HIV work

Presentation SMBE 2017

I gave a presentation today in Austin at SMBE 2017. Here is a link to the (hand drawn) slides:


Screen Shot 2017-07-05 at 6.45.04 PM

Selective sweeps in 24 years of HIV sequence data

Together with several collaborators at Stanford, I published a paper on the BioRxiv on selective sweeps in HIV. We find that when treatments didn’t work well (think AZT in the 1980s), sweeps were very soft, but with better treatments sweeps are getting “harder.” The first author on the paper, Alison Feder, has done most of the work on this paper including making all the figures (and they are cool!).

Alison F Feder, Soo-Yon Rhee, Robert W Shafer, Dmitri A Petrov, Pleuni S Pennings. 2015. More efficacious drugs lead to harder selective sweeps in the evolution of drug resistance in HIV-1. BioRxiv, doi:


In the early days of HIV treatment, drug resistance occurred rapidly and predictably in all patients, but under modern treatments, resistance arises slowly, if at all. The probability of resistance should be controlled by the rate of generation of resistant mutations. If many adaptive mutations arise simultaneously, then adaptation proceeds by soft selective sweeps in which multiple adaptive mutations spread concomitantly, but if adaptive mutations occur rarely in the population, then a single adaptive mutation should spread alone in a hard selective sweep. Here we use 6,717 HIV-1 consensus sequences from patients treated with first-line therapies between 1989 and 2013 to confirm that the transition from fast to slow evolution of drug resistance was indeed accompanied with the expected transition from soft to hard selective sweeps. This suggests more generally that evolution proceeds via hard sweeps if resistance is unlikely and via soft sweeps if it is likely.

Figure 3

Figure 3: Drug resistance mutations are correlated with diversity reduction differently in different types of treatments. Treatment efficacy from literature review (% of patients with virologic suppression after 48 weeks) showed positive correspondence with clinical recommendation among RTI regimens (A) and PI+RTI regimens (B). DRM SE lower among the more efficacious and clinically recommended treatments among RTI treatments (C) and RTI+PI treatments (D). Mixed effect model shows significantly different slopes for NNRTI treatments versus NRTI treatments (E) and PI/r treatments versus PI treatments (F). Each line in (EF) represents the fitted decay in diversity with each DRM for a different treatment from the full mixed effects model and p-value labeling indicates the difference between the plotted full model and the null not fitting slopes separately for treatment groups.

Review on HIV drug resistance published on arXiv

See here for the blog post. And here for the paper.

Screen shot 2012-11-26 at 6.02.30 PM

The slides of my talk at the Evolution meeting in Ottawa 2012

The slides can be downloaded here: 2012_07penningsevolution.pdf

I am particularly fond of this picture of a selective sweep.

Selective sweep in HIV. An NNRTI drug resistance mutation (K103N) goes to fixation in patient virus. The mutation (A to T) apparently occurred on one haplotype (genetic background) and as it went to fixation, genetic variation on was lost. Not all sweeps in HIV are hard sweeps, see next figure.

Selective sweep in HIV. An NNRTI drug resistance mutation (K103N) goes to fixation in patient virus. The mutation (A to T) apparently occurred on one haplotype (genetic background) and as it went to fixation, genetic variation on was lost. Not all sweeps in HIV are hard sweeps, see next figure.

Soft selective sweep in HIV. The K103N substitution is caused by an A to T or A to C mutation. In this patient both alleles are present. A clear example of a "multiple-origin soft sweep".

Soft selective sweep in HIV. The K103N substitution is caused by an A to T or A to C mutation. In this patient both alleles are present. A clear example of a “multiple-origin soft sweep”.

The data I talked about in Ottawa came from this study: Bacheler et al 2000.

Bacheler LT, Anton ED, Kudish P, Baker D, Bunville J, Krakowski K, Bolling L, Aujay M, Wang XV, Ellis D, Becker MF, Lasut AL, George HJ, Spalding DR, Hollis G, Abremski K. Human immunodeficiency virus type 1 mutations selected in patients failing efavirenz combination therapy. Antimicrob Agents Chemother. 2000 Sep;44(9):2475-84.

June 2012 The Scientist has a story about my PLoS Comp Bio paper


You can read the story on The Scientist’s website. The story explains what I have done and then asks what others from the field think of it, which shows that some people agree with me and some don’t. Here are some quotes from the piece by Sabrina Richards:

“The paper is interesting, and may be important for getting scientists to think about evolution of drug resistance,” said Andrew Read […].

Not everyone agrees that one mutation can establish drug resistance. “I feel it is too simplistic and not entirely plausible” that one mutation would be enough for multi-drug therapy to fail, said Roger Kouyos.

[…] But this simplicity may be good, argued Sergei Kosakovsky Pond […].  If models get too complicated, he explained, “you can make them do almost anything” by plugging in the right numbers.

Andrew Read is at Penn State, Roger Kouyos is at the ETH in Zurich and Sergei Kosakovsky Pond is at UC San Diego. 

June 2012 Paper on standing genetic variation and HIV drug resistance out in PLoS Comp Biol

Have a look here for the paper. And have a look here for my video abstract on Vimeo.

Standing genetic variation and the evolution of drug resistance in HIV from Pleuni Pennings on Vimeo.

“Behind” the PLoS Comp Biol paper

Why I study HIV drug resistance

HIV drug resistance is a major problem and it is an evolutionary problem. So you’d think that many evolutionary biologists work on it. But no. Surprisingly few evolutionary biologists work on drug resistance (see this interesting paper on why). But if we want to prevent the evolution of resistance, we need to understand how it evolves. One major question is whether drug resistance typically evolves from pre-existing mutations or from new mutations. 

HIV is also a very cool system to do evolutionary biology. I really enjoy looking at all the data that are available for HIV and it makes me very happy to have data for multiple patients and multiple timepoints. 

Standing genetic variation leads to drug resistance in 6% of patients

I used models from evolutionary biology and data from clinical trials to determine the role of pre-existing mutations for the evolution of drug resistance in patients who are treated with NNRTI-based anti-retroviral treatment. I found that pre-existing mutations lead to drug resistance in 6% of the patients who start treatment (have a look at this video to see how I estimated this number). 6% may not sound as a lot, but worldwide, this affects a lot of people. 

Why 6% is very relevant

In 2010, around 1.4 million HIV infected people started antiretroviral therapy (data from WHO website). Most of those patients are in low and middle-income countries. They usually start treatment with a first-line treatment, as recommended by WHO: a combination of an NNRTI plus two NRTI’s. If the results from my paper can be extrapolated to these patients, then in 6% of the 1.4 million patients, drug resistance will evolve due to pre-existing mutations. That means that after just a few months of treatment, the first-line drugs do not work anymore in 84000 patients per year. First-line drugs are the drugs that are cheapest & easiest to take (one pill a day). If first-line drugs no longer work, then the patients need to take second-line drugs, which may not even be available everywhere. 

Treatment interruptions

There was a time when HIV researchers thought that treatment interruptions may be good for patients. From this time (roughly 2000-2006), there are data available from clinical trials in which patients followed a schedule of “structured treatment interruptions”. I used these data to determine the probability that interruptions lead to resistance, and I found that long interruptions lead to the evolution of 6% of the patients when they restart treatment. Shorter interruptions lead to less resistance.

Starting therapy for the first time and starting therapy after a long treatment interruption are both associated with a probability of evolution of drug resistance of 6%. My analysis suggests that it is not a coincidence. In both cases, the population size of the virus is at the same level and it is the population size that determines to a large extent the probability that pre-existing resistance mutations are present.   

Pregnant women

When I submitted the paper for the first time, I had only looked at patients starting treatment and interrupting treatment. But one of the reviewers said that data on treatment of pregnant women showed that my analysis was clearly wrong. So I started looking at the evolution of drug resistance in women treated with “single dose nevirapine” treatments. Such treatment is so short that all resistance that comes up is probably due to pre-existing mutations. A perfect setting for testing my ideas on pre-existing mutations!

It turned out that the data on pregnant women fitted almost perfectly to the model I had proposed. So the paper became longer, but the main message stayed the same. In addition, now I had a way to estimate the effect of using 3 vs 1 drugs on the establishment probability of pre-existing mutations. Plus, maybe even more importantly, the data on pregnant women showed that starting treatment with a few weeks of AZT monotherapy can reduce the probability that nevirapine resistance evolves when treatment with nevirapine (or nevirapine plus two other drugs) is started. Starting treatment with AZT monotherapy may be a useful strategy for other patients too. 

Drug resistance due to pre-existing mutations is preventable

I believe that drug resistance due to pre-existing mutations is preventable. We need to find out what is the best way to start treatment in order to make sure that the pre-existing mutations do not get a chance to become established and lead to drug resistance. 

Figure from the paper


This figure shows the probability that resistance evolves due to a treatment interruption depending on the length of the interruption. Data are from 7 STI studies (STI = structured treatment interruption).

Successful treatment reduces the availability of drug resistance mutations. When treatment is interrupted, the viral population size grows, so that more variation will be available. When treatment is started again, selection can work on the (now) pre-existing mutations.

After a long interruption, the probability that pre-existing drug resistance mutations become fixed is the same as when treatment is started for the first time. This suggests that the population is back at its equilibrium, at least with respect to drug resistance mutations.

HFSP funding

I have a three-year postdoctoral fellowship (June 2010 until June 2013) from HFSP. Here is a short description of the project (from my grant application).

Drug resistance is a major problem in diseases such as cancer, malaria and HIV and despite strategies to avoid resistance (such as combination therapy) resistance still occurs and many questions remain. One question is why non-adherence (when a patient doesn’t takes his drugs as prescribed) sometimes increases and sometimes decreases the risk of evolution of resistance. In principle, non-adherence can both slow down the evolution of resistance, because selection pressure decreases, or it can speed up the evolution of resistance, because the evolving population (e.g., the virus population) will grow, leading to more genetic variation on which selection can act. Whether the net result of non-adherence is slower or faster evolution must therefore depend on factors such as drug or virus characteristics, but it is not yet understood how.
HIV is an ideal model to study the effect of the level of adherence on the evolution of drug resistance for two reasons: 1) because a patient usually gets infected with HIV once and patients stay infected for life. An HIV strain therefore evolves within a patient for many years, independently from HIV strains in other patients. Every patient therefore represents an independent evolutionary history of a viral population. 2) The second reason for studying HIV is that there are good data available on patient adherence, on drug characteristics, and resistance mutations. I will combine stochastic modeling approaches and data-analysis to reach a better understanding of adherence-resistance relationships. The results will allow for design of better treatment strategies.

One Response to “HIV work”


  1. Older news « Pleuni Pennings - November 2, 2012

    […] HIV project […]

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: