Disruption of dNTP homeostasis by ribonucleotide reductase hyperactivation overcomes AML differentiation blockade

Acute myeloid leukemia (AML) is characterized by differentiation arrest within bone marrow (BM).^1,2 Successful use of all-trans retinoic acid (ATRA) or inhibitors of mutant isocitrate dehydrogenase (IDH) highlights the achievement of differentiation therapy.^3-5 Other differentiating compounds whose effects are not limited to any specific leukemia subtype are those intervening nucleotide metabolism, including nucleoside analogs (eg, Ara-C)⁶ or inhibitors of dihydroorotate dehydrogenase.⁷ There is an unmet need to understand mechanisms underlying activity of those agents.

Proliferative cancer cells hijack de novo deoxynucleoside triphosphate (dNTP) biosynthesis to meet DNA replication demands.⁸ A key player in dNTP biosynthesis is ribonucleotide reductase (RNR).⁹ The functional RNR catalytic unit is a cytosolic heterotetramer consisting of 2 large subunits (RRM1) and 2 small subunits (RRM2 and RRM2B).⁹ RRM2 levels vary throughout the cell cycle; its transcription is minimal in G₀/G₁ and maximal in S phase, determining RNR activity.^10-12 RRM2 levels are tightly regulated by the DNA damage-induced ataxia telangiectasia and Rad3-related (ATR) pathway,^13,14 which is downstream of augmented R-loops, the triple-stranded DNA:RNA hybrids.^15,16 Moreover, high dNTP levels are reportedly mutagenic via perturbing DNA replication and impairing mitochondrial function.¹⁷ The dNTP hydrolase SAM domain and HD domain-containing protein 1 (SAMHD1), which degrades dNTP and functionally antagonizes RNR,^17-19 reportedly promotes resistance to nucleoside-based chemotherapies by hydrolyzing active triphosphate metabolites like Ara-CTP.^20,21 Last, RNR activation may have opposing effects on tumorigenesis,²² prompting us to explore the role of RNR hyperactivation in AML.

Herein, we identify nelarabine (NEL) from our differentiation compound screen. Its effects stem from R-loop–mediated replication stress and RNR hyperactivation-induced dNTP imbalance. We further characterize extracellular signal-regulated kinase (ERK) hyperactivation as downstream of the imbalance.

Peripheral blood or BM specimens were obtained from patients with AML at City of Hope (COH) Comprehensive Cancer Center. Patient characteristics are summarized in supplemental Table 4 available at the Blood Web site. Risk groups are based on World Health Organization classification. All subjects signed informed consent forms. Sample acquisition was approved by COH Institutional Review Board in accordance with the Declaration of Helsinki.

Immunodeficient NOD-scid IL2Rgnull-3/GM/SF (NSGS) mice used for human-in-mouse xenograft models were obtained from the Jackson Laboratory (stock no. 013062). Mll-AF9/Samhd1^−/− mice were generated by crossing Samhd1^−/− mouse²³ with Mll-AF9 (MA9) knock-in mouse (Jackson Laboratory, stock no. 009079). CD45.1⁺ congenic mice were from the National Cancer Institute (strain no. 01B96). Mouse care and experimental procedures complied with established institutional guidance and approved protocols from the Institutional Animal Care and Use Committee at COH.

Metabolites were extracted with methanol and subjected to targeted metabolomic profiling on the UltiMate 3000 UPLC chromatography system coupled with Q-Exactive orbitrap mass spectrometer. Targeted metabolites were quantified by area under the curve (AUC). Other details are provided in supplemental Materials.

Primary cells were treated as indicated and processed according to the Fluidigm protocol. Other analysis procedures are provided in supplemental Methods.

Total RNA was isolated from cells treated as indicated. RNA sequencing libraries preparation procedures are provided in supplemental Methods. Sequencing run was performed in the single read mode using Illumina HiSeq 2500.

Data obtained from multiple experiments were reported as mean ± standard error of the mean (SEM). Unpaired, 2-sided Student t test was used to compare means between 2 groups. One-way analysis of variance with multiple comparisons was used to compare means among 3 or more groups.

To define potential differentiation indicators for virtual screen, analysis of GSE125112 revealed 35 genes commonly upregulated by ATRA²⁴ (fold-change ≥ 1.5; P < .01; Figure 1A; supplemental Table 1). In parallel, analysis of gene expression profiles of differentiation-inducing agents from NCI-60,²⁵ including ATRA, zalcitabine,²⁶ and sodium butyrate,²⁷ revealed 55 commonly upregulated genes (r > 0.3, P < .01; Figure 1A; supplemental Table 2). There was an overlap of only 1 gene, CD38, whose high expression is seen at later stages of hematopoietic differentiation.²⁸ Interestingly, transcriptome analysis of the lineage⁻Sca⁺cKit⁺ (LSK) subset sorted from conditional Idh2^R140Q;Flt3^ITD knock-in mice revealed that Cd38 was upregulated by in vivo administration of AG-221, a potent IDH2 inhibitor²⁹ (supplemental Figure 1A). We confirmed CD38 induction in U937, KG1A, and NB4 after ATRA, zalcitabine, or sodium butyrate treatment (supplemental Figure 1B) and IDH2^R140Q-expressing TF-1 after AG-221 treatment as reported²⁸ (supplemental Figure 1C).

We next queried the developmental therapeutics program database (>20 000 compounds) with CD38 as input (Figure 1B). Among compounds retrieved from CellMiner,²⁵ 26 US Food and Drug Administration (FDA)-approved compounds were positively correlated with CD38 level (r > 0.3, P < .01), including ATRA (NSC-122758; r = 0.543, P < .01). We requested the top 79 compounds available from the National Cancer Institute (r > 0.6, P = 0) for phenotypic screen (Figure 1B; supplemental Table 3) using ER-HoxA9 cells, a murine differentiation-arrest model.⁷ At a fixed dose of 5 µg/mL, 3 compounds (NSC-641818, NSC-37641, and NSC-755985 [NEL]) exhibited remarkable differentiation (GFP⁺ percentage >25%; Figure 1C). NEL was the only FDA-approved compound identified.^30-34 Following NEL treatment, ER-HoxA9 cells underwent neutrophil-like changes (Figure 1D).

We next determined whether NEL promoted differentiation in human AML. NEL treatment upregulated myeloid marker expression levels within a clinically achievable concentration³⁵ (Figure 1E-F; supplemental Figure 1D-E; supplemental Table 4). NEL-treated cells also showed cytochemical changes and cellular morphology suggestive of neutrophil or monocyte maturation³⁶ (Figure 1G; supplemental Figure 1F). NEL treatment upregulated levels of transcription factors associated with myeloid differentiation³⁷ (Figure 1H), eventually resulting in apoptosis and growth inhibition (supplemental Figure 1G-H). NEL’s differentiation induction effects were correlated with viability inhibition in AML cells (supplemental Figure 1I-J); the treatment had less effects on viability of normal CD34⁺ cells (supplemental Figure 1I). Moreover, colony-formation capacity (CFC) of AML cells was suppressed (supplemental Figure 1K).

To evaluate in vivo effects of NEL treatment, we injected CD34⁺ or T cell–depleted AML cells into NSGS mice³⁸ (supplemental Figure 1L). NEL treatment³⁹ significantly decreased leukemic cell engraftment (Figure 1I,K; supplemental Figure 1M). In vivo differentiation effects were confirmed (Figure 1J). Specifically, loss of the primitive subpopulation (CD117⁺ or CD34⁺) and emergence of monocyte or neutrophil subpopulation (CD14b⁺/CD64⁺ or CD15⁺/CD49d⁻) were seen. We further extended our study to NSGS mice xenografted with U937 cells. Importantly, NEL administration significantly delayed leukemia onset and conferred a survival advantage relative to controls (Figure 1L-N).

Ara-GTP is NEL’s active metabolite^40,41 (supplemental Figure 2A). Accordingly, U937 cells that are more sensitive to NEL relative to THP1 accumulated higher levels of Ara-GTP after comparable treatment (supplemental Figures 1I and 2B). We hypothesized that NEL-induced differentiation was related to nucleotide metabolism perturbation. Metabolomic analysis of NEL-sensitive cells (U937 and KG1A) revealed that many deoxynucleotides increased to different extents after NEL treatment (Figure 2A; supplemental Figure 2C; supplemental Table 5). Moreover, the analyses were performed after treatment for 12 hours when the cell growth had not been altered yet (supplemental Figure 1H), suggesting that these changes were unlikely consequences of cell death. NEL treatment did not increase fractions of newly generated deoxynucleotides synthesized from glucose or glutamine (Figure 2B; supplemental Figure 2D; supplemental Tables 6-7), excluding the contribution of de novo synthesis.

Given that our initial metabolomics analysis cannot distinguish isobaric nucleotides, we performed a targeted nucleotide quantification assay to assess NEL treatment effects on a full panel of dNTPs/NTPs (Figure 2C-D; supplemental Figure 2E) and found that NEL treatment increased dNTP levels unequally. The resultant imbalance was featured by a dramatic increase in dGTP levels (Figure 2C-D; supplemental Figure 2F), in contrast to modest and symmetric increases of dNTPs during G₁/S transition when RNR activity increases^41-43 (supplemental Figure 2G-H).

We then asked whether the unequal increases in dNTPs and differentiation outcome were mediated by RNR. To test the possibility, we used an RNR inhibitor COH29, which targets the ligand-binding pocket of RRM2.⁴⁴ Supplementation with COH29 alleviated dNTP imbalance (Figure 2C-D; supplemental Figure 2F) and reversed differentiation after NEL exposure (Figure 2E-F). Similar rescue was seen after treatment of hydroxyurea (HU), another RNR inhibitor⁴⁵ (data not shown). Unlike short-term exposure (Figure 2C-D), long-term treatment of COH29 depleted dNTPs⁴⁴ (supplemental Figure 2I). Although COH29 treatment alone did not induce differentiation (supplemental Figure 2J), it did inhibit cell growth in both normal and AML cells (supplemental Figure 2K-L). To validate the viability rescue by diminishing dNTP imbalances, we established an “RRM2-low” line by introducing a doxycycline (DOX)-inducible RRM2-shRNA construct (ishRRM2) into U937 cells. DOX treatment decreased RRM2 protein levels by approximately 50% (Figure 2G), with minimal effects on cell cycling (supplemental Figure 2M). As anticipated, after NEL treatment, this line exhibited less dNTP imbalance (supplemental Figure 2N), differentiation (Figure 2H; supplemental Figure 2O), and viability inhibition (supplemental Figure 2P) than those seen in controls (ishCtrl).

We then tested whether supplementing cells with individual deoxynucleosides (dNs)¹⁸ to experimentally create a dNTP imbalance would phenocopy NEL treatment, given that exogenous dNs can be salvaged to form dNTPs.⁴⁶ Indeed, dG treatment induced robust differentiation and cell death (Figure 2I; supplemental Figure 2Q). dT or dA treatment induced differentiation at relatively higher doses, whereas dC did not antagonize cell viability as reported^18,47 (Figure 2I). dG treatment promoted a significant increase in total dGTP levels to an extent greater than that of any other dNTPs (Figure 2J; supplemental Figure 2R). Notably, addition of dC or dA could diminish dG’s effects (Figure 2K; supplemental Figure 2R-S). Furthermore, addition of forodesine, an inhibitor of purine nucleoside phosphorylase,⁴⁸ could exacerbate dG-triggered imbalance, whereas forodesine alone was less effective than dG (supplemental Figure 2T). dG treatment induced differentiation and inhibited viability of AML cells within a clinically achievable dosage⁴⁹ (Figure 2L; supplemental Figure 2U-V). We further depleted SAMHD1 as a safeguard of dNTP homeostasis¹⁹ in SAMHD1-proficient THP1 cells (supplemental Figure 2W-Y) and observed enhanced dG-mediated differentiation and growth inhibition relative to SAMHD1 wild-type (WT) controls (Figure 2M-N), confirming the importance of dNTP imbalance.

To evaluate the effects of dNTP pool imbalance on leukemia stem cell (LSC) activity, we conducted in vitro limiting dilution assay^50-52 using BM cells from Mll-AF9 transgenic mice⁵³ or primary AML cells. dG treatment resulted in a decrease in LSC frequency and differentiation induction (Figure 2O; supplemental Figure 2Z). Relative to vehicle controls, dG pretreatment markedly decreased engraftment of leukemic cells in NSGS mice at 12 weeks after transplantation (Figure 2P-Q). Notably, secondary transplantation of BM cells from mice receiving dG-pretreated cells resulted in nearly complete elimination of leukemia engraftment (Figure 2R) compared with those of control cells, highlighting the impairment of LSC self-renewal (Figure 2S).

We next asked whether treatment of NEL, a known genotoxin, upregulated RNR subunits through DNA damage response machinery.^9,54 Among RNRs, RRM2 levels were significantly upregulated by NEL treatment (Figure 3A-B). Immunofluorescence revealed enhanced cytoplasmic level and nuclear localization of RRM2 (Figure 3C). Notably, NEL treatment for 12 hours significantly altered levels of ATR signaling effectors, including p-CHK1 and E2F1, whereas modest alterations were seen on p-CHK2 (Figure 3B). Consistently, unlike ATM inhibition, ATR inhibition or E2F1 knockdown reduced NEL-mediated upregulation of RRM2 (Figure 3D-E; supplemental Figure 3A-C), confirming that NEL treatment upregulated RRM2 via the ATR/CHK1/E2F1 axis.⁵⁵ RRM2 levels correlate with NEL sensitivity in hematopoietic cancer cell lines in contrast to the role of SAMHD1 as reported⁵⁶ (Figure 3F; supplemental Figure 3D). Accordingly, NEL treatment caused more robust increased levels of RRM2 and DNA damage effectors in SAMHD1 knockout (KO) THP1 cells compared with WT controls (supplemental Figure 3E), consistent with SAMHD1’s function in preventing Ara-GTP accumulation and decreasing replication stress.²¹ 

The mechanism of Ara-C action is similar to NEL^41,57; thus, we asked whether differentiation induction seen after Ara-C treatment as reported previously^6,58 involved RRM2 activation. Indeed, Ara-C significantly increased RRM2, p-CHK1, E2F1, and dNTP levels in U937 cells but only induced modest changes in SAMHD1-proficient THP1 cells (Figure 3G; supplemental Figure 3F). Next, we assessed the role of RNR in Ara-C treatment effects by applying 2 strategies mentioned above. In 1, COH29 treatment alleviated dNTP imbalance (Figure 3H; supplemental Figure 3G) and weakened differentiation markers upregulation after Ara-C treatment (Figure 3I; supplemental Figure 3H-I). In another approach using ishRRM2-U937 as a model, Ara-C treatment resulted in less differentiation induction relative to that seen in ishCtrl-U937 cells (Figure 3J). Notably, RRM2 levels are strongly and positively correlated with Ara-C sensitivity (Figure 3K; supplemental Figure 3J), whereas SAMHD1 is a resistant factor.^20,21 Additionally, we asked whether patients with AML exhibiting high RRM2 levels achieve better outcomes after treatment with Ara-C–based standard care. Indeed, we observed a positive correlation between higher RRM2 levels and longer overall survival (Figure 3L; supplemental Figure 3K-L; Table 1) through retrospectively analyzing both TCGA⁵⁹ and GSE14468⁶⁰ cohorts.

To further define the differentiation-related DNA insult by NEL treatment, we tested whether the insult associated with excessive formation of R-loops that specifically activate ATR signaling.^15,16,61 Interestingly, Ara-C is known to trigger R-loop formation,⁶² likely because of transcription-replication conflicts. Similarly, NEL treatment enhanced R-loops formation evidenced by increased nuclear staining of the S9.6 antibody (Figure 3M-N). Overexpressing RNASEH1 to resolve R-loops remarkably abrogated ATR/CHK1 activation and RRM2 upregulation, thereby blocking differentiation induction and partially rescuing viability inhibition by NEL treatment (Figure 3M-Q). Moreover, HU did not induce R-loop formation (Figure 3M-N) or differentiation (supplemental Figure 3M), although it resulted in DNA damage as evidenced by CHK1 phosphorylation (Figure 3O).

We next asked whether non–DNA-incorporating compounds could induce differentiation in an RNR-dependent manner. Camptothecin (CPT), a topoisomerase I inhibitor, reportedly induces leukemia cell differentiation.⁶³ Similar to Ara-C, CPT treatment increased RRM2, p-CHK1, and E2F1 levels in AML cells (supplemental Figure 3N-O). Notably, the drug-induced effects on dNTP imbalance and differentiation were significantly rescued by RNR downregulation, whereas viability inhibition was partially reversed (supplemental Figure 3P-T). Moreover, CPT is also known to induce R-loop formation.⁵⁵ Consistently, overexpressing RNASEH1 abrogated ATR/CHK1 activation and RRM2 upregulation by CPT treatment (Figure 3O).

We asked whether direct upregulation of RRM2 would initiate differentiation. We first used THP1 as a model. Following CDK2-mediated phosphorylation of Thr33, WT RRM2 was recognized by Cyclin-F (CCNF) via the RxI motif (aa49-aa51) for degradation at G₂/M phase.¹⁴ Although ectopic expression of WT FLAG-RRM2 marginally affected RRM2 levels, expression of RRM2 mutants exhibiting less binding affinity to CCNF¹⁴ (RRM2-T33A [T33A] and RRM2-RxI/AxA [Rxl/AxA]) promoted RRM2 accumulation (supplemental Figure 4A), thereby decreasing cell viability (supplemental Figure 4B). Given that RRM2 protein accumulated mostly after Rxl/AxA overexpression (OE), we transduced an inducible RRM2-RxI/AxA mutant (iRxI/AxA) into THP1 for further analysis (Figure 4A; supplemental Figure 4C). After DOX induction, iRxI/AxA-transduced cells showed increased dNTPs, particularly a marked increase in dGTP, pronounced differentiation, and decreased viability relative to MOCK cells (Figure 4B-C; supplemental Figure 4D). To further enhance RRM2 activity, we exposed this inducible line to the combination of DOX plus NEL or DOX plus Ara-C. Relative to DOX alone, the combination further increased RRM2 levels, aggravated dNTP imbalance, enhanced CD11b induction, and inhibited viability (Figure 4A-C; supplemental Figure 4D). To assess outcomes in vivo, we transplanted these engineered cells into NSGS mice and treated mice with combination of DOX plus NEL or DOX plus phosphate-buffered saline (PBS). Mice injected with MOCK cells receiving combination of DOX plus PBS succumbed to systemic disease shortly, whereas mice injected with iRxI/AxA cells receiving combination of DOX plus PBS survived significantly longer and exhibited reduced leukemic burden (Figure 4D-E). Relative to PBS controls, NEL treatment further decreased leukemic burden and prolonged survival in mice engrafted with iRxI/AxA cells (Figure 4D-E). Similar results were seen in leukemic mice treated with Ara-C (supplemental Figure 4E-F).

To assess effects of stimulating RRM2 activity in primary AML cells, we used a strategy of CCNF depletion.¹⁴ We noted CCNF dependency score was negatively correlated with RRM2 basal levels in 94 hematologic cancer cell lines (Figure 4F). Indeed, CCNF knockdown (KD) in THP1 cells promoted RRM2 accumulation, created dNTP imbalance, induced differentiation, and decreased cell growth (Figure 4G-I; supplemental Figure 4G). To determine whether RRM2 was required for CCNF KD effects, we established an RRM2-low THP1 line as described above. After DOX treatment, ishRRM2-THP1 cells exhibited decreased RRM2 levels (supplemental Figure 4D). CCNF KD significantly increased RRM2 and CD11b in ishCtrl-THP1 cells, whereas it only marginally increased RRM2 in ishRRM2-THP1 cells, with no change in CD11b levels (supplemental Figure 4H-I), suggesting that CCNF KD effects were mainly dependent on RRM2. Next, we depleted CCNF in primary AML cells and observed varying increases in RRM2 levels among specimens (n = 6; Figure 4J), hinting additional RRM2 regulatory mechanisms other than CCNF.^64,65 We further assessed effects of combining CCNF KD with Ara-C or NEL treatment. CCNF KD decreased the half-maximal inhibitory concentration (IC₅₀) values of both drugs (Figure 4K; supplemental Figure 4J) and enhanced NEL-induced differentiation compared with NEL alone at a fixed dose (20 µM; supplemental Figure 4K).

We asked whether aggravating dNTP imbalance by deleting SAMHD1 could enhance RRM2 hyperactivation mediated inhibitory effects. To do so, we used Mll-AF9/Samhd1 KO mice (Samhd1-KO) generated through crossing Samhd1 KO mice²³ to Mll-AF9 transgenic mice. We expressed iRxI/AxA construct or MOCK control in Samhd1-KO or Samhd1-WT BM cells, respectively, resulting in 4 groups: Samhd1-WT/MOCK, Samhd1-KO/MOCK, Samhd1-WT/RRM2 OE, and Samhd1-KO/RRM2 OE. We then used these engineered cells for CFC re-plating. Interestingly, murine AML cells harboring iRxI/AxA showed significantly decreased CFC compared with MOCK cells in primary and serial re-platings; whereas Samhd1 KO alone barely affected CFC, it further compromised CFC in AML cells harboring iRxI/AxA (Figure 4L). To test the hypothesis in vivo, we transplanted engineered cells into CD45.1-expressing congenic recipient mice for leukemia assessment. iRxI/AxA-expressing MA9 leukemic mice exhibited reduced leukemic chimerism in BM and improved survival relative to MOCK controls (Figure 4M-N). Notably, BM cells from iRxI/AxA-expressing MA9 leukemic mice exhibited RRM2 overexpression and dGTP overproduction (supplemental Figure 4L-M). Moreover, Samhd1 KO combined with iRxI/AxA expression further decreased engraftment and extended leukemic mice survival compared with iRxI/AxA expression alone (Figure 4M-N). To test whether genetically elevating RRM2 alone or in combination with Samhd1 KO affected LSC activity, we conducted in vivo limiting dilution assays by establishing murine AML transplants using those engineered cells.⁵² Similarly, iRxI/AxA expression alone resulted in a remarkable decrease in LSC frequency, and the combination of Samhd1 KO and iRxI/AxA expression effectively inhibited in vivo repopulating capacity of AML cells (Figure 4O).

To define downstream pathway that conferred outcomes of dNTP imbalance, we assessed gene expression profiles in NEL or dG-treated AML cell lines (U937 and KG1A) by RNA-seq analyses (supplemental Tables 8-11). Ingenuity pathway analysis revealed Rho family GTPase signaling and ERK/MAPK signaling as top upregulated pathways in both treatments (Figure 5A-B). To validate whether these 2 pathways were effectors of RRM2 hyperactivation, we performed RNA-seq analyses of iRxI/AxA- and MOCK-THP1 cells after DOX treatment (supplemental Table 12). Gene set enrichment analysis revealed significant enrichment of RAS/ERK pathway signatures in iRxI/AxA-THP1 cells (Figure 5C), suggesting the role of ERK signaling as downstream of RRM2 hyperactivation. We further confirmed ERK activation after NEL, dG, or Ara-C treatment in AML cells (Figure 5D,F; supplemental Figure 5A). Of note, either iRxI/AxA OE or CCNF KD resulted in increased phospho-ERK in THP1 or primary AML specimens, respectively (Figure 5E,G; supplemental Figure 5B). To address the biological relevance of ERK activation, we established a U937 line with low basal ERK signaling using ERK2 shRNA. Although ERK2 knockdown slightly decreased cell growth relative to controls, it blocked NEL-induced growth inhibition effects (Figure 5H-I). We further injected U937 cells expressing either ERK2 shRNA or nontargeting control into NSGS mice and treated both groups with NEL. Relative to mice receiving ERK2-intact AML cells, mice receiving ERK2 KD cells showed less therapeutic responses following NEL treatment (Figure 5J-K). Moreover, pretreatment with ERK inhibitors to downregulate its basal signaling partially rescued the effects of NEL or dG treatment (supplemental Figure 5C-F), whereas long-term treatment of an ERK or MEK inhibitor alone eliminated AML CFC (supplemental Figure 5G), likely because of complete abrogation of ERK signaling.^66,67

Severe dNTP imbalance reportedly impairs mitochondrial DNA replication,^68,69 resulting in mitochondrial stress and release of reactive oxygen species (ROS), which leads to ERK activation.⁷⁰ Accordingly, dG or NEL treatment caused disruption of mitochondrial matrix morphology and loss of mitochondrial cristae (Figure 5L; supplemental Figure 5H). Moreover, dG treatment increased mitochondrial superoxide, decreased mitochondrial membrane potential, and induced leakage of mitochondrial DNA into cytoplasm (Figure 5M; supplemental Figure 5I-K). Notably, addition of ROS scavenger N-acetyl cysteine (NAC) reversed mitochondria superoxide increase, dampened ERK activation, and partially rescued cellular outcome by dG treatment in AML cells (Figure 5M; supplemental Figure 5K-L), indicating that mitochondrial ROS release links dNTP imbalance to ERK activation (Figure 5N). Moreover, CD34⁺CD38⁻ AML cells showed increased differentiation and reduced CFC after dG treatment, whereas these effects were partially rescued by NAC treatment (supplemental Figure 5M-N), suggesting that both differentiation and LSC impairment are downstream of mitochondrial dysfunction (Figure 5N).

Hyperactive RAS reportedly promotes commitment of differentiation through ERK hyperactivation.^71-74 To determine whether oncogenic RAS cooperated with NEL-induced ERK activation to induce differentiation and growth inhibition, we used splenocytes from mice bearing conditional oncogenic Kras (Kras^{Lox-Stop-Lox (LSL) G12D/+}/Vav-Cre⁺),^75,76 followed by Mll-AF9 transduction. Relative to Vav-Cre⁺ splenocytes transformed by MA9 alone, MA9/Kras^G12D doubly transformed cells were more vulnerable to NEL-induced inhibitory effects (Figure 5O; supplemental Figure 5O). Notably, ERK inhibition partially rescued NEL’s effects on MA9/Kras^G12D cells (Figure 5P; supplemental Figure 5P). We next asked whether RRM2 level predicted prognosis in patients carrying oncogenic RAS after Ara-C treatment, which also primed ERK signaling (Figure 5F; supplemental Figure 5A). To do so, we reanalyzed clinical outcomes of RAS mutant or RAS WT cases from the GSE14468 cohort containing sufficient RAS mutant cases.⁶⁰ Indeed, high RRM2 level predicted longer survival for AML patients carrying RAS mutations, whereas in patients with WT RAS, RRM2 level was less predictive (Figure 5Q-R; supplemental Table 13).

Drug combination is desired to achieve maximal efficacy. To identify NEL-based combination therapy, we performed a genome-wide clustered regularly interspaced short palindromic repeats (CRISPR) screen in a U937 subline resistant to NEL (U937/R; Figure 6A). This subline (IC₅₀ = 42.3 μM) was generated by gradual adaptation of parental cells (IC₅₀ = 3.8 μM) to 50 µM NEL over months.⁷⁷ After Cas9⁺ clonal derivation (supplemental Figure 6A), we transduced cells with GeCKO guide RNA (gRNA) library⁷⁸ and treated with 20 µM NEL for 7 days to negatively enrich candidate gRNAs associated with drug response.⁷⁹ Among the top hits, we identified NT5C2 to be significantly depleted in NEL-treated cells (supplemental Table 14), which encodes a nucleotidase reportedly to inactivate the active metabolites of purine analogs.^80-83 Herein, given that NEL’s effects associated with ERK activation, and hyperactivation of ERK can be deleterious to cancer cells,^84,85 we thus analyzed gRNAs targeting genes encoding those negative feedback regulators of ERK signaling^84,85 (Figure 6B), including the dual-specificity phosphatases (DUSPs), the sprouty proteins, and the sprouty-related EVH1 domain-containing proteins.^84,86 Among those genes, DUSP1 and DUSP6 were markedly depleted in NEL-treated vs vehicle-treated cells⁸⁷ (Figure 6B; supplemental Table 14). Four of 6 gRNAs targeting DUSP6 were depleted in NEL-treated cells (supplemental Figure 6B). Given its higher abundancy than DUSP1 (supplemental Figure 6C) and its adaptive upregulation after short exposure to drug treatment (Figure 6C), we prioritized DUSP6 for further investigation.

In U937/R cells, DUSP6 depletion potentiated NEL-induced ERK activation, differentiation, and apoptosis than did comparably treated DUSP6 WT cells (Figure 6D-G). To determine whether targeting DUSP6 would enhance NEL’s effects, we tested a small molecule inhibitor of DUSP6, 2-benzylidene-3-(cyclohexylamino)-1-Indanone hydrochloride (BCI).⁸⁸ Analogously, the addition of BCI re-sensitized U937/R cells to NEL-induced ERK activation, differentiation, and apoptosis (supplemental Figure 6D-F), whereas those effects were partially blocked by ERK2 knockdown (Figure 6H-I; supplemental Figure 6G). Enhanced leukemia targeting effects by combination were also seen in primary AML cells (Figure 6J-K; supplemental Figure 6H). To assess outcomes in vivo, we used NSGS models xenografted with MOLM13 cells, because its sensitivity to NEL was comparable to those of most primary AML specimens (supplemental Figure 6I). Although NEL treatment alone had modest effects, the combination significantly alleviated leukemia burden (Figure 6L-M; supplemental Figure 6I), with a comparable survival benefit to that of standard chemotherapy (supplemental Figure 6K), suggesting the response to NEL depends on reaching a lethal threshold of ERK activity, and DUSP6 may antagonize that effect (Figure 6N).

Our results highlight the importance of dNTP homeostasis in myeloid differentiation. We show that dNTP imbalance caused by RRM2 hyperactivation through either drug treatment or genetic manipulation overcomes the differentiation blockade. Specifically, dGTP level is most vulnerable to be perturbated. Accordingly, dG is most potent in inducing differentiation and LSC ablation among all dNs. We also note that SAMHD1 may underlie varying responses to dG treatment in AML cells; SAMHD1 KO in AML cells remarkably enhances dG or RRM2 OE-induced differentiation and self-renewal ablation (Figures 2M-N and 4L-O). Therefore, SAMHD1 may counteract dNTP pool changes initiated by drug treatment or RRM2 upregulation, lessening the cellular outcome.

Our study reveals a novel role of RRM2 in promoting AML differentiation. Specifically, treatment of NEL, Ara-C, or CPT induces RRM2 hyperactivation, in agreement with previous reports of increases in RNR subunit levels after replication stress.⁵⁵ In particular, this drug-induced RNR activation leads to imbalanced dNTPs, in contrast with dNTP changes after allosteric RNR activation during cell cycling.⁴² One explanation for this discrepancy in dNTP outcomes could be the duration of RRM2 upregulation, which may determine the extent of RNR activation. Moreover, SAMHD1 determines the extent to which RNR is activated, given its restriction on Ara-CTP or Ara-GTP accumulation.⁸⁹ Nevertheless, in SAMHD1-proficient AML cells, overexpression of a stable RRM2 mutant promotes tumor regression and impairs LSC self-renewal (Figure 4).

We identify augmented R-loop formation underlies differentiation induction by NEL treatment. R-loops specifically activate ATR signaling as previously reported.^15,16 Consistently, both Ara-C⁶² and CPT⁹⁰ reportedly trigger R-loop formation,⁹¹ whereas HU does not.¹⁶ These results highlight the role of R-loop–initiated DNA damage in promoting differentiation, whereas we do not exclude the possibility that other actions of these genotoxins (NEL, Ara-C) also serve as mechanisms of lethality (eg, incorporation into nascent DNA).

ERK is an important player in controlling hematopoietic cell activity.⁹² Complete repression of ERK signaling by MEK or ERK inhibitors causes lethality,^66,67 whereas excessive ERK activation also compromises viability of cancer cells.^84,85,93 Accordingly, our data extend the emerging concept that hyperactivation of ERK can be deleterious in the context of AML. Notably, RAS mutations predict a better therapeutic response in patients with AML treated with high-dose cytarabine^74,94 but show resistance to a venetoclax-based regimen.⁷² Moreover, we provide a therapeutic strategy to further enhance ERK signaling in AML cells primed by NEL treatment through cotargeting DUSP6, thereby surpassing an ERK hyperactivation-related lethal threshold (Figure 6). The combination of BCI and NEL not only suppresses leukemogenesis in an AML xenograft model but exhibits better tolerability than standard chemotherapy (supplemental Figure 6J-L).

In conclusion, we show that disrupting dNTP pool homeostasis overcomes differentiation blockade of AML cells and impairs LSC self-renewal. Our study prompts a reappraisal of therapeutic response in patients with RAS-mutated AML with considerations of factors controlling dNTP homeostasis, such as RRM2 and SAMHD1. Finally, we provide a rationale for further evaluation of the combination of nelarabine with a DUSP6 inhibitor against AML.

The authors thank COH Comprehensive Cancer Center, as well as patients, donors, and their physicians for providing primary specimens for this study, Jianjun Chen for providing Cas9-expressing THP1 cells, Jan Rehwinkel at University of Oxford for constitutive Samhd1 KO mice, Dong-er Zhang at University of California, San Diego, for RNASEH1 vectors, Johanna Ten Hoeve at UCLA metabolomics center for assistance with metabolites measurements, and Stemcyte Inc. for the cord blood (CB) sample. The authors also thank Elise Lamar and Marjorie Robbin for help in manuscript editing and proofreading.

This work was supported in part by National Institutes of Health (NIH) grant R01 CA152108 and Department of Defense impact award CA190124 to J.Z., NIH grant R01 CA2062101 and Cancer Prevention & Research Institute of Texas grant DP150061 to S.T., and Gehr Family Center for Leukemia Research support to L.L. The authors acknowledge the support of the Animal Resources Center, Analytical Cytometry Core, Bioinformatics, Light Microscopy, Electron Microscopy, Integrative Genomics Core, Mass Spectrometry & Proteomic Core at COH Comprehensive Cancer Center, supported by the NIH, National Cancer Institute under award P30CA33572.

The content is solely the responsibility of the authors and does not necessarily represent official views of the NIH.

Contribution: H.W. designed and performed experiments, interpreted results, and wrote the manuscript; X.H. designed and performed experiments including bioinformatic data mining, primer extension assay for dNTP measurements, immunofluorescence, and CRISPR-Cas9 screen, and reviewed the manuscript; Lei Zhang performed Western blot analyses and reviewed the manuscript; H.D. assisted with animal experiments and reviewed the manuscript; F.H. performed the CFC assay, Mll-AF9 retrovirus packaging and transduction, and animal experiments; J.X. performed flow cytometry analysis; M.L. performed statistical analyses; W.C. performed transcriptomic analyses; X.L. and K.V.P. performed quantitative MS analyses; W.H. assisted with flow cytometry analyses and animal experiments; Z.L. and Lianjun Zhang assisted with experiments; L.X.T.N. performed transmission electron microscopy experiments; L.Y. interpreted CRISPR screen results; L.F. reviewed RNA-seq results; D.J.G. provided ishRRM2/iRRM2-T33A constructs; J.Z. provided murine Kras^G12D/+ splenocytes; A.S. provided AML samples; D.H. provided compound COH29; D.B.S. provided ER-HoxA9 cells; P.P., C.-W.C., Y.-H.K., G.M., S.T., H.J., and X.W. reviewed and edited the manuscript; and L.L. designed the study, interpreted data, and prepared the manuscript with input from others.

Conflict-of-interest disclosure: D.B.S. is a cofounder and holds equity in Clear Creek Bio, is a consultant and holds equity in SAFI Biosolutions, and is a consultant for Keros Therapeutics. The remaining authors declare no competing financial interests.

Correspondence: Ling Li, Department of Hematological Malignancies Translational Science, Gehr Family Center for Leukemia Research, Hematologic Malignancies and Stem Cell Transplantation Institute, Beckman Research Institute of City of Hope, Monrovia Research Center 2005, 1500 E. Duarte Rd, Duarte, CA 91010; e-mail: lingli@coh.org; Hongchuan Jin, Laboratory of Cancer Biology, Provincial Key Laboratory of Biotherapy in Zhejiang, Sir Run Run Shaw Hospital, Zhejiang University, 3 East Qingchun Rd, Hangzhou 310016, China; e-mail: jinhc@zju.edu.cn; and Xian Wang, Department of Medical Oncology, Sir Run Run Shaw Hospital, Zhejiang University, 3 East Qingchun Rd, Hangzhou 310016, China; e-mail: wangx118@zju.edu.cn.