Persistent HIV-1 replication maintains the tissue reservoir during therapy

Abstract

Lymphoid tissue is a key reservoir established by HIV-1 during acute infection. It is a site associated with viral production, storage of viral particles in immune complexes, and viral persistence. Although combinations of antiretroviral drugs usually suppress viral replication and reduce viral RNA to undetectable levels in blood, it is unclear whether treatment fully suppresses viral replication in lymphoid tissue reservoirs. Here we show that virus evolution and trafficking between tissue compartments continues in patients with undetectable levels of virus in their bloodstream. We present a spatial and dynamic model of persistent viral replication and spread that indicates why the development of drug resistance is not a foregone conclusion under conditions in which drug concentrations are insufficient to completely block virus replication. These data provide new insights into the evolutionary and infection dynamics of the virus population within the host, revealing that HIV-1 can continue to replicate and replenish the viral reservoir despite potent antiretroviral therapy.

Figures

Extended Data Figure 1. The amounts of virus and the concentrations of drugs measured during antiretroviral therapy

Panels a-c show how the number of copies HIV-1 RNA per ml of blood, the number of the HIV-1 RNA particles bound to the follicular dendritic cell network per gram of lymphoid tissue, and the number HIV-1 RNA positive cells per gram of lymphoid tissue change over the first 6 months of treatment in subjects 1774, 1727 and 1679 (a, b and c, respectively). Filled circles represent detectable measures. Unfilled circles represent undetectable measures and are plotted at the limit of detection. Panels d-f show antiretroviral concentrations in cells from lymph node (dashed line) or blood (solid line) in subjects 1774, 1727 and 1679 (d, e and f, respectively) (see Methods). Intracellular TFV-diphosphate concentrations (fmol/106 cells) are shown in orange, FTC-triphosphate (fmol/106 cells) in green, ATV (ng/mL) in purple, and EFV (ng/mL) in blue. Samples with concentrations that were below the limits of quantification (2.5 fmol/106 cells, 2.5 fmol/106 cells, 0.014 ng/mL and 0.063 ng/mL, respectively) were assigned a value of 1 for graphical illustration purposes.

Extended Data Figure 2. Phylogenies and Highlighter plots for the Gag region of HIV-1

Maximum-likelihood trees were constructed using gene sequences from the Gag region of HIV-1 from lymph node and blood before and after the guanosines within all possible APOBEC3 trinucleotide sequence context of edited sites were masked in the alignments, regardless of their presence in hypermutant or non-hypermutant sequences, to avoid their distortion in the phylogenetic reconstructions. Branch tips are colored according to compartment sampled: red for plasma; gold for lymph node; and blue for blood. The progressive shading of the colors of the branch tips indicate the points in time sampled. Phylogenetic trees reconstructed from the haplotypes in which the guanosines in the APOBEC3 trinucleotide context of the edited sites are masked in the alignments correct the skewing effect caused by clustering of shared haplotypes that harbor repetitive G-to-A substitutions and longer branch lengths caused by a larger number of these mutations in the hypermutated sequences whilst retaining the phylogenetic information. The horizontal scale indicates the expected number of substitutions per nucleotide site per unit time with haplotypes from later time points having diverged more. The Highlighter plots show the haplotypes from the lymphoid tissue and blood time point clusters aligned to the plasma virus sequence from day 0. The particular nucleotide changes are color coded in the alignment (thymidine, red; adenosine, green; cytosine, blue; and guanosine, orange). Magenta circles represent APOBEC3-induced G-to-A change in a trinucleotide context of the edited sites, which are distinguishable from the more random error-prone viral reverse transcriptase and RNA polymerase II replicating enzyme induced mutations. Gene sequences from the Gag region of HIV-1 from Subject 1774, who continued to have measureable amounts of HIV-1 RNA in plasma on treatment, and Subjects 1727 and 1679 who were well-suppressed on treatment (a, b and c, respectively) before and after the guanosines within the particular APOBEC3 trinucleotide sequence context of edited sites were masked in the entire sequence alignment (left and right panels, respectively).

Extended Data Figure 3. Phylogenies and Highlighter plots for the Pol region of HIV-1

Maximum-likelihood trees were constructed using gene sequences from the Pol region (retrotranscriptase [pol2]) of HIV-1 from lymph node and blood before and after the guanosines within all possible APOBEC3 trinucleotide sequence context of edited sites were masked in the alignments, regardless of their presence in hypermutant or non-hypermutant sequences, to avoid their distortion in the phylogenetic reconstructions. Branch tips are colored according to compartment sampled: red for plasma; gold for lymph node; and blue for blood. The progressive shading of the colors of the branch tips indicate the points in time sampled. The horizontal scale indicates the expected number of substitutions per nucleotide site per unit time with haplotypes from later time points having diverged more. The Highlighter plots show the haplotypes from the lymphoid tissue and blood time point clusters aligned to the plasma virus sequence from day 0. The particular nucleotide changes are color coded in the alignment (thymidine, red; adenosine, green; cytosine, blue; and guanosine, orange). Magenta circles represent APOBEC3-induced G-to-A change in a trinucleotide context of the edited sites. Gene sequences from the Pol region of HIV-1 that spanned the genomic region encoding the viral enzyme reverse transcriptase from Subjects 1774, 1727, and 1679 (a, b and c, respectively) before and after the guanosines within the particular APOBEC3 trinucleotide sequence context of edited sites were masked in the entire sequence alignment (left and right panels, respectively).

Extended Data Figure 4. Alternative drug-dependent fitness landscape plots

a, Fitness landscape plot for a partially drug-resistant strain that confers relatively low-level resistance to drugs as compared with the fitness costs imposed by the drug-resistant mutations. The drug-resistant strain (blue line) does not outcompete the drug-sensitive strain (orange line) at any effective treatment concentration where it can grow. There are two phases to the dynamics. At lower effective drug concentrations (left of grey line) the drug-sensitive strain thrives. Beyond this threshold, neither strain can continuously replicate. b, Fitness landscape plot for a highly drug-resistant strain. This strain confers a high-level of drug resistance relative to the replicative fitness cost imposed by the resistance mutations. At low effective drug concentrations (left of grey line), the drug-sensitive strain outcompetes the drug-resistant strain. At high effective drug concentrations, the drug-resistant strain outcompetes the drug-sensitive strain and can continuously replicate. We argue that, typically, fully drug resistant mutants of this sort neither exist in the viral population of patients before treatment, nor arise through random mutation during the course of antiretroviral therapy (see Supplementary Information and Supplementary Table 2). Drug-resistant strains, which are capable of ongoing replication at high effective drug concentrations are not typically generated in individuals because: they are generated in a single step very rarely; and stepwise generation from partially resistant strains is also rare because partially resistant strains are outcompeted in the sanctuary site which constantly refills the pool. The strain specific effective reproductive numbers for the drug-sensitive RS (orange) and drug-resistant RR (blue) strains are shown. For simplicity, only the impact of changes to the effectiveness of a single drug in a single compartment is shown.

Extended Data Figure 5. Model of replication dynamics and treatment effectiveness in the viral reservoir fitted to the data

The model is fitted to the total inferred average body counts of free virus particles (green line), infected CD4+ T cells (orange line) and virus bound to the follicular dendritic cell network of B cell follicles (grey line). a, Demonstrates the dynamics over the first 200 days of treatment. Note that early on during antiretroviral therapy, HIV-1 RNA in plasma declines more rapidly than virus bound to the follicular dendritic cell network of B cell follicles. Circles demonstrate average data from the 3 patients discussed in detail in this study and an additional 9 patients presented elsewhere. Where the average value was indeterminate because of test sensitivity, the data are fitted below the upper limit of the average log10 infectious units. b, Demonstrates the dynamics over a longer period. The model predicts the persistent low-level viral RNA in plasma. The diamond symbol represents data relating to the long-term persistent virus, as measured using quantitative reverse transcription PCR (see Methods). The optimal model fit parameters are presented in Supplementary Table 1.

Figure 1. Time-structured phyloanatomic history of haplotypes in lymph nodes and blood

MCC phylogenetic trees constructed from the complete alignments of the haplotypes from the Gag region of HIV-1 for subjects 1774, 1727 and 1679 with all haplotypes (a, b and c, respectively) and with the haplotypes containing G-to-A hypermutations removed (d, e and f, respectively). Branch colors represent the most probable (modal) anatomic location of their descendent node inferred through Bayesian reconstruction of the ancestral state, along with the posterior probabilities for the location of their parental nodes.

Figure 2. Cartoon illustration of the drug concentration-dependent spatial model

In the main compartment (i=0; the majority of lymphoid tissue and the blood) drug concentration is high (grey and red). In the sanctuary site (i=1; a small fraction of the lymphoid tissue and localized extracellular fluid) drug concentration is low (pink). There are uninfected cells (Xi), long-lived infected cells (Yi), and short-lived infected cells (Qi), as well as virus particles (Vi) that can be bound by few (Fi) or many (Gi) receptors on the follicular dendritic cell network. The dashed lines represent the effect of treatment in blocking infection and production of infectious virus particles. For graphical simplicity, we do not show the emergence of drug resistance, the production of noninfectious virus particles, virus clearance, nor cell death.

Figure 3. Drug-dependent fitness landscape

Effective reproductive numbers for drug-sensitive (RS, orange line) and partially drug-resistant (RR, blue line) strains are driven by the effective drug concentration of a single drug in the relevant compartment. Grey lines mark thresholds separating three possible outcomes. When effective drug concentrations are low, the benefit of drug resistance does not overcome the fitness cost of mutations and drug-sensitive strains dominate. This ceases to be true at intermediate effective drug concentrations and drug-resistant strains dominate. At high concentrations, both RS and RR fall below one (red line), neither strain can grow and virus replication is halted.

Figure 4. Modelling replication dynamics and treatment effectiveness in the viral reservoir

In the main compartment, antiretroviral therapy is wholly effective against drug-sensitive virus (z0 = z̃0 = 1). In the sanctuary site, a, low treatment effectiveness (z1 = z̃1 = 0.3) favors drug-sensitive strains. Partially drug-resistant strains will exist, but at levels below those favoring stepwise evolution towards a fully drug-resistant strain. b, Intermediate treatment effectiveness (z1 = z̃1 = 0.6), favors partially drug-resistant virus at levels that may suffice for evolution towards a fully drug-resistant strain. c, high treatment effectiveness (z1 = z̃1 = 1) favors the decline of all strains and the cessation of virus replication.