HIV was transmitted from apes to humans multiple times before the virus began to spread among humans some time in the early 20^th century. The virus is now found around the globe and has evolved into several different subtypes. Also, HIV continues to evolve inside each infected individual so that different variations of the virus are often present in the same person at any given time.

When a person initially becomes infected, immune cells called killer T-cells seek out and destroy infected cells. However, HIV quickly acquires genetic mutations that allow it to escape the immune system and multiply in the body. There is a cost to this evasion strategy: the mutant forms of the virus often don’t multiply as rapidly as the variant that first infected the individual. This may explain why the virus often mutates back to its previous form in newly infected people. But many questions remain about how these pressures influence the evolution of the virus in individuals and in populations over time.

Now, Zanini et al. have sequenced the genomes of HIV samples collected from nine people with HIV in Sweden over the course of five to eight years. These individuals were diagnosed between 1990 and 2003 – when HIV medications were not as widely used as they are now – and did not receive any treatment during the course of the study. Therefore, the study provides a rare opportunity to look at how HIV evolves in the absence of drugs that target the virus.

Zanini et al. sequenced the entire genetic code of each form of the virus identified in the samples. The data show that HIV follows predictable patterns of evolution within individuals as well as across human populations. Mutations happen frequently all over the HIV genome. However, the mutant viruses often revert to a common or “optimal” form of the virus throughout the course of infection. This suggests that there is a tradeoff between the benefits of acquiring new mutations and maintaining a set of traits that have enabled the virus to spread so successfully in humans.

To make it easier for other researchers to explore the data, Zanini et al. created a web application that allows others to access and create visual representations of viral evolution.

Together these findings suggest that it will be possible to achieve a fuller understanding of RNA virus evolution that integrates the molecular biology of the virus and the immune response of the host with the evolutionary changes.

The human immunodeficiency virus 1 (HIV-1) is a paradigmatic example of a rapidly adapting population characterized by high diversity, strong selection, and recombination. HIV-1 has originated from multiple zoonotic transmissions from apes in the early part of the 20th century (Sharp and Hahn, 2011), one of which gave rise to the worldwide pandemic. This lineage is called group M and has diversified into different subtypes at a rate of about 1 in 1000 substitutions per year (Lemey et al., 2005; Li et al., 2015). Tens of thousand of HIV-1 group M variants are available in the Los Alamos National Laboratories HIV database (LANL) (Foley et al., 2013).

The evolution of HIV-1 ultimately takes place within infected individuals and can be observed directly in longitudinal samples of virus populations from the same individual. The detailed knowledge of HIV-1 biology paired with historical samples makes HIV-1 an ideal system to study general features of evolution at high mutation rates and strong selection that are otherwise only accessible in evolution experiments (Elena and Lenski, 2003; Miralles et al., 1999).

During the first few months of an HIV-1 infection, the viral population typically acquires several mutations that mediate escape from cytotoxic T-cells (CTL). The properties and dynamics of these mutations and their effect on epitope presentation and virus control have been extensively studied (Bernardin et al., 2005; Goonetilleke et al., 2009; Jones et al., 2004; Kearney et al., 2009; Liu et al., 2011). Such escape variants often compromise virus replication and have been shown to revert upon transmission of the virus to a human leukocyte antigen (HLA) mismatched host (Brockman et al., 2010; Leslie et al., 2004; Li et al., 2007). However, the quantitative contribution of escape and reversion to HIV-1 evolution and the degree to which costly mutations are compensated by additional mutations is less well understood (Lythgoe and Fraser, 2012; Schneidewind et al., 2009). Similar to CTL escape mutations, drug resistance mutations can spread through the viral population within weeks and longitudinal sequence data has been used to study emergence and fitness cost of resistance mutations (Hedskog et al., 2010; Little et al., 2008; Paredes et al., 2009).

Longitudinal sequence data has also been used to track evolution driven by the humoral immune response against HIV-1 which occurs throughout infection and results in high rates of evolution in the variable loops of HIV-1 surface proteins (Bar et al., 2012; Richman et al., 2003; Shankarappa et al., 1999). In a pioneering study, Shankarappa et al. (1999) characterized HIV-1 evolution of parts of the gp120 envelope protein over approximately 6–12 years in 11 patients, demonstrating consistent patterns of diversity and divergence. While this early study was limited to about 10 sequences per sample, next-generation sequencing technology today allows deep characterization of intrapatient HIV-1 variability including rare mutations (Fischer et al., 2010; Hedskog et al., 2010; Henn et al., 2012; Tsibris et al., 2009).

The earlier studies discussed above have either focused on early infection or particular regions of the HIV-1 genome. However, most virus evolution happens during chronic infection simultaneously at many locations in the genome. To develop a comprehensive and quantitative understanding of the evolution and diversification of HIV-1 populations, we generated a whole-genome, deep-sequencing data set covering nine patients over 5–8 years with 6–12 time points per patient. Importantly, the data set covers the entire genome such that no substitution is missed and includes early samples defining the initial population. To our knowledge, this is the only whole-genome deep sequencing data set with long follow-up of multiple patients. We provide interactive and intuitive web-based access to the data set and hope it will become a resource for many future studies, like the data set by Shankarappa et al. (1999) has been in the last years.

Below, we first describe the methodology we developed to sequence the entire HIV-1 genome at great depth. We then analyze the intrapatient evolution of HIV-1 and show that the minor variants in the virus population explore sequence space in predictable fashion at the single site level. At the same time, we observe a strong tendency for reversion towards the global HIV-1 consensus that is not limited to early infection but occurs at an approximately constant rate throughout chronic infection. Reversion is more frequent at sites that are more conserved at the global level, suggesting a direct relationship between intrapatient fitness cost and global conservation. Together with reproducible patterns of intrapatient variation, this link explains why HIV-1 fitness landscapes can be inferred from cross-sectional data (Dahirel et al., 2011; Ferguson et al., 2013; Mann et al., 2014). We further find frequent recombination, which allows the viral population to evolve independently in different regions of the genome. Nevertheless, recombination is not frequent enough to decouple mutations closer than 100 base pairs and we observe signatures of hitch-hiking at short distances (Maynard Smith and Haigh, 1974).

The study included nine HIV-1-infected patients who were diagnosed in Sweden between 1990 and 2003. Data from two additional patients (p4 and p7) were excluded from analysis because of suspected superinfection and failure to amplify early samples with low virus levels, respectively. The patients were selected to have a relatively well-defined time of infection and to have been treatment-naive for a minimum of 5 years. Patients diagnosed in recent years rarely fulfill these inclusion criteria because therapy is almost universally recommended, but this was not the case when the study patients were diagnosed. Basic characteristics of the patients are presented in Table 1. Additional data on the patients are presented in Supplementary file 1 and Figure 1—figure supplement 1. All patients progressed to moderate-to-severe immunosuppression during the study period. The final CD4 count within the study period for patients 1–3, 5–6 and 8–11 were 340, 369, 140, 228, 287, 378, 158, 200 and 251, respectively. Of the nine patients, eight were males, seven were men who have sex with men (MSM), two were heterosexually infected (HET), and eight were infected in Sweden. The method to estimate the date of infection (EDI) is described in Materials and methods. We retrieved longitudinal biobank plasma samples covering the time period from early infection (time since EDI 33–209 days) until end of study follow-up, which usually coincided with start of antiretroviral therapy (time since EDI 5.5–8.3 years). The number of samples per patient ranged between 6 and 12 and, in total, we investigated 73 samples. The median plasma HIV-1 RNA level was 12,000 copies/ml for the samples with available data on RNA levels. Some samples had low RNA levels, which partly is explained by the study design that required that the patients were treatment-naive for at least 5 years. For detailed information about each individual sample see Supplementary file 1. For a phylogenetic tree of minor variants from all patients see Figure 1—figure supplement 2.

HIV-1 whole-genome deep sequencing

The basic steps of our amplification and sequencing pipeline are illustrated in Figure 1 and explained in detail in Materials and methods. Briefly, viral RNA was extracted from 400 μl patient plasma and used for one-step RT-PCR amplification with six overlapping primer sets that span almost the complete HIV-1 genome, similar to the strategy developed by Gall et al. (2012). Sequencing libraries were made starting from 0.1–1.5 ng of DNA with a stringent size selection for long inserts (>400 bp). Sequencing was performed on the Illumina MiSeq platform and sequence reads were quality filtered and assembled using an in-house data processing pipeline. In total, approximately 100 million reads passed the quality filtering. The coverage varied considerably between samples and amplicons, but was mostly of the order of several thousands or more, see Figure 1B.

Figure 1 with 2 supplements see all

Download asset Open asset

Importantly, sequencing depth is determined not only by coverage but also by template availability and sequencing errors (Iyer et al., 2015). We performed a number of control experiments to quantify templates and assess the accuracy of estimates of frequencies of single nucleotide polymorphism (SNP). The results are summarized in the following section and described in detail in Materials and methods.

We quantified the number of HIV-1 genomes that contributed to each sequencing library by PCR-limiting dilution (median: 120 quartiles: 50–500). Hence template availability, rather than coverage, determined the sequencing depth, see Figure 1B. Comparison with routine plasma HIV-1 RNA level measurements performed at time of sampling showed that the median template complementary DNA (cDNA) recovery efficiency was 30%.

We estimated the error rate of the PCR and sequencing pipeline by amplifying and sequencing a plasmid clone. Figure 1C compares the SNP frequencies observed in a clone to those observed in a patient sample. After quality filtering, PCR and sequencing errors never exceeded 0.3% of reads covering a particular position. To detect and control for variation in PCR efficiency among fragments and skewed amplification of different variants, we compared frequencies of variants in overlaps between the six amplicons. A SNP in the overlap is amplified and sequenced twice independently and the concordance of the two measurements of variant frequencies was used to estimate the fragment specific depth, which is limited both by template input and PCR efficiency, see Materials and methods. Frequency estimates were often reproducible to within a few percent. Sometimes, however, variant recovery was poor (mostly in fragment 5) and frequency estimates less accurate. Those fragments are flagged on the website and can be excluded from analyses requiring large depth.

We minimized PCR recombination by reducing the number of PCR cycles and optimizing the reaction protocol (see Materials and methods, Di Giallonardo et al. (2013); Mild et al. (2011)). Control experiments using mixtures of two cultured virus populations showed that less than 10% of reads had experienced RT-PCR recombination.

Taken together, our control experiments show that depending on the sample and fragment, we could estimate frequencies of SNPs down to 1% accuracy (corresponding to several hundred effective templates). In some cases, however, the template number was low or template recovery poor such that only presence or absence of a variant could be called. Furthermore, SNPs remained linked through cDNA synthesis, PCR and sequencing.

Consistent evolution across the entire genome

In most patients, the virus populations were initially homogeneous and diversified over the years, as expected for an infection with a single or small number of similar founder viruses (Keele et al., 2008; Salazar-Gonzalez et al., 2009). In two patients, p3 and p10, the first sample displayed diversity consistent with the transmission of several variants from the same donor, see Figure 3—figure supplement 1. For each of the nine patients, we reconstructed the HIV-1 genome sequence of the first sample by an iterative mapping procedure described in Materials and methods. We use this initial consensus sequence to approximate the sequence of the transmitted founder virus(es). Our first sample is estimated to be between 1 and 7 months into infection and a few mutations (likely CTL escape mutations) had probably spread through the viral population by that time. Thus, the initial consensus sequence probably will differ slightly from the true founder virus. The number of differences, however, will be small compared to the sequence divergence in the 5 or more years of follow-up such that this initial consensus sequence is a useful approximation of the founder virus(es).

Figure 2 shows an example of the dynamics of frequencies of SNP relative to the founder sequence over time, where each dot (top) or line (bottom) represents the frequency of a nucleotide different from the founder sequence. Interactive versions of this graph are available for the entire genome of all patients at hiv.tuebingen.mpg.de.

Figure 2

Download asset Open asset

To measure the rate at which the virus population accumulates mutations, we calculated the average distance of each sample from the approximate founder sequence in 300 bp windows. Regressing this distance against time yields the rate of divergence in different regions of the genome, see Figure 3A. As expected, some regions such as the variable loops in gp120 and nef evolve faster, while enzymes – protease (PR), reverse transcriptase (RT) – and the rev response element (RRE) evolve more slowly. The rate of divergence varies by about a factor of 10 along the genome, but is consistent with typically about 1.5-fold differences across patients (standard deviation of log2(fold change) 0.6 ± 0.2). The overall pattern of the rate of mutation accumulation agrees with a recent map of HIV genome-wide variation from a population perspective (Li et al., 2015) and correlates well with entropy in a large HIV-1 group M alignment (Spearman’s p = 0.7 after the same smoothing).

Figure 3 with 2 supplements see all

Download asset Open asset

Minor variants reproducibly explore global HIV-1 diversity

Having found that coarse patterns of divergence are comparable among patients, we asked whether intrapatient diversity at individual sites in the viral genome follows general and predictable patterns. To this end, we compared diversity at each position to the diversity observed in HIV-1 group M (see Materials and methods).

Figure 4A shows the rank correlation between the site-by-site diversity in each patient and a global collection of HIV-1 sequences, both measured by Shannon entropy (see Materials and methods). In all cases, correlation with cross-sectional diversity was initially low, as expected for largely homogeneous populations. As diversity increases within patients, it tends to accumulate at positions that are not conserved, resulting in a rank correlation of about 0.4–0.5 after about 8 years. These correlations are individually significant and reproducible among different genomic regions (error bars in Figure 4).

Figure 4 with 2 supplements see all

Download asset Open asset

Figure 4—source data 1

Tab-delimited files with plotted data.

https://doi.org/10.7554/eLife.11282.012

Download elife-11282-fig4-data1-v2.zip

Figure 4B offers an alternative perspective on the exploration of sequence space by the HIV-1 populations. We classified nucleotide positions in the genome into four categories ranging from highly conserved positions to less conserved positions within group M (Q1–Q4) using the same alignment as above. Next we asked what fraction of sites within these categories show intrapatient variation at a level of at least 1%. For the least conserved positions, this fraction increased steadily to about 20% after 8 years, while less than 1% of the most conserved sites shows variation above the 1% level. This latter fraction rapidly saturates and does not increase over time. Since variant amplification, sequencing and variant calling does not use any information on cross-sectional conservations, the absence of variation above 1% at conserved sites is further evidence that our amplification and sequencing pipeline does not generate spurious variation. Other thresholds yield similar results.

Taken together, the observations in Figure 4 show that variation is not limited by the mutational input, but that HIV-1 populations accumulate diversity wherever mutations do not severely compromise virus replication. At the single nucleotide level, the spectrum of mutational possibilities is explored reproducibly and the level of within-host diversity is predicted by time since infection and cross-sectional diversity. Conserved positions are typically monomorphic even in deeply sequenced samples with high RNA template input.

The majority of nonsynonymous substitutions are positively selected

In addition to the reproducible patterns of minor diversity, virus evolution is characterized by adaptations that are specific to the host. Immune selection results in escape mutations that rapidly spread through the population (Richman et al., 2003; Walker and McMichael, 2012). Such mutations tend to be nonsynonymous, i.e., change the viral proteins, while evolution at synonymous sites is expected to be conservative. Nevertheless, synonymous mutations can be affected by ‘selective sweeps’ of linked nonsynonymous mutations (Maynard Smith and Haigh, 1974).

To quantify the degree at which the evolutionary dynamics of HIV-1 is dominated by selective sweeps, we calculated divergence and diversity separately at nonsynonymous and synonymous sites in different parts of the HIV-1 genome. Figure 5A compares nonsynonymous (solid lines) and synonymous divergence (dashed lines) in different regions of the genome. In agreement with the results presented in Figure 3, the observed rate of evolution at nonsynonymous sites differed substantially between genomic regions, with env being the fastest and pol the slowest. Divergence at synonymous sites, however, varied very little between different genomic regions indicating random accumulation of synonymous mutations (rather than positive selection).

Figure 5

Download asset Open asset

Figure 5—source data 1

Tab-delimited files with plotted data.

https://doi.org/10.7554/eLife.11282.017

Download elife-11282-fig5-data1-v2.zip

Figure 5B shows the corresponding plot for diversity, i.e., the distances between sequences from the same sample. Diversity at nonsynonymous sites (solid lines) saturates after about 2–4 years, suggesting that nonsynonymous SNPs either stay at low frequency because they are deleterious or rapidly increase in frequency and fix without contributing much to diversity. Synonymous diversity increases steadily in the 5’ part of the genome (structural and enzymes), while it saturates in the 3’ half of the genome after a few years – the exact opposite of nonsynonymous divergence.

Indeed, we observe a strong anti-correlation between synonymous diversity and nonsynonymous divergence, which is further quantified in Figure 5C. This suggests that frequent non-synonymous substitutions limit synonymous diversity because they drive linked synonymous mutations to fixation or to extinction (Maynard Smith and Haigh, 1974). We will quantify linkage and recombination below, but the differences in diversity accumulation already suggest that linkage is restricted to short distances.

The contrasting behavior of synonymous and nonsynonymous SNPs is also seen in the SNP frequency spectrum – the histogram of SNP abundance – shown in Figure 5D. While the spectra agree for frequencies below 20%, synonymous mutations are strongly underrepresented at higher frequencies (Fisher’s exact test at frequency 0.5, p-value <10^-10). This corroborates the interpretation that, due to substantial recombination, sweeping nonsynonymous mutations only occasionally ‘drag’ adjacent synonymous mutations to fixation. Synonymous mutations rarely rise in frequency because of their own effect on fitness, since they usually have small or deleterious phenotypic effects and do not contribute directly in immune evasion (Zanini and Neher, 2013). The about five-fold excess of nonsynonymous over synonymous SNPs at high frequencies (see Figure 5D) shows that the majority of common nonsynonymous mutations spread due to positive selection.

Next, we sought to quantify what fraction of nonsynonymous divergence is driven by escape from cytotoxic T-lymphocytes (CTLs). Four-digit HLA types were determined for all patients and a set of putatively targeted HIV-1 epitopes were determined using the epitope binding prediction tool MHCi (tools.immuneepitope.org/mhci). We then asked whether we observed more nonsynonymous substitutions in epitopes predicted to be targeted than expected by chance (excluding the variable loops of gp120 and the external part of gp41, see Materials and methods). We found a significant enrichment by a factor 1.9 in the putatively targeted region (p-value <3 × 10⁻⁶), corresponding to 5.5 excess nonsynonymous substitutions, whereby the total number of nonsynonymous substitutions per patient is on average 43 (median, quartiles 36–64). The set of predicted epitopes will contain false positives and lack true epitopes, hence the actual number of CTL driven substitutions could be higher as for example suggested by Allen et al. (2005) who report that roughly half of non-envelope mutations are associated with CTL responses.

Extensive reversion towards consensus

Many CTL escape mutations reduce the replicative capacity of the virus and it is known that such escape mutations often revert upon transmission to a host in which the corresponding epitope is not targeted (Friedrich et al., 2004; Herbeck et al., 2006; Leslie et al., 2004). The balance between escape and reversion results in association between specific escape mutations and the HLA types of the hosts (Kawashima et al., 2009; Palmer et al., 2013). In a diverse population of hosts, the most common state at a specific site is likely the preferred state, while rare alleles tend to be escape mutations that reduce viral replicative capacity (Carlson et al., 2014).

To quantify patterns of reversion and fitness cost, we classified sites in the approximate founder sequence of the viral populations in each subject as being identical or different from the HIV-1 group M consensus. Figure 6A shows the fraction of sites where the founder nucleotide is replaced by a mutant during the infection. This fraction is about ten-fold higher if the founder nucleotide differs from the group M consensus than if it is identical to the group M consensus. Reversion towards group M consensus occurs at a roughly constant rate throughout the observation time (5–8 years).

Figure 6 with 1 supplement see all

Download asset Open asset

Figure 6—source data 1

Tab delimited files with plotted data.

https://doi.org/10.7554/eLife.11282.019

Download elife-11282-fig6-data1-v2.zip

Of all changes accumulated by the viral populations, 30 ± 2.5% are reversions towards group M consensus (mean and standard deviation of patient bootstraps after 4–7 years). Similar results are found when comparing with the subtype consensus of each patient virus (24 ± 2.5%). Reversions are between 4 and 5 times more frequent than expected in the absence of a reversion bias (7.8 and 4.5%, respectively). These findings agree with results by Allen et al. (2005), who report that about 20% of amino acid substitutions are reversions.

By focusing on sites where the founder virus differed from the group M consensus, we are predominantly looking at weakly conserved sites. To control for conservation, we carried out the same analysis after stratifying sites by overall level of conservation. Figure 6B shows the result of this analysis, focusing on samples after 5–6 years for the sake of clarity. We find that the rate of reversion is highest at the most conserved sites. Almost 50% of all non-consensus positions at highly conserved sites had reverted to consensus after about 5 years – an almost 1000-fold excess. Even at the least conserved sites, divergence towards group M consensus exceeded divergence away from group M consensus by a factor of 3. These results suggest that the global HIV-1 group M consensus sequence represents an ‘optimal’ HIV-1 sequence, which acts as an attractor for the evolutionary dynamics within hosts. This attraction is strongest at conserved sites, but extends to the least conserved sites in the genome.

Lack of long-range linkage due to frequent recombination

To quantify the decoupling of SNPs by recombination, we calculated linkage disequilibrium (LD) between SNPs as a function of distance for each of the six fragments, see Figure 7. For most fragments, we observed a consistent decrease of LD over the first 100–200 bps, with fragment 5 being an exception with linkage of mutations at longer distances. Importantly, our linkage control (a 50/50 mix of two distinct virus isolates and a total of 1250 RNA molecules per PCR fragment) shows no decay of LD with distance, suggesting negligible RT-PCR recombination.

Figure 7

Download asset Open asset

Figure 7—source data 1

Tab-deliminated files with plotted data.

https://doi.org/10.7554/eLife.11282.023

Download elife-11282-fig7-data1-v2.zip

The observed decay of LD in patient samples is consistent with a recombination rate of 10⁻⁵/bp/day as estimated in (Batorsky et al., 2011; Neher and Leitner, 2010). Our reasoning proceeds as follows. Figure 5B indicates that diversity accumulates over a time frame of 2–4 years, i.e., about 1000 days. Recombination at a rate of 10⁻⁵/bp/day hits a genome on average every 100 bps in 1000 days. Mutations further apart than 100 bps are hence often separated by recombination and retain little linkage consistent with the observed decay length in Figure 7. The longer linkage in fragment 5 (env) might have several reasons that extend beyond our simple argument: (i) homologous recombination might be suppressed in the most variable regions, (ii) the accuracy of SNP frequency estimates is lower in F5 due to poorer amplification, and (iii) the rapid evolution of env due frequent substitutions and sweeps gives less time to break up linkage. In particular, as shown in Figure 5C, frequent and strong selective sweeps affect synonymous diversity in physical proximity along the genome, confirming the presence of linkage at short distances.

For phylogenetic analysis, we can extract haplotypes from the sequencing reads up to 500 bp in length. Only in the more diverse regions are 500 bp sufficient for well-resolved phylogenies (see Figure 8). However, we find that linkage does not extend beyond 100–200 bp. Hence the read length is not a limiting factor. Only during rapid population shifts such as drug resistance evolution, long read technologies such as PacBio would be necessary to capture the evolutionary dynamics (Nijhuis et al., 1998).

Figure 8

Download asset Open asset

We have presented a comprehensive portrait of intrapatient evolution of HIV-1 that covers almost the entire genome of the virus, characterizes minor genetic variants, and tracks the fate and dynamics of these variants over a follow-up period of up to 8 years in nine patients. We find that, during the infection, HIV-1 explores the sequence space surrounding the founder virus systematically; similar mutational patterns are observed within different, unrelated patients. Linkage between mutations is limited to approximately 100 bp, so the virus population can accumulate substitutions independently in different regions of the genome as suggested by theoretical models (Mostowy et al., 2011; Rouzine and Coffin, 2005). Nonetheless, local dynamics of SNPs is often dominated by hitch-hiking between neighboring mutations, resulting in an anticorrelation between nonsynonymous divergence and synonymous diversity. A large fraction of all substitutions are reversions towards the global HIV-1 consensus sequence, and these reversions steadily accumulate throughout infection.

The evolutionary dynamics of HIV-1 populations is the result of stochastic forces like mutation and frequent bottlenecks, deterministic fixation of favorable mutations, and recombination. The relative importance of these forces remains unclear (Brown, 1997; Frost et al., 2000; Kouyos et al., 2006; Maldarelli et al., 2013; Pennings et al., 2014; Rouzine and Coffin, 1999). Our observation that intrapatient diversity recapitulates diversity seen across HIV-1 group M and the strong tendency to revert towards consensus suggest that, in chronic infection, selection determines diversity. The reproducible exploration of sequence space can coexist with frequent adaptation only in frequently recombining large populations (Neher et al., 2013). We observe that mutations further apart than 100 bp are effectively shuffled by recombination in most parts of genome, consistent with previous estimates of the HIV-1 recombination rate (Batorsky et al., 2011; Neher and Leitner, 2010). Linkage and stochastic effects become stronger with increasing frequency of strength of selection, consistent with lower synonymous diversity and more LD in env.

While rapid CTL escape at 5–10 sites over the first 2 years of infection has been documented in detail (Allen et al., 2005; Goonetilleke et al., 2009; Herbeck et al., 2006; Jones et al., 2004; Liu et al., 2011) and population level associations between specific HLA types and escape variants suggest widespread CTL escape (Kawashima et al., 2009), the effect of escape and reversion on long-term evolutionary trends is less clear (Lythgoe and Fraser, 2012; Roberts et al., 2015). We find a strong tendency for viral populations to revert towards the global HIV-1 consensus. At sites where the founder sequence differs from the subtype consensus, substitutions are almost five-fold overrepresented: Instead of ≈5% reversions expected based on the fraction of sequence at which the founder virus differs from consensus of the HIV-1 subtype, almost 25% of substitutions are reversions, in agreement with earlier reports on reversion of CTL escapes (Allen et al., 2005; Li et al., 2007). This tendency to revert increases with the level of conservation of the site, suggesting a quantitative relationship between fitness cost and conservation. While reversion is particularly prevalent in acute infection (Li et al., 2007), we show that reversion is not limited to early infection but happens throughout chronic infection.

The bias towards reversion results in a two- to three-fold reduction of the long-term evolutionary rate of HIV, a trend that is reinforced by selection during transmission (Carlson et al., 2014; Sagar et al., 2009). Inter-individual evolutionary rates of HIV-1 are two to six times lower than intra-individual rates, and a number of possible mechanisms have been suggested to explain this discrepancy (Lythgoe and Fraser, 2012). Our results strongly indicate that most of this mismatch can be explained by steady reversion during infection; other factors such as retrieval of ‘stored’ latent variants or stage-specific selection might also contribute to the rate mismatch (Immonen and Leitner, 2014; Lythgoe and Fraser, 2012).

The high rate of reversion has implications for phylogenetic dating. Given the five-fold excess of reverting minor variation, reversion would balance divergence once the typical distance from the consensus sequence equals 17%, corresponding to a nucleotide diversity of about 30%; this is remarkably close to the actual divergence between HIV-1 groups M, N and O (Li et al., 2015). On longer distances, this simple argument will have to be modified due to compensatory mutations resulting in gradual shift of the preferred state at some positions; nonetheless, it indicates a dramatic slowing down of divergence at a scale of the HIV-1-SIVcpz divergence. This apparent deceleration of evolution could explain the contradictory findings of attempts to date the age of HIV-1 and SIV (Worobey et al., 2010). The strong and lasting preference for specific nucleotides needs to be accounted for in phylogenetic analysis, as has recently been shown using experimentally determined fitness landscapes of influenza virus proteins (Bloom, 2014).

Our observations are consistent with results by Ashenberg et al. (2013); Doud et al. (2015), who showed that amino acid preferences are mostly conserved between related influenza strains. Similarly, divergence between HIV-1 subtypes is small enough that epistatic interactions have not yet changed the majority of the preferred states. With increasing evolutionary distance between strains, the molecular context and preferred amino acid at a particular position is more and more likely to differ (Pollock et al., 2012). Nevertheless, a subset of amino acids preferences are conserved almost universally (Risso et al., 2015) and enable sequence based homology search.

The concordance between intrahost variation and patterns of conservation across HIV-1 group M hints at universal fitness costs of mutations. Recently, cross-sectional conservation has been used as a proxy for fitness costs in models of HIV-1 fitness landscapes (Dahirel et al., 2011; Ferguson et al., 2013; Mann et al., 2014). Since reproducible intrapatient diversity likely reflects fitness costs of mutations in vivo, our results provide a direct justification for this approach. However, nine patients are insufficient to extend this analysis to fitness interactions between mutations.

One limitation of this study was the availability of samples from patients with sufficiently long follow-up without therapy after a well-defined time of infection. The majority of patients were MSM infected with subtype B virus. Thus, we cannot exclude that the aspects of HIV-1 evolution that we have investigated may differ between transmission routes or HIV-1 subtypes. While substitution and recombination errors of our optimized protocol for HIV-1 RNA extraction and an RT-PCR are low, the other main limitation was that the number of available template molecules was small in some samples (see also Iyer et al., 2015). In principle, the Primer ID method, which labels and resequences each individual template, allows quantification of templates and almost complete elimination of experimental substitution and recombination errors (Jabara et al., 2011). However, we are not aware of a Primer ID protocol for genome-wide sequencing, which was essential to our study.

The analysis and means of data sharing of next-generation sequencing data from viral populations is still in its infancy. Raw reads require substantial post-processing before the data can be used to answer biological questions and technology is changing rapidly such that no standardized pipelines exist. To facilitate the exploration and further analysis of our data, we have developed a web server that allows to browse characteristics of individual patients, e.g., graph the frequencies of single nucleotide polymorphism through time, and download different aspects of the data in convenient format. It is, for example, possible to select an arbitrary region of the genome with length <500 bp, such as the frequently investigated V3 region, and extract an alignment of haplotypes covering this region along with their frequency from all time points belonging to one patient. We hope that the convenient access to processed data will facilitate follow-up studies on other aspects of viral evolution.

HIV-1 and other microbial populations evolve in a constant struggle between adaptation to a changing environment and maintenance of functionality. Large mutation rates and population sizes generate standing genetic diversity that is limited by the fitness costs rather than mutation rates. Hence the limiting factor for adaptation is not generating the useful mutations, but combining multiple mutations necessary to survive – e.g. escape mutations and reversions – and pruning deleterious mutations. In HIV-1, this process is facilitated by frequent recombination. We expect that the systematic exploration of sequence space, the reproducible patterns of minor variation, and frequent reversion will be characteristic of other RNA viruses. Properties of linkage between mutation will differ since mechanisms of recombination are diverse. But even though selective forces, recombination, and time scales will vary among different microbial populations, theoretical models of rapid adaptation population have shown that many features of the evolutionary dynamics are independent of the specific system (Fisher, 2013; Neher, 2013). Intrapatient evolution of HIV-1 is a unique opportunity to study this evolutionary dynamics directly in vivo.

1. 1. Zanini F
   2. Brodin J
   3. Thebo L
   4. Lanz C
   5. Bratt G
   6. Albert J
   7. Neher RA

   (2015) Whole genome deep sequencing of HIV-1

   Publicly available at the EBI European Nucleotide Archive (Accession no: PRJEB9618).

   http://www.ebi.ac.uk/ena/data/view/PRJEB9618
2. 1. Zanini F
   2. Brodin J
   3. Thebo L
   4. Lanz C
   5. Bratt G
   6. Albert J
   7. Neher RA

   (2015) Population genomics of intrapatient HIV-1 evolution

   Interactive web application facilitating exploration and access to the data.

   http://hiv.tuebingen.mpg.de

1. 1. Allen TM
   2. Altfeld M
   3. Geer SC
   4. Kalife ET
   5. Moore C
   6. O'Sullivan KM
   7. DeSouza I
   8. Feeney ME
   9. Eldridge RL
   10. Maier EL
   11. Kaufmann DE
   12. Lahaie MP
   13. Reyor L
   14. Tanzi G
   15. Johnston MN
   16. Brander C
   17. Draenert R
   18. Rockstroh JK
   19. Jessen H
   20. Rosenberg ES
   21. Mallal SA
   22. Walker BD

   (2005) Selective escape from CD8+ t-cell responses represents a major driving force of human immunodeficiency virus type 1 (hIV-1) sequence diversity and reveals constraints on HIV-1 evolution

   Journal of Virology 79:13239–13249.

   https://doi.org/10.1128/JVI.79.21.13239-13249.2005

   • Google Scholar
2. 1. Ashenberg O
   2. Gong LI
   3. Bloom JD

   (2013) Mutational effects on stability are largely conserved during protein evolution

   Proceedings of the National Academy of Sciences of the United States of America 110:21071.

   https://doi.org/10.1073/pnas.1314781111

   • Google Scholar
3. 1. Bar KJ
   2. Tsao Chun-yen
   3. Iyer SS
   4. Decker JM
   5. Yang Y
   6. Bonsignori M
   7. Chen X
   8. Hwang K-K
   9. Montefiori DC
   10. Liao H-X
   11. Hraber P
   12. Fischer W
   13. Li H
   14. Wang S
   15. Sterrett S
   16. Keele BF
   17. Ganusov VV
   18. Perelson AS
   19. Korber BT
   20. Georgiev I
   21. McLellan JS
   22. Pavlicek JW
   23. Gao F
   24. Haynes BF
   25. Hahn BH
   26. Kwong PD
   27. Shaw GM
   28. Trkola A

   (2012) Early low-titer neutralizing antibodies impede HIV-1 replication and select for virus escape

   PLoS Pathogens 8:e1002721.

   https://doi.org/10.1371/journal.ppat.1002721

   • Google Scholar
4. 1. Batorsky R
   2. Kearney MF
   3. Palmer SE
   4. Maldarelli F
   5. Rouzine IM
   6. Coffin JM

   (2011) Estimate of effective recombination rate and average selection coefficient for HIV in chronic infection

   Proceedings of the National Academy of Sciences of the United States of America 108:5661–5666.

   https://doi.org/10.1073/pnas.1102036108

   • Google Scholar
5. 1. Bernardin F
   2. Kong D
   3. Peddada L
   4. Baxter-Lowe LA
   5. Delwart E

   (2005) Human immunodeficiency virus mutations during the first month of infection are preferentially found in known cytotoxic t-lymphocyte epitopes

   Journal of Virology 79:11523–11528.

   https://doi.org/10.1128/JVI.79.17.11523-11528.2005

   • Google Scholar
6. 1. Bloom JD

   (2014) An experimentally determined evolutionary model dramatically improves phylogenetic fit

   Molecular Biology and Evolution 31:1956.

   https://doi.org/10.1093/molbev/msu173

   • Google Scholar
7. Software

   1. Bostock M

   (2015)

   D3.js

   D3.js.
8. 1. Brockman MA
   2. Brumme ZL
   3. Brumme CJ
   4. Miura T
   5. Sela J
   6. Rosato PC
   7. Kadie CM
   8. Carlson JM
   9. Markle TJ
   10. Streeck H
   11. Kelleher AD
   12. Markowitz M
   13. Jessen H
   14. Rosenberg E
   15. Altfeld M
   16. Harrigan PR
   17. Heckerman D
   18. Walker BD
   19. Allen TM
   20. Rosato PC
   21. Kadie CM
   22. Carlson JM
   23. Markle TJ
   24. Streeck H
   25. Kelleher AD
   26. Markowitz M

   (2010) Early selection in gag by protective HLA alleles contributes to reduced HIV-1 replication capacity that may be largely compensated for in chronic infection

   Journal of Virology 84:11937.

   https://doi.org/10.1128/JVI.01086-10

   • Google Scholar
9. 1. Brodin J
   2. Krishnamoorthy M
   3. Athreya G
   4. Fischer W
   5. Hraber P
   6. Gleasner C
   7. Green L
   8. Korber B
   9. Leitner T

   (2013) A multiple-alignment based primer design algorithm for genetically highly variable DNA targets

   BMC Bioinformatics 14:255.

   https://doi.org/10.1186/1471-2105-14-255

   • Google Scholar
10. 1. Brown AJL

    (1997) Analysis of HIV-1 env gene sequences reveals evidence for a low effective number in the viral population

    Proceedings of the National Academy of Sciences of the United States of America 94:1862–1865.

    https://doi.org/10.1073/pnas.94.5.1862

    • Google Scholar
11. 1. Carlson JM
    2. Schaefer M
    3. Monaco DC
    4. Batorsky R
    5. Claiborne DT
    6. Prince J
    7. Deymier MJ
    8. Ende ZS
    9. Klatt NR
    10. DeZiel CE
    11. Lin TH
    12. Peng J
    13. Seese AM
    14. Shapiro R
    15. Frater J
    16. Ndung'u T
    17. Tang J
    18. Goepfert P
    19. Gilmour J
    20. Price MA
    21. Kilembe W
    22. Heckerman D
    23. Goulder PJ
    24. Allen TM
    25. Allen S
    26. Hunter E

    (2014) HIV transmission. selection bias at the heterosexual HIV-1 transmission bottleneck

    Science 345:1254031.

    https://doi.org/10.1126/science.1254031

    • Google Scholar
12. 1. Cock PJA
    2. Antao T
    3. Chang JT
    4. Chapman BA
    5. Cox CJ
    6. Dalke A
    7. Friedberg I
    8. Hamelryck T
    9. Kauff F
    10. Wilczynski B
    11. de Hoon MJL

    (2009) Biopython: freely available python tools for computational molecular biology and bioinformatics

    Bioinformatics 25:1422–1423.

    https://doi.org/10.1093/bioinformatics/btp163

    • Google Scholar
13. 1. Cohen MS
    2. Shaw GM
    3. McMichael AJ
    4. Haynes BF

    (2011) Acute HIV-1 infection

    The New England Journal of Medicine 364:1943.

    https://doi.org/10.1056/NEJMra1011874

    • Google Scholar
14. 1. Dahirel V
    2. Shekhar K
    3. Pereyra F
    4. Miura T
    5. Artyomov M
    6. Talsania S
    7. Allen TM
    8. Altfeld M
    9. Carrington M
    10. Irvine DJ
    11. Walker BD
    12. Chakraborty AK

    (2011) Coordinate linkage of HIV evolution reveals regions of immunological vulnerability

    Proceedings of the National Academy of Sciences of the United States of America 108:11530–11535.

    https://doi.org/10.1073/pnas.1105315108

    • Google Scholar
15. 1. Di Giallonardo F
    2. Zagordi O
    3. Duport Y
    4. Leemann C
    5. Joos B
    6. Künzli-Gontarczyk M
    7. Bruggmann R
    8. Beerenwinkel N
    9. Günthard HF
    10. Metzner KJ
    11. Wainberg M

    (2013) Next-generation sequencing of HIV-1 RNA genomes: determination of error rates and minimizing artificial recombination

    PLoS ONE 8:e74249.

    https://doi.org/10.1371/journal.pone.0074249

    • Google Scholar
16. 1. Doud MB
    2. Ashenberg O
    3. Bloom JD

    (2015) Site-specific amino acid preferences are mostly conserved in two closely related protein homologs

    Molecular Biology and Evolution 32:2944–2960.

    https://doi.org/10.1093/molbev/msv167

    • Google Scholar
17. 1. Döring A
    2. Weese D
    3. Rausch T
    4. Reinert K

    (2008) SeqAn an efficient, generic C++ library for sequence analysis

    BMC Bioinformatics 9:11.

    https://doi.org/10.1186/1471-2105-9-11

    • Google Scholar
18. 1. Elena SF
    2. Lenski RE

    (2003) Microbial genetics: evolution experiments with microorganisms: the dynamics and genetic bases of adaptation

    Nature Reviews Genetics 4:457–469.

    https://doi.org/10.1038/nrg1088

    • Google Scholar
19. 1. Ferguson AL
    2. Mann JK
    3. Omarjee S
    4. Ndung’u T
    5. Walker BD
    6. Chakraborty AK

    (2013) Translating HIV sequences into quantitative fitness landscapes predicts viral vulnerabilities for rational immunogen design

    Immunity 38:606–617.

    https://doi.org/10.1016/j.immuni.2012.11.022

    • Google Scholar
20. 1. Fiebig EW
    2. Wright DJ
    3. Rawal BD
    4. Garrett PE
    5. Schumacher RT
    6. Peddada L
    7. Heldebrant C
    8. Smith R
    9. Conrad A
    10. Kleinman SH
    11. Busch MP

    (2003) Dynamics of HIV viremia and antibody seroconversion in plasma donors

    AIDS 17:1871–1879.

    https://doi.org/10.1097/00002030-200309050-00005

    • Google Scholar
21. 1. Fischer W
    2. Ganusov VV
    3. Giorgi EE
    4. Hraber PT
    5. Keele BF
    6. Leitner T
    7. Han CS
    8. Gleasner CD
    9. Green L
    10. Lo CC
    11. Nag A
    12. Wallstrom TC
    13. Wang S
    14. McMichael AJ
    15. Haynes BF
    16. Hahn BH
    17. Perelson AS
    18. Borrow P
    19. Shaw GM
    20. Bhattacharya T
    21. Korber BT

    (2010) Transmission of single HIV-1 genomes and dynamics of early immune escape revealed by ultra-deep sequencing

    PloS One 5:e12303.

    https://doi.org/10.1371/journal.pone.0012303

    • Google Scholar
22. 1. Fisher DS

    (2013)

    Asexual evolution waves: fluctuations and universality

    JSTAT 2013:P01011.

    • Google Scholar
23. Book

    1. Foley B
    2. Leitner T
    3. Apetrei C
    4. Hahn B
    5. Mizrachi I
    6. Mullins J
    7. Rambaut A
    8. Wolinsky S
    9. Korber B

    (2013)

    Theoretical Biology and Biophysics Group

    Los Alamos: Los Alamos National Laboratory.

    • Google Scholar
24. 1. Friedrich TC
    2. Dodds EJ
    3. Yant LJ
    4. Vojnov L
    5. Rudersdorf R
    6. Cullen C
    7. Evans DT
    8. Desrosiers RC
    9. Mothé BR
    10. Sidney J
    11. Sette A
    12. Kunstman K
    13. Wolinsky S
    14. Piatak M
    15. Lifson J
    16. Hughes AL
    17. Wilson N
    18. O'Connor DH
    19. Watkins DI

    (2004) Reversion of CTL escape-variant immunodeficiency viruses in vivo

    Nature Medicine 10:275.

    https://doi.org/10.1038/nm998

    • Google Scholar
25. 1. Frost SDW
    2. Nijhuis M
    3. Schuurman R
    4. Boucher CAB
    5. Brown AJL

    (2000) Evolution of lamivudine resistance in human immunodeficiency virus type 1-infected individuals: the relative roles of drift and selection

    Journal of Virology 74:6262–6268.

    https://doi.org/10.1128/JVI.74.14.6262-6268.2000

    • Google Scholar
26. 1. Gaines H
    2. von Sydow M
    3. Pehrson PO
    4. Lundbegh P

    (1988) Clinical picture of primary HIV infection presenting as a glandular-fever-like illness

    BMJ 297:1363–1368.

    https://doi.org/10.1136/bmj.297.6660.1363

    • Google Scholar
27. 1. Gall A
    2. Ferns B
    3. Morris C
    4. Watson S
    5. Cotten M
    6. Robinson M
    7. Berry N
    8. Pillay D
    9. Kellam P

    (2012) Universal amplification, next-generation sequencing, and assembly of HIV-1 genomes

    Journal of Clinical Microbiology 50:3838–3844.

    https://doi.org/10.1128/JCM.01516-12

    • Google Scholar
28. 1. Goonetilleke N
    2. Liu MKP
    3. Salazar-Gonzalez JF
    4. Ferrari G
    5. Giorgi E
    6. Ganusov VV
    7. Keele BF
    8. Learn GH
    9. Turnbull EL
    10. Salazar MG
    11. Weinhold KJ
    12. Moore S
    13. Letvin N
    14. Haynes BF
    15. Cohen MS
    16. Hraber P
    17. Bhattacharya T
    18. Borrow P
    19. Perelson AS
    20. Hahn BH
    21. Shaw GM
    22. Korber BT
    23. McMichael AJ
    24. CHAVI Clinical Core B

    (2009) The first t cell response to transmitted/founder virus contributes to the control of acute viremia in HIV-1 infection

    Journal of Experimental Medicine 206:1253–1272.

    https://doi.org/10.1084/jem.20090365

    • Google Scholar
29. 1. Hedskog C
    2. Mild M
    3. Jernberg J
    4. Sherwood E
    5. Bratt G
    6. Leitner T
    7. Lundeberg J
    8. Andersson B
    9. Albert J
    10. Nixon DF

    (2010) Dynamics of HIV-1 quasispecies during antiviral treatment dissected using ultra-deep pyrosequencing

    PLoS ONE 5:e11345.

    https://doi.org/10.1371/journal.pone.0011345

    • Google Scholar
30. 1. Henn MR
    2. Boutwell CL
    3. Charlebois P
    4. Lennon NJ
    5. Power KA
    6. Macalalad AR
    7. Berlin AM
    8. Malboeuf CM
    9. Ryan EM
    10. Gnerre S
    11. Zody MC
    12. Erlich RL
    13. Green LM
    14. Berical A
    15. Wang Y
    16. Casali M
    17. Streeck H
    18. Bloom AK
    19. Dudek T
    20. Tully D
    21. Newman R
    22. Axten KL
    23. Gladden AD
    24. Battis L
    25. Kemper M
    26. Zeng Q
    27. Shea TP
    28. Gujja S
    29. Zedlack C
    30. Gasser O
    31. Brander C
    32. Hess C
    33. Günthard HF
    34. Brumme ZL
    35. Brumme CJ
    36. Bazner S
    37. Rychert J
    38. Tinsley JP
    39. Mayer KH
    40. Rosenberg E
    41. Pereyra F
    42. Levin JZ
    43. Young SK
    44. Jessen H
    45. Altfeld M
    46. Birren BW
    47. Walker BD
    48. Allen TM

    (2012) Whole genome deep sequencing of HIV-1 reveals the impact of early minor variants upon immune recognition during acute infection

    PLoS Pathogens 8:e1002529.

    https://doi.org/10.1371/journal.ppat.1002529

    • Google Scholar
31. 1. Herbeck JT
    2. Nickle DC
    3. Learn GH
    4. Gottlieb GS
    5. Curlin ME
    6. Heath L
    7. Mullins JI

    (2006) Human immunodeficiency virus type 1 env evolves toward ancestral states upon transmission to a new host

    Journal of Virology 80:1637.

    https://doi.org/10.1128/JVI.80.4.1637-1644.2006

    • Google Scholar
32. 1. Hoof I
    2. Peters B
    3. Sidney J
    4. Pedersen LE
    5. Sette A
    6. Lund O
    7. Buus S
    8. Nielsen M

    (2009) NetMHCpan, a method for MHC class i binding prediction beyond humans

    Immunogenetics 61:1.

    https://doi.org/10.1007/s00251-008-0341-z

    • Google Scholar
33. 1. Hunter JD

    (2007)

    Matplotlib: a 2D graphics environment

    Computing in Science and Engineering 9:90–95.

    • Google Scholar
34. 1. Immonen TT
    2. Leitner T

    (2014) Reduced evolutionary rates in HIV-1 reveal extensive latency periods among replicating lineages

    Retrovirology 11:81.

    https://doi.org/10.1186/s12977-014-0081-0

    • Google Scholar
35. 1. Iyer S
    2. Casey E
    3. Bouzek H
    4. Kim M
    5. Deng W
    6. Larsen BB
    7. Zhao H
    8. Bumgarner RE
    9. Rolland M
    10. Mullins JI
    11. Paraskevis D

    (2015) Comparison of major and minor viral SNPs identified through single template sequencing and pyrosequencing in acute HIV-1 infection

    PLOS ONE 10:e0135903.

    https://doi.org/10.1371/journal.pone.0135903

    • Google Scholar
36. 1. Jabara CB
    2. Jones CD
    3. Roach J
    4. Anderson JA
    5. Swanstrom R

    (2011) Accurate sampling and deep sequencing of the HIV-1 protease gene using a primer ID

    Proceedings of the National Academy of Sciences of the United States of America 108:20166–20171.

    https://doi.org/10.1073/pnas.1110064108

    • Google Scholar
37. 1. Jones NA
    2. Wei X
    3. Flower DR
    4. Wong M
    5. Michor F
    6. Saag MS
    7. Hahn BH
    8. Nowak MA
    9. Shaw GM
    10. Borrow P

    (2004) Determinants of human immunodeficiency virus type 1 escape from the primary CD8+ cytotoxic t lymphocyte response

    Journal of Experimental Medicine 200:1243–1256.

    https://doi.org/10.1084/jem.20040511

    • Google Scholar
38. 1. Kawashima Y
    2. Pfafferott K
    3. Frater J
    4. Matthews P
    5. Payne R
    6. Addo M
    7. Gatanaga H
    8. Fujiwara M
    9. Hachiya A
    10. Koizumi H
    11. Kuse N
    12. Oka S
    13. Duda A
    14. Prendergast A
    15. Crawford H
    16. Leslie A
    17. Brumme Z
    18. Brumme C
    19. Allen T
    20. Brander C
    21. Kaslow R
    22. Tang J
    23. Hunter E
    24. Allen S
    25. Mulenga J
    26. Branch S
    27. Roach T
    28. John M
    29. Mallal S
    30. Ogwu A
    31. Shapiro R
    32. Prado JG
    33. Fidler S
    34. Weber J
    35. Pybus OG
    36. Klenerman P
    37. Ndung'u T
    38. Phillips R
    39. Heckerman D
    40. Harrigan PR
    41. Walker BD
    42. Takiguchi M
    43. Goulder P

    (2009) Adaptation of HIV-1 to human leukocyte antigen class I

    Nature 458:641.

    https://doi.org/10.1038/nature07746

    • Google Scholar
39. 1. Kearney M
    2. Maldarelli F
    3. Shao W
    4. Margolick JB
    5. Daar ES
    6. Mellors JW
    7. Rao V
    8. Coffin JM
    9. Palmer S

    (2009) Human immunodeficiency virus type 1 population genetics and adaptation in newly infected individuals

    Journal of Virology 83:2715.

    https://doi.org/10.1128/JVI.01960-08

    • Google Scholar
40. 1. Keele BF
    2. Giorgi EE
    3. Salazar-Gonzalez JF
    4. Decker JM
    5. Pham KT
    6. Salazar MG
    7. Sun C
    8. Grayson T
    9. Wang S
    10. Li H
    11. Wei X
    12. Jiang C
    13. Kirchherr JL
    14. Gao F
    15. Anderson JA
    16. Ping L-H
    17. Swanstrom R
    18. Tomaras GD
    19. Blattner WA
    20. Goepfert PA
    21. Kilby JM
    22. Saag MS
    23. Delwart EL
    24. Busch MP
    25. Cohen MS
    26. Montefiori DC
    27. Haynes BF
    28. Gaschen B
    29. Athreya GS
    30. Lee HY
    31. Wood N
    32. Seoighe C
    33. Perelson AS
    34. Bhattacharya T
    35. Korber BT
    36. Hahn BH
    37. Shaw GM

    (2008) Identification and characterization of transmitted and early founder virus envelopes in primary HIV-1 infection

    Proceedings of the National Academy of Sciences of the United States of America 105:7552–7557.

    https://doi.org/10.1073/pnas.0802203105

    • Google Scholar
41. 1. Kouyos RD
    2. Althaus CL
    3. Bonhoeffer S

    (2006) Stochastic or deterministic: what is the effective population size of HIV-1?

    Trends in Microbiology 14:507–511.

    https://doi.org/10.1016/j.tim.2006.10.001

    • Google Scholar
42. 1. Lemey P
    2. Rambaut A
    3. Pybus O

    (2005)

    HIV evolutionary dynamics within and among hosts

    AIDS Rev 8:125.

    • Google Scholar
43. 1. Leslie AJ
    2. Pfafferott KJ
    3. Chetty P
    4. Draenert R
    5. Addo MM
    6. Feeney M
    7. Tang Y
    8. Holmes EC
    9. Allen T
    10. Prado JG
    11. Altfeld M
    12. Brander C
    13. Dixon C
    14. Ramduth D
    15. Jeena P
    16. Thomas SA
    17. John AS
    18. Roach TA
    19. Kupfer B
    20. Luzzi G
    21. Edwards A
    22. Taylor G
    23. Lyall H
    24. Tudor-Williams G
    25. Novelli V
    26. Martinez-Picado J
    27. Kiepiela P
    28. Walker BD
    29. Goulder PJR

    (2004) HIV evolution: CTL escape mutation and reversion after transmission

    Nature Medicine 10:282–289.

    https://doi.org/10.1038/nm992

    • Google Scholar
44. 1. Li B
    2. Gladden AD
    3. Altfeld M
    4. Kaldor JM
    5. Cooper DA
    6. Kelleher AD
    7. Allen TM

    (2007) Rapid reversion of sequence polymorphisms dominates early human immunodeficiency virus type 1 evolution

    Journal of Virology 81:193–201.

    https://doi.org/10.1128/JVI.01231-06

    • Google Scholar
45. 1. Li G
    2. Piampongsant S
    3. Faria NR
    4. Voet A
    5. Pineda-Peña AC
    6. Khouri R
    7. Lemey P
    8. Vandamme AM
    9. Theys K

    (2015) An integrated map of HIV genome-wide variation from a population perspective

    Retrovirology 12:18.

    https://doi.org/10.1186/s12977-015-0148-6

    • Google Scholar
46. 1. Li H
    2. Handsaker B
    3. Wysoker A
    4. Fennell T
    5. Ruan J
    6. Homer N
    7. Marth G
    8. Abecasis G
    9. Durbin R
    10. 1000 Genome Project Data Processing Subgroup

    (2009) The sequence Alignment/Map format and SAMtools

    Bioinformatics 25:2078.

    https://doi.org/10.1093/bioinformatics/btp352

    • Google Scholar
47. 1. Little SJ
    2. Frost SD
    3. Wong JK
    4. Smith DM
    5. Pond SL
    6. Ignacio CC
    7. Parkin NT
    8. Petropoulos CJ
    9. Richman DD

    (2008) Persistence of transmitted drug resistance among subjects with primary human immunodeficiency virus infection

    Journal of Virology 82:5510.

    https://doi.org/10.1128/JVI.02579-07

    • Google Scholar
48. 1. Liu Y
    2. McNevin JP
    3. Holte S
    4. McElrath MJ
    5. Mullins JI

    (2011) Dynamics of viral evolution and CTL responses in HIV-1 infection

    PloS One 6:e15639.

    https://doi.org/10.1371/journal.pone.0015639

    • Google Scholar
49. 1. Lunter G
    2. Goodson M

    (2011) Stampy: a statistical algorithm for sensitive and fast mapping of illumina sequence reads

    Genome Research 21:936.

    https://doi.org/10.1101/gr.111120.110

    • Google Scholar
50. 1. Lythgoe KA
    2. Fraser C

    (2012) New insights into the evolutionary rate of HIV-1 at the within-host and epidemiological levels

    Proceedings. Biological Sciences / the Royal Society 279:3367.

    https://doi.org/10.1098/rspb.2012.0595

    • Google Scholar
51. 1. Maldarelli F
    2. Kearney M
    3. Palmer S
    4. Stephens R
    5. Mican J
    6. Polis MA
    7. Davey RT
    8. Kovacs J
    9. Shao W
    10. Rock-Kress D
    11. Metcalf JA
    12. Rehm C
    13. Greer SE
    14. Lucey DL
    15. Danley K
    16. Alter H
    17. Mellors JW
    18. Coffin JM

    (2013) HIV populations are large and accumulate high genetic diversity in a nonlinear fashion

    Journal of Virology 87:10313–10323.

    https://doi.org/10.1128/JVI.01225-12

    • Google Scholar
52. 1. Mann JK
    2. Barton JP
    3. Ferguson AL
    4. Omarjee S
    5. Walker BD
    6. Chakraborty A
    7. Ndung'u T
    8. Regoes RR

    (2014) The fitness landscape of HIV-1 gag: advanced modeling approaches and validation of model predictions by in vitro testing

    PLoS Computational Biology 10:e1003776.

    https://doi.org/10.1371/journal.pcbi.1003776

    • Google Scholar
53. 1. McKinney W

    (2010)

    Data structures for statistical computing in python

    Proceedings of the 9th Python in Science Conference pp. 51–56.

    • Google Scholar
54. 1. Mild M
    2. Hedskog C
    3. Jernberg J
    4. Albert J

    (2011) Performance of ultra-deep pyrosequencing in analysis of HIV-1 pol gene variation

    PloS One 6:e22741.

    https://doi.org/10.1371/journal.pone.0022741

    • Google Scholar
55. 1. Miralles R

    (1999) Clonal interference and the evolution of RNA viruses

    Science 285:1745–1747.

    https://doi.org/10.1126/science.285.5434.1745

    • Google Scholar
56. 1. Mostowy R
    2. Kouyos RD
    3. Fouchet D
    4. Bonhoeffer S
    5. Vartanian J-P

    (2011) The role of recombination for the coevolutionary dynamics of HIV and the immune response

    PLoS ONE 6:e16052.

    https://doi.org/10.1371/journal.pone.0016052

    • Google Scholar
57. 1. Moutaftsi M
    2. Peters B
    3. Pasquetto V
    4. Tscharke DC
    5. Sidney J
    6. Bui HH
    7. Grey H
    8. Sette A

    (2006) A consensus epitope prediction approach identifies the breadth of murine T(CD8+)-cell responses to vaccinia virus

    Nature Biotechnology 24:817.

    https://doi.org/10.1038/nbt1215

    • Google Scholar
58. 1. Neher RA
    2. Kessinger TA
    3. Shraiman BI

    (2013) Coalescence and genetic diversity in sexual populations under selection

    Proceedings of the National Academy of Sciences of the United States of America 110:15836–15841.

    https://doi.org/10.1073/pnas.1309697110

    • Google Scholar
59. 1. Neher RA
    2. Leitner T

    (2010) Recombination rate and selection strength in HIV intra-patient evolution

    PLoS Computational Biology 6:e1000660.

    https://doi.org/10.1371/journal.pcbi.1000660

    • Google Scholar
60. 1. Neher RA

    (2013)

    Annual review of ecology

    Evolution, and Systematics 44:195.

    • Google Scholar
61. 1. Nielsen M
    2. Lundegaard C
    3. Worning P
    4. Lauemøller SL
    5. Lamberth K
    6. Buus S
    7. Brunak S
    8. Lund O

    (2003) Reliable prediction of t-cell epitopes using neural networks with novel sequence representations

    Protein Science 12:1007.

    https://doi.org/10.1110/ps.0239403

    • Google Scholar
62. 1. Nijhuis M
    2. Boucher CAB
    3. Schipper P
    4. Leitner T
    5. Schuurman R
    6. Albert J

    (1998) Stochastic processes strongly influence HIV-1 evolution during suboptimal protease-inhibitor therapy

    Proceedings of the National Academy of Sciences of the United States of America 95:14441–14446.

    https://doi.org/10.1073/pnas.95.24.14441

    • Google Scholar
63. 1. Palmer D
    2. Frater J
    3. Phillips R
    4. McLean AR
    5. McVean G

    (2013)

    Proceedings of the Royal Society of London B

    20130696, Proceedings of the Royal Society of London B, 280.

    • Google Scholar
64. 1. Paredes R
    2. Sagar M
    3. Marconi VC
    4. Hoh R
    5. Martin JN
    6. Parkin NT
    7. Petropoulos CJ
    8. Deeks SG
    9. Kuritzkes DR

    (2009) In vivo fitness cost of the M184V mutation in multidrug-resistant human immunodeficiency virus type 1 in the absence of lamivudine

    Journal of Virology 83:2038–2043.

    https://doi.org/10.1128/JVI.02154-08

    • Google Scholar
65. 1. Pennings PS
    2. Kryazhimskiy S
    3. Wakeley J

    (2014) Loss and recovery of genetic diversity in adapting populations of HIV

    PLoS Genetics 10:e1004000.

    https://doi.org/10.1371/journal.pgen.1004000

    • Google Scholar
66. 1. Peters B
    2. Sette A

    (2005) Generating quantitative models describing the sequence specificity of biological processes with the stabilized matrix method

    BMC Bioinformatics 6:132.

    https://doi.org/10.1186/1471-2105-6-132

    • Google Scholar
67. 1. Pollock DD
    2. Thiltgen G
    3. Goldstein RA

    (2012) Amino acid coevolution induces an evolutionary stokes shift

    Proceedings of the National Academy of Sciences of the United States of America 109:E1352.

    https://doi.org/10.1073/pnas.1120084109

    • Google Scholar
68. 1. Price MN
    2. Dehal PS
    3. Arkin AP

    (2009) FastTree: computing large minimum evolution trees with profiles instead of a distance matrix

    Molecular Biology and Evolution 26:1641.

    https://doi.org/10.1093/molbev/msp077

    • Google Scholar
69. 1. Richman DD
    2. Wrin T
    3. Little SJ
    4. Petropoulos CJ

    (2003) Rapid evolution of the neutralizing antibody response to HIV type 1 infection

    Proceedings of the National Academy of Sciences of the United States of America 100:4144.

    https://doi.org/10.1073/pnas.0630530100

    • Google Scholar
70. 1. Risso VA
    2. Manssour-Triedo F
    3. Delgado-Delgado A
    4. Arco R
    5. Barroso-delJesus A
    6. Ingles-Prieto A
    7. Godoy-Ruiz R
    8. Gavira JA
    9. Gaucher EA
    10. Ibarra-Molero B
    11. Sanchez-Ruiz JM

    (2015) Mutational studies on resurrected ancestral proteins reveal conservation of site-specific amino acid preferences throughout evolutionary history

    Molecular Biology and Evolution 32:440–455.

    https://doi.org/10.1093/molbev/msu312

    • Google Scholar
71. 1. Roberts HE
    2. Hurst J
    3. Robinson N
    4. Brown H
    5. Flanagan P
    6. Vass L
    7. Fidler S
    8. Weber J
    9. Babiker A
    10. Phillips RE
    11. McLean AR
    12. Frater J
    13. Gojobori T
    14. SPARTAC trial investigators

    (2015) Structured observations reveal slow HIV-1 CTL escape

    PLOS Genetics 11:e1004914.

    https://doi.org/10.1371/journal.pgen.1004914

    • Google Scholar
72. Software

    1. Ronacher A

    (2015)

    Flask

    Flask.
73. 1. Rouzine IM
    2. Coffin JM

    (1999)

    Linkage disequilibrium test implies a large effective population number for HIV in vivo

    Proceedings of the National Academy of Sciences of the United States of America 96:10758.

    • Google Scholar
74. 1. Rouzine IM

    (2005) Evolution of human immunodeficiency virus under selection and weak recombination

    Genetics 170:7–18.

    https://doi.org/10.1534/genetics.104.029926

    • Google Scholar
75. 1. Sagar M
    2. Laeyendecker O
    3. Lee S
    4. Gamiel J
    5. Wawer MJ
    6. Gray RH
    7. Serwadda D
    8. Sewankambo NK
    9. Shepherd JC
    10. Toma J
    11. Huang W
    12. Quinn TC

    (2009) Selection of HIV variants with signature genotypic characteristics during heterosexual transmission

    The Journal of Infectious Diseases 199:580.

    https://doi.org/10.1086/596557

    • Google Scholar
76. 1. Salazar-Gonzalez JF
    2. Salazar MG
    3. Keele BF
    4. Learn GH
    5. Giorgi EE
    6. Li H
    7. Decker JM
    8. Wang S
    9. Baalwa J
    10. Kraus MH
    11. Parrish NF
    12. Shaw KS
    13. Guffey MB
    14. Bar KJ
    15. Davis KL
    16. Ochsenbauer-Jambor C
    17. Kappes JC
    18. Saag MS
    19. Cohen MS
    20. Mulenga J
    21. Derdeyn CA
    22. Allen S
    23. Hunter E
    24. Markowitz M
    25. Hraber P
    26. Perelson AS
    27. Bhattacharya T
    28. Haynes BF
    29. Korber BT
    30. Hahn BH
    31. Shaw GM

    (2009) Genetic identity, biological phenotype, and evolutionary pathways of transmitted/founder viruses in acute and early HIV-1 infection

    Journal of Experimental Medicine 206:1273–1289.

    https://doi.org/10.1084/jem.20090378

    • Google Scholar
77. 1. Schneidewind A
    2. Brumme ZL
    3. Brumme CJ
    4. Power KA
    5. Reyor LL
    6. O'Sullivan K
    7. Gladden A
    8. Hempel U
    9. Kuntzen T
    10. Wang YE
    11. Oniangue-Ndza C
    12. Jessen H
    13. Markowitz M
    14. Rosenberg ES
    15. Sekaly R-P
    16. Kelleher AD
    17. Walker BD
    18. Allen TM

    (2009) Transmission and long-term stability of compensated CD8 escape mutations

    Journal of Virology 83:3993–3997.

    https://doi.org/10.1128/JVI.01108-08

    • Google Scholar
78. 1. Shankarappa R
    2. Margolick JB
    3. Gange SJ
    4. Rodrigo AG
    5. Upchurch D
    6. Farzadegan H
    7. Gupta P
    8. Rinaldo CR
    9. Learn GH
    10. He X
    11. Huang XL
    12. Mullins JI

    (1999)

    Consistent viral evolutionary changes associated with the progression of human immunodeficiency virus type 1 infection

    Journal of Virology 73:10489.

    • Google Scholar
79. 1. Sharp PM
    2. Hahn BH

    (2011) Origins of HIV and the AIDS pandemic

    Cold Spring Harbor Perspectives in Medicine 1:a006841.

    https://doi.org/10.1101/cshperspect.a006841

    • Google Scholar
80. 1. Skar H
    2. Albert J
    3. Leitner T

    (2013) Towards estimation of HIV-1 date of infection: a time-continuous IgG-model shows that seroconversion does not occur at the midpoint between negative and positive tests

    PloS One 8:e60906.

    https://doi.org/10.1371/journal.pone.0060906

    • Google Scholar
81. 1. Smith JM
    2. Haigh J

    (1974) The hitch-hiking effect of a favourable gene

    Genetical Research 23:23.

    https://doi.org/10.1017/S0016672300014634

    • Google Scholar
82. 1. Tsibris AM
    2. Korber B
    3. Arnaout R
    4. Russ C
    5. Lo CC
    6. Leitner T
    7. Gaschen B
    8. Theiler J
    9. Paredes R
    10. Su Z
    11. Hughes MD
    12. Gulick RM
    13. Greaves W
    14. Coakley E
    15. Flexner C
    16. Nusbaum C
    17. Kuritzkes DR

    (2009) Quantitative deep sequencing reveals dynamic HIV-1 escape and large population shifts during CCR5 antagonist therapy in vivo

    PloS One 4:e5683.

    https://doi.org/10.1371/journal.pone.0005683

    • Google Scholar
83. 1. van der Walt Stéfan
    2. Colbert SC
    3. Varoquaux Gaël

    (2011)

    The NumPy array: a structure for efficient numerical computation

    Computing in Science & Engineering 13:22–30.

    • Google Scholar
84. 1. Walker B
    2. McMichael A

    (2012) The t-cell response to HIV

    Cold Spring Harbor Perspectives in Medicine 2:.

    https://doi.org/10.1101/cshperspect.a007054

    • Google Scholar
85. 1. Worobey M
    2. Telfer P
    3. Souquiere S
    4. Hunter M
    5. Coleman CA
    6. Metzger MJ
    7. Reed P
    8. Makuwa M
    9. Hearn G
    10. Honarvar S
    11. Roques P
    12. Apetrei C
    13. Kazanji M
    14. Marx PA

    (2010) Island biogeography reveals the deep history of SIV

    Science 329:1487.

    https://doi.org/10.1126/science.1193550

    • Google Scholar
86. 1. Zanini F
    2. Neher RA

    (2013) Quantifying selection against synonymous mutations in HIV-1 env evolution

    Journal of Virology 87:11843.

    https://doi.org/10.1128/JVI.01529-13

    • Google Scholar

Author details

1. Fabio Zanini

   Evolutionary Dynamics and Biophysics, Max Planck Institute for Developmental Biology, Tübingen, Germany

   Contribution

   FZ, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   No competing interests declared.

2. Johanna Brodin

   Department of Microbiology, Tumor and Cell Biology, Karolinska Institute, Stockholm, Sweden

   Contribution

   JB, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   No competing interests declared.

3. Lina Thebo

   Department of Microbiology, Tumor and Cell Biology, Karolinska Institute, Stockholm, Sweden

   Contribution

   LT, Acquisition of data, Analysis and interpretation of data

   Competing interests

   No competing interests declared.

4. Christa Lanz

   Evolutionary Dynamics and Biophysics, Max Planck Institute for Developmental Biology, Tübingen, Germany

   Contribution

   CL, Conception and design, Acquisition of data

   Competing interests

   No competing interests declared.

5. Göran Bratt

   Department of Clinical Science and Education, Stockholm South General Hospital, Stockholm, Sweden

   Contribution

   GB, Conception and design, Contributed unpublished essential data or reagents

   Competing interests

   No competing interests declared.

6. Jan Albert

   1. Department of Microbiology, Tumor and Cell Biology, Karolinska Institute, Stockholm, Sweden
   2. Department of Clinical Microbiology, Karolinska University Hospital, Stockholm, Sweden

   Contribution

   JA, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   For correspondence

   Jan.Albert@ki.se

   Competing interests

   No competing interests declared.

7. Richard A Neher

   Evolutionary Dynamics and Biophysics, Max Planck Institute for Developmental Biology, Tübingen, Germany

   Contribution

   RAN, Conception and design, Analysis and interpretation of data, Drafting or revising the article

   For correspondence

   richard.neher@tuebingen.mpg.de

   Competing interests

   RAN: Reviewing editor, eLife.

   "This ORCID iD identifies the author of this article:" 0000-0003-2525-1407

• 9,383

  views

• 1,446

  downloads

• 208

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.