When clinical heterogeneity exceeds genetic heterogeneity: thinking outside the genomic box in chronic myelomonocytic leukemia

Chronic myelomonocytic leukemia (CMML) is a lethal myeloid neoplasm hallmarked by clonal peripheral monocytosis and ineffective hematopoiesis accompanied by bone marrow dysplasia. CMML has long been classified as a subtype of the myelodysplastic syndromes (MDSs) until 2001 when the World Health Organization (WHO) consolidated a group of diseases, including CMML, into a new category known as myelodysplastic/myeloproliferative neoplasms (MDS/MPNs).^1,2  These diseases lie at the interface of WHO-defined myeloid neoplasms because they have pathoclinical features of both MDS and MPNs. Although CMML is broadly classified as MDS-CMML and MPN-CMML based on the presence of leukocytosis (defined as >13 × 10³ cells per dL), the clinical presentation of an individual CMML patient can be much more heterogeneous.³  Patients may present with varied constellations of MDS and MPN symptomatology making CMML an extremely diverse clinical entity. Further, 30% of cases transform to acute myeloid leukemia (AML), and nearly all cases are refractory or relapse to standard therapies.⁴ 

Recent preclinical studies have begun to elucidate the molecular underpinnings of CMML. Genomic analysis, including 2 exome sequencing studies, identified 3 highly recurrent mutated genes (ASXL1, TET2, and SRSF2) along with other mutated genes that implicate alternative pre–messenger RNA splicing, epigenetic modifications, and cytokine signaling pathways as key components of CMML molecular pathology.^5-10  Despite the extremely diverse clinical presentations and refractoriness to standard therapies, we now know that the mutatome of CMML, similar to that of AML, contains a relatively small number of somatic mutations compared with other leukemias and adult cancers. Further, the ratio of clinical heterogeneity to genetic heterogeneity in CMML is even higher than that of AML given the lower number of somatic mutations per exome and more diverse clinical symptomatology seen in the latter.

Unlike chronic myeloid leukemia, it could be argued that the clinical course of CMML and AML would not have predicted this molecular finding. AML is a proliferative neoplasm whose clinical course is characterized by relapsed disease after successful suppression of chemosensitive clones suggesting a large clonal and mutational diversity. The fact that AML exomes harbor ∼10 somatic mutations per megabase, logarithmically lower than that seen in melanoma and lung cancer, remains surprising to many clinicians who have experienced unrelenting therapeutic disappointments in the clinic over the course of decades. An optimistic view could postulate that a relatively low mutation burden in CMML and AML reflects the presence of tractable molecular vulnerabilities to genomically targeted therapies, but recent history argues against this. For example, BRAF inhibitors have been associated with dramatic responses in patients with metastatic melanoma deemed a “Lazarus effect” in some.¹¹  Although targeted therapies for AML such as FLT3, CKIT, and IDH inhibitors have promise, their clinical impact has been comparatively less apparent.^12-15 

The restricted genetic heterogeneity balanced by profound clinical diversity present in CMML emphasizes the need to delineate other molecular modifiers. Herein, we will discuss the clinical and genetic heterogeneity of CMML, explore the case for expanded and alternative modes of genomic profiling, and summarize recent nongenomic preclinical findings with the potential for rapid translation to the clinic. The goal of this review is not to provide a comprehensive overview of CMML pathogenesis, but instead to highlight nongenomic molecular discoveries that may offer further insight into the clinical heterogeneity of CMML and provide opportunities to improve CMML patient care.

Clinical diversity is appreciated at all levels for CMML ranging from pathologic diagnosis, patient symptomatology, natural history, and response to therapy, making it a uniquely heterogeneous leukemia. The peripheral blood findings in patients with CMML may include leukocytosis with a monocyte or neutrophil predominance (while maintaining a white blood cell count >1 × 10³ cells per dL). Leukopenia and neutropenia are less frequently appreciated in CMML, but macrocytic anemia and/or thrombocytopenia are common.⁴  In some cases, thrombocytopenia can be immune related, whereas in others reflecting the underlying disease process.¹⁶  The bone marrow aspirate often demonstrates dysplasia in any or all hematopoietic lineages. However, dysplasia is often minimal and not necessary for the diagnosis so long as clonal, persistent monocytosis is evident.²  Hypercellularity is universal, but reticulin fibrosis, ring sideroblasts (associated with the presence of SF3B1 mutations),^17-19  and concomitant mast cell infiltrates are seen in a minority of cases.²⁰  Myeloblast percentages can range from 1% to 19% and are highly prognostic.

A minority of patients are identified following incidental findings of abnormal peripheral counts, whereas others present with constitutional symptoms that may include profound fatigue, night sweats, and fever similar to that of MPNs.²¹  Symptoms related to cytopenias are common and spleen related complaints occur in roughly 25% of cases.⁴  Rarely, CMML cases can present with hyperleukocytosis, which requires urgent therapy and hospitalization. Myelomonocytic infiltrates may also occur involving the skin, lymph nodes, and other extramedullary sites.^22,23  Although the majority of patients die of their disease, many patients have a quiescent prodrome that can last months to years, which cannot be fully predicted by mutation profiling.^24,25  Together, these presentations can occur in varied combinations, further expanding the diversity in clinical presentation.

For the purposes of this review, we propose 2 definitions of genomic heterogeneity in CMML. One, which has been published in landmark genomic studies across cancers, is defined as the average number of somatic missense mutations per exome for any given disease. An alternative distinction is the number of recurrently mutated genes above a certain percentage among specific disease studies. In Figure 1, we aggregate data from Lawrence et al²⁶  and Merlevede et al⁵  to illustrate that the number of somatic mutations per exome are comparable among MDS, AML, and CMML, thereby establishing them among the most mutationally homogenous adult cancers. Further, analysis of the most common recurrent mutations demonstrates that the mutational landscape of AML is characterized by several recurrent mutations above a frequency of 10%. However, CMML is characterized by a more restricted set of mutated genes, albeit at a higher frequency suggesting that the number of highly recurrent genes mutated in CMML is less than that of their myeloid counterparts. Recent exome sequencing studies confirm the restricted mutatome of CMML. Both Merlevede et al and Mason et al identified somatic gene mutations previously reported in targeted sequencing studies.^5,10  They additionally confirm the presence of other previously identified genes in ETNNK and SETBP1; however, no additional mutations were identified with a frequency exceeding 10%.^27,28 

Several studies contoured our knowledge of CMML genomics. Among the earliest studies, Kohlmann et al profiled 81 CMML patients and identified TET2, CBL, RAS, and/or RUNX1 gene mutations in >70% of cases, demonstrating the genetic homogeny of CMML.²⁹  A subsequent landmark study by Yoshida et al identified mutations in genes associated with alternative splicing to be among the most frequent mutations in myeloid neoplasms.⁸  Of interest, patients with CMML had enrichment of mutations in the RNA splicing gene SRSF2, which has been linked to peripheral monocytosis in several studies.¹⁷  Subsequent sequencing studies showed that by restricting sequencing to only 9 genes with the inclusion of SRSF2, one can identify a clonal abnormality in >90% of CMML cases.⁷  Additional studies have explored the clinical relevance of these gene mutations in CMML. At least 3 studies have identified frame shift or nonsense mutations in ASXL1 as the only gene with prognostic relevance in CMML.^4,6,24  A recent study identified RUNX1, NRAS, and SETBP1 as additional somatic gene mutations with prognostic relevance.³⁰  Combinations of mutations appear to have specificity for monocytosis and are highly enriched in CMML. For example, comutation of TET2 and SRSF2 was associated with monocytosis in >90% of cases in on series.³¹ 

Juvenile myelomonocytic leukemia (JMML) is a rare pediatric myeloid neoplasm that is phenotypically comparable to proliferative CMML.³²  Notably, the genomic landscape of JMML is well described and distinct from that of its adult counterpart^33,34  (Figure 1A). Although both malignancies have mutations that upregulate the RAS pathway, these are seen in ∼90% of JMML and at a much lower frequency in CMML. CMML cases have been demonstrated to have mutations in genes regulating RNA splicing and epigenetics in well over 50% of cases, whereas these are rare in JMML. Together, this comparison illustrates diseases with convergent clinical phenotypes that manifest via distinct mutational landscapes.

Despite a relatively homogeneous mutatome at presentation, recent studies in CMML and AML suggest that additional genomic discovery affords greater biologically and clinically informative data. One approach to expanding genomic profiling may center on sequencing a large, clinically well-defined CMML cohort with the goal of identifying recurrent, diagnostically informative genetic combinations. Diagnostic classification of myeloid neoplasms has begun to move from a focus on morphologic classification to inclusion of mutationally defined entities with pathogenetic relevance.²  For example, a recent study in AML sequenced 111 genes from 1540 patient specimens in well-defined clinical trial cohorts testing intensive therapy approaches. Unsupervised cluster analysis of somatic mutations identified 11 distinct genetic subgroups with unique gene-gene and complex gene interactions associated with clinical outcome, partially accounting for the clinical diversity in AML.³⁵  Although the incidence of CMML is 10-fold lower than that of AML (0.4 vs 4 per 100 000), a similar study may be feasible in CMML, perhaps leveraging the fact that CMML cases have decreased genetic diversity compared with AML.^36,37  Such studies would be especially relevant in CMML given the dynamic classification systems currently present (French-American-British [FAB] vs WHO vs Prognostic Scoring Systems) as it may resolve or consolidate ambiguous morphologic definitions of CMML and MDS/MPNs with more definitive, genetic lesions as a backbone for ontogeny.

A second approach may center on sequential sequencing studies with a particular focus of changes at relapse and at transformation in an attempt to identify novel genetic events that are acquired and drive clinical consequences. Several studies support the notion that mutational architecture is distinct at relapse in AML and that the loss of mutations after primary therapy may have positive clinical ramifications.³⁸  For example, 1 study recently demonstrated that AML cases achieving a complete molecular response in preexisting somatic mutations after induction therapy had a clinically significant decrease in relapse-free survival.³⁹  Similar studies have begun in CMML testing patients treated with azanucleosides, the de facto standard of care. Merlevede et al have sequenced patients before and after azanucleosides and, surprisingly, identified no clear difference in mutational architecture at the time of response suggesting a noncytotoxic mechanism of action in CMML.⁵  However, patients sequenced at disease progression have clear changes in mutational architecture consistent with the clinical observation that postazanucleoside therapy CMML cases are chemorefractory with a survival measured in months.⁴⁰  Finally, a third approach may be to leverage single-cell sequencing to identify clonal heterogeneity obscured by conventional bulk leukemia sequencing. Although whole genome sequencing can faithfully infer clonal architecture compared with single cell sequencing, it has been hypothesized that exome and targeted panel sequencing, which represent the vast majority of sequencing experiments performed in CMML, grossly underestimate the true genetic diversity in any given sample. This concept is exemplified by a recent study in AML where Paguirigan et al genotyped AML single cells derived from 6 patients (∼705 cells per patient) known to harbor mutations in NPM1 and FLT3. Although all cases were predicted to have 1 clone containing both heterozygous mutations of NPM1 and FLT3 via bulk sequencing, single cell genotyping of only those 2 genes revealed 9 distinct clonal populations in each patient suggesting that bulk sequencing grossly underestimates genetic diversity.⁴¹  Such a study in CMML could uncover unrecognized genetic diversity overlooked by conventional sequencing, further aligning with its clinical heterogeneity.

Among the most striking preclinical evidence favoring nongenomic molecular features relevant to CMML heterogeneity is that learned from monogeneic mouse models (Table 1). Several informative mouse models based on genetic lesions that recapitulate many feature of human CMML include TET2 deletion, NRAS, and CBL mutation.^42-44  However, 2 mouse models based on the most commonly mutated genes in CMML (ASXL1 and SRSF2) do not recapitulate the disease and are more consistent with a bone marrow failure syndrome.^45,46  Further, several mouse models unexpectedly develop a CMML phenotype by perturbing long non coding RNAs (lncRNAs), tyrosine kinases, or nonmutated epigenetic modifiers, perhaps supporting alternative modes of molecular pathogenesis. Mouse models with deregulation of NOTCH, XIST, CREBBP, TAK1, and NR4A are just some of these examples.^47-50  Subsequently, we highlight recent preclinical evidence arising from nongenomic modalities that offer insight into the clinical diversity in CMML.

Initial CMML genomic studies identified recurrent somatic mutations in TET2, IDH2, and DNMT3A, as well as ASXL1 and EZH2, key modifiers of DNA methylation and histone modification, respectively.^6-9  These mutations lead to genotype-specific transcriptome-wide changes by globally augmenting epigenetic regulatory mechanisms.^51,52  However, given the high frequency of epigenetic gene mutations in CMML and other myeloid neoplasms, some have hypothesized that there may exist mutation-independent mechanisms of epigenetic dysregulation. To that end, Li et al have recently reported in AML that the presence and dynamics of “epi-alleles” (a measure of epigenetic heterogeneity) behave discordantly with the presence or type of somatic mutations. Epi-alleles also contain information relevant to clinical prognosis that are independent of mutation status.⁵³  Although what drives the underlying epigenetic heterogeneity is unclear, it appears to be contributory to the molecular heterogeneity in AML and should be explored in CMML.

Meldi et al have recently reported 1 clinically practical application of mutation-independent epigenetic heterogeneity. In this report, an epigenetic signature was identified that was highly predictive of response to the azanucleoside decitabine. The signature was derived from enhanced reduced representation bisulfite sequencing that permitted investigators to annotate the epigenetic marks in promoters and nonpromoter regions of responding and nonresponding CMML patients. Using this approach, they identified 167 differentially methylated regions that were robustly capable of discriminating responders from nonresponders. These findings were validated in an external data set, which demonstrated no association to somatic mutation but, rather, was associated with the expression of the inflammatory proteins CXCL4 and CXCL7.⁵⁴ 

Several lines of evidence implicate inflammatory, cell extrinsic pathways in the pathogenesis and perhaps clinical heterogeneity of CMML. CMML is epidemiologically linked to autoimmune disorders such as immune thrombocytopenic purpura and Hashimoto’s thyroiditis suggesting convergent pathophysiology.¹⁶  Moreover, paraneoplastic syndromes that clinically resemble idiopathic autoimmune disorders are reported to accompany CMML at a higher frequency compared with MDS highlighting underlying inflammatory pathophysiology. Interestingly, these phenomenon have been reported to improve with CMML therapy.^55-59  Pathologically, a recent study by Solary et al demonstrated that the characteristic monocytosis is, in fact, an expansion of classical monocytes with reduction of other nonclassical monocytic subtypes.⁶⁰  This phenotype was present across all patients studied and was capable of distinguishing CMML from other entities with benign causes of monocytosis.⁶⁰  Classical monocytes represent the inflammatory monocytic subset and are preferentially recruited to sites of tissue injury or infection via the CCR2-CCL2 axis, which is upregulated in CMML. The association between classical monocytosis and CMML is so robust that this assay is now being validated as a diagnostic modality in CMML.

Granulocyte-macrophage colony-stimulating factor (GM-CSF), a critical cytokine implicated in inflammation and myeloid differentiation is also associated with CMML pathology. We and others have demonstrated that CMML bone marrow mononuclear cells are hypersensitive to GM-CSF and require its presence for efficient engraftment in immune-compromised mice.^61-63  This concept has been tested clinically using ruxolitinib, a JAK 1/2 inhibitor in patients with CMML. The Janus Kinase 2 (JAK2) is the sentinel kinase involved in GM-CSF receptor signaling, and ruxolitinib potently downregulates inflammatory cytokines in other myeloid neoplasms.^64-66  In a phase 1 study in CMML, responses were associated with dramatic reduction in inflammatory cytokine generation.²¹  Although the inflammatory cytokine profiles of CMML have not been specifically categorized, investigations are ongoing that may influence the molecular and clinical heterogeneity.

Mutations in genes encoding splicing proteins constitute the largest fraction of recurrently mutated genes in CMML. Approximately 40% of patients harbor mutations in SRSF2, and 10% to 15% of patients harbor mutations in SF3B1 or ZRSR2.⁸  Recent work has demonstrated that point mutations in SRSF2 elicit transcriptome-wide aberrancies resulting from augmentation of the consensus RNA recognition motif that results in several misspliced transcripts. Phenotypically, the SRSF2 murine model (Table 1) partially recapitulates the CMML phenotype as it results in cytopenias without monocytosis.^45,67,68  Although much remains to be investigated regarding the role of splicing aberrancies in myeloid malignancy, lessons from epigenetic leukemia research form the foundation for questions that could shed light on yet additional layers of nongenomic molecular heterogeneity as a result of RNA splicing abnormalities. For example, are there mutation-independent mechanisms of alternative RNA splicing? Do alternative transcripts contribute to clinical diversity, and can this be harnessed for therapeutic purposes? This last question has been partially addressed in elegant work reported by Lee et al.⁶⁹ 

Genomic studies have shown that mutations in splicing genes are faithfully mutually exclusive. Two hypotheses have been proposed that could potentially explain this observation. One suggests that multiple splicing mutations were evolutionary redundant, and therefore, cells containing this combination was not selected during disease initiation. A second hypothesis postulates that multiple splicing mutations are not tolerated and are therefore lethal to clones acquiring a second splicing mutation. Leveraging models of SRSF2 and SF3B1 mutation, the latter hypothesis was tested by genetically deleting the wild-type allele of a SRSF2^P95H/WT in a murine model or wild-type controls.⁶⁹  This approach indeed identified a genotype-specific synthetic lethality in SRSF2 (and other splicing mutations) mutant mice that may lead to a potential therapeutic window for inhibition of splicing in SRSF2-mutated CMML. This concept has been successfully preclinically tested with pladienolide derivatives that bind to the SF3B complex augmenting spliceasomal function. This class of splicing inhibitor is in clinical development (H3B 8800) paving the way for a genotype-specific clinical trial.

Although comprehensive mutational analysis has identified previously unknown mechanisms governing CMML pathogenesis, the molecular features associated with CMML clinical heterogeneity have not been fully elucidated. Somatic gene mutations in recurrently mutated genes have identified epigenetic, RNA splicing, and cytokine signaling as critical pathways with therapeutic vulnerabilities to exploit in future CMML clinical trials. Clinical association studies identified specific mutations, such as ASXL1, as independently prognostic and fostered the application of next-generation sequencing panels in the clinic.^6,70  However, the number of recurrent somatic mutations in CMML is far lower than solid tumors and perhaps even lower than that observed in other myeloid neoplasms. This and other lines of evidence reviewed here should remind us that other molecular pathogeneses should be translated with the same fervor as that observed with somatic genetic profiling.

Although fundamentally guided by somatic genetics, other areas of molecular heterogeneity provide great promise in eradicating CMML. Recent studies exploring epigenomic, inflammatory, and alternative RNA splicing heterogeneity, when integrated with genetic data, provide the most comprehensive platform to understand the breadth of molecular heterogeneity in CMML. For example, several studies suggest that cytokines can augment alternative splicing without the presence of somatic mutations in splicing genes.^71,72  Other studies have implicated mutations in genes that were thought to solely regulate glucose metabolism (IDH1/2) in the pathology of DNA methylation suggesting that integrating molecular data sets that profile both somatic DNA events and other molecular aberrancies will yield the most comprehensive understanding of CMML biology.⁵¹  It is these integrated investigations that are also likely to yield the most clinically meaningful discoveries. Multiple reports have recently identified promising signatures of azanucleoside response and refined the definition of monocytosis based on this approach, which merit validation and, if substantiated, full implementation in the clinic. “Bedside to bench” approaches may also offer insight into the molecular underpinnings of CMML heterogeneity as an increasing number of clinical trials for drugs that target the GM-CSF, RAS, and JAK pathways are being performed with informative correlative analysis. These studies and the formation of CMML consortia enabling collaborative molecular efforts have made it now possible to generate critical integrative data sets for future study.

Contribution: All authors wrote the manuscript and contributed to its scientific outline.

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

Correspondence: Eric Padron, Malignant Hematology Program, H. Lee Moffitt Cancer Center and Research Institute, 12902 Magnolia Dr, Tampa, FL 33612; e-mail: eric.padron@moffitt.org.