Philippe Lemey, et al.: HIV Evolutionary Dynamics
HIV Evolutionary Dynamics Within and Among Hosts
Philippe Lemey1,2, Andrew Rambaut1 and Oliver G. Pybus1
1Department of Zoology, University of Oxford, Oxford, UK; 2Rega Institute, Katholieke Universiteit Leuven, Leuven, Belgium
The HIV evolutionary processes continuously unfold, leaving a measurable footprint in viral gene
sequences. A variety of statistical models and inference techniques have been developed to recon-
struct the HIV evolutionary history and to investigate the population genetic processes that shape
viral diversity. Remarkably different population genetic forces are at work within and among hosts.
Population-level HIV phylogenies are mainly shaped by selectively neutral epidemiologic processes,
implying that genealogy-based population genetic inference can be useful to study the HIV epi-
demic history. Such evolutionary analyses have shed light on the origins of HIV, and on the
epidemic spread of viral variants in different geographic locations and in different populations. The HIV
genealogies reconstructed from within-host sequences indicate the action of selection pressure.
In addition, recombination has a significant impact on HIV genetic diversity. Accurately quantifying
both the adaptation rate and the population recombination rate of HIV will contribute to a better
understanding of immune escape and drug resistance. Characterizing the impact of HIV transmission
on viral genetic diversity will be a key factor in reconciling the different population genetic pro-
cesses within and among hosts. (AIDS Reviews 2006;8:125-40)
Corresponding author Philippe Lemey, email@example.com
HIV. Evolution. Population dynamics. Coalescent. Selection. Recombination.
AIDS Reviews 2006;8:125-40
Department of Zoology
University of Oxford
Oxford, England, UK
Simian immunodeficiency viruses (SIV) have frequent-
ly moved among primate species, and one such event
resulted in the devastating HIV-1 pandemic in humans.
Because the virus was not recognized until the early
1980s1, our documented record of the AIDS epidemic
is largely limited to the past two decades of its trans-
However, HIV-1 strains circulating today carry in their
genome sequences a significant amount of information
about the evolutionary and epidemiologic history of the
virus. The high mutation rates2,3 and short generation
times4-6 of HIV are the constant fuel for its rapid evolution-
ary change. The long-term fate of these abundant ge-
netic changes depends on the interplay of effective
population size and natural selection, resulting in an
extremely high rate of HIV genomic evolution. Population-
level processes such as selection, migration, population
dynamics, and recombination shape HIV genetic diver-
sity both among and within hosts. The genetic footprint
of these processes may be complex, obscured or scram-
bled, which means that realistic evolutionary models are
necessary to recover the useful information contained in
HIV gene sequences sampled within and among hosts.
Understanding the processes that determine viral
genetic diversity will undoubtedly assist in the struggle
against viral infections and will contribute to our knowl-
edge of past epidemiologic events. Here, we discuss
evolutionary models and inference methods for HIV
population genetic processes. In particular, we high-
light some recent computation techniques that have
particular utility for comparing HIV-1 epidemics among
populations and for investigating the dynamics of intra-
host HIV diversity7-10.
AIDS Reviews 2006;8
Since HIV populations evolve at a rate that is several
orders of magnitude faster than that of their human
hosts, HIV sequences sampled longitudinally will usu-
ally accumulate a significant amount of evolutionary
change. Longitudinally sampled or “heterochronous”11
sequences can be obtained in one of two ways: either
from a single patient over the course of infection, or from
different patients over the duration of an epidemic (Fig. 1).
Interestingly, phylogenies reconstructed from such
sequences have distinctive features that reveal the
differences in the dynamics of HIV evolution at the in-
ter-host and intra-host levels12.
Within each host, the viral population is targeted by
both cellular and humoral immune responses, resulting
in relatively strong diversifying selection that is most
noticeable in the variable regions of the envelope (env)
gene. It has been demonstrated that the rate of amino
acid substitution in env correlates with the rate of phe-
notypic escape from neutralizing antibodies13. This im-
plies that neutralizing antibody responses cause the
relative fitness of different strains within an infection to
vary, thus constituting a major force that drives rapid
lineage turnover. As a result, intra-host phylogenies of
heterochronous env sequences exhibit an asymmetri-
cal or “ladder-like” shape, with limited diversity at any
one time (Fig. 1)12.
In contrast, HIV evolution at the inter-host level shows
little evidence that HIV transmission is driven by a
similar selective process14,15. Inter-host phylogenies
of HIV sampled through time are not ladder-like and
show the persistence of multiple lineages through time
(Fig. 1)12. The shape of inter-host phylogenies is pri-
marily determined by (selectively) neutral demographic
In summary, HIV lineages within a host vary in their
ability to survive and infect new cells, whereas different
HIV lineages among hosts show little genetic variation
in their ability to infect new individuals. Some lineages
Figure 1. A: HIV-1 within-host phylogeny with branch lengths in time units: partial env gene longitudinally sampled from a single patient over
155 months (subtype B, 129 sequences, 516 bp; patient 9)34,53. B: HIV-1 population phylogeny with branch lengths in time units: full length
env gene sampled during the U.S. HIV epidemic from 1981-1995 (subtype B, 39 sequences, 2396 bp)35. C: Root-to-tip divergence as a
function of sampling time for the within-host phylogeny. The R2 value is indicated above the regression line. D: Root-to-tip divergence
as a function of sampling time for the population phylogeny: divergence estimates for the full length and C2V5 env gene are shown with
black squares and open diamonds respectively.
Time (months since seroconversion)
R2 = 0.67
R2 = 0.27
R2 = 0.89
0 1 2 3 4 5 6 7 8
Time (years since seroconversion)
-1-2 9 10 12 1311
204060 800 100 140 160120
1985 1990 1995 19801975 1970 19651960
1990 1992 19941988198619841982
Philippe Lemey, et al.: HIV Evolutionary Dynamics
may have more opportunity for onward infection than
others, but such variation is not heritable and therefore
does not generate natural selection. One possible ex-
ception may be subtype C, which has been hypothe-
sized to be more sexually transmissible than other
strains16,17, but this has yet to be confirmed.
The two patterns outlined above suggest that diffe-
rent evolutionary and population genetic processes
should be inferred from within and among host se-
quence data. Within hosts, we can focus on selection
and adaptation and the variables involved in these
processes, such as effective population size, virus
generation time, and the distribution of selection coef-
ficients. These processes are important from a clinical
perspective since they contribute to the variability in
disease progression and the development of drug re-
Among hosts, we can use the information contained
in population-level phylogenies to investigate the move-
ment of HIV lineages among locations and risk groups,
or to estimate change in viral effective population size
over time. Since the latter depends on the changes in
the number and density of infected hosts through time14,
such analyses can help to elucidate the origin and
epidemic spread of different HIV variants20.
Phylogenetic and statistical models
for HIV evolution
To extract useful information from gene sequences
we need (a) accurate models of evolutionary and popu-
lation genetic processes, and (b) statistical methods
to infer evolutionary parameters and their confidence
limits. Tremendous advances have been made on both
fronts in recent years. Evolutionary models for HIV can
be classified as having either a phylogenetic or a popu-
lation genetic basis. In reality, both types of model are
closely related and are becoming increasingly inte-
grated. The former models represent the shared an-
cestry of gene sequences using a bifurcating tree, and
typically include complex descriptions of the pattern of
nucleotide substitution along lineages. Population
genetic methods are most interested in the distribution
and frequency of polymorphic nucleotide sites within a
set of sequences.
The advantages and disadvantages of each ap-
proach are clear: recombination is easier to accom-
modate in population genetic studies (see later),
whereas accurate models of sequence evolution
through time are more easily implemented in a phylo-
Methods of phylogenetic estimation and models of
molecular evolution have both been extensively re-
viewed elsewhere21,22 and will not be covered here.
Typically, phylogenies are estimated directly from se-
quence data and make no assumptions about how
population-level processes influence the shape of the
phylogeny23. To obtain phylogenies with branch lengths
that are measured on a calendar timescale (months or
years) the model of molecular evolution needs to as-
sume some form of relationship between genetic diver-
gence and time. The simplest is the so-called “strict”
molecular-clock model, which assumes a precise linear
relationship between the two, such that the rate of
molecular evolution remains constant through time. There
are several methods to infer constant rates of molecular
evolution (substitution rates) from heterochronous se-
quence data24, the simplest being the linear regression
approach depicted in figure 1. However, the shared
ancestry among the sampled sequences means that
the standard regression assumption of independent
data points is violated, and appropriate confidence
intervals are very difficult to obtain24. This and other
problems have now been overcome by the develop-
ment of genealogy-based probabilistic methods25,26.
Even these approaches, however, still assume a strict
molecular clock, and although constant rates of HIV
evolution might be realistic over small timescales
(e.g. within hosts, Fig. 1), statistical testing has de-
monstrated that this assumption is generally unrealistic
for RNA viruses at epidemic scales27. Fortunately, more
sophisticated methods that can accommodate evolutio-
nary rate variation among lineages have recently been
developed28,29, including, most promisingly, Bayesian
“relaxed” molecular clocks30-32.
Divergence appears to accumulate at a fairly con-
stant rate within hosts, but at a more variable rate
among hosts (Fig. 1). This can be explained by a
combination of different replication rates among pa-
tients (and thus different HIV generation times; Lemey
P, unpublished work) and different levels of immune
response, which can influence non-synonymous sub-
stitution rates18 (Lemey P, unpublished work)33.
Using a molecular clock when evolutionary rates vary
considerably may lead to underestimation of the rate
and overestimation of divergence times29, so caution
should be taken when assessing differences in evolu-
tionary rate. This is illustrated in figure 1 by regression
plots of root-to-tip divergence for env sequences sam-
pled (i) throughout the infection history of a single
patient and (ii) across different patients during the
U.S. epidemic34,35. Although the timescale of sampling
AIDS Reviews 2006;8
is comparable for both datasets, the within-host regres-
sion plot for the env C2V5 gene region suggests less
divergence-rate variability compared to the complete
env sequences sampled among patients.
When the among-patient regression is based on the
same C2V5 region as was used for the within-host
regression, the rate variability is even more pronounced.
Although C2V5 is one of the most divergent gene re-
gions in env, this is not reflected in a steeper regres-
sion slope (faster evolutionary rate). Hence the consid-
erable rate-variability among hosts might result in an
underestimation of the evolutionary rate at this level, if such
rate variability is not explicitly modeled using a relaxed-
clock approach (simulations indicated that there was no
evidence for substitution saturation; data not shown).
It has been said that the effect of convergent or re-
version mutations in C2V5 is to reduce estimated di-
vergence among patients36. However, the results of
such approaches are likely to be sensitive to the posi-
tion of the phylogenetic root used for each patient and
uncertainty in root position must be considered. Rever-
sion substitutions are likely more common in C2V5 than
other HIV genome regions, hence among-host studies
should aim to consider multiple genome regions. In
addition, the problem of reversionary changes can be
avoided by considering synonymous sites only.
To illustrate the effect of rate variability more pre-
cisely, evolutionary rate estimates under both strict and
relaxed clock assumptions are listed in table 1. Using
a strict molecular clock, the estimate of the C2V5 rate
is similar to that for the complete env rate. When a
relaxed clock is used, however, the C2V5 rate is some-
what higher than the complete env rate, as expected
for this gene region. The coefficient of variation, which
quantifies the variability of the substitution rate among
branches under a relaxed clock, indicates that the rate
varies more among patients in the C2V5 gene region.
This is also evident from the greater deviations from
the fitted regression line (Fig. 1). The within-host coef-
ficient of variation suggests that substitution rates vary
considerably. This might initially seem at odds with the
regression analysis (Fig. 1). However, the ladder-like
structure of the within-host tree has a greater degree
of non-independence of root-to-tip distances due to
shared evolutionary history and will result in a lower
apparent variance in rates24.
Importantly, HIV substitution rate variability within
hosts will be affected by transient polymorphisms8.
These mutations, which are likely to be deleterious,
usually segregate on external branches and give rise
to a faster rate for these branches (Lemey P, unpub-
lished work). The regression plot, however, will not be
affected considerably by different rates for internal and
external branches, and will mainly be determined by
mutations fixed between time points. Transient delete-
rious mutations might also explain why the within-host
rate is faster than the among-host rate for the C2V5
gene region (Table 1). However, further statistical eval-
uation is required to test this hypothesis.
The methods outlined above describe how phylog-
enies measured on a real timescale can be estimated
from sampled sequences. A separate set of evolution-
ary models can be used to estimate population-level
processes from such phylogenies – models that pro-
vide a mathematic description of the statistical proper-
ties of phylogenies under different population scenarios.
The theoretic foundation of this technique was origi-
nally developed by Kingman, and is generally known
as “coalescent theory”37,38. Central to population ge-
netic theory is the relationship between genetic diver-
sity (or patterns of polymorphism) and effective popu-
lation size. Effective population sizes (Ne) are used in
Strict clock rate
Relaxed clock rate
Coefficient of variation
(3.54E-3 - 4.70E-3)
(3.22E-3 - 6.01E-3)
(2.45E-3 - 5.28E-3)
(2.88E-3 - 7.65E-3)
(5.66E-3 - 8.12E-3)
(6.26E-3 - 1.01E-2)
Evolutionary rate estimates were obtained using Bayesian MCMC implemented in BEASTv.1.350. Relaxed clock estimates were estimated using an uncorrelated relaxed
clock model31. In this approach, rates are drawn independently and identically from an underlying distribution, in this case a lognormal distribution. The data sets are
described in the legend of figure 1.
Philippe Lemey, et al.: HIV Evolutionary Dynamics
population genetic models to avoid unnecessary math-
ematic complexity and can be thought of as the “ge-
netic” size of a population.
One way to understand effective population sizes is
to note that a real-life, complex, biologic population
with effective size X loses genetic diversity by drift at
the same rate as a simple “theoretically perfect” popu-
lation with actual population size X. In the latter, effec-
tive and actual population sizes are the same; in the
former the effective size is generally smaller. A theo-
retically perfect population has no natural selection,
non-overlapping generations, and no recombination
within the genome region under investigation. If the
evolutionary dynamics of an organism can be mea-
sured on a real timescale of months or years (as out-
lined above), then effective population size will also be
a function of the organism’s generation time.
The original coalescent model has been subsequent-
ly extended so that it is now able to reflect the popula-
tion processes of recombination39,40, population subdi-
vision41, and changing population size42,43, and can
also be extended to heterochronous sequences44.
The manner by which population processes affect the
shape of phylogenies is illustrated in figure 2, which de-
picts two theoretically perfect populations with different
demographic histories. Because the shape of the sample
phylogeny depends on the demographic history of the
population, the former can be used to estimate the latter.
How exactly can the model illustrated in figure 2 be
used to infer the epidemic history of a viral strain?
The first step is to consider that each individual in the
population represents one infection, so that population
size equals the number of infected individuals. (If the
phylogeny represents within-host evolution then each
Figure 2. Representation of the coalescent process for variable population sizes. In one, the population size has been exponentially increas-
ing (left), and in the other it has recently decreased (right, a population ‘bottleneck’). Moving back in time, from the present to the past,
sampled lineages join together, or coalesce. The rate at which they do this is inversely proportional to the population size, as there are
fewer possible ancestors for each lineage when the population size is small. The genealogies for five samples drawn at different times from
this coalescent process are shown underneath. If substitutions accumulate during this genealogic process, then we can estimate the geneal-
ogy using phylogenetic approaches applied to gene sequences.
AIDS Reviews 2006;8
individual represents one infected cell). Coalescence
events occur when two infections in a sampled lineage
are both infected by the same donor. In essence, the
method assumes that the phylogeny estimated from
virus sequences accurately reflects the underlying
transmission tree45. This situation is analogous to the
problem in evolutionary biology of estimating species
trees from gene trees46. This assumption is unlikely to
be too restrictive because (i) virus transmission, or
among-virus competition after transmission, typically
generates a strong population bottleneck, making it
very unlikely that multiple viral lineages will be repeat-
edly transmitted47-49, and (ii) viral gene trees often span
several decades so the time when two viral lineages
coalesce is close (relative to the timescale of the phy-
logeny) to the actual transmission times. This may not
be true when small transmission chains occurring over
a short timescale are analyzed128.
Probabilistic inference under many population ge-
netic models has been achieved using standard max-
imum likelihood (ML) estimation, where the aim is to
find the population parameters that give maximum
probability to the observed genealogy. The genealogy
is, however, never directly observed, but usually it is
itself the result of phylogenetic inference based on the
sequence data. Therefore, if population genetic infer-
ence is conditioned on a single ML phylogeny, any
stochastic or systematic errors in the phylogenetic es-
timation procedure are ignored.
Computationally intensive methods have recently
been proposed to tackle this problem by averaging
over a set of plausible genealogies using Monte Carlo
integration. For serially sampled populations, full
probabilistic genealogy-based modeling and Bayesian
inference using Markov Chain Monte Carlo (MCMC)
sampling have proven most useful26. The examples of
HIV demography discussed below were analyzed using
BEAST50, a Bayesian MCMC program for genealogy-
based population genetic inference that includes the
recently developed Bayesian skyline plot model as well
as relaxed clock models7,31.
Within-Host HIV population dynamics
Determining the effective population size of HIV
populations within patients is a key goal towards un-
derstanding within-patient evolutionary dynamics. It
decides whether genetic drift or natural selection is the
most important evolutionary process, and also deter-
mines whether HIV genetic variation should be mo-
deled stochastically or deterministically.
Evolutionary theory predicts that in small popula-
tions, mutations will be produced more rarely and
their fixation will largely depend on chance stochastic
events (genetic drift). In large populations, however,
mutations occur more frequently, but their fate will
ultimately be decided by the deterministic action of
Distinguishing between these scenarios is essential
to understanding processes such as drug resistance
and immune escape. Although the number of HIV-in-
fected cells within hosts is estimated51 around 107-108,
the fraction of cells that contribute to future generations
of viruses is unknown (this fraction is a key determinant of
effective population size).
Estimates of Ne using coalescent approaches ap-
plied to HIV within-host gene sequences have typi-
cally been much lower than the number of infected
cells (~103), and this has been interpreted as support
for stochastic models of HIV evolution52,53. Figure 3
shows within-host Bayesian skyline plots for three, dif-
ferent, longitudinally sampled patients with estimates
of Ne ranging between 102 and 104. Proponents of
stochastic within-host HIV evolution might also con-
sider these estimates as being in favor of neutrality.
However, all coalescent-based estimators of Ne as-
sume neutrality a priori, but when this assumption
cannot be upheld, conclusions about the evolutionary
processes based on the size of Ne will be inappropri-
ate8,54. There are also several problems in trying to
justify this assumption using standard neutrality tests,
including weak statistical power to reject the neutral
regime, and non-independency when applied to serial
With these limitations in mind, Rouzine and Coffin
(1999) adopted a different approach, inspired by clas-
sical population genetics, that evaluates the pattern of
linkage disequilibrium in HIV sequence data. The basis
of their test is the deviation of the frequency at which
four possible genetic variants at two loci are observed
from the product of the corresponding one-locus frequen-
cies. The extent of linkage disequilibrium is critically de-
pendent on Ne, and according to simulation experi-
ments, the pattern observed in HIV sequence data54 was
compatible with values of Ne ranging between 2*10-4
to 5*10-5. These estimates are closer to values that can
be expected under a deterministic regime. They note,
however, that the population size might be significantly
reduced under HAART, which could explain the variability
in time to develop drug resistance19.
More recently, Edwards, et al. (2006) used genea-
logy-based statistics to measure deviations from neu-
Philippe Lemey, et al.: HIV Evolutionary Dynamics
trality, explicitly accounting for the potential bias of
demography. These statistics revealed a clear signal
of selection that could not be generated by recombina-
tion8. Under such deviations from neutrality, estimates
of Ne, like the ones plotted in figure 3, should merely
be regarded as a measure of diversity.
The strength of natural selection in HIV gene se-
quences has also been repeatedly demonstrated using
the non-synonymous/synonymous substitution rate ratio
(d N/d S), especially in the env gene region55,56. In par-
ticular, site-specific estimates of dN/dS using genealogy-
based codon substitution models have proven useful
in investigating the distribution of selective coefficients in
protein coding sequences. These approaches are prone
to overestimation of the number of positively selected
sites when recombination is in effect57,58. Recently, a
population genetic approximation to the coalescent
with recombination, rather than a phylogenetic approach,
has been proposed to estimate diversifying selection
in the presence of recombination from isochronous
sequences10, highlighting the importance of adopting
a population genetic perspective to tackle complex
evolutionary problems. An extension of the nonpara-
metric McDonald-Kreitman test59 has also been ap-
plied to HIV to estimate within-host adaptation rates,
assuming free recombination, and revealed a stag-
gering adaptation rate of, on average, one adaptive
fixation event in env every ~2.5 months18. Such esti-
mates, in addition to the strongly asymmetrical shape
of within-host genealogies and the values of genealo-
gy-based statistics8, argue for strong selection acting
on the immunodominant env gene. It has been argued
that genes coding for targets of the immune response
can be affected by frequency dependent selection,
where haplotypes coding for a new epitope will have
a selective advantage until an appropriate immune
response has been developed60. Model-based infe-
rence will be necessary to obtain quantitative insights
into this process, which has already been reported to
occur for HIV in vivo and in vitro61,62.
Both experimental and epidemiologic studies have
provided compelling evidence that genetic recombina-
tion is a significant factor in shaping HIV diversity63,64.
Abundant coinfection of spleen cells65, the diploid HIV
genome, and reverse transcriptase strand transfer
events constitute the mechanism by which recombina-
tion within hosts occurs. Not surprisingly, high recom-
bination rates have been estimated using coalescent
Time (months since seroconversion)
effective population size
0 20 406080 100
Figure 3. Within-host Bayesian skyline plots: partial env gene longitudinally sampled from three different patients over 105, 91, and 105 months
respectively (patient1: 129 sequences, 516 bp; patient 3: 109 sequences, 516 bp; patient 7: 198 sequences, 516 bp)34,53. The population
estimate obtained by BEAST50, Ne *generation time, was rescaled to Ne using a HIV generation time of two days. The bold line represents
the mean population size, while the grey area represents the 95% credibility interval.
AIDS Reviews 2006;8
methods applied to sequence data sampled across
hosts as well as within hosts66,67.
Population genetic methods estimate the population
recombination rate, i.e. the rate at which recombinant
genome regions become fixed in the population. This
rate is a function of both the per-generation “molecu-
lar” recombination rate and effective population size68.
Shriner, et al. (2004) studied the HIV recombination
rate within a single individual and applied stringent
experimental procedures to avoid artifactual recombi-
nation whilst amplifying the 3’ half of the genome67. By
rescaling an estimate of the population recombination
rate, using the experimentally determined mutation rate
(µ = 2.5 x 10-5 per site per generation2) and the co-
alescent estimate of θ (= Ne * µ), they obtained a mean
estimate of 1.38 x 10-4 recombination events per adja-
cent sites per generation67. The study also revealed
considerable variation between two different coales-
cent estimators of the recombination rate66,69, indica-
ting that caution should be taken when interpreting
Several biologic factors might give rise to biases in
the estimate of population recombination rates. At-
tempts have been made to implement more complex
nucleotide substitution models in recombination rate
estimators70, but the specifics of within-host demo-
graphic history and deviations from neutrality outlined
above need to be accounted for. Simulations have
shown that these processes do affect estimates of
Progress is being made towards the goal of co-esti-
mating multiple evolutionary and population genetic
forces10,71,72, but there is a need for these approaches
to be validated and extended to serially sampled data.
It appears that HIV is well adapted to recombine
within hosts and it has been proposed that the recom-
bination process is a form of sexual reproduction. In
asexual populations, the accumulation rate of benefi-
cial mutations in a gene will be restricted by their link-
age to other segregating mutations73. It has been sug-
gested that the biologic role of recombination is to
counteract the adverse effects of linkage and accelerate
adaptation rates. However, recombination can both
create and break up favorable combinations of viruses,
so the net effect of recombination depends on the
interaction of the fitness effects of different mutations
within the genome (referred to as ‘epistasis’).
Recombination would be beneficial in situations of
negative epistasis, where the combination of two det-
rimental mutations results in a greater loss of fitness
than expected from the single mutations (synergy), and
where beneficial mutations may act antagonistically.
However, an extensive analysis of HIV protease and
partial RT sequences with associated fitness values
indicated that there is a predominant signal of positive
epistasis (the opposite scenario)74, but the statistical
support for this has been recently questioned75.
Rouzine and Coffin (2005) presented a model for HIV
dynamics under selection and weak recombination,
and derived an accumulation rate of beneficial muta-
tions that increases with higher recombination rates
and increased population size. They also provided im-
portant predictions for HIV evolution under antiviral
therapy by showing that drug-resistance evolution can
be prevented if the HIV population is suppressed be-
low a critical value, and that drug concentrations re-
quired to prevent rebound of resistant virus can be
significantly decreased if the number of target sites in
the HIV genome is large76.
HIV forms distinct subpopulations in different ana-
tomical sites as well as in different cell types, often
referred to as compartmentalization. Even within a spe-
cific organ, genetic analysis has revealed appreciable
The pattern of HIV migration and colonization in dif-
ferent body tissues and cell types has important evo-
lutionary and clinical repercussions. For example, HIV
dynamics between blood and brain tissue, and within
different brain compartments, have been implicated
in the pathogenesis of HIV-associated dementia and
the independent development of drug resistance77-79.
Within-host HIV migration events have typically been
displayed graphically using phylogenetic trees, and
sometimes quantitatively assessed using parsimony
techniques78. Although the impact of migration on ex-
pected phylogenetic tree shape has been adequately
modeled and several software tools to estimate migra-
tion rates are available80,81, such coalescent approach-
es have yet to be seriously applied to HIV gene se-
quences. This might change now that migration rates,
substitution rates, and population sizes can be simul-
taneously estimated from heterochronous sequences9,
even when these parameters and the number of sub-
populations change over time82. Moreover, co-estimating
migration rates, population growth and recombination
rates has been made available for isochronous se-
It should be noted that all current implementations as-
sume an “island model”, which does not allow for extinc-
tion of subpopulations. Although this might be a reason-
able assumption in many applications, within-host HIV
dynamics have been shown to fit a meta-population
Philippe Lemey, et al.: HIV Evolutionary Dynamics
Cercocebus atys atys
Pan troglodytes troglodytes
Pan troglodytes schweinfurthii
19001940 1960 1980 2000
19001940 1960 1980 2000
1900 1940 1960 1980 2000
Figure 4. Geographic range of two chimpanzee subspecies (Pan troglodytes Schweinfurthii and Pan troglodytes troglodytes) and sooty
mangabey species (Cercocebus atys atys), phylogenetic tree of the SIVcpz/HIV-1 lineage and the SIVsmm/HIV-2 lineage and Bayesian
skyline estimates for HIV-1 group M, HIV-1 group O and HIV-2 group A. The SIVcpz/HIV-1 phylogenetic tree was reconstructed from pol
amino acid sequences, including recently obtained Ptt samples from Cameroon102; The SIVsmm/HIV-2 phylogenetic tree was reconstructed
from gag nucleotide sequences. The color of the branches in the trees matches the color of the geographic ranges of the chimpanzee sub-
species. The timeframe of the independence war in Guinea-Bissau (1963-1974) is superimposed as a magenta rectangle onto the HIV-2
group A population dynamics. For HIV-1 group O, the skyline plot for the concatenated gag, int and env gene sequences and the env se-
quences separately are shown in black and blue respectively.
AIDS Reviews 2006;8
model in particular cases13. Allowing for local extinction
(e.g. as a consequence of a high turnover of produc-
tively infected T-cells13) might therefore be an interesting
extension of structured coalescent models for HIV.
Besides population structuring, infection of different
cell-types that have different turnover rates can also
considerably influence viral generation times85. Diffe-
rences in cellular turnover first became evident when
decay curves of plasma viremia following antiretroviral
treatment were investigated5,86. After an initial rapid
decay for 1-2 weeks, largely representing the decline
in productively infected CD4+ T-lymphocytes, plasma
virus decreases at a lower rate and ultimately drops
below the viral-load detection limit5,86. The second de-
cay phase might be due to macrophages, which are
less sensitive to viral cytopathogenic effects and de-
layed HIV release from dendritic cells. Even under the
viral-load detection limit, a stable HIV reservoir remains
in the form of resting memory T-cells, which can still
release replication-competent virus upon reactivation87.
The extremely long half-life of this compartment gua-
rantees lifelong viral persistence and destroys the hope
for eradication using the current antiretrovirals88. If viral
lineages have gone through infection rounds in those
cellular reservoirs with slow turnover, mean replication
rates will considerably decrease85.
It has been shown that replication rate (and its re-
ciprocal, generation time) can be inferred using an
estimate of the synonymous substitution rate and the
in vitro estimate of the mutation rate per replication
cycle. These estimates agree with mathematic model-
ing of virologic data when subpopulations of latently
infected cells are taken into account85. It should be
noted, however, that the estimate of the mutation rate
used in this study might have been too low (µ = 5.33
x 10-4 per site per replication cycle vs. 2).
Kelly, et al. (2003) provided a more formalized frame-
work to bridge the gap between dynamic models and
population genetic models of HIV infection, confirming
the prediction of a reduced evolutionary rate when
infection involves multiple cell-types89. Serial sample
coalescent theory has also been employed to infer
HIV-1 generation time in vivo44, this time with an up-
dated estimate of the mutation rate2, providing esti-
mates more agreement with virologic data (1.2 days/
generation vs. 1.8 days/generation5).
HIV population dynamics among hosts
The ability to infer the dates of origin of epidemics
and to investigate historic patterns of transmission from
viral gene sequences now plays a key role in molecu-
lar epidemiology. Many studies have contributed to
what can be considered a reasonably clear picture of
how different HIV variants have spread in the past.
HIV-1 comprises three different lineages, groups M, N,
and O, each of which is the result of a separate cross-
species transmission of SIV from chimpanzees90-92.
Group M has successfully founded epidemics worldwide
and these founder events have led to the generation of
several subtypes15. Evolutionary analyses using molecu-
lar clock techniques suggest that a common ancestor of
HIV-1 group M existed around 193093,94. The immediate
precursor of HIV-1 group M infects chimpanzees of the
subspecies Pan troglodytes troglodytes in West Central
Africa90,91 (Fig. 3). Not surprisingly, the highest degree
of group M diversity has been found close to this area.
An epidemiologic survey by Vidal, et al. (2000) re-
vealed a very high diversity of HIV strains present in
the Democratic Republic of Congo (DRC)98; these
strains showed much less distinction between intra-
and intersubtype diversity in comparison to group M
strains sampled globally95. Coalescent analysis of
these sequences (albeit based on a single-tree esti-
mate) suggested a logistic growth of HIV-1 population
size over time96,97.
In figure 4, we show a more up-to-date estimate for
the same dataset, obtained using the Bayesian skyline
plot method7. This method not only incorporates phy-
logenetic uncertainty but also takes into account the
uncertainty in the estimate of evolutionary rate, where-
as in the original analysis both the phylogeny and the
rate were fixed to a point estimate96.
Figure 4 confirms an early period of slow HIV-1
spread, followed by a subsequent period of more rapid
epidemic transmission in more recent decades. The
timescale of this estimate, which suggests an origin of
HIV-1 group M around 1940, is slightly more recent
than in previous coalescent analyses96. The slightly
different timescale results from a different treatment of
the evolutionary rate parameter in the two analyses.
Here, we used Bayesian inference to estimate evolu-
tionary rate from the full length env gene subtype B
data set of Robbins, et al. 2003, whilst allowing for a
separate rate for the V3-V5 gene region.
In the previous analysis, the evolutionary rate was
fixed to 0.0023 nucleotide substitutions per site per
year (CI: 0.0016-0.0033)96, which was based on analy-
sis of the V3-V5 region of the Korber, et al. (2000)
dataset93. Although this region is one of the most
variable in the HIV genome, it gave a roughly similar
rate to that obtained for the complete env gene, sug-
Philippe Lemey, et al.: HIV Evolutionary Dynamics
gesting that this estimate may be subject to the prob-
lem of underestimation arising from the assumption
of rate constancy, as discussed above and illustrated
in figure 1.
Given that heterochronous data is now available for
this DRC population98-100, it would be more appropriate
to estimate the rate from the data rather than fixing it to
one value or specifying a strong prior distribution.
The geographic location, timing, and phylogenetic
structure of the HIV-1 epidemic all provide evidence
against the hypothesis that HIV has emerged due to
SIVcpz-contaminated oral polio vaccines in the DRC in
the late 1950s93,95,96. In the past, low SIV infection
rates in chimpanzees have prevented a very strong
case being made for the natural transfer hypothesis of
SIV to humans. The finding that all three HIV-1 lineages
were spawned by an SIV progenitor carried by the central
chimpanzee subspecies, Pan troglodytes troglodytes
(Ptt, Fig. 4), and that Pan troglodytes Schweinfurthii
only carry an SIV distantly related to HIV-1101, high-
lighted the importance of screening the Ptt subs-
Very recently, an analysis of about 600 fecal samples
revealed several Pan troglodytes troglodytes communi-
ties in southern Cameroon with widespread SIVcpz-Ptt
infection, indicating that the outbreak began in rural
Cameroon and then traveled to Kinshasa, DRC102. This
agrees with our genetic estimate of HIV-1 group M
epidemic history (Fig. 4). We hypothesize low and un-
recognized HIV transmission in remote African areas
during the early phase of the history of the epidemic,
followed by more rapid epidemic spread within a
changing and increasingly connected African popula-
tion, possibly assisted by some level of iatrogenic hu-
In contrast to HIV-1 group M, group O infections
have mostly remained restricted to Cameroon, with
some movement to neighboring countries in West Cen-
tral Africa. Current group O seroprevalence is rela-
tively modest; however, a higher genetic diversity for
group O viruses compared to group M viruses prompt-
ed the suggestion group O has been circulating in
Central Africa for longer104.
A Bayesian skyline plot estimated from concatena-
ted gag, int and env gene sequences does indeed
suggest an earlier common ancestor for group O than
for group M (Fig. 4). The group O estimate of epi-
demic history has been plotted on the same scale as
the group M estimate; the plots indicate that group O
in Cameroon has not undergone the same explosive
spread as group M in the DRC.
Previous analyses of this heterochronous data used
a multi-locus model that allowed each gene region to
have a different phylogenetic history and suggested
that such estimates are not heavily biased by recom-
bination among gene regions105. Although replication
assays have found group O to be significantly less “fit”
than group M106, these in vitro differences are not re-
flected in in vivo differences in viral load107 and there-
fore should not be extrapolated to the epidemiologic
level, despite the temptation to ascribe the different
epidemic outcomes of groups O and M to potential
differences in transmissibility.
The relative contributions of viral genetic differ-
ences and epidemiologic circumstances to the varia-
tion in epidemic history among strains must therefore
remain an unanswered question. The skyline plot re-
sults indicate that the number of effective infections
were roughly similar around 1960, but that group O did
not benefit as greatly from the extrinsic factors that led
to the increased transmission of group M infection after
1960. HIV-1 group N might have found itself in even
less favorable conditions for epidemic spread.
The second type of HIV, HIV-2, clusters with SIV from
sooty mangabeys (Cercocebus atys atys) to form a
separate lineage in primate lentivirus phylogeny. Dif-
ferent HIV-2 lineages appear to be the result of sepa-
rate cross-species transmissions, but only two HIV-2
strains (A and B) give rise to an appreciable number
of infections in humans. Viral load within asymptom-
atic patients and transmission probability between in-
dividuals are both significantly lower for HIV-2 than for
HIV-1 group M108,109. It is therefore unsurprising that HIV-2
has not spread much further than the West African
countries that coincide with the historic range of the
sooty mangabey (Fig. 4).
The densest focus of HIV-2 prevalence is in Guinea-
Bissau. A community study carried out in northwestern
Guinea-Bissau revealed a seroprevalence as high as
10% among adults110,111. In this and other populations,
HIV-2 prevalence consistently peaked in older age
groups112,113; this observation plus further investigation
of HIV-2 risk factors led to the hypothesis that HIV-2
was mainly disseminated by a generation that was
sexually active during the independence war in Gui-
nea-Bissau, which took place during the 1960s and
early 1970s112. Coalescent analysis enabled this hy-
pothesis to be tested using genetic data and revealed
that HIV-2 subtype A switched from endemic trans-
mission to epidemic growth sometime around 1955-
1970114. We estimated a Bayesian skyline plot from
HIV-2 env sequences sampled in northwestern Guinea-
AIDS Reviews 2006;8
Bissau; this confirms that the period of epidemic growth
coincides with the time frame of the Guinea-Bissau
independence war (1963-1974, Fig. 4). In this analysis,
a prior distribution for the rate of evolution was pro-
vided by an analysis of the same gene region of seri-
ally sampled HIV-2 sequences, published by Shi, et
al129. (2005). Both sexual and blood-borne HIV-2 trans-
mission might have drastically increased during the
independence war, and large-scale inoculation cam-
paigns have been recorded at the local hospital where
the sequence data was obtained.
Coalescent analyses similar to those presented
above have been used several times to study spe-
cific subtypes and epidemic strains of HIV-1 group M,
both in Africa and in other continents. The methods
have also been used to compare the epidemic poten-
tial of different subtypes co-circulating in the same
Although subtype distributions in different countries
are continually changing116, the developed world has
mainly been burdened by HIV-1 subtype B infections.
Until the early 1980s, this strain spread unnoticed
among high-risk groups, notably homosexual men, in
the USA. Phylogenetic-based reconstruction of the
epidemic history of subtype B in the USA, including
strains sampled relatively early in the epidemic, re-
vealed an explosive spread in the 1970s that slowed down
towards the present (a logistic growth trend; Fig. 5).
The estimated time for the population to double in size
at the onset of the epidemic is amongst the shortest
reported for HIV population dynamics (0.84 years-1,
0.74-0.96)35, consistent with its propagation through
standing networks of injecting drug-users and homo-
Interestingly, similar growth rates were estimated for
several UK homosexual transmission clusters, empha-
sizing again the importance of high-risk group dynam-
ics in the onset of an epidemic. Although the virus
established an epidemic in the UK by several indepen-
dent and more recent introductions, the inferred de-
mography in the homosexual “sub-epidemics” is qua-
litatively (logistic growth) and quantitatively (growth rate
and ratio of effective number of infections over preva-
lence) similar to the U.S. epidemic history (Fig. 5)118.
The leveling off towards an equilibrium state is con-
sistent with behavioral interventions and HIV preven-
tion strategies in both populations. However, it should
be noted that all Bayesian skyline plots show some
signal of steady-state dynamics towards the present
(Fig. 4 and 5), which could be at least partly attributed
MRCA of all
Effective number of infections
Figure 5. Bayesian skyline plots for HIV-1 subtype B. The skyline plot representing the U.S. epidemic history was reconstructed using the
serially sampled data set analyzed in Robbins K, et al.35. The date for the most recent common ancestor of sequences from US origin is
indicated with an arrow. The UK skyline plot was inferred using the largest cluster (‘Cluster 2’) identified in Hue, et al. 2005118; the same
prior distribution for the evolutionary rate was used in our analysis.
Philippe Lemey, et al.: HIV Evolutionary Dynamics
to within-host HIV evolution. If all sequences are sam-
pled from different hosts, then transmission events (and
thus coalescent events) are expected to occur at least
some years into the past. In addition, such sequences
probably carry recent (slightly) deleterious mutations,
which have not yet been eliminated at the population
level by purifying selection, leading to an overestima-
tion of the time to the most recent coalescence event
in the tree.
Further research is needed to quantify and model
these factors. For the time being, we recommend that
the uncertainty of any estimate is always taken into
account when coalescent analyses are being inter-
preted, and we suggest that very recent epidemic his-
tory should be interpreted with caution, especially
when estimated phylogenies contain long external
What is the influence of recombination on HIV demo-
graphic inferences? Although recombination is un-
doubtedly pervasive within hosts65,67, this does not
necessarily invalidate estimating a transmission tree
using HIV gene trees for three main reasons:
(i) The ability to reconstruct genealogies of se-
quences sampled across hosts will mainly be
hampered by recombination events between dis-
tinct variants harbored by different patients,
therefore requiring coinfection or superinfection.
Although cases of superinfection have been re-
ported and several mosaic HIV genomes are the
“circulating” proof of their occurrence119,120, esti-
mated rates of superinfection are generally very
low121-123. Rates of superinfection will also vary con-
siderably among risk groups, being greater in
high-risk groups such as injecting drug users,
and commercial sex workers, who make up only
a small fraction of overall prevalence.
(ii) The more divergent parental sequences of re-
combinants are, the more impact they are ex-
pected to have on tree reconstructions. Those
recombinants are, however, the easiest to iden-
tify using recombination detection programs and
can be omitted from the analysis (although this
is an ad hoc way to deal with, or rather ignore,
the problem of recombination).
(iii) HIV demographics are generally characterized
by exponential growth, at least in some stage of
the epidemic history, generating star-like trees
with long external and short internal branches
(Fig. 1 B). In these growing populations, fewer
recombination events can scramble the topo-
logic information contained in the sequences
data70,105,124. It is therefore not surprising that the
assumption of a single phylogenetic history
across the genome, or unlinked phylogenies for
different genes, did not lead to large differences in
the demographic estimates for HIV-1 group O105.
(iv) Recombination that occurs among lineages with-
in a single infection will not bias the topology of
an among-host phylogeny. Such recombination
may increase the variance in evolutionary rate
among lineages, but this can now be adequate-
ly modeled using relaxed clock approaches.
In the context of our point (i) above, an interesting
application of estimating intra-subtype population re-
combination rates was provided by Taylor and Korber
(2004)125. They simulated sequence data using a struc-
tured coalescent model with recombination, reflecting
transmission dynamics with varying levels of superin-
fection. By comparing population recombination rates
inferred from these simulations with estimates from real
sequence data, they concluded that superinfection
rates might be as high as 15% of infections. This con-
trasts with much lower rates observed in epidemio-
logic surveys121-123. This difference could be the result
of assumptions made in the simulations, such as con-
stant numbers of infected individuals, homogeneous
substitution rates among sites, neutral evolution, or
epidemiologic heterogeneity125. Small networks of indi-
viduals, with large superinfection rates relative to the
total population, can severely impact genealogic esti-
mates125. Recent findings of a significantly higher fre-
quency of dual infections in high-risk populations seem
to confirm this126. Interestingly, simulation studies have
specifically assessed the impact of epidemiologic mi-
xing patterns on demographic inference127. Although
the networks of behavior that spread HIV can affect the
relationship between Ne and census population size,
parametric models for estimating growth rate – the
parameter of interest from an epidemiologic perspec-
tive – seem to be highly robust to violations of pan-
mixis, even for small samples127. More research is
needed, however, to evaluate the influence of social
network structures together with geographic distance
on HIV diversity in larger populations.
Conclusions and perspectives
Different population genetic processes are shaping
viral diversity within and between hosts. Model-based
inference of HIV gene sequences now enables quan-
titative insights into the effects of these processes to
be obtained. Among hosts, the change in viral effective
AIDS Reviews 2006;8
population size over time can be modeled, revealing
historic changes in transmission dynamics. In contrast
to within-host HIV dynamics, there is little influence of
immune-driven natural selection at this level. Extensive
variation in partner exchange and transmission-associa-
ted bottlenecks can both contribute to genetic drift at
the population level14,15. In addition, selectively advan-
tageous mutations might “miss the boat” for transmis-
sion if they occur late in infection14,15. Modeling and
analysis of transmission chain data might help to ex-
plain how intra-host evolution is transformed into HIV
evolution at the population level. For example, a recent
coalescent analysis of a homosexual transmission pair
revealed that transmission was associated with a se-
vere loss of diversity (> 99%)47. Whether HIV transmis-
sion is selectively neutral is, however, still the subject
The population genetic inferences discussed in this
review indicate that model complexity required de-
pends on the level at which the virus population is
sampled. Different processes need to be modeled within
and among hosts. Although the complexity of HIV intra-
host, inter-host and transmission dynamics is not unique
among human pathogens (cfr. Hepatitis C)12, HIV is by
far the most extensively studied pathogen and is rep-
resented by the greatest amount of genetic data.
Therefore, HIV presents an opportunity to truly understand
viral genetic diversity and the population genetic pro-
cesses that shape it.
We would like to thank Beatrice Hahn for providing
the figure of the natural range of sooty mangabey and
two chimpanzee subspecies. We thank Stephane Hué
for providing the sequence data of the UK transmission
cluster. Philippe Lemey was supported by a long-term
EMBO fellowship. Andrew Rambaut and Oliver G. Py-
bus were supported by the Royal Society.
1. CDC. Pneumocystis pneumonia–Los Angeles. Morb Mortal Wkly
2. Mansky L. Forward mutation rate of HIV-1 in a T lymphoid cell line.
AIDS Res Hum Retroviruses 1996;12:307-14.
3. Mansky L, Temin H. Lower in vivo mutation rate of HIV-1 than that
predicted from the fidelity of purified reverse transcriptase. J Virol
4. Ho D, Neumann A, Perelson A, Chen W, Leonard J, Markowitz M.
Rapid turnover of plasma virions and CD4 lymphocytes in HIV-1
infection. Nature 1995;373:123-6.
5. Perelson A, Neumann A, Markowitz M, Leonard J, Ho D. HIV-1
dynamics in vivo: virion clearance rate, infected cell life-span, and
viral generation time. Science 1996;271:1582-6.
6. Wei X, Ghosh S, Taylor M, et al. Viral dynamics in HIV-1 infection.
7. Drummond A, Rambaut A, Shapiro B, Pybus O. Bayesian coales-
cent inference of past population dynamics from molecular se-
quences. Mol Biol Evol 2005;22:1185-92.
8. Edwards C, Holmes E, Wilson D, et al. HIV-1 envelope gene evolu-
tion during chronic infection is deterministic and dominated by
negative selection. Genetics (In press).
9. Ewing G, Nicholls G, Rodrigo A. Using temporally spaced se-
quences to simultaneously estimate migration rates, mutation rate
and population sizes in measurably evolving populations. Genetics
10. Wilson D, McVean G. Estimating diversifying selection and func-
tional constraint in the presence of recombination. Genetics 2006;
11. Drummond A, Pybus O, Rambaut A, Forsberg R, Rodrigo A. Mea-
surably evolving populations. Trends in Ecology and Evolution
12. Grenfell B, Pybus O, Gog J, et al. Unifying the epidemiological and
evolutionary dynamics of pathogens. Science 2004;303:327-32.
13. Frost S, Dumaurier M, Wain-Hobson S, Brown A. Genetic drift and
within-host metapopulation dynamics of HIV-1 infection. Proc Natl
Acad Sci USA 2001;98:6975-80.
14. Holmes E. The phylogeography of human viruses. Mol Ecol 2004;
15. Rambaut A, Posada D, Crandall K, Holmes E. The causes and
consequences of HIV evolution. Nat Rev Genet 2004;5:52-61.
16. John-Stewart G, Nduati R, Rousseau C, et al. Subtype C is associ-
ated with increased vaginal shedding of HIV-1. J Infect Dis 2005;
17. Iversen A, Learn G, Skinhoj P, Mullins J, McMichael A, Rambaut A.
Preferential detection of HIV subtype C’ over subtype A in cervical
cells from a dually infected woman. Aids 2005;19:990-3.
18. Williamson S. Adaptation in the env gene of HIV-1 and evolutionary
theories of disease progression. Mol Biol Evol 2003;20:1318-25.
19. Frost S, Nijhuis M, Schuurman R, Boucher CA, Brown A. Evolution
of lamivudine resistance in HIV1-infected individuals: the relative
roles of drift and selection. J Virol 2000;74:6262-8.
20. Pybus O. Inferring evolutionary and epidemiologic processes from
molecular phylogenies. University of Oxford, Oxford 2000.
21. Swofford D, Olsen G, Waddell P, Hillis D. Phylogenetic inference.
In: Molecular Systematics. Ed. Hillis D, Moritz C, Mable B: Sinauer
22. Salemi M, Vandamme A-M. The phylogenetic handbook: a practical
approach to DNA and protein phylogeny. Cambridge University
23. Wilson D, Falush D, McVean G. Germs, genomes and genealogies.
TRENDS in Ecology and Evolution 2005;20:39-45.
24. Drummond A, Pybus O, Rambaut A. Inference of viral evolutionary
rates from molecular sequences. Advances In Parasitology 2003;
25. Rambaut A. Estimating the rate of molecular evolution: incorporating
non-contemporaneous sequences into maximum likelihood phylog-
enies. Bioinformatics 2000;16:395-9.
26. Drummond A, Nicholls G, Rodrigo A, Solomon W. Estimating mutation
parameters, population history and genealogy simultaneously from
temporally spaced sequence data. Genetics 2002;161:1307-20.
27. Jenkins G, Rambaut A, Pybus O, Holmes E. Rates of molecular
evolution in RNA viruses: a quantitative phylogenetic analysis. J Mol
28. Sanderson M. A nonparametric approach to estimating diver-
gence times in the absence of rate constancy. Mol Biol Evol
29. Yoder A, Yang Z. Estimation of primate speciation dates using local
molecular clocks. Mol Biol Evol 2000;17:1081-90.
30. Aris-Brosou S, Yang Z. Effects of models of rate evolution on estima-
tion of divergence dates with special reference to the metazoan 18S
ribosomal RNA phylogeny. Syst Biol 2002;51:703-14.
31. Drummond A, Ho S, Phillips M, Rambaut A. Relaxed phylogenetics
and dating with confidence. PLoS Biol 2006;4.
Philippe Lemey, et al.: HIV Evolutionary Dynamics Download full-text
32. Thorne J, Kishino H, Painter I. Estimating the rate of evolution of the
rate of molecular evolution. Mol Biol Evol 1998;15:1647-57.
33. Frost S, Liu Y, Pond S, et al. Characterization of HIV-1 envelope
variation and neutralizing antibody responses during transmission
of HIV-1 subtype B. J Virol 2005;79:6523-7.
34. Shankarappa R, Margolick J, Gange S, et al. Consistent viral evo-
lutionary changes associated with the progression of HIV-1 infec-
tion. J Virol 1999;73:10489-502.
35. Robbins K, Lemey P, Pybus O, et al. U.S. HIV-1 epidemic: date of
origin, population history, and characterization of early strains. J
36. Herbeck J, Nickle D, Learn G, et al. HIV-1 env evolves toward ances-
tral states upon transmission to a new host. J Virol 2006;80:1637-44.
37. Kingman J. The coalescent. Stochastic Processes and their Appli-
38. Kingman J. On the genealogy of large populations. J Appl Probab
39. Hudson R. Gene genealogies and the coalescent process. In: Ox-
ford Surveys in Evolutionary Biology. Ed. Futuyama D, Antonovics
J. Oxford University Press 1990.
40. Griffiths R, Marjoram P. Ancestral inference from samples of DNA
sequences with recombination. J Comput Biol 1996,3:479-502.
41. Nath H, Griffiths R. The coalescent in two colonies with symmetric
migration. J Math Biol 1993;31:841-51.
42. Slatkin M, Hudson R. Pairwise comparisons of mitochondrial DNA
sequences in stable and exponentially growing populations. Genet-
43. Griffiths R, Tavare S. Sampling theory for neutral alleles in a
varying environment. Philos Trans. R Soc Lond B Biol Sci 1994;
44. Rodrigo A, Felsenstein J. Coalescent approaches to HIV population
genetics. In: The Evolution of HIV. Ed. Crandall K. Baltimore: John
Hopkins University Press 1999.
45. Leitner T, Fitch W. The Phylogenetics of Known Transmission His-
tories. In: The Evolution of HIV. Ed. Crandall K. Baltimore: Johns
Hopkins University Press 1999:315-45.
46. Slowinski J, Page R. How should species phylogenies be inferred
from sequence data? Syst Biol 1999;48:814-25.
47. Edwards C, Holmes E, Wilson D, et al. Population genetic estimation
of the loss of genetic diversity during horizontal transmission of
HIV-1. BMC Evol Biol 2006;6:28.
48. Derdeyn C, Decker J, Bibollet-Ruche F, et al. Envelope-constrained
neutralization-sensitive HIV-1 after heterosexual transmission. Sci-
49. Wolinsky S, Wike C, Korber B, et al. Selective transmission of HIV-
1 variants from mothers to infants. Science 1992;255:1134-7.
50. Drummond A, Rambaut A. BEAST v1.3, Available from http://evolve.
51. Haase A, Henry K, Zupancic M, et al. Quantitative image analysis
of HIV-1 infection in lymphoid tissue. Science 1996;274:985-9.
52. Brown A. Analysis of HIV-1 env gene sequences reveals evidence
for a low effective number in the viral population. Proc Natl Acad
Sci USA 1997;94:1862-5.
53. Shriner D, Shankarappa R, Jensen M, et al. Influence of random
genetic drift on HIV-1 env evolution during chronic infection. Genet-
54. Rouzine I, Coffin J. Linkage disequilibrium test implies a large ef-
fective population number for HIV in vivo. Proc Natl Acad Sci USA
55. Bonhoeffer S, Holmes E, Nowak M. Causes of HIV diversity. Nature
56. Yamaguchi Y, Gojobori T. Evolutionary mechanisms and population
dynamics of the third variable envelope region of HIV within single
hosts. Proc Natl Acad Sci USA 1997;94:1264-9.
57. Anisimova M, Nielsen R, Yang Z. Effect of recombination on the
accuracy of the likelihood method for detecting positive selection
at amino acid sites. Genetics 2003;164:1229-36.
58. Shriner D, Nickle D, Jensen M, Mullins J. Potential impact of recom-
bination on sitewise approaches for detecting positive natural selec-
tion. Genet Res 2003;81:115-21.
59. McDonald J, Kreitman M. Adaptive protein evolution at the Adh
locus in drosophila. Nature 1991;351:652-4.
60. Nielsen R. Changes in ds/dn in the HIV-1 env gene. Mol Biol Evol
61. Yuste E, Moya A, Lopez-Galindez C. Frequency-dependent selec-
tion in HIV-1. J Gen Virol 2002;83:103-6.
62. Holmes E, Zhang L, Simmonds P, Ludlam C, Brown A. Convergent
and divergent sequence evolution in the surface envelope glyco-
protein of HIV-1 within a single infected patient. Proc Natl Acad Sci
63. Levy D, Aldrovandi G, Kutsch O, Shaw G. Dynamics of HIV-1 re-
combination in its natural target cells. Proc Natl Acad Sci USA
64. Robertson D, Sharp P, McCutchan F, Hahn B. Recombination in
HIV-1. Nature 1995;374:124-6.
65. Jung A, Maier R, Vartanian J, et al. Multiply infected spleen cells in
HIV patients. Nature 2002;418:144.
66. McVean G, Awadalla P, Fearnhead P. A coalescent-based method
for detecting and estimating recombination from gene sequences.
67. Shriner D, Rodrigo A, Nickle D, Mullins J. Pervasive genomic re-
combination of HIV-1 in vivo. Genetics 2004;167:1573-83.
68. Stumpf M, McVean G. Estimating recombination rates from popula-
tion-genetic data. Nat Rev Genet 2003;4:959-68.
69. Kuhner M, Yamato J, Felsenstein J. Maximum likelihood estimation
of recombination rates from population data. Genetics 2000;156:
70. Carvajal-Rodriguez A, Crandall K, Posada D. Recombination esti-
mation under complex evolutionary models with the coalescent
composite-likelihood method. Mol Biol Evol 2006;23:817-27.
71. Kuhner M. LAMARC 2.0: maximum likelihood and Bayesian estima-
tion of population parameters. Bioinformatics 2006;22:768-70.
72. Przeworski M, Charlesworth B, Wall J. Genealogies and weak puri-
fying selection. Mol Biol Evol 1999;16:246-52.
73. Fisher R. The Genetic Theory of Natural Selection. Oxford: Claren-
don Press 1930.
74. Bonhoeffer S, Chappey C, Parkin N, Whitcomb J, Petropoulos C.
Evidence for positive epistasis in HIV-1. Science 2004;306:1547-50.
75. Wang K, Mittler J, Samudrala R. Comment on “Evidence for positive
epistasis in HIV-1”. Science 2006;312:848 [author reply 848].
76. Rouzine I, Coffin J. Evolution of HIV under selection and weak re-
combination. Genetics 2005;170:7-18.
77. Shapshak P, Segal D, Crandall K, et al. Independent evolution of
HIV-1 in different brain regions. AIDS Res Hum Retroviruses 1999;
78. Salemi M, Lamers S, Yu S, de Oliveira T, Fitch W, McGrath M.
Phylodynamic analysis of HIV-1 in distinct brain compartments pro-
vides a model for the neuropathogenesis of AIDS. J Virol 2005;
79. Smit T, Wang B, Ng T, Osborne R, Brew B, Saksena N. 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.
80. Beerli P, Felsenstein J. Maximum-likelihood estimation of migration
rates and effective population numbers in two populations using a
coalescent approach. Genetics 1999;152:763-73.
81. Beerli P, Felsenstein J. Maximum likelihood estimation of a migra-
tion matrix and effective population sizes in n subpopulations by
using a coalescent approach. Proc Natl Acad Sci USA 2001;
82. Ewing G, Rodrigo A. Coalescent-based estimation of population
parameters when the number of demes changes over time. Mol Biol
83. Bahlo M, Griffiths RC. Inference from gene trees in a subdivided
population. Theor Popul Biol 2000;57:79-95.
84. Nielsen R, Wakeley J. Distinguishing migration from isolation: a
Markov chain Monte Carlo approach. Genetics 2001;158:885-96.
85. Kelly J. Replication rate and evolution in HIV. J Theor Biol 1996;