Deep sequencing affords multiple advantages over current sequencing techniques for the detection of minority HIV variants. Co-linear sequence can be obtained in areas of significant genetic heterogeneity, characteristic of HIV env
, and at a faster speed and greater depth than conventional cloning or single genome sequencing. We developed a rigorous methodology for the filtering and validation of quantitative deep sequencing in the context of viral infections that can readily be applied to other deep sequence data sets. Quantitative deep sequencing of plasma samples from subjects failing a CCR5 antagonist-containing regimen provided orders of magnitude greater coverage than previously possible and revealed that minor V3 loop sequence variants have significant clinical implications in chronic HIV-1 infection. We directly demonstrated the previously hypothesized vast diversity of intra-patient HIV sequences, in this case within a variable determinant of viral coreceptor usage, and quantified the dynamic sequence variation over time 
Although variants using either CXCR4 or CCR5 are known to coexist in plasma, deep sequencing detected a multitude of distinct co-circulating V3 forms and highlighted the extraordinary diversity of V3 sequences available to the virus as it adapted to changing fitness environments. Variants with predicted CXCR4 usage as well as VVC-resistant variants that emerged during VVC treatment were shown to exist as uncommon or rare forms in baseline samples in three of the four subjects we studied. These rare forms differ from the dominant V3 form at multiple amino acid positions and exist at greater frequencies than the single cycle mutation rate of HIV (10−5
per base per cycle) or the observed in vivo frequencies of non-nucleoside reverse transcriptase or protease inhibitor resistance mutations (0.03% and 0.03%, respectively) in treatment-naïve HIV-infected patients 
. This finding suggests that these V3 loops are not generated and lost with each replication cycle, but in fact may be maintained as minority variants during chronic infection. The documentation of extensive V3 sequence diversity, coupled with the rapid expansion of rare variants in response to a drug selection pressure highlights the extraordinary challenge HIV diversity poses for preventive or therapeutic vaccines targeting areas of similar sequence heterogeneity. These data further provide compelling evidence that minority HIV variants present at less than 1% of the HIV-1 population in plasma are clinically relevant. The level of a minority variant that is clinically relevant most likely varies as a function of the potency of the background antiretroviral regimen, underscoring the importance of combining several potent agents when constructing antiretroviral regimens.
The reason subject 47 experienced treatment failure remains unclear. The low GSS and PSS of this subject's antiretroviral regimen, determined at baseline and week 2, suggest that limited potency of the antiretroviral regimen was the most likely explanation for failure to achieve or maintain a 1-log10
reduction in viral load. Subject adherence and plasma VVC levels were adequate, and no changes were observed in coreceptor tropism, phenotypically or genetically. Although most mutations that confer CCR5 antagonist resistance map to the V3 loop, sequence changes in other regions of env
have been noted (M Lewis, et al, 15th
Conference on Retroviruses and Opportunistic Infections, February 5–8, 2008, Boston, MA, Abstract 817) 
. It is possible that sequence changes in env outside of the V3 loop could have contributed to virologic failure, but phenotypic testing with full-length env
constructs from subject 47 did not demonstrate any loss of VVC susceptibility.
Our analysis has important limitations. Some error is inevitably introduced through these experimental methods and it is not obvious precisely which of the rare forms are the consequence of this sequencing error. Furthermore, the longitudinal nature of the data allowed tracking of frequency shifts in interesting sequences that were subject to VVC selection pressure. Because of the limited length of the sequences, homoplasy rather than shared lineages may underlie some of the observed clustering patterns in the trees, and in vitro recombination during amplification may also have contributed to some of the apparent clades 
. The trees are not expected to provide accurate renditions of the phylogenies as they are by necessity constrained by limitations imposed by the nature of the data (very short sequences, convergence in a scenario of emerging drug resistance, recombination, critical mutations occurring by insertion and deletion), and are based on thousands of short fragments with some experimental error. However, they do capture the complexity of the samples, as well as the rapid temporal shifts in the evolutionary landscape as a consequence of selective pressure. Clear evolutionary progression over time was evident in each subject. Multiple co-circulating minor CXCR4-using viruses were found to exist in chronic infection, and the virus simultaneously explored many escape routes as the drug-susceptible form was selected against. The process of escape may push the virus towards sampling unexplored regions of the sequence space. Taken together, our data demonstrate the feasibility and utility of harnessing deep sequencing platforms to comprehensively assess viral diversity, quantify minor sequence variants, and provide insight into the mechanisms of viral escape from novel CCR5 antagonists.