Linking Cancer Metabolism to DNA Repair and Accelerated Senescence Free

Otto Warburg was first to describe a diminished Pasteur effect in tumors, which actively take up and ferment glucose even in the presence of oxygen (1). Prior models ascribed deregulated glucose fermentation in cancer cells primarily to compensation for mitochondrial defects or adaptation to tumor hypoxia (2), but the focus has shifted to roles for metabolic products beyond ATP. Indeed, aerobic glycolysis promotes accumulation of intermediates that can serve as precursors for the proteins, lipids, and nucleic acids needed to support rapid cancer cell growth (3, 4). In turn, glutamine “addiction” in cancer, first observed by Eagle as an elevated requirement for cells in culture (5), has similarly been ascribed to answering increased demand for building blocks for cell proliferation (6–8). Beyond biosynthesis, recent attention has focused on potential regulatory functions for metabolic intermediates produced by glycolysis and/or glutaminolysis, via their roles as co-factors and inhibitors of chromatin-modifying enzymes (9–12). Relevant chromatin-modifying enzyme/coenzyme pairs include histone acetyltransferase (HAT) and acetyl-CoA, PARP and NAD⁺, histone lysine methyltransferases (HMT) and S-adenosyl methionine, Jumonji-domain containing histone lysine demethylases (JmjC HDM) and α-ketoglutarate (α-KG), O-linked N-acetylglucosamine (O-GlcNAc) transferase (OGT) and GlcNAc, and, of course, Ser/Thr and Tyr protein kinases and ATP. These considerations have raised the hypothesis that via its epigenetic effects, cancer metabolic reprogramming may influence gene expression to drive oncogenesis and maintain cancer cell identity. For example, glycolytic metabolism in cancer cells impacts global chromatin structure by modulating histone acetylation (13), potentially altering transcription but also impinging on DNA repair. Indeed, along with their well-studied roles in epigenetic regulation of transcription, HATs, PARPs, and HMTs are also key regulators of DNA damage response (DDR; ref. 14), suggesting a mechanism by which cancer metabolism might directly influence genomic instability and resistance to genotoxic stress.

In particular, specific patterns of histone modification are associated with ionizing radiation induced foci (IRIF), the multikilobase chromatin domains that form rapidly at sites of chromosomal double-strand breaks (DSB) and mark eroded telomeres (15–18). Although DSBs are difficult to visualize in intact nuclei, IRIF are easily detected and can serve as a proxy for DNA damage (19). To probe the interaction of cancer metabolism and DNA repair, we used small molecule inhibitors, cell-permeable metabolic intermediates, and RNAi to perturb metabolic pathways in MCF7 breast adenocarcinoma cells. We observed that inhibition of glycolysis before irradiation allowed IRIF to form but blocked their timely resolution. Detecting residual DNA breaks by comet assays confirmed a defect in DSB repair. In the face of persistent damage, rather than undergoing apoptosis, many cells entered accelerated senescence. Additional chemical probes pointed to two pathways downstream of glycolysis, the hexosamine biosynthetic pathway (HBP) and tricarboxylic acid (TCA) cycle, mediating opposing effects on IRIF persistence, DSB repair, and cell senescence. Finally, we were able to implicate Polycomb Repressive Complex (PRC) 1 and 2 as the ultimate targets of cancer metabolic reprogramming in DSB repair. Taken together, these findings reveal critical connections between cancer cell metabolism, DSB repair, and senescence with implications for genomic instability, carcinogenesis, and therapeutic resistance.

The MCF7 Tet-On Advanced cell line was obtained from Clontech. The generation and characterization of the MCF7^GFP-IBD cell line has been described (20) and was used with further authentication by IDEXX BioResearch within the last 6 months. Panc 02^GFP-IBD, U-87 MG^GFP-IBD, and hTERT-HME1^GFP-IBD cell lines were developed similarly from parent cell lines purchased from the ATCC. Briefly, GFP-IRIF binding domain (IBD) cloned into the pLVX-Tight-Puro vector was transduced along with pLVX-Tet-On Advanced vector (Clontech) into each cell line. Following G418 and puromycin selection, cells were induced with 1 μg/mL doxycycline and sorted to establish the IRIF reporter cell lines. The Panc 02^GFP-IBD and U-87 MG^GFP-IBD cell lines were maintained in RPMI (Invitrogen), supplemented with 10% Tet system approved FBS (Clontech). The hTERT-HME1^GFP-IBD cell line was maintained in mammary epithelial cell basal medium (MEBM) supplemented with mammary epithelial cell growth medium (MEGM SingleQuot, Lonza). For studies requiring glucose and glutamine limitation, media was prepared using DMEM base, D-glucose (Sigma), and L-glutamine solutions (Gemini Bioproducts) at appropriate concentrations with 10% Tet system approved FBS.

MCF7^GFP-IBD cells were seeded at 2.5 × 10⁴ per well in 24-well plates on cover glass (Fisher Scientific) in 4.5 g/L glucose media. GFP-IBD expression was induced with 1 μg/mL doxycycline (Sigma) for 48 hours. For glucose- and glutamine-limiting conditions, media was exchanged 1 day after plating. Cells were incubated with small molecule inhibitors or cell-permeable metabolites (Supplementary Table S1) for 1 hour prior to 6 Gy irradiation by a GammaCell (MDS Nordion) ⁶⁰Co source unless otherwise noted and IRIF persistence was evaluated at 24 hours. Control, nonirradiated cells treated with each inhibitor or metabolite were examined at 24 hours to confirm lack of toxicity and no increase in IRIF formation. In turn, cells treated with inhibitors or metabolites were examined at 2 hours after irradiation to detect any suppression of IRIF formation.

For imaging, cells were fixed with 2% paraformaldehyde in PBS for 5 minutes, followed by two washes with PBS. Slides were mounted with ProLong Gold (Invitrogen) after staining with 5 μg/mL Hoechst 33342 (Sigma) or mounted with SlowFade Gold anti-fade reagent with DAPI (Invitrogen). Images were captured on a Zeiss Axiovert 40 CFL microscope with a 40× Plan-Neofluar objective and Axiocam digital camera controlled by AxioVision 4.8 software and pseudo-colored in Adobe Photoshop or ImageJ (http://imagej.nih.gov/ij/). Numbers of foci per nucleus were determined using ImageJ, and means ± SEM were plotted. Statistical significance of IRIF phenotypes was determined by two-tailed, unpaired t test with Welch correction using GraphPad Prism 6 software. P values of ≤0.05 are considered to be statistically significant [***, P ≤ 0.001; **, P ≤ 0.01; *, P ≤ 0.05. P > 0.05 is not significant (n.s.)].

Sets of three validated gene-specific Trilencer-27 siRNA duplexes targeting expression of OGT and OGA (MGEA5) and the Trilencer-27 Universal scrambled negative control siRNA duplex were obtained from OriGene Technologies.

The siRNA sequences used in this study were:

• OGT(a) - ACUACUCAGAUCAACAAUAAGGCTG;

• OGT(b) - CCUACUCUAAUAUGGGAAACACUCT;

• OGT(c) - GGCACAUCGAGAAUAUCAGGCAGGA;

• MGEA5(a) - CCUCUAGAAUGGUAACAAAUCAGCC;

• MGEA5(b) - GCACGAGAAUAUGAGAUAGAGUUCA;

• MGEA5(c) - CGAGCAAAUAGUAGUGUUGUCAGTG.

For siRNA analysis, MCF7^GFP-IBD cells were seeded in 6-well plates to achieve 60% to 80% confluence after 24 hours. Transfections of the individual duplex siRNAs, mixtures of three gene-specific duplex siRNAs or the scrambled control were performed using FuGENE HD (Promega) according to the manufacturer's instruction. After 24 hours, transfected cells were seeded in 24-well plates with cover glasses in high or low glucose media with 1 μg/mL doxycycline with or without PUGNAc. After overnight incubation, cells were irradiated with 6 Gy. After 24 hours, cells were fixed and GFP-IBD foci were imaged. OGT protein expression and O-GlcNAc protein modification were analyzed by Western blot analysis 48 hours after siRNA transfection. OGT antibody from Thermo Fisher Scientific was used for Western blot analysis of OGT protein expression and anti-O-linked N-acetylglucosamine antibody from Abcam for detection of O-GlcNAc protein modification.

MCF7^GFP-IBD cells were seeded at 3 × 10⁴ per 35 mm Fluorodish (World Precision Instruments) in 4.5 g/L glucose media. After 1 day, cells were treated with small molecule agents for 1 hour prior to irradiation at 6 Gy. To monitor effects of glucose or glutamine limitation on senescence induction, growth media was changed to the appropriate media 1 day after seeding and cells cultured overnight before further treatment. The SA-β-Gal assay was performed as described (21), fixing cells 5 days after irradiation. Images were captured on a Zeiss Axiovert 200M microscope with 20× Plan-NeoFluar objective and Axiocam digital camera controlled by OpenLab software. Images were corrected for white balance using an ImageJ macro (http://digital.bsd.uchicago.edu/%5Cimagej_macros.htmL).

To estimate the level of SA-β-Gal staining for each condition, SA-β-Gal–positive and SA-β-Gal–negative cells were counted in multiple fields, the percentage of SA-β-Gal–positive determined, and these values were averaged. The percentage of positive staining is indicated in each SA-β-Gal image as mean ± SEM.

MCF7^GFP-IBD cells were seeded at 1 × 10⁶ per 10 cm dish. 2-DG, 2-FDG, and mannose were added to the media 1 hour before irradiation. Cells were harvested the next day, and lysed in lysis buffer (50 mmol/L Tris-HCl pH 7.5, 150 mmol/L NaCl, 1% NP-40, 1 mmol/L EGTA, 0.05% SDS) with HALT protease and phosphatase inhibitor cocktail (Thermo Scientific). A total of 10 μg of total cell lysates were loaded per lane on NuPage Bis-Tris precast gels (Invitrogen), transferred to a 0.45-μm nitrocellulose membrane (Bio-Rad), and probed with an anti-BiP antibody (C50B12, Cell Signaling). For the detection of poly(ADP-ribose) (PAR) chains, cells were treated with 50 mmol/L 2-DG and/or 10 μmol/L PARP inhibitor veliparib (ChemieTek) 1 hour before the induction of DNA damage using 1 mmol/L H₂O₂ for 10 minutes. A total of 20 μg of total cell lysate was loaded per lane, and analyzed by Western blotting using an anti-PAR antibody (10H, GeneTex).

Neutral comet assays were performed according to the manufacturer's protocol (CometAssay, Trevigen). Briefly, MCF7^GFP-IBD cells were seeded at 2 × 10⁵ per well on 6-well plates, treated as indicated for IRIF imaging and harvested at 24 hours by trypsinization, washed once with PBS, and re-suspended at 1 × 10⁵ cells/mL. Cells were then mixed with Comet LM agarose at 1:10, and 5,000 cells were spotted per area on CometSlides. After incubation in lysis solution for 1 hour at 4°C, electrophoresis was performed as recommended. Slides were fixed with 70% methanol for 30 minutes at room temperature, dried, and stained with SYBR green for imaging using a Zeiss Axiovert 40 CFL with 10× Plan-NeoFluar objective, and an Axiocam digital camera controlled by AxioVision 4.8 software. Images were analyzed using an Image J comet assay macro (http://www.med.unc.edu/microscopy/resources/imagej-plugins-and-macros/comet-assay), and pseudo-colored in Adobe Photoshop.

Ester-protected analogues of (R)-2-HG and (S)-2-HG were synthesized as previously described (22). (R) or (S)-5-oxotetrahydrofuran-2-carboxylic acid (650 mg, 5.0 mmol/L) was dissolved in acetonitrile (15 mL), followed by addition of i-Pr2NEt (1.05 mL, 6.0 mmol/L, 1.2 equiv) and 3-(trifluoromethyl)benzyl bromide (0.92 mL, 6.0 mmol/L, 1.2 equiv). The mixture was refluxed for 15 minutes, allowed to cool to room temperature and then stirred overnight. The solvent was removed under vacuum and the resulting white residue was redissolved in ethyl acetate (50 mL). The organic layer was washed with 10% HCl (50 mL), 10% sat NaHCO₃ (50 mL), and brine (50 mL), and dried with Na₂SO₄. Rotary evaporation yielded a yellow oil, which was purified using flash chromatography, eluting with 1:1 hexanes:ethyl acetate to give the pure compound as an oil [(R)-2-HG: 1.26 g, 89%; or (S)-2-HG: 1.08 g, 75%], which solidified to a white solid after co-evaporation with ether.

¹H NMR (500 MHz, CDCl₃) δ 7.65–7.56 (m, 2H), 7.56–7.51 (m, 1H), 7.48 (t, J = 7.9 Hz, 1H), 5.24 (s, 2H), 5.00–4.94 (m, 1H), 2.61–2.47 (m, 3H), 2.33–2.23 (m, 1H). ¹³C NMR (126 MHz, CDCl₃) δ 175.83, 169.59, 135.77, 131.67, 129.35, 125.59, 125.56, 125.53, 125.50, 125.06, 125.03, 125.00, 124.97, 75.56, 66.58, 26.65, 25.75.

The constitutively nuclear-localized DNA damage and repair signaling adapter protein 53BP1 is recruited to IRIF within minutes after DNA damage and then disperses as DSBs are repaired (23). We have exploited GFP fused to the 53BP1 minimal IRIF binding domain (IBD), which consists of a nuclear localization domain, dimerization domain, paired Tudor domains, and an ubiquitination-dependent recruitment motif (24–26), as a live-cell reporter for DNA DSB formation and repair in MCF7 human breast cancer cells (MCF7^GFP-IBD; ref. 20). Upon irradiation of MCF7^GFP-IBD cells with 6 Gy, GFP-IBD relocalizes to form several dozen IRIF. In response to the DNA damage signal, cells arrest proliferation, primarily in G₁. A phase of rapid repair ensues during the first 2 hours and most remaining IRIF resolve over 48 hours. Although most surviving cells return to proliferation, a fraction fail to complete repair and remain arrested. Over several days, these cells eventually adopt features characteristic of senescent cells including a flat morphology, altered ploidy, and expression of senescence-associated β-galactosidase (SA-β-Gal). When MCF7^GFP-IBD cells are treated with higher radiation doses and/or radiation sensitizers, such as the PARP inhibitor veliparib, IRIF persistence and the proportion of senescent cells each increase (20, 27). Consistent with the known role of PARP in nonhomologous end-joining (NHEJ) DSB repair (28), treating MCF7^GFP-IBD cells with 10 μmol/L veliparib for 1 hour prior to 6 Gy irradiation significantly increased both IRIF persistence (P ≤ 0.001; Supplementary Fig. S1A and S1B) and DNA fragmentation measured by neutral comet assays (P ≤ 0.001) at 24 hours (Supplementary Fig. S1C and S1D).

In searching for other agents that might similarly impair the DDR, we observed that treating MCF7^GFP-IBD cells with the novel glucose transporter (GLUT1) inhibitors Compounds 11 and 12 (ref. 29; see Supplementary Table S1 for structures and references for all small-molecule probes) prior to irradiation induced both significant IRIF persistence (P ≤ 0.001) and increased senescence at 5 days (Fig. 1A and Supplementary Fig. S2A). Suggesting a role for glucose uptake in DNA repair, limiting glucose transport with the conventional inhibitors 2-deoxy-D-glucose (2-DG) and phloretin as well as simply lowering media glucose from 4.5 g/L to 1 g/L or removing glucose altogether each recapitulated this effect (Fig. 1A–C and Supplementary Fig. S2A and S2B). In turn, conventional small-molecule probes of glycolysis, 3-bromopyruvate (3-BP) to inhibit hexokinase, 3-(3-pyridinyl)-1-(4-pyridinyl)-2-propen-1-one (3-PO) to target phosphofructokinase-2 (PFK-2), alizarin red S to block phosphoglycerate mutase 1 (PGAM1), or oxamate to inhibit lactate dehydrogenase (Fig. 1D), each promoted IRIF persistence and senescence (P ≤ 0.001, Fig. 1E and Supplementary Fig. S2C), establishing a requirement for glucose metabolism in IRIF resolution. Consistent with effects observed with MCF7^GFP-IBD cells, the glucose uptake inhibitor phloretin and hexokinase inhibitor 3-BP similarly promoted IRIF persistence in other cell lines. When either the mouse pancreatic cell line Panc 02^GFP-IBD or human glioma cell line U-87 MG^GFP-IBD were irradiated with 6 Gy in 4.5 g/L glucose media, both resolved IRIF by 24 hours. Although treatment with phloretin or 3-BP prior to irradiation did not induce IRIF on its own, blocking glycolysis delayed IRIF resolution at 24 hours after irradiation in both cell lines (Supplementary Fig. S2E and S2F).

In some cells, glycolysis is critical for maintaining ATP levels, raising the concern that glycolysis inhibitors might delay DNA repair via ATP depletion. However, that MCF7 cells display active respiration and avid utilization of alternate fuels (30) suggests they can compensate for decreased glycolytic flux. Alternatively, glucose fermentation might augment NAD⁺ pools, supporting PARP activity to promote NHEJ repair. However, neither media glucose limitation nor treatment with 2-DG blocked PARylation upon H₂O₂ treatment (Supplementary Fig. S3A–S3D). To explain his observation that elevating glycolysis blocked senescence (31), Kondoh invoked the Crabtree effect, where increased fermentation suppresses respiration. He proposed that the resulting decrease in mitochondrial reactive oxygen species (ROS) would protect cells from DNA damage. To test this hypothesis, cells were irradiated in 1 g/L glucose in the presence of antioxidants N-acetyl-L-cysteine, butylated hydroxyanisole, or EUK134. The antioxidants failed to restore IRIF kinetics (Supplementary Fig. S4A and S4B), suggesting a direct role for glucose metabolism in DNA repair.

As a potential confounding factor linking glucose limitation to IRIF persistence, we considered the potential impact of inhibition of secretory pathway N-glycosylation and resulting activation of the unfolded protein response (UPR; ref. 32). Much like 1 g/L glucose media, both the N-glycosylation inhibitor tunicamycin (TM) and SERCA inhibitor thapsigargin (THA) induced IRIF persistence (Fig. 2A and B). Arguing against a primary role for UPR in IRIF kinetics, 1 g/L glucose media failed to induce upregulation of the UPR marker GRP78/BiP (ref. 33; Fig. 2C). To dissect this further, we examined suppression of UPR by mannose in cells treated with 2-DG versus the negative control 2-FDG (ref. 34; Fig. 2D). Even though it blocked GRP78/BiP induction in 2-DG–treated cells (Fig. 2E), mannose failed to restore IRIF resolution (Fig. 2F and G), unlinking the UPR and DSB repair pathways.

In glucose-depleted cells, the core glycosylation subunit N-acetylglucosamine (GlcNAc) can enter the HBP to restore levels of UDP-GlcNAc (Fig. 3A) but fails to repopulate glycolysis, the pentose phosphate pathway, or TCA cycle (35). Beyond secretory pathway glycosylation, UDP-GlcNAc serves as a co-factor for nucleocytoplasmic OGT in O-linked protein GlcNAcylation (refs. 36, 37; Fig. 3A). Suggesting a role for the HBP in DNA repair, the addition of GlcNAc to 1 g/L glucose media restored IRIF resolution and shortened comet tails in irradiated MCF7^GFP-IBD cells (Fig. 3B and C). Confirming a role for the HBP and OGT in DNA repair, inhibiting the HBP rate-limiting enzyme glutamine fructose-6-phosphate amidotransferase with azaserine or directly blocking OGT with either alloxan, BADGP, or ST060226 each significantly induced IRIF persistence in 4.5 g/L glucose (P ≤ 0.001, Fig. 3D and E). In turn, blocking the deglycosylating enzyme O-GlcNAcase (OGA) with PUGNAc restored IRIF resolution in 1 g/L glucose (P ≤ 0.01, Fig. 3D and E, Supplementary Fig. S5B). Consistent with the IRIF kinetics, neutral comet assays demonstrated that GlcNAc and PUGNAc restored DSB repair in 1 g/L glucose (P ≤ 0.001), although alloxan had an opposing effect (P ≤ 0.01, Fig. 3F and G).

To confirm links between the HBP and DNA repair in a nontransformed cell line, we used hTERT-HME1^GFP-IBD, a human telomerase reverse transcriptase immortalized mammary epithelial cell line expressing the IRIF reporter. Like MCF7^GFP-IBD, hTERT-HME1^GFP-IBD displayed greater IRIF persistence at 24 hours after irradiation when treated with glucose uptake inhibitor Compound 11 or OGT inhibitor ST060266 and faster IRIF resolution when treated with GlcNac or the OGA inhibitor PUGNAc (Supplementary Fig. S5C).

As a complementary strategy to analysis with chemical probes, we examined targeting O-GlcNAcylation via RNAi to knock down the expression of OGT or OGA in MCF7^GFP-IBD cells. When cells were treated with 27-mer duplex siRNAs targeting OGT to impair nucleocytoplasmic O-GlcNAcylation prior to irradiation, IRIF persistence at 24 hours was increased in both 4.5 and 1 g/L glucose media, recapitulating the effects of OGT inhibitors alloxan, BADGP, and ST060226 (P ≤ 0.01, Fig. 3H and I and Supplementary Fig. S5D–S5G). In turn, duplex siRNA targeting OGA enhanced IRIF resolution in 1 g/L glucose media, much like the OGA inhibitor PUGNAc. Further confirming the analogy between the chemical probes and siRNA knockdown, the OGA inhibitor PUGNAc failed to restore IRIF resolution in cells treated with siRNA to silence OGT.

Given the requirement for glutamine in the conversion of fructose-6-P to glucosamine-6-P to enter the HBP (Fig. 3A), we expected that glutamine metabolism might also impinge on IRIF kinetics. Indeed, restricting media glutamine enhanced IRIF persistence in 4.5 g/L glucose media, but paradoxically restored IRIF resolution in 1 g/L glucose (P ≤ 0.001, Fig. 4A and Supplementary Fig. S6A). Glutamine also enters the TCA cycle via conversion to glutamate and α-ketoglutarate (α-KG), joining the glycolytic product pyruvate in formation of acetyl-CoA (Fig. 4B). Downregulation and/or mutation of TCA cycle enzymes are observed in cancer (38) and activation of the TCA cycle may promote senescence (39). Toward establishing links between the TCA cycle and DSB repair, we augmented TCA cycle activity via the addition of cell permeable pyruvate or inhibition of pyruvate dehydrogenase kinase with dichloroacetate. Both perturbations resulted in IRIF persistence (P ≤ 0.01 and ≤ 0.001, respectively, Fig. 4C and Supplementary Fig. S6B). Similarly, the cell-permeable TCA intermediates oxaloacetate and malate each induced IRIF persistence (P ≤ 0.001, Fig. 4C and Supplementary Fig. S6B). Dimethyl 2-oxoglutarate (DM-α-KG), a cell-permeable analogue of α-KG, induced IRIF persistence in both 4.5 and 1 g/L glucose media (P ≤ 0.001, Fig. 4D and Supplementary Fig. S6A). In turn, blocking glutamine-dependent anaplerosis with the glutaminase inhibitor Compound 968 or glutamate dehydrogenase inhibitor epigallocatechin gallate restored normal IRIF kinetics in 1 g/L glucose media (P ≤ 0.001, Fig. 4E and Supplementary Fig. S6A). Comet assays confirmed that IRIF persistence after treatment with DM-α-KG and Compound 968 reflected inhibition of DSB repair (P ≤ 0.001, Fig. 4F and G).

By virtue of their catalytic requirement for molecular oxygen and α-KG as co-factors, dioxygenases such as the hypoxia-inducible factor (HIF)-1α prolyl hydroxylases, TET DNA demethylases, and Jumonji domain-containing (JMJD) HDM are considered key epigenetic enzymes linking metabolism to gene expression (refs. 10, 40, 41; Fig. 5A). Extending this paradigm to metabolic control of DDR, a 1 hour pretreatment with the dioxygenase inhibitor IOX1 prior to irradiation in 1 g/L glucose media restored IRIF resolution (P ≤ 0.001, Fig. 5B and Supplementary Fig. S7A). Similarly, the oncometabolite and α-KG antagonist 2-hydroxyglutarate (2-HG; refs. 42, 43) also restored IRIF resolution (P ≤ 0.001, Fig. 5B and Supplementary Fig. S7A). In turn, IOX1 and 2-HG partly blocked induction of senescence after irradiation in 1 g/L glucose (Fig. 5C).

Toward determining the presumptive enzyme target of α-KG, we found that the S enantiomer of 2-HG that can inhibit HIF prolyl hydroxylases and the R enantiomer that cannot confer similar effects on IRIF resolution (refs., 22, 43; P ≤ 0.001, Fig. 5B and Supplementary Fig. S7A), suggesting a minor role if any for HIF1α. Although TET-dependent DNA demethylation has been linked to excision repair, a role for TET proteins in DSB repair remains to be established. By contrast, activation of JMJD proteins to demethylate histones H3 at K4, K9, K36, K79, and/or H4 at K20 could antagonize well-known protein–protein interactions critical to IRIF function and DSB repair (17). Indeed, treating cells with either PBIT to inhibit H3 K4 demethylase Jarid1A/B or with GSKJ4 to block H3 K27 demethylase JMJD3 each restored IRIF resolution after irradiation in 1 g/L glucose (P ≤ 0.01 and ≤ 0.001, respectively, Fig. 5B and Supplementary Fig. S7A), decreased senescence (Fig. 5C), and accelerated DNA repair as measured by comet assay (P ≤ 0.001, Supplementary Fig. S7B). In turn, R-2-HG, IOX1 and GSKJ4 each blocked the effect of DM-α-KG, restoring IRIF resolution and DNA repair (P ≤ 0.001, Fig. 5D and E and Supplementary Fig. S7B).

Toward determining the order of function between O-GlcNAcylation and histone demethylase activity in IRIF resolution, we examined relevant combinations of cell-permeable metabolites and specific inhibitors. Both GlcNAc and PUGNAc restored IRIF resolution and DNA repair in cells treated with DM-α-KG, suggesting that O-GlcNAcylation might promote methylation or protect against demethylation (P ≤ 0.001, Fig. 5D and E and Supplementary Fig. S7C and S7D). In a reciprocal experiment, the demethylase inhibitors IOX1 and R-2-HG overcame the effects on IRIF resolution of OGT inhibitors alloxan and ST060266 (P ≤ 0.001, Fig. 5F and Supplementary Fig. S7E). A conservative interpretation places α-KG-dependent lysine demethylation downstream of O-GlcNAcylation in DNA repair and IRIF resolution.

Confirming a role for protein methylation in DSB repair, blocking the H3K9 methyltransferase G9a with BRD4770 or the H3K27 methyltransferase EZH2 with GSK126 each promoted IRIF persistence and delayed repair in both 4.5 and 1 g/L glucose (P ≤ 0.001, Fig. 6A–D). We placed O-GlcNAcylation upstream of histone methylation, insofar as neither GlcNAc nor PUGNAc could overcome inhibition of IRIF resolution after treatment with methyltransferase G9a inhibitor BRD4770 (Supplementary Fig. S7F and S7G).

We were particularly struck by the IRIF persistence observed upon treatment with the H3K27 methyltransferase inhibitor GSK126. Its target, the SET domain protein EZH2, serves as a catalytic subunit for PRC2. PRC2-dependent H3K27 trimethylation marks genes for epigenetic silencing, partly via recruitment and activation of the PRC1 histone H2A K119 ubiquitylation complex (44). Notably, beyond its key role in Polycomb-mediated gene repression, the PRC1 ubiquitin ligase catalytic subunit BMI1 rapidly re-localizes to IRIF where it regulates DSB repair via ubiquitylation of H2A and H2AX (45, 46). In line with the results of Ismail and colleagues (47), treating cells with the BMI1 inhibitor PRT4165 markedly delayed IRIF resolution (P ≤ 0.001, Fig. 6E and F). Importantly, IRIF persistence induced by PRT4165 was not reversed by increasing GlcNAcylation with PUGNAc or blocking demethylation with IOX1 (P ≤ 0.001, Fig. 6E and F). These results are consistent with Polycomb-group (PcG) proteins and specifically BMI1/PRC1 as the ultimate target of cancer metabolism mediating epigenetic control of DSB repair, IRIF dynamics, and senescence (Fig. 7).

Among the many hallmarks of cancer (48), altered metabolism has returned to the forefront as a potential therapeutic target. From Warburg's initial description of aerobic glycolysis to the present, the defining feature of cancer metabolic reprogramming has remained the deregulated uptake and utilization of glucose beyond cellular needs for ATP production. This apparent inefficiency is commonly ascribed to the elevated demands for biosynthetic intermediates required for rapid cell proliferation (3, 6, 7, 49) and/or as a means of resistance to cell death (50). Supporting the former model, a moderate correlation has been observed across multiple studies between ¹⁸F-fluorodeoxyglucose (FDG) uptake as measured by PET in human tumors and fraction of Ki-67 positive nuclei or other measures of cell proliferation in biopsies. However, recent discoveries that the oncogenicity of mutations of TCA cycle enzymes IDH1 and IDH2 is mediated by excess production of the oncometabolite 2-HG, leading to inhibition of histone and/or DNA demethylases (10, 51, 52), have lent support to an alternative model that implicates metabolic reprogramming in deregulation of gene expression. Our work extends this model beyond epigenetic regulation of transcription to effects of chromatin modification on DNA repair. We found that disrupting glycolysis and/or glutaminolysis impaired the response to ionizing radiation, thereby linking cancer metabolism to enhanced DNA DSB repair. The increased tolerance to genotoxic stress induced by metabolic reprogramming may promote genomic instability and cell immortality in cancer cells by evading DNA-damage induced senescence, independent of cell proliferation per se.

We note that our primary strategy was to exploit small molecule probes rather than genetic tools to perturb cell metabolism. Beyond offering finer control of dose and timing compared with RNAi (53), chemical probes are particularly suited to the analysis of metabolic networks, where intermediates and metabolites are often considered nodes connected by metabolic enzymes as edges. Although RNAi, CRISPR, or other nucleic acid-targeted perturbations are powerful tools for analysis of specific gene products and linear pathways, they are poorly matched to analysis of cell metabolism, given the complex web of reactions that affect many intermediates. By contrast, a single chemical probe may target multiple isoforms or whole classes of binding sites. In turn, probes are readily used in combination and temporal order, allowing straightforward interrogation of order-of-function, functional redundancy, and other pathway relationships (54, 55). Of translational significance, many of the small molecules used here have been evaluated for therapeutic use (46).

Among the most dramatic results was the block to DNA repair induced by TCA cycle intermediates including the cell permeable analogue of α-KG. In turn, the TCA cycle enzyme-derived oncometabolite 2-HG blocked all these effects. Providing a strong argument that α-KG and its competitive inhibitor 2-HG mediate their effects via epigenetic targets, (1) α-KG serves as a co-factor for multiple dioxygenases including the Jumonji domain-containing (JMJD) HDM, while 2-HG is a dioxygenase and JMJD inhibitor, (2) α-KG effects were recapitulated by G9a HMT inhibitor BRD4770 and EZH2 HMT inhibitor GSK126, and (3) α-KG effects were blocked by Jarid1A/B HDM inhibitor PBIT and JMJD3 demethylase inhibitor GSKJ4. These results appear to provide a link between mitochondrial metabolism and genomic instability that may be independent of oxidative phosphorylation and ROS.

Among our results with the greatest potential for translation, we implicated the HBP in regulation of DNA repair and senescence. We found that activating OGT by treating cells with GlcNAc, a readily available neutraceutical used to relieve inflammation, or targeting OGA with its inhibitor PUGNAc or by knockdown with siRNA each promote DSB repair. Blocking the pathway with small-molecule OGT inhibitors or by OGT knockdown had the opposite effect. Of potentially relevant OGT targets, surveys of the O-GlcNAc-ome have revealed modification of histones, DNA damage repair proteins, histone methyltransferases, and PcG proteins (57–59). Specifically, OGT mediates O-GlcNAcylation of EZH2 on Ser75 in MCF7 cells, enhancing protein stability and maintaining H3 K27 trimethylation (60). Depletion of EZH2 is sufficient to delay DSB repair (61). However, given the multiple targets of O-GlcNAcylation in epigenetic regulation, its precise roles in DSB repair and IRIF kinetics remain to be elucidated.

Multiple lines of evidence point to cancer metabolism accelerating DSB repair via activation of NHEJ. First, our use of GFP-IBD as a reporter may have skewed our study toward tracking end-joining, given the known role of 53BP1 in promoting NHEJ over homologous recombination (HR; refs. 23, 62). In turn, NHEJ inhibitors promote IRIF persistence and senescence after irradiation (63), although candidate HR inhibitors fail to promote either response (our unpublished results). Providing a potential mechanistic link, several PcG proteins have been linked to NHEJ by detection of physical interactions and/or functional studies. One implication is that cancer metabolic reprogramming may speed resolution of DSBs by rapid NHEJ repair before they can initiate slow, but accurate, HR repair. Cancer cells may gain an advantage not only from more efficient DSB repair that supports proliferation in the face of genotoxic stress, but also from the increased genomic instability due to error-prone repair.

The Warburg effect has been widely recognized as a driver of cancer cell growth, tumorigenicity, and therapeutic resistance. Our chemical genetics approach revealed that cancer metabolism also promotes DNA repair and thereby blocks accelerated senescence, inactivating a recognized tumor suppressor mechanism that serves as a critical barrier to carcinogenesis (64–66). These studies identified roles for both glycolysis and glutaminolysis via their dual contributions to the TCA cycle and HBP as key determinants of cellular responses to DNA damage. Our data also suggest a re-examination of the mechanisms of carcinogenesis by oncometabolites to include mechanisms beyond deregulation of gene expression. Indeed, 2-HG alone appears sufficient to recapitulate the benefits of metabolic reprogramming in promoting DNA repair and blocking cell senescence. Similarly, that treating cells with GlcNAc mirrors the effects of the oncometabolite 2-HG supports a reevaluation of O-GlcNAcylation as a mediator and target in cancer and aging.

Taken together, our results suggest that along with an established role in supporting cell proliferation, increased glycolysis and glutaminolysis may also support cancer cell survival in the face of genotoxic stress. By promoting DNA repair by error-prone NHEJ, metabolic reprogramming may serve a previously unrecognized role in tumor progression to increase genomic instability and maintain cellular immortality.

No potential conflicts of interest were disclosed.

Conception and design: E.V. Efimova, N.A. Shamsi, S.J. Kron

Development of methodology: E.V. Efimova, S. Takahashi, N.A. Shamsi, E. Labay, O.A. Ulanovskaya

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E.V. Efimova, S. Takahashi, E. Labay, O.A. Ulanovskaya

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E.V. Efimova, S. Takahashi, N.A. Shamsi, D. Wu, E. Labay, O.A. Ulanovskaya, R.R. Weichselbaum, S.J. Kron

Writing, review, and/or revision of the manuscript: E.V. Efimova, S. Takahashi, N.A. Shamsi, E. Labay, O.A. Ulanovskaya, R.R. Weichselbaum, S.J. Kron

Study supervision: S.A. Kozmin, S.J. Kron

The authors thank S. Bond, C. Labno, and V. Bindokas in The University of Chicago Integrated Microscopy Core Facility for technical assistance; V. Boilot and A. Ramamurthy for technical assistance and proofreading; and W. Lu and J. Rabinowitz at Princeton University for helpful discussions.

This work was generously supported by NIH R01s CA164492 and CA176843, metabolomics supplement CA164492-S1, and a grant from Grant Achatz and the Alinea team (to S.J. Kron), by a Susan G. Komen Postdoctoral Fellowship KG101224 (to S. Takahashi), by P50 GM086145 (to S.A. Kozmin), and by funds from the Ludwig Center for Metastasis Research (to R.R. Weichselbaum).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.