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

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 ACMG/AMP guidelines for the ataxia telangiectasia mutated (ATM) gene according to the Food and Drug Administration (FDA)-approved ClinGen protocol. These gene-specific rules for ATM were modified from the American College of Medical Genetics and Association for Molecular Pathology (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 reevaluation 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 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.


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 and VUS 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 2fold 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 disease Ataxia Telangiectasia (A-T) (MIM# 208900)], a severe, early-onset disorder characterized by progressive cerebellar ataxia and ocular telangiectasias 12 and increased cancer risk most commonly for leukemia and lymphomas 13 .Large epidemiological and molecular studies have demonstrated that variants that cause A-T in the biallelic state are also expected to cause increased risk of breast and pancreatic cancer 14,15 .As such, variants that cause A-T in the biallelic 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 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 variants of uncertain significance (VUS) 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 3/19/2024).
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.

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 FDA-approved 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 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 patients 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, Variant of Uncertain Significance, Likely Pathogenic and Pathogenic) and evidence combinations with a few modifications that are supported by a Bayesian framework 16 .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, PS4_Moderate, PM1, PM6, PP1, PP2, PP4, PP5, BS2, BS4, BP1, BP3, BP5, BP6) The HBOP VCEP chose not to adopt numerous ACMG/AMP codes for ATM for several reasons (Table 1).First, breast cancer is relatively common and the majority of it is nonhereditary, 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 two-fold risk which leads to substantial phenocopy and unaffected carriers of pathogenic variants within a family.The codes that were not adopted are detailed below.

PS2/PM6: De novo
The observation of a de novo variant in the setting of a new disease is evidence towards 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.

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

PS3
Well-established in vitro or in vivo functional studies supportive of a damaging effect •Protein functional studies (Table 5) •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: radiosensitivty 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.

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 or present in ≤ .001% in all sub-populations •EXCEPTION: under-represented sub-populations with N=1 but frequency >.001% •Not considered a conflicting piece of evidence for variants that otherwise are likely benign/benign •Use as PM2_Supporting (not moderate)

PM3
For recessive disorders, detected in trans with a pathogenic variant.

PM4
Protein length changes due to in-frame deletions/insertions in a nonrepeat 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.

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

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 •Use caution in applying the wrong type of computational evidence (protein vs. RNA) towards 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, see PM3 for specific phenotype considerations ( Table 6)

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 .Regarding the use of BS4 (lack-of-segregation) in families with biallelic 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 biallelic-unaffected patient.

PP4/BS2: Phenotype
A patient who has a phenotype that is highly specific for a disease or an unaffected patient 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 a patient with an alternate mechanism for disease This rule does not apply, because there are numerous examples of patients 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 were carriers of a single pathogenic variant 18 .

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 3/19/2024).

PP2 and BP1: Low Rate of Benign Missense Variation
Pathogenic missense variants in ATM have been described, and there is not a specified low rate of benign missense variation.

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 19 Population Based Rules (BA1, BS1 and PM2_Supporting) BA1 and BS1.The HBOP VCEP compared parameters for both the autosomal dominant and autosomal recessive conditions ascribed to ATM to estimate population allele frequency thresholds using the Whiffen/Ware calculator 20 ; 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 autosomal recessive inheritance was selected along with a prevalence of 1:40,000 [21][22][23][24][25] .As ATM is the only gene 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 26 .
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 carriers 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 carrier is still considered eligible for PM2_Supporting.

PVS1
LoF is the mechanism of disease for ATM 22,27,28 .The rules governing the application and appropriate weight of PVS1 are based on the ClinGen SVI recommendations 29 .There are five variant types that fall under the PVS1 category: nonsense and frameshift alterations; canonical (+/-1,2) 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 [30][31][32][33] .This transcript contains a non-coding first exon (Exon 1) and 62 subsequent coding exons: Exon 2-63 (Figure 1).Of note, ATM may be annotated with four non-coding first exons leading to legacy nomenclature references in historical data.
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) 34 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 [35][36][37][38] .The N-Solenoid domain is thought to be important for protein function because there are numerous patients affected with A-T who carry alterations that are known to lead to in-frame losses in the N-Solenoid domain (Supplementary Figure 1) 30,32,[39][40][41][42][43][44][45][46][47][48][49] .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 3/19/2024).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 biallelic 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 45,47,50-53   .In addition, LoF alterations lying between p.Met1? and the next downstream, inframe methionine at p.Met94 have also been observed in A-T patients supporting that downstream methionine residues are unable to serve as an alternate start codon that would produce a rescue effect [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 patients with A-T 30,43,45,47,57-62   .Therefore, LoF alterations impacting codons between p.Met1 and p.Arg3047 are eligible for PVS1 at varying weights according to the PVS1 Decision Tree (Figure 2).

ATM Exon Numbering and Reading Frame. The ATM gene is depicted exon-by-exon. The amino acid size of each exon is depicted within the boxes in black text. The two major functional domains are outlined in red (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 twonucleotide 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.
5. Based on clinical and structural data, we have considered in-frame alterations targeting HEAT repeats as PVS1_Strong, the only exception being any very small in-frame alterations with PROVEAN score suggesting pathogenic, that were considered PVS1_Supporting 6.Based on clinical and structural data, we have considered in-frame alterations targeting FATKIN as PVS1, the only exception being very small in-frame alterations with PROVEAN score suggesting pathogenic, that were considered PVS1_Supporting 7. As far as we know, p.Arg3047Ter is the last PTC variant known to be pathogenic 8.The existence of experimental data (literature and/or personal communication from HBOP VCEP members) supporting the PVS1 weight are denoted by red-underline in the PVS1 decision tree.

PVS1 (variants listed in A)
Exon skipping or use of a cryptic splice site preserves reading frame

PVS1_Strong (variants listed in B)
Special case: use of a cryptic splice site preserving reading frame + very small Indel alteration + in silico supporting pathogenic (PROVEAN)

PVS1_Supporting (variants listed in C)
(downgraded from PVS1_Strong) 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 29 .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 (UTRs) 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 +1,2 baseline weight provided in the PVS1 Decision Tree (Figure 2).Each possible +/-1,2 splice variant is parsed into a PVS1 list (A to F) depending on reading frame and impact on the N-Solenoid or FATKIN domains (Figure 2). Figure 2 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 (Figure 2, 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 wildtype transcripts 63 .There is also one splice site at ATM c.7515+2 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 which 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 judgement of the quality and quantitative result of the RNA assay according to recent recommendations by the SVI 64 .In contrast, RNA functional studies establishing a lack of aberrant splicing can be coded as BP7(RNA).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, BP7) PS1.A variant that produces the same protein change as a known pathogenic alteration can be given PS1 towards 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, PS1 can be used as an RNA hotspot and the weight applied is per the ClinGen SVI recommendations (Table 2) 64 .

PS1_Moderate PS1_Supporting
Prerequisite for all: the predicted event of the VUA must precisely match the predicted event of the comparison (likely) pathogenic 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 [likely] pathogenic variant).For an exonic variant, predicted or proven functional effect of missense substitution(s) encoded by the VUA and (likely) pathogenic 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.
a If 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.
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 are eligible for PM4 due to the identification of numerous A-T patients harboring such pathogenic alterations 45,56 .
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 [65][66][67] .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 65 .This was further supported by application to prediction of damaging effect in large functional datasets for multiple cancer genes 68 .
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 69 .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 64 .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) 64 .
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 applied 64 (See Table 3).Using these modifications, many synonymous and deep intronic variants can be classified towards benign by applying both BP7 and BP4, in the absence of conflicting data.
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 4) 46,70,71 .Because many of the published assays have only a few variants, they contain insufficient known-pathogenic and known-benign controls for Bayesian validation 72 .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 Not applicable because the codes are unrelated 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 towards 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: 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 pathogenic variant carriers 10,11 .ORs should be statistically significant with a p-value <0.05 and a lower 95% confidence interval >1.5.
PM3. Biallelic pathogenic variants in ATM cause A-T.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 patients to meet a 'confident' or 'consistent' ATM-associated A-T phenotype with additional weight afforded to those with a 'confident' phenotype.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 5 to assign a final weight.
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 a 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 of very mild homozygous A-T patients (who are affected with dystonia and not cancer) is caused by the founder alteration ATM c.6200C>A (p.Ala2067Asp) 73 .Excepting homozygous cases, multiple cases of biallelic adult patients unaffected by A-T are additive and can be ascribed BP2 weight based on the cumulative points defined in Table 5 up to a maximum weight of BP2_Strong.Oculomotor apraxia (inability to follow an object across visual fields) or abnormal ocular saccades (rapid refixation from one object to another).d.Choreoathetosis or dystonia (involuntary movements; twisting and repetitive movements, abnormal postures).e. Peripheral axonal neuropathy OR Anterior horn cell neuronopathy 2. Oculocutaneous telangiectasia of the conjunctivae, ears, or face.3. Immunodeficiency (often frequent infections) and/or leukemia/lymphoma.

Laboratory Results:
1. ATM protein levels ≤ 15% of controls in patient fibroblast or lymphoblastoid cell lines.If ATM protein levels are slightly greater than 15%, the ATM kinase activity must be shown to be "negative or low or residual" (see notes). 2. Elevated serum alpha-fetoprotein (AFP) levels >65ug/L in a patient ≥ 2 years old.3. Increased sensitivity to ionizing radiation in patient fibroblast or lymphoblastoid cell lines.

Notes:
1. ATM protein levels ≤15% of control levels show >95% sensitivity and >98% specificity for diagnosing (A-T).Protein levels >15% may arise due to a missense variant, a leaky splicing variant, a variant resulting in a kinase-dead protein (where protein levels may not be affected), or a diagnosis other than A-T. 2. When assigning case report criteria based solely on laboratory results (i.e., presence of TWO or more of laboratory results 1-4), there is a greater likelihood that the most specific laboratory results #1 and #2 will be available, and that there will be some clinical indication that the individual(s) has A-T. 3. When assessing homozygous or in trans variants (with a likely pathogenic or pathogenic ATM variant) for possible downgrade in an unaffected individual, the individual should be 18 years or older with no evidence of A-T.

Modified Evidence Code Combinations
The HBOP VCEP adopted the original ACMG-AMP categorical evidence code combinations 1 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 likely pathogenic.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 SVI 16 (Table 6).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.

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); 7 intronic variants; and 3 synonymous variants.Of these 9 had a consensus B/LB classification in ClinVar, 13 had a consensus P/LP classification in ClinVar; 6 had conflicting interpretations, and 5 were considered a VUS.After developing and applying the VCEP rules, the final classifications achieved were 9 benign variants, 2 likely benign variants, 4 likely pathogenic variants, 12 pathogenic variants, and 6 variants of uncertain significance (Figure 3, Table 7).
Among the variants considered (likely) benign in ClinVar (n=9), the VCEP classified six as (likely) benign and three as VUS (ATM c.5556_5557delinsGA (p.Asp1853Asn), ATM c.7919C>G (p.Thr2640Ser) and ATM c.331+7G>A).Among the variants that were (likely) pathogenic in ClinVar (n=13), the VCEP classified all 13 as (likely) pathogenic.Among 11 variants classified as VUS or conflicting in ClinVar, the VCEP classified five as (likely) benign (four due to application of BA1 or BS1, and one due to the combination of BP4 and BP7), three as (likely) pathogenic (two with PM3_Strong or PM3_Very Strong; and one with the application of PVS1), and three as VUS due to limited evidence (Figure 3, Table 7).The final classifications asserted by the VCEP were submitted to the ClinGen VCI and deposited in ClinVar.

Discussion
The routine employment of Next-Generation Sequencing represents major advancement in the detection of pathogenic variants in hereditary cancer genes.However, a concomitant and seemingly exponential increase in the detection of variants of uncertain significance 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 FDA-approved ClinGen VCEP process.This body of work describes the decisions made by the VCEP towards that goal with the ultimate benefit of improving patient outcomes.
The Spanish ATM Working Group (SpATM-WG) defined gene-specific ACMG/AMP style rules for ATM, with many similar decisions on rules specifications (Supplementary Table 1)

74
. However, this VCEP also has substantial departures from the SpATM-WG rules that result largely from a more in-depth analysis related to the FDA-approved process that requires ClinGen SVI and HC-CDWG oversight and collaboration related to rules development.For example, this VCEP has justified the up-weighting of the PM3 and BP2 biallelic 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 patients who do not have sufficient ATM expression 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, rather a molecular confirmation of the disease-state of the patient.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 patients 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.
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=11; 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 towards the application of PM3 and BP2 biallelic 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 patients under PM3 leading to the increased number of LP/P variants.The VCEP is performing ongoing curation and further rule modifications taking into consideration any new information that may be forthcoming, including the development of any new ATM functional studies.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.

Figure 1 .
Figure 1.ATM Exon Numbering and Reading Frame.The ATM gene is depicted exon-by-exon.The amino acid size of each exon is depicted within the boxes in black text.The two major functional domains are outlined in red (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 twonucleotide 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.

38 ) 63 ) 1 .
Ser1893-Val3056) (exons 38 to 63) p.(Arg3047Ter) in exon 63 the most C-terminal variant known to be pathogenic Exon skipping/cryptic site disrupts reading frame Predicted to undergo NMD (p.Ser1893_Glu2979) PVS1 (variants listed in D) Exon skipping or cryptic splice site disrupts reading frame Not predicted to undergo NMD (p.Leu2980_Val3056) PVS1 (variants listed in E) (upgraded from PVS1_Strong) Exon skipping or use of a cryptic splice site preserving reading frame Special case: use of a cryptic splice site preserving reading frame + very small Indel alteration + in silico supporting pathogenic (Arg3047Ter) in exon 63 the most C-terminal variant known to be pathogenic Exon skipping or use of a cryptic splice site disrupts reading frame (all predicted to undergo NMD) PVS1 (list A) PVS1 (list A) PVS1 (list A) PVS1 (list D) PVS1 (list D) Exon skipping or use of a cryptic splice site preserves reading frame Very small Indel predicted damaging by PROVEAN Exon skipping or use of a cryptic splice site preserves reading frame, or PTC not predicted to undergo NMD Very small Indel predicted damaging by PROVEAN Choi, Y., and Chan, A.P. (2015).PROVEAN web server: a tool to predict the functional effect of amino acid substitutions and indels.Bioinformatics 31, 2745.10.1093/BIOINFORMATICS/BTV195.

Figure
Figure 2. PVS1 Decision Tree for ATM.PVS1 eligible variant types are split into five categories: initiation codon variants, nonsense and frameshift variants, ≥1 exon deletions, ≥1 exon duplications and lastNT/canonical splice variants.Considerations related to NMD, N and C terminal boundaries, domain involvement, tandemness, and splice prediction/observation inform the weight that can be afforded to the PVS1 criterion.Small in-frame events were predicted with PROVEAN and scores are provided.Nucleotides in red-underline have splice effects reported in the literature.

Figure 3 .
Figure 3. ATM Pilot Variant Categorization.33 pilot variants are displayed as community classification in ClinVar (left) 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 right.Granular detail of the type of conflict and the type of variant are presented in the table.PTC: Premature Termination Codon; CV: ClinVar

Table 3 PS1
•PVS1_Variable(RNA): Observed splice defect •NOTE: PVS1 and PVS1(RNA) has code combination restrictions: See 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 4 )
PS2De novo (paternity confirmed) in a patient with the disease and no family history.Do not use for AD or AR disease

TABLE 1 . SUMMARY OF ATM -SPECIFIC RULES SPECIFICATIONS .
BS2Observed in a healthy adult individual for a dominant (heterozygous) 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) 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 towards 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.

31+2T>|C, G, A c.-30-2A>|G, C, T c.-31-1G>|A C, T N-Solenoid (HEAT repeats)
2. PVS1 Decision Tree for ATM.PVS1 eligible variant types are split into five categories: initiation codon variants, nonsense and frameshift variants, ≥1 exon deletions, ≥1 exon duplications and lastNT/canonical splice variants.Considerations related to NMD, N and C terminal boundaries, domain involvement, tandemness, and splice prediction/observation inform the weight that can be afforded to the PVS1 criterion.Small in-frame events were predicted with PROVEAN and scores are provided.Nucleotides in red-underline have splice effects reported in the literature.