ALKBH overexpression in head and neck cancer: potential target for novel anticancer therapy

The nine identified human homologues of E. coli AlkB 2-oxoglutarate (2OG) and Fe(II)-dependent dioxygenase, ALKBH1-8 and FTO, display different substrate specificities and diverse biological functions. Here we discovered the combined overexpression of members of the ALKBH family in head and neck squamous cell carcinomas (HNSCC). We found direct correlation of ALKBH3 and FTO expression with primary HNSCC tumor size. We observed unidentified thus far cytoplasmic localization of ALKBH2 and 5 in HNSCC, suggesting abnormal role(s) of ALKBH proteins in cancer. Further, high expression of ALKBHs was observed not only in HNSCC, but also in several cancerous cell lines and silencing ALKBH expression in HeLa cancer cells resulted in dramatically decreased survival. Considering the discovered impact of high expression of ALKBH proteins on HNSCC development, we screened for ALKBH blockers among newly synthetized anthraquinone derivatives and demonstrated their potential to support standard anticancer therapy.

Head and neck squamous cell carcinomas (HNSCC) include a variety of tumours of different epidemiology and ethology, originating in the oral cavity, hypopharynx, nasopharynx, oropharynx, or larynx. HNSCC is the seventh most common malignancy worldwide, responsible for approximately 400 000 deaths per year 1 In cancer tissues increased dynamic of metabolic processes result in high level of nucleic acid modifications and in consequence, induction of DNA repair systems 2,3 . These system expression leads to the removal of DNA lesions before they become toxic to the rapidly dividing cancer cell, ensuring tumour welfare and creating a major mechanism of resistance to anticancer therapy 3 .
In Escherichia coli, AlkB (EcAlkB) protein is a prototypical, one-protein DNA/RNA repair system, part of the so-called "adaptive response". It is the best known member of the 2-oxoglutarate (2OG) and Fe(II)-dependent dioxygenase superfamily 4 . The human genome encodes nine EcAlkB homologs: ALKBH 1-8 and FTO (fat mass and obesity associated protein) 5 . They fulfil a variety of biological functions, such as DNA repair, involvement in the metabolism of different RNA species, histone demethylation, acting as methyltransferases or taking a part in fatty acid metabolism 6 .
Overexpression of individual ALKBH proteins has been detected in overwhelming majority of various types of cancer 7 , suggesting a pro-carcinogenic role of these proteins (Table 1), e.g. FTO overexpression correlates with increased cancer cell survival in breast cancer, under the conditions of glutamine deficiency 8 . Likewise, Alkbh2 knockdown in bladder cancer tissue limited tumour development, while ALKBH2 down-regulation sensitized cells to alkylating agents in glioma 9,10 and cisplatin in lung cancer 11 . Considering these findings, inhibitors of ALKBH proteins could act as anticancer substances. The group of anthraquinones, including rhein and emodin, is particularly promising because these compounds exhibit anti-inflammatory, anti-bacterial, and anti-cancer properties 12 . Moreover, rhein has been shown to inhibit EcAlkB, ALKBH 2, 3, and FTO activity 13,14 .
For the first time, we detected overexpression of a majority of the ALKBH proteins in clinical HNSCC samples and high levels of expression in several cancer cell lines. Importantly, some of the proteins changed localization in cancer cells, indicating specific function in different cell compartments, and correlated with clinical tumour parameters. Finally, we modified emodin to improve its inhibitory effect against ALKBHs and show anticancer activity of some of the new derivatives obtained on different cancer cell lines.

Results expression of ALKBH proteins correlates with HnScc development. ALKBH levels are increased
in HNSCC. The clinicopathological characteristics of the 41 patients are shown in Table 2. The table includes main data concerning patients, TNM tumour parameters, and other diseases.
Using western blot (WB) analysis, we detected overexpression of seven out of the nine ALKBH proteins (ALKBH1, 2, 3, 4, 5, 8, and FTO) in HNSCC cancer tissues, as compared to the surrounding, unaffected tissue (Figs 1A and S1-7). The expression of ALKBH 6 and 7 was not detected by our WB method. We used TET2 dioxygenase expression as the negative control and observed that it was expressed at equal levels in cancer and normal tissue (data not shown). The highest expression levels were observed for ALKBH4, 5, and FTO (Fig. 1C). Further analysis indicated that, among overexpressed ALKBH proteins, the greatest difference between tumour and surrounding tissue was observed for ALKBH2 (5-fold), FTO (4-fold), ALKBH1 (3-fold), and 5 (2-fold) (Fig. 1D). Thus, simultaneous overexpression of the indicated dioxygenases in cancer tissue may be used in cancer diagnosis as a meta-marker. Towards this aim, the best candidates are ALKBH1 and FTO, according to receiver operating characteristic (ROC), where false positive rate were below 26% (Fig. S8).
Analysing the levels of ALKBHs in HNSCC tumours, we observed the simultaneous increase in expression of at least four ALKBH proteins in approx. 88% of patients and at least five ALKBHs in over 50% of patients. More than 25% of patient samples tested showed high expression of all seven ALKBH proteins (Fig. 1E). This implicates similar substrate specificity and/or common regulation of these proteins. To find out if ALKBH overexpression is a common phenomenon in cancer cells, we also assessed expression of ALKBH proteins in various cancerous cell lines: HeLa (cervix), U87 (brain), BICR 18 (larynx), Jurkat E6-1 (peripheral blood) and H929 (bone marrow), and embryonic HEK293 (kidney). As a model of normal tissue we used EUFA30 (fibroblast) (Fig. 1B). In almost all of the cancerous cell lines-despite different origin-we observed expression of all of the ALKBH proteins tested. The highest expression of ALKBH proteins was detected in HeLa, Jurkat, H929, and HEK293, while the lowest was detected in BICR18 and EUFA30, the control.
Correlation between the levels of expression of ALKBH proteins. Next, we assessed the relationship between the expression levels of individual ALKBHs protein in normal surrounding tissue and in HNSCC ( Fig. 2A,B). In the healthy periphery, we identified three correlated pairs, with the highest Spearman's rank correlation coefficients (ρ > 0.55, p < 10 −3 ) calculated for ALKBH5-FTO, ALKBH1-FTO, and ALKBH3-ALKBH5 pairs. In the cancer tissue, we observed a similar correlation involving ALKBH5-FTO, ALKBH3-ALKBH5, and ALKBH1-ALKBH5, but also noted that the level of ALKBH2 was correlated with that of ALKBH5 and ALKBH1 with ALKBH3.
To investigate further these correlations, we used small interfering RNA (siRNA) for ALKBH1, 2, 3, 4, 5, FTO genes and using WB analysis, we recognized the expression levels of ALKBH1, 3, 5 and FTO (Fig. 2C). The HeLa cells were chosen as highly expressing the ALKBH proteins. In the case of ALKBH2, 4, and 8 genes, their silencing led to the downregulation of almost all ALKBHs tested. Silencing of ALKBH1 or FTO did not influence the level of the investigated proteins. Interestingly, only in two cases we observed the opposite effect, namely, ALKBH8 silencing led to the elevated level of ALKBH1, while FTO silencing increased ALKBH5 level. Moreover, we noticed downregulation of ALKBH3 and FTO after the silencing of ALKBHs in the most cases. tionship between ALKBH level and cancer progression, we compared protein expression levels with tumour parameters. We detected significantly higher levels of ALKBH3 and FTO in larger primary tumours (type T4 in TNM classification), as compared to smaller ones (type T2 in TNM classification) (Fig. 3A). Interestingly, in type T2 tumours, the level of ALKBH3 was similar or slightly higher than in adjacent tissue, whereas in type T4 tumours ALKBH3 level was strongly elevated in comparison to unaffected surrounding tissue. Meanwhile, increased levels of FTO were observed in all examined groups (Fig. 3C). Thus, the content of ALKBH3 and FTO increases along with the tumour size. We did not observe any other correlations between tumour size or primary tumour extension and the other ALKBH proteins; ALKBH levels were not correlated with tumour invasiveness (G parameter), metastasis (N parameter), or TNM stage (Fig. 3A). Finally, we assessed the impact of ALKBH expression level on HeLa viability. Using siRNA, we silenced single/double ALKBH proteins and found that silencing ALKBH1, 4, 5, and FTO significantly influenced cell viability (Fig. 3D). On the other hand, silencing ALKBH2, 3, 8 and the combined silencing of ALKBH2 and 3 did not influence cell viability. When silencing performed in EUFA30, HeLa, and Jurkat cell lines with siRNAs targeted towards selected ALKBHs, we observed that downregulation of ALKBH1, 4, or FTO slightly promoted necrosis in HeLa cells to the values of 25.5, 21.7, and 30.6%, respectively (Fig. 3E). On the other hand, Jurkat cells did not show any sensitivity to siRNAs used. Although in EUFA30 cells, ALKBH1 and FTO siRNAs induced apoptosis (9 and 7%, respectively), the levels of cell viability were still in the range of the control.

Elevated level of RNA N 6 -methyladenosine in HNSCC.
Having discovered high level of ALKBH proteins in HNSCC, we decided to assess the level of N 6 -methyl-adenosine (N 6 meA), the most abundant RNA modification in eukaryotic cells and the substrate for FTO and ALKBH5 proteins. Surprisingly, despite high expression of FTO/ALKBH5 in cancer tissue, the level of N 6 meA was also significantly elevated, in comparison to normal tissue (Fig. 4A). Nevertheless, we observed a great deal of variation in the FTO/ALKBH5 vs N 6 meA ratio in individual patients (Fig. 4B).  the cellular localization of ALKBHs in HnScc cells. Confocal microscopy analysis was used to examine any changes in the subcellular localization of ALKBH proteins in HNSCC cells, as compared to the normal, surrounding tissue (Table S1). For ALKBH1, 4, 8, and FTO we observed the expected, previously-described localization. However, for ALKBH2 and 5, we found that these nuclear proteins were also highly expressed in cytoplasm. Additionally, we observed that ALKBH3 was expressed only in cytoplasm (Fig. 5). These changes in protein localization suggest that ALKBHs may play different roles in cancer and normal cell metabolism.

Design of ALKBH inhibitors
The western blot analysis and immunofluorescence on clinical HNSCC samples and RNA interference in human cell line indicate the involvement of ALKBH proteins in cancer survival and growth. As a result, the ALKBH inhibitors may offer a new approach to anticancer therapy. Towards this aim, we performed molecular modelling of ALKBH proteins with a set of selected antraquinones with proven anti-cancer activity, as well as newly synthetized chloro-derivatives.

In silico modelling of the ALKBH -anthraquinone (enzyme-ligand) complexes. Because no struc-
tures of the proper forms of holoenzyme complexes with anthraquinone derivatives are accessible in PDB (Fig. S9), we performed in silico screening for all ALKBH proteins by means of homology modelling, followed by molecular docking. The binding affinities estimated for 60 various complexes of the ten ALKBH proteins (ALKBH1-8, FTO, and EcAlkB) with six anthraquinone derivatives ranged from 0.02 to 10 μM (Fig. 6A). EcAlkB www.nature.com/scientificreports www.nature.com/scientificreports/ and FTO were the most preferred targets for all ligands, ALKBH2 and ALKBH3 were clearly less preferred, while the remaining six proteins did not bind the tested anthraquinones (K diss > 1 μM). These results are in agreement with previous experimental data demonstrating that rhein inhibits bacterial EcAlkB and eukaryotic ALKBH 2, 3, and FTO proteins 13,14 .
The structure of the FTO-rhein complex was close to that accessible in PDB, but the ligand was placed deeper at the binding site (RMSD < 2 Å). In contrast to ALKBH3-for which the carboxyl group of rhein interacts with Arg118 and Arg131 (Fig. S10) -no salt-bridges were identified for FTO-ligand interactions; thus, not surprisingly, 7-chloroemodin was found to be the best ligand, whereas emodin and 5-chloroemodin scored slightly lower (Figs S11 and S12).

Impact of anthraquinones on stability and enzymatic activity of AlkB, ALKBH3, and FTO.
We tested the affinity of our newly synthesized anthraquinones, 7-chloroemodin and 5,7-dichloroemodin, (Fig. S13) for ALKBH3 and FTO and assessed their effect on the enzymatic activity of targeted proteins using the thermal shift assay (TSA). The thermodynamic parameters for thermal denaturation of both proteins were measured for their holo forms, in the presence of 2OG and Mn 2+ , an analogue of Fe 2+ .
The strongest ligand for ALKBH3 was rhein (Figs 6B and S14A), because its addition increased the melting temperature the most. Emodin, aloe-emodin, and 5,7-dichloroemodin bound to ALKBH3 slightly more weakly.
All of the tested ligands increased the melting temperature of the FTO holo form (52.8 ± 0.6 °C) by more than 10 °C (Fig. 6C). The most pronounced effect was observed for 7-chloroemodin and 5,7-dichloroemodin, with these complexes melting at 71.5 ± 0.2 °C and 67.4 ± 0.1 °C, respectively.
For EcAlkB, ALKBH3, and FTO, the inhibitory effect was further quantified by measuring inhibition of demethylation activity in vitro (Fig. 6D). We used 3-methylcytosine (3meC) on TT(3meC)TT pentamer as a substrate and the anthraquinones listed in Fig. 6D as inhibitors. The best EcAlkB inhibitors were 7-chloroemodin and 5,7-dichloroemodin, with IC 50 of 19.8 μM and 16.4 μM, respectively, followed by rhein with IC 50 of 20.5 μM. The remaining anthraquinones displayed at least 10-fold lower activity. Enzymatic activity assay also confirmed that rhein is the best inhibitor of ALKBH3, with IC 50 of 40 nM (Fig. S14). Aloe-emodin displayed lower activity  www.nature.com/scientificreports www.nature.com/scientificreports/ (IC 50 8.8 μM) and the other anthraquinones were inactive (IC 50 > 100 μM). For the case of FTO, activity that was monitored using N 6 meA demethylation assay and rhein was the most efficient inhibitor with IC 50 of 7.1 μM, followed by 7-chloroemodin with IC 50 of 28.1 μM, which was more active than emodin. In summary, rhein, 7-chloroemodin, and 5,7-dichloroemodin are the most effective ALKBH inhibitors, but possess different affinities towards the individual proteins.

Cytotoxic effect of anthraquinones on normal and cancer cells.
We evaluated the cytotoxic efficacy of the compounds under study, on cancerous HeLa, U87, BICR18, embryonic HEK293, and normal EUFA30 cells (Fig. 7A). We found that emodin is the most cytotoxic among all natural anthraquinones tested. Moreover, it is more toxic to cancerous than healthy cells. The IC 50 values for emodin were 110.1, 42.7, and 81.1 μM for HeLa, U87, and BIRC18, respectively. Its therapeutic index (TI) varied from 1.25 to 3.2. Within newly synthetized compounds, 5,7-dichloroemodin was the most potent, with IC 50 equal to 10.6 and 39.8 μM and a therapeutic indexes of 9.6 and 2.6 for HeLa and BICR18, respectively. Summarizing, emodin and 5,7-dichloroemodin showed higher cytotoxicity against cancerous than healthy cell lines. Noteworthy, the biggest therapeutic indices, equal 14.6 μM, showed aloe-emodin against HeLa cells. Unfortunately, this compound was not cytotoxic against HEK293 and U87 cell lines. Additionally, HeLa cells were less resistant to anthraquinones.
In summary, the natural anthraquinone emodin and our novel derivative 5,7-dichloroemodin demonstrate promising potential as inhibitors for anticancer therapy.

Discussion
In the last decade, the family of AlkB dioxygenases has been extensively studied, especially in the context of new substrates, but also in terms of their role in the development of so-called "diseases of affluence", including obesity, type-2 diabetes, cancer etc. Several recent publications have reported increased levels of individual ALKBHs in various types of human cancer (Table 1). Here we demonstrated that in HNSCC, the expression level of seven out of nine ALKBHs (1, 2, 3, 4, 5, 8, and FTO) was significantly increased in 68-90% of the samples. This shows, www.nature.com/scientificreports www.nature.com/scientificreports/ for the first time, that not just one single ALKBH, but the whole group of ALKBH proteins is overexpressed in HNSCC tumours. At least four different ALKBH proteins were highly expressed in 88% of tested tumours and in 25% of cases, all seven proteins were overexpressed. We found highly expressed ALKBH proteins in several cancerous cell lines, suggesting that the simultaneous overexpression of several members of ALKBH family proteins is a common process accompanying cancerogenesis. This hypothesis is supported by the observation of strong co-expression of particular ALKBH pairs in HNSCC, namely ALKBH1 with 3 and ALKBH2 with 5. Moreover, in the case of HeLa cells, the silencing of ALKBH2, 4, or 8 genes reduced expression of other homologs (Fig. 2C). Keeping in mind that FTO and ALKBH5 share the same substrate specificity 15,16 , not surprisingly that the silencing of the FTO gene leads to upregulation of ALKBH5 protein. Furthermore, one can suppose that the ALKBHs co-express in the similar way and can be co-regulated at the expression level. Further, silencing of ALKBH1, 4, 5, and FTO markedly decreased HeLa cancer cell survival (Fig. 3), suggesting involvement of these ALKBHs in cancer development by DNA/RNA/protein modification. In contrast, silencing of ALKBH2, 3, 8 or both, ALKBH2 and 3, did not affect cancer cell viability. This is in agreement with prior reports showing that knocking down ALKBH2 or 3 did not influence mice phenotype and viability 17 . We can conclude that overexpression/co-expression of the whole group of ALKBH proteins correlates with cancer development, including its different types, indicated by: (i) the high expression of almost all of the ALKBHs in HNSCC and in various cancer cell lines; (ii) decreased cancer cell viability following ALKBHs silencing; (iii) previously reported high expression of single ALKBHs in different types of cancer; (iv) the siRNA-treated mice with developed cancer show diminishing tendency of cancerous tissue. Summing up, overexpression of a single or a group of ALKBH proteins could be used as a marker for cancer diagnosis 18 . The samples obtained by biopsy (or surgery) could be subjected to Western blot analysis and adequate chemotherapy, e.g. combination of alkylating agents and antraquinone derivatives, could be established.
Further, we demonstrated that ALKBH3 and FTO levels are correlated with primary tumour size (T in TNM classification): protein expression was higher in larger T4 tumours, as compared to smaller T2 tumours. The correlation between increased FTO expression and tumour size may suggest a new regulatory role for this protein in tumour development. On the other hand, the high amount of the repair protein ALKBH3 present in larger tumours may be explained by accumulation of alkylation damage to DNA/RNA, due to metabolic instability of the developed cancer. We can presume that other ALKBH proteins may also promote or sustain tumour welfare, considering their high expression levels. www.nature.com/scientificreports www.nature.com/scientificreports/ Interestingly, we also observed an increased global level of N 6 meA in HNSCC, the most abundant internal modification in mammalian mRNA, suggesting an abnormal methylation profile in HNSCC. N 6 meA is involved in mRNA stability, translation efficiency, splicing, RNA-protein interactions, and more 15 . It is also responsible for www.nature.com/scientificreports www.nature.com/scientificreports/ cancer stem cell pluripotency, cell differentiation, proliferation and metastasis 19 . Our findings reflect a discrepancy in FTO/ALKBH5 action as the demethylase regulating N 6 meA level in mRNA in normal vs cancer cells. These findings are supported by the results of Merkeinestein group who has shown that an additional copy of FTO gene did not influence the global level of N 6 meA 20 . However, FTO may impact demethylation of selected transcripts, like in acute myeloid leukaemia where FTO plays a critical oncogenic role through demethylation of ASB2 and RARA mRNA (protein that are responsible for blood cell differentiation) 21 , also demethylation of RUNX1T1 (negative regulator of adicipogenesis) 15 , C/EBPβ (stimulator of hepatocytes and adipocytes growth) 22 , PPARγ (adipocytes differentiation) 23 , GAP-43 (growth of axons) 24 . Collectively, these findings prove that FTO is involved in various cancer development.
The different role of ALKBHs in normal vs cancer cells was also confirmed by the results of subcellular localization using immunofluorescence. In HNSCC we observed the unexpected localization of nuclear proteins ALKBH2 and 5 in the cytoplasm, while ALKBH3 was found only in cytoplasm. These changes in subcellular localization suggest that these proteins can be involved in new, unknown functions that may be connected with tumour growth. Further, changes in localization of the remaining ALKBH proteins may be the result of a specific pathological state and can be taken under consideration as a prognostic marker for cancer transformation 25 .
The relatively high expression of ALKBHs in tumor tissues correlate with cancer development 26,27 making these proteins the potential targets for anticancer therapy. We focused on natural antraquinones as inhibitors of ALKBHs and to improve the solubility and the inhibitory effect of these compounds, we substituted hydrogen(s) with chloride(s) atoms in emodin. Molecular docking analysis and in vitro activity assay revealed that all tested antraquinones strongly interact with EcAlkB, FTO, ALKBH2, and ALKBH3. Rhein shows the strongest inhibitory effect among compounds under study; however, emodin and newly synthetized 5,7-dichloemodin were the most toxic against cancer cell lines. These data are in agreement with the results published by Li and co-workers (2016), who established the IC 50 values for rhein against EcAlkB and ALKBH3 were 12.7 and 5.3 μM, respectively 14 . Similarly to our observation, the authors did not found any cytotoxic effect of rhein against U87 cells. Chen and co-workers (2012) calculated the rhein IC 50 against FTO protein activity in the range of about 30 μM 13 . The similar results were obtained by Wang and co-workers, where emodin has shown higher cytotoxic effect than rhein on human proximal tubular epithelial HK-2 cell line, and only emodin induced apoptosis 28 . On the other hand, the facts that Jurkat cells, although expressing ALKBHs at higher level than HeLa cells, were more resistant to antraquinones, and that EUFA30 cells poorly expressing ALKBHs were least sensitive imply that the emodin derivatives act not only by ALKBHs inhibition but also interfere with other metabolic pathways. The assessment of apoptosis/necrosis in EUFA30, HeLa, and Jurkat cells treated with siRNAs directed towards selected ALKBHs confirmed that at least, but not last, the inhibition of the dioxygenases studied leads to cell death, since the only visible impact of siRNAs could be observed in the case of HeLa cells, similarly to the results shown for anthraquinones. If this were not the case, the Jurkat cells would be more sensitive than HeLa. Indeed, it was shown that rhein and emodin also interact with other proteins. Emodin inhibits VEGFR2, PI3K, and p-AKT expression in colon colorectal HCT116 cells 29 . Moreover, emodin inhibits tumor necrosis factor alpha-induced calcification via the NF-kB pathway 30 . Additionally, rhein influences cytoskeleton regulation, protein folding, or transcription control in breast cancer MCF-7 cell lines 31 . Also, the synergistic way of action cannot be excluded.
After the anthraquinone treatment, we observed mainly apoptosis in EUFA30 and Jurkat cells, and necrosis in HeLa cells. Varela-Ramirez and co-workers (2011) obtained similar results testing diphenylmethyltin chloride (DPMT) on various cell lines. They observed that DPMT induces necrosis (more than 40% of cells) on HeLa cells and mostly apoptosis on Kit-225 (human T lymphoma) or EL4 (murine T lymphoblastic) 32 .
For the first time we proved majority of human ALKBH proteins overexpress in HNSCC. Based on the key role of ALKBHs in HNSCC development, ALKBHs should be considered as promising new targets for the diagnosis and treatment of various type of cancer that exhibit high level of ALKBH proteins. In particular, the synthesis and use of new, more powerful ALKBH inhibitors could be a promising strategy supporting anti-cancer therapies based on alkylating agents. This new, modern approach could be used not only in head and neck cancers, but also in other types of cancer that demonstrate ALKBH overexpression.

Materials and Methods
clinical samples. 41 tissues (blinded samples) of head and neck carcinomas (29 males and 12 females) were collected at the Department of Otorhinolaryngology, Faculty of Medicine and Dentistry at Medical University of Warsaw between 2011 and 2014. 73.3% of these patients suffered from laryngeal cancer. All of the cancers were histologically verified as squamous cell carcinomas. The research was performed in accordance with the relevant guidelines and regulations. Tumours were classified according to the TNM staging system. Patients were not treated with chemo-or radiotherapy. Samples after histopathological examination were available in archives. Patients agreed to use the samples according to current procedure. Ethical Comity accepted biochemical tests on the samples prior to subsequent utilization according to routine procedures of biohazard. Samples were anonymized with no access to patient personal data. The clinicopathological data are presented in Table 2. www.nature.com/scientificreports www.nature.com/scientificreports/ with horse-radish peroxidase. All incubations were performed in 5% milk/PBST (PBST -phosphate buffer saline with 0.1% TWEEN-20). Chemiluminescence was measured using the ChemiDoc MP Imaging System (Bio-Rad). Total protein was standardized in four steps: (i) equal masses of the tissue were taken for extraction in RIPA buffer (50 mg); (ii) extract was then assayed by Bradford assay for protein content; (iii) equal amounts were loaded on the gel and verified by Coomassie blue staining; (iv) protein transferred to the nitrocellulose membrane was visualized by Ponceau-S reversible staining prior to the final western blot. Additionally, in the case of several proteins, primary antibodies recognized epitopes of different molecular mass. To define which band corresponds to which examined ALKBH protein, we decreased the protein levels by using RNA interference. N 6 meA RnA methylation assay. Frozen samples were homogenized in liquid nitrogen. Total RNA was extracted by TRIzol Reagent (ThermoFisher) according to the manufacturer's protocol. Concentration and purity of RNA was measured spectrophotometrically by Nanodrop 2000c (ThermoScientific). Two hundred ng of total RNA was used in quantification of N 6 meA level with use of N 6 meA RNA Methylation Quantification Kit (EpiQuik). The assay was carried out according to the manufacturer's protocol. Absorbance at 450 nm was measured using a scanning multiwall multimode spectrophotometer (DTX 880, Beckman Coulter). To determine relative methylation status of RNA samples, following Eqs. (1) and (2) were used: Expression and purification of hFTO, AlkB, and ALKBH3 proteins. The plasmid with hFTO was constructed using StarGate cloning system (IBA Life Science) and introduced into E.coli BL-21. Bacteria were cultured in 37 °C to OD 600 = 1, induced with 1 mM IPTG, cultured for 16 h at 16 °C, harvested, resuspended in PBS buffer with protein inhibitor cocktail and lysozyme. After 30 min, cells were sonicated 6 times for 30 sec with ultrasound (250 W). After centrifugation (20 000 x g, 10 min), the His-hFTO protein was purified using Ni + -Sepharose and size exclusion chromatography (Enrich SEC. 650). Purity of the protein was verified using SDS-PAGE. The purified protein was stored in the lysis buffer with 50% glycerol at 80 °C. Expression and purification of AlkB and ALKBH3 dioxygenases was performed as previously described 34,35 . thermal shift assays for ligand binding. The assay was developed previously to monitor ligand binding to AlkB 36 . Fluorescence experiments were performed using a Varian Cary Eclipse spectrofluorimeter. Emission was monitored at 345 nm (excitation at 280 nm) over a temperature range of 40-95 °C with heating rate of 1 °C/ min. The protein samples were dissolved in HEPES buffer at pH 7.5 with 2 mM dithiotreitol (DTT) and 150 mM NaCl to a final concentration of 1.2 µM. The process of thermal unfolding was performed at a concentration of 5 µM of the tested compound, either in the presence or absence of cofactors (50 µM 2OG and 100 µM Mn(II)). The thermal unfolding of the holo form of the protein was always monitored and further used as the reference. All the calculations were performed using Origin 9.0 software, assuming the simplest model of two-state (Folded-Unfolded) transition.
In vitro inhibition assay. AlkB Table 2). 1 × 10 6 cells were seeded onto a 100 mm diameter dishes. After 24 h of incubation, siRNAs mixtures (with lipofectamine, in OPTI-MEM medium) were added. After 48 h, cells were collected for WB or flow cytometry analysis. For WB analysis, adherent cells were rinsed three times with PBS, scraped, and centrifuged (800 x g, 10 min, 4 °C). Next, cells were re-suspended in 100 μL RIPA buffer with a 1 x mammalian protease inhibitor cocktail (Sigma-Aldrich). The resulting lysates were spun down at 30 130 x g to remove insoluble particles, assayed for protein content using the Bradford assay, and loaded onto the SDS-PAGE at the protein amount of 20 μg per line. The strength of WB signals corresponding to the particular ALKBHs were calculated by densitometry. The expression changes by 30% were considered as significant.
Viability assay. Exponentially growing cells at the density of 3 × 10 3 cells/well were seeded onto a 96-well plate, cultured for 18 h followed by the respective treatment of antraquinones or its derivatives in various concentrations or 34 nM RNA interference of ALKBH proteins, and then cultured for a an additional 48 h. Alamar Blue (Invitrogen) viability assay was performed according to manufacturer's protocol: after 4 h incubation, emission at 590 nm was measured with excitation at 560 nm using a scanning multiwall multimode spectrophotometer (DTX 880, Beckman Coulter). The experiment was carried out at least three times, with three replicates for each inhibitor concentration. After background subtraction, inhibition rates, IC50, values were calculated as the concentration of the component that inhibits cell growth by 50%. All the calculations were done using Origin 9.0 software. flow cytometry. The Annexin V-FITC apoptosis detection kit (ApoAlert Annexin V-FITC Apoptosis Kit, TaKaRa) was used to detect apoptosis by flow cytometry. Cells were seeded at 6-well plates at concentration of 5 × 10 5 cells/well, cultured for 18 h, and then the tested agent was applied for the indicated periods. Afterward, cells were washed with PBS, resuspended in binding buffer, and anti-Annexin V FITC-conjugated antibody and propidium iodide were added to 100 µl aliquots. The mixtures were incubated for 15 min at room temperature, supplemented with binding buffer to 500 µl, and processed using BD FACSCalibur (BD Biosciences). Data were analyzed in Flowing Software version 2.5.1 (Flowing Software, http://www.uskonaskel.fi/flowingsoftware).

Molecular modelling.
All calculations were performed with the aid of Yasara Structure package (ver. 17.1.28) using amber03 force-field. Initial structures of human ALKBH2 (3s57), ALKBH3 (2iuw), ALKBH5 (4oct), ALKBH7 (4qkd) ALKBH8 (3thp) and FTO (4zs2), rice ALKBH1 (5xeg) and E.coli (3i3q) were adopted from the appropriate PDB records. Each of these structures carried 2OG and metal cation (either Mn or Fe) in the catalytic centre, while all other ligands were removed. The structures of ALKBH1, ALKBH4 and ALKBH6 were modelled by homology on the basis PDB structures selected automatically by the combination of blast E-value, sequence coverage and structure quality. For each template, up to 5 alignments with the target sequence were used, and up to 50 different conformations were tested for each modelled loop. The resulting models were evaluated according to the structural quality (dihedral distribution, backbone and side-chain packing) and that with the highest score of these covering the largest part of the target sequence was used as the template for a www.nature.com/scientificreports www.nature.com/scientificreports/ hybrid model, which was further iteratively improved with the best fragments (e.g. loops) identified among the highly-scored single-template models.
Molecular Docking was done for each protein/ligand pair with the aid of VINA algorithm with 128 rounds of flexible docking applied against 16 replicas of protein structures (each representing re-optimized random ensemble of protein sidechain rotamers), and the resulting 2048 complexes were clustered with 2 Å threshold.
Statistical analyses. All analyses were performed using R software (version 3.3.0, www.r-project.org) with outliers and gplots packages. The significance level α of 0.05 was assumed in all statistical tests. The Shapiro-Wilk test was used to assess agreement of ALKBHs content with the Gaussian distribution. Because even after filtering of extreme values with the Grubbs' test for putative outliers 38 the vast majority of the distributions were found to be non-Gaussian, further analyses were based on the non-parametric methods. The statistical significance of the differences observed in protein levels were tested using Wilcoxon tests; the paired (signed-rank) or non-paired versions (rank-sum, Mann-Whitney) were applied accordingly to the hypothesis being tested (i.e. the differences concerning either individuals or whole group analysed, respectively). Correlation matrixes were calculated for the levels of seven proteins using Spearman's rank correlation coefficients. Hierarchical cluster analysis of these matrixes was performed according to the Ward criterion. To estimate the correlation between ALKBHs levels and G/TNM parameters, analysis of variance (ANOVA) and post-hoc pairwise t-student test with Benjamini Hochberg adjustment was used. In that case, samples were divided into groups according to G or TNM parameters. Distribution analysis for each group was also performed. Normal distribution of data sets of examined proteins was not be excluded (Shapiro-Wilk test) 39 . ethics approval and consent to participate. All participants were informed about the purpose of the study and gave their written informed consent. The study was approved by Warsaw Medical University Bioethical Committee permission for working with human tissues/tumours no. KBO/17/11 April 12th 2011. All research was performed in accordance with the relevant guidelines and regulations.

Data Availability
Please contact the corresponding author for all data requests.