Detectability of HIV Residual Viremia despite Therapy Is Highly Associated with Treatment with a Protease Inhibitor-Based Combination Antiretroviral Therapy

HIV persistence despite therapy contributes to chronic immune activation and inflammation, increasing the risk of aging-associated events in HIV-infected individuals. We sought here to better understand the complex link between clinical and treatment features and HIV persistence despite therapy.

ABSTRACT

HIV persistence despite therapy contributes to chronic immune activation and inflammation, increasing the risk of aging-associated events in HIV-infected individuals. We sought here to better understand the complex link between clinical and treatment features and HIV persistence despite therapy. A total of 11,045 samples from 1,160 individuals under combination antiretroviral therapy (cART) with an unquantifiable viral load (VL; limit of quantification, 20 copies/ml) were categorized as detectable or undetectable depending on the detection of a PCR signal using a commercially available assay. Generalized estimating equation (GEE) regression was used to model viral load detectability and to assess the determinants of residual viremia (RV; VL detected below 20 copies/ml) despite therapy. A high VL zenith was associated with a higher probability to have a detectable viremia under cART. Conversely, the probability to have a detectable viral load below 20 copies/ml decreased with time under therapy. Of therapy regimens, protease inhibitor (PI)-based cART was associated with a significantly higher probability of detectable RV compared to nonnucleoside transcriptase inhibitor- or integrase inhibitor-based cART. We found that a PI-based treatment regimen is highly associated with an increased frequency of RV, supporting previous evidence suggesting that PI-based cART regimens could favor ongoing viral replication in some individuals.

INTRODUCTION

The advent of improved combination antiretroviral therapy (cART) allows clinicians to successfully manage HIV-infected individuals and to decrease side effects. However, a new set of HIV-associated complications is emerging. HIV infection is now seen as a chronic disease that for many will span several decades of life (1). Effectively treated HIV-infected adults are at greater risk for aging-associated events, such as cardiovascular disease, neurocognitive disease, osteoporosis, and liver and kidney diseases (1, 2). Non-AIDS-defining cancers are increasing in number and now constitute the majority of cancers diagnosed in the HIV-infected population (3).

Several factors contribute to the excess risk of these non-AIDS events, including a high prevalence of traditional risk factors (such as substance abuse, obesity, and hypertension), chronic immune activation (IA), and inflammation (4). The etiology of persistent IA in the context of successful cART is considered multifactorial and involves microbial translocation, coinfections, metabolic disorders (as for instance oxidation of low-density lipoproteins), regulatory T cell (Treg) deficiency, and decreased thymic function (5–8). More importantly, HIV persistence appears to be a critical factor driving IA (9). Indeed, HIV persists during cART and residual viremia below the limit of detection of clinical assays can be detected in most individuals using improved single-copy assays (10). Persistent HIV residual viremia in cART-treated individuals arises from reactivation of viral expression from latently infected cells that constitute the viral reservoirs (11, 12) and may also originate from ongoing viral replication in anatomical compartments where drug penetration is suboptimal (13–15). The latter mechanism of HIV persistence would be facilitated by cell-to-cell transmission of HIV (16, 17). The two models are not mutually exclusive and may coexist to different extents in different individuals (18).

Viral persistence and residual inflammation are interdependent and fuel each other in a “vicious circle” that seems difficult to interrupt (9). On the one hand, current cART does not inhibit viral transcription from integrated viral genomes and does not prevent reactivation from latency or the inflammation that might result from this phenomenon. On the other hand, chronic inflammation might lead to HIV persistence by generating new target cells, enabling infection of activated and resting target cells, and/or increasing the proliferation of infected cells (6). Of note, HIV persistence would play a key role in IA as it supports the other elements triggering this deleterious process. For instance, residual release of viral particles in the gut may be responsible for local inflammation, leading to epithelium permeability and associated microbial translocation (19). HIV infection also downregulates Treg function (20).

The understanding of factors associated with HIV production or replication despite cART is thus one of the main gaps that need to be bridged since suppressing the latter processes could lead to a decrease in IA and associated damaging consequences. The HIV treatment regimen is one factor that would likely play a role in HIV replication under cART. For instance, it has been shown that intensification of a three-drug suppressive ART regimen with the integrase inhibitor (INI) raltegravir resulted in a transient increase in HIV episomal DNA and a decrease in IA markers. Of note, these effects were mostly observed in individuals treated with protease inhibitors (PIs), suggesting that PI-based regimens could favor ongoing virus replication (21, 22).

In this report, we analyzed more than 11,000 samples originated from 1,160 participants and performed an in-depth analysis of factors associated with detectable viremia below 20 copies/ml in cART-treated individuals to better understand the complex link between clinical features and HIV persistence despite therapy. The objective was to investigate whether clinical features (such as age, gender, or the CD4 T cell nadir) or the ART regimen used were associated with HIV RNA detectability below 20 copies/ml.

RESULTS

We have recorded 11,045 unquantifiable HIV VLs (below 20 copies/ml) from 1160 participants with 1 to 35 samples per individual (median and interquartile range [IQR]; 8 and 4 to 14 samples/individual), depending on the follow-up time of each individual. The median (IQR) follow-up duration was 47 (16 to 82) months. A total of 29% of the samples were classified as “detectable” (n = 3,161), as shown in Table S1 in the supplemental material. The median (IQR) VL detectability rate was 28.6% (12.5 to 50.0) per individual and was not linked to the follow-up duration (P = 0.76).

The characteristics of the participants are presented in Table S2 in the supplemental material. A total of 45% of the individuals were female, and 53% of the studied population was of African origin. The median age at diagnosis was 33.1 years, and the median delay between diagnosis and treatment introduction was 93 days (Table S2).

Participants’ characteristics at sampling time are presented in Table S3 and include the age at sampling (median, 44.3 years old), the treatment regimen, the pill number, the daily dosage requirements, and the duration of last treatment without interruption. We also included the total duration under therapy defined as the sum of different treatment durations if a participant experienced one or more treatment interruptions (Table S3). The CD4⁺ T-cell nadir (median, 259 cells/mm³) and the CD4⁺ T-cell count at sampling (median, 638 cells/mm³) are also included in Table S3.

The detailed ARV regimens used are presented in Table 1 . In summary, the most frequently used nucleotide reverse transcriptase inhibitor (NRTI) combinations were TDF/FTC (47.2%), followed by ABC/3TC (38.2%). As a third agent, DTG was the most prescribed molecule (20.9%), followed by nevirapine (NVP; 17.0%), efavirenz (EFV; 14.7%), and ATV/r (13.4%). Regarding triple combinations, ABC/3TC/DTG was the most frequent option, followed by FTC/TDF/EFV and FTC/TDF/ATV/r (Table 1).

TABLE 1.

Treatment regimen at sampling (n = 11,045 samples)

Treatmenta	No. (%) of samples
2NRTI
3TC/AZT	625 (5.7)
3TC/TDF	63 (0.6)
3TC/d4T	7 (0.1)
ABC/3TC	4,225 (38.3)
ABC/TDF	54 (0.5)
AZT/TDF	27 (0.2)
FTC/TAF	801 (7.3)
FTC/TDF	5,211 (47.2)
ddI/3TC	15 (0.1)
ddI/AZT	2 (0.0)
ddI/FTC	1 (0.0)
ddI/TDF	12 (0.1)
ddI/d4T	2 (0.0)
Third molecule
INI (± booster)
DTG	2,308 (20.9)
RAL	321 (2.9)
EVG/c	926 (8.4)
NNRTI
EFV	1,621 (14.7)
NVP	1,881 (17.0)
RPV	707 (6.4)
ETR	1 (0.0)
PI (± booster)
IDV	1 (0.0)
IDV/r	2 (0.0)
NFV	9 (0.1)
ATV	100 (0.9)
ATV/r	1,485 (13.4)
DRV/r	705 (6.4)
DRV/c	56 (0.5)
LPV/r	750 (6.8)
SQV/r	10 (0.1)
FPV/r	162 (1.5)
Most frequently used treatment combinations
ABC/3TC/DTG	2,052
FTC/TDF/EFV	1,118
FTC/TDF/ATV/r	1,003
FTC/TDF/NVP	938
ABC/3TC/NVP	699
FTC/TDF/RPV	567
FTC/TAF/EVG/c	536
ABC/3TC/ATV/r	434
FTC/TDF/EVG/c	390
FTC/TDF/DRV/r	349
ABC/3TC/DRV/r	346
ABC/3TC/EFV	308
FTC/TDF/LPV/r	301

^a3TC, lamivudine; AZT, zidovudine; TDF, tenofovir disoproxil fumarate; d4T, stavudine; ABC, abacavir; FTC, emtricitabine; TAF, tenofovir alafenamide; ddI, didanosine; DTG, dolutegravir; RAL, raltegravir; EVG/c, elvitegravir-cobicistat; EFV, efavirenz; NVP, nevirapine; RPV, rilpivirine; ETR, etravirine; IDV/r, indinavir-ritonavir; NFV, nelfinavir; ATV, atazanavir; DRV, darunavir; LPV, lopinavir; SQV, saquinavir; FPV, favipiravir.

Clinical factors independently associated with detectable HIV VL below 20 copies/ml.

We used the GEE regression to model VL detectability in order to determine which clinical factors are independently associated with a detectable HIV VL below 20 copies/ml (Table 2 ). A high VL zenith was associated with a higher probability to have a detectable viremia (OR = 1.2 [95% CI = 1.1 to 1.3]). Conversely, the probability to have a detectable HIV VL below 20 copies/ml decreased with time under therapy (per 10 days, OR = 0.998 (95% CI = 0.997 to 0.998) (Table 2).

TABLE 2.

Impact of participant characteristics and cART regimens on the probability to detect HIV RNA below 20 copies/ml (n = 10,227 samples)

Parameter	Coefficient ± SE	P	OR (95%CI)
Intercept	−1.4 ± 0.25
Age at sampling (per 10 years)	–0.018 ± 0.028	0.53	0.98 (0.93–1.04)
Gender (Ref=female)
Male	0.14 ± 0.075	0.062	1.1 (0.99–1.3)
Ethnicity (Ref=Caucasian)
African	0.15 ± 0.077	0.053	1.0 (0.73–1.5)
Others	0.12 ± 0.18	0.51	1.2 (0.998–1.4)
Delay between HIV diagnosis and first treatment (per 10 days)	0.0003 ± 0.0003	0.36	1.0 (0.999–1.001)
CD4 T cell nadir at sampling (per 100 copies/mm³)	–0.029 ± 0.018	0.11	0.97 (0.94–1.01)
Treatment duration (per 10 days)	–0.0021 ± 0.0002	<0.0001	0.998 (0.997–0.998)
VL zenith at sampling (log10 of copies/mm³)	0.20 ± 0.042	<0.0001	1.2 (1.1–1.3)
Treatment regimen at sampling (Ref=2 NRTI, 1PI)
2 NRTI, 1 INI	–0.25 ± 0.063	<0.0001	0.78 (0.69–0.88)
2 NRTI, 1 NNRTI	–0.17 ± 0.068	0.013	0.85 (0.74–0.97)

Male gender and African ethnicity were associated with a higher probability to have a detectable HIV viremia below 20 copies/ml, although it did not reach statistical significance (P values of 0.062 and 0.053, respectively) (Table 2). The age at sampling or the delay between HIV diagnosis and treatment initiation did not independently affect the probability to detect a VL below 20 copies/ml (Table 2).

Treatment features independently associated with detectable HIV VL below 20 copies/ml.

The probability to have a detectable VL was higher when the third molecule was a PI compared to an INI (odds ratio [OR] = 1.3; 95% confidence interval [95%CI] = 1.1 to 1.5]) (Table 2). Non-nucleoside reverse transcriptase inhibitor (NNRTI)-based treatment was also associated with a significantly lower probability of detectable VL compared to PI-based cART (OR = 0.85 [95%CI = 0.74 to 0.97]) (Table 2).

We did not observe any significant difference between INI- and NNRTI-based cART (Table 3). Importantly, these observations could not be explained by differences in NRTI combinations because such combinations were not associated with the probability to detect HIV viremia below 20 copies/ml (Table S4). Nor was the pill burden associated with a higher chance of detecting residual viremia (Table S5). Finally, we also compared the most widely used INI in our cohort (RAL and DTG) and did not observe any statistically significant difference (Table S6).

TABLE 3.

Impact of participant characteristics and cART regimens on the probability to detect HIV RNA below 20 copies/ml (n = 10,227 samples)

Parameter	Coefficient ± SE	P	OR (95%CI)
Intercept	−1.6 ± 0.25
Age at sampling (per 10 years)	−0.018 ± 0.028	0.53	0.98 (0.93–1.04)
Gender (Ref=female)
Male	0.14 ± 0.075	0.062	1.1 (0.99–1.3)
Ethnicity (Ref=Caucasian)
African	0.15 ± 0.077	0.053	1.0 (0.73–1.5)
Others	0.12 ± 0.18	0.51	1.2 (0.99–1.4)
Delay between HIV diagnosis and first treatment (per 10 days)	0.0003 ± 0.0003	0.36	1.0 (0.99–1.001)
CD4 T cell nadir at sampling (per 100 copies/mm³)	−0.029 ± 0.018	0.11	0.97 (0.94–1.01)
Treatment duration (per 10 days)	−0.0021 ± 0.0002	<0.0001	0.998 (0.997–0.998)
VL zenith at sampling (log10 of copies/mm³)	0.20 ± 0.042	<0.0001	1.2 (1.1–1.3)
Treatment regimen at sampling (Ref=2 NRTI, 1 INI)
2 NRTI, 1 NNRTI	0.084 ± 0.062	0.18	1.1 (0.96–1.2)
2 NRTI, 1 PI	0.25 ± 0.063	<0.0001	1.3 (1.1–1.5)

DISCUSSION

In the present retrospective study, which included a very high number of samples from more than 1,100 individuals under triple cART, we showed that the probability for an individual to have a detectable residual HIV viremia below 20 copies/ml is correlated with VL zenith and inversely correlated with the duration of therapy. PI-based treatment was also associated with a higher chance of detectable VL.

Persistent HIV residual viremia in cART-treated individuals could arise from reactivation of HIV reservoirs (11, 12) and/or from ongoing viral replication in anatomical compartments where drug penetration is suboptimal (13, 14). The latter cause is highly controversial and is the source of a permanent debate in the field. However, solid arguments suggest that ongoing replication occurs at least in some individuals despite therapy (14, 21–26). Convergent data show the stronger ability of NVP, compared to efavirenz, to better suppress residual viremia, possibly related to the good penetration of NVP in anatomic compartments (23, 27, 28). Moreover, raltegravir intensification of a three-drug suppressive ART regimen resulted in an increase in episomal DNAs and a decrease in IA markers in a large percentage of cART-suppressed subjects (21, 22). The effect of raltegravir intensification was particularly evident in PI-treated individuals, suggesting that ongoing viral replication occurs in these individuals. Remarkably, our study showed that individuals under PI-based triple ART have a higher chance of having detectable VL below 20 copies/ml. We therefore obtained an additional piece of evidence that ongoing viral replication occurs in some individuals and could be related to specific treatment regimen.

Using samples from 2009 to 2013, Lambert-Niclot et al. very recently showed that VL suppression to <50 copies/ml is associated with INI-class ART initiation (29). We demonstrated the superior ability of INI to suppress VL compared to PI-based regimens, using a more stringent VL detection limit (undetectable below <20 copies/ml) and taking into account the current cART regimen for every VL measurement. Moreover, since we included samples up to 2018, we could include the most recent INIs in our analysis (DTG, etc). Importantly, we showed that PI-based ART regimens are associated with less optimal virological suppression, also compared to NNRTI-based cART. These are major findings of this study as residual replication despite cART would not only induce IA and chronic inflammation but also replenish the HIV latent reservoirs that are considered the main hurdle to an HIV cure (30–32). It should be noted that, using a different study design and statistical approach, Gianotti et al. also observed that regimens based on PIs were associated with a greater percentage of time spent with RV after achieving virological suppression (defined as VL < 50 copies/ml) (33). Moreover, Morón-López et al. observed a decrease in residual viremia following a switch from PI-based to DTG-based regimen in a randomized clinical trial (34).

Besides pharmacokinetic properties and penetration in anatomical compartment, adherence and tolerability issues may explain the higher frequency of detectable residual viremia in PI-treated individuals, although Pasternak et al. did not show any difference in adherence to cART between PI and NNRTI in a recent cross-sectional analysis of HIV-infected individuals with electronically measured adherence (35).

We also showed that the probability to detect a residual viremia decreased with time on therapy and was associated with high VL zenith. This likely reflects the evolution of the HIV reservoirs over time. Indeed, HIV total DNA load decays primarily during the initial 3 to 4 years of treatment (36–41). The decreasing frequency of detectable viremia during cART could therefore reflect the reduction in the HIV reservoir size. Less frequent reactivation from the latent HIV reservoirs could also result from the evolution of the molecular mechanisms underlying HIV latency. For instance, DNA methylation at the HIV promoter progressively accumulates in latently infected cells isolated from cART-treated individuals (42). This phenomenon would likely cause a deep HIV latency and could explain a reduction in reactivation events.

This study has some limitations that deserve discussion. The main limitation is the observational context. Clinicians prescribed different cART regimens to individuals with diverse clinical characteristics. Although we did adjust for these factors in multivariable analysis, residual confounding cannot be entirely excluded. Our study also lacks data on adherence and cART-associated adverse events, which can impact on RV detection. Finally, a more sensitive quantitative assay (ideally single-copy assay [SCA]) would have provided more precise VL measurements. However, SCAs require large amounts of blood and are costly and time-consuming, thus precluding their use for studying RV in large groups of participants.

In conclusion, we demonstrated that the detectability of residual viremia in cART-treated HIV-infected individuals (below 20 copies/ml) is associated with high VL zenith and decreases with time under therapy, very likely reflecting the size of the HIV reservoirs and their ability to be reactivated. These results reinforce the idea that HIV-infected individuals should be promptly treated following HIV diagnosis. We also showed that PI-based treatment regimen was highly associated with an increased frequency of residual viremia, supporting the previous evidences suggesting that PI-based cART regimens could favor ongoing virus replication in some individuals. Given its potential deleterious consequences on IA and on latent reservoir replenishment, this finding should be taken into consideration for the choice of treatment.

MATERIALS AND METHODS

Sample selection.

This was a retrospective cohort study. The laboratory database was screened for individuals of at least 18 years of age that have been followed in Liege university hospital (Belgium) and treated with a three-drug regimen containing two nucleoside/nucleotide reverse transcriptase inhibitors (NRTIs) plus a third drug belonging to one of three following classes: an NNRTI, a PI, or an INI. In addition to a single third drug, a booster drug was allowed. We selected only samples with unquantifiable HIV viral load (VL) (below 20 copies/ml) measured by Roche Cobas AmpliPrep/Cobas TaqMan HIV test version 2.0 (CAP/CTM v2.0).

The CAP/CTM v2.0 assay targets the gag and LTR genes of group M and O HIV strains and quantifies the VL over a range of 20 to 10,000,000 copies/ml (43). The assay also reports qualitative RNA detection below the lower limit of quantification (HIV RNA detected but not quantifiable) (43). HIV RNA unquantifiable VLs were thus categorized as undetected or detected. We included samples from June 2009 to July 2018 since the CAP/CTM v2.0 assay was used to measure all HIV RNA VLs during this period.

The available demographic and clinical data for the participants included the date of birth, gender, ethnicity, date of HIV diagnosis, start date of first antiretroviral (ARV) treatment, CD4⁺ T-cell nadir, and plasma VL zenith. The CD4⁺ T-cell count and information regarding ARV regimens were also collected at the moment of sampling. Using these data, the age at sampling and the duration of treatment were calculated for each sample. In cases where participants temporarily interrupted therapy, we indicated both the total treatment duration (TTD; corresponding to the sum of the previous and current treatment durations) and the last treatment duration without interruption (TD).

The Ethical Committee of the Liège University Hospital approved the study protocol (reference 2019/16). Participants were informed of data collection by their treating physician and could object to further collection of clinical data according to an opt-out procedure. All participants included were assigned unique identification numbers to anonymize the data and protect confidentiality. All methods were carried out in accordance with relevant guidelines and regulations (44).

Statistical analysis.

Data were summarized as means and standard deviations, medians and IQRs, and extreme values for continuous variables, while frequency tables were used for the categorical variables.

GEE regression was used to model the binary VL detectability (0 = “undetected,” 1 = “detected”) as dependent variable in function of sociodemographic, clinical, and treatment characteristics. GEE allows repeated measurements (multiple samples for the same individual) to analyze the impact of the considered covariates on the probability of VL detectability. GEE for binary data with logit link function and exchangeable covariance structure was used. In the models, time-varying variables (age, treatment regimen, treatment duration, pill number, VL zenith, CD4 T cell nadir, and CD4 T cell count) were considered at the sample collection time, while gender, ethnicity, and delay between HIV diagnosis and first treatment were fixed. The results are presented using estimated regression coefficients, as well as their standard errors (SE), P values, and odd ratios (ORs). The VL detectability rate (%) was calculated as the number of samples with “detected” VL × 100/total number of samples for each individual during available follow-up time at each treatment regimen.

P values of <0.05 were considered statistically significant. Missing values were not replaced. Data analysis was carried out using SAS (v9.4 for Windows).