**3. Discussion**

The spectrum of atazanavir-selected mutations has been largely influenced by data published in the earliest in vitro passage experiments and clinical trials. During in vitro passage experiments with three subtype B clones, the most commonly emerging DRMs were V32I, M46I, I50L, I84V, and N88S [42]. The initial reports of the in vivo selection of PI-associated DRMs, based on the use of unboosted atazanavir in ART-naïve patients, demonstrated that I50L and G73S were the most commonly occurring mutations in patients with VF [24,43]. A few cases of VF and emergen<sup>t</sup> PI-associated DRMs have been reported in the clinical trials of ART-naïve patients receiving boosted atazanavir [16,44], consistent with the hypothesis that PI-resistance mutations develop only in viruses exposed to a narrow window of suboptimal drug concentrations that both exert selective pressure on the virus and allow virus replication [45]. Nonetheless, PI resistance in previously PInaïve patients receiving lopinavir/r for second-line therapy has increasingly been reported, usually beginning after 12–18 months of therapy [46]. In addition, phenotypic studies have shown that DRMs selected by other PIs confer atazanavir cross resistance particularly when they occur in combination [47,48].

In the years since atazanavir has been introduced, there has been a gradual accumulation of data on the spectrum of mutations emerging in previously PI-naïve patients with VF on an ART-regimen containing boosted and less commonly unboosted atazanavir. In contrast to the earliest clinical trials of boosted atazanavir, these studies included cohorts of patients who were ART experienced at the time atazanavir was administered and who may not have been monitored as closely for VF thus enabling their viruses to evolve for longer period of time under atazanavir selection pressure. Moreover, these studies have included an increasing proportion of sequences from patients with non-subtype B viruses.

Our analysis confirmed that the five major DRMs selected in vitro by atazanavir (V32I, M46I, I50L, I84V, and N88S) were among the most commonly occurring major DRMs. Four additional DRMs also occurred commonly, including M46L, I54V, V82A, and L90M. I50L is a signature atazanavir-associated DRM because it has only been reported in patients receiving atazanavir and it increases susceptibility to PIs other than atazanavir [48,49]. N88S is also considered a signature atazanavir-associated DRM because it is rarely selected by other PIs and it does not significantly reduce susceptibility to other PIs, with the exception of nelfinavir and indinavir [48]. L/M89T may also be a signature atazanavir-associated mutation because it appears to occur more commonly in patients receiving atazanavir than in patients receiving any other PI; for example, it has only been reported in three previously PI-naïve patients receiving lopinavir/r (https://hivdb.stanford.edu/cgi-bin/Mutations. cgi?Gene=PR; accessed on 1 November 2021). In contrast, each of the remaining atazanavirselected mutations appear to be commonly selected by other PIs, in particular lopinavir/r.

With the exception of L10F and L33F, each of the above 17 most commonly selected major or accessory DRMs was significantly associated with reduced atazanavir susceptibility in a previously published weighted least squares regression analysis of 1644 sequences [48]. Few published phenotypic data are available on sequences containing L89T.

Some limitations of our review should be discussed. First, most of the sequences that we reviewed were obtained from retrospective cohort studies and case series. For these studies, the duration of therapy, accompanying ARVs, frequency of virological monitoring and genotypic resistance testing, and duration of virological failure were generally not available. Therefore, the extent to which these factors were associated with emergen<sup>t</sup> PI-associated DRMs could not be explored. Second, the dataset contained an underrepresentation of subtypes other than subtype B. Third, we could not be sure that every sequence was obtained from a patient receiving atazanavir as his/her first PI as treatment histories are often incomplete. Nonetheless, we emailed the authors of those studies containing the largest numbers of DRMs and received confirmation that, to the authors' knowledge, the patients had just received atazanavir. Fourth, at least 32 studies in PubMed that contained sequences from 1000 patients receiving boosted or unboosted atazanavir could not be included in our analysis because the primary sequence data and complete list of protease mutations were not available.

In conclusion, to our knowledge, this is the only comprehensive analysis of atazanavirselected mutations. Our analysis shows that the spectrum of atazanavir-selected mutations extends beyond those mutations observed in the earliest clinical trials in which patients received either boosted or unboosted atazanavir. The expanded spectrum is likely due to the large number of sequences in our analysis and the likelihood that many of the patients in the studies we reviewed had prolonged VF and ongoing replication while receiving atazanavir. The study also identified one novel nonpolymorphic atazanavirselected mutation that predominantly occurred in non-subtype B sequences. The relatively low cross-resistance to darunavir/r combined with preliminary data suggests that boosted atazanavir can be an efficacious regimen for second-line therapy. However, comparative clinical trials are required to determine the optimal boosted PI to use for second-line and potentially later-line therapy in LMICs.

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

#### *4.1. Study Selection Criteria*

We analyzed publicly available HIV-1 group M protease nucleotide sequences obtained from previously PI-naïve patients receiving boosted or unboosted atazanavir. Sequences were obtained from HIVDB, which is populated with sequences from GenBank annotated with the ART history of the patients from whom the sequences were obtained [13]. The analysis was last updated 31 December 2021. We supplemented the data in HIVDB with previously unpublished sequences performed at SUH and with previously unpublished sequences from two collaborating research groups: the EIDB [14] and the RHIVDB [15]. Additionally, we performed a PubMed search to identify studies describing HIV-1 group M protease sequences that were not present either in HIVDB or GenBank.

Publications reporting eligible protease sequences were reviewed to determine the treatment history of the patient from whom each sequence was obtained to confirm that the patient had received no PI prior to atazanavir and to distinguish those patients receiving unboosted atazanavir from boosted atazanavir. Each sequence was annotated with the year and country of virus isolation, the type of sample (e.g., PBMCs), the sequencing method (Sanger dideoxynucleoside sequencing versus NGS), and the nature of the study population. HIV-1 subtype was determined using the HIVDB subtyping program [50].

We also characterized each study according to whether it included patients in a clinical trial or in a treatment cohort for whom genotypic resistance testing was routinely available for patients with VF as opposed to a case series or case reports for which the indications for genotypic resistance testing were not reported. Studies that performed routine genotypic resistance testing on all patients with VF provide information on how often PI resistance arises in patients receiving atazanavir. In contrast, the remaining studies were considered likely to be enriched for patients with acquired PI resistance.
