Variant Interpretation & ACMG Classification

Every genetic test you order — whether it's an epilepsy panel, an exome, or a genome — will come back with variants classified into one of five tiers. Understanding this system is essential because the classification directly determines what you can and cannot do with the result.

The ACMG/AMP 2015 guidelines created a standardized five-tier system: Pathogenic (P), Likely Pathogenic (LP), Variant of Uncertain Significance (VUS), Likely Benign (LB), and Benign (B). Think of these as a probability spectrum — Pathogenic means ≥99% chance the variant causes disease, while VUS means we genuinely don't know (the 10–90% probability range).

The critical clinical distinction you need to internalize: both P and LP are actionable — they can inform treatment, guide surveillance, and justify cascade family testing. A VUS is NOT actionable — it means the lab reviewed all available evidence and couldn't determine whether this variant matters. You should never change management based on a VUS alone. One important caveat: for autosomal recessive conditions, a single heterozygous P/LP variant identifies the patient (or family member) as a carrier — the 'actionable' result in that context is carrier status counseling and recurrence risk assessment, not a diagnosis of the condition itself. A second pathogenic variant on the other allele (confirmed in trans) is required to establish a diagnosis.

Behind each classification is a structured evidence evaluation using 28 criteria organized into a point system (Tavtigian 2018). Each piece of evidence — population frequency, computational predictions, functional data, de novo status — contributes positive or negative points. The total score maps to a tier. You don't need to memorize every criterion code, but understanding the point system helps you read reports critically and know when parental testing or other follow-up could change a classification.

When a lab classifies a variant, they're weighing several categories of evidence. You don't need to memorize every criterion code, but understanding the main evidence types helps you read reports critically and know when to push back or request additional testing.

Population Frequency

The simplest and most powerful question: how common is this variant in healthy people? The gnomAD database (>250,000 exomes and genomes from multiple ancestries) is the standard reference. If a variant is present at >5% frequency in any population, it's classified as Benign automatically — the only stand-alone criterion in the whole system. If the variant is absent or extremely rare in gnomAD, that's supporting evidence for pathogenicity (+1 point), but rarity alone is weak evidence — plenty of rare variants are harmless.

Computational Predictions

Software tools predict whether a variant is likely damaging based on evolutionary conservation, protein structure, and other features. The key tools you'll see on reports: REVEL (ensemble score 0–1; ≥0.644 is the ClinGen threshold for moderate pathogenic evidence), CADD (Phred-scaled; >20 = top 1% most deleterious), and SpliceAI (for splice effects; >0.5 = likely effect). No single tool is diagnostic — when tools conflict, computational evidence is counted as neutral (0 points).

Gene Constraint: pLI and LOEUF

You'll see these metrics on reports and they're important for interpreting truncating variants. Constraint metrics measure how intolerant a gene is to loss-of-function (LoF) variants within the human population — based on comparing the observed number of LoF variants in gnomAD to the number expected by chance. pLI (probability of LoF intolerance) >0.9 means the gene is highly intolerant to heterozygous LoF variants — people who lose one copy of this gene tend not to survive or reproduce normally. LOEUF (LoF observed/expected upper bound fraction) is the newer metric: LOEUF <0.35 indicates strong constraint. These are gene-level measures of within-species selection pressure — distinct from cross-species evolutionary conservation (PhyloP, GERP++), which asks whether a specific DNA position has been preserved across vertebrates over millions of years. A gene can be constrained (high pLI) without every individual position being conserved, and vice versa. In practice: high constraint supports applying PVS1 to truncating variants; low constraint (low pLI, high LOEUF) suggests the gene tolerates LoF and PVS1 may not apply.

De Novo Status

Finding that a variant arose spontaneously (present in the child, absent from both parents) is one of the strongest single lines of evidence (+4 points when confirmed). Each person carries only ~1–2 new coding variants, so the probability of one landing by chance in the exact gene causing the child's phenotype is very low. This is why trio sequencing is so powerful for severe early-onset conditions. Important: 'assumed' de novo (parents not tested) is weaker (+2 points) — you get the full +4 only when both parents are confirmed negative.

Functional Studies

Lab experiments showing that a variant disrupts gene function can provide strong evidence. But the bar is high: the assay must measure a disease-relevant function, include proper controls, and be validated against known pathogenic and benign variants. An in vitro experiment without these controls doesn't meet the standard.

Segregation and Phasing

Does the variant track with disease in the family? For recessive conditions, confirming that two variants are on different alleles (in trans) is essential. Co-segregation through multiple affected relatives strengthens pathogenicity evidence proportional to the number of informative meioses.

Null Variants and PVS1: The Strongest Single Criterion

PVS1 is the single most powerful pathogenic criterion in the ACMG framework (+8 points at full strength — Very Strong). It applies to 'null' variants that are predicted to completely eliminate the gene product: nonsense (stop-gain) variants, frameshift insertions/deletions, canonical splice site variants (at the ±1,2 positions), initiation codon variants, and whole-gene or multi-exon deletions. The concept behind PVS1 is haploinsufficiency — the idea that having only one working copy of the gene is not enough for normal function, so losing one copy causes disease. Haploinsufficiency is established through multiple lines of evidence: high gene constraint (pLI >0.9, LOEUF <0.35), known pathogenic LoF variants in ClinVar, and ClinGen gene-disease curation confirming LoF as the disease mechanism. Critically, PVS1 does NOT apply to genes where the disease mechanism is gain-of-function (GoF) — a truncating variant in a GoF gene may be benign or protective, because destroying the protein is not the same as making it overactive. In neurogenetics, this distinction is vital: in some ion channel genes (e.g., SCN8A, SCN1A, GRIN2A), some variants cause disease through GoF missense changes, not LoF mechanisms — missense variants may contribute to either GoF or LoF at the channel level, such that functional data or other lines of evidence may be required to understand the specific mechanism for a specific or novel variant. Always verify the gene's established mechanism before applying PVS1.

Let's walk through a realistic genetic testing report — the kind you'll receive on service — and apply what we've learned. This is a representative teaching example based on real clinical scenarios.

The patient: An 8-month-old boy with infantile epileptic spasms (West syndrome), developmental plateau, and no family history of epilepsy. Brain MRI shows no structural abnormality. EEG shows hypsarrhythmia. The treating neurologist orders a comprehensive epilepsy gene panel (300+ genes) as a singleton test (parents not included).

The report: The panel identifies one variant of interest — an SCN2A missense variant classified as VUS. See the report excerpt and criteria walkthrough below.

Let's evaluate this ourselves. The lab found two pieces of positive evidence: the variant is absent from the general population (supporting) and computational tools predict it is damaging (moderate). But without parental testing, de novo status is unknown — and that's where the biggest opportunity lies.

The total score is 3 points — firmly in VUS territory (need ≥6 for LP). But look what happens if we order parental testing: if this variant turns out to be de novo, that adds +4 points, bringing the total to 7 — which crosses the threshold into Likely Pathogenic. One blood draw from each parent could change this from 'uncertain' to 'actionable.' This is the single most important practical lesson in variant interpretation: always consider whether parental testing could resolve a VUS.

Now let's look at a trio whole exome sequencing report — where both parents are sequenced alongside the child. This is the preferred strategy for severe early-onset conditions because it immediately identifies de novo variants. But as we'll see, even confirmed de novo status doesn't always push a variant past the VUS threshold.

The patient: A 3-year-old girl with global developmental delay, drug-resistant epilepsy (onset 4 months), and no distinctive features. Both parents are neurologically healthy. Brain MRI shows mild cerebral atrophy. The team orders trio WES.

The report: The trio identifies one de novo variant in STXBP1 — a well-established developmental and epileptic encephalopathy gene. Despite being confirmed de novo, the lab has classified it as VUS. Let's understand why.

The variant scores PS2 (+4 for confirmed de novo) and PM2_Supporting (+1 for absent from gnomAD) = 5 points total. That's one point short of the 6-point LP threshold. Why didn't computational evidence push it over? Because the REVEL score (0.48) falls below the ClinGen threshold (0.644) — the computational tools aren't confident this amino acid change is damaging. Without that computational support, the variant sits at 5 points.

This teaches an important lesson: de novo is strong evidence, but it's not a guarantee of pathogenicity. Each person carries ~1–2 de novo coding variants, and most of them are harmless. De novo + rarity gives you 5 points — you still need at least one more piece of evidence (computational support, functional data, additional affected individuals, or phenotype specificity) to reach LP.

Variant interpretation in neurogenetics has several unique pitfalls that don't apply as strongly in other specialties. Knowing these will save you from misinterpreting results on service.

Gain-of-function vs. loss-of-function genes

This is the single most important concept. Some ion channel genes cause disease through gain-of-function (GoF) missense variants — the protein works abnormally, not less. If you see a truncating (nonsense or frameshift) variant in a GoF gene, it may actually be benign or even protective, because destroying the protein isn't the same as making it overactive. Classic neurogenetics examples: SCN8A (some contexts), KCNQ3, and certain GRIN2A variants cause disease through GoF. Always check: what is this gene's established disease mechanism before interpreting a truncating variant?

Repeat expansions are invisible to standard sequencing

Standard exome sequencing does NOT detect trinucleotide repeat expansions. This is a critical blind spot for child neurology: Friedreich ataxia, most spinocerebellar ataxias, myotonic dystrophy, Huntington disease, Fragile X, and C9orf72 ALS/FTD are all caused by repeat expansions that WES cannot see. If the phenotype suggests a repeat disorder and the exome is negative, you have NOT excluded these diagnoses — dedicated repeat testing is required.

Splice variants need extra evidence

Variants at canonical splice sites (the ±1,2 nucleotides of the intron) are well-characterized and almost always disruptive. But variants deeper in the intron (positions ±3 to ±8 and beyond) have uncertain effects. Tools like SpliceAI can predict splice disruption, but RNA studies (RT-PCR from patient tissue) provide much stronger evidence.

Mosaicism can fool you

A variant present in only a fraction of cells may have a variant allele fraction well below 50%, potentially falling below standard bioinformatic filters and being missed entirely. Low-level mosaicism in a parent can also make a variant look de novo when the parent actually carries it at low levels in gonadal tissue. If a family has recurrence of an apparently de novo condition, consider parental gonadal mosaicism.

VUS results are the most common source of confusion and anxiety for families — and honestly, for clinicians too. Here's a practical framework.

The key message to families: 'We found a change in a gene that could be related to your child's condition, but right now the scientific evidence isn't strong enough to say for sure. This is called a variant of uncertain significance. It doesn't confirm a diagnosis, and it doesn't rule one out. We'll monitor this over time as more information becomes available.'

What to do next

• Order parental testing — this is often the single highest-yield follow-up step, especially if de novo status would push the variant toward LP (as we saw in Report 1)
• Document the VUS clearly in the medical record with explicit language about its uncertain status — this prevents future providers from misinterpreting it as a confirmed diagnosis
• Schedule reanalysis in 1–2 years — labs can re-evaluate existing data as databases grow and new gene-disease associations are validated
• Refer to genetic counseling for detailed family discussion

What NOT to do

• Never start or change treatment based on a VUS alone
• Never tell the family 'it's nothing' — it might be reclassified later
• Never tell the family 'it's probably the cause' — that creates false certainty that may lead to inappropriate clinical decisions

Reanalysis and reclassification: Variant interpretation is not static. Reanalysis of previously negative or VUS-only exomes yields new diagnoses in ~10–25% of cases. Most VUS reclassifications (~10–20% within 5 years) move toward benign, but a meaningful fraction are upgraded to LP/P. ClinGen Variant Curation Expert Panels (VCEPs) publish gene-specific guidelines that can change how individual variants are classified — these replace the generic ACMG rules for the specified gene. See the Genetic Epilepsies and Pharmacogenetics modules for examples of how variant interpretation directly guides treatment decisions.