I guess it is hard to publish something like: "We sequenced it but it had no effect". Many amino acid substitutions have very low impact on protein function, this why everyone of us has such a high degree of variation in it's genome, also in coding regions. 

For example: In exome sequencing (I've fished one example of an exome seq study at our lab) studies we usually see ~20,000 mutations per human genome in coding regions. While some of them do not lead to an amino acid substitution, about 50% actually lead to a different amino acid, and of course it's impossible that all of them have an effect. Especially when the original and the substitute amino acid share characteristics (hydrophobicity, polarity, charged, ...).

However, I have no publication at hand that has tested this for a specific amino acid (but my guess are that there are plenty)

sure, where you can find out of 60k variants in a sample's exome, the goal of the analysis is to find effective mutations. try the journal os negative results and maybe you'll find some examples (https://jnrbm.biomedcentral.com).

fred

Yes, I figured a negative result issue may be part of it, but if you are, for example, testing or verifying a way of predicting which AA changes are important, wouldn't you want to test some that are predicted to be neutral, too?

of course all variants are interesting and should be analyzed but the amount of work and money command you to apply filters to focus on short lists. among all filters you can apply, there are lot of prediction tools as SIFT, but I guess you know them.

fred

There is an example in a melon TILLING population in which there are two different mutations in a gene only one of which causes the a mutant phenotype (firm flesh). http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0015776

There are examples in medical case-studies of the type "We found three patient mutants, and this one 1 is the one that actually does anything". In such cases positive and negative results are in the same paper.