HIV Populations Are Large and Accumulate High Genetic Diversity in a Nonlinear Fashion

INTRODUCTION

Infection with human immunodeficiency virus type 1 (HIV) results in lifelong persistent infection. In most cases, HIV infection results from expansion of a single or limited number of viral variants (1–4), producing an initially uniform virus population. From the time of infection, HIV genetic diversity emerges as a function of mutation, drift, recombination, selection, and population size. Early in infection, genetic diversity increases in a linear fashion (5), at rates somewhat lower than that predicted by the rapid (generation time, 1 to 2 days) and error-prone replication program (with the unselected reverse transcription mutation rate of 3 × 10⁻⁵ to 5 × 10⁻⁵ mutations/base/replication cycle [6]). HIV variants emerge with mutations at a number of positions; distribution of genetic distances in these early populations largely approximates a Poisson distribution, suggesting that in general, sites undergo mutation at random. Emergence of variants with mutations at specific cytotoxic T-lymphocyte (CTL) sites is relatively frequent and suggests that although mutations may occur at random, individual variants emerge as escape mutations (1, 5). These data (1, 5) demonstrate a strong role for both mutation and selection in the formation of initial populations in infected individuals. After years of infection, substantial genetic diversity accumulates, and this highly diverse population can rapidly respond to selective pressures, facilitating immune escape and resistance to antiviral drugs. Understanding how new mutations emerge and become fixed in HIV populations is critical to designing effective strategies for the prevention and suppression of these sequelae (7–14).

Although much has been learned regarding establishing HIV infection in vivo, critical gaps in our understanding persist that limit our understanding of the dynamics of HIV populations and the emergence of drug resistance. In particular, the size of the replicating HIV population in vivo remains uncertain. Relatively small population sizes (<1,000 infected cells/replication cycle/infected individual) have been reported, implying that stochastic effects and genetic drift predominate, with the potential for rapid emergence of mutations and shifts in population structure. In contrast, we and others have suggested a relatively large replicating population in vivo, with considerable contribution of deterministic effects, including slow shifts in population structure (15, 16). Most prior studies of HIV diversity in infected patients focused on env. env data sets are characterized by high diversity and are rich in strongly selected immune response sites, but they do not offer potential to understand detailed emergence of antiretroviral drug resistance. In addition, high genetic diversity presents substantial challenges in obtaining data sets that are not biased by selective amplification; genetic diversity and an excess of insertions and deletions also render env data sets difficult to align with confidence, complicating detailed phylogenetic and population genetic analyses.

To investigate HIV population genetics parameters, including population size in regions relevant to antiviral drug resistance to the majority of antiretrovirals, we investigated a 1.3-kb amplicon in pro-pol, which includes positions where mutations conferring escape from the CTL response as well as resistance to commonly used treatment regimens are found (12, 17). This region has a degree of genetic diversity (0.8% to 2% average pairwise difference [18–20]) in chronically infected individuals that is suitable for detailed fine-structure analyses of HIV populations using phylogenetic (18–20) and population genetics (15, 19, 21–23) approaches. We investigated pro-pol diversity in 33 treatment-naive individuals by analyzing large collections of individual HIV sequences, in many cases at multiple time intervals after infection. pro-pol diversity varied almost 100-fold, from 0.02% in recently infected individuals to more than 2% in individuals infected for more than 15 years. This new data set permitted a detailed analysis of HIV genetic variation, from which robust measures of diversity, divergence, and population size were obtained. Total sequence diversity (including synonymous and nonsynonymous changes) was strongly correlated with duration of infection even during chronic infection. Increases in genetic diversity over time correlated with increases in synonymous but not nonsynonymous mutations and did not correlate with plasma HIV RNA level or CD4⁺ T cell count. Studies of 11 patients sampled over 1 to 14 years revealed that the genetic composition of HIV populations changed slowly; significant shifts in HIV populations occurred only after 100 to 1,000 viral generations. We estimated that the effective replicating virus population is at least 10-fold larger than previous measurements derived using analysis of env sequences. The size and diversity of the replicating populations suggest that both selection and drift are important mechanisms leading to the emergence of HIV variants in vivo.

MATERIALS AND METHODS

RESULTS

We used single-genome sequencing (SGS) (20, 25, 33) to study pro-pol evolution in 33 HIV-infected treatment-naive patients, all with unprotected sex as their risk category (Table 1). Study patients were predominantly male, with an average age of 35.1 years. Patients were infected from an estimated 9 days to more than 15 years prior to the first sample based on patient history and laboratory studies; 22 patients were infected for <1 year. All but one (patient 1) had a positive Western blot at the time of phlebotomy. All patients had CD4 lymphopenia with a median of 401 CD4 cells/μl blood, and viral RNA levels ranged from 3.1 to 6.1 log₁₀ copies/ml of plasma. As described previously (20, 25), SGS produces a data set of individual sequences derived from single HIV genomes that is ideally suited to investigate genetic diversity because of its low error rate, undetectable assay-based recombination, and absence of founder effects due to resampling. We obtained an average of 22 (range, 9 to 51) sequences for each time point. To investigate the precision of these determinations, we constructed theoretical populations, which we sampled with multiple replicates of increasing sample sizes. As shown in Fig. 1, increasing sample sizes above 10 sequences yielded adequately precise measurements of genetic diversity (to within 1% of theoretical value, with a standard error of mean of 0.11). This level of sampling yields reproducible measurements of genetic diversity.

Among these study patients, pro-pol nucleotide diversity, as measured by percent average pairwise distance (APD), ranged nearly 100-fold from 0.02% in early (<1 year's duration) infection to slightly more than 2% after 15 years of infection (Table 1). Notably, all but one (patient 1) of the early infection patients had positive Western blots, demonstrating that a strong serologic response was already present. We first compared the minimum duration of infection with genetic diversity of each patient sample tested. Overall, there was a significant correlation between the minimum duration of infection and diversity, measured as APD (r² = 0.47, P < 0.001), indicating a progressive increase in diversity with time. Detailed analyses revealed that the rate of accumulation of APD was not uniform, however. As shown in Fig. 2A, analysis of all samples from the 33 patients revealed that early in infection (patients 1 to 17), APD increased relatively rapidly, at an overall rate (0.006%/day; r² = 0.45, P = 0.002) which approximated that expected from the mutation rate of reverse transcriptase (corresponding to an increase of 0.004%/day, assuming 1 replication cycle/day [6, 34, 35]) (Fig. 2B) and which was similar to previously published data (5). As Keele et al. (1) and others have reported, analysis of HIV genetic diversity in early infections revealed that pairwise differences were Poisson distributed, indicating that overall, mutations occurred randomly throughout the sequence. Consistent with earlier findings, we identified several individuals with recent HIV infection with HIV populations with higher genetic diversity than expected assuming a single infecting virus, indicating infection with more than one founder (Fig. 1A).

In contrast to recent infections, when analyses were restricted to the patients infected for more than 1 year, APD increased 0.0002%/day (r² = 0.49, P = 0.005) (Fig. 2C), indicating ongoing accumulation of new mutations, albeit at a rate about 30-fold lower than in early infection. During the period where accumulation of diversity slowed (1 to 2 years), we noted considerable range in diversity among patients (Fig. 1A and Fig. 2D), suggesting variable effects of selection and drift.

The period approximating 1 year of infection included samples with a relatively wide spectrum of genetic diversity. To investigate whether mutations were distributed randomly throughout pro-pol, we analyzed the distribution of pairwise differences. As previously described, random accumulation of mutations yields distributions according to Poisson statistics, while nonrandom mutation results in skewed pairwise differences. Analysis of the distribution of pairwise differences in HIV populations from chronically infected individuals revealed distributions with strong Poisson characteristics, but with deviations from ideal Poisson populations (F. Maldarelli, unpublished data). These data suggest that mutations continue to accumulate in random fashion during chronic infection, but specific changes may occur as well.

To further characterize the accumulation of new mutations, we compared changes in synonymous and nonsynonymous diversity over time. As shown in Fig. 2E, both nonsynonymous and synonymous diversity increased sharply during early months of infection; however, approximately 8 months later, nonsynonymous diversity stabilized and synonymous diversity continued to increase. These data indicate that PR and RT are undergoing change largely under purifying selection most likely as a result of constraints on protein structure.

Although we detected a significant correlation between genetic diversity and duration of infection, the correlation coefficients for recent and chronic infection (r² = 0.45 and 0.49, respectively) indicated that duration of infection explained only a portion of the variability in genetic diversity. To look for other correlates, we compared virologic and immunologic measures. No correlation was found between diversity and plasma HIV RNA or CD4 count in individuals with established HIV infection (duration of infection of >3 months, r² = 0.04 and 0.07, respectively) (Fig. 3A), indicating that overall HIV pro-pol genetic variation was not associated with the level of viremia or extent of immunodeficiency.

We further investigated the relationship between genetic diversity and time with longitudinal sampling of 12 patients with HIV infection and various baseline diversities. To determine the relative tempo of HIV variation, we compared sequences from samples obtained on daily, monthly, and yearly bases by phylogenetic analysis. As we and others have shown (1–5), HIV population structure was relatively monomorphic during early infection (Fig. 4, patient 2, panel D), which arose from the few mutations that appeared over the relatively short period of observation. Early in infection (<1 year), diversity increased approximately as predicted by the mutation rate (Fig. 4, patient 10), as previously noted (5). In contrast, during chronic HIV infection, diversity remained relatively stable (Fig. 4, patients 14, 24, 29, 30, and 31, panels B), even during progressive decline of CD4 cell counts (patients 24 and 30) and more than 10-fold increases in HIV RNA levels (patient 30). As expected from cross-sectional data (Fig. 2A), increases in diversity were nonetheless detectable in temporally spaced samples obtained from individual patients, although consistent rates of divergence among all patients were, in general, not discernible (data not shown). Analysis of daily samples from two patients revealed no variation in HIV diversity over a 10-day observation period (Fig. 4, patient 29, daily samples, panel B; data for the second patient are not shown), excluding rapid fluctuation due, for example, to differential seeding of the population from genetically distinct tissue sites of replication.

Neighbor-joining analysis revealed that temporally spaced pro-pol sequences remained highly related to one another. Samples obtained on a daily basis (Fig. 4, patient 29) or over 5 years revealed that only a few (1 to 6) sequences or clusters of sequences from individual times had bootstrap values (>75%) sufficient to support the observed branching (Fig. 4, patients 10, 14, 29, 30, and 31, panels D, thick colored bars). Of the 12 patients with longitudinal sampling, one (Fig. 4, patient 30, panel D, blue branches) had evidence for divergence in a subset of 6 sequences after a sampling interval exceeding 5 years, and a second (patient 24) had evidence of emergence of a distinct lineage after nearly 14 years. In the remaining patients, phylogenetic topologies of temporally spaced samples suggested a shared common ancestry for HIV sequences; the most recent sequences did not demonstrate progressive accumulation of diversity compared to the earliest sequences.

Temporally spaced data were also useful in providing a detailed view of HIV polymorphisms and identify changes in individual allele frequencies over time. As shown in Table 2, for 10/11 patients, a relatively small number of alleles underwent change during the observation period (median, 9%; range, 0 to 18%). None of the alleles that emerged or underwent fixation were linked to alleles that underwent significant change in allele frequency, indicating that fixation did not result in a selective sweep that carried other alleles. Rather, the occurrence of unlinked polymorphisms emerging or undergoing fixation in this fashion indicates that populations are highly diverse, and certain lineages were simply lost or emerged as result of new mutation. Most sites did not undergo changes in allele frequency, suggesting that selection at these sites was not sufficiently strong to change the frequency. HIV from one patient (patient 24) underwent significant change during a prolonged observation period (5,099 days), with 43% of 90 polymorphisms undergoing significant change, 21 of which were new or lost alleles, a number of which were linked (Table 2). As shown in Fig. 4, the HIV population structure in this patient was distinct, with all of the sequences from the later time point on a distinct lineage with strong bootstrap support, accounting for the number of new changes. Patient 30 also had a new bootstrap-supported lineage emerge after a long period (5.7 years) but also had a number of variants present.

Recombination is a common phenomenon in HIV replication; as we previously reported, approximately 6% of infected cells are likely infected with more than one provirus (36), providing the opportunity for recombination to occur. In HIV-infected patients, recombinants accrue during the entire course of infection. As a result, demonstration of recombination using standard phylogenetic techniques (37) detected frequent evidence of recombination with recombination intervals of 36 to 120 nt (Maldarelli, unpublished).

Despite the absence of clear phylogenetic evidence of divergence and the relatively stable intrapatient viral diversity, substantial population shifts were detectable when we applied an adaptation of the geographic subdivision test (21) to identify patterns of population structure. Population shift is indicated by a loss of panmixia (a population characteristic in which all sequences in the sample comparison belong to a single replicating group); in comparing sequences from two different time points, a low (1 × 10⁻⁹) probability of panmixia indicates population divergence. In contrast to the relatively homogeneous populations indicated by the NJ analyses, the geographic subdivision test showed clear evidence of population shift in pro-pol sequences from patients with HIV infection sampled over prolonged periods (Fig. 4, patients 10, 14, 24, 29, 30, and 31, panels C), whereas at short intervals (Fig. 4, patient 2 or patient 29, daily samples), no evidence of population shift was detectable. Cumulative analysis of all intrapatient pairwise comparisons revealed that the median time to population shift (defined as a probability of panmixia of <10⁻⁹) was 1,017 days, and the minimum duration before shift was detected was 193 days (Fig. 5). These data are consistent with our initial report of the population subdivision adaptation (21) and indicate that significant change in HIV pro-pol population structure takes place with a time scale that is 100- to 1,000-fold longer than the replication cycle time of HIV in vivo (1 to 2 days).

The relatively low rate of population shift in HIV population structure implies relatively large replicating populations in vivo. Therefore, we used two tests to investigate further the effective size (N_e) of the HIV populations. As shown in Fig. 6, coalescent analyses (diamonds) yielded uniformly low measures of effective population size, on the order of 100 to 1,000, similar to estimates previously reported (33, 38–41), a surprising result in light of the slow change in population structure detected by the population subdivision analyses. This difference may be due to the fact that this method ignores the contributions of selection and recombination, both of which can lead to underestimation of population size (15, 42). Therefore, we next determined population size using a phylogeny-independent method described by Tajima and Nei (31) and Waples (32). This method estimates population size based on the rate of change of individual allele frequencies over time and thus yields a range of population sizes; assuming no selection, large changes in allele frequencies yield the smallest estimates of population size, and relatively small changes in allele frequency yield the largest population sizes. As shown in Fig. 6A (whisker plot), N_e estimates varied more than 10- to 100-fold among individual patients, reflecting a wide range of changes in allele frequency among HIV pro-pol polymorphisms. HIV populations from two individuals (patients 10 and 14) had relatively narrow quartile distributions of population sites, reflecting a restricted range of allele frequency changes.

The median N_e estimates obtained using the latter method were in the range of 10³ to 10⁴ (Fig. 6B), or >30-fold higher than that measured by coalescence-based methods, and are more consistent with, although still less than, population sizes estimated from linkage equilibrium analyses (15). Even the minimum estimates of allele frequencies obtained by this method were, in general, greater than those estimated by coalescent methods. To investigate the relative contributions of selection and drift on N_e, we further analyzed the type of variability on a site-by-site basis (Fig. 6B). We expected that nonsynonymous polymorphisms resulting in changes in amino acids would be subject to greater selective forces and would yield smaller values for N_e, whereas estimates of N_e using synonymous polymorphisms would be less subject to selection and more influenced by genetic drift and would yield large values for N_e. Consistent with this expectation, the overall population size measured using synonymous sites was greater than that measured using nonsynonymous sites; the difference, however, was modest and of marginal statistical significance (5,600 versus 4,500 transmitting cells per generation for nonsynonymous and synonymous sites, respectively; two-sided t test, P = 0.035). We investigated the estimates of population sizes by nucleotide position of polymorphisms within pro-pol to investigate the role of synonymous and nonsynonymous sites and to determine whether there were region-specific effects of drift or selection. As shown in Fig. 6B, the nonsynonymous and synonymous alleles that contributed to large and small population size estimates were distributed throughout pro-pol and were not localized by gene or domain.

The observation that some nonsynonymous sites yielded high population sizes suggests that some sites are not undergoing selection; alternatively, it is possible that such polymorphisms are maintained by frequent mutation at specific sites. If individual sites were undergoing frequent mutation, we would expect to identify such sites as repeatedly polymorphic in several individuals. However, of 56 nonsynonymous sites yielding population estimates of >20,000, only 2 (3.6%) were present more than once. As a result, it is unlikely that frequent mutation at individual sites is responsible for persistence of stable polymorphisms; these polymorphisms are more likely to be stably maintained over time because of relatively large population sizes. Taken together, these data indicate that measurements of HIV effective population sizes are heavily influenced by variations in allele frequency and change over time and from one site to the next due to variation in selection and drift. Our estimates should, therefore, be taken as a lower bound, and the true values are likely to be much higher.

DISCUSSION

HIV genetic diversity within individuals is the substrate upon which immune and antiretroviral drug selection act. Previous studies (1–5, 13) have reported that diversity in most recently infected individuals is very low, consistent with initiation of infection with a single variant. In patients with established infection, pro-pol diversity accumulated at a much lower rate than in recently infected individuals, and over the course of infection, diversity increased in a nonlinear fashion (Fig. 2A). The strength of the correlation between diversity and time for both early and established HIV infection (r² = 0.47 to 0.55) suggests that duration of infection explains only a portion of the variability in diversity. All of the participants in this study were infected with subtype B virus; a recent study sequencing single genomes from early-postinfection subtype C-infected individuals has identified a similar increase in genetic diversity in nonstructural genes, including vif, vpu, tat, and rev (43).

The absence of association between pro-pol diversity and viral RNA level that we observed is similar to a previous analysis of env diversity and viral RNA levels (44) and implies that despite 100- to 1,000-fold differences in the level of viremia, the number of productively infected cells must be sufficiently large to sustain a highly diverse population of virus. Furthermore, we found no instances of a sudden shift in the HIV population that would suggest a bottleneck due to a selective sweep or other strong limitation on the infected-cell population size. Additionally, the absence of short-term fluctuations in diversity implies that the virus in blood is a well-mixed population derived from a constant, steady source, rather than localized bursts of virus from sites infected with genetically distinct populations. Finally, in a related study, we have found that diversity of the virus population is maintained throughout the course of infection, even following reductions in the number of productively infected cells 10,000-fold following antiretroviral therapy, indicating a large population of infected cells (M. Kearney, J. Spindler, S. Yu, W. Shao, A. O'Shea, C. Rehm, C. Poethke, J. W. Mellors, J. M. Coffin, and F. Maldarelli, presented at the 17th Conference on Retroviruses and Opportunistic Infections, San Francisco CA, 16 to 19 February 2010). As previously observed (5), genetic diversity early in HIV infection accumulated at a rate approximating that expected from its mutation rate. In contrast, accumulation of diversity slowed more than 30-fold in chronically infected individuals, suggesting a restriction on accumulation of new mutations. Differential accumulation of synonymous and nonsynonymous mutations is consistent with limitation of diversity due to purifying selection. In general, only a small proportion of polymorphisms underwent change over time, fewer still were fixed, and in only one patient (patient 19), with strong phylogenetic evidence of emergence of a distinct variant more than 13 years after infection, were these fixed polymorphisms linked (Table 2). Previous reports of accumulation of variation in env according to a strict (17) or relaxed (45) molecular clock were not reflected in our overall analysis of pro-pol. Instead, diversity increased asymptotically, with maximum APD values on the order of 2% about 15 years after infection, suggesting a limit to the amount of diversity that can accumulate within an individual. Similar conclusions (15) on the lack of temporal structure in HIV sequences have been drawn from analyses of env sequences in several patients (46). Maximum intrapatient pro-pol diversity during chronic infection was still substantially lower than the corresponding interpatient pairwise comparisons, which typically exceeded 5% (reference 5 and data not shown). In addition, it is not clear why accumulation of diversity slowed markedly after 9 to 18 months of infection. It is unlikely that slowing in diversity accumulation was the result of onset of immune responses, as accumulation of diversity occurred after development of serologic and cellular immune responses. These data indicate that within an individual, HIV genetic variation remains restricted by strong purifying selective forces.

All of the participants in this study were infected with subtype B virus. It will be of great interest to determine whether other subtypes have similar intrapatient diversity and accumulate diversity at rates comparable to subtype B. Recently, Rossenkhan and coworkers (43) conducted a detailed analysis of subtype C-infected individuals, sequencing single genomes from early-postinfection individuals to obtain diversity estimates for HIV accessory genes, including vif, vpu, tat, and rev. Similar to the case with subtype B, genetic diversity was restricted in these early infection samples and accumulated over time. A comprehensive analysis of subtype-specific genetic variation will yield new insights in understanding HIV pathogenesis.

The relative size of the replicating HIV population (N_e) remains uncertain but is a critical parameter in understanding the spread of new mutations conferring resistance and immune escape (8, 9, 13). In relatively small populations (≪1/mutation rate or ≪3 × 10⁴), new mutations spread in stochastic fashion, while in large populations (≫1/mutation rate or ≫3 × 10⁴), emergence of new variants approaches a deterministic limit (16). Estimating replicating population sizes typically uses coalescent approaches. Coalescent theory is an inherently retrospective approach rooted in neutral population genetics theory that reconstructs a genetic history based on present population structure. The model assumes that mutations arise according to a constant mutation rate in a strict molecular clock-like fashion; all alleles are neutral and reassort in random mating in populations that remain constant in size. Using a contemporaneous set of polymorphisms with measured allele frequencies in populations, coalescence uses probability analyses to reconstruct an entire population history, identifies times when genealogies “coalesce” to a most recent common ancestor (MRCA) of the population, and describes the most probable pathway to the ancestor, depicted in dendrograms that are measured in time (rather that genetic distances present in phylogenetic analyses). Based on genetic diversity determinations, a replicating population size can be estimated. Coalescence theory generally underestimates population size but represents a powerful approach to reconstructing genetic histories of diverse variants, including HIV (47), over long periods, where genetic diversity is substantial. In analysis of intrapatient data, however, the genetic diversity is more restricted, and coalescent approaches may be more sensitive to the effects of selection, yielding lower estimates for population size. In our estimates, standard coalescent approaches yielded uniformly low replicating population sizes, in the range of 10 to 100 (Fig. 3). Additional analyses using allele frequency variation to estimate N_e yielded replicating population sizes that were 30-fold greater than coalescence-based estimates, and these estimates varied greatly from one site to the next.

Site-by-site analysis also revealed that both synonymous and nonsynonymous polymorphisms underwent relatively slow change, indicating that some nonsynonymous sites are subject to relatively little selection. In addition, we also observed nonsynonymous and synonymous sites that underwent change at relatively high rates, suggesting that such sites were undergoing selection compared to others. Constraints on nonsynonymous sites have been well described: additional selective forces, including RNA structure and codon preference, may affect the allele frequency of synonymous sites. One consequence of large population sizes is a relatively long time to detectable genetic shift. The median time of approximately 1,000 days (corresponding to about 1,000 virus generations) for population genetic shift to appear suggests that prior to therapy, HIV replication proceeds as a large, well-mixed population without selective sweeps or rapid changes in composition. Since many, if not most, of the nonsynonymous changes in HIV that become fixed during all phases of infection are in sites recognized by the cellular or humoral immune response (5, 48–50), the absence of detectable bottlenecks in the population associated with their appearance implies that the selective force imposed by the immune response to any given epitope, although readily detectable by the selection of escape mutations, is not sufficiently strong to influence the overall population size or structure.

Our finding of relatively large population sizes contrasts sharply with previous studies that concluded the existence of relatively small population sizes using env sequences for analyses. Earlier env data sets available for study, such as those of Shankarappa et al. (17), are extensive but have relatively few individual plasma-derived sequences compared to the larger numbers of sequences used here to determine population size. For comparison purposes, we did carry out a site-by-site analysis on two patients in the data set of Shankarappa et al. with 10 or 11 sequences/time point. Our analysis revealed median population sizes of 2,736 (range, 2,362 to 53,702) and 5,688 (range, 3,197 to 62,571), similar to what we have identified in pro-pol; the high upper boundaries of these determinations represent the contribution of alleles with relatively stable allele frequency over time and reflect the presence of a relatively large population size. New studies with more single-genome sequences will be useful in directly estimating population sizes using env and pro-pol sequences.

Population sizes in the range of 1 × 10⁴ to 1 × 10⁵ approximate the inverse of the estimated unselected mutation rate of 3 × 10⁻⁵ to 4 × 10⁻⁵ (6); the HIV mutation rate in vivo has not been well studied, and actual mutation frequencies are likely to be strongly influenced by both selection and genetic drift (15, 39, 51–53). This conclusion is consistent with the detection of alleles with rapid (selection) and slow (drift) change and with the overall slowing in accumulation of diversity in chronic HIV infection. The issue becomes particularly important in considering the frequency of drug resistance mutations in untreated individuals. The rapid and reproducible appearance of such mutations following monotherapy with antiviral drugs such as lamivudine (3TC) (11) and single-dose nevirapine (54) implies their presence in the replicating virus population in most or all infected individuals prior to therapy. Their frequency will be determined by the balance between mutation, counterselection, and drift (16) but must be at least the inverse of the replicating population size, on average. Studies to date using sensitive allele-specific PCR methods, however, have failed to reproducibly detect such mutations, suggesting that the population size may be substantially larger than estimated here. Further development of very sensitive mutation detection technology as well as advances in mathematical modeling will be needed to resolve this important issue and provide critical tests of the selection-drift hypothesis and a better understanding of the virus population size and structure, which can be directly applied to understanding the emergence of drug resistance.

The population studies reported here have broad implications for understanding the pathogenesis and therapeutic responses in other chronic viral infections, especially so for those viruses with constantly replicating populations in chronic infection and new and expanding therapeutic agents, such as hepatitis B and C viruses. Hepatitis B virus has a number of effective therapeutic agents, although determinants of viral control and resistance are poorly understood. Genetic diversity is substantial, but the relationships between genetic diversity, population size, and emergence of resistance have not been extensively investigated (55, 56). Therapy for hepatitis C has expanded with additional targets and therapeutic agents; cure rates have improved, but the virologic correlates of eradication are incompletely understood. Population genetics studies have demonstrated that hepatitis C virus populations are highly genetically diverse, even relative to HIV, so it is likely that, similar to the case with HIV, drug-resistant mutations will preexist prior to therapy. Intrapatient genetic variation has been investigated (57–61), although population sizes have not been extensively investigated and it is not known how fast new drug-resistant mutations may be expected to emerge. Additional studies such as those reported here will have direct applications in the design of clinical trials and the composition of combination therapy necessary to eradicate viral infection.