Phylogenetics and Phyloanatomy of HIV/SIV Intra-Host Compartments and Reservoirs: The Key Role of the Central Nervous System

Abstract

Background:

The ability of the human immunodeficiency virus type 1 (HIV-1) to persist in anatomic compartments and cellular reservoirs is a major obstacle for eradication of replicationcompetent virus in the infected host.

Approach:

We extensively review recent advancements in phylogenetic and phylogeographic techniques that provide a unique opportunity for studies of intra-host HIV-1 compartmentalization and the detection of potential reservoirs.

Conclusion:

We show that infected macrophages in the central nervous system (CNS) harbor viral subpopulations that play a key role in the emergence of escape variants and viral rebound following discontinuation of antiretroviral therapy. An HIV cure, therefore, cannot be achieved without the effective targeting of the virus in the CNS, for which in depth knowledge of viral population dynamics contributing to the development and maintenance of latent reservoirs is critical.

1. EVIDENCE OF CELLULAR AND ANATOMICAL HIV-1 RESERVOIRS

1.1. HIV-1 Reservoirs and the Challenge of Eradication

Thirty years after the discovery of HIV-1 as the causative agent of AIDS, a cure has yet to be found. Highly active anti-retroviral therapy (HAART) has significantly reduced AIDS-related morbidity and mortality, with improved regimens that are more potent, have fewer side effects, and lower medicinal burden. Unfortunately, treatment does not fully restore human health but simply prolongs the onset of this destructive disease and, if interrupted, invariably leads to viral rebound even in patients with low-to-undetectable plasma viremia [1–3]. In addition to lifelong treatment, replacing certain drugs within the HAART regimen has also become eminent owing to the emergence of drug resistant viral variants over the course of infection [4]. The identification and quantification of critical sources of lowlevel viremia and viral rebound, as well as the emergence of drug-resistant variants in the face of treatment has been a well-studied area of research since the development of HAART and has expanded with recent advancements in phylogenetic and sequencing techniques. This research has contributed significantly to our understanding of the complexity of HIV-1 infection and replication and, importantly, the ability of the virus to evolve and adapt to both the host immune response and antiretroviral therapy.

HIV-1 latency was initially attributed to proviral DNA in resting memory CD4⁺ T cells [5], which have since been shown to contribute to low-level persistent viremia during HAART, emergence of antiviral escape variants, and viral rebound after treatment interruption [6–14]. These cells do not release infectious virus in the resting state but can do so following cellular activation, which can occur under a variety of conditions [5, 6]. Virus production in latently infected cells, occurring in the absence of new rounds of infection of surrounding cells [15–18], is typical of viral reservoirs, defined as “a cell type or anatomical site in association with which a replication-competent form of the virus persists with more stable kinetics than the main pool of actively replicating virus” [19, 20].

Several studies have indicated that in addition to resting naïve and memory CD4⁺ T cells, viral reservoirs include, but may not be limited to, a recently discovered immature memory T cell population with stem-cell like properties, termed CD4⁺ T memory stem cells [21–23], as well as monocytes/macrophages [24], tumor-associated macrophages [25], and astrocytes [24]. Follicular dendritic cells are an additional, unique reservoir in that they contribute to viral persistence by trapping virus particles on the cell surface, rather than harboring latent provirus, and this interaction can remain stable for as long as nine months [26, 27]. Another potential reservoir is gut mucosa, although its contribution to viral rebound after treatment interruption seems to be limited [28]. On the other hand, persistence of HIV-1-infection in the spleen [29] has been shown to contribute to viral rebound in patients with incomplete suppression by HAART [9]. Moreover, the ability of the virus to switch from the use of the CCR5 co-receptor for entry to CXCR4 in ~50% of patients [30] has been attributed to increasing replication in resting, naïve CD4⁺ T cells [31] and subsets of CXCR4-expressing macrophages [32], although higher affinity for the CD4 receptor has specifically been associated with macrophage tropism, as macrophages display a lower CD4 surface density [20]. The timing of the switch from CCR5 to CXCR4 is particularly important for CCR5 antagonist monotherapies used in developing countries, as the re-emergence of pre-treatment CXCR4-using variants has been reported [33].

It has been shown that viral sequences obtained from circulating CD4⁺ T cells during low-level viremia and viral rebound post-therapy withdrawal are phylogenetically distinct from circulating cell-free virus in plasma [34], as well as viral sequences from monocytes [35]. Such findings not only implicate monocytes and/or additional reservoir cell populations in the low-level production of virus but also indicate that viral reservoirs are characterized by specific evolutionary dynamics due to their ability to remain hidden from both synthetic and host-mediated antiviral response.

1.2. HIV-1 Compartmentalization and Metapopulation Structure

Although difficult to sample in humans, several anatomic “compartments” have been identified that can harbor tissue-specific HIV-1 subpopulations phylogenetically distinct from those circulating in peripheral plasma or other tissues [36]. According to the accepted definition, an anatomic compartment is “a site for which there is limited exchange of viral genetic information with other sites” [20]. Moreover, if the compartment is characterized by suboptimal free drug concentrations, it is referred to as an anatomical sanctuary [20, 37, 38]. Since the first demonstration that genetic differences exist between blood- and brain-derived viral sequences [39], a large proportion of viral compartment studies have focused on the central nervous system (CNS). The CNS is often considered an “immune privileged” site due to its low-density T cell population and weak adaptive immune response attributed to the blood brain barrier (BBB) and blood cerebrospinal fluid (CSF) barrier, which restrict migration of certain cells and other materials [40]. This also renders the CNS a potential sanctuary [41], where poor drug penetration owing to tightly regulated anatomical barriers plays a major role in limiting drug efficacy [42].

CNS compartmentalization is likely the result of founder effects, either during primary or late infection, and tissue-specific selective pressures that shape distinct viral populations, followed by restricted viral trafficking [7, 43–49]. HIV-1 production in the CNS mainly occurs in perivascular macrophages and macrophage-like microglia [50, 51], and has been associated with both the emergence of drug resistance [52, 53] and AIDS-related neuropathology [54, 55], a spectrum of disorders collectively known as HIV-associated neurocognitive disorders (HAND), ranging from asymptomatic to HIV-associated dementia (HAD) [56]. Subjects with and without HAD often harbor viral populations in the CSF that are genetically distinct from virus in the blood [57–62] and exhibit characteristics of macrophage/microglia tropism [63], in which case both latency characterized by low-level replication and anatomical isolation may contribute to the formation of a reservoir during primary infection [64, 65].

The compartmentalization of the virus in the CNS exemplifies HIV-1 metapopulation structure (i.e. division in distinct sub-populations) within different cells and anatomic sites of the infected host. Indeed, tissue-specific viral populations have been detected within female genital tract [44, 66–68], male foreskin [69] and semen [70]. Latent viral forms associated with compartmentalization within the female genital tract and/or semen and foreskin of the male may be important in transmission, since they appear to be preferentially transmitted over more recent variants [71–74]. Limited compartmentalization has also been observed in breast milk [75–77], gut mucosa [78], and lung tissues [79]. The existence of distinct intra-host viral subpopulations has been confirmed by studies using the SIV-macaque model, exhibit compartmentalization in spleen and secondary lymph nodes during primary and early chronic infection [80] as well as in the genital tract [81].

2. PHYLOGENETIC METHODS OF DETERMINING VIRAL COMPARTMENTS AND POTENTIAL RESERVOIRS

2.1. Testing Viral Compartmentalization

The phenomenon of HIV-1 compartmentalization that can ultimately give rise to a reservoir is probably the result of local conditions affecting viral evolution at specific anatomical sites limiting viral trafficking and/or exerting pressures that alter the fitness landscape of the viral population [82, 83]. Compartmentalization can be inferred in a phylogenetic tree of viral sequences amplified from different tissues and/or cell types of an infected subject by the presence of a separate monophyletic clade including all or most of the sequences from a specific site [84]. In addition, sequences sampled from a viral reservoir are expected to lack temporal structure – i.e. sequences sampled at different time points during the infection will intermix since they persist in the reservoir – and show less mean divergence from the most recent common ancestor (MRCA), i.e. the root node of the tree, than other sequences in the tree [36]. Confidence for the monophyletic clades is usually assessed through the use of bootstrapping [85] or Bayesian posterior probabilities estimated by Markov chain Monte Carlo (MCMC) methods [86–88]. Bootstrapping is a technique based on the evaluation of distance based (e.g. neighbor-joining) or maximum likelihood trees inferred from alignment replicates, usually 500 to 1000, built by randomly sampling (with replacement) sites from the original alignment. A bootstrap value represents the proportion of trees in which a particular group of sequences clustered together, with values >80% considered as a significant support for the cluster. Alternatively, Bayesian posterior probabilities are inferred from a posterior distribution of possible phylogenies obtained using an MCMC algorithm that incorporates prior knowledge of evolutionary model parameters [89]. Monophyletic clades present in >90% of the posterior distribution of trees are considered significantly supported.

Phylogenetic analysis of HIV/SIV sequences sampled from post mortem brain tissues and longitudinal plasma samples and/or peripheral tissues has shown a clear-cut example of compartmentalization in the CNS, wherein all brain-derived sequences cluster within a distinct and well-supported monophyletic clade (Fig. 1) [48, 54, 55, 90]. However, real data sets may also display partial compartmentalization, characterized by a certain degree of intermixing between sequences obtained from different cell or tissue types, such as, for example, sequences derived from different anatomic sites in the brain (Fig. 1). Therefore, quantitative methods for assessing compartmentalization have been developed that take into account both intracompartmental genetic distances and distances from sequences outside of the proposed compartment (refer to Zárate et al., 2007 [91] for a detailed review). The most commonly used methods can be divided into two main categories: (1) tree-based and (2) distance-based methods. Three well-known tree-based methods are the SlatkinMaddison (SM) test [92], Tree Correlation Coefficients (TCC) [93], and Simmonds Association Index (SAI) [94]. The SM test determines the minimum number of migration events between pre-assigned compartments (i.e. group of viral sequences amplified from different sites) consistent with the topology of the phylogenetic tree and based on the maximum parsimonious reconstruction of the tissue/cell type of origin of ancestral sequences (i.e. the internal nodes of the tree). Statistical support is based on the comparison of observed migrations with a null distribution of randomly intermixed sequences (panmictic population). Significantly less migrations than 99.9% of the null distribution indicate compartmentalization. TCCs (r, and r_b) represent correlated distances between two sequences in a phylogenetic tree with the information about whether or not they were isolated from the same compartment. The distance between two sequences can be either the number of tree branches separating the sequences (r_b) or the cumulative genetic distance between the sequences (r). Statistical significance is achieved by estimating the distribution of these coefficients by permuting sequences between compartments. The SAI assesses the degree of population structure in the phylogenetic tree by weighting the contribution of each internal node based on its depth in the tree and evaluating the significance of the observed value using a bootstrap sample both over the population structure and tree topology. Distance-based methods include Wright’s measure of population subdivision (F_ST) [44, 95, 96], nearest neighbor statistic (S_nn) [97], and analysis of molecular variance (AMOVA) [98]. F_ST compares the mean pairwise genetic distance between two sequences sampled from different pre-assigned compartments to the mean distance between sequences sampled from the same compartment, in which statistical significance is derived via population-structure randomization. S_nn is simply a measure of how often the nearest neighbors of each sequence are isolated from the same or different assigned compartments. AMOVA calculates an association based on the genetic diversity of the sequences between and within compartments and is an extension of F_ST, in which the distances are restricted to Euclidean and the variability is calculated from the sum of the squared distances between the sequences.

Fig. (1). HIV-1 maximum likelihood (ML) tree of brain and peripheral tissue derived sequences from a patient with HIV-associated dementia.

HIV-1 gp120 sequences were amplified from tissues sampled at necropsy and are a subset of the sequences described in Salemi et al. (2005) [54]. The ML tree was inferred with the HKY+G model. Values along the branches represent bootstrap support (500 replicates); only values >80% are shown. Branch lengths are scaled in nucleotide substitutions per site according to the scale bar at the bottom. Sequences from different brain (frontal lobe, spinal chord, occipital lobe) and peripheral (lymph node, seminal vesicle) tissues are indicated in blue and red, respectively, and by different symbols according to the legend in the figure. Brain sequences are completely compartmentalized (bootstrap support 100%). Sub-compartmentalization of individual CNS as well as peripheral tissues is also evident, although not all supported by bootstrapping.

Comparison of different compartmentalization tests based on 92 separate HIV-1 data sets and 1,500 simulated data sets not surprisingly resulted in disagreement between tree-based and distance-based methods [91]. Tree-based methods are expected to be more robust, since they include phylogenetic information otherwise ignored. However, they can also be misleading if branches are unresolved and have low support, since in such cases they lead to an overestimation of topological distance, or degree of segregation, between sequences. Furthermore, each method can be affected by sampling bias, for example underrepresentation of sequences from a particular compartment, which is often the case for difficult-to-sample anatomical locations prior to autopsy, such as the brain, or sites/cell populations with low-level replication, such as peripheral monocytes. The largest change (>70%) in the proportion of sequences classified as compartmentalized with skewed sampled size was observed for the TCC and SM tests, with SAI next in line (56%), indicating an increased effect of sampling underrepresentation on the power of tree-based over distance-based compartmentalization detection methods. Also, as with other phylogenetic analyses, recombination can greatly bias the results. Introduction of recombinant sequences results in reduced power, particularly for tree-based methods, to detect compartmentalization due to increased intermixing between clades of the tree, thereby increasing phylogenetic uncertainty. Finally, since the development and improvement of sequencing methods using single genome amplification (SGA) [99–103], comparisons of SGA with sequencing methods that rely on earlier PCR/cloning techniques have shown that the latter can introduce significant errors that influence the degree of diversity and thus compartmentalization [100, 104], although the general validity of this conclusion has been questioned [105]. In summary, given the potential limitations of each method, compartmentalization should always be tested using more than one algorithm and discrepancies should be evaluated by taking into account the specific assumptions of each test in relation with the actual data set under investigation.

2.2. Phyloanatomy

As discussed above, restricted exchange of genetic information with other sites is the defining property of a viral compartment. Recently, Bayesian phylogeography methods, originally developed to study the spread of viral epidemics [106, 107], have been successfully applied to the study of viral trafficking between different anatomical sites and cell populations, also referred to as viral gene flow or phyloanatomy [49, 108a]. Phyloanatomy can provide significant insights into the establishment and maintenance of potential viral reservoirs, and it is based on the incorporation of spatial information (sampled tissue) into the evolutionary analysis of an organism that is evolving according to a population structure, or coalescent, model. Earlier implementations of phyloanatomy based on maximum parsimony have also been utilized for the investigation of HIV-1 intra-host evolution [48], and have been described elsewhere [108b-110], but are not the focus of this review.

By using the Bayesian phylogeography framework implemented in the BEAST software package [111, 112], information on intra-host migration rates, origin and timing of specific migration events, as well as the number of migrations (“Markov jumps”) and time spent in a given location (“Markov rewards”) can be inferred from an alignment of heterochronous sequences from various anatomical locations [113]. Such migrations, supported in the form of Bayesian posterior probabilities, are interpreted as the probability that sequences from one anatomical location gave rise to, or shared ancestral history with, sequences in another location at a given point in time. Although seemingly complex, the use of the continuous-time Markov Chain (CTMC) model to trace transitions between discrete (or continuous) geographical traits along branches of a phylogenetic tree is analogous to the use of the general time-reversible nucleotide substitution model of transitions between nucleotide character traits along the same tree, both of which are inherent viral evolutionary processes naturally embedded in the genealogy.

Unfortunately, like other epidemiological studies, even for a moderately small number of spatial locations, most migration events might not even be sampled. Therefore, recent improvements of the Bayesian phylogeographic methodology, using hierarchical modeling, have incorporated joint analysis of paralleled migration processes, such as similar data sets from multiple patients, for a more statistical approach to gene flow analysis. Referred to as the Bayesian hierarchical phylogenetic model (HPM), this framework allows for a more accurate interpretation of summary data [114] than would be possible for consensus [115, 116] or combined [117, 118] approaches. Essentially, viral sequence data from multiple subjects can be used in a single analysis, akin to a strict combined approach, while simultaneously allowing for different phylogenetic parameters in the individual paralleled partitions, e.g. patients, as in the consensus approach. Results from the individual partitions are combined to provide overall summaries of each parameter, such as migration rates. The overall model and individual partition models are fitted simultaneously, enabling the overall model to feed back information, in the form of a prior, into the estimation of the individual paralleled partitions. This feedback mechanism results in a more precise estimate of each evolutionary parameter. Details of the application of HPM methodology to HIV-1 phylogeography and phyloanatomy are outlined in Cybis et al. (2013) [107]. Although the study assessed viral gene flow only between CD8+ and CD4+ T cell compartments, the application of this statistical method to phyloanatomy has the potential to greatly impact future investigations of HIV-1 compartmentalization and reservoir dynamics.

2.3. Analyzing the Temporal Structure of Longitudinally Sampled Viral Sequences

Compartmentalization and restricted viral migration are necessary but not sufficient conditions to infer the presence of a viral reservoir. The flow of virus and/or infected cells between compartments can mask low-level viral persistence, as is the case for the memory CD4+ T-cell population in lymphoid tissues during chronic infection [11, 80]. Mannioui et al. (2009) used molecular biology techniques to detect in vivo viral replication during low-level viremia, represented by the presence of episomal DNA, indicative of recent infection [119] and, thus, persistent viral replication within a population. However, evidence is also needed of the contribution of a potential reservoir to the reemergence of archival variants [120] and/or viral rebound that could be determined based on shared ancestry between sequences during rebound and sequences obtained prior to treatment withdrawal. From a phylogenetic perspective, the clustering pattern of a viral reservoir would be characterized by lack of significant cumulative genetic divergence between sequences sampled at different time points [36]. To which extent real phylogenies differ from perfectly temporally structured trees – representing the progressive increase in viral diversity and divergence from the MRCA during the course of the infection [121, 122] – can be accurately quantified by the temporal structure statistic, defined between 0 (no temporal structure) and 1(perfect temporal structure) [123]. In particular, the clustering of sequences sampled late in infection with earlier ones would reduce temporal structure and indicate the reemergence of latent viruses (Fig. 2). Similar to the previously mentioned SM test of compartmentalization, topological temporal structure can be evaluated by measuring the extent of clustering whereby sequences are categorized as discrete states corresponding to sampling time, rather than anatomical location [123]. The number of state changes within the inferred phylogeny is then compared to that of a null distribution to assess statistical significance.

Fig. (2). Bayesian maximum clade credibility (MCC) tree highlighting reduced intra-host temporal structure.

The coalescent framework implemented in BEAST [111, 112] was used for a serially sampled SIV-infected macaque dataset to obtain a posterior distribution of trees scaled in substitutions/site, from which the MCC tree was derived using Tree Annotator [112]. Branches are colored according to sampling time point in days post-infection (dpi), with temporally clustered sequences (>3) collapsed for illustrative purposes. Three separate instances of shared ancestry between sequences sampled relatively far apart (92 and 223 dpi) have been highlighted, with corresponding clade posterior probabilities reported to the right for which the temporal states have been randomized. The final result is referred to as the temporal clustering statistic, ranging from 0 (absence of temporal structure) and 1 (perfect temporal structure).

2.4. Intra-Host Evolutionary Rate Variation

In addition to temporal structure, analyzing evolutionary rate variation among phylogenetic lineages can be used to investigate temporal signal. In the case of latent virus reactivation, shared direct ancestry (clustering) between sequences sampled during early and late infection will significantly reduce the evolutionary rate during that period of time [84, 124]. In contrast, migration of virus between anatomical sanctuaries may increase the evolutionary rate. Both incidences act to reduce the ‘clock-like’ signal associated with the molecular clock hypothesis, which assumes a linear increase in the number of substitutions with time along all phylogenetic lineages [125]. Temporal signal can be quantified using a simple regression of root-to-tip genetic distances in the phylogeny against sampling time (Fig. 3) with the program Path-O-Gen (v1.3; available from http://tree.bio.ed.ac.uk/). Path-O-Gen-derived R² values indicate ‘clock-likeness’ of the sequence data, wherein high R² values (>0.5) are indicative of sequences that evolve under a strict or relaxed molecular clock.

Fig. (3). Result of temporal signal analysis in Path-O-Gen.

(A) A maximum likelihood tree was inferred using RaxML for serially sampled sequences from a SIV-infected macaque dataset. Branches are colored according to sampling time, with interior branches treated as temporal ancestral states inferred using maximum parsimony in MacClade (available from http://macclade.org). (B) A linear regression model was used to measure the temporal signal, or “clocklikeness,” in Path-O-Gen for root-to-tip genetic divergence (y-axis) for each sequence (yellow dot) in the phylogeny above against sampling time (x-axis). Colored boxes again correspond to sampling time for visual comparison.

Immomen & Leitner (2014) have recently developed a method to quantify the impact of potential reservoirs on viral evolutionary rates by identifying lineages within a serially sampled phylogeny that evolve at a significantly lower rate than at least one other lineage in the tree [126]. The method assumes that HIV mutates according to a Poisson process with a single molecular clock rate when replicating but does not mutate when latent. For sequence pairs that share a common ancestry, latency is calculated based on the Poisson probability of accumulating the number of mutations observed on the shorter lineage d_Short, given the evolutionary rate of the longer lineage d_Long. Thus, latency is defined when the shorter lineage provides a Pr[(λk)/(k!e^-λ)] ≤ 0.05, where λ=d_Long*L and k<=d_Short*L for sequence length L. However, because hypermutation, recombination, and selection can have substantial effect on evolutionary rates, sequences experiencing these evolutionary events should be removed prior to analysis and only 3^rd codon positions analyzed.

3. RECENT EVIDENCE OF CNS AS A KEY VIRAL RESERVOIR

Each of the phylogenetic methods described above is not sufficient alone to identify with certainty a viral reservoir. Additionally, over- or under-sampling of sequences from specific cell types or tissues, as well as the number and time intervals of available longitudinal sequences, can greatly influence the results. Experimental design and optimization of sampling schemes is crucial for any phylogeny-based investigation of viral reservoirs [124]. However, when properly designed phylogenetic analyses are used in combination to and complemented with studies of viral replication kinetics within specific tissues and/or cell populations, they can be powerful tools for the advancement of reservoir investigation and drug treatment strategies.

Such an approach, applied to the studies of HIV/SIV brain infection and neuroAIDS, has recently provided several lines of evidence that the CNS, along with its cellular components, are a key viral reservoir, actively contributing to both the re-seeding of peripheral virus and emergence of drug resistance. For example, parenchymal-derived sequences from patients diagnosed with HAD display reduced evolutionary rates [55], which may be explained by the low-level persistent expression of unintegrated viral DNA in CNS-resident macrophages for prolonged periods of time (at least 30 days) [127]. Levels of unintegrated viral DNA relative to the integrated form are 10-fold elevated in brain of patents with HAD [128, 129], and CNS-resident macrophage populations that harbor latent, integrated forms of viral DNA (perivascular macrophages and microglia) have relatively low turnover rates, ranging from months to years. As previously mentioned, anatomical barriers such as the BBB and blood-CSF barrier inhibit viral gene flow and drug penetration, resulting in isolated, or compartmentalized, viral replication. Viral compartmentalization in the CNS may be a critical phenomenon from the standpoint of neurological impairment, but migration of the virus, although limited, in and out of this compartment is of more importance when considering the reservoir potential of the CNS. Clear evidence of reseeding peripheral tissues by CNS-derived viral variants was found for HIV-1-infected patients, as it can be inferred from the emergence of peripheral sequences from CNS-associated ancestral nodes in Bayesian genealogical reconstructions (Fig. 4) [55]. Furthermore, use of the well-characterized SIV_mac251-infected rhesus macaque model of neuroAIDS demonstrated that this recirculation was not simply limited to terminal illness but can also happen during early stages of infection (Fig. 4) [49]. Evolution of drug resistance has also been identified in the CNS compartment of treated patients [51, 53], which poses a significant challenge to treatment optimization, since specific antiretroviral regimens are based on resistance testing of circulating plasma virus.

Fig. (4). Bayesian maximum clade credibility (MCC) trees with highlighted viral migration events from brain into circulation and estimated ancestral state posterior probabilities using phyloanatomy algorithms.

The phylogeographic framework implemented in the BEAST package [111, 112] was used to obtain a posterior distribution of trees from which MCC trees were derived using Tree Annotator [112]. The trees in the figure include sequences from an HIV-1-infected patient (analyzed in Lamers et al. (2010) [55] and reanalyzed here with the Bayesian phylogeography framework (left panel)) and sequences from two SIV-infected CD8+ lymphocyte-depleted macaques (middle and right panels) [49]. Branches are colored according to the sampled tissue/cell location (exterior) or highest posterior probability location of the reconstructed ancestral sequence (interior), with blue representing brain-derived sequences and red representing sequences derived from peripheral tissues or blood. Peripheral blood mononuclear cells (PBMCs) and cell populations in the SIV trees refer to flow cytometry-sorted CD3+ lymphocyte and CD14+ monocyte cell populations. Collapsed branches represent monophyletic clades consisting of sequences from the same tissue/cell type and sampling time point. Cross-sectional sequences from the HIV-1-infected patient (left panel) and one of the SIV-infected, CD8+ lymphocyte-depleted macaques (right panel) were taken at autopsy/necropsy, the time in days post-infection (dpi) depicted for the macaque. Longitudinal sequences were derived from the macaque depicted in the middle panel prior to necropsy following euthanization due to the onset of SAIDS [49]. Stars represent inferred viral entry into the periphery from the brain and are located at the midpoint of the branches along which these events occur. The posterior probability, expressed as a percent of total posterior probability, of the ancestral state at the preceding node is reported at the bottom for each migration event. Ancestral state posterior probabilities can be observed using FigTree (available from http://tree.bio.ed.ac.uk/software/figtree/).

Based on the accessibility of CSF relative to parenchymal brain tissue, researchers have proposed that sampling of viral sequences from the CSF could provide genetic information on viral populations in the brain reservoir as early as primary infection [64, 65]. However, other studies have indicated that CSF viral sequences are not representative of the viral population dynamics observed in the parenchyma [130, 131], wherein individual parenchymal lobe tissue may harbor distinct viral sub-populations [48, 54]. Moreover, extensive sequence analyses in SIV-infected macaques have identified heterogeneous populations within the CNS, defined also by genetically distinct subcompartments, such as meninges and parenchyma [49], basal ganglia, cerebellum, and hippocampus [132] (Fig. 1). Although subpopulations of virus found in the CSF are likely derived from regions within the brain, trafficking of virus between brain parenchyma compartments and peripheral blood is not well understood. This lack of knowledge, together with evidence of the complexity of viral genetic composition within the brain not accurately mirrored in the CSF underlines the significant limitation of using sequences from the CSF to characterize the CNS reservoir to a sufficient depth or pre-screen for drug-resistant variants within tissue sanctuaries. Therefore, although phylogenetic methods have identified the CNS as a key viral reservoir and potentially crucial barrier to treatment and eradication, more studies are urgently needed to improve our understanding of the contribution of individual CNS components in HIV reservoir establishment and maintenance.

4. FUTURE DIRECTIONS

The past few years have witnessed significant advances in the development of novel and powerful phylogenetic methods applicable to the study of intra-host viral compartments and reservoirs of fast evolving viruses, such as SIV/HIV [25, 48, 49, 54] and HCV [133, 134], and made clear the central role played by CNS infection in HIV persistence and rebound. The use of such analysis tools in AIDS research, however, is limited by obvious ethical restraints. HIV evolutionary dynamics of deep tissues are extremely difficult to investigate in vivo since the required sampling often impractical, when not unethical. In addition, most of the findings mentioned in this review highlight the importance of investigating the establishment of viral reservoirs during primary infection, which is difficult to detect in humans. On the other hand, non-human primate models provide a viable alternative. SIV infection of rhesus macaques (Macaca mulatta) is one of the most widely used models for HIV-related pathogenesis studies [135]. Infected macaques exhibit similar clinical manifestations as HIV-1 infection, albeit on a shorter time scale of approximately one to three years [136]. More importantly, they offer the unique opportunity for a wide range of infected tissue and cell population sampling that is critical to HIV-1 reservoir studies and can be performed in the presence or absence of treatment. Much can, and still needs to be, learned from this model with respect to viral latency and the contribution of viral evolutionary and population dynamics to reservoir formation.