The Impact of Inflammation-Induced Tumor Plasticity during Myeloid Transformation

Hematopoiesis is a tightly regulated process where hematopoietic stem cells (HSC) differentiate into functional and mature subsets evenly, with the progressive plasticity to meet the needs of various physiologic demands such as regeneration and infections (1). Through aging, the acquisition of somatic mutations in hematologic malignancy–associated genes leads to the clonal expansion of progenitors, resulting in an indolent condition referred to as clonal hematopoiesis (CH) of indeterminate potential (CHIP), also known as age-related CH (ARCH; refs. 2–4). As a consequence, the mutant clone gains selective advantage, leading to an increased contribution of cell populations within the normal range (defined with a variant of allele frequency >2%; ref. 5). Interestingly, the rate of progression to a bona fide myeloid neoplasm in individuals with CHIP is only ∼1% per year (2), suggesting that additional contributing factors collaborate to promote neoplasia. Still, the specific events preventing or causing the progression to malignancy are mainly unknown.

The epigenetic regulator ten-eleven-translocation enzyme 2 (TET2) is among the most commonly mutated genes in both CHIP and myeloid malignancies. TET2 mutations occur in ∼30% to 50% of myelodysplastic syndromes (MDS), ∼13% of myeloproliferative neoplasms (MPN), and ∼10% to 30% of acute myeloid leukemia (AML) cases (6, 7). Together with TET1 and TET3, TET2 belongs to a family of proteins that comprise 2-oxoglutarate– and FE2(II)-dependent dioxygenase activity and catalyze the conversion of 5-methylcytosine (5mC) to 5-hydroxymethylcytosine (5hmC) and its iterative products (8). Murine modeling has shown that deletion of Tet2 is associated with an increase in HSC self-renewal, arrest in differentiation, myelomonocytic bias, and myeloid transformation (9–12). However, the precise function of TET2 in suppressing CH and leukemia remains mainly unknown.

To better characterize the molecular basis underlying the phenomenon of TET2-driven clonal evolution leading to aberrancies in hematopoiesis, we have generated a genetically engineered mouse strain carrying one of the most frequent TET2 missense mutations found in human myeloid neoplasms (p.H1881R) and individuals with CHIP. We leveraged single-cell technologies to delineate the transcriptional landscape of an aging-induced progression between an indolent state (similar to human CHIP) and bona fide myeloid malignancy. Surprisingly, our data demonstrate that an enhanced inflammatory response is sufficient to collaborate with Tet2 deficiency in promoting myeloid neoplasia. Specifically, our results reveal that inflammation transcriptionally rewires Tet2 mutant progenitors to differentiate into cells with the characteristics of inflammatory monocytes that egress the bone marrow to infiltrate peripheral tissues. Interestingly, our data suggest that the inflammatory transcriptional program expressed by these aberrant monocytes has importance beyond TET2-associated leukemogenesis, as its presence is strongly associated with poor prognosis in patients with AML. Together, our work provides novel insights into the mechanisms of inflammation-induced myeloid neoplasia, potentially relevant for the generation of early prognostic markers that may predict the progression to myeloid malignancy from a precursor CHIP state and for the development of target therapies.

In addition to nonsense/frameshift mutations, patients with CH/MDS/AML frequently present with a broad spectrum of TET2 missense mutations, most commonly within the two highly conserved regions comprising the catalytic domain (AA1140–1478 and AA1845–2002; Fig. 1A). The impact of these diverse mutations in the catalytic and noncatalytic functions of TET2 and malignant transformation is currently unclear, as the vast majority of studies utilize Tet2 loss-of-function models (9–11, 13). One of the most frequent missense mutations affects Histidine 1881, essential for coordinating the active site iron, a critical cofactor for TET2 catalytic activity (6). Previous and our own in vitro studies have suggested that the overexpression of TET2^H1881R in 293T cells leads to a decreased 5hmC compared with wild-type (WT), suggesting that this mutant is catalytic dead (ref. 8; Fig. 1B and C). To better characterize the effects of the H1881R mutation on the structure and function of TET2, molecular dynamics (MD) simulations (14) were carried out on WT TET2 and the TET2^H1881R mutant with three possible oxidized substrates: 5mC, 5hmC, and 5fC (15). Root mean squared fluctuations (RMSF), sugar puckering, correlated motions, and energy decomposition analyses (EDA) were calculated from the resulting trajectories. EDA provides a qualitative approach to investigate the change in average interaction energy over the calculated ensemble. This analysis showed a reduction in the total interaction (destabilization) of 10.1 (32.2, 10.5) kcal/mol for 5mC (5hmC, 5fC) for the H1881R mutant compared with WT TET2. The largest destabilization was observed for 5hmC, in which the substrate is stabilized overall in WT TET2 by 32.2 kcal/mol compared with the TET2^H1881R mutant. Several residues around the active site and beyond exhibit significant changes in the interaction between WT and H1881R TET2 (Fig. 1D). The addition of a positive charge drives the decrease in stability of 5hmC upon mutation at position 1881 and promotes differences in how the substrate base is oriented in the active site. By contrast, the H1881 residue becomes more stable upon mutation to Arg by 24.4 kcal/mol, owing to the favorable interaction between the positive charge on the residue and the DNA substrate, even in the presence of the divalent cation in the active site. The disruption of the active site alignment is reflected in the changes to the backbone puckering and delta and chi angles of the ribose on the substrate base (Fig. 1E–G), and in changes in RMSF and correlated motions of TET2 (SI), in which less correlated motions are observed for the cancer variant compared with WT. Further, the intermolecular interaction energy differences per residue of WT-5hmC and H1881R-5hmC (ΔE) presented in Fig. 1D show that R1261 significantly stabilizes WT by ∼18 kcal/mol, in agreement with our previous studies on TET2 oxidation using QM/MM modeling (16, 17). Additionally, liquid chromatography/tandem mass spectrometry (LC/MS-MS) analysis of replicate transfections using the H1881R mutant and controls (WT TET2, catalytically inactive TET2, and an empty vector) demonstrated that the H1881R mutant is inactive, with no detectable 5hmC/5fC/5caC generated upon expression (Fig. 1H). Together, these results underscore the importance of H1881 in the coordination of the divalent cation, as well as in the stability and orientation of the DNA substrate in the active site, suggesting that the TET2^H1881R mutant is unlikely to be able to catalyze the oxidation reaction on any of the oxidation states.

To better understand how such missense mutations of TET2 promote CH and delineate the molecular mechanisms inducing progression to malignancies, we generated a mouse model that expresses HuTET2 p.H1881R (Mm Tet2 H1794R) from the endogenous locus (referred to as HR; Supplementary Fig. S1A). RNA expression levels of the WT and HR allele were comparable in stem and progenitor Tet2^HR/+ (here referred to as Tet2^HR) cells [hematopoietic stem and progenitor cells (HSPC): lineage c-Kit⁺ cells; Fig. 2A]. Likewise, WT and Tet2^HR showed similar Tet2 protein levels, indicating that the mutation does not affect the protein stability (Fig. 2B). In contrast, the endogenous expression of the Tet2^HR mutant in one allele was sufficient to decrease the levels of 5-hydroxy-methylation of the DNA in these progenitor cells (Fig. 2C), supporting the MD predictions illustrated in Fig. 1. Tet2^HR and Tet2^HR/HR mice were born in the expected mendelian ratios and showed normal development. Similar to Tet2^KO cells (9), Tet2^HR HSPCs collected from mice 8 to 16 weeks old showed increased colony-forming replating capacity in vitro (Supplementary Fig. S1B) and increased competitive reconstitution capacity in vivo (Supplementary Fig. S1C). However, we observed that young Tet2^HR mice showed normal hematopoiesis. In contrast, upon aging, approximately 90% of the Tet2^HR mutant cohorts developed a spectrum of myeloid neoplasms (with latencies between 1.2 and 2 years) and increased mortality (Fig. 2D). To evaluate the potential role of Tet2 missense mutations in leukemic transformation, we assessed for cooperativity between Tet2^HR mutation and genetic lesions that have previously been shown to co-occur in human patients with AML (18–20). We observed that the Tet2^HR mutation cooperates with both Flt3-ITD and AML–ETO9a translocations as well as cohesin (Stag2) mutations leading to AML with significantly faster kinetics (Supplementary Fig. S1D–S1K). Surprisingly, the whole-exome sequencing of Tet2^HR mouse tumor cells (from eight different Tet2^HR mice) showed that not all tumors carry additional somatic mutations in genes frequently mutated in hematologic malignancies (Supplementary Fig. S1L.) Such findings suggested that progression to transformation can be in some cases independent of acquisition of secondary mutations in agreement with emerging notions of epigenetic and environmental triggers in human cancer.

The myeloid neoplasms developing in these animals with long latency showed a broad spectrum of aggressiveness and disease manifestation. Overall, Tet2^HR-induced myeloid malignancies were associated with leukocytosis, monocytosis, frequent anemia, and thrombocytopenia (Fig. 2E). Peripheral blood (PB) immunophenotyping of Tet2^HR mice older than 14 months revealed a significant expansion of myeloid cells (CD11b⁺, Gr-1⁺; Fig. 2F and G). Additionally, in the bone marrow, the HSPC compartment (LSK; Lineage^neg c-Kit⁺ Sca-1⁺ cells) was significantly increased in old Tet2^HR mice (Supplementary Fig. S2A), correlating with an expansion of myeloid multipotent progenitors (MPP3) and a loss of lymphoid multipotent progenitors (MPP4; Supplementary Fig. S2B–S2D). Compared with Tet2^WT and Tet2^HR young mice, old Tet2^HR showed a significantly larger spleen size (Supplementary Fig. S2E) driven by extramedullary hematopoiesis (Supplementary Fig. S2F–S2H). Morphologic evaluation of the bone marrow and extramedullary tissues of the Tet2^HR mice (spleen, liver, and lymph node) showed a spectrum of disease ranging from a more chronic phase of the disease with features of myelodysplastic/myeloproliferative neoplasm (characterized by atypical megakaryocytic hyperplasia and increased myeloid:erythroid ratio with preserved maturation of the myeloid lineage in the bone marrow, often with various degrees of extramedullary involvement predominantly involving the spleen and liver; Fig. 2H, middle) to a fully “transformed” AML with the involvement of extramedullary tissues manifesting frequently as myeloid sarcoma (Fig. 2H, bottom). Finally, sublethally irradiated mice transplanted with bone marrow cells from aged mice with an AML phenotype developed myeloid neoplasms more, and the mice more rapidly succumbed to the disease (Fig. 2I), confirming that Tet2^HR leukemic cells were fully transformed. Together, our data indicate that although the loss of Tet2 catalytic activity increases the self-renewal potential of the progenitor cells, it is not sufficient to promote transformation in Tet2^HR young mice and suggest that secondary events, acquired through aging, collaborate with the loss of Tet2 catalytic activity to induce myeloid neoplasms, including AML.

To delineate transcriptional events contributing to the progression to AML in the Tet2^HR mutant animal model, we performed single-cell RNA sequencing (scRNA-seq) on total bone marrow cells collected from young animals (8–10 weeks old) with no signs of disease and older animals (>60 weeks old) showing both chronic (MDS/MPN-like) and transformed myeloid disease (AML-like). Cells with more than 10% of mitochondrial reads and cells expressing fewer than 400 genes or more than 6,000 genes were removed from the analysis. Following this quality control, we collected 37,355 transcriptomes from young and old WT and Tet2^HR mice (Fig. 3A). Seurat anchor-based integration taking the 2,000 most variable genes as input was used to account for batch effect (21). The resulting scaled data were used to calculate principal component analysis (PCA) followed by uniform manifold approximation and projection (UMAP) visualization (22) using the first 20 principal components. Next, cells were assigned to their specific cell-type identity in an iterative fashion. First, unbiased clusters were given their respective hematopoietic lineage based on the expression of previously established lineage markers: HSPCs (Meis1 and Kit), monocytes and dendritic cells (Irf8 and Ly6c), neutrophils (Cebpe and Ly6G), B cells (Cd79, Vpreb3, and Bach2), T cells (Cd3, Cd8, Cd4, and Cd28), and erythroid cells (Klf1, Hba-a2, and Rhd; Supplementary Fig. S3A–S3C). Additionally, each cell type was subset, reclustered, and assigned to a more specific cell type based on the expression of well-known markers (Supplementary Fig. S3D; Supplementary Tables S1 and S2). This analysis resulted in 19 different cell states, including HSPCs, common lymphoid progenitors, B cells, plasma cells, T cells, natural killer cells, erythroid cells, myeloid progenitors, monoblasts, classical monocytes, nonclassical monocytes, conventional dendritic cells, plasmacytoid dendritic cells, basophils, proneutrophils, preneutrophils, and neutrophils (Fig. 3B). Both WT and Tet2^HR young transcriptomes appeared intermingling through the different cell types, showing no significant differences in their distribution in the UMAP. Moreover, differential expression analysis of young animal transcriptomes showed no statistical significance (Fig. 3C). Cell-cycle assignments per cell were determined, but no differences between conditions were found (Supplementary Fig. S3A). In contrast, our analysis revealed significant transcriptional differences later in life in animals that develop transformed disease (AML-like phenotype) compared with age-matched WT counterparts (Fig. 3C and D). In agreement with the immunophenotype analysis, loss of Tet2 catalytic activity in old mice resulted in an increased percentage of monocytes and a decreased percentage of granulocytes and B cells, especially in mice with the transformed phenotype (Fig. 3C and D; Supplementary Fig. S3E). Furthermore, the mutant granulocytes and monocytes from old mice accumulated in occupied distinct regions of the UMAP, suggesting that these cells shared expression profiles distinct from the WT controls. These results indicate that aging-mediated acquisition of aberrant transcriptional programs promotes transformation in TET2 mutant—associated myeloid neoplasms.

We next focused our analysis on the myelomonocytic compartment. To identify potential oncogenic pathways in Tet2^HR cells, we performed differential expression analysis within each cell population between age-matched mutant and WT HSPCs and the different myeloid cell states. Tet2^HR cells from young mice showed no significant differences in gene expression compared with young WT cells (data not shown). In contrast, Tet2^HR cells from aged mice showed significantly differentially expressed genes (DEG) in multiple cell states. Specifically, we identified 31 DEGs in HSPCs and 262 in classic CD14⁺ monocytes (Fig. 3E). We did not observe upregulation of established oncogenic pathways. However, gene set enrichment analysis identified inflammation-related signatures such as interferon gamma, interferon alpha, and TNFα hallmark signatures as the primary upregulated signatures in Tet2^HR with transformed disease. We created an overarching inflammatory response gene signature combining several inflammatory response signatures from the hallmark database. We found this inflammatory response pathway upregulated in different cell populations, from the HSPCs to the differentiated lineages of the myeloid compartment (Fig. 3F–H). Furthermore, we plotted the inflammatory response score on the UMAP representation, and we observed that the highest values correlated with the regions with increased density of Tet2^HR myeloid cells (Fig. 3F) representing the major transcriptional differences in the transformed samples. To gain deeper transcriptional coverage, we expanded the transcriptional profile of the inflammatory monocytes. We sorted total monocytes from Tet2^WT and Tet2^HR mice and performed bulk RNA sequencing (RNA-seq). This analysis helped confirm our finding of increased inflammation in the monocyte compartment of the Tet2^HR mice, but this time using bulk RNA-seq of the monocyte compartment of Tet2^WT and Tet2^HR mice. Indeed, genes related to IFN response, inflammatory response, and JAK/STAT signaling were enriched in Tet2^HR monocytes (Supplementary Fig. S4A and S4B).

Finally, and due to this induction of inflammatory gene signatures, we quantified the levels of proinflammatory cytokines in the bone marrow fluid of aged WT and Tet2^HR animals with aggressive neoplasms. We found an accumulation of a number of inflammatory cytokines, including IFNγ, IL6, IL1b, TNFα, and IL12, suggesting the establishment of a proinflammatory microenvironment in the bone marrow of these animals (Fig. 3I). Overall, these data indicate that the progression to myeloid transformation correlates with an increased inflammatory response signaling in the myeloid compartment of the diseased mice.

We next reanalyzed only the transcriptomes progressing from the HSPC to the differentiated monocytes and generated a new UMAP (Fig. 4A). Unbiased clustering coupled with cell calling revealed seven states: HSPCs, myeloid progenitors, monoblasts/promonocytes, classical monocytes, nonclassical monocytes, macrophages, and a distinct “monocytic-like” population present predominantly in Tet2^HR mice with transformed disease (Fig. 4A–C; Supplementary Fig. S4A and S4B; Supplementary Table S3). This separate cluster of monocytes (referred to henceforth as MHC II^hi monocytes due to its high levels of expression for MHC II) expressed transcriptional markers similar to the classical monocytes and different from nonclassical monocytes and dendritic cells (Fig. 4B and C; Supplementary Fig. S4C–S4E). In addition, MHC II^hi monocytes were characterized by a unique inflammatory signature including genes related to the IFN response pathway (Ly6a, Irf1, Irf3, Ifi204, Ifi209, Ifitm3, Isg15, Cxcl9, and Cxl10), components of MHC class II (Cd74, H2-Aa, H2-Ab1, etc.), as well as crucial inflammatory transcription factors including Stat1, Jun, Junb, Fos, and Fosb. Furthermore, we observed that this unique inflammatory response signature was also imprinted at earlier stages of monocyte development (in a subset of HSPCs, myeloid progenitors, and monoblast cells) in Tet2^HR mice (Fig. 4B), suggesting that the emergence of inflammation is also apparent at earlier differentiation stages and could shape their differentiation potential. To assess how inflammatory response affects the differentiation of the myeloid lineage, we used Monocle3 to calculate a pseudotime ordering along monocytic differentiation and to project differentiation trajectories onto the UMAP (23). We found that Tet2^HR mice with histologic-confirmed AML/myeloid sarcoma disease showed a breakpoint in the transition of the monoblasts leading to the Tet2 mutant MHC II^hi monocytes (Fig. 4D and E). We applied local regression (loess) to model variations of gene expression through pseudotime. Interestingly, the expression levels of myeloid markers through pseudotime showed no differences between the different groups (Fig. 4F). In contrast, inflammatory genes were progressively upregulated through the different transcriptional states until reaching the MHC II^hi monocytic subset in the Tet2^HR mice with severe transformed disease (Fig. 4G). These results suggest that aging and inflammatory signals dramatically alter the developmental trajectory of the Tet2^HR mutant myeloid cells that are imprinted early in the differentiation stages and promote the emergence of MHC II^hi monocytes.

Recent studies have described that the bone marrow is the source of an IFNγ-mediated subset of inflammatory monocytes (characterized by the high expression of MHC II and Sca-1 surface markers) in response to gastrointestinal infections (24). Furthermore, in a model of multiple sclerosis, another study has identified bone marrow–mediated generation of pathologic monocytes expressing inflammatory mediators like the chemokine Cxcl10 and the inflammatory acute phase reactant serum amyloid A3 gene (Saa3; refs. 25, 26). Interestingly, the Tet2^HR MHC II^hi monocytes showed transcriptional programs very similar to these subsets of inflammatory and pathologic monocytes previously described (Fig. 4H; refs. 24, 25, 27). Given that Tet2^HR MHC II^hi monocytes overexpressed Cxcl10 and Saa3, we interrogated the levels of these ligands in the bone marrow and the serum of aged mice and found that both inflammatory mediators were significantly increased in Tet2^HR mice with transformed disease (Fig. 4I and J). Taken together, our results indicate that enhanced inflammatory signaling in Tet2 mutant mice leads to the generation of a distinct subset of an inflammatory-primed monocytic population with pathologic potential.

To gain further insights into the biological and potential clinical significance of these findings and to assess the impact of the MHC II^hi monocytes in patients with AML, we generated an MHC II^hi monocyte–specific gene signature composed of genes specific to the MHC II^hi population from the single-cell analysis, genes upregulated in Tet2^HR monocytes from the bulk RNA-seq (Supplementary Fig. S4F and S4G), and 10 monocytic genes, that define the cell type (CTSS, KLF4, ZEB2, IRF5, F13A1, CSF1R, CD14, FCGR3A, ITGAM, and CCR2; Supplementary Tables S4–S6). Next, we compared the survival outcomes of patients with TET2-mutated AML with high or low expression of the MHC II^hi signature in a younger (<60 years old, n = 60) and older (≥60 years old, n = 186) cohort using the Alliance study protocols (27). We found that the MHC II^hi monocyte gene signature is significantly associated with survival in the younger cohort of TET2-mutated patients [3-year overall survival (OS) high vs. low; <60 years 24% vs. 58% P < 0.001; ≥60 years, 0% vs. 7% P < 0.22; Fig. 5A; Supplementary Table S7]. This result gives additional credence to our model, in which leukemia progression is supported not only by mutations and aberrant cytogenetics (28) but also by the induction of inflammation and a population of inflammatory monocytes. Notably, the adverse prognostic impact of the MHC II^hi monocyte signature was also applicable in the setting of AML in general, in both <60-year-old (n = 686) and ≥60-year-old (n = 186) patients with de novo AML. Similarly, in this cohort, the MHC II^hi monocyte signature was associated with significantly shorter survival (3-year OS high vs. low; <60 years 32% vs. 58% P < 0.001; ≥60 years, 4% vs. 21% P < 0.001; Fig. 5B; Supplementary Table S8). When considered with the European LeukemiaNet (ELN), the MHC II^hi monocyte signature was an independent predictor of OS in younger (<60 years) and older (≥60 years) adult patients with AML, demonstrating a “biomarker potential” for the presence of the MHC II^hi population (Supplementary Table S9).

Emerging studies are connecting inflammation to myeloid neoplasms, supporting our Tet2^HR animal findings (29). This notion prompted us to search for a monocytic population similar to the MHC II^hi monocytes described here in human samples. We thus analyzed several (n = 36) bone marrow samples with mutations in the catalytic domain of TET2, including 11 samples with a TET2^H1881 mutation using flow cytometry. As a control, we used a cohort of human AML samples with WT TET2 and mutated NPM1. We found that cases with TET2 mutation had an increased percentage of CD64⁺CD4⁺ monocytes with increased expression of HLA-DR (Fig. 5C and D) compared with TET2 WT NPM1-mutated samples (P = 0.041). Future studies are required to test the biological role of such monocytes in human AML.

Given the high levels of expression of the components of the MHC II^hi cluster and inflammatory response genes like Sca-1, both encoding for proteins expressed on the cell surface, we next assessed whether these inflammatory monocytes can be detectable in mice by flow cytometry. By immunophenotyping the blood and bone marrow myeloid compartment of aged animals (>16 months old), we observed a progressive expansion of an MHC II^hi, Sca-1^hi monocyte (MHC II^hi) population in the Tet2^HR mice with AML (Fig. 6A and B). Interestingly, we found that MHC II^hi monocytes can also be detected in aged Tet2^KO animals (Supplementary Fig. S5A; ref. 29), suggesting that this inflammatory population is not exclusive to the Tet2^HR mutation and correlates with Tet2 deficiency. More­over, MHC II^hi monocytes were also increased in the PB and the spleen of leukemic Tet2^HR mice (Fig. 6C; Supplementary Fig. S5B). In agreement with such animal studies, as described above (Fig. 5C and D), one could detect such monocytes in the bone marrow of patients carrying TET2 H1881 mutations using a combination of antibodies against CD45, CD4, CD64, and HLA-DR. However, we believe that emergence of this MHC II^hi monocytic population is not a universal phenomenon in murine myeloid leukemia, as we failed to detect such cells in carrying a number of additional mutations (Npm1/Flt3-ITD and Jak2^V617F; Supplementary Fig. S5C).

To define whether the emergence of inflammatory MHC II^hi monocytes is a cell-intrinsic Tet2-driven phenomenon, we generated hematopoietic chimeras by transplanting CD45.2⁺ bone marrow cells from young (8 weeks) Tet2^WT or Tet2^HR mice into lethally irradiated CD45.1⁺ recipient mice. At early time points after reconstitution, both Tet2^WT and Tet2^HR chimeras showed the emergence of lymphoid and myeloid reconstitution and no presence of inflammatory monocytes in the PB of any of the groups. However, 1 year after transplantation, we were able to see that MHC II^hi monocytes emerged and expanded in Tet2^HR chimeras (Fig. 6D and E), demonstrating that the emergence of this population is driven by the mutation of Tet2 in the hematopoietic compartment.

Interestingly, morphologic assessment of the liver and spleen of a mouse with histologically aggressive disease had shown marked infiltration by leukemic immature mononuclear cells, predominantly with an intrasinusoidal pattern of distribution, and some areas with morphologic features suggestive of monocytic differentiation (characterized by ample eosinophilic cytoplasm and highly convoluted nuclear contours, best appreciated in the spleen in Supplementary Fig. S5D and S5E). Therefore, we wondered whether this monocytic component of the leukemic infiltrates corresponded to the MHC II^hi monocytes identified in our transcriptional analysis and also by flow cytometry. To test this, we applied a multiplex immunofluorescence platform using Cd11b (a mature myeloid cell marker), MHC II antigen, and antibodies recognizing active modifications (phosphorylation) of the inflammatory transcription factors Stat1 and Stat3 (Fig. 6F; Supplementary Fig. S5E). This analysis showed that monocytic cells from the leukemic infiltrate (Cd11b⁺ cells) also expressed high levels of MHC II, p-Stat1, and p-Stat3, indicating that the inflammatory monocytes identified in our transcriptional analysis egress the bone marrow and infiltrate peripheral tissues. To evaluate if the MHC II^hi monocytes are part of the tumor, we transplanted bone marrow cells from three Tet2^HR old mice with signs of myeloid leukemia into lethally irradiated CD45.1 recipient animals. Six weeks after transplantation, we confirmed that the MHC II^hi monocytes expanded in the recipient mice (Fig. 6G and H). To test whether the MHC II^hi monocytes are transformed cells, we transplanted c-Kit⁺–enriched bone marrow cells as well as MHC II^hi–sorted cells from CD45.2 Tet2^WT and Tet2^HR mice into lethally irradiated CD45.1 recipients. Six weeks after transplantation, we confirmed that the c-Kit⁺ HSPC cells expanded in the recipient mice; however, MHC II^hi monocytes had no ability to transplant (Fig. 6I and J). Overall, our results demonstrate that inflammation in conjunction with Tet2 mutations promotes the emergence of a monocytic population (MHC II^hi), which acquires the ability to infiltrate peripheral tissues.

To further prove that inflammation can induce the emergence of this MHC II^hi population, we induced a chronic inflammatory environment by administering consecutive doses of lipopolysaccharide (LPS) by an intraperitoneal injection into young mice (8 weeks old; Fig. 7A). Interestingly, we found that LPS promoted the expansion of MHC II^hi monocytes in Tet2^HR mice but not in WT controls (Fig. 7B). Furthermore, the expansion of this inflammatory population remained sustained for at least 8 months after the last LPS injection (Fig. 7B). At the endpoint (8 months following the first injection), we observed overt splenomegaly in Tet2^HR mice treated with LPS (Fig. 7C and D). Morphologic assessment of spleen tissue of these mice showed extramedullary hematopoiesis, whereas no involvement of the spleen was observed in neither the Tet2^WT mice treated with LPS nor the Tet2^HR mice treated with PBS (Fig. 7E). The PB analysis at 8 months confirmed the increased levels of monocytes in Tet2^HR mice treated with LPS together with the decreased red blood cells (RBC) and white blood counts (WBC; Fig. 7F and G; Supplementary Fig. S6A). Immunophenotyping of bone marrow and spleen samples showed significant expansion of the LSK compartment (Fig. 7H; Supplementary Fig. S6B). Moreover, the expansion of MHC II^hi monocytes correlated with an increased level of Saa3 in the PB of Tet2^HR mice injected with LPS (Supplementary Fig. S6C). Interestingly, we observed increased colony formation capacity, consistent over 4 consecutive passages in the Tet2^HR LPS-treated mice compared with controls (Fig. 7I). This finding demonstrates that inflammatory signals such as LPS can accelerate the production of MHC II^hi monocytes in Tet2^HR mice, suggesting that an enhanced inflammatory response cooperates with Tet2 loss-of-function genetic events in promoting malignant transformation.

LPS triggers the TLR signaling pathway, and previous studies have shown that targeting factors in this signaling pathway, such as IRAK1/4, suppresses inflammatory activation in MDS and AML (30). Additionally, studies have shown that infections drive preleukemic myeloproliferation (PMP) and that antibiotic treatment prevented and reversed PMP (31). To investigate if antibiotic and IRAK1/4 inhibitor treatments can prevent the emergence of the MHC II^hi monocytes, we transplanted CD45.2 Tet2^HR mice into lethally irradiated CD45.1 Tet2^WT mice and allowed them to recover for 4 weeks after transplantation. Mice were treated with either PBS, antibiotic cocktail, or an IRAK1/4 inhibitor (Supplementary Fig. S6D). An antibiotic cocktail was provided by oral gavage for 1 week (given at a 1:50 concentration in drinking water; ref. 31). An IRAK1/4 inhibitor was injected intraperitoneally 5 days a week for 4 weeks (32). We observed that antibiotic and IRAK1/4 inhibitor treatment prevented the expansion of MHC II^hi monocytes in transplanted mice but not in the PBS-treated controls (Supplementary Fig. S6E and S6F). Taken together, our results demonstrate that blocking inflammatory signals disrupts the generation of MHC II^hi monocytes in leukemic mice.

During the last decade, a number of studies have suggested a relationship between dysregulation of inflammatory and immune response–related pathways and the development of myeloid neoplasms (MDS, MPN, and AML; ref. 33). However, the cellular and molecular mechanisms by which inflammation interacts with or even induces carcinogenesis in blood malignancies are under investigation. It is known that in the bone marrow, inflammatory signals activate HSPCs, inducing their proliferation and rapid differentiation to increase the myeloid output under stress conditions (34). Therefore, several studies have postulated that inflammation is detrimental to WT progenitor cells but could preferentially promote the emergence of clones that carry CHIP mutations, endowing cells with enhanced self-renewal (34). In agreement with this notion, we show here that a TET2 mutation found in CHIP and myeloid neoplasms can lead to myeloid malignancy and that this process of cellular transformation is coupled to organismal aging as well as the induction of inflammatory gene expression programs and an inflamed microenvironment.

A number of elegant studies have recently challenged the notion that monocyte function is modulated only by local signals at the infiltrating tissues (24, 25, 35–38). They have proposed that inflammation, particularly IFN signaling, increases the myeloid output in the bone marrow and rewires stem and progenitor cells (HSPC), resulting in the generation of inflammatory-primed monocytes, which respond faster to the exogenous stimuli. For instance, a subset of bone marrow inflammatory monocytes, characterized by their elevated levels of the MHC II and Sca-1 surface antigens, show increased regulatory and effector responses during Toxoplasma gondii intestinal infection (24). Work presented here reveals that there could be parallels between infection and early stages of transformation in the bone marrow. Specifically, our single-cell transcriptomic profiling suggests that inflammation reshapes the differentiation trajectory of preleukemic HSPCs, giving rise to a novel subset of inflammatory MHC II^hi monocytes. Although they lack leukemia-initiating capacity (LIC), Tet2^HR inflammatory monocytes acquire the ability to infiltrate peripheral tissues such as the liver and the spleen. They express inflammatory response genes, proteins of the MHC class II antigen presentation cluster, and a number of secreted factors. Such gene expression profiles would suggest that these inflammatory monocytes could affect the microenvironment and support disease progression (Fig. 7J). Indeed, it was recently suggested by van Galen and colleagues that myeloid cells in human AML could play an immunomodulatory role (35). Further supporting the notion that inflammatory monocytes also occur in human AML, we were able to show that a gene expression signature composed of genes highly expressed in mouse inflammatory Tet2^HR monocytes is tightly associated with poor prognosis in patients with AML.

It is well established that CH (including CHIP carrying TET2 mutations) is associated with inflammatory pathologies like cardiovascular disease, pulmonary disease, and type 2 diabetes (39–42). Previous studies have shown that Tet2-deficient macrophages promote inflammation, aggravating atherosclerosis (43–45). Specifically, myeloid Tet2-deficient cells showed the enhanced production of IL6 following an inflammatory trigger due to the direct effect of Tet2 in repressing the IL6 promoter by recruiting HDAC2 (46). Likewise, in response to bacterial infection–induced IL6, Tet2-deficient HSPCs show increased proliferation and myeloid differentiation (45). Additionally, Tet2-deficient cells have a strong proliferative and intrinsic antiapoptotic advantage in response to TNFa and IL6 activation (47, 48). These studies suggest an enhanced cell-extrinsic and cell-intrinsic component of the Tet2 response during inflammation and transformation. Our transcriptional analysis reveals that the inflammatory response pathway is aberrantly upregulated in Tet2 mutant leukemic cells, suggesting that Tet2 may regulate a large inflammatory program and not only IL6. Interestingly, inflammatory mediators (such as Cxcl10 and Saa3) are expressed by Tet2 mutant monocytes, and they are also increased in the bone marrow fluid and plasma of the Tet2 mutant mice with AML. Previous studies have shown that Saa3 has a protumorigenic role in pancreatic cancer and colon cancer, as well as promotes metastasis in breast cancer models (26, 49, 50), suggesting similar roles also in blood malignancy. Future studies on the cross-talk between inflammatory monocytes and their microenvironment will be crucial to delineate potential roles in supporting the malignant self-renewal progenitors and modulating the immune response, thereby facilitating tumor progression (51).

To construct the targeting vector with the mutation H1794R, we digested the plasmid K427, previously generated by Oliver Bernard to target exon 12 of mouse Tet2, with Kpn I and Nhe I enzymes (52). Next, we subcloned the targeting sequence into a pCDNA 3.1 plasmid and generated the mutation H1794R by using the Quick Change Stratagene kit. The mutated targeting sequence was then cloned back to the plasmid K427. The targeting vector K427-H1794R was electroporated to mouse embryonic stem cells by Applied Stem Cells. We PCR-screened for the clones carrying the proper homologous recombination using the primers GTAGAAGGTGGCGCGAAGGGGC and CACCCATAATTTTGTGTCTGAAGC, placed in the Neo cassette and outside the 3′ homology arm. The Rodent Genetic Engineering Laboratory at NYU Langone generated the Tet2HR-Frt mice by tetraploid complementation of two different targeted clones. Finally, the Flp cassette was removed by crossing the Tet2HR-Frt mice with Flp mice (JAX stock 003946). The primers CACCCATAATTTTGTGTCTGAAGC and GAGGCATGTTGAATGACTCTGG were used for PCR-geno­typing the mutant allele, discriminating a WT allele of 111 bp and an HR allele of 250 bp. All mice were housed at New York University School of Medicine under pathogen-free conditions. All procedures were conducted in accordance with the Guidelines for the Care and Use of Laboratory Animals and were approved by the Institutional Animal Care and Use Committees at New York University School of Medicine.

Blood was collected by either retro-orbital or a transterminal intracardiac bleed into EDTA-coated tubes. Blood samples were analyzed on an element HT5 Hemavet (Heska) according to the manufacturer's instructions.

Total bone marrow cells were collected from flushing the femur and tibia of CD45.2⁺ mice. Cell suspension was treated with lysate using 2 mL of ACK lysing buffer (Quality Biological). Donor cells (0.5 × 10⁶ per genotype per mouse) were transplanted via retro-orbital injection into lethally irradiated (2 × 450 rad) congenic CD45.1⁺ recipients (B6.SJL-Ptprca). Reconstitution was monitored 4 weeks after transplantation.

Mouse bone marrow cells were enriched using a CD117 microbead kit (Miltenyi Biotec) and separated using a MACS (Miltenyi Biotec) cell separator. After enrichment of CD117 cells, 2,000 cells/well were seeded in the MethoCult (M3434) medium (STEMCELL Technologies). The MethoCult medium cell mixture was plated into 12-well plates with water supply in the interwell chamber. After 7 days (1 passage) of incubation at 37°C in 5% CO₂, colonies were counted in three replicates per sample. The process was repeated for a total of 5 passages.

PB was collected via retro-orbital bleeding and lysed using 2 mL ACK lysing buffer. Bone marrow cells were collected from the femur and tibia of mice, and RBCs were lysed using 2 mL ACK lysing buffer (Quality Biological). For obtaining splenocytes and hepatocytes for flow cytometry, spleens and livers were harvested from mice and physically dissociated by pressing through a 0.40-μm cell strainer (Falcon) and were lysed with 5 mL ACK lysing buffer (Quality Biological). Single-cell suspensions were resuspended in PBS with 2% FCS and blocking solutions. All samples were gated on viable cells using forward scatter (FSC) and side scatter (SSC) size discrimination, followed by exclusion of doublets. Surface staining of mouse HSCs was performed in FACS blocking buffer using monoclonal antibodies against APC-Cy7–conjugated Gr-1, Cd11b, Cd4, Cd8, Ter119, Cd45r, and Nk1.1 (BioLegend), APC-conjugated Cd117 (BioLegend), PE-Cy7–conjugated Sca-1 (BioLegend), FITC-conjugated Cd150, and Pacific Blue–conjugated Cd48. Surface staining of mouse progenitor cells was performed in FACS blocking buffer using monoclonal antibodies against APC-Cy7–conjugated Gr-1, Cd11b, Cd4, Cd8, Ter119, Cd45r, and Nk1.1 (BioLegend), APC-conjugated Cd117 (BioLegend), PE-Cy7–conjugated Sca-1 (BioLegend), FITC-conjugated Cd34 (Invitrogen), and Pacific Blue–conjugated Cd16/32 (Invitrogen). Surface staining of mouse lineage cells was performed in FACS blocking buffer using monoclonal antibodies against FITC-conjugated Cd4 (Pharmingen) and Cd8 (BioLegend), PE-conjugated Cd71 (BioLegend), PECy7-conjugated Cd11b, and Pacific Blue–conjugated CD45r (BioLegend). Surface staining of mouse MHC II^hi cells was performed in FACS blocking buffer using monoclonal antibodies against APC-Cy7–conjugated Ly6c (BioLegend), BV605-conjugated Ly6 g (BioLegend), Pacific Blue–conjugated Cd11b (BioLegend), PE-conjugated I-A/I-E (BioLegend), and PE-Cy7–conjugated Ly6a (BioLegend). Samples were incubated with antibodies for 1 hour in the dark at 4°C. Samples were acquired using an LSRFortessa (BD Biosciences), and data were analyzed with FlowJo. Cell sorting was performed on an SY3200 Cell Sorter (Sony).

Tissue samples were fixed in 4% paraformaldehyde for 48 hours in 4°C. Samples were washed 3 times with PBS (Corning) and were placed in 70% ethanol solution. Hematoxylin and eosin staining was performed using standard methods.

Cell lysate was generated from WT and Tet2 mutant mouse bone marrow and was collected from spine, hips, femur, and tibia. A single-cell suspension was generated after the addition of RBC lysis ACK buffer and was treated with 2× Laemmli buffer and β-mercaptoethanol and quantified using BCA protein assay (Pierce). An equal amount of proteins was applied to a 4% to 12% Bis-Tris Gel (Invitrogen), and SDS-PAGE was performed. Wet transfer was performed following gel electrophoresis, followed by 1 hour blocking with 5% nonfat milk and TBST and overnight primary antibody incubation. The primary antibody was incubated at a 1:1,000 concentration. After washing the excess primary antibody, secondary antibody incubation was performed for 1 hour at room temperature before developing. A secondary antibody was incubated at a 1:2,000 concentration.

For collecting bone marrow fluids, the four long bones (two femurs and two tibiae) of each mouse were flushed with the same 200 μL of PBS. Cells were spun down by centrifugation at 1,000 RPM for 5 minutes. Next, supernatants were clarified by a second centrifugation at 12,000 RPM for 5 minutes. IL1α and IL1β were measured by ELISA (mouse IL1β and mouse IL1α ELISA kit; R&D Systems). IFNα, IL1β, and other cytokines were measured using the LEGENDplex antivirus response panel (BioLegend).

Blood was collected by retro-orbital bleed into a BD Microtainer Serum Separator Tube. Blood was allowed to clot for 30 minutes at room temperature. After incubation, samples were centrifuged at 15,000 × g for 3 minutes. Serum was collected from the supernatant. The SAA Elisa Kit (Abcam) was used to quantify the levels of Saa3 in the serum following the direction of the manufacturer.

Mice were treated with an antibiotic cocktail as previously described (31). All antibiotics used are listed as follows: concentration used, company, catalog number. In detail, in the first week, mice received a daily intragastric gavage with 100 μL of a mixture of kanamycin (4 mg/mL, Sigma-Aldrich, 60615), gentamicin (0.35 mg/mL, Sigma-Aldrich, G1914), colistin (8500 U/mL, Sigma-Aldrich, C4461), metronidazole (2.15 mg/mL, Sigma-Aldrich, M3761), and vancomycin (0.45 mg/mL, Sigma-Aldrich, V2002). For the 3 weeks that followed, antibiotics were administered in the autoclaved (nonacidified) drinking water at a 50-fold dilution except for vancomycin, which was maintained at 0.5 mg/mL. Autoclaved (nonacidified) water was used to dissolve anti­biotics. Antibiotic water was prepared fresh and replaced weekly.

Mice were treated with an IRAK1/4 inhibitor as previously described (30, 32). The IRAK1 inhibitor (IRAK1/4 inhibitor; Amgen Inc.) was purchased from Sigma-Aldrich (I5409). Animals were intraperitoneally injected with NCGC-1481 (30 mg/kg).

Bone marrow cells were isolated from the spine, hips, femur, and tibia of mice. A single-cell suspension was generated after the addition of RBC lysis ACK buffer. Cells were tagged with cell hashing (53) oligo-tagged antibodies (BioLegend) according to the manufacturer's instructions and pooled on a 10X Chromium instrument (10X Genomics) to generate barcoded single-cell beads-in-emulsion (GEM) according to the manufacturer's protocol. scRNA-seq libraries were prepared using the Chromium Single-Cell’ v3 Reagent Kit (10X Genomics) following the direction of the manufacturer. The libraries were quantified using Qubit fluorometric quantification (Thermo Fisher Scientific). The quality was assessed on an Agilent Bioanalyzer 2000, obtaining an average library size of 455 bp. Libraries were pooled and normalized to 2 nmol/L and clustered on a HiSeq4000 high-output mode on a paired-end read flow cell and sequenced for 28 cycles on R1 (10X barcode and unique molecular identifiers), followed by 8 cycles of i7 index (sample index) and 98 bases on R2 (transcript), with a coverage of around 130M reads per sample.

scRNA-sequencing data were demultiplexed and converted to FASTQ format using Illumina bcl2fastq software and then processed using Cell Ranger for barcode processing and gene counting using the mm10 reference. Quality control of the data were done using the R package Seurat (19). Low-quality cells were excluded with a cutoff of a minimum of 400 genes and a maximum of 6,000 genes expressed per cell, as well as a cutoff of a maximum of 10% of the reads coming from mitochondrial reads. Samples that were multiplexed using hashtag oligos were demultiplexed using Seurat function HTODemux using a positive quantile of 0.99. Cells determined to be doublets or singlets after demultiplexing were removed. After filtering, we obtained 37,355 cells. Data were then normalized by total expression per cell, multiplied by 10,000, and then log-transformed.

Samples split between several libraries and run at the same time were combined using the Seurat function merge. The Seurat anchor-based integration method was used to account for technical differences between libraries run on different days, using the 2,000 most variable genes. The resulting scaled data were used to calculate PCA followed by UMAP using 20 neighbors and 20 principal components. The resulting principal components were also used as input to the Smart Local Moving clustering algorithm used to partition the data set. A range of resolutions (0.1–10) was used to explore clusters that corresponded to cell populations and biological processes.

UMAP plots colored by density ratios were determined by calculating densities for each condition, by splitting the UMAP plane into 2D bins and counting the percentage of points in each bin. The log₂ fold change between the values for each condition was calculated and used to color each cell.

Pseudotime analysis was conducted using Monocle3 (22) on the monocyte subset of the integrated and scaled matrix. Loess curves were fit to gene expression values for each condition along the pseudotime axis.

Cells were assigned cell-type identity in an iterative fashion. First, all cells were clustered. Clusters were assigned known broad cell types based on the expression of established markers. Specifically, the cluster-average expression values were filtered for genes expressed in any clusters based on a detection threshold of 10 counts per million (∼3 transcripts per cell), the values for each gene were normalized relative to the maximum level across all clusters, and the rescaled values were averaged across all markers for each population, with each cell assigned to the population with the highest value. Following separation into the broad cell types, each cell type was subset into its own data set and reclustered. Clusters within each cell type were then assigned granular cell-type labels based on the expression of established markers using the average expression levels of every signature and subtracting the aggregated expression of randomly selected control genes. Cell-cycle analysis was performed per cell based on the expression of established markers using the average expression levels of each signature and subtracting the aggregated expression of all genes.

Box plots of cell-type abundances were created by calculating per mouse, the percentage of cells in each cell-type, relative to the total number of cells per mouse. Statistical analysis was conducted using the ggpubr package to calculate the Wilcox test statistic between conditions, and ggplot was used to create the plots (54).

Differential expression between conditions or between cell types was performed using the Wilcox test statistic, using the FindMarkers function from the package Seurat. The MHC II monocyte gene signature was determined by comparing, within the monocytes, genes significantly expressed in the MHC II^hi monocytes.

Inflammatory response genes were defined as genes included in either the Hallmark Interferon Alpha Response, Hallmark Interferon Gamma Response, Hallmark Inflammatory response, Hallmark IL6– JAK–STAT3 Signaling, Hallmark IL2–STAT5 signaling, or Hallmark TNFA Signaling via NFKB gene sets from the Molecular Signatures Database (MSigDB). The inflammatory response score was calculated by taking the difference between the mean expression of genes within the gene list, compared with the mean expression of all genes.

Analysis of bulk RNA-seq was performed using the pipelines provided in the Seq-N-Slide pipeline (ref. 55; https://github.com/igordot/sns). Briefly, after trimming using Trimmomatic (55), reads were aligned to the mm10 reference genome using STAR (56), and count tables were generated using featureCounts (57). DESeq2 (58) was then used to identify DEGs between the Tet2^HR and Tet2^WT conditions.

Analysis of whole-exome sequencing data was performed using the Seq-N-Slide pipeline (ref. 55; https://github.com/igordot/sns). In short, adapters were trimmed using Trimmomatic (55) and aligned to reference genome mm10 using BWA-MEM (59). Duplicate reads were removed using Sambamba (60). Somatic point mutations and small indels were identified using Strelka (61).

CHIP data were acquired from Bick and colleagues, Zink and colleagues, and Jaiswal and colleagues (45, 62, 63). AML data were acquired from the Tet2 COSMIC database (64). Tet2 missense mutations were identified and plotted using the trackViewer package (65). Tet2 domains were defined according to Hu and colleagues (14).

Average gene expression of classical monocytes, nonclassical monocytes, and MHC II^hi monocytes was calculated and used in an enrichment analysis for gene signatures characterizing inflammatory and pathologic monocytes from Villani and colleagues, Giladi and colleagues, and Askenase and colleagues (24, 25, 27). The gene set variation analysis (GSVA) package was used to calculate enrichment (66).

The H1881R mutation was incorporated into the 4NM6 crystal structure using the LEaP program of AMBER (14). The structure was prepared and compared with WT as previously described with Mg(II) in place of Fe(II) using the published 5-position modified cytosine parameters (15). Three systems, with either 5mC, 5hmC, or 5fC as the everted base, were simulated in triplicate for 200 ns of production dynamics each with TIP3P water and the ff99SB force field (58). Production dynamics were performed with the Langevin thermostat in the NPT ensemble using the pmemd.cuda AMBER18 code (16). The H1881R results were compared with our previous WT simulations extended out to 500 ns (15). Cpptraj was used for structural analysis, including calculating the root mean square fluctuations and backbone angles (67). An open-source program was used to perform the by-residue energy decomposition analysis with respect to the 5-position cytosine (https://zenodo.org/record/4469899#.YSGFmx0pBZo). Data were further processed using R v4.0.3 with the abind, data.table, and tidyverse packages and Python v3.6.7 with the NumPy, pandas, snakemake, and matplotlib modules (54, 68–70). UCSF Chimera was used to create images (71).

The human TET2-CS variant (1129–1936 Δ1481–1843) with an N-terminal FLAG tag was expressed using a pLEXm expression vector as previously described (10, 15). The H1881R mutation or mutation of the catalytic H1382Y and D1384A (HxD mutant) was generated by standard means. HEK293T cells were cultured in DMEM with GlutaMAX (Thermo Fisher Scientific) and 10% FBS (Sigma). Cells were transfected with WT hTET2-CS, mutant hTET2-CS, or an empty pLEXm vector control using Lipofectamine 2000 (Thermo Fisher Scientific) according to the manufacturer's protocol. The medium was changed 24 hours after transfection, and cells were collected by trypsinization 48 hours after transfection and resuspended in PBS. Genomic DNA was isolated from the cells using the DNeasy Blood and Tissue Kit (Qiagen).

For LC/MS-MS, 1 μg of genomic DNA from each sample was degraded to component nucleosides with Nucleoside Digestion Mix (New England Biolabs) at 37°C overnight. The nucleoside mixture was diluted 10-fold into 0.1% formic acid and injected onto an Agilent 1200 Series HPLC with a 5 μm, 2.1 × 250 mm Supelcosil LC-18-S analytical column (Sigma) equilibrated to 45°C in buffer A (0.1% formic acid). The nucleosides were separated by column affinity over 10 minutes in buffer A at a flow rate of 0.5 mL/minute, and then the column was washed in 0 to 100% buffer B [0.1% formic acid, 30% (v/v) acetonitrile]. Tandem MS/MS was performed by positive ion mode ESI on a 6,460 triple-quadrupole mass spectrometer (Agilent) with a gas temperature of 225°C, a gas flow of 12 L/minute, a nebulizer pressure of 35 psi, a sheath gas temperature of 300°C, a sheath gas flow of 11 L/minute, a capillary voltage of 3,500 V, a fragmentor voltage of 70 V, and a delta EMV of +1,000 V. Collision energies were optimized to 10 V for 5mC and 5fC; 15 V for 5caC; and 25 V for 5hmC. Multiple reaction monitoring mass transitions were 5mC 242.11→126.066 m/z; 5hmC 258.11→124.051; 5fC 256.09→140.046; 5caC 272.09→156.041; and T 243.10→127.050. Standard curves were generated using standard nucleosides (Berry & Associates, Inc.) ranging from 1.85 μmol/L to 113 pmol/L (9.25 pmol to 0.565 fmol total). The sample peak areas were fit to the standard curve to determine the amount of each modified cytosine in the genomic DNA sample. Amounts are expressed as the percentage of total cytosine modifications.

We investigated 872 adult patients with de novo AML who were enrolled on the CALGB/Alliance study protocols and received treatment regimens, as detailed below. Patients were excluded from outcome analyses if they received an allogeneic HSC transplantation in first complete remission (CR).

All patients gave written informed consent for participation in the studies. All study protocols were in accordance with the Declaration of Helsinki and approved by Institutional Review Boards at each treatment center. All patients were enrolled on the CALGB 8461 (cytogenetic studies), CALGB 9665 (leukemia tissue bank), and CALGB 20202 (molecular studies) companion protocols. Patients were treated on CALGB/Alliance protocols [refs. 72–81; CALGB 8525 (n = 28), 8621 (n = 1), 8721 (n = 1), 8821 (n = 3), 8923 (n = 10), 9022 (n = 3), 9120 (n = 1), 9222 (n = 53), 9420 (n = 9), 9621 (n = 101), 9720 (n = 66), 10201 (n = 62), 10502 (n = 19), 10503 (n = 227), 10603 (n = 54), 11001 (n = 5), 11002 (n = 7), and 19808 (n = 232)].

Clinical endpoints were defined according to generally accepted criteria. A CR was defined as recovery of morphologically normal bone marrow and blood counts (i.e., neutrophils ≥1.5 × 10⁹/L and platelets >100 × 10⁹/L), and no circulating leukemic blasts or evidence of extramedullary leukemia, all of which had to persist for ≥4 weeks. Disease-free survival (DFS) was measured from the date of achievement of a CR until the date of relapse or death from any cause; patients not known to have relapsed or died at last follow-up were censored on the date they were last examined. OS was measured from the date of diagnosis to the date of death from any cause; patients not known to have died at last follow-up were censored on the date they were last known to be alive. Event-free survival (EFS) was measured from the date of study entry until the date of failure to achieve CR, relapse, or death. Patients alive and in CR at last follow-up were censored.

Low and high monocyte signature groups were compared using the Fisher exact test for categorical variables. Estimated probabilities of DFS, OS, and EFS were calculated using the Kaplan–Meier method (23), and the log-rank test evaluated differences between survival distributions. Cox proportional hazard models were used to calculate HR for DFS, OS, and EFS. All statistical analyses on Alliance patients were performed by the Alliance Statistics and Data Center.

The MHC^hi monocyte signature for the survival analysis in the Alliance patients was calculated using 204 genes identified either as marker genes of the MHC II^hi monocyte population in the single-cell data or as differentially expressed in the Tet2 monocytes profiled by bulk RNA-seq. Additionally, 10 known monocyte genes were added to the list (CTSS, KLF4, ZEB2, IRF5, F13A1, CSF1R, CD14, FCGR3A, ITGAM, and CCR2). The signature was derived as a linear combination of the expression of the 214 genes. The monocyte signature for patient i was c_i = Σ w_j x_ij, where x_ij was the expression value for gene j in patient i, and w_j was the weight assigned to gene j. The univariable Cox regression coefficients for OS for each of the 214 genes included in the signature were used as the weights (w_j) in the monocyte signature.

Viable cryopreserved bone marrow or blood cells of patients enrolled onto the CALGB 9665 tissue bank protocol were stored for future analyses prior to starting treatment. Mononuclear cells were enriched through Ficoll-Hypaque gradient centrifugation and cryopreserved until use. Genomic DNA was extracted using the DNeasy Blood and Tissue Kit (Qiagen). The mutational status of 80 protein-coding genes was determined centrally at The Ohio State University by targeted amplicon sequencing using the MiSeq platform (Illumina). Furthermore, testing for CEBPA mutations was performed with the Sanger sequencing method, thus adding up to a total of 81 genes analyzed.

All data generated in this study are publicly available in the Gene Expression Omnibus (GEO) under accession numbers GSE182615 and GSE201659 as well as the Sequence Read Archive (SRA) with the BioProject ID PRJNA833651. Patient data used in survival analysis were obtained from Alliance Statistics and Data Center and are available upon reasonable request; generated data are deposited in the GEO under the accession numbers GSE137851 and GSE63646. The code used for the single-cell data analysis is available at https://github.com/igordot/scooter.

A. Yeaton reports grants from the National Center for Advancing Translational Sciences (NCATS), the NIH, through grant award number TL1TR001447 during the conduct of the study, as well as other support from Microsoft Research and the Broad Institute outside the submitted work. V. Paraskevopoulou reports grants from EMBO outside the submitted work. D.T. Starczynowski reports personal fees from Kurome Therapeutics outside the submitted work; serves on the scientific advisory board for and has equity in Kurome Therapeutics; and has consulted for Kymera Therapeutics, Kurome Therapeutics, Captor Therapeutics, and Tolero Therapeutics. O. Abdel-Wahab reports grants from Nurix Theraeputics, Loxo Oncology, and personal fees from Pfizer (Boulder) outside the submitted work. A.D. Viny reports other support from Nooma Bio outside the submitted work. R.M. Stone reports other support from AbbVie, Actinium, Agios, Astellas, BioLineRx, Celgene, Daiichi Sankyo, GEMoaB, Janssen, Macrogenics, Novartis, Takeda, Trovagene, Bristol Myers Squibb, Boston Pharmaceuticals, CTI Pharma, GSK, Jazz, Syros, Syntrix/ACI, Elevate Bio, BerGenBio, Foghorn Therapeutics, Aprea, Innate, OncoNova, Aptevo, AvenCell, Epizyme, and Kura Oncology outside the submitted work. A.-K. Eisfeld reports other support from Karyopharm Therapeutics outside the submitted work. I. Aifantis reports nonfinancial support and other support from Foresite Labs outside the submitted work. M. Guillamot reports grants from the American Society of Hematology and Gilead during the conduct of the study. No disclosures were reported by the other authors.

A. Yeaton: Conceptualization, data curation, software, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. G. Cayanan: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. S. Loghavi: Investigation, writing–review and editing. I. Dolgalev: Software. E.M. Leddin: Investigation, visualization, writing–review and editing. C.E. Loo: Resources, formal analysis, visualization. H. Torabifard: Investigation. D. Nicolet: Investigation. J. Wang: Investigation. K. Corrigan: Investigation. V. Paraskevopoulou: Resources, investigation. D.T. Starczynowski: Conceptualization, resources. E. Wang: Resources, investigation. O. Abdel-Wahab: Resources, supervision. A.D. Viny: Resources, formal analysis, investigation, visualization. R.M. Stone: Conceptualization, resources, supervision, funding acquisition, investigation, methodology, project administration, writing–review and editing. J.C. Byrd: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. O.A. Guryanova: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, writing–original draft, project administration, writing–review and editing. R.M. Kohli: Formal analysis, visualization. G.A. Cisneros: Conceptualization, resources, investigation. A. Tsirigos: Conceptualization, resources, supervision, funding acquisition, methodology, project administration, writing–review and editing. A.-K. Eisfeld: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, writing–original draft, project administration, writing–review and editing. I. Aifantis: Conceptualization, resources, formal analysis, supervision, funding acquisition, visualization, methodology, project administration, writing–review and editing. M. Guillamot: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, writing–original draft, project administration, writing–review and editing.

We thank all members of the Aifantis laboratory for discussions throughout this project; A. Heguy and the NYU Genome Technology Center (GTC) for expertise in sequencing as well as the Applied Bioinformatics Laboratories (ABL) for providing bioinformatics support and helping with the analysis and interpretation of the data (the GTC and ABL are shared resources partially supported by Cancer Center Support Grant P30CA016087 at the Laura and Isaac Perlmutter Cancer Center); the NYU Histology Core (5P30CA16087-31) for assistance; C. Loomis for immune multiplex experiments; the Rodent Genetic Engineering Laboratory at NYU Langone for electroporating the ESC clones and generating the Tet2 mutant mice; the NYU Flow Cytometry facility for expert cell sorting; Olivier Bernard (Gustave Roussy and Université Paris-Saclay) for the Tet2 targeting vector; and Kristian Helin (Memorial Sloan Kettering Cancer Center) for the mouse Tet2 antibody. The results in this work here are in part based upon data generated by The Cancer Genome Atlas Research Network: https://www.cancer.gov/tcga. This work has used computing resources at the NYU School of Medicine High-Performance Computing Facility. A. Yeaton is supported by a Translational Research TL1 Grant. G.A. Cisneros is supported by the NIH (grant no. R01GM108583). A. Tsirigos is supported by the NCI/NIH R01CA260028-01, NCI/NIH P01CA229086, and NCI/NIH R01CA252239-01 and is a St. Baldrick Scholar. A.-K. Eisfeld is supported by Pelotonia, the American Society of Hematology, the Leukemia & Lymphoma Society (LLS), and the Leukemia Research Foundation. V. Paraskevopoulou is an EMBO postdoctoral fellow (ALTF 742-2020). M. Guillamot is supported by the Ramon Areces Foundation, is supported by the American Society of Hematology (ASH Restart Research Award), and is a Gilead Research Fellow. I. Aifantis is supported by the NIH/NCI (RO1CA216421, RO1CA173636, RO1CA228135, RO1CA242020, RO1HL159175 and RO1CA271455), The EvansMDS Foundation, and the LLS. U10CA180821, U10CA180882, and U24CA196171 have supported the Alliance for Clinical Trials in Oncology.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.