Methods Ethics statement
The SHCS has been approved by the following ethical committees of all participating institutions: Kantonale Ethikkommission Bern; Ethikkommission beider Basel; comite d’ethique du ??departement de medicine de Hopitaux Universitaires de Geneve; ??^ ` commission d’ethique de la recherche clinique, Lausanne; ?comitato etico cantonale, Bellinzona; Ethikkommission des Kanton St.Gallens; and Ethik-Kommission Zurich, all Switzer?land. Written informed consent has been obtained from all participants [13].

Study population
We compared genotypic drug resistance tests from individuals included in the SHCS who failed first-line cART. The SHCS is a nationwide, multicenter, clinic-based cohort with continuous enrolment and semi-annual study visits. The last considered follow-up was the 18 October 2011. The SHCS drug resistance database is linked to the SHCS and includes .14,000 sequences from genotypic drug resistance tests performed by one of the four authorized laboratories in Switzerland [14]. Sequences are stored in SmartGene’s (Zug, Switzerland) Integrated Database Network System (IDNS version 3.6.6).to the first-line cART for at least 180 days (no viral load ,50 copies/mL). These individuals are further termed non-responders. If cART was changed before 90 days or if the last viral load during cART was undetectable, it was assumed that the treatment change was due to toxicity reasons and the next treatment was considered for analysis. We calculated the time with replicating virus after virological failure until the resistance test was performed. The time with replicating virus was defined as the time period when patients had detectable viral loads (.50 copies/mL). If viral loads changed between two measurements from undetectable to detectable or vice versa, it was assumed that viruses were replicating half of the time. We grouped individuals with ,3, 3? and .6 months with replicating virus. Characteristics were compared with Fishers exact test (categorical variables) and Wilcoxon rank-sum test (continuous variables). The loss of genotypic activity was estimated using the Stanford algorithm (version 6.1.1). The activity of a drug was defined as lost when the Stanford penalty score was $30 (http://hivdb.stanford. edu/). Drug resistance associated mutations were defined by IASUSA [15]. Minor PI mutations were not considered for analysis. Thymidine-analogue mutations (TAMs) were categorized in TAM 1 (M41L, L210W, T215Y) and TAM 2 (D67N, K70R, T215F, K219E/Q). We identified risk factors for an early accumulation of mutations (,6 months with replicating viruses). We performed univariable and multivariable logistic regression analyses. The following variables were included in the multivariable model: sex, transmission group, age, subtype, square root CD4 cell and viral load at the time when the resistance test was performed, NRTI treatment, PI/r or NNRTI use, and the year of cART initiation. Likelihood ratio tests did not indicate significant departures from linearity for continuous variables. Adherence is an additional potential confounder. Self-reported adherence is documented only since May 2003 in the SHCS, therefore a sensitivity analysis including adherence data was performed with patients who failed cART after this date [16]. Not only the time with replicating virus but also the viral load might be predictive for the number of emerging mutations, therefore copy-years viremia was used in a sensitivity analysis instead of the time with replicating virus [17]. Copy-years viremia is a way to express the amount of exposure an individual has had to the virus over a period of time (akin to pack-years of smoking). The mean viral load of two successive measurements is multiplied by the time they are apart. Statistical analyses were performed with Stata 11 SE software (StataCorp, College Station, TX, USA). All confidence intervals were two-sided, and the level of significance was set at 0.05.

Results Baseline characteristics
We included 129 patients with a virological failure on a NNRTI-containing and 99 on a PI/r-containing cART (Table 1). The fraction of non-responders (patients who did not reach ,50 copies/mL) was 20.2% and 22.2% (p = 0.745), respectively. Most baseline characteristics were similar between groups, although PI/ r-treated patients started cART later (median: 2007 vs. 2004, p,0.001). The co-administered NRTIs varied slightly. The most commonly used NRTI combination in individuals treated with NNRTIs was zidovudine (AZT) and lamivudine (3TC) (41.1%). Most PI/r-treated patients received tenofovir (TDF) with either 3TC or emtricitabine (FTC) (45.5%, p = 0.061). The median time with replicating virus was 144.5 days (IQR: 87.5?33) and 141Patient selection and statistical analysis
We did a cross-sectional analysis and restricted our study to individuals who started first-line cART with nucleoside reverse transcriptase inhibitors (NRTIs) and either a ritonavir-boosted protease inhibitor (PI/r) or a non-nucleoside reverse transcriptase inhibitor (NNRTI) and who had a genotypic drug resistance test performed after virological failure but before treatment change to second-line cART. A treatment failure was defined if at least one HIV-1 RNA was detectable ($50 copies/mL) after previous suppression (,50 copies/mL) or when individuals did not respondTable 1. Baseline characteristics of individuals who failed combination antiretroviral therapy (cART) containing a non-nucleoside reverse transcriptase inhibitor (NNRTI) or a ritonavir boosted protease inhibitor (PI/r).

Individuals were categorized by the time the resistance test was performed after virological failure: 34 and 34, 49 and 28, 46 and 37 treated with NNRTI or PI, respectively, had a resistance test performed after ,3, 3? and .6 months with replicating virus. The median time with replicating virus in the category .6 months was similar between groups: 277.3 days (IQR: 226?06.5) and 292 days (IQR: 234.5?28) for NNRTIand PI/r-treated patients (p = 0.916), respectively. The median viral load at the time when the genotypic resistance test was performed was considerably higher among patients treated with a NNRTI compared to a PI/r (HIV-1 RNA: log10 3.5 copies/mL [IQR: 2.8?.6] vs. log10 2.8 copies/mL [IQR: 2.3?3.6], p,0.001). In addition, the copy-years viremia was also substantially higher in patients treated with NNRTI (641.2 years * copies/mL [IQR: 114.5?348.1], p,0.001) compared to PI/r (216.9 years * copies/mL [IQR: 33.9?454.6]). The numbers of transmitted NRTI mutations were similar in NNRTI- and PI/r treated individuals. A resistance test had been performed prior to any cART initiation among 146 of 228 individuals. The prevalence of transmitted NRTI mutations was 1.5% and 5.0% in PI/r and NNRTI-treated individuals (p = 0.378), respectively. Restricting the analysis to patients with known baseline resistance data and without transmitted drug resistance mutations did not alter conclusions (data not shown). Adherence data was available for 157 of 169 (92.9%) individuals who failed cART after May 2003. Adherence was similar between patients treated with PI/r- or NNRTI-containing cART: 76.9% and 79.8% (p = 0.503) never missed a drug dose six months before reporting adherence.