Resistance to decitabine and 5-azacytidine emerges from adaptive responses of the pyrimidine metabolism network

Mechanisms-of-resistance to decitabine and 5-azacytidine, mainstay treatments for myeloid malignancies, require investigation and countermeasures. Both are nucleoside analog pro-drugs processed by pyrimidine metabolism into a nucleotide analog that depletes the key epigenetic regulator DNA methyltranseferase 1 (DNMT1). We report here that DNMT1 protein, although substantially depleted (~50%) in patients’ bone marrows at response, rebounded at relapse, and explaining this, we found pyrimidine metabolism gene expression shifts averse to the processing of each pro-drug. The same metabolic shifts observed clinically were rapidly recapitulated in leukemia cells exposed to the pro-drugs in vitro. Pyrimidine metabolism is a network that senses and preserves nucleotide balances: Decitabine, a deoxycytidine analog, and 5-azacytidine, a cytidine analog, caused acute and distinct nucleotide imbalances, by off-target inhibition of thymidylate synthase and ribonucleotide reductase respectively. Resulting expression changes in key pyrimidine metabolism enzymes peaked 72-96 hours later. Continuous pro-drug exposure stabilized metabolic shifts generated acutely, preventing DNMT1-depletion and permitting exponential leukemia out-growth as soon as day 40. Although dampening to activity of the pro-drug initially applied, adaptive metabolic responses primed for activity of the other. Hence, in xenotransplant models of chemorefractory AML, alternating decitabine with 5-azacytidine, timed to exploit compensating metabolic shifts, and addition of an inhibitor of a catabolic enzyme induced by decitabine/5-azacytidine, extended DNMT1-depletion and time-to-distress by several months versus either pro-drug alone. In sum, resistance to decitabine and 5-azacytidine emerges from adaptive responses of the pyrimidine metabolism network; these responses can be anticipated and thus exploited. GRAPHICAL ABSTRACT


INTRODUCTION
The deoxycytidine analog pro-drug decitabine and the cytidine analog pro-drug 5-azacytidine can increase lifespans of patients with myeloid malignancies, shown by randomized trials in patients with myelodysplastic syndromes (MDS) and acute myeloid leukemia (AML)(reviewed in 1 ). Both pro-drugs are processed by pyrimidine metabolism into a deoxycytidine triphosphate (dCTP) analog, Aza-dCTP, that depletes the key epigenetic regulator DNA methyltransferase 1 (DNMT1) in dividing cells 2 . DNMT1-depletion terminates malignant selfreplication but maintains normal hematopoietic stem cell self-replication 3-14a vital therapeutic index when treating myeloid malignancies, since recovery by functional hematopoiesis is needed to reverse low blood counts, the cause of morbidity and death. Also clinically significant, the cell cycle exits induced by DNMT1depletion do not require the p53 apoptosis axis, and can hence occur even in TP53-mutated, chemotherapyresistant malignant cells (reviewed in 15 ): Accordingly, decitabine or 5-azacytidine can benefit even patients with high risk, chemorefractory disease 1 . Nevertheless, only ~40% of treated patients benefit overall, and even in responders, relapse is typical. There is therefore a need to understand the mechanisms by which malignant cells resist decitabine or 5-azacytidine, and to use such knowledge to improve response rates and durations. An important piece of this puzzle could be how these pro-drugs are processed into DNMT1 targeting Aza-dCTP.
Decitabine and 5-azacytidine have an identical pyrimidine base modification -replacement of carbon at position 5 with nitrogen -but the sugar moiety is deoxyribose in decitabine and ribose in 5-azacytidine. This channels their metabolism differently, with critical roles for the following enzymes: deoxycytidine kinase (DCK), uridine cytidine kinase 2 (UCK2), cytidine deaminase (CDA), and carbamoyl-phosphate synthetase (CAD). The initial, rate-limiting step in the processing of decitabine toward Aza-dCTP is its phosphorylation by DCK 16,17 .
DCK-null AML cells thus resisted decitabine, even at a concentration of 360µM 18 , and sensitivity was restored by transfection with an expression vector for DCK 16,19 . The initial phosphorylation of 5-azacytidine on the other hand is by UCK2 20,21 . Thus, AML cell lines resistant to >50µM of 5-azacytidine contained inactivating mutations in UCK2 22 , and sensitivity was restored by transfection with an expression vector for UCK2 22 . Despite such in vitro data, contributions of altered DCK and/or UCK2 to clinical relapse has been minimally investigated: one study of 14 decitabine-treated patients measured DCK expression in peripheral blood or bone marrow at relapse vs diagnosis, with inconclusive results 23 ; another study of 8 decitabine-treated patients did find that DCK expression was significantly decreased at relapse 24 .
CDA is another pyrimidine metabolism enzyme shown to mediate resistance to decitabine or 5azacytidine in vitro: CDA rapidly catabolizes both pro-drugs into uridine counterparts that do not deplete DNMT1 25 and that instead cause off-target anti-metabolite effects, e.g., by misincorporating into DNA 26 . Thus, introduction of expression vectors for CDA into malignant cells conferred decitabine-resistance 27,28 . High CDA expression in tissues such as the liver underlies the brief in vivo plasma half-lives of decitabine and 5-azacytidine of ~15 minutes vs 9-16 hours in vitro at 37°C 29,30 . In a pre-clinical in vivo model, CDA-rich tissue microenvironments (e.g., liver) provided sanctuary to AML cells from decitabine 31 . In clinical analyses, higher CDA expression in males appeared to contribute to inferior responses of their MDS to decitabine or 5-azacytidine [32][33][34] . Nevertheless, as for DCK and UCK2, a contribution of CDA to clinical relapse is neither established nor addressed by current clinical practice.
CAD is the first enzyme in the de novo pathway that synthesizes dCTP from glutamine and aspartate: de novo synthesized dCTP can compete with Aza-dCTP for incorporation into DNA, and CAD upregulation has been implicated in resistance to 5-azacytidine in vitro 20, 35,36 . Again, however, a contribution of CAD to clinical resistance/relapse has not been examined. Altogether therefore, DCK, UCK2, CDA and CAD expression changes are known to mediate resistance to decitabine or 5-azacytidine in vitro, but there is little information and no countermeasures for their individual or collective contributions to clinical resistance. Here, upon a first serial analyses of DNMT1 levels in patients' bone marrows on clinical decitabine or 5-azacytidine therapy, we found that this target was not being engaged at clinical relapse and showing why, bone marrows at relapse exhibited shifts in DCK, UCK2, CDA and CAD expression in directions adverse to pro-drug conversion to Aza-dCTP.
Pyrimidine metabolism is a network that senses and regulates nucleotide levels 37 , and we found that decitabine and 5-azacytidine cause distinct nucleotide imbalances, that in turn trigger specific, adaptive changes in expression of key pyrimidine metabolism enzymes. The consistency and predictability of pyrimidine metabolism network responses to decitabine or 5-azacytidine perturbations enabled their anticipation and exploitation to instead enhance pro-drug effects: simple, practical treatment modifications, evaluated in pre-clinical in vivo models of aggressive chemo-refractory AML, preserved the favorable therapeutic index of non-cytotoxic DNMT1depletion and markedly improved efficacy. Bone marrow samples for research were obtained from patients with AML with written   informed consent on a study protocol approved by the Cleveland Clinic Institutional Review Board (Cleveland,   Ohio). Murine experiments were approved by the Cleveland Clinic Institutional Animal Care and Use Committee (Cleveland, Ohio).

DNMT1 is not depleted at clinical relapse or with in vitro resistance
We measured DNMT1 protein levels in patients' bone marrows before and during therapy with decitabine or 5azacytidine (39 serial bone marrow samples from 13 patients, median treatment duration 372 days, range 170-1391). Serial bone marrow biopsies from the same patient were cut onto the same glass slide and stained simultaneously to facilitate time-course comparison of DNMT1 protein levels quantified by immunohistochemistry and ImageIQ imaging and software (Figure 1A). At time-of-response, DNMT1 protein was markedly and significantly decreased by ~50% compared to pre-treatment ( Figure 1A). At the time-of-relapse on-therapy, however, DNMT1 protein levels had rebounded to levels comparable to or exceeding pre-treatment levels ( Figure 1A).
The proliferation marker MKI67 was increased at relapse in almost all the patients, consistent with active progression of disease ( Figure 1B).
We then evaluated if AML cells resist clinically relevant concentrations of decitabine in vitro in a similar way. AML cells (THP1, K562, OCI-AML3, MOLM13) were cultured in the presence of decitabine 0.5 -1.5M.
AML cells that proliferated exponentially in the presence of decitabine emerged as early as 40 days after the first addition of decitabine ( Figure 1C). As per clinical relapse, DNMT1 was not depleted from the decitabineresistant AML cells despite the presence of decitabine ( Figure 1D). UCK2 and CDA protein levels were markedly elevated in decitabine-resistant vs vehicle-treated parental AML cells ( Figure 1D). Also upregulated was CAD, by total protein levels and by S1856 phosphorylation, a post-translational modification linked with its functional activation ( Figure 1D). In contrast, DCK protein levels were suppressed ( Figure 1D). In short, the in vitro resistance resembled the clinical resistance, with preserved DNMT1 and reconfigured pyrimidine metabolism.

DCK is important for maintaining dCTP and UCK2 for maintaining dTTP levels
To examine DCK and UCK2 roles in dCTP and/or dTTP maintenance, we measured nucleotide levels in DCK knock-out (KO), UCK2 KO and wild-type control leukemia cells (HAP1). Knock-out of DCK significantly decreased dCTP but not dTTP ( Figure 3A, B). Thus, DCK is important to dCTP maintenance, consistent with DCK upregulation as a response to dCTP suppression by 5-azacytidine (Figure 2). Knock-out of UCK2 significantly decreased dTTP (Figure 3A, B). Thus, UCK2 is important to dTTP maintenance, consistent with UCK2 upregulation as a response to dTTP suppression by decitabine (Figure 2).

Resistance countermeasures evaluated in vivo
We then examined solutions to resistance in a patient-derived xenotransplant (PDX) model of chemorefractory AML (summarized in Table 1): Schedule decitabine administration to avoid DCK troughs: Immune-deficient mice were tail-vein innoculated with 1 million human AML cells each obtained from a patient with AML that was refractory/relapsed to decitabine then cytarabine. On Day 9 after innoculation, mice were randomised to treatment with (i) vehicle; (ii) decitabine timed to avoid DCK troughs (Day 1 and Day 2 each weekdecitabine-Day1/2); or (iii) decitabine timed to coincide with DCK troughs (Day 1 and Day 4 each weekdecitabine-Day1/4) ( Figure S2A). Vehicle-treated mice showed distress on Day 45, at which point all mice were euthanized or sacrificed for analyses. The bone marrows of vehicle and also decitabine-Day1/4-treated mice, were replaced by AML cells by light microscope examination, but normal murine myelopoiesis was evident with decitabine-Day1/2 treatment ( Figure S2B). This impression was confirmed by flow cytometry analyses of the bone marrows: human CD45+ (hCD45+) cells were ~92% with PBS, ~63% with decitabine-Day1/4 and ~26% with decitabine-Day1/2 treatment ( Figure S2C). Spleens were enlarged and had effaced histology with vehicle or decitabine-Day1/4 treatment but had mostly preserved histology with decitabine-Day1/2 ( Figure S2D,E). Spleen weights as another measure of AML burden were >5fold greater with vehicle vs decitabine-Day1/4 but were lowest with decitabine-Day1/2 treatment ( Figure S2F).
Inhibit CDA / Inhibit ribonucleotide reductase in the de novo pyrimidine synthesis pathway / Schedule decitabine administrations to increase overlaps between drug exposure windows and malignant cell S-phase entries: CDA can be inhibited by THU, while de novo pyrimidine synthesis can be inhibited with deoxythymidine (dT) that inhibits ribonucleotide reductase in this pathway 41 . NSG mice tail-vein innoculated with 1 million AML cells each were randomised to (i) vehicle; (ii) THU+dT, (iii) decitabine; (iv) THU+decitabine; or (v) THU+dT+decitabine ( Figure 4A). PBS and THU/dT-treated mice developed signs of distress, and were euthanized, on day 42. Mice receiving other treatments were sacrificed 3 weeks later (day 63) to increase chances of seeing differences in AML burden between these treatments ( Figure 4A). Visual inspection of femoral bones of vehicle or THU/dT treated mice showed replacement of reddish functional hematopoiesis with whitish leukemia (Figure 4B), whereas femurs from THU/decitabine or THU/dT/decitabine-treated mice retained a normal reddish appearance ( Figure 4B). Accordingly, flow cytometry demonstrated replacement by human AML cells in PBS and THU/dTtreated mice (>90% hCD45+), that was improved to some extent by decitabine alone (~85% hCD45+) but and even greater extent by either THU/decitabine (~35% hCD45+) or THU/dT/decitabine (~42% hCD45+)( Figure   4C)(dT did not add further to the benefit from THU). Murine hematopoiesis was completely suppressed with PBS and THU/dT (0% murine Cd45+), almost completely suppressed with decitabine-alone (~5% mCd45+) but similarly preserved with THU/decitabine (~40% Cd45+) or THU/dT/decitabine treatment (~27% Cd45+)( Figure   4C). Hemoglobin and platelet levels were most suppressed, and white cell (peripheral leukemia) counts most elevated, with vehicle or THU/dT treatment, but only mildly to moderately suppressed with any of the decitabine containing regimens ( Figure 4D). Spleen weights as a measure of AML burden were ~4-fold greater with vehicle or THU/dT vs decitabine-alone treatment, and lowest with THU/decitabine-or THU/dT/decitabine ( Figure 4E).
Spleen histology confirmed replacement by AML cells (with necrotic areas) with vehicle or THU/dT ( Figure 4E), AML infiltration with decitabine-alone, but normal-appearance with THU/decitabine or THU/dT/decitabine treatment ( Figure 4E). We also evaluated the use of hydroxyurea to inhibit ribonucleotide reductase: hydroxyurea 100 mg/kg IP was administered on Day 1 before THU/decitabine on days 2 and 3 -hydroxyurea addition did not add benefit ( Figure S4). Thus, using THU to inhibit CDA augmented decitabine anti-AML activity but incorporation of dT or hydroxyurea to inhibit de novo pyrimidine synthesis did not. DNMT1-depletion by decitabine or 5-azacytidine is S-phase dependent, suggesting frequent, distributed administration, to increase chances of overlap between malignant S-phase entries and drug exposures, could be better than historical pulse-cycled administration. Accordingly, bone marrow AML burden was lowest with frequent, distributed administration of THU/decitabine 2X/week (Day 1,2) compared to pulse-cycled administration of THU/decitabine for 5 days every 4 weeks ( Figure S4).

In vivo treatment-resistance
Bone marrow cells harvested at day 63 when the AML-innoculated mice were doing well on-therapy demonstrated DNMT1-depletion, with the greatest DNMT1-depletion with THU/decitabine alternating with THU/5-azacytidine week-to-week (~65% DNMT1-depletion) vs THU/decitabine alone (~50%) or decitabine alone (~35%) or vehicle (~15%) (Figure 6E, S7). By contrast, bone marrow AML cells harvested after euthanasia for distress while receiving these same therapies demonstrated preserved DNMT1 levels measured by flowcytometry ( Figure 6E). These in vivo treatment-resistant AML cells demonstrated significant upregulations of CDA and CAD vs AML cells from mice treated with vehicle, with the greatest upregulations in cells from mice that received the alternating regimen and survived the longest ( Figure 6F). Thus, in vivo treatment-resistance was again by pyrimidine metabolism shifts adverse to DNMT1 target-engagement ( Figure S8).

DISCUSSION
Malignant myeloid cells proliferating through decitabine or 5-azacytidine therapy in vitro, in mice and in patients, evaded DNMT1-depletion via pyrimidine metabolism shifts adverse to pro-drug processing into the DNMT1depleting nucleotide Aza-dCTP. Critically, the protective metabolic shifts are induced acutely -decitabine added to AML cells rapidly suppressed DCK and upregulated UCK2 and CDA, and 5-azacytidine rapidly suppressed UCK2 and upregulated DCK and CDA, with the protein expression changes peaking 72-96 hours after a single pro-drug exposure. Others have reported that decitabine upregulated CDA in leukemia and solid cancer cells by 6 to 1000-fold within 96 hours 46,47 , while 5Aza upregulated DCK in leukemia cells by ~30% within 48 hours 18 .
These acute reconfigurations of pyrimidine metabolism arise from off-target actions of the agents that cause nucleotide imbalances: Decitabine inhibition of TYMS has been previously reported [38][39][40] , and here we found that TYMS protein is acutely depleted. A portion of administered decitabine (Aza-dC), after phosphorylation by DCK to Aza-dCMP, is deaminated by deoxycytidine deaminase (DCTD) into a deoxyuridine monophosphate (dUMP) analog Aza-dUMP. dUMP is the substrate for TYMS that rate-limits dTTP production. Aza-dUMP depletes TYMS (TYMS, like DNMT1, methylates carbon #5 of the pyrimidine ring that is substituted with a nitrogen in decitabine).
In this way, decitabine decreases dTTP that in turn increases dCTP (via less dTTP inhibition of ribonucleotide reductase-mediated reduction of CDP into dCDP [38][39][40] ). 5-azacytidine on the other hand depletes RRM1 protein and decreases dCTP 48 , presumably again as a result of the active nitrogen-substitution in the pyrimidine ring, although this has not been definitively evaluated. Stated simply, decitabine and 5-azacytidine drive dCTP levels in opposite directions, triggering distinct adaptive responses by pyrimidine metabolism, a network that senses and regulates nucleotide amounts 37 . DCK is particularly important for preserving dCTP levels, as shown by the decrease in dCTP in DCK-KO cells (shown also by others 49 ), consistent with DCK upregulation as an appropriate adaptive metabolic reponse to dCTP suppression by 5-azacytidine. UCK2 on the other hand is particularly important for dTTP maintenance, shown by the decrease in dTTP in UCK2-KO cells, consistent with UCK2 upregulation as a metabolic adaptation to dTTP suppression by decitabine. Both decitabine and 5-azacytidine acutely upregulated CDA, and acutely downregulated CAD. CAD, however, was upregulated in patients' and murine bone marrows at MDS/AML relapse/progression, the only discrepancy we found between pyrimidine metabolism patterns induced acutely versus found in stable resistant cells.
This mode of learned resistance, emerging from adaptive responses of the pyrimidine metabolism network to pro-drug perturbations, does not require mutation at the genetic level, perhaps explaining why several studies that have looked for correlations between MDS/AML genetics and decitabine/5-azacytidine resistance have generated inconclusive and even contradictory results 4,50-56 . Given that resistance-causing metabolic reconfigurations appear emergent, pre-treatment pyrimidine metabolism expression levels may also not necessarily predict response 4,50-58 . Consistent and predictable, however, were the automatic adaptive responses of pyrimidine metabolism to pro-drug exposures, enabling anticipation, outmaneuvering and even exploitation: first, serial administrations of decitabine scheduled to avoid decitabine-induced troughs in DCK expression was strikingly superior to schedules that coincided with DCK troughs, and alternating decitabine with 5-azacytidine week-to-week, timed (at least approximately) to exploit priming of each agent for activity of the other (UCK2 and DCK are maximally upregulated ~96 hrs after decitabine and 5-azacytidine respectively), was significantly superior to administration of either pro-drug alone. The timing of alternation was crucialalternating the prodrugs in 4 week cycles, or their simultaneous administration, did not add benefit.
Second, frequent, distributed schedules of administration, as a strategy to increase chances for overlap between malignant cell S-phase entries and drug exposure windows, were superior to conventional pulse-cycled scheduling. Such frequent, metronomic administration is feasible in mice and humans because we selected prodrug doses to deplete DNMT1 without off-target cytotoxicity 3,4,59 . Consistent with these observations, RNAsequencing analysis of patients' baseline bone marrows found that a gene expression signature of low cell cycle fraction predicted non-response to standard 5-azacytidine therapy 57 , and regulatory approval of decitabine and 5-azacytidine to treat myeloid malignancies involved lowering doses from initially evaluated, toxic high doses, and the administration of these lower doses more frequently 1 .
Third, combining THU, to inhibit the catabolic enzyme CDA, with decitabine and/or 5-azacytidine produced substantial extensions in anti-AML efficacy in vivoan important detail in such combinations was that the decitabine and 5-azacytidine doses were reduced to preserve a non-cytotoxic DNMT1-targeting mode of action 3,31,44,45 . Stated another way, simple dose-escalations of decitabine or 5-azacytidine are not a solution for resistance since this compromises the therapeutic-index foundation for success: high Cmax of these agents is cytotoxic via off-target anti-metabolite effects including depletion of TYMS (reviewed in 1 ), and while AML cells that survive initial pro-drug exposures get progressively educated for resistance as they indefinitely selfreplicate/proliferate, polyclonal normal myelopoiesis proliferates then differentiates in successive waves, each exposure-naïve and hence potentially vulnerable to the anti-metabolite effects of high doses of decitabine or 5azacytidine.
The key de novo pyrimidine synthesis enzyme CAD was downregulated in AML cells upon initial challenge by the pro-drugs but was upregulated in exponentially proliferating stably resistant cells by expression and by protein S1359 phosphorylation. Although we did not find a benefit from combining decitabine with dT or hydroxyurea to inhibit ribonucleotide reductase (that is in the de novo pathway) we and others have found promise in countering resistance by inhibiting other molecular targets in the pathway, e.g., using PALA to inhibit CAD or leflunomide to inhibit DHODH 60 . Thus, in next steps, we plan to evaluate inhibitors for different targets in the de novo pyrimidine synthesis pathway, and also to increase doses of the CDA inhibitor THU, since eventual AML progression to the optimized regimen in our pre-clinical in vivo studies here was characterized by even greater upregulations of CDA.
Thus, decitabine-and 5-azacytidine-resistance emerges from adaptive responses of the pyrimidine metabolism network to the perturbations caused by these pyrimidine nucleoside analog pro-drugs. These compensatory metabolic shifts individually and collectively impede engagement of the DNMT1 molecular target of therapy. These metabolic responses, being pre-programmed and consistent, can be anticipated, outmaneuvered and even exploited, using simple and practical treatment modifications that preserve the vital therapeutic index of non-cytotoxic DNMT1-depletion.

ACKNOWLEDGEMENTS
We acknowledge technical or other assistance from Quteba Ebrahem, Reda Mahfouz and Tae Hyun Hwang, and administrative support from JoAnn Bandera.

17.
Wang H, Chen P, Wang J, et al. In vivo quantification of active decitabine-triphosphate metabolite: a novel pharmacoanalytical endpoint for optimization of hypomethylating therapy in acute myeloid leukemia.   (Fig.S2, S3). THU+Dec/THU+5Aza alternated week to week (Fig.6) was superior to THU+Dec or THU+5Aza alone (Fig.5), or THU+Dec/THU+5Aza alternated month to month (Fig.5)  Figure 1. DNMT1 is not depleted at clinical relapse or with in vitro resistance. A) Decitabine (Dec) or 5azacytidine (5Aza) therapy decreased bone marrow DNMT1 at response (green) but DNMT1 rebounded to pre-treatment levels (dark blue) at relapse (red). Serial bone marrow biopsies from the same patent were cut onto the same slide, stained for DNMT1, and the number of DNMT1-positive nuclei was quantified objectively using ImageIQ software, in 13 individual patients and positive/negative controls (tissue blocks of HCT116 wildtype and DNMT1-knockout cells respectively). D = days of therapy. Pre-Rx = pre-treatment; HI = hematologic improvement; CR = complete remission; SD = stable disease; Rel. = relapse. Mean±SD of 3 image segments (cellular regions) per sample; p-value paired t-test, 2-sided. B) Expression of key pyrimidine metabolism enzymes at relapse/progression on Dec vs baseline. Cartoon shows key enzymes favoring (green) or impeding (red) Dec or 5Aza conversion into the DNMT1-depleting Aza-dCTP. Bone marrow cells aspirated pretreatment and at relapse/progression on Dec (13 patients, median duration of therapy 175 days, range 97-922) or 5Aza (14 patients, median duration of therapy 433 days, range 61-1155) were analyzed by QRT-PCR. Dec-resistant AML cells (THP1, K562, OCI-AML3 and MOLM13), and in parental THP1 AML cells treated with vehicle, Dec or 5-azacytidine (5Aza). Western blots.      Figure S2. Impact of decitabine scheduling to avoid vs coincide with DCK troughs. Figure S3. Impact of decitabine scheduling to avoid or coincide with DCK troughs. Figure S4. Impact of (a) using hydroxyurea to inhibit ribonucleotide reductase, and (b) scheduling THUdecitabine for 5 days every 4 weeks instead of 2 days every week.      Xcalibur was used to process and quantify raw data. Briefly, a processing method was built using MRM transitions and peak retention times from standards. All samples were processed with the same method to generate integrated total ion intensity (integrated peak area) for each analyte. Manual inspection was performed to confirm the peak assignment and integration. The final report value was normalized to the internal standards and total number of cells used to generate the extract.

Treatment of a patient-derived xenotransplant model of treatment-resistant AML. Patient-derived primary
AML cells from a patient with AML that had progressed on standard chemotherapy then decitabine salvage therapy, were transplanted by tail-vein injection (1.0 x10 6 /mouse) into non-irradiated 6-8 week old NSG mice.
Mice were anesthetized with isofluorane before transplantation. Mice were randomized to different treatments on Day 9 after inoculation, with treatments as indicated in each figure and legend. Doses of drugs used were: intra-peritoneal tetrahydrouridine (THU) 10 mg/kg given intra-peritoneal up to 3X/week; subcutaneous decitabine 0.2 mg/kg up to 3X/week (or 0.1 mg/kg when combined with THU); subcutaneous 5-azacytidine 2 mg/kg up to 3X/week (or 1 mg/kg when combined with THU); intra-peritoneal dT 2 g/kg up to 2X/week. Tail-vein blood samples for blood count measurement by HemaVet were obtained prior to leukemia inoculation, and at intervals thereafter as indicated in the figures. Mice were observed daily for signs of pain or distress, e.g., weight loss that exceeded 20% of initial total body weight, lethargy, vocalization, loss of motor function to any of their limbs, and were euthanized by an IACUC approved protocol if such signs were noted.
Bioinformatic and statistical analysis. Wilcoxon rank sum, Mann Whitney, and t tests were 2-sided unless otherwise stated because of apriori literature-based hypotheses (dCTP level analyses) and performed at the 0.05 significance level or lower (Bonferroni corrections were applied for instances of multiple parallel testing).
Standard deviations (SD) and inter-quartile ranges (IQR) for each set of measurements were calculated and      Figure 7B). Figure S7. DNMT1 was not depleted from AML cells at progression (time-of-distress as shown in Figure 7B) but was depleted at time-of-response (bone marrow harvested at Day 63 in a separate experiment). Flow cytometry. Positive and negative controls for DNMT1 were HCT116 colon cancer cells with wild-type DNMT1 and DNMT1 knock-out (KO).