Single phylogenetic reconstruction method is insufficient to clarify relationships between patient isolates in HIV-1 transmission case

Abstract

The accuracy of the molecular phylogenies has important implications for the understanding of HIV-1 evolution and transmission. Most studies to date on HIV-1 transmission using phylogenetic analysis have relied on the V3 loop region of the envelope, and to a lesser degree on fragments of the gag gene. The polymerase gene has recently been considered to hold sufficient genetic variability to allow the useful study of potential routes of transmission [1]. Stürmer and colleagues [2] demonstrated the relationship of HIV-1 strains derived from two patients using a phylogenetic reconstruction technique. Two phylogenetic trees were constructed based on the polymerase (entire protease and first 293 amino acids of reverse transcriptase) and the envelope (C2V3 regions) gene sequences using the neighbor-joining method with the Kimura-2-parameter substitution module. They compared the distribution patterns of the references and the sample sequences between the two trees, and came up with the conclusion that the polymerase gene sequence could not be used on its own to provide the relationship information between patient isolates. We agree that the polymerase gene is insufficient to clarify this relationship. However, we would like to raise a few issues with the computational protocols used that may affect the interpretation of their results. The appropriate method for phylogenetic analysis in an HIV-1 transmission case is a crucial issue that may or may not provide plausible routes of infection. We have learned from the forensic investigation of previous transmission cases that at least two phylogenetic reconstruction methods are usually utilized and are accepted by court in a pretrial hearing as meeting the criteria for the admissibility of evidence [3,4]. Several studies have suggested that many phylogenetic analysis methods, including Fitch–Margoliash, neighbor-joining, minimum evolution, maximum likelihood, maximum parsimony, an unweighted pair group method using arithmetic averages and a Fitch–Margoliash method assuming a molecular clock (KITSCH) should be considered to reach persuasive conclusions [3–7]. Therefore the conclusion reached by Stürmer et al. [2] using the neighbor-joining method alone is not absolutely correct, although there is no flaw in their phylogenetic reconstruction protocols. The conclusion of that study would be different if the sequences were analysed by other methods, i.e. maximum likelihood or maximum parsimony. In addition, we propose that tree reconstruction should be repeated with different seed numbers. An identical distribution pattern should be obtained regardless of the seed numbers chosen or the reconstruction method used. The consensus tree generated from different methods and seed numbers would provide stronger evidence than that derived from a single method with a single seed number.