Visual Abstract
TO THE EDITOR:
Acute myeloid leukemia (AML) is a high-risk disease with inadequate treatments and suboptimal outcomes, with less than a third of patients surviving at 5 years.1 Myeloid sarcomas (MSs) are extramedullary tumors of myeloid blasts disrupting normal tissue architecture, with a highly variable frequency.2,3 MSs are associated with driver oncogenes such as NPM1, MLL rearrangements, and inv(16).4 The prognosis of AML in patients with MS is unclear,5 although some studies suggest that MS is associated with worse prognosis.6 In the allogeneic hematopoietic stem cell transplant (HSCT) setting, MS can escape the intended graft-versus-leukemia effect, potentially from immune-based resistance.7,8 Understanding the molecular underpinnings of MS and how it differs from medullary disease will identify potential therapeutic strategies that may improve outcomes.
Patient characteristics are shown in supplemental Table 1. Bulk RNA sequencing (RNA-seq) from formalin-fixed, paraffin-embedded (FFPE) samples of 5 paired MSs and marrow samples demonstrated differential gene expression, with the majority of genes more highly expressed in the MS vs bone marrow (BM) samples (supplemental Table 2). Gene Set Enrichment Analysis (GSEA) analysis9 was performed using Hallmark Molecular Signatures Database (MSigDB) and identified 17 gene sets significantly enriched in MS (supplemental Table 3); the most highly enriched gene set was epithelial mesenchymal transition (EMT; Figure 1A).
To identify transcription factors that may drive EMT, we mined our RNA-seq and identified TWIST1 messenger RNA (mRNA) as the most differentially expressed EMT-associated transcription factor in the MS vs marrow with log2fold change 6.257 (P-adjusted [Padj] < .001; Figure 1B). SNAI2 also demonstrated increased expression in the MS with log2fold change 3.980 (Padj < .016). Independent samples were processed for histology and demonstrated disrupted tissue architecture by myeloid blast cells and surrounding stromal tissue (Figure 1C). Immunohistochemistry demonstrated TWIST1 expression in MS and BM AML, with higher expression in MS (Figure 1C; supplemental Figure 1; supplemental Table 6). This suggests that in samples from patients with AML, there is likely a TWIST1-driven EMT-like signature promoting extramedullary disease.
Given the genetic and anatomic heterogeneity in our samples, we sought a mouse model to confirm key findings. Our previous work10,11 used a conditional allele of Npm1flox-cA/+, which develops MS (Figure 2A-B). We isolated RNA from both fresh and FFPE MS and BM AML specimens to assess whether a signature similar to humans would be seen using a single oncogenic driver. In fresh tissue, the top hit was an EMT-like signature (Figure 2C), with increased expression of TWIST1 in MS (Figure 2D; supplemental Table 4). In FFPE tissues, the same EMT-like signature was the second-most enriched gene set (supplemental Figure 2), indicating that differences between fresh and FFPE-derived samples did not affect the enrichment for an EMT-like signature (supplemental Table 5). mRNA for the canonical EMT transcription factor Twist1 was also significantly differentially expressed with log2fold change 4.86 (Padj = .01) in the fresh mouse samples. We next hypothesized that the AML cells, which are cKit+, would show reduced expression in the BM compared with cKit– cells, whereas in MS, we would see the opposite pattern. To address this, we transplanted AML cells into lethally irradiated hosts, and when they became moribund, we euthanized them, harvested both BM and MS, fractionated the cells based upon cKit (Figure 2E), and demonstrated the expected pattern. The mouse studies confirmed an EMT-like signature in patient-derived samples and a mouse model, irrespective of tissue preparation method. These results suggest that this EMT-like signature and TWIST1 expression are potentially critical to MS development.
Although our previous observations indicated that BM from moribund Npm1cA animals were capable of developing leukemia and MS in secondary transplantation (Meyer et al10 and Vassiliou et al11 and data not shown), it was unclear whether the MSs themselves displayed plasticity and could generate leukemia in secondary transplants. To evaluate the plasticity of the MS phenotype, a single-cell suspension of MS cells from an isolated murine tumor was transplanted into sublethally irradiated mice, and recipients became moribund 3 to 4 weeks later. All recipients (4/4) demonstrated >3 separate MSs in axilla and subcutaneous tissue, circulating peripheral blast cells, and evidence of marrow disease on the evaluable sample (data not shown), along with hepatosplenomegaly. This demonstrated that MS cells could revert to a medullary disease phenotype upon transplantation and establish distant tumor sites, suggesting cell-intrinsic plasticity. This suggests as well that an isolated tumor does not preclude reversion back to a more systemic disease process.
Although MS is a well-described clinical entity in AML, its effects on prognosis and molecular underpinning are unknown. Using a combination of patient-derived samples and an animal model, we demonstrate that MS samples display an EMT-like signature. We refer to the signature as “EMT-like” because, unlike classic EMT, which occurs normally in development or during metastatic spread of carcinomas,12 AML is a mesodermal derivative. Nonetheless, we consistently found an enrichment of an EMT-like signature, regardless of species or sample source. In addition, we were able to confirm increased levels of TWIST1 mRNA and protein in human samples, potentially indicating it as a critical driver of MS development. Importantly, verification that TWIST1 alone is driving the EMT-like signature and is necessary for MS development are required.
Our MS transplantation data suggest that the leukemia cells are capable of changing their phenotype because the transplanted MS cells generate both intramedullary and extramedullary disease, consistent with cell-intrinsic plasticity. This is supported by the fact that patients with MS benefit from systemic therapy regardless of BM involvement.13 Although NPM1c and TET214 deleterious mutations are clinically associated with MS, it is unclear whether TET2 loss promotes a similar EMT-like signature, which will require additional studies to uncover.
Our RNA-seq analysis did not reveal any clear somatic mutational drivers of the MS phenotype (data not shown), but care must be taken not to overinterpret this finding because FFPE-derived RNA-seq may not provide adequate coverage. In addition, given the limited number of matched samples (5) with 4 of 5 specimens representing leukemia cutis, our patient-derived data may be biased toward a single anatomic location. Finally, we were unable to correct for the degree of tumor infiltration in the tissue, which may have skewed the results toward only the most highly enriched genes or gene sets. Given the agreement with our mouse model, in which the MSs were not leukemia cutis, we suspect that the EMT-like signature is broadly present independent of MS location, but additional studies with higher resolution approaches are needed to identify potential somatic mutations as well as site-specific gene expression changes.
Although MS is typically understudied clinically, in the allogeneic HSCT setting, it remains a clinical challenge, and anecdotal evidence suggests that relapse after HSCT can be the result of MS alone, in the absence of BM disease. This would potentially suggest that MS endorses an immunosuppressive microenvironment, protecting MS cells from a graft-versus-leukemia effect. An EMT-like signature would be consistent with an immunosuppressive microenvironment.15 In epithelial-derived cancers, Transforming Growth Factor β (TGFb) is a critical upstream mediator of EMT, with a known role as an immunosuppressive cytokine.16 In our own data, we noted increased expression in the MS of the receptor ligand pair TGFb1:TGFbR3 (supplemental Table 3), which could indicate a mechanistic link in the development of an immunosuppressive microenvironment in MS. Future studies are needed to better define this mechanism and identify potential therapeutic vulnerabilities.
Acknowledgments: This work was supported by grants from the National Cancer Institute (R01CA204231), National Heart, Lung, and Blood Institute (P01HL14962), and National Institute of Digestive and Diabetes and Kidney Diseases (Dk134064) [S.R.].
Contribution: K.E.Z., A.E.M., E.A., and S.R. conceived and designed the study; K.E.Z., J.A.-A., A.E.M., A.C., M.H., E.G., and C.S. performed the research; and G.V. provided an essential reagent (Npm1flox-cA/+) and assisted with editing of the manuscript.
Conflict-of-interest disclosure: The authors declare no competing financial interests.
The current affiliation for K.E.Z. is Center for Cancer and Blood Disorders, Phoenix Children’s Hospital, Phoenix, AZ.
Correspondence: Sridhar Rao, Versiti Blood Research Institute, 8733 West Watertown Plank Rd, Milwaukee, WI 53266; email: [email protected].
References
Author notes
K.E.Z. and J.A.-A. contributed equally to this study
Mouse-derived data are available through the Gene Expression Omnibus database (accession number GSE264700).
Processed human data files are available on request from the corresponding author, Sridhar Rao ([email protected]).
The full-text version of this article contains a data supplement.