Beyond CYP2D6: SMN1, GBA and PMS2, where short reads fail

CYP2D6 is the textbook case of a locus where a standard short-read pipeline fails without warning. It is not the only one. Any gene that sits beside a near-identical pseudogene or paralogue creates the same conditions: reads that could come from either copy, ambiguous placement, and a caller that either miscounts or stays silent. Three genes show three different ways this goes wrong.

SMN1: the copies are the answer

Spinal muscular atrophy is mostly caused by the absence of functional SMN1. The difficulty is that SMN1 and SMN2 are almost identical, so the clinically relevant question is not “is there a variant?” but “how many copies of SMN1 are there?”. Read depth over a paralogous pair cannot answer that reliably, and a variant caller has nothing to call when the problem is a missing gene. It needs dedicated copy-number calling.

Even then, a normal copy number is not the end of the story. Some carriers have two copies of SMN1 on one chromosome and none on the other, so a normal count does not equal “not a carrier”. A responsible report states the residual risk instead of implying it is zero.

GBA: recombinants between the gene and its pseudogene

GBA lies next to GBAP1, a pseudogene with high sequence similarity. Some pathogenic alleles are recombinants, in which part of the gene has been replaced by pseudogene sequence. Short reads from such an allele are placed on the pseudogene or on the gene depending on the aligner, so the signal is split and a standard pipeline can see neither half convincingly. Resolving it needs locus-specific analysis across the homology region rather than gene-by-gene calling.

PMS2: when the honest answer is “not evaluable”

PMS2, relevant to Lynch syndrome, has a pseudogene that shares more than 95% identity with part of the gene, including exons 9 and 11 to 15. In those exons, short reads cannot reliably say which copy they come from. There is no clever caller that fixes this: the information is not in the reads. What a platform can do is say so, marking those exons as not evaluable on short-read data instead of returning an empty, normal-looking result.

Three genes, one rule

In the first case the answer is a copy number, in the second a haplotype across a homology region, and in the third a declared limit. What they share is the rule that governs the rest of the platform: a result states its boundary. What was evaluated, what could not be, and why must be visible to the geneticist who signs, because the alternative is a normal that nobody computed.