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

The principal barrier to an HIV cure is the presence of a latent viral reservoir (LVR) made up primarily of latently infected resting CD4+ (rCD4) T-cells. Studies in the United States have shown that the LVR decays slowly (half-life=3.8 years), but this rate in African populations has been understudied. This study examined longitudinal changes in the inducible replication competent LVR (RC-LVR) of ART-suppressed Ugandans living with HIV (n=88) from 2015–2020 using the quantitative viral outgrowth assay, which measures infectious units per million (IUPM) rCD4 T-cells. In addition, outgrowth viruses were examined with site-directed next-generation sequencing to assess for possible ongoing viral evolution. During the study period (2018–19), Uganda instituted a nationwide rollout of first-line ART consisting of Dolutegravir (DTG) with two NRTI, which replaced the previous regimen that consisted of one NNRTI and the same two NRTI. Changes in the RC-LVR were analyzed using two versions of a novel Bayesian model that estimated the decay rate over time on ART as a single, linear rate (model A) or allowing for an inflection at time of DTG initiation (model B). Model A estimated the population-level slope of RC-LVR change as a non-significant positive increase. This positive slope was due to a temporary increase in the RC-LVR that occurred 0–12 months post-DTG initiation (p<0.0001). This was confirmed with model B, which estimated a significant decay pre-DTG initiation with a half-life of 7.7 years, but a significant positive slope post-DTG initiation leading to a transient estimated doubling-time of 8.1 years. There was no evidence of viral failure in the cohort, or consistent evolution in the outgrowth sequences associated with DTG initiation. These data suggest that either the initiation of DTG, or cessation of NNRTI use, is associated with a significant temporary increase in the circulating RC-LVR.

This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted May 16, 2023. ; https://doi.org/10.1101/2023.05.12.23289896 doi: medRxiv preprint Frozen PBMC were thawed and rCD4 T cells were isolated through negative selection and used for the QVOA to obtain an estimate of the infectious units per million (IUPM) rCD4 T cells using a maximum-likelihood based method, as described previously [7]. Any sample that experienced contamination during cell culture was excluded. 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. The resulting rCD4 T cell QVOA results (n=254) were used for all subsequent analyses, and included samples from 88 participants (Table 1).
Participants were majority female (63.6%), and had been on ART for a median of 7.7 years when the study began (range = 1.5 -12.1). There was a range of time points per participant included in the study (n=19, 13, 20, 31, and 5 with one, two, three, four and five time points, respectively) with a median of three time points. Of the 254 QVOA measurements included, 152 were obtained prior to DTG-initiation (59.8%). Temporal changes in IUPM estimates for all participants were visualized by time on ART and time since DTG-initiation ( Figure 1A and B).
QVOA well-level data was used in two Bayesian hierarchical linear models (A and B) to examine the longitudinal change in the RC-LVR of the cohort. Both models used time since ART initiation as the time variable. However, model B incorporated an inflection in the model at the time of DTG initiation, allowing for different rates of change pre-and post-DTG. In addition to the longitudinal change, the models also examined the effects of biological sex, and were fit using the full dataset (n=88) and a subset that included only individuals with multiple time points (n=69; Table 1). These hierarchical models included both individual-level and cohort-level parameters (i.e. hyperparameters). Individual-level parameters estimated the linear longitudinal changes in the RC-LVR for each participant while the cohort-level parameters related individuallevel parameters between the participants within the cohort. The hierarchical structure facilitates the interpretation of cohort-level features despite individual-level differences between the participants.
for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Using the full cohort dataset, the single individual-level rate model (model A) estimated that the size of the RC-LVR for the group was increasing in a non-significant manner (doubling time = 106.5 years; IQR = -41.4 -23.4; Figure 2B). The individual participant estimated slopes were highly variable, ranging from significant decreases to significant increases. Female sex was associated with a smaller RC-LVR, supporting previous findings from this same population (median = -1.23; IQR = -1.6 --0.9, Figure 2C) [6]. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted May 16, 2023. ; https://doi.org/10.1101/2023.05.12.23289896 doi: medRxiv preprint IUPM in the first year after DTG initiation, with 25% (11/44) of time points in that group being higher than what was estimated by the model (p<0.0001, chi-square test; Table 2). The percentage of significantly higher maximum likelihood IUPM estimates pre-DTG, and post-DTG (>1 year) were 2.2% (3/139) and 3.5% (2/57), 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 D (p<0.0001; Table S1). There were not enough significantly different time points observed in the subtype A or recombinants and other categories to examine these groups for significant differences.
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 (n=88) 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 [8]. The assay can distinguish if a positive well contains one or more viral outgrowth species, and allows for phylogenetic analysis of these outgrowth viruses to examine if there is any evidence of ongoing evolution in the RC-LVR post-DTG initiation. Subjects with sufficient longitudinal NGS outgrowth data derived from rCD4 QVOA (n=41) were examined for any signs of continuing evolution by for use under a CC0 license.
This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted May 16, 2023. ; https://doi.org/10.1101/2023.05.12.23289896 doi: medRxiv preprint examining change in pairwise distance overtime using a linear regression model as previously described [9]. 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 (Figure 4). There were eight individuals (19.5%) with significant changes in their pairwise distances in both pol and env over time (Table 3). 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. 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), which supports the viral load analysis that the increases seen in the IUPM were not due to underlying ongoing viral replication.

Discussion:
These data are the first detailed longitudinal analysis 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 for use under a CC0 license.
This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted May 16, 2023. ; https://doi.org/10.1101/2023.05.12.23289896 doi: medRxiv preprint in this Ugandan population versus comparable North Americans [5]. 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 allowed in model B there was a significantly negative slope pre-DTG, suggests that the natural pattern of the RC-LVR in this population was declining, if not for the temporary increase seen post-DTG initiation.
In addition, the modeled rate of decline pre-DTG switch corresponds to a half-life of nine years, which is slower than previous estimates using QVOA derived IUPM data for North American populations [3,4]. In the North American population, the half-life of overall LVR decline was approximately 3.7 years, which would appear to be significantly faster than the rate of decline observed here. 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 [6,10,11]. In addition, most individuals in our study were on ART for >7 years prior to study enrollment, and some studies have recently found a tri-modal decay pattern of intact proviral species, with the third phase that consists of very slow proviral decay starting approximately seven years after ART initiation (half-life=18.7 years). It is possible that the relatively longer half-life observed here reflects this third, very slow phase of decay [12].
The finding of an increase in the RC-LVR for approximately one-year post-DTG initiation was a surprising observation, and the return to lower levels after one year suggests that this increase was temporary. Interestingly, other research studies have also found short-term effects related to DTG-initiation in people living 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 [13]. In addition, in a small exploratory study of individuals who were virally suppressed on non-INSTI regimens it was for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted May 16, 2023. ; https://doi.org/10.1101/2023.05.12.23289896 doi: medRxiv preprint 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 [14].
This same study found that there is also a temporary decrease in viral diversity post-DTG initiation [15]. 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 [16]. 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 virallysuppressed 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 [17]. 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 [16,18]. Understanding the possible immunological mechanism behind the temporary increase in the RC-LVR seen here is currently being explored.
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 [19]. 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 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted May 16, 2023. ; https://doi.org/10.1101/2023.05.12.23289896 doi: medRxiv preprint 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. Lastly, 19 individuals in the study only contributed one RC-LVR measurement to the analysis. This was due to missed sample collections, contamination, or, excluding any sample that used bulk CD4+ T-cells for the QVOA.
To address this, the analysis was repeated using only people who contributed more than one time point, with virtually identical results.
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. 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.

Materials and Methods:
Study Population: The details of the population examined for this analysis have been discussed in detail previously [5,6]. Briefly, people living with HIV-1 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 for use under a CC0 license. This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted May 16, 2023. ; https://doi.org/10.1101/2023.05.12.23289896 doi: medRxiv preprint annually in 2016, 2017, 2019, and 2020 [5]. 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 [6]. For each study time point, blood was collected and separated into peripheral blood mononuclear cells (PBMC) and plasma, which were both stored for later analysis. 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. Quantitative viral outgrowth assay: QVOA was performed as previously described [5,6]. Briefly, frozen PBMC were thawed and CD4+ cells were isolated using negative selecting bead  [20]. For 2015 samples, all rCD4 cells were plated in a limiting dilution as previously described [20]. A portion of 2015 QVOA were 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. In addition, for use under a CC0 license.
This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
Next-generation sequencing of outgrowth viruses: Outgrowth viral sequences from p24 positive wells were obtained as previously described [8]. Briefly, viral RNA was isolated from culture supernants 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 ( Figure 4) [8,21]. 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 [9]. 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 R 2 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: Viral load and QVOA data were linked to ART regimens and visit dates and merged into a data frame in the R statistical computing environment. We assume the number of p24 positive outgrowth wells after 21 days is a binomial outcome with probability = 1 - This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted May 16, 2023. ; https://doi.org/10.1101/2023.05.12.23289896 doi: medRxiv preprint probability p 14d . Assuming wells with positive outgrowth at 14d are always positive at 21d, the QVOA results can modeled as multinomial outcomes with probabilities p (p 14d ) (outgrowth at 14d and 21d), p (1 -p 14d ) (outgrowth at 21d 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 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 ( ( ) 1 -( ) ) = + + ( + * ( )) + where ( ) is the IUPM for the i-th subject at t days after ART initiation, subject-specific intercept and slope are normally distributed random effects, is the categorical fixed effect of sex , * ( ) is the change in IUPM growth/decay rate upon switching to an integrase inhibitor-containing regimen, and is the error term. 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 1,000 steps and then 3,000 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.
for use under a CC0 license.
This article is a US Government work. It is not subject to copyright under 17 USC 105 and is also made available (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.