I recently wrote a blog post about the strategies used to translate T-cell receptor nucleotides en masse and extract (what can arguable be considered) the useful bit: the CDR3.
In that talk I touched on the IMGT-definition of the CDR3: it runs from the second conserved cysteine in the V region to the conserved FGXG motif in the J. Nice and easy, but we have to remember that it's the conserved bit that's key here: there are other cysteines to factor in, and there are a few germline J genes that don't use the typical FGXG motif.
However even that paints too simple a picture, so here's a quick follow up point:
These are human-imposed definitions, based more on convenience for human-understanding than biological necessity. The fact is that we might well produce a number of TCRs that don't make use of these motifs at all, but that are still able to function perfectly well; assuming the C/FGXG motifs have function, it's possible alternative motifs might compensate for these.
I have examples in my own sequence data that appear to clearly show these motifs having been deleted into, and then replaced with different nucleotides encoding the same. Alternative residues must certainly be introduced on occasion, and I'd be surprised if none of these make it through selection; we just don't see these because we aren't able to generate rules to computationally look for these.
I actually even recently found such an example with verified biological activity: this paper sequenced tetramer-sorted HIV-reactive T-cells, revealing one that contained an alpha chain using the CDR3 'CAVNIGFGNVLHCGSG'.*
For the majority of analyses, looking for rare exceptions to rules probably won't make much difference. However as we increase the resolution and throughput of our experiments, we're going to find more and more examples of things which don't fit the tidy rules we made up when we weren't looking so deeply. If we're going to get the most out of our 'big data', we need to be ready for them
* I was looking through the literature harvesting CDR3s, which reminds me of another point I want to make. Can I just ask, from the bottom of my heart, for people to put their CDR3s in sensible formats so that others can make use of them? Ideally, give me the nucleotide sequence. Bare minimum, give me the CDR3 sequence as well as which V and J were used (and while I stick to IMGT standards, I won't judge you if you don't - but do say which standards you are using!). Most of all, and I can't stress this enough, please please PLEASE make all DNA/amino acid sequences copyable.**
** Although spending valuable time copying out or removing PDF-copying errors from hundreds of sequences drives me ever so slightly nearer to a breakdown, it does allow me to play that excellent game of "what's the longest actual word I can find in biological sequences". For CDR3s, I'm up to a sixer with 'CASSIS'.
