**3. Discussion**

In this study, a cross-sectional survey was conducted among 174 patients with ART interruption in five areas more heavily affected by HIV/AIDS in China. The prevalence of HIVDR was 19.5% at Sanger sequencing, which was obviously higher than a pretreatment HIVDR survey conducted among ART naïve patients in 2017 [23]. A cross-sectional survey of patients with a median of 13.9 months on antiretroviral therapy in eight provinces in China showed that the prevalence of HIVDR in patients receiving ART was 4.3% [24]. Other studies have also suggested that the prevalence of drug resistance in patients receiving ART in some areas of China during the same period is less than 5% [25–27]. Compared with patients who have been receiving antiretroviral therapy, patients whose ART has been interrupted have a higher HIVDR prevalence. Furthermore, the rate of resistance to EFV and/or NVP (15.5%) was more than three times that (4.6%) in those patients. These findings were consistent with those in other countries [28], providing more evidence that patients with ART interruption should consider substituting EFV/NVP for PI or integrase inhibitors when re-initiating antiretroviral treatment.

It was unexpected to ge<sup>t</sup> higher rates of HIVDR at NGS than through Sanger sequencing. However, the increase in HIVDR was not statistically significant until the detection threshold was lowered to 5%. Moreover, DRMs detected in protease would have limited impacts on such PIs as Lopinavir/r (LPV/r) in second-line regimens. Four DRMs including D30N (1), M46I (2) and N88D (1) were detected in protease in four participants at a 5%–15% frequency, in addition to one Q58E at a frequency above 90%. The former two confer high or intermediate resistance to Nelfinavir, which is seldomly used at present. N88D and Q58E are PI accessary resistance mutations. There were slight increases in NRTI and NNRTI resistance at the threshold of 5%, when compared with the threshold of 20% or SS, which was consistent with other previous studies [29,30].

Despite the same primer sets for reverse transcription and PCR amplification, there were three NNRTI DRMs in three patients with frequencies between 23% and 44% detected by NGS, not SS. These discrepancies were more likely to be PCR resampling produced at the first several rounds of amplification, and less likely to be the artifacts during NGS sequencing, as NGS artifacts had a <1% frequency and were randomly distributed. Compared with that at the threshold of 5%, the rates of HIVDR were almost doubled when the threshold decreased to 2%, and tripled when down to 1%, as very-low-frequency DRMs were dramatically increased between 1% and 5%. The minority mutations detected by NGS may have resulted from apolipoprotein B mRNA editing catalytic polypeptide (APOBEC)- mediated G-to-A hypermutation such as E138K, M184I and G190E in RT or spontaneous mutation during viral replication, or from PCR error, which was introduced by error-prone reverse transcriptase or PCR enzymes [31]. The sharp increase in low-frequency DRMs agrees with the 5% detection threshold mentioned in other studies [32,33]. At present, the optimal threshold for the NGS detection of low-frequency mutation sites is still inconclusive. Studies have shown that a 5% threshold can provide a reproducible clinically relevant treatment result [34], suggested using a detection threshold of 5% to minimize technical artifacts [31,35]. Using a 5% threshold to report low-frequency variation allows NGS to achieve a moderately increased sensitivity to detect low-frequency DRMs, without compromising sequence accuracy [36,37].

The most common NNRTI DRM above a 5% frequency was K103N (14.4%) in RT, which is consistent with the ART history of EFV/NVP-based first-line therapy. In NRTI DRMs, K65R was the highest (1.2%) at a >5% frequency, which was similar to other research results [29,38]. In addition, this mutation was dramatically increased to 20.7% at 2%–5% frequency. We also found that the distribution of the K65R mutation was related to the HIV subtypes CRF07\_BC and CRF08\_BC. The K65R mutation causes high-level resistance to tenofovir and intermediate resistance to lamivudine, which are the backbone of first- and second-line regimens. The minority K65R mutations may become major mutations when re-initiating ART, giving rise to virological failure. The subtype-specific

differences in the K65R mutation distribution are mainly due to differences in codon usage [39], as it is subtype C in this region in the CRF07\_BC and CRF08\_BC strains.

The disease progression in patients with HIV is usually monitored by measuring their CD4+ T cell counts. After the interruption of antiretroviral therapy, the CD4+ T cell count drops rapidly in the first few months, then declines slowly [40,41]. A previous study did not find a significant difference in the CD4+ T cell count between patients with and without HIVDR variants [42]. In this study, the patients were divided into three groups across mutation frequencies detected by NGS at a 15% threshold: patients without variants, patients with low-frequency variants and patients with high-frequency variants. Interestingly, the Kruskal–Wallis test found significant differences in the CD4+ T cell counts among the three groups. After the interruption of ART, the viral load level increased exponentially, reaching a peak of 10<sup>6</sup> copies/mL. Then, the viral load usually drops to 103–105 copies/mL within a few weeks, which can remain relatively stable for many years [43]. In this study, the viral load of patients under ART interruption was above 10<sup>3</sup> copies/mL, and there were statistical differences in the viral loads among the three groups.

Generally speaking, the drug resistance rate will decrease over time in patients after ART interruption. This trend is mainly due to the fact that, in the absence of drug selection pressure, wild-type strains with high viral fitness replace resistant strains to become the main strains in patients [44]. Compared with the baseline, the HIVDR mutations, such as N88D, K65R, M184VI, K103N, E138AG, V179D and P225H, still existed but the frequency gradually decreased, consistent with earlier studies [18,42]. Nevertheless, we found new low-frequency mutations such as I84V, F77L and V106A at follow-ups in this study. A study also showed that new mutations emerge in the absence of drug selection pressure [18]. The emergence of new mutations may be due to the reason that, without the drug selection pressure, virus strains carrying resistance-related mutations may have reduced replication fitness, so they cannot become dominant strains. The results of this study show that new mutations emerge in the form of low frequency, but whether they will continue to exist as the main drug-resistant strains over time is still unknown, and further research is needed.

There are limitations to this study. Firstly, the number of participants was limited. The findings could not be simply extrapolated to other patients with ART interruption. However, the fall or rise in DRMs would be similar to the other settings without ART. The minority DRMs detected by NGS were consistent with other pre-treatment surveys. Secondly, the participants in this study mainly used NNRTI-based first-line regimens before ART interruption. Thus, the findings were not able to reflect the situations with the interruption of ART on PI or integrase inhibitors. Thirdly, the rate of follow-up was low, as patients with ART interruption also dropped out of the HIV care program and were reluctant to keep the follow-up appointments.

In conclusion, patients with ART interruption had higher HIVDR, especially to EFV/NVP. NGS is able to detect more DRMs in such patients than SS, but largely at frequencies of less than 5%. Studies have shown that the drug resistance reports based on a 5% threshold can predict the failure of ART [21,45]. Due to the limitations of this study, the optimal NGS threshold for predicting the failure of antiviral therapy has not ye<sup>t</sup> been determined. Further studies are needed to define the NGS detection threshold for providing a predicative of ART failure.

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

#### *4.1. Study Design and Study Participants*

We conducted a cross-sectional study among adult patients with ART interruption between 1 January and 31 December 2016, in four provinces and one municipality of China, including Guangxi, Henan, Sichuan, Yunnan and Chongqing, where in China the HIV epidemic was more concentrated, as described previously [46]. Inclusion criteria were: patients aged eighteen years or older, had discontinued ART for at least one month at survey, had received ART for at least three months before ART interruption and had agreed

to provide written informed consent. These patients were followed up in 2017. Plasma samples were handled locally and transported in dry ice to the laboratory at the National Center for AIDS/STD Control and Prevention (NCAIDS), Chinese Center for Disease Control and Prevention (China CDC), where the HIV viral load was measured, and HIV drug resistance was genotyped on samples with a viral load ≥1000 copies/mL.

#### *4.2. HIV RNA Extraction and Sample Amplification*

Viral RNA was extracted using the QIAamp viral RNA mini kit (Qiagen, Hilton, Germany), following the manufacturer's protocol. Extracted RNA was used for amplification of the PR region (4–99 amino acids) and the partial RT region (1–251 amino acids) of the HIV-1 *pol* gene region using an in-house method [47]. Briefly, the first-round PCR was performed in 25 μL volume reactions using the one-step RT-PCR kit (Promega, Madison, WI, USA) with the primers PRTM-F1 and RT-R1. The second-round PCR was performed in 50 μL volume reactions using the primers PRT-F2 and RT-R2.

#### *4.3. Sanger Sequencing and Drug Resistance Analysis*

The amplified products were sequenced using Sanger sequencer ABI 3730 (Applied Biosystems, Carlsbad, CA, USA). Obtained raw sequences were edited and assembled using Sequencher (version 4.10.1, Gene Codes Corporation, Ann Arbor, MI). The secondary peak threshold was set to 20%. The edited sequence was aligned with the HXB2 reference sequence through BioEdit (version 7.0.9, Informer Technologies Inc.). Sample contamination and other sequence quality controls were monitored using the WHO HIVDR QC tool (https://recall.bccfe.ca/who\_qc/, accessed on 23 September 2019) and through constructing Neighbor joining phylogenetic trees using MEGA (version 6.06). HIV-1 subtypes were determined using reference sequences from HIV Databases (https://www.hiv.lanl.gov/content/index, accessed on 8 October 2019) through phylogenetic trees. The drug resistance mutations were identified and interpreted using the HIVdb version 8.9-1 (https://hivdb.stanford.edu/hivdb/, accessed on 10 October 2019).

#### *4.4. Next-Generation Sequencing and Drug Resistance Analysis*

Second-round amplicons were cleaned using KAPA PureBeads (Roche, Basel, Switzerland) and quantified using the Qubit dsDNA HS assay kit (Thermo Fisher Scientific, Carlsbad, CA, USA). Sequencing libraries were prepared using the 96-sample KAPA HyperPlus Kit (Roche, Basel, Switzerland). The samples were pooled in each run with a 40% Phix control library (v3, Illumina, San Diego, CA, USA) to increase the diversity of libraries and then sequenced on the Illumina Miseq system [48] using a v3 600-cycle reagen<sup>t</sup> kit (Illumina, San Diego, CA, USA). The raw data were analyzed using the HyDRA Web tool (http://hydra.canada.ca/, accessed on 20 January 2020) according to *HyDRA Web User Guide* [49], producing lists and frequencies of drug resistance mutations, and then interpreted by using the HIVdb algorithm.

## *4.5. Statistical Analysis*

DRMs were identified according to the 2019 International Antiviral Society (IAS) list. HIV drug resistance was defined as low-, intermediate- or high-levels of drug resistance according to the HIVdb algorithm 8.9-1. HIV drug resistance was determined based on NGS with detection thresholds of 20%, 15%, 10%, 5%, 2% and 1%, and was compared with that detected based on Sanger sequencing by the McNemar test, respectively. Quantitative data are presented as medians and were analyzed by the Kruskal–Wallis test. All data were analyzed using the Statistical Analysis System version 9.4 (SAS Institute Inc., Cary, NC, USA). Two-sided *p* values of <0.05 were considered statistically significant.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/2076-0817 /10/3/264/s1, Table S1: The HIV drug resistance mutations of each sample (*n* = 174) were detected by Sanger sequencing and the next-generation sequencing; Table S2: The drug resistance mutations of each sample (*n* = 49) at baseline and follow-up detected by the next-generation sequencing, and the value in brackets is the mutation frequencies.

**Author Contributions:** M.L., H.X., Y.R., Y.S. and L.L. (Lingjie Liao). were responsible for study design and planning. S.L. (Shujia Liang), C.Z., M.C., S.L. (Shu Liang) and C.L. contributed to data collection. M.L., L.L. (Lei Liu), Z.Z., Y.F., C.S., H.X., Y.R., Y.S. and L.L. (Lingjie Liao) contributed to data analysis and interpretation. M.L., H.X., Y.R., Y.S. and L.L. (Lingjie Liao) contributed to writing the manuscript. All authors read and approved the final version of the manuscript.

**Funding:** This work was supported by National Natural Science Foundation of China (11971479), the Ministry of Science and Technology of China (grant 2017ZX10201101, 2018ZX10721102-006), the State Key Laboratory of Infectious Disease Prevention and Control, the Department of Science and Technology of Guangxi Zhuang Autonomous Region (AB16380213), the Key technical posts for HIV/AIDS prevention and control of Guangxi Bagui scholars, and special funds for AIDS prevention and control. The fund providers had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.

**Institutional Review Board Statement:** Institutional review board approval was granted by the National Center for AIDS/STD Control and Prevention, Chinese Center for Disease Control and Prevention.

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** Data are contained within the article.

**Conflicts of Interest:** The authors declare that there are no conflict of interest.
