architectureof the tree, with subpopulations diverging most significantlyin different major limbs. From Fig. 2, one can immediately perceivethat different PPV lines become isolated in majors limbs, withhaplotypes that occur frequently in samples from one limb, withits associated branches and leaves, being totally absent insamples from another limb. To statistically sustain this hypothesis,we applied NCA to evaluate the association between haplotypeand geographical range within the host. First of all, the phylogenynetwork shows that haplotypes abundant in more than one majorlimb (H1 and H4) are likely the most ancestral, suggesting thatthey existed prior to limb development and corresponding PPVline isolation. Moreover, the inferred evolution process forclades 1-1, 1-4, 1-6, and 2-2 is contiguous range expansion,defined previously by Templeton (41) as a gradual, moving frontof expansion for a given population. The possibility of rangeexpansion for a virus population in a tree that is already systemicallyinfected is limited to the newly formed branches and leaves(and roots) appearing over time. This result is consistent withthe observation that PPV haplotypes seem to be unable to recolonizepreviously infected cells and confirms our conclusion that thevirus population is expanding in one direction, toward youngextremities as they develop, with no possibility of return andmixing via the vascular system in previously infected organsand tissues. We believe that this mechanism satisfactorily explainshow several lines become isolated in different parts of thetree. Because neutrality tests detected no selection actingon any haplotypes, it is also likely that upon isolation, thelines differentiate rapidly due to the drastic bottlenecks discussedabove and the associated genetic drift. Contiguous range expansionis the most likely inference, as it concerned 282 of the 333samples analyzed. Other clades gave inconclusive inferences,except clade 1-7, the evolutionary process of which was inferredas allopatric fragmentation. We believe that this latter caseis not reliable since clade 1-7 groups a very limited numberof samples.

Overall, the results presented in this report have importantimplications, as they challenge some views commonly adoptedamong plant virologists and beyond. While panmixy is being deniedfor within-host HIV populations, with the concept of the metapopulationapplying better, we demonstrate here that differentiation cango further in other virus species such as PPV, with severalpopulations becoming isolated and evolving independently foryears in a single host. PPV is a member of the genus Potyvirus,which alone accounts for nearly 25% of described plant virusspecies. It will be interesting to evaluate how the situationreported here applies to other potyviruses and to other virusgenera infecting annual or perennial plants. Finally, the factthat the leaves harbor a tremendous collection of clones fromthe virus population(s) is also a striking situation that hasbeen largely overlooked, with huge implications for both theoreticaland practical aspects of virus biology, epidemiology, and evolution;how these findings apply to other tree-infecting virus speciesappears an interesting prospect, as it is, in large majority,from the leaves that viruses are acquired during transmissionby aerial vectors.

ACKNOWLEDGMENTS

We thank Fabien Halkett and the group CaGeTE for critical readingof the manuscript.

This work was supported by the government of Tunisia, by theFrench ministry of Agriculture, and by a grant from DADP-INRA/RégionLanguedoc-Roussillon.

FOOTNOTES

* Corresponding author. Mailing address: UMR Biologie et Génétique des Interactions Plantes-Parasites, CIRAD-INRA-ENSAM, TA 41/K, Campus International de Baillarguet, 34398 Montpellier Cedex 05, France. Phone: 33 (0)4 99 62 48 04. Fax: 33 (0)4 99 62 48 22. E-mail: blanc{at}ensam.inra.fr.

Supplemental material for this article may be found at http://jvi.asm.org/.

REFERENCES

Top