HIV-1 evolves by introducing mutations (substitutions, indels, recombination) through a “sloppy” replication mechanism, mainly due to the unfaithful replication by the viral reverse transcriptase. These mutations are often deleterious 
or otherwise detrimental to virus fitness 
. However, some mutants have an advantage as they may allow escape from immune surveillance 
or more effective infection of certain tissue compartments or cell types, such as cells in the brain or the genital tract 
or naïve CD4+ T-cells, which express CXCR4 
. Here we show that in addition to the mutational processes, HIV-1 can alter its population structure by frequency shifts among subpopulations. Because we analyzed a relatively small number of sequences per time point, we were careful to include the sampling into our analysis method. Over short time (days, weeks, months) these fluctuations were consistent with a constant population size, and most mutations that occurred at this time scale were neutral or only weakly selected. On longer time scales we noticed that the fluctuations became significant movements.
Here, we focused on short-term evolutionary processes (days, weeks, and months), whereas earlier studies as well as the generation of, and escape from, neutralizing antibody responses involve time frames of months to years 
. Clinically, our patient was classified as a slow disease progressor. Genetically, the virus population in our patient was described by co-existing subpopulations. Thus, it is interesting to compare the HIV population genetics of our patient to previously published patients with normal and slow disease progression. In a study by Shankarappa et al, 5 patients had slow disease progression (p2, p3, p7, p9, p11) and 5 had normal progression (p1, p5, p6, p8) 
. These patients were followed over many years, but interestingly over a sampling period equivalent to ours (522 days, but with fewer samples), patients in both clinical groups showed subpopulation structure qualitatively similar to our patient (Figure S6
). Thus, the short-term evolution we study here is likely representative for many patients regardless of disease progression rate.
One might have expected that the persisting subpopulations found in this patient were controlled by balancing selection 
. Directional selection would have favored the fittest of the subpopulations and it would have been unexpected to see them coexist for so long, let alone to have several well separated subpopulations, which implies that they have existed for longer than the study period. Hence, some type of frequency-dependent selection, where the fitness of a variant/subpopulation is dependent on its relative frequency, would be the alternative hypothesis to neutral drift. Here we show that although the immune system partly controls virus replication during the chronic phase of the disease, particularly well in a slow progressor, and where one would expect escape mutants to dominate in env
, the genetic evolution is consistent with a neutral process, at least over the time period studied here. In agreement with this, it was recently shown that genetic drift was a main contributor to HIV evolution in culture 
. Similarly, in several other virus systems with large population sizes and high mutation rates, where deterministic processes are expected, genetic drift was shown to have a larger effect than expected 
. Furthermore, stochastic evolution during drug treatment of HIV-1 has previously been demonstrated 
. Thus, also in vivo
evolution of HIV-1 during the chronic phase may be largely described by neutral and stochastic processes. We speculate that this might be due to that the immune system “hits” all subpopulations with near equal efficiency.
Our tests for neutrality of the subpopulation frequency fluctuations are of necessity informal. A more formal procedure would assess the likelihood of the data under neutral models with varying Ne
and compare with models that additionally include either balancing or directional selection. There exist methods for estimating Ne
from multi-allele temporal data (e.g. 
), as well as methods for inferring directional selection from two-allele temporal data (e.g. 
). However, we are not aware of likelihood methods that include balancing selection and multiple alleles, and their development is beyond the scope of the present study. Hence, we have relied on a less formal method that may not have optimal power, but nevertheless is informative. In addition, our Ne
estimate from sequence data were in the order of previously estimated Ne
of HIV-1 in chronic infection 
, however, subpopulation structure or non-neutral evolution may bias these estimates, therefore we included a large range of plausible Ne
in our test of neutrality (Ne
We have sampled free HIV-1 viral particles in plasma but we do not know where these virions were produced. The degree of compartmentalization of HIV-1 replication is uncertain; some researchers have found evidence of compartmentalization whereas others have not 
. However, in untreated patients most virus in plasma is produced by short-lived activated CD4+ T-lymphocytes 
and there is no or limited compartmentalization between virus in plasma and lymphocytes 
. Thus, the plasma virus population should be competing for the same resources, which would justify our analysis of whether balancing selection exists. However, we cannot exclude that the frequency fluctuations we see may be due to differential production from different compartments. The subpopulations were present in actively replicating virus since 1) the subpopulations were detected over time at high frequencies (i.e., detected in 7–11 single molecules per time point), 2) the molecules sequenced must represent virions which were replication competent at least in the previous generation, and 3) subpopulation s6 evolved at a measurable evolutionary rate.
It was interesting to note that PNGS were significantly over-represented among positively selected sites. Glycosylation and movement of glycans have been suggested to be an important immune escape mechanism of HIV-1 
. Our data are compatible with such a scenario, which suggests that immune escape and positive selection on PNGS may have contributed to the evolution of the genetic subpopulations in our patient. Previously, it was demonstrated quantitatively that a wide range in strength of the autologous neutralizing antibody response between patients and corresponding differences in the impact on the viral population 
. The fact that we observed positive selection on PNGS, and earlier studies have shown consecutive replacement HIV-1 env
and continuous neutralization escape 
, do not contradict our observation that evolution in our patient overall was consistent with a neutral model of evolution.
The estimation of the evolutionary rate may be misled if phylogenetic relationships are not accounted for. This may occur in a naïve analysis where genetic distances between two (or more) time points are directly compared. Without accounting for the phylogeny, especially when the population is divided into clear subpopulations, frequency shifts between existing variants will masquerade as de novo
mutations. We show that in env
, which has the highest evolutionary rate in the HIV-1 genome, the number of de novo
mutations accumulated a significant distance after about one month within a subpopulation. Hence, this suggests that sampling more frequently than this may not be useful to estimate the evolutionary rate in patients during the chronic phase. This is in good agreement with previous estimates of significant temporal changes at 22 months based on genomic sequences of similar length (~1100 nt) in gag-pol 
, which evolves much slower than env
. Note, however, that selection during drug treatment may potentially act upon existing variants/subpopulations at a much faster rate 
, but, as we show here, during chronic, asymptomatic, untreated viral infection evolution proceeds mostly by neutral drift over shorter time frames.
In conclusion, we have performed high-frequency sampling of HIV-1 evolution in a chronically infected, untreated patient with slowly progressing disease. We shown that multiple well-separated subpopulations may persist for years, and over weeks and possibly months their frequencies remained constant. Over the time period of years, however, their fluctuations became significant, but were still consistent with a neutral model of evolution. However, sequence-based methods showed that individual sites had experienced positive selection, possibly as the subpopulations were being formed over several years. While the subpopulation frequencies fluctuated consistent with neutrality, the divergence within a subpopulation showed a temporal trend that was resolved at about one month's time.