Drug Resistance Patterns and Virus Re-Suppression among HIV-1 Subtype C Infected Patients Receiving Non-Nucleoside Reverse Transcriptase Inhibitors in South Africa

Ziad El-Khatib1,2*, Allison K. DeLong3, David Katzenstein4, Anna Mia Ekstrom1, Johanna Ledwaba2, Lerato Mohapi5, Fatima Laher5, Max Petzold6, Lynn Morris2 and Rami Kantor7 1Division of Global Health (IHCAR), Karolinska Institutet, Stockholm, Sweden 2AIDS Virus Research Unit, National Institute for Communicable Diseases (NICD), Johannesburg, South Africa 3Center for Statistical Sciences, Brown University, Providence, Rhode Island, USA 4Division of Infectious Diseases, Stanford University, California, USA 5Perinatal HIV Research Unit (PHRU), University of the Witwatersrand, Soweto, South Africa 6Nordic School of Public Health (NHV), Gothenburg, Sweden 7Division of Infectious Diseases, Brown University Alpert Medical School, Providence, Rhode Island, USA


Background
More than 1 million HIV-1 subtype C infected patients in South Africa are receiving antiretroviral therapy (ART) [43]. In 2004 a national treatment program was initiated, including a first-line regimen containing a non-nucleoside reverse transcriptase inhibitor (NNRTI), either efavirenz or nevirapine, in combination with NRTI, stavudine or zidovudine, and lamivudine [27] The virologic outcomes of firstline regimens among subtype C-infected people in South Africa are comparable to those among subtype B infected patients in Switzerland, where approximately 10% of patients experience virologic failure after 12 months and up to 25% experience virologic failure by two years on ART [20].
Previously [11], we surveyed viremia and drug resistance prevalence among 998 patients in Soweto, South Africa and found that 94/883 (11%) receiving first-line regimens for a median of 42 months were viremic and 78/94 (83%) had drug resistance. Here, we obtained retrospective data on 431 of these patients, enrolled at a single clinic, and examined factors associated with the evolution and patterns of DRM. Additionally, we followed 25 of these viremic patients to explore the implications of DRM on continued NNRTI-based treatment.

Study sites and patients
The study was done at the Perinatal HIV Research Unit (PHRU), a non-governmental organization (NGO) research clinic in Chris Hani Baragwanath Hospital, Soweto, outside Johannesburg, South Africa (11,29). At the time of the study, the clinic staff consisted of five medical doctors, two nursing assistants and two counselors managing around 1500 ART recipients, with 50 daily visits [11].
Consenting patients were enrolled from March through September 2008 if they were ≥18 years old and >12 months on a first-line regimen [11]. At study enrollment, viral load (VL), CD4 cell count and HIV-1 reverse transcriptase (RT) genotype were assessed and basic demographic information obtained. Virologic failure was defined as VL>400 copies/mL at study enrollment [11]. Persistent virologic failure and re-suppression were defined as VL>400 copies/mL and return to ≤ 400 copies/mL, respectively, at follow-up ( Figure 1).
Medical records were reviewed retrospectively to extract information on potential risk factors that may be associated with DRM or re-suppression, including age, gender, year of HIV diagnosis, ART regimens and dates, history of sdNVP or other ART exposure, tuberculosis (TB) treatment, pre-ART initiation VL and CD4 cell counts, WHO stage prior to ART initiation and any treatment interruptions in the last six months prior to study enrollment. Virologic failure prior to study enrollment, was defined as VL>400 copies/mL at either of the prior two visits in the last 12 months. Poor adherence was considered to be returning more than seven days late for the drug refill appointment pre-study enrollment. An instrument was designed and tested to extract information from medical records using Epi Data [9,10]. The study and consent forms were approved by the University of the Witwatersrand Human Research Ethics Committee in South Africa and the Regional Medical Ethics Board in Stockholm, Sweden.

Laboratory assessments
CD4 cell counts were performed by FACSCountTM (Becton Dickinson BioSciences, Immunocytochemistry Systems, San Jose, California, USA) and VL was measured using the Roche Amplicor, version 1.5 (Roche Molecular Diagnostics, Basel, Switzerland) with a lower limit of detection of 400 copies/mL. Both assays were performed at the National Institute for Communicable Diseases (NICD), Johannesburg. VL data extracted from medical records that were done as part of routine clinical care used the Versant HIV-1 RNA 3.0 (Siemens Deerfield, IL, USA) bDNA technology. For HIV-1 drug resistance testing, an in-house genotyping assay was performed at the NICD [35].

Data analysis and statistics
Risk factors were examined for associations with two drug resistance outcomes: (i) number of all-class DRM at enrollment; and (ii) continued viremia versus re-suppression at follow-up. Associations between viral re-suppression and presence of any DRM, number of NRTI and NNRTI DRM and the total number of DRM at enrollment were examined. Due to the small sample size, bivariate analysis was performed without adjusting for confounding variables and the results must be interpreted with this in mind.
Ordinal logistic regression was used to examine the association between risk factors and the number of DRM at enrollment expressed as odds ratios (OR), and 95% confidence intervals (95% CI). Unlike Poisson regression, ordinal logistic regression can be fitted to zeroinflated data and does not assume that the events (i.e. accumulation of DRM) are independent and occur at a constant rate. Each model was checked to ensure the assumption of proportional odds between successive DRM categories was met. To examine risk factors associated with persistent virologic failure at follow-up, Fisher exact tests were used for categorical risk factors (OR, 95% CI) and Wilcoxon rank sum tests for continuous risk factors (difference in median, 95% CI). Analysis was performed using Stata/SE College Station, Texas (version 10.1) (38) and R (version 2.11.1) (40). P-values less than 0.05 were considered statistically significant.   (Table 1) displays characteristics of the 431 patients who had been on ART for at least 12 months where 75% were females, 96% were born in South Africa, 90% had above primary school education and the median age at study enrollment was 38 years. Ninety-one percent were receiving efavirenz-based therapy and 9% a nevirapine-based therapy.

Patient characteristics
Before ART initiation, the median VL and CD4 were 71,995 (range 1,078 to >500,000) copies/mL and 93 (range 1 to 444) cells/ mm3 respectively. At study enrollment, patients had received ART for a median of 45 months (range 13 to 152) and the CD4 cell count increased to 419 (range 16 to 1,270).
Forty-three patients (10%) had VL>400 copies/mL (median 6,510; range 407 to >500,000) at enrollment, almost all (98%) previously suppressed on ART. Median time on ART was similar among patients with VL < and >400 copies/mL (p=0.86). However at study enrollment, those with VL>400 copies/mL had a significantly lower CD4 cell count compared to patients with VL<400 copies/mL (p<0.01).
Among 40/43 patients for whom there was a record of previous ART exposure: 8/31 (26%) females had received sdNVP prior to initiating ART and 5/40 (13%) patients had been exposed to other ART before initiating the current ART regimen. Compared to unexposed individuals, there was a borderline association between viremia at study enrollment and exposure to sdNVP (p=0.05).
Several risk factors were significantly associated with increased numbers of DRM (Table 3). Patients with prior exposure to either sdNVP or other ART had more mutations than those not previously exposed. The ordinal regression OR was 3.8 (95%CI 1.1 to 15.2; p=0.03), i.e. it was 3.8 times more likely for patients with prior ART exposure to have ≥1 vs. 0, ≥2 vs. ≤1, ≥3 vs. ≤2 DRM and so on. Number of DRM was positively associated with being female (OR 5.6; 95%CI 1.3 to 24.5; p=0.02), having a higher CD4 cell count (OR 1.7 per 100 CD4 cells; 95%CI 1.1 to 2.7; p=0.02) and having detectable VL at one of two earlier scheduled visits (OR 8.4; 95%CI 1.9 to 42.4; p<0.01). The association with gender was mainly explained by prior exposure to pMTCT, mainly sdNVP, among the women. Only one male had any prior exposure to ART. The number of DRM was negatively associated with coming late for the drug refill visit in the last month (OR 0.1; 95%CI 0 to 0.5; p=0.01) and with VL such that for participants with 1-log unit higher VL the odds of having a higher number of DRM was 0.5 (95%CI 0.2 to 1.0; p=0.04). Finally the median VL of the seven patients with no DRM was 83,000 copies/mL compared with a median VL of 6,510 copies/mL among those with at least one DRM, providing evidence for existing but incomplete drug pressure amongst those with DRM.

Risk factors for persistent virologic failure and drug resistance mutations at follow-up
Follow-up data and samples were available for 25/43 (58%) of the viremic patients, after a median of 8 (range 4 to 10) months ( Figure  1). Persistent virologic failure at follow-up, in 13/25 (52%) patients, was associated with a detectable VL in the two visits prior to study enrollment (p<0.01) and the number of DRM at study enrollment (OR 2.36; 95%CI 1.11 to 5.02; p=0.04), particularly NRTI mutations (OR 3.68; 95%CI 1.11 to 12.17; p=0.05) ( Table 3). All 13 patients had genotypic resistance with six additional DRM acquired at follow-up, leading to high level predicted resistance to efavirenz and/or nevirapine (100%) and lamivudine (100%) and intermediate to low predicted resistance to etravirine in 7/13 (54%) ( Table 4a). Although viremic, patient number 35 did not have any DRM at study enrollment, but with continued treatment and presumably better adherence, three DRM were selected at follow-up with a persistent, albeit lower, VL for nine months. At study enrollment nine of these patients had failed first-line regimens with full predicted susceptibility to NRTIs.
Twelve of the 25 (48%) patients re-suppressed at follow-up, after a median of eight additional months on treatment with the same NNRTI (mostly efavirenz)-based regimen. Comparison of clinical and laboratory characteristics of these 12 patients with the 13 who had persistent virologic failure showed no significant differences in sex, median CD4 and VL prior to ART initiation. At study enrollment 3/12 (25%) re-suppressed patients could not be amplified and three (25%) had no DRM. However, the remaining six patients who were resuppressed had NNRTI DRM, three had K103N, one had K103N and M184V and one had K103N, V106M and M184V (Table 4b). The sixth patient had three NNRTI and 3 NRTI mutations. Thus, six patients with high level NNRTI resistance and three patients with high level NNRTI and lamivudine resistance achieved re-suppression while continuing the same first-line regimen.

Discussion
HIV-1 drug resistance is a potential cause and is often a consequence of virologic failure. In this study, we examined drug resistance in HIV-1 subtype C infected South African patients failing first-line regimens after a minimum of 12 months on ART. Of 43 long-term ART recipients with viremia, most had multiple DRM. Among 25 of these patients followed on continued NNRTI-based ART, 12 achieved virologic suppression and 13 had persistent VL>400 copies/mL. We examined the characteristics and patterns of DRM, the estimated drug resistance and implications for further therapies among these ART experienced, subtype C infected patients.
In examination of co-occurrence of DRM, 82% of 38 available RT sequences demonstrated a wide spectrum (range 1 to 7) of cooccurring, DRM. The majority (63%) had ≥3 DRM, mostly including K103N accompanied by M184V/I, at times with complex mixtures of additional NRTI and NNRTI mutations. Patients with prior ART drug exposure and those with detectable VL and higher CD4 cell counts prior to or at study enrollment, respectively, tended to have a higher number of DRM. These results are likely to reflect the selection of drug resistance due to lapses in adherence, reduced drug exposure and inadequate drug pressure. Conversely, patients who were late to pharmacy or who had higher VL at study enrollment tended to have fewer DRM, perhaps reflecting a very low adherence.
Drug resistance mutations among subtype C NNRTI recipients have been identified after virologic or immunologic failure based on  Here, the frequency of DRM upon virologic failure using a threshold of >400 copies/mL was similar to thresholds of VL >1,000 or >5,000 copies/mL [14,28,33,42] or immunologic failure criteria (8,15). We compared the patterns of DRM found here to 418 published sequences from adult patients failing first-line regimens in subtype C studies, mostly from Southern Africa and India (8,15,19,28,35,37,42), accessed, in July 2010 at the Stanford HIV Sequence Database (13). Patterns of DRM, including the overall frequency of any DRM (82% in the current study vs. 83% among published sequences), ≥1 NRTI resistance mutations (11% vs 9%), the prevalence of K103N (55% vs 42%), M184V/I (66% vs. 74%), and K65R (3% vs. 6%) were not significantly different. However, the data presented here compared to the published sequences demonstrated significant differences in the frequency of ≥1 NNRTI mutation (58% vs 40%, p=0.03) and a lower rate of TAM (21% vs 37%, p=0.05). These modest differences may be ascribed to differences in clinical management strategies, specific drug combinations or duration of virologic failure. Among eight patients with TAMs, the most common pathway (seen in 6/8 patients) was TAM-2 related (D67N, K70R, K219Q/R/E),; and the rest (2/8) were mixed with the TAM-1 pathway (M41L, L210W, T215Y) extending similar prior Southern African observations [28,31] .
Although K103N and other NNRTI resistance mutations confer high level NNRTI resistance [12,13] , six patients in our study who harbored such mutations, re-suppressed VL after 4 to 10 months with no change in their first-line regimen. Three of these patients had high level resistance to two drug classes, with the addition of the M184V mutation, conferring resistance to lamivudine. These findings extend observations by Hoffmann and colleagues who reported 11 males with either NNRTI and NRTI mutations, who re-suppressed with continued first-line regimens, raising questions about potential cautious re-use or continuation of those medications in certain circumstances [14] . It is plausible that improved adherence was a factor in this observed resuppression, though this could not be confirmed in this patient population. The observation that successful re-suppression was strongly associated with recent failure and a low number of DRM seems logical, but should be confirmed prospectively in larger studies.
Thirteen patients found to be viremic both at enrollment and at   follow-up, had a longer duration of virologic failure on treatment and a higher number of DRM. These findings substantiate the observation that resistance evolves as a function of continued, albeit suboptimal, drug pressure due to reduced adherence, treatment interruptions or both. Despite the relatively short time between sequences -8 months, and a median of 3 DRM per patient, mutations accumulated with a rise in high level predicted 2-class resistance. This is consistent with observations in HIV-1 subtype B [5,18]. Some of the accumulated DRM were associated with etravirine resistance, conferring intermediate resistance to this drug after first-line regimen failure, suggesting the need for further studies of the use of this NNRTI in subsequent regimens [23].
Among women, we found a borderline association between exposure to sdNVP and virologic failure, albeit years afterwards and with a period of suppression, consistent with the results of a recent study from the Western Cape, South Africa [6] and another report of a significant association between detection of minority NNRTI mutations and treatment failure, even after 18 months had elapsed since sdNVP [4] . These findings are not entirely in line with other reports  The ordinal logistic regression model could not be fitted because there was too little variability in the risk factor or the assumption of proportional odds ratios was not met (i.e. only 4 were on TB therapy, 4 had taken other ARV prior to starting ART, 3 had treatment interruption, and the assumption was not met for EFV vs. NVP. d One female was exposed to both of sdNVP and other ARV, pre-ART initiation, and she had VL <400 copies/mL at study enrollment.   from sub-Saharan Africa and Asia that suggest that in the short-term, administration of ART >12 months after sdNVP may not jeopardize the efficacy of NNRTI based ART [17,26].
Where genotyping is routinely available, the interpretation of DRM following the failure of first-line regimens drives decisions about clinical management, adherence counseling, switching to second-line regimens and the need for new ART combinations. Among these HIV-1 subtype C infected patients, a proportion of those failing firstline regimens responded to continuing their current regimen despite predicted high-level resistance. In resource-limited settings, where genotyping is not available, the practice of reinforcing adherence, continuing a first-line regimen and repeating a VL test after 6 months may be justified. However, it will be important to monitor such patients closely to determine if they ultimately fail due to the presence of archived resistance mutations. In such cases, switching to a second-line regimen, as mandated by treatment national guidelines, should not be deferred.
There are several limitations to this study. First, the number of patients with drug resistance, longitudinal sequences and re-suppression was relatively small and longer term CD4 cell counts and VLs from 25 failing patients were not available. Second, a small proportion of patients' samples, particularly among those who resuppressed, could not be successfully amplified and sequenced. Third, genotyping at onset of ART was not available. However, the prevalence of transmitted ART resistance, in South Africa, is still low (<5%) [24,34] . Finally, confirmation of VL was not immediately available, and clinical practice included continued first line-treatment for 4-6 months between detection of virologic failure, reinforcing adherence and re-testing VL. The fact that patients were not switched to second-line regimens, despite repeated virologic failure, reflects the current limited treatment options in this setting.
In summary, this study extends observations of the range of DRM patterns among HIV-1 subtype C patients in South Africa receiving long-term first-line regimens. Surprisingly, a number of patients with DRM that predict high-level resistance, K103N with or without M184V, were successfully re-suppressed on the same first-line regimen. Reinforcing adherence without changing treatment among patients with first-line virologic failure could spare the expense and toxicity of second-line regimens in resource-limited settings with high HIV burdens. While additional studies are needed to confirm these observations and examine their longevity, detection of viremia on treatment should prompt repeat testing, adherence counseling and, if viremia is persistent, provision of second-line therapy.