Alternative genetic codes in bcftools csq for mitochondrial variant annotation

[etwatson]

Hello,

I’ve been using snpEff for variant effect prediction and find it very useful. However, I have noticed some instances of multi-nucleotide polymorphisms that nullify effect predictions based on single nucleotides - so I have become very interested in using bcftools csq in order to perform this analysis in a haplotype-aware manner.

I frequently analyze mitochondrial variants, which often require the use of alternative genetic codes. Fun mitochondria in fungi that contain mobile introns! From what I can tell, bcftools csq does not currently support specifying different genetic codes for annotation.

Would it be possible to add an option to allow users to specify an alternative genetic code when running bcftools csq? This would be particularly helpful for those working with mitochondrial genomes, where the standard nuclear genetic code does not apply.

If there is already a workaround for this, I would appreciate any guidance. Otherwise, I’d love to hear if this could be considered as a feature request.

Thanks for your time, and I appreciate all the work that goes into maintaining bcftools!

[pd3]

It is not supported but we can think about it, if you can provide detailed descriptions with specific examples.

[etwatson]

That is excellent. There are 33 well established alternative genetic codes as described by Anjay Elzanowski and Jim Ostell at the National Center for Biotechnology Information (NCBI). Anyone working on these organisms, especially on mitochondrial or chloroplast genetics, would be very appreciative.

As I mentioned, in snpEff, you can set the mitochondrial codon table in the config file, and even make changes to it, since there are some lineage specific codon tables. For example AUA and AUU are initiation codons in human mitochondria, while AUC also acts as an initiation codon in mice mitochondria. However, as you know, snpEff is limited to single nucleotide effect predictions, so it would be excellent if we could use bcftools csq for predicting variant effects.

For some examples here are a few of the major differences from the standard code:

Codon	Alternate	Standard	Taxonomy	Translation Table
AGA	Ter *	Arg R	Human mito	2
AGG	Ter *	Arg R	Human mito	2
AUA	Met M	Ile I	Human mito	2
UGA	Trp W	Ter *	Human mito	2
AUA	Met M	Ile I	Yeast mito	3
CUU	Thr T	Leu L	Yeast mito	3
CUC	Thr T	Leu L	Yeast mito	3
CUA	Thr T	Leu L	Yeast mito	3
CUG	Thr T	Leu L	Yeast mito	3
UGA	Trp W	Ter *	Yeast mito	3
UGA	Trp W	Ter *	Mold/Fungi/Protozoan mito	4
AGA	Ser S	Arg R	Invertebrate mito	5
AGG	Ser S	Arg R	Invertebrate mito	5
AUA	Met M	Ile I	Invertebrate mito	5
UGA	Trp W	Ter *	Invertebrate mito	5

[pd3]

I can imagine adding a new switch to use an alternative genetic code.
What are the numeric values in the Translation Table column?

[etwatson]

That's great. The numeric values in the translation table column represent the annotation given to a specific genetic code by the International Nucleotide Sequence Database Collaboration. Most programs that utilize a switch for alternative genetic codes will use this standardized number to distinguish from the standard genetic code (transl_table=1).

[pd3]

Thank you. Yes, we can add that.

One question though, I am not familiar enough with the translation. Looking at the tables, I see there are non-standard start codons, even in the standard table. How prevalent is it? It raises a question, should the variant, say, ATG(M) > CTG(L) be considered as start_retained?? I hope not, it would be a complication.

[etwatson]

Yes, it complicates things. As you mention, with the standard genetic code (translation table 1), AUG is the canonical start codon, but alternate start codons like UUG, CUG, AUU, AUA, and GUG can also be used in specific contexts to initiate translation depending on the context surrounding the start codon. One example is the mammalian NAT1 gene in which the non-canonical start codon GUG is evolutionarily conserved.

In your example with ATG(M) > CTG(L), initiation is still possible, but it is highly context dependent. The sequence flanking the start codon (especially at −3 and +4) heavily affects ribosome recognition. A strong Kozak sequence (e.g., GCCA/GCCCUGG) might compensate somewhat for the lack of ATG. But this sort of thing requires functional study to confirm. We are talking about in silico predictions of variant effects, so I think it is very likely to be "start_lost".

However, CTG(L) > ATG(M) in a gene known to be initiated by a non-canonical start codon would be "start_retained", since the initiator tRNA will add a methionine at the correct position. In prokaryotes, the initiator tRNA (tRNA-fMet) carries a formyl-methionine and can pair with many non-canonical start codons better than the tRNA-Met in eukaryotes, albeit at different efficiencies. However, I should note that a prokaryotic gene that undergoes a CTG(L) > ATG(M) mutation will likely increase in expression as ATG is a more efficient initiator of translation - in experimental systems, researchers often alter the start codons to tune expression.

All of this is why NCBI/GenBank includes a translation table option in the CDS region of the GenBank file since the standard code does not work with prokaryotes.

[pd3]

OK, I added the feature, please try it out.