Specifications of the ACMG/AMP variant curation guidelines for the analysis of germline ATM sequence variants

Summary

The ClinGen Hereditary Breast, Ovarian, and Pancreatic Cancer (HBOP) Variant Curation Expert Panel (VCEP) is composed of internationally recognized experts in clinical genetics, molecular biology, and variant interpretation. This VCEP made specifications for the American College of Medical Genetics and Association for Molecular Pathology (ACMG/AMP) guidelines for the ataxia telangiectasia mutated (ATM) gene according to the ClinGen protocol. These gene-specific rules for ATM were modified from the ACMG/AMP guidelines and were tested against 33 ATM variants of various types and classifications in a pilot curation phase. The pilot revealed a majority agreement between the HBOP VCEP classifications and the ClinVar-deposited classifications. Six pilot variants had conflicting interpretations in ClinVar, and re-evaluation with the VCEP’s ATM-specific rules resulted in four that were classified as benign, one as likely pathogenic, and one as a variant of uncertain significance (VUS) by the VCEP, improving the certainty of interpretations in the public domain. Overall, 28 of the 33 pilot variants were not VUS, leading to an 85% classification rate. The ClinGen-approved, modified rules demonstrated value for improved interpretation of variants in ATM.

This paper details ACMG/AMP-style guidelines for ataxia telangiectasia mutated (ATM) variant interpretation and shows the results of pilot classifications. Several phenotype-driven rules were omitted or modified due to low-penetrance and commonness of breast cancer. This work aims to harmonize the classification of ATM variants and avoid misinterpretation by using standard rules.

Introduction

The widespread adoption of low-cost, high-throughput, next-generation-sequencing (NGS)-based, multi-gene panel tests has led to a substantial increase in the detection of germline sequence variants. In 2015, in response to this increase, the American College of Medical Genetics and Genomics and the Association for Molecular Pathology (ACMG/AMP) provided a substantial update to their variant interpretation guidelines, addressing many of the new challenges for variant interpretation.^1,2 Because these guidelines are intended for use with any Mendelian disorder, gene- and disease-specific modifications may be needed to develop a tailored approach to variant classification. The process of tailoring variant interpretation guidelines is overseen by the National Institute of Health-funded Clinical Genome Resource (ClinGen) whose mission it is to develop an authoritative, comprehensive, central resource for expert-guided, gene- and variant-level information.^3,4,5 As part of this ClinGen initiative, the Hereditary Breast, Ovarian, and Pancreatic Cancer (HBOP) Variant Curation Expert Panel (VCEP) formed in 2018, with a goal of specifying criteria of the 2015 ACMG/AMP baseline guidelines for clinical classification of variants in ATM (MIM: 607585), BARD1 (MIM: 601593), BRIP1 (MIM: 605882), CHEK2 (MIM: 604373), PALB2 (MIM: 610355), RAD51C (MIM: 602774), and RAD51D (MIM: 602954) (https://clinicalgenome.org/affiliation/50039/). Based on the large number of variants, especially variants of uncertain significance (VUSs), in ClinVar, the ataxia telangiectasia mutated (ATM) tumor suppressor gene was selected for initial work of this VCEP.

ATM encodes a serine-threonine kinase involved in the cellular response to DNA damage.⁶ Heterozygous loss-of-function (LoF) variants in ATM are associated with approximately 2-fold increased lifetime risks for breast cancer (MIM: 114480) with a penetrance of 20%–30% and a 6.5-fold increased risk for pancreatic cancer.^7,8,9,10,11 Biallelic pathogenic variants in ATM lead to the autosomal-recessive (AR) disease ataxia telangiectasia (A-T) (MIM: 208900), a severe, early-onset disorder characterized by progressive cerebellar ataxia and ocular telangiectasias and increased cancer risk most commonly for leukemia and lymphomas.¹² Large epidemiological and molecular studies have demonstrated that variants that cause A-T in the bi-allelic state are also expected to cause increased risk of breast and pancreatic cancer.^13,14 As such, variants that cause A-T in the bi-allelic state are also considered by the HBOP VCEP to cause increased risk of breast and pancreatic cancer in the heterozygous state. Given the demonstrated increased risk for autosomal-dominant (AD)and -recessive disease, individuals with likely pathogenic/pathogenic (LP/P) variants in ATM may elect to increase cancer surveillance and/or be counseled for family planning. However, there are currently over 7,500 VUSs deposited to ClinVar, many of which are missense and non-coding variants (https://www.ncbi.nlm.nih.gov/clinvar/?term=atm%5Bgene%5D&redir=gene; accessed March 19, 2024).

Therefore, the HBOP VCEP selected ATM for development of a validated set of variant classification rule specifications modeled on the baseline 2015 ACMG/AMP guidelines. The gene-specific rules for ATM along with application of these rules to curation of a series of ATM variants are described herein.

Methods

ClinGen HBOP VCEP

The HBOP VCEP formed in 2018 and is comprised of an international team of experts with relevant backgrounds in basic science research, including protein functional analysis, clinical genetics, tumor pathology, computational principles, and/or variant interpretation. All members declared conflicts of interest as required by the ClinGen process, including several members who are full-time employees at clinical diagnostic laboratories.

The HBOP VCEP convened bi-weekly to consider the applicability, weight modifications, and gene-specific nuances of each of the categorical ACMG/AMP guidelines for ATM.¹ Initial rules were drafted based on evidence in the literature, internal laboratory data, and expert opinion and were approved for pilot phase by ClinGen’s Sequence Variant Interpretation (SVI) group, who oversees this process.

Pilot phase

The ATM-specific rules were applied in a pilot test of 33 variants comprised of multiple different types (frameshift, nonsense, synonymous, intronic, canonical intronic, missense, and structural variants) with different applicable evidence (high frequency variants, rare variants, variants identified in individuals with A-T, variants in different functional domains, and variants tested in published functional assays), and/or selected for a variety of clinical assertions in ClinVar. Relevant clinical and allelic data from unpublished sources were solicited from the membership ahead of curation. Two curators independently evaluated variants and compared results. Differences were resolved first by discussion and agreement in a separate biocurator working group that convened monthly. Differences were then escalated for a secondary review and consensus from the whole HBOP VCEP by vote. If needed, rules were modified or clarified in response to this process.

Final ATM rules

Modifications made in response to the pilot study were submitted to the ClinGen SVI for review. The final round of modifications, as recommended by the SVI, were implemented, and resubmitted for approval. Final interpretations for each of the pilot variants were curated into the Variant Curation Interface (VCI) and ultimately deposited to ClinVar. Classifications followed the original five-tier model (benign, likely benign, VUS, LP, and P) and evidence combinations with a few modifications that are supported by a Bayesian framework.¹⁵ The most recent ATM guidelines can be found on the Criteria Specification Registry and will be updated periodically as the HBOP VCEP continues their work (https://cspec.genome.network/cspec/ui/svi/doc/GN020).

Results

Rules not adopted for ATM by the HBOP VCEP (PS2, PM1, PM6, PP1, PP2, PP4, PP5, BS2, BS4, BP1, BP3, BP5, and BP6)

The HBOP VCEP chose not to adopt numerous ACMG/AMP codes for ATM for several reasons (Table 1).¹ First, breast cancer is relatively common, and the majority of it is non-hereditary or sporadic. Second, hereditary and sporadic breast cancer cannot be distinguished from each other at this time. And third, ATM has low penetrance for breast cancer, conveying only 2-fold risk, which leads to substantial phenocopy and unaffected individuals harboring a pathogenic variant within a family. The codes that were not adopted are detailed below.

Table 1.

Summary of ATM-specific rules specifications

Code	Original application	ATM-modified application
PVS1	null variant in a gene where LoF is a known mechanism of disease	per ATM exon map (Figure 1) and ATM PVS1 decision tree (Table S1) •PVS1_Variable: predicted splice defect •PVS1_Variable(RNA): observed splice defect •note: PVS1 and PVS1(RNA) have code combination restrictions; see Table 3
PS1	same amino acid change as a previously established pathogenic variant regardless of nucleotide change	•protein: this rule may be applied only when a splice defect is ruled out for both alterations either by RNA analysis and/or in silico splice predictions •RNA (use as PS1_ Variable) per SVI guidelines: see PS1 table (Table 2)
PS2	de novo (paternity confirmed) in an individual with the disease and no family history	do not use for AD or AR disease
PS3	well-established in vitro or in vivo functional studies supportive of a damaging effect	•protein functional studies (Table S2) •PS3_Moderate: A-T (ATM null cell line) failure-to-rescue studies (typically target phosphorylation) plus confirmatory radiosensitivity assay •PS3_Supporting: A-T (ATM null cell line) rescue study only •no weight: radiosensitivity only (non-specific) •RNA functional studies shall be coded as PVS1(RNA) (where RNA is for “observed”)
PS4	the prevalence of the variant in affected individuals is significantly increased compared with the prevalence in controls	•do not use for proband-counting studies •case-control studies; p value ≤0.05 and (OR, hazard ratio, or relative risk ≥2 or lower 95% CI ≥ 1.5).
PM1	located in a mutational hot spot and/or critical and well-established functional domain	•do not use: benign and pathogenic variants are known to occur within the same domains and germline mutational hotspots are not well defined at this time
PM2	absent/rare from controls in an ethnically matched cohort population sample	•variant absent in gnomAD V2.1.1 or present in ≤0.001% in all sub-populations •exception: under-represented sub-populations with N = 1 but frequency >0.001% •not considered a conflicting piece of evidence for variants that otherwise are LB/B •use as PM2_Supporting (not moderate)
PM3	for recessive disorders, detected in trans with a pathogenic variant	per A-T PM3 tables (Tables 4 and S3)
PM4	protein length changes due to in-frame deletions/insertions in a non-repeat region or stop-loss variants	•do not use for in-frame insertions and deletions as no data are available for this rule at this time •PM4 can be used for stop-loss variants (including frameshifts leading to protein extensions).
PM5	missense change at an amino acid residue where a different missense change determined to be pathogenic has been seen before	•do not use for hotspot: multiple amino acid substitutions at the same residue can be pathogenic or benign and bioinformatic tools cannot yet confidently distinguish them •apply to frameshifting or truncating variants as PM5_supporting for variants with premature termination codons upstream of p.Arg3047, which are expected to be more severe than the most C-terminal pathogenic variant p.Arg3047∗
PM6	confirmed de novo without confirmation of paternity and maternity	do not use for AD or AR disease
PP1	co-segregation with disease in multiple affected family members	•AD condition: co-segregation analysis in lower-penetrance genes can lead to false positive results •AR condition: informative instances of co-segregation in A-T families are too rare to be formally analyzed at this time; however, this VCEP supports approaching this similarly to the ITGA2B/ITGB3 and hearing-loss VCEPs who have outlined PP1 criteria for these AR disorders
PP2	missense variant in a gene that has a low rate of benign missense variation and where missense variants are a common mechanism of disease	do not use: ATM does not have a specified low-rate of benign missense variation
PP3	multiple lines of computational evidence support a deleterious effect on the gene or gene product	•protein analysis: metapredictor REVEL score ≥0.733 •RNA: SpliceAI score ≥0.2 •do not use in conjuction with PVS1(RNA) •use caution in applying the wrong type of computational evidence (protein vs. RNA) toward the cumulative body of evidence for the opposite mechanism
PP4	phenotype specific for disease with single genetic etiology	•do not use for AD disorder •for AR disorder, PP4 is built into PM3 specifications/weights (Table S3)
PP5	reputable source recently reports variant as pathogenic but the evidence is not available to the laboratory to perform an independent evaluation	do not use
BA1	allele frequency is above 5% in Exome Sequencing Project, 1000 Genomes, or ExAC	>0.5% (0.005)
BS1	allele frequency is greater than expected for disorder	>0.05% (0.0005)
BS2	observed in a heterozygous state in a healthy adult individual for an autosomal dominant disorder with full penetrance expected at an early age	do not use: ATM has reduced penetrance
BS3	well-established in vitro or in vivo functional studies shows no damaging effect on protein function	•protein functional studies (BS3) (Table S2) BS3_Moderate (protein): both radiosensitivity and ATM-null cell line rescue (usually phosphorylation of multiple substrates) are normal. Note “moderate” does not exist in the current ACMG weights for benign but can be considered as two supporting benign lines of evidence toward final classification •BS3_Supporting (protein): either radiosensitivity or ATM-null cell line rescue (usually phosphorylation of multiple substrates) are normal •note: BP4 protein predictions may be used in conjunction with BS3 for protein effects •RNA: do not use; see code BP7_Variable(RNA)
BS4	lack of segregation in affected members of a family	do not use: ATM has reduced penetrance
BP1	missense variant in gene where only LoF causes disease	do not use: ATM has known missense pathogenic variation
BP2	observed in trans with a pathogenic variant for a fully penetrant dominant gene/disorder	•See A-T PM3|BP2 table (Tables 4 and S3)
BP3	in-frame deletions/insertions in a repetitive region without a known function	do not use
BP4	multiple lines of computational evidence suggest no impact on gene or gene product	•protein analysis: metapredictor REVEL score ≤0.249 •RNA: SpliceAI score ≤0.1
BP5	variant found in an individual with an alternate molecular basis for disease	do not use
BP6	reputable source recently reports variant as benign, but the evidence is not available to the laboratory to perform an independent evaluation	do not use
BP7	a synonymous (silent) variant for which splicing prediction algorithms predict no impact to the splice consensus sequence nor the creation of a new splice site, and the nucleotide is not highly conserved	BP7: synonymous and deep intronic •can be used for deep intronic variants beyond (but not including) +7 (donor) and −40 (acceptor) •may also apply BP4 to achieve LB •is not considered a conflicting piece of evidence against a body of evidence supporting a pathogenic splice defect BP7_Variable(RNA): RNA functional studies •lack of aberrant splice defect: please see PVS1_Variable(RNA) section (above) for guidance on baseline weights and modifications of weight based on quality for RNA assays •note: BP4 splice predictions may not be used in conjunction with BP7

Points-to-weights conversions—PM3: supporting: 1 to 1.5 point, moderate: 2 to 3.5 points, strong: 4 to 7.5 points, very strong: ≥8 points; BP2: supporting: −1 to −1.5 points; moderate: −2 to −3.5 points; strong: ≤−4 points; very strong: not applicable.

PS2/PM6: De novo

The observation of a de novo variant in the setting of a new disease is evidence toward pathogenicity. The use of de novo instances is not informative for ATM because breast cancer as a “new disease” cannot be confidently established given the commonness of sporadic breast cancer.

PP1/BS4: Co-segregation

Segregation of a disease and a variant within the same family is evidence for pathogenicity. However, genes conferring lower risk (relative risk = 2) for AD conditions, should not be considered for co-segregation analysis because an unrealistic number of pedigrees is needed to obtain a true positive result while avoiding a false positive result. For example, in a gene with a relative risk of 2, like ATM, the probability of obtaining a true positive result for PP1 (as supporting strength) caps at 80% with 40 pedigrees; however, that same circumstance also comes with an ∼3.5% chance of obtaining a false positive BP4 result.¹⁶ The use of PP1 for multiple bi-allelic family members affected with A-T was considered. One pilot variant, ATM c.7271T>G (GenBank: NM_000051.3) (p.Val2424Gly), had PP1 applied as a unique exception due to the atypical high penetrance reported for this variant; however, no such instances were identified in more typical variants in the pilot phase. As such, in preparation for the occurrence of this rare possibility, the VCEP has proposed to consider previously approved segregation guidelines from other AR diseases, such as for Glanzmann thrombasthenia, if this situation arises during curation of ATM variants with typical penetrance (https://cspec.genome.network/cspec/ui/svi/doc/GN011). Regarding the use of BS4 (lack of segregation) in families with bi-allelic A-T is theoretically feasible; however, siblings with the same two variants as the A-T-affected proband would be captured under the BP2 code, which is a bi-allelic unaffected individual.

PP4/BS2: Phenotype

An individual who has a phenotype that is highly specific for a disease or an unaffected individual who has not manifested disease can be used in the pathogenic (PP4) or benign (BS2) direction, respectively. However, since hereditary and sporadic breast and pancreatic cancer cannot be distinguished and because the penetrance is low for breast cancer, neither situation can be satisfied for ATM.

BP5: Variant present in an individual with an alternate mechanism for disease

This rule does not apply because there are numerous examples of individuals carrying both an ATM LP/P variant in addition to a second LP/P in other genes whose phenotype is not different than if they harbored a single pathogenic variant.¹⁷

PM1: Variant in a functional domain without benign variation

Although ATM has well established functional domains, there are many benign variants described in these domains based on allele frequency and homozygous occurrences alone (https://gnomad.broadinstitute.org/gene/ENSG00000149311?dataset=gnomad_r2_1; accessed March 19, 2024).

PP2 and BP1: Low rate of benign missense variation

Both pathogenic and benign missense variants have been identified in ATM, and there is no indication of missense constraint; therefore, neither PP2 nor BP1 applies (https://www.deciphergenomics.org/gene/ATM/browser).

BP3: In-frame indels in a repetitive region

There are insufficient data to support the use of in-frame deletions/insertions in a repetitive region without a known function.

PP5 and BP6: Reputable source

These rules regarding reputable sources have been discontinued at the recommendation of the SVI.¹⁸

Population based rules (BA1, BS1, and PM2_Supporting)

BA1 and BS1

The HBOP VCEP compared parameters for both the AD and AR conditions ascribed to ATM to estimate population allele frequency thresholds using the Whiffen/Ware calculator¹⁹ (https://cardiodb.org/allelefrequencyapp/). Because LP/P variants in ATM are considered a relatively infrequent cause of hereditary breast cancer, the genetic heterogeneity was set to 0.02: in other words, as if 2% of hereditary breast cancer cases are caused by ATM LP/P variants. The allelic heterogeneity was conservatively set to 1.0: in other words, assuming that there is only one LP/P variant that causes ATM-related breast cancers. Lastly, the penetrance for ATM and breast cancer was conservatively set to 0.2 based on data from multiple studies of hereditary breast cancer.^10,11 Using these parameters, and a prevalence of 1:8 women for breast cancer, the maximum credible allele frequency was 0.625%. Similarly, for A-T, the AR inheritance was selected along with a prevalence of 1:40,000.^{20,21,22,23,24} As ATM is the only gene with variants that causes A-T, the genetic heterogeneity was set to 1.0, and penetrance was set to 0.90. Using these parameters, the maximum credible allele frequency is 0.527%. Given the conservative parameters put into the calculator and to simplify, the BA1 threshold was set to 0.5%. For BS1, all parameters remained the same except for the extremely conservative allelic heterogeneity value, which was dropped to 0.10, leading to an order-of-magnitude decrease in the maximum credible allele frequency of 0.05%. In applying these frequency codes, statistical models should be considered to account for error related to sample size such as the filtering allele frequency in gnomAD v2.1.1.²⁵

PM2

ClinGen has deviated from the Richards et al. ACMG/AMP guidelines for PM2 and now recommends that this evidence code be uniformly down weighted to PM2_Supporting (https://www.clinicalgenome.org/site/assets/files/5182/pm2_-_svi_recommendation_-_approved_sept2020.pdf). This recommendation was adopted for rare ATM variants. Due to the incomplete penetrance, it is reasonable to expect that unaffected individuals harboring a pathogenic variant are present in the general population. As such, a variant does not need to be absent in the general population to apply PM2_Supporting. For ATM, rarity is considered as a general population frequency of ≤0.001% in each subpopulation. Any alteration that exceeds 0.001% in a large general population database but for which there is only one individual is still considered eligible for PM2_Supporting.

LoF codes (PVS1 and PVS1[RNA])

PVS1

LoF is the mechanism of disease for ATM.^14,20,26 The rules governing the application and appropriate weight of PVS1 are based on the ClinGen SVI recommendations.²⁷ There are five variant types that fall under the PVS1 category: nonsense and frameshift alterations, canonical (+/−1 and 2) splice site alterations (and some last-nucleotide alterations), gross deletions, gross duplications, and initiation codon alterations. Several features influence the weight ascribed to PVS1 including (1) nonsense-mediated RNA decay (NMD), (2) the impact of an NMD-escaping effect on a critical functional domain, (3) the size of the NMD-escaping effect relative to the size of the protein, and (4) gene-specific features.

ATM canonical transcript

The reference transcripts considered by this VCEP are NM_000051.3/ENST00000278616.8. All exons from this transcript are considered constitutive exons without major alternative splice isoforms that could result in a rescue of PVS1-eligible variants.^28,29,30,31 This transcript contains a non-coding first exon (exon 1) and 62 subsequent coding exons (exons 2–63) (Figure 1). Of note, ATM may be annotated with four non-coding first exons leading to legacy nomenclature references in historical data.

Figure 1.

ATM exon numbering and reading frame

ATM is depicted exon by exon. The number of amino acids encoded by each exon is depicted within the boxes in black text. The two major functional domains are outlined in blue (N-solenoid, comprised of sub-domains HEAT repeat and TAN domain) and green outline (FATKIN domain comprised of the FAT and FAT-C sub-domains). Each exon is shaped to indicate the number of overhanging nucleotides at either end, which will assist in determining any reading-frame changes from gross deletions or duplications of whole single- or multi-exons. A vertical line indicates a blunt start or end with no overhanging nucleotides. An upper overhang on either side represents a two-nucleotide overhang; a lower overhang represents a single-nucleotide overhang on that side. To use this diagram, a line drawn at the start and end of a deletion or duplication will be either parallel (in-frame event) or non-parallel (frameshift) as in the examples.

ATM functional domains

ATM is comprised of two main functional domains: the N-solenoid domain and the FATKIN domain. The N-terminal half of the protein is an α-solenoid structure (N-solenoid) (amino acids 1–1892)³² that is able to interact with nucleic acids and various protein partners. The phosphoinositide 3-kinase domain (PI3-K), the focal adhesion targeting (FAT), and the focal adhesion targeting carboxyterminal (FATC) collectively comprise the FATKIN domain of ATM (Figure 1). The FATKIN domain is directly responsible for ATM kinase function, which is essential for tumor suppressor activity. Therefore, the FATKIN domain is considered critical for protein function and NMD-escaping variants, including in-frame losses and truncations between p.Leu2980 and p.Arg3047, that adversely affect the FATKIN domain are given PVS1 as Very_Strong.^33,34,35,36 The N-solenoid domain is thought to be important for protein function because there are numerous individuals affected with A-T who carry alterations that are known to lead to in-frame losses in the N-solenoid domain (Figure S1).^{28,30,37,38,39,40,41,42,43,44,45,46,47} However, compared to the FATKIN domain, there are relatively few missense pathogenic mutations (https://www.ncbi.nlm.nih.gov/clinvar/?term=atm%5Bgene%5D&redir=gene; accessed March 19, 2024). Because of this, in-frame single- or multi-exon losses impacting the N-solenoid domain can receive PVS1_Strong.

PVS1 eligibility boundaries

Because pathogenic variants in ATM cause both A-T in a bi-allelic state and cancer predisposition in a heterozygous state, this VCEP was able to leverage evidence from A-T cohorts to inform PVS1 boundaries. At the N terminus, it was determined that variants destroying the initiation codon are ascribed PVS1 as Very_Strong due to the identification of numerous A-T-affected individuals harboring p.Met1? variants.^{43,45,48,49,50,51} In addition, LoF alterations lying between p.Met1? and the next downstream, in-frame methionine at p.Met94 have also been observed in A-T probands supporting that downstream methionine residues are unable to serve as an alternate start codon that would produce a rescue effect.^{52,53,54,55,56} At the C terminus, p.Arg3047 is considered the last critical amino acid based on many reports of a nonsense variant at this position in individuals with A-T.^{28,41,43,45,50,55,56,57,58,59} Therefore, LoF alterations impacting codons between p.Met1 and p.Arg3047 are eligible for PVS1 at varying weights according to the PVS1 decision tree (Table S1).

Gross deletions

Single-to-multi-exon deletions that are frameshifting and NMD-prone receive PVS1 weight at Very_Strong as per the original 2018 PVS1 guidelines.²⁷ Alterations producing NMD-escaping transcripts that adversely affect the N-solenoid receive PVS1_Strong, and those adversely affecting the FATKIN domain receive PVS1 as Very_Strong. The HBOP VCEP has made a diagram to assist with discerning the reading frame disruption of gross deletions and duplications (Figure 1).

Gross duplications

Single-to-multi-exon duplications that do not involve either the 5′ or 3′ untranslated regions are eligible for PVS1 weight whether they are confirmed or presumed in tandem. PVS1 (as Very_Strong) and PVS1_Strong can be applied for in-frame events confirmed or presumed to disrupt the FATKIN domain, respectively; and PVS1_Strong and PVS1_Moderate can be applied for in-frame events confirmed or presumed to disrupt the N-solenoid domain, respectively. Care should be taken to ensure that the functional domains are disrupted by the duplication, which means both the 5′ and 3′ breakpoint of the duplication must be within the same domain. Duplications that have one breakpoint in the N-solenoid domain and one breakpoint in the FATKIN domain do not disrupt either domain and do not receive any PVS1 weight.

Splice variants

Canonical splice variants are defined as the +/−1 and 2 positions in the introns surrounding an exon, as well as some alterations at the last nucleotide of the exon. If the sequence does not conform to the consensus U2 donor site of Ggtrrgt (where the capital G is the last-nucleotide position of the exon and where r is any purine) then the impact of a last-nucleotide substitution on splicing is expected to be greater. Such alterations are eligible for PVS1 weight but are reduced by one strength level from the corresponding +1 and 2 baseline weight provided in the PVS1 decision tree (Table S1). Each possible +/−1 and 2 splice variant is parsed into a PVS1 list (A to F) depending on reading frame and impact on the N-solenoid or FATKIN domains (Table S1). Table S1 was informed by in silico score from SpliceAI and/or PROVEAN, in conjunction with published and unpublished splicing data. Of note, there are several variants that receive PVS1_Supporting because they are predicted to make use of an in-frame alternate splice site that preserves the reading frame and leads to a small insertion or deletion that is predicted by PROVEAN to be deleterious (Table S1, lists C and F). There are also several candidate variants that do not receive any PVS1 weight because they are +2T>C alterations that do not have a predicted splice impact by SpliceAI. Although rare, +2T>C alterations are known to produce predominantly wild-type transcripts.⁶⁰ There is also one splice site at ATM c.7515+2 (GenBank: NM_000051.3) that is atypical in that it has a native cytosine instead of the consensus thymine. Therefore, a C>T substitution here is predicted to improve the native splice sequence, and it receives no PVS1 weight.

PVS1(RNA)

Any spliceogenic variant, whether canonical, exonic, or deeper intronic, that is confirmed by RNA studies to have a deleterious splice defect can be coded as PVS1(RNA). The application of PVS1(RNA) supplants any other predictive lines of evidence (PVS1 or PP3). Of note, PS3, the code for functional data supporting a pathogenic event, is not used for RNA data because it is reserved for downstream (e.g., protein) functional effects that can be observed in conjunction with an RNA defect and applied in addition to PVS1(RNA). The weight for PVS1(RNA) can be variably ascribed based on curator judgment of the quality of the RNA assay and quantity of the resulting effect, according to recent recommendations by the SVI.⁶¹ In contrast, RNA functional studies establishing a lack of aberrant splicing can be coded as BP7(RNA). In contrast to PVS1(RNA), BP7(RNA) should be applied only to synonymous, intronic, and other non-coding variants because there can still be a protein level impact that needs to be ruled out before classifying it as (likely) benign. The weight for BP7(RNA) can be variably ascribed from supporting to strong based on curator judgment of the quality of the RNA assay.

Computational/predictive data-driven rules (PS1, PM4, PM5, PP3, BP4, and BP7)

PS1

A variant that produces the same protein change as a known pathogenic alteration can be given PS1 toward pathogenicity. This rule may only be applied when a splice defect is ruled out for both the known LP/P alteration and the variant under evaluation by in silico splice predictions or RNA evidence. If splicing is a factor for both variants, and both variants have exactly the same impact on splicing (predicted or experimentally shown), then PS1 can be used as an RNA hotspot, and the weight applied is per the ClinGen SVI recommendations (Table 2).⁶¹

Table 2.

PS1 code weights for variants with same predicted splicing event as known LP variant

Variant under assessment (VUA)	Baseline computational/predictive code applicable to VUA	Position of comparison variant relative to VUA	PS1 code applicable to VUA
			with P comparison variant	with LP comparison variant
Located outside splice donor/acceptor ±1,2 dinucleotide positions	PP3	same nucleotide	PS1	PS1_Moderate
	PP3	within same splice donor/acceptor motif (including at ±1,2 positions)	PS1_Moderate	PS1_Supporting
Located at splice donor/acceptor ±1,2 dinucleotide positions	PVS1	within same splice donor/acceptor ±1,2 dinucleotide	PS1_Supporting	N/A
	PVS1	within same splice donor/acceptor region, but outside ±1,2 dinucleotidea	PS1_Supporting	PS1_Supporting
	PVS1_Strong, PVS1_Moderate, or PVS1_Supporting	within same splice donor/acceptor ±1,2 dinucleotide	PS1	N/A
	PVS1_Strong, PVS1_Moderate, or PVS1_Supporting	within same splice donor/acceptor region but outside ±1,2 dinucleotidea	PS1_Moderate	PS1_Supporting

Prerequisite for all: the predicted event of the VUA must precisely match the predicted event of the comparison LP variant (e.g., both predicted to lead to exon skipping, or both to lead to enhanced use of a cryptic splice motif, and the strength of the prediction for the VUA must be of similar or higher strength than the strength of the prediction for the comparison LP variant). For an exonic variant, predicted or proven functional effect of missense substitution(s) encoded by the VUA and LP variant should also be considered before application of this code. Dinucleotide positions refer to donor and acceptor dinucleotides in reference transcript(s) used for curation. Designated donor and acceptor motif ranges should be based on position weight matrices for intron category (see methods). For GT-AG introns, these are defined as follows: the donor motif, last 3 bases of the exon and 6 nucleotides of intronic sequence adjacent to the exon; acceptor motif, first base of the exon and 20 nucleotides upstream from the exon boundary. Consider other motif ranges for non-GT-AG introns.

^aIf relevant, splicing assay data for a pathogenic variant outside a ±1,2 dinucleotide position may be used to update a PVS1 decision tree and hence the applicable PVS1 code for a ±1,2 dinucleotide variant. Table is reproduced with permission from Walker et al.⁶¹.

PM4

In-frame deletions and insertions as well as variants disrupting the native stop codon may be eligible for PM4. However, for ATM, there are no data available at this time to inform the use of in-frame insertions or deletions. Stop-loss variants in ATM, including frameshifts leading to protein extensions, are eligible for PM4 due to the identification of numerous A-T probands harboring such pathogenic alterations.^43,54,62

PM5

This rule is ascribed to missense variants at an amino acid residue where another pathogenic missense alteration has been identified. However, amino acid substitutions at a single residue in ATM can be pathogenic or benign. Thus, the use of this rule is not recommended. However, this rule has been co-opted as PM5_Supporting to increase the evidence for pathogenicity for LoF alterations being ascribed PVS1 or PVS1(RNA) as Very_Strong. This rule is governed by ATM’s lack of alternative splicing events that would produce a functional protein leading to a putative rescue of LoF alterations by splicing the variant out. In this manner, the use of PVS1 and PM5_Supporting will classify all ATM LoF variants as likely pathogenic even if they do not meet PM2_Supporting. PVS1/PVS1(RNA)-eligible variants (applied as Very_Strong) that do meet rarity (PM2_Supporting) will be classified as likely pathogenic with the addition of PM5_Supporting.

PP3/BP4 protein

This VCEP favors the use of the metapredictor REVEL for single-nucleotide variation and Provean for small in-frame indels as a single predictor to anticipate the impact of a protein change.^63,64,65 A REVEL score ≥0.733 is considered damaging (PP3). And a score ≤0.249 is considered neutral. This threshold is based on the general recommendation and not derived as an ATM-specific threshold at this time.⁶³ This was further supported by application to prediction of damaging effect in large functional datasets for multiple cancer genes.⁶⁶

PP3/BP4 RNA

The VCEP uses SpliceAI as a sole predictor due to its ability to accurately predict loss of native splice sites and creation of cryptic sites.⁶⁷ This VCEP did not declare gene-specific thresholds for SpliceAI but recommends those set forth by the SVI in applying PP3 to non-canonical splice variants with a SpliceAI score of ≥0.2 and BP4 to variants with a SpliceAI score ≤0.1.⁶¹ In the event that RNA data are available, and they reflect a substantial variant-specific impact, do not use both PVS1(RNA) and PP3 or BP4. However, in the event that RNA data are available, and they reflect no variant-specific impacts, PP3 or BP4 may be applied in conjunction with BP7(RNA) (see Table 3). BP4 may also be used in conjunction with BP7 (see below).⁶¹

Table 3.

Restrictions on combining criteria

	PP3	PS3|BS3	PS1	PVS1	PVS1(RNA)1	BP4	BP7	BP7(RNA)
PP3	–	✓	✓	X	X	N/A	N/A	✓
PS3|BS3	✓	–	N/A	✓	✓	✓	✓	✓
PS1	✓	N/A	–	N/A	N/A	N/A	N/A	N/A
PVS1	X	✓	N/A	–	X	X	X	X
PVS1(RNA)	X	✓	N/A	X	–	X	X	X
BP4	N/A	✓	N/A	X	X	–	✓	✓
BP7	X	✓	N/A	X	X	✓	–	X
BP7(RNA)	✓	N/A	N/A	X	X	✓	X	–

N/A, not applicable because the codes are unrelated.

BP7

This rule was originally intended for synonymous variants; however, the VCEP applied the rule to deep-intronic variants beyond (but not including) +7 at the donor site and −40 at the acceptor site. Per the SVI’s recent guidance, this code is to be applied only when BP4 is met, in which case both BP4 and BP7 would be applied⁶¹ (see Table 3). Using these modifications, many synonymous and deep intronic variants can be classified toward benign by applying both BP7 and BP4, in the absence of conflicting data. Of note, since the writing and piloting of these rules, the SVI has provided guidance that the boundaries for splice sites should be +7 and −21.⁶¹ Since the HBOP VCEP embraces these guidelines we intend to update the rules to conform in our next revision.

Functional evidence (PS3/BS3)

PS3/BS3

This is applied to protein functional studies or studies that are downstream of RNA effects. For ATM, there are multiple well-established functional studies that employ the use of ATM-null cell lines to observe the general rescue of radiosensitivity and/or ATM-specific events such as phosphorylation of ATM substrates (Table S2).^44,68,69 Because many of the published assays have only a few variants, they contain insufficient known-pathogenic and known-benign controls for Bayesian validation.⁷⁰ However, because variant controls in several studies behave as expected in these assays the VCEP has approved a maximum weight of PS3_Moderate and BS3_Moderate for a combination of functional studies that are concordant for a non-functional or functional result, respectively. For non-functional results to be used as PS3_Moderate, both an ATM-specific functional result and a non-specific radiosensitivity functional result should agree. If there is disagreement between results, then no weight should be applied toward PS3. If only the ATM-specific-study (e.g., ATM auto- or trans-phosphorylation at specific residues) result is available and reflects non-functional, a maximum weight of PS3_Supporting can be given. However, because a non-functional result from a radiosensitivity assay is not specific to an ATM defect, a non-functional result in a radiosensitivity assay alone does not achieve any PS3 weight. In the benign direction, a neutral result in either an ATM-specific assay or a radiosensitivity assay can be ascribed BS3_Supporting per each. Of note, both PP3/BP4 in silico protein predictions and PS3/BP4 protein functional studies can be co-applied.

Note that RNA functional studies reflecting aberrant splicing should be coded as PVS1(RNA) and lack of aberrant splicing as BP7(RNA). Because PS3/BS3 eligible observations measure effects downstream of splicing, it may be appropriate to apply these codes in conjunction with PVS1(RNA)/BP7(RNA).

Phenotype-related rules (PS4, PM3, and BP2)

PS4

Case-control studies with ATM pathogenic variants are expected to yield odds ratio (OR) > 2 based on the known increased lifetime breast cancer risks for individuals harboring pathogenic variants.^10,11 ORs should be ≥ 2.0 and should be statistically significant with a p value <0.05 and a lower 95% confidence interval >1.5. Of note, with rare variants where a case control analysis cannot be statistically powered, an approximation called “proband counting” can be used instead. In this method, affected probands can be weighted/counted toward pathogenicity once they reach a certain number that is designed to accommodate the disease and penetrance. It is most useful for pathognomonic gene-disease relationships with high penetrance. Because many genes cause breast cancer predisposition, and because penetrance is low, proband counting does not apply to ATM.

PM3

Biallelic pathogenic variants in ATM cause A-T (https://www.ncbi.nlm.nih.gov/books/NBK26468/). Laboratory studies are available to help rule out differential diagnoses of other ataxia-associated diseases. Of note, A-T can manifest in an atypical fashion, often called variant A-T, that usually presents in childhood with similar features but has a slower progression. The VCEP has created criteria for individuals to meet a “confident” or “consistent” ATM-associated A-T phenotype with additional weight afforded to those with a confident phenotype (Tables 4 and S3). There are several considerations in addition to phenotype that need to be reviewed when weighting and applying PM3, including identification of a second ATM variant, phase of the second variant or zygosity, and general population frequency of the variant under consideration. For the application of PM3, points ascribed to multiple probands are additive, and the cumulative points can be used as in Table 4 to assign a final weight.

Table 4.

PM3 and BP2 bi-allelic code strengths

PM3a	Classification/zygosity of the co-occurring variant	Points per unrelated A-T proband (PM3)
		Confirmed in transc	Phase unknown	Second variant unidentified or VUS	Homozygous
	Phenotype “confident”b	4	2	1	2
	Phenotype “consistent”b	2	1	0.5	1
BP2		Points per unaffected adult (>18 years old) proband (BP2)
		Confirmed in trans		Phase unknown	Homozygous (max −2.0)
	P or LP variant in a patient	−4		−2	laboratory setting
					−2
					database setting
					−1

^aVariants may not exceed general population frequency >.01%.

^bSee Table S3 for phenotype categorization of 'confident' and 'consistent'

^cDo not use observations in cis.

BP2

Each adult (over 18 years of age) without features of A-T that has an ATM variant under consideration in the homozygous state, in trans, or phase unknown with an LP/P ATM variant contributes to BP2 evidence. There are two important considerations in the application of BP2: (1) the source of the data, where a laboratory setting gets stronger weight than a database setting, due increased rigor in the former and risk of a false positive result due to technical issues like allele drop out in the latter; and (2) homozygous individuals have a maximum total weight of −2 points (equivalent to BP2_Moderate) no matter how many independent instances there are. This protects against the assumption that a variant is benign when in reality it might be hypomorphic and pathogenic, but an individual may have sub-clinical or very mild features that may be overlooked by a cancer ascertainment bias. The risk of such a phenomenon is reduced in a compound heterozygous state where the other allele is more likely to have typical risks and stronger presentation. One example is the founder alteration ATM c.6200C>A (GenBank: NM_000051.3) (p.Ala2067Asp), which causes a milder form of A-T in which homozygotes are affected with dystonia and not cancer.⁷¹ Excepting homozygous cases, multiple cases of bi-allelic adult individuals unaffected by A-T are additive and can be ascribed BP2 weight based on the cumulative points defined in Table 4 up to a maximum weight of BP2_Strong.

Modified evidence code combinations

The HBOP VCEP adopted the original ACMG-AMP categorical evidence code combinations¹ with two modifications. To achieve a minimum likely pathogenic classification for PVS1-eligible alterations, the combination of PVS1 plus one additional supporting line of pathogenic evidence is allowed to achieve LP. In addition, one strong line of evidence in the benign direction is sufficient to achieve a likely benign classification. Both specific modifications are in line with a Bayesian model of variant interpretation published by the SVI¹⁵ (Box 1). The use of several code combinations is explicitly permitted or restricted by this VCEP and/or the SVI, and these are listed in Table 3.

Box 1. Rules for combining criteria.

Pathogenic criteria

Pathogenic

• Very strong (PVS1, PVS1[RNA] PM3_VeryStrong) and

  •   a) ≥1 strong (PS1-PS4, PM3_Strong, PP1_Strong) or
  •   b) ≥2 moderate (PM1-PM6, PP4_Moderate, PP1_Moderate) or
  •   c) 1 moderate (PM1-PM6, PP4_Moderate, PP1_Moderate) and 1 supporting (PP1-PP5, PM3_Supporting) or
  •   d) ≥2 supporting (PP1-PP5, PM3_Supporting)

•   2) ≥2 strong (PS1-PS4, PM3_Strong, PP1_Strong) or

•   3) 1 strong (PS1-PS4, PM3_Strong, PP1_Strong) and

  •   a) ≥3 moderate (PM1-PM6, PP4_Moderate, PP1_Moderate) or
  •   b) 2 moderate (PM1-PM6, PP4_Moderate, PP1_Moderate) and ≥2 supporting (PP1-PP5, PM3_Supporting) or
  •   c) 1 moderate (PM1-PM6, PP4_Moderate, PP1_Moderate) and ≥4 supporting (PP1-PP5, PM3_Supporting)

•   4) 1 moderate (PM1-PM6, PP4_Moderate, PP1_Moderate) and ≥4 supporting (PP1-PP5, PM3_Supporting)

Likely pathogenic

•   1) 1 very strong (PM3_VeryStrong) and 1 moderate (PP1-PP5, PM3_Supporting) or

•   2) 1 very strong (PVS1, PM3_VeryStrong) and 1 supporting (PP1-PP5, PM3_Supporting) or

•   3) 1 strong (PS1-PS4, PM3_Strong, PP1_Strong) and 1 to 2 moderate (PM1-PM6, PP4_Moderate, PP1_Moderate) or

•   4) 1 strong (PS1-PS4, PM3_Strong, PP1_Strong) and ≥2 supporting (PP1-PP5, PM3_Supporting) or

•   5) ≥3 moderate (PM1-PM6, PP4_Moderate, PP1_Moderate) or

•   6) 2 moderate (PM1-PM6, PP4_Moderate, PP1_Moderate) and ≥2 supporting (PP1-PP5, PM3_Supporting) or

•   7) 1 moderate (PM1-PM6, PP4_Moderate, PP1_Moderate) and ≥4 supporting (PP1-PP5, PM3_Supporting)

Benign criteria

Benign

•   1) 1 stand-alone (BA1) or

•   2) ≥2 strong (BS1-BS4)

Likely benign

•   1) 1 strong or

•   2) 1 strong (BS1-BS4) and 1 supporting (BP1-BP7, BS3_Supporting, BP7_Supporting[RNA]) or

•   3) ≥2 supporting (BP1–BP7, BS3_Supporting, BP7_Supporting[RNA])

Pilot

Biocurators evaluated 33 variants of varying type and ClinVar classification in a pilot study. Clinical data were collected regarding co-occurrence data from participating clinical diagnostic laboratories and disseminated in a deidentified fashion to the biocurators. Each variant was reviewed independently by two biocurators who applied lines of evidence for a final classification. Evidence codes and classifications were compared among the biocurator group and reviewed by the HBOP VCEP. Evidence codes and classifications approved by the VCEP were submitted for SVI approval and ultimately deposited to ClinVar. The pilot curation set consisted of 10 non-splicing PVS1-eligible alterations (of a variety of variant types), 13 missense alterations (including one generated by an indel), seven intronic variants; and three synonymous variants. Of these, 10 had a consensus benign/likely benign (B/LB) classification in ClinVar, 13 had a consensus P/LP classification in ClinVar, five had conflicting interpretations, and five were considered VUSs. After developing and applying the VCEP rules, the final classifications achieved were nine B variants, two LB variants, four LP variants, 12 P variants, and six VUSs (Figure 2; Table S4).

Figure 2.

ATM pilot variant categorization

Thirty-three pilot variants are displayed as community classification in ClinVar (top) where VUS/LP/P conflicting interpretation variants and VUS/LB/B conflicting interpretation variants are binned along with consensus VUS as “ClinVar VUS/Conflict.” Interpretation with the HBOP rules specifications for ATM are on the bottom. Granular detail of the type of conflict and the type of variant are presented in Table S4.

Among the variants considered (likely) benign in ClinVar (n = 10), the VCEP classified seven as LB and three as VUSs (ATM c.5556_5557delinsGA [GenBank: NM_000051.3] [p.Asp1853Asn], ATM c.7919C>G [GenBank: NM_000051.3] [p.Thr2640Ser], and ATM c.331+7G>A [GenBank: NM_000051.3]). Among the variants that were LP in ClinVar (n = 13), the VCEP classified all 13 as LP. Among the 10 variants classified as VUSs or conflicting in ClinVar, the VCEP classified four as LB (three due to application of BA1 or BS1, and one due to the combination of BP4 and BP7), three as LP (two with PM3_Strong or PM3_Very Strong and one with the application of PVS1), and three as VUSs due to limited evidence (Figure 2; Table S4). The final classifications asserted by the VCEP were submitted to the ClinGen VCI and deposited in ClinVar.

Discussion

The routine employment of NGS represents major advancement in the detection of pathogenic variants in hereditary cancer genes. However, a concomitant and seemingly exponential increase in the detection of VUSs is an unfortunate discomfort for many patients and care providers. While it is not possible to resolve the classification of all variants, the development of a set of rules to harmonize classifications across diagnostic and research laboratories can decrease uncertainty related to differential classifications within the public domain. The HBOP VCEP was tasked to define such ACMG/AMP guidelines for ATM under the ClinGen VCEP process. This body of work describes the decisions made by the VCEP toward that goal with the ultimate benefit of improving patient outcomes.

Comparison to the Spanish ATM Working Group ATM rules

The Spanish ATM Working Group (SpATM-WG) defined gene-specific ACMG/AMP style rules for ATM, with many similar decisions on rules specifications (Table S5).⁷² However, this VCEP also has substantial departures from the SpATM-WG rules that result largely from a more in-depth analysis related to the ClinGen process that requires SVI and Hereditary Cancer-Clinical Domain Working Group oversight and collaboration related to rules development. For example, this VCEP has justified the up-weighting of the PM3 and BP2 bi-allelic codes while the SpATM-WG adopted the original SVI-expounded recommendations (https://clinicalgenome.org/site/assets/files/3717/svi_proposal_for_pm3_criterion_-_version_1.pdf). Another difference is the SpATM-WG assignment of PS3 to variants identified in A-T individuals who do not have sufficient ATM levels or substrate phosphorylation. The HBOP VCEP considers this a phenotypic line of evidence rather than a functional line of evidence as this result is not necessarily variant specific but is rather a molecular confirmation of the disease state of the individual. This concept is incorporated into the VCEP interpretation for PM3. Lastly, among other differences, the HBOP VCEP has elected to omit certain codes for A-T individuals that SpATM-WG does apply, including de novo codes PM6 and PS2, co-segregation codes BS4 and PP1, and PS4 proband counting, which this VCEP applies as PM3.

Limitations and future considerations

Rules governing variant interpretation are never finished: they must be revisited on a regular basis to accommodate the new information that is continually being identified, as has already happened since the writing of this original set of guidelines. In the next major revision of the ATM guidelines, the HBOP VCEP will revisit population-based rules in light of a new version of gnomAD (v4), will formally adopt the SVI’s recommendations that −20 is the BP7 boundary on the splice acceptor side, and will eventually need to undertake the conversion of the guidelines to the points-based system that is currently under development with the ClinGen SVI. Additional considerations for rules modifications may be considered as variants are curated. One example may be to increase the weight applied to PM4 as we curate variants that shift the reading frame after the C-terminal boundary and that also elongate the protein. Another future consideration as we accumulate VCEP-approved LP/P alterations will be to evaluate the application of PM5 (at a moderate weight instead of a supporting weight) on an exon-by-exon basis using a pooled case-control analysis to determine if we can apply this at greater-than supporting strength. In addition to revising the rules and updating them based on new data, the biocurators will continue to classify variants with these rules which may, in turn, result in the need to re-evaluate earlier-classified variants for the application of rules such as PS1 “hotspot.”

Variant interpretation resolution

The careful in-depth consideration of each rule has had an impact on ClinVar classified variants leading to a substantial decrease in the conflicting/VUS rate by nearly 50% (ClinVar n = 10; VCEP n = 6). The improvement of this VUS rate is likely related to three major features: (1) data sharing of otherwise siloed clinical data among participating clinical diagnostic laboratories toward the application of PM3 and BP2 bi-allelic codes, (2) the establishment of BA1 and BS1 frequency thresholds leading to the increased number of LB/B variants, and (3) the justified increase in weight applied to A-T probands under PM3 leading to the increased number of LP/P variants. While the decrease in VUS rate is laudable, we recognize that this can still improve. As is endemic to genes like ATM that have lower penetrance and that cause a relatively common disease, they suffer a relatively high VUS rate due to the restriction of responsibly applying variant interpretation rules. However, we are at a crossroads where the in silico tools are becoming more sophisticated, the functional studies are becoming more comprehensive, and we are amassing enough variant-level proband data that we may look forward to applying some rules with increased confidence to alleviate the VUS rate in the near future. The VCEP is performing ongoing curation and further rule modifications, taking into consideration any new information that may be forthcoming. Using this method, this VCEP aims to further reduce VUS rates and discordance in variant interpretations submitted to ClinVar with the ultimate goal of improving risk assessment and family genetic counseling.

Data and code availability

The published article includes all datasets generated or analyzed during this study. Any unpublished data, including deidentified laboratory internal clinical and RNA data can be made available upon reasonable request. The interpretations for pilot ATM variants reported in this paper are available in ClinVar: https://www.ncbi.nlm.nih.gov/clinvar.

Acknowledgments

We wish to thank the entire VCEP for thoughtful discussion and sharing of unpublished data. In addition, we wish to thank Lidia Feliubadaló, Patrick Concannon, R. Malcolm Taylor, Melissa Southey, Kelly McGoldrick, Sarah Brnich, Sarah Nielson, Huma Rana, Deborah Ritter, Steven Harrison, Sharon Plon, Heidi Rehm, Leslie Biesecker, ClinGen SVI, and the Hereditary Cancer Clinical domain WG executive committee.

In addition, the HBOP VCEP is supported, in part, by U24CA258058 (A.B.S., M.C., F.J.C., and M.d.l.H.). In addition, M.H. is supported by NIH grant 5U24CA258058; M.d.l.H. is supported by the Spanish Ministry of Science and Innovation, Plan Nacional de I+D+I 2013-2016, ISCIII co-funded by FEDER from Regional Development European Funds (Spain and the European Union) and by NIH grant 5U24CA258058-02; A.B.S. is supported by NHMRC Investigator Fellowship (APP177524); L.C.W. is supported by the New Zealand Health Research Council (22/187); C.T. is supported by Cancer Research UK (EDDPGM-Nov22/100004); M.T. was supported by the NIHR Cambridge Biomedical Research Centre (NIHR203312); W.D.F. was supported by the Canadian Institutes of Health Research FDN 148390; M.C. was supported by the Basser Center for BRCA Research; A.N.M. is supported by Breast Cancer Research Foundation; and F.J.C. is supported by NIH grants R35CA253187 and P50CA116201 and the Breast Cancer Research Foundation.

Declaration of interests

M.J.A. was a paid employee of Invitae. M.E.R., T.B., T.P., and C.C.Y. were paid employees of Ambry Genetics. L.Z. was a paid employee of Natera. S.H. was a paid employee of GeneDx.

Published: September 23, 2024