Gene-specific ACMG/AMP classification criteria for germline APC variants: recommendations from the ClinGen InSiGHT Hereditary Colorectal Cancer / Polyposis Variant Curation Expert Panel

Abstract

Purpose:

The Hereditary Colorectal Cancer/Polyposis Variant Curation Expert Panel (VCEP) was established by the International Society for Gastrointestinal Hereditary Tumours (InSiGHT) and the Clinical Genome Resource (ClinGen), who set out to develop recommendations for the interpretation of germline APC variants underlying Familial Adenomatous Polyposis (FAP), the most frequent hereditary polyposis syndrome.

Methods:

Through a rigorous process of database analysis, literature review, and expert elicitation, the APC VCEP derived gene-specific modifications to the ACMG/AMP (American College of Medical Genetics and Genomics and Association for Molecular Pathology) variant classification guidelines and validated such criteria through the pilot classification of 58 variants.

Results:

The APC-specific criteria represented gene- and disease-informed specifications, including a quantitative approach to allele frequency thresholds, a stepwise decision tool for truncating variants, and semiquantitative evaluations of experimental and clinical data. Using the APC-specific criteria, 47% (27/58) of pilot variants were reclassified including 14 previous variants of uncertain significance (VUS).

Conclusions:

The APC-specific ACMG/AMP criteria preserved the classification of well-characterised variants on ClinVar while substantially reducing the number of VUS by 56% (14/25). Moving forward, the APC VCEP will continue to interpret prioritised lists of VUS, the results of which will represent the most authoritative variant classification for widespread clinical use.

INTRODUCTION

Heterozygous germline pathogenic variants in the tumour suppressor gene APC (Adenomatous Polyposis Coli, HGNC:583) lead to classic or attenuated Familial adenomatous polyposis (FAP, MONDO: 0021057), an autosomal dominant condition characterised by the growth of hundreds and thousands of colorectal adenomatous polyps, which almost invariably progresses to early-onset colorectal cancer (CRC) if left untreated (1, 2). As a result, clinical management guidelines for endoscopic surveillance and risk-reducing surgery are in place (3–5). Pathogenic germline APC variants demonstrate variable expressivity manifested in different age of onset, polyp burden, and the presence of extra-colonic features, which include duodenal adenoma, duodenal carcinoma, and gastric carcinoma with an increasing incidence noted in recent years (6). Other extra-intestinal manifestations include osteomas, desmoids, epidermoid cysts, congenital hypertrophy of the retinal pigment epithelium (CHRPE), adrenal adenomas, hepatoblastomas, medulloblastomas, and papillary thyroid carcinomas (7).

Historically, the term attenuated FAP (MONDO: 0016362) was used to distinguish a milder form of the disease from classic FAP (MONDO: 0021055). However, it has been increasingly recognised that the dichotomy between classic and attenuated FAP is somewhat arbitrary and does not fully capture the continuous spectrum of the colorectal phenotype and the complexity of extra-colonic lesions. Hence, attenuated FAP (AFAP) is often regarded as a legacy description and is no longer recommended, as are other historical nomenclatures for specific phenotypes such as Gardner or Turcot syndromes. As a result, the terms classic and attenuated FAP are combined and treated as one entity (MONDO: 0021057) when discussing the pathogenicity of APC variants in relevant phenotypes.

During the last three decades, thousands of rare or private pathogenic APC germline variants have been identified in FAP families. In parallel, advances in high-throughput sequencing, expansion in large hereditary cancer gene panels, and genome-scale screening in individuals with unrelated phenotypes or healthy controls have led to the detection of rare APC variants at a rate several orders of magnitude higher than in targeted sequencing, adding to the challenge of variant pathogenicity interpretation. The lack of existing data, information sharing, and consensus on variant classification have rendered most of these findings as either variants of uncertain clinical significance (VUS) or variants with conflicting interpretation. ClinVar currently lists 10,212 APC germline variants, 66% of which are VUS and only 8% overlap with the APC locus-specific database (LSDB) (retrieved 04/05/2022).

To address this issue, expert bodies are curating actively under the governance of Clinical Genomic Resource (ClinGen) – an NIH-funded effort dedicated to building a central resource that defines the clinical relevance of genes and variants (8). For well-defined genes and diseases, ClinGen variant curation expert panels (VCEP) submit variant classification with their accompanying evidence to ClinVar and the ClinGen Evidence Repository – the first regulatory grade human variant database. The works of other VCEPs are summarised previously (9, 10), which notably including a VCEP for PTEN, another established polyposis gene leading to PTEN Harmatoma tumour syndrome (11).

The International Society for Gastrointestinal Hereditary Tumours (InSiGHT) is a multidisciplinary consortium formed in 2005 by the merger of the Leeds Castle Polyposis Group and the International Collaborative Group on Hereditary Non-Polyposis Colorectal Cancer. The group established standardised variant interpretation guidelines for germline mismatch repair (MMR) variants, the underlying cause of Lynch syndrome. This led to the consistent and systematic evaluation of 2,360 MMR variants independent of the ACMP/AMP framework (12). InSiGHT also houses the world’s largest curated LSDBs of variants in gastrointestinal polyposis predisposing genes on a Leiden Open Variation Database (LOVD) format, which currently lists 1867 different and 5628 total APC variants (retrieved 12/05/2022).

Building on the existing connection between InSiGHT and ClinGen, a Hereditary Colorectal Cancer/Polyposis VCEP was convened with the aim to improve accuracy and consistency in variant interpretation in APC, the MMR genes, and other polyposis genes including MUTYH (HGNC:7257), STK11 (HGNC:11389), POLD1 (HGNC:9175), POLE (HGNC:9177), SMAD4 (HGNC:6770) and BMPR1A (HGNC:1076). Here we describe the work of the APC VCEP in the development of APC-specific classification guideline and its validation through pilot variant classification. The criteria were designed to capture disease relevance of APC variants in the pathogenesis of FAP but not other rare phenotypes with specific molecular mechanisms or clinical presentations (e.g, gastric adenocarcinoma and proximal polyposis of the stomach (GAPPS, MONDO: 0017790) and isolated desmoids).

METHODS

The APC VCEP

The APC subcommittee of the ClinGen InSiGHT Hereditary Colorectal Cancer/Polyposis VCEP (refer to here as the APC VCEP) consists of 46 specialists with a balanced representation of expertise including gastroenterologists, medical geneticists, genetic counsellors, research scientists, bioinformaticians and clinical laboratory diagnosticians. Members are from 14 countries and diverse institutions worldwide. In three separate monthly meetings, the APC VCEP devoted focused discussions in functional, computational, and clinical subgroups, which was further reviewed and synthesised in another monthly meeting with the whole committee. Overall virtual conferences were conducted over a 2-year course and in-person meetings were held at the InSiGHT Biennial Conference in Auckland in 2019 and New Jersey in 2022.

Specification of the ACMG/AMP criteria

To provide standardised terminology and guidelines for variant classification, the ACMG and AMP jointly developed criteria for pathogenic (P) and benign (B) variants based population, experimental, computational, and clinical evidence (13). The criteria are assigned weights based on a hierarchy of Benign stand-Alone (BA), Pathogenic Very Strong (PVS), Benign/Pathogenic Strong (BS/PS), Benign/Pathogenic Moderate (PM/BM), and Benign/Pathogenic Supporting (PP/BP) evidence, which are combined to reach a 5-tier classification verdict ranging from Pathogenic (P), Likely Pathogenic (LP), VUS, Likely Benign (LB), to Benign (B) (Table 3). The assignment of evidence weight and rule combination are based on a quantitative framework using a Bayesian method, which provides statistical validation and enables further refinement of the ACMG/AMP criteria (14). Publicly available databases, predictive tools, published and unpublished data (experimental results, clinical laboratory data and case-level information) were acquired through systematic literature searching and information provided by committee members. The APC VCEP followed the general recommendations and feedback from the ClinGen Sequence Variant Interpretation (SVI) working group (15–19) and ClinGen VCEP procedures which were further revised by results of the pilot study. The APC-specific criteria and any subsequent updates are available at: https://cspec.genome.network/cspec/ui/svi/doc/GN089.

Table 3.

Rules for combining criteria APC-specific ACMG/AMP variant classification criteria

In addition to the original ACMG/AMP rules for combining pathogenic criteria, the following additional rules apply: (1) the combination of one Pathogenic Very Strong criterion and one Pathogenic Supporting criterion reach a classification of Likely Pathogenic; (2) the fulfillment of one Benign Strong criterion reaches Likely Benign; (3) if a rare variant fulfilling only PM2_Supporting but no other pathogenic codes also meets criteria for classification as (Likely) Benign, the population data is not considered conflicting and the variant can be classified as (Likely) Benign; (4) PVS1 cannot be applied in conjunction with splicing predictions (PP3) or RNA assays (PS3); (5) if RNA assay findings conflict with splice predictors, RNA findings override computational predictions (i.e., BS3 over PP3, and PS3 over BP4); (6) PS4_Variable and PP1_Variable should not be applied to a variant if BA1 or BS1 is met; however, meeting PM2_Supporting is not compulsory for pathogenic variants so that clinical criteria may be applied for such pathogenic variants with some levels of population data.

Selection of transcript

The preferred reference APC transcript for coding, intronic and promoter 1A variants is NM_000038.6 (MANE select transcript). This transcript contains 16 exons, including a non-coding exon 1. The NM_001127510.3 transcript contains one additional and one overlapping “non-coding” exon in the 5’ region compared to NM_000038.6. For promoter 1B deletions, the preferred transcript is NM_001127511.3, which has an alternative coding exon 1. The LRG_130 summarises all three “additional” exons, resulting in 18 exons (Supplemental Table S1).

Variants for pilot classification

A balanced spectrum of 58 APC variants were chosen from ClinVar and the InSiGHT LSDB based on the following eligibility requirements: (1) variants covering different types such as nonsense, frameshift, splice site, missense, synonymous, intronic, stop loss, in-frame indels, and large duplications / deletions, including presumed missense or synonymous variants which are in fact splice variants; (2) variants with conflicting interpretations within ClinVar and between ClinVar and the InSiGHT LOVD; (3) variants encompassing a broad range of criteria and different combinations of criteria; (4) variants distributed throughout the APC gene, including regions associated with a milder polyposis phenotype (Figure 1); and finally (5) variants with a range of classifications in ClinVar (Supplemental Table S2). Phenotype data from routine diagnostic testing were acquired for all pilot variants from VCEP members and documented in a standardised, anonymised format with the bare minimum of information required for phenotypic scoring.

Figure 1. APC gene, APC protein, and criteria boundaries and genotype-phenotype correlations.

Representation of the APC gene and its main protein product (middle) on the reference sequence NM_000038.6 (non-coding exon 1 not shown). The figure shows the boundaries for the application of PVS1, BS3 and BP1 (top), and genotype-phenotype correlations (bottom). The APC protein comprises several domains and motifs as shown. The 15-aa repeats confer high affinity binding to β-catenin whereas the 20-aa repeats both bind and promote β-catenin phosphorylation, ubiquitination and subsequent proteolytic degradation by a cytoplasmic destruction complex. Abbreviations: AA – amino acid; SAMP motifs - Serine-Alanine-Methionine-Proline motifs; EB1 - end-binding protein; DLG domain: discs large domain.

Each variant was independently curated by at least two of the eight collaborating biocurators using ClinGen’s Variant Curation Standard Operating Procedure (SOP) Version 3 in the Variant Curation Interface (20). The disparities between codes used and final classifications were examined first among the biocurators and then with the wider VCEP, enabling an iterative process by which further modifications to the evidence codes were agreed upon to enhance their usability and accuracy. The first 58 variants classified by the APC specifications of the ACMG/AMP criteria are now publicly accessible on ClinVar, with the designation of a 3-star review status indicating expert panel consensus and FDA recognition of evidence quality (https://www.ncbi.nlm.nih.gov/clinvar/submitters/508966/). The detailed evidence used for each curation of these APC variants is also available in the ClinGen Evidence Repository (https://erepo.clinicalgenome.org/evrepo/).

RESULTS

The APC-specific modifications to the ACMG/AMP codes are summarised in Table 1. Further comments to all criteria are found in Supplemental Table S3, including the explanations for excluding eight of the 28 original ACMG/AMP criteria (PM1, PM3, PM4, PP2, PP4, PP5, BP3, BP6). For the remaining 20 criteria, gene-based and/or strength modifications were made. The rules for combining criteria to reach a final classification based on Bayesian reasoning is shown in Table 3 (14).

Table 1.

APC-specific ACMG/AMP variant classification criteria

(a) Note that de novo score is distinct from phenotype points and are not equivalent to the points used to classify a variant in Tavtigian et al (60). The parents are unaffected if they: have less than five colorectal adenomas in a colonoscopy and are without phenotype consistent with APC; or they are older than 60 years of age, have no signs of gastrointestinal tumors (e.g., rectal bleeding), no phenotype consistent with APC and the family history is unremarkable. (b) Reports of exon deletion/skipping/loss, insertion of intronic nucleotides. General: General modifications; Gene-Specific: Gene-specific modifications based on what is known about APC/FAP; N/A: not applicable for APC; Strength: modified strength of criteria based on the quality and quantity of evidence, CHRPE: Congenital hypertrophy of the retinal pigment epithelium.

Minor allele frequency-driven rules (BA1, BS1 and PM2_Supporting)

The Whiffin/Ware allele frequency calculator was used to calculate APC-specific minor allele frequencies (MAF) (21). Assuming an estimated FAP prevalence of 1:6,850 to 1:31,250 live births (22), the value of 1:10,000 was used for the calculation of PM2_supporting. To define “allelic heterogeneity”, the frequency of the most common pathogenic APC variant NM_000038.6:c.3927_3931del p.(Glu1309AspfsTer4) was used (0.06, found in 325 of 5527 APC variant records on InSiGHT LSDB, retrieved 15/12/2021). Penetrance of APC-associated FAP was specified as 0.9 to account for the occurrence of a milder phenotype spectrum. Based on these values, the calculated MAF suggestive of pathogenicity is ≤ 0.0003% (PM2_Supporting). Using an equally conservative approach, an estimated prevalence of 1:5,000 people and penetrance of 0.8 were used to account for milder cases of APC-associated FAP in the calculation of BA1 threshold.

Depending on the severity and specificity of the phenotype, the detection rate of a pathogenic APC germline variant in families with colorectal adenomatous polyposis ranges between 20–80% (23–25). To reflect this, “genetic heterogeneity” was set at 0.5, denoting the assumption that a (L)P APC variant is identified in approximately 50% of unselected patients with adenomatous polyposis. The MAF threshold for BA1 was computed to be ≥ 0.006%. Since BA1 is a stand-alone criterion that yields in an uncontested Benign classification, to be even more strict, the final MAF threshold for BA1 was determined to be ≥ 0.1%. Similarly, based on a prevalence of 1:5,000 and allelic heterogeneity of 0.06, the MAF threshold for BS1 was ≥ 0.001% (rounded), which is close to the MAF of the most frequent pathogenic APC variant c.3927_3931del; this variant was found in 2 out of 236,524 alleles in the non-cancer dataset from gnomAD v2.1.1 (0.0008%, retrieved 15/12/2021). This also aligned with Zastrow et al. who suggested the use of MAF of the most frequent pathogenic variant in the general population as the threshold for BS1 (26).

Computational / predictive data-driven rules (PVS1, PS1, PP3, BP4, BP7, PM5)

Null variant in a gene where loss of function is a known mechanism of disease (PVS1)

The majority of pathogenic APC variants are protein truncating (nonsense, frameshift, splice, and single/multi-exon deletions, and duplications), which leads to the disruption of β-catenin regulatory domains and subsequent loss of APC tumour suppressor function. The APC VCEP derived considerations to nonsense-mediated decay (NMD), alternative transcript, variant type, and strength-level adjustment based on known genotype-phenotype correlation (Figure 2A and Supplemental Table S3). While NMD represents an important contributor to variant pathogenicity for other genes, it is less relevant for APC as its last exon (exon 16) comprises 77% of the protein (codons 653–2843) including several important functional domains (Figure 1). Indeed, truncated APC alleles were consistently detected in the transcript analyses of leukocyte RNA without NMD blockade (27, 28). Allele-specific expression demonstrated that premature termination in exon 16 did not trigger NMD and in other exons only partial NMD (29, 30).

Figure 2. PVS1 decision tree (A) and canonical splice variant modified weights (based on reference sequence NM_000038.6) (B).

(a) Splice variants must not have any detectable nearby (+/− 20 nucleotide) strong consensus splice sequence that may reconstitute in-frame splicing. (b) For details refer to Figure 2(B). PVS1_variable is applicable to listed variants only. (c) For Guanine to non-Guanine last nucleotide changes, evidence strengths are downgraded by one level. (d) For +2T>C changes where native splice site is not or weakly predicted, strengths are one level down from the other canonical ±1/2 splice variants at the same site. (e) For full gene deletions of a known haploinsufficient gene, a pathogenic classification is warranted in the absence of conflicting evidence with PVS1 alone. (f) Not applicable if promoter 1A and 1B are also deleted. Abbreviations: NT – nucleotide, Mod – moderate, Supp – supporting, N/A – not applicable.

Most pathogenic APC variants in FAP families are located in the 5´half of the gene. Well known statistical genotype-phenotype relationships include pathogenic variants 5’ of codon 168, between codons 312 and 412 (alternatively spliced part of exon 10), or 3’ of codon 1580, which tend to be associated with a milder (attenuated) colorectal phenotype (less or later onset adenomas), whereas pathogenic variants between codons 1250–1464 usually cause a severe, early-onset disease (23) (Figure 1). To reflect this, the APC VCEP defined the applicability of PVS1 at the extremities of the gene by evaluating the 5’-most and 3’-most variants. The variant NM_000038.6:c.147_150del p.(Lys49AsnfsTer20) was absent from population databases (PM2_Supporting), reported in nine index patients meeting three phenotype points (PS4_Moderate), and segregated with FAP in three meioses in two families (PP1) (25, 31–33) (unpublished data). Based on a cautious assumption that protein truncation provides only moderate evidence for pathogenicity with relative odds of 4.33:1 (14), this was the 5’-most LP variant based on available evidence (combination of two supporting and two moderate criteria). Under the same rationale, NM_000038.6:c.7932_7935del p.(Tyr2645LysfsTer14) was the 3’-most variant that could be classified as LP based on the combination of one strong, one moderate and one supporting criteria: This variant was absent from gnomAD (PM2_Supporting) and reported in nine index patients meeting four phenotype points (PS4) (34–38) (unpublished data). Because no truncating variant upstream/downstream of these two can be classified as (L)P, the current inclusive boundaries for the application of PVS1 were defined by their corresponding codons.

A combination of RNA analysis and splice prediction data from SpliceAI, MaxEntScan and VarSeak were considered in the assignment of PVS1 strength to canonical ±1/2 splice single nucleotide variants and guanine (G) to non-G change at the last nucleotide of each coding exon (Figure 2B and Supplemental Table S3). The impact on reading frame was interpreted only when the in silico predictions were concordant, and the more conservative prediction was always used unless RNA evidence was available to corroborate the prediction. Based on this, canonical ±1/2 splice variants were assigned to List A to E with decreasing level of evidence strength from very strong to not applicable. G to non-G last nucleotide changes were also evaluated and weighed with one level downgrade in strength from the corresponding canonical sites if the splicing predictions were up to the same standards.

Full gene and frameshifting single-/multi-exon deletions fulfilled PVS1, as well as in-frame deletion of exon 13 and/or 14, where there was convincing phenotypic data (27), (unpublished data). Full gene deletions were considered Pathogenic by default. For other single-/multi-exon deletions with preserved reading frame the strength level of PVS1 was downgraded to PVS1_Moderate. Proven tandem duplication with disruption of reading frame reached PVS1, while presumed tandem duplications only reached PVS1_Strong. Finally, since there is another transcript (NM_001127511.3) with an alternative first coding exon located 5’ of the first coding exon of NM_000038.6 and there are no reports of variants at the initiation codon in patients with relevant phenotype (internal data), PVS1 was deemed not applicable to variants affecting the initiation codon. Given the complexity in the mechanism of disease and phenotype variability of promoter variants, the VCEP did not allow the use of PVS1 for variants in the promoter region and recommend that these variants should be assessed on a case-by-case basis. To evaluate the usability of the criteria for promoter variants, an FAP-associated promoter 1B deletion was also included in the pilot study.

Missense variant in gene where only LOF causes disease (BP1)

As APC is a gene for which primarily truncating variants are known to cause disease (39), the missense variant type was regarded as evidence for benign classifications by the APC VCEP (BP1). The central and C-terminal domains of the APC protein are natively unfolded by bioinformatics predictions and verified experimentally by some studies, which likely explains the resistance of the APC protein to missense variation (40). However, this criterion was not applicable to missense variants located within the first 15-amino acid repeat of the β-catenin binding domain (codon 1021–1035) due to the presence of two LP variants in this region: NM_000038.6:c.3077A>G p.(Asn1026Ser) and NM_000038.6:c.3084T>A p.(Ser1028Arg) (Supplemental Table S3).

Same or other amino acid change at the same position (PS1, PM5)

The APC VCEP allowed the application of PS1 and PS1_Moderate for missense variants that resulted in the same amino acid change as previously established P and LP variants, respectively. Similarly, the use of PM5 and PM5_Supporting was allowed for missense variants at amino acid positions where a different missense change determined to be (L)P has been seen before. There are currently only two missense variants in APC that can be classified as LP (c.3077A>G p.(Asn1026Ser) and c.3084T>A p.(Ser1028Arg)), as detailed in the explanation to PS1 in Supplemental Table S3. Other variants leading to the same missense change at these positions meet PS1_Moderate. No apparent missense variant has been classified as Pathogenic based on current evidence. The APC VCEP further specified that PS1 and PS1_Moderate can also be used for a splice variant when it occurs at the same nucleotide position as a previously established (L)P variant and has comparable or worse splice predictions.

Protein-related in silico predictive tools (PP3, BP4, and BP7)

The large, unstructured central region of the APC protein poses unique challenge to in silico tools, which rely heavily on the accurate alignment of nucleotide sequences for the prediction of variant pathogenicity. Pathogenicity predictions by five protein-related computational tools (Align-GVGD, SIFT, PolyPhen2, MAPP, REVEL) differed widely in their predictions of pathogenicity (range 17.5–75.0%) and benignity (range 25.0–82.5%) for APC missense variants in ClinVar (41). Moreover, the predictions for the only known LP APC missense variants (c.3077A>G p.(Asn1026Ser) and c.3084T>A p.(Ser1028Arg)) did not show an unequivocally deleterious effect across different tools. As a result, the APC VCEP did not recommend the use of protein-related computational prediction models (based on amino acid intrinsic features, sequence conservation, evolution, etc.) for missense variants (PP3 and BP4) at this time. However, splicing predictors should be used for presumed missense variants to reveal any splicing effect (PP3). For synonymous and intronic variants, the APC VCEP encouraged the use of approved splicing predictors including SpliceAI, MaxEntScan, and varSEAK to assess splicing, and the use of PP3/BP4 was permitted with ≥ 2 splicing predictors showing consistent splicing consequences. Synonymous and deep intronic variants (beyond +7/−21) variants without apparent effect on splicing could be classified as LB (BP4, BP7).

Experimental data-driven rules (PS3 and BS3)

The APC gene encodes a large multifunctional protein which is involved in several biological and developmental processes (Figure 1) (42). Germline loss-of-function variants in APC cause FAP through activation of the canonical Wnt/β-catenin signalling pathway (43). Wnt/β-catenin-regulated transcription drives cell proliferation, survival and the maintenance of an undifferentiated state, which becomes overactivated in the absence of APC and lead to the development of colorectal adenomas. The APC VCEP systematically reviewed the literature for all published functional data of APC variants, evaluated the validity of different types of assays used in the field and derived gene-specific recommendations for their applicability and evidence strength level for variant classification in line with current guidelines (18, 44) (Supplemental Table S4).

In the context of careful experimental design, the APC VCEP viewed β-catenin regulated transcriptional assays and surface plasmon resonance binding analysis of β-catenin as acceptable supporting evidence for APC variant interpretation under specific circumstances. These assays were considered applicable to APC variants located within the β-catenin binding domain between codons 959–2129 (42). In addition, RNA assays in germline patient-derived samples have been well-established for the detection of abnormal splicing, which represents an appreciable disease mechanism in APC. The strength level of RNA evidence has been modified to reflect this (Table 1).

Clinical data-driven rules (PS4, BS2, PS2, PM6, PP1, BS4, BP2 and BP5)

Increased prevalence of a variant in affected individuals compared to controls (PS4)

Due to the intra- and interfamilial variability of the colorectal phenotype, genotype-phenotype correlations, extraintestinal manifestations, and other polyposis syndromes resembling FAP, the APC VCEP performed a rigorous review of the available evidence and established a point-based system for scoring phenotypic information relevant to criteria PS4, PS2/PM6 and PP1 (Table 2). Given the exceedingly low allele frequency of most pathogenic APC variants, no case-control studies of FAP cohorts reaching statistical significance were available. The APC VCEP therefore defined the absolute number phenotype points required in affected individual for different PS4 strength.

Table 2.

Phenotype scoring relevant to criteria PS2, PS4, PM6, PP1 and BS4 (max. 1 point per proband)

Phenotypic consistency	Phenotype highly specific for APC	Phenotype consistent with APC but not highly specific
Phenotype point per proband	1	0.5
Polyposis	Typical colorectal phenotype: 20–99 colorectal adenomas (a) and ≤ 20 yr OR ≥ 100 colorectal adenomas (a) and ≤ 30 yr OR ≥ 1000 colorectal adenomas (a) at any age OR other accepted descriptor (b) of colorectal adenomas (a) at any age	Other colorectal phenotype: ≥ 20 colorectal adenomas (a) at 20 to 70 yr OR a documented diagnosis of FAP / AFAP OR ≥ 100 / any accepted descriptor (b) of colorectal polyps without histological confirmation
Desmoid(s)	without somatic CTNNB1 variant	Unknown CTNNB1 status
Medulloblastoma	WNT subtype without somatic CTNNB1 variant	Unknown subtype and/or CTNNB1 status
Hepatoblastoma	without somatic CTNNB1 variant	Unknown CTNNB1 status
CHRPE		Multifocal/bilateral
Multiple gastric adenomas		Presence (≥ 2 gastric adenomas)
Multiple duodenal adenomas		Presence (≥ 2 duodenal adenomas)
Osteoma(s)		Presence
Family history		Typical FAP family history (dominant pedigree pattern) (c)

^(a)Histologically confirmed adenomas, description of colorectal polyps without confirmation of histology is not accepted.

^(b)Other accepted descriptors include uncountable, innumerable, countless, and carpeting, which refers to the coverage of the entire colon with distinct polyps. A single laterally spreading lesion covering a local area is not accepted.

^(c)Excluded from scoring for PS2 / PM6 and not applicable if PP1 is already used; can only be used if at least one variant heterozygote from the family and one additional relative each fulfill at least 0.5 points.

Observed in healthy adult individual (BS2)

In APC VCEP’s terms, a healthy individual must be ≥ 50 years old, and either (i) had no CRC/polyposis-related indication for genetic testing or (ii) had less than five colorectal adenomas detected in a colonoscopy, but no other relevant phenotypic features (for details regarding this definition see Supplemental Table S3). A variant heterozygote reported in a control, non-cancer, normal, or unaffected population, but lacking the above information, was counted as half a healthy individual points, thus requiring more individuals to satisfy BS2. BS2 was met when a variant was observed with ≥ 10 healthy individual points and BS2_Supporting with ≥ 3 healthy individual points. The use of the non-cancer dataset of gnomAD was not considered a valid source of healthy heterozygous adult individuals due to the lack of phenotypic information (e.g., insidious gastrointestinal polyps) and to avoid evidence double counting with BA1 or BS1.

Based on our knowledge, there are no reports of homozygous pathogenic germline APC variants in FAP patients, likely due to the lethal nature of homozygosity as observed in embryonic mouse development (45). Therefore, the observation of a germline variant in a homozygous state ≥ two times in the non-cancer dataset of gnomAD was also considered strong evidence for benign classification (BS2).

De novo data (PS2 and PM6)

APC encodes a large multifunctional protein comprising 2,843 amino acids and is prone to spontaneous variation. Up to one quarter of APC variants occur de novo, which counteract the survival disadvantage of FAP and maintain its disease prevalence in a variant-selection balance (46–49). Bona fide de novo occurrence was ascertained when parents lacked phenotypic features as described in Table 2. The definition of an unaffected parent was set to be more stringent than that of a healthy unaffected individual as described for BS2. Somatic and postzygotic mosaicism needs to be considered, as they are frequently associated with a milder colorectal phenotype in index patients (49) and can also be present in (asymptomatic) parents (50, 51). The APC VCEP argued that both mosaicism in index patients and parents can be used for PS2. For low-level somatic / postzygotic mosaicism (<10%) in index patients the presence of the variant should be confirmed in at least one affected tissue sample.

Co-segregation with disease in multiple affected family members (PP1) or lack of segregation (BS4)

For segregation data, family members are deemed affected if they meet at least 0.5 phenotype points as described in Table 2 or if they have ≥ 10 or “multiple” colorectal adenomas. Only genotype- and phenotype- positive individuals and phenotype-positive obligate heterozygotes should be included when counting meioses for PP1. Heterozygotes who have received chemoprevention may have a milder phenotype and may also be included.

When a particular variant segregates with a phenotype in a family, it provides evidence for association of the locus with the disease but not evidence for the deleteriousness of the variant itself. The pathogenicity of the variant can be inferred from such evidence, with the caveat that the variant under interrogation may be in linkage disequilibrium with the true pathogenic variant in the family. Multigene panel testing and full gene sequencing can also reduce the confounding effects of linkage disequilibrium and ascertainment bias. To qualify for lack of segregation, one or more affected genotype-negative members of the family must reach in total at least 1 phenotype point (BS4) or one genotype-negative member has at least 0.5 phenotype points (BS4_supporting).

Co-occurrence with pathogenic variants (BP2) or with alternative molecular causes for disease (BP5)

In the context of a fully penetrant dominant disorder, the detection of an APC variant in trans with a (L)P variant could be considered supporting evidence for benign classification. The observation of a variant in an unknown phase with ≥ 3 different (L)P variants would also satisfy BP2. Established genetic causes for other molecular subtypes of the colorectal polyposis phenotype include heterozygous germline variants in POLD1 or POLE (Polymerase-proofreading-associated polyposis), biallelic variants in MUTYH (MUTYH-associated polyposis), NTHL1 (NTHL1-associated tumor syndrome), MSH3, MBD4, and the MMR genes MLH1, MSH2, MSH6 or PMS2 (germline mismatch repair deficiency) (BP5). (49, 52–57).

Validation through pilot variant classification

Representative APC variants (n=58) were selected to encompass a range of variant types, including 25 presumed missense, 7 presumed synonymous, 8 truncating (nonsense/frameshift), 1 stop-loss, 4 splice site, 7 intronic, 3 inframe deletion/insertion variants, and 3 large deletions or duplications including a promoter 1B deletion. Collectively, all applicable APC-specific ACMG/AMP codes were utilised in the classification of the pilot variants except BS4 (lack of segregation in affected relatives). The most frequently applied code was PM2_supporting (n=39), which showed the rarity of APC variants in general. The gnomAD v2.1.1. non-cancer population database contained 17 of the pilot variants, for which either BA1 (n=9) or BS1 (n=8) was applied. A total of eight institutions submitted clinical data for 50 variants, which aided in the classification of 39 variants through the application of PS4, BS2, PS2, PM6, and/or PP1. Experimental evidence was validated with corresponding codes (PS3/BS3) applied for 16 variants. PVS1 was used in 11, PS1 in one, PM5 in five, PP3 in four, BP4 in nine, and BP7 in seven variants. A list of all pilot variants, their assertions by ClinVar submitters, and their APC VCEP-approved classifications by the APC rule specifications with evidence codes applied are listed in Supplemental Table S2.

The classification of pilot variants by APC-specific ACMG/AMP criteria were compared with their respective classification on ClinVar, which depending on the number and quality of submissions, could be considered as a standard for validation of gene-specific rules. There were 15 (L)B variants, 18 (L)P variants and 25 VUS on ClinVar, which included nine variants with conflicting assertions, defined by multiple discordant interpretations by ClinVar submitters without an overwhelming majority (≥3). Specifically, these included two variants with (L)B vs VUS and seven variants with (L)P vs VUS classifications. The classification outcome of the pilot variants by the APC-specific rules, as compared with their overall ClinVar classification, is shown in Figure 3. In summary, classification by APC-specific ACMG/AMP criteria was largely consistent with ClinVar classification. All six ClinVar B variants remained B after reclassification. 67% LB variants (6/9) were reclassified as B, while one variant NM_000038.6:c.754A>G p.(Thr252Ala) as VUS due to the paucity of clinical data. Three of the 12 P variants were downgraded to LP, and one to VUS. One of the six LP variants were reclassified as P and one as VUS. The three P variants reclassified as LP were NM_000038.6:c.423–11A>G (PS3_moderate, PS4_moderate, PM2_supporting, and PP3), NM_000038.6:c.835–8A>G (PS3_moderate, PS4_moderate, PM2_supporting, and PP3), and a frameshift deletion from exons 4–7 (NC_000005.10:g.(?_112775619)_(112801393_?)del, PVS1 and PM2_supporting). Each of these variants were interpretated by a single submitter only in ClinVar. The strict control of evidence quality inherent to the APC-specific criteria may have resulted in the use of experimental and clinical codes at lower weights than ClinVar submitters and therefore a less definitive classification. Although it is worth noting that an LP classification is nevertheless possible with a compilation of evidence from different domains. In practical terms, an LP classification has a posterior probability of pathogenicity of 0.9 to 0.99, which still warrants clinical action (14). The P variant reclassified as VUS was NM_000038.6:c.32dup p.(Gln12AlafsTer3) and the LP variant was NM_000038.6:c.8514C>A p.(Tyr2838Ter), which both met PS2_supporting and were located at the extremities of the protein and therefore outside of the boundaries for PVS1 application. Notably, c.32dup has been observed in heterozygous state in 3 healthy unrelated adult individuals (BS2_Variable not met; unpublished data). All three variants reclassified as VUS by the APC-specific criteria had only one or two submissions on ClinVar, which suggested the deficiency in evidence behind their initial ClinVar classification. Among the 25 VUS by ClinVar assertions the application of the APC-specific criteria allowed the reclassification of 56% of the VUS (14/25) into a clinically meaningful pathogenicity class (20% each were reclassified to B and LB (5/25), and 16% to LP (4/25)). Importantly, these included the reclassification of 56% variants with conflicting interpretation (5/9).

Figure 3.

Classification of the 58 selected pilot APC variants by the original ClinVar assertion (left) and the APC-specific ACMG/AMP guidelines (right).

DISCUSSION

As the paradigm of modern genetics shifts from variant identification to interpretation, characterising the clinical significance of variants becomes imminent for the translation of genetic testing into medical practice. The standardised terminology and guidelines developed by the ACMG/AMP provided the fundamental backbone for up-to-date variant classification, but not enough granularity for the precise interpretation of variants in specific genes and diseases. At the same time, as a guideline designed to have universal applicability, some of the original ACMG/AMP criteria are unavoidably ambiguous, making it prone to subjectivity and user-to-user variability.

In this study, we assembled a multidisciplinary consortium of clinicians and scientists in relevant fields leveraging the depth of disease expertise in the InSiGHT consortium to conduct evidence-based expert panel review of the APC gene using the ClinGen variant curation expert panel process. In alignment with the ACMG/AMP parent framework, the APC-specific variant interpretation guidelines assume a single-variant disease relationship for a high penetrance monogenic condition. The criteria in general cannot be applied to frequent low/moderate penetrant variants such as NM_000038.6:c.3920T>A p.(Ile1307Lys) and NM_000038.6:c.3949G>C p.(Glu1317Gln), where the clinical presentation and disease mechanism are more heterogenous and complex (58, 59).

The APC VCEP paid particular attention to ensure the mutual exclusivity of the classification codes so that the same evidence is not counted twice in the gene-specific criteria. The original ACMG/AMP codes were extended with meticulous details, especially in the clinical and experimental domains, in an effort to accurately depict the phenotypic variability of FAP and the functional diversity of the APC protein. As a medically actionable gene, the classification of an APC variant into either the benign or pathogenic category has important and long-lasting clinical implications. The detailed specifications for the evidence in the APC-specific criteria therefore serve as a quality assurance tool and reduces the risk of false-positive interpretation. At the same time, the APC VCEP acknowledged that certain requirements in the gene-specific criteria were quite restrictive and high-quality data might be difficult to obtain. To not dismiss evidence lightly and avoid misinterpretation of variants with clinical consequences, the VCEP also allowed strength downgrade for evidence wherever possible to accommodate for the design of cohort studies, the data structure of reference population databases and the set-up in routine diagnostic and screening context.

Overall, the APC-specific ACMG/AMP codes performed satisfactorily in the pilot study, resulting in largely consistent interpretation of well-documented benign and pathogenic variants in ClinVar, and a reclassification of 56% (14/25) of VUS into 10 (L)B and 4 LP variants. Application of the gene-specific rules help to reclassify a substantial portion of all APC VUS into a clinically relevant pathogenicity class which is particularly important for the large number of VUS listed in ClinVar. While two of the 18 (L)P and 1 of the 15 (L)B pilot variants were reclassified as VUS, this proportion is likely to be lower in the large group of APC variants because the pilot variants belonged to a selected group of variants that covered a wide range of classification scenarios and are not representative for the distribution of variants as a whole.

While functional assessments of variants, especially RNA-based analyses, are relatively well published in the literature, the clinical data needed for classification (phenotype, proband count, segregation, and de novo status) are less well-described or remained private for internal use by individual laboratories. Our work highlighted a process for standardised aggregation of case-level information from a range of different laboratories, which was paramount in the classification of the pilot variants and provided incentives for data sharing. The validity of clinical data depends heavily on the documentation of well-phenotyped individuals prepared by clinicians and genetic service providers, and the competency of biocurators at analysing phenotypic information. Establishing the infrastructure for standard variant reporting and proficient variant interpretation training would facilitate accurate and consistent application of clinical evidence. To lay the groundwork for perspective expert panel approval for the substantial number of VUS and conflicting APC variants submitted to ClinVar, the next step will be the design of a streamlined algorithm to systematically and comprehensively evaluate a variant and to implement this strategy in a large-scale classification approach including the use of variant prioritisation features of the ClinGen Variant Curation Interface. Prioritised lists of promising causative APC variants that remain at VUS will be subjected to a data mining and molecular-driven work-up to collect further clinical and experimental evidence.

To resolve the interpretative challenges of variants in the post-genomic era, an APC subcommittee of the InSiGHT and ClinGen Hereditary Colorectal Cancer / Polyposis VCEP was constituted and APC-specific variant classification criteria were developed. Future steps of the APC VCEP include the curation of variants in the ClinVar and InSiGHT LSDB with the outcome of an expert panel-approved status. The APC-specific specifications will evolve as more evidence underlying variant pathogenicity is discovered, and as the general recommendation for the ACMG/AMP guidelines from the ClinGen SVI working group or other entities continues to develop. The most up-to-date version of the VCEP specifications are made publicly available at www.clinicalgenome.org. Moving forward, the APC VCEP will proceed with standardised interpretations of prioritised lists of VUS, the results of which will represent the most authoritative variant classification for widespread clinical use.

Supplementary Material

1

Supplemental Table S1 APC exon numbering (exon numbering of NM_000038.6 is used)

2

Supplemental Table S2 Pilot variants

4

Supplemental Table S3 Further comments from the APC VCEP regarding APC-specific modifications

5

Supplemental Table S4 Functional Tests

List of abbreviations:

ACMG

American College of Medical Genetics and Genomics

AMP

Association for Molecular Pathology

APC

adenomatous polyposis coli

B

Benign

BA

Benign stand-Alone

BS

Benign Strong

BP

Benign Supporting

ClinGen

Clinical Genome Resource

CHRPE

congenital hypertrophy of the retinal pigment epithelium

EMBL-EBI

European Molecular Biology Laboratory and European Bioinformatics Institute

gnomAD

Genome Aggregation Database

FAP

familial adenomatous polyposis

FDA

Food and Drug Administration (United States)

HGVS

Human Genome Variation Society

InSiGHT

International Society for Gastrointestinal Hereditary Tumours

LB

Likely Benign

LP

Likely Pathogenic

LOF

loss of function

LOVD

Leiden Open Variation Databases

LSDB

locus-specific databases

MAF

minor allele frequency

MANE

matched annotation from NCBI and EMBL-EBI

NCBI

National Center for Biotechnology Information

NGS

next-generation sequencing

P

Pathogenic

PVS

Pathogenic Very Strong

PS

Pathogenic Strong

PM

Pathogenic Moderate

PP

Pathogenic Supporting

SNV

single nucleotide variant

SVI WG

Sequence Variant Interpretation Working Group

VCEP

variant curation expert panel

VUS

variant of uncertain significance

Data Availability

Data are available upon request. All variants reviewed and reclassified by the ClinGen-InSiGHT Variant Curation Expert Panel in this study have been submitted to the ClinVar Database (https://www.ncbi.nlm.nih.gov/clinvar/). The detailed evidence used for the classification of these variants is available in the ClinGen Evidence Repository (https://erepo.clinicalgenome.org/evrepo/). These data may also become available upon a data transfer agreement approved by the local ethics committee and can be obtained after contacting the corresponding author (X.Y.) upon request.