Research ArticleMetabolism

The induction of HAD-like phosphatases by multiple signaling pathways confers resistance to the metabolic inhibitor 2-deoxyglucose

See allHide authors and affiliations

Science Signaling  03 Sep 2019:
Vol. 12, Issue 597, eaaw8000
DOI: 10.1126/scisignal.aaw8000

Resisting a metabolic poison

Once imported into cells and phosphorylated, the glucose analog 2-deoxyglucose (2DG) inhibits glycolysis, leading to the proposal of using 2DG as a cancer treatment. Using yeast as a model, Defenouillère et al. investigated how cells become resistant to 2DG. Exposure to 2DG activated several signaling pathways that resulted in the increased expression of the gene encoding the phosphatase Dog2. In contrast, glucose availability transcriptionally repressed DOG2 expression. When overexpressed, a human homolog of Dog2 conferred 2DG resistance to human cells, suggesting that cancer cells with increased abundance of this phosphatase could escape the toxic effects of 2DG.

Abstract

Anti-cancer strategies that target the glycolytic metabolism of tumors have been proposed. The glucose analog 2-deoxyglucose (2DG) is imported into cells and, after phosphorylation, becomes 2DG-6-phosphate, a toxic by-product that inhibits glycolysis. Using yeast as a model, we performed an unbiased mass spectrometry–based approach to probe the cellular effects of 2DG on the proteome and study resistance mechanisms to 2DG. We found that two phosphatases that target 2DG-6-phosphate were induced upon exposure to 2DG and participated in 2DG detoxification. Dog1 and Dog2 are HAD (haloacid dehalogenase)–like phosphatases, which are evolutionarily conserved. 2DG induced Dog2 by activating several signaling pathways, such as the stress response pathway mediated by the p38 MAPK ortholog Hog1, the unfolded protein response (UPR) triggered by 2DG-induced ER stress, and the cell wall integrity (CWI) pathway mediated by the MAPK Slt2. Loss of the UPR or CWI pathways led to 2DG hypersensitivity. In contrast, mutants impaired in the glucose-mediated repression of genes were 2DG resistant because glucose availability transcriptionally repressed DOG2 by inhibiting signaling mediated by the AMPK ortholog Snf1. The characterization and genome resequencing of spontaneous 2DG-resistant mutants revealed that DOG2 overexpression was a common strategy underlying 2DG resistance. The human Dog2 homolog HDHD1 displayed phosphatase activity toward 2DG-6-phosphate in vitro and its overexpression conferred 2DG resistance in HeLa cells, suggesting that this 2DG phosphatase could interfere with 2DG-based chemotherapies. These results show that HAD-like phosphatases are evolutionarily conserved regulators of 2DG resistance.

INTRODUCTION

Most cancer cells display an altered metabolism, with an increased glucose consumption to support their proliferative metabolism that is based on aerobic glycolysis (Warburg effect) (1, 2). Inhibiting glycolysis has been proposed as a strategy to target cancer cells, and various metabolic inhibitors have been considered (3, 4).

2-Deoxy-d-glucose (2DG) is a derivative of d-glucose that is imported by glucose transporters and is phosphorylated by hexokinase into 2DG-6-phosphate (2DG6P), but cannot be further metabolized owing to the 2-deoxy substitution, triggering a decrease in cellular adenosine 5′-triphosphate (ATP) content in tumors (5). Mechanistically, 2DG6P accumulation hampers glycolysis by inhibiting hexokinase activity in a noncompetitive manner (6, 7) and by inhibiting phospho-glucose isomerase activity in a competitive manner (8). Because cancer cells rely on an increased glycolysis rate for proliferation, 2DG has been of interest for cancer therapy, particularly in combination with radiotherapy or other metabolic inhibitors (911). These features led to a phase 1 clinical trial using 2DG in combination with other drugs to treat solid tumors (12). Its derivative, 18fluoro-2DG, is also used in cancer imaging (positron emission tomography scans) because it preferentially accumulates in tumor cells due to their increased glucose uptake (13). In addition, because of its structural similarity to mannose, 2DG (which could also be referred to as to 2-deoxymannose, because mannose is the C2 epimer of glucose) also interferes with N-linked glycosylation and causes endoplasmic reticulum (ER) stress (1416), which has been proposed to be the main mechanism by which 2DG kills normoxic cells (17). 2DG toxicity has also been linked to the depletion of phosphate pools after 2DG phosphorylation (18). Last, interference of 2DG with lipid metabolism and calcium homeostasis through unknown underlying mechanisms has been previously described (19). Resistance to 2DG has been detected in cell cultures (20).

Because these metabolic and signaling pathways are evolutionarily conserved, simpler eukaryotic models such as the budding yeast Saccharomyces cerevisiae can be used to understand the mode of action of 2DG. Moreover, yeast is particularly well suited for these studies because of its peculiar metabolism (21). Akin to cancer cells, S. cerevisiae preferentially consumes glucose through glycolysis over respiration, regardless of the presence of oxygen. This occurs through glucose-mediated repression of genes involved in respiration and alternative carbon metabolism, which operates at the transcriptional level. This glucose-mediated repression mechanism is relieved upon activation of the yeast ortholog of adenosine 5′-monophosphate–activated protein kinase (AMPK) Snf1, which phosphorylates the transcriptional repressor Mig1 and leads to its translocation out of the nucleus (2225). Mutations that render yeast cells more tolerant to 2DG have been identified (2631). These findings suggest the existence of cellular mechanisms that can modulate 2DG toxicity, which are important to characterize if 2DG is to be used therapeutically.

In yeast, 2DG was initially used to identify genes involved in glucose repression because 2DG, like glucose, causes Snf1 inactivation and thus prevents the use of alternative carbon sources (3234). The characterization of mutants that can grow in 2DG-containing sucrose medium has allowed the identification of components of the glucose repression pathway (26, 29, 35) and has revealed that mutations in HXK2, which encodes hexokinase II, also render yeast cells more tolerant to 2DG, perhaps by limiting 2DG phosphorylation and 2DG6P accumulation (2931, 36, 37). Last, several 2DG-resistant mutants display increased 2DG6P phosphatase activity, which could detoxify the cells of this metabolite and dampen its negative effects on cellular physiology (38, 39). Two 2DG6P phosphatases named DOG1 and DOG2 have been cloned, and their overexpression triggers 2DG resistance and prevents 2DG-mediated repression of genes (33, 40, 41).

The toxicity of 2DG has also been studied in the context of cells grown in glucose-containing media, which may be more relevant for the understanding its mode of action in mammalian cells. Under these conditions, 2DG toxicity is independent of its effect on the glucose repression of genes and involves different mechanisms, such as a direct inhibition of glycolysis and other cellular pathways (30, 42, 43). Accordingly, several mutations leading to 2DG resistance in cells grown in sucrose medium do not lead to resistance in cells grown in glucose medium (30, 44). Deletion of REG1, which encodes a regulatory subunit of protein phosphatase 1 (PP1) that inhibits Snf1 (45, 46), leads to 2DG resistance (30). The resistance of the reg1∆ mutant depends on the presence of Snf1, and the single deletion of SNF1 also renders yeast hypersensitive to 2DG (31), thus demonstrating that Snf1 activity is crucial for 2DG resistance. A model has been proposed in which the 2DG sensitivity displayed by the snf1∆ mutant involves misregulated expression and localization of the low-affinity glucose transporters Hxt1 and Hxt3 (47). In addition, the deletion of LSM6, which encodes a component of a complex involved in mRNA degradation, also leads to 2DG resistance in an Snf1-dependent manner, but the mechanism by which this occurs is unknown (31). Thus, many aspects of the pathways mediating 2DG sensitivity or resistance remain to be explored.

In the present study, our unbiased, mass spectrometry (MS)–based approach in yeast revealed that the main 2DG6P phosphatase, Dog2, was induced upon exposure to 2DG and participated in 2DG detoxification in glucose medium. We found that 2DG induced Dog2 (and Dog1, to a certain extent), by triggering unfolded protein response (UPR) and mitogen-activated protein kinase (MAPK)–based stress-responsive signaling pathways. Moreover, the expression of DOG2 was additionally regulated by Snf1 and the glucose-repression pathway through the action of downstream transcriptional repressors and contributed to the resistance of glucose-repression mutants to 2DG. The partial characterization of 24 spontaneous 2DG-resistant mutants revealed that most mutants displayed increased DOG2 expression, suggesting a common strategy used to acquire 2DG resistance. Particularly, genome resequencing identified that mutations in CYC8, which encodes a transcriptional corepressor, caused 2DG resistance through increased Dog2 expression. The identification of a potential human homolog of the Dog1/2 proteins, HDHD1 [haloacid dehalogenase (HAD)–like hydrolase domain containing 1, also named PUDP (pseudouridine-5′-phosphatase)], which displays 2DG6P phosphatase activity in vitro and which caused 2DG resistance in HeLa cells upon overexpression, suggests that HAD-like phosphatases are conserved regulators of 2DG resistance.

RESULTS

A proteomic assessment of the cellular response to 2DG reveals increased expression of metabolic enzymes

As a first step to characterize the cellular response of yeast cells to 2DG treatment in an unbiased and quantitative manner, we performed proteome-wide MS-based proteomics on wild-type (WT) cells treated with 0.2% 2DG, a concentration that prevents growth of WT cells on plates (30). Untreated cells were used as a negative control. Overall, 78 proteins were significantly more abundant, whereas 18 proteins were less abundant after 2DG treatment [false discovery rate (FDR) of 0.01] (Fig. 1A and data file S1). Among the up-regulated candidates, proteins involved in various metabolic processes were significantly enriched, including those involved in the metabolism of glucose, glucose-6-phosphate, and other carbohydrates (Fig. 1B and fig. S1), in line with 2DG interfering with glycolysis (5).

Fig. 1 Proteomics analysis of the response to 2DG in yeast reveals transcriptional induction of the 2DG6P phosphatases Dog1 and Dog2.

(A) Volcano plot representing changes in protein abundance in total protein extracts of wild-type (WT) yeast in response to 2DG (0.2%), obtained by MS-based proteomics and analyzed with MaxQuant software. The x axis corresponds to the log2 value of the abundance ratio [label-free quantification (LFQ)] between 2DG treatment and the negative control. The y axis represents the ─log10 of the P value of the statistical t test for each quantified protein (n = 3 independent biological replicates). Lines: threshold with an FDR of 0.01. (B) Gene ontology (GO) analysis of the proteins identified as up-regulated in response to 2DG treatment along with their P value and the proteins included in each category. (C) Western blot on total protein extracts of yeast cells expressing endogenously tagged Dog1-GFP or Dog2-GFP, before and after 2DG addition for the indicated times, using an anti-GFP antibody. A longer exposure is displayed for Dog1-GFP cells to highlight the higher abundance of Dog1 after 2DG addition. Rsp5, whose levels did not change upon 2DG addition in all of our experiments, is used as a loading control (n = 2 independent experiments). (D) β-Galactosidase assays of WT yeast cells expressing LacZ under the control of the pDOG1 or pDOG2 promoters, before and after 2DG treatments for 3 hours (±SEM, n = 3 independent experiments, t test). A.U., arbitrary units. (E) Serial dilutions of cultures from the indicated strains were spotted onto SD plates containing no DG or 0.05% 2DG and grown for 3 days at 30°C (n = 2 independent experiments). (F) β-Galactosidase assays of WT and hog1∆ strains expressing LacZ under the control of the pDOG2 promoter, before and after 2DG treatments for 3 hours (±SEM, n = 3 independent experiments, t test). (G) Western blot on total protein extracts of WT and hog1∆ cells endogenously expressing a Dog2-GFP fusion, before and after 2DG addition for 3 hours, using an anti-GFP antibody. Total protein was visualized in gels using a trihalo compound. Glc, glucose. (H) Relative expression of Dog2-GFP under the same conditions as (G) after normalization to total protein and using WT/untreated as a reference (±SEM, n = 3 independent experiments, t test).

These proteomics data revealed an increase in the abundance of the 2DG6P phosphatases Dog1 or Dog2, which could not be discriminated at the MS level because of their high sequence identity (92%). Although these phosphatases can promote 2DG resistance (33, 40, 41), the increased expression of Dog1 and/or Dog2 in response to 2DG was intriguing because it raised the question of how exposure to this synthetic molecule could trigger an adaptive resistance mechanism in yeast. To first confirm that 2DG induced Dog1 and Dog2, we GFP (green fluorescent protein)–tagged each at their chromosomal locus to maintain endogenous regulation. Western blotting revealed that both Dog1 and Dog2 expression levels were increased in the presence of 2DG, and that Dog2-GFP was more abundant than Dog1-GFP (Fig. 1C).

To evaluate the contribution of transcription in the regulation of DOG1 and DOG2 genes by 2DG, we fused the corresponding promoters to a β-galactosidase reporter. 2DG increased the activity of the DOG1 and DOG2 promoters (Fig. 1D), suggesting that the expression of Dog1 and Dog2 was at least, in part, due to increased transcription. The deletion of DOG2 sensitized yeast cells to 2DG, but that of DOG1 had little effect (Fig. 1E), indicating that Dog2 is functionally more important than Dog1, perhaps because of its higher expression level (Fig. 1C). These results indicate that Dog2 participates in the natural tolerance of WT cells to low concentrations (0.05%) of 2DG.

The expression of Dog2, whose endogenous function in yeast metabolism is unknown, is induced by various stresses, such as oxidative and osmotic stresses, through the stress-responsive MAPK ortholog of p38, Hog1 (48). The deletion of HOG1 prevented the maximal induction of pDOG2-LacZ expression (Fig. 1F) and Dog2 induction (Fig. 1, G and H) without affecting the induction of Dog1 (fig. S2A). When we tested the effect of 2DG on Hog1 phosphorylation using an antibody directed against the phosphorylated form of mammalian p38 (49), we found that Hog1 appeared partially activated upon 2DG addition, but to a lower extent compared to the activation induced by hyperosmotic shock (fig. S2B). These data show not only that the stress-activated protein kinase Hog1 participates in Dog2 induction but also that the maximal expression of Dog1 and Dog2 by 2DG involves at least one additional level of regulation.

The expression of Dog1 and Dog2 is induced by the UPR pathway through 2DG-induced ER stress

Exposure of cancer cells to 2DG interferes with N-linked glycosylation, likely due to the structural similarity of 2DG with mannose, a constituent of the N-glycan structures (14). This interference results in ER stress and, consequently, in the induction of the UPR pathway in mammalian cells (14). Treatment of yeast with 2DG also induced a defect in the glycosylation of carboxypeptidase Y (CPY) (Fig. 2A), a vacuolar (lysosomal) protease whose membrane-anchored precursor is N-glycosylated in the course of its intracellular trafficking (50). This defect was not as extensive as that observed upon treatment of cells with tunicamycin, an inhibitor of the first step of glycosylation that also causes ER stress and is a strong UPR inducer (Fig. 2A) (51, 52). The addition of exogenous mannose in the medium suppressed the glycosylation defects caused by 2DG, but not those caused by tunicamycin (Fig. 2A). This result supports the idea that 2DG mimics mannose and interferes with its incorporation in N-glycans, as has been proposed in mammalian cells (14, 17).

Fig. 2 2DG treatment induces Dog2 expression through glycosylation defects that trigger ER stress and the UPR.

(A) WT cells were grown overnight to mid-log phase in synthetic complete (SC) medium, centrifuged, and resuspended in SC medium containing mannose (2%) or not, and treated with 0.2% 2DG or tunicamycin (Tm; 1 μg/ml) for 4 hours. Total protein extracts were Western-blotted for carboxypeptidase Y (CPY) (n = 3 independent experiments). Glc, glucose. Glycos., glycosylation. (B) Schematic of the UPR signaling pathway in yeast showing how ER stress triggers Ire1-mediated splicing of the pre-mRNA encoding the transcription factor Hac1 and the subsequent induction of UPR target genes. (C) β-Galactosidase assays on WT and hac1∆ cells expressing LacZ under the control of a UPR-inducible promoter (pUPRE1) and treated with 0.2% 2DG or tunicamycin (1 μg/ml) for 3 hours (±SEM, n = 3 independent experiments, t test). (D) β-Galactosidase assays on WT cells expressing LacZ under the control of the DOG2 promoter and treated as in (C) (±SEM, n = 3 independent experiments, t test). (E) β-Galactosidase assays on WT and hac1∆ cells expressing LacZ under the control the DOG2 promoter, before and after 3-hour 2DG treatments (±SEM, n = 3 independent experiments, t test). (F) Western blot for GFP on total protein extracts of WT and hac1∆ cells endogenously expressing a Dog2-GFP fusion, before and after 3-hour treatment with 2DG or tunicamycin. (G) Relative expression of Dog2-GFP under the same conditions as (F) after normalization to total protein and using WT/untreated as a reference (±SEM, n = 3 independent experiments, t test). (H) Serial dilutions of cultures from the indicated strains were spotted onto SC plates (supplemented with 2% mannose when indicated) containing no DG or 0.05% 2DG, and were grown for 3 days at 30°C (n = 2 independent experiments). (I) Serial dilutions of cultures from the indicated strains overexpressing DOG2 (pGPD-DOG2) or not (Ø) were spotted onto SC-Ura plates containing 0, 0.05, or 0.2% 2DG. The plates were scanned after 3 days of incubation at 30°C (n = 2 independent experiments).

In yeast, the UPR pathway is initiated by a multifunctional ER membrane protein named Ire1. Upon sensing ER stress and after dimerization, Ire1 splices pre-mRNAs encoding the transcription factor Hac1, leading to its translation and the trans-activation of Hac1 targets carrying a UPRE (UPR element) in their promoter (Fig. 2B) (53). To address whether 2DG can induce the UPR pathway in yeast, we used a UPRE-driven reporter, UPRE1-LacZ (52). 2DG elicited the expression of this reporter in a manner dependent on Hac1, confirming that 2DG is a bona fide UPR inducer (Fig. 2C).

We then tested whether 2DG-mediated induction of Dog2 involved the UPR pathway. First, we found that tunicamycin strongly induced the pDOG2-LacZ reporter (Fig. 2D), suggesting that glycosylation defects and/or the ensuing ER stress promotes DOG2 expression. Moreover, the induction of this reporter by 2DG was reduced by more than twofold in hac1∆ mutant cells, suggesting that UPR contributed to this induction (Fig. 2E). Comparable results were obtained for pDOG1-LacZ (fig. S3, A and B). These results were confirmed at the protein level for Dog2 (Fig. 2, F and G).

UPR-compromised mutants such as hac1∆ and ire1∆ were both hypersensitive to 2DG (Fig. 2H), suggesting that they cannot cope with ER stress caused by 2DG. The addition of exogenous mannose in the medium, which can alleviate the glycosylation defects caused by 2DG (Fig. 2A), restored 2DG tolerance of these mutants to a comparable level as that of WT cells (Fig. 2H), but not that of snf1∆, a 2DG-hypersensitive mutant (31). Together, these results confirm that 2DG induces ER stress by interfering with N-glycosylation, that the subsequent activation of the UPR pathway increased Dog1 and Dog2 expression, and that UPR mutants display an increased sensitivity to 2DG, which correlates with a lower level of Dog1 and Dog2 expression. This was further supported by the ability of Dog2 overexpression to restore the growth of hac1∆ or ire1∆ cells at various concentrations of 2DG (Fig. 2I).

2DG also activates the MAPK-based cell wall integrity pathway, which additionally contributes to Dog1 and Dog2 expression

To address a possible additional contribution of other signaling pathways that would contribute to the 2DG-induced expression of Dog1 and Dog2, we ran a bioinformatics analysis on the proteins that were more abundant upon 2DG treatment (Fig. 1A and data file S1) using YEASTRACT (54) to evaluate whether the observed variations in protein expression could reveal a potential transcriptional signature. The best hit (corresponding to 35 of 78 candidates, P = 0) was the MADS-box transcription factor Rlm1, a downstream target of the cell wall integrity (CWI) signaling pathway (55). The CWI pathway is activated by several stresses such as cell wall damage and involves plasma membrane–localized sensors, the guanosine triphosphatase Rho1 and its cofactors [guanine nucleotide exchange factors and accessory proteins], protein kinase C, and a cascade leading to the activation of the MAPK Slt2 and downstream transcription factors, such as Rlm1 (Fig. 3A) (55). There are well-established links between CWI signaling and ER stress (5660); in particular, the CWI pathway is required for viability upon tunicamycin-induced ER stress. Moreover, 2DG causes cell wall defects (43, 6163) and as such might trigger CWI pathway activation. We found that the deletion of many genes in the CWI pathway that act upstream of Rlm1, such as those encoding the sensors MID2 and WSC1, caused an increase in 2DG sensitivity (Fig. 3B). Thus, much like the UPR pathway, the CWI pathway is required for 2DG tolerance. 2DG treatment induced the phosphorylation of the MAPK Slt2 within the first hour of exposure (fig. S4A). 2DG also triggered the induction of a reporter consisting of an Rlm1-regulated promoter fused to LacZ (fig. S4B). Thus, the CWI pathway is activated by 2DG treatment.

Fig. 3 2DG activates the MAPK-based CWI pathway, which is required for 2DG tolerance and additionally contributes to the regulation of Dog2 expression.

(A) Schematic of the CWI pathway showing the various components and their requirement for growth on 2DG (see color code in the inset) based on drop tests shown in (B). (B) Serial dilutions of cultures from the indicated deletion strains were spotted onto SD plates containing no DG or 0.05% 2DG and grown for 3 days at 30°C (n = 2 independent experiments). (C) β-Galactosidase assays on WT and slt2∆ cells expressing LacZ under the control of the DOG2 promoter, before and after 3-hour 2DG treatments (±SEM, n = 3 independent experiments, t test). (D) Western blot on total protein extracts of WT and slt2∆ cells expressing an endogenously tagged Dog2-GFP fusion, before and after 3-hour treatment with 2DG, using an anti-GFP antibody. (E) Relative expression of Dog2-GFP under the same conditions as (D) after normalization to total protein and using WT/untreated as a reference (±SEM, n = 3 independent experiments, t test). (F) Serial dilutions of cultures from the indicated strains were spotted onto SC plates (supplemented with 2% mannose when indicated) containing no DG or 0.05% 2DG and were grown for 3 days at 30°C (n = 2 independent experiments).

We thus questioned whether CWI activation by 2DG contributed to Dog1 and Dog2 induction. The activity of both promoters was decreased in the slt2∆ mutant (Fig. 3C and fig. S4C), which was confirmed for Dog2 by analyzing the expression levels of Dog2-GFP (Fig. 3, D and E). Thus, similarly to tunicamycin (58), 2DG induces CWI signaling and promotes increased Dog1 and Dog2 expression, in addition to inducing the UPR pathway. These effects may, in turn, contribute to 2DG tolerance. However, the addition of exogenous mannose did not improve tolerance of CWI mutants to 2DG (Fig. 3F). Thus, ER stress relief is not sufficient to suppress 2DG toxicity in these mutants, suggesting that 2DG has other cellular effects beyond triggering ER stress.

A third pathway inhibits the expression of Dog2, but not Dog1, by glucose availability and participates in the resistance of glucose-repression mutants to 2DG

2DG inhibits the activity of hexokinase and phospho-glucose isomerase (68), thereby impairing glycolysis and leading to energetic stress. Accordingly, 2DG treatment activates AMPK in mammals (64). Several lines of evidence indicate that the activity of the yeast AMPK ortholog Snf1 is important for 2DG tolerance. Whereas the snf1∆ mutant is hypersensitive to 2DG, the reg1∆ mutant, in which Snf1 is hyperactive, displays increased 2DG resistance that depends on Snf1 activity (31). We questioned whether some of the resistance or sensitivity phenotypes associated with this pathway (Fig. 4A) could be due to an altered level of Dog1 or Dog2 expression, particularly because Snf1 promotes Dog2 expression (48). To confirm the effect of Snf1 on Dog1 and Dog2 expression, glucose-grown cells were transferred to a medium containing lactate as a sole carbon source, which should trigger Snf1 activation and, consequently, the de-repression of glucose-repressed genes (Fig. 4A) (65). As expected, switching cells to lactate medium induced Snf1 activation, as determined using antibodies directed against the activated (phosphorylated) form of human AMPKα, which cross-reacts with yeast Snf1 (Fig. 4B) (66). The expression of Dog2, but not that of Dog1, was increased under these conditions in an Snf1-dependent manner (Fig. 4B). In line with these data, the pDOG2-LacZ reporter was induced in cells transferred to lactate medium, but this response was decreased in the snf1∆ mutant, suggesting that DOG2 is repressed by glucose and is thus constantly repressed in the snf1∆ mutant (Fig. 4C), which may explain why the snf1∆ mutant is sensitive to 2DG. Although Dog1/2 overexpression in an snf1∆ mutant has been reported to not rescue resistance to 2DG (47), we thought that this might be because the authors used a high-copy plasmid in which Dog2 expression was still under the control of its endogenous, Snf1-regulated promoter. When overexpressed using a strong and constitutive promoter (pGPD), Dog2 rescued Snf1 growth in 2DG-exposed cells (Fig. 4D).

Fig. 4 DOG2, but not DOG1, is negatively controlled by glucose availability through transcriptional repression by Mig1-Mig2 and the kinase Snf1.

(A) Schematic of the glucose repression pathway showing how Snf1 (the yeast homolog of AMPK), PP1 (composed of Glc7 and Reg1 subunits), and their downstream transcriptional repressors Mig1/Mig2 regulate glucose-repressed genes in response to glucose availability or absence (such in the presence of lactate). (B) WT and snf1∆ strains, both expressing endogenously tagged Dog1-TAP and Dog2-GFP fusions, were grown overnight in SC medium and then either treated with 0.2% 2DG or switched to an SC lactate medium for 4 hours. Dog1-TAP was detected with the peroxidase–anti-peroxidase (PAP) complex and Dog2-GFP with anti-GFP antibodies, phosphorylated (p) Snf1 with anti-phospho-AMPK and total Snf1 with an anti-polyHis tag (because Snf1 contains a stretch of 13 histidine residues that can be used for its detection) (n = 2 independent experiments). (C) β-Galactosidase activity on WT and snf1∆ cells expressing LacZ under the control of the DOG1 or the DOG2 promoter, before and after 3-hour growth in lactate. The fold induction after transfer to lactate is indicated for each promoter in each strain (±SEM, n = 4 independent experiments). (D) WT and snf1∆ strains, transformed either with a genomic clone containing both DOG1 and DOG2 under the control of its own promoter (pend:DOG) or with a vector containing DOG2 under the control of the strong GPD promoter (pGPD:DOG2), were grown, serially diluted, and spotted onto SC plates (SC-Leu or SC-Ura) with or without 0.2% DG, and grown for 3 days at 30°C (n = 2 independent experiments). (E) β-Galactosidase assays on WT and the indicated deletion mutants expressing LacZ under the control the DOG2 promoter after overnight growth in SC medium (exponential phase) (±SEM, n = 6 independent experiments). (F) The indicated strains, all expressing an endogenously tagged Dog2-GFP fusion, were grown overnight in SC medium (to exponential phase). Total protein extracts were immunoblotted with anti-GFP antibodies. Rsp5 was used as a loading control (n = 2 independent experiments). (G) Serial dilutions of cultures of the indicated mutants were spotted on yeast extract-peptone-dextrose (YPD) plates containing 0, 0.2, or 0.5% of 2DG. Plates were scanned after 3 days of growth at 30°C (n = 3 independent experiments).

In contrast, Dog2 expression was increased in glucose-repression mutants that display increased Snf1 activity, such as mutants lacking the hexokinase Hxk2 or the PP1 regulatory phosphatase Reg1 (44, 67), both at the promoter and the protein levels (Fig. 4, E and F). In addition, lack of the Snf1-regulated transcriptional repressors Mig1/Mig2 (Fig. 4A) also led to an increase in Dog2 expression (Fig. 4, E and F). A regulatory sequence in the promoter that is present between −250 and −350 bp relative to the ATG start codon, combined with a proposed Mig1-binding site located at ~200 bp (40, 48), was critical for this glucose-mediated repression (fig. S5, A and B). The reg1∆ or hxk2∆ mutants, which are more tolerant to 2DG than to WT (30, 31), showed increased Dog2 expression (Fig. 4, E and F), which was also the case for the double mutant mig1mig2∆ (Fig. 4G). We deleted DOG1 and DOG2 in these mutants to evaluate their contribution to 2DG resistance. The absence of Dog1 and Dog2 sensitized all strains to 2DG (Fig. 4G); in particular, the mig1mig2dog1dog2∆ showed a level of sensitivity that was comparable to that of the WT, demonstrating that the resistance of the mig1mig2∆ mutant was due to increased DOG expression. In contrast, the reg1dog1dog2∆ and the hxk2dog1dog2∆ strains remained more resistant to 2DG than the WT, suggesting additional mechanisms of resistance. Last, we observed that the snf1mig1mig2∆ mutant was resistant to 2DG despite the absence of Snf1, in line with the idea that snf1∆ is 2DG sensitive because of the constitutive repression of Mig1/Mig2 target genes, such as Dog2 and possibly other genes (Fig. 4F). Overall, we conclude that Dog2 is also regulated by glucose availability through Snf1 activity, which contributes to the resistance of glucose-repression mutants to 2DG.

Increased expression of DOG2 is frequently observed in spontaneous 2DG-resistant clones

Previous screens have identified 2DG-resistant mutants (26, 29, 35). The initial purpose of these screens was to identify mutants that were insensitive to the repressive effect of 2DG or that were impaired for glucose phosphorylation (because only 2DG6P is toxic to cells) and, consequently, were performed on media containing carbon sources other than glucose. The mechanisms involved in 2DG resistance on glucose medium were tackled later by screening of the deletion library (30), which led to the identification of resistant mutants, some of which were subsequently confirmed (31). In the course of our experiments, we often found that 2DG-resistant clones could spontaneously arise from WT or even 2DG-susceptible strains (Fig. 1E). Yeast acquires 2DG resistance at a high frequency, which can be accompanied by an increase in 2DG6P phosphatase activity (28, 38). To study whether Dog2, which is functionally critical for 2DG resistance, was up-regulated during the emergence of spontaneous 2DG-resistant mutants, we spread ~2 × 106 cells on glucose-based medium containing 0.2% 2DG and selected clones that had appeared after 6 days of growth. In total, 24 clones were obtained, whose resistance to 2DG was confirmed (fig. S6A). Using the pDOG2-LacZ reporter, we found that 13 clones displayed significantly increased DOG2 promoter activity as compared to WT strains (Fig. 5A), suggesting that Dog2 overexpression is a frequent feature of 2DG-resistant clones. Among those, we expected to isolate reg1 and hxk2 mutants because both are resistant to 2DG in glucose medium [see Fig. 5G; (30, 31)]. To identify them, we first tested whether some of the isolated resistant clones displayed phenotypes typical of reg1 mutants, such as sensitivity to tunicamycin (68, 69) or selenite (which enters the cell through the glucose-repressed transporter Jen1) (70, 71). Three of the isolated clones displayed these phenotypes (fig. S6A). Sequencing of the REG1 coding sequence in these clones revealed the presence of nonsense mutations, which likely explain Reg1 loss of function (fig. S6B). To identify potential hxk2 mutants, we performed a complementation test by crossing the 2DG-resistant clones with either a WT strain or an hxk2∆ strain and tested the ability of the resulting diploids to grow on 2DG. All diploids generated by the cross with the WT strain lost their ability to grow on 2DG, except for two (clones #23 and #24) (fig. S6C), indicating that 2DG resistance was generally caused by recessive mutations. In contrast, 12 diploids obtained by the cross with the hxk2∆ strain maintained their ability to grow on 2DG (fig. S6C), suggesting that 2DG resistance of the initial strains was caused by a deficiency in HXK2 function. Sequencing of the HXK2 open reading frame (ORF) in these 12 mutants (fig. S6D) revealed that 2 did not display any mutation in the HXK2 ORF (clones #1 and #8) and may represent regulatory mutants in cis, possibly in regions that were not sequenced. In favor of this hypothesis, we found that their 2DG sensitivity was restored by re-expression of HXK2 using a multicopy, genomic clone (fig. S6E). The other 10 mutants carried at least one mutation in the HXK2 coding sequence (fig. S6D). Four mutants acquired nonsense mutations (clones #12, #15, #16, and #20), five mutants (#2, #3, #11, #13, and #14) carried a missense mutation in residues conserved across all hexokinase sequences tested (fig. S7, alignment), and one mutant (#21) bore two missense mutations. All of these missense mutations occurred in residues involved in glucose or ATP binding or were located near such residues (fig. S7) (7274).

Fig. 5 The characterization of spontaneous 2DG-resistant strains identifies mutants showing increased Dog2 expression, including a new mutant allele of CYC8.

(A) Twenty-four clones showing a spontaneous resistance to 0.2% 2DG were isolated. The β-galactosidase activity of these mutants, due to the expression of the LacZ reporter driven by the DOG2 promoter, was measured after overnight growth in SC medium (to exponential phase) (±SEM, n = 4 independent experiments, t test). Colors represent the identity of the mutants as determined in fig. S6 (A and B) for reg1 and fig. S6 (C to E) for hxk2 and (B) to (G) for cyc8. (B) Schematic of the domain organization of the Cyc8 protein, showing the Poly(Q) and Poly(QA) repeats and the N-terminal tetratricopeptide (TPR) repeats. Red: mutation identified by whole-genome resequencing of the spontaneous 2DG-resistant mutants #9 and #10. (C) A WT strain, the mutant strains #9 and #10, and the reg1∆ mutant (used as a positive control) were grown in SC medium (to exponential phase). Total protein extracts were immunoblotted for invertase [invertase is heavily glycosylated and migrates as a smear (116)] (n = 2 independent experiments). (D) WT and mutants #9 and #10 were transformed with low-copy (centromeric) plasmid either empty or containing WT CYC8 or mutant cyc8 (Gln320*), spotted on SC-Leu or SC-Leu + 0.2% 2DG medium, and grown for 3 days at 30°C. Middle: the control plate was scanned and then washed for 1 min under a constant flow of water, and then scanned again (n = 2 independent experiments). (E) β-Galactosidase activity on WT and mutants #9 and #10 expressing LacZ under the control the DOG2 promoter and transformed with an empty vector or a low-copy vector containing WT CYC8, after growth in SC medium (normalized to the value of the WT, ±SEM, n = 3 independent experiments, t test). (F) Western blot on total protein extracts of WT and mutants #9 and #10 cells expressing an endogenously tagged Dog2-GFP fusion and transformed with either an empty plasmid or a low-copy (centromeric) plasmid containing WT CYC8 after growth in SC medium, using an anti-GFP antibody. (G) Relative expression of Dog2-GFP under the same conditions as (F) after normalization by total proteins and using the WT control as a reference (±SEM, n = 3 independent experiments, t test). (H) Serial dilutions of cultures of the indicated mutants were spotted on SC medium or SC + 0.2% 2DG medium and grown for 3 days at 30°C (n = 2 independent experiments). mut, mutant.

Of the 2DG-resistant clones that expressed pDOG2-LacZ at a high level, two clones (#9 and #10) lacked mutations in either REG1 or HXK2, suggesting that their resistance involved a mechanism that did not involve these proteins. Whole-genome resequencing of the genomic DNA isolated from these clones and comparison with that of the parent strain revealed several single-nucleotide polymorphisms (table S1), including a nonsense mutation in CYC8 (C958>T) that was identified in both strains, causing a premature stop codon at position 320 (Fig. 5B). The CYC8 gene is also known as SSN6 (Suppressor of snf1), and mutations in this gene lead to constitutive expression of the glucose-repressed gene encoding invertase (SUC2), even in an snf1 mutant (75, 76). CYC8 encodes a transcriptional co-repressor that controls the expression of glucose-regulated genes (77). Because the repression of DOG2 expression by glucose is controlled by Snf1 and Mig1/Mig2 (Fig. 4, E and F), we hypothesized that CYC8 could also take part in DOG2 regulation, such that a mutation in CYC8 could lead to increased 2DG resistance through DOG2 overexpression. We observed that mutants #9 and #10 expressed invertase even when grown in glucose medium (repressive conditions), in agreement with a mutation in CYC8 (Fig. 5C). We also observed that these mutants displayed increased adhesion or persistence of colony structures upon washing the plate (78) that is often associated with mutants that are prone to flocculation, such as cyc8 mutants (Fig. 5D) (79, 80). Introduction of a low-copy (centromeric) plasmid containing CYC8 under the control of its endogenous promoter suppressed the 2DG resistance of mutants #9 and #10 as well as their adhesion to agar plates (Fig. 5D). This was not the case when mutants #9 and #10 expressed a truncated version of CYC8 identified in this screen (Fig. 5D), although these mutants did not display slow growth in contrast to the cyc8∆ mutant, suggesting that it is at least partially active for other functions. Last, we tested whether the strong activity of the DOG2 promoter observed in mutants #9 and #10 (Fig. 5A) was also due to the lack of a functional CYC8. The re-introduction of low-copy vector containing CYC8 led to a decreased reporter expression in these mutants (Fig. 5E). This result was confirmed by examining Dog2-GFP expression at the protein level (Fig. 5, F and G). Therefore, DOG2 expression is repressed by Cyc8 and spontaneous cyc8 mutants display an increased Dog2 expression. The increased resistance of these mutants to 2DG also depended on the expression of Dog1 and Dog2 as the deletion of DOG1 and DOG2 restored their sensitivity to 2DG (Fig. 5H). Together, this initial characterization revealed that DOG2 overexpression is a common phenomenon within spontaneous 2DG-resistant clones, both in known 2DG-resistant mutants (reg1 and hxk2) and in the 2DG-resistant cyc8 mutants that we isolated.

HDHD1, a human member of the HAD-like phosphatase family, is a 2DG6P phosphatase involved in 2DG resistance

The increase in Dog2 expression in various spontaneously 2DG-resistant mutants reminded us of an earlier study in HeLa cells, showing increased 2DG6P phosphatase activity in isolated 2DG-resistant clones (20). Dog1 and Dog2 belong to the family of HAD-like phosphatases, which are conserved from bacteria to human (Fig. 6A). The bacterial homolog of Dog1/Dog2, named YniC, can also dephosphorylate 2DG6P in vitro (81), and we found that the expression of YniC in the double dog1dog2∆ yeast mutant also restored 2DG resistance (Fig. 6B). We used this phenotype to identify potential human homologs (Fig. 6B). A Position-Specific Iterated Basic Local Alignment Search Tool (PSI-BLAST) analysis (82) on the human proteome using the Dog2 protein sequence retrieved HDHD1-isoform a (NP_036212.3) as the candidate with highest (39%) homology. HDHD1 is a HAD-like phosphatase with in vitro activity towards phosphorylated metabolites such as pseudouridine-5′-phosphate (83). When expressed in the double dog1dog2∆ mutant, HDHD1 partially rescued growth on 2DG-containing medium (Fig. 6B). Yeast growth on 2DG was not restored by the expression of HDHD4 (also named NANP or N-acylneuraminate-9-phosphatase) (84)), which belongs to the same subfamily as HDHD1 within the HAD-phosphatase family (37% homology), or of PSPH [phosphoserine phosphatase; (85)], another close family member (fig. S8, A and B). Moreover, among the four predicted isoforms of HDHD1 (isoforms 1 to 4), only HDHD1-1 rescued the growth of the dog1dog2∆ on 2DG-containing medium (fig. S8C) although all isoforms were expressed in yeast, as confirmed by Western blotting using HDHD1 antibodies (fig. S8D).

Fig. 6 The human phosphatase HDHD1 has a 2DG6P phosphatase activity and its overexpression leads to 2DG resistance in HeLa cells.

(A) Multiple protein sequence alignment of yeast Dog1, Dog2, E. coli yniC, and the human proteins HDHD1, HDHD4, and PSPH aligned with ClustalX 2.0. The highly conserved catalytic aspartates are displayed in yellow. The first six amino acids of PSPH were truncated to optimize the N-terminal alignment of its catalytic aspartates with the other phosphatases. (B) Serial dilutions of WT and dog1dog2∆ strains transformed with the indicated plasmids were spotted on SC-Ura medium with or without 0.05% 2DG and were scanned after 3 days of growth at 30°C (n = 2 independent experiments). (C) Serial dilutions of dog1dog2∆ strains transformed with an empty vector or vectors allowing the expression of an HDHD1 or its predicted catalytic mutant, HDHD1-DD>AA (in which the N-terminal catalytic aspartates were mutated to alanines), were spotted on SC-Ura medium with or without 0.05% 2DG and were scanned after 3 days of growth at 30°C (n = 2 independent experiments). (D) Recombinant, His-tagged HDHD1 and HDHD1-DD>AA were expressed in bacteria and purified for in vitro enzymatic tests; 0.7 μg was loaded on a gel to show homogeneity of the protein purification. (E) In vitro 2DG6P phosphatase activity of HDHD1 and HDHD1>DDAA as measured by assaying glucose release from 2DG6P (n = 3 independent experiments). (F) Growth of HeLa cells transfected with an empty vector (□) or with a construct allowing the overexpression of HDHD1 (○) over time in the absence (open symbols) or presence (filled symbol) of 5 mM 2DG. The number of cells is normalized to that of the untransformed/untreated cells after 3 days (±SEM, n = 3 independent experiments, t test).

The mutation of conserved aspartate residues (Asp12 and Asp14; Fig. 6A), predicted to be essential for the catalytic activity of HAD phosphatases (86), abolished the ability of HDHD1 to restore growth of the dog1dog2∆ mutant on 2DG (Fig. 6C and fig. S8E), suggesting that it may act as a 2DG6P phosphatase. This notion was confirmed by the ability of purified recombinant HDHD1 (Fig. 6D) to dephosphorylate 2DG6P in vitro, an activity that required its putative catalytic residues (Fig. 6E). We then tested whether HDHD1 overexpression in HeLa cells could lead to an increased resistance to 2DG. Low concentrations of 2DG (5 mM) in the presence of glucose (25 mM) were sufficient to inhibit the growth of HeLa cells transfected with an empty vector, whereas those that overexpressed HDHD1 were insensitive to 2DG treatment (Fig. 6F). These results suggest that dysregulated expression of HDHD1 could modulate 2DG resistance in human cells.

DISCUSSION

Although the first studies examining the effect of 2DG on glycolysis in yeast and in normal and cancer tissues were published more than 60 years ago (8789), the mode of action for 2DG is not fully understood. 2DG can be considered to be a general competitor of glucose (90) and a glycolysis inhibitor because of its inhibition of glycolytic enzymes (68). However, this view has been challenged because exposure of cancer cells to 2DG interferes with N-linked glycosylation, likely because of the structural similarity of 2DG with mannose (91), resulting in ER stress and UPR induction in mammalian cells (14, 16). Therefore, 2DG interferes with other cellular functions beyond glycolysis.

In this study, we used an unbiased MS approach to analyze the effects of 2DG on the total cellular proteome. We found that the abundance of many glycolytic enzymes was increased, likely due to impaired glycolysis, as well as many genes regulated by the MAPK-based CWI pathway, suggestive of its activation. Using the list of 2DG–up-regulated proteins, we also identified the phosphatases Dog1 and/or Dog2, which have been shown to play a role in 2DG resistance (33, 40, 41). Dog2 induction upon 2DG treatment was intriguing because it was not clear how a synthetic molecule such as 2DG could trigger a resistance mechanism. It should be noted that the cellular functions of Dog1 and Dog2, beyond 2DG dephosphorylation, are unknown. They can dephosphorylate various phosphorylated sugars in vitro (40), in agreement with the ability of other members of the HAD family of phosphatases to accommodate various substrates (81). Therefore, it seemed plausible that Dog1/2 induction was a response to one or more of the cellular consequences of 2DG treatment, rather than to 2DG itself. Our study revealed that the expression of Dog1 and Dog2 was actually controlled by multiple signaling pathways, each of which is activated as a response to 2DG (Fig. 7).

Fig. 7 Working model.

Glucose phosphorylation triggers the onset of the glucose-repression pathway in which PP1 inactivates Snf1. This leads to the lack of phosphorylation of Mig1 and Mig2, which remain in the nucleus to mediate the glucose repression of genes such as DOG2. The deletion of REG1, HXK2, or MIG1 and MIG2 or a mutation in CYC8 leads to 2DG resistance, which is at least partially mediated through increased expression of DOG2, leading to the dephosphorylation of 2DG6P. In contrast, the deletion of SNF1 causes an increased sensitivity to 2DG, which can be rescued by the deletion of MIG1 and MIG2 or by Dog2 overexpression. In parallel, 2DG6P causes (i) ER stress and triggers the UPR pathway, which stimulates DOG2 expression through the transcription factor Hac1, and (ii) the CWI pathway, likely through interference with polysaccharide and cell wall synthesis, which also induces DOG2 through the transcription factor Rlm1.

This study also probed the signaling pathways that are activated upon 2DG exposure and determined the causes for these activations. First, we confirmed that 2DG triggered ER stress and the onset of the UPR pathway. This finding was not unexpected given the proposed mode of action of this drug and the available data in the literature on mammalian cells (14, 16), but had not been formally proven in yeast with current tools. The induction of a UPR reporter upon 2DG treatment and the hypersensitivity of UPR mutant strains such as hac1∆ or ire1∆ support this conclusion. The latter phenotype was alleviated by restoring N-glycosylation through the addition of mannose to the culture medium. The consolidation of the links between 2DG and ER stress, notably the identification of UPR mutants as 2DG-hypersensitive mutants, suggests that drugs targeting the UPR may synergize with 2DG in a glycolytic cancer background. We also showed that Dog1 and Dog2 are induced by ER stress, in line with high-throughput studies, suggesting that Dog1 and Dog2 are UPR-induced genes (92).

Second, we showed that 2DG activates the MAPK-based CWI pathway, which mediates the cellular response to cell wall alteration and other stresses. Several studies reported connections between ER stress signaling and the CWI pathway (5660), and 2DG-elicited N-glycosylation defects and ER stress may have repercussions on the CWI pathway. For instance, CWI pathway mutants are sensitive to tunicamycin because they lack an ER surveillance pathway, which normally prevents the inheritance of stressed ER to the daughter cell during cell division (58). Many CWI mutants were sensitive to low concentrations (0.05%) of 2DG, but unlike UPR mutants, their growth was not restored by the addition of exogenous mannose, suggesting that an N-glycosylation defect is not the sole reason why CWI mutants are hypersensitive to 2DG. However, 2DG interferes with the synthesis of structural polysaccharides that make up the yeast cell wall because 2DG acts as an antagonist of mannose and glucose incorporation into these polymers (42, 6163). 2DG exposure leads to yeast cell lysis at sites of growth, which is where glucan synthesis occurs (43) and where the major glucan synthase Fks1 is localized (93). 2DG-induced weakening of the cell wall could trigger the CWI pathway (Fig. 7), which would explain why the main sensors responsible for sensing cell wall damage, Wsc1 and Mid2 [reviewed in (55)], are required for growth on 2DG. In addition, we report that the expression of Dog1 and Dog2 was decreased in CWI mutants, which could further sensitize these strains to 2DG. The effect of 2DG on cell wall synthesis suggests an interference of 2DG with UDP-glucose metabolism that is likely to be conserved in metazoans (94), where it may affect metabolic pathways involving these precursors, such as glycogen synthesis (95, 96).

Our data also revealed that Dog2 expression was regulated at the transcriptional level by glucose availability through the glucose-repression pathway, which is controlled by the kinase Snf1 and the regulatory subunit of the PP1 phosphatase Reg1. These findings explain why reg1 mutants have increased 2DG6P phosphatase activity (28, 38, 39), because the glucose-mediated repression of Dog2 is defective in these mutants. The negative regulation of Dog2 by this pathway may also contribute to the 2DG resistance of hxk2 and mig1 mutants, which participate in the glucose-repression pathways and have been identified in previous screens (29, 35). The 2DG resistance displayed by the reg1∆ and hxk2∆ strains partially depended on increased expression of Dog2, because even when these mutants lacked both DOG genes, they still grew better than a WT strain on 2DG-containing media. In the hxk2∆ mutant, the lack of hexokinase 2 is compensated for by the expression of other glucose-phosphorylating enzymes (such as Hxk1 and the yeast glucokinase homolog, Glk1) (97), which may be less prone to phosphorylate 2DG (36), and thus could lead to a lower accumulation of 2DG6P in the cell. Hxk2 also has nonmetabolic roles beyond sugar phosphorylation [reviewed in (98)], and loss of these nonmetabolic roles may additionally contribute to this phenotype. However, many of the hxk2 point mutants isolated during the characterization of spontaneous 2DG-resistant clones were affected at or near glucose-binding residues, suggesting a primary metabolic role for Hxk2 in 2DG resistance. Concerning the reg1∆ strain, additional mechanisms of resistance beyond an increased expression of Dog2 also remain to be investigated. Previous work suggested that the resistance of the reg1∆ mutant depends on Snf1 hyperactivity, because the additional deletion of SNF1 in this background restores 2DG sensitivity to the reg1∆ strain (31). Thus, additional Snf1-dependent mechanisms cooperate with the increased expression of DOG2 to allow reg1∆ cells to resist 2DG.

After 2DG import and phosphorylation, the amount of 2DG6P in cells has been reported to exceed that of G6P by up to 80-fold (99), although this number should probably be confirmed by more direct and modern methods. Overexpression of Dog2 can revert the repressive effect of 2DG (33), suggesting that it can clear the 2DG6P pool up to a point at which 2DG6P is no longer detected by a yet unknown cellular glucose-sensing mechanism. Thus, Dog2 overexpression appears to be a good strategy for acquiring 2DG resistance, and Dog2 was overexpressed in the majority of the spontaneously resistant mutants we isolated, including strains with mutations in REG1 and HXK2. In addition, we identified two mutants carrying a point mutation in the gene encoding the transcriptional repressor Cyc8, leading to a truncated protein at codon 320, within its eighth predicted TPR repeat. The TPR repeats are involved in the interaction of Cyc8 with its co-repressor Tup1 and in its recruitment to specific promoters, possibly through pathway-specific DNA-binding proteins (100). Point mutations in TPR units 9 and 10 affect Cyc8 function in a similar manner to the mutation we isolated (namely, an effect on glucose repression but not global growth), suggesting that alterations at this region only affect a subset of Cyc8 functions (101). This finding confirms the differential requirement of TPR repeats for the various functions of Cyc8 and, in particular, the involvement of the TPR repeats 8 to 10 in glucose repression (102). Together, these data describe DOG2 overexpression as a successful strategy to overcome 2DG toxicity.

On the basis of sequence similarity, we identified HDHD1 as an enzyme displaying in vitro 2DG6P phosphatase activity and whose overexpression in both yeast and HeLa cells allowed resistance to 2DG. Whether HDHD1 is responsible for the 2DG resistance previously reported in human cells (20) remains to be investigated.

Together, our work shows that 2DG-induced activation of multiple signaling pathways can rewire the expression of endogenous proteins that target 2DG6P to promote 2DG tolerance, and whose increased expression can lead to 2DG resistance. Because of the existence of endogenous genes that can confer 2DG resistance when overexpressed, such as HDHD1, and of the perturbation of several cellular pathways in mammalian cells by 2DG, it is possible that similar resistance strategies occur in human cells. Such strategies likely superimpose on other resistance mechanisms that should be scrutinized in the future.

MATERIALS AND METHODS

Yeast strain construction and growth conditions

All yeast strains used in this study derive from the S. cerevisiae BY4741 or BY4742 background and are listed in table S2. Apart from the mutant strains obtained from the yeast deletion collection (Euroscarf) and the fluorescent GFP-tagged strains originating from the yeast GFP clone collection (103), all yeast strains were constructed by transformation with the standard lithium acetate–polyethylene glycol protocol using homologous recombination and verified by polymerase chain reaction (PCR) on genomic DNA prepared with a lithium acetate (200 mM)/SDS (0.1%) method (104).

Yeast cells were grown in YPD medium (2%) or in SC medium [containing yeast nitrogen base (1.7 g/liter; MP Biomedicals), ammonium sulfate (5 g/liter; Sigma-Aldrich), the appropriate drop-out amino acid preparations (MP Biomedicals), and 2% (w/v) glucose, unless otherwise indicated]. Alternatively, SC medium could contain 0.5% lactate as a carbon source (from a 5% stock adjusted to pH 5; Sigma-Aldrich). Precultures of 4 ml were incubated at 30°C for 8 hours and diluted in fresh medium on the evening to 20-ml cultures grown overnight with inoculation optical densities at 600 nm (OD600) of 0.0003 for YPD and 0.001 for SC medium, giving a culture at mid-log phase the next morning.

For glucose-depletion experiments, cultures were centrifuged and resuspended in an equal volume of SC/lactate medium and incubated at 30°C during the indicated times. For 2DG, NaCl, and tunicamycin treatments, the compounds were added to mid-log phase yeast cultures grown overnight to respective final concentrations of 0.2% (w/v), 400 mM, and 1 μg/ml and incubated for the indicated times. 2DG and tunicamycin were purchased from Sigma-Aldrich. The mannose-supplemented medium (Fig. 2A) consisted of an SC medium that contained all the elements indicated above plus 2% (w/v) mannose.

Plasmid construction

All the plasmids presented in this study are listed in table S3 and were directly constructed in yeast using plasmid homologous recombination (105). DNA inserts were amplified by PCR using 70-mer primers containing 50-nt homology overhangs and Thermo Fisher Phusion High-Fidelity DNA Polymerase, and receiver plasmids were digested with restriction enzymes targeting the insertion region. Competent yeast cells rinsed with lithium acetate were incubated for 30 min at 30°C with 20 μl of the PCR product and 1 μl of the plasmid digestion product, followed by a heat shock at 42°C for 20 min and a recovery phase in rich medium (YPD) for 90 min at 30°C and plated on synthetic medium without uracil. The pDOG1/2-LacZ vectors were generated using the pJEN1-LacZ vector obtained from Bernard Guiard (106) and cloning the pDOG1 or pDOG2 promoters (1 kb) at Bgl II and Eco RI sites. The DOG1, DOG2, HDHD1 (all four isoforms), HDHD4, PSPH, and yniC overexpression vectors were obtained by digesting a pRS426 vector (2μ, URA3) containing the pGPD (glyceraldehyde-3-phosphate dehydrogenase gene, TDH3) (107) promoter with Eco RI and Bam HI enzymes. In parallel, inserts were PCR-amplified using the following DNA templates: DOG1 and DOG2 from yeast genomic DNA preparations (primers: oSL1166/1167 and oSL1141/1142), the yniC ORF from Escherichia coli DH5α cells (oSL1172/1173), and the human gene ORFs (UniProt identifiers: HDHD1 = Q08623-1/2/3/4, HDHD4 = Q8TBE9, and PSPH = P78330) from DNA sequences generated by gene synthesis after codon optimization for yeast expression (Eurofins Genomics) (HDHD1: oSL1170/1171 for all isoforms except isoform 3: oSL1170/1216; HDHD4: oSL1214/1215; PSPH: oSL1212/1213). These PCR products were designed to include a 50-bp overlap with the digested plasmid to enable their cloning by homologous recombination in yeast after co-transformation. The pGPD-HDHD1-DDAA vector was obtained by PCR amplification (oSL1155/1171) of the HDHD1 DNA sequence obtained by gene synthesis using a specific 5′ primer carrying two D>A mutations and insertion of this insert into the pRS426 vector digested with Eco RI and Bam HI as described above. This construct was verified by sequencing. The DDAA mutant was PCR-amplified (oSL1297/oSL1298) and subcloned at Nde I/Bam HI sites into pET15b-6His-HDHD1, a gift of Dr. E. Van Schaftingen. The WT CYC8 gene and the mutant allele present in mutant #9 were obtained by PCR amplification on the corresponding genomic DNAs with primers (oSL1369/1370) containing a 50-bp overlap with a pRS415 vector (CEN, LEU2). The resulting PCR products were digested with Xba I and Bam HI and co-transformed for cloning by homologous recombination in yeast as described above. The plasmids generated in yeast were rescued by extraction [lithium acetate/SDS method (104)] and electroporation in bacteria, and then amplified and sequenced before being re-transformed in the appropriate strains.

MS and proteomics analyses

Samples used for the proteome-wide analysis of 2DG treatments were prepared from six liquid cultures (WT strain, BY4741) grown overnight at 30°C in 100 ml of rich medium with 2% glucose to mid-log phase. On the next morning, 2DG was added to three of the cultures to a final concentration of 0.2% to obtain triplicates treated with 2DG and triplicates without drug treatment (negative control). After 2.5 hours of incubation at 30°C, the six cultures were centrifuged at 4000g for 5 min at 4°C, resuspended in 500 μl of 10% trichloroacetic acid (TCA; Sigma-Aldrich), and lysed by shaking after addition of glass beads (0.4 to 0.6 mm, Sartorius) for 10 min at 4°C. Cell lysates were retrieved by piercing under the 1.5-ml tubes and brief centrifugation. Precipitated proteins were centrifuged at 16,000g for 10 min at 4°C, supernatants were discarded, and pellets were rinsed four times in 1 ml of 100% cold acetone.

Proteins were then digested overnight at 37°C in 20 μl of 25 mM NH4HCO3 containing sequencing-grade trypsin (12.5 μg/ml; Promega). The resulting peptides were sequentially extracted with 70% acetonitrile and 0.1% formic acid. Digested samples were acidified with 0.1% formic acid. All digests were analyzed by an Orbitrap Fusion equipped with an EASY-Spray nanoelectrospray ion source and coupled to an Easy nano-LC Proxeon 1000 system (all from Thermo Fisher Scientific, San Jose, CA). Chromatographic separation of peptides was performed with the following parameters: Acclaim PepMap100 C18 precolumn [2 cm, 75 μm inner diameter (i.d.), 3 μm, 100 Å], Pepmap-RSLC Proxeon C18 column [50 cm, 75 μm i.d., 2 μm, 100 Å], 300 nl/min flow, using a gradient rising from 95% solvent A (water, 0.1% formic acid) to 40% B (80% acetonitrile, 0.1% formic acid) in 120 min, followed by a column regeneration of 20 min, for a total run of 140 min. Peptides were analyzed in the orbitrap in full-ion scan mode at a resolution of 120,000 [at m/z (mass/charge ratio) 200] and with a mass range of m/z 350 to 1550 and an AGC target of 2 × 105. Fragments were obtained by higher-energy C-trap dissociation activation with a collisional energy of 30% and a quadrupole isolation window of 1.6 Da. MS/MS data were acquired in the linear ion trap in a data-dependent mode, in top-speed mode with a total cycle of 3 s, with a dynamic exclusion of 50 s and an exclusion duration of 60 s. The maximum ion accumulation times were set to 250 ms for MS acquisition and 30 ms for MS/MS acquisition in parallelization mode.

Raw MS data from the Thermo Fisher Orbitrap Fusion were analyzed using the MaxQuant software (108) version 1.5.0.7, which includes the Andromeda peptide search engine (109). Theoretical peptides were created using the S. cerevisiae S288C proteome database obtained from UniProt. Identified spectra were matched to peptides with a main search peptide tolerance of 6 parts per million. After filtering of contaminants and reverse identifications, the total amount of yeast proteins identified among the six samples was equal to 3425. Protein quantifications were performed using MaxLFQ (110) on proteins identified with a minimum amount of two peptides with an FDR threshold of 0.05. LFQ values were then analyzed using Perseus (version 1.5.0.15). For the statistical analysis of yeast proteomes treated with 2DG compared to negative control samples, each group of triplicates was gathered into a statistical group to perform a Student’s t test. Results are presented in the form of Volcano plots (111) and statistically significantly up-regulated and down-regulated candidates were determined by setting an FDR of 0.01 and an S0 of 2.

GO-term analyses of proteomics data

The 79 statistically significantly up-regulated candidates obtained in the proteomics analysis were used as input for the FunSpec web interface (http://funspec.med.utoronto.ca/) with default settings and a P value cutoff of 0.01 to determine the GO biological processes that are enriched in this list of 79 genes compared to the total S. cerevisiae genome annotation (number of total categories, 2062). The complete list of enriched GO biological processes with P < 0.01 and the genes included in each category are shown in Fig. 1B.

Protein extracts and immunoblotting

Yeast cultures used for protein extracts were all grown in SC medium. For each protein sample, 1.4 ml of culture was incubated with 100 μl of 100% TCA for 10 min on ice to precipitate proteins, centrifuged at 16,000g at 4°C for 10 min, and broken for 10 min with glass beads, as described for liquid chromatography–MS/MS sample preparation. Lysates were transferred to another 1.5-ml tube and centrifuged for 5 min at 16,000g at 4°C, supernatants were discarded, and protein pellets were resuspended in 50 μl*(OD600 of the culture) of sample buffer [50 mM tris-HCl (pH 6.8), 100 mM dithiothreitol, 2% SDS, 0.1% bromophenol blue, and 10% glycerol, complemented with 50 mM tris-base (pH 8.8)]. Protein samples were heated at 95°C for 5 min and 10 μl was loaded on SDS–polyacrylamide gel electrophoresis (PAGE) gels (4 to 20% Mini-PROTEAN TGX Stain-Free, Bio-Rad). After electrophoresis, gels were blotted on nitrocellulose membranes for 60 min with a liquid transfer system (Bio-Rad) and membranes were blocked in 2% milk for 20 min and incubated for at least 2 hours with the corresponding primary antibodies. Primary and secondary antibodies used in this study as well as their dilutions are listed in table S4. Membranes were washed three times for 10 min in tris-borate-SDS-Tween 20 0.5% buffer and incubated for at least an hour with the corresponding secondary antibody (coupled with horseradish peroxidase). Luminescence signals were acquired with the LAS-4000 imaging system (Fujifilm). Rsp5 was used as a loading control; alternatively, total proteins were visualized in gels using a trihalo compound incorporated in SDS-PAGE gels (stain-free TGX gels, 4 to 20%; Bio-Rad) after 1-min ultraviolet-induced photoactivation and imaging using a Gel Doc EZ Imager (Bio-Rad).

β-Galactosidase assays

β-Galactosidase assays were performed using 1 ml of mid-log phase yeast cultures carrying the pDOG1-LacZ or pDOG2-LacZ plasmids, grown overnight in SC medium without uracil with 2% glucose and switched to the specified conditions. The OD (600 nm) of the culture was measured, and samples were taken and centrifuged at 16,000g at 4°C for 10 min. Cell pellets were snap-frozen in liquid nitrogen and resuspended in 800 μl of buffer Z (pH 7, 50 mM NaH2PO4, 45 mM Na2HPO4, 10 mM MgSO4, 10 mM KCl, and 38 mM β-mercaptoethanol). After addition of 160 μl of ONPG (ortho-nitrophenyl-β-d-galactopyranoside, 4 mg/ml; Sigma-Aldrich), samples were incubated at 37°C. Enzymatic reactions were stopped in the linear phase (60-min incubation for pDOG2-LacZ and 120-min incubation for the pDOG1-LacZ plasmid, as per initial tests) by addition of 400 μl of Na2CO3, and cell debris were discarded by centrifugation at 16,000g. The absorbance of clarified samples was measured with a spectrophotometer set at 420 nm. β-Galactosidase activities (arbitrary units) were calculated using the formula 1000*[A420/(A600*t)], where A420 refers to the enzyme activity, A600 is the turbidity of the culture, and t is the incubation time. Each enzymatic assay was repeated independently at least three times.

Drop tests

Yeast cells grown in liquid rich or SC medium for at least 6 hours at 30°C were adjusted to an optical density (600 nm) of 0.5. Serial 10-fold dilutions were prepared in 96-well plates and, using a pin replicator, drops were spotted on plates containing rich or SC medium containing 2% (w/v) agar and, when indicated, 2DG [0.05 or 0.2% (w/v)], sodium selenite (200 μM), or tunicamycin (1 μg/ml). Mannose plates (Figs. 2I and 3F) were prepared as regular SC plates containing 2% mannose (w/v) in addition to glucose. Plates were incubated at 30°C for 3 to 4 days before scanning. For the adhesion test (Fig. 5D), the plates were scanned and the colonies were then washed with tap water under a constant flow of water for 1 min as previously described (78). Excess water was removed before the plates were scanned again.

Isolation of spontaneous mutants and characterization

WT cells transformed with a pDOG2-LacZ plasmid (pSL410) were grown overnight in SC-Ura medium and ca. 2 × 106 cells were spread on SC-Ura plates containing 0.2% 2DG and grown for 6 days at 30°C. The clones obtained (24 clones) were restreaked on SC-Ura to isolate single clones. Resistance to 2DG was confirmed by drop tests (fig. S6A). β-Galactosidase enzyme assays and total protein extracts were performed on cultures grown to the exponential phase. For β-galactosidase assays, the results were statistically tested using an unpaired t test with equal variance, assuming a normal distribution of the values.

Diploids were obtained by crossing each resistant mutant with a WT or hxk2∆ strain of the opposite mating type (BY4742, Matα) and selecting single diploid clones on selective medium (SC-Met-Lys). Sequencing of the REG1 and HXK2 loci was done after PCR amplification on genomic DNA isolated from the corresponding clones. For whole-genome sequencing, genomic DNA of the WT, clone 9, and clone 10 was purified using the Qiagen genomic DNA kit (Genomic-tip 20/G) using 30 OD equivalents of material following the manufacturer’s instructions after zymolyase treatment (Seikagaku). A PCR-free library was generated from 10 μg of gDNA and sequenced at the Beijing Genomics Institute (Hong Kong) on Illumina HiSeq 4000. The mutations in each clone were identified through comparative analysis of the variants detected by mapping their reads to the reference genome (BY4741) (112, 113) and those detected by mapping the WT reads to the reference. The differential variants were filtered by quality (vcf QUAL > 1000) and manually inspected through Integrative Genomics Viewer (IGV) for validation (114).

Cell culture and transfection

HeLa cells were maintained at 37°C and 5% CO2 in a humidified incubator and grown in Dulbecco’s modified Eagle’s medium, supplemented with 10% fetal calf serum. Cells were regularly split using Trypsin-EDTA to maintain exponential growth. HeLa cells were transfected with plasmid pCMV-Sport6-HDHD1 and pCS2 (empty control) using Lipofectamine 2000 according to the manufacturer’s instructions. All culture media reagents were from Thermo Fisher Scientific. 2DG (Sigma-Aldrich) was used at a final concentration of 5 mM, and tunicamycin (from Streptomyces sp.; Sigma-Aldrich) was used at a final concentration of 5 μg/ml. For 2DG resistance assays, cells were grown in a 10-cm2 flask and split in a 24-well plate in the absence or presence of 5 mM 2DG. Cells were counted each day with a hemocytometer after trypsinization and labeling with trypan blue. Total extracts were prepared by incubating cells (10 cm2) on ice for 20 min with 400 μl of TSE Triton buffer [50 mM tris-HCl (pH 8.0), 150 mM NaCl, 0.5 mM EDTA, and 1% Triton X-100] containing protease inhibitors (cOmplete protease inhibitor cocktail, EDTA-Free, Roche Diagnostics). Cells were then lysed mechanically with scrapers, and the lysate was centrifuged at 13,000g at 4°C for 30 min. Proteins were assayed in the supernatant using the Bio-Rad Protein Assay reagent (Bio-Rad) and 40 μg of proteins was loaded on SDS-PAGE gels.

Recombinant His-tagged HDHD1 and HDHD1-DDAA protein purifications

E. coli BL21 bacteria were transformed with plasmids allowing the expression of His-tagged HDHD1 or HDHD1-DDAA. A 100-ml preculture was grown overnight in LB + ampicillin (100 μg/ml), diluted 50-fold into 1 liter of culture. The OD reached 0.7 to 0.9. Isopropyl-β-D-thiogalactopyranoside (1 mM) was then added to induce the recombinant protein and cells were further grown at 25°C for 3 hours. Cells were harvested, and the pellet was frozen in liquid N2 and thawed on ice. The pellet was resuspended in 20 ml of lysis buffer [25 mM Hepes (pH 6.7), 300 mM NaCl, 15 mM imidazole, 2 mM β-mercaptoethanol, 10% (v/v) glycerol, and protease inhibitor cocktail (cOmplete protease inhibitor cocktail, EDTA-Free, Roche Diagnostics)]. Cells were then sonicated and Triton X-100 was added to a final concentration of 1%. The lysate was centrifuged at 12,000 rpm in an SW-32 rotor (Beckman Coulter) for 15 min at 4°C and then the supernatant was further centrifuged at 35,000 rpm for 1 hour at 4°C. The supernatant was incubated with 800 μl of Ni-NTA bead slurry (Qiagen) and rotated overnight at 4°C. The beads were collected by centrifugation (1000g, 2 min, 4°C), resuspended in lysis buffer, and washed with 50 ml of lysis buffer at 4°C, and then washed again with 50 ml of thrombin cleavage buffer (50 mM Hepes, 5 mM CaCl2, 100 mM NaCl, and 10% glycerol) at 4°C. The His-tag was removed by cleavage with 16 U of thrombin (27-0846-01; Sigma-Aldrich) added directly onto the beads for 2 hours at 25°C. The eluate was then collected and incubated with 500 μl of benzamidine-Sepharose 6B (GE Healthcare) at room temperature for 30 min to remove thrombin. The supernatant was collected and protein content was assayed by SDS-PAGE and colloidal blue staining (Brilliant Blue G-colloidal, Sigma-Aldrich), and protein concentration was assayed by the Bradford method (Bio-Rad protein assay, Bio-Rad).

Enzyme assays

2DG6P phosphatase assays were performed in 250 μl of reaction containing 1.5 mM 2DG6P (#17149, Cayman Chemical, Ann Arbor, Michigan, USA) in 50 mM Hepes (pH 6.7), 10 mM MgCl2, and 10% glycerol and 30 μg of recombinant HDHD1 or HDHD1-DDAA. Samples were incubated at 37°C for various times (0, 5, 10, and 15 min) and the reaction was stopped by adding 150 μl of EDTA (0.5 M). Then, the 2DG generated was assayed by adding 500 μl of glucose assay reagent (GAGO20, Sigma-Aldrich) and further incubating at 37°C for 30 min. The reaction was stopped by adding 500 μl of H2SO4 (12N), and the absorbance of the reaction was measured at 540 nm. A slope (A540 over time) was calculated to assess enzyme activity and to make sure that the reaction was in the linear range. The measurements were repeated three times.

Statistical analysis

Mean values were calculated using a minimum of three independent measurements from three biological replicates and are plotted with error bars representing SEM. Statistical significance was determined using a t test for paired variables assuming a normal distribution of the values, as follows: *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001; ns, P > 0.05.

SUPPLEMENTARY MATERIALS

stke.sciencemag.org/cgi/content/full/12/597/eaaw8000/DC1

Fig. S1. Characterization of other candidates showing increased abundance after 2DG treatment.

Fig. S2. Hog1 signaling responds to 2DG, but DOG1 expression is not regulated by Hog1.

Fig. S3. DOG1 expression is regulated by the UPR pathway.

Fig. S4. Slt2 participates in the regulation of DOG1 expression.

Fig. S5. Cis regions involved in the regulation of the DOG2 promoter by glucose.

Fig. S6. Identification of reg1 and hxk2 mutants within the isolated spontaneous 2DG-resistant mutants.

Fig. S7. Multiple protein sequence alignment of Hxk2 orthologs and positions of the mutations identified.

Fig. S8. HDHD1 but not its close homologs HDHD4 or PSPH allow resistance to 2DG.

Table S1. Single-nucleotide variants in clones #9 and #10 as compared to the WT strain, as identified by whole-genome resequencing.

Table S2. Yeast strains used in this study.

Table S3. Plasmids used in this study.

Table S4. Antibodies used in this study.

Data file S1. Proteomic response to 2DG treatment.

References (117121)

REFERENCES AND NOTES

Acknowledgments: We thank N. Joly (Institut Jacques Monod, Paris, France) for discussions about the HDHD1 in vitro assay; G. Lelandais (I2BC, Gif-sur-Yvette, France) for help with the use of Yeastract analysis and advice on statistical analysis; P. Sanz (Instituto de Biomedicina de Valencia, Valencia, Spain) and J. A. Prieto (Instituto de Agroquímica y Tecnología de Alimentos, Valencia, Spain) for the gift of the GST-Dog2 construct; E. Van Schaftingen (de Duve Institute, Université Catholique de Louvain, Louvain-la-Neuve, Belgium) for the gift of the His-tagged HDHD1 construct; P. Walter (UCSF, San Francisco, CA, USA) for the gift of the pUPRE1:lacZ reporter; D. Levin (Boston University, Boston, MA, USA) for the gift of the pCYC(2xRlm1):lacZ reporter; C. Stirling (University of Manchester, Manchester, UK) for the gift of the anti-invertase antibody; and M. Schmidt (University of Pittsburgh, Pittsburgh, PA, USA) for advice regarding gDNA extraction for sequencing. We also thank T. Léger and the Proteomics facility of the Institut Jacques Monod [supported by the Region Ile-de-France (SESAME), the Paris-Diderot University (ARS), and CNRS] for assistance. We thank A. Babour (IUH, Hôpital Saint Louis, Paris, France), A. ČopiČ (Institut Jacques Monod, Paris, France), M. Ruault (Institut Curie, Paris, France), E. Chevet (CLCC Eugène Marquis, Rennes, France), and members of the Léon and Chevet labs for insightful comments and critical reading. Funding: This work was supported by fellowships from the Fondation pour la Recherche Médicale (SPF20150934065 to Q.D.) and the Ligue contre le cancer (TAZK20115 to C.L.), and by grants from the Agence Nationale de la Recherche (P-Nut, ANR-16-CE13-0002 to S.L.) and the Fondation ARC pour la recherche sur le cancer (PJA20181208080 to S.L.). Author contributions: Q.D., A.V., C.L., and S.L. contributed reagents, performed experiments, and acquired and analyzed the data. A.F. and J.S. analyzed the genome sequencing data. S.L. and Q.D. wrote the manuscript. S.L. directed the work. Competing interests: The authors declare that they have no competing interests. Data and materials availability: The MS proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE (115) partner repository with the dataset identifier PXD014373. All other data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials.
View Abstract

Stay Connected to Science Signaling

Navigate This Article