Extensive Recombination among Human Immunodeficiency Virus Type 1 Quasispecies Makes an Important Contribution to Viral Diversity in Individual Patients

Although recombination during human immunodeficiency virus type1 (HIV-1) replication in vitro and in vivo has been documented,little information is available concerning the extent that recombinationcontributes to the diversity of HIV-1 quasispecies in the courseof infection in individual patents. To investigate the impactof recombination on viral diversity, we developed a techniquethat permits the isolation of contemporaneous clonal viral populationsresulting from single infectious events by plasma-derived viruses,thereby permitting the assessment of recombination throughoutthe viral genomes, including widely separated loci, from individualpatients. A comparison of the genomic sequences of clonal virusesfrom six patients, including patients failing treatment withantiretroviral therapy, demonstrated strong evidence for extensiverecombination. Recombination increased viral diversity throughtwo distinct mechanisms. First, evolutionary bottlenecks appearedto be restricted to minimal segments of the genome requiredto obtain selective advantage, thereby preserving diversityin adjacent regions. Second, recombination between adjacentgene segments appeared to generate diversity in both pol andenv genes. Thus, the shuffling of resistance mutations withinthe genes coding for the protease and reverse transcriptase,as well as recombination between these regions, could increasethe diversity of drug resistance genotypes. These findings demonstratethat recombination in HIV-1 contributes to the diversity ofviral quasispecies by restricting evolutionary bottlenecks togene segments and by generating novel genotypes in pol and env,supporting the idea that recombination may be critical to adaptiveevolution of HIV in the face of constantly moving selectivepressures, whether exerted by the immune system or antiretroviraltherapy.

INTRODUCTION

Top

Introduction

Human immunodeficiency virus type 1 (HIV-1) encapsidates twoRNA genomes in progeny viral particles (30, 41) and, duringthe ensuing replicative cycle, recombination between the twogenomes can occur (3, 8, 28, 35). To the extent that the twoRNA genomes are distinct, recombination generates viral DNAcoding for genetic information different from that of the twoparental strains (9, 12, 16, 17). Numerous studies performedin vitro have documented the occurrence of recombination duringHIV-1 replication (6, 27, 38, 39, 46, 47, 54). Although thesestudies have found that the rate of recombination can be influencedby the nature of the target cell and that certain genomic regionsmay be hot spots for recombination, there is general agreementthat recombination occurs frequently throughout the genome (2to 20 events/genome/replicative cycle) and has the capacityto rapidly shuffle genomic segments from parental viruses (21,27, 46, 54). Numerous viral strains derived through recombinationbetween distinct HIV-1 subtypes have been identified in infectedindividuals, indicating that recombination also occurs in vivo(26, 42, 49), and studies performed using limiting-dilutionPCR suggest that such events are occurring frequently (51; F.Maldarelli, M. Kearney, S. Palmer, M. Polis, J. Mican, R. Stephens,D. Rock, J. Mellors, and J. Coffin, Abstr. 12th Conf. RetrovirusesOpportunistic Infect., abstr. 667, 2005).

The advantages that accrue through recombination and the impactof recombination on HIV-1 pathogenesis in individual patientsare not clearly defined (45). The shuffling of polymorphismsfound in distinct viral quasispecies could play a role in generatingviral diversity (33, 45). The ability to maintain extensivediversity may be extremely important for viral pathogenesis,because it ensures the availability of viral quasispecies ableto escape changes in the selective pressures exerted by theimmune response or by antiretroviral therapy. Indeed, resultsobtained with other models support the idea that diversity generatedby recombination can be beneficial for adaptation to changingevolutionary pressures (15, 40).

In addition to generating diversity, viral recombination couldalso be useful in preserving existing diversity. In the courseof infection in an individual patient, diversity can be threatenedby evolutionary bottlenecks. Although reflective of an initialselective advantage, the emergence of a genetically homogeneouspopulation could prove disadvantageous, because the progenymight later become susceptible to elimination by an immune responsefocused against shared antigenic determinants. An evolutionarybottleneck resulting from the emergence of a unique viral specieswith high resistance to antiretroviral agents would pose thisrisk. If, however, during or subsequent to emergence, such strainscould recombine at high rates with preexisting strains, muchviral diversity could be maintained in regions outside thoseresponsible for the bottleneck.

Despite its potential importance, the role of recombinationin generating and preserving viral diversity in vivo has beendifficult to define for technical reasons. The analysis of sequencesobtained by amplification and cloning of viral genomes fromplasma cannot address this question, because recombination isknown to occur during amplification by PCR (23, 32), and individualsequences can be over- or underrepresented in such bulk amplificationproducts due to differences in the efficiency of amplificationof individual sequences with a given pair of primers. Recombinationduring amplification can be avoided by performing limiting-dilutionPCR, but this technically demanding approach also has certainliabilities. In particular, the possibility of recombinationoccurring during the reverse transcription reaction is difficultto exclude, some sequences may not be amplified using any givenpair of primers and therefore would be lost from the analysis,and the amplification of a single fragment containing widelyseparated genomic regions (e.g., pol and env) is not feasible.

To further evaluate the extent that recombination contributesto the generation and preservation of the diversity of HIV-1in the course of infection, we developed a technique that permitsthe isolation of contemporaneous clonal viral populations resultingfrom single infectious events by plasma-derived viruses andwhich therefore allows the assessment of recombination betweendistant regions of viral genomes from individual patients. Usingthis technique, we generated such clones from patients experiencingvirological treatment failure, and sought evidence that (i)high rates of recombination occur throughout the viral genomein vivo, (ii) the restriction of diversity resulting from evolutionarybottlenecks in regions coding for the enzymes targeted by antiretroviraldrugs need not extend to other genomic regions, and (iii) therearrangement between gene segments directly contributes togenetic diversity.

MATERIALS AND METHODS

Top

Materials and Methods

Patient population. Clonal viral populations were obtained from six HIV-1-infectedpatients; clinical information for these patients is summarizedin Table 1. Because we were interested in evaluating patientsin whom viruses may have been subjected to evolutionary bottlenecks,five of the patients we evaluated were failing treatment withantiretroviral agents. The treated patients had initially receivednucleoside analog reverse transcriptase (RT) inhibitors onlyand had developed resistance to these agents prior to treatmentwith regimens that included protease inhibitors and nonnucleosideRT inhibitors. A decrease in viral load of >50% resultingfrom antiretroviral treatment was not documented for any patient,despite the fact that viral load had been regularly monitoredover the interval of time the patients received treatment withprotease inhibitors.

View this table:
[in this window]
[in a new window]
TABLE 1. Clinical characteristics of patients at the time of study and summary of the number of clones obtained^a

Clonal viral populations. To obtain clonal viral populations, the following protocol wasused: MT4 cells expressing CCR5 and CXCR4 receptors (MT4-R5,obtained from O. Schwartz, Pasteur Institute), were maintainedin RPMI 1640 medium containing 10% fetal bovine serum, 100 U/mlpenicillin G, and 100 µg/ml streptomycin (complete medium).On the day of infection, cultures were resuspended at 2 x 10⁶cells/ml in complete medium containing 1% (vol/vol) dimethylsulfoxide, and 0.25-ml aliquots were distributed in 24-wellplates. Plasma was diluted (usually 1:4) in complete mediumcontaining (final concentration) 1% dimethyl sulfoxide and 2µg/ml DEAE-dextran, and 0.25 ml of this mixture was addedto each well. The plates were centrifuged (860 x g, 2 h, 22°C),and cultured for 4 h at 37°C to permit viral entry. Cellswere recovered and washed once, and 200-µl aliquots containing2 x 10⁴ cells were distributed into 96-well plates. The cultureswere maintained at 37°C in 5% CO₂ and were passaged witha 1:10 dilution every 7 days for up to 40 days. Cultures wereinspected frequently by light microscopy, and when patent cytopathicchanges were observed, the culture supernatant and the cellpellet from infected wells were recovered separately and frozenat –80°C. If viral replication was observed in >20%of the wells, the experiment was repeated after further dilutionof the plasma.

Sequencing of viral genomes. DNA was extracted from infected cell pellets using a QIAampViral DNA mini kit (QIAGEN, Valencia, CA) and used to amplifyfour segments of proviral DNA, two adjacent segments in gag-pol(total of ${cong}$ 1,300 bp) and two adjacent segments in env (totalof ${cong}$ 1,300 bp). For several patients, the testing of several setsof primers was required to identify pairs that amplified DNAfrom all clones of a given patient. In difficult cases, a fragmentsurrounding the area of interest was cloned and sequenced, permittingthe design of suitable primers for amplification of proviralDNA. Amplification products were directly sequenced in bothdirections using an ABI automated sequencing platform (AppliedBiosystems, Foster City, CA). All chromatograms were visuallyinspected to exclude the presence of sequences with ambiguousor polymorphic bases. Sequences were aligned using CLUSTAL X(version 1.81), and alignments in regions with insertions wereverified manually.

Detection of multiple infectious events. To evaluate the relationship between the proportion of infectedwells in limiting-dilution experiments and the proportion ofwells infected with more than one virus, a suspension containingapproximately equal numbers of two distinct clonal viruses (Aand B) was prepared, and MT4-R5 cells were infected with serialdilutions of this suspension. DNA was isolated from cell pelletsfrom infected wells, and a 670-bp fragment of RT and/or a 579-bpfragment of env were amplified by PCR. Amplified fragments werepurified and digested with PstI (RT) or HindIII (env), whichrecognized sites present only in the fragments from virus Aor virus B, respectively, thereby allowing the identificationof wells infected with one or both viruses. This approach couldreliably detect both sequences in mixtures in which the minoritysequence represented 10% of total viral DNA. Exploring the tworegions using DNA from the same infected well always gave concordantresults (data not shown).

Evaluation of the evolution of viral sequences during culture. To examine whether the viral sequence, as determined by bulksequencing, was subject to change during culture, MT4-R5 cellswere infected under limiting-dilution conditions with virusesobtained by transfection of HeLa cells with an infectious proviralmolecular clone derived from pNL4-3, but carrying protease resistancemutations M46I and I54V and V82A or I54V and V82A, combinationsof mutations known to severely impair viral replicative capacity(29). DNA from 36 infected wells was extracted, proviral DNAencompassing the protease region was amplified, and the bulkproducts were sequenced.

Assessment of recombination. Phylogenetic trees were obtained by maximum likelihood usingthe DNAML program in the Phylip package (version 3.6). The congruencebetween phylogenetic trees was evaluated using the incongruencelength difference (ILD) test as implemented in PAUP* (version4.0).

Rmin was determined using the RecMin software designed by Myersand Griffiths (34). The output of this program includes theposition of gene segments with obligate recombination events(in the absence of homoplasy), permitting an assessment of thedistribution of recombination events.

To estimate recombination rates, the Pairwise program in theLDhat package was used (31). This program is a modificationof that proposed by Hudson (18). It estimates recombinationrates by combining the coalescent likelihoods of pairwise comparisonsof segregating sites and has been extended to take into accountthe occurrence of recurrent mutations at a single site. In implementingthe test, the population mutation parameter ( ${theta}$ = 4N_eµ,where N_e is the effective population size and µ is theper-site mutation rate) was determined for each sample set (52),and confidence limits for ${theta}$ were determined by coalescent simulationusing DnaSP (48). Likelihood matrices were then constructedfor each sample using the estimated value of ${theta}$ for populationrecombination parameters ( ${rho}$ = 4N_er, where r is the per-site recombinationrate) ranging from 0 to 1,000. To assess confidence intervalsof ${rho}$ , values whose log probability was ±2 of the valuewith maximum probability were determined.

Assessment of nucleotide divergence. To evaluate nucleotide divergence between sequences for a givensegment of the viral genome, the number of synonymous substitutionsper potential synonymous site (dS) was calculated pairwise forall clones by using the modified Nei-Gojobori method (36). Inthis analysis, a transition/transversion ratio of 2.0 was used,the Jukes-Cantor correction was applied, and gaps, when present,were handled by pairwise deletion. The mean distance of eachclone relative to all other clones from the same patient andthe overall mean divergence were determined. We were interestedin comparing the nucleotide divergence of different regionsof the viral genome for clones from the same patient. It isknown, however, that a variety of factors, such as codon usage,amino acid composition, and RNA secondary structure, can influencedS, independent of differences in mutation rate (1). To controlfor such differences, a training set of 11 HIV-1 subtype B sequencesfrom the Los Alamos database was used to identify regions ingag, protease, RT, and env which met the following criteria:(i) all were coding regions, (ii) insertions and deletions didnot prevent unambiguous alignment, (iii) mean dS was similaramong the regions. The following regions were identified (positionscorrespond to the HXB2 reference strain): gag, 1963 to 2160;protease, 2253 to 2327 and 2412 to 2486; RT, 2865 to 3203; envelopeC1, 6324 to 6617; envelope C3, 7110 to 7379. The overall similarityin dS of these regions was confirmed (see Results) by analyzingall HIV-1 subtype B viruses in the Los Alamos alignment databasefor which sequences of all regions were available (n = 43).These regions were then evaluated for all clones from each patient.

Evaluation of viral tropism. The tropism of envelope sequences was evaluated using the algorithmdescribed by Jensen et al. (20). Variants with predicted X4tropism were detected for 3 of 6 patients (patient 1, 5 of 25clones; patient 3, 5 of 12 clones; patient 4, 1 of 19 clones).For a given patient, the sequence of the V3 region was identicalin the different clones. The tropism of representative clonesbearing these V3 sequences was evaluated by generating recombinantviruses expressing the envelopes (gp120 plus extracellular domainof gp41) and evaluating their ability to infect U373-CD4-CCR5cells and U373-CD4-CXCR4 cells in a single-cycle colorimetricassay. The recombinant viruses from patients 3 and 4 showedexclusive X4 tropism. The recombinant viruses from patient 1had low infectivity in this assay, and predicted tropism couldnot be confirmed.

Data presentation and statistical methods. All viral nucleotide positions correspond to the HXB2 referencestrain. Resistance mutations are as defined by the IAS-USA DrugResistance Mutations Group (22). Statistical comparisons ofdiversity (dS) were made using the Kruskal-Wallis test; posttestcomparisons, performed only if P was <0.05, were made usingthe Dunn test.

RESULTS

Top

Results

Isolation of clonal viruses from plasma of patients infected with HIV-1. To generate clonal viral populations resulting from single infectiousevents by plasma-derived viruses, we developed the novel limiting-dilutiontechnique described in Materials and Methods. For these studies,it was important that viruses were derived from single infectiousevents. Limiting the analysis to experiments where <20% ofwells were infected reduces the probability of there being wellsinfected with more than one virus. Nevertheless, in an experimentwhere 20 of 100 wells are infected, the Poisson distributionpredicts that 18 wells would be infected with a single virusand 2 wells would be infected with two viruses. Thus, it wasimportant to demonstrate that multiply infected wells were notoccurring more frequently than predicted by chance alone and,above all, that cells infected with more than a single viruscould be identified and excluded from analysis. To study thisquestion, clonal viruses derived from two different clinicalspecimens were mixed in equal proportions, cells were infectedwith dilutions of this mixture using our protocol, and infectedwells were evaluated for the presence of the two strains byrestriction analysis (Table 2). The proportion of infected wellsdecreased with increasing dilutions of the mixture. As expected,the number of wells infected with both strains was high whenthe proportion of infected wells was high, but decreased instrict accordance with the Poisson distribution as the percentageof infected wells decreased. Importantly, when viral DNA fromwells infected with both strains was amplified and sequenced,such doubly infected wells were easily identified in all cases(six of six wells) by the presence of ambiguous bases at allpositions where the sequences of the two clones were different,providing an easy means to exclude from analysis wells infectedwith more than one virus.

View this table:
[in this window]
[in a new window]
TABLE 2. The proportion of wells infected with single or multiple viruses depends on the proportion of infected wells, as predicted by the Poisson distribution^a

It was also important to demonstrate that the DNA sequencesfor the clonal viral populations, obtained by bulk sequencingof proviral DNA, did not evolve during culture in vitro. Totest this, viruses with a known sequence were produced by transfectionusing infectious proviral molecular clones derived from pNL4-3but carrying protease resistance mutations known to impair viralfitness. Clonal viral populations were generated using our technique,and the protease region was sequenced for 36 clones. Withoutexception, these sequences were identical to those of the originalplasmid (14,256 total bases). In particular, no evidence ofreversion of the resistance mutations was observed despite thestrong replicative disadvantage for these viruses in the absenceof drug.

Clonal viruses generated from plasma of HIV-1-infected patients. Clonal viral populations were generated from five chronicallyinfected patients who were failing antiretroviral therapy (Table1, patients 1 to 5), as well as one treatment-naïve patient(patient 6). For all plasma samples studied, the large majorityof infectious events were detected between 13 and 29 days afterinitiation of culture (81% of 165 total clonal viruses). Tocharacterize the genome of clonal viruses, DNA was extractedfrom infected cell pellets, and genomic regions of ${cong}$ 1,300 bpin gag-pol and env were amplified from proviral DNA and sequenced.For the 127 clonal viruses evaluated (25, 19, 12, 19, 34, and18 clones, respectively, for patients 1 to 6), base calls wereunambiguous at all positions sequenced. In the course of thestudies, proviral DNA from only two infected wells producedsequences in which ambiguous bases were identified, and thesesamples were excluded from analysis. Clonal viruses from a singlepatient grouped closely together in phylogenetic trees but werewidely separated from the viruses obtained from other patientsor laboratory strains, regardless of the region evaluated (datanot shown). A comparison of the results of population-basedgenotyping of plasma viruses (available for the protease andRT for four patients) and the sequences of the clonal virusesindicated that the clonal viruses were representative of thethose in the total population. For positions that were monomorphicin the plasma genotype, the consensus sequence of the clonalviruses was, without exception, identical (3,622 total bases).For the 62 positions that were polymorphic in the plasma genotype,the same polymorphisms were identified among the sequences fromclonal viruses in 59 of 62 cases. In contrast, many polymorphismsidentified in low proportions among the sequences of the clonalviruses were not detected by bulk genotyping of plasma viruses.

Evidence for recombination in contemporary viruses. Visual inspection of aligned sequences from contemporary clonalviruses revealed numerous examples consistent with recombination.To systematically evaluate the recombination between differentregions of the viral genomes, a phylogenetic method was employed(14). For each patient, phylogenetic trees corresponding todifferent parts of the genome were generated using the maximumlikelihood method, and the topologies were compared using theincongruence length difference test. Rejection of the null hypothesisindicates significant differences between the topologies andstrongly supports the occurrence of recombination. An exampleis shown in Fig. 1, in which sequences that clustered togetherin the phylogenetic tree of the envelope C2-V4 region are indicatedby different colors. In the phylogenetic tree for the envelopeC1-V2 region of the same viruses, some clones remain closelyassociated (for example, the pairs of sequences marked by anasterisk). More commonly, however, viruses that are clusteredin the C2-V4 tree are dispersed throughout the C1-V2 tree, andthe topologies of these two trees are strongly incongruent bythe ILD test (P < 0.001). Similarly, viruses that are clusteredin the C2-V4 tree are also dispersed in the tree comparing thegag-protease region of the same viruses (P was <0.001 usingthe ILD test). As shown in Table 3, phylogenetic trees for differentgenomic regions were usually incongruent (P < 0.05) whenviruses from all six patients were evaluated using this approach,a result compatible with extensive recombination between allgenomic regions evaluated. For several cases, the analysis wasrepeated after systematic removal of single sequences or evenpairs of sequences, but no such deletions resolved the incongruencebetween regions, indicating that incongruence was not determinedby only a small number of recombination events (data not shown).The ILD test showed significant incongruence between RT andprotease for only three of the six patients evaluated. It isnoteworthy that the three patients for whom this test was notsignificant for these regions are the three patients found tohave very low diversity in the protease region (see below),and limited diversity is known to impair the ability of theILD test to detect recombination (44).

View larger version (35K):
[in this window]
[in a new window]
FIG. 1. Phylogenetic trees corresponding to different parts of the genome of clonal viruses from a single patient are incongruent. Genomic regions corresponding to protease + RT (2253 to 3338), C1-V2 envelope (6221 to 6871), and C2-V4 envelope (6872 to 7567) for 25 clonal viruses obtained from patient 1 were sequenced, and phylogenetic trees were generated using the maximum-likelihood method. Clonal viruses that clustered together in the C2-V4 envelope tree are highlighted with the same color. Although certain clones remain clustered in the other trees (e.g., those marked with an asterisk in the middle panel), clonal viruses that cluster together in one tree are generally dispersed throughout the other trees. The trees were incongruent using the ILD test (P < 0.001 for all pairwise comparisons).

View this table:
[in this window]
[in a new window]
TABLE 3. Evaluation of the congruence between phylogenetic trees obtained using nucleotide sequences of clonal viruses from different regions of the viral genome^a

To calculate a lower bound for the minimum number of recombinationevents necessary to construct a history of the sample sequencesin the absence of homoplasy, (Rmin), a modification of Hudson'sfour-gamete test was used (34). Consistent with the idea thatrecombination had occurred frequently, Rmin was elevated forall patients, with values of 49, 24, 33, 28, 47, and 33, respectively,for patients 1 to 6. The positions of these putative recombinationevents were scattered throughout the portions of the HIV genomesequenced.

Estimations of the rates of mutation and recombination. To estimate the contribution of mutation and recombination tosequence diversity, the coalescent-based technique describedby McVean et al. (31) was used to determine the population recombinationparameter ( ${rho}$ = 4N_er), where r is the per-site recombination rateper generation and N_e is the effective population size. By comparisonwith the population mutation parameter ( ${theta}$ = 4N_eµ), whereµ is the per-site mutation rate per generation, the relativerate of recombination compared to mutation (r/µ) can bedetermined. As shown in Fig. 2, the population mutation parameterswere not strikingly different in a comparison of the six patients(median, 0.021; range, 0.018 to 0.032). Considerably greatervariation was seen in the population recombination parameters(median, 0.03; range, 0.002 to 0.11). Thus, the rates of recombinationrelative to the rates of mutation were 4.1, 1.8, 0.8, 0.5, 5.7,and 0.1, respectively, for the six patients. The recombinationparameter did not correlate with nucleotide diversity. No obviouscorrelations were observed between the clinical characteristicsof the patients and the population recombination parameter,although this parameter was considerably lower for the singleuntreated patient studied. More sequences were available fromthe two patients with the highest population recombination parameter(patients 1 and 5), but when analyses were repeated using randomlyselected subsets of these sequences similar in number to thoseavailable for the other patients, no significant effect on therecombination parameter was observed, and the recombinationparameter remained elevated (data not shown).

View larger version (22K):
[in this window]
[in a new window]
FIG. 2. Contribution of mutation and recombination to sequence diversity. For each patient, the population recombination parameter and the population mutation parameter were determined using the coalescent-based approach described by McVean et al. (31).

In an effort to conform to the assumption of neutral evolutionin this model, analyses in which only silent mutations wereconsidered were performed. After correction for the number ofsynonymous sites present in the sequence ( ${cong}$ 22%), the populationmutation parameters was somewhat higher than those obtainedwhen the entire nucleotide sequence was considered but remainedsimilar among the patients studied (median, 0.047; range, 0.033to 0.053). Restriction of the analysis to silent mutations hadonly a small effect on the population recombination parameters(median, 0.034; range, 0.008 to 0.123). In this analysis, therates of recombination relative to the rates of mutation were2.0, 0.7, 0.7, 1.0, 2.7, and 0.2, respectively, for the sixpatients. We conclude that the rate of recombination among contemporaryviruses is often high and can occur at a rate that approachesor surpasses the mutation rate.

Diversity in different genomic regions. To evaluate sequence divergence between clones from a givenpatient, the dS was determined for different regions in thegenome. In this analysis, the mean distance of each clone relativeto all other clones from the same patient and the overall meandivergence were determined. To permit comparisons between differentregions in the genome, this analysis was restricted to segmentsin the genome for which the extent of diversity due to silentmutations was shown to be similar when sequences of virusesinfecting 43 unrelated patients were evaluated using the sameapproach (Fig. 3A). As shown in Fig. 3B, the extent of diversitydue to the presence of synonymous substitutions could be quitedifferent in different portions of the genome for viruses fromthe same patient. When the patterns of sequence divergence indifferent portions of the genome for all the patients are compared,however, several common features emerge (Fig. 4). In particular,sequence diversity in the protease region was often lower thanthat observed in other regions, and it was strikingly low inthree cases (patients 1, 2, and 4). This pattern was not observedfor some patients, including the individual who had not receivedantiretroviral therapy (patient 6). As shown above, the diversityobserved in the protease region was not different from thatof other regions when the sequences from 43 heterologous virusesfrom the Los Alamos database were compared (Fig. 3A). Similarly,the mean pairwise diversity between the consensus sequencesfor the six patients was, as expected, considerably higher thanthat seen between clones from a single patient, but no significantdifference was found in a comparison of the different genomicregions (data not shown). These findings argue against the existenceof selective pressure operating on synonymous sites in the segmentof the protease that reduces diversity that was analyzed inthese studies. Because all the patients for whom reduced diversityin the protease region was observed had been treated with proteaseinhibitors, the results are compatible with the possibilitythat this region of the genome had passed through one or morebottlenecks associated with the emergence of viruses carryingresistance mutations. To further test this idea, a similar analysiswas performed using sequences in the protease region obtainedfrom the same patient prior to and during the evolution of proteaseresistance (patient C in reference 5). Phylogenetic analysissuggested the occurrence of bottlenecks associated with theemergence of viruses carrying the I54V and A71V mutations between3 and 23 months and the fixation of the I93L mutation between24 and 34 months. As shown in Fig. 5, considerable diversitywas observed in the protease region prior to and immediatelyfollowing treatment with protease inhibitors but, at later times,a significant decrease in the diversity in the protease regionwas observed. Taken together, these findings suggest that evolutionarybottlenecks can limit diversity in the protease region but thatsuch reductions need not extend to adjacent genomic regions.

View larger version (20K):
[in this window]
[in a new window]
FIG. 3. Sequence diversity in different portions of the viral genome. The dS was determined for selected regions in genes coding for the C-terminal portion of gag (gag), protease (prot), RT, envelope C1 (C1), and envelope C3 (C3) using the modified Nei-Gojobori method (36). Each point represents the average pairwise distance of each sequence to all other sequences analyzed, and the bar represents the overall mean divergence. Panel A shows the analysis of all HIV-1 subtype B viruses in the Los Alamos alignment database for which sequences of all regions were available (n = 43), and panel B shows results of the analysis of the 19 clonal viruses from patient 4. The small arrows in panel B indicate groups of clones whose gag sequences were similar at all silent positions but whose sequences differed from those of the other groups by 3 or 4 silent mutations.

View larger version (21K):
[in this window]
[in a new window]
FIG. 4. Patterns of sequence diversity in different portions of the viral genome observed for clonal viruses. For each patient, the dS was determined for selected regions in genes coding for the C-terminal portion of gag (gag), protease (prot), RT, envelope C1 (C1), and envelope C3 (C3) as described in the legend to Fig. 3. The result for each of the six patients is presented as the mean ± standard deviation of the average pairwise distance of each clonal virus compared to those of all other clonal viruses obtained from that patient.

View larger version (13K):
[in this window]
[in a new window]
FIG. 5. Selection of protease resistance mutations is associated with a loss of sequence diversity. The dS was determined for the region of the protease gene described in the legend to Fig. 3 by using sequences obtained at different intervals after initiation of treatment with protease inhibitors. Previously published phylogenetic studies indicated that this patient experienced bottlenecks associated with the emergence of viruses carrying the I54V and A71V mutations between 3 and 23 months and fixation of the I93L mutation between 24 and 34 months (5). Results are presented as means ± standard deviations of the average pairwise distances of each sequence compared to those of all other sequences obtained at the same time point.

Less striking reductions in diversity were also observed inRT segments in some patients, consistent with the prior occurrenceof bottlenecks in this region (Fig. 4). All the patients studiedhad developed resistance to RT inhibitors prior to receivingprotease inhibitors, and it is possible that the additionaltime since the occurrence of treatment-related bottlenecks hadbeen sufficient for more silent mutations to have accumulatedin the RT, thereby explaining the somewhat greater diversityseen in comparisons of RT and protease. Evidence for restrictionin diversity in portions of the env gene was also observed.Patients could show either significantly greater diversity inthe C1 envelope region than the C3 envelope region (patients4 and 5, P < 0.001), or greater diversity in the C3 regionthan in the C1 region (patient 3, P < 0.001).

Factors influencing apparent diversity between clones. Although dS is commonly used as a measure of "evolutionary distance"resulting from the accumulation of synonymous mutations, a secondfactor that influenced this parameter in our data was identified.In the example shown in Fig. 3B, the mean value of dS for thegag fragment analyzed was 0.03, suggesting moderate diversity.When the distribution of values for each clone relative to allother clones is examined, however, the values fall into threedistinct groups with one outlier (Fig. 3B, arrows). Within eachof these groups, the nucleotide sequence of the gag fragmentwas identical but differed from that of the other groups by3 or 4 silent mutations. Thus, the finding of a low value fordS was useful in identifying regions that were essentially homogeneousdue to prior bottlenecks, but apparent high diversity often,but not always, reflected situations in which a small numberof alternative sequences were used repeatedly by different clones(see below).

Recombination between gene segments contributes to the diversity of env. The findings presented above indicate that the diversity withinshort segments of the HIV-1 genome could be either variable,restricted to a relatively small number of alternative choices,or even essentially homogeneous. Nevertheless, considerablediversity could be maintained by splicing together these segmentsin novel combinations. To test this idea further, the aminoacid sequences of the various envelope domains for clones froma given patient were compared in order to determine the numberof distinct genotypes that were present in each domain and theextent to which clones that shared a similar genotype in onedomain did or did not also express similar genotypes in adjacentdomains. For this analysis, genotypes that differed by a singleamino acid change not identified in another clone were consideredto be similar.

The results obtained for patient 3 are shown in Fig. 6A. Threeclearly distinct genotypes were observed in the V3 domain, onegroup with exclusive X4 tropism and two distinct groups withR5 tropism, and the clones have been separated into groups onthis basis. The number of distinct sequences within a givendomain ranged from two (C2 domain) to nine (V1 domain). Forall domains, clones sharing similar genotypes were always present.Some tendency for coevolution of env domains was observed forX4 viruses, which shared closely related C3 domains (data notshown) and often used V4 domains not identified in R5 viruses(Fig. 6A). This association was not absolute, however, and oneX4 virus (clone 11) used a V4 domain also identified for bothof the R5 populations. In contrast, the same V1 or V2 domaincould be used by viruses expressing different V3 domains, andthe same V3 domain could be associated with a variety of differentV1, V2, and C2 domains. As a consequence of this recombinatorialdiversity, no two clones expressed the same combination of V1-to-V4segments.

View larger version (47K):
[in this window]
[in a new window]
FIG. 6. Recombination between envelope domains contributes to the diversity of env. At the top of each panel, the consensus amino acid sequences of the V1, V2, V3, and V4 regions of envelope for the 12 clonal viruses from patient 3 (panel A) and the 25 clonal viruses from patient 1 (panel B) are shown. Only amino acid changes different from the consensus sequence are shown for each clone. For each domain, sequences that are identical or that differ by a single amino acid substitution not identified in another sequence are highlighted with the same color. The tropism of the clonal viruses predicted by the algorithm of Jensen et al. is also indicated.

The results for patient 1 are shown in Fig. 6B. Four distinctgenotypes were present in the V3 domain, including one populationwith predicted X4 tropism. The number of distinct genotypeswithin a given domain ranged from 4 (V2 and V3 domains) to 13(V4 domain). For all domains, clones sharing the same genotypewere always present. The same V1, V2, or V4 segment could beused by viruses expressing different V3 domains. In this patient,no clear evidence of coevolution between V3 domains and othervariable domains was observed for clones with putative X4 tropism.Among the 25 clones evaluated, only three (clones 5, 9, and10) used the same combination of V1-to-V4 segments. When variabilityin the C1, C2, C3, and N-terminal portion of C4 was includedin this analysis, none of the 25 clones used the same combinationof envelope domains.

The results for the other four patients were generally similarto those presented in Fig. 6. The number of distinct genotypeswithin a given domain ranged from 1 (V3 domain, patients 2,5, and 6; V1 domain, patient 2) to 12 (C2 domain, patient 5).The proportion of clones using the same combination of envelopesegments was 0/19 (patient 2), 0/18 (patient 4), 3/35 (patient5), and 4/18 (patient 6).

Some, but not all, of the apparent diversity within a givendomain appeared to result from recombination events. For example,the V4 genotype of clone 7 from patient 3, although scored as"distinct," could be produced by recombination between clone1 and clone 10. Many other examples were present and were observedin all domains except for the V3 domain. The V3 domain is shortand often showed the least diversity; both of these factorswould reduce the likelihood of identifying recombination events.Nevertheless, it is possible that recombination events thatchange the amino acid sequence in this domain often have a negativeimpact on viral infectivity and tend to be underrepresentedwhen infectious viral clones are evaluated.

It is also noteworthy that some clones used a sequence for onedomain that was markedly different from that of the other virusesfrom that patient (e.g., V1 domain in clone 7 from patient 3,and V4 segment in clone 1 from patient 1), but other domainsof these viruses were identical to those of other viral clones,suggesting that these "outlier" segments may have been derivedfrom recombination events with minority or archived viral populations.

Drug resistance genotypes and recombination. As shown above, diversity, as measured by the accumulation ofsilent mutations, was often limited in the protease and couldalso be low in the RT. Consistent with this finding, only limiteddiversity in the resistance genotypes was observed. For threetreated patients (patients 1, 2, and 4), diversity was restrictedto the gain or loss of one or two resistance mutations by singleclones. For example, for the 25 clones of patient 1, one clonehad gained the F53L mutation in protease, while other cloneshad lost either the V118I or L100I and K103N mutations in RT.Otherwise, all clones expressed the same 8 protease resistancemutations and the same 11 RT resistance mutations.

For two of the patients studied, two resistance mutations inprotease and two resistance mutations in RT were expressed byonly a subset of clones. For these four pairs of mutations,some clonal viruses expressed each mutation individually, whileothers expressed both mutations simultaneously (data not shown).This pattern cannot be explained by the sequential appearanceof the mutations alone, and either recombination or the occurrenceof more than one mutation or reversion at a single site (homoplasy)must be invoked. Considerable shuffling of protease and RT resistancegenotypes was also observed for these two patients. Proteasegenotypes represented by more than a single clone were usuallyfound in association with several different RT genotypes. Thus,when the protease and RT regions were considered together, themost abundant resistance genotype was expressed by only 25%(patient 3) and 38% (patient 5) of total clonal viruses. Clonalviruses isolated from patient 3 had both R5 and X4 tropism.Clonal viruses in both populations could express the same resistancegenotype, and several different genotypes were identified amongX4 and R5 viruses. Thus, tropism and resistance genotype didnot cosegregate in this patient.

DISCUSSION

Top

Discussion

We generated clonal viral populations resulting from singleinfectious events by viruses present in a single specimen ofplasma from HIV-1-infected individuals. A comparison of thegenomic sequences of these clonal viruses from each of six patientssupports the idea that recombination events strongly influenceviral diversity during the evolution of infection in a givenpatient. Our results demonstrate that (i) the overall frequencyof recombination is very high; (ii) when bottlenecks occur withina selected region of the genome, recombination may allow themaintenance of diversity in adjacent regions; (iii) shufflingof resistance mutations within the protease and RT and betweenthese regions can generate diversity in resistance genotypes;(iv) recombination between different gene segments also makesan important contribution to diversity in the env gene; and(v) recombination between viruses with distinct tropism is clearlyoccurring.

The analysis of clonal viruses proved to be a useful methodto evaluate recombination among contemporaneous viruses frominfected patients. Control experiments indicated that underlimiting-dilution conditions, most wells had been infected witha single virus, and the wells in which more than a single infectiousevent had occurred could be identified and excluded after sequenceanalysis. In studies evaluating the replication of viruses carryingprotease resistance mutations that impaired viral replicativecapacity, no evidence for the reversion of mutations or theappearance of compensatory mutations was found, suggesting thatthe time in culture necessary to produce viral clones was notsufficient for variants occurring in vitro to emerge to theextent that their presence was perceptible in bulk amplificationproducts obtained by amplifying proviral DNA. The sequencingof proviral DNA from infected wells in the course of these studiesfurther supported these conclusions. In only 2 of the 129 wellsevaluated were ambiguous bases identified in any of the genomicregions that were sequenced. The low frequency of sequencescontaining ambiguous signals further supports the idea thatmost wells had been infected with a single virus and that theevolution of viral sequences during culture had little or noimpact on the results. By using this approach, any segment ofthe genome is accessible to study, and evidence of recombinationor linkage between distant regions can be evaluated.

Our results confirmed studies performed using limiting-dilutionPCR that contemporaneous viruses show evidence of extensiverecombination. For the six patients studied here, estimatesof the rate of recombination relative to the rate of mutationobtained using a coalescent approach ranged from 0.1 to 5.7,and the recombination rate exceeded the mutation rate for threeindividuals. Shriner et al., analyzing a different genomic regionusing sequences obtained by limiting-dilution PCR from a singlepatient, found that the recombination rate was 5.3-fold higherthan the mutation rate (51). The method used in these studiesassumes neutral evolution, and extensive homoplasy could resultin an overestimation of the recombination rate. Analyses restrictedto the evaluation of synonymous mutations, however, gave similarresults, supporting the conclusion that this bias did not stronglyinfluence the results. Further analysis of our clones suggestedthat recombination could have an impact on viral diversity inseveral distinct but important ways.

Recombination allows regeneration of viral diversity following bottlenecks. The extent of diversity in different portions of the genome,as measured by the accumulation of silent mutations, could bequite variable. In particular, low levels of diversity wereoften encountered for the protease region in patients receivingantiretroviral treatment. The presence of such bottlenecks isconsistent with the idea that during the evolution of drug resistance,the occurrence of specific combinations of resistance mutationsthat give a decisive fitness advantage is followed by strongselection of viruses carrying this genotype. Previous studiesevaluating the impact of treatment-induced bottlenecks occurringin the protease or RT on diversity in other genomic regionshave given varying results, with evidence of reduced diversity,transient reductions in diversity, or no effects having beenobserved (4, 7, 10, 13, 19, 24, 37, 53). Our results suggestthat because of the extensive recombination between viral quasispecies,bottlenecks in one region did not preclude the preservationof considerable diversity in other genomic regions. It shouldbe emphasized, however, that none of the patients we evaluatedhad marked reductions in viral load resulting from treatment.Thus, during the time necessary for selection of an optimalprotease sequence, ample opportunity was available for recombinationbetween other loci. In contrast, patients in whom reductionsin env diversity has been noted have usually demonstrated markedreductions in viral load prior to the emergence of resistantstrains (10, 19, 24, 37), thereby limiting potential partnersfor recombination. We also observed evidence that viruses hadrecovered env segments from highly divergent strains (possiblyfrom archived viruses or minority populations replicating inprotected sites) through recombination, suggesting a mechanismthrough which env diversity could be reconstituted, even incases where transient bottlenecks had occurred. Findings consistentwith this scenario have recently been presented (24). Thus,our results suggest that recombination can help restrict evolutionarybottlenecks to minimal segments of the genome carrying the selectiveadvantage while preserving diversity in adjacent regions.

Recombination contributes to the generation of diversity. Our results also emphasize that recombination plays an importantrole in the generation of diversity. When individual envelopedomains were evaluated, we found that the diversity of a givendomain was often limited to a relatively small number of distinctgenotypes. For example, no more than four distinct V3 loop sequenceswere detected for any patient; for several patients, only asingle V3 loop sequence was identified. Most patients studiedhere were known to have been infected with HIV-1 for more than10 years and therefore had probably passed the time when peakdiversity in envelope is seen (50). Recombination between envelopedomains, however, was extensive. Thus, even when only limiteddiversity was observed in individual domains, envelope sequences,when considered in their entirety, showed considerably greaterdiversity. When all variable and constant envelope regions wereincluded in the analysis, each clone from patients 1 to 4 expresseda unique combination of envelope domains and, although clonesusing the same combination of envelope domains were observedfor patients 5 and 6, these populations represented only 3 of34 and 4 of 18 total clones, respectively.

Similarly, recombination contributed to the generation of diversityin resistance genotypes, as has previously been observed invitro (33). Viruses from a given treated patient expressed incommon most of the protease and RT resistance mutations identified,consistent with the existence of prior bottlenecks in theseregions. In most patients, however, some positions associatedwith resistance were polymorphic. In cases where such polymorphicsites were expressed by more than a single clone, considerableshuffling of mutations, both within the protease or RT and betweenthese regions, was also observed. Thus, patient 5 had two polymorphicsites in RT and two in protease, but nine distinct resistancegenotypes were observed, and the most common genotype was expressedby only 38% of the viral clones studied.

It should be emphasized that our analysis was restricted toviral clones that were infectious for the MT4-R5 target cells.As previously described by others (11, 25, 43), we found thatthe proportion of viruses in plasma that gave rise to infectiousevents was extremely low, varying from 0.002 to 0.016% for thesix patients studied here. Both viruses using CCR5 and CXCR4coreceptors were identified, suggesting that tropism did notrestrict infectivity in MT4 cells. Numerous other factors maycontribute to the apparent low infectivity of plasma viruses,including the presence of lethal mutations in the genome, structuraldefects in the particles, neutralization by antibodies, andloss of infectivity in the interval between venipuncture andinfection of target cells. Viral recombination may also havea deleterious effect on infectivity of plasma viruses by associatingincompatible gene segments in individual virions; given thehigh rates of recombination observed in this study, this factormay play an important role. In this regard, it is noteworthythat among the infectious clonal viruses studied here, no exampleof recombination in the envelope V3 loop that modified the aminoacid sequence of this region was identified, whereas apparentrecombination within other envelope segments was frequent. Studiescomparing the envelope sequences of infectious clonal virusesand sequences derived by limiting-dilution PCR from the samesample of plasma coupled with evaluation of the functional propertiesof the corresponding gene segments would help define the extentthat recombination may impair the infectivity of viruses invivo.

The reasons for the prevalence of recombinogenic "sexual" reproductionin biology, a characteristic shared by HIV-1, have not beendefined. Currently, two general models are often invoked: recombinationcould increase the rate of adaptive evolution, and/or improvethe efficiency of eliminating the effects damaging mutations(15, 40). Recently, an evaluation of the impact of combinationsof drug resistance mutations on fitness in the absence of drugpressure suggested that epistasis between individually deleteriousmutations in HIV-1 is positive (antagonistic) (2). Althoughthe epistasis between beneficial mutations was not examined,these findings do not offer support for the second alternative.Our findings demonstrate that recombination in HIV-1 makes animportant contribution to the diversity of viral quasispecies,both by restricting evolutionary bottlenecks to gene segmentsand by generating novel genotypes in pol and env. This increaseddiversity resulting from recombination may be critical to adaptiveevolution in the face of changes in immune pressure or antiretroviraltherapy. Further studies evaluating the contribution of recombinationto the sequential evolutionary changes observed for individualpatients will be helpful in exploring this possibility.

ACKNOWLEDGMENTS

This work was supported in part by the Agence Nationale de Recherchessur le SIDA (ANRS) grant 2005/002. C.C. is a fellow of the ANRS.

FOOTNOTES

* Corresponding author. Mailing address: INSERM U 552, Hôpital Bichat—Claude Bernard, 46, rue Henri Huchard, 75018 Paris, France. Phone: 33-1-40-25-63-55. Fax: 33-1-40-25-63-70. E-mail: hance{at}bichat.inserm.fr.

C.C. and T.N. contributed equally to this work.

REFERENCES

Top