Background
A key hallmark of cancer cells is their metabolic reprogramming consisting of enhanced aerobic glycolysis over oxidative respiration. This so called “Warburg effect” leads to the accumulation of methylglyoxal (MG), a highly toxic and reactive dicarbonyl, that spontaneously glycates proteins, nucleic acids and lipids [
1]. The reaction of MG with the amino group of proteins leads to the formation of advanced glycation end-products (AGEs) such as hydroimidazolones (MG-Hs) and argpyrimidines [
2]. To fight MG, mammalian cells developed a detoxifying system consisting of glyoxalases 1 and 2 (GLO1 and GLO2) that transform MG into D-lactate in the presence of reduced glutathione [
3]. The imbalance between MG production and its detoxification results in MG adducts accumulation in cells identified as dicarbonyl stress. The consequences of this cellular stress have been mainly studied in the context of diabetes mellitus and its complications [
4]. Collagen, laminin and mitochondrial proteins are examples of MG targets [
5]. Because of their slow turnover, long-living proteins such as collagens are more prone to being glycated and cross-linked by MG [
6]. Previously, we reported the consistent accumulation of MG protein adducts in patients with breast [
7] and colon [
8] cancers compared to their respective normal counterparts. Initially, MG was mostly considered a toxic molecule for both normal and cancer cells. Recently, using breast and glioblastoma cell lines, our group has demonstrated that low doses of MG promote tumor growth rather than inhibit it. This hormesis effect of MG seemingly reconciles contrasting data in the literature [
9]. In fact, at sub-toxic doses, MG turns out to be beneficial to cancer cells as they acquire resistance to apoptosis and enhanced growth properties.
Furthermore, it remains controversial whether GLO1 in cancer acts as a tumor suppressor or oncogene. Several studies depicted GLO1 as an amplified and/or overexpressed oncogene associated with a poor prognosis in various types of malignant tumors, thus considering the inhibition of GLO1 activity as a potential anti-cancer therapy (for review [
10]). Zender and collaborators [
11] identified and validated GLO1 as a tumor suppressor gene in hepatocellular carcinoma and knockdown of GLO1 using short hairpin RNAs (shRNAs) increased tumor growth in vivo. In line with these data, we have recently reported that breast cancer cells stably depleted in GLO1 and, consequently, bearing a higher level of MG, exhibit significantly enhanced tumor growth and metastatic potential [
12]. We have showed that MG glycates HSP90, which leads to nuclear accumulation of Yes-associated protein (YAP) and the blockade of the Hippo tumor suppressor pathway [
12]. If this mechanism has shed some light on the pro-growth effect of MG, it did not provide any clues to explain how this metabolite might enhance the metastatic phenotype of breast cancer cells. Accordingly, the aim of the present study was to ascertain whether MG could specifically have an impact on essential processes and pathways governing migration and metastasis.
Comprehensive RNA-sequencing (RNASeq) analysis of GLO1-depleted cancer cells identified a metastatic transcriptomic signature involving genes tightly associated with cell migration and extracellular matrix (ECM) remodeling, such as collagens and tenascin C. We have demonstrated that MG-induced pro-metastatic signature is linked with the activation of the mitogen-activated protein kinase kinase (MEK)/extracellular signal-related protein kinase (ERK) pathway, which signals through activated SMAD1. We have further demonstrated that MEK/ERK-sustained activation in GLO1-depleted cells notably occurs through the down-regulation of dual specificity phosphate 5 (DUSP5) phosphatase expression upon MG stress in breast cancer cells. Overall, our study demonstrated for the first time that MG stress is able to change the ECM and to regulate migratory signaling pathways, both in favor of enhanced metastatic dissemination of breast cancer cells.
Methods
Cell culture and reagents
MDA-MB-231 and MCF7 cancer cell lines were obtained from the American Type Culture Collection (ATCC, Manassas, VA, USA). MDA-MB-468 cells were kindly provided by Dr C. Gilles (University of Liège, Belgium). Cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) (Lonza) containing 10% fetal bovine serum (FBS, ThermoFisher Scientific) and 2 mM L-glutamine (Lonza). For cell signaling analysis, cells were cultured in DMEM with L-glutamine and without FBS. L-carnosine (C9625), aminoguanidine (396494) and methylglyoxal (MG, M0252) were from Sigma. We excluded the presence of significant formaldehyde contamination (< 3%) in MG (lot #BCBQ9416V) by nuclear magnetic resonance (NMR) analysis. Transforming growth factor (TGF)β (#10021) was from Peprotech. U0126 (S1102) was from Selleckchem. Anti-argpyrimidine antibody (mAb6B) specificity has been previously confirmed by competitive enzyme-linked immunosorbent assay (ELISA) and shown to not react with other MG-arginine adducts [
13].
MG measurements were performed as previously described [
12]. Briefly, culture medium was collected from GLO1-silenced MDA-MB-231 cells and the corresponding attached cells were counted to normalize MG measurements. Levels of MG were determined by derivatization with O-phenylenediamine (oPD) and analyzed by stable isotope dilution ultra-performance liquid chromatography tandem mass spectrometry (UPLC-MS/MS) as described previously [
14].
All animal experimental procedures were performed according to the Federation of European Laboratory Animal Sciences Associations (FELASA) and were reviewed and approved by the Institutional Animal Care and Ethics Committee of the University of Liège. Animals were housed in the GIGA-accredited animal facility (University of Liège). Eight-week-old female NOD-SCID mice were inoculated via the tail vein with MDA-MB-231 shNT, shGLO1#1 and #2 cells (500000cells/200 μl phosphate-buffered saline (PBS)) and were monitored for 6 weeks for development of metastases. Mice received intra-peritoneal (IP) injection of carnosine (100 mg/kg) three times per week from the day of the engraftment until the end of the experiment (n = 12–14 per condition). Lungs were collected and embedded in paraffin for immunohistochemical analysis (IHC).
Immunohistochemical analysis (IHC)
IHC was performed using a standard protocol as described previously [
12] using the following antibodies: anti-vimentin (1:4, clone V9, Ventana Medical Systems), anti-tenascin C (1:1000, ThermoFisher Scientific) and anti-collagen VI α3 (1:200, Sigma). Positive cells were visualized using a 3,3′-diaminobenzidine substrate and the sections were counterstained with hematoxylin. Tissue sections incubated without primary antibody showed no detectable immunoreactivity. Tissue was stained with hematoxylin eosin and Masson's trichrome using standard protocols.
Evaluation of IHC staining
For vimentin IHC, three different whole immunolabeled tissue sections per lung were digitalized at high resolution (Ventana iScan HT, Ventana Medical Systems). Vimentin-positive metastatic cells were quantified by computerized counts (QuPath 0.1.2 open source software) and verified by manual counting. The number of positive cells to lung area was reported, yielding a count expressed as the number of cells per square millimeter. For tenascin C IHC and Masson’s trichrome staining, a score of 0 was associated with negative staining and a score of 1 with positive staining in tumor cells.
Plasmids, shRNA and small interfering RNA (siRNA) transfection, quantitative reverse transcription-PCR (qRT-PCR), western blot and immunofluorescence
Standard protocols are detailed in Additional file
1: Supplementary methods.
Collagen retraction assay
Collagen gels were made as previously described [
15]. Briefly, DMEM containing 10% fetal calf serum (FCS) was prepared at the appropriate concentration and at pH 8.3 to accommodate the addition of bovine skin collagen (3 mg/ml final concentration) in HAc 0.1% and MDA-MB-231 (300.000 cells/gel). After polymerization of the collagen, the gels were detached from the walls of the dishes by gentle tapping and swirling. Gel diameter was measured at regular intervals and collagen gel areas were calculated.
Migration and invasion assays
MDA-MB-231 and MDA-MB-468 cells (2 × 105 cells) were suspended in serum-free DMEM medium (0.1% bovine serum albumin (BSA), 1% penicillin/streptomycin (pen/strep)) and seeded into the upper part of Transwell filters (pore size 8 μm, Costar) for migration or Transwell filters precoated with Matrigel for invasion (BD Biosciences). The lower compartment was filled with DMEM containing 1% BSA, 1% pen/strep and 10% FBS. After 16 h or 24 h incubation at 37 °C for migration and invasion, respectively, migrating cells were stained with Diff-Quick (Medion Diagnostics). Inserts were scanned at 10× magnification (Eclipse Ti, Nikon). The area covered by migrating/invading cells was quantified by densitometry (ImageJ). Data are expressed as relative migration/invasion ability compared to control cells.
MCF7 (15,000 cells/well) and MCF7-M (60,000 cells/well) cells were seeded in Seahorse XFp mini-plates (Agilent) and analyzed using the Seahorse glycolysis stress test according to manufacturer’s recommendations. Cells were first challenged with glucose (10 mM) then successively stressed with oligomycin (1 μM) and 2-deoxyglucose (50 mM). All results were normalized to protein quantification.
Glucose uptake
Cells were incubated in glucose-free medium in the presence of 2-NBDG (ThermoFisher Scientific) for 30 min and further analyzed by flow cytometry (BD Biosciences). Median fluorescence intensity (MFI) was calculated.
GLO1 activity assay
GLO1 activity was assessed as previously described [
7]. Briefly, proteins were extracted with radioimmunoprecipitation (RIPA) buffer, quantified and mixed with a pre-incubated (15 min at 25 °C) equimolar (1 mM) mixture of MG and GSH (Sigma) in 50 mM sodium phosphate buffer, pH 6.8. S-D-lactoylglutathione formation was followed spectrophotometrically by the increase in absorbance at 240 nm at 25 °C. GLO1 activity data are expressed as arbitrary units (A.U.) of enzyme per milligram of proteins.
Statistical analysis
Data from two groups were compared using the unpaired Student’s t test with or without Welsch’s correction according to homoscedasticity. Experimental data from more than two groups were compared using one-way or two-way analysis of variance (ANOVA) depending on the number of grouping factors. Dunnett’s test was applied for simple comparisons while Student-Newman-Keul’s (one-way ANOVA) or Bonferroni’s (two-way ANOVA) tests were used for multiple comparisons. In the case of discrete variables (IHC scores) or non-normally distributed variables, groups were compared using Mann-Whitney’s U test. Outliers were detected using whisker box plots. A bilateral p value <0.05 was considered statistically significant with a 95% confidence interval.
Discussion
MG dicarbonyl stress has recently been associated with the aggressiveness of various malignancies including colon and breast carcinoma. However, the molecular mechanisms underlying these effects have not been clarified so far, particularly regarding MG-driven metastatic phenotype acquisition. Thanks to a comprehensive transcriptomic approach we identified a pro-metastatic signature associated with the MG stress condition, which notably comprised genes associated with ECM remodeling, and migration, two traits that are essential for metastasis.
Stromal cells are generally considered the main contributors to tumor ECM deposition and remodeling that favor tumor growth and invasion. In this study, we showed for the first time that cancer cells under MG stress directly contribute to major ECM changes. Differentially expressed genes in GLO1-depleted cells include several ECM components such as collagens, tenascin C and lumican. On the one hand, this finding corroborates recent studies implicating tumor cells in the expression of ECM components (i.e., collagens and fibronectin) that fulfill pro-tumoral and pro-metastatic functions at both primary and metastatic sites. On the other hand, an interesting parallel can be drawn with diabetes mellitus, where hyperglycemia results in the accumulation of MG and its protein adducts. In fact, GLO1-depleted myoblasts grown under a hyperglycemic condition display MG-induced elevated expression of collagens and increased fibrosis [
28]. Elevated amounts of collagens secreted by MG-stressed tumor cells could be cross-linked by MG and thus generate specific mechanotransduction signals on tumor cells to favor their acquisition of a pro-invasive and pro-metastatic phenotype. In good accordance with this hypothesis, GLO1-depleted cancer cells show augmented ECM reorganization, anchorage-independent growth and increased migration/invasion ability; all inhibited in the presence of MG scavengers. Next to collagens, tenascin C is another remarkable cancer cell-derived ECM component overexpressed upon MG stress. Tenascin C induces epithelial-to-mesenchymal transition (EMT) changes in breast cancer cells [
29] and promotes their survival and outgrowth at secondary organs, such as the lung [
30]. An example consistent with the acquisition of an invasive phenotype upon MG stress is the up-regulation of CD24, a mucin-like adhesion molecule that enhances the metastatic potential of malignant cells and is shown to be a marker of poor prognosis in breast carcinomas [
31]. It is noteworthy that MG-stress-driven fine-tuning of the ECM consisted of the up-regulation of pro-metastatic genes but also the down-regulation of specific components such as lumican the reduced expression of which has been associated with poor prognosis in patients with breast cancer [
32]. Further studies will help us better understand the role of MG in pro-tumoral ECM remodeling through a direct effect on cancer cells but also on surrounding cells such as fibroblasts and endothelial cells.
Several pro-tumorigenic signaling pathways that are activated by ECM stiffness promote aerobic glycolysis [
33]. Emerging evidence indicates that metabolic alterations and an abnormal ECM cooperatively drive cancer cell aggressiveness and treatment resistance [
20,
33]. In this study, we connected MEK/ERK/SMAD1 signaling activation and ECM remodeling with the major and underestimated consequence of the glycolytic switch - MG stress. We propose that MG stress could be sustained through a positive regulatory loop induced by ECM quantitative and qualitative changes. In this study, MCF7-M cells that had undergone EMT transition and glycolytic switch had accumulation of MG adducts, GLO1 and NRF2 overexpression and a higher migratory capacity than parental cells, this latter being reversed using MG scavengers.
SMAD4 is a common transcriptional partner for activated SMADs, the loss of which predicts liver metastasis in colon cancer [
34] and bad prognosis in patients with breast cancer [
35]. We showed that upon dicarbonyl stress, breast cancer cells had a significant decrease in SMAD4 and had enhanced migratory capacity. Our previous data [
12] and this study suggest that MG stress mimics TGFβ control in ECM remodeling and the EMT process. Consistent with the hypothesis of a similarity between the pro-tumor effects of TGFβ and MG stress is the loss of SMAD4, which has been shown to abolish TGFβ tumor-suppressive functions while maintaining its role as an EMT inducer [
36]. Massagué’s group originally reported that BMP and ERK signaling inactivate SMAD1 through the phosphorylation of specific serine residues [
25]. This mechanism has notably been invoked to explain how oncogenic RAS could override TGFβ tumor suppressive effects [
37]. In other studies, RAS-ERK signals enhanced the transcriptional activity of SMAD1 in response to BMPs [
38]. Mechanistically, we provided evidence that MG stress induces sustained activation of the MEK/ERK pathway that signals through SMAD1 independently of KRAS mutation status. Accordingly, either the inhibition of MEK or SMAD1 impacts on the expression of MG stress pro-metastatic signature and decreases the migratory potential of breast cancer cells. We further pointed out for the first time that MG significantly impacts on MAPK-mediated response to stress through the inhibition of DUSP catalytic activity in breast cancer cells.
The role of GLO1 in cancer progression remains controversial. In this study, GLO1 silencing favored a pro-migratory and metastatic phenotype, suggesting a tumor-suppressing role of GLO1. A recent study published by Guo and colleagues reported that inhibition of GLO1 promoted apoptosis and decreased invasion of MDA-MB-231 cells [
39]. These seemingly contradictory results both point to the importance of determining the MG concentrations achieved when using the GLO1 silencing strategy. Indeed, we recently demonstrated that MG exerts an hormetic effect on cancer cells that is defined by low-dose stimulation and high-dose inhibition of tumor growth [
9].
The significant anti-metastatic effect of carnosine designates MG stress as a promising target for therapeutic intervention in aggressive tumor cells that have undergone energy metabolic switch. To date, inhibitors of glycolysis have had only modest therapeutic efficacy in cancer. These unsatisfactory results may be related to the large heterogeneity reported for several glycolytic enzymes, as a consequence of different genes, splice forms and post- translational modifications. These latter make cancer cells react differently when challenged with anti-glycolytic agents [
40]. Based on our results, we postulate that the use of MG scavengers to prevent the formation of protein and DNA adducts might be a more adaptive strategy to target highly glycolytic tumors. Importantly, carnosine reversed the activation of MEK/ERK MAPK signalization, thus further highlighting the therapeutic potential of MG scavengers across different malignancies. Despite the recent recognition of MG as an oncometabolite [
41], contemplating the targeting of dicarbonyl stress as a promising strategy for cancer prevention and/or therapy is still in its early days and merits further studies.
Acknowledgements
The authors are thankful to S. Nüchtern, N. Maloujahmoum, F. Agirman and V. Hennequière for expert technical assistance. The authors are particularly grateful to S. Dubois (University of Liège) for her valuable technical support in the tail vein injection model. We are also grateful to Dr A. Chariot and Dr P. Close (University of Liège) for kindly making available MCF7-M cells. We thank Dr Koji Uchida (University of Tokyo) for kindly providing anti-argpyrimidine antibody. We acknowledge the scientific and technical support provided at the following technology platforms of the GIGA-R institute (University of Liège): animal, imaging and flow cytometry, genomics, viral vectors and immunohistology.