Therapeutic prediction of HIV-1 DNA decay: a multicenter longitudinal cohort study

Background Factors predicting peripheral blood total HIV-1 DNA size in chronically infected patients with successfully suppressed viremia remain unclear. Prognostic power of such factors are of clinical significance for making clinical decisions. Methods Two sets of study populations were included: 490 China AIDS Clinical Trial (CACT) participants (Training cohort, followed up for 144 to 288 weeks) and 117 outpatients from Peking Union Medical College Hospital (PUMCH) (Validation cohort, followed up for more than 96 weeks). All patients were chronically HIV-1-infected and achieved successful HIV-1 plasma RNA suppression within week 48. Total HIV-1 DNA in blood at baseline, 12, 24, 48, 96, 144 and 288 weeks after combined antiretroviral therapy (cART) initiation were quantified. Generalized estimating equations and logistic regression methods were used to derive and validate a predictive model of total HIV-1 DNA after 96 weeks of cART. Results The total HIV-1 DNA rapidly decreased from baseline [median = 3.00 log10 copies/106 peripheral blood mononuclear cells (PBMCs)] to week 24 (median = 2.55 log10 copies/106 PBMCs), and leveled off afterwards. Of the 490 patients who had successful HIV-1 plasma RNA suppression by 96 w post-cART, 92 (18.8%) had a low total HIV-1 DNA count (< 100 copies/106 PBMCs) at week 96. In the predictive model, lower baseline total HIV-1 DNA [risk ratio (RR) = 0.08, per 1 log10 copies/106 PBMCs, P < 0.001] and higher baseline CD4+ T cell count (RR = 1.72, per 100 cells/μL, P < 0.001) were significantly associated with a low total HIV-1 DNA count at week 96. In an independent cohort of 117 patients, this model achieved a sensitivity of 75.00% and specificity of 69.52%. Conclusions Baseline total HIV-1 DNA and CD4+ T cell count are two independent predictors of total HIV-1 DNA after treatment. The derived model based on these two baseline factors provides a useful prognostic tool in predicting HIV-1 DNA reservoir control during cART. Supplementary Information The online version contains supplementary material available at 10.1186/s12879-021-06267-5.


Introduction
A critical goal of antiretroviral therapy is to reduce the size of the total HIV-1 DNA reservoir which poses a major obstacle for HIV-1 eradication [1][2][3]. Total HIV-1 DNA load has been associated with disease progression, treatment efficacy, and HIV-1 co-morbidities such as HIV-1-associated dementia and malignancy [4][5][6][7]. High total HIV-1 DNA levels during cART have been associated with faster viral rebound after structured treatment interruption (STI) [8][9][10], while low total HIV-1 DNA levels may increase the possibility of prolonged viral remission after cART interruption, which was observed in the Mississippi baby [11] and post-treatment controllers (PTCs) [12,13]. In the VISCONTI cohort study, the median total HIV-1 DNA level was near 100 copies/10 6 PBMCs in the 14 PTCs when cART was discontinued, which served as a milestone of relatively successful HIV-1 reservoir control, potentially leading to prolonged remission or functional cure [12].
It is generally accepted that early treatment is critical for the viral suppression of PTCs [12]. Numerous studies have investigated factors related to the dynamics of total HIV-1 DNA decay during cART, including treatment initiation time, antiretroviral therapy intensity, demographic data, baseline HIV-1 plasma RNA and DNA, CD4+ and CD8+ cell count and CD4/CD8 ratio. Early initiation of cART during primary infection, low baseline total HIV-1 DNA and high CD4+ nadir were shown to increase the likelihood of achieving low total HIV-1 DNA outcome [14][15][16][17]. In addition, a cross-sectional study has reported that 28% out of 522 patients who are initiated cART during the chronic phase can achieve low total HIV-1 DNA (< 150 copies/10 6 PBMCs) after long-term viral suppression [18]. Thus, as it is associated with ART-free remission, total HIV-1 DNA could serve as a usefully virologic parameter for monitoring therapeutic effects.
For clinicians, to translate the HIV-1 DNA test result into a prognostic prediction of certain ART is intriguing, especially when the HIV-1 viral load remains undetectable. Our previous study has shown that after 96 weeks' ART, low HIV-1 DNA level is significantly associated with higher CD4/CD8 ratio [19]. However, baseline factors that predict low total HIV-1 DNA level in chronic patients after treatment remain elusive.
As such, in this study, we included 490 chronically HIV-1-infected treatment-naïve patients enrolled in multicenter HIV-1 cohort studies [20][21][22] and 117 outpatients in China. For the first time, a statistical model predicting the possibility of achieving a low total HIV-1 DNA outcome (< 100 copies/10 6 PBMCs) for patients with chronic HIV-1 infection under suppressive cART was derived and validated in two independent cohorts.

Subjects
The Institutional Review Board of the Peking Union Medical College Hospital approved this study, and written informed consents were obtained from all the study participants. All methods were performed in accordance with the relevant guidelines and regulations. This study consists of two sets of study populations: 490 China AIDS Clinical Trial (CACT) participants and 117 outpatients from Peking Union Medical College Hospital (PUMCH).
In addition, another cohort consists of 117 HIV-1infected adult outpatients, who were successfully treated (treatment was the same as CACT patients) between 2009 to 2016 at PUMCH and followed up for more than 96 weeks, was used as an independent validation cohort. The inclusion criteria for both the training and validation cohorts were: (1) complete baseline characteristics including age, gender, transmission route, and subtype, CD4+ and CD8+ T cell count, and HIV-1 plasma RNA levels; (2) total HIV-1 DNA levels must be available at baseline and week 96, while for other time points of 12 w, 24 w, and 48 w, only less than 2 missing values were allowed; (3) baseline total HIV-1 DNA level ≥ 100 copies/10 6 PBMCs; and (4) Successful HIV-1 plasma RNA suppression within week 48.

Total HIV-1 DNA quantitation
Total HIV-1 DNA was extracted from peripheral blood containing 0.25-1 million PBMCs using the Qiagen QIAsymphony DNA Mini Kit (QIAGEN, Hilden, Germany). The SUPBIO total HIV-1 DNA Quantitative PCR Kit (SUPBIO, Guangzhou, China) was used for simultaneously quantitating total HIV-1 DNA and cell number, following the manufacturer's instructions. The linear quantification range of the SUPBIO total HIV-1 DNA quantitative kit was 20 copies/10 6 PBMCs to 100, 000 copies/10 6 PBMCs.

Statistical methods
Total HIV-1 DNA and RNA values were log-transformed in all inferential analyses. Generalized linear models of the longitudinal data using a generalized estimating equation (GEE) were performed to select the factors that predict the binary treatment outcome-total HIV-1 DNA < 100 copies/10 6 PBMCs at week 96. Logistic regression models were used to characterize how the binary treatment outcome can be predicted by the baseline values of factors identified as a significant predictor in the final GEE model. The receiver operating characteristic (ROC) curve was utilized to evaluate the performance of the prediction power of the logistic regression model in the training set. The software packages SPSS 22.0 (IBM Corporation, Armonk, New York, USA) and MATLAB (R2014a, MathWorks, Natick, MA) were used for data analyses. A P value of < 0.05 was considered statistically significant.

Baseline characteristics
Baseline characteristics were summarized for both the training (n = 490) and validation (n = 117) cohorts in Table 1. For the training cohort, the majority of patients were aged 30-40 years, male (69.8%), infected via sexual transmission [men who have sex with men (MSM) 31.4%, heterosexual 52.7%], and infected by the CRF01_ AE subtype (44.5%). The median CD4+ T cell count was 226 cells/μL and median CD4/CD8 ratio was 0.25. The median HIV-1 plasma RNA level was 4.66 log 10 copies/ mL, and median total HIV-1 DNA was 3.00 log 10 copies/10 6 PBMCs. For the validation cohort, the majority were male (92.3%) and 71.8% of them were MSM. However, the clinical characteristics, including CD4+ and CD8+ T cell counts and HIV DNA levels, were similar between the two cohorts.
Total HIV-1 DNA decay during 288 weeks of cART During 288 weeks of cART, total HIV-1 DNA levels rapidly decreased from a median value of 3.00 log 10 copies/ 10 6 PBMCs at baseline to 2.55 log 10 copies/10 6 PBMCs at week 24 (P < 0.001, Fig. 1), then leveled off near the level of 2.55 log 10 copies/10 6 PBMCs during weeks 24 to  Fig. 1). In the following study, we consider the total HIV-1 DNA value at week 96, when all 490 patients had HIV-1 DNA data, as the HIV-1 DNA outcome after cART.

Statistical model based on baseline DNA and CD4+ T cells to predict total HIV-1 DNA under cART
Further, generalized estimating equations (GEE) and logistic regression methods were used to derive a predictive model of total HIV-1 DNA < 100 copies/10 6 PBMCs after 96 weeks of cART. With the GEE analysis, we found that longitudinal total HIV-1 DNA, CD4+ T cells, and CD4/CD8 ratio values were correlated with the total HIV-1 DNA outcome (Supplementary Table S1). Univariate and multivariate logistic regression analyses were then performed to relate the binary outcome with each identified predictor and combinations thereof. Total HIV-1 DNA and CD4 count were screened out as significant predictors of the treatment outcome of total HIV-1 DNA. Interestingly, the highest area under the ROC curve (AUC) of 0.82 was observed in the multivariate model with baseline total HIV-1 DNA and CD4 count other than each univariate model ( Fig. 2A). As such, the best-performed predictive model based on baseline total HIV-1 DNA and CD4 count was derived: logit(θ) = 4.579-2.518 × DNA + 0.477 × CD4 (Eq. 1).
Where θ represents the probability of achieving total HIV-1 DNA < 100 copies/10 6 PBMCs at week 96 post-cART; DNA represents the baseline total HIV-1 DNA level in log 10 copies/10 6 PBMCs; and CD4 represents the baseline CD4+ T cell count in 100 cells/μL. For example, if a patient starts cART treatment with 300 (2.48 log 10 ) copies/10 6 PBMCs of baseline total HIV-1 DNA and a CD4 count of 500 cells/μL, there is a 72% (θ = 0.72) chance that his total HIV-1 DNA will fall below 100 copies/10 6 PBMCs after 96 weeks of treatment according to eq. 1.
With Eq. 1, we validated the prediction model with the validation cohorts, which contained 117 outpatients from the real hospital settings. At cutoff 2, the greatest true positive rate (TPR) and false positive rate (FPR) distance, the predictive model performed similarly in both training and validation cohorts. TPRs were 83.70 and 75.00% while FTRs were 29.40 and 30.48% in the training and validation cohorts, respectively (Fig. 2B). In this independent cohort of 117 patients, this model achieved a sensitivity of 75.00% and specificity of 69.52%.

Discussion
In this study, the characteristics and impact factors of HIV-1 DNA decay was studied with longitudinal cohorts. A probability model that predicts the outcome of HIV-1 DNA reservoir size after 2 years of cART using baseline values of total HIV-1 DNA and CD4+ T cell count was established for the first time. The model performed well and similarly in both the training cohort and an independent outpatient validation cohort, which indicated the robustness of the model.
Globally, most HIV-1 patients on treatment are chronically infected. Achieving remission or a potential functional cure in such patients is more challenging than in those who are acutely infected [10,12]. Studies of the HIV-1 reservoir in chronic patients were limited by small populations or were cross-sectional [18,23,24]. To our knowledge, the data in the present study represent the largest longitudinal cohort study focusing on chronically HIV-1-infected individuals. We found that for chronically HIV-1-infected patients, total HIV-1 DNA decay after initiation of cART demonstrated a Cutoff 1, the lowest threshold to obtain a TPR greater than 90%; Cutoff 2, the greatest TPR and FPR distance; Cutoff 3, the smallest threshold to obtain an FPR less than 20% biphasic reduction within 96 weeks. Most decline happened in the first 6 months and then began to plateau, which was consistent with other findings in chronically HIV-1-infected patients [25,26]. Compared with the studies by Besson et al. [25] and Gandhi et al. [27], our sample size was larger, and we detected the HIV-1 DNA at week 12 and 24. We suggested for the first time that the HIV-1 DNA decline was mainly within 24 weeks rather than 48 weeks (1 year). In addition, Gandhi et al. found that there was a positive correlation between the level of HIV-1 DNA before treatment and on treatment.
Our study further confirmed that baseline HIV-1 DNA was an independent predictive factor of HIV-1 DNA after 2 years of treatment. Furthermore, 18.8% of chronic patients achieved < 100 total HIV-1 DNA copies/10 6 PBMCs after 96 weeks of cART. In comparison, with prolonged treatment as long as the median duration of 13 years, it was reported that 28% of such patients could achieve~150 total HIV-1 DNA copies/10 6 PBMCs [18].
Our data showed that the pre-treatment total HIV-1 DNA burden and CD4+ T cell count correlate to long term total HIV-1 DNA control in real hospital settings. This model could be applicable for clinical decisionmaking on therapy strategies. Our findings suggest that chronically infected patients initiating cART with lower total HIV-1 DNA and higher CD4+ T cell baseline levels have a higher possibility of achieving low total HIV-1 DNA level with effective cART. In 2016, HIV-1 DNA has been used as one of the criteria for screening subjects in treatment-interruption study by Calin et al. [28].
In future studies such as treatment-interruption, deescalation therapy, or cell or gene therapy, clinicians could select suitable subjects who may benefit more by preliminary prediction of HIV-1 DNA outcome prior to patients' initial treatment. This study also supports initiatives for early treatment. Although delayed treatments can also keep viral plasma RNA within the detection limits, they maintain high DNA levels.
One limitation existed in our study was that we did not account for other possible mechanisms contributing to total HIV-1 DNA decay, including duration of infection, antibody responses [29], and polyfunctionality of HIV-1-specific T cells [30]. These present interesting directions for future hypothesis-driven investigations. In addition, the predictive model was established based on specific HIV-1 DNA testing methods and specific populations. The results of the study necessitate further verification in different settings and with different methods.

Conclusions
In summary, this is a longitudinal study in a large population of chronically HIV-1-infected patients to evaluate the predictive factors associated with total HIV-1 DNA decay after cART. Based on baseline total HIV-1 DNA and CD4, we established a predictive model for treatment-naïve patients to estimate their likelihood of achieving total HIV-1 DNA < 100 copies/10 6 PBMCs after suppressive cART. This model will be applicable for identifying potential cART-cessation candidates who are relatively more likely to achieve functional cure before cART initiation.  The funders had no role in study design, data collection, data analyses, preparation of the manuscript, or decision to publish.

Availability of data and materials
The datasets used and/or analyzed in the current study are available from the corresponding author on reasonable request.