Subtyping was based on a partial HIV-1 pol sequence of 987 nucleotides, encoding the whole protease and amino acids 1–230 of RT. This region was amplified by RT-polymerase chain reaction and sequenced using infrared-labelled primers learn more as previously described

[21,22]. Sequences were first analysed using the National Center for Biotechnology Information HIV-1 subtyping tool to quickly discriminate between B and non-B strains. Non-B sequences were subsequently aligned with sequences from the most recent reference data set from the Los Alamos National Laboratory website (http://hiv.lanl.gov/) using BioEdit 7.0.5 (http://www.mbio.ncsu.edu/BioEdit/bioedit.html) and ClustalX 1.83 (http://bips.u-strasbg.fr/fr/Documentation/ClustalX/). The resulting alignment was analysed with the Phylip package version 3.67 (http://evolution.genetics.washington.edu/phylip.html)

and a neighbour-joining tree was built based on the F84 substitution model. The reliability of the tree topology was assessed by bootstrapping using 1000 replicate data sets. Sequences that could not be unequivocally assigned ABT-199 mouse to a pure subtype or CRF were considered as possible recombinants and examined using Simplot 3.5.1 (http://sray.med.som.jhmi.edu/SCRoftware/simplot/) to identify the recombination pattern. Similarity plots and bootscans were generated by comparing the sequence under investigation with those of reference strains using a sliding window of 300 nucleotides with 20-nucleotide steps. Subtype assignment of each recombination

fragment was confirmed through phylogenetic analysis using the same parameters as for the whole sequences. For URFs, we further examined the degree of similarity of the pol sequences to other HIV-1 sequences, using the BLAST search engine (http://www.ncbi.nlm.nih.gov/blast/) with default settings. Standard nonparametric methods (the Wilcoxon signed-rank test) were used to compare median age, HIV-1 RNA levels and CD4 cell counts in patients with B and non-B subtypes. Categorical variables in these two groups were compared using χ2 or Fisher’s exact test. The crude and Mantel–Haenszel adjusted odds ratios of having a non-B subtype Thymidine kinase were also calculated. Univariate analysis was performed using χ2 and logistic regression. A subsequent multivariate analysis was performed on all variables, using the same tests with a full model. The Cochran–Armitage test for trend was used to determine if an association, when present, was linear. In all tests, a P-value below 0.05 was considered significant. HIV-1 subtype was determined for all patients, revealing an overall prevalence of non-B clades of 11.4% (417 of 3670 patients). Continent of origin (92.2% Europe, 4.5% Africa and 3.3% other), route of infection (35.6% heterosexual, 32.9% IDU, 26.3% MSM and 5.2% other) and gender (70.4% male) were known for 97% (n=3561), 53.5% (n=1963) and 98.1% (n=3602) of individuals, respectively.