Combination antiretroviral therapy (cART) has dramatically reduced HIV related morbidity and mortality. Potent new drugs for patients with multiple drug resistance have been introduced [1–5]. Nevertheless, virological failure in treatment-experienced patients is still a major concern and therefore HIV drug resistance testing has a key role for the optimal choice of active drugs in patients with multiple drug failure. Accordingly, current guidelines recommend resistance testing for patients with multiple drug failure, but also for newly infected individuals and for pregnant women as transmission of resistant HIV mutants to therapy naïve individuals are a rising concern [6–9].

Two technical principles are in use today for resistance testing: Genotypic resistance tests (GRT) and phenotypic resistance tests (PRT). GRT is based on population gene sequencing of defined DNA segments, typically to detect mutations, which represent at least 20% of the virus population and confer HIV-1 drug resistance [10, 11]. As a special form of genotyping, virtual PRT (vPRT) correlates genotypic data for plasma HIV-1 RNA of a candidate gene with a large database of paired biological and clinical phenotypes [12–15]. Numerous genotypic interpretation systems have become available during the past decade, which provide excellent prediction of drug response. On the other hand, comparing different algorithms, some very significant differences and opposite predictions continue to be observed for the interpretation of the impact of mutational patterns (T. Klimkait, manuscript in preparation).

PRT assesses viral expression. A special form of it, the replicative phenotypic resistance test (rPRT) utilizes several replication cycles of a recombinant infectious virus to follow viral propagation in the presence of antiretroviral drugs [16, 17]. By permitting several cycles of viral replication in vitro rPRT can detect viral minorities below one percent [18]. However, rPRT is more costly, and takes longer than GRT.

Several studies have demonstrated the clinical benefit and cost-effectiveness of GRT [19–25] compared to standard of care. This study was designed to analyse whether the dissecting, sensitive format of PRT may provide a diagnostic benefit over GRT. Analyses comparing virtual PRT to GRT have thus far not been able to document a clear clinical advantage for PRT with a higher proportion of patients achieving a suppressed viral load [14, 15, 26–30]. Our first retrospective single centre analysis of GRT combined with a highly sensitive rPRT already suggested, although statistically underpowered, that patients being switched to new cART based on drug choice from a combination of both tests tended to have better virological response than those with only GRT-based resistance information [31].

In the present study we included all available data of prospectively conducted resistance tests for patients enrolled in the much larger multicentre Swiss HIV Cohort Study (SHCS) and compared the virological outcome in patients initiating a new antiretroviral drug regimen based on results of either GRT alone or rPRT combined with GRT. The highly sensitive format of rPRT used in the SHCS allows the detection of less than 1% of resistant virus in a clinical sample with a mixed virus population [18]. We therefore explored whether the complementing information of rPRT improves patient outcome when used routinely in the clinical setting.

In this multicentre cohort study of prospectively assessed HIV-1 drug resistance in patients the addition of rPRT to GRT as compared to GRT alone showed a trend towards improved success rates for treatment with increasing levels of antiretroviral pre-treatment. In the subgroup of heavily pre-treated patients with multiple drug failures the addition of rPRT significantly improved virological outcome with a 70% increased odds for achieving treatment success after adjusting for confounders and SHCS centre. The clinical benefit of resistance testing must be critically evaluated in its clinical context. Between 1999 and 2007 resistance declined overall in the SHCS [38]. This decrease was mainly driven by two mechanisms, the loss to follow-up or death of high-risk patients exposed to mono- or dual-nucleoside reverse transcriptase inhibitor therapy and the continued enrolment of low risk patients who were taking cART that contained boosted protease inhibitors or NNRTI as first-line therapy.

From a virologist's point of view the add-on benefit of rPRT is of particular relevance in patients with multiple drug failure and archived mutations. In patients with multiple virological drug failure and multiple therapy changes the genomic complexity of deposited HIV sequences increases. Growing resistance coincides with a rise of viral quasispecies [18, 39, 40]. Although GRT provides relevant information to clinicians for optimal drug choices it has important limitations for mixed virus populations and for the detection of emerging or residual virus variants. The interpretation of a GRT results becomes particularly challenging for therapy-experienced patients where specific mutations have to be assigned to distinct HIV genomes. Today several unique rule based algorithms are very well established e.g. Stanford (HIV drug resistance database, Stanford university; USA), ANRS (National Agency for AIDS Research, France), Rega (Institute for Medical Research and University Hospitals, Belgium), and G2P (geno2pheno system, Max-Planck-Institute, Germany). However, the agreement among these algorithms tends to decrease in parallel to the growing complexity of viral mutation patterns [41]. Interpretation and choice of the optimal regimen becomes particularly difficult for heavily pretreated patients, where the clinical treatment options become scarce or in situations where drug pressure after longer treatment interruptions is absent.

One intrinsic potential limitation of this study lies in the fact that the choice of requesting GRT or GRT+PRT was largely centre-dependent, thereby introducing a possible centre bias and depending on any centre's preference for certain regimens. However by using a multilevel or hierarchical model, the effect of resistance testing was estimated after adjusting for the measured or unmeasured effect of centre.

Our study has several strengths. We used stringent and very conservative criteria to define the target population of this observational cohort study. The cohort represents an unselected population of HIV-infected individuals, which is larger than the populations included in previously published observational studies and clinical trials. In addition this study includes a relatively large number of females and IDU making it more representative. We were able to include important variables in our analysis known to be related to virological outcome. For example, our data indicate that the study population included a relatively large group of patients with adherence problems in comparison to the general patient population in the SHCS. Roughly one third of patients had indicated that they had missed more than 2 doses in the previous four weeks and one fifth of patients stated to have missed more than 2 consecutive doses. Thus, our findings should be interpreted in the context of a patient population that poses real challenges for optimal clinical management and most likely makes it more difficult to demonstrate an add-on benefit of rPRT to GRT than one would have seen in a clinical trial with a more selected patient population.

High molecular diversity of HIV is a common problem in long-term treated, highly therapy experienced patients. In such patients with complex resistances rPRT is able to assign resistances to several co-existing viruses rather than placing the gene mutations onto one single virtual virus genome as done by GRT, and thus gives more conservative estimates of antiretroviral drug resistance. In contrast, GRT in such patients leads to over-simplification by indicating cross resistance patterns per viral genome that tend in reality to be more complex. As a consequence, GRT may in these patients over-interpret the viral resistance and erroneously indicate to clinicians and their patients a lower number of remaining treatment options. The higher percentage in the GRT group of PI containing regimes, paralleled by a reduction in NNRTI-containing ART combinations suggests two likely reasons: on the one hand a centre effect for the favoured therapy-scheme, on the other, due to their low genetic barrier, the prompt stop of NNRTIs after virological failure. This study did, however, not assess whether or not this decision was always based on the formal demonstration of predominant NNRTI-related resistance mutations.

Previous studies on GRT and PRT have investigated virological outcome with mixed findings, either resulting in non-significant gains [19–21] or in only a small benefit [25] and cost savings [22–24] from GRT-guided therapy adjustment. Moreover, several clinical trials have investigated different types of PRT, but until now a possible advantage of providing PRT remains unclear. In a randomized trial of heavily pre-treated patients PRT did not result in an intention to treat analysis in a greater proportion of virological suppression when compared to standard of care. In the as treated analysis a statistically significant 16% difference of patients with less than 400 c/mL was found [26]. In another randomized trial by Meynard et al. a less sensitive single-cycle PRT was used. The combination of PRT with GRT compared to GRT did not result in a higher rate of HIV-1 suppression [28]. GRT plus vPRT was compared to GRT in a large Australian trial but the investigators found at 48 weeks no difference in virological outcome [27]. In one trial patients with drug failure were randomized either to access to routine GRT, vPRT, or "no testing". No difference in the time to virological failure was found between groups. However, in the subgroup of patients with more than four previous failures patients with vPRT did have significantly prolonged time to treatment failure [29]. In another randomized controlled trial by Dunn et al. there was no difference between GRT alone and GRT plus PRT [30]. Both trial groups worked with a less sensitive method of PRT compared to the one used in this study.

Evidence from clinical trials investigating whether GRT, PRT or the combination of both improve virological outcome is limited. Subgroup analyses from trials suggest that PRT may improve clinical outcome in patients with multiple previous failure. Our findings are in line with those trials. Our study shows that rPRT, when added to GRT, may indeed lead to improved virological outcome, particularly in the population of heavily pre-treated patients. This is corroborated by the finding that a therapy status "no treatment at time of testing" is significantly less frequent for the GRT + rPRT group. This indicates that GRT + rPRT was more often chosen in complex therapy situations. As scientific basis: replicative PRT functionally dissects resistant virus populations and may reveal remaining viable regimens, particularly in patients with limited options and thereby increase the chance for virological success. In contrast GRT tends to place for analysis all mutations on one viral "consensus" genome.

Our study suggests that a stepwise testing strategy adding replicative PRT for patients with multiple drug failure provides benefit for better clinical decision-making. Further studies are needed to confirm whether this strategy translates into improved virological outcome in patients with limited treatment options.