Clonal hematopoiesis landscape in frequent blood donors Free

Access to blood products is the backbone of modern medicine. The burden of blood donation is shouldered by a small group of altruistic healthy volunteers, with only ∼4% of eligible individuals donating.¹ To protect the iron stores of donors from critical depletion, whole blood donation is limited to 4 (females) or 6 units (males) per year in Europe and North America,^2,3 each unit representing ∼10% of a donor’s total blood volume.⁴ Different environmental stressors, such as infection, cytokines, chemotherapy, or blood loss, can trigger active proliferation of hematopoietic stem cells (HSCs),^5-8 which in turn affects the acquisition and propagation of genetic lesions.⁹ The expansion of HSC clones (and their progeny) carrying lesions is referred to as clonal hematopoiesis (CH), and its prevalence increases with age.^10-12 Considering that all red blood cells (RBCs) are replaced 3 times per year,^13-15 and each donated unit of blood adds 2.5% to 4% excess erythropoiesis for that year, we studied this specific and unconfounded type of systemic stress that manifests molecularly in form of recurrently increased levels of erythropoietin (EPO),¹⁶ on the clonal composition of the hematopoietic system. Because specific environmental cues can favor the outgrowth of HSCs carrying certain mutations,^17-20 we hypothesized that repeated large-volume phlebotomy may contribute to the clonal landscape by promoting clones with functionally distinct types of mutations.

Detailed information is provided in supplemental Materials, available on the Blood website.

We performed next-generation sequencing of DNA from peripheral blood leukocytes of male patients aged >60 years with an extensive blood donation history (frequent donor [FD] cohort). The control cohort consisted of age-matched donors with few lifetime donations (control donor [CD] cohort). A total of 217 FD and 212 CD samples were acquired and analyzed as cohort 1 and cohort 2. Thus, buffy coats, which are waste products of component preparation from whole blood donations, of selected healthy male volunteer blood donors who donated between December 2019 and June 2020 (FD and CD cohort 1), December 2020 and November 2021 (consecutive samples), and May and August 2023 (FD and CD cohort 2) at the German Red Cross Blood Service Baden Württemberg-Hessen were used for the study. All donors provided written informed consent, allowing for anonymous processing of the samples as approved by the ethics committee (vote number 329/10).

Patients were included at the Medical University Innsbruck, with approval from the ethics committee of the Medical University Innsbruck (EK-Nr, 1182/2020) in the setting of an acute ischemic stroke. For a detailed explanation, see “Single-cell DNA and immunophenotype analysis of primary samples.”

Umbilical cord blood was obtained from full-term donors with parental informed consent at the Royal London Hospital (London, United Kingdom) under approval by the East London Ethical Research Committee. Mononuclear cells were isolated by density centrifugation using Ficoll-Paque (GE HealthCare). Anonymized human bone marrow (BM) samples from consenting healthy volunteers were provided with approval from the ethics committee of Goethe University Frankfurt (number 329/10). Lin^–CD34⁺CD38^– cells were cultured in StemSpan SFEM (Stemcell Technologies) with 100 ng/mL rhFLT-3L, 100 ng/mL rhSCF, and 100 ng/mL rhTPO for 48 hours. CRISPR editing was then performed with the NEON Transfection system to introduce the ribonucleoprotein complex using the indicated small guide RNAs and donor templates (supplemental Table 9). For long-term culture assays, 48 hours after CRISPR-induced modification, 1000 hematopoietic stem and progenitor cells (HSPCs) were transferred to MS-5 plates. Once a week, 500 μL of medium was replaced with the following stimuli: human EPO (3 U/mL), human interferon gamma (IFN-γ; 100 ng/mL), or lipopolysaccharide (LPS; 1 μg/mL). After 4 weeks, cells were collected and analyzed by flow cytometry. For generation of humanized mice reconstituted with edited hHSCs harboring DNMT3A W305∗ or R882H mutations, NBSGW mice aged between 8 to 12 weeks received IV injections of hHSC (10 000-20 000 Lin^–CD34⁺CD38^– cells per mouse). Engraftment of the reconstituted human hematopoietic system was validated for each mouse by BM aspiration at 6 weeks before starting the regimen of erythropoietic stress.

Distinct from most published CH studies,^10-12,21-25 our cohorts consisted exclusively of closely monitored and exceptionally healthy, age- and sex-matched individuals. We first acquired and analyzed a cohort of 105 FD and 103 CD individuals (cohort 1). Subsequently, an independent second cohort of 112 FD and 109 CD individuals (cohort 2) was collected using the same criteria as those set for the first cohort (Figure 1A; supplemental Table 1A-B). No significant difference in overall CH prevalence between the FD and CD cohorts was found. This was true in both cohorts and regardless of the variant allele frequency (VAF) cutoff used, whether the sensitive cutoff of 0.5% or the conventional VAF cutoff of 2%^10-12 (Figure 1B). The VAFs of the detected variants did not significantly differ between the FD and CD cohorts (supplemental Figure 1A). For all subsequent analyses, the 0.5% VAF cutoff was used, and both cohorts were combined.

Most individuals with CH had a single identified mutation: 46 of 70 (70.0%) and 64 of 94 individuals (68.1%) in the CD and FD cohorts, respectively (supplemental Figure 1B). Consistent with previous studies,^10-12,28,29 mutations in DNMT3A and TET2 were the most prevalent in both the FD and CD cohorts (supplemental Tables 2A-B, 3A-B, and 4A-B; supplemental Figure 1C). VAF distributions of mutations in these 2 genes did not significantly differ between the cohorts (supplemental Figure 1D; supplemental Tables 3A-B and 4A-B).

Within the DNMT3A gene, mutations were distributed throughout the length of the gene (Figure 1C), as reported previously for CH-associated DNMT3A mutations.^28,30 The frequency of acute myeloid leukemia (AML) hot spot mutations at position 882 in DNMT3A was low and similar in both cohorts (supplemental Table 3A-B). Interestingly, within the FD cohort, we observed a trend toward a higher proportion of frameshift variants, variants resulting in a premature stop, or structural variants in the DNMT3A gene (Figure 1C).

We next used the recently introduced stability score²⁷ to characterize the DNMT3A variants. A total of 20 and 13 DNMT3A variants were matched from the FD and CD cohorts, respectively. FD DNMT3A variants had significantly lower stability scores than CD cohort variants (Figure 1D; supplemental Table 3A-B). Decreased stability has been directly linked to the regulated degradation of the DNMT3A protein, resulting in a quantitatively reduced enzymatic activity as opposed to a functionally aberrant activity.²⁷ This was further supported by in silico structural predictions conducted for a selected set of destabilizing nonsense DNMT3A variants from the FD cohort. Three FD DNMT3A nonsense variants, W305∗, S663fs, and E733∗, were chosen for structural modeling (supplemental Figure 1E) compared with full-length DNMT3A based on their type of mutation and their high VAF (>5%). The early stop variant W305∗ (exon 8) is predicted to be degraded because of nonsense-mediated messenger RNA decay (NMD).³¹ If translated, the truncated protein lacks two-thirds of its length, including the entire methyltransferase domain. The S663 frameshift mutation results in a premature stop in exon 18 (amino acid position 704) and is also likely degraded via NMD.³¹ In case of NMD, S663fs DNMT3A is predicted to have reduced methyltransferase activity and to be incapable of complexing with wild-type (WT) DNMT3A protein and sequestering the latter, as has been reported for the R882 variant.³² Lastly, the E733∗ variant has a premature stop in exon 19 and is also expected to undergo NMD. Upon translation, E733∗ DNMT3A will contain the catalytic loop residues yet lack the target recognition domain and homodimer binding capacity. Compared with previously described missense mutations capable of interacting with WT DNMT3A protein and causing aberrant methylation, including the AML hot spot variant R882H, the 3 FD variants, W305∗, S663fs (704∗), and E733∗, are predicted solely to cause a quantitative reduction in DNA methylation levels.^33,34 Thus, both stability scores and small-scale in silico structural predictions performed for this selected set of DNMT3A variants from FDs point toward a quantitatively diminished yet qualitatively normal enzyme activity.

Using the fitness score (s), introduced based on a retrospective analysis of large-scale CH studies,²⁸ we next matched 25 DNMT3A mutations in the FD and 17 DNMT3A mutations in the CD cohorts (supplemental Table 5A-B). s values of 4% to 10% per year are categorized as moderate, and s values >10% per year indicate that a mutation is likely to take over the BM under normal aging.²⁸ For example, the malignant R882C and R882H variants have reported fitness scores of 19% and 14% per year, respectively.²⁸ Interestingly, the mean fitness score (s value)²⁸ was only 10.5% per year for FD DNMT3A variants compared with 13.5% per year for CD DNMT3A variants (P < .001; Figure 1E), suggesting that the DNMT3A variants observed in the FD cohort are, on average, less likely to expand during normal aging. Furthermore, the site-specific mutation rate showed a tendency to be lower in the FD cohort (supplemental Figure 1F). This suggests that mutations expanding with erythropoietic stress associated with blood donation display only moderate fitness in the general population and in the absence of additional stimuli. In line with this, the VAFs of most DNMT3A mutations remained stable within short periods of time between 2 consecutive donations (supplemental Figure 1G-H).

For selected donors and DNMT3A variants, digital droplet polymerase chain reaction was performed (supplemental Figure 2A) on several mature cell fractions (B cells, T cells, and monocytes) along with the immature CD34⁺ compartment. Consistent with previous studies,^35-37 all 5 mutations were detected in all 4 sorted populations, including T cells, implying their acquisition and selection in multipotent HSCs (supplemental Figure 2B).

After whole blood donation, which removes ∼10% of the total circulating hemoglobin mass,^26,38 one of the initial responses of the body is to increase the production of EPO to stimulate BM erythropoiesis.^39-42 Accordingly, approximately twofold increased concentrations of EPO are detected in the serum of blood donors and remain elevated for up to 120 days after blood donation.^16,43-45 Therefore, we sought to functionally investigate a potential link between the variants found in FDs and EPO. The same 3 DNMT3A mutations we had previously subjected to structural analysis (supplemental Figure 1E) were also reconstructed and functionally analyzed in vitro.

Using CRISPR/Cas9, monoallelic W305∗, S663fs (704∗), and E733∗ mutations were introduced into primary human HSCs (supplemental Figure 3A). DNMT3A-edited HSCs were cultured using long-term culture assays in the presence or absence of EPO or inflammatory stimuli (IFN-γ or LPS), and the VAFs of the introduced DNMT3A mutations were assessed after 4 weeks (Figure 2A). Of note, we observed that FD DNMT3A mutations expanded in EPO-rich culture (Figure 2B), which promoted robust erythroid differentiation characterized by the expansion of CD235a⁺CD71⁺ cells (supplemental Figure 3B-C) as expected. The twofold expansion of FD DNMT3A variants was observed using 2 different doses of EPO (supplemental Figure 3D). By contrast, allele frequency of FD DNMT3A variants was not increased upon LPS exposure (supplemental Figure 3E). Two known preleukemic DNMT3A variants, R882C and R882H, were also engineered and similarly analyzed. In sharp contrast with FD DNMT3A variants, allelic frequency of both R882 clones increased upon IFN-γ exposure but remained stable in EPO-rich culture (Figure 2C). Similar observations were made with human BM-derived CD34⁺ cells (supplemental Figure 3F). This mutation-specific pattern of responsiveness to different environmental cues was also observed when HSPCs harboring FD variants or preleukemic variants were cocultured (supplemental Figure 3G), pointing toward a cell-intrinsic mechanism to explain the disparity between the DNMT3A clones’ growth dynamics.

We next sought to explore the molecular mechanisms underlying the selective outgrowth of FD variants within EPO-rich environments. The W305∗ variant from the FD cohort was chosen along with preleukemic DNMT3A mutations to perform single-cell deposition of human erythroleukemic cell line K562, into which these mutations had been previously introduced by CRISPR-Cas9 editing. After colony screening, we selected the monoclonal colonies harboring specific mutations and performed RNA sequencing (Figure 3A). Pathway enrichment analysis revealed that in W305∗ mutated clones, transcriptional programs associated with heme metabolism were selectively upregulated (supplemental Figure 4A; supplemental Table 10). Interestingly, when analyzing the differentially expressed genes associated with the W305∗ mutation, we observed that DNMT3A itself was downregulated (supplemental Figure 4B; supplemental Table 11). Considering this observation and based on the different location and predicted changes in the reading frame caused by FD mutations compared with preleukemic R882 mutations (Figure 1C), we analyzed the abundance of alternatively spliced DNMT3A transcript isoforms (Figure 3B). Indeed, W305∗ mutant K562 cells expressed the protein-encoding transcript 2 at lower levels, whereas transcripts annotated to undergo NMD were highly expressed compared with the WT or R882 DNMT3A (Figure 3C), which is consistent with the in silico structural predictions described above.

Our current model suggested canonic yet quantitatively attenuated DNMT3A activity as a potential mechanism of EPO responsiveness of the FD mutants. If this was true, then knockdown of WT DNMT3A should similarly convey EPO responsiveness, which was therefore tested in the cord blood–derived hematopoietic progenitor cell line HUDEP-2. We used lentivirus-based CRISPRi targeting of the DNMT3A promoter (supplemental Figure 5A). This line has the ability to undergo erythroid differentiation.^46,47 Transcriptomes of CRISPRi-DNMT3A–targeted HUDEP-2 cells were distinct from controls in which a nontargeting guide RNA was used (Figure 3D; supplemental Figure 5B). More specifically, and consistent with the downstream effects of the W305∗ mutation in K562 cells (supplemental Figure 4A; supplemental Table 10), the heme metabolism gene set was enriched upon downregulation of DNMT3A in HUDEP-2 cells, whereas immune response–associated genes were depleted (Figure 3E; supplemental Figure 5C; supplemental Tables 12 and 13). Moreover, expression of hemoglobin genes, along with that of the EPO receptor itself, was upregulated in the DNMT3A-targeted HUDEP-2 cells (Figure 3F). Ingenuity pathway analysis identified EPO signaling as the pathway showing the strongest upregulation upon silencing of DNMT3A (supplemental Figure 5D). This erythroid priming as a result of DNMT3A downregulation was functionally confirmed in a competitive culture setting. Of note, under erythroid differentiation conditions, DNMT3A-downregulated HUDEP-2 cells showed superior growth dynamics (Figure 3G; supplemental Figure 5E), which overall is in line with the notion that FD DNMT3A variants respond with superior fitness in EPO-rich environments.

To study the impact of different DNMT3A mutations in CD34-enriched blood cells of healthy donors, we analyzed these by concurrent assessment of their genotype and the surface marker expression at single-cell resolution using the Tapestri platform^48,49 (supplemental Figure 6A). We took advantage of our access to primary samples from blood donors known to carry different DNMT3A mutations (supplemental Tables 3A-B and 4A-B). Two additional samples from hematologically clinically normal individuals, positive for DNMT3A R882H CH, were included. The full list of samples and corresponding characteristics is shown in supplemental Table 14. All samples were enriched for CD34⁺ cells before surface staining with a 50-antibody panel (supplemental Table 15B), followed by droplet-based analysis of specific DNMT3A mutations in single cells. Projection of the determined cell surface protein expression pattern onto published single-cell proteogenomic reference maps⁵⁰ revealed 15 different cell clusters (Figure 4A). In the samples from the 2 FDs harboring the variants W305∗ and E773∗, the relative contribution of mutant and WT cells to each of the 15 cell clusters was indistinguishable (Figure 4B; supplemental Figure 6B). In sharp contrast, in all 3 R882 (2 × R882H and 1 × R882C) CH samples, an expansion of the monocytic fraction at the expense of mature lymphoid cell types (B and T cells) fraction was apparent within the mutated compartment compared with the WT cells in the same donor (Figure 4C; supplemental Figures 6C-E and 7A-B). The DNMT3A R366H variant, identified in 1 of the FD donors and found to be EPO nonresponsive, exhibited a similar lineage bias as the R882 variants (supplemental Figure 7B). The myeloid bias was furthermore apparent in the granulocyte-monocyte progenitor fraction (supplemental Figure 7B). Here, an up to sixfold higher percentage of cells with a R882H/C or R366H mutation was assigned to the granulocyte-monocyte progenitor cluster compared with the corresponding WT cells.

The numbers of immature hematopoietic stem and progenitor cells in primary samples were generally low, as expected for nonmobilized peripheral blood samples, which was particularly the case for R882 donors. To better characterize the effects of the different DNMT3A mutations within the stem and progenitor fraction, we next developed a xenograft model. A xenograft mouse model engrafted with human HSCs carrying DNMT3A W305∗ or R882H mutations provided additional resolution into the erythroid lineage, as well as all the immature populations within the hematopoietic system. To model the environmental pressure of the FD cohort, mice were subjected to a stringent erythropoiesis-inducing regimen of serial bleeding combined with intravascular hemolysis and human EPO injections (Figure 4D). We then sorted and determined the frequency of each mutation in different stem and progenitor compartments as well as mature lineages (supplemental Figure 7C). Consistent with the single-cell data from primary patient samples shown above, we observed no significant bias in the presence of the W305∗ mutation between myeloid (common myeloid progenitor, granulocyte, and monocyte) and lymphoid (common lymphoid progenitor and B cells) lineages. However, as we gained more resolution, we observed that W305∗ frequency was underrepresented in myeloid and lymphoid progenitors compared with W305∗ frequency in HSCs as well as to the erythroid lineage (megakaryocyte-erythtroid progenitor, erythtroid progenitor, and red blood cell; Figure 4E). By contrast, R882H mutant DNMT3A cells showed a preferential expansion into the myemono- and granulocytic lineage (common myeloid progenitor, granulocyte, and monocyte), consistent with previous reports and the lineage bias detected in R882 mutant CH blood samples (Figure 4F), despite the erythroid stress incurred. Collectively, these data illustrate that the EPO-responsive DNMT3A W305∗ variant facilitates stable and balanced blood development during homeostasis, yet it promotes preferential erythroid reconstitution under stress induced by serial blood loss and EPO treatment. This clonal behavior sharply contrasts with the known preleukemic R882 DNMT3A mutation, which drives a pronounced myeloid bias both during homeostasis and under erythroid stress.

Although CH has been associated with aging and linked to RBC disorders, such as anemia^23,25 and erythrocytosis,²⁴ the impact of repetitive blood donation–associated erythropoietic stress on hematopoiesis and selection of HSC clones has not been studied. In this work, a total of 429 sexagenarians, including 217 frequent whole blood donors (>100 lifetime donations) and 212 sporadic donors (<10 lifetime donations), who served as an age- and health status–matched benchmark cohort, were analyzed for CH, one of the hallmarks of an aging hematopoietic system, in 2 independent cohort pairs. To our knowledge, this is the first work investigating the effect of frequent blood donations on clonal dynamics of a healthy hematopoietic system. Male blood donors were chosen because of their higher limit of maximum donations per year,² as well as due to the fact that the decades-long history of menstrual period bleeds, which are highly variable in volume, would have represented a substantial, unquantifiable confounding factor in females. Sexagenarians were selected due to the high background frequency of CH in this age group as well as the higher likelihood of achieving the set number of cumulative blood donations.

A first key observation is that we found no significant increase in CH prevalence among frequent whole blood donors compared with age-matched controls. In agreement with most aging-associated CH,^10-12,28,29,DNMT3A and TET2 were the 2 most commonly mutated genes in both cohorts, with no significant differences regarding overall incidence or VAF. The majority of mutations in these genes are considered to be “early hits” and deemed “low risk” with regard to their malignant potential.^36,51-53 This is consistent with a recent report that found, if anything, a reduced risk of AML was associated with blood donation.⁵⁴ Interestingly, and differing from historic cohorts, no JAK2 mutations were identified in the blood donors. We propose the makeup of our cohorts as the most likely reason for the absence of JAK2 mutations: blood donors are exquisitely healthy individuals. By contrast, all historic sequencing cohorts comprised both patients and healthy individuals, whereas JAK2 CH has been strongly associated with disease. JAK2 CH was found in patients with myeloproliferative neoplasm (essential thrombocythemia and polycythemia vera), cardiovascular disease, thromboembolism, and other disorders.^55-59 All these are conditions that would make an individual ineligible as blood donor. We did not detect any increase in known pathogenic variants, including premalignant hits such as the DNMT3A^R882 mutation. However, it is essential to emphasize that given the relatively modest size of our cohorts and the very low frequency of known pathogenic variants, no definitive statement with regard to safety in terms of the incidence of pathogenic hits in frequent blood donors can be made at this point. Future studies, ideally with cohort sizes similar to those of the UK Biobank,⁶⁰ will provide a more definitive answer to the question and a more comprehensive picture of the mutational spectra in healthy aged individuals. We hope that our work will inspire new prospective long-term studies of hematopoietic health of blood donors and other donor cohorts. It would be equally important to also monitor donors who eventually discontinue blood donations to avoid deselection bias of donors and ensure that donors developing medical conditions are appropriately followed.

According to the current paradigm, specific environmental cues favor the outgrowth of HSPCs carrying certain previously acquired mutations.^20,61-63 Different environmental stressors will have a varied impact on the fate of the diverse mutant clones.^17,64-66 Blood donation alters the proliferative behavior of HSPCs, skewing their differentiation toward erythropoiesis.^39,40 All RBCs are replaced 3 times per year,^13-15 that is, an RBC mass equivalent to 12 to 18 liters of whole blood,^4,67 corresponding to 24 to 36 blood units,^3,38 is generated. Each donated unit thus adds 2.5% to 4% excess erythropoietic stress in that year, with the donation of 6 units increasing it by 6 times this number. We hypothesized that exposure to this specific and persistent type of stress of the erythroid system represents a novel type of selection pressure, that is, frequent blood donation could favor the outgrowth of mutant clones responding to EPO stimulation. We re-created this stress by treating CRISPR-edited human HSPCs with EPO in culture. We tested the effects of EPO on HSCs carrying DNMT3A variants found in frequent blood donors and compared EPO with LPS and IFN-γ. The latter have been shown to support the outgrowth of preleukemic DNMT3A clones (R882H and R882C).⁹ In vitro, we observed mutation-specific responses to distinct environmental cues: frequency of FD DNMT3A variants increase in EPO-rich environments but not in response to inflammatory signals, whereas R882 mutant cells exhibited the opposite behavior. In vivo, our analysis of DNMT3A variants at the single-cell level in donor samples and reconstitution in humanized mice revealed the distinct impact on lineage contribution from HSCs carrying FD DNMT3A variants compared with the premalignant R882 mutation during homeostasis and erythroid stress, respectively. It further supports the notion that different stressors have divergent effects on HSCs harboring specific DNMT3A mutations. Importantly, our human HSC transplant model of RBC loss combined with exogenous bursts of human EPO represents a rather acute and aggressive scenario of human erythropoietic stress. Refined in vivo models are necessary to better reflect the subtle long-term erythropoietic pressure that FDs experience during their life span. The limited life span of humanized mice and the lack of suitable models for reconstitution of all human hematopoietic lineages (including erythropoiesis) in mice need to be considered when mimicking late effects of low-grade environmental stress.

In the light of the limitations associated with an in vitro and xenotransplant-based assessment of HSPC properties, our insight from primary donor samples using single-cell multiomics analysis is particularly informative. Comparison of mutant and nonmutant cells within the same sample confirmed distinct features of selected FD DNMT3A mutations. Premalignant R882 mutant HSCs display a marked myelomonocytic bias even in otherwise healthy individuals. By contrast, HSCs with FD DNMT3A variants did not show abnormal changes in their blood lineage contribution during homeostasis. Consistent with our findings, single-cell transcriptome and methylome analysis from DNMT3A-KO (resembling the effect of FD DNMT3A variants) mice point toward the expansion and transcriptional bias of HSPCs toward an erythro-megakaryocytic fate.^35,49,68,69 The different types of DNMT3A variants may share a proneness for megakaryocyte-erythropoietic skewing of HSPCs. Indeed, HSCs express the EPO receptor, and a direct shortcut differentiation trajectory from HSCs directly toward megakaryocyte-erythroid progenitors has been reported in mice.^50,70-72 

Overall, our experimental data provide a possible explanation for the enrichment of EPO-responsive clones over time; that is, for the higher presence of this class of DNMT3A variants in frequent blood donors. Whether this new class of DNMT3A variants promotes a proliferative advantage to HSCs only in the presence of EPO remains to be identified. Along these lines, EPO is by no means the only molecular player likely involved in the HSPC response in the setting of large-volume phlebotomy. For example, long-lasting depletion of iron stores in blood donors is well documented,^73-76 manifesting in the form of iron-deficient erythropoiesis and inherent reduction of ferritin levels, which can have direct and secondary effects on HSPCs. Further studies are required to comprehensively analyze the molecular interplay of the direct and indirect factors altered in FDs compared with CDs and how they may orchestrate the selection pressure and changes within the immature hematopoietic compartment. Another relevant aspect to be considered in the context of our in vitro experiments is the contribution of specific growth factor cocktails designed to maintain stemness⁷⁷ to the selection of certain clones.

Functional differences between various DNMT3A mutations have been comprehensively studied.^27,78 Thus, a new class of DNMT3A mutations, which result in decreased protein stability, has been defined and functionally validated recently.²⁷ These variants are of particular interest compared with the dominant-negative, stable DNMT3A variants, such as R882H.²⁷ Indeed, the average stability score²⁷ of DNMT3A variants enriched in the FD cohort was significantly lower than the variants enriched in the CD group. Moreover, the predicted reduced DNMT3A protein levels of the 3 EPO-responsive variants analyzed were corroborated by studies in K562 clones demonstrating that FD variant W305∗ generates reduced DNMT3A transcript levels and is associated with a shift in abundance of transcripts undergoing NMD. A phenotype of reduced DNMT3A activity of the blood donation–associated variants, as opposed to the aberrant methylation observed with preleukemic R882 variants, may explain the fundamentally different growth dynamics after cytokine exposure. DNMT3A transcript 2 was previously reported to be associated with active proliferation and malignancy and we observed that, in contrast to the reduced expression associated with W305∗ variant, R882 variants express higher levels of this DNMT3A transcript (Figure 3C; supplemental Figure 4C). To our knowledge, distinct effects of different DNMT3A mutations on the abundance of different DNMT3A transcripts have not been reported before. Future work will help define the role of the altered transcript balance on the growth dynamics not only for DNMT3A variants predicted to undergo NMD but also for other types of DNMT3A variants.

Interestingly, when the recently described fitness score (s)²⁸ was applied, the DNMT3A variants from the FD cohort had a significantly lower score than matched variants identified in the control cohort. This may suggest that DNMT3A mutations expanding with bleeding associated stress display only moderate fitness in the general population and in the absence of additional stimuli. In line with this and for the ones tested, the VAFs of most DNMT3A mutations remained stable within short periods of time between 2 consecutive donations. In contrast, in the few cases in which we analyzed a known premalignant variant (DNMT3A R882C and SRSF2 P95H) its pathogenic potential was indeed reflected in a twofold to threefold expansion of the clone during the same interval (supplemental Table 6 and data not shown). Nevertheless, an observation period of 1 year (on average) is certainly too short for conclusive assessment of the growth kinetics of a variant.

The increased frequency of EPO-responsive DNMT3A variants in FDs suggests causality. Although the acquisition of a given variant is stochastic, microenvironment-driven evolution in the context of extensive blood donation, a previously undescribed selection pressure, appears to favor this novel class of DNMT3A mutations with an otherwise low to moderate growth potential. This phenomenon is an example of ongoing nuanced Darwinian evolution at the level of somatic stem cells in healthy individuals. The EPO-responsive DNMT3A mutations described here suggest that different stressors have divergent effects on HSCs harboring specific DNMT3A mutations. Furthermore, our study introduces an important and uniform reference CH data set from a novel population of exclusively healthy individuals distinguished by regular exposure to a highly specific systemic stress. Because most recently the age limit for blood donation was lifted in Europe and experienced donor continue, their health status permitting, to donate until well into their eighties, we will seek to corroborate both quality and quantity of CH in older populations. Moreover, longitudinal sampling of buffy coats of selected donors over decades will inform about long-term clonal dynamics.

First and foremost, the authors thank all volunteer blood donors for their dedication and altruism with which they continue to provide a lifesaving resource. They thank Stefanie Müller, the physicians, and the technical staff involved in collection and processing of the whole blood units, as well as the IT Department at the German Red Cross Blood Donation Service Baden-Württemberg-Hessen for their assistance in identifying and supplying the samples. The authors thank Steffen Schmitt, Marcus Eich, Klaus Hexel, Tobias Rubner, and Florian Blum from the German Cancer Research Center (DKFZ) Flow Cytometry Core Facility, as well as the DKFZ Genomics and Proteomics Core Facility and the DKFZ Single-cell Open Lab for their assistance. The authors further thank the Genome Technology Access Center and McDonnell Genome Institute at the Washington University School of Medicine. They also kindly acknowledge Andre Lieber from University of Washington (Seattle, Washington) for providing the HUDEP-2 cell line; Alexander Waclawiczek and Moritz Gerstung (DKFZ) for critical revision of the manuscript, helpful comments, and discussion; and Shawn Clouthier, Brittney Otero, Ohimai Unoje, Marc Arribas-Layton, and Robert Durruthy-Durruthy from Mission Bio for their support in setting up Tapestri platform–based single-cell analysis of the samples and support during bioinformatic analysis.

This work was partly supported by the SPP2036 and SFB873 funded by the Deutsche Forschungsgemeinschaft, the ERC Advanced Grant SHATTER-AML (AdG-101055270) and the DKTK joint funding project RiskY-AML, the Integrate-TN Consortium from Deutsche Krebshilfe, and the Dietmar Hopp Foundation (A.T.). This work was partly supported by Cancer Research UK (CC10045), the UK Medical Research Council (CC10045), the Wellcome Trust (CC10045 [D.B.]). H.H.E. was supported by the Kay Kendall fellowship (KKL1397) and Cancer Research UK Early Detection and Diagnosis Project Award grant (Nov23/100002). M.S. received funding through the Deutsche Forschungsgemeinschaft Walter Benjamin Fellowship (493935791).

Contribution: D.K., H.H.E., D.B., H.B., and A.T. designed the research; D.K., H.H.E., E.D., A-M.L., R.W., P.S., I.K., S.H., D.P., A.F., S.N., J.R., and A.P. performed experiments; D.K., H.H.E., M.S., A.-M.L., J.P., and M.L.-S. analyzed data; I.K. performed structural modeling; S.C. and A.K.-S. performed statistical analysis; D.K., H.H.E., I.K., S.C., M.S., A.-M.L., and J.P. directly accessed and verified the underlying data reported in this manuscript; K.Z., D.W., and T.T. provided critical reagents for the study; L.V., J.F.D., T.N.W., and A.K.-S supervised certain aspects and D.B., H.B., and A.T. supervised the study overall; D.K., H.H.E., T.N.W., H.B., and A.T. wrote the original draft; D.K., H.H.E., E.D., A-M.L., T.N.W., D.B., H.B., and A.T. reviewed and edited the manuscript; and all authors had full access to all the data, discussed, commented on, and approved the final version of the manuscript, and accept responsibility to submit for publication.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Andreas Trumpp, Division of Stem Cells and Cancer, German Cancer Research Center, Im Neuenheimer Feld 280, 69120 Heidelberg, Germany; email: a.trumpp@dkfz.de; Dominique Bonnet, Haematopoietic Stem Cell Laboratory, The Francis Crick Institute, 1 Midland Road, London NW1 1AT, United Kingdom; email: dominique.bonnet@crick.ac.uk; and Halvard Bonig, German Red Cross Blood Donation Centre, Department for Cell Therapeutics, Sandhofstrasse 1, 60528 Frankfurt am Main, Germany; email: h.boenig@blutspende.de.