A boy of four with a recessive condition has one pathogenic nonsense variant in the expected gene and, after filtering, nothing on the other allele. The pipeline's default filter removed synonymous variants in the first step, as it does for every case. Among the removed variants is a rare synonymous change in the same gene, two bases from an exon boundary, which a splice predictor scores highly.
It is found eighteen months later, when a scientist, on a hunch, lifts the filter for that one gene. RNA from blood confirms exon skipping. The variant had been in the first variant file, under a label that meant it would never be looked at.
What the field knows
A synonymous variant changes a codon without changing the amino acid, and for most of the history of molecular genetics that was taken to mean it changed nothing. Sauna and Kimchi-Sarfaty (2011) reviewed the evidence that this is wrong often enough to matter: synonymous changes can alter splice sites and splicing enhancers or silencers, change messenger RNA structure and stability, and shift the speed of translation through codon usage, each of which has been shown to cause or modify human disease. Hunt and colleagues (2014) catalogued the mechanisms and the examples, and Zeng and Bromberg (2019) argued from the accumulated cases that the class deserves routine assessment rather than routine exclusion.
Splicing is the main route. A synonymous change in the last or first bases of an exon can weaken the splice site as surely as an intronic change beside it, and changes deeper in an exon can disrupt the enhancer sequences that splicing depends on. Splice predictors trained on sequence (Jaganathan et al., 2019) score synonymous variants like any other, and the ClinGen splicing recommendations (Walker et al., 2023) apply to them without distinction: a synonymous variant with a strong splice prediction is a candidate for RNA analysis, and one with a demonstrated splicing effect can be classified as pathogenic.
The share is small and the consequences are not. Across large case series the fraction of pathogenic variants that are synonymous is a few in a hundred, concentrated in recessive genes where the second allele is otherwise unexplained and in genes with well-described splicing-sensitive exons. A filter that removes all of them removes a small number of answers from a large number of cases, and the cases it affects are exactly the ones that stay open.
Codon usage is the quieter mechanism. A synonymous change to a rare codon can slow translation enough to alter how a protein folds, and changes to messenger RNA structure can alter how much protein is made. These effects are harder to predict and harder to test than splicing, and for a clinical case they mostly matter as a reason not to call a synonymous variant benign by reflex.
Where practice breaks
The filter is the first step and nobody sees past it. Removing synonymous variants is a default in most annotation pipelines, inherited from a time when the variant list had to be made small enough to read by hand. The removal happens before any scientist looks, so the scientist never sees what was removed and has no occasion to question it. A filter that is never seen is never reviewed, and this one has outlived the constraint that justified it.
The exception is left to hunches. A scientist who knows the gene, or who remembers a case, lifts the filter; the next scientist does not. The decision to look at synonymous variants in a particular gene is one of the most consequential in an unresolved recessive case and it is taken, when it is taken, by individual memory.
When a synonymous variant is kept, it is under-assessed. It arrives labelled as synonymous, which reads to a tired reviewer as benign, and the splice score beside it has to overcome the label. The same score on a missense variant would be acted on. The label does the work the evidence should.
The RNA route is missing for the same reasons it is missing for intronic variants: no standing pathway to an RNA assay, no record of whether a suitable tissue exists, and a case that closes on uncertainty rather than on a plan. The synonymous candidate has the extra disadvantage that it was never on the list, so there was no uncertainty to close on.
How Tulparus handles it
The workbench does not discard a variant class before a case has a shape. Synonymous variants are retained and scored through the same channels as every other class, with the splice predictors' outputs recorded by tool and version. In the general ranking they sit low, because most of them deserve to; in the right context they rise, because the context is what the ranking reads.
The context that matters most is the one-allele recessive case. When a hypothesis holds one established pathogenic allele and lacks a second, the search for the second allele explicitly includes synonymous variants in the gene, ordered by splice prediction and by position relative to exon boundaries, alongside copy-number, deep intronic and structural candidates. The boy's variant would have appeared on that list on the first day, with its score and its position, as a proposed candidate for RNA confirmation.
Each such candidate carries a route to resolution: the gene's expression in accessible tissues, the samples the case already holds, and the evidence weight an RNA result would carry under the current guidance. The report, rendered from the case, would say that a second allele was unconfirmed, name the synonymous candidate, and state the assay proposed, rather than reporting a carrier with an unexplained phenotype.
No variant is called pathogenic on a prediction. The splice score is shown with its limits, the hypothesis stays open until the RNA result arrives, and the reviewer decides. What changes is that the variant is seen at all, by every scientist on every case, rather than by the one who happened to remember.
What you can do tomorrow
The default filter is a configuration decision, and it can be changed this week.
- For every one-allele recessive case, re-run the gene with synonymous variants retained and sort by splice prediction. It takes minutes, it needs no new software, and it has solved cases.
- Change the pipeline default so that synonymous variants with a splice prediction above the published threshold survive the first filter in every case.
- Report synonymous candidates with the splice score, the tool and the version, and treat them under the same evidence rules as any other splice-region variant.
- Keep a list of the one-allele recessive cases, with the gene and whether the synonymous search has been done. It is a short list and a productive one, and it is the first list to revisit when a new splice predictor is released.
Sources
- Sauna ZE, Kimchi-Sarfaty C. Understanding the contribution of synonymous mutations to human disease. Nature Reviews Genetics, 2011.
- Hunt RC, Simhadri VL, Iandoli M, Sauna ZE, Kimchi-Sarfaty C. Exposing synonymous mutations. Trends in Genetics, 2014.
- Zeng Z, Bromberg Y. Predicting functional effects of synonymous variants: a systematic review and perspectives. Frontiers in Genetics, 2019.
- Jaganathan K, Kyriazopoulou Panagiotopoulou S, McRae JF, et al. Predicting splicing from primary sequence with deep learning. Cell, 2019.
- Walker LC, Hoya M, Wiggins GAR, et al. Using the ACMG/AMP framework to capture evidence related to predicted and observed impact on splicing: recommendations from the ClinGen SVI Splicing Subgroup. American Journal of Human Genetics, 2023.