Temporary increase in circulating replication-competent latent HIV-infected resting CD4+ T cells after switch to an integrase inhibitor based antiretroviral regimen

Summary

Background

The principal barrier to an HIV cure is the presence of the latent viral reservoir (LVR), which has been understudied in African populations. From 2018 to 2019, Uganda instituted a nationwide rollout of ART consisting of Dolutegravir (DTG) with two NRTI, which replaced the previous regimen of one NNRTI and the same two NRTI.

Methods

Changes in the inducible replication-competent LVR (RC-LVR) of ART-suppressed Ugandans with HIV (n = 88) from 2015 to 2020 were examined using the quantitative viral outgrowth assay. Outgrowth viruses were examined for viral evolution. Changes in the RC-LVR were analyzed using three versions of a Bayesian model that estimated the decay rate over time as a single, linear rate (model A), or allowing for a change at time of DTG initiation (model B&C).

Findings

Model A estimated the slope of RC-LVR change as a non-significant positive increase, which was due to a temporary spike in the RC-LVR that occurred 0–12 months post-DTG initiation (p < 0.005). This was confirmed with models B and C; for instance, model B estimated a significant decay pre-DTG initiation with a half-life of 6.9 years, and an ∼1.7-fold increase in the size of the RC-LVR post-DTG initiation. There was no evidence of viral failure or consistent evolution in the cohort.

Interpretation

These data suggest that the change from NNRTI- to DTG-based ART is associated with a significant temporary increase in the circulating RC-LVR.

Funding

Supported by the NIH (grant 1-UM1AI164565); Gilead HIV Cure Grants Program (90072171); Canadian Institutes of Health Research (PJT-155990); and Ontario Genomics-Canadian Statistical Sciences Institute.

Research in context.

Evidence before this study

HIV is a largely incurable infection despite the use of highly successful antiretroviral therapy (ART) due to the presence of a population of long-living resting CD4+ T cells, which can harbor a complete copy of the virus integrated into the host cell’s DNA. Previous work in North American populations has shown that this latent reservoir decreases very slowly for years after ART initiation, and may stabilize or even increase after seven to ten years of therapy. Changes in the latent reservoir have been poorly studied in African populations, which harbor the vast majority of the global HIV burden. We examined changes in the levels of these cells, referred to as the latent viral reservoir, in a group of ART-suppressed Ugandans living with HIV. During this examination, Uganda authorities switched the backbone drug used in ART regimens to a different class of drug, integrase inhibitors, that blocks the ability of the virus to integrate into the cell’s DNA.

Added value of this study

We found that for approximately a year after the switch to the new integrase inhibitors, there was a temporary spike in the size of the latent viral reservoir despite the new drug continuing to completely suppress viral replication.

Implications of all the available evidence

This study provides critical data on the make-up and maintenance of the latent reservoir in Africans, and suggests that for approximately a year after switching to integrase inhibitors a portion of individuals may experience significant changes to their HIV reservoir as well.

Introduction

HIV persists in individuals treated with fully suppressive antiretroviral therapy (ART) due in large part to the presence of a population of latently infected immune cells that are collectively referred to as the latent viral reservoir (LVR).¹ The LVR is primarily made up of resting memory CD4+ (rCD4) T cells, which can be found in the circulating blood and tissues throughout the body.² While ART effectively blocks new cells from becoming infected in people with HIV, it has no direct effect on the stably integrated proviruses that comprise the LVR. The LVR is maintained in the presence of ART through clonal expansion of latently infected cells due to a combination of homeostatic proliferation and antigenic stimulation, which together with the relatively long-lived nature of rCD4 T cells, creates an estimated half-life of the entire LVR of approximately 44 months.^3,4 In addition, long-term analysis of the LVR in individuals treated for 10+ years from the United States suggest that the LVR may stabilize later in treatment and remain constant or possibly increase.⁵

However, the vast majority of work examining the LVR has been focused on high-income countries, even though more than two-thirds of people with HIV are African. Earlier work by our group demonstrated that the replication-competent (RC) LVR, as measured by the quantitative viral outgrowth assay (QVOA), was significantly lower in Ugandans than in North Americans.⁶ One possible explanation for this difference could be that there might be a more rapid decay of the RC-LVR in this Ugandan population. To examine this hypothesis, this previous study was extended to include longitudinal large blood draws (2015–2020), which were subsequently used for determination of changes in the size of the RC-LVR over time.

During this longitudinal study, Uganda medical authorities began a nationwide roll-out of integrase strand transfer inhibitor (INSTI) based ART in 2018, consisting of Dolutegravir (DTG), Tenofovir (TDF), and Lamivudine (3TC). This change provided an opportunity to examine for any effects the switch from a non-nucleoside reverse transcriptase inhibitor (NNRTI) based regimen, the primary ART backbone prior to the switch, to one containing an INSTI, may have on the size of the LVR. To our knowledge, no study to date has found significant differences in LVR maintenance by ART regimen.

Methods

Study population

The details of the population examined for this analysis have been discussed in detail previously.^6,7 Briefly, people with HIV from the Rakai and Kyotera Districts of Uganda who were virally suppressed (viral load < 40 copies/mL) for at least 18 months were recruited for annual large blood draws (∼180 mL) to examine the size and make-up of their LVR (Table 1). An initial group of participants were enrolled in 2015 (n = 70), and were followed annually in 2016, 2017, 2019, and 2020.⁶ An additional group of participants with estimated dates of seroconversion (n = 20) were later enrolled in the study in 2016 and followed annually in 2017, 2018, and 2019.⁷ For each study time point, blood was collected and separated into peripheral blood mononuclear cells (PBMC) and plasma, and stored. All participants underwent a clinical examination, viral load, and CD4+ cell count at each study visit. One participant was excluded from the analysis due to a loss of viral suppression during the study due to being incarcerated.

Table 1.

Summary of study participants.

	Full cohort (n = 88)	Individuals with multiple timepoints (n = 71)
Switched to DTG (n, %)	72 (82%)	66 (93%)
Female sex (n, %)	56 (64%)	46 (65%)
HIV-1 Subtype (n, %)
D	48 (54.5%)	40 (56.3%)
A1	16 (18.2%)	14 (19.7%)
C	2 (2.3%)	1 (1.4%)
Recombinants	17 (19.3%)	15 (21.1%)
No subtype determined	5 (5.7%)	1 (1.45%)
Pre-ART viral load, log10 copies/ml	4.58 (2.91–6.57)	4.67 (3.00–6.57)
Nadir CD4, cell/mm3	203 (3–939)	182 (3–690)
Age at baseline, years	43 (26.6–59.6)	43.5 (30–59.6)
BMI at baseline, kg/m2	22.7 (15.8–34.2)	22.7 (15.8–34.2)
ART duration at baseline, years	7.1 (1.5–11.7)	7.5 (2.2–11.7)
Observation time, years	3.5 (1.0–5.3)	3.5 (1.0–5.3)
Pre-DTG	2.8 (0.0–4.1)	2.8 (0.0–4.1)
Post-DTG	1.6 (0.6–2.0)	1.6 (0.6–2.0)
Pre-DTG visits (mean)	1.8	1.9
0	1 (1.1%)	1 (1.4%)
1	35 (39.8%)	18 (25.4%)
2	40 (45.5%)	40 (56.3%)
3	9 (10.2%)	9 (12.7%)
4	3 (3.4%)	3 (4.2%)
Post-DTG visits (mean)	1.8	1.6
0	24 (27.3%)	7 (9.9%)
1	16 (18.2%)	16 (22.5%)
2	48 (54.5%)	48 (67.6%)

Median and range presented, unless otherwise stated. DTG = Dolutegravir; ART = Antiretroviral therapy; BMI = Body mass index.

Ethics

All participants provided written informed consent, and ethical approval was obtained from the Institutional Review Boards at National Institutes of Allergy and Infectious Diseases (14IN123), Uganda Virus Research Institute (GC/127/461), Uganda National Council for Science and Technology (HS1651), and Johns Hopkins University (CR00044266).

Quantitative viral outgrowth assay

QVOA was performed as previously described.^6,7 Briefly, samples were stored and run in small batches of 2–4 samples at a time. There were no differences in sample collection methods between years. Frozen PBMC were thawed and CD4+ cells were isolated using negative selecting bead purification (CD4+ T-cell Isolation Kit II, Miltenyi). The number of viable CD4+ T cells were counted and if the sample contained sufficient numbers, rCD4 cells were isolated using negative selecting bead purification (anti-CD25, anti-biotin MicroBeads, anti-CD69 MicroBead Kit II, and anti–HLA-DR MicroBeads; Miltenyi). The resulting rCD4 cells were stimulated with Phytohaemagglutinin (PHA) and γ-irradiated PBMC and co-cultured with MOLT-4 cells transfected with CCR5 and naturally expressing CD4 and CXCR4 (MOLT4/CCR5; National Institutes of Health AIDS reagent program). The co-culture supernatants were tested for presence of HIV p24 by ELISA (PerkinElmer) after 14 or 21 days, to indicate outgrowth of replication-competent provirus.⁸ A portion of 2015 QVOA were only tested for outgrowth viruses at 14 days, which was included and considered in both of the models. All time points collected after 2015, were measured at 21 days, with some also being tested at 14 days.

Next-generation sequencing of outgrowth viruses

Outgrowth viral sequences from p24 positive wells were obtained as previously described.⁹ Briefly, viral RNA was isolated from culture supernatants of p24 positive wells, next-generation site-directed sequencing libraries were created for reverse transcriptase (pol; HXB2 position 2723–3225) and gp41 (env; HXB2 position 7938–8256), and sequenced (Illumina Inc, San Diego CA). Identical sequence reads were combined and the prominent strains (sequences with >2.5% of the total sequence reads analyzed for a given sample) were identified, cleaned of possible intra-well recombinants, and aligned for all time points from a given person.^9,10 These pol and env sequences were used to determine HIV subtype by phylogenetic analysis, and the alignments were used to calculate the amount of change in the pairwise distance overtime using a linear regression model as previously described.¹¹ Briefly, raw pairwise differences between all sequences in the alignment were calculated and compared to the time between the two sequences. These values were then examined using a linear regression to determine if there was a significant association overtime. The R² and p-values were collected for all individuals who had sequence data available from two or more time points, and compared between individuals who experienced a significant increase in their IUPM 0–12 months post-DTG initiation and everyone else with available sequence data. Differences were examined by chi-square analysis.

Data analysis and statistics

Viral load and QVOA data were linked to ART regimens and visit dates. We assume the number of p24 positive outgrowth wells after 21 days is a binomial outcome with probability $p=1-\exp \left(-n_i\lambda \right)$, where n_i is the number of cells in the i-th well and $\lambda$ is the rate of infected cells. This outcome is further partitioned into wells that do or do not have positive outgrowth after 14 days with a probability p_14d. Assuming wells with positive outgrowth at 14 d are always positive at 21 d, the QVOA results can be modeled as multinomial outcomes with probabilities p × p_14d (outgrowth at 14 d and 21 d), p × (1−p_14d) (outgrowth at 21 d only), and 1−p (no outgrowth), respectively. Given this formula and outgrowth data, we used the non-linear minimization algorithm in R (function nlm) to estimate the maximum likelihood (ML) IUPM for each subject and timepoint. We calculated 95% confidence intervals from the Hessian matrix for each estimate. Next, we fit a mixed effects logistic regression to these data:

$$\log \left(\frac{{\lambda }_i\left(t\right)}{1-{\lambda }_i\left(t\right)}\right)={\alpha }_i+{\beta }_i^{sex}+{\beta }_i^{DTG}\left(t\right)+{\beta }_{i}t$$

where ${\lambda }_i\left(t\right)$ is the rate of infected cells for the i-th subject at t days after ART initiation, subject-specific intercept ${\alpha }_i$ and slope ${\beta }_i$ are normally distributed random effects, ${\beta }_i^{sex}$ is the categorical fixed effect of sex, and ${\beta }_i^{DTG}\left(t\right)$ is the categorical fixed effect for samples after switching to an integrase inhibitor-containing regimen. We implemented this model in the RStan package, which provides an R interface to Stan, a Bayesian statistical programming language (https://mc-stan.org), to generate a posterior sample of the model parameters. We ran four replicate chain samples for a warm-up period of 1000 steps and then 3000 steps, recording every 4 steps. Further details about the Stan analysis are provided as Supplementary Material. In addition, we ran posterior predictive checks by simulating QVOA data, i.e., number of positive wells at 14 and 21 days, from the multinomial distribution parameterized from the Stan outputs.

Differences between the ML and Bayesian model IUPM estimates for a given time point were determined to be significant if the ML point-estimate and median Bayesian estimate were both greater or less than the 95% confidence interval or highest density intervals of the ML or Bayesian model, respectively. Note that the ML estimates use only well count data for a given participant and time point, whereas the Bayesian model estimates are derived from data across participants and time points.

Role of funders

The funders had no role study design, data collection, data analysis, interpretation, or writing of the report.

Results

The population used for the longitudinal analysis of the RC-LVR has been described in detail previously (Table 1).^6,7 Participants were majority female (63.6%), and had been on ART for a median of 7.1 years when the study began (range = 1.5–11.7). There was a range of time points per participant included in the study with a median of three time points (Table 1). PBMC samples that had insufficient CD4+ T cells isolated to warrant resting cell isolation were plated as bulk CD4 cells (n = 20) and excluded from this analysis. rCD4 T cell QVOA results (n = 266) were used for the subsequent analyses and included samples from 88 participants (Table 1). Of the 266 QVOA measurements, 154 were obtained prior to DTG-initiation (57.9%). Temporal changes in ML IUPM estimates for all participants were visualized by time on ART and time since DTG-initiation (Fig. 1A and B).

Fig. 1.

Change in Infectious units per million (IUPM) over time. Data are maximum likelihood estimates (MLEs) of IUPM based on quantitative viral outgrowth assays (QVOA, 21 days outgrowth). Participants are aligned based on time of ART initiation (A) or time of dolutegravir (DTG) initiation (B). Open points correspond to samples taken after the participant switched to a DTG-containing drug regimen. Points representing participants with only one sample are reduced in size. Longitudinal samples from the same study participant are connected by line segments and coloured with respect to biological sex (see colour legend). The grey line on (A) is the regression trend for the population-level posterior median values of IUPM over time (mixed-effects logistic regression model); light grey region is the 95% highest posterior density.

QVOA well-level data was used in Bayesian linear mixed effects models to examine the longitudinal change in the RC-LVR of the cohort. First, we fit a linear model of a linear growth or decay of IUPM over time, allowing for variation among participants in the slope and intercept (model A). This model was fit using either the full QVOA dataset (n = 88) or a subset that included only individuals with multiple time points (n = 71; Table 1). Using the full cohort dataset, the individual-level estimates of intercepts and slopes in model A indicated that the overall size of the RC-LVR for the group was slightly increasing over time; however, this rate of change was not statistically significantly different from zero (median doubling time = 49.7 years; IQR = −19.0 to 76.1; Fig. 2B). Estimated slopes among individual participants were highly variable, ranging from significant decreases to significant increases in IUPM over time. Female sex (fixed effect) was associated with a smaller RC-LVR, supporting previous findings from this same population (median ${\beta }_{sex}$ = −1.2; IQR = −1.6 to −0.9, Fig. 2D).⁷

Fig. 2.

Posterior densities of logistic regression model parameters for the full cohort analysis. The curves (green for model A [no-switch] and purple for model B [switch]) represents the posterior density for the mean IUPM at initiation (A), the mean IUPM change (B), and the effect of DTG initiation (C) and biological sex (D) as indicated by the axis label. “Switch” indicates that regression model B was used, which includes a fixed effect for after DTG initiation. Standard deviations for the posterior densities are shown in the supplementary (Supplementary Fig. S2).

To investigate model A’s fit to the data and why the estimated overall slope implied that the size of the RC-LVR for the group was not measurably decaying over time, we checked whether the day 21 ML IUPM estimates were significantly different from those predicted by the model for each timepoint per individual (Supplementary Fig. S4), split by time since DTG-initiation. The day 21 pre-DTG initiation timepoints (n = 140) were unbiased with an approximately even split between ML IUPM estimates that were higher or lower than the model estimates (Table 2a, Table 2b). However, in the first year post-DTG initiation there was a significant bias for ML IUPMs to be higher than model predictions. Specifically, 73% (40/55) of the ML IUPMs were higher than the model predictions, which reverted back to 42% (24/57) in samples taken past one year after DTG-initiation (Table 2; p < 0.005, chi-square test).

Table 2a.

Proportion of IUPM maximum likelihood (ML) estimates that were higher/lower than model A estimates at day 21 by time point pre- and post-DTG initiation.

Time window	ML > Model n (%)	ML < Model n (%)	Total (n = 252)
Pre-DTG initiation	68 (49%)	72 (51%)	140
Post-DTG (0–1 year)	40 (73%)	15 (27%)	55
Post-DTG (>1 year)	24 (42%)	33 (58%)	57

Chi-square significance p < 0.005; Full cohort dataset analysis.

Table 2b.

Significant differences between maximum likelihood and model A estimated IUPM measurements at day 21 by time point pre- and post-DTG initiation.

Time window	Observed > Model n (%)	Observed < Model n (%)	Total (n = 252)
Pre-DTG initiation	5 (3.6%)	7 (5.0%)	140
Post-DTG (0–1 year)	12 (21.8%)	2 (3.6%)	55
Post-DTG (>1 year)	2 (3.5%)	4 (7.0%)	57

Chi-square significance p < 0.001, Full cohort dataset analysis.

A relatively smaller number of timepoints, 12.7% (32/252), had significantly different estimates of IUPM when the ML point estimates and confidence intervals (95%) and Bayesian median and highest density intervals (95%) were compared (posterior predictive checks). Of the 252 individual time points, 19 (7.5%) and 13 (5.2%) had ML estimates significantly higher or lower than the Bayesian model predictions, respectively (Table 2). When examining the distribution of these outlying observations with respect to DTG-initiation, however, there was a significantly higher percentage of individual time points with significantly higher ML IUPM estimates in the first year after DTG initiation, with 22.2% (12/54) of time points in that group having higher ML IUPM estimates than what was predicted by model A (p < 0.001, chi-square test; Table 2). The percentage of significantly higher ML IUPM estimates pre-DTG, and post-DTG (>1 year) were 3.6% (5/140) and 3.4% (2/58), respectively (Table 2). This analysis was also performed stratified by subtype of the infecting strain (determined by outgrowth sequencing of pol and env) and similar patterns were seen in subtype A and D (Supplementary Table S1). Moreover, the ML IUPM estimates for the timepoint one year after DTG initiation were significantly greater than the estimates preceding DTG initiation for the same individuals (paired Wilcoxon test, p = 0.0073).

To determine if DTG initiation explained the increasing size of the RC-LVR for the group implied by model A (Fig. 2B), we incorporated a fixed effect term associated with the time of DTG initiation. Specifically, model B adds a population-level fixed effect that shifts the intercept upon switching to an integrase inhibitor-containing regimen (Supplementary Fig. S5). Adding the DTG fixed effect resulted in significantly decreasing estimated RC-LVR sizes pre- and post-DTG initiation (half-life = 6.9 years; IQR = 5.2–10; Fig. 2B). In addition, DTG initiation was associated with a significant increase in RC-LVR (${\beta }_{DTG}$ = 0.52 logit units; IQR = 0.43–0.61; Fig. 2C). This increase resulted in an approximate 1.7-fold increase in IUPM immediately post-DTG initiation.

Sex had a similar effect in model B as in model A, with female sex associated with a significantly smaller reservoir (${\beta }_{sex}$ = −1.2 logit units). Similar to model A, the individual participant slopes were highly variable, ranging from significant decreases to significant increases. Similar findings were found for both models using the reduced set of individuals with multiple time points (Supplementary Figs. S1–S3). In addition, we fit a third model in which a fixed effect term associated with DTG initiation altered the slope instead of the intercept (model C; see Supplementary Fig. S5). Again, we obtained significant decays in the RC-LVR with a significant increase in the associated with DTG initiation (half-life post-DTG = 14.6 years, IQR = 9.4–33.2 versus half-life pre-DTG = 8.2 years, IQR = 6.0–12.8). Because both slopes share a common intercept at the time of ART initiation, this fixed effect term in model C has the same interpretation as model B of a transient increase in RC-LVR in association with DTG initiation.

One possible explanation for the apparent increase in the RC-LVR in the time points immediately post-DTG initiation would be a short-term loss of viral control during this period. Viral load measurements for all participants were examined for any failures during the study period, but no cases of failure were documented (viral load > 200 copies/mL). In addition, a portion of wells where viral outgrowth was detected in the QVOA were sequenced using a site-directed next-generation sequencing (NGS) assay for the reverse-transcriptase (RT) portion of the pol gene and the gp41 region of the env gene, as previously described.⁹ Subjects with sufficient longitudinal NGS outgrowth data derived from rCD4 QVOA (n = 41) were examined for any signs of continuing evolution by examining change in pairwise distance overtime using a linear regression model as previously described.¹¹ Of the subjects with available NGS data from pre- and post-DTG initiation, there was a median of three time points available per individual (range = 2–5), and a total of 909 wells were sequenced (Fig. 3 and Supplementary Table S2). There were eight individuals (19.5%) with significant changes in their pairwise distances in both pol and env over time (Supplementary Table S2). There were 11 (26.8%) individuals who had a significant increase in pairwise distance of either pol or env, but not the other, and 22 individuals (53.7%) had no significant changes in either region (Supplementary Table S2). It should be noted that due to the large number of comparisons that are generated in pairwise distance measurements of this nature, even small changes, may be found to be significant. Therefore, estimates of pairwise changes for individuals with significant increases in their IUPM 0–12 months post-DTG initiation were compared to those who did not have significant increases during that time. Individuals with increased IUPM measurements were not more likely to have significantly changed pairwise distances (p = 0.33, chi-square test, Fig. 3C), which supports the viral load analysis that the increases seen in the IUPM were not due to underlying ongoing viral replication.

Fig. 3.

Representative neighbor-joining phylogenetic tree of viral pol sequences derived from next-generation sequencing of viral outgrowth viruses from Donor 17, who experienced a significant increase post-DTG initiation. Sequences derived from outgrowth viruses from a specific well are shown as individual circles, and were isolated from pre-DTG switch [baseline (0, red) and years 1 (blue), 2 (green)] and post-DTG switch time points [years 4 (purple), and 5 (gold)], respectively (A). A representative sequence for HIV-1 subtypes B (HXB2) is shown in grey. Genetic distance is indicated by scale at the bottom. These data were used to calculate the pairwise distance over time for Donor 17 and examined for significant changes by linear regression (B; r² = 0.00, p = 0.181). A table of changes in longitudinal pairwise distance in pol and env versus significant increases in IUPM 0–12 months post-DTG for cohort is shown (C).

Discussion

These data are one of the most detailed longitudinal analyses of the LVR in an African population, and identified a possible temporary effect of DTG-treatment initiation on the size of the RC-LVR in this population. The initial observation that the RC-LVR was slightly increasing was the opposite of the expected observation given the previous findings of a smaller RC-LVR in this Ugandan population versus comparable North Americans.⁶ However, given that the majority of individual time points that were significantly higher than the model prediction were found immediately post-DTG initiation, and that when an inflection point was introduced to the regression models, there were significantly negative slopes in both pre-DTG and post-DTG intervals, suggests that the natural pattern of the RC-LVR in this population was declining, if not for significant temporary increase seen post-DTG initiation. In addition, the modeled rate of decline pre-DTG switch corresponds to a half-life of about seven years or more, which is slower than the earlier previous estimates using QVOA derived IUPM data for North American populations.^3,4 However, there are some important caveats that make a direct comparison difficult. First, the Ugandan population was primarily female, while HIV latency studies done in North America and Europe include predominantly male cohorts. The current analysis and other studies suggest that there are important sex-based differences in the RC-LVR.^7,12,13 In addition, most individuals in our study were on ART for >7 years prior to study enrollment, and a recent study by McMyn et al. found a tri-modal decay pattern of intact proviral species in North Americans, with a third phase that consists of very slow decay or even an increasing of the LVR size starting approximately seven years after ART initiation.⁵ It is possible that the relatively longer half-life observed pre-DTG initiation reflects this third phase of LVR maintenance.¹⁴ In addition, given the strong significant increase seen in this Ugandan population post-DTG initiation it would be interesting to examine the role of INSTI initiation in the LVR estimates observed in the long-term North American population examined by McMyn et al.⁵

The finding of a temporary increase in the RC-LVR for approximately one-year post-DTG initiation was a surprising observation. Interestingly, other research studies have also found short-term effects related to DTG-initiation in people with HIV. In particular, the RESPOND study, a large analysis of Europeans and Australians with HIV, found that the risk of a cardiovascular disease event almost doubled for the first six months in individuals who initiated an INSTI, and that this risk decreased over the next 18 months to return to normal.¹⁵ In addition, in a small exploratory study of individuals who were virally suppressed on non-INSTI regimens it was found that there are subtle shifts in the overall make-up of the rCD4 cell population 48 weeks post-DTG initiation that lead to a decrease in total HIV DNA levels in effector memory cells.¹⁶ This same study found that there is also a temporary decrease in viral diversity post-DTG initiation.¹⁷ In all of these cases, the effects observed were temporary, similar to the increase in RC-LVR seen here. It should be noted that the INDOOR study found no changes in total HIV DNA or cell-associated RNA post-DTG initiation in a small group of virally suppressed individuals on protease-inhibitor containing regiments.¹⁸ This highlights an important caveat of our findings since the majority of the Ugandans were on NNRTI-based regimens pre-DTG initiation. It is possible that it is the removal of the NNRTI and not the addition of DTG that is causing the effect seen here. In addition, with only one follow-up time point post the temporary increase in the RC-LVR, it is unclear what the long-term effects on the LVR will be in these individuals, but this work is ongoing.

One possible explanation for this temporary increase in the RC-LVR is a temporary change in the circulating immune cell make-up post-DTG initiation. This is supported by an analysis of the relatively large-scale SWORD 1 and 2 studies, where it was found that in virally-suppressed individuals who switched to a two-drug regimen of DTG and an NNRTI (Rilpivirine), there was temporary increase of soluble CD14 (sCD14), which is a marker of monocyte activation.¹⁹ However, the INDOOR study did not see changes in cytokines, and a smaller study of individuals switching from an NNRTI (Efavirenz) to DTG found a decrease in sCD14 eight weeks post-switch.^18,20 Understanding the possible immunological mechanism behind the temporary increase in the RC-LVR seen here is currently being explored.

In addition, it is unknown from these data what contribution general CD4+ T cell homeostatic proliferation, clonal expansion, or possibly pathogen specific proliferation may be playing a role in the increase in RC-LVR observed post-DTG initiation. This is in part due to the fact that the current methods available for examining clonal expansion in the LVR lack statistical accuracy and robustness.²¹ To this end, our group is currently working to adjust the Bayesian models used here to accurately incorporate proviral sequence data to properly examine changes in clonality.

A limitation of the current study is our inability to use the intact proviral DNA assay (IPDA) in the Ugandan cohort, because that assay has only been validated for HIV subtype B, but we are currently working to validate a subtype-adjusted IPDA to determine if the changes seen here post-DTG initiation are signs of a general increase in the overall size or a change of the cellular make-up of the LVR.²² Another limitation is that the QVOA has a high level of inter- and intra-subject variability. The effect of this variability was mitigated by modeling well-level outgrowth data, as well as using the group and individual level data together. In addition, the size of this cohort helps to mitigate QVOA variability. While there were no documented cases of loss of viral suppression (outside the excluded individual), this prospective study was not designed to capture the effect of DTG initiation, and therefore the timing between DTG initiation and the subsequent viral load/QVOA measurements were not measured directly after post-DTG initiation. However, there was no consistent evidence of viral evolution in individuals who experienced significant increases in their RC-LVR, suggesting that low-level viral failure is not contributing to this finding. Another limitation is that a portion of the baseline QVOA assays were measured only at day 14 of viral outgrowth (based on protocols existing at that time) and not continued to day 21. However, this was included and accounted for in both models.

There are several possible confounders which could have also affected these findings including assay batch effects, differences in sampling, or other population level infectious disease outbreaks. The QVOA protocol used during the study was largely unchanged except for the day 14 and day 21 testing during year 0 stated above, which was accounted for in the model. There were no apparent batch effects in IUPM results, or ARV stock outs during the study. In addition, there were no apparent large-scale outbreaks of other infectious diseases in the Rakai region during the study period with the exception of the COVID-19 pandemic which occurred during the sample collection period of the final sample year. Given that these time points were 18+ months post-DTG switch it should not have any effect on the short-term increase seen 6–12 months post-DTG switch.

It will be critical to examine if the temporary increase seen in these Ugandans is found in other populations of suppressed individuals who switched to DTG or another INSTI, as well as what possible mechanism is causing this increase. In addition, it will be important to examine if this change is due to an overall increase in the total LVR or a shifting of the latently-infected rCD4 T cell population to a more inducible phenotype for 0–12 months post-DTG switch, which could have important implications for possible HIV cure strategies.

Contributors

RCF, SJR, TCQ, JLP, AFYP, ADR designed and oversaw the study; AAC, OB, EEB, EK, JM, JL, SS, CK, BL, JH generated laboratory data; ST, SJ, AA, TK, PB, RMG, SJR oversaw clinical study and subject care; DB, CM, RR, SLL, ADR oversaw sequence analysis; RCF, AFYP performed full data analysis; RCF, AFYP, JLP, ADR verified underlying data and wrote the manuscript; and all authors reviewed, edited, and approved manuscript.

Data sharing statement

Sequence data is available in Genbank (accession numbers OR069776-OR072555). Full relevant study data is available upon request from the corresponding author, and with approval of all pertinent institutional review boards.

Declaration of interests

RR and SLL are employed by BioInfoExperts, LLC. No other authors have any conflicts to report.