**DOI:**10.1128/JVI.77.3.2271-2275.2003

## ABSTRACT

Three of five virally suppressed human immunodeficiency virus type I (HIV-1)-infected patients treated with highly active antiretroviral therapy and followed intensively with a supersensitive reverse transcriptase PCR assay with a lower limit of quantitation of 5 copies/ml showed statistically significant viral load decays below 50 copies/ml, with half-lives of 5 to 8 months and a mean of 6 months. This range of half-lives is consistent with the estimated half-life of the latent HIV-1 reservoir in the peripheral blood. Those patients without decay of viral load in plasma may have significant cryptic HIV-1 residual replication.

Highly active antiretroviral therapy (HAART) has been shown to have potent effects on human immunodeficiency virus type I (HIV-1) infection and has led to successful treatment of many infected individuals in the developed world. Many of these patients have demonstrated so-called undetectable levels of viral RNA in peripheral blood plasma after treatment with HAART regimens. Here, using a supersensitive PCR-based assay, we examine the dynamics of HIV-1 RNA change in a select subset of patients with viral loads in plasma of <50 copies/ml.

HIV-1 replicates and is cleared in vivo at extremely high rates (17, 26, 30, 33, 39). HAART, by interfering with viral replication, causes declines in plasma virus, with a rapid first phase bringing viral loads in plasma down by 1 or 2 orders of magnitude followed by a slower second phase (26, 27). In diverse cohorts, HAART may lead to viral loads in the peripheral blood plasma of <400 to 500 copies/ml, with somewhat fewer patients obtaining <50 copies of plasma HIV-1 RNA/ml (24). However, in recent clinical trials 80% of patients have achieved <50 copies/ml (13). The first- and second-phase declines are related to loss of productively infected CD4^{+} T lymphocytes and the loss of a long-lived infected cell pool, respectively (26), although loss of virions trapped on follicular dendritic cells may also contribute to the second phase (2, 15, 16). The release of virus by activation of latently infected cells has been suggested as a possible third phase (11), but viral decay kinetics consistent with the hypothesized half-life of latently infected cells of approximately 6 months have never been observed (31, 40).

While HAART induces profound declines in plasma virus, several studies have demonstrated that resting CD4^{+} T lymphocytes and other cells in the body, including seminal cells, may maintain replication-competent proviruses in patients on virally suppressive HAART (4, 5, 10, 25, 29, 34, 35). Also, very-low-level cryptic viral replication has been demonstrated in the peripheral blood and lymphoid tissue of most patients on virally suppressive HAART (2, 12, 18, 19, 21, 28, 40; V. Natarajan, M. Bosche, J. A. Metcalf, D. J. Ward, H. C. Lane, and J. A. Kovacs, Letter, Lancet **353:**119-120, 1999). Ongoing replication and the lack of sensitive assays have hindered the quantitative assessment of possible additional phases of plasma virus decay.

Recently, we have utilized a supersensitive reverse transcriptase PCR (RT-PCR) to evaluate the blood plasma of patients on virally suppressive HAART with fewer than 50 copies/ml of plasma viral RNA by clinical assay systems (6). We demonstrated that the vast majority of patients have low but detectable levels of viral RNA in peripheral blood plasma, even during virally suppressive HAART regimens (6). In the present study, the decay rates of low levels of plasma viral RNA were analyzed for a subset of patients on virally suppressive HAART, who not only had <50 copies/ml by clinical assays but also were known to be compliant and had no intercurrent illnesses. This subset we felt best represented patients with little to no viral replication for whom we might be able to discern continuing plasma viral decay.

## Patient selection.

Five HIV-1-infected men from a larger group of over 80 patients with <50 copies/ml of viral RNA, monitored at the Thomas Jefferson University Medical Center and the Center for Human Virology, were analyzed. These patients were not representative of the total group and were selected on the basis of possibly having the greatest degree of viral suppression. Each of these patients was on a stable HAART regimen and had not experienced intercurrent illness or changes in antiretroviral therapies. As well, each of these patients had relatively long follow-up with repeated supersensitive RT-PCR analyses of his plasma and was known to be adhering to antiretroviral drug therapy during the time of follow-up. All patients had consistently fewer than 400 to 500 copies of viral RNA/ml in peripheral blood plasma in the months prior to being selected for this cohort and subsequent follow-up by supersensitive RT-PCR for peripheral blood plasma viral RNA. These patients were found to have peripheral plasma viral RNA levels of fewer than 50 copies/ml on at least two occasions prior to entry into this cohort for follow-up by the laboratory-based supersensitive RT-PCR assay. All patients were stable immunologically (i.e., regarding CD4^{+} T-lymphocyte counts) and virologically during HAART. The screening and analysis of these patients were approved by the Thomas Jefferson University Institutional Review Board, and each patient signed an informed consent form.

## Plasma HIV-1 RNA and RT-PCR.

Eighty milliliters of peripheral blood was obtained, and Ficoll-Hypaque gradient centrifugation was used to separate cells from plasma. The blood plasma was concentrated via ultracentrifugation at 45,000 rpm for 1 h using an NVT90 rotor on a Beckman ultracentrifuge. The supernatant was discarded, and virion-associated genomic RNA was extracted from the subsequent pellet by using a guanidinium thiocyanate method (Promega, Inc.) (3). The methodology for the supersensitive RT-PCR for HIV-1 RNA was described in detail previously (6, 36-38). The amplified PCR products were hybridized with a probe labeled with phosphorus 32, SK 19, and Southern blotting was then used to visualize the specific bands of the amplicons. A standard curve was developed by using an in vitro-transcribed *gag* RNA construct, as described previously (36). Comparison of the test samples with this serially diluted standard curve of the amplified in vitro-transcribed standard was used to quantify unspliced viral RNA to 5 copies/ml within the linear amplification range of this assay. Viral transcripts below 5 copies/ml were also detected but could not be quantified by this assay system (6). In some samples no viral transcripts could be detected, and they were treated as <1 copy/ml. Quantitation of the viral transcripts was performed via analysis by PhosphoImager (Molecular Dynamics, Sunnyvale, Calif.).

## Linear regression analysis.

Viral load (VL) data were fitted using a maximum-likelihood procedure that allows for ‘censored data' (20). Viral transcripts <5 copies/ml were treated as censored data in the range of 1 to 5 copies/ml if detected or as censored data below 1 copy/ml if undetected.

Let *Y _{j}
* be the VL measurement at time

*t*

_{j}. Also, suppose we have a model that given a set of parameter values predicts the viral load. Let

*θ*be the vector of parameters in the model and let

*f*(θ,

*t*) be the predicted VL at time

_{j}*t*. Finally, assume that

_{j}*Yj*=

*f*(θ,

*t*) + ε

_{j}*, where ε*

_{j}*is the error between the theoretical model and the experimental data, which we assume is normally distributed with variance σ*

_{j}^{2}: ε

_{j}≈

*N*(0,σ

^{2}). We fitted the model to the data by using the following maximum-likelihood procedure.

An uncensored measurement contributes the term
$$mathtex$$\[\frac{1}{\sqrt{2{\pi}{\sigma}^{2}}}\ e^{{-}\frac{1}{2{\sigma}^{2}}\ [y_{j}\ {-}\ f({\theta},t_{j})]^{2}}\]$$mathtex$$
to the likelihood. A censored measurement for a VL where the value is only known to be below a given threshold (α_{1}) has a contribution to the likelihood of
$$mathtex$$\[P(Y_{k}{\leq}{\alpha}_{1}){=}{{\int}_{{-}{\infty}}^{{\alpha}_{1}}}\ \frac{1}{\sqrt{2{\pi}{\sigma}^{2}}}\ e^{{-}\frac{1}{2{\sigma}^{2}}\ [u\ {-}\ f({\theta},t_{k})]^{2}}\ du{=}{\Phi}_{1}({\alpha}_{1},\ {\theta},\ {\sigma},\ t_{k})\]$$mathtex$$
whereas a censored measurement for a VL where the value is known to be only in a given range (α_{1} to α_{2}) has a contribution to the likelihood of
$$mathtex$$\[P({\alpha}_{1}{\leq}Y_{k}{\leq}{\alpha}_{2}){=}{{\int}_{{\alpha}_{1}}^{{\alpha}_{2}}}\ \frac{1}{\sqrt{2{\pi}{\sigma}^{2}}}\ e^{{-}\frac{1}{2{\sigma}^{2}}\ [u\ {-}\ f({\theta},t_{k})]^{2}}du{=}{\Phi}_{2}({\alpha}_{1},\ {\alpha}_{2},{\theta},\ {\sigma},\ t_{k})\]$$mathtex$$

The likelihood function is obtained as the product of the contributions provided by each measurement:
$$mathtex$$\[\left[{{\prod}_{y_{j}{\in}U}}\ \frac{1}{\sqrt{2{\pi}{\sigma}^{2}}}\ e^{{-}\frac{1}{2{\sigma}^{2}}{\ }[y_{j}\ {-}\ f({\theta},\ t_{j})]^{2}}\right]{\cdot}\ \left[{{\prod}_{y_{k}{\in}C_{{\alpha}1}}}\ {\Phi}_{1}({\alpha}_{1},\ {\theta},\ {\sigma},\ t_{k})\right]\]$$mathtex$$
(1)
$$mathtex$$\[{\cdot}\ \left[{{\prod}_{y_{j}{\in}C_{{\alpha}1,{\alpha}2}}}\ {\Phi}_{2}({\alpha}_{1},\ {\alpha}_{2},\ {\theta},\ {\sigma},\ t_{i})\right]\]$$mathtex$$
where *U* = uncensored data, *C*
_{α1} = censored data whose value is only known to be below the threshold α_{1}, and *C*
_{α1,α2} = censored data whose value is known to be only in a given range (α_{1} to α_{2}).

To analyze viral decay, we assume that the logarithm of the VL decays according to a straight line, such that log (*Y*
_{j})= −*mt*
_{j} + *q+* ε*
_{j}
*, where

*m*is the slope of the decay curve. Maximum-likelihood estimates for this linear regression model are then obtained by searching for parameters (

*m*,

*q*, σ

^{2}) that maximize the likelihood function (equation 1).

The decay slope *m*, the initial viral load *q*, and the variance σ^{2} of the error between theory and data were estimated with corresponding 95% confidence intervals (CI). The 95% CI was computed by bootstrapping the pairs (*t*
_{k}, *V*
_{k}) with replacement, where *V*
_{k} is the viral load measured at time *t*
_{k} (8). The significance of the slope *m* being different from zero with *n* observations was tested by a *t* test using the variable *m*/*S*, where *S* is the standard error of the estimated parameter *m* and *n* − 2 is the number of degrees of freedom (32). Since the censored data do not allow a direct computation of *S*, the value was estimated from the bootstrapped data (7).

## Viral load decay rates.

The baseline clinical and virological characteristics of the patients are given in Table 1. All patients had viral loads in plasma below 50 copies/ml on entry into the study. In Fig. 1, the data and the regression lines, when significant, are shown for each of these patients. Table 2 summarizes the estimate of the viral load decay slope *m*, its 95% CI, and the probability of *m* being non-zero. The corresponding half-lives *t*
_{1/2} were computed from the estimate of *m* by the formula *t*
_{1/2} = ln (2/*m*) = 0.693/*m*. The decay slope was statistically different from zero only for patients 1, 2, and 5. For these patients the viral load decay yielded half-lives of 256, 149, and 138 days, respectively, with a mean of approximately 6 months. Patients 3 and 4 did not show statistically significant decay of viral load in plasma.

## Implications.

HAART-induced decay of plasma virus occurs with a rapid first phase followed by a slower second phase (26, 27). Assuming that the second phase of decay continues unabated, virus was predicted to be eliminated from long-lived cells in 2 to 3 years of completely suppressive antiretroviral therapy. Thus, the failure of therapy to eradicate the virus in large cohorts of patients treated for much longer than 3 years suggests that either there exist additional very slowly decaying viral reservoirs or HAART does not fully suppress ongoing viral replication, or both. The existence of an additional phase of decay has been difficult to ascertain because of the limitations involved in quantifying extremely low levels of plasma virus, and alternative approaches were pursued. Refined coculture methods showed that, in individuals whose plasma viremia levels are well suppressed by antiretroviral therapy, peripheral blood mononuclear cells containing replication-competent HIV-1 decayed with a mean *t*
_{1/2} of approximately 6 months (40), close to the decay characteristics of memory lymphocytes in humans and monkeys (14, 22, 23). However, slower decays or no evidence of decay was observed in less selective patient populations (9), possibly due to ongoing cryptic viral replication in some patients.

Here, using a new supersensitive RT-PCR assay with a quantitative threshold of 5 copies/ml, we were able to reveal continuing decay of viral load in plasma below the threshold of 50 copies/ml with a mean *t*
_{1/2} of 6 months and a range of 5 to 8 months. Although this decay was observed for only 3 of the 5 patients studied, it seems significant that the estimated mean half-life of 6 months is the same as the estimated half-life of resting CD4^{+} T lymphocytes harboring replication-competent HIV-1 in individuals consistently maintaining plasma HIV-1 RNA levels of fewer than 50 copies/ml (31, 40). Further, the observation of these decays suggests that the supersensitive assay is providing quantitatively reliable measurements in the range of 5 to 50 copies/ml.

While viral decay continued for three patients, for the other two patients analyzed, no statistically significant decay trend was observed. Thus, for these two patients the viral load in plasma may have reached a new quasi-steady state in which low-level viral replication was balanced by viral clearance or may be so close to steady state that the rate of viral decay could not be reliably established. The existence of such low viral steady states was suggested by the work of Furtado et al. (12), in which constant low levels of *tat*, *rev*, and *gag* mRNA were detected in HAART-treated patients with viral loads in plasma below 50 copies/ml and are predicted by dynamic models of HIV-1-infection and treatment in which drug sanctuaries exist (1).

In the three patients with viral decay continuing below 5 copies/ml, there may still be ongoing cryptic viral replication. This is due to the fact that even if replication is ongoing, as long as the rate of virion production is lower than the rate of clearance, viral loads will continue to decay. An equivalent condition for continuing decay is that the reproductive number be less than 1, i.e., that the number of cells infected by each currently infected cell be less than 1. Thus, the 6-month half-life that we observed for this new phase of plasma virus decay and the 6-month half-life previously observed for latently infected CD4^{+} T lymphocytes harboring replication-competent virus (31, 40) may still overestimate the rate of decay that could be obtained if all replication were stopped. In addition, if replication is ongoing in these patients, it is possible that the currently observed decay will ultimately stop with the attainment of a new very low steady state in which the viral load in plasma is less than 5 copies/ml.

This analysis shows that statistically significant decays of viral load in plasma can be observed at least for some patients in the range of measurements below 50 copies/ml. Due to the presence of resting CD4^{+} T lymphocytes that harbor replication-competent HIV-1, it has been postulated that a third phase of viral decay should be present (11). Here we have provided evidence of this decay for a subset of patients and have shown that it occurs with a half-life of 6 months, the presumptive half-life of the latent reservoir. Ongoing cryptic viral replication from CD4^{+} T lymphocytes, monocytes, macrophages, and/or other cell types occurring at a level sufficient to balance clearance may account for those patients for which no decay of plasma HIV-1 RNA levels was demonstrable (29).

## ACKNOWLEDGMENTS

This work was performed under the auspices of the U.S. Department of Energy and supported by NIH grants RR06555, AI28433 (A.S.P.), and AI46289 (R.J.P.).

We thank C. Han and R. Ribeiro for helpful discussions and Rita Victor and Brenda Gordon for excellent secretarial assistance.

## FOOTNOTES

- Received 7 March 2002.
- Accepted 30 October 2002.

- Copyright © 2003 American Society for Microbiology