I have sequenced a gene, mutations in which confer resistance to an antibiotic treatment, from the bacteria that are able to grow in the presence of antibiotic. The bacterial clones all share the same genetic background (i.e. closely related) and there are around 60 sequences in total, all of which have non-synonymous substitution(s). I would like to test whether there is evidence of positive/negative selection in this alignment. I used Datamonkey server to perform SLAC, FEL, REL and PARRIS analyses (using an automatic model selection tool) and MEGA Tajima’s Test of Neutrality. SLAC and FEL found no selected sites, while REL found that 230/385 sites were positively selected and PARRIS shows that there is an overall evidence of positive selection. Tajima’s Neutrality Test has a D value of -2.4, indicating a recent selective sweep. Considering that SLAC is a very conservative method, while REL tends to produce high rates of false-positives, I have a hard time interpreting the results. Taking into account the experimental setup, my expectation is that in a population that was under a strong selective pressure (antibiotic treatment) and went through a genetic bottleneck (as Tajima D demonstrates), one could probably expect to have evidence of diversifying selection. Can anyone advise on what method would be the most appropriate to use in this situation or help with interpretation?
Dear Olga Makarova,
You can ask this question to some scientists of population genetics. I just suggest the two names and their email id is as following
both people are from Ludwig Maximilians University Munich
Dear Olga,
I think you should apply the DH test which is a combination of the Tajima D test and the normalized Fay and Wu's H-test. DH test has been shown to be very effective in detecting positive selection. In addition, it is sensitive primarily to directional selection and no other driving forces. For more details on the test, please find attached the paper of Zeng et al. 2006.
Good luck!
I have used HYPHY and the datamonkey server to assess selection among bacterial populations, namely cholera and S. aureus. When researchers publish the result from datamonkey (SLAC, FEL, IFEL, and REL), they usually show a comparison of all the sites/branches where selection was significant and then discuss sites which were identified consistently across all methods. However, if you read the methods papers by Pond and Murrell, MEME and FUBAR are actually preferred to those other methods (as they are reportedly better performing extensions for assessing site and branch specific selection). There are also another approach you can take. You can download HYPHY and then run compartmentalization tests (e.g. tree correlation coefficient) to identify whether clades on the phylogeny cluster by a phenotype (i.e. Abx resistant vs susceptible). This may give you a better idea about the population structure based on a trait.
Overall, it may better if we knew more about your dataset. Are the genes from isolates from a single geographic area, hospital unit, laboratory strain, or just a random sample from NCBI? Do you expect there to be selection in the sample? Last, while comparisons of dN/dS were developed to assesses interspecies selection, you note the limitation of comparing intraspecies isolates. However, I don't know how this translates to looking at a single intraspecies gene. Perhaps the selection occurs outside of the gene due to the metabolic cost the resistance mechanism, for example. Maybe time for some WGS?
Thanks a lot for a very detailed reply, Taj!
I have just tried MEME and FUBAR, as you suggested, MEME finds 0 sites under episodic diversifying selection and FUBAR finds 24 sites under pervasive diversifying selection. I will give a try to HYPHY compartmentalization tests too.
Regarding the dataset, the sequences come from the same lab strain, which was originally susceptible to the antibiotic, but for the purpose of the experiment was exposed (in the presence of antibiotic) to several treatments some of which may promote the strain's mutagenesis. This effectively resulted in antibiotic resistance in all clones and mutations in the gene that confers that resistance with distinctive mutation hot spots differing between the treatments. These mutations are all non-synonymous (actually, there are no synonymous substitutions at all), and although we know that this gene is the only one that confers resistance (that is why went for sequencing of one gene rather than WGS), it is non-essential and there don't seem to be any significant costs associated with losing it in direct competition. My expectation is that there should be sings of diversifying selection, but I had my doubts about the methods to test it.
Thanks again for suggestions!
It seems like you have tried a lot of site specific tests of positive selection, but I'm always wary of site tests, especially the interpretation of the results. Your tajima's D results suggest a selective sweep, so I don't think you want to test for pervasive positive selection at sites, but episodic directional selection on branches. You will need to download the hyphy developer distribution from github, but look in res/TemplateBatchFiles and you will find a batch file call DirectionalREL.bf. This is a modified branch-site method to work with amino acid data for the purpose of detecting where in the tree a selective sweep took place. Sergei used this on some influenza data a few years back and compared it to FEL at the time (I think this is the correct citation: Pond et al. 2008. Mol. Biol. Evol. 25(9): 1809-1824). This is a different spin than the tests you've been trying, and you might want to compare it to your FUBAR results.
Generally, I think branch and branch-site tests are more informative, as we are generally interested in something that occurred at a specific place or places on an evolutionary tree. You should be able to test for a proportion of sites under positive selection in the antibiotic resistant strains or a branch leading up to antibiotic resistent clades with a branch-site method, and then only use site methods as a way to confirm sites on the branch or branches of interest. For me, site method results alone are very difficult to interpret and write about.
Dear Olga,
PAML is the standar software employed to test positive selection on specific residues. Inside PAML structure you will find the complement CODEML that you need to reach your goal.
It is really easy to use either Linux or windows stations. Next, you will find a full tutorial that will help you a lot with your research. Here is a link in which you will find the best tutorial about CODEML and PAML.
Best,
Andrés
http://www.molecularevolution.org/resources/activities/paml_activity
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Bioinformatics