IDH1/2 Mutations Sensitize Acute Myeloid Leukemia to PARP Inhibition and This Is Reversed by IDH1/2-Mutant Inhibitors

Somatic mutations in genes encoding for isocitrate dehydrogenase 1 and 2 (IDH1/2^MUT) occur in various types of cancer, such as glioma, cholangiocarcinoma, and certain myeloid neoplasms, including acute myeloid leukemia (AML), myelodysplastic syndromes, and myeloproliferative neoplasms (1–5). Wild-type IDH1/2 (IDH1/2^WT) converts isocitrate to α-ketoglutarate (αKG) with concomitant reduction of NADP⁺ to NADPH. IDH1/2^MUT result in a neomorphic function, where IDH1/2^MUT enzymes convert αKG and NADPH to 2-hydroxyglutarate (D2HG) and NADP⁺ (6). D2HG accumulation is oncogenic because it inhibits various αKG-dependent dioxygenases involved in epigenetic regulation, thus inducing cellular dedifferentiation and leukemogenesis (7, 8). Appreciation of the causative role of IDH1/2^MUT in AML formation and maintenance (9–11) led to the development of agents such as the IDH1^MUT inhibitor ivosidenib (AG-120; ref. 12) and the IDH2^MUT inhibitor enasidenib (AG-221), which was recently FDA approved for the treatment of relapsed/refractory IDH2^MUT AML (13, 14). Although ivosidenib or enasidenib monotherapy was effective in some patients with difficult-to-treat AML, the majority of treated patients either did not have deep responses, or did not have durable responses, indicating the need to combine these drugs with other antileukemic agent(s) (14–16).

Other effects of D2HG besides inhibition of αKG-dependent dioxygenases include the inhibition of the DNA repair enzyme alkB homolog (ALKBH; refs. 17, 18) and the DNA damage response proteins lysine-specific demethylase 4A/B (KDM4A/B; refs. 19–21) and ataxia-telangiectasia mutated (ATM; ref. 22). Decreased ATM function, due to mutational inactivation, transcriptional repression or posttranslational depletion, leads to decreased DNA double-strand break (DSB) repair (23), increased DNA damage, and sensitivity to DNA repair inhibitors, such as PARP inhibitors, in prostate (24), breast (25), colorectal (26) and lung cancers (27), and lymphomas (28). Accordingly, compared with IDH1/2^WT cells, IDH1/2^MUT cells show increased levels of DNA damage and sensitization to olaparib and talazoparib, either as monotherapy or in combination with DNA-damaging agents (21, 22, 29, 30). These results have been described using genetically engineered cancer cells or murine hematopoietic stem cells (HSC), but not using models relevant for human AML. We investigated the levels of DNA damage and sensitivity to PARP inhibitors and DNA damage–inducing chemotherapy in IDH1^MUT, IDH2^MUT, and IDH1/2^WT primary AML cells.

Peripheral blood and bone marrow samples were obtained from AML patients treated in the Cleveland Clinic (Cleveland, OH). Diagnosis was confirmed according to the 2008 WHO classification criteria. These samples were subjected to next-generation sequencing (NGS) and copy number variation (CNV) analysis targeting approximately 60 genes that are frequently mutated and/or lost in AML and genes involved in DNA damage responses, including IDH1/2, TET2, ATM, BRCA1, BRCA2, XRCC2-5, and RAD50-52. Cancer and germline DNA was obtained from AML cells and paired CD3⁺ T cells or buccal swabs, respectively. Sequencing and bioinformatic analyses were conducted as described previously (4). Variant allelic frequencies (VAF) were calculated as the fraction of mutated reads divided by the total number of reads for the gene. VAFs were adjusted to CNVs at the locus of each mutation. Informed consent was obtained from patients according to protocols approved by the Cleveland Clinic Institutional Review Board and in accordance with the Declaration of Helsinki. Clinical details of the patients were obtained from their medical records.

From the AML patients genotyped by NGS, we selected those with the following somatic mutation configurations: IDH1^MUT/IDH2^WT/TET2^+/+, IDH1^WT/IDH2^MUT/TET2^+/+, IDH1^WT/IDH2^WT/TET2^−/− or IDH1^WT/IDH2^WT/TET2^+/+, hereafter referred to as IDH1^MUT, IDH2^MUT, TET2^−/−, and IDH1/2^WT AML samples, respectively. Using copy number–adjusted VAFs, the clonal architecture of IDH1^MUT, IDH2^MUT, and TET2^−/− AML samples was reconstructed and cases wherein the classifying mutations were clonal/ancestral with a mutational load of >80% were selected (n = 5 primary human AML samples for each group). Only IDH1/2^MUT that are known D2HG producers were included. TET2^−/− patients included those with hemizygous or homozygous TET2 mutations (TET2^MUT/− or TET2^MUT/MUT).

In all cell culture experiments, primary human AML cells from the aforementioned bone marrow samples were cultured in Iscove's Modified Dulbecco's Medium (Gibco, Life Technologies, Thermo Fisher Scientific) supplemented with 10% FBS (HyClone, Thermo Fisher Scientific), 10 ng/mL IL3, 50 ng/mL stem cell factor, 3 U/mL erythropoietin, and 10 ng/mL granulocyte-macrophage colony–stimulating factor in 5% CO₂ at 37°C and were simultaneously used in various experiments (Supplementary Fig. S1). For colony formation assays (CFA), cells pretreated in the presence or absence of AGI-5198 (the preclinical version of the IDH1^MUT inhibitor ivosidenib; ref. 12), AGI-6780 (the preclinical version of the IDH2^MUT inhibitor enasidenib; ref. 31), 10 mmol/L D2HG, or 5 μmol/L N-acetyl cysteine (NAC) were seeded at a density of 1 × 10⁴ to 1 × 10⁶ cells/mL, in 3 mL Methocult methylcellulose medium (Stem Cell Technologies). The seeding density depended on the concentration of the cytotoxic agent. Cells were treated for 48 hours with 10 to 50 nmol/L daunorubicin, 200 to 1,000 nmol/L cytarabine, 10 to 50 μmol/L 5-azacytidine, 1 to 10 μmol/L decitabine (all for 48 hours), or 2 to 6 Gy ionizing radiation (IR) using a ¹³⁷Cs source. PARP inhibitors (0 to 25 μmol/L olaparib or 0 to 25 nmol/L talazoparib) were given during 48 hours before the start of the CFA and for 7 days during the CFA. Thus, treatment with all cytotoxic agents lasted for at least 48 hours, in which period >99% of investigated AML cells underwent at least one cell cycle (Supplementary Fig. S2). Isogenic HCT116 IDH1^WT/WT and IDH1^WT/R132H knock-in cells, generated by AAV-targeting technology GENESIS (32), were kindly provided by Horizon Discovery Ltd. and cell culture and CFAs were performed as described previously (30). Colonies (>50 cells) were counted at 7 days after treatment, and results were analyzed to determine the clonogenic fraction. This is the number of colonies counted, divided by the number of cells plated and corrected for the plating efficiency, as described previously (30). Cell survival at 3 days after treatment was determined by MTT assays. AGI-5198 and AGI-6780 were purchased from MedChemExpress. D2HG, NAC, 5-azacytidine, cytarabine, daunorubicin, decitabine, and MTT were purchased from Sigma-Aldrich. Olaparib (AZD-2281) and talazoparib (BMN-673) were purchased from SelleckChem.

Quantitative enzyme cytochemistry (metabolic mapping) of AML cells was performed and analyzed as described previously (30, 33, 34). The specific NADP⁺-dependent IDH1/2 activity (EC# 1.1.1.42), NAD⁺-dependent IDH3 activity (EC# 1.1.1.43), and NADP⁺-dependent activity of glucose-6-phosphate dehydrogenase (G6PD; EC# 1.1.1.49) were determined against 10 mmol/L isocitrate or glucose-6-phosphate (Serva) and 3 mmol/L NAD⁺ or 0.8 mmol/L NADP⁺ (Boehringer) in the presence of nitroblue tetrazolium chloride (Sigma-Aldrich). Incubation was performed at 37°C for 60 minutes. Control reactions were performed in the absence of substrate but in the presence of cofactors to assess nonspecific enzyme activity. To detect the impact of cytoplasmic IDH1^MUT and mitochondrial IDH2^MUT on NADP⁺-dependent IDH1/2 activity, 1-methoxy-5-methylphenazinium methylsulfate (methoxy-PMS) and 5-methylphenazinium methylsulfate (PMS, both Sigma) were used, respectively, because the former does not pass mitochondrial membranes, whereas the latter does (34). Photomicrographs were made on a Leica microscope at ×40 magnification using Qwin software.

Enantiomer-specific mass spectrometry analysis of D2HG levels of AML cell lysates was performed as described before (35).

AML cells were analyzed using a colorimetric NADP⁺:NADPH ratio assay (Abcam), a fluorometric GSH:GSSG ratio assay (Abcam), and a fluorometric CellROX Deep Red ROS assay (Life Technologies), in 96-well plates using a POLARStar Galaxy microplate reader (BMG Labtech).

qRT-PCR was performed as described previously (36). Each sample was assayed in triplicate and normalized to ABL expression (37). Primers are listed in Supplementary Table S1.

ATM protein expression was measured by immunoblotting using primary antibodies against ATM (Genetex) and β-actin (Cell Signaling Technology). Immunoblots were analyzed using a Li-Cor Odyssey system (Li-Cor Biotechnology). Two sets of “Silencer Select” siRNAs against ATM mRNA (s530444 and s5304445) and one negative control siRNA (#4390843) were obtained from Life Technologies and transfected into AML cells using standard protocols. siRNA efficacy was confirmed by immunoblotting against ATM. ATM-siRNA s530444 was selected to be used in CFAs.

IDH1, IDH2, and TET2 mutational data and ATM, IDH1, IDH2, and G6PD mRNA expression data (RNASeq v2 RSEM or RPKM) for AML, low-grade glioma, and glioblastoma cases were extracted from The Cancer Genome Atlas (TCGA) via cBioPortal (38, 39) and correlated with each other as described previously (40).

DNA DSBs were determined using immunofluorescence staining of γH2AX (Millipore). The number of γH2AX⁺ foci per cell was quantified from deconvoluted stacks of photomicrographs using custom-made software, as described previously (4).

Data were processed and analyzed using R and visualized using GraphPad Prism. Two-sided tests were used with significance defined as α < 0.05.

The clinical, cytogenetic, and molecular characteristics of the selected IDH1^MUT, IDH2^MUT, TET2^−/−, and IDH1/2^WT AML patient samples (n = 5 for each group) are shown in Supplementary Tables S2–S4. The clinical characteristics of the selected IDH1^MUT and IDH2^MUT AML patients were representative for those described in a previous cohort study of IDH1/2^MUT AML patients (4).

Motivated by earlier reports that genetically engineered and primary IDH1/2^MUT cancer cells have decreased levels of ATM expression (22) and increased levels of DNA damage (21, 22, 29, 30), we investigated these phenomena in primary human AML cells. We observed decreased ATM mRNA and protein expression in IDH1/2^MUT AML cells as compared with IDH1/2^WT AML cells. Administration of an IDH1/2^MUT inhibitor restored ATM expression in IDH1/2^MUT AML cells. ATM mRNA expression in TET2^−/− cells was not significantly lower than in TET2^+/+ AML cells (Fig. 1A). We determined D2HG concentrations in cell lysates of each IDH1/2^MUT sample. D2HG concentrations were higher in IDH1^MUT AML cells than in IDH2^MUT AML cells, as has been described previously (41), and were potently suppressed by AGI-5198 and AGI-6780, respectively (Fig. 1B). Using TCGA data, we confirmed that ATM mRNA expression is severely decreased in IDH1^MUT AML and not significantly decreased in IDH2^MUT and TET2^−/− AML (Fig. 1C). We observed more γH2AX⁺ foci (which recognize DNA DSBs) in IDH1/2^MUT than in IDH1/2^WT AML cells under steady-state conditions. Furthermore, the number of γH2AX⁺ foci was higher in IDH1/2^MUT than in IDH1/2^WT AML cells after daunorubicin treatment. To confirm the causal relationship between IDH1/2^MUT and increased levels of DNA damage, we pretreated IDH1/2^MUT cells with an IDH1/2^MUT inhibitor prior to treatment with IR or daunorubicin, which reversed the number of γH2AX⁺ foci in IDH1/2^MUT to levels observed in IDH1/2^WT AML cells in a time-dependent fashion (Fig. 1D).

The relationship between increased DNA damage, decreased ATM function, and sensitivity to PARP inhibitors (24–28, 42) prompted us to compare the responses of IDH1/2^MUT and IDH1/2^WT AML cells to the PARP inhibitors olaparib and talazoparib. After treatment with olaparib or talazoparib, the surviving fraction of IDH1/2^MUT AML cells was lower than that of IDH1/2^WT AML cells in CFAs (Fig. 2A and B). To investigate whether a causal relationship existed between IDH1/2^MUT and this sensitization to PARP inhibitors, we pretreated IDH1^MUT AML cells with AGI-5198 and IDH2^MUT AML cells with AGI-6780 before cytotoxic treatment (Fig. 2C). Pharmacologic inhibition of IDH1/2^MUT for at least 7 days protected IDH1/2^MUT AML cells against PARP inhibitors (Fig. 2D–G). In addition, IDH1/2^MUT inhibitors did not affect the sensitivity of IDH1/2^WT AML cells to PARP inhibitors (Fig. 2H and I). We also observed reversible sensitivity to PARP inhibitors using another model of isogenic IDH1^WT/R132H HCT116 cells, as compared with IDH1^WT/WT HCT116 cells (Supplementary Fig. S3).

Given that IDH1/2^MUT decrease the DNA damage response and cause sensitivity to PARP inhibitors, we hypothesized that IDH1/2^MUT also sensitize AML cells to other DNA damage–inducing agents. Relative to IDH1/2^WT and TET2^−/− AML cells, we observed a significantly reduced surviving fraction of IDH1/2^MUT AML cells after treatment with daunorubicin or IR in CFAs (Fig. 3A and B). In addition, pharmacologic inhibition of IDH1/2^MUT for at least 7 days protected IDH1/2^MUT AML cells, but not IDH1/2^WT AML cells, against subsequent treatment with daunorubicin or IR (Fig. 3C–H). We confirmed these results in isogenic IDH1^WT/WT and IDH1^WT/R132H HCT116 colorectal cancer cells (Supplementary Fig. S4). Pretreatment with the ROS scavenger NAC during 3 days did not affect the survival of IDH1/2^MUT AML cells after treatment with daunorubicin or IR (Supplementary Fig. S5). We did not observe survival differences between IDH1/2^MUT and IDH1/2^WT AML cells after treatment with cytarabine, 5-azacytidine, or decitabine, which are antimetabolites and hypomethylating agents that do not induce DNA damage. We also did not observe survival differences between IDH1/2^MUT and IDH1/2^WT AML cells after treatment with daunorubicin or IR in short-term (3-day) cell viability assays that do not capture the long-term effects of treatment-induced DNA damage as adequately as CFAs (Supplementary Fig. S6; ref. 43).

We hypothesized that combined treatment with a PARP inhibitor and a DNA-damaging agent has additive effects on IDH1/2^MUT AML cells. Combined treatment with olaparib or talazoparib and daunorubicin was more lethal to both IDH1/2^WT and IDH1/2^MUT AML cells than daunorubicin treatment alone, but the effect was significantly larger in IDH1/2^MUT AML cells (Fig. 4).

To investigate causality between IDH1/2^MUT, ATM suppression, and therapy sensitivity, we knocked down ATM in AML cells using siRNA (Supplementary Fig. S7). ATM knockdown did not affect the sensitivity of IDH1/2^MUT AML cells to daunorubicin or IR (Fig. 5A–D) but sensitized IDH1/2^WT AML cells to these treatments (Fig. 5E and F). After 7 days of pretreatment with D2HG, untransfected IDH1/2^WT AML cells were sensitized to daunorubicin or IR, but IDH1/2^WT AML cells were not further sensitized when ATM was knocked down (Fig. 5G and H). IDH1/2^MUT inhibitors protected untransfected IDH1/2^MUT AML cells against daunorubicin or IR (Fig. 3C–F), but did not protect IDH1/2^MUT AML cells when ATM was knocked down (Fig. 5I and J). Another siRNA with a lower knockdown efficiency of siRNA sensitized IDH1/2^WT AML cells less for daunorubicin or IR (Supplementary Fig. S8).

In glioma and colorectal cancer cells, IDH1^MUT inhibits IDH1/2^WT function, which perturbs redox states and sensitizes these cells to irradiation (30). We interrogated whether or not IDH1/2^WT function and redox states play a role in the therapy sensitization of IDH1/2^MUT AML cells. NADP⁺-dependent IDH1/2 activity was significantly lower in IDH1/2^MUTAML cells than in IDH1/2^WT and TET2^−/− AML cells. However, the impact of the decreased IDH1/2-mediated NADPH production capacity on the total cellular NADPH production capacity in AML cells was limited, because the NADPH production capacity by G6PD was approximately 4-fold larger than that of IDH1 and IDH2 combined (Fig. 6A). IDH1/2^MUT were not associated with changes in NAD⁺-dependent IDH3 activity or NADP⁺-dependent G6PD activity in AML cells (Fig. 6A and B). Pretreatment with an IDH1/2^MUT inhibitor for 3 days restored NADP⁺-dependent IDH1/2 activity in IDH1/2^MUT AML cells (Fig. 6C). In addition, D2HG administration of 10 mmol/L D2HG [which achieved D2HG levels in IDH1/2^WT AML cells that were similar to untreated IDH1/2^MUT AML cells (Fig. 1B)] decreased NADP⁺-dependent IDH1/2 activity in IDH1/2^WT cells, which supports a causative role of D2HG accumulation in the suppression of IDH1/2-mediated NADPH production (Fig. 6C). In agreement with the modest effects of IDH1/2^MUT on the total cellular NADPH production, we observed similar NADP⁺:NADPH ratios, GSH:GSSG ratios and ROS levels between IDH1/2^MUT and IDH1/2^WT AML cells under steady-state conditions and after pretreatment with daunorubicin or IR (Fig. 6D–F). In TCGA data, the mRNA expression of IDH1, IDH2, and G6PD enzymes was unchanged in IDH1/2^MUT versus IDH1/2^WT AML, whereas mRNA expression of these enzymes was lower in IDH1/2^MUT versus IDH1/2^WT glioma (Supplementary Fig. S9).

We found that primary IDH1/2^MUT AML cells have reduced DNA damage responses and suppressed expression of ATM. As a consequence, they are sensitized to a PARP inhibitor, daunorubicin, or IR, and this is negated by pretreatment with an IDH1/2^MUT inhibitor, which also restores ATM expression and decreases DNA damage. In mechanistic experiments using siRNA and exogenous D2HG, we obtained further evidence of a cascade wherein D2HG accumulation leads to ATM suppression and decreased DNA damage responses, resulting in increased IDH1/2^MUT AML therapy responses. Although our results suggest that PARP inhibitors enhance responses of IDH1/2^MUT AML to daunorubicin, they also suggest that PARP inhibitors or daunorubicin should not be combined with IDH1/2^MUT inhibitors in AML, because IDH1/2^MUT inhibitors disrupt the D2HG–ATM–DNA damage cascade. These findings, in combination with earlier findings in IDH1^MUT glioma, are summarized in a model shown in Fig. 5I.

Our results corroborate other studies showing that compared with IDH1/2^WT counterparts, IDH1(/2)^MUT human glioma, human colorectal cancer, and murine HSCs are sensitized to treatment with daunorubicin, IR, or PARP inhibitors due to ATM suppression and increased DNA damage levels (21, 22, 29, 30). Mechanistic studies have provided evidence that IDH1/2^MUT decrease ATM expression by increasing methylation of the repressive histone mark H3K9 that may rely on inhibition of the histone demethylases KDM4A and/or KDM4B by D2HG (21, 22). TET2 is a major downstream target of D2HG in IDH1/2^MUT AML (44), but ATM downregulation was not observed in TET2^−/− mice (22), nor was homologous recombination significantly impaired after treatment of U2OS DR-GFP cells with an siRNA against TET2 (21). This is supported by the finding that restoration of TET2 function sensitizes, rather than protects, TET2^+/− AML cells to PARP inhibitors (45), and it may also be supported by our finding that ATM mRNA and protein expression were not different in TET2^−/− AML cells as compared with TET2^+/+ AML cells.

Several mechanistic results from IDH1/2^MUT AML cells in the current study contrast our earlier findings in IDH1^MUT glioma and colon carcinoma cells, where pretreatment with an IDH1^MUT inhibitor or the ROS scavenger NAC for 3 days achieved radioprotection due to restored NADPH production and ROS detoxification (30). In IDH1/2^MUT AML cells, such protection against cytotoxic therapy required incubation with an IDH1/2^MUT inhibitor for 7 days and was not achieved by using NAC. Similar to the profound metabolic effects of IDH1/2^MUT in glioma (40, 46), IDH1/2^MUT reduced IDH1/2-mediated NADPH production in IDH1/2^MUT primary AML cells. However, this did not affect therapy responses in AML cells wherein IDH1/2 provides <20% of the cell's NADPH; in contrast, in glioma, IDH1/2 provides approximately two thirds of the cell's capacity to produce NADPH (47). IDH1/2^MUT were associated with decreased mRNA expression of IDH1/2 and G6PD in glioma but not in AML, suggesting that IDH1/2^MUT may alter the metabolism of AML cells to a lesser extent than that of glioma. Relative to glioma and colon carcinoma cells, slower protection of IDH1/2^MUT AML cells by IDH1/2^MUT inhibitors is likely due to it being mediated by a different mechanism that involves slow epigenetic alterations needed to suppress ATM expression. Reversing redox states in glioma and colon carcinoma cells is likely to be much faster. Theoretically, the increased DNA damage in IDH1/2^MUT AML cells can be explained by differences in cell doubling times between IDH1/2^MUT and IDH1/2^WT cells (48), and IDH1^MUT inhibitors are reported to affect cell cycle duration (49). However, we found no differences in doubling time between IDH1/2^MUT and IDH1/2^WT AML cells and increased γH2AX⁺ foci argue against cell-cycle perturbations as being responsible for our results.

Patients with IDH1/2^MUT glioma have longer survival times than IDH1^WT counterparts (2, 50, 51), probably by virtue of improved responses to chemotherapy and IR (30, 52). Although our data suggest that IDH1/2^MUT AML cells are sensitive to chemotherapy and IR, there is no difference between the survival of patients with IDH1/2^MUT AML or IDH1/2^WT AML (4, 53). This discrepancy between our results and the data from observational studies might be inherent to limitations of our in vitro data, such as a relatively small sample size, the inclusion of AML with ancestral IDH1/2^MUT only or different behavior of IDH1/2^MUT AML cells in vitro and in vivo. The latter issue would be interesting to investigate in a translational clinical trial wherein the therapy response of IDH1/2^MUT AML patients (in vivo) and their primary samples (in vitro) are compared. Alternatively, the aforementioned discrepancy may be inherent to limitations of observational studies. Of note, high-dose IR is nowadays rarely used in the treatment of AML and only 30% to 40% of elderly AML patients (aged ≥65 years) are reported to receive any type of chemotherapy, of which not all receive intensive treatment regimens, such as daunorubicin (54, 55). Indeed, examining data from the NCI's Surveillance, Epidemiology and End Results (SEER) program, we found that only approximately 60% of AML patients of all ages receive any type of chemotherapy (see Supplementary Data). As a possible explanation for the absence of survival differences between patients with IDH1/2^MUT AML or IDH1/2^WT AML, low use of daunorubicin and IR may prevent the putative predictive effects of IDH1/2^MUT to materialize into a significant prognostic association in retrospective studies.

In summary, this study is the first to show that IDH1/2^MUT AML is vulnerable to PARP inhibition as monotherapy, but especially when combined with daunorubicin treatment. IDH1/2^MUT inhibitors protect IDH1/2^MUT AML cells against PARP inhibitors, daunorubicin, or IR, which suggests that combinations of IDH1/2^MUT inhibitors and DNA-damaging agents should be avoided. Our data are crucial to the rational design and analysis of clinical trials with IDH1/2^MUT inhibitors, especially for clinical trials that investigate combinations of IDH1/2^MUT inhibitors with conventional chemotherapy for AML (e.g., ClinicalTrials.gov NCT02632708). Instead, our results show that exploiting impaired DNA repair in IDH1/2^MUT AML cells using a PARP inhibitor, ideally combined with a DNA-damaging agent, may be a better strategy for the treatment of IDH1/2^MUT AML. The investigation of PARP inhibitor monotherapy in clinical trials in AML patients with somatic IDH1/2^MUT is warranted.

M. A. Sekeres is a consultant/advisory board member for Celgene. No potential conflicts of interest were disclosed by the other authors.

Conception and design: R.J. Molenaar, Y. Nagata, M.A. Sekeres, C.J.F. van Noorden, J.P. Maciejewski

Development of methodology: R.J. Molenaar, C.J.F. van Noorden, J.P. Maciejewski

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R.J. Molenaar, Y. Nagata, M. Xu, H.E. Carraway

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R.J. Molenaar, T. Radivoyevitch, M. Khurshed, B. Przychodzen, H. Makishima, H.E. Carraway, S. Mukherjee, C.J.F. van Noorden, J.P. Maciejewski

Writing, review, and/or revision of the manuscript: R.J. Molenaar, T. Radivoyevitch, Y. Nagata, H. Makishima, M. Xu, F.E. Bleeker, J.W. Wilmink, H.E. Carraway, S. Mukherjee, M.A. Sekeres, C.J.F. van Noorden, J.P. Maciejewski

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Khurshed, S. Mukherjee

Study supervision: R.J. Molenaar, F.E. Bleeker, C.J.F. van Noorden, J.P. Maciejewski

This work was supported by AMC PhD Scholarship (R.J. Molenaar), the Dutch Cancer Society (KWF; UVA 2014-6839 and AMC2016.1-10460, to R.J. Molenaar, M. Khurshed, F.E. Bleeker, J.W. Wilmink, and C.J.F. van Noorden), the NIH (Bethesda, MD; NIH) grants R01HL118281, R01HL123904, R01HL132071, R35HL135795, a grant from the AA & MDS International Foundation (Rockville, MD), the Robert Duggan Charitable Fund (Cleveland, OH, all to J.P. Maciejewski), a Scott Hamilton CARES grant (Cleveland, OH; H. Makishima), and a grant from the AA & MDS International Foundation (H. Makishima).

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.