A standard WGS run yields ~4–5 million variants per individual. The ACMG/AMP 2015 guidelines (Richards et al.) established a consensus five-tier classification system with 28 evidence criteria, later quantified by the Tavtigian 2018 Bayesian point system (see slides). LP and P are both clinically actionable — they trigger follow-up, cascade testing, and counselling. VUS must not drive clinical action; a VUS means 'we don't know,' not 'it's probably fine.' Strength levels can be applied at reduced or increased levels (e.g., PVS1 at 'Moderate' = +2 points), enabling nuanced evidence integration.

**Population Frequency Evidence** Population allele frequency is one of the most powerful benign evidence criteria. The gnomAD database — aggregating exome and genome data from >250,000 individuals across multiple ancestries — is the gold-standard reference for population frequency. - **BA1 (Benign — Stand-Alone)**: Allele frequency >5% in any gnomAD subpopulation. At this frequency, the variant is too common to be pathogenic for virtually any Mendelian disease. - **BS1 (Benign — Strong)**: Allele frequency greater than expected for the disease prevalence. Gene- and disease-specific thresholds apply — calculated from disease prevalence, genetic heterogeneity, and penetrance. - **PM2 (Pathogenic — originally Moderate)**: Absent or extremely rare in gnomAD (AF <0.0001 for dominant; <0.001 for recessive conditions). Note: ClinGen downgraded PM2 to Supporting strength in 2023 (PM2_Supporting, +1 point), reflecting that rarity alone is weak evidence of pathogenicity. **Critical gnomAD caveats**: gnomAD contains individuals with severe disease (including neuropsychiatric conditions), so rare pathogenic variants for recessive conditions may be present at low carrier frequencies. Always check subpopulation frequencies — a variant rare overall may be common in a specific population. Filtering allele frequency (FAF) accounts for population structure and is preferred over raw AF. **Key Databases** - **ClinVar**: Public repository for variant-condition interpretations from submitting laboratories. Contains millions of records. Assess star ratings: 1-star = single submitter; 2-star = criteria provided; 3-star = reviewed by expert panel; 4-star = practice guideline. Discordant classifications between labs are common. - **DECIPHER**: Database of genomic variants in patients with developmental disorders; particularly useful for CNV interpretation. - **OMIM**: Online Mendelian Inheritance in Man — the reference for gene-disease relationships and inheritance patterns. **Computational / In Silico Evidence (PP3/BP4)** Computational tools predict variant pathogenicity based on evolutionary conservation, protein structure, and functional impact. PP3 (Supporting pathogenic) and BP4 (Supporting benign) require **multiple lines** of computational evidence — no single tool is diagnostic. - **REVEL**: Ensemble meta-predictor integrating 13 tools; score 0–1; >0.5–0.7 typically threshold for pathogenic support - **CADD (Combined Annotation Dependent Depletion)**: Phred-scaled score integrating conservation, regulatory, and protein-level annotations; CADD >20 = top 1% most deleterious variants genome-wide - **AlphaMissense (Google DeepMind)**: Structure-based pathogenicity predictor leveraging AlphaFold; scores 0–1 - **SpliceAI**: Deep learning splice effect predictor; >0.2 = possible splice effect, >0.5 = likely effect - **Conservation tools** (PhyloP, PhastCons, GERP++): Measure cross-species evolutionary conservation at a position — high conservation suggests functional importance - **Constraint metrics** (pLI, LOEUF): Gene-level measures of intolerance to variation within the human population. pLI >0.9 = gene is intolerant to heterozygous LoF variants. These measure within-species selection pressure — distinct from cross-species conservation. **Critical distinction**: Conservation is about evolutionary time across species; constraint is about population genetics within humans. A gene can be constrained without being conserved at every position, and vice versa. Conflicting in silico predictions default to neutral evidence (VUS).

**Functional Studies (PS3/BS3)** Functional evidence evaluates whether a variant disrupts the gene product's function in a laboratory assay. PS3 (Strong pathogenic) requires **well-established functional studies** that demonstrate a damaging effect on the gene product. BS3 (Strong benign) requires well-established studies showing no functional impact. **What qualifies as 'well-established'?** - The assay must measure a function relevant to the disease mechanism (e.g., ion channel electrophysiology for channelopathies, enzyme activity for metabolic disorders) - The assay must include appropriate positive and negative controls - The assay must have been validated against known pathogenic and benign variants to establish its predictive value - In vitro overexpression systems without validation against known variants do NOT meet the PS3 standard - Animal models with orthologous phenotypes can provide PS3 evidence but require careful interpretation of species-specific differences **Strength calibration for PS3**: When the assay is less well-established or results are borderline, PS3 can be downgraded to Moderate (+2) or Supporting (+1). Gene-specific VCEP guidelines often define exactly which assays qualify at which strength level. **Case-Level Evidence (PS4)** PS4 (Strong pathogenic) applies when a variant is observed at a statistically significantly higher frequency in affected individuals compared to controls. This requires: - Well-powered case-control studies with odds ratio (OR) ≥5.0 and confidence interval not overlapping 1.0 - For rare variants without formal case-control data, PS4 can be applied at reduced strength based on the number of unrelated probands with the same variant and matching phenotype - ClinGen recommends: ≥2 unrelated probands = PS4_Moderate; ≥6 unrelated probands = PS4 at full Strong strength **Phenotype Specificity (PP4)** PP4 (Supporting pathogenic) applies when the patient's phenotype or family history is highly specific for a disease with a single genetic aetiology. For example: - A child with Dravet syndrome phenotype and an SCN1A variant — Dravet is highly specific for SCN1A - PP4 is less applicable to non-specific phenotypes like 'intellectual disability' or 'epilepsy' which have hundreds of genetic causes - In neurogenetics, PP4 is most informative for electroclinical syndromes with distinctive seizure types, EEG patterns, or developmental trajectories that point to a specific gene

**De Novo Evidence** De novo variants — those arising spontaneously in the proband and absent from both biological parents — are among the most powerful evidence for pathogenicity in neurogenetics, where many disorders (e.g., SCN1A Dravet syndrome, KCNQ2 neonatal epilepsy) are caused predominantly by de novo variants. - **PS2 (Strong, +4 points)**: De novo occurrence **confirmed** by parental testing in a patient with the disease and no family history. Both parents must be tested and confirmed to not carry the variant. This is Strong evidence because de novo variants in constrained genes are far more likely to be pathogenic than inherited variants. - **PM6 (Moderate, +2 points)**: **Assumed** de novo — maternity and paternity are not confirmed, or parental testing was not performed. PM6 is weaker than PS2 because non-paternity, mosaicism in a parent, or laboratory error could explain the apparent de novo status. **Segregation Evidence (PP1/BS4)** Co-segregation of a variant with disease in a family provides evidence proportional to the number of informative meioses observed. - **PP1 (Supporting to Strong)**: Variant co-segregates with disease in affected family members. Each additional affected carrier strengthens the evidence. Quantitative LOD-score approaches are now preferred: - LOD ≥1.5 (approximately 3 informative meioses): Supporting (+1) - LOD ≥3.0 (approximately 6+ informative meioses): Moderate (+2) - LOD ≥5.0: Strong (+4) - **BS4 (Strong benign)**: Variant does NOT segregate with disease — an affected family member does not carry the variant. This is powerful evidence against pathogenicity, though reduced penetrance and phenocopies must be considered. **Allelic Evidence (PM3/BP2)** - **PM3 (Moderate, +2 points)**: For autosomal recessive disorders, the variant is detected **in trans** with a known pathogenic variant. This establishes that both alleles are affected, supporting pathogenicity. PM3 strength increases with the number of unrelated probands observed in trans: PM3 at Moderate for 1 observation, PM3_Strong for multiple independent observations. - **BP2 (Supporting benign)**: Observed **in trans** with a known pathogenic variant in a **dominant** disorder (the individual has two hits but the disease is dominant — the second variant is likely benign), OR observed **in cis** with a pathogenic variant (both variants are on the same allele — the second variant is likely a passenger, not independently pathogenic).

**PVS1: Loss-of-Function Variants** Null variants (nonsense, frameshift, canonical splice site ±1,2, initiation codon, single/multi-exon deletion) in haploinsufficiency genes are among the most interpretively powerful variants. PVS1 is the strongest single criterion in the ACMG/AMP framework, but its application requires careful calibration. **PVS1 strength calibration (Abou Tayoun 2018)**: - **PVS1 Very Strong (+8)**: Canonical splice site ±1,2; nonsense/frameshift predicted to trigger NMD; complete gene deletion — in a gene where LoF is an established disease mechanism - **PVS1_Strong (+4)**: Nonsense/frameshift in the last exon or that escape NMD; splice variants where effect on splicing is established but not canonical - **PVS1_Moderate (+2)**: Splice region variants where effect is uncertain; initiation codon variants (p.Met1?) where no alternative in-frame start codon is confirmed - **PVS1_Supporting (+1)**: Variants where the LoF mechanism is not well-established for the gene **PVS1 caveats**: PVS1 does NOT apply to gain-of-function (GoF) genes. In neurogenetics, this is critical: genes like KCNQ3, SCN8A (in certain contexts), and GRIN2A may cause disease through gain-of-function missense variants, not LoF. A truncating variant in a GoF gene may be benign or protective. Always verify the gene's established disease mechanism before applying PVS1. Check gnomAD for LoF tolerance — genes tolerant to LoF variants (low pLI, high LOEUF) may not support PVS1. **Splice Variants** Splice variants present unique interpretation challenges: - Canonical ±1,2 positions: PVS1 applies (Very Strong) — these almost always abolish the splice site - Extended splice region (positions ±3 to ±8, or deeper intronic): Require SpliceAI or other tools to predict effect; PVS1 at reduced strength (Moderate or Supporting) - RNA studies (RT-PCR from patient tissue) can confirm aberrant splicing and upgrade PVS1 strength - Cryptic splice activation can create novel exons from intronic sequence — requires RNA analysis to detect **In-Frame Indels and Repeat Expansions** - **PM4 (Moderate)**: In-frame insertions/deletions that change protein length in a non-repetitive region, or stop-loss variants. PM4 is Moderate because in-frame changes may or may not disrupt protein function. - **BP3 (Supporting benign)**: In-frame insertion/deletion in a repetitive region of unknown function — likely tolerated. - **Repeat expansions** (e.g., trinucleotide repeats in Huntington disease, FMR1, C9orf72 hexanucleotide expansion in ALS/FTD) fall outside the standard ACMG/AMP framework. These require gene-specific interpretation guidelines and specialised testing methods (Southern blot, repeat-primed PCR, long-read sequencing). **Somatic Variants and Mosaicism** Mosaicism — the presence of a variant in only a fraction of cells — poses significant challenges: - Low-level somatic mosaicism in a parent can mimic de novo inheritance (variant appears absent in parental blood but is present in gonadal tissue) - Somatic mosaicism in the proband may result in variant allele fractions significantly below 50%, potentially causing the variant to be missed by standard filtering - Deep sequencing (>500x) or specialised mosaic detection tools may be needed - The ACMG framework was not designed for somatic variants; cancer-specific frameworks (AMP/ASCO/CAP 2017) apply to somatic tumour variants **Hotspot Residues (PM1)** PM1 (Moderate) applies when a variant falls in a well-established mutational hotspot or critical functional domain without benign variation. In neurogenetics: voltage sensor domains (S4 segments) of ion channels (KCNQ2, SCN1A), selectivity filter residues, and ligand-binding domains of neurotransmitter receptors (GRIN2A, GRIN2B) are established PM1 regions.

**Report Structure** A clinical variant interpretation report should include: 1. **Variant identification**: HGVS nomenclature (genomic, coding, and protein-level), gene name, transcript used, zygosity 2. **Classification**: The 5-tier ACMG/AMP classification with the specific evidence criteria applied and their strength levels 3. **Evidence summary**: A narrative describing each line of evidence — population frequency, in silico predictions, functional data, segregation, de novo status, and phenotype match 4. **Clinical correlation**: How the variant relates to the patient's phenotype and the known gene-disease relationship 5. **Recommendations**: Suggested follow-up (cascade testing, functional studies, segregation analysis, re-analysis timeline) **VUS Management** When a variant of uncertain significance is identified, clear communication is essential: - A VUS is NOT a diagnosis — it means current evidence is insufficient to classify the variant as pathogenic or benign - **Never use a VUS for clinical decision-making** — management should be guided by clinical findings alone, not uncertain genetic data - Sample framing: 'This variant is currently classified as uncertain, which means we do not yet have enough scientific evidence to know whether it contributes to your child's condition or is a harmless change.' - Document the VUS in the medical record with explicit language about its uncertain significance to prevent downstream misinterpretation by other providers **Re-Analysis Workflows** - Recommended re-analysis interval: every **1–2 years**, or upon emergence of new clinical features - Re-analysis of previously negative or VUS-only exomes yields new diagnoses in approximately **10–25% of cases** - The majority of VUS reclassifications (~10–20% within 5 years) move toward **benign or likely benign** - Sources of new evidence: growth of gnomAD and ClinVar, new functional studies, additional affected families, ClinGen VCEP publications - Encourage **family segregation testing** to generate PP1 or BS4 evidence that can shift classification **Variant Reclassification** Variant interpretation is iterative and collaborative: - ClinVar serves as the central repository for variant-condition interpretations; laboratories are expected to update their submissions as evidence evolves - Discordant ClinVar classifications between labs are common — a known quality issue that ClinGen expert panels work to resolve - When a VUS is reclassified to LP/P, the laboratory should proactively notify the ordering clinician so that clinical management can be updated - When a previously LP/P variant is downgraded, affected families must be re-counselled **Gene-Specific Guidelines (ClinGen VCEPs)** ClinGen Variant Curation Expert Panels publish gene-specific rule modifications that **replace** the generic ACMG/AMP 2015 rules for the specified gene. VCEP guidelines adjust evidence thresholds, define which functional assays qualify at which strength level, add gene-specific criteria, and restrict inapplicable rules. Always check for a published VCEP SOP before finalizing classification — examples include SCN1A, CDH1, and RASopathy gene VCEPs. **Neurogenetics-Specific Considerations** - Most de novo variants identified by trio sequencing in NDD patients are initially VUS — reclassification over time is expected - Ion channel genes may have both LoF and GoF disease mechanisms — the same gene can cause different phenotypes depending on whether the variant causes loss or gain of function - Electroclinical syndrome specificity (PP4) is particularly powerful in epilepsy genetics — distinctive seizure types, EEG patterns, and developmental trajectories can point to specific genes