Our study cohort included 57 consecutive cases of AML (median age: 68 years, range 29–83, 23 female and 34 male), 5 patients with acute promyelocytic leukemia (APL) (median age: 57 years, range 44–73, 3 female and 2 male) and 7 with myelodysplastic syndrome (MDS) (median age: 82 years, range 74–88, 1 female and 6 male) diagnosed and treated at the Hematology Unit of Policlinico Tor Vergata, Rome, Italy, during the years 2017-2020. AML patients aged <75 years received intensive chemotherapy according to ELN-2017 (20) and Gruppo Italiano Malattie EMatologiche dell’Adulto (GIMEMA) protocol (27), while patients aged >75 years received hypomethylating treatment or supportive care according to their performance status. Based on established risk criteria (28)), APL patients were treated with ATRA plus standard chemotherapy (n=1) or with the all trans-retinoic acid (ATRA) and arsenic trioxide (ATO) combination (n=4) as shown previously (29). Based on IPSS-R at diagnosis (30), MDS patients were classified as lower-risk (LR, ≤ 3.5; n=2) and higher-risk (HR, > 3.5;n=5) and were treated according to the European Society for Medical Oncology (ESMO) guidelines with either supportive measures or disease-modifying agents, respectively (31, 32). Fifteen healthy donors (HDs) were also included as a control group (median age: 48 years, range 33–66). Clinical and biological features of our patient cohort are shown in Table 1 . According to the declaration of Helsinki, all patients and controls gave informed consent to the study, which was approved by our Institutional Review Board.

Routine morphologic, immunophenotypic and genetic analyses were carried out in all patients at presentation using standard techniques, as detailed elsewhere (27, 33). Conventional karyotyping was performed on bone marrow (BM) diagnostic aspirates after short-term culture and analyzed after G-banding. The description of the karyotypes was done according to the International System for Human Cytogenetic Nomenclature (34). As for molecular analysis, total RNA was extracted from Lympholyte-H (Cedarlane) isolated BM and peripheral (PB) mononuclear cells (MNCs) collected at diagnosis and in selected cases during follow-up using standard procedures, and reverse-transcribed with random hexamers as primers (35). Molecular diagnostic analyses for recurrent gene translocations were performed in all AML patients at the time of initial diagnosis (36, 37). DNA was extracted using a column based Qiagen kit protocol (Qiagen, Hilden, Germany). Mutations of NPM1 and FLT3 genes were investigated at diagnosis on DNA extracted from BM-MNCs, using protocols reported elsewhere (38, 39). Patients’ genetic and cytogenetic features were used to stratify our population according to the 2017 ELN guideline version.

Based on availability of fresh primary AML blasts, we selected for further analysis 8 AML samples at diagnosis. MNCs samples were separated using Ficoll gradient. The cells were then labeled with the following anti-human antibodies: CD45 APC Vio 770 (Miltenyi Biotec); CD3 and CD19 Brilliant Violet 780, CD34 APC and CD33 PE-CF594 (Becton Dickinson); CD14 PE-Cy5.5 (Beckman Coulter). In order to exclude dead cells, we used the LIVE/DEAD Fixable Aqua Dead Cell Stain Kit (Invitrogen). The different populations were sorted by high-speed cell sorting, in the 6-way sorting MoFlo Astrios (Beckman Coulter) using a sequential gating strategy described in Supplementary Figure 1 . After doublets and dead cells exclusion, lymphocyte and monocyte population were used as controls and leukemia cells were selected for CD34 or CD33 expression.

Evaluation of the vitamin A, E and C was performed by isocratic high performance liquid chromatographic method with UV detection from serum and plasma. Vitamins purification was carried out using commercially available kits from Chromsystems Instruments & Chemicals GmbH (Grafälfing, Germany).

We designed specific quantitative TaqMan realtime PCR assays (RQ-PCR) for SLC2A1, SCL23A2, SLC2A3 and TET2 transcript quantification. For the generation of plasmid standard curves, RT-PCR products of each transcript were cloned into the plasmid vector pCR II-TOPO (Invitrogen, Groningen, The Netherlands). After a blue/white selection, plasmid DNAs were purified by Nucleospin Plasmid DNA purification kit (Macherey-Nagel) from recombinant colonies, and sequenced using the bigdye® terminator v3·1 cycle sequencing kit (Thermo Fisher Scientific). Sequencing analysis was performed on an ABI 3130 automated sequencer (Thermo Fisher Scientific), and data was analyzed using the seqscape software version 2·5. Plasmid DNA concentration was determined by absorbance measurement and 6 serial plasmid dilutions (107, 106, 105, 104, 103, 102 copies) were prepared for RQ-PCR assays in order to generate a standard curve. Primer and probe sequences used in this study (designed using the Software Primer Express, Thermo Fisher Scientific, Foster City CA, USA) are shown in Table 2 . RQ-PCR reactions and fluorescence measurements were carried out on the ABI PRISM 7700 Sequence Detection System (Thermo Fisher Scientific). Briefly, the reaction mixture of 25 µl contained 1×Master Mix (Thermo Fisher Scientific, Foster City, CA), 300 nM of each primer, 200 nM of Taqman probe and 5 µl of cDNA. Amplification conditions were: 2 min at 50°C, 10 min at 95°C followed by 60 cycles at 95°C for 15 s and at 60°C for 1 min. A threshold value of 0.2 was used and baseline was set to 3-15. The level of each transcript was normalized on the number of Abelson (ABL1) gene and expressed as copies number every 104 copies of ABL1. Following serial dilution experiments with water, the quantitative assays showed maximum reproducible sensitivity at 10-4 for SLC2A1, SLC23A2 and TET2, and 10-5 for SLC2A3 transcripts. The coefficient of the standard curves for the 4 assays ranged from 0.998 to 0.999 and the slope ranged from 3.296 to 3.359.

To assess the mutational profile of AML patients included in the study, we evaluated 30 genes, known to be frequently mutated in myeloid malignancies, using NGS and the Myeloid Solution by SOPHiA GENETICS (Saint-Sulpice, Switzerland). The resulting captured libraries were further processed on a MiniSeq sequencing platform (Illumina, San Diego, CA, USA). FASTQ sequencing files were then uploaded on the SOPHiA DDM platform for analysis by the specific technology. Only alterations identified as highly or potentially pathogenic were considered for analysis. The sensitivity of this NGS assay was ≥ 1%. The FASTQ files were further processed using the Sequence Pilot software version 4.1.1 (JSI Medical Systems) for alignment and variant calling. The functional significance of the somatic mutations was checked on the public Catalogue Of Somatic Mutations In Cancer (COSMIC) v69 database. Functional interpretation was performed using SIFT 1.03, PolyPhen 2.0 and MutationTaster 1.0 algorithms. Single‐nucleotide polymorphisms (SNPs), annotated according to the National Center for Biotechnology Information Single Nucleotide Polymorphism Database (NCBI dbSNP) were deleted from the analysis. The detection limit for variants was 5% variant allele frequency (VAF).

Primary PB-MNCs of 5 AML patients with blast proportion ≥ 60% and PB-MNCs from 2 healthy donors, were selected for culture studies. Briefly, 106 primary AML and HDs cells/ml were seeded into a culture flask and cultured in RPMI-1640 medium (Sigma-Aldrich, St. Louis, MO, USA) supplemented with 2 glutamine (EuroClone, Pero, Milan, Italy), 1% penicillin/streptomycin (Euroclone), 20% FBS (Signa-Aldrich), 10 ng/ml each of IL-3, SCF and FLT3LG (Peprotech) at 37°C in a humidified CO2 incubator.

Ascorbate (L-ascorbate, Sigma-Aldrich) was diluted in RPMI-1640 medium at 250 mM concentration. Drug aliquots were stored at −80°C and for each experiment a new aliquot was thawed and used at a final concentration of 1mM. Cultured cells and supernatant were collected at indicated times and analysed for vitamin C content. Ascorbate 1mM corresponds to 198.1 mg/L of ascorbate or 176.12 mg/L of vitamin C in its oxidized form.

To study intracellular vitamin C content, primary PB-MNCs from AML and healthy individuals pellets (range 0.5-5 x106 cells) were lysed on ice using 100 µl of ice-cold internal solution (Chromsystems Instruments) for 20 min before centrifugation at 13,000 rpm for 10 min at 4°C. The supernatant was collected and stored at -80°C until HPLC analysis. The amount of total vitamin C in the 100 µl of extraction buffer (internal standard) was divided by the number of cells, and expressed in femtomol/cell.

Statistical analysis was performed by means of GraphPad Prism software version 5.0 (LaJolla, CA). Non parametric tests (Mann Whitney and Kruskal-Wallis) were used to evaluate the association between vitamin A, E and C levels and patient characteristics (age, sex, karyotype, FLT3 and NPM1 mutational status, and 2017-ELN risk groups). Correlations between vitamin A, E and C levels, and white blood cells (WBC) or BM/PB blast counts was carried out using the Spearman method and the relative Rho coefficient. Survival curves were estimated according to the Kaplan-Meier method and were tested for significance by the log-rank test. All tests were 2-sided, accepting p<0.05 to indicate statistically significant differences, and confidence intervals were calculated at 95% level. Survival curves were built by using the software package SPSS version 17.0 (Chicago, IL).

We initially analysed the PB levels of vitamins A, C and E in a cohort of 29 AML (including 5 APL cases) and 7 MDS patients at diagnosis and in 8 HDs. As shown in Figure 1A , there were no significant differences in vitamin A and E levels between patient with myeloid neoplasms and HDs, except for lower level of vitamin E observed in MDS as compared with HDs (p=0.01). Conversely, vitamin C concentration was significantly lower in patients with AML and MDS as compared to controls (AML p<0.0001 and MDS p=0.006). Based on these preliminary results, we extended the analysis of vitamin C plasma level to a validation cohort including additional 7 HDs and 33 AML samples at diagnosis confirming the lower level of circulating vitamin C in AML patients at onset (p<0.0001, Figure 1B ).

Correlations between vitamin C level and clinical-biological characteristics were studied in the total cohort of 62 AML samples at diagnosis ( Table 1 ). Conventional karyotype or fluorescence in situ hybridization (FISH) were evaluable in 61/62 AML (98%) and 7 MDS (100%) cases. NPM1 and FLT3-ITD mutations were present in 15 of 57 AML cases (26.3% for both).

No significant correlations were found between vitamin C plasma concentrations, sex (p=0.851) and age in AML patients (p=0.104, Spearman’s Rho correlation: -0.218) and HDs (sex p=0.77, Spearman’s Rho correlation: -0.009; age p=0.84, Spearman’s Rho correlation: -0.05). There was an inverse correlation between circulating vitamin C levels and PB blast proportion (p=0.029, Spearman’s rank correlation= -0.298) ( Figure 2A ), while no relation was found with total WBC (p=0.232, Spearman’s rank correlation= -0.154) ( Figure 2B ), BM blast proportion (p=0.315, Spearman’s Rho correlation: -0.138), FLT3 (p=0.284) and NPM1 (p=0.547) molecular status ( Supplementary Figure 2A ), cytogenetics (p=0.367), and 2017-ELN risk stratification groups (p=0.754) ( Supplementary Figures 2B, C ). As to survival outcomes, baseline vitamin C levels did not play a predictive role for overall survival (OS) nor disease-free survival (DFS) ( Supplementary Figure 3A, B ).

To explore the dynamics of vitamin C plasma levels with regards to disease status we longitudinally track its changes upon disease remission and relapse. To that end, we studied 13 AML patients at the time of achievement of CR as well as 10 patients with relapsed/refractory disease. As shown in Figure 2C , in patients who achieved CR, vitamin C levels significantly increased as compared to those at the time of diagnosis in serial patients (p=0.04), but overall were lower than those found in HDs (p=0.006). In contrast, we observed a further significant reduction of vitamin C concentration in the 10 relapsed/refractory patients when compared to corresponding samples at AML onset (p=0.002) ( Figure 2D ).

Data on mutational status of common myeloid driver genes were available for 22 AML samples. A total of 54 mutations were found (median of 2.5 per patients) at a median VAF of 36.4% (range: 5.5-50.2%). Most common lesions were identified in NPM1 (n=8, 36%), FLT3 (n=5, 23%), DNMT3A (n=4, 19%) ASXL1, IDH1, NRAS, PTPN11, RUNX1 and TP53 (n=3 each, 14%) genes ( Figure 3 ). No correlation between mutational profile and vitamin C concentration was found.

To explore whether the observed differences in vitamin C levels may support dysfunctions of physiological uptakes mechanisms, a quantitative assessment of SCL23A2, SLC2A1, and SLC2A3 transcripts was performed on PB-MNCs samples harvested at diagnosis from 22 AML cases and compared to those of 15 HDs. As reported in Figure 4A , expression of the main vitamin C transporters SCL23A2, SLC2A1, and SLC2A3 was significantly reduced in AML patients as compared to HDs (p<0.0001, p=0.0009 and p=0.001, respectively). Moreover, since it has been shown that vitamin C drives the expression of a TET2-dependent gene signature in human leukemia cell lines (25), we investigated, in the same cohort of AML patients and HDs, the TET2 transcripts level. We did not find any differences in TET2 expression between patients and controls (p=0.842) ( Figure 4B ). Moreover, no correlations between vitamin C concentration and SCL23A2, SLC2A1, SLC2A3 and TET2 levels in AML patients with available paired samples (data not shown) were found.

Cytoplasmic vitamin C concentration was investigated in MNCs purified from AML samples at diagnosis (n=9), lymphocytes and blast cells (n=8) and HDs (n=7). As shown in Figure 5A , the intracellular concentration of vitamin C was significant lower in MNCs purified from AML patients as compared to HDs (p=0.002). To further identify the specific cellular fractions, AML blasts contained significantly less vitamin C than corresponding lymphocyte fractions (p=0.015) ( Figure 5B ).

The flux of vitamin C through the cells is controlled by regulation of active and facilitated diffusion transport, mediated by SVCT and GLUT transporters, respectively. Since expression of the main vitamin C transporters and intracellular vitamin C concentration were low in AML patients, we investigated the ability of primary MNCs purified from 5 AML patients at onset, to absorb vitamin C by measuring its uptake across the cell membrane at 1, 2, 3, 5 and 7 days after incubation with 1mM ascorbate. As reported in Figure 6A , the vitamin C cellular uptake increased in a time-dependent manner, with maximum ascorbate incorporation after 72h. Vitamin C intracellular concentration progressively decreased, reaching the basal level value of untreated blasts after 5 and 7 days. The same experiment in PB-derived HD-MNCs showed a similar ascorbate uptake with a peak at 72h, albeit at a lower rate ( Figure 6B ). These results suggest that although primary AML blasts are characterized by low levels of SVCT and GLUT transporters, neoplastic cells are still able to absorb vitamin C and that vitamin C transporters are not the only factors dictating the ability of leukemic blasts to regulate such process. Moreover, vitamin C was undetectable in the culture medium just after 24h in the absence of cells, due to natural decay but, by contrast, was detectable in medium with cultured blasts until 72h indicating an active role of blasts to maintain vitamin C in the culture medium. Altogether, these findings underline a specific role of AML blasts in vitamin C turnover ( Figure 6C ).