Background:  Our goal is the integration of biological models of disease transmission based on phylogenetic analysis of viral sequence data into a model of HIV transmission. Here we derive time-dependent viral phylogenies from sequence data collected for drug-resistance screening to estimate the HIV transmission network.

Methods:  We analyzed 3053 sequences collected from 2140 patients during 1997-2003 for routine clinical testing at the Chelsea and Westminster Hospital. Synonymous distances were estimated using the Nei-Gojobori/JC model. Initial phylogenies were constructed by Neighbor-Joining. Maximum likelihood branch lengths were obtained by fitting the MG94xREV GDD 3x3 model to that phylogeny using HyPhy on a Linux  cluster. Detailed phylogenies were constructed using Bayesian MCMC (MrBayes), and timescaled phylogenies estimated using variable rate/relaxed clock approaches in R8S and BEAST.

Results:  We identified 542 subtype-B sequences with close similarity to at least 1 other sequence from a different patient in this same group. Analysis of pairwise synonymous distances suggested that a large number of distinct transmission clusters may be present. We analyzed 4 groups (91 sequences) in greater detail, both with MrBayes and using variable rate phylogenies. Each of the 4 groups includes several closely related transmissions. In one cluster, all but 1 of the patients appear to have been infected during the same short period. In another, staggered transmissions occur with a similar average time between initial infection and subsequence transmission, although this cluster also suggests that groups of transmissions (>2 individuals) occurred within shorter periods of time. Other possible transmission clusters were identified by searching for clusters of sequences which were of a similar depth. Of 435 individuals, 179 (41.1%) are probable members of a transmission cluster, defining clusters as 3 or more people with sequences as close in time as these. A particular feature of interest is the substantial difference in size between clusters (see the table), which has important implications for epidemiological models.

Conclusions:  Use of time-resolved phylogenies based on drug-resistance genotypes obtained from clinical monitoring has allowed a substantial reconstruction of the HIV transmission network in this community. A transmission network of this form is complementary to a sexual network, but more directly relevant for viral epidemiology.