Someone suggested that the ratio between mean non-synonymous substitutions per non-synonymous site (dN) and mean synonymous substitutions per synonymous site (dS) should be calculated considering at least three sequences from each of the species. However, the explanation given was not understood by me. Can anyone please elucidate why should we consider multiple sequences from each species while calculating dN/dS?
Another query is, what if only one sequence is available for some of the species for that particular coding sequence under study?
I guess the more sequences you have, the more confidence you have in your results, especially if you are dealing with means. With one sequence you cannot tell if there are differences between dN or dS in different sequences for the same species. But if you have only one sequence for a species, you just have to go with the assumption that that sequence is uniform irrespective of the number of sequences, and that there are no mutations that may change your dN/dS ratio.
A single sequence from one species has to be compared to sequences from other species. You cannot calculate changes from one sequence. It has often been claimed that a dN:dS ratio of 1.0 indicates neutral drift or no selection pressure on a protein, but nearly all proteins are under negative and/or positive selection pressures with some codons being selected for changes while other codons are selected to remain constant. The HIV-1 envelope protein, for example, has quite a few absolutely invariant sites and some sites that evolve very rapidly due to host immune selection pressure.
If there is a 3D structure known for your protein, or a very similar protein (similar enough to do a sequence alignment of your sequence to the 3D structure sequence) you can use the ConSurf tool (http://consurf.tau.ac.il/overview.html ) to color the 3D structure by rate of evolution. For most proteins, the key catalytic sites are invariant while the rest of the molecule is free to evolve.
Even within a species, you may need to carefully consider how you want to calculate the scores for the data. If you have ten sequences from each of two isolated sub-populations, for a hypothetical example, and one population has GAA = Glu for one codon and the other has GAG = Glu for that same codon, it is more likely that this was a single G to A or A to G mutation in one of the founders of one of the populations, than many independent events. The phylogenetic methods like PAML as discussed by Adam Retchless, aim to help with this. But the calculation may assume that the data came from random sampling of a single well-mixed population. Very often, the data we have to work with violates the simple assumptions of the models used to analyze the data. This does not male the analyses worthless, but does mean that we should carefully consider how the results may be biased by such things as non-random sampling, lack of mixing in the population, etc...
I agree with Adam that intraspecific variation/polymorphism may behave differently from interspecific /divergence one. So that having several sequence from a species is a way to set appart fixed differences from polymorphisms. However it would be relevant only if you do a branch specific dN:dS analyses. Moreover, intraspecific dN:dS component may be largely affected by recombination, since methods assume a single tree. As an alternative to PAML, I recommend datamonkey, the user friendly servor HYPHY that is also more flexible.
Replicates within species give a timescale for the type of selection that you are observing. If your gene is like MHC, with balancing selection maintaining lots of variation within species, as well as adaptive evolution among taxa, variants within species should also show the effect, provided sample size is large enough. In contrast, deep comparisons can start to show loss of significance, as synonymous substitutions start to accumulate while AA substitutions start to saturate. I think the best solution is to analyse character state change in a phylogenetic context, breaking up any long branches as far as is possible.
We have been examining very large numbers of haplotypes for mt sequences in some very recently diverged species. We find that synonymous changes are very frequent within species and nonsynonymous changes are mostly restricted to species level divergences. This suggests purifying selection within species and perhaps some adaptive changes surrounding speciation. So, in terms of your original question, you might get some interesting data if you sequence multiple specimens. But as other commentators have noted, if you have only limited sequence data you might be unlucky to encounter bp changes that have not yet been subjected to selection. Also, be aware that inferences of dN/dS depend strongly on your ability to infer ancestral states. A small number of outgroup sequences will compromise this ability. You need to step carefully here and be aware of the limitations that small data sets impose. More is better, so get as many sequences as you can.
- Badges
- Science topic
- Similar topics
- Biological Science
- Evolutionary Biology
- Phylogenetics