NMDrules: A VEP plugin to predict NMD in stop codon generating variants

Motivation

• NMD prediction is becoming increasingly important for understanding the actual molecular mechanism driving gene-disease association.
• Protein truncating variants (PTVs) that escape NMD can exert their pathogenic effects through gain-of-function or dominant-negative mechanisms. Unfortunately, they are often confused with loss-of-function variants, which can lead to incorrect genetic diagnosis.
• After noticing the shortcomings* in predicting NMD escape variants of open-source variant annotators (e.g. VEP NMD plugin, SnpEff, LOFTEE) or the complex integration of specialised software (i. aenmd, NMDEscPredictor, NMDetective, ALoFT, nmd-escape) into VEP based annotation pipelines, I decided to implement a simple VEP plugin myself that performs NMD prediction of stop codon generating variants (i.e. stop_gained, stop_loss, some frameshift variants).
• In addition to applying the current canonical and non-canonical rules for NMD prediction, the plugin extracts information about the genomic context surrounding the generated stop codon, and leverages an hypothetical translation reinitiation.

* The biggest shortcoming I've found is the inability to correctly infer the location of the stop codon generated by a frameshift variant. For example, variant EVL(NM_016337.3):c.573_574del:p.Val192ProfsTer105 generates a stop codon 315 bps downstream of the dinucleotide deletion. As implemented, most of the annotators mentioned above would consider the affected exon to be the same one where the deletion occurs, when in fact the stop codon is generated a couple of exons downstream.

Description

This is a plugin for Ensembl Variant Effect Predictor (VEP) software that predicts whether a stop codon generating variant (i.e. stop_gained, stop_loss, some frameshift variants) triggers nonsense-mediated mRNA decay (putative_NMD_triggering) or not (canonical_NMD_escaping, noncanonical_NMD_escaping) based on the following rules (1) (NMD_prediction => NMD_rule):

• The variant is located in an intronless transcript: canonical_NMD_escaping => intronless
• The variant is located in the last exon: canonical_NMD_escaping => last_exon
• The variant is located in the last 50 bp of the penultimate exon: canonical_NMD_escaping => 50bp_penult_exon
• The variant is located in the first 150 coding bases: noncanonical_NMD_escaping => first_150bp
• The variant is located in an exon larger than 407 bp: noncanonical_NMD_escaping => lt_407bp_exon

Then, it calculates the distance in bps to the next ATG (start codon) after the stop codon to leverage an hypothetical translation reinitiation (3) (Next_ATG). If no ATG is found downstream of the stop codon, the output value is -1.

The plugin also shows the 2 codons/amino acids before the novel stop codon, plus the next nucleotide, for a detailed analysis, as they may influence the NMD efficiency (2) (Stop_context).
Format: -2codon(-2aa)-1codon(-1aa)stop_codon(Stop)fourth_letter

Example:

Variant: 10-102509528-C-CG (GRCh37)
HGVS: PAX2(NM_000278.5):c.76dup:p.Val26GlyfsTer28
Annotation: noncanonical_NMD_escaping|first_150bp|169|GCC(Ala)CTG(Leu)TGA(Stop)C

Implementation

The plugin starts by filtering in all variants with a stop codon (Ter) in their HGVSp notation. Exceptions are synonymous variants at the stop codon (e.g. p.Ter811=) or when no stop codon is inferred downstream of a frameshift variant (e.g. p.Ter257GlufsTer?). It then internally applies the mutation to the original cDNA to determine the location of the new stop codon. All insertions and deletions must be contained within the cDNA sequence.

Splicing variants and deletions involving intron/exon boundaries are not considered because VEP cannot properly infer the mutated protein.

This plugin has been tested on all the gnomAD 2.1.1 exomes variants without throwing any errors. Be aware that more complex variants may cause unobserved bugs.

References

1. The impact of nonsense-mediated mRNA decay on genetic disease, gene editing and cancer immunotherapy (Lindeboom et al. 2019, Nature Genetics)
2. Systematic analysis of nonsense variants uncovers peptide release rate as a novel modifier of nonsense-mediated mRNA decay efficiency (Kolokada et al. 2024, bioRxiv)
3. Advanced variant classification framework reduces the false positive rate of predicted loss-of-function variants in population sequencing data (Singer-Berk et al. 2023, Am J Hum Genet)

Usage

mv NMDrules.pm  ~/.vep/Plugins
./vep -i variations.vcf --plugin NMDrules

Credit

@marcpybus - https://github.com/marcpybus/