Activation of a Subset of Evolutionarily Young Transposable Elements and Innate Immunity Are Linked to Clinical Responses to 5-Azacytidine

The DNA methyltransferase inhibitor (DNMTi) 5-azacytidine has been approved by the FDA and the European Medicines Agency to treat higher-risk myelodysplastic syndrome (MDS), chronic myelomonocytic leukemia (CMML), and acute myeloid leukemia (AML). Except for allogeneic bone marrow transplantation, 5-azacytidine is the only drug that has been shown to extend the survival of patients with MDS. Yet only about half of the patients respond to the treatment, and most patients become resistant over time (1, 2). Despite intense investigation, the molecular factors associated with the variable clinical responses have not yet been identified (3, 4).

Numerous studies have reported that aberrant DNA methylation is one of the most common molecular changes in cancer cells (5). Hypomethylation of abnormally silenced tumor suppressor genes has been suggested to be the most likely mechanism for response to DNMTis (5–7), but upregulation of cancer testis antigens (CTA) has also been suggested to play important roles in the activation of the immune system and in killing cancer cells (8–10). Recently, we and other groups have described “viral mimicry,” in which the destruction of cancer cells is mediated by the innate immune system in response to single- and double-stranded transcripts of transposable elements (TE) that are induced by DNMTi treatment (11, 12).

Activation of the viral mimicry pathway has also been demonstrated in cancer cell lines treated with other types of drugs that target the epigenome, such as inhibitors of G9a (histone lysine 9 methyltransferase) or histone lysine-specific demethylase I (13, 14). Therefore, epigenetic mechanisms including DNA methylation and histone modifications keep TEs silenced (13–15). Despite these extensive in vitro data, the expression profiles of genes involved in the viral mimicry pathway have not yet been extensively investigated or validated in patients with cancer treated with DNMTis.

Here we first confirmed that the three major classes of TEs, which are present in all human cells, were indeed upregulated in patients in response to the same drug (5-azacytidine) in different hematologic malignancies. We then show that eight upregulated LTRs are capable of activating genes in the type I IFN pathway providing direct mechanistic evidence for a role for viral mimicry in patients.

The Nordic cohort included 12 patients (MDS, n = 6; CMML, n = 5; AML, n = 1), all treated with 5-azacytidine from whom consecutive peripheral blood or bone marrow samples were available. All patients had provided written informed consent in accordance with the Declaration of Helsinki, and the study was approved by the local ethics committees and IRB. 5-azacytidine treatment was performed according to the Nordic MDS guidelines, in which 5-azacytidine was administered subcutaneously (100 mg/m²) for five consecutive days on a 28-day schedule. Patient characteristics at treatment initiation are shown in Supplementary Table S1. All patients had been diagnosed with higher-risk disease except for one, who had lower-risk MDS but elevated blast counts before the onset of treatment. Clinical response to 5-azacytidine was evaluated using IWG criteria (16) and defined as the best response during the whole treatment course. Patients with complete remission (CR), partial remission (PR), marrow complete remission (mCR), and hematologic improvement (HI) were considered as responders. The patient characteristics of the Australian (MDS, n = 7; CMML, n = 7) and the pediatric [AML, n = 11; acute lymphoblastic leukemia (ALL), n = 3] cohorts were described in our and others' previous reports, respectively (17, 18).

Peripheral blood (PB) and bone marrow (BM) mononuclear cells were isolated immediately upon sample collection using Ficoll-Paque PLUS (GE Healthcare) or Lymphoprep (StemCell Technologies) density gradient centrifugation and cryopreserved at –196°C for later use. The cells were thawed and immediately isolated into either CD14⁺ monocytes, CD3⁺ T cells, or CD34⁺ stem and progenitor cells using magnetic bead-based separation on a RoboSep magnetic cell separator device (StemCell Technologies) or using MACS columns (Miltenyi Biotec) according to the manufacturer's instructions. For all 5 patients with CMML, CD14⁺ cells were used. CD34⁺ cells were used for 2 patients with MDS with BM samples available. For three of the patients with MDS and AML, the CD3 and CD19 double-negative population was used for further analysis of mainly myeloid cells. These were isolated by stepwise positive selection of CD3⁺ and CD19⁺ cells. The isolated cells were subjected to total RNA extraction using the AllPrep DNA/RNA Kit (Qiagen) according to the manufacturer's instructions, including stringent DNase I cleanup of any contaminating genomic DNA, and the concentration was measured using the Qubit fluorometer with RNA HS assay (Thermo Fischer Scientific). For 2 patients with MDS, total RNA was extracted directly from PB mononuclear cells without further isolation. For each individual patient, the same malignant cell populations were isolated and used for RNA sequencing (RNA-seq) at each time point before and during treatment (Supplementary Table S1).

Total RNA-seq was performed on the 63 RNA samples from the 12 Nordic patients. cDNA libraries were prepared using RNA HyperPrep kits with RiboErase (KAPA Biosystems) according to the manufacturer's instructions and were sequenced as single-end 75 bases on a NextSeq 500 instrument (Illumina). Sequencing reads were aligned against the human GRCh37 reference genome to report results consistent with our previous study (19, 20) by using Bowtie2 and TopHat2 v.2.1.0 (21) with the following parameters: “bowtie2-2.2.5/bowtie2/bowtie2-build –f ∼/genomes/hg19.fa Index-hg19” and “tophat-2.1.0.OSX_x86_64/tophat2 -g 1 -I 5000 -o tophat_out bowtie2/hg19 sequence_file.fastq.” The number of overlapping mapped reads with coding genes and TEs was obtained using Samtools-1.2 and BEDTool2-2.24.0 (22) with the following parameters: “samtools-1.2/samtools index accepted_hits.bam” and “bedtools2/bin/coverageBed -a reference_file.bed -b sequence_file_hits.bed -split -sorted > hits.count.txt.” RPKM values (reads per kilobase per million mapped reads) were calculated using uniquely mapped reads. The raw and processed data sets from this study have been submitted to the NCBI Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo/) under accession number GSE118558.

Gene ontology analysis of upregulated genes was performed using biological processes in the Gene Ontology database (http://www.geneontology.org; refs. 23, 24). Upregulated genes were defined as those increased by 2-fold or more posttreatment relative to pretreatment using median RPKM values of the nonresponders and responders (the highest values from the multiple time points were used). Gene set enrichment analysis (GSEA) was performed on the basis of P values obtained by a two-tailed Wilcoxon signed-rank test comparing pretreatment and posttreatment RPKM values (the highest values from the multiple time points were used) in the nonresponders and responders. A standard Kolmogorov–Smirnov test (25) was used to test whether a specific gene set showed differential expression between the pretreatment and posttreatment sample. The significance of GSEA was evaluated by permutating the target gene set. Tumor suppressor genes were identified in the NCG 5.0 database (n = 38; ref. 26). Ninety-nine of 208 CTAs identified by CTdatabase (27) were used, because the other 109 CTAs were not expressed. Immune response-related genes (n = 2,312) were identified following Gene Ontology terms (GO:0002376 and GO:0045321).

The number of reads from TE transcripts in the Nordic and Australian cohorts was obtained using BEDTool after unique mapping by TopHat2 v.2.1.0. A majority of TEs can be detected by the mapping, as reported in our previous study (20). The definitions of TEs were based on RepeatMasker (http://www.repeatmasker.org). The TEs overlapping with coding genes or with long noncoding RNAs were excluded from the analysis to determine the transcripts emanating from the TE elements, as described in our previous report (20). The TEs expressed at low level (less than 20 read counts) were also filtered out from this study. For the analysis of endogenous retroviruses (ERV), we only included reads mapped to their LTR regions, because LTRs are putative promoters of ERVs capable of initiating transcription.

cDNA synthesis using RNA samples from the pediatric cohort was performed with an iScript cDNA Synthesis Kit (Bio-Rad) after digestion of genomic DNA by TURBO DNase (Thermo Fisher Scientific) according to the manufacturer's instructions. qPCR reactions were carried out in CFX96 Real-Time PCR detection system (Bio-Rad) with a KAPA SYBR FASTqPCR Kit (KAPA Biosystems). The conditions of qPCR reaction were as follows: 3 minutes at 95°C, followed by 40 cycles of 10 seconds at 95°C, 20 seconds at 60°C, and 20 seconds at 72°C. Primer sequences are listed in Supplementary Table S2.

Eight individual LTRs were transcribed from a T7 promoter in PCR products using the mMESSAGE mMACHINE T7 Transcription Kit (Thermo Fisher Scientific). Transfection was performed using Lipofectamine 2000 Transfection Reagent (Thermo Fisher Scientific) using 500 ng of in vitro transcribed LTRs for 5 × 10⁴ HCT116 cells. Total RNA from HCT116 was purified 48 hours after treatment using a Direct-zol RNA MiniPrep Kit (Zymo Research). Primer sequences are listed in Supplementary Table S2.

We first performed RNA-seq on sorted malignant myeloid cells and T cells from the Nordic cohort of patients with MDS, CMML, and AML (malignant cells: four nonresponders and eight responders; T cells: five responders; see Supplementary Table S1; Supplementary Fig. S1). In addition, we downloaded RNA-seq data on CD34⁺-sorted MDS and CMML hematopoietic stem and progenitor cells from an Australian cohort in a publicly available database (five nonresponders and nine responders; see Supplementary Table S1; Supplementary Fig. S1; ref. 17).

We analyzed RNA-seq data by supervised cluster analysis to identify differential expression patterns between nonresponders and responders. We concentrated on upregulated genes for the following analyses, because previous reports from us and others using cultured cells showed that upregulation of tumor suppressor genes, CTAs, and innate immune genes might be involved in antitumor effects of the drug in vitro (5, 28). In the nonresponders, 385 coding genes were identified as being highly upregulated in the malignant cells when comparing posttreatment to pretreatment values [fold change (FC) ≥ 2; Supplementary Fig. S2A; Supplementary Table S3). Among responders, 1,117 coding genes were identified as being highly upregulated (FC ≥ 2; Supplementary Fig. S2A; Supplementary Table S3). Gene ontology analysis of the upregulated genes showed that immune response gene sets, including those involved in defense response and innate immune response, were upregulated in samples from malignant cells of patients responding to 5-azacytidine treatment (Supplementary Fig. S2B; Supplementary Table S4), but not from malignant cells of nonresponders or T cells from responders consistent with recent findings in cancer cell lines after DNMTi treatment (11–13, 19, 20). Earlier studies reported that treatment with DNMTis alone or together with histone deacetylase inhibitors enhance T-cell rejuvenation accompanied by tumor T-cell infiltration facilitated by cytokine production by the cancer cells (29, 30). We confirmed the upregulation of leukocyte activation gene sets in the malignant cells from responders (Supplementary Fig. S2B; Supplementary Table S4); however, these gene sets were not upregulated in the gene ontology analysis of the nonresponders and T cells from responders (Supplementary Fig. S2B; Supplementary Table S4). GSEA confirmed that immune response–related genes were significantly upregulated following 5-azacytidine treatment in responders (P = 3 × 10⁻⁶) but not in nonresponders (Fig. 1A). The activation of the innate immune system in normal cells may lead to off-target effects for patients (5); however, we found no upregulation of these gene sets among the 3,164 genes in the presumably normal T cells from responders (FC ≥ 2; Supplementary Figs. S2A and S2B; Supplementary Table S4). Thus, the 5-azacytidine-inducible innate immune response might be cancer cell–specific. This specificity may be due to the fact that 5-azacytidine has an absolute requirement for incorporation into DNA to inhibit DNA methylation (31). Because normal cells often have lower proliferation rates and many are quiescent, this may help explain the preferential effects on cancer cells (5). Activation of the innate immune system may therefore play a key role in the clinical response to 5-azacytidine in patients with hematologic cancer.

Abnormal silencing of tumor suppressor genes by DNA methylation has long been suggested to be the most likely target for epigenetic therapy, because reactivation of such genes by epigenetic drugs might inhibit the growth of cancer cells (6, 7, 32). However, most tumor suppressor genes identified in the NCG 5.0 database (n = 38; ref. 26) did not show significant upregulation in GSEA (Fig. 1B; Supplementary Fig. S3A). Interestingly, both CDKN2A and CDKN2B (which encode the cyclin dependent kinase inhibitors p16 and p15, respectively) were upregulated at low levels in responders (Supplementary Figs. S3A and S3B), but not in nonresponders (Supplementary Fig. S3B). Earlier studies have reported abnormal silencing of CDKN2B (but not CDKN2A) by promoter hypermethylation in patients with myeloid cancer (32). The mechanisms of CDKN2A reactivation in this study are unclear, but CDKN2B protein expression can be upregulated by promoter demethylation during DNMTi treatment (33). Although the reactivation of CDKN2A and CDKN2B might also influence the differential clinical responses to 5-azacytidine treatment, the absolute expression levels of the genes were relatively low following 5-azacytidine treatment in responders (median of RPKM; CDKN2A = 0.37, CDKN2B = 0.30). Therefore, activation of the innate immune system is more likely associated with the clinical responses. This is supported by recent findings in cancer cell lines (11, 12) showing that knocking down genes in the type I IFN signaling pathway can abolish the antitumor effect of DNMTis, suggesting that the activation of the innate immune system may play a key role in DNMTi effectiveness as therapeutic drugs.

The derepression of CTAs by DNMTis has also been associated with activation of an immune response (8, 9). However, GSEA showed that the CTAs were not significantly upregulated following 5-azacytidine treatment in responders (Fig. 1C). Our data therefore suggest that CTAs are unlikely to trigger immune response genes in this group of patients.

We also applied unsupervised analyses for type I IFN signaling pathway genes (GO:0060337), which are a subset of innate immune response genes (GO:0045087). These genes were preferentially upregulated in all types of malignant cells from responders in both the Nordic and Australian patient cohorts (Supplementary Figs. S4A–S4G). When the RNA-seq data were assembled into one data set, a differential upregulation of the genes in responders was also apparent (Fig. 2A and B; P < 0.001). The RNA-seq data included the genes for IFI27, a type I IFN inducible protein, and IRF4, an IFN regulatory transcription factor (Fig. 2C; P < 0.05). Responder-specific upregulation was also found in the leukocyte activation genes LEF1, encoding a T-cell receptor-alpha enhancer-binding transcription factor, and CCL5, encoding a chemokine also known as RANTES (Fig. 2D; P < 0.05). However, the upregulation of type I IFN signaling pathway genes and leukocyte activation genes was not linked to clinical responses in all patients, and other molecular changes may also be important for the responses to 5-azacytidine. Furthermore, some of the type I IFN signaling pathway genes showed downregulation following 5-azacytidine treatment by an unknown mechanism (Fig. 2A; Supplementary Figs. S4B and S4G). The average mapping rate in the samples from the Nordic patients (average; 91%, highest; 94%, lowest; 77%) was comparable to that of Australian patients (average: 83%; highest: 87%; lowest: 78%). Because, the variation in mapping rates might have effects on the results obtained for expression changes of type I IFN signaling pathway genes, we reanalyzed the data using only those samples with high mapping rates (>90%). The preferential upregulation of type I IFN signaling pathway genes in responders was confirmed (Supplementary Fig. S4H). Thus, the variations in RNA quality as seen in the mapping rates did not change the conclusion.

Recent studies in cell lines and mouse models have reported that the innate immune system is activated by the presence of transcripts from LTRs, an effect we have called “viral mimicry” (11, 12). However, non-LTR retrotransposons such as long and short interspersed nuclear elements (LINE and SINE) may also affect the activation of the antiviral innate immune response (34, 35). Chiappinelli and colleagues (11) showed a correlation between the expression of LTRs and of type I IFN signaling pathway genes in patients with epithelial ovarian cancer without any history of anticancer treatment in a study based on The Cancer Genome Atlas (TCGA) consortium data. Unnikrishnan and colleagues (17) showed that immune and inflammatory response genes were induced by 5-azacytidine treatment in CMML and MDS responders. Furthermore, we previously reported the upregulation of LTRs by 5-azacytidine treatment for separated CD34⁺ stem and progenitor cells cultured in vitro from bone marrow cells of patients with MDS (36). However, there are no reports of the upregulation of TEs (including LTRs, LINEs, and SINEs) in patients with cancer after DNMTi treatment. Because TEs are widely distributed throughout the human genome, we simplified the analyses by focusing on TEs in intergenic regions, as we have done in a previous comprehensive in vitro analysis (Supplementary Figs. S5–S7; ref. 20).

Although numerous TEs were upregulated in the patient samples following 5-azacytidine treatment, as reported for cancer cell lines (19, 20, 37), we found no significant differences between nonresponders and responders in terms of the global upregulation of TEs, that is, the total number of TE transcripts was not correlated with activation of the innate immune system (Supplementary Figs. S5B–S5C, S6B–S6C, and S7B–S7C). However, as we previously found that evolutionarily young LTRs might be responsible for activation of the innate immune system in vitro (20), we examined the evolutionary ages of responder-specific upregulated TEs (Fig. 3A) based on levels of genetic divergence (Fig. 3B), which can be considered indices of evolutionary age (38, 39). The responder-specific upregulated LTRs had significantly lower divergence levels than LTRs that were not upregulated (P < 0.001). These results are remarkably in line with our previous in vitro data, which identified evolutionarily young LTRs as potential triggers for the activation of the innate immune system (20). Moreover, evolutionarily young LINEs and SINEs were also found in the responder-specific upregulated elements (Fig. 3B). We also observed specific classes of evolutionarily young TEs in responder-specific upregulated TEs in both CD14⁺ cells from patients with Nordic CMML at C1D5 (cycle 1, day 5) and mononuclear cells from patients with Nordic MDS at C3D1 (Supplementary Fig. S8; P < 0.001).

We next categorized the responder-specific upregulated TEs into different TE families (Supplementary Figs. S5D, S6D, and S7D). Some specific families were upregulated (FC ≥ 2) in 5-azacytidine responders, such as LTR12C and LTR7 (Supplementary Figs. S5D and S5E). We previously showed that LTR12 superfamilies such as LTR12C and LTR12 might play key roles in activation of the type I IFN signaling pathway, because these evolutionarily young super-families were among the most abundantly expressed LTRs in cancer cell lines treated with DNMTis (19, 20, 37). Ahmad and colleagues (34) reported that cytoplasmic Alu:Alu hybrids, overexpressed as a result of adenosine deaminase ADAR1 deficiency, can act as endogenous ligands for the viral RNA sensor MDA5. These earlier reports and the present study suggest that not only 5-azacytidine-induced evolutionarily young LTRs, but also LINEs and SINEs, may potentially trigger activation of the innate immune system in responders to 5-azacytidine treatment.

As shown in the Nordic and Australian cohorts of adult patients with aggressive myeloid cancers, activation of the innate immune system is responder-specific. To extend the generality of this finding, we analyzed samples from pediatric leukemia patients (AML, n = 11; ALL, n = 3; Supplementary Table S1; Fig. 4A and B; Supplementary Figs. S9, S10A–S10C) from The Therapeutic Advances in Childhood Leukemia & Lymphoma (TACL) consortium (18). We were not able to perform total RNA-seq on these samples; therefore, we used RT-qPCR for selected immune response genes, leukocyte activation genes, and LTRs.

Consistent with the results from the adult myeloid cancers, both the innate immune system genes (IFI27 and LEF1) and evolutionarily young LTRs (LTR12C and LTR7) were significantly upregulated in the eight responders (P < 0.05), but not in the six nonresponders, of the pediatric cohort (Fig. 4A and B). Upregulation of CDKN2A and CDKN2B were also observed in some responders (P < 0.01), but the absolute expression of these two induced genes following 5-azacytidine treatment in responders was low (Supplementary Fig. S9). The low expression, as observed in some patients with ALL, may be due to genomic deletions of the genes as reported with previously (40). Further, CDKN2B was also increased in some nonresponders by 5-azacytidine treatment, although this was not significant (P = 0.125). The pediatric patients also received chemotherapy (fludarabine and cytarabine) along with 5-azacytidine. It is possible that the combination treatment may have an effect on enhanced innate immune activity, but the activation of the innate immune system could still distinguish a variable clinical response to anticancer therapies. These results support the hypothesis that the activation of the innate immune system is associated with clinical responses to 5-azacytidine across different types of hematological malignancies independently of exact treatment conditions, suggesting a common mechanism of action.

Our analyses in treated patients suggested that upregulated TEs might be responsible for the activation of the host innate immune system. To confirm that specific examples of these LTRs could in fact trigger viral mimicry, in vitro transcribed RNAs for eight selected LTRs strongly induced in responders, were transfected into the HCT116 colon cancer cell line, which we have previously shown is competent to respond to induced TE expression (Fig. 5; ref. 12). The expression of type I IFN genes was increased by transfection of all of the LTRs. However, the abilities of individual LTRs to induce genes in the type I IFN pathway were quite variable. For example, MSTB, MTL2A2, and LTR7 induced the highest levels of expression of IFIT2, whereas LTR12D, LTR57, and LTR6B induced the highest levels of IFI6. The abilities of individual LTRs to induce specific viral mimicry responses are therefore highly dependent on the LTR. The tested in vitro transcribed LTRs are single-stranded RNAs; therefore, the variable responses for LTR-derived RNA molecules may be due to differences in secondary structures such as hairpin loops or RNA modifications such as tri-phosphorylation (41). Further studies are needed to elucidate the potential function of the responder-specific LTRs. Nevertheless, the data provide mechanistic support for the concept that activation of LTRs in patients is capable of triggering an innate immune response.

Recent studies in vitro have shown that the innate immune system is activated by the presence of transcripts from LTRs and other types of TEs (20, 34, 35). Here, our analysis of gene expression profiles in patients before and after 5-azacytidine treatment shows responder-specific activation of the innate immune system in three heterogenous cohorts of patients with different types of hematologic malignancies, one of which was solely pediatric patients. The activation of the innate immune system is likely due to the induced expression of evolutionarily young TEs. This is possibly because evolutionarily young TEs are more dependent on DNA methylation for silencing and hence show a more prominent induction by DNMTis, as described in our previous in vitro study (20). We and others have recently shown that disruption of genes in the type I IFN signaling pathway abolishes the antitumor effect of DNMTis (11, 12). Moreover, the potential functions of evolutionarily young LTRs, LINEs, and other noncoding RNAs as tumor-intrinsic triggers of innate immune signaling pathways have been reported in earlier studies (20, 34, 35). Chiappinelli and colleagues (11) showed the upregulation of type l IFN signaling pathway genes following overexpression of plasmid vectors containing LTRs. Further, Guler and colleagues (35) revealed that silencing of evolutionarily young LINEs using siRNAs targeting conserved regions of the LINEs contributed to the establishment of lethal cancer cells that were tolerant of antitumor drugs. We therefore suggest that the viral mimicry pathway plays a role in the patients' response to treatment with 5-azacytidine. Although our data in vitro also suggested that evolutionarily young LTRs are responsible for the activation of the host innate immune system, further evidence will be needed to confirm this finding.

Additional molecular mechanisms may clearly also be important for the clinical response to 5-azacytidine. Numerous studies have reported that reactivation of silenced tumor suppressor, DNA repair, or cell differentiation genes could normalize growth behavior in DNMTi-treated cells (42, 43). Moreover, upregulation of CTAs following DNA demethylation might increase the immunogenicity of, and consequently induce apoptosis of, cancer cells (5). In this study, the activation of these gene sets was not associated with the clinical response to 5-azacytidine. However, the gene ontology analysis and GSEA were performed on the FC of median expression values in all 26 patients. Therefore, the reactivation of silenced tumor suppressor, DNA repair, and differentiation genes might also contribute to an improved clinical response in some patients.

Our study was limited to 40 patients who had hematologic cancers and were grouped according to their clinical response rather than classical immuno-morphological classification subtypes. We were able to confirm a statistically robust responder-specific activation of the innate immune system in two adult cohorts (Nordic and Australian) and one pediatric cohort of leukemia patients and in different cell types. The fact that 5-azacytidine may target the many thousands of TEs, which make up a large proportion of the human genome, may indicate a generalized mechanism of action. Therefore, the statistically significant correlation between viral mimicry pathway activation and drug response also suggests the possibility of a common mechanism in different cell types.

Our results support the concept that the activation of the innate immune system following 5-azacytidine treatment is important in patients. Further large-scale prospective clinical and translational studies are indeed required to clearly validate this association, and we have recently opened such a clinical trial (#NCT03999723). Our data also show for the first time that although all three major classes of TEs (LTRs, LINEs, and SINEs) are upregulated following 5-azacytidine treatment, only specific classes of evolutionarily young TEs seem to be associated with induction of the innate immune system and clinical responses to 5-azacytidine. Therefore, we suggest that the activation of evolutionarily young TEs and the innate immune system may be common factors that are important for the clinical responses to 5-azacytidine.

P.A. Jones is a consultant at Zymo Corporation and is a reviewer at CPRIT. K. Grønbæk is an advisory board member at Celgene and is an advisory board member/teacher at Otsuka Pharma. No potential conflicts of interest were disclosed by the other authors.

Conception and design: H. Ohtani, A.D. Ørskov, A.S. Helbo, M. Liu, P.A. Jones, K. Grønbæk

Development of methodology: H. Ohtani, A.S. Helbo, M. Liu, G. Liang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.D. Ørskov, A.S. Helbo, L. Gillberg, J. Ungerstedt, E. Hellström-Lindberg, W. Sun, G. Liang, K. Grønbæk

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Ohtani, A.D. Ørskov, A.S. Helbo, W. Zhou, G. Liang, K. Grønbæk

Writing, review, and/or revision of the manuscript: H. Ohtani, A.D. Ørskov, A.S. Helbo, L. Gillberg, M. Liu, J. Ungerstedt, E. Hellstrom-Lindberg, W. Sun, G. Liang, P.A. Jones, K. Grønbæk

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.D. Ørskov, K. Grønbæk

Study supervision: M. Liu, P.A. Jones, K. Grønbæk

We thank the Van Andel Research Institute (VARI) Genomics and Bioinformatics Core; David Nadziejka for technical editing of the article; and all members in Grønbæk and Jones group for discussion. This work was supported by the NCI (grant no. R35CA209859), the Folz Fund of the Grand Rapids Community Foundation, and by Van Andel Institute through the Van Andel Institute – Stand Up to Cancer Epigenetics Dream Team. Stand Up to Cancer is a division of the Entertainment Industry Foundation. Research grants are administered by the American Association for Cancer Research, the Scientific Partner of SU2C. The study was also funded by the Lundbeck Foundation (grant no. R180-2014-3675), the Danish Cancer Society (grant no. R124-A7695), and the Novo Nordisk Foundation (grant no. NNF13OC0003435). The Grønbæk Lab is funded by center grants from The Danish Cancer Society (Danish Research Center for Precision Medicine in Blood Cancer; grant no. 223-A13071-18-S68), the Novo Nordisk Foundation (Novo Nordisk Foundation Center for Stem Cell Biology, DanStem; grant no. NNF17CC0027852), and from Greater Copenhagen Health Science Partners (Clinical Academic Group in Translational Hematology).

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.