Acute HIV-1 infection viremia associate with rebound upon treatment interruption

Summary

Background

Analytic treatment interruption (ATI) studies evaluate strategies to potentially induce remission in people living with HIV-1 but are often limited in sample size. We combined data from four studies that tested three interventions (vorinostat/hydroxychloroquine/maraviroc before ATI; Ad26/MVA vaccination before ATI and VRC01 antibody infusion during ATI).

Methods

The statistical validity of combining data from these participants was evaluated. Eleven variables including HIV-1 viral load at diagnosis, Fiebig stage and CD4+ T cell count were evaluated using pairwise correlations, statistical tests, and Cox survival models.

Findings

Participants had homogeneous demographic and clinical characteristics. Because an antiviral effect was seen in participants who received VRC01 infusion post-ATI, these participants were excluded from the analysis, permitting a pooled analysis of 53 participants. Time to viral rebound was significantly associated with variables measured at the beginning of infection: pre-ART viral load (HR=1.34, p=0.022), time to viral suppression post-ART initiation (HR=1.07, p<0.001) and area under viral load curve (HR=1.34, p=0.026).

Conclusions

We show that higher viral loads in acute HIV-1 infection were associated with faster viral rebound, demonstrating that the initial stage of HIV-1 infection prior to ART initiation has a strong impact on viral rebound post-ATI years later.

Graphical Abstract

eTOC blurb

Mdluli et al. analyzed data from 53 persons who initiated treatment upon diagnosis of acute HIV-1 infection and underwent treatment interruption three years later. Viral load dynamics at the beginning of HIV-1 infection, rather than variables measured prior to treatment interruption, were associated with time to rebound.

Introduction

While antiretroviral therapy (ART) is undeniably effective at limiting HIV-1 replication, prolonged treatment is not sufficient to eradicate the viral reservoir in people living with HIV-1 (PLWH). Upon treatment interruption, plasma viral rebound occurs in a relatively short time frame in most individuals (two to four weeks)¹. Nonetheless, some individuals can become post-treatment controllers and maintain undetectable viral loads for years after treatment interruption^2,3. Analytical treatment interruption (ATI) studies, where ART is temporarily stopped under close supervision, are key to identifying features that may be associated with ART-free viral control. The identification of such biomarkers would allow for assessment of new treatment modalities without the risk associated with interrupting ART⁴.

Previous studies have shown that individuals who initiated ART in acute/early HIV-1 infection were more likely to control viremia following treatment cessation^2,3,5,6. The ATI studies that enrolled participants from the RV254 prospective acute HIV-1 infection cohort in Bangkok, Thailand offer a unique opportunity to investigate factors associated with viral control after treatment interruption. Four ATI studies, including three randomized controlled trials (RCT), enrolled RV254 participants. These ATI participants started ART upon diagnosis in Fiebig stage I to IV^7,8 and were virally suppressed for a median of three years before enrollment in an ATI study (range 1.78–6.57 years). In the RV411 ATI study, all eight participants had initiated ART in Fiebig I and had undetectable plasma viral load for at least 2.4 years. Post-ATI, viral rebound (RNA ≥ 1,000 copies/mL) occurred a median of 31 days post treatment interruption (range 15–68)⁹. In the RV409 RCT (open label), participants received vorinostat/hydroxychloroquine/maraviroc (VHM, n=9) or placebo (n=5) prior to treatment interruption. Viral rebound ocurred at a median of 28 days (range 14–49) in the active arm versus 35 days (range 28–112) in the placebo arm, with no difference between groups (p=0.425)¹⁰. In the RV405 RCT (double blind), participants received a therapeutic HIV-1 Ad26/MVA mosaic vaccine regimen (n=18) or placebo (n=9) prior to treatment interruption. There was a modest delay in time to rebound in vaccine recipients (median: 28, range 13–46 days) compared to placebo recipients (median: 20, range 13–24 days) (p=0.003)¹¹. In the RV397 RCT (double blind), participants were randomized to receive a broadly neutralizing human monoclonal antibody (VRC01, n=13) or placebo (n=5) infusion on the day ART was stopped. There was a modest delay in time to rebound in VRC01 recipients, (median 33, range 13–305 days) compared to placebo recipients (median 14, range 14–32 days, p=0.010)¹² and no evidence that VRC01 lead to viral escape¹³.

While these ATI studies could evaluate strategies that can enable HIV-1 remission, the limited number of participants in each study and the known individual variability complicate the identification of features associated with ART-free control. We first tested whether it was statistically justifiable to combine data from the four ATI studies to determine if certain parameters were associated with viral rebound kinetics. Second, we investigated whether variables measured either during the first weeks of infection or pre-ATI were more determinant in predicting time to viral rebound. By following individuals from their diagnosis in acute HIV-1 infection, we demonstrated that individual clinical parameters may be predictive of time to viral rebound post-ATI years later.

Results

Participants had homogeneous demographic and clinical characteristics

Sixty-seven individuals were diagnosed in acute HIV-1 infection (Fiebig I-IV) and were enrolled in the RV254 study and initiated ART a median of two days after diagnosis (range 0–8 days). After a median of 3.1 years of viral suppression (range 1.78–6.57), they were enrolled in one of four ATI studies (Table S1). To investigate whether data from the four ATI studies could be combined, participants were divided into four groups constituted by participants who received 1) no intervention (all RV411 participants and placebo recipients in each study, n=27), 2) VHM (n=9), 3) Ad26/MVA vaccination (n=18) and 4) VRC01 infusion (n=13) (Table S1). Participants showed similar characteristics with the majority being Thai (95.5%), male (97%), MSM (94%), with a median age of 27 (IQR 25–32) years and infected with CRF01_AE viruses (73.2%) (Table S2). When participants were stratified by administered intervention, there were no significant differences between groups, indicating that the groups were homogeneous in population demographics (p≥0.250) (Table S2). Antiretroviral drug regimens varied across individuals and over time in a given individual, yet, until four weeks prior to ATI, there was no significant difference when the treatments were compared by intervention group (p=0.058) (Figure S1). In the four weeks preceding the ATI, ART regimens differed significantly (p<0.001) with a majority of participants switching to TDF/3TC/DRV/RTV except for the participants who received VHM intervention.

Finally, we evaluated eleven clinical variables. Seven variables were collected at the beginning of infection, either prior to ART initiation (viral load at diagnosis, Fiebig stage, CD4 count, CD8 count and CD4/CD8 ratio) or following ART initiation (total viral load AUC and time to viral suppression) (Figure 1, Figure S2). These initial data points were collected at the RV254 study enrolment visit. Four variables were collected immediately prior to the ATI: ART duration, CD4 count, CD8 count and CD4/CD8 ratio. We note that the total viral load AUC was calculated to reflect the AUC imputed to account for the ‘missed’ viremia prior to diagnosis and the AUC calculated based on viral load measurements post-diagnosis (Figure S3, Figure 1). The eleven clinical variables measured at the beginning of infection or pre-ATI did not differ across the four study groups (p≥0.107) except for Fiebig stages (p=0.001), as the RV411 study exclusively enrolled participants who initiated ART during Feibig stage I (Table 1). Participants who did not receive an intervention had been diagnosed earlier in acute infection, with nine participants diagnosed in Fiebig I (33%) compared to zero or one Fiebig I participant in the intervention groups (none for VHM or Ad26/MVA vaccination and one in the VRC01 infusion arm).

Figure 1. Viral Load Area Under the Curve (VL AUC) at HIV-1 diagnosis.

The upper panel (A) is a schematic description of the VL AUC calculations as a function of the Fiebig stage and viral load at HIV-1 diagnosis. Viral load kinetics are figured in blue with a representative curve. Diagnosis is marked with a dotted red line. VL measures obtained following diagnosis are shown in black and serve to calculate the real AUC. The imputed AUC corresponds to the period prior to diagnosis (without VL measurements) and is shaded in grey. The lower panel represents the imputed VL AUC (B), actual VL AUC (C) and total VL AUC (D) calculated for the 67 participants in our study. The total VL AUC used for analysis is the sum of the imputed and actual AUC.

Table 1.

Clinical characteristics of HIV-1 acutely infected participants from RV254 ATI studies, stratified by administered intervention.

Study		No intervention	VHM	Ad26/MVA	VRC01	P
Total N		27	9	18	13
Fiebig Stage	FI	9 (33.3)			1 (7.7)	0.001
	FII	7 (25.9)		6 (33.3)	7 (53.8)
	FIII	11 (40.7)	7 (77.8)	7 (38.9)	5 (38.5)
	FIV		2 (22.2)	5 (27.8)
ART duration (years)	Median (IQR)	2.8 (2.6 – 3.7)	4.7 (3.1 – 5.5)	3.0 (2.8 – 3.5)	3.1 (2.4 to 4.0)	0.602
VL at diagnosis (log10 copies/mL)	Median (IQR)	5.6 (4.5 – 6.4)	5.6 (4.9 – 6.7)	5.9 (5.1 – 6.7)	5.8 (5.4 – 6.5)	0.725
Time to viral suppression (weeks)	Median (IQR)	8 (4 – 12)	8 (8 – 16)	12 (9 – 19)	12 (8 – 12)	0.107
Total viral load AUC (log10 copies/mL/day)	Median (IQR)	6.5 (5.4 – 7.2)	6.8 (6.0 – 7.7)	7.0 (6.3 to 7.5)	6.8 (6.3 to 7.4)	0.559
Pre-ART CD4 (cells/mm^3)	Median (IQR)	303 (227 – 468)	412 (308 – 490)	354 (284 – 536)	388 (265 – 515)	0.626
Pre-ART CD8 (cells/mm^3)	Median (IQR)	447 (301 – 694)	638 (500 – 701)	601 (417 – 795)	820 (343 – 1127)	0.211
Pre-ART CD4/CD8	Median (IQR)	0.8 (0.6 – 0.9)	0.5 (0.4 – 0.9)	0.5 (0.4 – 0.9)	0.6 (0.5 – 1.0)	0.447
Pre-ATI CD4 (cells/mm^3)	Median (IQR)	562 (485 – 716)	643 (555 – 752)	632 (548 – 738)	769 (622 – 864)	0.277
Pre-ATI CD8 (cells/mm^3)	Median (IQR)	610 (478 – 739)	564 (511 – 591)	465 (393 – 629)	670 (530 – 886)	0.121
Pre-ATI CD4/CD8	Median (IQR)	1.1 (0.9 – 1.3)	1.1 (1.0 – 1.2)	1.1 (1.0 – 1.5)	1.1 (0.8 – 1.2)	0.544
Days to HIV-1 RNA ≥ 20 copies/mL	Median (IQR)	19 (14 – 26)	21 (20 – 28)	23 (17 – 28)	29 (19 – 34)	0.413
Days to HIV-1 RNA ≥ 1000 copies/mL	Median (IQR)	22 (17 – 33)	28 (28 – 35)	28 (21 – 33)	33 (27 – 36)	0.260

Participants without post-ATI interventions were pooled for correlates analysis

The ATI studies were evaluated with two endpoints corresponding to the number of days until a confirmed HIV-1 RNA measurement at ≥20 copies/mL (Figure S4A) or ≥1000 copies/mL (Figure 2A). Participants who received no intervention (other than ART) could be pooled because the population was homogeneous in demographic and clinical variables; time to viral rebound (HIV-1 RNA ≥1000 copies/mL) ranged between 13 and 112 days. Time to viral rebound distributions were compared for the three interventions groups. The median and variance of the time to rebound were not statistically different between the four treatment groups (Kruskal-Wallis p=0.34, Brown-Forsythe p=0.28, respectively) (Figure 2B; similar results for time to rebound at HIV-1 RNA ≥20 copies/mL, Figure S4B).

Figure 2. Participants without intervention, VHM and Ad26/MVA interventions were pooled for correlates analysis with time to rebound measured at HIV-1 RNA ≥1000 copies/mL.

(A) The length of time each participant spent on ART before ART interruption (ATI) is shown on the left and the time to viral rebound on the right. ART duration is in years while time to viral rebound is in days. (B) Time to viral load rebound measured at HIV-1 RNA ≥1000 copies/mL did not differ across interventions with respect to the median (Kruskal-Wallis test) and the variance (Brown-Forsythe test) of the outcome (N = 26, N = 9, N = 18, and N = 13 for the no intervention, VHM, Ad26/MVA and VRC01 groups, respectively). (C) Effect of each intervention against the pooled ‘no intervention’ group tested by Cox regression multivariate model adjusted by Fiebig stage as Fiebig stages differed across interventions (N = 26, N = 9, N = 18, and N = 13 for the no intervention, VHM, Ad26/MVA and VRC01 groups, respectively). Intervention effects were tested using all the participants per study and an adjustment was added to account for potential confounding by Fiebig stages which differed significantly across the studies.

In addition, to test for statistical heterogeneity across studies in all participants, we estimated study-specific intervention effects against the no intervention group while adjusting for the study level covariates which differed across studies. A Cox regression model was used to investigate the effect of ATI study-specific interventions with respect to the no intervention group using all the participants per study. The Cox regression model was adjusted by Fiebig stage because Fiebig stages varied significantly across intervention groups (Table 1). While later Fiebig stages showed expected trends towards shorter times to viral rebound when compared to Fiebig I, Fiebig stages did not significantly influence time to viral rebound when considered with interventions (Figure 2C). The adjusted intervention effect for each ATI study against the pooled placebo group showed a trend towards delayed time to viral rebound for participants with an intervention but this was not significant for participants who received VHM (p=0.424) or Ad26/MVA vaccination (p=0.598) (Figure 2C). However, participants who received VRC01 showed a significant delay in time to viral rebound compared to participants with no intervention (p=0.027) (Figure 2C; for time to rebound at HIV-1 RNA ≥20 copies/mL, Figure S4C). Thus, we could not statistically justify pooling participants who received VRC01 with participants who received VHM or Ad26/MVA. In addition, we note that VRC01 infusions were initiated at the time of ART interruption while VHM or Ad26/MVA vaccination occurred prior to the ATI. One post-treatment controller identified in the placebo arm of the Ad26/MVA vaccination study was also excluded from the analysis because the participant presented an HLA allele associated with viral control (HLA B*57:01:01)¹¹. There was no other individual with HLA B*57 alleles in the cohort. This resulted in a total of 53 participants (26 without and 27 with an intervention) pooled for analysis of factors associated with time to viral rebound.

Time to rebound did not differ across Fiebig stages

Correlations between the eleven clinical variables measured at the beginning of infection and pre-ATI were analyzed in the pooled data. Variables measured in acute infection prior and post-ART initiation were significantly correlated to each other (Figure 3A). Viral load at HIV-1 diagnosis was positively correlated to total viral load AUC (rho=0.99), and time to viral suppression (rho=0.52) and was negatively correlated with CD4 (rho=−0.41) (Figure 3A). Four variables were significantly different across Fiebig stages: viral load at diagnosis, viral load AUC, pre-ART CD8 and pre-ART CD4/CD8 ratio (Figure 3B–3E). Comparison of viral load at diagnosis across Fiebig stages showed a significant difference (Kruskal-Wallis test, p=0.0005); however, only Fiebig I showed different viral loads than the other Fiebig stages (Mann–Whitney U test), p<0.05) (Figure 3B). Similarly, viral load AUC differed when Fiebig I was compared to the other Fiebig stages (Figure 3C). Pre-ART CD8 showed that Fiebig I differed from Fiebig II and Fiebig IV stages (Figure 3D), while pre-ART CD4/CD8 showed that Fiebig I differed from Fiebig III and Fiebig IV stages (Figure 3E), Comparisons across Fiebig stages II, III and IV showed that viral loads did not differ (Kruskal-Wallis test, p>0.05) with a wide range of viral loads and overlapping distributions. Next, we investigated whether time to viral rebound varied across Fiebig stages using a Cox regression, visualized with Kaplan-Meier curves. There was no significant difference in time to rebound. across Fiebig stages with: p=0.64 and p=0.72 for time to viral rebound corresponding to an HIV-1 RNA measurement of ≥1000 copies/mL and ≥20 copies/mL, respectively (Figures 4A and 4B). Similarly, comparisons of time to viral rebound across all Fiebig stages or between Fiebig stage I and other stages (Kruskal-Wallis test and Mann Whitney U test, respectively) showed no significant difference: the median time to rebound was 28, 24, 28 and 30 days when considering rebound as an HIV-1 RNA measurement of ≥1000 copies/mL, respectively, for Fiebig stages I through IV (p=0.49) (Figure 4C). Results were similar when time to rebound to a measurement of HIV-1 RNA ≥20 copies/mL was considered with a median of 23, 20, 21 and 28 days for each Fiebig stage from I to IV (p=0.54) (Figure 4D).

Figure 3. Multiple correlations among baseline variables at HIV-1 diagnosis.

(A) Spearman correlations between continuous baseline variables are shown for pairwise variables. Significant correlations are shown in red (positive) or blue (negative) (N = 53). (B - E) Comparison of baseline variables by Fiebig stage (N = 9, N = 12, N = 23 and N = 7 for FI, FII, FIII and FIV, respectively). Tested variables were viral load (VL) at HIV-1 diagnosis, VL AUC, pre-ART CD8 cell counts and pre-ART CD4/CD8 T cell ratio. The total VL AUC used for analysis is the sum of the imputed and actual AUC. Kruskal-Wallis test was used to compare across all Fiebig stages and Mann Whitney U test was used to compare each Fiebig stage to Fiebig stage I (marked with asterisks).

Figure 4. Time to rebound did not differ across Fiebig stages.

Kaplan-Meier survival curves for Fiebig stage compared the time to viral rebound of Fiebig stages I to IV for (A) HIV-1 RNA ≥1000 copies/mL and (B) HIV-1 RNA ≥20 copies/mL. Kruskal-Wallis test was used to compare across all Fiebig stages and Mann Whitney U test (represented with asterisks) used to compare each Fiebig stage to Fiebig stage I for time to rebound at (C) HIV-1 RNA ≥1000 copies/mL and (D) HIV-1 RNA ≥20 copies/mL. (N = 9, N = 12, N = 23 and N = 7 for FI, FII, FIII and FIV, respectively).

Viral load at HIV-1 diagnosis was associated with time to viral rebound

Further analysis with survival methods evaluated variables measured at the beginning of infection (pre- and post-ART initiation) and pre-ATI. Cox regression univariate models showed that several variables measured in the first weeks of HIV-1 infection were associated with time to viral rebound while none of the variables measured prior to ATI were associated with rebound kinetics (Figures 5A and 5B and Figures S5A and S5B). Three correlated viral load specific variables were linked to time to rebound; viral load at diagnosis (p=0.022), time to viral suppression (p<0.001) and viral load AUC (p=0.026). The optimal cut off in the viral load distribution (selected using the maximally selected rank statistics method) was 4.7 log10 copies/mL and Kaplan-Meier survival curves were used to display that participants with viral loads higher than 4.7 log10 copies/mL had a shorter time to rebound than those with lower viral loads (p=0.011) (Figure 5C). We replicated this finding when testing the commonly used threshold of 5 log10 copies/mL^14–17 which yielded a p-value = 0.038. Similar results were observed for the two other viral load-specific variables: viral load AUC (p=0.011) and time to viral load suppression (p=0.008) (Figure 5E and 5F). Correlations with time to rebound at HIV-1 RNA ≥20 copies/mL corroborated these results: Cox regression univariate models showed significant correlations with viral load at diagnosis (p=0.035), time to viral suppression (p=0.011) and viral load AUC (p=0.033) (Figure S5A). Kaplan Meier survival curves were used to display that participants with higher viral load at diagnosis (Figure S5C), higher viral load AUC (Figure S5E) and higher time to viral load suppression (Figure S5F) had a shorter time to viral rebound measured at RNA ≥20 copies/mL (p=0.009, p=0.009 and p=0.037, respectively). In accord with the fact that pre-ART CD4 counts associated negatively with viral load at diagnosis (rho = −0.41) (Figure 3A), participants who had pre-ART CD4 counts greater than 326 cells/mm^3 showed a longer time to rebound than those with lower CD4 counts (p=0.024 for HIV-1 RNA ≥1000 copies/mL; p=0.063 for HIV-1 RNA ≥20 copies/mL) (Figure 5D and Figure S5D). We reran analyses while including the participant with HLA allele B*57. Similar results were obtained for time to viral suppression. However, significance was lost for the associations with viral load at diagnosis (p=0.051 for HIV-1 RNA ≥1000 copies/mL (Figure S6) and p=0.075 for HIV-1 RNA ≥20 copies/mL (Figure S7)) and viral load AUC (p=0.057 for HIV-1 RNA ≥20 copies/mL).

Figure 5. Viral load variables at diagnosis impacted time to viral rebound (HIV-1 RNA ≥1000 copies/mL).

Cox regression univariate models show the impact of clinical variables measured at HIV-1 diagnosis (A) or prior to the ATI (B) on the time to viral rebound (N = 53). (C-F) Kaplan-Meier survival curves for the baseline variables associated with time to rebound selected through Cox regression model for (C) viral load at diagnosis, (D) pre-ART CD4, (E) total viral load AUC and (F) weeks to viral suppression. Categories were divided using maximally selected rank statistics (N = 26, N = 9 and N = 18 for the no intervention, VHM and Ad26/MVA groups, respectively).

Sensitivity analyses were conducted by removing each treatment group to confirm the robustness of the association of viral load variables with time to viral rebound (Figure 6 and Figure S8). The significant associations between pre-ART variables and time to viral rebound were usually maintained when the VHM, Ad26/MVA or both treatment groups were removed. However, the association with total viral load AUC became insignificant when the Ad26/MVA group was removed (p=0.067) and the association with pre-ART CD4 counts became insignificant when the VHM group (p=0.137) or both groups (p=0.067) were removed. While the no intervention group showed higher point estimates of the hazard ratios, the 95% confidence intervals (CI) were wider, indicating the limited precision of the estimates due to smaller sample size (Table S3). As such, analyses using all the pooled participants showed hazard ratios with slightly narrower 95% confidence intervals (CI) than analyses where the treatment groups (VHM and/or Ad26/MVA) were removed.

Figure 6. Sensitivity analysis of the Cox regression results to the treatment interventions.

Results are shown for time to rebound at HIV-1 RNA ≥1000 copies/mL. The sensitivity of the correlates is evaluated by removing each treatment at a time or simultaneously to test if the model remains significant.

Discussion

By combining data from four ATI studies, we evaluated the impact of eleven clinical variables on the time to viral rebound in 53 participants who underwent ATI. We showed that viral load variables measured in the first weeks of HIV-1 infection (from HIV-1 diagnosis until viral suppression through ART) were associated with time to rebound, with higher viral loads in acute infection and longer time to viral suppression post ART initiation linked to shorter time to viral rebound post-ATI.

Two key advantages of this study are that the four ATI studies included participants from the same parent cohort and variables were measured not only pre-ATI but also pre-ART initiation at HIV-1 diagnosis in acute infection. Like most ATI studies, the four ATI studies included a limited number of participants and often more parameters were evaluated than there were participants in the study. Hence, pooling participant data from the same parent cohort results in less heterogeneity than when analyzing data from independent cohorts. For example, the RV254 cohort enrolls predominantly Thai MSM which limits both the HLA genotype and virus diversity. In addition, variables were measured in the different ATI studies using the same assays and laboratories. We first statiscally tested that the ATI studies could be combined: pooling participants with the VHM and Ad26/MVA interventions together with those who received no intervention was possible because the time to rebound endpoint did not significantly differ between these participants. Combining these studies increased the sample size to a total of 53 participants which improved the correlates results, improved precision and consistency of intervention effects, and tightened confidence intervals of hazard ratios.

The second advantage of this study design was that variables were measured not only pre-ATI but also in the first weeks of infection including immediately prior and following ART initiation, about three years prior to the ATI. Our work leveraged the design of the RV254 cohort that enrolls PLWH in the earliest stages of HIV-1 infection and who initiate ART immediately. Participants were diagnosed in Fiebig stages I to IV, with multiple measurements in the first weeks of infection just prior to and following ART initiation. While different studies enrolled PLWH in chronic infection and showed that there was a correlation between viremia prior to ART and viremia post ATI^18–21, few studies enrolled participants who initiated treatment during acute or early infection and these studies did not typically have precise data collected at diagnosis in acute infection. Different studies showed associations between viremia pre-ART and viremia post-ATI but there was not necessarily a relationship with time to rebound^20,21. Volberding and colleagues showed that participants in acute or recent infection with viral loads below 100,000 copies/mL were more likely (22/46, 48%) to maintain low viral loads (less than 5,000 copies/mL) after 24 weeks than those with viral lads above 100,000 copies/mL (7/27, 26%)¹⁵. A similar correlation between low viral loads in primary infection and control post treatment interruption was reported by Goujard and colleagues²². It has also been shown that individuals who initiated treatment early in infection are more likely to control viremia following treatment cessation². In a meta-analysis of data collected from 728 participants who underwent treatment interruption in 14 studies, Namazi and colleagues showed that participants who initiated ART in acute or early infection were more likely to become post-treatment controllers than participants who initiated ART in chronic infection (13% versus 4%, p < 0.001)³. Similar findings were reported in a prior meta-analysis (n=235 participants from 6 ATI studies): 9% of those who initiated ART during acute or early infection controlled viremia (viral load < 200 copies/mL at week 12) compared to 3% of those who initiated ART during chronic infection (P=0.01)²³. The higher likelihood of delayed rebound in individuals who initiated treatment early could be associated with their smaller viral reservoirs when compared to patients in chronic infection²⁴, although there is not necessarily an association between reservoir size and post-treatment control^22,23.

Importantly, by pooling a cohort of individuals followed longitudinally since ART initiation in acute infection, we can start to decipher features that may be driving viremia upon treatment cessation. While viral rebound likely has multifactorial causes, it is conceivable that in the first week of infection, there are fewer interactions between parameters making it easier to identify a key variable than it would be after weeks or months of infection when viral and host features have already co-evolved. It is significant that time to viral rebound was not associated with clinical variables measured pre-ATI but with those measured in acute HIV-1 infection including prior to ART initiation - variables that are rarely measured but may be key for our understanding of the rebound susceptibility of the reservoir. We also modeled the viral load AUC by imputing the pre-diagnosis viral load peak for the participants who were diagnosed at a time likely to correspond to their downslope. Although this imputation ensures that individuals diagnosed at the viral load nadir do not have the same AUC as individuals diagnosed in the upslope in Fiebig I, our cohort size was likely too small to characterize the effect of this modeling. In our pooled analysis, we showed that the variability across interventions was within what would be expected given inter-individual variability. Our results demonstrate the importance of viral load in the first weeks of infection as a predictor of time to rebound in a cohort who initiated ART upon diagnosis in acute infection. Three correlated viral load-specific measures showed significant associations: viral load at diagnosis, the time to viral suppression post ART initiation and the viral load AUC which we imputed to account for the ‘missed’ viremia prior to diagnosis. The threshold that was found to statistically divide the distribution of viral loads at diagnosis with respect to time to rebound was 4.7 log10 copies/mL - whether this threshold can serve as a biomarker will need to be tested. Interestingly, in a non-human primate study, peak viral load prior to initiation of ART (for a year) was also shown to be the best predictor of time to viral rebound²⁵.

In conclusion, our study emphasizes that clinical parameters during acute infection have long term consequences on rebound kinetics for individuals who initiated ART early and underwent treatment interruption on average three years later. It has already been reported in a therapeutic vaccine study that pre-ART variables can predict study endpoints²⁶. Although obtaining pre-ART data is challenging, characterizing this under-appreciated aspect of reservoir establishment may facilitate the elucidation of better predictors of viral rebound.

Limitations of the study

There are certain limitations to our study. Almost all participants were Thai (except three), males (except two) and MSM (except four), reflecting the configuration of the parent cohort RV254. While our study identified the impact of viremia in the first weeks of infection (prior to or around ART initiation) on time to viral rebound post ATI, specific mechanisms behind delayed rebound were not evaluated and further studies are needed to measure the potential role of reservoir characteristics and host immune responses. There was some heterogeneity in the cohort in terms of ART duration prior to ATI (from about two to six years) and drug regimens, we did not identify any impact of these factors on time to rebound; yet, it is plausible that the ART duration or regimen prior to ATI associate with other proviral reservoir characteristics and an integrated analysis may further explain why our study revealed the role of acute and early infection variables but not pre-ATI factors. The ATI studies in our pooled analysis were designed to ensure that participants resumed ART rapidly upon viral rebound – the limited duration of uncontrolled viremia was crucial as many participants were still HIV-1 seronegative at the time of the ATI (early initiation of ART in acute infection had prevented seroconversion). Since participants restarted ART after two viral load measurements at 1,000 HIV-1 RNA copies/mL, our study was geared toward understanding of short-term control of viremia rather than potential long-term virologic control. As such, we excluded from our analysis the placebo individual who suppressed viremia for a prolonged period likely as a consequence of HLA-B*57-mediated control. Furthermore, as most diagnoses of HIV-1 infection occur during the chronic stage, it will be important to determine whether similar correlations are obtained from ATI studies in this population. The ultimate goal would be to determine a quantifiable laboratory measurement at the time of treatment interruption that correlates with time to viral rebound. Research is ongoing to determine whether a single assay or combination of assays would reliably serve this purpose, as this type of signature would permit the evaluation of research interventions targeting HIV-1 cure without the need for additional treatment interruption studies, thus reducing potential for HIV-1 disease progression or transmission during the ATI period²⁷.

STAR Methods

RESOURCE AVAILABILITY

Lead contact.

Further information and requests for resources should be directed to and will be fulfilled by the lead contact, Morgane Rolland (mrolland@hivresearch.org).

Materials availability.

This study did not generate new unique reagents.

Data code and availability.

• Data reported in this paper will be shared by the lead contact upon request.

• Code is available at https://www.hivresearch.org/publication-supplements.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODELS AND SUBJECT DETAILS SECTION

The RV254 Acute HIV-1 Infection study (clinicaltrials.gov NCT00796146) is a prospective cohort study designed to follow adults with Fiebig I to V acute HIV-1 infection in Bangkok, Thailand²⁸. Individuals seeking voluntary HIV-1 counseling and testing at the Thai Red Cross Anonymous Clinic, Institute of HIV Research and innovation (IHRI) and community-based organizations in Bangkok, Thailand are screened for acute HIV-1 infection using sequential immunoassay and pooled nucleic acid testing^7,29. ART is offered to all participants as part of a separate protocol (NCT00796263)^8,30. Participants are followed for 2 to 15 years and undergo clinical evaluations and blood collection for CD4, HIV-1 RNA and safety monitoring as well as archiving of peripheral blood mononuclear cells and plasma. As of June 17, 2022, there were 669 participants under active follow up.

This work included participants who had been enrolled in four RV254 substudies: RV411 (Viral Suppression After Analytic Treatment Interruption in Thai Patients Who Initiated Highly Active Antiretroviral Therapy During Acute HIV Infection, NCT02614950)⁹; RV409 (A Randomized Study to Compare the Efficacy of Vorinostat/Hydroxychloroquine/Maraviroc (VHM) in Controlling HIV After Treatment Interruption in Subjects Who Initiated ART During Acute HIV Infection, NCT02475915)¹⁰; RV405 (A Combined Phase 1/2a, Exploratory Study of a Therapeutic Vaccine Using an Adenovirus Type 26 Vector Prime and Modified Vaccinia Ankara Boost Combination With Mosaic Inserts in HIV-1 Infected Adults Who Initiated Antiretroviral Treatment During Acute HIV Infection, NCT02919306)¹¹; and RV397 (Viral Suppression After Analytic Treatment Interruption in Thai Patients Who Initiated Highly Active Antiretroviral Therapy During Acute HIV Infection, NCT02614950)¹².

Participants information on sex at birth, age and race was self-reported. Information on gender and socioeconomic status was not reported for this study.

Ethics statement

All participants signed written informed consent and participated in protocols approved by Thai and US (Walter Reed Army Institute of Research) Institute Review Boards. The investigators have adhered to the policies for protection of human subjects as prescribed in AR 70–25.

METHOD DETAILS

Quantifying heterogeneity across studies

Three types of variability were systematically assessed; i) the clinical heterogeneity or variability in the study participants, ii) the methodological heterogeneity or variability in the study designs and types of intervention and iii) the statistical heterogeneity or variability in the intervention effects, which is a consequence of both clinical and methodological heterogeneity (Figure S7). Clinical variability was evaluated by comparing the distributions of demographic and baseline clinical variables across studies using Kruskal-Wallis (for continuous explanatory variables) and chi-squared tests (for categorical explanatory variables). Significant clinical variables (p<0.05) may be potentially confounding to study-specific interventions. Methodological heterogeneity was evaluated by objectively comparing study designs such as ATI-study intervention types, timing of the intervention and types of outcomes. Statistical heterogeneity (i.e, study outcome endpoints and intervention effects) was evaluated using Kruskal-Wallis test to compare study medians and Brown-Forsythe test to compare study variances. A Cox regression model assessed the intervention effects of each ATI study adjusted by baseline variables that varied significantly across the intervention groups. Any significance in statistical heterogeneity (p<0.05) would suggest that a study-specific intervention led to substantial effect in the study population. However, it is statistically justifiable to pooled data from studies that are sufficiently homogeneous in terms of participants, interventions and outcomes for analysis of covariates of time to viral rebound.

Viral load Area Under the Curve (AUC) calculation

For each participant, the acute infection viral load AUC was calculated as the sum of the imputed AUC, which accounts for the viremia that was missed before diagnosis, and the post-diagnosis AUC, which is based on viral load measurements between diagnosis and viral load suppression. This was done to address the fact that an individual diagnosed in Fiebig I could have similar viral loads as an individual diagnosed in Fiebig IV during the viral load downslope. The imputed pre-diagnosis AUC was modeled based on data from the RV217 cohort: we used 1,263 viral load measurements obtained from 96 participants (from East Africa (n=60) and from Thailand (n=36)) with more than three viral load measured after acute HIV-1 diagnosis³¹. Participants with consecutive Fiebig staging information (I-II to III, III to IV and IV to V-VI) were used to estimate that Fiebig stage I-II ended at a median of 2.5 days before peak viremia while Fiebig stages III and IV ended at a median of 1.5 and 4 days after peak viremia, respectively. Individual viral load curves were obtained by fitting a nonlinear viral dynamics model³² (Figure S3A). The viral load (VL) AUC for each modeled VL from the 96 RV217 participants was calculated using trapezoidal rule as a function of the corresponding viral load and the number of days from peak viremia (Figure S3B):

$$\mathit{\text{VLau}}{c}_{ij}^{RV217}=g\left(V{L}_{ij}^{RV217},t_{ij}^{RV217}\right)\text{where i is individual, j is days from peak}$$

The imputed viral load AUC for each RV254 participant for a given viral load and Fiebig stage was the mean of the viral load AUC from all individuals given the corresponding viral load and Fiebig stage fitted by a local polynomial regression per Fiebig stage (Figure S3C–S3D):

$$E\left[\mathit{\text{VLau}}{c}_k^{\mathit{\text{imputed}}}|VL,\text{Fiebig}\right]\left(x,f\right)=\mathit{\text{mean}}\left(\left\{\mathit{\text{VLau}}{c}_i^{RV217}:VL=x,\text{Fiebig}=f\right\}\right)\text{where k is individual}$$

The post-diagnosis viral load AUC was calculated by trapezoidal rule using all viral load measurements from diagnosis until viral suppression (defined as <50 copies/mL of HIV-1 RNA). The post-diagnosis viral load AUC was added to the imputed pre-diagnosis AUC to obtain the total viral load AUC used for analysis (Figure 1).

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analysis

A flowchart describes the statistical analysis plan (Figure S9). Virological variables were compared using the Kruskal-Wallis test for continuous variables (wilcox_test) and chi-squared test for categorical variables (chisq_test) in Tables 1 and S2 using the ‘finalfit’ R package. Kruskal-Wallis test was used for comparisons across multiple groups by Fiebig stage (Figures 3 and 4) or by intervention (Figures 2, S4) and Mann Whitney U test (wilcox_test) was used for pairwise comparisons (Figure 3) using the R ‘rstatix’ package. Brown-Forsythe test was used to compare study variances in Figures 2 and S4 using the function levene_test from the R ‘rstatix’ package. Associations between virological parameters (Figure 3) were determined using Spearman correlations, through the R package ‘correlation’. Cox regression univariate models were used to determine intervention effects (Figures 2 and S4) and association of baseline variables with time to rebound (Figures 5–6, S5–S8) using the coxph function in the ‘finalfil’ R package. The function cox.zph under the ‘survival’ package was used to check for the proportionality assumption. All p-values were > 0.05 showing that none of the covariates were time dependent and did not violate the proportionality assumption. Kaplan-Meier survival curves (Figures 4–5, S5–S7) were plotted and compared by the log-rank test using the ggsurvplot function in the ‘survminer’ R package. Maximally selected rank statistics were used to robustly estimate the optimal cutoff that allows to divide the continuous variables into groups and visualize differences in the Kaplan-Meier estimates of survival curves. This test selects the cutoff that provides the greatest discrimination power between groups to improve visualization of the curves. The threshold was selected using the surv_cutpoint function and curves were visualized through the R package ‘survminer’. Statistical significance was defined as a P-value <0.05. All analyses were carried out in R (version 4.0.0; R Foundation for Statistical Computing, Vienna, Austria) using RStudio (version 1.3.959; Integrated Development for R. RStudio, PBC, Boston, MA).

Highlights.

• 53 participants, HIV-suppressed since acute infection, underwent treatment interruption

• Time to rebound did not differ across Fiebig stages

• Higher viral load at HIV-1 diagnosis was associated with faster viral rebound

Context and Significance.

Antiretroviral therapy (ATI) decreases the quantity of circulating virus, or viral load, in people living with HIV-1. Analytic treatment interruption (ATI) is an approach to study the mechanisms that regulate viral load in these individuals by letting it rebound in a controlled manner. While many ATI studies involved participants that had already started ART, here the authors had the unique opportunity to access clinical data from the time of diagnosis of acute HIV-1 infection, before ART initiation, in 53 individuals who were virally suppressed for a median of three years before treatment interruption. They analyzed multiple factors and observed that viral loads measured during acute HIV-1 infection, rather than variables measured prior to treatment interruption, associates with time to rebound.

Inclusion and diversity

One or more of the authors of this paper self-identifies as an underrepresented ethnic minority in science. One or more of the authors of this paper self-identifies as a member of the LGBTQ+ community. The author list of this paper includes contributors from the location where the research was conducted who participated in the data collection, design, analysis, and/or interpretation of the work.