Clonal Hematopoiesis

From an incidental sequencing finding to a disease of aging blood

Ashwin Kishtagari, MBBS, MSCI

Division of Hematology/Oncology, Vanderbilt University Medical Center

2026-08-12

Outline

1. Defining the object — what clonal hematopoiesis is, and how often it is there
2. Who progresses — quantitative risk models and what they miss
3. Beyond the marrow — cardiovascular and inflammatory disease
4. Managing the person — workup, surveillance, clinics, and trials

Frame the talk around one tension: clonal hematopoiesis is nearly universal in aging blood, and myeloid neoplasia is rare. Everything interesting in this field lives in the gap between those two facts.

Defining the object

Clonal hematopoiesis, and the states it names

Clonal expansion of hematopoietic stem and progenitor cells carrying a somatic mutation that confers a fitness advantage, in the absence of a diagnosed hematologic neoplasm.

	No clonal marker	Clone at VAF ≥2%
Normal counts	normal	CHIP
Persistent unexplained cytopenia	ICUS	CCUS
Dysplasia meeting criteria	—	MDS

Both WHO-HAEM5 and the ICC formally recognized CHIP and CCUS as precursor states in 2022. Cytopenia must persist ≥4 months; X-linked genes in males require VAF ≥4%.

The WHO and ICC CHIP/CCUS definitions are essentially harmonized. Where they diverge is the CCUS/MDS boundary: ICC allows MDS to be diagnosed without dysplasia when an MDS-defining genetic lesion is present, so a cytopenic patient with monosomy 7 and no dysplasia is MDS by ICC and CCUS by WHO5.

The 2% threshold came from sequencers, not from biology

• Set by the analytic sensitivity of clinically available assays in 2015
• A heterozygous variant at VAF 2% means roughly 4% of nucleated blood cells carry it
• The line sits nowhere near the inflection point of risk

Clone size	Risk of myeloid neoplasm
any clone detected (floor VAF 0.035)	HR 11.1 (95% CI 3.9–32.6)
VAF ≥0.10	HR 49 (95% CI 21–120)
small vs very large clones	0.03% vs 0.5% per year

Prevalence rises with age, and with sequencing depth

By age, exome-scale assays

Age	Prevalence
<50 y	~1%
60–69 y	5.6%
70–79 y	9.5%
80–89 y	11.7%
≥90 y	18.4%

By detection floor

Assay	Finding
WES, VAF ≥2%	3.4–5.5% of UK Biobank
smMIPs, VAF ≥0.1%	>20% at 60–69 y
UKB exomes, any VAF	23% carry a clone
error-corrected, n=20	19 of 20

Reported prevalence is a property of the assay as much as of the population.

The 95% figure from Young 2016 rests on twenty people. Say the sample size out loud. It is directionally right and replicated in spirit by Acuna-Hidalgo, but it is not a population estimate.

Clonal diversity collapses in the eighth decade

• 3,579 single-cell-derived colonies from 10 people aged 0 to 81, whole-genome phylogenies
• Younger adults sustain hematopoiesis from an estimated 20,000 to 200,000 stem and progenitor cells
• Under 65 the system is massively polyclonal
• Over 75, 12 to 18 clones account for 30 to 60% of blood production, most founded decades earlier

This reframes the epidemiology. The age-70 inflection is not a step change in how many people carry a mutation. It is a collapse of clonal diversity, and a large share of that expanded clonality carries no known driver at all. Our gene panels see a fraction of what is happening.

Which genes, before and after cytotoxic exposure

De novo clonal hematopoiesis

Epigenetic regulators account for roughly 70%.

• DNMT3A, TET2, ASXL1 together carry two-thirds to 90% of variants
• Splicing factors (SF3B1, SRSF2, U2AF1) are rare before age 50
• TET2 overtakes DNMT3A as the leading driver in the eighth decade

After chemotherapy or radiation

The DNA damage response genes take over.

Gene	OR with prior therapy
TP53	2.7
PPM1D	3.6
CHEK2	4.6

Radiation OR 1.4, cytotoxic chemotherapy OR 1.2. Targeted agents and immunotherapy show no association.

Germline background selects which somatic clone wins

• The TCL1A promoter variant rs2887399 slows expansion of TET2, ASXL1, SF3B1, and SRSF2 clones, and does nothing to DNMT3A clones
• Across 731,835 people, screening 236 cancer-predisposition genes identified 22 new clonal hematopoiesis predisposition genes. CHEK2 carriers develop DNMT3A clones preferentially; ATM carriers accumulate mosaic chromosomal alterations
• Germline-selected clones progress at 2.7 times the rate of unselected clones for myeloid malignancy, and 13.1 times for lymphoid malignancy
• In the highest-risk stratum, 25-year absolute myeloid malignancy risk reaches 46% for CHEK2 and 30% for ATM, against 4% in non-carriers

One inherited SNP determines which of two competing somatic clones expands. This is the cleanest illustration in the field that clonal hematopoiesis is a germline-somatic interaction, not a purely stochastic somatic process.

Who actually progresses

The arithmetic that breaks “0.5 to 1% per year”

• The figure comes from roughly 31 incident hematologic cancers in cohorts assembled for diabetes and psychiatric genetics
• In unselected biobank populations the myeloid neoplasm rate with gene-mutation CHIP is 0.17% per year
• The 88% of carriers in the lowest risk tier run at 0.07% per year
• Recursive partitioning spreads 10-year risk across 0.78% to 85%

CHIP prevalence in the elderly runs around 10,000 per 100,000. Annual myeloid neoplasm incidence runs around 12 to 75 per 100,000. An average across a hundredfold-skewed distribution is wrong for almost every individual patient.

The clonal hematopoiesis risk score

Variable	0.5	1.0	1.5	2.0	2.5
Single DNMT3A	yes	no
High-risk gene		no			yes
Mutations		1		≥2
VAF		<0.20		≥0.20
RDW		<15%			≥15%
MCV		<100 fL			≥100 fL
Cytopenia		CHIP	CCUS
Age		<65 y	≥65 y

Score runs 7.5 to 15.5. High-risk genes are SRSF2, SF3B1, ZRSR2, IDH1, IDH2, FLT3, RUNX1, JAK2, TP53.

Derived in 193,743 UK Biobank exomes, validated in a further 245,147, then in clinical cohorts at DFCI/BWH and Pavia.

Worth pointing out at the podium: after high-risk genotype, the two strongest predictors are MCV and RDW. Two numbers off a CBC outrank clone size and mutation number.

Two omissions matter. U2AF1 is missing because of a GRCh38 mismapping artifact, not because it is low risk. PPM1D is missing because there is no signal in an untreated population, so the score is not calibrated for patients on chemotherapy. The authors say so explicitly.

Three tiers, three different futures

Tier	Score	Share	10-year risk of MN
Low	≤9.5	88.4%	0.67% ± 0.08
Intermediate	10–12	10.5%	7.83% ± 0.81
High	≥12.5	1.1%	52.2% ± 4.96

c-index 0.807 in derivation, 0.799 in validation, 0.788 and 0.727 in the two clinical cohorts.

CHIVE, our own prospective registry

181 enrolled, 99 clonal, serial sampling over 2.5 years.

CHRS tier	Progressed
Low (n=50)	2%
Intermediate (n=32)	3.1%
High (n=14)	28%

Clonal cytopenia is a different animal

The clonal cytopenia risk score was built in 357 CCUS patients across 17 centers and validated in Pavia.

Component	Points
Splicing factor mutation	2
Platelets <100 ×10⁹/L	2.5
≥2 mutations	3

Tier	2-year risk of MN
Low (<2.5)	6.4%
Intermediate (2.5–5)	14.1%
High (≥5)	37.2%

Note the timescale. High-risk CCUS reaches 37% at two years where high-risk CHIP reaches 52% at ten. In the prospective Leeds cohort of 400 CCUS patients, mutation number was the dominant predictor.

What the risk scores do not see

• U2AF1 is absent from the high-risk gene list because the locus is duplicated in GRCh38 and reads mismap. Malcovati gives it a positive predictive value of 0.88 to 0.97
• PPM1D is absent because there is no signal in an untreated population, yet it is the single most enriched gene in therapy-related clonal hematopoiesis
• Zygosity is nowhere in any model. JAK2 V617F with 9p copy-neutral LOH carries HR 54.8 against 44.1 for the heterozygous state
• Growth rate is a second axis entirely. DNMT3A clones grow at roughly 5% per year, SRSF2 P95H at over 50%. Two patients at VAF 8% can be on trajectories an order of magnitude apart

The zygosity paper is our BioVU work. It is the cleanest argument that the current models are incomplete in a way that is fixable today, since zygosity is already inferable from the sequencing data every clinical lab generates.

Beyond the marrow

The finding that redirected the field

In 17,182 people sequenced for reasons having nothing to do with blood, carriers died more often, and leukemia did not explain it.

Outcome	Effect
Hematologic cancer	HR 11.1 (3.9–32.6)
All-cause mortality	HR 1.4 (1.1–1.8)
Incident coronary heart disease	HR 2.0 (1.2–3.4)
Ischemic stroke	HR 2.6 (1.4–4.8)

Confirmed three years later in 4,726 coronary cases and 3,529 controls: incident CHD HR 1.9 (1.4–2.7), early-onset MI OR 4.0 (2.4–6.7), with increased coronary calcification in carriers.

From TET2 loss to interleukin-6

TET2 loss

↓ promoter methylation silences Dusp10

↓ JNK1 phosphorylation rises

↓ BRCC3 deubiquitylates NLRP3

↓ inflammasome activation, IL-1β release

↓ IL-6 → STAT3 → CSF1R, macrophage survival

↓ plaque

Partial marrow reconstitution with Tet2-deficient cells is sufficient to enlarge plaque in Ldlr^−/− mice, and an NLRP3 inhibitor is preferentially atheroprotective in those chimeras.

The same NLRP3 dependency has since been shown for heart failure, atrial fibrillation, renal hypertension, and steatohepatitis.

Two honest nuances. Dnmt3a loss produces a convergent inflammatory macrophage phenotype despite opposite effects on methylation, so this is not a TET2-only story. And JAK2 V617F appears to work through AIM2 rather than NLRP3.

Two lines of human evidence for the IL-6 axis

CANTOS, canakinumab in TET2 carriers

3,923 samples, 338 with CHIP, 103 with TET2.

MACE HR 0.38 (0.15–0.96) in TET2 carriers, HR 0.93 (0.78–1.10) in non-carriers.

P for interaction 0.14. Post hoc, 24% power, no prospective stratification. Hypothesis-generating.

IL6R p.Asp358Ala, a natural experiment

35,416 UK Biobank participants free of cardiovascular disease.

Large clones carry HR 1.59 (1.21–2.09). In those carriers the IL6R variant gives HR 0.46 (0.29–0.73), and in non-carriers HR 0.95 (0.89–1.01).

P for interaction 0.003, replicated at P = 0.036.

The wider phenome

Systematic review and random-effects meta-analysis of 88 studies.

Outcome	Pooled estimate
Hematologic malignancy	4.28 (2.29–7.98)
Coronary heart disease	1.76 (1.27–2.44)
Cancer mortality	1.46 (1.13–1.88)
All-cause mortality	1.34 (1.19–1.50)
Heart failure	1.27 (1.15–1.41)
Renal impairment	1.25 (1.18–1.33)
Stroke	1.16 (1.05–1.28)
Cardiovascular mortality	1.09 (0.97–1.22)

Gene specificity matters more than the composite.

• JAK2 drives the venous thrombosis signal, HR 4.2 (2.18–8.08)
• TET2 drives gout, HR 1.53 (1.05–2.22), which is mechanistically apt
• Chronic liver disease OR 2.01, supported by Mendelian randomization at OR 2.37
• Isolated DNMT3A showed no excess myeloid malignancy, mortality, or renal risk

Two results that should keep us honest

63,700 patients, five TIMI trials

Any cardiovascular event: aHR 1.07 (0.99–1.16), P = 0.08.

First MI aHR 1.31 (1.05–1.64). Recurrent MI aHR 0.94 (0.79–1.13). P for interaction 0.008.

No treatment-effect heterogeneity by CHIP status for any of five therapies.

ZEUS, ziltivekimab, July 2026

Over 6,300 patients with atherosclerosis, chronic kidney disease, and hs-CRP ≥2 mg/L.

MACE HR 0.99 (0.88–1.11). Free IL-6 and hs-CRP fell as expected. The trial was not CHIP-selected.

Clonal hematopoiesis may matter more for atherogenesis than for event recurrence under maximal secondary prevention. A prospective genotype-stratified trial has still never been done.

Managing the person, not the variant

Most clonal hematopoiesis is found by accident

• Plasma cell-free DNA. With matched white-cell sequencing, 53.2% of cfDNA variants in metastatic cancer patients and 81.6% in cancer-free controls were clonal hematopoiesis rather than tumor
• Tumor-only panels. Among 17,469 solid tumor patients, 26.5% carried clonal hematopoiesis in blood and 5.2% had a CH variant present in the tumor specimen. Half looked oncogenic by OncoKB, and 98.7% were absent from germline polymorphism databases
• A specific trap. Five of six JAK2 V617F calls and two of 58 KRAS G12X calls on plasma panels in lung cancer were clonal hematopoiesis. The KRAS call would falsely imply EGFR-TKI resistance

Working up an incidental finding

1. Exclude the ordinary causes of cytopenia first. Iron, B12, folate, renal function, thyroid, autoimmune disease, medications, alcohol, viral infection. Detection of a clone is the start of the algorithm, not the end
2. Confirm persistence for at least four months before calling it CCUS
3. Confirm somatic origin against non-hematopoietic tissue. Fingernail clippings, hair, or cultured fibroblasts. Avoid buccal swabs, since leukocyte contamination produces false germline calls when clonal burden is high
4. Consider germline predisposition when VAF sits near 50%, the patient is young, or there is a family history. DDX41 carries roughly 50% lifetime risk by age 90 and is present in about 1 in 400 people
5. Marrow biopsy for CCUS, cytosis, or a high-risk genotype. Not for low-risk CHIP with normal counts

Surveillance intervals are consensus, not evidence

Risk tier	Complete blood count	Bone marrow
Low-risk CHIP, CHRS ≤9.5	annual	not indicated
Intermediate, CHRS 10–12	every 6 to 12 months	case by case
High-risk, CCUS, therapy-related	every 3 to 6 months	consider periodic evaluation

No professional society has issued a management guideline for clonal hematopoiesis. The AHA has published a scientific statement, and ASH an education program chapter. Every interval above is extrapolated from progression modeling. No surveillance schedule has ever been shown to improve survival.

What a clonal hematopoiesis clinic actually sees

CHIVE, Vanderbilt

181 enrolled, 99 clonal. Referrals came from cardiology (69) more than hematology (63), with genetics at 45.

Comorbidity	CH+	CH−
Coronary disease	55.7%	32.9%
Hypertension	79.4%	52.4%
Heart failure	22.7%	9.8%
Chronic kidney disease	41.2%	18.3%

HALT, Ohio State

143 referrals over three years, 80 for suspected clonal hematopoiesis.

After evaluation, 76 confirmed CHIP or CCUS, 3 reclassified as MDS, 1 benign variant.

43.4% had prior chemotherapy. CHRS distribution ran 37% low, 43% intermediate, 20% high, far enriched over the population.

Cardiology as the leading referral source is the strongest published argument for the multidisciplinary structure. Also worth naming the economics: an academic CH clinic runs over $500,000 a year with no reimbursement model.

The harms of knowing

• In 528 young breast cancer survivors, 28.2% reported at least moderate anxiety simply from learning that CHIP exists. Only 5.9% had heard of it before
• Testing preference was highly malleable. Showing baseline population risks moved 11.1% away from testing; describing a hypothetical treatment moved 14.1% toward it
• GINA covers health insurance and employment. It does not cover life, disability, or long-term care insurance, and whether a somatic variant is even protected genetic information is unsettled
• Current data do not support routine CH testing in breast cancer, and CH should not change breast cancer therapy in most patients

Do not order the test without a plan for what a positive result will change.

What is actually being tested

Trial	Agent	Population	Status
IMPACT (NCT05641831)	canakinumab vs placebo	high-risk CCUS	recruiting, open at VUMC
PIMMs (NCT05030441)	ivosidenib	IDH1-mutant CCUS	reported ASH 2025
NCT05102370	enasidenib	IDH2-mutant CCUS	completed
COLCOT / LoDoCo2	low-dose colchicine	coronary disease	substudies reported

Ivosidenib produced 82% hematologic response and IDH1 clearance in 4 of 10 patients treated a year or more. Colchicine at 0.5 mg daily lowered overall VAF against placebo, with TET2 down 10.3% and SF3B1 down 19.9%. High-dose intravenous ascorbate in TET2-mutant CCUS produced no responses.

Where this leaves us

• Clonal hematopoiesis is close to universal in aging blood, and the clinically dangerous genotypes sit in about 0.18% of the population
• We can now stratify risk well enough to run prevention trials, and two randomized datasets show a clone can be pharmacologically modified
• The inflammatory hypothesis got harder in 2026. ZEUS was null, and no CHIP-selected outcome trial has ever been run
• Zygosity, clonal growth rate, mosaic chromosomal alterations, and germline background all carry information that no current model uses

The open question is whether detecting clonal hematopoiesis changes any outcome. Nothing in the literature answers that yet.

Close on the trial question. IMPACT and a colchicine prevention trial are the studies that would change the answer to the screening question, and both are enrolling or fundable now.

Selected reading

• Weeks LD, et al. Prediction of risk for myeloid malignancy in clonal hematopoiesis. NEJM Evid 2023;2(5):EVIDoa2200310
• Beeler JS, Walter MJ, Bolton KL. Clonal hematopoiesis and its progression to myeloid neoplasms. Haematologica 2026;111(5):1536–1551
• Vanner RJ, et al. Clinical decisions in clonal hematopoiesis. Haematologica 2026;111(5):1584–1609
• Rhee JW, et al. Clonal hematopoiesis and its cardiovascular implications. AHA Scientific Statement. Circulation 2026;153(11):e940–e952
• Xie Z, et al. Demystifying the diagnosis and management of ICUS, CHIP, and CCUS. Hematology Am Soc Hematol Educ Program 2025;2025(1):682–690
• Shannon ML, Heimlich JB, … Bick AG, Savona MR. CHIVE, a prospective longitudinal clonal hematopoiesis cohort and biorepository. Blood Adv 2024;8(13):3453–3463