USING PHYLOGENETICS TO ACCURATELY INFER HIV-1 TRANSMISSION DIRECTION

Inferring the direction of transmission between linked individuals living with HIV provides unparalleled power to understand the epidemiology that determines transmission. State-of-the-art approaches to infer directionality use phylogenetic ancestral state reconstruction to identify the individual in whom the most recent common ancestor of the virus populations originated. However, these methods vary in their accuracy when applied to different datasets and it is currently unclear under what circumstances inferring directionality is inaccurate and when bias is more likely. To evaluate the performance of phylogenetic ancestral state reconstruction, we inferred directionality for 112 HIV transmission pairs where the direction of transmission was known, and detailed additional information was available. Next, we fit a statistical model to evaluate the extent to which epidemiological, sampling, genetic and phylogenetic factors influenced the outcome of the inference. Third, we repeated the analysis under real-life conditions when only routinely collected data are available. We found that the inference of directionality depends principally on the topology class and branch length characteristics of the phylogeny. Specifically, directionality is most correctly inferred when the phylogenetic diversity and the minimum root-to-tip distance in the transmitter is greater than that of the recipient partner and when the minimum inter-host patristic distance is large. Similarly, under real-life conditions, the probability of identifying the correct transmitter increases from 52%--when a monophyletic-monophyletic or paraphyletic-polyphyletic tree topology is observed, when the sample size in both partners is small and when the tip closest to the root does not agree with the state at the root--to 93% when a paraphyletic-monophyletic topology is observed, when the sample size is large and when the tip closest to the root agrees with the state at the root. Our results support two conclusions. First, that discordance between previous studies in inferring transmission direction can be explained by differences in key phylogenetic properties that arise due to different evolutionary, epidemiological and sampling processes; and second that easily calculated metrics from the phylogenetic tree of the transmission pair can be used to evaluate the accuracy of inferring directionality under real-life conditions for use in population-wide studies. However, given that these methods entail considerable uncertainty, we strongly advise against using these methods for individual pair-level analysis.