Background:  Sequences used to define HIV-1 resistance are often applied to online databases to define subtype to improve patient care. This study aims to compare 3 common analytical tools using sequential clinical samples.   

Methods: Subtype analyses were performed using PR and RT gene sequences from samples with epidemiological evidence of genetic diversity (2002 - end Aug 2005). Three clinical comparative tools were used: Stanford://hivdb.Stanford.edu/, NCBI://www.ncbi.nlm.nih.gov/, REGA://dbpartners.stanford.edu/RegaSubtyping/. Comparative and statistical analyses were performed using SPSS.   

Results: 1646 sequences were submitted for subtyping. Subtype B=827(50.2%); non-B (NB)=800(48.6%); indeterminate=19(1.2%). All three analytical tools provided concordant results for 1206(73.3%) of sequences submitted, B=742, NB=464, including A=41, C=337, D=9, F=7, G=19, H=2, CRF02_AG=49.  REGA was unable to assign a significant number of samples n=311,18.9% relative to Stanford and NCBI (p<0.001). Of the 311, Stanford and NCBI provided concordant subtype analyses for 146(46.9%) and discordant for 165(53.1%). Reasons for non-assignment by REGA: no cluster with pure subtype, no detection of recombination (n=65:20.9%); cluster with pure subtype, detection of recombination but failure to classify as CRF (n=115:37%); cluster with CRF, detection of recombination in pure subtype but failure to classify as CRF (n=39:12.5%); cannot explain (n=92:29.6%). Of the remaining 129 results, REGA and Stanford were concordant in 12 cases, REGA and NCBI in 30, Stanford and NCBI in 25 and there were 62 discordant results across the 3 tools. Discordant results were associated with subtype A/CRF01_AE (n=48:77.4%), C (n=6:9.7%), F (n=3:4.8%), CRF10_CD (n=2:3.2%), CRF06_cpx, CRF13_cpx, J (all n=1:1.6%).

Conclusions: In 73% of cases there was good agreement between the analytical approaches. Unassignment of 18.9% of sequences in REGA probably reflects the stringency of the process: query sequences submitted to REGA are interrogated by phylogenetic analysis, bootstrap support, bootscanning analysis and phylogenetic signal detection. The NCBI tool was more subjective as the user had to interpret the graphical output produced. Stanford was easy to use but compares query sequences to approx. 94 reference sequences only, and therefore may not detect new and recombinant subtypes. Careful consideration is required when using these tools in isolation for clinical care.