Residency · Residency · Medical Genetics Genomics

Variant Classification: ACMG/AMP Guidelines in Practice

The Five-Tier Classification System

Overview

The ACMG and the Association for Molecular Pathology (AMP) developed a standardized framework in 2015 (Richards et al.) for classifying germline sequence variants into five categories. Pathogenic (P) variants are clearly disease-causing with sufficient evidence for clinical action. Likely Pathogenic (LP) variants have strong evidence for disease causation, defined as greater than 90% certainty. Variants of Uncertain Significance (VUS) have insufficient evidence to classify as either benign or pathogenic. Likely Benign (LB) variants have strong evidence against pathogenicity, also at greater than 90% certainty. Benign (B) variants are clearly not disease-causing. In clinical practice, Pathogenic and Likely Pathogenic classifications are considered actionable, while Likely Benign and Benign should not prompt clinical action. VUS should not be used for clinical decision-making but should prompt continued surveillance and periodic reanalysis.

Evidence Categories

Population Data (PM2, BA1, BS1, BS2)

Population frequency data is among the most objective forms of evidence. BA1 (Benign Stand-Alone) applies when a variant has an allele frequency greater than 5% in any general population database such as gnomAD or ExAC, and this alone is sufficient for a benign classification. BS1 (Benign Strong) applies when the allele frequency is greater than expected for the disorder. PM2 (Pathogenic Moderate) applies when the variant is absent or extremely rare in population databases, though ClinGen now recommends applying this at the supporting rather than moderate level (PM2_Supporting). BS2 (Benign Strong) applies when the variant is observed in a healthy adult for a fully penetrant early-onset condition. Key databases include gnomAD, which comprises over 140,000 exomes and over 76,000 genomes. It is essential to consider ancestry-matched populations, as a variant may be common in one population but absent in others. Filtering allele frequency thresholds should be disease-specific, ranging from 0.01% for severe childhood conditions to 1% for common adult-onset diseases.

Computational/Predictive Data (PP3, BP4, BP7)

PP3 (Pathogenic Supporting) applies when multiple in silico tools predict a deleterious effect. BP4 (Benign Supporting) applies when multiple tools predict no impact. BP7 (Benign Supporting) applies to synonymous variants with no predicted splicing impact. Available in silico prediction tools for missense variants include REVEL (the most widely validated meta-predictor), CADD, PolyPhen-2, SIFT, MutationTaster, and BayesDel. For splicing prediction, SpliceAI (a deep learning tool), MaxEntScan, dbscSNV, and SQUIRLS are available. Conservation is assessed using PhyloP, GERP++, and phastCons scores. ClinGen recommends REVEL scores of 0.773 or above for PP3 and 0.290 or below for BP4, though gene-specific thresholds may vary. It is important to note that computational evidence alone is never sufficient for a pathogenic classification.

Functional Data (PS3, BS3)

PS3 (Pathogenic Strong) applies when well-established functional studies demonstrate a damaging effect, such as through enzymatic assay, protein stability measurements, or cellular phenotype assessment. BS3 (Benign Strong) applies when well-established functional studies show no damaging effect. For an assay to be considered "well-established," it must measure the function relevant to the disease mechanism, include positive and negative controls using known pathogenic and benign variants, and be published in peer-reviewed literature or generated through large-scale functional assays such as multiplexed assays of variant effect (MAVE). ClinGen gene-specific recommendations define which functional assays qualify at which evidence strength level. The strength of PS3 can be downgraded to moderate or supporting if the assay has limitations.

Segregation Data (PP1, BS4)

PP1 (Pathogenic Supporting) applies when a variant co-segregates with disease in affected family members. The evidence strength increases with the number of segregations: 3 segregations support a Supporting level, 5 segregations reach Moderate, and 7 or more reach Strong. BS4 (Benign Strong) applies when there is lack of segregation with disease. Segregation analysis must account for reduced penetrance and phenocopies, and LOD score-based approaches are preferred for formal analysis.

De Novo Data (PS2, PM6)

PS2 (Pathogenic Strong) applies when a de novo variant is identified in a patient with disease and no family history, provided both maternity and paternity have been confirmed. PM6 (Pathogenic Moderate) applies when de novo status is assumed without formal confirmation. The evidence strength increases with the number of unrelated de novo observations.

Allelic Data (PM3, BP2)

PM3 (Pathogenic Moderate) applies for recessive conditions when the variant is detected in trans with a known pathogenic variant, with strength increasing when observed in multiple unrelated individuals. BP2 (Benign Supporting) applies when a variant is observed in trans with a pathogenic variant for a dominant condition, or in cis with a known pathogenic variant.

Other Patient/Phenotype Data

PP4 (Pathogenic Supporting) applies when the patient's phenotype is highly specific for a particular gene, such as an achondroplasia phenotype with an FGFR3 variant. PS4 (Pathogenic Strong) applies when the variant prevalence in affected individuals is statistically increased compared to controls based on case-control data.

Variant Type and Predicted Effect

PVS1 (Pathogenic Very Strong) applies to null variants such as nonsense, frameshift, canonical splice site (+/-1,2), initiation codon loss, and single or multi-exon deletions in a gene where loss of function is a known mechanism of disease. This criterion must be applied carefully, checking for rescue by alternative transcripts, last exon location, and NMD prediction. The ClinGen PVS1 decision tree provides specific guidance for downgrading based on context. PM1 (Pathogenic Moderate) applies when a variant is located in a mutational hotspot or well-established functional domain without benign variation. PM4 (Pathogenic Moderate) applies to in-frame deletions or insertions in non-repeat regions, or to stop-loss variants. PM5 (Pathogenic Moderate) applies to a novel missense change at a position where a different pathogenic missense has been established. PP2 (Pathogenic Supporting) applies to missense variants in genes with low rates of benign missense variation where missense is a common mechanism of disease.

Combining Evidence: Rules for Classification

Pathogenic

A Pathogenic classification can be reached through several evidence combinations: 1 Very Strong plus at least 1 Strong; 1 Very Strong plus at least 2 Moderate; 1 Very Strong plus 1 Moderate plus 1 Supporting; 1 Very Strong plus at least 2 Supporting; at least 2 Strong; 1 Strong plus at least 3 Moderate; 1 Strong plus 2 Moderate plus at least 2 Supporting; or 1 Strong plus 1 Moderate plus at least 4 Supporting.

Likely Pathogenic

A Likely Pathogenic classification requires: 1 Very Strong plus 1 Moderate; 1 Strong plus 1 to 2 Moderate; 1 Strong plus at least 2 Supporting; at least 3 Moderate; 2 Moderate plus at least 2 Supporting; or 1 Moderate plus at least 4 Supporting.

Benign

A Benign classification requires either 1 Stand-Alone criterion (BA1) or at least 2 Strong benign criteria.

Likely Benign

A Likely Benign classification requires 1 Strong benign criterion plus 1 Supporting, or at least 2 Supporting benign criteria.

VUS

A VUS classification results when criteria for either pathogenic or benign classification are not met, or when conflicting evidence exists.

ClinGen Gene-Specific Variant Curation

ClinGen Expert Panels

The ACMG/AMP guidelines are intentionally broad, and ClinGen Variant Curation Expert Panels (VCEPs) develop gene-specific adaptations. These modifications include adjusted population frequency thresholds, approved functional assays with defined evidence strength levels, phenotype-specific PP4 criteria, and modified PVS1 application. Notable examples include the ENIGMA consortium for BRCA1/2, as well as expert panels for RASopathies, CDH1, hearing loss, Rett syndrome (MECP2), and cardiomyopathies. ClinGen-approved classifications are deposited in ClinVar as "expert panel reviewed" with a 3-star rating.

ClinVar

ClinVar is a public archive of variant-disease relationships maintained by the NCBI. It receives submissions from clinical laboratories, research groups, and expert panels. A star rating system reflects review status, ranging from 0 to 4 stars, with 4 stars indicating practice guideline level. Conflicts between submitters are common, especially for VUS classifications, and ClinGen expert panels help resolve these discordances. Clinicians should check ClinVar regularly for reclassifications of previously reported variants.

Variant Reclassification

Frequency and Impact

Approximately 7 to 10% of VUS are reclassified annually as new evidence accumulates. Most reclassifications, about 90%, move VUS toward Likely Benign or Benign. VUS to Likely Pathogenic or Pathogenic occurs in approximately 10% of reclassifications. Rarely, Pathogenic or Likely Pathogenic variants are downgraded, often due to updated population frequency data. Laboratories have a responsibility to update classifications and notify ordering providers when significant reclassifications occur.

Challenges

There is no standardized system for re-contacting patients or providers after reclassification. The process creates a significant time and resource burden on clinical laboratories. Patient anxiety during the period of uncertainty is a real concern. There are also medicolegal implications of acting or not acting on a VUS that is later reclassified.

Common Pitfalls in Variant Interpretation

Several common errors should be avoided in variant interpretation. These include over-reliance on a single in silico predictor rather than using multiple tools, failure to check population databases with appropriate ancestry matching, applying PVS1 to genes where loss of function is not an established disease mechanism, using clinical laboratory classifications without reviewing the underlying evidence, treating a VUS as pathogenic for clinical management or conversely ignoring a VUS in a compelling clinical context without appropriate follow-up, not phasing two variants in a recessive gene to determine whether they are in cis or trans, and ignoring gene-disease validity before proceeding with variant interpretation.

<image>A flowchart showing the ACMG/AMP variant classification decision process. Starting with a sequence variant identified on genetic testing, the flowchart branches into two main paths: evidence for pathogenicity (left, red) and evidence for benign classification (right, green). Each path shows the specific criteria codes (PVS1, PS1-4, PM1-6, PP1-5 on the pathogenic side; BA1, BS1-4, BP1-7 on the benign side) organized by evidence strength (Very Strong, Strong, Moderate, Supporting). At the bottom, a combining rules table shows how criteria combinations lead to final classifications of Pathogenic, Likely Pathogenic, VUS, Likely Benign, or Benign.</image>

<image>A detailed decision tree for applying PVS1 (null variant criterion). Starting with "Is the variant a predicted loss-of-function type?" (nonsense, frameshift, canonical splice, initiation codon loss, exon deletion), the tree branches through key considerations: (1) Is LOF an established mechanism of disease for this gene? (2) Is the variant in the last exon or the last 50 bp of the penultimate exon (potential NMD escape)? (3) Does the gene have biologically relevant alternative transcripts that rescue function? (4) For splice variants, does the variant affect a constitutive or alternative exon? Each terminal node shows the recommended evidence strength: PVS1 (Very Strong), PVS1_Strong, PVS1_Moderate, or PVS1_Supporting, with explanatory annotations.</image>

<image>A before-and-after infographic illustrating variant reclassification. A timeline shows a VUS reported in 2019 with a patient counseled about uncertainty. Over time, new evidence accumulates: a 2020 functional study demonstrating protein loss of function (PS3), a 2021 gnomAD update confirming absence in 150,000 individuals (PM2), and a 2022 ClinGen expert panel review incorporating segregation data from multiple families (PP1). The VUS is reclassified to Likely Pathogenic in 2022, triggering updated clinical recommendations. The diagram highlights the workflow for laboratory reclassification notification and clinical follow-up action.</image>

Clinical Pearls

A VUS should never be used as the sole basis for clinical management decisions such as prophylactic surgery or reproductive decisions. However, clinical surveillance based on family history and phenotype should continue regardless of variant classification. Approximately 90% of VUS reclassifications move toward benign rather than pathogenic, and patients should be counseled accordingly to reduce anxiety. PVS1 is the most powerful single pathogenicity criterion but requires careful application: one must verify that loss of function is the disease mechanism, check for NMD escape in last-exon variants, and ensure the variant affects all clinically relevant transcripts. Population frequency data is among the most objective evidence available, and a variant present at greater than 5% in any population (BA1) can be classified as benign based on this evidence alone, regardless of in silico predictions or case reports. Gene-disease validity must be established before variant interpretation; applying ACMG/AMP criteria to a gene without sufficient evidence of disease association is a common error. Different clinical labs may classify the same variant differently, and when discordances exist in ClinVar, the underlying evidence should be reviewed or inter-laboratory reconciliation requested through ClinGen. Proactive reanalysis of VUS every 1 to 3 years is recommended, especially for patients who remain undiagnosed, as new population data, functional studies, or expert panel classifications may resolve the uncertainty.

References

• Richards S et al. Standards and guidelines for the interpretation of sequence variants: a joint consensus recommendation of the ACMG and AMP. Genet Med. 2015;17(5):405-424.
• Tavtigian SV et al. Fitting a naturally scaled point system to the ACMG/AMP variant classification guidelines. Hum Mutat. 2020;41(10):1734-1737.
• Abou Tayoun AN et al. Recommendations for interpreting the loss of function PVS1 ACMG/AMP variant criterion. Hum Mutat. 2018;39(11):1517-1524.
• Brnich SE et al. Recommendations for application of the functional evidence PS3/BS3 criterion using the ACMG/AMP sequence variant interpretation framework. Genome Med. 2019;12(1):3.
• Harrison SM et al. Overview of specifications to the ACMG/AMP variant interpretation guidelines. Curr Protoc Hum Genet. 2019;103(1):e93.
• ClinGen. https://clinicalgenome.org/. Accessed 2025.

Read this lecture as Markdown