Iron chelators target both proliferating and quiescent cancer cells

Poorly vascularized areas of solid tumors contain quiescent cell populations that are resistant to cell cycle-active cancer drugs. The compound VLX600 was recently identified to target quiescent tumor cells and to inhibit mitochondrial respiration. We here performed gene expression analysis in order to characterize the cellular response to VLX600. The compound-specific signature of VLX600 revealed a striking similarity to signatures generated by compounds known to chelate iron. Validation experiments including addition of ferrous and ferric iron in excess, EXAFS measurements, and structure activity relationship analyses showed that VLX600 chelates iron and supported the hypothesis that the biological effects of this compound is due to iron chelation. Compounds that chelate iron possess anti-cancer activity, an effect largely attributed to inhibition of ribonucleotide reductase in proliferating cells. Here we show that iron chelators decrease mitochondrial energy production, an effect poorly tolerated by metabolically stressed tumor cells. These pleiotropic features make iron chelators an attractive option for the treatment of solid tumors containing heterogeneous populations of proliferating and quiescent cells.

Scientific RepoRts | 6:38343 | DOI: 10.1038/srep38343 molecular mechanism of action of VLX600. We here report that VLX600 binds iron and that this property is the underlying mechanism of the ability of VLX600 to reduce cell proliferation and to decrease mitochondrial OXPHOS. We show that also other iron chelators have the ability to affect the viability of MCTS, albeit with lower potency than VLX600. The ability of iron chelators to reduce mitochondrial energy production adds to the evidence of this class of compounds as having attractive anti-neoplastic activities.

Results
VLX600 is an iron chelator. The molecular structure of VLX600 is shown in Fig. 1A. The precise molecular mechanism of action of VLX600 was unknown and we therefore performed a Connectivity Map-based mechanistic exploration by examining the gene expression profile of drug-treated tumor cells 14 . We employed two different cellular models; the breast cancer cell line MCF-7 and colon carcinoma cell line HCT116, grown as 2-D monolayer and 3-D MCTS, respectively. MCF-7 cells were chosen since it is the most frequently used cell model in the Connectivity Map database. We selected MCTS HCT116 to investigate if the response is the similar when cells were grown in 3-D cell culture. The gene expression signature induced by VLX600 was most similar to that of ciclopirox olamine (CPX; 6-cyclohexyl-1-hydroxy-4-methyl-2(1H)-pyridone 2-aminoethanol), the ChemBridge compound 5109870 (2-hydroxy-3-methoxybenzaldehyde 2-pyridinylhydrazone), and deferoxamine (Fig. 1B). All of these compounds were previously described as iron chelators [15][16][17] , suggesting that the anticancer activity of VLX600 could be attributed to iron chelation and sequestering. Complex formation between VLX600 and different metals was examined using spectrophotometry (Fig. 1C). VLX600 was indeed found to form complexes with ferrous and ferric iron as well as with Co 2+ , whereas only minor changes in the UV/Vis spectra were observed with Cu 2+ , Zn 2+ , Ni 2+ or Al 3+ (Fig. 1C). Addition of extracellular iron (Fe 2+ and Fe 3+ ) abolished the cytotoxic activity of the compound (Fig. 1D), supporting the hypothesis of iron chelation being instrumental to biological activity of VLX600. As previously described, VLX600 is able to reduce mitochondrial OXPHOS, particularly uncoupled respiration 10 . We found that addition of iron chloride restored OXPHOS capacity in cells exposed to VLX600 (Fig. 1E).
Characterization of iron chelation. By means of density functional theory (DFT) calculations on Fe(II) (VLX600) 2 , the N2 nitrogen in the 1,2,4-triazine moiety was found to preferentially coordinate iron over the N4  HCT116 cells (multicellular spheroid culture) exposed to VLX600 were uploaded to the CMAP data base to identify other compounds with similar mechanism of action. (C) Analysis of metal binding by VLX600 using spectrophotometry as described 16 . Note the reduction in A 340 after addition of Fe 2+ , Fe 3+ and Co 2+ , whereas Cu 2+ and other metal ions do not affect A 340 . Representative of three independent experiments (D) Antiproliferative activity of VLX600 on HCT116 cells is abrogated by addition of iron chloride (FeCl 2 and FeCl 3 ). Cells were grown for 72 h in the presence or absence of VLX600 and iron chloride and viability was assessed by MTT assay. Mean ± S.D. (n = 4), representative repeated experiments. (E) The reduction of oxygen consumption by VLX600 in HCT116 cells is reversed by the addition of iron. Mean ± S.D. (n = 4), representative of two independent experiments. nitrogen by as much as 18 kcal/mol ( Fig. 2A). Using this coordination mode, the geometries were calculated for both oxidation states and spin states of Fe(VLX600) 2 (Supplementary Table 1). Extended X-Ray Absorption Fine Structure (EXAFS) confirmed that iron(II) binds two VLX600 ligands to form a bis-complex where iron binds six nitrogen atoms in a pseudo-octahedral configuration at very short Fe-N bond distances, 1.90 and 1.915 Å for iron(II) and iron(III), respectively. This is in good agreement with reported iron(II)-terpyridine complexes, Supplementary Table 3 and the DFT calculated distances. Furthermore, Fe-C/N distances at ~2.80, 3.35 and 4.05 Å were identified and fit with distances observed in the iron(II)-terpyridine complexes and those in the calculated model presented in Fig. 2A. The very short Fe-N bond distances are consistent with both iron(II) and iron(III) being present as low-spin complexes, and with significant back-bonding for iron(II) making the bond distances even shorter and the complex being significantly stabilized. The refined structure parameters are given in Supplementary Table 2, and the model fits of the experimental EXAFS and corresponding Fourier transforms ( Supplementary Figures 1 and 2, respectively). The positions of the absorption edges clearly show that the VLX600 complexes of both iron(II) and iron(III) in principle remain in their oxidation states after 48 hours of storage, see Figure S3. The iron(III) complex seems to be completely stable, while there is maybe a sign that the iron(II) complex is partly oxidized as seen in the slightly longer mean Fe-N bond distance (Supplementary Table 2), and the somewhat different shape of the absorption edge, see Figure S3. The pre-edge at 7111-7114 eV is a strong indicator of the oxidation state of iron 18 . However, in these cases the pre-peak is observed at 7112 eV for both complexes, likely due to that they are low-spin complexes and with significant back-bonding in the iron(II) VLX600 complex, Figure S3.
The chemical structure of VLX600 was modified by altering the position of the nitrogen in the pyridine moiety (VLX641 and VLX642; Fig. 2B). According to the model describing the coordination mode ( Fig. 2A), this modification would result in loss of tridentate iron binding and, thus, biological activity of the molecules. Indeed, the VLX600 analogues showed no cytotoxic activity (Fig. 2C), suggesting limited off-target effects of the molecular scaffold in this biological system. Furthermore, the chemical alterations abolished the effect on mitochondrial respiration (Fig. 2D).  Inhibition of tumor cell proliferation by VLX600. VLX600 was identified in a screen for compounds active on 3-D tumor spheroids but also shows antiproliferative activity on colon cancer cell lines in monolayer culture 10 . We examined the effect of VLX600 on a number of colon cancer cell lines and generally found IC 50    (Fig. 3H,I). The catalytic activity of ribonucleotide reductase is dependent on an iron-binding site in the M2 subunit of the enzyme and iron chelators have been found to inhibit this enzyme 20 . As expected, VLX600 inhibited ribonucleotide reductase activity in vitro (Fig. 3J).

Inhibition of mitochondrial activity and spheroid viability by iron chelators. We examined
whether also other iron chelators have the ability to decrease mitochondrial oxygen consumption. Indeed, CPX, Triapine and deferoxamine all decreased uncoupled OXPHOS (Fig. 4A). CPX and Triapine reduced HCT116 MCTS viability (i.e. GFP signal; previously validated to be a reliable surrogate marker of viability 11 ), although with considerably lower potency compared with VLX600 (Fig. 4B). MCTS formed from HT-29 were only affected by VLX600 and higher concentrations of Triapine whereas deferoxamine did not show detectable activity even at concentrations > 300 μ M. Compound lipophilicity is particularly important for activity in MCTS systems 21 as well as in solid tumors in vivo 22 . The higher 3-D cytotoxic potency of VLX600 compared to other iron chelators examined here may at least be partially explained by the higher lipophilicity of VLX600 (XlogP ~ 3) (Supplementary Table 4).

Decreased hypoxia in spheroids and in tumors.
The level of oxygen consumption in spheroids is expected to be reflected in the degree of hypoxia 10 . Exposure of spheroids to a mitochondrial uncoupler does indeed increase the pimonidazole hypoxic fraction (pHF) whereas VLX600 decreases the pHF (Fig. 5A). Similar to VLX600, ciclopirox reduced the pHF, albeit to lower degree than VLX600 (Fig. 5A), whereas deferoxamine did not cause any observable effect (Fig. 5A). These alterations of pHF are consistent with the observed degree of changes in MCTS viability (Fig. 4B).  VLX600 is well tolerated by rodents and shows anti-tumor activity in animal tumor models which are associated with decreased proliferative indices 10 . To address the question whether VLX600 affects tumor hypoxia in vivo, we injected VLX600 into mice bearing HCT116 colon cancer xenografts. VLX600 exposure significantly reduced the pHF 4 hours after injection of the drug (Fig. 5B,C). [10][11][12][13] . OXPHOS is known to be functional at oxygen concentrations as low as 0.5% 23 but it is not known whether electron transport components are expressed in core regions of spheroids. Cytochrome oxidase (complex IV) is rate-limiting in the electron transport chain 24 and its activity was reported to decrease under hypoxic/hypoglycemic conditions 25 . We examined the distribution of cytochrome oxidase IV (COXIV) immunoreactivity and cytochrome oxidase in situ activity in colon cancer spheroids (Fig. 5D). Immunoreactivity and enzymatic reactivity were strongest in the area of 50-100 μ m from the spheroid surface containing Ki67-positive, proliferating cells (Fig. 5D). However, COXIV activity was clearly discerned in the core regions of the spheroids (Fig. 5D,E). Exposure to VLX600 for 24 hours led to a uniform reduction of cytochrome oxidase activity in MCTS and also to a decreased immunoreactivity for COXIV (Fig. 5D).

Discussion
The cells in the cores of MCTS have the ability to regrow after cytotoxic therapy 26 and have in this respect similarities to cancer initiating stem cells. Recently, several studies have identified inhibition of mitochondrial OXPHOS as a promising strategy to combat quiescent cancer cells in hypoxic and nutritionally compromised environments 10,11,[27][28][29] . There is a growing body of observational evidence to support this idea. Importantly, and contrary to the Warburg hypothesis, mitochondria are the main source of ATP production in cancer cells 30 and OXPHOS is the predominant source of ATP in also hypoxic layers of MTCS 31 . Furthermore, increased OXPHOS activity is observed in human tumors in situ compared with adjacent normal tissue 32 . Components involved in mitochondrial biogenesis, mitochondrial translation and mitochondrial lipid biosynthesis are transcriptionally upregulated in human breast cancer epithelial cells and downregulated in adjacent stromal cells 33 . Elevated expression of the mitochondrial markers such as TIMM17A and TOMM34 is associated with poor clinical outcome and may be predictive of higher tumor grade and metastasis [34][35][36] .
Mitochondrial inhibitors have long been reported to possess anti-tumor activity [37][38][39] . Examples of such compounds are Rho123 37 , which inhibits OXPHOS 40 , and mitochondria-targeted lipophilic cations 41 . Metformin is a commonly prescribed anti-diabetic drug that increases cellular glucose uptake 42 . Metformin inhibits mitochondrial OXPHOS, likely through inhibition of complex I activity 43,44 and metformin exposure appears to diminish tumor formation in diabetic patients 45 and in mouse animal models 46,47 . VLX600 is a mild inhibitor of OXPHOS, primarily inhibiting uncoupled respiration (i.e total respiratory capacity) 10 . It is interesting to note that the limited effect on OXPHOS elicited by VLX600 on monolayer cultures is associated with major effects on hypoxia in spheroids. A possible explanation for this phenomenon is that total available, albeit limited, mitochondria capacity is utilized in the cells of the core area, and that mild inhibition has large consequences in oxygen utilization and energy production.
Iron chelators have complex effects on tumor cells 1,5-9 . Inhibition of ribonucleotide reductase results in inhibition of cell proliferation by depleting deoxyribonucleotide pools and we here show that iron chelators affect OXPHOS. We previously reported that VLX600 induces HIF-1α expression 10 , an observation explained by the present finding that the drug chelates iron. Prolyl hydroxylases, that negatively regulates HIF-1α protein stability, are iron-dependent enzymes 48 and iron chelators are known to induce HIF-1α 49 VLX600 induces a HIF-1α -dependent shift to glycolysis in exposed cells 10 , and it is possible that this leads to further depletion of glucose in tumors and a strengthened bioenergetic catastrophe. Some iron chelators form redox-active complexes with metals, notably cupper, and this has been reported to be essential for their ability to induce apoptosis 50 . Hovever, exposure to VLX600 does not result in the generation of increased levels of reactive oxygen species in cells 10 , and the antiproliferative effects of this drug is therefore most likely due to depletion of cellular labile iron pools.
The ability of iron chelators to inhibit the proliferation and decrease the viability of heterogeneous populations of tumor cells in solid tumors opens interesting possibilities for future therapies. VLX600 is presently in a phase-I clinical trial (ClinicalTrials.gov NCT02222363) for patients with refractory advanced solid tumors which will hopefully provide proof of concept for this treatment strategy.
Scientific RepoRts | 6:38343 | DOI: 10.1038/srep38343 Compounds. VLX600, VLX641 and VLX642 were were acquired from OnTarget Chemistry (Uppsala, Sweden). Deferoxamine, ciclopirox and triapine were from Sigma-Aldrich (St. Louis, MO, USA). For iron chelation experiments, Cu 2+ , Co 2+ , Al 3+ , Ni 2+ , Zn 2+ (nitrate salts) and FeCl 2 and FeCl 3 were obtained from Sigma Aldrich (St Louis, MO, USA). Connectivity Map. The Connectivity Map (CMAP) (www.broad.mit.edu/cmap) build 02 contains genome-wide expression data for 1300 compounds (6100 instances, including replicates, different doses and cell lines). We followed the original protocol using MCF-7 breast cancer cells as described by Lamb et al. 14 . Briefly, cells were plated in 6-well plates at a density of 0.4 × 10 6 cells per well and were left to attach for 24 h prior to addition of drugs. We also performed CMAP-based mechanistic exploration in MCTS HCT116. MCTS HCT116 were grown as previously described 10 and were exposed to VLX600 (10 μ M) or to vehicle control (DMSO). After 6 hours treatment, the cells were washed with PBS and total RNA was prepared using RNeasy miniprep kit (Qiagen, Chatsworth, CA). Starting from two micrograms of total RNA, gene expression analysis was performed using Genome U133 Plus 2.0 Arrays according to the GeneChip Expression Analysis Technical Manual (Rev. 5, Affymetrix Inc., Santa Clara, CA). Raw data was normalized with MAS5 (Affymetrix) and gene expression ratios for drug treated vs. vehicle control cells were calculated to generate lists of regulated genes. Filter criteria were present calls for all genes in both the VLX600 treated DMSO (vehicle) exposed control cells and an expression cut-off of at least 200 arbitrary expression units. Only probes present on HG U133A were used, for CMAP compatibility. The 30 most up and down regulated genes (i.e., probes) for each compound were uploaded into CMAP and compared to the 6100 instances in the CMAP database, to retrieve a ranked compound list. Raw and normalized expression data have been deposited at Gene Expression Omnibus with accession number GSE84051 (MCF-7) and GSE53777 (MCTS HCT116).
Cytotoxicity assay. The Fluorometric Microculture Cytotoxicity Assay, FMCA, described in detail previously 51 , was used for measurement of the cytotoxic effect of library compounds and the established standard drugs. The FMCA is based on measurement of fluorescence generated from hydrolysis of fluorescein diacetate (FDA) to fluorescein by cells with intact plasma membranes. Cells were seeded in the drug-prepared 384-well plates using the pipetting robot Precision 2000 (Bio-Tek Instruments Inc., Winooski, VT). In each plate, two columns without drugs served as controls and one column with medium only served as blank. Cell viability was also monitored using the MTT assay 52 (in Fig. 1D).
Ribonucleotide reductase assay. Ten million (1 × 10 7 ) HCT116 cells grown in 150 mm plates were exposed to indicated drugs for 24 hours. The ribonucleotide assay was performed as previously described 19 . Immunohistochemistry. Multicellular spheroids produced by the hanging drop method in 96 well plates were fixed in 2% buffered formalin, dehydrated, embedded in paraffin and sectioned. Each sample contained 24 spheroids (spheroids from each 96 well plate were pooled into 4 groups). The sections were deparaffinized with xylene, rehydrated and microwaved and then incubated overnight with the monoclonal primary antibodies diluted in 1% (wt/vol) bovine serum albumin and visualized by standard avidin-biotin-peroxidase complex technique (Vector Laboratories, Burlingame, CA, USA). Counterstaining was performed with Mayer's haematoxylin. Antibody MIB-1 (against the nuclear proliferation-associated antigen Ki67) was from Immunotech SA, Marseille, France and used at a dilution 1:400.
LogP predictions. XlogP3 values were retrieved from Pub Chem https://pubchem.ncbi.nlm.nih.gov/ compound/. Assessment of cellular DNA synthesis. The fluorescence microscope ArrayScan V HCS system (Cellomics Inc.) was used for measurement of 5-ethynyl-2′-deoxyuridine (EdU) incorporation. HCT116 cells were seeded into 96-well plates (PerkinElmer Inc., Wellesley, MA, USA), left to attach over night, before test compounds were added. Cells were treated with VLX600 for 24 h or vehicle control. Cells were stained using Click-iT EdU HCS assay (C10354, Invitrogen, Molecular Probes Inc., OR, USA) according to the manufacturer's instructions. Processed plates were loaded in the ArrayScan and analyzed. Images were acquired for each fluorescence channel, using suitable filters with 10x objective and in each well at least 1,000 cells were analyzed. Average total intensity in the EdU channel was measured.
Scientific RepoRts | 6:38343 | DOI: 10.1038/srep38343 Measurement of oxygen consumption rates (OCR). Measurements of cellular oxygen consumption were performed using a Seahorse XF analyzer as recommended by the manufacturer (Agilent Technologies, Santa Clara, CA). Briefly, cells were pleated in 100 μ L culture medium in XF 24-well cell plates with indicated blank control wells. Plates were first placed at room temperate for 1 hour and then moved to an incubator for another 1-5 hours to facilitate attachment of cells. Another 150 μ l of culture medium was then added to each well followed by overnight incubation. Before the OCR measurements, medium was removed from each well and replaced with 500 μ L Seahorse assay media (pyruvate 1 mM, glucose 25 mM) at 37 °C without CO 2 for 1 h. OCR were measured using an XF24 Extracellular Flux Analyzer. A cell mitochondrial stress kit from Seahorse Bioscience was used for mitochondrial stress tests and experiments were performed as described previously 10 .
Assessment of pimonidazole adducts accumulation in mouse xenografts. Once HCT116 tumors in SCID mice had grown to a size of 200 mm 3 , mice were injected with VLX600. Pimonidazole was injected 1 hour before the animals were sacrificed. Tumors were sectioned and stained for pimonidazole adducts. Quantification was performed double blind from photographs of the stained slides.
Computational Details. All density functional calculations were done using Jaguar 8.5 53 (Schrodinger, Inc., New York, NY, 2014). Structures were optimized using the LACVP* basis set 54 and the M06 functional 55 . Single-point energy calculations were performed using the LACVP** basis set with a PBF solvation model 56 for water using geometries from M06/LACVP*. C2 symmetry was used in all complexes where applicable.
Statistics. IC 50 -values and EC 50 -values (inhibitory/effective concentration 50%) were calculated by using non-linear regression and a standard sigmoidal dose-response model in the GraphPad Prism program (GraphPad Software, Inc. San Diego, CA, USA). Statistical significance was assessed using Student's t-test (Microsoft Excel).