ATM Germline Variants: What They Mean and How to Read Them in Your Own Sequence Data

A heterozygous pathogenic variant in ATM is a moderate-penetrance cancer risk allele. Roughly 1 in 200 to 1 in 300 people carry one. It raises lifetime female breast cancer risk to something in the 20–30% range, against a population baseline near 12–13%. It raises pancreatic cancer risk several-fold off a small base. It also raises prostate cancer risk, with a tilt toward aggressive disease. It does not shorten life expectancy in any measured, general way, and it is not comparable to BRCA1 or BRCA2, where lifetime breast cancer risk runs 50–70%. Two pathogenic ATM alleles is a different condition entirely: ataxia-telangiectasia, a recessive neurodegenerative and immunodeficiency syndrome that presents in early childhood.

The interesting part for anyone with their own sequence data is that ATM is a hard gene to call and a harder gene to interpret. Most of the missense variants you will find are variants of uncertain significance. A meaningful fraction of pathogenic alleles are large deletions that standard short-read pipelines drop. The single most consequential ATM allele in breast cancer is a missense variant. It is not a truncation.

Getting the variant out of your own data

ATM is on chromosome 11q22.3, 66 exons, encoding a 3,056-residue PI3K-family kinase. Work against the MANE Select transcript NM_000051.4 / ENST00000675843, and write every variant in HGVS c. notation against it. Half the confusion in ATM literature comes from people quoting protein positions from a different transcript version.

Pull the region and annotate:

# exact coordinates from your GTF, not from memory
awk '$3=="gene" && /gene_name "ATM"/' gencode.v44.annotation.gtf \
  | cut -f1,4,5 > atm.bed

bcftools view -R atm.bed -f PASS sample.g.vcf.gz -Oz -o atm.vcf.gz

vep -i atm.vcf.gz --assembly GRCh38 --cache --offline \
    --mane_select --hgvsc --hgvsp --transcript_version \
    --plugin SpliceAI,snv=spliceai_scores.raw.snv.hg38.vcf.gz,indel=... \
    --plugin dbNSFP,REVEL_score,CADD_phred,gnomADe_AF \
    --custom clinvar.vcf.gz,ClinVar,vcf,exact,0,CLNSIG,CLNREVSTAT \
    -o atm.annotated.vcf --vcf

Two annotation traps here. First, C11orf65 overlaps ATM on the opposite strand, so --pick without --mane_select will sometimes hand you a consequence against the wrong gene. Second, always read CLNREVSTAT alongside CLNSIG. A single-submitter “Pathogenic” with no assertion criteria is not the same object as a 2-star criteria_provided,_multiple_submitters,_no_conflicts call, and ATM has a long tail of stale single-submitter entries.

Coverage first, then variants

Before you trust an absence of findings, check that you could have seen one. Run mosdepth over the exons:

mosdepth --by atm_exons.bed --thresholds 10,20,30 atm_cov sample.bam

At 30x mean WGS you should have near-complete coverage of ATM’s coding exons. Exome data is a different story: ATM’s GC-rich 5’ exons and several small internal exons routinely drop below 20x on hybrid capture, and that is exactly where a heterozygous variant becomes a coin flip.

The bigger failure mode is copy number. A few percent of pathogenic ATM alleles are multi-exon deletions or duplications, which a GATK HaplotypeCaller SNV/indel pipeline will not report at all. If your only data is short-read WGS, run a read-depth caller and an SV caller over the locus:

cnvkit.py batch sample.bam -r reference.cnn -p 8
manta --bam sample.bam --referenceFasta GRCh38.fa --runDir manta_out

Single-exon deletions from 30x short reads are genuinely unreliable. If a CNV call lands in ATM and matters to you, it needs orthogonal confirmation (MLPA or long-read) in a clinical lab. This is one place where “my consumer report said negative” is often just “nobody looked.”

One more artifact to check: variant allele fraction. A germline heterozygote sits near 0.5. An ATM variant in blood at VAF 0.05–0.20 is more likely clonal hematopoiesis than a germline finding, and ATM is among the genes seen in clonal expansions and in the germline of people with myeloid malignancies ¹. Distinguishing the two requires a non-blood tissue (buccal, fibroblast) or at minimum a repeat draw. A rare germline ATM variant has also been reported segregating with familial myeloproliferative neoplasms, which is why the germline-versus-somatic question is not academic ².

Which variants count

The classification tiers that matter in practice:

Truncating alleles. Nonsense, frameshift, and canonical ±1/±2 splice variants across nearly the whole coding sequence. ATM loss-of-function is a well-established mechanism, so PVS1 applies to most of these. Exceptions: variants in the final exon or the last 50 bp of the penultimate exon escape nonsense-mediated decay, and given that ATM’s kinase domain sits at the extreme C-terminus, an NMD-escaping truncation there can still be damaging. Do not apply PVS1 mechanically.

Missense. The large majority are VUS. REVEL and CADD are inputs. They are not verdicts. Case-control data on rare ATM missense variants shows the aggregate signal is real but individual-variant resolution is poor ³. The one exception worth knowing by name is c.7271T>G (p.Val2424Gly), a dominant-negative missense allele associated with breast cancer risk well above the typical truncating-variant estimate.

Common variants. If your report flags p.Phe858Leu (c.2572T>C) or p.Pro1054Arg (c.3161C>G), those are common polymorphisms. Large case-control work found common variants in ATM and other susceptibility genes unlikely to increase breast cancer risk ⁴. Filter anything above roughly 0.5% gnomAD popmax allele frequency out of your pathogenic candidate set.

Intronic and synonymous. Run SpliceAI. Treat delta scores above 0.5 as worth flagging and above 0.8 as strong. Remember that a high SpliceAI score is computational evidence (PP3). It is not a classification. RNA-seq from blood can sometimes settle it directly: look for exon skipping or intron retention in ATM reads, though ATM’s expression in whole blood is modest and you want deep coverage to see allele-specific effects.

What the risk numbers support

Breast cancer is the best-characterized association, with truncating ATM variants conferring roughly two- to three-fold relative risk, translating to a lifetime absolute risk in the 20–30% band ³. An international working group convened specifically to reconcile ATM risk estimates across studies. The consensus is a moderate-risk gene with variant-specific heterogeneity, and no single number describes it ⁵.

Prostate cancer: rare ATM variants are associated with risk in large consortium analysis, with the signal concentrated in truncating alleles ⁶. Pancreatic cancer risk is elevated several-fold, but the population baseline is low, so absolute lifetime risk lands in the single digits to low teens.

Biallelic ATM loss (ataxia-telangiectasia) carries a large leukemia and lymphoma risk, and the tumor genomes show a striking structural phenotype: ALL arising in A-T patients is tightly linked to chromothripsis ⁷. That is a mechanistic window into what ATM does (double-strand break sensing and repair coordination) rather than anything a heterozygote should extrapolate from.

Radiosensitivity and the DNA-repair question

Two separate things get conflated. ATM-deficient tumors are radiosensitive: somatic ATM mutations turn up enriched in exceptional responders to radiotherapy ⁸. Whether a heterozygous germline carrier has clinically meaningful normal-tissue radiosensitivity, including from diagnostic imaging, remains unresolved, and the working-group literature treats it as an open question ⁵.

Similarly, ATM sits in the homologous recombination pathway, which invites the assumption that ATM-altered tumors behave like BRCA2-altered ones. They do not. Direct comparison in prostate cancer found substantially lower HRD scores in ATM-altered than in BRCA2-altered tumors ⁹. If you are reading a tumor report, that difference matters for how you interpret HRD-based biomarkers.

Where this stops being a data problem

Research-grade WGS is not a diagnostic test. A pathogenic ATM finding in your own data needs confirmation in a CLIA/CAP laboratory before anyone acts on it, and interpretation belongs with a genetics professional who can take a family history. Surveillance decisions (imaging modality, starting age, interval) and anything involving a relative’s testing are clinical decisions. We do not make them from a VCF, and neither should you.

Questions people also ask

Is the ATM gene worse than BRCA? No. ATM is moderate penetrance, with breast cancer risk roughly two- to three-fold over baseline ³, while BRCA1/2 are high penetrance with lifetime breast cancer risks around 50–70%. The exception is the c.7271T>G missense allele, which behaves more aggressively than typical ATM truncations.

What is the life expectancy of someone with an ATM mutation? For a heterozygous carrier, there is no established reduction in general life expectancy. The risk profile is cancer-specific. It is not a whole-body shortening. Ataxia-telangiectasia from two pathogenic alleles is a severe childhood-onset condition with markedly reduced survival, and it is a separate clinical entity.

What cancers does ATM raise risk for? Female breast, pancreatic, and prostate cancer are the best-supported associations ⁶³⁵. Hematopoietic malignancies show an association in the germline context as well, and the somatic/clonal overlap complicates interpretation of ATM variants found in blood ¹².

Why did my report say VUS? Because most rare ATM missense variants have insufficient evidence to classify. This is expected for a large gene with a high missense tolerance and no simple functional assay in routine use. A VUS should not drive any decision, and classifications are periodically revisited as case data accumulates.

Can I tell from my own data whether it’s germline or clonal? Partially. Check VAF: near 0.5 is consistent with germline, and 0.05–0.20 in blood suggests clonal hematopoiesis. Confirming requires sequencing a second tissue.

Footnotes

1. Ryan J. Stubbins, Sophia C. Korotev, Lucy A. Godley. Germline CHEK2 and ATM Variants in Myeloid and Other Hematopoietic Malignancies. Current Hematologic Malignancy Reports, 2022. https://doi.org/10.1007/s11899-022-00663-7 ↩ ↩²

2. Shiyu Wang, Hang Chen, Eddie L. Imada, et al. Whole Genome Sequencing Uncovers a Rare Germline Variant in ATM Associated with Familial Myeloproliferative Neoplasms. Blood, 2021. https://doi.org/10.1182/blood-2021-153317 ↩ ↩²

3. David E. Goldgar, Sue Healey, James G. Dowty, et al. Rare variants in the ATMgene and risk of breast cancer. Breast Cancer Research, 2011. https://doi.org/10.1186/bcr2919 ↩ ↩² ↩³ ↩⁴

4. Caroline Baynes, Catherine S. Healey, Karen A. Pooley, et al. Common variants in the ATM, BRCA1, BRCA2, CHEK2 and TP53 cancer susceptibility genes are unlikely to increase breast cancer risk. Breast Cancer Research, 2007. https://doi.org/10.1186/bcr1669 ↩

5. Fabienne Lesueur, Douglas F. Easton, Anne‐Laure Renault, et al. First international workshop of the ATM and cancer risk group (4-5 December 2019). Familial Cancer, 2021. https://doi.org/10.1007/s10689-021-00248-y ↩ ↩² ↩³

6. Questa Karlsson, Mark N. Brook, Tokhir Dadaev, et al. Rare Germline Variants in ATM Predispose to Prostate Cancer: A PRACTICAL Consortium Study. European Urology Oncology, 2021. https://doi.org/10.1016/j.euo.2020.12.001 ↩ ↩²

7. Milind B. Ratnaparkhe, Mario Hlevnjak, Thorsten Kolb, et al. Genomic profiling of Acute lymphoblastic leukemia in ataxia telangiectasia patients reveals tight link between ATM mutations and chromothripsis. Leukemia, 2017. https://doi.org/10.1038/leu.2017.55 ↩

8. Jennifer Ma, Jeremy Setton, Luc G.T. Morris, et al. Genomic analysis of exceptional responders to radiotherapy reveals somatic mutations in ATM. Oncotarget, 2016. https://doi.org/10.18632/oncotarget.14400 ↩

9. Tamara L. Lotan, Harsimar Kaur, Daniela C. Salles, et al. Homologous recombination deficiency (HRD) score in germline BRCA2- versus ATM-altered prostate cancer. Modern Pathology, 2021. https://doi.org/10.1038/s41379-020-00731-4 ↩