• We are sorry, but NCBI web applications do not support your browser and may not function properly. More information
Logo of jvirolPermissionsJournals.ASM.orgJournalJV ArticleJournal InfoAuthorsReviewers
J Virol. Jun 2007; 81(12): 6643–6651.
Published online Apr 11, 2007. doi:  10.1128/JVI.02268-06
PMCID: PMC1900087

Comparative Study of Methods for Detecting Sequence Compartmentalization in Human Immunodeficiency Virus Type 1[down-pointing small open triangle]

Abstract

Human immunodeficiency virus (HIV) infects different organs and tissues. During these infection events, subpopulations of HIV type 1 (HIV-1) develop and, if viral trafficking is restricted between subpopulations, the viruses can follow independent evolutionary histories, i.e., become compartmentalized. This phenomenon is usually detected via comparative sequence analysis and has been reported for viruses isolated from the central nervous system (CNS) and the genital tract. Several approaches have been proposed to study the compartmentalization of HIV sequences, but to date, no rigorous comparison of the most commonly employed methods has been made. In this study, we systematically compared inferences made by six different methods for detecting compartmentalization based on three data sets: (i) a sample of 45 patients with sequences gathered from the CNS, (ii) sequences from the female genital tract of 18 patients, and (iii) a set of simulated sequences. We found that different methods often reached contradictory conclusions. Methods based on the topology of a phylogenetic tree derived from clonal sequences were generally more sensitive in detecting compartmentalization than those that relied solely upon pairwise genetic distances between sequences. However, as the branching structure in a phylogenetic tree is often uncertain, especially for short, low-diversity, or recombinant sequences, tree-based approaches may need to be modified to take phylogenetic uncertainty into account. Given the frequently discordant predictions of different methods and the strengths and weaknesses of each particular methodology, we recommend that a suite of several approaches be used for reliable inference of compartmentalized population structure.

Understanding intrapatient human immunodeficiency virus (HIV) dynamics and evolution is crucial to the design of more efficient therapies for infected individuals. However, the processes shaping viral populations are complex and poorly understood. The evolution of HIV is strongly affected by a number of viral and host factors. During early infection, the virus colonizes different organs within the body and can form somewhat separate viral populations, driven to adapt to their particular environments and subjected to different selective pressures (22, 44). If trafficking and gene flow between viral subpopulations is significantly restricted, then each subpopulation can become genetically distinct from others, i.e., compartmentalized. Compartmentalization has been defined in different ways, for example, as genetic heterogeneity between subpopulations (1), as the result of independent micro-evolution (8), as the result of restricted viral gene flow (39), or as the presence of distinct but phylogenetically related genotypes (43). Various mechanisms can contribute to the compartmentalization of HIV populations. The high rate of mutation of HIV in vivo (53, 56) can rapidly increase the molecular distance between subpopulations. Differences in selective pressures imposed by the immune system and disparities in local concentrations of antiviral drugs can result in divergent evolution of the virus (58, 61). Furthermore, compartmentalized viral populations have been shown to possess distinct phenotypic characteristics, such as cellular tropism (68), drug resistance (58, 64, 69), and level of pathogenesis (14).

HIV type 1 (HIV-1) populations isolated from the genital tract in both men and women have been reported as compartmentalized (12, 29, 43, 45, 47, 48, 65, 73) when compared to the viruses isolated from the blood or lymphoid tissue. Since the most common route of HIV transmission worldwide is genital exposure (52), the study of HIV populations replicating in the genital tract will help us develop strategies to prevent transmission. Several studies reported that the rate of transmission via genital exposure is related to the plasma viral load and the CD4+ cell count in the source, as well as to the stage of the infection (21, 49, 71), suggesting a role for the immune system in the transmission and evolution of HIV populations replicating in the genital tract. Moreover, differences in divergence and variability of the sequences (15), pattern of drug resistance (58), and coreceptor usage (29) have been reported for viruses isolated from the genital tract compared to those obtained from the bloodstream.

Different tissues in the central nervous system (CNS) can harbor distinct viral populations (1, 8, 20, 27, 32, 35, 38, 41, 42, 46, 51, 54, 57, 61, 62, 63, 64, 68, 69, 70). Samples of viral populations are collected either by examining brain tissue from infected individuals post mortem or by drawing samples from the cerebrospinal fluid (CSF) in HIV-positive patients. The CNS offers a unique environment for HIV replication, because the presence of the blood-brain barrier (BBB) and the blood-CSF barrier restricts viral trafficking between the bloodstream and the CNS, giving rise to at least two segregated HIV populations. In addition, antiretroviral drugs have different levels of permeability through the BBB and the blood-CSF barrier that largely depend on their biochemical characteristics, including lipo-solubility and molecular weight (16). For example, some nucleoside analogs can cross the BBB and hamper the replication of HIV in the CNS, whereas protease inhibitors are pumped out by P-glycoproteins present in the BBB (30) and their ability to reach the viruses replicating in the CNS may be impaired. However, in both cases suboptimal inhibitor concentrations are attained. Furthermore, because the immune response in the CNS includes microglial and T cells as well as macrophage/monocytes (31, 72), changes in cellular tropism of the virus from CCR5 to CRCX4 receptors can confer a selective advantage. Such shifts in receptor usage have been correlated with the rise of neurovirulent viruses in the CNS and development of AIDS-dementia complex (4, 13, 35, 51, 62, 68).

Regrettably, analytical methods used to evaluate the degree of compartmentalization among viral populations lack consistency in rigor and selection of procedure across studies. In some cases, the mere observation that sequences obtained from the same compartment clustered together in a phylogenetic tree has been interpreted as evidence of compartmentalization, whereas others have relied upon more formal statistical approaches, such as the Slatkin-Maddison (SM) test (60). To our knowledge, there has been no attempt to compare the performance of different methods available for testing compartmentalization in HIV sequences and to investigate when and to what degree the methods agree. In this study we examined published HIV sequences isolated from different compartments within the same patient, focusing on sequences derived from the CNS and the female genital tract, and used six previously published methods to detect compartmentalization. For the purposes of our study we divided the methods in two categories: those which used a phylogenetic tree to detect compartmentalization (tree based) and those based on pairwise genetic distances between viral clones (distance based). Based upon 92 biological data sets obtained from GenBank and 1,500 simulated data sets, we concluded that tree-based methods were more sensitive in detecting compartmentalization than distance-based methods. However, to guard against false positives due to the uncertainty in phylogenetic reconstruction, distance-based methods should also be taken into account.

MATERIALS AND METHODS

Sequence data.

We downloaded all available (as of February 2006) HIV sequences isolated from the CNS and the female genital tract from GenBank. This sample contained 3,236 sequences from the CNS and 1,393 from the female genital tract. As a selection criterion for inclusion in this study, we required that sequences from at least two compartments with at least five sequences each be available from the same patient. Samples from 62 patients met these criteria and were included in the analyses. Aligned sequences and phylogenetic trees used in this study can be downloaded from http://www.hyphy.org/pubs/zarate_comp.tgz. The number of patients used in this work, the sources of the virus, the viral genes sequenced, and the medians of the diversity and of the length of the alignment, organized by publication, are shown in Table Table11.

TABLE 1.
Sequences used for comparing compartmentalization detection methods

CNS data.

We downloaded viral clones from 44 patients (1, 20, 27, 32, 38, 54, 57, 62, 64). Of those, 42 had developed some level of neuropsychiatric and/or neuropathologic diagnosis specifically related to HIV infection. Thirty-one patients had received some antiretroviral treatment, and 10 of those were on therapy at the time of death or sampling. A total of 1,958 sequences from CNS and blood (with a median of 11 and a range of 5 to 27 sequences per patient per compartment) were grouped in 62 data sets by the source patient and the gene or gene region sequenced.

Female genital tract data.

A total of 584 sequences (median of 7 and range of 5 to 11 sequences per patient per compartment) derived from 18 patients (29, 43, 48) were grouped in a total of 30 data sets by the source patient and the gene sequenced. Fourteen women included in this study were clinically asymptomatic, in 14 cases the infection risk was genital exposure, and 4 subjects had been treated with antiretrovirals.

Data analyses.

We utilized three tree-based methods and three distance-based methods, briefly described below.

(i) Slatkin-Maddison test.

The SM test (tree based) (60) determines the minimum number of migration events between the separated populations consistent with the structure of the reconstructed phylogenetic tree. Statistical support is based on the number of migration events that would be expected in a randomly structured population, derived by permuting sequences between compartments.

(ii) Simmonds association index.

The Simmonds association index (AI; tree based) (70) 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 (progressively less for nodes near the root) and evaluating the significance of the observed value using a bootstrap sample both over the structure of the population and the shape of the phylogenetic tree.

(iii) Correlation coefficients.

Correlation coefficients (r, rb; tree based) (11) are a way to correlate 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 (rb) or the cumulative genetic distance between the sequences (r). To assess whether the computed coefficient was statistically significant, we estimated the distribution of these coefficients by permuting sequences between compartments. A P value of 0.05 or less was considered statistically significant.

(iv) Wright's measure of population subdivision.

Wright's measure of population subdivision (FST; distance based) (24, 25, 59) compares the mean pairwise genetic distance between two sequences sampled from different compartments to the mean distance between sequences sampled from the same compartment. Statistical significance is derived via a population-structure randomization test. We calculated this score using three different approaches, two estimates of FST (25, 59) and an estimate of KST (24), and did not observe any differences in the statistical significance of the results. For our analyses, the distance matrices were calculated using the TN93 genetic distance (66).

(v) Nearest-neighbor statistic.

The nearest-neighbor statistic (Snn; distance based) (26) is a measure of how often the nearest neighbors of each sequence were isolated from the same or different compartments. The distance between sequences is measured using the TN93 metric (66) (not the number of sites in which two sequences differ, as in the original description).

(vi) AMOVA.

Analysis of molecular variance (AMOVA; distance based) (18) calculates an association based on the genetic diversity of the sequences between and within compartments. AMOVA is an extension of Wright's F statistics, in which the distances are restricted to Euclidean and the variability is calculated from the sum of the squared distances between the sequences.

SM, AI, FST, Snn, and correlation indices tests were reimplemented and run using HyPhy (33). AMOVA was carried out using the package ADE4 (67).

Simulations.

Using sequence data simulated with Serial Simcoal (3, 17), we studied the effects of the migration rate and the sample size on the abilities of the methods to detect compartmentalization. Based on two subpopulations with an effective population size of 5,000 each and 15 migration rates ranging from 0.00005 to 0.1 migrations per generation, we drew a random sample of 20 sequences per compartment. One hundred replicates were simulated for a given value of the migration rate. We also ran a series of simulations with the migration rate fixed at 0.0005 migrations per generation and set the sample size at 5, 10, 20, and 50 sequences, with 100 replicates for a fixed sample size. In addition, we ran simulations with a fixed migration rate of 0.0005 migrations per generation drawing a sample of 5 sequences from one compartment and 20 from the other. As a control, we simulated 100 data sets evolved over random trees using the program Seq-Gen (50). The sequences were generated using the substitution model HKY85 (23), equal nucleotide frequencies, and a transition/transversion ratio of 2. These data sets consisted of 40 sequences of 500 nucleotides in length, which were equally divided randomly between two compartments.

Comparisons between methods.

In order to quantify the level of agreement obtained with the different approaches used in this study, we calculated the κ score (10) for each pair of methods. Briefly, if two procedures are used independently to produce a dichotomous (yes/no) classification of N observations, then κ = (p0pe)/(1pe), where p0 is the proportion of times the methods agree and pe estimates the probability that the two independent methods agree by chance. If fyy is the proportion of N cases which both tests rate as “yes,” fnn is the proportion of those which both methods rate as “no,” and fyn/fny is the proportion of discordant tests, then p0 = fyy + fnn and pe = (fyy + fyn)(fyy + fny) + (fnn + fny)(fnn + fyn). κ ranges between −1 and 1, with positive values indicating more agreement than expected by chance. We used a simple qualitative scale (2) to interpret the level of agreement: κ < 0.2, poor; 0.2 ≤ κ < 0.4, fair; 0.4 ≤ κ < 0.6, moderate; 0.6 ≤ κ < 0.8, good; 0.8 ≤ κ ≤ 1.0, excellent.

RESULTS

We begin by presenting side-by-side comparisons of compartmentalization classification decisions made by the different methods.

Tree-based versus distance-based methods.

For the 62 CNS data sets, SM and FST classified 38 data sets as compartmentalized and 14 as noncompartmentalized. Hence, the SM test and FST were in good agreement, based on the κ score of 0.62. When applied to the female genital tract data, both methods agreed in 22/30 cases, finding compartmentalization in 18 data sets and rejecting compartmentalization for 4 data sets. The κ statistic value of 0.35 indicated fair agreement. When the results from the SM analysis were compared to those obtained with AMOVA, we found poor agreement between the methods. The κ score for the SM test and AMOVA in the CNS was 0.09, and in the genital tract it was 0.14. Meanwhile, the AI and FST comparison yielded κ scores of 0.56 (CNS, moderate agreement) and 0.29 (genital tract, fair). For the CNS data, the degrees of agreement between SM or AI and Snn were similar to those between SM or AI and FST. However, for the female genital tract data, Snn had higher levels of agreement with the tree-based methods than the other distance methods.

The κ scores for each pair of methods are shown in Table Table2.2. In general, when comparing methods from different classes, the level of agreement was poor to fair for the female genital tract data sets. However, for the Snn distance-based method, the level of agreement was moderate. For the CNS data sets, the level of agreement was fair to good when comparing tree-based methods against FST or Snn and poor to fair when comparing against AMOVA.

TABLE 2.
Levels of agreement between methods as measured by pairwise κ scores

Comparison between same-class methods.

For comparisons between methods of the same class, first we compared the SM test with the AI. Both methods are based on measuring the degree of phylogenetic segregation between sequences from different compartments. SM and AI concurred to a moderate extent on both data sets. For the CNS sequences, SM and AI detected compartmentalization in 38 data sets and classified 11 sets as noncompartmentalized (κ = 0.48). For the genital tract samples, both methods classified 22 data sets as compartmentalized and only 3 as noncompartmentalized (κ = 0.44).

Second, we compared distance-based FST and Snn and found that they agreed on 51 of the CNS data sets and on 26 of the female genital tract data sets. These results indicate moderate agreement for the CNS (κ = 0.48) and fair agreement for the female genital tract (κ = 0.35). When we compared FST and AMOVA, we found poor agreement between the two methods. In the CNS, both methods agreed that 9 of the cases were classified as compartmentalized and 16 were not, yielding a κ score of 0.03. For the genital tract, six cases were concordantly classified as compartmentalized and nine as not compartmentalized (κ = 0.11). Similar poor results were obtained between Snn and AMOVA, with κ values of 0.05 for the CNS and 0.14 for the female genital tract.

The poor agreement observed for FST and Snn when compared with AMOVA can be attributed to the propensity of the AMOVA test to reject compartmentalization (80 of all the analyzed data sets), which in turn raised the level of disagreement with FST and Snn. This phenomenon of observer bias (9) is known to lower κ scores. AMOVA may lack power to detect compartmentalization when the level of sequence divergence is low, as was the case in many of the test cases analyzed here.

Effect of branch lengths.

One possible cause for disagreement between tree-based and distance-based methods is that the former ignore branch length information when computing compartmentalization scores. SM treats the topology as given, and AI performs bootstrapping to incorporate the uncertainty in tree topology, but the actual scoring does not incorporate branch lengths. This can be misleading, because short interior branches cannot be unequivocally resolved and have a low degree of phylogenetic support and, when not taken into consideration, can lead to the overestimation of the topological distance (i.e., the degree of segregation) between the sequences. To investigate possible effects of short branch lengths, we calculated two coefficients which correlated the ordinal variable measuring whether or not two sequences are from the same compartment with either the genetic distance between them in the tree, r, or the number of branches separating the sequences in the tree, rb (11). We found that r and rb were in good agreement when used to classify sequences as compartmentalized (κ = 0.67 for CNS and κ = 0.66 for genital tract), suggesting that the inclusion of branch lengths does not systematically alter the classification.

Despite broad agreement between the correlation coefficients, in certain “difficult” cases, when other classifications strongly disagreed, we found discordance between r and rb (Table (Table3),3), suggesting that inclusion of branch lengths can have a strong effect on the conclusions of compartmentalization analyses in certain situations.

TABLE 3.
Examples of different results obtained with correlation coefficients r and rb and with SM and FSTa

Effect of recombination.

Recombination is known to occur during HIV infection, and this phenomenon can interfere with the detection of compartmentalization. The rationale behind this is that if the sequences obtained from an infected patient were a product of recombination, this would result in a lower power to detect compartmentalization because the clades of the tree would appear to be mixed and the phylogenetic reconstruction would be uncertain. With this in mind, we analyzed all data sets for recombination using a genetic algorithm approach (34). We found evidence of recombination in 36 of 92 data sets, 19 from the CNS and 17 from the female genital tract. Next, we took the alignments and trees of the nonrecombinant fragments and analyzed each of the fragments separately for compartmentalization. We show two examples of how recombination can affect compartmentalization detection. We found recombination breakpoints at nucleotides 171 and 215 of the alignments of the env viral sequences obtained from patients C and J, respectively (64). After analyzing the two fragments of the alignments separately, we found no evidence of compartmentalization in the 5′ segment, whereas the 3′ section was compartmentalized. The results are shown in Table Table4.4. We stress that it is imperative that recombination screening be a part of careful compartmentalization analyses.

TABLE 4.
Examples of the effect of recombination on compartmentalization detectiona

Simulated data.

In analyzing simulated data, first we measured the power of all competing methods to detect compartmentalization as the function of the migration rate between the two compartments. We found that tree-based methods were more powerful in detecting compartmentalization at low migration compared with distance-based AMOVA and FST, whereas Snn behaved very similar to SM regardless of being a distance-based method. As shown in Fig. Fig.1,1, tree-based SM and AI and distance-based Snn were able to detect compartmentalization in at least 80% of the simulated replicates with migration rates not exceeding 0.0007 migrations per generation. FST had at least an 80 power for migration rates lower than 0.0003. AMOVA performed poorly, only detecting compartmentalization in 80% of the replicates at migration rates lower than 0.0001, with the power to detect compartmentalization falling quickly after that point. Interestingly, FST and r behaved almost identically, but rb outperformed r when the migration rate was between 0.0001 and 0.005. This finding suggests that relatively weakly segregated sequences may be better classified using topological structure alone, since mean pairwise distances within and between compartments may be more difficult to distinguish. When the migration rate exceeded 0.0009, all methods experienced a precipitous drop in the proportion of samples classified as compartmentalized. To rule out the possibility that the tree-based methods could show higher power because they are too liberal, i.e., suffer from high rates of false positives, we simulated 100 data replicates using an unstructured population and randomly assigning sequences to two compartments with 20 sequences each. We found that none of the methods detected compartmentalization in any of the 100 replicates (data not shown), suggesting that all six tests are conservative and, hence, unlikely to suffer from a high rate of false positives.

FIG. 1.
Compartmentalization on the simulated data sets. The graphic shows the proportion of data sets classified as compartmentalized plotted against the migration rate (migration events per generation) used to run the simulation. The curve corresponding to ...

Pairwise comparisons between the methods on simulated data (Table (Table2)2) tended to show moderate to good agreement and to exceed the correlations calculated from biological data sets, especially when comparing distance-based AMOVA to FST or Snn.

Second, we examined the effect of sample size for a fixed migration rate of 0.0005 migrations per generation. Predictably, larger sample sizes increased the detection power of all methods. For example, replicates with 50 sequences per compartment were consistently classified as compartmentalized by all the methods, whereas those with 20 sequences were correctly classified only by tree-based methods and, lastly, samples with 5 sequences could not be consistently classified by any method (results not shown).

Finally, we evaluated the effect of having very different sample sizes for each compartment by running simulations of two subpopulations at a fixed migration rate of 0.0005 migrations per generation and drawing a sample of 5 sequences from one compartment and 20 from the other. One hundred replicates were run, and the results were compared with those obtained in the simulations where the sample size in both compartments was the same (20 sequences). As shown in Table Table5,5, the proportion of samples classified as compartmentalized by all the methods is lower when the sample sizes are very different compared to the ones observed for equal sample sizes, indicating that the power to detect compartmentalization of all the methods tested can be affected by having skewed sample sizes.

TABLE 5.
Proportion of simulated data sets classified as compartmentalized when equal and different sample sizes are drawn from the compartmentsa

DISCUSSION

A complete picture of HIV evolution, pathogenesis, and epidemiology must necessarily include a comprehensive characterization of how the virus colonizes, adapts to, and migrates between various tissues and organs in the host. To date, most published molecular studies of HIV have focused on the virus circulating in the bloodstream (19).

When samples from multiple organs or tissues are available, it is useful to investigate whether the viral population is structured and, if this is the case, whether compartment-specific evolutionary pressures contribute to viral diversity and divergent evolution within the host. Reliable detection of viral compartmentalization is nontrivial. Compartmentalization can be transient, and at certain points during infection the virus may be more likely to migrate between the compartments, e.g., when the viral load increases dramatically in one of the compartments, increasing the chances of migration to other compartments. Possible temporal fluctuation in the degree of compartmentalization can introduce sampling bias and confound our ability to detect population structure. For example, disease progression and sampling time can both have a strong effect on the structure of sampled populations. In addition, because it is easier to isolate viruses when the viral load is high, viral samples may be biased toward time points in which the compartmentalization signal is weakened.

Our analyses of two distinct samples of viral sequences, one coming from the CNS, including brain tissues of primarily late-stage disease or fatal cases, and the other from the genital tract of asymptomatic or stable cases, revealed surprising similarities between the distributions of compartmentalized and noncompartmentalized patients. The majority of women in the female genital tract cohort did not show lesions or inflammation in the genital tract, whereas most of the patients from whom the CNS data were obtained showed symptoms of neurological impairment, as well as neuropathology (inflammation). Hence, compartmentalization of HIV populations is not necessarily confined to any specific stage of disease. Furthermore, compartmentalization also may not be related to increased pathogenicity of viruses in a particular compartment, since compartmentalization was detected in patients with no lesions (genital tract) or at an advanced stage of pathogenesis (patients that developed neuro-AIDS). However, it should be noted that some suggestive correlations, e.g., between compartmentalization of HIV sequences in the CNS and dementia (57, 62), have been reported. While it is intuitive to consider that adaptation of HIV to replicate more efficiently in the CNS may be responsible for neurovirulence and that adaptive mutations are more likely to reach high frequencies within a compartmentalized subpopulation, it is yet to be established that the correlation between compartmentalization and dementia is present in a significant number of cases. Additionally, a sampling bias towards patients with clinical neuropathological symptoms may influence such correlational studies.

There is no reason to believe that tissue-specific adaptations in HIV are limited to the CNS. For example, a difference in transmission rates through genital exposure between men and women has been documented (37), a potential effect of viral evolution specific to the male and female genital tracts that includes factors such as virus load of donor and cell and tissue type of recipient. Differences in the patterns of drug resistance in viruses localized in the genital tract and circulating in the blood have also been reported (12). Hence it is possible that certain compartments can act as reservoirs for drug-resistant viruses, especially if optimal concentrations of drugs cannot be attained in those compartments. Suboptimal pharmacokinetics have been reported for both the female genital tract (58) and the CNS (4, 61, 69). Well-documented differences between viruses sampled from different compartments reinforce the need for more comprehensive analyses of within-host viral populations.

Given the difficulty and cost of obtaining representative viral clones from multiple organs and tissues, it is imperative to select the best available methods for subsequent molecular studies. Presently, there is no accepted “gold standard” for detecting compartmentalization in viral populations; hence, the need for a rigorous comparison between cited methodologies is obvious. When fundamentally different approaches arrive at the same conclusion, we may be more confident in the results. Our results showed that different methods to detect compartmentalization frequently disagree. In order to justify giving preference to the conclusions of one method or class of methods, we benchmarked them on a set of simulated sequence alignments. Overall, we found that the sensitivity of the methods fell as the rate of migration between two compartments increased. For a fixed migration rate, up to a certain threshold (in the case of our simulated data this threshold is 0.0007 migrations per generation, but this is not necessarily related to actual migration rates), methods based on examining the shape of the phylogenetic tree (the SM test and AI) had more power to detect compartmentalization than the methods based solely on pairwise genetic distances between sequences (FST and AMOVA). However, tree-based methods can be sensitive to topological uncertainty or recombination and place too much weight on phylogenetic segregation achieved via poorly supported short interior branches. Phylogenies constructed from within-host viral samples are often poorly resolved, and choosing a particular branching order may overestimate the extent of phylogenetic separation.

Small sample size can also adversely affect power to detect compartmentalization. For instance, when we analyzed simulated data with five sequences per compartment, all methods performed poorly. Consequently, we argue that more than five sequences per compartment are needed to gain any confidence in the results and, therefore, future studies should aim to exceed this threshold. Based our simulation results, we consider 20 sequences per compartment to be an adequate sample size, at least for sequences with a level of diversity and length similar to those included in this study. It is also important to analyze similar numbers of clones from each compartment, because highly skewed samples can reduce power and accuracy of the methods. As the techniques for isolating viral clones and sequencing improve and more sequences per compartments are routinely included in molecular studies, we expect that uncertainty due to small sample sizes will become less common.

Finally, the strategy employed to obtain clonal sequences is critically important. In most previous studies, sequence samples were gathered by isolating the genetic material from a tissue sample, performing PCR, and sequencing multiple clones. In two cases (27, 43), limiting dilution PCR was used, whereby one clone from each PCR was sequenced. Other authors (20, 57, 64) reported pooling multiple PCR products before cloning and sequencing. To what extent the sample is representative of the viral diversity will clearly be affected by the number of independent PCRs and the procedures used for clone selection. For instance, if several samples are taken from the compartment but only one sequence from each aliquot is obtained, the most prevalent sequence is likely to be detected repeatedly without adding new information about the lower-frequency viruses present in the population. On the other hand, if multiple clones are extracted from a sample, we can gain more knowledge in the diversity of the viral population infecting a specific compartment. Ideally, longitudinal samples are preferable so that a dynamic picture of the compartmentalization status of a patient can be resolved. A universally accepted procedure for sampling a viral population does not exist, and we shall not endeavor to propose one, apart from the general recommendation that one has to choose a sampling method that is best suited to a particular problem and to weigh these requirements against the cost and the amount of work needed to obtain the most appropriate samples.

We examined the level of agreement between methods using the standard κ statistic. In general, we found that all methods had better pairwise agreement on simulated data than they did on biological samples. This finding was not unexpected, since simulated data sets were generated under the same simple model of structured populations, whereas within-host evolution is likely to be much more complex and variable between patients. The length and level of diversity of the sequences obtained are likely to play a role in the ability to detect compartmentalization. Interestingly, the CNS data sets had in general lower diversity and shorter length than the female genital tract data, and the levels of agreement of the methods were in general higher for the CNS. However, there was no correlation between diversity and the classification of any specific data set as compartmentalized (data not shown). Tree-based and distance-based methods were more congruent with methods from the same class than with those from the other class. The exception to this observation is Snn; while this method behaved similarly to FST when applied to the CNS data sets, the results obtained with the female genital tract and the simulated data indicated that it was more sensitive than FST and had a power similar to that of tree-based SM, consistent with the previous suggestion that Snn is a more powerful statistic than other FST metrics (26). However, small sample size combined with high diversity can affect its performance.

Even though distance-based methods appeared to be less sensitive, positive compartmentalization results based on pairwise distance alone reflect substantial accumulation of compartment-specific mutations in different subpopulations. In many cases, however, a few point mutations can drastically alter viral phenotype, for instance, conferring resistance to fusion (36) or to nucleoside (28, 55) and nonnucleoside (40) reverse transcriptase inhibitors. If the populations are segregated by only a few key mutations, fixed in one of the populations, there may be insufficient signal for distance-based methods to detect compartmentalization, but because such mutations can result in complete phylogenetic segregation of samples from different compartments, tree-based methods are able to detect population structure with confidence.

Other methods to detect segregation and differentiation between two subpopulations exist and have been employed in the study of population dynamics of various organisms. For instance, MIGRATE (5, 7) can be used to infer population parameters such as effective population size and migration rates, while LAMARC (6) can also incorporate recombination. In this study we focused on the analysis of those methods that have been applied to the study of intrapatient compartmentalization of HIV, but a further investigation of alternative techniques may prove fruitful.

In summary, we observed that many published samples that were reported as compartmentalized or not compartmentalized might have been classified differently using an alternative method. In light of this finding, we espouse a method consensus approach, where all available tools are used to classify a given sample. When discordant results are obtained, further analysis, or additional sampling, may be recommended. Screening for recombination and evaluating different substitution models can also help to discern some of the contradictions. If a reliable phylogenetic tree (with well-supported internal branches) can be inferred for an entire sample, then tree-based methods such as the SM test or association index appear to be preferable to distance-based methods (FST and AMOVA), due to better power to detect stabilizing selection within compartments. The combined use of different methods will result in a more reliable determination of intrapatient viral compartmentalization status, which is required for a better understanding of virus infection dynamics and pathogenesis as well as molecular epidemiology.

Acknowledgments

This research was supported in part by the National Institutes of Health (AI43638, AI47745, and AI57167), the University of California Universitywide AIDS Research Program (grant number IS02-SD-701), and by a University of California, San Diego Center for AIDS Research/NIAID Developmental Award to S.D.W.F. and S.L.K.P. (AI36214). In addition, P.S. was supported by NIH grants DA14533, DA12580, and GM056529.

Footnotes

[down-pointing small open triangle]Published ahead of print on 11 April 2007.

REFERENCES

1. Abbate, I., G. Cappiello, R. Longo, A. Ursitti, A. Spano, S. Calcaterra, F. Dianzani, A. Antinori, and M. R. Capobianchi. 2005. Cell membrane proteins and quasispecies compartmentalization of CSF and plasma HIV-1 from AIDS patients with neurological disorders. Infect. Genet. Evol. 5:247-253. [PubMed]
2. Altman, D. 1997. Practical statistics for medical research. Chapman and Hall, London, England.
3. Anderson, C. N. K., U. Ramakrishnan, Y. L. Chan, and E. A. Hadly. 2005. Serial SimCoal: a population genetics model for data from multiple populations and points in time. Bioinformatics 21:1733-1734. [PubMed]
4. Antinori, A., A. Cingolani, M. L. Giancola, F. Forbici, A. De Luca, and C. F. Perno. 2003. Clinical implications of HIV-1 drug resistance in the neurological compartment. Scand. J. Infect. Dis. 35:41-44. [PubMed]
5. Beerli, P. 2006. Comparison of Bayesian and maximum-likelihood inference of population genetic parameters. Bioinformatics 22:341-345. [PubMed]
6. Beerli, P., and J. Felsenstein. 1999. Maximum-likelihood estimation of migration rates and effective population numbers in two populations using a coalescent approach. Genetics 152:763-773. [PMC free article] [PubMed]
7. Beerli, P., and J. Felsenstein. 2001. Maximum likelihood estimation of a migration matrix and effective population sizes in n subpopulations by using a coalescent approach. Proc. Natl. Acad. Sci. USA 98:4563-4568. [PMC free article] [PubMed]
8. Burkala, E., J. He, J. West, C. Wood, and C. Petito. 2005. Compartmentalization of HIV-1 in the central nervous system: role of the choroid plexus. AIDS 19:675-684. [PubMed]
9. Byrt, T., J. Bishop, and J. Carlin. 1993. Bias, prevalence and kappa. J. Clin. Epidemiol. 46:423-429. [PubMed]
10. Cohen, J. 1960. A coefficient of agreement for nominal scales. Educ. Psych. Meas. 20:37-46.
11. Critchlow, D. E., S. Li, K. Nourijelyani, and D. K. Pearl. 2000. Some statistical methods for phylogenetic trees with application to HIV disease. Math. Comput. Model. 32:69-81.
12. De Pasquale, M. P., A. J. Leigh Brown, S. C. Uvin, J. Allega-Ingersoll, A. M. Caliendo, L. Sutton, S. Donahue, and R. T. D'Aquila. 2003. Differences in HIV-1 pol sequences from female genital tract and blood during antiretroviral therapy. J. Acquir. Immune Defic. Syndr. 34:37-44. [PubMed]
13. Di Stefano, M., L. Monno, J. Fiore, G. Buccoliero, A. Appice, L. Perulli, G. Pastore, and G. Angarano. 1998. Neurological disorders during HIV-1 infection correlate with viral load in cerebrospinal fluid but not with virus phenotype. AIDS 12:737-743. [PubMed]
14. Donaldson, Y. K., P. Simmonds, A. Busuttil, J. E. Bell, J. W. Ironside, R. P. Brettle, and J. R. Robertson. 1994. Redistribution of HIV outside the lymphoid system with onset of AIDS. Lancet 343:382-385. [PubMed]
15. Ellerbrock, T. V., J. L. Lennox, K. A. Clancy, R. F. Schinazi, T. C. Wright, M. Pratt-Palmore, T. Evans-Strickfaden, C. Schnell, R. Pai, L. J. Conley, E. E. Parrish-Kohler, T. J. Bush, K. Tatti, and C. E. Hart. 2001. Cellular replication of human immunodeficiency virus type 1 occurs in vaginal secretions. J. Infect. Dis. 184:28-36. [PubMed]
16. Enting, R. H., R. M. Hoetelmans, J. M. Lange, D. M. Burger, J. H. Beijnen, and P. Portegies. 1998. Antiretroviral drugs and the central nervous system. AIDS 12:1941-1955. [PubMed]
17. Excoffier, L., J. Novembre, and S. Schneider. 2000. Computer note. SIMCOAL: a general coalescent program for the simulation of molecular data in interconnected populations with arbitrary demography. J. Hered. 91:506-509. [PubMed]
18. Excoffier, L., P. E. Smouse, and J. M. Quattro. 1992. Analysis of molecular variance inferred from metric distances among DNA haplotypes: Application to human mitochondrial DNA restriction data. Genetics 131:479-491. [PMC free article] [PubMed]
19. Frost, S. D. W., M.-J. Dumaurier, S. Wain-Hobson, and A. J. Leigh Brown. 2001. Genetic drift and within-host metapopulation dynamics of HIV-1 infection. Proc. Natl. Acad. Sci. USA 98:6975-6980. [PMC free article] [PubMed]
20. Gatanaga, H., S. Oka, S. Ida, T. Wakabayashi, T. Shioda, and A. Iwamoto. 1999. Active HIV-1 redistribution and replication in the brain with HIV encephalitis. Arch. Virol. 144:29-43. [PubMed]
21. Gray, R. H., M. J. Wawer, R. Brookmeyer, N. K. Sewankambo, D. Serwadda, F. Wabwire-Mangen, T. Lutalo, X. Li, T. vanCott, and T. C. Quinn. 2001. Probability of HIV-1 transmission per coital act in monogamous, heterosexual, HIV-1-discordant couples in Rakai, Uganda. Lancet 357:1149-1153. [PubMed]
22. Haase, A. T., K. Henry, M. Zupancic, G. Sedgewick, R. A. Faust, H. Melroe, W. Cavert, K. Gebhard, K. Staskus, Z. Q. Zhang, P. J. Dailey, H. H. J. Balfour, A. Erice, and A. S. Perelson. 1996. Quantitative image analysis of HIV-1 infection in lymphoid tissue. Science 274:985-989. [PubMed]
23. Hasegawa, M., H. Kishino, and T. A. Yano. 1985. Dating of the human ape splitting by a molecular clock of mitochondrial-DNA. J. Mol. Evol. 22:160-174. [PubMed]
24. Hudson, R., D. Boos, and N. Kaplan. 1992. A statistical test for detecting geographic subdivision. Mol. Biol. Evol. 9:138-151. [PubMed]
25. Hudson, R., M. Slatkin, and W. P. Maddison. 1992. Estimation of levels of gene flow from DNA sequence data. Genetics 132:583-589. [PMC free article] [PubMed]
26. Hudson, R. 2000. A new statistic for detecting genetic differentiation. Genetics 155:2011-2014. [PMC free article] [PubMed]
27. Hughes, E., J. Bell, and P. Simmonds. 1997. Investigation of the dynamics of the spread of human immunodeficiency virus to brain and other tissues by evolutionary analysis of sequences from the p17gag and env genes. J. Virol. 71:1272-1280. [PMC free article] [PubMed]
28. Kavlick, M. F., T. Shirasaka, E. Kojima, J. M. Pluda, F. Hui, Jr., R. Yarchoan, and H. Mitsuya. 1995. Genotypic and phenotypic characterization of HIV-1 isolated from patients receiving (−)-2′,3′-dideoxy-3′-thiacytidine. Antivir. Res. 28:133-146. [PubMed]
29. Kemal, K. S., B. Foley, H. Burger, K. Anastos, H. Minkoff, C. Kitchen, S. M. Philpott, W. Gao, E. Robison, S. Holman, C. Dehner, S. Beck, W. A. Meyer, A. Landay, A. Kovacs, J. Bremer, and B. Weiser. 2003. HIV-1 in genital tract and plasma of women: compartmentalization of viral sequences, coreceptor usage, and glycosylation. Proc. Natl. Acad. Sci. USA 100:12972-12977. [PMC free article] [PubMed]
30. Kim, R. B., M. F. Fromm, C. Wandel, B. Leake, A. J. Wood, D. M. Roden, and G. R. Wilkinson. 1998. The drug transporter P-glycoprotein limits oral absorption and brain entry of HIV-1 protease inhibitors. J. Clin. Investig. 101:289-294. [PMC free article] [PubMed]
31. Koenig, S., H. E. Gendelman, J. M. Orenstein, M. C. Dal Canto, G. H. Pezeshkpour, M. Yungbluth, F. Janotta, A. Aksamit, M. A. Martin, and A. S. Fauci. 1986. Detection of AIDS virus in macrophages in brain tissue from AIDS patients with encephalopathy. Science 233:1089-1093. [PubMed]
32. Korber, B. T., K. J. Kunstman, B. K. Patterson, M. Furtado, M. M. McEvilly, R. Levy, and S. M. Wolinsky. 1994. Genetic differences between blood- and brain-derived viral sequences from human immunodeficiency virus type 1-infected patients: evidence of conserved elements in the V3 region of the envelope protein of brain-derived sequences. J. Virol. 68:7467-7481. [PMC free article] [PubMed]
33. Kosakovsky Pond, S. L., S. D. W. Frost, and S. V. Muse. 2005. HyPhy: hypothesis testing using phylogenies. Bioinformatics 21:676-679. [PubMed]
34. Kosakovsky Pond, S. L., D. Posada, M. B. Gravenor, C. H. Woelk, and S. D. Frost. 2006. Automated phylogenetic detection of recombination using a genetic algorithm. Mol. Biol. Evol. 23:1981-2001. [PubMed]
35. Langford, D., A. Grigorian, R. Hurford, A. Adame, R. Ellis, L. Hansen, and E. Masliah. 2004. Altered P-glycoprotein expression in AIDS patients with HIV encephalitis. J. Neuropathol. Exp. Neurol. 63:1038-1047. [PubMed]
36. Lohrengel, S., F. Hermann, I. Hagmann, H. Oberwinkler, L. Scrivano, C. Hoffmann, D. von Laer, and M. T. Dittmar. 2005. Determinants of human immunodeficiency virus type 1 resistance to membrane-anchored gp41-derived peptides. J. Virol. 79:10237-10246. [PMC free article] [PubMed]
37. Long, M. E., H. L. Martin, J. K. Kreiss, S. M. J. Rainwater, L. Lavreys, D. J. Jackson, J. Rakwar, K. Mandaliya, and J. Overbaugh. 2000. Gender differences in HIV-1 diversity at time of infection. Nat. Med. 6:71-75. [PubMed]
38. Morris, A., M. Marsden, K. Halcrow, E. S. Hughes, R. P. Brettle, J. E. Bell, and P. Simmonds. 1999. Mosaic structure of the human immunodeficiency virus type 1 genome infecting lymphoid cells and the brain: evidence for frequent in vivo recombination events in the evolution of regional populations. J. Virol. 73:8720-8731. [PMC free article] [PubMed]
39. Nickle, D. C., M. A. Jensen, D. Shriner, S. J. Brodie, L. M. Frenkel, J. E. Mittler, and J. I. Mullins. 2003. Evolutionary indicators of human immunodeficiency virus type 1 reservoirs and compartments. J. Virol. 77:5540-5546. [PMC free article] [PubMed]
40. Nunberg, J. H., W. A. Schleif, E. J. Boots, J. A. O'Brien, J. C. Quintero, J. M. Hoffman, E. A. Emini, and M. E. Goldman. 1991. Viral resistance to human immunodeficiency virus type 1-specific pyridinone reverse transcriptase inhibitors. J. Virol. 65:4887-4892. [PMC free article] [PubMed]
41. Ohagen, A., A. Devitt, K. Kunstman, P. Gorry, P. Rose, B. Korber, J. Taylor, R. Levy, R. Murphy, S. Wolinsky, and D. Gabuzda. 2003. Genetic and functional analysis of full-length human immunodeficiency virus type 1 env genes derived from brain and blood of patients with AIDS. J. Virol. 77:12336-12345. [PMC free article] [PubMed]
42. Petito, C. 2004. Human immunodeficiency virus type 1 compartmentalization in the central nervous system. J. Neurovirol. 10:21-24. [PubMed]
43. Philpott, S., H. Burger, C. Tsoukas, B. Foley, K. Anastos, C. Kitchen, and B. Weiser. 2005. Human immunodeficiency virus type 1 genomic RNA sequences in the female genital tract and blood: compartmentalization and intrapatient recombination. J. Virol. 79:353-363. [PMC free article] [PubMed]
44. Pilcher, C. D., D. C. Shugars, S. A. Fiscus, W. C. Miller, P. Menezes, J. Giner, B. Dean, K. Robertson, C. E. Hart, J. L. Lennox, J. J. J. Eron, and C. B. Hicks. 2001. HIV in body fluids during primary HIV infection: implications for pathogenesis, treatment and public health. AIDS 15:837-845. [PubMed]
45. Pillai, S. K., B. Good, S. Kosakovsky Pond, J. K. Wong, M. C. Strain, D. D. Richman, and D. M. Smith. 2005. Semen-specific genetic characteristics of human immunodeficiency virus type 1 env. J. Virol. 79:1734-1742. [PMC free article] [PubMed]
46. Pillai, S. L., S. L. Kosakovsky Pond, Y. Liu, B. M. Good, M. C. Strain, R. J. Ellis, S. Letendre, D. M. Smith, H. F. Gunthard, I. Grant, T. D. Marcotte, J. A. McCutchan, D. D. Richman, and J. K. Wong. 2006. Genetic attributes of cerebrospinal fluid-derived HIV-1 env. Brain 129:1872-1883. [PubMed]
47. Ping, L.-H., M. S. Cohen, I. Hoffman, P. Vernazza, F. Seillier-Moiseiwitsch, H. Chakraborty, P. Kazembe, D. Zimba, M. Maida, S. A. Fiscus, J. J. Eron, R. Swanstrom, and J. A. E. Nelson. 2000. Effects of genital tract inflammation on human immunodeficiency virus type 1 V3 populations in blood and semen. J. Virol. 74:8946-8952. [PMC free article] [PubMed]
48. Poss, M., A. G. Rodrigo, J. J. Gosink, G. H. Learn, D. de Vange Panteleeff, J. Martin, L. Harold, J. Bwayo, J. K. Kreiss, and J. Overbaugh. 1998. Evolution of envelope sequences from the genital tract and peripheral blood of women infected with clade A human immunodeficiency virus type 1. J. Virol. 72:8240-8251. [PMC free article] [PubMed]
49. Quinn, T. C., M. J. Wawer, N. Sewankambo, D. Serwadda, C. Li, F. Wabwire-Mangen, M. O. Meehan, T. Lutalo, R. H. Gray, and T. R. P. S. Group. 2000. Viral load and heterosexual transmission of human immunodeficiency virus type 1. N. Engl. J. Med. 342:921-929. [PubMed]
50. Rambaut, A., and N. C. Grassly. 1997. Seq-Gen: an application for the Monte Carlo simulation of DNA sequence evolution along phylogenetic trees. Comput. Appl. Biosci. 13:235-238. [PubMed]
51. Ritola, K., K. Robertson, S. A. Fiscus, C. Hall, and R. Swanstrom. 2005. Increased human immunodeficiency virus type 1 (HIV-1) env compartmentalization in the presence of HIV-1-associated dementia. J. Virol. 79:10830-10834. [PMC free article] [PubMed]
52. Royce, R. A., A. Sena, W. Cates, and M. S. Cohen. 1997. Sexual transmission of HIV. N. Engl. J. Med. 336:1072-1078. [PubMed]
53. Saag, M. S., B. H. Hahn, J. Gibbons, Y. Li, E. S. Parks, W. P. Parks, and G. M. Shaw. 1988. Extensive variation of human immunodeficiency virus type-1 in vivo. Nature 334:440-444. [PubMed]
54. Salemi, M., S. L. Lamers, S. Yu, T. de Oliveira, W. M. Fitch, and M. S. McGrath. 2005. Phylodynamic analysis of human immunodeficiency virus type 1 in distinct brain compartments provides a model for the neuropathogenesis of AIDS. J. Virol. 79:11343-11352. [PMC free article] [PubMed]
55. Schinazi, R. F., R. M. Lloyd, Jr., M. H. Nguyen, D. L. Cannon, A. McMillan, N. Ilksoy, C. K. Chu, D. C. Liotta, H. Z. Bazmi, and J. W. Mellors. 1993. Characterization of human immunodeficiency viruses resistant to oxathiolane-cytosine nucleosides. Antimicrob. Agents Chemother. 37:875-881. [PMC free article] [PubMed]
56. Shankarappa, R., J. B. Margolick, S. J. Gange, A. G. Rodrigo, D. Upchurch, H. Farzadegan, P. Gupta, C. R. Rinaldo, G. H. Learn, X. He, X.-L. Huang, and J. I. Mullins. 1999. Consistent viral evolutionary changes associated with the progression of human immunodeficiency virus type 1 infection. J. Virol. 73:10489-10502. [PMC free article] [PubMed]
57. Shapshak, P., D. M. Segal, K. A. Crandall, R. K. Fujimura, B.-T. Zhang, K.-Q. Xin, K. Okuda, C. K. Petito, C. Eisdorfer, and K. Goodkin. 1999. Independent evolution of HIV type 1 in different brain regions. AIDS Res. Hum. Retrovir. 15:811-820. [PubMed]
58. Si-Mohamed, A., M. D. Kazatchkine, I. Heard, C. Goujon, T. Prazuck, G. Aymard, G. Cessot, Y. H. Kuo, M. C. Bernard, B. Diquet, J. E. Malkin, L. Gutmann, and L. Belec. 2000. Selection of drug-resistant variants in the female genital tract of human immunodeficiency virus type 1-infected women receiving antiretroviral therapy. J. Infect. Dis. 182:112-122. [PubMed]
59. Slatkin, M. 1993. Isolation by distance in equilibrium and non-equilibrium populations. Evolution 47:264-279.
60. Slatkin, M., and W. P. Maddison. 1989. A cladistic measure of gene flow inferred from the phylogenies of alleles. Genetics 123:603-613. [PMC free article] [PubMed]
61. Smit, T., B. Brew, W. Tourtellotte, S. Morgello, B. Gelman, and N. Saksena. 2004. Independent evolution of human immunodeficiency virus (HIV) drug resistance mutations in diverse areas of the brain in HIV-infected patients, with and without dementia, on antiretroviral treatment. J. Virol. 78:10133-10148. [PMC free article] [PubMed]
62. Smit, T. K., B. Wang, T. Ng, R. Osborne, B. Brew, and N. K. Saksena. 2001. Varied tropism of HIV-1 isolates derived from different regions of adult brain cortex discriminate between patients with and without AIDS dementia complex (ADC): evidence for neurotropic HIV variants. Virology 279:509-526. [PubMed]
63. Staprans, S., N. Marlowe, D. Glidden, T. Novakovic-Agopian, R. Grant, M. Heyes, F. Sweeka, S. Deeks, and R. Price. 1999. Time course of cerebrospinal fluid responses to antiretroviral therapy: evidence for variable compartmentalization of infection. AIDS 13:1051-1061. [PubMed]
64. Strain, M. C., S. Letendre, S. K. Pillai, T. Russell, C. C. Ignacio, H. F. Gunthard, B. Good, D. M. Smith, S. M. Wolinsky, M. Furtado, J. Marquie-Beck, J. Durelle, I. Grant, D. D. Richman, T. Marcotte, J. A. McCutchan, R. J. Ellis, and J. K. Wong. 2005. Genetic composition of human immunodeficiency virus type 1 in cerebrospinal fluid and blood without treatment and during failing antiretroviral therapy. J. Virol. 79:1772-1788. [PMC free article] [PubMed]
65. Sullivan, S. T., U. Mandava, T. Evans-Strickfaden, J. L. Lennox, T. V. Ellerbrock, and C. E. Hart. 2005. Diversity, divergence, and evolution of cell-free human immunodeficiency virus type 1 in vaginal secretions and blood of chronically infected women: associations with immune status. J. Virol. 79:9799-9809. [PMC free article] [PubMed]
66. Tamura, K., and M. Nei. 1993. Estimation of the number of nucleotide substitutions in the control region of mitochondrial-DNA in humans and chimpanzees. Mol. Biol. Evol. 10:512-526. [PubMed]
67. Thioulouse, J., D. Chessel, S. Dolédec, and J.-M. Olivier. 1997. ADE-4: a multivariate analysis and graphical display software. Stat. Comput. 7:75-83.
68. Thompson, K., M. Churchill, P. Gorry, J. Sterjovski, R. Oelrichs, S. Wesselingh, and C. McLean. 2004. Astrocyte specific viral strains in HIV dementia. Ann. Neurol. 56:873-877. [PubMed]
69. Venturi, G., M. Catucci, L. Romano, P. Corsi, F. Leoncini, P. Valensin, and M. Zazzi. 2000. Antiretroviral resistance mutations in human immunodeficiency virus type 1 reverse transcriptase and protease from paired cerebrospinal fluid and plasma samples. J. Infect. Dis. 181:740-745. [PubMed]
70. Wang, T. H., Y. K. Donaldson, R. P. Brettle, J. E. Bell, and P. Simmonds. 2001. Identification of shared populations of human immunodeficiency virus type 1 infecting microglia and tissue macrophages outside the central nervous system. J. Virol. 75:11686-11699. [PMC free article] [PubMed]
71. Wawer, M. J., R. H. Gray, N. K. Sewankambo, D. Serwadda, X. Li, O. Laeyendecker, N. Kiwanuka, G. Kigozi, M. Kiddugavu, T. Lutalo, F. Nalugoda, F. Wabwire-Mangen, M. P. Meehan, and T. C. Quinn. 2005. Rates of HIV-1 transmission per coital act, by stage of HIV-1 infection, in Rakai, Uganda. J. Infect. Dis. 191:1403-1409. [PubMed]
72. Wiley, C. A., R. D. Schrier, J. A. Nelson, P. W. Lampert, and M. B. Oldstone. 1986. Cellular localization of human immunodeficiency virus infection within the brains of acquired immune deficiency syndrome patients. Proc. Natl. Acad. Sci. USA 83:7089-7093. [PMC free article] [PubMed]
73. Zhu, T., N. Wang, A. Carr, D. Nam, R. Moor-Jankowski, D. Cooper, and D. Ho. 1996. Genetic characterization of human immunodeficiency virus type 1 in blood and genital secretions: evidence for viral compartmentalization and selection during sexual transmission. J. Virol. 70:3098-3107. [PMC free article] [PubMed]

Articles from Journal of Virology are provided here courtesy of American Society for Microbiology (ASM)
PubReader format: click here to try

Formats:

Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...

Links

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...