Clonal Hematopoiesis as a Confounder: Why a Variant Can Be Somatic and Still From the Wrong Tissue

Zetobit · Bioinformatics Insight Series Clinical Genomics

Bioinformatics Insight Series

Clonal Hematopoiesis as a Confounder: Why a Variant Can Be Somatic and Still From the Wrong Tissue

Every quality-control mechanism a somatic pipeline has is designed to answer one question: is this variant real? Clonal hematopoiesis defeats all of them by satisfying the criterion. A CH variant is a genuine somatic mutation, present in the specimen, reproducible on orthogonal testing, and correctly called. It is simply not from the tumor. The pipeline has no filter for provenance, because provenance was never something a variant caller was asked to determine.

Kanna Nandakumar, PhD · Zetobit, LLC

Somatic variant calling is built around a distinction between signal and artifact. Strand bias, orientation bias, read position, mapping quality, duplicate collapse, population-database filtering — the entire apparatus exists to separate mutations that are physically present in the sample from errors introduced by chemistry, amplification, alignment, or sequencing. A pipeline that passes a variant through all of it has established something specific and useful: the alternate allele is really there.

Clonal hematopoiesis is the failure mode that argument does not cover. Hematopoietic stem cells accumulate somatic mutations with age, and when one acquires a fitness advantage its descendants expand into a detectable clone. The resulting variants are somatic, real, and present in the blood. They are not errors in any sense a filter recognizes.

The definitional problem is stated cleanly in the liquid biopsy literature: because CH variants derive from hematopoietic cells, they are biologically real, present, and detectable in plasma, and in typical laboratory validation studies they would be considered true positives when confirmed with an orthogonal plasma-only test — yet they are not part of the tumor or its pathogenic mechanisms and must be considered false positives for therapeutic decision-making.¹ The proposed term for these is clinical false positives: variants erroneously assigned a tumor origin when they come from non-tumor cells.¹

That phrase is worth holding onto, because it names a category most pipelines do not have. A variant is usually true or false. This one is both, depending on which question you asked.

An orthogonal confirmation assay will reproduce a CH variant perfectly. Reproducibility was never the thing in doubt. The tissue of origin was, and no amount of re-testing the same specimen answers it.

How much of the plasma variant pool this accounts for

The scale is larger than the framing "confounder" suggests. In a prospective study of 124 patients with metastatic cancer and 47 controls, using a high-intensity assay covering 508 genes at over 60,000× raw depth with matched white blood cell sequencing, 81.6% of cfDNA mutations in controls and 53.2% in cancer patients had features consistent with clonal hematopoiesis.²

In the same work, of the somatic mutations identified in the plasma of cancer patients, only 24.4% were also present in the matched tumor.³ The majority of what a deep plasma assay detects in a cancer patient is not tumor.

At clinical scale the picture is consistent. Across 16,812 liquid profiles spanning 49 cancer types, analyzed with an assay that independently sequences plasma cell-free nucleic acids and buffy-coat white blood cell DNA, 42.3% of patients carried at least one CH variant among reportable clinical genes.⁴

The genes where it matters most are the actionable ones

If CH variants fell in genes nobody reports, this would be a curiosity. They do not. The overlap between CH driver genes and clinically actionable cancer genes is the reason the problem has teeth.

In that 16,812-patient series, the proportion of detected variants that were of CH rather than tumor origin was 39% for BRCA2, 37.9% for CHEK2, 27.4% for BRCA1, 20.1% for ATM, and 18.5% for TP53 — against 7.3% for NRAS, 5.8% for BRAF, 2.1% for EGFR, and 2.1% for KRAS.⁴

The pattern is not random. DNA damage response genes are both recurrently mutated in CH and the gatekeepers for PARP inhibitor eligibility. A prostate cancer study found CHIP variants accounted for close to half of the somatic DNA repair mutations detected by liquid biopsy, with prevalence rising steeply with patient age.⁵ Roughly half the time plasma appeared to contain a mutation that would direct PARP inhibitor therapy, it contained a CHIP variant rather than prostate cancer DNA.⁵

Conversely, hotspot-driven decisions are relatively protected. EGFR is not recurrently mutated in CH, so an EGFR hotspot assay is largely unaffected; larger panels including genes like TP53 are where false positives concentrate.⁶ The exposure scales with panel breadth — which is the opposite of the intuition that a bigger panel is a safer one.

Proportion of detected variants of CH rather than tumor origin

Gene	CH origin	Why it matters clinically
BRCA2	39%	PARP inhibitor eligibility
CHEK2	37.9%	DNA damage response; hereditary risk context
BRCA1	27.4%	PARP inhibitor eligibility
ATM	20.1%	PARP inhibitor eligibility
TP53	18.5%	Prognostic and trial-stratifying in many tumor types
NRAS	7.3%	Targeted therapy selection
BRAF	5.8%	BRAF/MEK inhibitor eligibility
EGFR / KRAS	2.1% each	Hotspot-driven decisions, comparatively protected

Figures from a 16,812-patient liquid profiling series across 49 cancer types.⁴ Proportions are specific to that cohort and assay; the gene ranking is the transferable finding, not the exact percentages.

Tissue sequencing is not exempt

It is tempting to file this as a liquid biopsy problem. It is not. Solid tumor specimens contain infiltrating leukocytes, and those leukocytes carry the same clonal mutations.

In a retrospective analysis of 17,469 patients with solid cancers sequenced with matched tumor and peripheral blood, mutational analysis of hematopoietic cells identified 7,608 presumptive somatic non-silent mutations across 396 genes in 4,628 patients; 1,075 CH mutations were detected in the solid tumor specimens of 912 patients.⁷ Annotating those variants with OncoKB classified 534 of them — 49.7% — as oncogenic or likely oncogenic.⁸ These are exactly the variants a report would surface.

The VAF signature is instructive. Median variant allele fraction for these mutations was 0.04 in the tumor specimens against 0.16 in the matched blood.⁷ A CH variant appearing in tissue is typically present at low allele fraction, because it is carried by the leukocyte fraction of the specimen rather than the tumor cells — which places it squarely in the band a pipeline would otherwise interpret as subclonal tumor biology.

The published example makes the consequence concrete: a KRAS G12R mutation in a 77-year-old woman with colorectal cancer, identified as CH-derived because its allele fraction was higher in blood than in tumor (0.087 versus 0.041). Since FDA-approved therapies in that setting require wild-type KRAS status, misattributing that mutation to the tumor without a matched blood sample would have changed the treatment decision.⁸

This is the direct link to the tumor-only problem the series has covered before. Database-based germline filtering already inflates the somatic call set by letting private germline variants through. CH adds a second population of non-tumor variants that germline filtering cannot touch, because they are not germline — they will not appear in gnomAD, they are not present in every cell, and their allele fractions do not look heterozygous. A pipeline built to subtract germline noise has no mechanism for subtracting hematopoietic somatic noise.

Figure 1Allele fraction alone does not carry provenance. Three variants indistinguishable in plasma resolve immediately against matched white blood cell DNA — one absent from blood and therefore tumor-derived, two present and therefore hematopoietic. Values are schematic, chosen to show the shape of the problem rather than drawn from a specific case. Note the third row: a CH variant can sit at higher allele fraction in blood than in plasma, the signature that identified a CH-derived KRAS mutation in a documented case.⁸

Why the obvious heuristics fail

Several plausible shortcuts get proposed for identifying CH variants without sequencing blood. Each has a documented limitation worth knowing before relying on it.

Filtering by gene list

Restricting suspicion to canonical CH drivers — DNMT3A, TET2, ASXL1, PPM1D — catches the common cases and misses the consequential ones. The variants that change therapy decisions are in BRCA1/2, ATM, and CHEK2, which are not on the canonical list but carry substantial CH fractions.⁴ A gene-list filter is also unstable in the other direction: most CH cases in population-scale analysis have no known driver mutation at all.⁹

Filtering by allele fraction

CH variants are often low-VAF, but so are subclonal tumor variants and ctDNA in low-shedding disease. The distributions overlap directly. In tissue the median CH VAF was 0.04 against 0.16 in matched blood⁷ — a threshold set to exclude the former excludes real subclonal biology, which is the same trade the series described for FFPE artifact suppression.

Assuming young patients are unaffected

Prevalence is strongly age-dependent — approaching 50% by age 80 in population-scale whole-genome analysis⁹ — but oncology patients are not the general population. Cytotoxic chemotherapy and radiation selectively expand mutant clones, particularly PPM1D and TP53.¹⁰ In treated glioma patients, 17% of a 135-patient cohort had CH-type cfDNA mutations, and prior temozolomide with concurrent radiation was significantly associated with their increase.¹¹ A pretreated patient of any age carries elevated risk, and the therapy that expanded the clone is often the reason the liquid biopsy was ordered.

Assuming a low-level filter will catch it

Attempts to build CHIP filters from conventional blood sequencing run into their own error floor. A one-read CHIP filter based on conventional NGS of peripheral blood cell DNA carries a high false-positive rate, particularly below 0.1% VAF and for base changes with high background artifactual noise; an effective filter at that sensitivity needs error-control strategies of its own.¹² The confounder does not disappear by lowering a threshold — it acquires a new error mode.

What actually resolves it

The problem has a clean solution, which is unusual for the failure modes this series covers. Sequence the blood.

Matched white blood cell sequencing from the same draw allows the variant's compartment of origin to be determined directly, and this is now the consensus recommendation: CAP and AMP recommend that cfDNA assays incorporate whole blood controls to differentiate CH from tumor-derived variants.¹³ The high-intensity study concluded the same, emphasizing the importance of matched cfDNA–white blood cell sequencing for accurate variant interpretation.²

The infrastructure is maturing around this. The BLOODPAC Consortium established a CH/CHIP Working Group in 2024 specifically to address accurate identification and removal of CH from liquid biopsy results, beginning with a standardized lexicon — including the distinction between tumor-derived and non-tumor-derived somatic mutations.¹⁴ That a consortium needed to standardize the vocabulary indicates how recently the field converged on treating this as a defined analytical category rather than an incidental nuisance.

Practical guidance

Treat matched blood as an assay requirement, not an enhancement. For any plasma assay broader than a hotspot panel, and for tumor-only tissue profiling in DNA damage response genes, the buffy coat is the control that answers the question the report is asking. Where a matched normal is genuinely unavailable, that limitation belongs in the report rather than in the methods.

Record the compartment, not just the variant. A variant record should carry which compartment it was observed in and at what allele fraction in each. The plasma-versus-blood comparison is the evidence; storing only the plasma value discards it, and no downstream reanalysis can reconstruct it.

Scale suspicion to panel breadth and gene identity. A BRCA2 truncating variant at low VAF in plasma from a previously treated 74-year-old warrants a different prior than an EGFR L858R in the same sample. The gene-level CH fractions are published and should inform how much confirmatory evidence a call needs before it drives therapy.⁴

Capture treatment history as an analysis variable. Prior cytotoxic chemotherapy and radiation shift both CH prevalence and its gene spectrum toward PPM1D and TP53.¹⁰ That is metadata the pipeline can act on, and it is usually available.

Report CH findings rather than silently discarding them. Once identified, a CH variant is not merely noise removed. CH carries an established risk of progression to hematologic neoplasia, including therapy-related myeloid neoplasms.⁶ In one glioma cohort, CH-type cfDNA mutations were an independent prognostic factor for shorter survival.¹¹ A filter that deletes them converts a clinically meaningful incidental finding into nothing.

The counterpoint

Two things keep this in proportion.

First, the problem is largely solved where matched sequencing is performed. This is not an open methodological question in the way that, say, deconvolution reference completeness is. The mechanism is understood, the assay design that fixes it is established, and professional bodies recommend it. The gap is implementation, not knowledge.

Second, hotspot-driven testing is substantially protected. Assays focused on genes not recurrently mutated in CH carry limited exposure,⁶ and the gene-level data show single-digit CH fractions for EGFR, KRAS, and BRAF.⁴ The risk concentrates in comprehensive panels, DNA damage response genes, older patients, and pretreated patients — which is a specific enough profile to act on rather than a diffuse warning about all liquid biopsy.

It is also worth noting that CH is increasingly understood as informative in its own right rather than purely as interference — the same literature that treats it as a confounder now describes it as a risk factor generating distinct cfDNA signals that can be used diagnostically.¹⁴

The shape of the problem

This series has repeatedly described pipelines that answer a well-posed question about the wrong thing. A swapped sample produces a clean BAM and a correct answer for the wrong person. A benchmark reports high accuracy for a region that excludes where calling is hard. A deconvolution reports proportions over a reference that lacks the cell type actually present.

Clonal hematopoiesis is the somatic-calling version. Every filter fires correctly. The variant is real, the read support is sound, the orthogonal confirmation succeeds. What the pipeline never established — because it was never asked to, and because a VCF has no field for it — is which population of cells the mutation came from. That question is answerable, cheaply, by sequencing a second compartment from the same tube.

The number on the report says a mutation is present. It does not say whose cells it is in.

Related in this series: the tumor-only piece covers the other major source of non-tumor variants in an unmatched call set, and CH is additive to it; the ctDNA detection-limits piece covers what a negative plasma result can and cannot establish; the sample identity piece covers the more general case of a technically perfect pipeline answering for the wrong source; and the FFPE piece covers a third population of variants that are chemically real but not tumor biology.

References

1. Characterization of plasma cell-free DNA variants as of tumor or clonal hematopoiesis origin in 16,812 advanced cancer patients. Clinical Cancer Research. 2025;31(13):2710. https://aacrjournals.org/clincancerres/article/31/13/2710/763079/ (cited for the "clinical false positive" framing)
2. Razavi P, Li BT, Brown DN, et al. High-intensity sequencing reveals the sources of plasma circulating cell-free DNA variants. Nature Medicine. 2019;25:1928–1937. https://www.nature.com/articles/s41591-019-0652-7
3. Sun K. Clonal hematopoiesis: background player in plasma cell-free DNA variants. Annals of Translational Medicine. 2019;7(Suppl 8):S384. https://atm.amegroups.org/article/view/33616/html (commentary on ref. 2; cited for the 24.4% tumor-concordance figure)
4. Characterization of plasma cell-free DNA variants as of tumor or clonal hematopoiesis origin in 16,812 advanced cancer patients — gene-level CH fractions and 42.3% patient-level prevalence. Clinical Cancer Research. 2025;31(13):2710. https://aacrjournals.org/clincancerres/article/31/13/2710/763079/
5. Jensen K, Konnick EQ, Schweizer MT, et al. Association of clonal hematopoiesis in DNA repair genes with prostate cancer plasma cell-free DNA testing interference. JAMA Oncology. 2021;7(1):107–110. Summary coverage: https://newsroom.uw.edu/news-releases/blood-cell-mutations-confound-prostate-cancer-liquid-biopsy
6. Clinical significance of clonal hematopoiesis of indeterminate potential in hematology and cardiovascular disease. Diagnostics. 2022;12(7):1613. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9317488/
7. Ptashkin RN, Mandelker DL, Coombs CC, et al. Prevalence of clonal hematopoiesis mutations in tumor-only clinical genomic profiling of solid tumors. JAMA Oncology. 2018;4(11):1589–1593. https://pubmed.ncbi.nlm.nih.gov/29872864/
8. Ptashkin RN, et al. (ref. 7), full text — OncoKB annotation of CH variants and the KRAS G12R case. https://jamanetwork.com/journals/jamaoncology/fullarticle/2683808
9. Genetics and epidemiology of mutational barcode-defined clonal hematopoiesis. Nature Genetics. 2023. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10703693/
10. Clinical impact of clonal hematopoiesis on patients with solid tumors: a systematic review and meta-analysis. Frontiers in Oncology. 2026. https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2026.1770012/full
11. High prevalence of clonal hematopoiesis-type genomic abnormalities in cell-free DNA in invasive gliomas after treatment. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8048515/
12. Clonal haematopoiesis: a source of biological noise in cell-free DNA analyses. https://pmc.ncbi.nlm.nih.gov/articles/PMC6442654/
13. Caris Life Sciences. Incidental clonal hematopoiesis — assay documentation citing CAP/AMP recommendations for whole blood controls in cfDNA assays. https://www.carislifesciences.com/physicians/physician-tests/caris-assure/incidental-clonal-hematopoiesis/ (vendor documentation; the underlying CAP/AMP guidance should be cited directly where available)
14. Tell R, et al. Lexicon for clonal hematopoiesis in liquid biopsy. Clinical and Translational Science. 2026. https://ascpt.onlinelibrary.wiley.com/doi/10.1111/cts.70463