I am not sure you can find something about selective pressure by site doing pure pairwise analysis (runmode = -2), this is usually used for global distances.
On the other hand doing site analysis you can get this info by site, but it is generally a bad idea to use this result when analyzing less than 4 sequences (if you try it, have a look to associated errors of the estimations).
have a look to this post:
an other program that can do this for you is SLR http://www.ebi.ac.uk/goldman/SLR/ but I am afraid that you will find same restrictions...
I think the best solution, if you can, is to group your sequences by 4.