Stop-Gain Mutation Predicted As Benign By Sift/Polyphen2?

Question

[michealsmith]

I've found a stop-gain mutation(in coding region) very close to the 5' end of mRNA (the second exon while the whole protein may have 10+ exons) in my patient sample, which means the majority of the protein is NOT translated, making this mutation at the top of my disease-causing gene list. (Translation starts from 5' mRNA, right?) However, when I predict deleteriousness using SIFT and Polyphen2, both algorithms show it's benign, which puzzles me a lot.

If you look at the score in database:

SIFT:
A       C      0.15      K     Q
A       G      0.01      K     E
A       T      0.15      K      *

K to *, SIFT score=0.15, which means benign; while K to E single mutation seems much more deleterious!

Polyphen2:
A        C        0.879        D
A        G        0.958        D
A        T        0.691        NA

The same here that stop-gain mutation is the least deleterious.

I'm wondering, is Polyphen2/SIFT designed to cover stop-gain/loss mutation? And generally speaking is such prediction always reliable? (I came across such problem before, some very well-studied/annotated deafness-causing SNP is predicted by both algorithms to be benign.)

Also, biologically speaking, is it possible that stop-gain/loss mutation can be not deleterious at all? I check this gene, which is highly conserved across species and none of discovered SNP/indel is stop-gain/loss.

Thanks!

Answer 1

[DG]

I understand possible reluctance to name the gene as this is likely a research project, and I don't think PolyPhen or SIFT takes this in to account, but yes there are plenty of genes in which a Loss of Function (LOF) variant is perfectly benign. There has been some work cataloging this, but the obvious examples are things like Olfactory Receptors and such. Genes were we have lots of in-paralogs are often also able to be rescued by a similar enough paralog in the genome.

Anyway as for your particular problem, I never use PolyPhen or SIFT to categorize stop-gain and stop-loss, only predict the possible impact of amino acid substitutions. Generally we categorize Stop-Gains and Losses on their own anyway as being very likely to be disease causing.

If you are doing an automated prioritization of variants you need to program something more robust than just sorting PolyPhen/SIFT output to categorize appropriately.

[Bpow]

I reiterate that LOF of one allele does not always lead to a phenotype, it depends on whether haploinsufficiency of that gene is sufficient to affect its function. In some cases, the loss of an allele can be less severe than a missense mutation.

The example that springs to mind is mutations of COL1A1 leading to osteogenesis imperfecta. There is some genotype-phenotype correlation. Often a nonsense mutation or frameshift leading to nonsense mediated decay (a "null allele") will have a less-severe phenotype (due to reduced production of collagen. On the other hand, substitution of a glycine in a Gly-X-Y repeat can lead to a dominant negative effect where the abnormal protein product interferes with multimer formation. (for another description, see section 16.6.4 of http://www.ncbi.nlm.nih.gov/books/NBK7574/)

[DG]

Even a homozygous loss of function, where the whole gene is lost, can be benign. Again, Olfactory receptors being a prime example

[michealsmith]

Thanks. So basically Polyphen/SIFT is not designed for nonsense mutation? I'm using Annovar pipeline, which actually defines some gene as "dispensable" because of frequent stop-gain/loss mutations withint that gene. I've checked my gene, almost no reported SNPs are stop-gain/loss, and the whole gene is away from segmental duplication

[DG]

I'd also cross-check against local sequencing projects and not just SNP databases. But yes I wouldn't generally rely on PolyPhen/SIFT values for nonsense mutations, just to prioritize missense mutations.

Answer 2

[Matt Shirley]

Is your stop codon mutation located in the 5' untranslated region (UTR) of the mRNA sequence? If so, then the stop codon will make no difference, as translation initiates downstream from it. Both of these programs, I believe, consider whether a snp is located in a coding region of a gene.

[michealsmith]

No, sorry, it's located within the coding region of mRNA, just very close to 5' end (but not 5'UTR!)....Since translation starts from 5' end of mRNA, then stop codon very close to 5' end will truncate the protein to nearly nothing, right?

[Matt Shirley]

When you say that translation starts at the 5' end of the mRNA you are not entirely accurate. Translation starts from an AUG codon. If your mutation is before this codon, I doubt it will have any effect. I'm not saying this is what you are seeing, but just pointing it out as a possibility.