Healthy dynamics of CD4 T cells may drive HIV resurgence in perinatally-infected infants on antiretroviral therapy.

In 2019 there were 490,000 children under five living with HIV. Understanding the dynamics of HIV suppression and rebound in this age group is crucial to optimizing treatment strategies and increasing the likelihood of infants achieving and sustaining viral suppression. Here we studied data from a cohort of 122 perinatally-infected infants who initiated antiretroviral treatment (ART) early after birth and were followed for up to four years. These data included longitudinal measurements of viral load (VL) and CD4 T cell numbers, together with information regarding treatment adherence. We previously showed that the dynamics of HIV decline in 53 of these infants who suppressed VL within one year were similar to those in adults. However, in extending our analysis to all 122 infants, we find that a deterministic model of HIV infection in adults cannot explain the full diversity in infant trajectories. We therefore adapt this model to include imperfect ART adherence and natural CD4 T cell decline and reconstitution processes in infants. We find that individual variation in both processes must be included to obtain the best fits. We also find that infants with faster rates of CD4 reconstitution on ART were more likely to experience resurgences in VL. Overall, our findings highlight the importance of combining mathematical modeling with clinical data to disentangle the role of natural immune processes and viral dynamics during HIV infection.

Introduction

In 2019 there were 490,000 children under five living with HIV, and 150,000 newly diagnosed cases [1]. Although infants receiving antiretroviral treatment (ART) can suppress viral load (VL), eventually the cessation of treatment leads to HIV rebound, due to reactivation of latently-infected cells. Nevertheless, early initiation of ART can lead to extended periods of suppression in the absence of treatment—for example, over 22 months in the case of the ‘Mississippi Child’ and 8.75 years in a South African participant of the Children with HIV Early antiRetroviral therapy (CHER) trial [2, 3]. Therefore, understanding the dynamics of HIV suppression and rebound following ART initiation in young infants is crucial for optimizing treatment strategies and increasing the likelihood of achieving and sustaining viral suppression.

We previously showed that a simple biphasic model of VL decay captures the early dynamics of HIV decline in perinatally-infected infants on ART and that these dynamics are similar to those in adults [4]. However, models applied to dynamics of infection in adults over longer timescales typically encode assumptions that do not extend to infants [5-12]. First, CD4 T cell dynamics in adults are typically described as a balance between a constant total rate of influx and a constant per capita rate of loss, leading to steady trajectories in the absence of infection. In contrast, HIV-uninfected infants experience a natural, exponential decline in CD4 T cell numbers per unit volume of blood as the immune system matures [13]. Second, perinatally-infected infants undergo a transient period of CD4 T cell reconstitution upon ART initiation, during which numbers quickly recover to those of HIV-uninfected infants [14]. This short-lived process cannot be captured by the constant CD4 recruitment term exploited in many models of adult infection. Third, the standard assumption that ART is completely effective in blocking new infection of cells may not hold true for young infants, due to challenges in treatment adherence. Thus, canonical models of HIV suppression and rebound in adults must be modified for infants to include potential reductions in ART efficacy, and more complex dynamics of CD4 T cell numbers.

Here we model the dynamics of HIV infection in a cohort of perinatally-infected infants from Johannesburg, South Africa who initiated ART early in life. We extend a simple deterministic model of HIV suppression and rebound in adults to incorporate incomplete treatment adherence and dynamics of natural CD4 T cell decline and reconstitution. By fitting this model to longitudinal viral RNA and CD4 T cell data, we estimate rates of reactivation and reconstitution. We also show that individual variation in CD4 reconstitution rates and VL persistence are important factors driving variation in HIV suppression and resurgence characteristics across infants, in addition to ART adherence. Overall, our results demonstrate the complex interplay between natural immune processes and HIV dynamics, and highlight the importance of mathematical modeling in disentangling these factors.

Materials and methods

Ethics statement

All protocols for the LEOPARD study were approved by the Institutional Review Boards of the University of the Witwatersrand and Columbia University. Written informed consent was obtained from mothers for their own and their infants’ participation.

Data

The LEOPARD study has been described previously [4, 15]. Briefly, 122 perinatally-infected infants were enrolled at the Rahima Moosa Mother and Child Hospital in Johannesburg, South Africa, between 2014 and 2017. The majority began ART within two weeks of birth (median age: 2.5 days; interquartile range (IQR): 1–8), and were followed for up to four years. VL (HIV RNA copies ml−1) and CD4 T cell concentrations (cells μl-1) in the blood were sampled over time, and various clinical covariates were also recorded, including the infant’s pre-treatment CD4 percentage, the mother’s VL and CD4 count after delivery, and the mother’s prenatal ART history (S1(A) Text).

With these data, we previously identified a subset of 53 infants who successfully suppressed VL within one year [4, 16], with suppression defined as having at least one VL measurement below the 20 copies ml−1 detection threshold of the RNA assay. Here, we are interested in the interplay between natural CD4 T cell dynamics and infection processes, and whether and how this interplay determines whether an infant achieves suppression and/or experiences VL rebound. We have therefore broadened our analysis to all 122 infants.

In addition to the data described previously, we include information relating to ART adherence that was obtained at each study visit (S1(B) Text and S1 Fig). For each drug in each infant’s ART regimen—with the recommended, and most common, being zidovudine (AZT), lamivudine (3TC), and (i) nevirapine (NVP) in the first four weeks of treatment, then (ii) ritonavir-boosted lopinavir (LPV/r) after four weeks—we estimated a percentage adherence by comparing the weight of medicine returned to the expected amount returned assuming perfect adherence. Less than 100% adherence can result from missed doses or ‘under-dosing’ (giving too little medicine at each dose), whereas greater than 100% adherence can occur through over-dosing or problems with drug tolerance (infants may spit up bad-tasting medicine, therefore requiring repeat dosing). For many visits, adherence could not be calculated because leftover medicine was not returned. The adherence estimates are therefore influenced by many unobserved factors and, given uncertainty in how the quantitative estimates map to actual adherence, we instead defined a categorical variable that labeled adherence estimates greater than 90% as ‘good’, and estimates less than 90% as ‘poor’. Using 85% and 95% as alternative thresholds for good adherence did not alter our findings. With this course-grained approach, some missing values could be manually labelled based on physician commentary from accompanying questionnaires (for example, if substantial gaps in dosing were noted, adherence was labeled as poor). We then summarized the average adherence of each infant as the most frequently reported category (good or poor) across their time series.

Model

We describe HIV dynamics in an infant on ART using a deterministic ordinary differential equation (ODE) model (Fig 1A) [7, 12]. We assume that CD4 target cells, T(t) (measured as the concentration of cells per μl of blood, but from here on referred to as ‘cells’ or ‘counts’ for brevity), undergo background growth and loss at net rate θ(t, T), and are infected by free infectious virus, VI(t), with a per capita transmission rate β (Table 1). During ART this transmission is blocked with efficacy ϵ1, where ϵ1 < 1 reflects incomplete adherence to reverse transcriptase inhibitors (for example, NVP and AZT) and/or their failure to completely block infection of new cells. A proportion (1—ϕ) of the newly infected target cells either die through abortive infection [17, 18] or seed the latent reservoir; the remaining fraction, ϕ, become productively infected. These infected cells I(t) are lost at an average rate δ, and produce infectious virus at an average per capita rate (1 − ϵ2)p, where ϵ2 < 1 reflects incomplete adherence to protease inhibitors (for example, LPV/r) and/or incomplete blocking of the production of infectious virus. The infected cell population is also boosted by reactivation of the latent reservoir, at total rate a. We do not explicitly model the number of latently infected cells due to uncertainties in the rates of proliferation and loss in this population, and a lack of available data to estimate these parameters. Finally, free virus is lost at rate c. This system is represented by the following equations;

Fig 1

Model framework and analysis schematic.

(A) The infection model with rate constants. CD4 target cells (T) are infected by free infectious virus (VI) and either become productively infected cells (I), die though abortive infection, or become latently infected. Productively infected cells produce both infectious (green) and non-infectious (red) virus, both of which are cleared at rate c, and latently infected cells can become reactivated at a later point to join the productively infected cell population. CD4 target cells also undergo reconstitution and natural decline processes at total net rate θ(t, T). Further details are given in the text. (B) Schematic illustrating the definition of viral resurgence (top) compared to no resurgence (bottom). The timing of resurgence is defined as the time at which viral load first starts increasing (vertical red line), and the size of resurgence is the total integrated viral load during the upslope period (blue shaded region). The dashed horizontal line represents the detection threshold of the assay.

Table 1

Model parameters and population-level estimates.

Parameter	Description	Units	Value, if fixed	Mean (SE), if estimated*
β0 = (1 − ϵ1)(1 − ϵ2)β	Per-cell effective transmission rate	(copies ml−1)−1 day−1		4.0 × 10−8 (9.5×10−9)
ϕ	Proportion of infections that are productive	–	0.05–0.35† [17, 18]
p¯=p/c	Ratio of viral production to loss	copies ml−1 cell−1		1210 (174)
δ	Average rate of loss of infected cells	day−1		0.06 (0.007)
a	Total rate of latent reservoir reactivation	cells day−1		2.1 × 10−4 (5.7×10−5)
b 0	Extent of natural CD4 T cell decline	cells	-2354 [13]
-1/b1	Timescale of natural CD4 T cell decline	days	1003 [13]
r	Rate of CD4 T cell reconstitution	cells day−1		7.5 (0.5)
T R	Age at reconstitution plateau	days	210–235†[14]
T 0	Initial number of CD4 T cells	cells		1659 (76)
V 0	Initial viral load	copies ml−1		7386 (1838)

*Estimates taken from the model with lowest AIC

†A range of values around previous estimates was explored

SE = standard error of the fixed effect; cells = cells μl−1

The basic reproduction number for this model at ART initiation is where T0 is the initial CD4 count, and we assume the contribution of the latent reservoir to the production of infected cells is negligible at this time.

In the simplest case we assume all rate parameters are constant over time. We also investigated an extension of this model that incorporates a delay in reactivation of the latent reservoir, such that where T is the time to reactivation in days. Assuming the rate of virus turnover is faster than that of CD4 T cells [19], we reduce the model to the following system (S1(C) Text); where V is total free virus VI + VNI, β0 = (1 − ϵ1)(1 − ϵ2)β, and . The compound parameter represents the contribution of each infected cell to the total viral load, through its rate of production of both infectious and non-infectious virions (p) and the average time they persist in circulation (1/c). The compound parameter β0 incorporates both transmission and the total efficiency of both modes of action of ART (through ϵ1 and ϵ2), and thus includes the extent of each infant’s adherence to treatment. In this model, . Note that because β0 is a single parameter which we estimate from the data, we cannot use this expression to estimate the pre-treatment value of R0.

Finally, we extend the model for young infants through the term governing background CD4 T cell growth and loss, θ(t, T). In models of infection dynamics in adults, θ(t, T) typically takes the form λ − dT, where λ and d are constant rates representing cell influx and natural decay processes, respectively. These forms lead to steady trajectories in the absence of infection. For infants, we propose an alternative θ(t, T) that instead accounts for (i) the exponentially declining concentration of CD4 T cells that is observed as HIV-uninfected infants age [13], and (ii) the transient recovery in CD4 counts experienced by HIV-infected infants during the early stages of ART [14]. First, the natural decline in CD4 T cells can be captured by an exponential decay function where c0, b0 and b1 are constant parameters that have been independently estimated in a cohort of 80 uninfected children in Germany, including 39 aged between 2 months and 4 years [13]. This function also captured CD4 T cell dynamics in 381 South African children, of whom 300 were aged between 2 weeks and 5 years [20]. Second, the additional reconstitution of the CD4 T cell pool in HIV-infected infants can be modeled as a transient increase in cell counts during the early stages of ART, i.e. where r is the constant rate of reconstitution and T is the time to reach healthy levels [14]. Combining these processes of reconstitution and the natural decline of CD4 T cell counts in infants gives , and where

Model fitting and comparisons

We fit Eqs 1 and 2 to the VL and CD4 T cell data from all 122 infants using a nonlinear mixed effects approach. All VL observations below the detection threshold were treated as censored values, and we assumed both V(t) and T(t) were lognormally distributed [21, 22]. Given the relative infrequency of CD4 T cell measurements, we fixed four parameters across all individuals (Table 1): three that governed the reconstitution and natural dynamics of target cells (T, b0 and b1), and the proportion of newly infected cells that become productively infected (ϕ). All other parameters were estimated and allowed to have both fixed and random effects. In subsequent analyses we estimated fixed and random effects for T and ϕ. We also examined the importance of individual variation in adherence, VL persistence and CD4 T cell recovery by comparing the best fit model to three alternative models in which or r were fixed across all infants.

Following exploratory fits, each estimated parameter was assumed to follow a lognormal distribution, with the exception of a which followed a logit-normal distribution with pre-specified upper bound, and T which followed a normal distribution. We verified that the random effects for all estimated parameters were normally distributed, using the Shapiro-Wilk test. Guided by the exploratory fits, we allowed β0 and d to be correlated, and assumed all other parameters were independent. We confirmed the structural identifiability of all parameters [23], detailed in S1(D) Text, and conducted additional sensitivity analyses by varying each chosen parameter in turn and re-simulating the model, while keeping all other parameters fixed. We used these simulations to assess the sensitivity of model predictions to our choice of fixed parameters. Model fitting and parameter estimation were implemented in Monolix 2020R1 [22], detailed in S1(E) Text. Downstream analyses and plotting were conducted in R version 4.03 [24], with the deSolve, cowplot, patchwork and tidyverse packages [25-28].

We compared the statistical support for different models using the Akaike Information Criterion (AIC). For model i, AIC = 2k − 2 ln L, where k is the number of estimated parameters, ln L is the maximum log-likelihood, and lower AIC values indicate stronger statistical support. We assessed the relative support for model i using ΔAIC = AIC—AIC, where AIC is the minimum AIC value across all models. Differences greater than five indicate substantially greater support for the model with AIC = AIC. For the favored model, we used the individual-specific parameter estimates to predict VL and CD4 T cell trajectories for each child. These trajectories extended either to the end of our study period or two years after their last observation, whichever was earlier. We then compared how viral infection and the natural decline in CD4 T cells mediated the overall VL and CD4 T cell dynamics. We calculated the relative contributions of new viral infection and natural decline to decreases in CD4 T cell numbers as respectively. Similarly, the relative contributions of new viral infection and latent reservoir reactivation to increases in the number of productively infected cells were respectively.

Statistical analyses

We tested for statistical associations between model parameters, clinical covariates (S1(A) Text), and the risk of VL resurgence—defined as any predicted increase in VL following initiation of ART (Fig 1B). We chose VL resurgence as our indicator of imperfect viral control rather than VL rebound (any predicted increase in VL following initial suppression of HIV) due to the small number of infants experiencing the latter (7/122 infants experienced rebound compared to 53/122 experiencing resurgence). We defined the timing of VL resurgence as the first point at which the model predicted an increase in VL, and the size of resurgence as the total integrated VL during the upslope period (Fig 1B). We then tested for associations using Spearman correlations between pairs of quantitative variables, the Kruskal Wallis test between quantitative and categorical variables, and Chi-squared tests between pairs of categorical variables. We adjusted for multiple testing using the Benjamini-Hochberg correction.

Results

The model for adult infection, with θ(t, T) = λ − dT, was a poor fit to the infant data, particularly the CD4 T cell counts (S2 Fig, Table 2). We therefore used the model adapted for infant infection, with , in all further analyses. First, we verified that the infant model with fixed time to reconstitution plateau, T, and constant rate of latent reactivation, a, was structurally identifiable (see Table 1 and S1(D) Text) [23]. We initially fixed T = 222 days across all infants, following previous modeling of CD4 reconstitution in another cohort of HIV-infected infants who initiated ART 82 days after birth, on average [14]. We refitted the model with different fixed values and verified that the best fits were obtained when T = 230 days (S3 Fig). We also varied ϕ between 0.05 and 0.35 [17, 18] and found the best fit for ϕ = 0.35. Including random effects for T and ϕ did not improve model fits, nor did estimating the fixed effects (i.e., estimating the population average of T and ϕ; S1(F) Text). This is likely due to the increased complexity introduced by estimating these additional parameters. Similarly, including a delay in reactivation of the latent reservoir did not improve model fits (Table 2). We therefore focus on the model with a constant rate of reactivation, fixed T = 230 days, and fixed ϕ = 0.35. With this model, VL predictions were marginally sensitive to T and ϕ (S4 Fig), whereas CD4 T cell dynamics were only sensitive to T (S5 Fig).

Table 2

Model comparisons.

AIC values (ΔAIC) are quoted relative to the minimum AIC value across all models. The model with ΔAIC = 0 is the model with lowest AIC and thus has most statistical support. See Table 1 for parameter definitions.

Model*	ΔAIC
Constant reactivation, a	0.0
Time-varying reactivation, a = a(t)	212.9
θ(t, T) = λ − dTT	373.4
No reactivation, a = 0	419.3

*Unless stated otherwise, with T = 230 days and ϕ = 0.35.

Strikingly, our relatively simple deterministic model captured the wide variation in infant VL trajectories, including monotonic decreases to suppression, eventual suppression following transient increases in VL, and brief periods of suppression with a subsequent rebound in VL (Fig 2). Later, or multiple, rebound occurrences were generally not so well captured. These behaviors may be due to repeated fluctuations in treatment adherence or stochastic processes driving delayed reactivation of the latent reservoir, neither of which are included in the model. Initially, new infections were the major contributor to growth of the productively infected cell population (Eq 4; S6 Fig). However, in almost all infants the importance of new infection events was eventually superseded by reactivation from the latent reservoir, although this displacement was delayed by viral resurgence events (S7 Fig).

Fig 2

Model fits for viral RNA observations.

Each panel represents a different infant; points represent the data; and solid lines are the model fits. The dashed horizontal line is the detection threshold of the RNA assay, and red crosses are censored observations below this threshold. Panels shaded in red are infants who experienced viral resurgence (i.e. at least one period of increasing VL).

The majority of infants experienced a transient increase in CD4 T cell counts followed by a steady decline; these patterns were well captured by the model (Fig 3). The decline in CD4 T cells was almost always driven by natural processes, although the contribution of new infections increased during periods of VL resurgence (Eq 3; S8 Fig).

Fig 3

Model fits for CD4 T cell observations.

Each panel represents a different infant, ordered as in Fig 2; points represent the data; and solid lines are the model fits.

The fixed effects for all estimated parameters, and the standard error of the fixed effects, are given in Table 1. The average lifespan of productively infected cells (1/δ) across all individual infant estimates was 16 days (2.5%—97.5% percentile range = 3–61 days), and R0 at ART initiation was 0.48 (0.29–1.00), reflecting an initial decrease in VL across most infants. The rate of CD4 T cell reconstitution, r, was positively correlated with the initial number of CD4 T cells, T0 (S9 Fig), in contrast to the negative correlation reported elsewhere in other young cohorts [14, 29]. The basis of this difference is unclear, although we note that the infants in the LEOPARD cohort all initiated ART at a much younger age than the children in those studies. This association is unlikely to be driven by poor parameter identifiability, which would instead cause negative correlations through compensatory mechanisms. Notably, we found that infants with higher reconstitution rates, r, and higher VL production to decay ratios, , were more likely to experience increases in VL after ART initiation (p < 1 × 10−6; Fig 4A and 4B). For those infants who did experience increases in VL, larger and earlier increases were associated with higher VL production to decay ratios (p < 10−4; Fig 4C and 4D), but not reconstitution rates (p > 0.1). Overall, while it is intuitive that rebound dynamics would be positively associated with , which reflects both the average rate of virus production by infected cells and the persistence time of free virus in circulation, we find that the CD4 reconstitution rate r is also a key parameter determining an infant’s propensity for resurgence events.

Fig 4

VL resurgence is associated with rates of CD4 reconstitution, VL production and decay, and ART history of infant and mother.

(A–B) Relationship between the occurrence of VL resurgence (defined as any increase in VL following initiation of ART) and the CD4 reconstitution rate, r, in cells μl-1 day-1 (A) and the ratio of virus production to decay, , in copies ml−1 cell−1 (B). Each point represents a different infant and p < 0.0001(****) in both cases. Seven infants whose resurgence was a viral rebound event are highlighted in orange. (C–D) Relationship between and the size of VL resurgence in RNA copies ml−1 (C) and timing of VL resurgence in days (D). Each point represents an infant who experienced resurgence. Correlations are 0.66 and -0.75, respectively, and p < 0.0001 in both cases. (E) Relationship between the occurrence of VL resurgence and the timing of maternal ART initiation (p < 0.01). The size of each box reflects the proportion of infants in the corresponding category and the numbers show the corresponding sample size. (F) Relationship between infant ART adherence and the occurrence of VL resurgence (p < 0.05 for AZT and LVP/r; p = 0.06 for NVP). Adherence was classified as ‘good’ if the majority of adherence estimates were 90% or more, and ‘poor’ otherwise.

In addition to the model parameters, we found a longer duration of maternal prenatal ART was associated with risk of VL resurgence (p < 0.01; Fig 4E), as were poor LVP/r and AZT adherence (p < 0.05; Fig 4F). However, maternal ART and our binary adherence covariates were also associated with higher VL production to decay ratios (S10 Fig; all p < 0.01), suggesting potential colinearity. All other associations between VL resurgence characteristics and clinical covariates, including pre-treatment CD4 percentage and age at ART initiation (S1(A) Text), were not significant at the α = 0.05 level.

Variation in viral persistence, reported adherence, and the natural dynamics of CD4 T cells dictate infant trajectories

The most obvious explanation for the wide variety of VL trajectories we have identified here is variation in ART adherence, which may be more pronounced in infants than adults. Indeed, we found that reported poor adherences to LVP/r or AZT, which are the treatments most commonly administered long-term, were associated with resurgence (Fig 4F). In the model, adherence is reflected in the parameters ϵ1 and/or ϵ2, which dictate the efficiency of treatment at blocking new infection and virus production by infected cells, respectively. These parameters were not individually identifiable with these data, but instead are subsumed in the compound parameter β0 = (1 − ϵ1)(1 − ϵ2)β. If adherence were the main driver of variation in VL trajectories, then inter-individual variation in β0 should be the most crucial component of our model. Puzzlingly, however, we found no association between β0 and any measure of resurgence, or between β0 and average LVP/r or AZT adherence (p>0.05 in all cases), although we did find a negative association between β0 and adherence to the early-phase treatment NVP (p<0.05). Instead, within the model we found r and to be the key predictors of resurgence (Fig 4A–4D), suggesting that variation in CD4 T cell and VL persistence dynamics are important. We speculate that the strong correlation between β0 and the infected cell death rate δ across children (S9 Fig) masks the effect of inter-individual variation in the parameters ϵ1 and/or ϵ2, which is better captured by the adherence data derived from questionnaires. To explore this issue, we refit the model while removing the random effects for β0, and r in turn. Fixing any of these parameters resulted in substantially poorer fits (ΔAIC > 40; Table 3), suggesting that variation across individuals in CD4 T cell dynamics, VL persistence, and ART adherence all drive variation in HIV suppression and resurgence characteristics across infants.

Table 3

Model comparisons of adherence and CD4 recovery parameters.

AIC values (ΔAIC) are quoted relative to the minimum AIC value across all models. The model with ΔAIC = 0 is the model with lowest AIC and thus has most statistical support. See Table 1 for parameter definitions.

Model	ΔAIC
Fixed and random effects for β_0,p¯ and r *	0.0
Fixed and random effects for β0 and p¯; only fixed effects for r	44.5
Fixed and random effects for β0 and r; only fixed effects for p¯	53.1
Fixed and random effects for p¯ and r; only fixed effects for β0	358.8

*Corresponds to the best-fit model in Table 2.

Discussion

In this study we modeled the dynamics of HIV suppression and rebound in perinatally-infected infants receiving ART. Our framework extends previous models of rebound in adults [7, 12] by incorporating mechanisms of the natural decline and infection-induced reconstitution of CD4 T cells in young infants [14]. We found that new infection events were initially the major contributor to growth of the productively infected cell population, but that reactivation of the latent reservoir became more important once VL levels were low. We also identified natural processes as the long-term driver of declining CD4 T cell frequencies in blood. What was perhaps unexpected was that our simple framework can capture large variations in infant VL trajectories, including monotonic decreases to sustained suppression, resurgences in VL, and suppression with subsequent rebound. Although the canonical explanation for erratic VL patterns is imperfect ART adherence, we found that incorporating variation in CD4 reconstitution rates was also required to capture the complexity in our infant data. We demonstrate that within a deterministic framework the interplay of natural CD4 dynamics, constant levels of latent reservoir reactivation, and constant ART efficacy can recapitulate intricate infection dynamics. Our analyses indicate that with the levels of adherence achieved in this study, resurgence may in fact be inevitable for infants with certain virological and CD4 T cell parameter combinations.

Although our estimates of the average CD4 T cell reconstitution rate (r = 7.5 cells μl-1 day-1) is greater than those from another cohort of HIV-infected infants (r = 3.8 cells μl-1 day-1; ref. [14]), it is within the interquartile range. Notably, infants from this other cohort initiated ART later, on average, than the infants in our cohort (median = 82 days, 25th percentile = 34, 75th percentile = 121), and all eventually achieved viral suppression. We also found that higher rates of reconstitution were associated with a greater probability of experiencing a resurgence in VL. This relationship was not confounded by the immunological status of infants at the beginning of the study as we found no association between the reconstitution rate and pre-treatment CD4 percentage or counts, or between the risk of VL resurgence and pre-treatment CD4 percentage or counts. Our finding raises the possibility that rapid recovery of CD4 T cells, despite suggesting an improved clinical state, can also increase the risk of VL resurgence in some individuals by repopulating the target cell pool. Although it could also be that VL resurgence triggers more rapid CD4 reconstitution through increased anti-viral immune activity or density-dependent responses to CD4 depletion [14], the latter seems unlikely in this cohort as we did not detect a negative association between the initial number of CD4 T cells (T0) and r. Nevertheless, further investigation is needed to determine the directionality of this relationship, and whether the extent of CD4 T cell recovery may be used as a biomarker for individuals at increased risk of VL resurgence.

Our estimate of another key parameter, the rate of latent cell reactivation (a = 2 × 10−4 cells μl-1 day-1), is within the range of estimates obtained from adults during ART interruption (2 × 10−6—1 × 10−3 cells μl-1 day-1[7]). Biologically, a higher burden of reactivation (a) may reflect a larger latent reservoir in these infants and/or an increased per-cell rate of latent cell reactivation. Dynamically, larger reactivation estimates may compensate for significant fluctuations in treatment adherence that are not included in the model. We did not find any associations between a and the occurrence or size of VL resurgence. This is perhaps not surprising as a effectively represents the total contribution of latent cell reactivation averaged over the entire study, and its contribution to changes in VL relative to those of de novo infection events soon after ART initiation tends to be small (S6 Fig).

One counterintuitive result is that longer durations of maternal prenatal ART were associated with VL resurgence. This result could not be explained by worse adherence in this group (p > 0.2). However, we could not disentangle the effects of this variable from that of VL production to decay ratios, . Another study of this cohort found that longer exposure to maternal prenatal ART is associated with a larger viral reservoir [30]. It was speculated that maternal ART could lead to a larger representation of infants who acquired infection earlier during the pregnancy, potentially before ART was initiated, and hence have had a longer time to progress. Alternatively, there may be an enrichment of immuno-genetic risk factors in infants who become infected despite maternal ART.

There are a number of caveats to our modeling approach. First, our model does not differentiate between short- and long-lived productively infected cells, the loss of which underpin the multiphasic decline of VL in adults and infants on ART [4, 7, 31–34]. Instead, our estimate of the mean lifespan of a productively infected cell (1/δ) is effectively a weighted average of the mean lifespans of all productively-infected subpopulations. Our estimate (16 days) is consistent with the median lifespan of 17 days we estimated previously from the VL dynamics in a subset of these infants who achieved suppression [4], and roughly in line with a recent estimate of the loss rate of infected CD4 T cells with intact proviruses [35]. Second, we do not explicitly model the dynamics of latently infected cells as they are not directly observed. Instead, the parameter a in our model is effectively a ‘force of reactivation’, which combines the effects of reservoir size and the per-cell rate of reactivation. Third, we fit the peripheral CD4 T cell data to the number of target cells predicted by the model (T(t)), rather than the predicted sum of target cells, productively infected cells and latently infected cells. This approach is reasonable as the frequency of infection among CD4 T cells is typically small (S11 Fig and ref. [36]), and the majority of infected cells most likely reside in lymphoid tissues where infection-induced depletion of CD4 T cells is greatest [37]. We also assume all CD4 T cells are equally susceptible to infection, although in reality activated cells may be more susceptible than resting cells [38, 39]. However, this heterogeneity is implicitly incorporated within the transmission parameter, β, if the proportion of CD4 T cells that are susceptible remains approximately constant over time.

Finally, we acknowledge that the ART regimens used in the LEOPARD trial may not be optimal. Although considered most effective at the time of study design and implementation, more potent treatments—for example, integrase inhibitors and/or broadly neutralizing antibodies—have since been approved for young infants. It will be important to determine whether infants starting these newer treatments are also at risk of resurgence, as we have identified here.

In conclusion, we have extended the canonical framework for HIV suppression and rebound to include more realistic dynamics of CD4 T cell decline and reconstitution in young infants on ART. We estimated rates of reactivation and reconstitution, and identified distinct phases in which dynamics were either dominated by new infection of CD4 T cells, or by reactivation of the latent reservoir. We also demonstrated the importance of incorporating variation in CD4 reconstitution rates to capture the diversity of infant VL trajectories. Overall, our results suggest that VL resurgence in perinatally-infected infants may be inevitable in certain parameter regimes, and highlight the utility of mathematical modeling in understanding the dynamics of infant HIV infection.

Descriptions of clinical covariates and assessments of infant adherence, the derivation of the model, its structural identifiability, and our approach to parameter estimation using Monolix.

A—List of clinical covariates. B—Additional information regarding infant adherence. C—Model. D—Structural Identifiability. E—Nonlinear Mixed Effects Modeling in Monolix.

(PDF)

Click here for additional data file.

Reported adherence trajectories.

Adherence estimates greater than 90% were labeled ‘good’; and all others ‘poor’. Each panel represents a different infant.

(PDF)

Click here for additional data file.

The standard model for adult CD4 T cell dynamics does not capture infant data.

Each panel represents a different infant, points represent the data, and solid lines are the model fits. Here θ(t, T) = λ − dT, with λ and d assumed to have lognormal distributions. Initial estimates for the population mean were 1000 cells μl−1 day−1 and 0.25 day−1, respectively, and for the standard deviation were 1 and 0.1, respectively.

(PDF)

Click here for additional data file.

Comparing models with different fixed values of T and ϕ across all infants.

The AIC difference for model i was calculated as AIC AICmin, where AICmin is the minimum AIC value across all models. The model with zero difference is the model with lowest AIC and thus is the most strongly favored.

(PDF)

Click here for additional data file.

Sensitivity of VL predictions to model parameters.

Each fixed (A) or estimated (B) parameter was varied within 20% of its original value while keeping all other parameters at their original values. Original values for the estimated parameters were the population-level means from the best-fit model.

(PDF)

Click here for additional data file.

Sensitivity of CD4 T cell predictions to model parameters.

Each fixed (A) or estimated (B) parameter was varied within 20% of its original value while keeping all other parameters at their original values. Original values for the estimated parameters were the population-level means from the best-fit model.

(PDF)

Click here for additional data file.

Relative contributions of new infection events (grey) and LR reactivation (blue) to the generation of productively infected cells.

Each panel represents an infant, and red shaded regions show their VL scaled by its maximum value.

(PDF)

Click here for additional data file.

VL and CD4 T cell dynamics influence the time at which reactivation contributes most to productively infected cell growth.

Each point represents a different infant with respect to the time at which reactivation became the major contributor to productively infected cell growth and the time at which: (A) their VL started increasing (if applicable) and (B) their VL finished increasing (if applicable); p = 0.6 and p < 0.001, respectively, and the Spearman’s rank correlation coefficient for (B) is 0.61.

(PDF)

Click here for additional data file.

Relative contributions of healthy dynamics (green) and depletion through infection (grey) to the decline in CD4 T cell densities in blood.

Each panel represents an infant, and red shaded regions show the periods of increasing VL (from start to peak, as shown in Fig 1B).

(PDF)

Click here for additional data file.

Correlations between estimated parameters.

The color and magnitude of each point shows the strength of the correlation; those with p-values greater than a significance threshold of 0.05 are crossed out. p-values were adjusted using the Benjamini-Hochberg correction. The strong correlation between β0 and d was included in the nonlinear mixed effects model framework.

(PDF)

Click here for additional data file.

Longer duration of maternal ART, and poor adherence, are associated with greater VL production to decay ratios.

The ratio is given by , in copies ml−1 cell−1. Each point represents a different infant.

(PDF)

Click here for additional data file.

The frequency of infection in CD4 T cells is usually small.

Distribution of the maximum proportion of total CD4 T cells that are infected (I(t)/(I(t)+ T(t)) across all infants.

(PDF)

Click here for additional data file.

29 Mar 2022

Dear Prof. Yates,

Thank you very much for submitting your manuscript "Healthy dynamics of CD4 T cells may drive HIV resurgence in perinatally-infected infants on antiretroviral therapy" for consideration at PLOS Pathogens. As with all papers reviewed by the journal, your manuscript was reviewed by members of the editorial board and by several independent reviewers. The reviewers appreciated the attention to an important topic. Based on the reviews, we are likely to accept this manuscript for publication, providing that you modify the manuscript according to the review recommendations.

All three authors agree this is an excellent and important paper, which develops a mathematical model that takes into account many parameters of viral dynamics, and fits it to unique pediatric HIV infection data. However, all three reviewers all made multiple (and often overlapping) comments on clarity and structure of the paper, which I am quite confident can be addressed to make this paper clearer. Reviewer 1 noted that the data that the authors rely on is not accessible - this is of course an important issue as well.

Please prepare and submit your revised manuscript within 30 days. If you anticipate any delay, please let us know the expected resubmission date by replying to this email.

When you are ready to resubmit, please upload the following:

[1] A letter containing a detailed list of your responses to all review comments, and a description of the changes you have made in the manuscript.

Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out

[2] Two versions of the revised manuscript: one with either highlights or tracked changes denoting where the text has been changed; the other a clean version (uploaded as the manuscript file).

Important additional instructions are given below your reviewer comments.

Thank you again for your submission to our journal. We hope that our editorial process has been constructive so far, and we welcome your feedback at any time. Please don't hesitate to contact us if you have any questions or comments.

Sincerely,

Susan R. Ross, PhD

Section Editor

PLOS Pathogens

Susan Ross

Section Editor

PLOS Pathogens

Kasturi Haldar

Editor-in-Chief

PLOS Pathogens

​orcid.org/0000-0001-5065-158X

Michael Malim

Editor-in-Chief

PLOS Pathogens

orcid.org/0000-0002-7699-2064

***********************

All three authors agree this is an excellent and important paper, which develops a mathematical model that takes into account many parameters of viral dynamics, and fits it to unique pediatric HIV infection data. However, all three reviewers all made multiple (and often overlapping) comments on clarity and structure of the paper, which I am quite confident can be addressed to make this paper clearer. Reviewer 1 noted that the data that the authors rely on is not accessible - this is of course an important issue as well.

Reviewer Comments (if any, and for reference):

Reviewer's Responses to Questions

Part I - Summary

Please use this section to discuss strengths/weaknesses of study, novelty/significance, general execution and scholarship.

Reviewer #1: This is an interesting well-done analysis of CD4 T cell dynamics in perinatally-infected infants on ART.

The authors develop a mathematical that incorporates drug adherence, the kinetics of CD4 T cell resurgence in infants on ART as well as previously described declines on CD4 T cell populations with age. They fit their new model to data from the LEOPARD study and find that their new model agrees well with the data, whereas earlier simpler models did not. The paper increases our understanding of pediatric HIV infection and the effects of early ART and introduces new equations to describe T cell dynamics in HIV-infected infants.

Despite being very enthusiastic about this work, I do have some technical comments that need to be addressed.

1. In your description of the model on lines 97 and 98 you say �  �  is the fraction of infected cells that become productively infected or die through abortive infection. However, if they die then they should not contribute to the productively infected cells, I. I think you should have the abortively infected cells in the (1-�  �  �  fraction and that you should refer to I as the productively infected cells

2. You correctly point out that protease inhibitor blocks the production of infectious virus. Thus, shouldn’t your V equation refer to infectious virus not total virus? You could then add an equation for non-infectious virus as in Perelson et al. Science 1996 and properly fit the measured RNA to the sum of infectious and non-infectious virus.

3. To be consistent with previous literature it would be helpful to use d for the death rate of productively infected cells rather than d.

4. The compound parameter c bar is really a scaled viral production rate as suggested by your description of it as the contribution of each infected cell to the total VL. As such, why not use the symbol p bar. To avoid any confusion with the parameter representing clearance.

5. Please explain how you decided to use a logit-normal distribution for a and normal distributions for r and TR based on your preliminary fits. Also, what was the pre-specified upper bound for a?

6. I was surprised that you r estimate of the death rate of infected cells was so much smaller than estimates in adults, 0.05 per day vs ~ 1 per day (cf Markowitz et al JVI 77:5037-5038 (2003). Are there other estimates of the death rate of infected cells in infants that are consistent with your estimate?

7. I was also surprised that your estimate of R0 at ART initiation was 0.35 given that the median age at ART initiation was 2.5 days. If infection occurred at or near the time of birth then I would expect that on average the virus to be growing exponentially and not decaying before ART. Are there studies showing viral dynamics in untreated infants? In SHIV infection of rhesus macaques viral load increase for the first 1-2 weeks after infection – Fig 1 in JCI Insight. 2021;6(23):e152526. Please discuss this issue.

8. It is not clear how one can get access to the data used in this analysis. A data availability statement is needed

Reviewer #2: This is a very well-written paper, fitting data from young HIV-infected children with an impressive and careful mathematical fitting procedure. The mathematical model is extremely well-chosen as it is simple enough to keep parameters identifiable and rich enough to be able to account for recovery of CD4+ T cells and the natural decline in CD4 T cell counts in young children. It is impressive how well this simple model captures the wide variation in trajectories of the viral load and the CD4 count in all these children.

Reviewer #3: In this paper, the authors analyze an unique data set of HIV treatment in infants with a mathematical model for the viral dynamics, which also includes the dynamics of the target cells (in this case, CD4+ T cells). The paper is well written and the modeling elegantly done. The main strengths include the data set used, the novel modeling approach, and the interesting results. The paper is clearly written and the scholarship is of high caliber. Still, the work could be improved with some clarifications and also reference to some previous key publications.

Overall, the study is important in the relatively less studied field of pediatrics HIV, with focus on viral dynamics, and could be important to better understand the responses to treatment in this subgroups of infected people. The study is also somewhat novel, due to the quality of the data set and the simultaneous consideration of viral loads and CD4 counts, and treatment adherence.

**********

Part II – Major Issues: Key Experiments Required for Acceptance

Please use this section to detail the key new experiments or modifications of existing experiments that should be absolutely required to validate study conclusions.

Generally, there should be no more than 3 such required experiments or major modifications for a "Major Revision" recommendation. If more than 3 experiments are necessary to validate the study conclusions, then you are encouraged to recommend "Reject".

Reviewer #1: not applicable

Reviewer #2: No major issues.

Reviewer #3: There are no extra experiments needed to complete the manuscript. However, I think that there are some important issues that would benefit from some clarification.

One of the main assumptions is implicit in the use of CD4 T-cell numbers per unit volume. Although this is common in models of viral dynamics, I wonder if the absolute changes occurring in infants are properly captured this way. For example, the CD4+ T-cell concentration (per unit volume) decreases in this age range, but the absolute number may be increasing. HIV infection may be more dependent on the total available target cells and not their concentration. At least some thoughtful discussion of this issue would be welcome.

Related, in part, to the previous comment, there is important work in this area that should be referenced in the current paper. See for example the papers by Lewis, Callard et al in Frontiers Immuno 2017 and in J Infect Dis 2012, among others for their treatment of CD4+ T-cell levels; and the paper by Nagaraja, Gopalan, et al Scientific Reports 2021; or the older Melvin, Rodrigo et al JAIDS 1999.

One of the most difficult things in a long-term study of this kind in the change in therapy protocols that occur either by medical prescription or adherence issues. It would be useful to have some discussion of what the impact of using the average adherence for each infant is, as opposed to using their time series of adherence, which might not be possible (the model would be too complex for the data). However, I must say that it was not clear at all how you actually used the adherence (even the “average adherence”) of each infant in the model – do epsilon1 and/or epsilon2 (or their surrogates beta0 or c-bar) vary in time or are they estimated differently for different children? Was this used as a covariate (although it is not indicated in Table S1)? I couldn’t find how you used the adherence information. The available information itself is nicely shown in figures S9 and S10, and described in the first section of the supplementary material. But how was it used in the modeling?

It is not clear what is the impact and the need to consider the latent reservoir, namely the inclusion in the model of rho and a. Can you fit the data when rho= 1 and/or when a= 0?

Related with the previous comment, you state in line 199 that “in almost all infants the importance of new infection events was eventually superseded by reactivation from the latent reservoir”. However, this is only true when there is no more measurable virus (in the next sentence it even says that “this displacement was delayed by viral resurgence events…”). So, how relevant is the infection from the latent reservoir in the context of your study. Are these needed only for virus rebound when treatment is stopped?

One of your findings is that longer duration of maternal prenatal ART was associated with larger production of virus and with viral resurgence. This seems counter intuitive, and a little more discussion would be helpful. Is it that longer treatment may lead to transmission of resistant strains? In the discussion (line 286), you do say that longer prenatal ART is associated with larger viral reservoir in the infants. This is perhaps more puzzling that explanatory.

In your final conclusion, you state that resurgence may be inevitable in certain parameter regimes. This was not clear. What about increasing the adherence or improving efficacy of the drug protocol? The latter is a composed parameter, so perhaps you can't check in your model. But you could simulate the first effect for a couple of infants with resurgence to check what would happen assuming perfect adherence and keeping other parameters the same.

**********

Part III – Minor Issues: Editorial and Data Presentation Modifications

Please use this section for editorial suggestions as well as relatively minor modifications of existing data that would enhance clarity.

Reviewer #1: No editorial or data presentation issue

Reviewer #2: Line 104 Model: The cells undergoing abortive infection are included into the population of infected cells (that in this paper have a long expected life span). An alternative way of writing this would be to include the abortive infections in the rho parameter, i.e., cells undergoing abortive infection fail to arrive in the compartment of productively infected cells (which has been done previously). Would that make any difference, and what is the most natural way to write this?

Line 151: Now rho=0.9999. If this were to include abortive infections it would be much smaller.

Line 208: What is the R0 in the absence of ART according to these parameters? Is that a reasonable value?

Line 218 and Fig 4B: Is there some circularity in this reasoning: children experiencing a resurgence in the VL necessarily have a slower average downslope of the VL, i.e., a lower production to decay ratio?

Lines 267-268: Is drug adherence not correlated for these 3 compounds?

Fig S5c: I agree this correlation is due to the outliers. Are these parametric correlations? A non-parametric test could be more careful.

Fig S9: Difficult to get a message from this, is Fig S10 not sufficient?

Reviewer #3: Line 60: interquartile range is formally a single number, the range, although I do recognize that it is often used in the way presented here, i.e., the 25th and 75th percentiles.

Line 111: Using c-bar is not a great choice, because it gives the impression that this parameter is related to c (the clearance rate), but in fact it is inversely proportional to c, and you interpret it as the “contribution of each infected cell to the total viral load”, so more like p (production rate). It seems that p-bar would be easier for the reader.

Line 145: please add the word “structural” to your description of “We confirmed the identifiability of all parameters…”.

Table 1 (and also in some parts of the text, e.g., line 206): you present the “SE= standard error (random effect)”. But Monolix usually (by default) presents the SE of the fixed effect, and the standard deviation of the random effect. Please confirm which one of these you are discussing, or if indeed it is the standard error of the random effects, which is something different and indeed probably not very appropriate for this situation.

Line 153: You used the general AIC formula. Usually, it is recommended to use the small sample corrected AIC, even when the sample seems large enough, because you are also estimating many parameters (in Table 1, 19 parameters or so, including error and correlation parameters). However, technically the small sample AIC formula does not apply directly to multivariate models (and fits), as is the case here and further corrections are needed. See for example the book by Burnham and Anderson (section 7.7.6). This is just something for you to consider what is the best option in this case.

Figures with time courses: I suggest that you plot the figure with the line on top of the data, rather than the data on top of the line. The latter makes it difficult to see the fits when there is a concentration of data points.

Line 219: what do you mean with “deterministic model parameters”, deterministic as opposed to what?

Figure 4: Please define the labels in the x axis: "Before" - before being pregnant? Define "12+ wks" more than twelve weeks of treatment or initiated at more than 12 weeks of pregnancy? This only become clear when consulting table S1.

Figure 4 caption: description of panel (F), the words “is associated” seem extra.

Line 310: You state that “frequency of infection in CD4 T cells is usually small”, but was this true in your case. That is, what is the time course of I(t)/(T(t)+I(t)) in your results?

Table S1: It is not clear that “Variable treatment: continuous” means when you describe the variable as grouped. For example, “Birth weight” was divided into two groups more or less than 2500g. How do you treat this as a continuous covariate in Monolix? The same question for “Mother’s viral load”, etc…

Figure S7: The symbols seem to have different sizes, but this is not described in the caption.

**********

PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: Yes: Rob de Boer

Reviewer #3: No

Figure Files:

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org.

Data Requirements:

Please note that, as a condition of publication, PLOS' data policy requires that you make available all data used to draw the conclusions outlined in your manuscript. Data must be deposited in an appropriate repository, included within the body of the manuscript, or uploaded as supporting information. This includes all numerical values that were used to generate graphs, histograms etc.. For an example see here: http://www.plosbiology.org/article/info%3Adoi%2F10.1371%2Fjournal.pbio.1001908#s5.

Reproducibility:

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. Additionally, PLOS ONE offers an option to publish peer-reviewed clinical study protocols. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols

References:

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

24 Jun 2022

Submitted filename: Response-to-reviewers.pdf

Click here for additional data file.

14 Jul 2022

Dear Prof. Yates,

Thank you very much for submitting your manuscript "Healthy dynamics of CD4 T cells may drive HIV resurgence in perinatally-infected infants on antiretroviral therapy" for consideration at PLOS Pathogens. As with all papers reviewed by the journal, your manuscript was reviewed by members of the editorial board and by several independent reviewers. The reviewers appreciated the attention to an important topic. Based on the reviews, we are likely to accept this manuscript for publication, providing that you modify the manuscript according to the review recommendations.

In addition to the minor revisions requested by reviewer 3, please make sure that there is a statement in the manuscript regarding the public availability of your data.

Please prepare and submit your revised manuscript within 30 days. If you anticipate any delay, please let us know the expected resubmission date by replying to this email.

When you are ready to resubmit, please upload the following:

[1] A letter containing a detailed list of your responses to all review comments, and a description of the changes you have made in the manuscript.

Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out

[2] Two versions of the revised manuscript: one with either highlights or tracked changes denoting where the text has been changed; the other a clean version (uploaded as the manuscript file).

Important additional instructions are given below your reviewer comments.

Thank you again for your submission to our journal. We hope that our editorial process has been constructive so far, and we welcome your feedback at any time. Please don't hesitate to contact us if you have any questions or comments.

Sincerely,

Susan R. Ross, PhD

Section Editor

PLOS Pathogens

Susan Ross

Section Editor

PLOS Pathogens

Kasturi Haldar

Editor-in-Chief

PLOS Pathogens

​orcid.org/0000-0001-5065-158X

Michael Malim

Editor-in-Chief

PLOS Pathogens

orcid.org/0000-0002-7699-2064

***********************

Reviewer Comments (if any, and for reference):

Reviewer's Responses to Questions

Part I - Summary

Please use this section to discuss strengths/weaknesses of study, novelty/significance, general execution and scholarship.

Reviewer #1: This is a thorough analysis of the dynamics of CD4 T cells in perinatally infected infants on ART. The approach used is novel and should be of general interest.

Reviewer #2: See my previous review: this is an excellent paper.

Reviewer #3: (No Response)

**********

Part II – Major Issues: Key Experiments Required for Acceptance

Please use this section to detail the key new experiments or modifications of existing experiments that should be absolutely required to validate study conclusions.

Generally, there should be no more than 3 such required experiments or major modifications for a "Major Revision" recommendation. If more than 3 experiments are necessary to validate the study conclusions, then you are encouraged to recommend "Reject".

Reviewer #1: No major issues or new experiments needed.

Reviewer #2: See my previous review: this is an excellent paper.

Reviewer #3: (No Response)

**********

Part III – Minor Issues: Editorial and Data Presentation Modifications

Please use this section for editorial suggestions as well as relatively minor modifications of existing data that would enhance clarity.

Reviewer #1: No issues

Reviewer #2: All my recommendations have been handled carefully.

Reviewer #3: I thank the authors for the careful revision of the manuscript. I have only a couple of very minor questions.

1) Page 9, line 231: I believe this should be "standard error of the fixed effects" and not random effects. Please re-check throughout.

2) Page 9, line 234: are these the limits of the interval, or the range of multiple intervals across the children.

3) Page 9, line 258: instead of p, it would me more appropriate to say "not significant at the alpha=0.05 level."

4) Page 15, line 348: regarding your estimate of 16 days, there is a new paper that may be relevant, see White et al. PNAS 2022: https://doi.org/10.1073/pnas.2120326119

**********

PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: Yes: Rob J de Boer

Reviewer #3: No

Figure Files:

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org.

Data Requirements:

Please note that, as a condition of publication, PLOS' data policy requires that you make available all data used to draw the conclusions outlined in your manuscript. Data must be deposited in an appropriate repository, included within the body of the manuscript, or uploaded as supporting information. This includes all numerical values that were used to generate graphs, histograms etc.. For an example see here: http://www.plosbiology.org/article/info%3Adoi%2F10.1371%2Fjournal.pbio.1001908#s5.

Reproducibility:

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. Additionally, PLOS ONE offers an option to publish peer-reviewed clinical study protocols. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols

References:

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

Submitted filename: Morris et al review2.docx

Click here for additional data file.

18 Jul 2022

Submitted filename: Corrections-V2.rtf

Click here for additional data file.

19 Jul 2022

Dear Prof. Yates,

We are pleased to inform you that your manuscript 'Healthy dynamics of CD4 T cells may drive HIV resurgence in perinatally-infected infants on antiretroviral therapy' has been provisionally accepted for publication in PLOS Pathogens.

Before your manuscript can be formally accepted you will need to complete some formatting changes, which you will receive in a follow up email. A member of our team will be in touch with a set of requests.

Please note that your manuscript will not be scheduled for publication until you have made the required changes, so a swift response is appreciated.

IMPORTANT: The editorial review process is now complete. PLOS will only permit corrections to spelling, formatting or significant scientific errors from this point onwards. Requests for major changes, or any which affect the scientific understanding of your work, will cause delays to the publication date of your manuscript.

Should you, your institution's press office or the journal office choose to press release your paper, you will automatically be opted out of early publication. We ask that you notify us now if you or your institution is planning to press release the article. All press must be co-ordinated with PLOS.

Thank you again for supporting Open Access publishing; we are looking forward to publishing your work in PLOS Pathogens.

Best regards,

Susan R. Ross, PhD

Section Editor

PLOS Pathogens

Susan Ross

Section Editor

PLOS Pathogens

Kasturi Haldar

Editor-in-Chief

PLOS Pathogens

​orcid.org/0000-0001-5065-158X

Michael Malim

Editor-in-Chief

PLOS Pathogens

orcid.org/0000-0002-7699-2064

***********************************************************

Reviewer Comments (if any, and for reference):

9 Aug 2022

Dear Prof. Yates,

We are delighted to inform you that your manuscript, "Healthy dynamics of CD4 T cells may drive HIV resurgence in perinatally-infected infants on antiretroviral therapy," has been formally accepted for publication in PLOS Pathogens.

We have now passed your article onto the PLOS Production Department who will complete the rest of the pre-publication process. All authors will receive a confirmation email upon publication.

The corresponding author will soon be receiving a typeset proof for review, to ensure errors have not been introduced during production. Please review the PDF proof of your manuscript carefully, as this is the last chance to correct any scientific or type-setting errors. Please note that major changes, or those which affect the scientific understanding of the work, will likely cause delays to the publication date of your manuscript. Note: Proofs for Front Matter articles (Pearls, Reviews, Opinions, etc...) are generated on a different schedule and may not be made available as quickly.

Soon after your final files are uploaded, the early version of your manuscript, if you opted to have an early version of your article, will be published online. The date of the early version will be your article's publication date. The final article will be published to the same URL, and all versions of the paper will be accessible to readers.

Thank you again for supporting open-access publishing; we are looking forward to publishing your work in PLOS Pathogens.

Best regards,

Kasturi Haldar

Editor-in-Chief

PLOS Pathogens

​orcid.org/0000-0001-5065-158X

Michael Malim

Editor-in-Chief

PLOS Pathogens

orcid.org/0000-0002-7699-2064