**3. Discussion**

The present study showed an increasing trend of PDR to efavirenz in Mexico City from 2017 to 2020, crossing the 10% threshold by the end of the study period, while PDR to dolutegravir and bictegravir remained low and under 0.5%. Given the high genetic barrier of second-generation INSTI-based ART regimens [13], as well as the low impact of baseline DRMs observed in the NRTI backbone in the effectiveness of these regimens [14], HIV PDR surveillance has become less of a priority in regions widely using dolutegravir and bictegravir. Nevertheless, given the historical rapid increase in NNRTI resistance in LMICs worldwide, it is important to maintain surveillance in order to detect increasing PDR trends in time and save important ART options for future managemen<sup>t</sup> of HIV, strengthening HIV programs with focus on HIVDR-associated early warning indicators. Focused HIVDR surveillance to identify specific regions and populations with specific vulnerability issues and programmatic gaps is also important to improve the HIV care continuum with focused

interventions and to optimize both first- and second-line ART regimens, as well as pre- (PrEP) and post-exposure prophylaxis (PEP) regimens.

As expected, PDR to efavirenz/nevirapine continued to grow in Mexico City across the study period, maintaining the previously observed trend [10]. Importantly, no changes in this trend were observed after the nationwide implementation and procurement of second-generation INSTI-based regimens as preferred first-line options since the second half of 2019, which suggests onward transmission of efavirenz resistance even in the context of a significantly reduced use of the drug. Indeed, when analyzing Mexico City's HIV transmission network for the study period, we observed 88 clusters for which at least 50% of the nodes shared NNRTI PDR mutations. These included large clusters formed mainly by young cisgender men sharing the K103NS mutation and other NNRTI PDRassociated mutations. It is also important to consider the possible impact of ongoing DRM transmission on the growing cross-resistance trend to rilpivirine observed in the present study for the future use of long-acting regimens as part of both first- and second-line ART as well as PrEP regimens.

As expected, efavirenz/nevirapine resistance and, in general, PDR to any ARV drug were strongly associated with prior exposure to ARVs. The group of persons with prior exposure to ARVs in Mexico City was mostly formed by persons that, after being lost to clinical follow-up, later returned to clinical care to restart ART, keeping in mind that PrEP is not ye<sup>t</sup> widely available in Mexico. This group of ART restarters represents an important challenge and target for possible interventions, and it showed specific characteristics that were different from the rest of the study population, including higher representation of cisgender women and transgender women, higher age, lower education in general and higher representation of injectable drug users (Table S1). Some of these characteristics were described previously in a nationwide PDR study [9]. Unexpectedly, the odds of having PDR were not specifically higher in cisgender women, as observed in previous studies in Mexico and in other countries [5,9]. In the context of Mexico City, contrasted with other regions of Mexico, the epidemic is highly concentrated in MSM [15] who are generally younger, have a higher education level, and arrive earlier to clinical care than the heterosexual population [16,17]. In this context, MSM had significantly higher odds of efavirenz/nevirapine PDR compared to heterosexual cisgender men, possibly reflecting ongoing transmission of NNRTI resistance mutations such as K103NS and E138A. On the other hand, it is reassuring that both PDR and ADR to dolutegravir and bictegravir in 2020 remained low, although further surveillance is warranted given the short time of implementation of ART optimization nationwide. Taking this into consideration, most ADR cases observed in the present study could still be associated with failures to offer efavirenz-based first-line regimens in persons who had not ye<sup>t</sup> switched to INSTI-based options, which is also consistent with the high frequency of K103NS and M184VI observed in persons with ADR. Also noteworthy is the fact that persons arriving to care with advanced infection had higher odds of PDR. This could be associated with ART defaulters who, years later, return to clinical care because of complications associated with opportunistic infections, but could also sugges<sup>t</sup> a subpopulation of MSM characterized by late arrival to clinical care. Interestingly, this study identified a subset of individuals that did not answer several of the questions in the computer-based questionnaire and that were characterized overall by significantly higher PDR (Table 2). Further studies including in-depth interviews could be highly valuable for understanding this group, as it may present common vulnerability issues. Finally, although the level of ADR to dolutegravir/bictegravir observed in 2020 was low, it still warrants strengthening of strategies to improve adherence both at the clinic and community levels, especially in groups with a high risk of ART defaulting. This is especially relevant given that the overall level of ADR observed in this study was lower than that observed in other countries [5], possibly suggesting more frequent interruption of ART in persons without viral suppression. Also important is the high prevalence of cross-resistance to doravirine in persons with ADR. This drug is not ye<sup>t</sup> available in Mexico, but the current level of cross-resistance observed could limit its use locally.

The present study has important limitations worth mentioning. First, all participants were enrolled at a single institution, which may cause selection bias. Even though the Condesa clinic encompasses an important proportion of new diagnoses in the metropolitan area of Mexico City, a fraction of the population living with HIV could be underrepresented, especially persons living in areas with local social security clinics that offer HIV testing within the city. Nevertheless, even though the Condesa clinics care for persons lacking health insurance, 42% of HIV diagnoses performed at the clinics are in persons with social security [15], from which approximately 20% are lost to follow-up and most probably later return to clinical care in subsequent years or die. This fact strengthens the representativeness of the study population, even when coming from a single center. Second, information bias could exist, especially given that metadata collection was poor, especially at the beginning of the study period. Still, given its role as the most important HIV diagnosis center in Mexico City and the size of the clinic, we expect the study population to be highly representative of the population of persons living with HIV locally. Third, the present study excludes important populations that warrant further studies in both their specific vulnerabilities and structural challenges and their contribution to HIV PDR, especially male adolescents who have sex with men.

#### **4. Materials and Methods**

## *4.1. Study Population*

A cross-sectional, observational study was conducted at the Condesa Specialized Clinic, with two branches in Mexico City (located in the municipalities of Cuauhtémoc and Iztapalapa), between 2017–2020. The Condesa Specialized Clinic is the largest primary HIV care clinic and one of the main HIV diagnostic centers in Mexico, having diagnosed nearly 3500 individuals in 2020, approximately 70% of all the new infections in Mexico City's metropolitan zone [15]. All adults (>18 years) attending the Condesa Clinic for an HIV test, having received a positive result, including new diagnoses, referrals from other institutions, and persons returning to care after at least 3 months of ART defaulting, were invited to participate in the study between January 2017 and December 2020. These inclusion criteria were defined according to the WHO PDR definition [5]. Participants gave written informed consent to participate in the study. Since 2020, a computer-based self-administered questionnaire including demographic and clinical data was applied (a paper-based version of the questionnaire was available for participants preferring this option). Analysis of the variables included in the questionnaire was performed in a subset of the participants enrolled from June 2020 to December 2020. All participants donated a blood specimen for HIV sequencing and DR testing. HIV sequencing was performed at the Center for Research in Infectious Diseases of the National Institute of Respiratory Diseases (CIENI/INER), a reference center for HIV genotyping, following strict quality assurance processes. The study was reviewed and approved by the Institutional Review Board of the National Institute of Respiratory Diseases (project codes E12-17 and E02-20) and was conducted according to the principles of the Declaration of Helsinki.

Analysis of the characteristics and prevalence of acquired drug resistance was performed in an independent study group, including all clinically indicated HIV genotypes recorded at the National Ministry of Health HIV Database (SALVAR) from January 2020 to December 2020, as part of the national HIV program. HIV genotyping is recommended by the Mexican ART Guidelines for switching to second- and third-line regimens. All persons cared for at the Condesa clinic with two consecutive viral load values > 1000 copies/mL, whose second viral load test was performed in 2020, were included in the analysis. HIV sequencing was performed at CIENI/INER from the same blood specimen donated for the second viral load test, using internally validated Sanger sequencing methods, as explained below.

#### *4.2. HIV Amplification and Sequencing*

Sequences were obtained by next generation sequencing from a single amplicon including the complete protease (PR), reverse transcriptase (RT) and integrase (IN) genes (HXB2: PR 1-99, RT 1-560 and IN 1-288), using an in-house-validated method with Illumina sequencing technology on a MiSeq instrument (San Diego, CA, USA), as previously described [16,17]. A minority of the specimens in which amplification of this longer amplicon was not successful, as well as clinically indicated HIV genotypes for the ADR analysis, were amplified using a validated protocol developed by the US Centers for Disease Control and Prevention for the PR-RT region (HXB2 positions: PR 6-99, RT 1-251) [18] and an in-house developed and validated protocol for IN (HXB2: IN 1-288) [19]. These shorter amplicons were sequenced using NGS with standard Illumina protocols or by Sanger sequencing on a 3730xl Genetic Analyzer (ThermoFisher, Waltham, MA, USA) as previously described [10].

Next generation sequencing reads were filtered and assembled using HyDRA (Public Health Agency of Canada, Winnipeg, MB, Canada) [17,20]. Twenty percent consensus sequences (previously validated as Sanger-like sequences) were obtained and used for the HIVDR analyses [21]. Sanger sequences were assembled and edited using ReCall (BC Centre for Excellence in HIV/AIDS, Vancouver, BC, Canada) [22].

#### *4.3. HIV Drug Resistance Assessment*

Quality controls were applied to the sequences included in the database using the WHO HIVDR quality control tool [23,24]. Sequences not compliant with quality control were excluded from the study. Reasons for exclusion included inadequate sequence length, presence of stop codons, frameshift insertions/deletions, excess Apolipoprotein B mRNA-Editing Catalytic Polypeptide-like (APOBEC) or unusual mutations [23,24]. For participants with more than one sequence available, the first sequence was selected.

PDR was estimated using the Stanford HIVdb tool V.9.0 [24] and reported by drug class and individual drugs. Sequences with HIVDR were defined as those with a Stanford score ≥ 15 (at least low-level resistance) for efavirenz, nevirapine, any nucleoside reverse transcriptase inhibitor (NRTI), boosted darunavir, lopinavir, or atazanavir, raltegravir, elvitegravir, dolutegravir, or bictegravir, according to WHO standardized protocols [25]. PDR to INSTI was also reported referring to first-generation (raltegravir and elvitegravir) and second-generation INSTI (dolutegravir, bictegravir, and cabotegravir). HIVDR prevalence was estimated using a predefined Excel template developed for the WHO HIV ResNet Laboratory Network [26] (available by request from the corresponding author). HIV subtype was inferred using the REGAHIV-1 Subtyping Tool version 3.0 [27].

#### *4.4. Associations between Pretreatment Drug Resistance and Epidemiological Variables*

Exploratory analyses were performed comparing persons with and without PDR to any drug or with PDR to efavirenz/nevirapine, using Mann–Whitney U, chi square, or Fisher's exact tests, in accordance with the type of variable. Age and viral load were analyzed as continuous variables; gender, state of residence, marital status, education, social class, sexual risk, injectable drug use, use of venues for sex, and previous exposure to ARV were analyzed as categorical variables. CD4+ T cell count was stratified according to CDC clinical categories. Univariate associations between demographic, clinical and behavioral variables available and the presence of PDR to any ARV drug or PDR to efavirenz/nevirapine were explored with logistic regression, including only participants enrolled from June to December 2020, when completion of the metadata was best. Multivariable logistic regression models were constructed using all variables significantly associated with PDR in the univariate analyses. Additional variables were included a priori, owing to previous interest in HIVDR development. The best model was selected using Akaike information criterion, Bayesian information criterion, and Hosmer–Lemeshow goodness of fit test. Analyses were performed using STATA v16.

#### *4.5. HIV Transmission Network Inference*

The network was defined using a genetic matrix method based on PR-RT sequences, with Seguro HIV-TRAnsmission Cluster Engine (Seguro HIV-TRACE) [16], a locally adapted and secured version of the HIV-TRACE tool [28]. Clusters were defined when sequences showed pairwise Tamura–Nei 93 genetic distance < 1.5%. Although IN sequences were not used for network inference, the presence or absence of HIVDR to INSTI was considered as an additional attribute of each node.
