Trends and correlates of HIV-1 resistance among subjects failing an antiretroviral treatment over the 2003–2012 decade in Italy

Background Despite a substantial reduction in virological failures following introduction of new potent antiretroviral therapies in the latest years, drug resistance remains a limitation for the control of HIV-1 infection. We evaluated trends and correlates of resistance in treatment failing patients in a comprehensive database over a time period of relevant changes in prescription attitudes and treatment guidelines. Methods We analyzed 6,796 HIV-1 pol sequences from 49 centres stored in the Italian ARCA database during the 2003–2012 period. Patients (n = 5,246) with viremia > 200 copies/mL received a genotypic test while on treatment. Mutations were identified from IAS-USA 2013 tables. Class resistance was evaluated according to antiretroviral regimens in use at failure. Time trends and correlates of resistance were analyzed by Cochran-Armitage test and logistic regression models. Results The use of NRTI backbone regimens slightly decreased from 99.7% in 2003–2004 to 97.4% in 2010–2012. NNRTI-based combinations dropped from 46.7% to 24.1%. PI-containing regimens rose from 56.6% to 81.7%, with an increase of boosted PI from 36.5% to 68.9% overtime. In the same reference periods, Resistance to NRTIs, NNRTIs and PIs declined from 79.1% to 40.8%, from 77.8% to 53.8% and from 59.8% to 18.9%, respectively (p < .0001 for all comparisons). Dual NRTI + NNRTI and NRTI + PI resistance decreased from 56.4% to 33.3% and from 36.1% to 10.5%, respectively. Reduced risk of resistance over time periods was confirmed by a multivariate analysis. Conclusions Mutations associated with NRTIs, NNRTIs and PIs at treatment failure declined overtime regardless of specific class combinations and epidemiological characteristics of treated population. This is likely due to the improvement of HIV treatment, including both last generation drug combinations and prescription guidelines.


Background
Despite substantial reductions in AIDS-related morbidity and mortality [1,2], advancements in antiretroviral therapy have always been challenged by the development of drug resistance. Evolution of drug-resistant HIV variants remains a major limitation for the long-term control of HIV-1 infection as wide resistance to multiple drug classes is associated with clinical deterioration and death [3]. A multitude of drug-resistant strains have been reported that differ considerably in their susceptibility to the three major classes of antiretrovirals in use since the introduction of the highly active aniretroviral treatment (HAART): nucleos(t)ide reverse-transcriptase inhibitors (NRTIs), non-nucleoside reverse-transcriptase inhibitors (NNRTIs) and protease inhibitors (PIs) [4]. The prevalence of resistant strains rose to high levels in high-income countries in the past, when treatment often made of suboptimal regimens became available [5][6][7][8]. Drug resistance has emerged recently also in low/middle-income countries, as a consequence of the growing access to first-generation drug combinations coupled with the high burden of HIV infection [9,10].
Treatment failing patients appear to be the major cause of the transmission of resistant strains [11] which may compromise the efficacy of combination antiretroviral treatment (cART) in drug-naive patients [12]. Nevertheless, when the initial cART is tailored to the viral genotype to ensure full activity, patients can achieve virological responses comparable to those harbouring wild type virus [13]. Therefore, it is expected that current optimal regimens, selected on the basis of HIV pol genotype, control viral replication even in patients with primary resistance.
Available results on the prevalence of drug resistance are often difficult to compare as they differ in resistance associated mutations taken into account, timing of samples and selection of study populations. The latter could vary from all subjects on antiretroviral treatment, to patients failing ART, to subjects with available resistance test results. Moreover, these studies have estimated the prevalence of HIV drug resistance using a variety of analytical methods resulting in a wide range of estimates, ranging from 50% to 80% of subjects failing an antiretroviral treatment [5][6][7][8][14][15][16].
A reduction of acquired resistance has been recently reported in Italy until 2009 [17,18]. This may have derived from the latest prescription attitudes and, likely, to the introduction of more potent new drugs in salvage therapies. Nevertheless, considerable proportions of treated individuals are still likely to select for resistance mutations while on antiretroviral treatment, which may result in an ongoing transmission of HIV-1 resistant variants.
The aim of this study was to monitor acquired resistance to understand present trends and correlates of class resistance in subjects failing cART regimens in a multicenter Italian network based over the 2003-2012 period. Previous reports considered resistance prevalence among the whole studied population, regardless of the regimen administered at the time of resistance testing [16][17][18]. Since treatment changes can influence drug pressure on previously selected virus variants, we chose instead to evaluate resistance trends according to the antiretroviral drugs taken at time of failure.

Patients
Patients included in the study were adult HIV-infected individuals enrolled in 49 Italian clinical centres during the 2003-2012 period. All the clinical centres contributed data to the Antiretroviral Resistance Cohort Analysis (ARCA, www.hivarca.net) database, a nationwide repository used for non-profit research purposes and stored on a central server. Written informed consents had been obtained by patients. The research did not require approval from the Ethics Committees, according to the Italian law at the time when the study was conducted, since it was performed as an observational study in the context of clinical routines (art.1, Low. Decree 211/2003).

Inclusion criteria
Cases were selected according to DHHS Guidelines [19] on the basis of the concomitant detection of HIV-1 viral load over 200 copies/ml after at least 6 months of ongoing therapy and the availability of an HIV-1 genotypic test obtained while on treatment. The cART regimen was defined as any combination of three or more drugs including an NNRTI and/or a PI. When more than one sequence was available from the same subject in the same year of study, the first sequence was considered.

HIV-1 genotype and class resistance evaluation
Genotyping was based on a partial HIV-1 pol sequence including RT and protease and ranging from 1,000 to 1,280 nucleotides, depending on the sequencing protocol used at the contributing laboratory. Emergence of resistance at failure was evaluated according to the latest International AIDS Society (IAS) mutation list [20]. NRTI, NNRTI and PI class resistance was evaluated according to NRTI, NNRTI or PI based treatment used at time of virological failure.

Statistical analysis
The Cochrane-Armitage test was used to evaluate temporal trends. The crude and Mantel-Haenszel adjusted odds ratios (OR) of class resistance detection with 95% confidence interval (CI) were calculated. Univariate analysis was performed using χ 2 and logistic regression. A subsequent multivariate analysis was done on all variables, using the same tests with a full model. Analyses were done with SAS Software version 9.1.
The study included 6,796 pol sequences from subjects with viral load above 200 copies/ml while on treatment
Antiretroviral classes were administered in several combinations during the study period: the prevalence of the association of NRTIs + NNRTI decreased from 40.4% to 18.3% (p < .0001), while combinations of NRTIs + PI(s) increased from 53.4%, to 75.9% (p < .0001). The combination of all the three classes showed a significant declining trend from 6.0% to 3.2% (p < .0001) (Figure 1, Part A). The association between an NNRTI and a PI(s) increased significantly from 0.3% to 2.6% (p < .0001) in the study period.
Frequencies of specific mutations and resistance patterns are shown in Table 1 for NNRTI-and PI-based cART. Among NNRTI-based treatments, a significant decline was found for K103N/S, V106A/M, V108I, Y181C/I and G190S/A. No significant variations were detected for L100I, 188C/L/H, P225H and M230L. In PI-containing regimens, a reduction of frequency was found for D30N, L33F, M46I/L, G48V, L76V, V82A/F/T/S, I84V, N88D/S and L90M. A significant increase was detected for I50L and I54L/M. No significant variation was found for V32I, I47A/V, I50V and Q58E. Table 2 shows the uni-and multivariate logistic regression analysis investigating possible predictors of class resistance development on treatment. NRTI resistance in patients assuming any drug of this antiretroviral class was found in association with several independent predictors, including gender, risk factor, age, HIV-1 subtype, viral load at failure, previous virological failures, number of previous antiretroviral regimens, prior exposure to suboptimal NRTI therapy, associated antiretroviral class in use at failure and period of study.

Predictors of class resistance
Independent predictors of NNRTI resistance in NNRTIassuming subjects were gender, viral load at failure, previous virological failures.
Predictors of PI resistance were gender, risk factor, viral load at failure, previous virological failures, number of previous antiretroviral regimens, prior exposure to suboptimal NRTI therapy, use of TA and period of study. The usage of boosted rather than unboosted PIs was not associated with the risk of evolution on PI resistance.
The risk of resistance to any antiretroviral class was not influenced by the use of TA-including or sparing regimens or by the use of FTC rather than 3TC in the multivariate model, even though a lower prevalence of any NNRTI resistance was found in subjects assuming FTC compared to 3TC (52.6% vs. 61.2%, p = 0.012).

Discussion
Our data provide a solid evidence of a marked decrease in the prevalence of class resistance among treated individuals in Italy, as observed from a retrospective analysis including 49 clinical centres participating in the ARCA cohort during the last decade. Studies exploring resistance trends in European countries have observed an initial decrease over the 2005-2008 period [15][16][17][18]21]. However, HIV resistance to antiretrovirals still deserves continuous monitoring, extending the observation to more recent years and providing updated understanding of the correlates of resistance.
Differently from previous works [14,15] the design of our study focused on HIV drug resistance in patients assuming the corresponding drugs at the time of failure, providing a more accurate estimate with respect to using the whole pre-treated population [16][17][18]. This approach was adopted mainly to correct for the large observed variation in antiretroviral usage over time, specifically taking into account the antiretroviral regimen in use at failure.
This analysis allowed to detect a twofold and threefold decrease in resistance prevalence to NRTIs and PIs, respectively during the 2003-2012 decade. The study periods included two biennia and two triennia to balance the distribution of subjects overtime. NNRTI resistance declined less markedly, from 77.8% to 53.8%. Nevertheless, the reduction of NNRTI resistance overtime is more relevant than that observed in previous studies [18,22] even though a rapid selection of resistant variants is observed in patients failing NNRTI based regimens [23]. We observed an important reduction of specific NRTI mutations, such as K65R, L74V, M184V and TAMs in subjects administered drugs selecting for these mutations. This decrease may be due to multiple factors including the introduction of compact dual NRTI formulations and association with high genetic barrier boosted PIs. Several NNRTI mutations did not significantly vary overtime: a stable rate of selection was observed for L100I, 188C/L/H, P225H and M230L. A declining prevalence of K103N/S and Y181C/I overtime may be explained by the reduced usage of EFV and NVP which preferentially select for such mutations [24]. A similar declining trend can be observed when exploring the variations of specific PI mutations, which are largely influenced by the different variations in use of specific PIs overtime. For example, we observed an increase of I50L and I54L/M, which have been reported to be major mutations for atazanavir and darunavir, respectively [25].
Several epidemiological correlates were found to have an impact over the selection of resistance mutations only for NRTIs and PIs. An increased risk of resistance was found for sexual routes of transmission (heterosexual and homosexual) compared to IDU, in agreement with other Italian studies [17,18]. This finding may be explained by different adherence patterns in these subsets of patients as the higher rates of resistance among subjects who acquired the infection through sexual route may be due to suboptimal, even though intermediate to high, levels of adherence in this population. By contrast, among IDUs, very low levels of compliance to medical prescriptions can lead to an important reduction of drug levels, which may not result in sufficient drug selective pressure leading to resistance to NRTIs or PIs [22]. We also observed a higher risk of resistance to each class of antiretrovirals among males rather than females. This association with gender should take into account a possible lower adherence observed in female subjects [26,27]. Lower drug resistance rates were associated with plasma HIV-1 viremia above 100,000 copies/ml. This finding is different from previous observations obtained from the ARCA database [17] and it is in agreement with findings in patients undergoing resistance testing in routine clinical practice [28]. This may be due to cases of very low adherence, when low drug levels reduce both the efficacy of antiretroviral therapy and emergence of resistant strains [27]. Our observation underlines the need of genotypic testing at the very early detection of virological failure, when the HIV-1 viremia is still at a low level, as suggested by recent guidelines [19].
As expected, the occurrence of a previous virological failure increased the risk of resistance to any antiretroviral class. Other treatment history features, such as the number of cART regimens, have shown a role in the emergence of class resistance to NRTIs and PIs, but not to NNRTIs. The fact that NNRTI resistance shown no association with the temporal length of antiretroviral treatment can be explained by the characteristic pattern of resistance evolution of this antiretroviral class, whose use can be compromised by a single substitution leading to cross resistance [29].
The risk of resistance to NRTIs was higher when the patients failed NNRTI rather than PI based treatments. This is in agreement with previous findings and confirms that a higher protection against NRTI resistance is obtained by cART regimens including antiretrovirals with a higher genetic barrier [17,18,20]. However, this protection against the emergence of NRTI resistance was observed either for boosted or unboosted PIs compared to NNRTIs when the entire time period of the study was considered. Similarly, no difference in risk of PI mutations evolution could be found when comparing boosted and unboosted PIs overtime. Enven though unexpected, these observations are in agreement with data demonstrating that comparable virological outcome was achieved when the same PI, boosted or unboosted, was used in antiretroviral regimens in specific clinical settings and when allowed by guidelines [30,31]. Interestingly, M184V and NNRTI resistance were less common in subjects failing regimens including emtricitabine rather than lamivudine, as observed by previous studies [32]. This finding may be also partly related to the availability and increasing use of fixed dose combinations containing FTC, which have been shown to implement adherence [33,34]. Nevertheless, no significant impact on NNRTI resistance was observed when considering the concomitant use of specific NRTI backbones (3TC vs. FTC) or NNRTIs (EFV vs. FTC) in the multivariate model. Of note, TA-sparing regimens did not appear to decrease the risk of resistance. Even though the virological efficacy of TAs is generally confirmed in several clinical settings [35], our finding does not reduce the usefulness and virological efficacy of TA sparing compounds. In fact, our data show a continuous drop in resistance prevalence in the Italian HIV-infected population during a period when TAsparing regimens substantially substitute TA-including combinations, mainly due to toxicity issues.
Some limitations of our work should be acknowledged. First of all, Italian data on precise extent of HIV-1-treated population, specific treatment prescriptions and related overall achievement of virological suppression on cART are not available at present. In fact, an increase of treatment effectiveness could explain both the reduction of the available resistance testing and the extent of resistance to antiretrovirals. Moreover, we were not able to consider the impact of self-reported adherence in our study population, due to the lack of information regarding this issue for a large number of patients, referring to several clinical centers in Italy. Furthermore, patients with resistant strains in early years of the study could have been lost at follow-up, due to several reasons including death, contributing to the decline of overall resistance. As guidelines prescribe to perform a genotypic resistance testing at failure, almost all patients failing an antiretroviral regimens have been included in the analysis. Nevertheless, the full information about the number of subjects failing antiretroviral regimens is not available, and we cannot exclude a partial loss of information regarding subjects who were not tested for antiretroviral resistance. Even though it was not possible to evaluate all the data regarding the antiretroviral history of patients, our analysis included main covariates such as previous virological failures, number of previous antiretroviral regimens, prior exposure to suboptimal NRTI therapy. Importantly, this work includes an evaluation of resistance according to the antiretroviral regimen in use at failure, representing an achievement in the analysis of acquired resistance not performed in previous works [17,18,21].
In spite of the limitations mainly due to its retrospective and multicenter design, our study shows a substantial decrease in drug resistance after stratification for antiretroviral class in use at failure. This finding is encouraging for the goal of a more effective treatment of the HIV infected population, in order to achieve a stable reduction of the potential virus transmitters. Moreover, the reduction we observed in resistance prevalence among treated individuals is probably the main driving factor leading to the observed reduction in primary resistance in our country [36].

Conclusion
In conclusion, our analysis shows a reduction of drug resistance overtime, even when correcting for several other or covariates or confounders. This is likely the consequence of time-associated factors including the availability of new potent drugs, updated guidelines for the treatment of HIV infection, the availability and frequent use of resistance testing to select for the optimal antiretroviral treatment at baseline and whenever a virological failure occurs.