Patterns of acquired HIV-1 drug resistance mutations and predictors of virological failure in Moshi, Northern Tanzania

Emergence of HIV drug resistance poses a serious risk of inactivity to all currently approved antiretroviral drugs. Profiles of HIV drug resistance mutations (HIVDRM) and virological failure (VF) are not extensively studied in Tanzania. This study aimed to determine HIVDRM and predictors of VF in HIV-infected individuals failing first-line HIV drugs in Moshi, Northern Tanzania. A case-control study was conducted at Kilimanjaro Christian Medical Centre, Mawenzi, Pasua and Majengo health facilities with HIV-care and treatment clinics from October, 2017 to August, 2018. Cases and controls were HIV-infected individuals with VF and viral suppression (VS) respectively. HIV-1 reverse transcriptase and protease genes were amplified and sequenced. Stanford University’s HIV drug resistance database and REGA subtyping tool 3.0 determined HIVDRM and HIV-1 subtypes respectively. Odds ratios (OR) with 95% confidence interval (95% CI) investigated predictors of VF. P-value < 5% was considered statistically significant. A total of 124 participants were recruited, of whom 63 (50.8%) had VF, 61 (49.2%) had VS and 82 (66.1%) were females. Median [IQR] age and duration on ART were 45 [35–52] years and 72 [48–104] months respectively. Twenty-five out of 26 selected samples from cases were successfully sequenced. Twenty-four samples (96%) had at least one major mutation conferring resistance to HIV drugs, with non-nucleoside analogue reverse transcriptase inhibitor (NNRTI)-resistance associated mutations as the majority (92%). Frequent NNRTI-resistance associated mutations were K103N (n = 11), V106M (n = 5) and G190A (n = 5). Prevalent nucleoside analogue reverse transcriptase inhibitors-resistance associated mutations were M184V (n = 17), K70R (n = 7) and D67N (n = 6). Dual-class resistance was observed in 16 (64%) samples. Thirteen samples (52%) had at least one thymidine analogue-resistance associated mutation (TAM). Three samples (12%) had T69D mutation with at least 1 TAM. Two samples (8%) had at least one mutation associated with protease inhibitor resistance. Age [aOR = 0.94, 95% CI (0.90–0.97), p < 0.001] and occupation [aOR = 0.35, 95% CI (0.12–1.04), p = 0.059] associated with VF. In conclusion, HIV drug resistance is common among people failing antiretroviral therapy. Resistance testing will help to guide switching of HIV drugs.

. The incidence-prevalence ratio (IPR) of HIV globally has declined considerably from 11.2% in 2000 to 4.6% in 2018. In the same year, HIV IPR in sSA declined to about 3.9% and 5.5% in Eastern and Southern Africa; and Western and Central Africa respectively. Despite the decline in IPR, HIV-infection remains a public health problem in sSA with approximately 68% of the global burden of PLHIV and a substantial number of new HIV-infections [1].
Highly active antiretroviral therapy (HAART) in sSA countries including Tanzania has helped to reduce HIV/AIDS-related mortality from around 900,000 in 2010 to 470,000 in 2018 [1] with improved quality of life [2]. The "90-90-90 target", set by the United Nations Member States, aims for 90% of PLHIV to know their HIV status, 90% of PLHIV who know their HIV status to be on ART, and also 90% of PLHIV on ART to have sustainable viral suppression by the year 2020 [3]. In the race to achieve the second UNAIDS '90' target, about 51% of PLHIV are on HAART in Western and Central Africa (WCA); and 67% in Eastern and Southern Africa (ESA). The proportions of HIV-infected individuals on treatment with viral suppression is better in WCA (79%) compared to ESA (58%); however, both proportions are below the third UNAIDS '90' target [1]. One of the key barriers to achieve the third UNAIDS '90' in sSA is the emergence of HIV drug resistance (HIVDR).
HIVDR is defined by the World Health Organisation (WHO) as the presence of one or more mutations in HIV drug-targeted genes that compromise the ability of a specific drug or combination of drugs to block replication of HIV [4]. HIV-infected individuals failing firstline HAART regimens are reported to have 50 to 97% of non-nucleoside analogue reverse transcriptase inhibitors (NNRTI) resistance worldwide [4]. In sSA, more than 80% of HIVinfected individuals with virological failure (VF) on first-line HAART have HIV-drug resistance [5]. Also, sSA has been reported to accommodate tenofovir resistance in more than a half of people with first-line HAART failure on tenofovir-based regimens. Cytosine analogue resistance is common in sSA as compared to Western Europe. Eastern Africa more commonly reported lamivudine and emtricitabine resistance than NNRTI resistance [6].
Tanzania introduced antiretroviral therapy (ART) program in 2004. As of 2016, 839,544 people were on ART in Tanzania. The recommended triple first-line ART was 2 nucleoside analogue reverse transcriptase inhibitors (NRTI) + 1 NNRTI or 2 NRTI + 1 PI. The preferred (default) first-line ART regimen for adults, adolescents, pregnant/lactating mothers at the time of the present study was tenofovir (TDF) + lamivudine (3TC) + efavirenz (EFV). Prescription of ART at initiation and when switching is based on several factors, including co-morbidities, clinical condition, co-administered drugs, pregnancy status, age, and drug tolerability. Other recommended first-line ART combination regimens at the time of the study included TDF + emtricitabine (FTC) + EFV; TDF + (3TC or FTC) + dolutegravir (DTG); abacavir (ABC) + 3TC + (EFV or DTG); zidovudine (AZT) + 3TC + (EFV or DTG); AZT + 3TC + nevirapine (NVP); TDF + (FTC or 3TC) + atazanavir boosted by ritonavir (ATV/r). In case of first-line HAART failure, the recommended second-line HAART for adults and adolescents was zidovudine/lamivudine + ritonavir-boosted atazanavir or tenofovir/emtricitabine + ritonavirboosted atazanavir [7]. In Tanzania, the markers for treatment failure are based on immunological, clinical and WHO recommended virological criteria [8]. Clinical and immunological criteria were extensively described to be less sensitive and less effective [9] and may reduce early notifications of VF. Tanzania continues to scale-up HIV viral load (HVL) testing which is currently done at a few selected settings. Of more public health importance is that HIVinfected individuals confirmed to have VF are switched to second-line HAART without programmatic HIV drug resistance testing and monitoring, a practice which may transfer crossresistance patterns to newly switched drugs and limit available treatment options.
A wide range of first-line HIV drug resistance mutations (HIVDRMs) has been reported in Tanzania, including the thymidine analogue associated mutations which compromise susceptibility to multi-NRTIs [10][11][12]. To complement the efforts of unravelling the burden of HIVDR and preserve the integrity of limited second-line HIV drugs in Tanzania, this study aimed to examine genotypic drug resistance in reverse transcriptase (RT) and protease genes from selected individuals with VF on a wide range of first-line HAART in Moshi municipality, Northern Tanzania. The study also explored independent predictions for VF.

Study design and settings
This was an unmatched case-control study done in four HIV/AIDS care and treatment clinics (CTC) in Moshi municipality, situated in Kilimanjaro region, Northern Tanzania from October to August 2018. Moshi municipality is one of the seven districts of Kilimanjaro region. The municipality has a total of 18 health facilities with CTC. Four out of 18 CTCs with high client volume were purposively selected to participate in the study. CTCs included were Kilimanjaro Christian Medical Centre (KCMC), Mawenzi Regional Referral Hospital, Majengo and Pasua Health Centres. KCMC provides tertiary care medical services to around 6.8 million people living in the northern zone of Tanzania (Tanga, Kilimanjaro, Arusha and Manyara) and other referrals from nearby settings. Mawenzi Regional Referral Hospital provides referral medical services to approximately 1.6 million people in the Kilimanjaro region.
At the time of study recruitment, management of HIV-infected individuals was done inline with National Guidelines for the management of HIV and AIDS of 2017 [8]. CTCs routinely provided HIV counselling and testing services, ART care and treatment, treatment monitoring, laboratory investigations and treatment adherence support to HIV-infected individuals. Prescription of antiretroviral drugs was done by clinicians and in some cases by trained nurses in the health centres. HVL testing was routinely done for all HIV-infected individuals who have been on ART for at least 6 months. Those found with HVL > 1,000 copies/ ml were offered enhanced adherence counselling (EAC) by trained health care providers on a monthly basis for three months followed by retesting of HVL to confirm VF. Clinic visits for adolescents and youths were scheduled at different times from adults to maximize adherence counselling and support. Drug adherence levels of � 95% and < 95% was regarded as good and poor respectively [8]. Drug adherence was monitored by pill counting during the participant visit at the CTC. Pills missed were determined by physical counting of the remained pills against the pills prescribed in the previous visit. Percentage (%) of pills missed was calculated by using the formulae below; % of pills missed ¼ Number of pills remaining Total number of pills prescribed � 100 Percentage of drug adherence was calculated as follows; % drug adherence ¼ 100 À % of pills missed

Study population
The study population consisted of HIV-infected individuals attending the CTCs for routine care and who were on first-line ART treatment for one year or more. HIV-infected individuals with VF and viral suppression (VS) were included in the study as cases and controls respectively. The cases were defined as HIV-infected individuals with > 1000 copies/ml of HIV plasma RNA confirmed by < 0.5 logarithmic difference between initial and second HVL with three EAC in between. The controls were HIV-infected individuals attending the same CTCs along with cases but have viral suppression (HVL < 1000 copies/ml) that was measured in at least two different clinic visits. The cases found to have less than three EAC were excluded from the study. The study further excluded cases with a significant difference in HVL measured at the date of the interview and that measured before starting EAC (HIV-1 plasma RNA viral log 10 drop greater than 0.5 at three-month interval with 3 EAC in between).

Sample size determination
We assumed that the study had 80% power and the proportion exposed in the control group was 20%. With an equal number of cases and controls (ratio of 1:1), we could manage to detect an odds ratio (OR) of 3.0 or greater with the following sample size calculation as described by Charan and Biswas [13].
n = sample size in each group, r = ratio of controls to cases, � p = a measure of variability (average proportion exposed), p 1 = proportion exposed in cases, p 2 = proportion exposed in controls, p 1 -p 2 = Effect size (the difference in proportions), Z α/2 = level of two-tailed statistical significance, Z β = standard normal variate for power of the study. Our study power was 80%, Z β = 0.84; Level of significance was 0.05, Z α = 1.96; r = 1 (equal number of cases and controls), OR = 3.0 and p 2 = 20% or 0.2. The study thus aimed to recruit 63 cases and 63 controls making a total sample size of 126.

Enrolment and study procedures
Cases were recruited consecutively after confirmation of VF in the CTCs. Systematic sampling was applied to recruit controls since nearly all of the CTC attendees at every clinic visit had viral suppression. Study participants who met the inclusion criteria and consented to participate were interviewed to collect demographic and clinical information. After the interviews, 8-10 ml of EDTA whole blood was collected and centrifuged at 2200 revolutions per minute for 10 minutes with brake-offs to separate plasma from buffy coat and red blood cells within 4 hours post-collection. HVL was enumerated in-vitro from plasma by reverse transcriptionpolymerase chain (PCR) reaction as per the Abbott m2000rt system and was expressed in copies/ml of plasma. Due to financial constraints, 26 out of 63 samples of cases were selected for sequencing. The selected subset of samples represented the large cohort of 63 by age, duration on ART, gender, CD4 count, HAART in use, switching of ART, and HVL. The characteristics of the selected 26 samples and unselected 37 samples were similar (p > 0.05, Fisher's Exact test) (S1 Table). Out of 63 cases, 37 (58.7%) were 35 or more years of age, of these, 16 (43.2%) were selected for sequencing. Out of 63 cases, 58 (92.1%) were on ART for more than 36 months, of these, 25 (43.1%) were selected for sequencing. Out of total 63 cases, 37 (58.7) were females, of these, 12 (32.4%) were chosen for sequencing. Thirty-three (53.2%) out of total 63 cases had CD4 count � 350 cells/uL, of these, 16 (48.5%) were chosen for sequencing. The selected subset of samples had HVL > 14,000 copies/ml. S1 Table further describes the proportional characteristics of the selected and not selected samples for sequencing against the larger cohort of 63.
Plasma HIV-1 RNA was extracted using the PureLink1 Viral RNA/DNA Mini Kit (Invitrogen, ThermoFisher Scientific, USA) as per manufacturer's instructions. RT and protease genes of the extracted HIV-1 RNA were reversely transcribed into complementary DNA (cDNA) and nested PCR was done according to manufacturer's instructions prescribed in the HIV-1 genotyping kit: Amplification module (Applied Biosystems, Life Technologies, Warrington, UK). The amplified cDNA was purified using the ExoSAP-IT™ PCR product clean-up reagent (Applied Biosystems, ThermoFisher Scientific, Inc.). Reactions were performed using the HIV-1 genotyping kit: Cycle sequencing module (Applied Biosystems, Life Technologies, Warrington, UK) based on Sanger sequencing method using BigDye™ Terminator v3.1 cycle sequencing kit (Applied Biosystems, ThermoFisher Scientific, Inc.). Sequencing was done using 3500xl genetic analyser (Applied Biosystems) with a 24 capillary 50 cm array. The study profile is described in

Data analysis
Sequence data analysis. The raw sequence data were assembled, aligned and edited using automated base-calling software, RECall version 3.05 [14], available at (http://pssm.cfenet.ubc. ca) to generate individual consensus sequences. The sequence data were submitted to Gen-Bank and obtained accession numbers MT347616-MT347640. HIVDRMs were analysed using the HIV drug resistance database of Stanford University available at (http://hivdb. stanford.edu) and HIVDRM mutation list from the 2019 updates of drug resistance (AIDS Society). To get HIV-1 subtypes, generated consensus sequences were analysed using REGA subtyping tool version 3.0 [15] available at (http://dbpartners.stanford.edu:8080/ RegaSubtyping/stanford-hiv/typingtool/). REGA confirmed a subtype after clustering with a pure subtype in a database by > 800 base pairs with a bootstrap confidence of more than 70% in absence of recombination in the boot scan. The evidence for HIV-1 recombination was identified using Recombinant Identification Program version 3.0 [16] with window size 400 and 90% confidence threshold.
The maximum likelihood phylogenetic tree for HIV-1 diversity was generated using the Randomized Axelerated Maximum Likelihood (RAxML) program version 8.2.12 as previously described by Stamatakis [17]. Phylogenetic tree annotations were made using Interactive Tree Of Life (iTOL) [18] version 5 available at (https://itol.embl.de/). Statistical analysis. Data analysis was performed using STATA version 14.0 (College Station, Texas 77845-4512, USA). Categorical variables were summarized using frequency and proportion while median with interquartile range was used to summarize numerical variables. A Chi-square test was used to compare the differences in proportions between groups. Odds ratios

PLOS ONE
HIV-1 drug resistance mutations and predictors of virological failure in Moshi, Northern Tanzania (OR) and 95% confidence interval (CI) for predictors of VF were estimated using multivariable logistic regression model. A p-value < 0.05 (2 tails) was considered statistically significant.

Ethics consideration
The study was approved by the Kilimanjaro Christian Medical University College Research Ethics Review Committee in 2017 with ethical clearance certificate number 2028. The Kilimanjaro Regional Administrative Secretary granted permission to conduct the study at Pasua Health Centre, Majengo Health Centre and Mawenzi Regional Referral Hospital. The Executive Director of the KCMC granted permission to conduct the study at KCMC. All study participants consented/assented in writing to take part in the study. The HIV drug resistance testing results were shared with the respective CTCs for further management of the HIV-infected individuals.  (Fig 2). Sequencing of RT and protease genes of HIV-1 was successful on 25 out of 26 selected samples, one sample failed amplification. Of the 25 samples, almost all (n = 24) had at least one major mutation conferring resistance to HIV drug (Table 3). Dual-class resistance was observed in 16 (64%) samples. In general, thirteen samples (52%) had at least one thymidine analogue-resistance associated mutation (TAM) while three samples (12%) had T69D mutation together with at least 1 TAM. Identified TAMs were T215YF (n = 9), K219QE (n = 7), K70R (n = 7), D67N (n = 6), M41L (n = 4), and L210W (n = 3) ( Table 3). The presence of T69D mutation along with at least 1 TAM reduces the susceptibility of all currently NRTIs in use. One sample was fully susceptible (no detectable HIV drug resistance mutation) despite having VF and substantial-high HVL at the date of the interview.
Two samples (8%) had at least one mutation associated with protease inhibitors (PI) resistance (Table 3). One participant (K38) who had previously used ritonavir-boosted lopinavir had L24I, L33F, I54V, and V82A mutations which together reduce the susceptibility of ritonavir-boosted lopinavir. The same participant (K38) who was on ritonavir-boosted atazanavir at the date of interview had N88S mutation which confers a great effect on atazanavir. Only M46I mutation was found in both samples.

HIV-1 diversity
The proportions of HIV-1 diversity in the RT and protease genes are shown in Table 4. HIV-1 subtype C was the most prevalent subtype (48%) followed by A1 (36%), D (8%) and Circulating Recombinant Forms (CRF) (8%). The recombinants identified were CRF10_CD and

PLOS ONE
HIV-1 drug resistance mutations and predictors of virological failure in Moshi, Northern Tanzania CRF35_AD. A maximum-likelihood phylogenetic tree (Fig 5) describes the HIV-1 diversity with annotations. The two CRFs were only reported at KCMC. Subtype C was reported in all study sites except Mawenzi RRH (Fig 5).

Predictors of VF
The predictors of VF are shown in Table 5. In bivariate analysis, participant age, occupation, HAART initiated and HAART in use were significantly associated with VF. In the multivariable logistic regression analysis, occupation had marginal statistical significance [aOR = 0.35, 95% CI (0.12-1.04), p = 0.059] while age was an independent predictor of VF. Being employed in either formal or informal sector was 65% protective against VF. Furthermore, one unit

Discussion
This study aimed at determining HIVDRMs and predictors of VF in HIV-infected individuals failing to respond to first-line ART in Moshi municipality, Northern Tanzania. High profiles of HIVDRMs conferring acquired HIV drug resistance to NRTIs, NNRTIs and PIs were found. While occupation had marginal statistical significance in predicting VF, age was an independent factor associated with VF.
Most (96%) of the successfully sequenced samples had at least one major mutation conferring resistance to HIV drugs. NNRTI resistance (92%) was greater than NRTIs resistance (68%) consistent with findings elsewhere [6]. The most frequent NNRTI resistance-associated mutations were K103N, V106M, and G190A that confer high-level resistance to efavirenz and nevirapine [19]. NNRTIs were previously reported to have a low genetic barrier and hence  highly vulnerable to resistance [20]. These findings further support the Tanzanian programmatic intervention to replace efavirenz with dolutegravir which has proven to have a high genetic barrier to resistance [21]. This study further reports frequent NRTI resistance-associated mutations as M184V, K70R, and D67N in line with other studies [10]. M184V associates with high-level resistance to abacavir, emtricitabine, and lamivudine; while K70R and D67N are among the TAMs. More than a half of sequenced samples had at least one TAM. Existence of TAMs reduces the susceptibility of all available NRTIs with an exception of emtricitabine and lamivudine [22]. This study reports co-existence of T69D mutation with at least one TAM in three samples. Two of these samples had M41L and T215Y TAMs, a combination which further enhances cross-resistance to all US FDA approved NRTIs, a phenomenon known as Multi-NRTI

PLOS ONE
HIV-1 drug resistance mutations and predictors of virological failure in Moshi, Northern Tanzania resistance [19,23]. As a result, the Tanzania National AIDS Control Program may have fewer options of main-stream HIV drugs both in first and second-line regimens. Continuous programmatic monitoring of HIV drug resistance in HIV-infected individuals failing to respond to first-line HIV drugs will preserve the integrity of few NRTI options in second-line. However, there is a need to scale-up NRTIs with substantial genetic barrier to resistance. This study describes full genotypic susceptibility of prescribed HAART to a 49 years old individual with ID K15, on ART for 12 years and with HVL 81,123 copies/ml. At the interview, he reported good adherence to the prescribed HAART. Although the participant self-reported good adherence to medication, we speculate the reality of adherence was low. Several factors hinder adherence to medications including stigma, and lack of social support [24]. Community-based HIV treatment supporters chosen by HIV-infected individuals may improve adherence to medications and lead to favourable treatment outcomes [25].

PLOS ONE
HIV-1 drug resistance mutations and predictors of virological failure in Moshi, Northern Tanzania adults on HAART previously reported to have inconsistency in adhering to antiretroviral medication due to anxiety, depression, forgetfulness, fear of disclosure, ART adverse events and abandoning medication when they feel better [34], consequences which contribute to VF. Therefore, special care and treatment for adolescents and young adults are paramount with an emphasis on health education particularly the importance of disclosure and adhering to medications. This study reports that HIV drug-experienced individuals in formal or informal employment were protected from VF. Unemployment among HIV-infected individuals was previously described as a stressful event which reduced adherence to prescribed HIV medications resulting in high rates of VF [35] and increased mortality [36]. Supporting and engaging PLHIV in income-generating activities may stabilize financial and social status and positively impact drug adherence and favourable treatment outcomes.
Although not statistically significant, this study reports the positive association between advancement of WHO clinical stage of AIDS and VF, a finding which was consistent with those reported by Jobanputra et al in (2015) [28]. VF in individuals with advanced HIV/AIDS can be explained by the severe state of immunodeficiency which attracts opportunistic infections (OIs) and replicative fitness of HIV [37]. Early diagnosis of HIV, effective treatment and monitoring of HIV/AIDS and related OIs among individuals on HAART is highly recommended [9].
Tenofovir (TDF)-based combination antiretroviral therapy is one of the recommended first-line antiretroviral drugs of choice for adolescents and adults living with HIV worldwide PLOS ONE [9] as well as in Tanzania [8]. Although not statistically significant, being on TDF-based firstline HAART was protective against VF, this is consistent with studies previously done in North America [38], Thailand [39] and in a systematic review [40]. TDF-based regimens were extensively reported in previous randomized clinical trials to be more effective than other combination therapies in bringing favourable virological and immunologic outcomes [41,42]. A possible explanation is that, compared to other back-borne antiretroviral drugs, TDF-based therapy is consumed once daily with minimal treatment-related toxicity [41,43], is more tolerable [44] and can be easily adhered to by PLHIV [45]. Therefore we complement the prescription of TDF-based HAART in PLHIV initiating HAART as the strategy of reducing VF.

Limitations of the study
One of the limitations of this study is reporting bias of adherence to HAART and pill counting at the clinic visit. Some of the study participants might have provided imprecise information about adherence. However, this was the common practice prescribed under the Tanzanian HIV and AIDS Treatment and Management Guidelines of October 2017. Also, recall biases are certain to some of the information requested from the VS group. Finally, the study would have had much more to offer by sequencing the entire cohort of 63 plasma samples from participants with VF. Unfortunately, due to financial constraints, only 26 out of 63 plasma samples were sequenced. However, selection of a subset of 26 samples proportionally represented the large cohort 63 by age, duration on ART, gender, CD4 count, HAART in use, switching of ART, and HVL.

Conclusion
There is a high rate of HIVDRMs in HIV-infected individuals failing to respond to first-line HAART in Moshi, Northern Tanzania. This is probably a result of no programmatic monitoring of HIV drug resistance in individuals with VF. Prompt intervention is required to safeguard the limited number of second-line HIV drug options, such as the implementation of HIV drug resistance monitoring before and after switching HAART. In addition, scale-up of HIV drugs with higher genetic barrier to resistance is of importance.