Limits...
Reduced evolutionary rates in HIV-1 reveal extensive latency periods among replicating lineages.

Immonen TT, Leitner T - Retrovirology (2014)

Bottom Line: The method removes alternative sources that may affect evolutionary rates, such as hypermutation, recombination, and selection, to reveal the contribution of generation-time effects caused by latency.Furthermore, we discovered extensive effects of latency in sequence data (gag, pol, and env) from reservoirs as well as in the replicating plasma population.These results suggest that cycling in and out of latency plays a major role in the evolution of HIV-1.

View Article: PubMed Central - PubMed

Affiliation: Theoretical Biology & Biophysics, Los Alamos National Laboratory, Los Alamos, NM, 87545, USA. tti@lanl.gov.

ABSTRACT

Background: HIV-1 can persist for the duration of a patient's life due in part to its ability to hide from the immune system, and from antiretroviral drugs, in long-lived latent reservoirs. Latent forms of HIV-1 may also be disproportionally involved in transmission. Thus, it is important to detect and quantify latency in the HIV-1 life cycle.

Results: We developed a novel molecular clock-based phylogenetic tool to investigate the prevalence of HIV-1 lineages that have experienced latency. The method removes alternative sources that may affect evolutionary rates, such as hypermutation, recombination, and selection, to reveal the contribution of generation-time effects caused by latency. Our method was able to recover latent lineages with high specificity and sensitivity, and low false discovery rates, even on relatively short branches on simulated phylogenies. Applying the tool to HIV-1 sequences from 26 patients, we show that the majority of phylogenetic lineages have been affected by generation-time effects in every patient type, whether untreated, elite controller, or under effective or failing treatment. Furthermore, we discovered extensive effects of latency in sequence data (gag, pol, and env) from reservoirs as well as in the replicating plasma population. To better understand our phylogenetic findings, we developed a dynamic model of virus-host interactions to investigate the proportion of lineages in the actively replicating population that have ever been latent. Assuming neutral evolution, our dynamic modeling showed that under most parameter conditions, it is possible for a few activated latent viruses to propagate so that in time, most HIV-1 lineages will have been latent at some time in their past.

Conclusions: These results suggest that cycling in and out of latency plays a major role in the evolution of HIV-1. Thus, no aspect of HIV-1 evolution can be fully understood without considering latency - including treatment, drug resistance, immune evasion, transmission, and pathogenesis.

Show MeSH

Related in: MedlinePlus

Success of phylogenetic latency detection in critical simulation scenarios. Detection of latency-affected lineages was done by pair-wise comparisons of all taxa, testing whether the shorter lineage was significantly shorter according to a Poisson test. (A) Sensitivity results from simulations with 100 taxa and one random branch affected by latency at flatent = 0.1–0.9 of the corresponding original non-latent genetic distance. Lines show moving average for general trends (flatent = 0.9–0.1, left to right). Individual simulation results are shown in the Supplement (Additional file 1: Figure S2). (B) The minimum latency fraction on an affected branch to achieve 95% sensitivity as a function of the mean height of non-latent taxa from the MRCA of all taxa. The height is in log10-units to facilitate reading short tree height performance. (C) Specificity as a function of the mean height of non-latent taxa from the MRCA of all taxa. Panels B and C have loess curves fitted to show general trends.
© Copyright Policy - open-access
Related In: Results  -  Collection

License 1 - License 2
getmorefigures.php?uid=PMC4201670&req=5

Fig1: Success of phylogenetic latency detection in critical simulation scenarios. Detection of latency-affected lineages was done by pair-wise comparisons of all taxa, testing whether the shorter lineage was significantly shorter according to a Poisson test. (A) Sensitivity results from simulations with 100 taxa and one random branch affected by latency at flatent = 0.1–0.9 of the corresponding original non-latent genetic distance. Lines show moving average for general trends (flatent = 0.9–0.1, left to right). Individual simulation results are shown in the Supplement (Additional file 1: Figure S2). (B) The minimum latency fraction on an affected branch to achieve 95% sensitivity as a function of the mean height of non-latent taxa from the MRCA of all taxa. The height is in log10-units to facilitate reading short tree height performance. (C) Specificity as a function of the mean height of non-latent taxa from the MRCA of all taxa. Panels B and C have loess curves fitted to show general trends.

Mentions: To explore the limitations of our method, we tested its performance in critical simulations. Figure 1A shows simulation results from random phylogenies with 100 taxa each, where one random branch was latent for a fraction of its evolutionary time (flatent). Note that in each simulated phylogeny, the length and position of the latent branch may differ, and thus also the number of taxa affected by latency (range 1—(N-1) latent taxa). Overall, we were able to recover the lineages affected by latency even when flatent approached small values; for branches >0.02 subst/site and flatent > 0.4 we observed >90% sensitivity (Figure 1A). Because estimating the mean height of non-latent taxa from the MRCA of all taxa per time point is easy in real data, we also simulated trees varying the tree height and identifying the minimum proportion of latency in lineages that were detected as latent to achieve 95% sensitivity (Figure 1B). This means that, for instance, if the mean non-latent height in a sample was 0.1 substitutions/site, we can detect at 95% sensitivity latency periods as short as 7% of the non-latent genetic distance, while in a population with small divergence of 0.01 substitutions/site height, the affected lineages must have been latent for at least 60% of the evolutionary time. Clearly, while it is easy to detect latency on longer branches, the effect of latency on short branches is small and therefore also difficult to detect. The specificity of our method was at an overall 98% for genetic distances at 0.1 substitutions/site, and decreased slowly to approximately 70% at 0.25 substitutions/site (Figure 1C), a distance one rarely detects in a within-patient single time point sample (Additional file 1:Table S2).Figure 1


Reduced evolutionary rates in HIV-1 reveal extensive latency periods among replicating lineages.

Immonen TT, Leitner T - Retrovirology (2014)

Success of phylogenetic latency detection in critical simulation scenarios. Detection of latency-affected lineages was done by pair-wise comparisons of all taxa, testing whether the shorter lineage was significantly shorter according to a Poisson test. (A) Sensitivity results from simulations with 100 taxa and one random branch affected by latency at flatent = 0.1–0.9 of the corresponding original non-latent genetic distance. Lines show moving average for general trends (flatent = 0.9–0.1, left to right). Individual simulation results are shown in the Supplement (Additional file 1: Figure S2). (B) The minimum latency fraction on an affected branch to achieve 95% sensitivity as a function of the mean height of non-latent taxa from the MRCA of all taxa. The height is in log10-units to facilitate reading short tree height performance. (C) Specificity as a function of the mean height of non-latent taxa from the MRCA of all taxa. Panels B and C have loess curves fitted to show general trends.
© Copyright Policy - open-access
Related In: Results  -  Collection

License 1 - License 2
Show All Figures
getmorefigures.php?uid=PMC4201670&req=5

Fig1: Success of phylogenetic latency detection in critical simulation scenarios. Detection of latency-affected lineages was done by pair-wise comparisons of all taxa, testing whether the shorter lineage was significantly shorter according to a Poisson test. (A) Sensitivity results from simulations with 100 taxa and one random branch affected by latency at flatent = 0.1–0.9 of the corresponding original non-latent genetic distance. Lines show moving average for general trends (flatent = 0.9–0.1, left to right). Individual simulation results are shown in the Supplement (Additional file 1: Figure S2). (B) The minimum latency fraction on an affected branch to achieve 95% sensitivity as a function of the mean height of non-latent taxa from the MRCA of all taxa. The height is in log10-units to facilitate reading short tree height performance. (C) Specificity as a function of the mean height of non-latent taxa from the MRCA of all taxa. Panels B and C have loess curves fitted to show general trends.
Mentions: To explore the limitations of our method, we tested its performance in critical simulations. Figure 1A shows simulation results from random phylogenies with 100 taxa each, where one random branch was latent for a fraction of its evolutionary time (flatent). Note that in each simulated phylogeny, the length and position of the latent branch may differ, and thus also the number of taxa affected by latency (range 1—(N-1) latent taxa). Overall, we were able to recover the lineages affected by latency even when flatent approached small values; for branches >0.02 subst/site and flatent > 0.4 we observed >90% sensitivity (Figure 1A). Because estimating the mean height of non-latent taxa from the MRCA of all taxa per time point is easy in real data, we also simulated trees varying the tree height and identifying the minimum proportion of latency in lineages that were detected as latent to achieve 95% sensitivity (Figure 1B). This means that, for instance, if the mean non-latent height in a sample was 0.1 substitutions/site, we can detect at 95% sensitivity latency periods as short as 7% of the non-latent genetic distance, while in a population with small divergence of 0.01 substitutions/site height, the affected lineages must have been latent for at least 60% of the evolutionary time. Clearly, while it is easy to detect latency on longer branches, the effect of latency on short branches is small and therefore also difficult to detect. The specificity of our method was at an overall 98% for genetic distances at 0.1 substitutions/site, and decreased slowly to approximately 70% at 0.25 substitutions/site (Figure 1C), a distance one rarely detects in a within-patient single time point sample (Additional file 1:Table S2).Figure 1

Bottom Line: The method removes alternative sources that may affect evolutionary rates, such as hypermutation, recombination, and selection, to reveal the contribution of generation-time effects caused by latency.Furthermore, we discovered extensive effects of latency in sequence data (gag, pol, and env) from reservoirs as well as in the replicating plasma population.These results suggest that cycling in and out of latency plays a major role in the evolution of HIV-1.

View Article: PubMed Central - PubMed

Affiliation: Theoretical Biology & Biophysics, Los Alamos National Laboratory, Los Alamos, NM, 87545, USA. tti@lanl.gov.

ABSTRACT

Background: HIV-1 can persist for the duration of a patient's life due in part to its ability to hide from the immune system, and from antiretroviral drugs, in long-lived latent reservoirs. Latent forms of HIV-1 may also be disproportionally involved in transmission. Thus, it is important to detect and quantify latency in the HIV-1 life cycle.

Results: We developed a novel molecular clock-based phylogenetic tool to investigate the prevalence of HIV-1 lineages that have experienced latency. The method removes alternative sources that may affect evolutionary rates, such as hypermutation, recombination, and selection, to reveal the contribution of generation-time effects caused by latency. Our method was able to recover latent lineages with high specificity and sensitivity, and low false discovery rates, even on relatively short branches on simulated phylogenies. Applying the tool to HIV-1 sequences from 26 patients, we show that the majority of phylogenetic lineages have been affected by generation-time effects in every patient type, whether untreated, elite controller, or under effective or failing treatment. Furthermore, we discovered extensive effects of latency in sequence data (gag, pol, and env) from reservoirs as well as in the replicating plasma population. To better understand our phylogenetic findings, we developed a dynamic model of virus-host interactions to investigate the proportion of lineages in the actively replicating population that have ever been latent. Assuming neutral evolution, our dynamic modeling showed that under most parameter conditions, it is possible for a few activated latent viruses to propagate so that in time, most HIV-1 lineages will have been latent at some time in their past.

Conclusions: These results suggest that cycling in and out of latency plays a major role in the evolution of HIV-1. Thus, no aspect of HIV-1 evolution can be fully understood without considering latency - including treatment, drug resistance, immune evasion, transmission, and pathogenesis.

Show MeSH
Related in: MedlinePlus