Does clonal haematopoiesis increase risk of serious bacterial infection?

A 22% higher hazard across 590,537 people

The paper asks whether somatic mutations that let a blood stem cell produce an expanded clone also raise the risk of serious bacterial infection, meaning infections such as sepsis and bacterial pneumonia. Across three discovery cohorts, the authors identified 19,822 mutations in 43 CH genes in 40,560 of 590,537 individuals, or 6.87%. Carriers had a higher hazard of developing SBI (HR = 1.22, P = 4.66 × 10⁻²⁹), and the association held after adjustment for inflammatory comorbidities and ancestry and was independent of malignancy.

Two measures in that sentence answer different questions and should not be merged. The hazard ratio compares how quickly SBI develops over follow-up between carriers and non-carriers, while the 28-day mortality result concerns what happens after an infection is diagnosed. The authors report a relative risk of 1.84 (P = 0.0012) for 28-day post-infection mortality, so CH is linked both to becoming infected and to dying within 28 days once infected.

Filtering variants to separate true clones from germline noise

The credibility of the 6.87% prevalence depends on how mutations were called. The authors used a multi-parameter somatic variant calling and filtering approach: they called likely somatic variants with Mutect2, then enriched for real somatic mutations by requiring a low median variant allele fraction (VAF) across carriers of novel mutations, performing read-backed phasing of known germline variants against putative somatic variants, accounting for local germline copy number, and keeping mutations with predicted functional or pathogenic effects.

The paper offers a check on this filtering. Individuals whose Mutect2 variants were removed by the filters had an age range similar to non-carriers (Cohen's d = 0.19) and were younger than CH carriers (Cohen's d = 0.55), which the authors read as consistent with a germline origin for the filtered-out variants. Because CH prevalence rises with age, a filter that leaves behind age-matched non-carriers is behaving as expected rather than inflating the carrier group.

Larger clones carry larger infection hazards

The dose-dependent part of the title refers to clone size, measured as VAF and grouped as small (0 < VAF ≤ 0.1), medium (0.1 < VAF ≤ 0.4) and large (VAF > 0.4). In the UK Biobank cohort of 448,083 people, SBI risk was higher at all clone sizes in carriers of single-nucleotide variant or small indel CH than in carriers of copy-number-variant CH, and the hazard rose with clone size.

Clone size also separated the recurrent hotspot mutations. In a mega-analysis across the discovery cohorts, 25% of individuals with large SRSF2 P95 clones developed SBI within 10 years of variant detection, compared with 18% for large SF3B1 K700 clones, 23% for large IDH2 R140 clones and 9% for large DNMT3A R882 clones. These are absolute risks over a 10-year window, not hazard ratios, and they apply to carriers with no diagnosed haematological malignancy.

Rare hotspots carry higher risk, common ones more cases

The mutation with the largest absolute risk is not the one that matters most at population scale. DNMT3A R882 occurred at roughly 3.5 times the frequency of SRSF2 P95 mutations, so although only 9% of large R882 clone carriers developed SBI within 10 years, the authors argue that R882 may be linked to a greater number of absolute SBI cases in a general population than the splicing-factor hotspots SRSF2 P95 and SF3B1 K700.

This is the paper's central comparison for readers weighing biomarkers: per-carrier risk and total case burden point in different directions. The splicing-factor mutations SRSF2 and SF3B1, which the authors note affect differentiation of blood stem and progenitor cells, coincided with higher rates of both haematological malignancy and SBI, while the more common DNMT3A, TET2 and ASXL1 mutations were less extreme per carrier.

The association survives excluding cytopenias

A plausible alternative explanation is that CH carriers develop SBI because their clones already depress blood counts. The authors tested this by repeating the UK Biobank analysis after excluding individuals with leukopenia, neutropenia, lymphopenia or thrombocytopenia by clinical standards. Hotspot mutations remained significantly associated with SBI in people with no detectable cytopenia: HR = 3.52 for SRSF2 P95, 2.36 for SF3B1 K700, 1.53 for DNMT3A R882, 3.60 for NRAS G12 and 5.00 for IDH2 R140.

A second sensitivity analysis excluded 26,821 individuals with anaemia-related disorders. Associations remained significant for SRSF2 P95, DNMT3A R882, NRAS G12 and IDH2 R140; SF3B1 K700 stayed directionally associated (HR = 1.62) but lost statistical significance (P = 0.24), which the authors attribute to the small number of remaining carriers (n = 58). That loss of significance is a limit on the SF3B1 result, not on the broader pattern.

Bacterial risk rises while viral risk does not

The infection signal was specific to bacteria in this analysis. CH was significantly associated with SBI but not with viral infections (HR = 1.05, P = 0.14), including severe COVID-19, which the authors note contrasts with earlier reports linking CH to adverse COVID-19 outcomes. They offer three possible explanations: different immune mechanisms for bacterial versus viral defence, variation in how viral endpoints were defined, or reduced statistical power for individual viral pathogens.

Within bacterial infections, the association was largely independent of strain. In the Alliance for Genomic Discovery cohort of 92,211 people, CH mutations were associated with SBI caused by Staphylococcus, Streptococcus, Klebsiella and E. coli, but not with Pseudomonas, Enterococcus or Mycobacteria, and the combined all-strain analysis was significant (P = 3.92 × 10⁻¹⁵). The authors describe the associations as largely independent of bacterial strain, so the strain-specific nulls should be read as underpowered rather than protective.

One blood sample cannot capture clone growth

The study reflects a single sampling event of clones with varying growth rates, which the authors flag as a limitation. They note that clonal expansions in SRSF2 P95H carriers become dominant in the blood within 11 years, so longitudinal sampling may be needed to capture the more immediate risk of CH across subtypes. A VAF measured once is therefore a snapshot of clonal burden, not a trajectory.

The authors also frame the mechanism cautiously. Proteomic profiling showed perturbation of the FLT3/FLT3LG axis in CH carriers, which they interpret as altered haematopoietic signalling rather than a simple systemic inflammatory cytokine signature, and they state that further mechanistic studies are required to define causal pathways linking CH subtypes to SBI risk. The associations reported here are observational, from a retrospective cohort design.

Prioritise SRSF2 P95 for risk, DNMT3A R882 for burden

For readers deciding which CH finding to act on, the paper's own comparison sets the priority. SRSF2 P95 and SF3B1 K700 had the most robust associations with SBI, with 25% and 18% of large clone carriers respectively developing SBI within 10 years, and these results persisted in people never diagnosed with haematological malignancy, MDS, a cytopenia disorder or aplastic anaemia. If the question is which carrier faces the steepest individual risk, the splicing-factor hotspots lead.

If the question is where most SBI cases will arise, the answer shifts to DNMT3A R882 because it is far more common, even though only 9% of its large clone carriers developed SBI within 10 years. The authors present these subtypes as readily measurable candidates for future risk stratification, and the practical split is between per-carrier risk and population-level case burden rather than a single ranking of mutations.

Sources used in this answer