Reframing—Renaming(?)—Myelodysplastic Syndromes/Neoplasms and Clonal Hematopoiesis of Indeterminate Potential

To generalize is to be an idiot. To particularize is the alone merit of distinction.

William Blake

Keine neue Welt ohne neue Sprache (No new world without new language)

Ingeborg Bachmann

This essay is a perspective, an alternative view of the current nosology—and conceptions—of myelodysplastic syndromes/neoplasms (MDS) and clonal hematopoiesis of indeterminate potential (CHIP), names that we will argue have outlived their historical trajectory.

Rethinking and potential renaming of MDS and CHIP offers important advantages. First, in the clinic, to diminish the confusion of physician and patient, often exacerbated by Dr. Internet. Second, to place emphasis on assessing and addressing the bone marrow failure that is immediate and frequent in its manifestations, and, conversely, on earlier and tailored strategies for specific treatment of nascent leukemia. Third, to highlight distinctive pathophysiologies, now obscured within complex classification schemes, to improve diagnosis and treatment, and to serve as model processes in less understood syndromes. Fourth, to replace subjective morphologic markers and squishy nomenclature, and vague, unhelpful terms of “indeterminate potential,” and “uncertain significance,” with more current knowledge from large, curated data sets, advanced genomics and proteomics, and sophisticated computational analyses.

We provide here a different stance towards familiar material, not a comprehensive review of the topic. We outline possible classification schemes, which would need to be substantiated as hematologists move towards a modern consensus. Our goal is to provoke new thinking, to stimulate a fresh look at the old concepts of MDS and more recent notions underlying CHIP, and to encourage the expert community to broadly consider new approaches to a large group of serious diseases. We acknowledge the utility of current algorithms, some conveniently available as internet applications, aimed at predicting prognosis.^1–3 Scoring systems in medical practice can be agnostic of causation (as they are in the online commercial world) and of practical utility even if arbitrary in their parameters (economic, social, and insurance status are rarely included). “Black box” tools will be increasingly employed in the age of artificial intelligence, but they should not be confused with understanding of mechanisms and are likely to confuse causation (as they have in the commercial world).^{4, 5}

In an important mid-19^th century volume on inductive methodologies in science, biology was aligned with—minerology!—as a catalogue of descriptions, in contrast with the more concrete “hard sciences” and also with paleontology and geology, in which causation could be inferred from temporal sequences. Two centuries later, in the DNA era and with exploitation of genomics and multi-omics, in routine laboratory testing, there is another impetus to shift from description to understanding. There are excellent publications by others of similar opinions, which argue the primacy of genomics and the inadequacy of morphology in diagnosis and prognosis.^{1, 6–9} Further, analogous conflicts in terminology and semantics have arisen elsewhere in modern medicine: examples range from the placement of sickle cell anemia among the hemoglobinopathies and the arrangements of lymphoproliferative diseases.¹⁰

Background.

The history of MDS is well described and can be briefly summarized.^11–13 During the 20^th century, anemia became easy to diagnose and bone marrow biopsies routine; discovery of the cause of pernicious anemia and its effective treatment impelled efforts to define other syndromes and cures; and hematology developed as a medical specialty. Despite advances, some patients suffering refractory anemia were placed in poorly defined, “leftover” categories, based mainly on the histologic appearance of blood forming cells in the marrow: odd megakaryocytes, myeloid precursors without granules, and “megaloblastoid” erythroid forms.

Hematopathologists struggled to make consistent, clinically relevant diagnoses in such cases. Bone marrows sign outs with some similarities to leukemia, for example increased monocytes and low percentages of myeloblasts, led to unsatisfying terms such as “pre-leukemia” and “smoldering leukemia.”¹⁴ In the late 1980s, a small French, American, and British team of laboratory pathologists, dubbed FAB, formulated a useful five part classification under the heading of myelodysplastic syndromes (MDS). The FAB identified patients who did not appear to require leukemia treatment, in whom anemia was accompanied by ring sideroblasts, or monocytosis, or cytopenias with peculiar histologic variants of early myeloid precursors or megakaryocytes.¹⁵ Vague descriptions and eponyms were replaced by terms derived from more obvious features of marrow pathology.

The FAB scheme proved popular. It evolved over decades, incorporating and occasionally discarding categories, as cell type identification and enumeration became automated, cytogenetics routine, and genomics incorporated into diagnostic testing. However, the numerical border of marrow blasts between MDS and leukemia was troublingly mutable; chronic myelomonocytic leukemia seceded to become its own disease in an entirely separate classification system; and some entities were forced into the MDS schema (vacuoles, E1 enzyme, X-linked, autoinflammatory, somatic [VEXAS] syndrome) or excluded (paroxysmal nocturnal hemoglobinuria [PNH]) for unclear reasons. Meanwhile, cooperative research efforts established large databases, with outcome correlations and scoring systems,^{16, 17} helpful in establishing natural history, assigning patients to risk categories, making decisions among therapies, and comparing treatments and outcomes. Yet the unprecedented, recent publication of two different classifications (with up to 11 subclassifications)^{18, 19} adds to the difficulties of the pathologist and treating physician. While the revisions have advanced molecularly defined diseases (del(5q), SF3B1),^{20, 21} the retention of historical labels, such as MDS with low blasts and SF3B1 mutation and/or ring sideroblasts (where RS ≥ 15% is a permissible substitute for SF3B1 mutation) in WHO HAEM5¹⁹ allows for genetic heterogeneity, such as SRSF2 mutated MDS.²²

The CHIP story is shorter. When DNA sequences from large population databases were examined for the presence of mutations in genes known to be recurrently mutated in myeloid cancers (acute myeloid leukemia [AML], myeloproliferative neoplasms [MPN], and MDS), these genes were mutated at high frequency in the blood cells of older individuals. The specific mutated genes were a small subset of more than 50 myeloid neoplasm genes, mainly DNMT3A, ASXL1, and TET2. Mutated clones were age-dependent in their frequency and size, and correlated to overall mortality.^{23, 24} Even more surprising than the prevalence of CHIP clones in the elderly was their strong association (especially for TET2-mutated clones) with non-hematologic diseases, especially atherosclerotic coronary and cerebrovascular events initially,^25–27 and subsequently many others (chronic obstructive pulmonary disease,²⁶ gout,²⁸ and protection from Alzheimer’s^{29, 30}).

CHIP was defined by a mutant allele frequency of 2%, practical, at the time, based on the sensitivity of sequencing in clinical laboratories rather than clinically important.³¹ Given observed cytopenias with CHIP, a second sub-term was coined, clonal cytopenia of undetermined significance (CCUS), a diagnosis between CHIP and MDS, analogies made to monoclonal gammopathy of undetermined significance (MGUS), smoldering multiple myeloma (SMM), and MM. Despite the weak association of CHIP with hematologic malignancies in the initial, large cohort studies,^{23, 24} CHIP clinics have sprung up at major (predominantly American) medical centers, and alarms sounded at the inevitability of leukemia with aging.

Defining disease in MDS and CHIP.

In retrospect, MDS as a grouping seems particularly problematic, and not only for such practical difficulties as the vagaries of histopathology, poor consensus among “blinded” pathologists,^32–34 and widely variable clinical and laboratory findings. The fundamental problem: the myeloblast, always the major predictor of risk of progression to leukemia and survival, and therefore the single feature most concerning to the treating physician (Figure 1A).^{16, 17} Myeloblasts can increase in the marrow and blood transiently, following growth factor therapy or in sepsis with a marked left shift, but their persistent presence almost always heralds poor prognosis and eventual leukemia.

Figure 1.

Top. Heterogeneity and category errors in myelodysplastic syndromes/neoplasms (MDS) and clonal hematopoiesis of indeterminate potential (CHIP). Two examples are shown, for MDS and CHIP; transformation to leukemia is the endpoint. A. For MDS, analysis of patients in the IPSS-M MDS cohort were stratified by bone marrow blast level and specific mutations, for which mutation groups were defined by a causal approach to their varying effects on bone marrow blasts (see Supplement S.2 to S.4).³⁵ Patients with bone marrow blasts >2% have much higher rates of leukemic transformation compared to those with minimal to no blasts. Note the extreme range of leukemic risk, as between TP53 biallelic mutations and SF3B1 mutations, both within MDS. B. For CHIP, without hematologic abnormalities by definition, commonly occurring mutations confer little clinical relevant value, only modestly increasing a rare event in an aged population (e.g., DNMT3A CH),¹ but some specific mutations appear to predict AML. Singular mutations, like JAK2, have high penetrance to MPN, not AML, altering the risk perception of acquiring leukemia when removed from CHIP/CCUS cohorts, as done in this panel. The proportion of patients in each group with mutated clones strongly predictive of leukemic progression increases, from 1% or less of CHIP to about 45% in MDS (in both, there is evidence of disease by definition). Y-axis is based on the proportion evolved: JAK2 to MPN and other CHIP/CCUS and single DNMT3A to MDS/AML.

Bottom. The kinetics of “acute” myeloid leukemia. Disease-free survival is plotted on a time scale in order to compare MDS, CHIP, and “benign” hematologic diseases, and for comparison some lymphoproliferative syndromes. C. Median disease free survival (DFS) or 10-year DFS are plotted based on representative studies with variable endpoints of AML, myeloid neoplasia (MN), MDS/AML, and MM.^{1, 7, 35–39} Scaled comparison highlights the heterogeneity in the time course of development of frank leukemia; CHIP and CCUS data are usually displayed over a decade or more, from identification of a mutated clone to AML. The high progression rate of CCUS with specific myeloid neoplasia mutations is similar to very high-risk MDS (despite different malignant entity endpoints),^{7, 35} interpretable as developing, early, or transitioning leukemia. Note the high rate of progression of high-risk (HR) CHIP,¹ which is greater than for the majority of MDS categories in IPSS-M (there are different malignancy endpoints).³⁵ From the lower left graph, low-risk CHIP transformation rate is much lower than for SMM or MGUS. Although CHIP has been termed a pre-malignant condition, the magnitude of difference in leukemic risk suggests that this is inaccurate as related to the vast majority of patients. On the lower right graph, “benign” hematologic conditions, such as immune aplastic anemia, are compared to “low risk” MDS, demonstrating similar or lower malignant potential; it would be reasonable to consider these conditions “benign.” Line type represents the end-point of disease: solid, AML; short dashed, MN, a composite of MDS, MPN, and AML; long dashed, AML and MDS; dotted, MM. Line colors represent disease: black, MDS; orange, CCUS; yellow, CHIP; green, aplastic anemia (AA); blue, myeloma precursors, SMM and MGUS. Abbreviations: AML, acute myeloid leukemia. CCUS, clonal cytopenia of uncertain significance. CHIP, clonal hematopoiesis of indeterminate potential. DFS, disease free survival. IPSS-M, international prognostic scoring system – molecular. MDS, myelodysplastic neoplasm/syndrome. MGUS, monoclonal gammopathy of undetermined significance. MN, myeloid neoplasia. SMM, smoldering multiple myeloma. MPN, myeloproliferative neoplasm.

The introduction of measurable residual disease (MRD), assessed by flow cytometry or sensitive genomic sequencing in AML, has revealed the inadequacy of visible myeloblasts in reflecting the vast numbers of leukemic cells, hidden from sight in the conventional “remission” marrow.^{40, 41} Indeed, the WHO and other expert organizations have repeatedly lowered the blast threshold required for an AML diagnosis. The thresholds are based less on biology than on the expected progression and likelihood of remission. Conversely, there is a fundamental and unsettled question of the definition of the cancer cell: cells with seemingly sufficient features of malignancy have been known to be present in large proportions of healthy persons, whether defined by now primitive methods to examine tissue mosaicism,^{42, 43} refined single cell methodologies,^{44, 45} or more simply monoclonal gammopathy.^{10, 46}

Indeed, the relationship of small proportions of blast cells to leukemia is unclear, and their importance increases as molecular assays, designed to detect measurable residual disease after cancer treatment, increase in sensitivity. A model exists in the recently defined genetic subtypes in the WHO5 classification: NPM1 mutations can constitute leukemia, regardless of blast count (with appropriate “judicious clinicopathologic correlation”) given the documented natural history of rapid progression.¹⁹ So, are blast cell counts or genomic signatures better at denoting tiny populations of cells at the very earliest stages of leukemia, and can we distinguish between the originating leukemia cell(s) and more indolent, near physiologic blasts unlikely to expand and manifest as disease? Will the differences between the two be intrinsic to molecular signatures, externally determined by the local environment and the immune system,^{47, 48} or a combination of the two? For more molecularly complex cases, can our understanding of somatic clonal evolution identify a “minimal initiating cell” (MIC), allowing leukemia identification before an arbitrary blast threshold? What would re-conception of leukemia from these perspectives leave of “MDS”? Would refocusing on molecular, cellular, and immune biology better identify leukemia risk, bone marrow dysfunction, and which therapeutic targets to exploit, in which patients?

The understanding of CHIP, benefiting from its discovery in the era of (and by means of) rapid, cheaper genomics and the availability of enormous databases and biobanks, has evolved quickly. Initially proposed as a descriptive label,³¹ it is now codified as a pre-malignancy in the WHO/ICC classifications.^{18, 19} However, clonality is not distinctive to cancer. Hematopoiesis with normal blood counts was recognized as clonal, long before the introduction of deep sequencing, through X-inactivation studies, somatic mosaicism, and colony formation.^49–51 The specificity of myeloid mutations as the sole, primary drivers of clonality has been challenged by analysis of whole genome sequences: clonal hematopoiesis was defined by CHIP (mutated myeloid neoplasia genes) or by recurrent mutations in non-driver genes (through barcoding) with highest hazard of hematologic malignancy and mortality in CHIP⁵² but still elevated in CH without (known) drivers.⁵³ Certainly, a proportion of CH without known drivers will have new, causal candidates identified outside the cassette of myeloid neoplasia-associated genes as our genomic analyses continue to progress.

Further, the risk of acquired clonal hematopoiesis has been linked to germline genes,^{52, 54–56} almost always variants related to cell cycling, cell proliferation, telomere maintenance, and hematopoietic stem cell self-renewal, functions themselves independently associated with neoplasia. For instance, splicing factors, canonically associated with MDS, typically develop after the seventh decade of life,^{57, 58} except in individuals with telomere biology disorders,⁵⁹ prompting investigation and discovery that telomere attrition with aging is a selection pressure for splicing mutations and may slow accumulation of proliferative mutations.⁶⁰ Paradoxically, telomere attrition, an anti-aging target and pathogenic when accelerated, would serve as a mitotic clock and also cellular “alarm clock,” to drive cells likely to have acquired somatic mutations in to senescence or death. Recently, TCL1A was described to be important to HSC self-renewal capacity and is aberrantly expressed in certain CH; a single nucleotide variant (rs2887399) appeared to reduce the prevalence and clonal expansion rates for cells mutated in TET2, ASXL1, SF3B1, and SRSF2.⁶¹ These CH drivers rely on TCL1A expression, which is sufficient to independently promote HSC expansion, with a proposed mechanism of maintaining proliferative capacity while avoiding the negative feedback of the integrated stress response. Questions remain: Does CHIP drive malignancy directly (Figure 1B), or is it secondary to increased genomic instability, an altered marrow environment, deficient immunity, a declining stem cell reserve and reduced population size—or to increased systemic inflammation? Accumulating data suggest that clones mutated in DNMT3A,⁶² TET2,⁶² UBA1,⁶³ thrive under inflammatory conditions, a functional trait which may be advantageous early in life in response to infectious challenges but deleterious in the setting of non-infectious inflammation.⁶⁴

Statistical difficulties: association versus causality.

Difficulties defining disease boundaries in MDS and CHIP are related to questionable conclusions from traditional statistics: agnostic, employing regression analysis with independent variables lumped and internal connections, other than as confounders, ignored (Figure 2A).⁷³ Time-to-event analysis in the MDS and CH literature frequently use the Cox proportional hazards model, only reporting a single hazard ratio derived from baseline features rather than several timepoints. While statistically efficient, this approach risks conflating instantaneous hazard with longitudinal risks and does not address the difficulties in predicting survival based on the time-varying covariates with disease progression,^{74, 75} and may leave the distinction of hazard ratios, often the only risk cited, from longitudinal cumulative risk unclear, confusing physicians and patients. A different statistical approach is causal inference: directional arrows help define entities and distinguish them from unrelated or modifying factors, from precursor events, and from sequelae (Figure 2B). Causal inference can be used to statistically verify—and quantify—outcomes from data.⁷⁶ For MDS and CHIP, datasets have been analyzed using classical statistics, but there is ample evidence (from cell biology, animal models, and clinical observation) to denote some associations as causal (a mutated gene), others subsequent (platelet count, red cell distribution width). By diagramming a clinically meaningful causal structure graph, causal inference can be applied, effect sizes or probabilities assigned, direct and indirect effects mapped, and mediators at convergence points or unknown factors considered based on prior knowledge (Figure 2B–G). Visualizations from causal inference can strikingly refocus research questions and impact the design and interpretation of observational and interventional clinical protocols. While obviously useful, conventional statistical applications may inadvertently produce complex, unintuitive results, of limited biological meaning.^{9, 77} As an example, in scenarios such as CH and cardiovascular disease, analyses have produced conflicting results as to whether entities are causally related—a causal analysis diagram would consider the differential effects of smoking on ASXL1 CH, whether mosaic chromosomal abnormalities should be included, and the striking role of systemic inflammation in cardiac risk.^{53, 78–81} Consideration of causal relationships can yield distinct, clinically meaningful clusters that include valuable patterns of origin of disease, its progression, and response to therapies.^{8, 82, 83}

Figure 2. Causal inference and disease.

Assigning arrows denoting causality fundamentally alters traditional statistics. A. A classic representation of associated risk factors for a disease, without regard to causality or temporal relationship (see Supplement S.1). For a given outcome (AML), any parameter is considered equivalent in arriving at statistical significance: previous chemotherapy, radiation, genetics, age—a mix of preceding events and circumstances, and often also (early) consequences of an endpoint, such as a symptom, sign, or laboratory value (for AML, perhaps type B symptoms or RDW^{1, 2}). B. Causal analysis diagram, identifying known etiologic factors, such as genetics, aging, radiation therapy, and chemotherapy, for the development of AML, through induction of somatic mutations (or chromosomal abnormalities, methylation profile changes, or selection pressure). In a causal diagram, nodes (circles) are linked by arrows. Nodes represent discrete states, not cells or cell populations; arrows indicate a unidirectional relationship. A causal diagram can disclose conceptual errors, as in the use of surrogates like “previous cancer” and that “RDW” is a consequence of mutations and altered hematopoiesis rather than a cause of acute leukemia. Causal inference allows the addition of facilitating and blocking factors (immune compromise or activation, “immune surveillance,” are examples). A causal diagram is suitable for machine learning and artificial intelligence approaches. C. A general causal diagram for the modeling of somatic mutation disease. P represents probabilities of state transitions: these are often known empirically or can be experimentally tested. Examples are germline genetics, chance (in the form of DNA replication errors), and environment shape somatic genetics.⁶⁵ P₁, as an example, is high, and the broad arrow represents somatic variants incompatible with cellular function (intrinsic) or are negatively selected (extrinsic), leading to cell death, senescence, and clonal extinction. Along the pathway labeled P₂, passenger mutations accumulate if there is no change in function, the baseline accumulation rate mutations (a natural method of “barcoding”). The lowest probability, P_3, points to a somatic mutation that has the potential for clonal expansion, in turn represented by arrows P_e. Clonal expansion is context dependent not only on intrinsic cellular “fitness,” but also likely intra-clonal dynamics, cell-cell interactions, cell-stroma interactions, immune surveillance, and the environment.⁶⁶ Somatic mutations (P₃, P₃’) occur at certain rates and transformation to blasts (P_b) can only occur on certain somatic genetic backgrounds. D. A simplified causal schematic of leukemogenesis; P_i, labels a specific initiating genetic event required for blast transformation, P_b.⁶⁷ Transformative mutations are context-dependent and require the correct initial state (e.g., FLT3^ITD mutations is absent in the clonal hematopoiesis literature).^68–70 E-G. Somatic mutations in benign hematologic “disease,” emphasizing the boundary of “disease” (orange nodes) and its recognition as a diagnosis (orange dotted lines). Canonical PIGA, UBA1, and JAK2 mutations are present in a large proportion of healthy individuals who do not demonstrate a clinical phenotype.^{71, 72} Selection in an autoimmune environment leads to PIGA-mutant clonal expansion and PNH. For myeloproliferative neoplasms (MPNs), originating JAK2 V617F mutations have been traced to a post-zygotic event in the embryo and can be detected over decades at low frequency, long before achieving clonal dominance, manifesting symptoms; yet clonal hematopoiesis driven by JAK2, CALR, and MPL appear to have high penetrance to clinical disease.^{1, 2} When probabilities of transiting states are high, can a diagnosis be made early, even at low levels of clonal expansion, warrant pre-emptive treatment? Abbreviations: AML, acute myeloid leukemia. CHIP, clonal hematopoiesis of indeterminate potential. ITD, internal tandem duplication. MPN, myeloproliferative neoplasm. Mut, mutation. PNH, paroxysmal nocturnal hemoglobinuria. RDW, red blood cell distribution width. VEXAS, vacuoles, E1 enzyme, X-linked, autoinflammatory, somatic.

Applied to MDS and CHIP, causal analysis could aid in charting the biologic progression of somatically mutated cells, from their appearance with normal aging to risk-bearing CH mutations, from both cellular functional deficits and progressive molecular changes underlying progression of CCUS to MDS, and of MDS to AML (Figure 2C–D). In this manner, MDS with high probability of eventual leukemic transformation would simply be considered (early) AML, and treated as such, while those that have a somatic background without transformative potential could be separated into a benign category with a different therapeutic focus: alleviation of cytopenias, symptoms and improving quality of life.

Ontogenic errors, category mistakes.

In retrospect, the inclusion of marrows showing excess blasts as a subtype of MDS appears a “category mistake,” the confusion of an entity with its features (an ancient philosophical and theosophical conundrum, separating essence from attribute!⁸⁴). The threshold for the diagnosis of acute leukemia has been lowered over time, from 30% marrow blasts to 20% to 10%—but perhaps not enough. A “safe” (or non-diagnostic) level of marrow blasts seems imaginary, as more leukemia can be diagnosed below blast numbers determined by morphology, as in the detection of MRD after cancer treatment.^{40, 41} Determinant requisite genetic events, causal in leukemic evolution, should be detectable with molecular methods to identify minimal “initial” disease (MID?) in leukemia or what is now termed “high risk” MDS (Figure 2C–D). Thus the historical emphasis in MDS classification and prognosis schemes on “progression” to AML is illogical if a large proportion of patients with the MDS diagnosis—those with blasts or a certain combination of genetic mutations causally linked to leukemic development—already have a form of early “AML,” the very entity to which future risk is being assessed, and if low levels of blasts in high risk MDS simply represent earlier stages in the progression of leukemia (not of MDS to leukemia; 1Figure 1C).⁹ There have been efforts in this direction. NPM1 AML is now genetically defined (and does not require a minimum blast count), but the evidence for this change emphasizes the high risk of progression from MDS to AML, a different paradigm than considering they are a singular biologic entity that presents on a spectrum of ever increasing blast count.^85–87 Conversely, the molecular or functional signature of blasts not determined to inexorably expand remains unclear (despite being the FAB’s main goal, to define more indolent cases, with the MDS classification¹⁵), but serial sample testing is outlining the causal mechanisms of clonal trajectories and therapeutic targets, as in ASXL1 and high-risk clonal evolution in aplastic anemia and IDH1/2 mutated CH and AML with IDH inhibition and RAS pathway resistance mutations.^{88, 89} Our historical emphasis on morphology and blast count predates our molecular characterization of leukemia, where they were the only tools available, but we continue to conflate the degree of blasts as defining (essence) of leukemia rather than a feature (attribute). To avoid the category mistake of conflating an entity (AML) with features (blasts), blasts, could be explored with a causal framework—what are blasts a surrogate for, somatic mutations or chromosomal translocations (or epigenetic changes) that cause a maturation defect and marrow failure or a proliferative advantage?

A different category error appears in CHIP, apparent in the first published analyses. In these reports, there was not a strong claim of a link between the presence of clones mutated in myeloid neoplasia genes and later occurrence of MDS and AML: statistical significance required grouping on “hematologic malignancy” (and inclusion of lymphoid cancers).^{23, 24} Initially, AML, MDS, and MPN were combined as a single outcome, despite MPN’s different genetic features, pathophysiology, and presentation.^{1, 23, 24} Distinct clinical trajectories were not visible in composite endpoints. Mutations can be mapped to their likely, different termini: AML, MDS, and/or MPN (Figure 3A), as done in recent studies.^{2, 53, 91} Once mapped, hazard ratio (HR) interpretation is not intuitive: a HR represents an instantaneous risk over the time period examined, and does not map directly to a risk reduction in myeloid neoplasm or overall survival.^{74, 75} To the patient and practicing physician, clinical calculators, such as CHRS, CCRS, and MN-predict offer more intuitive risk interpretation and come with their own limitations,^1–3 one of which is their single, static timepoint of evaluation given clonal dynamics (expansion rate) can mediate nearly half of the CHRS’ predictive value.⁹² For instance, a 63-year-old patient with a single non-R882 DNMT3A mutation (VAF 18%, MCV 96, RDW 16) has a HR of 8.45 for AML,⁹¹ but, in contrast, the CHRS and MN-predict would both estimate the patient is low-risk (<1% MN development and 1% AML, 1% MDS development, respectively)—seemingly disparate outcomes.

Figure 3. Mutated genes, clones, and their diseases.

A. CH somatic mutations mapped to the risk of specific myeloid malignancies. Analysis of the UK Biobank data demonstrates that certain mutations in clonal hematopoiesis are distinctly associated with the development of particular myeloid neoplasms.² For instance, a CALR or JAK2 somatic mutation in CH makes a myeloproliferative neoplasm much more likely than a myelodysplastic syndrome or acute myeloid leukemia. In comparison, DNMT3A is the majority of CH (circle size is scaled to prevalence) and has decreased risk (hazard ratio < 1). Nucleotide mutation site matters, as DNMT3A R882 “hotspot” mutations do confer increased risk of AML. The scale of the axes is based on hazard ratios (0–5). X-axis is acute myeloid leukemia (AML; red). Y-axis is myeloproliferative neoplasm (MPN; yellow); Z-axis is MDS (blue). Circle color corresponds to malignancy progression (green = MPN and MDS) based on hazard ratios.² Circle area represents prevalence,^{1, 90} and are to relative scale (other than the smallest size; e.g., RUNX1). B. Interactions are complex in clonal “benign” hematologic disease (such as hypoplastic MDS and SF3B1 sideroblastic anemia), benign multisystem disease (VEXAS), and malignancy. Three axes allow for multidimensional data with depiction of severity for these three variables, highlighting the clustering of diseases like hypoplastic MDS with immune aplastic anemia that could be more nosologically akin to each other than in the current, separate classifications in “benign” disease and myelodysplastic neoplasms. C. Clusters of diseases can be further stratified and examined by mechanism. For “benign” diseases affecting the hematologic system, there can be a multitude of pathophysiologies, including immune-mediated (immune AA, hMDS) and primary stem-cell dysfunction (SF3B1-mutated sideroblastic anemia), for the former treatment with immunosuppressive therapy is beneficial while for the latter targeted treatment on the mechanism of dysfunction is ideal. For multisystem “benign” disease, one could categorize entities by their pathophysiology with antibody-mediated, complement mediated (PNH), and cell mediated (VEXAS, TET2-opathy), hinting at treatment. Malignant predisposition can be viewed by time to malignancy (long latency, DDX41) and predisposition to hematologic cancer alone (DDX41, RUNX1) versus hematologic and solid cancers (telomere biology disorders, Fanconi anemia), which informs the pathophysiology of these diseases. Abbreviations: AML, acute myeloid leukemia. CH, clonal hematopoiesis. FA, Fanconi anemia. ET, essential thrombocytosis. iAA, immune aplastic anemia. hMDS, hypoplastic MDS. MDS, myelodysplastic syndrome/neoplasm. MPN, myeloproliferative neoplasm. PNH, paroxysmal nocturnal hemoglobinuria. PV, polycythemia vera. SA, sideroblastic anemia. TBD, telomere biology disorder. UK, United Kingdom. VEXAS, vacuoles, E1 enzyme, X-linked, autoinflammatory, somatic.

There is sufficient concordance in hazard ratios, odds ratios, and survival analysis amongst these studies to make estimations about low-risk and high-risk CHIP. “Low-risk” CHIP could include low variant allele frequency, non-R882H DNMT3A somatic mutations, but we must consider whether we have sufficient information to define this as disease causing—and the people with them patients—especially with animal models and single cell analyses in humans demonstrating an adaptive resistance to inflammation (associated with aging).⁶²

In “high risk” CHIP,¹ as for MDS, there is a rational argument that patients with a narrowly defined set of mutated clones may already have leukemia, but an AML remarkable for its slow development over time (Figure 1C). Clones of IDH2, RUNX1 and splicing factor gene-mutated cells were present in only 1% of CHIP cases, but they carried >80% cumulative risk of progression to myeloid neoplasm ten years after first detection.⁹³ In limited single cell analyses of IDH1/2 mutated CH, the affected cells demonstrate a similar phenotype to IDH1/2 mutated AML cells.⁸⁹ Similarly, the absence of NPM1 and FLT3^ITD mutations in CH and their importance in AML has suggested their role in explosive, transformative growth. The examination of CH mutations, in isolation and combination, guided by epidemiologic studies, will aid in purposefully redrawing disease boundaries informed by biology.

Similar reasoning has been applied to CCUS: cytopenic patients with bland marrow appearances but harboring mutations in specific genes alone (SRSF2, U2AF1), and especially in combination with DNMT3A, TET2, or ASXL1, and at high variant allele frequencies, could be diagnosed as MDS (Figure 1C).⁷ So, is CCUS with mutations in myeloid neoplasia genes equivalent to MDS at an earlier stage and before more evident morphologic and clinical manifestations (ultimately AML in some, severe cytopenias in most)?⁷ Moving CHIP beyond “indeterminate potential” should be good news for worried patients, and advancing CCUS from “undetermined significance” to true early leukemia or to marrow failure in most should lead to earlier recognition of disease, early intervention, and better genetic characterization of syndromes.

Reframing MDS and CHIP.

Labels are sticky. Relabeling might discomfit MDS doctors, and textbook chapters (or online guides) would need to be rewritten. Medicine is conservative, and traditional nomenclature may stubbornly persist in terms repurposed from their original meanings. Nevertheless, counterfactuals are useful for critical analysis of current nosology, in questioning the meaning of familiar labels. What is MDS without cases containing blasts, if they are indicators of the onset of leukemia? What is MDS, or CCUS, solely characterized by genomics, before morphologic changes arise from specific functional changes? How “acute” is some AML if certain variants progress to manifest disease over a decade or more from what is now considered high-risk CHIP? “MDS,” “CHIP,” and “AML” might remain usable symbols, abbreviations best not spelled out, of recognizable but highly variable syndromes.

An initial reframed view for MDS is from a (relatively) non-malignant perspective, separating bone marrow failure, the major clinical manifestation overall, from leukemia.^{1, 2} Much low risk and hypoplastic MDS resembles aplastic anemia and related immune marrow failure syndromes, in clinical presentation, response to therapies, inflammatory features, and genomic profile (Figure 3B).^{36, 94} Cytopenias predominate, but also there are varying (lower) probabilities of evolution to high risk cytogenetic abnormalities, second tier high-risk mutated clones, and frankly dysplastic and myeloblastic bone marrows based on disease pathophysiology. There are concrete genetic indications of pre-malignancy, as in post-chemotherapy, immune aplastic anemia, telomere biology disorders,^{59, 60} that are clinically useful even if mechanistically incomplete.³⁷ Visualizing, from a bone marrow failure perspective, severity, regenerative capacity, mutations, and leukemic potential invites questions as to empirical prognostic factors, mechanisms, and commonalities of effective therapies (Figure 3C).

Another perspective might be pathologic somatic mutations, familiar in malignancy as leading to dysregulated proliferation, blocked differentiation, and resistance to apoptosis, but also etiologic in “benign” yet serious diseases across medical specialties, affecting cell function but not causing cancerous growth.⁹⁵ We have limited knowledge of risk factors for the occurrence and tolerance of mutations, from the external environmental to intrinsic intracellular constraints, and we seldom have insight into intermediate stages in evolution (exaptation); the ordering of somatic mutations in neoplasia (Figure 2D), that imply transient benefit, and contributing differently under altered environments (Figure 2C).^{66, 96} Clonal expansion is not random but driven by selection, but selection is not defined beyond a few instances, as in the setting of autoimmunity,^97–100 post-chemotherapy,¹⁰¹ and telomere length (Figure 2E). PPM1D somatic mutations hold different contextual meaning: in telomere biology disorders or age-related telomere attrition, PPM1D somatic mutations can reduce cellular senescence^{59, 60} while PPM1D somatic mutations can be selected post-chemotherapy for resistance to apoptosis and lead to therapy-related myeloid neoplasms and treatment resistance.¹⁰¹ In VEXAS syndrome, caused by UBA1 somatic mutations and originally misclassified as MDS due to the unusual appearance of their bone marrow cells, we do not understand the factors that provoke expansion and/or clinical manifestation, as we know individuals with detected UBA1 pathogenic variants are asymptomatic in population level data and why it typically manifests in the seventh decade of life (similar to splicing factors) when it is detected in the fifth decade.^{72, 102} Teleologic thinking—“superior” cell proliferation, self-renewal, and resistance to apoptosis—neglects the “blindness” of evolution, undirected to any particular end (like cancer, like humans), the nature of adaptation (loss of a function may be adaptive),⁹⁹ and the ephemerality of selective conditions (regeneration, infection, and physiochemical factors, as for TP53 loss¹⁰³ or PPM1D mutations¹⁰¹).

Empty acronyms might serve to describe the less well characterized entities, after removing separate diseases: (early) leukemia and stereotypical syndromes in MDS or the bulk of harmless aging mutations in CHIP (Figure 4A). A large proportion of MDS as currently conceived comprises discrete syndromes, often with stereotypical features: aneuploidy, chromosome 7 defects or del(5q); somatic mutations, SF3B1 in sideroblastic anemia and UBA1 in VEXAS syndrome^{95, 104, 105} or a combination, as in biallelic TP53 loss.^{106, 107} There may be common features also to syndromes resulting from genetic constraints imposed by a initiating mutation, fostering certain secondary mutations but physiologically incompatible with others,^{108, 109} such as spliceosome gene or cohesion gene mutations, TP53 mutated clones, or somatic clonal hematopoiesis in the setting of germline diseases (Figure 4B).¹¹⁰ Promotion of these subtypes to discrete diseases would allow higher resolution than prompted by current MDS classifications (Figure 4C), which are disadvantaged in their infrequent updates, relatively arbitrary labels, and impact of therapeutic interventions on outcomes. As a prominent example: VEXAS is dominated by severe inflammatory manifestations, but in medical textbooks it is listed as a variant of MDS; the usually mildly dysmorphic marrow infrequently undergoes leukemic transformation. Focus on individual disease entities encourages consideration of their particular clinical trajectories over time, especially treatment response as an intrinsic feature of a syndrome (revealing “hidden” mechanisms), and allows a lexicon unfettered by the constraints of existing classifications and subclassifications.

Figure 4. Reframing perspectives of nosology.

A. The simplest, and likely most acceptable approach, subtracts well defined disease entities from vague or inaccurate comprehensive aggregates. Individual diseases that are clearly defined are easier to understand and explain than when buried within a complicated taxonomy, and encourage comparisons with other pathophysiologies. Less well characterized syndromes require different approaches in the clinic and the research laboratory. Most obviously, AML (at an earlier stage of development, but nevertheless leukemia), is removed from MDS, as morphologic features inadequately account for heterogenous clinical phenotypes and outcomes;^{6 1} conversely, “benign” syndromes like sideroblastic anemias, del(5q) and VEXAS are distinct diseases with low leukemia risk and featuring their own clinical manifestations. “MDS” can be employed as an empty acronym, containing the remaining more poorly characterized entities. The vast majority of CHIP is not of indeterminate potential and is age-related clonal hematopoiesis (ARCH) of no clinical significance. The area of the circles and pie slices are to scale.⁶ In the CHIP pie, the unlabeled slices (left to right) are: SRSF2, JAK2, SF3B1, Other, and HR CHIP. B. Sets are useful employing in medicine, to visualize relationships and manifestations. A Venn diagram illustrates overlap and stereotypical cases and extreme instances. Much of MDS, CCUS, and CHIP can be incorporated into the large clusters of primarily malignant or mainly bone marrow failure. For instance, SF3B1 sideroblastic anemia is predominantly a bone marrow failure disorder, but secondary mutations (such as in RUNX1) occur in a subpopulation of cases that favor leukemia development.¹⁰⁴ C. A hierarchy can be envisioned for diseases based on somatic mutations that lead to clonality. While there are limitations to the organ system model commonly employed in clinical practice, pathophysiologies are often shared across medicine. Malignancy mostly features proliferation, resistance to cell death, and dysregulated growth; “benign” disease in the hematopoietic stem and progenitor cell (HSPC) has different paradigms, hematopoietic malfunction or dysfunction. Certain clonal diseases are a response to immune-mediated mechanisms, to which treatment should be directed (immune aplastic anemia and hMDS with immunosuppression; PNH treated with complement inhibition). In others, clones exert autoimmune (somatic autoimmune lymphoproliferative syndrome [ALPS]) or autoinflammatory (VEXAS, TET2-opathy) effects on extrahematopoietic tissues as well as affecting blood cell production, and treatments include modifying the behavior or a clone (sirolimus in ALPS), ameliorating the multisystem effects of a clone (anti-IL6 in VEXAS), or elimination of a clone (transplant). Some mutations lead to clonality and impair differentiation and hematopoiesis, producing syndromes labeled as “MDS” in addition to widely varying proclivity to leukemic evolution, ameliorative therapies also are different (SF3B1 sideroblastic anemia with luspatercept, del(5q) with lenalidomide). Abbreviations: AML, acute myeloid leukemia. CCUS, clonal cytopenia of undetermined significance. CHIP, clonal hematopoiesis of indeterminate potential. DBA, Diamond Blackfan anemia. FA, Fanconi anemia. hMDS, hypoplastic MDS. HR, high risk. HSPC, hematopoietic stem and progenitor cell. iAA, immune aplastic anemia. MDS, myelodysplastic syndrome/neoplasm. MN, myeloid neoplasm. NOS, not otherwise specified. PNH, paroxysmal nocturnal hemoglobinuria. PRC, polycomb repressive complex. SDS, Shwachman Diamond syndrome. TBD, telomere biology disorder. VEXAS, vacuoles, E1 enzyme, X-linked, autoinflammatory, somatic.

Similarly for CHIP, if we recast the high-risk subset as an unusual “chronic AML,” and remove a mere 1% of patients, what remains (Figure 4A)? Low-risk CHIP constitutes the vast majority of individuals who are not at significant risk of any myeloid neoplasia—let alone leukemia. Their malignancy risk is an order of magnitude lower than individuals with MGUS (Figure 1C): CH in this low-risk population is no longer of indeterminate potential but determinedly of no potential.

Last and distinct from the focus on myeloid neoplasia progression in the field is a potential reframing of a set of CHIP that has causal (or clinical) significance rather than indeterminate potential, akin to monoclonal gammopathy of causal significance, and its placement on a continuum of plasma cell dyscrasias.¹⁰ An analogous perspective would frame certain clonal hematopoiesis as non-malignant, although not benign, and causative in non-hematologic disease(s) due to the altered functions of the somatically mutated HSPCs and their progeny. Entities could include tumor infiltrating clonal hematopoiesis (TI-CH), affecting a quarter of patients with CHIP and solid tumors, which is associated with worse clinical outcomes and a causal link demonstrated in organoid models of TET2-mutated myeloid cells remodeling the tumor microenvironment and facilitating tumor growth.¹¹¹ Another potential entity could be JAK2^V617F CH associated with arterial vascular disease and venous thromboembolism in population studies and platelet activation in animal models,^{25, 112, 113} highlighting a potential causal role that invites attempts at intervention with clone-directed therapy, antiplatelets, or anticoagulation (as for anticomplement and/or anticoagulation when PNH when clone size >50%¹¹⁴). Plasma cells and their product immunoglobulins, for now, are easily monitored, but so eventually will be specifically mutated clones, their variant allele frequency, and subclones constituting architecture and evolution.

Practical implications.

The medical definition of disease is more pragmatic than categorical, far less strict than Wikipedia’s “any deviation from normal.” Physicians recognize diseases from their training, textbook chapter headings and subspecialty traditions; by “family resemblance” and gestalt; and, sometimes, by pathophysiology and the need and availability of treatment. Not every abnormality is a disease, nor is it necessarily clear when a predisposition, proclivity, or combination of aberrant physiologies or risk factors substantiate a disease. Disease names lack a taxonomy, sometimes based on an etiology, pathophysiology, manifestations, or historic or geographic eponym. Our proposed nosologic schemas are meant to be heuristic, to lead to a different perspective(s) and practical improvement in diagnosis and institution (or withholding) of treatment.

In the hematology clinic, abandoning outdated, limited, and increasingly complicated and confusing diagnostic formulas would provide clarity and perhaps simplicity, as patients and physicians would confront a diagnosis of a specific disease (Figure 4C). For MDS, as occurs at present, del(5q) would imply a favorable prognosis and lenalidomide response; in the future, IDH inhibitors would provide an opportunity to rescue or prophylax specific leukemias in progress or predicted. Further genomic delineation of leukemic disease, as is underway, would allow physicians to focus on intensive chemotherapy and transplant planning.^{115, 116} The focus in VEXAS should be the inflammation, and its treatments, which are overwhelmingly responsible for the disease-specific morbidity and mortality, and not unusual cells in the marrow and usually modest blood count depression. For the large “refractory anemia” MDS category, a reorganized nosology might encourage better use of effective agents, especially underutilized immunosuppression regimens and other forms of immunomodulation,⁹⁴ and the diminished emphasis on leukemia ease the testing of thrombopoietin mimetics in patients suffering from cytopenias. Further, a separation of “refractory anemia” MDS from SF3B1 sideroblastic anemia in clinical trials could provide clarity to physicians and patients on the likely response to therapy, as low-risk MDS trials enrolling a large proportion of SF3B1 patients makes interpretation of results difficult.^{117, 118} For CHIP, almost all patients would be reassured that their findings are incidental—and do not need “monitoring.”

Research Strategies.

In clinical research, protocols already are developed for targeted therapy, not for “MDS” but for specific syndromes defined by their molecular lesions, clinical features, and natural history. Here, complexity might be welcome, using causal inference modeling to determine not only the pathways that might be interrupted but the modifying or converging factors that affect responsiveness to a trial drug. In CHIP, whether specialized clinics are necessary, for patient and for hematological outcomes, is questionable, but replacement of a “high risk” group with an extraordinarily slowly developing type of leukemia invites efforts at early intervention, and deepening our understanding with well-designed, prospective databases with serial sequencing, to elucidate the interaction of clones with their environment.⁹²

What accounts for the apparent extraordinary range of kinetics in myeloid leukemias, from months to years? Can clonal architecture be a guide to the evolution of more aggressive and clinically manifest disease? At what point do the most sensitive assays for miniscule clones assume predictive importance, or conversely represent only background noise? Is there a reliable, diagnostic threshold for minimal “initial” disease to complement measurable residual disease after treatment?¹¹⁹ Can we better relate the network of interactions among cells in the marrow, especially immune surveillance,¹²⁰ and the complexity within cells, as for example commutated clones and competing clones, and further determine the role of a multiplicity of germline variants, beyond the rare predisposition syndromes? What are the explanations for “malignant clonal evolution,” myeloid cancers that arise from immune marrow failure and in the (significant) minority of low risk MDS? Conversely, what biologic processes limit the growth of clones almost universally in CHIP and the aging population in general? Research paradigm shifts can derive from new language for these diseases, and speed the transition from the laboratory “handbook” to the medical “textbook.”^{121, 122}

Despite approval of new drugs for MDS, the diagnosis remains highly unfavorable and anxiety provoking, even in so called low risk patients, and treatments other than stem cell transplant are ameliorative, modestly impacting life expectancy and quality of life. Decisions for therapy are often complicated and patients are frustrated. Reframing CHIP, unsettling as it might be, should clarify what actions in the clinic are mandated, which patients or “patients” do not require treatment and in whom aggressive treatment is indicated early, and promote novel clinical research. Not all the basic biology questions raised above will be answered, and even if answered in the laboratory will not necessarily translate to treatment. But our queries need to be better formulated and less confusing to envision solutions: clarity in our thinking of these entities and the terms we use to describe them should ease conversations in the physician’s office and “MDS” and “CHIP” clinics, and our understanding of disease origins and curative strategies.