DR4-Ser424 O-GlcNAcylation Promotes Sensitization of TRAIL-Tolerant Persisters and TRAIL-Resistant Cancer Cells to Death

TNF-related apoptosis-inducing ligand (TRAIL) has an attractive ability to kill cancer cells selectively (1, 2). TRAIL research has expanded to the study of its cognate receptors and the development of various TRAIL receptor agonists (TRA) as therapeutic candidates for cancer prevention (3). However, clinical trials of TRAIL or TRAs have failed to achieve a beneficial anticancer activity because many primary cancers are resistant to TRAIL (4–6). In some cases, small subpopulations of drug-sensitive cancer cells also acquire a nonmutational drug-tolerant phenotype and enter a “persistent state” due to heterogeneity within cancer cell populations (7–9). Thus, the use of genetic or proteomic profiling to elucidate the mechanisms that cause TRAIL resistance in cancer cells (10) or the combination of TRAIL with radiotherapy or chemotherapy (11) has been explored in an effort to develop innovative TRAIL-based therapeutic concepts.

Unlike the single functional TRAIL receptor in murine (12), there are two TRAIL death receptors in humans, DR4/TRAIL-R1 and DR5/TRAIL-R2, which mediate TRAIL-induced apoptosis, necrosis, and gene-activating pathways including NF-κB signaling (13–17). Several posttranslational modifications (18, 19) and spatial dynamics (20) of the receptors have been reported. In addition, cancer cells show different preferences for receptor signaling in a cell type–dependent manner (21). Elucidating the differences between two receptors in terms of their activation and signal transduction within the pathophysiologic context is one of the main current concerns in the field of TRAIL research. Thus, the specific targeting and manipulating each receptor by using agonistic/antagonistic/neutralizing antibodies and TRAIL variants selectively binding and activating to either receptor have been considered for the TRAIL experiments and development of clinical therapies (3, 4, 22, 23). For these research avenues, the DR4-dominant (e.g., DLD-1 human colorectal adenocarcinoma cells) and/or DR5-dominant (e.g., SK-Hep1 human hepatic adenocarcinoma cells) type of cells has been employed.

The O-linked β-N-acetylglucosamine (O-GlcNAc) modification, which is termed O-GlcNAcylation, is a conserved posttranslational modification at serine/threonine residues (24). This modification serves as a nutrient sensor for metabolic status and is responsible for numerous cellular signaling events and disease processes (25–27). UDP-GlcNAc derived from glucose via the hexosamine biosynthetic pathway (HBP) is the high-energy donor that is added to target proteins by O-GlcNAc transferase (OGT) and is removed hydrolytically by O-GlcNAcase (25). Although O-GlcNAcylation occurs primarily on nuclear/cytoplasmic proteins (24), this modification can mediate critical signaling at the plasma membrane via the interaction of OGT with phosphatidylinositol trisphosphate (28). However, the modification of TRAIL death receptors by O-GlcNAcylation and its roles in tumorigenesis are unknown.

From a cell-based functional screen using a cancer cell–derived cDNA expression library, we unexpectedly isolated a DR4 mutant that suppressed TRAIL cytotoxicity in cancer cells. We characterized that O-GlcNAcylation of DR4-Ser424 is a crucial step to initiate DR4-mediated tumor-selective killing mechanisms and to determine TRAIL sensitivity in DR4-preferred cancer cells. Moreover, we propose prospective strategies for overcoming TRAIL resistance to facilitate the clinical application of TRAIL.

Human stomach cancer cells (SNU lines) were obtained from Cancer Research Institute of Seoul National Hospital, Seoul National University (Seoul, Korea; ref. 29), and all other cells were from the ATCC. All the cell lines, except SNU lines, were authenticated by short tandem repeat profiling and tested for Mycoplasma contamination using MycoAlert Mycoplasma Detection Kit (Lonza) more than once during this study. SNU lines were authenticated by their manufacturers. Cells were used within a maximum of 13 passages after thawing.

Reagents and specific antibodies used in this study are described in Supplementary Materials and Methods. IDN-6556 was kindly provided by Dr. C.-W. Chung (LG Biotech Inc., Korea).

The wtDR4 construct has been reported previously (30). All point mutations and SNPs of DR4 were generated by site-directed mutagenesis through PCR using specific primers containing the corresponding mutations. The sequences of primers are described in Supplementary Materials and Methods. Transfection was performed with iN-fect (iNtRON Biotechnology) or polyethylenimine (PEI; Sigma-Aldrich) following the manufacturer's instructions.

For genomic engineering of OGT through CRISPR/Cas9 system, the single-guide RNA sequences (sgRNA oligos) targeting OGT (OGT sg; Supplementary Materials and Methods) were annealed and cloned into the BsmBI (Fermentas) sites of lentiCRISPR transfer plasmid (pXPR_001; Addgene) for virus production. Viral vectors were produced in HEK293T cells, and cells of interest were infected for OGT knockdown as described previously (31). After lentiviral delivery of both Cas9 and OGT sg, cells were selected with puromycin (Sigma-Aldrich). After single-cell cloning, the clones were screened by Western blotting.

Human full-length cDNA expression library was prepared by cloning full-length cDNA of mRNA sources from various cancer cell lines into pME18S-FL3 expression vector using a PCR-based oligo-capping method (Korea Research Institute of Bioscience & Biotechnology, KRIBB; https://genbank.kribb.re.kr). After cotransfection of a library with pEGFP-N1 into HeLa cells, cells were treated with TRAIL and the viability of GFP-positive cells was assessed as a readout under a fluorescence microscope (Olympus). Bcl-2 and c-FLIP_L were used as positive controls to isolate negative regulators in TRAIL-induced cell death.

The Cancer Genome Atlas (TCGA) consequence for S424L mutation of TNFRSF10A in uterine corpus endometrial carcinoma [TCGA-UCEC/TCGA-B5-A1MR] was accessed from Genomic Data Commons (GDC) Data Portal (https://portal.gdc.cancer.gov).

Cell viability was examined by microscopic detection (32), FACS analysis, or Trypan blue exclusion assay. Briefly, nuclear chromatin was stained with 1 μg/mL Hoechst 33342 (Molecular Probes) and 1 μg/mL propidium iodide (PI; Sigma-Aldrich) for approximately 15 minutes. Apoptotic ratio was measured from cells with fragmented blue nuclei among PI-positive signals (n = 300–600 cells) under a fluorescent microscope. For the quantification of PI incorporation by FACS analysis, adherent and floating cells harvested by trypsinization and centrifugation, respectively, were resuspended in PI solution and analyzed by using a FACSCalibur (BD Biosciences). Necrotic cells were analyzed by flow cytometry after Annexin-V and 7AAD double-staining using FITC Annexin-V Detection Kit with 7AAD (640922, BioLegend) according to the manufacturer's instructions.

For flow cytometry analysis, phycoerythrin (PE)-conjugated mouse anti-human DR4 (DJR1), anti-human DR5 (DJR2-4), and anti-IgG isotype control (MOPC-21) antibodies (BioLegend) were used according to the manufacturer's instructions. Briefly, cells were trypsinized and collected, washed with Cell Staining Buffer (420201, BioLegend), and incubated with PE-conjugated antibodies for 1 hour on ice. After being washed and resuspended with Cell Staining Buffer, cells were analyzed by flow cytometry using a FACSCalibur.

For O-GlcNAc detection, cells were preincubated in a glucose-free medium for 10–30 minutes and then exposed to TRAIL in a glucose-supplemented medium. Cells were lysed with preheated 2.5% SDS-containing buffer: 10 mmol/L Tris-HCl (pH 8.0), 1 mmol/L EDTA, 2.5% SDS, and protease inhibitor cocktail (Quartett). After sonication (optional) and heat denaturing at 95°C, lysis buffer (50 mmol/L Tris-HCl pH 8.0, 150 mmol/L NaCl, 10% -glycerol, 1% Triton X-100, 1 mmol/L EDTA, 1 mmol/L EGTA, 10 μmol/L Thiamet-G, and protease inhibitor cocktail) was added for dilution of the final SDS concentration to under 1%. Samples were pipetted or incubated with rocking to relieve the viscous state. After centrifugation at 20,000 × g, O-GlcNAcylated proteins were immunopurified from the supernatant using an anti-DR4 antibody, an anti-Flag M2 affinity gel (Sigma-Aldrich), or GlcNAc-specific succinylated wheat germ (sWGA; Vector Laboratories).

Samples were prepared as described previously (33). Briefly, cells were preincubated with 500 ng/mL His-TRAIL (BML-SE721, Enzo Life Sciences) for 30 minutes at 4°C, and further incubated at 37°C for additional 30 minutes to 1 hour. For DISC purification, the resulting lysate was incubated with Ni-NTA agarose (QIAGEN) at 4°C for the enrichment of His-TRAIL–captured immunocomplexes. Lipid rafts were purified from the same cell lysates by discontinuous sucrose density gradient purification (33). For the isolation of TRAIL-DISC within lipid rafts, the resulting lipid raft fraction was further incubated with Ni-NTA agarose for ligand complex capture.

Statistical analysis was performed using Prism version 5.03 software (GraphPad Software). Data are expressed as mean ± SEM from at least three experiments. For statistical comparisons, data were analyzed by either one-way ANOVA, followed by Tukey post hoc test or two-way ANOVA, followed by Bonferroni post hoc test as appropriate, and P < 0.05 was considered statistically significant.

To isolate a genetic regulator of TRAIL-mediated cell death, we performed a gain-of-function screen using a cell death rescue assay. A total of 1,099 full-length cDNAs in a mammalian expression vector were prepared from patient-derived cancer cells and transfected into HeLa human cervical carcinoma cells, followed by TRAIL treatment (Fig. 1A, left). From the screen, four novel clones were identified to affect TRAIL cytotoxicity, and unexpectedly, DR4 was one of those clones (Supplementary Fig. S1A, TNFRSF10A). We found that this DR4 clone had four mutations (Fig. 1A and B, MUT), comprising three (L86R, H141R, and R209T) in the extracellular region and one (S424P) in the cytosolic death domain. Unlike wtDR4, DR4-MUT did not induce apoptotic morphologies, including cell shrinkage and membrane blebbing (Fig. 1A, graph). To identify which of these mutations caused the loss of cytotoxicity, we generated four revertants from DR4-MUT by individually substituting the mutations back into wild-type residues, so that each revertant harbored three mutations and one wild-type residue (Fig. 1B). Using reconstitution systems in DR4-null T98G cells, we first confirmed that all the revertants expressed equivalently on the cell surface as wtDR4 (Supplementary Fig. S1B). With ectopic expression assays in DLD-1 cells in which the TRAIL signaling is triggered dominantly through DR4, we found that only the DR4-MUT^P424S revertant fully restored the cytotoxic activity equal to that of wtDR4 (Fig. 1C).

Next, to examine whether DR4-Ser424 is essential for mediating TRAIL-induced cell death, we directly introduced a proline mutation at Ser424 of wtDR4 by substitution of the 1270th thymidine with cysteine. In addition to the DR4^S424P mutant isolated in our screen, we also found another DR4-Ser424 mutation. From searching the database of patients with cancer in TCGA (TCGA-B5-A1MR-10A), we discovered a DR4^S424L mutant in uterine corpus endometrial carcinoma, although the frequency was low (1 of 530 cases, 0.19%). Ectopic expression assays revealed that all these mutants showed loss of cytotoxicity after TRAIL stimulation and even suppressed TRAIL-induced apoptosis (Fig. 1D) despite of the equivalent membrane expressions to wtDR4 (Supplementary Fig. S1C). In addition, with reconstitution assays in DR4-null T98G human glioblastoma cells and SNU-216 human stomach adenocarcinoma cells, we confirmed a lack of caspase activation and PARP-1 cleavage by DR4-Ser424 mutants (Fig. 1E). Finally, because DR4 polymorphisms associated with cancer risk have been described in various human cancers (34), we examined the cytotoxicity of other DR4 polymorphisms (rs6557634, rs4871857, rs17088993, rs2230229) that have appeared at particularly high frequencies in cancers. Unlike DR4^1270T>C (DR4^S424P), the other polymorphisms described above actively induced apoptosis in DLD-1 cells (Supplementary Fig. S1D), emphasizing the notion that mutations at DR4-Ser424 are responsible for the resistance of tumor cells to TRAIL-induced apoptosis.

Because TRAIL induces not only apoptosis but also programmed necrosis (15, 16), we decided to test whether DR4-Ser424 mutants could also affect necrosis. Owing to the requirement of RIPK3 for necroptosis, we employed two different cancer cell lines from analyses of apoptosis, SNU-16 human stomach adenocarcinoma cells expressing endogenous RIPK3 and HeLa cells expressing exogenous RIPK3-HA (HeLa/RIPK3-HA), and treated them with TRAIL in the presence of SMAC mimetic SM-164 and a pan-caspase inhibitor IDN-6556 (T/S/I in figures). Similar to the results for apoptosis, necroptosis was also suppressed by DR4-Ser424 mutants (Fig. 1F). In addition, the active forms of necroptosis mediators, such as phosphorylated RIPK1/RIPK3/MLKL, were not observed after the ectopic expression of DR4-Ser424 mutants (Fig. 1G). Thus, DR4-Ser424 is the key residue for triggering necroptosis as well as apoptosis in response to TRAIL.

Serine generally undergoes posttranslational modifications, such as phosphorylation and O-GlcNAcylation (35). We thus hypothesized that DR4-Ser424 might undergo a posttranslational modification and monitored the phosphorylation and O-GlcNAcylation of DR4 upon TRAIL administration. Results from immunoprecipitation (IP) analysis using an anti-phosphoserine antibody showed that the overexpressed wtDR4 appeared to be phosphorylated (Fig. 2A, IP:p-Serine). However, the observation that DR4^S424A also seemed to be phosphorylated as much as wtDR4 revealed that DR4 phosphorylation might occur at residues other than Ser424 (Fig. 2A). Because proline's unique cyclic structure commonly causes conformational changes in the secondary structure of proteins (36), the weak phosphorylation of DR4^S424P may have been due to a structural perturbation (Fig. 2A). Meanwhile, we observed that wtDR4 was O-GlcNAcylated, but all DR4-Ser424 mutants, including DR4^S424P, DR4^S424A, and DR4^S424L, were not (Fig. 2A, IP:HA). Using GlcNAc-specific agarose-conjugated sWGA to enrich O-GlcNAcylated proteins by affinity purification, we confirmed that DR4-Ser424 is the only site of O-GlcNAcylation and there was no additional Ser/Thr residue for O-GlcNAcylation (Fig. 2B) or the modulation of cytotoxicity (Supplementary Fig. S1E) within the peptides from 425th residue to the end of DR4. In contrast to DR4, we found that DR5 was not O-GlcNAcylated under the same conditions (Fig. 2C).

We observed that a little proportion of endogenous DR4 proteins was modified by O-GlcNAc under basal conditions in untreated control cells, while the frequency of this modification markedly increased in response to treatment with either apoptotic TRAIL (Fig. 2D) or necrosis-inducing T/S/I (Fig. 2E). Accordingly, DR4 O-GlcNAcylation was abolished by the CRISPR/Cas9-mediated knockdown of OGT expression (Fig. 2E; Supplementary Fig. S2A). Furthermore, results from immunoprecipitation assays revealed that upon TRAIL stimulation, the binding of DR4 to OGT was enhanced in apoptotic DLD-1 and necrotic SNU-16 cells (Fig. 2F), or OGT bound to DR4 vice versa (Fig. 2G). Taken together, we concluded that DR4 was modified by O-GlcNAc upon TRAIL ligation in an OGT-dependent manner.

We further examined TRAIL-induced cell death under O-GlcNAcylation–deficient conditions. We found that TRAIL-induced apoptosis was accelerated in the presence of Thiamet-G, an inhibitor of O-GlcNAcase, but mitigated by ST045849, an OGT inhibitor, and by 6-diazo-5-oxo-l-norleucine (DON), a glutamine analogue that results in a decreased concentration of UDP-GlcNAc (Fig. 2H). In SK-Hep1 and HeLa cells, which transmit TRAIL signaling preferentially via DR5, blocking DR5 using a DR5-neutralizing antibody (αDR5) allowed only DR4-mediated TRAIL signaling. Under this condition, TRAIL/DR4-dependent cell death was undermined by OGT knockdown (Fig. 2I, TRAIL+αDR5). As expected, TRAIL-induced cytotoxicity was not influenced by OGT levels in the absence of DR5 blocking (Fig. 2I, TRAIL) because DR5 was not modified by O-GlcNAc (Fig. 2C). Consistently, the overexpression of wtDR4 evoked less apoptosis and caspase-8 activation in OGT-deficient SK-Hep1 cells than those of DR5, which was similar to the effect of O-GlcNAcylation-defective DR4^S424P (Fig. 2J). Expressions and surface localization of DR4 and DR5 were not affected by OGT knockdown (Fig. 2J, Flag; Supplementary Fig. S2B). Upon TRAIL-necrotic stimulation combined with blocking DR5, the phosphorylation-dependent activation of RIPK1/RIPK3/MLKL was also suppressed in OGT-knockdown HeLa/RIPK3-HA cells (Supplementary Fig. S2C). These observations indicate again that O-GlcNAcylation specifically enhances DR4-mediated cell death.

To examine the contribution of DR4 O-GlcNAcylation to the TRAIL-DISC formation, we used again Thiamet-G and ST045849 to potentiate and inhibit O-GlcNAcylation, respectively, and purified DISC. Because DR5 was not O-GlcNAcylated (Fig. 2C, I, and J), the amount of DR5 within the isolated DISC was not influenced by O-GlcNAcylation, while DR4 recruitment into the TRAIL-DISC was abolished by treatment of ST045849 (Fig. 3A). Levels of FADD and caspase-8 in the purified TRAIL-DISC were apparently increased after incubation with Thiamet-G, but were again diminished by treatment of ST045849 (Fig. 3A). Interestingly, OGT was also recruited into the purified DISC together with DR4. These results indicate that DR4 O-GlcNAcylation is required for DISC formation upon TRAIL stimulation.

Confirmatively, compared with wtDR4, the O-GlcNAcylation–defective DR4^S424P was little recruited into the DISC upon TRAIL stimulation and failed to form TRAIL-DISC in HeLa cells (Fig. 3B, His pull-down). DR4-Ser424 mutants also failed to bind to FADD (Supplementary Fig. S3A). While the similar level of DR5 was detected within TRAIL-DISC regardless of the endogenous DR4 O-GlcNAcylation (Fig. 3A), there was little DR5 within TRAIL-DISC in DR4^S424P-expressing cells as compared with that of wtDR4-expressing cells (Fig. 3B, His pull-down). Because DR4^S424P could also interact with DR5 upon TRAIL ligation (Supplementary Fig. S3B), we speculated that DR4^S424P might capture DR5 like a decoy receptor and hinder DR5 from being properly recruited into the TRAIL-DISC. We tested this hypothesis by purifying and analyzing DR4-enriched complex. When the protein complexes in wtDR4-Flag- and DR4^S424P-Flag–expressing cells were purified by a Flag pull-down, we found that OGT, FADD, and caspase-8 were not detected in the DR4^S424P complex, whereas DR5 was contained in the DR4^S424P complex as well as in the wtDR4 complex (Fig. 3B, Flag pull-down).

To further inquire whether O-GlcNAcylated DR4–containing DISC is indeed assembled within lipid rafts and whether DR4 O-GlcNAcylation is critical for DR4 translocation into and clustering within lipid rafts, we prepared lipid rafts using sucrose density gradient fractionation (33) and examined DISC formation, DR4 distribution, and receptor clustering. After TRAIL stimulation, DISC was formed within lipid rafts in DLD-1 cells (Fig. 3C) and wtDR4-expressing HeLa cells (Fig. 3D), but not in ST045849-treated cells (Fig. 3C) and DR4^S424P-expressing cells (Fig. 3D). Accordingly, unlike DR4-Ser424 mutants, only wtDR4 could translocate into lipid rafts (Fig. 3E). Unlike wtDR4, DR4-Ser424 mutants also failed to self-aggregate in both non-raft and lipid raft fractions under nonreducing conditions (Fig. 3F). Together, DR4 O-GlcNAcylation is an essential prerequisite to initiate TRAIL signaling via facilitating DISC formation and DR4 clustering in lipid rafts. More, when we analyzed the impact of DR4 O-GlcNAcylation on necrosome formation, we found that, unlike those in wtDR4-expressing HeLa/RIPK3-HA cells, necrosome (Fig. 3G) and MLKL trimers (Fig. 3H; ref. 37) were not observed in DR4-Ser424 mutant–expressing cells upon TRAIL-necrosis stimulation.

Given that O-GlcNAcylation is regulated according to the concentration of glucose (26), we assessed the effects of glucose on DR4 O-GlcNAcylation and TRAIL/DR4-dependent death. Compared with DLD-1 and HEK293T cells cultured in a glucose-rich medium, DR4 O-GlcNAcylation following TRAIL ligation was not observed in the same cell types cultured in a glucose-free medium (Fig. 4A–C). DR4 O-GlcNAcylation in HEK293T cells expressing wtDR4-Flag increased as the concentration of glucose was elevated and by treatment with Thiamet-G (Fig. 4B). Conversely, the modification was reduced by treatment with DON (Fig. 4B). Again, O-GlcNAcylation of DR4^S424P was undetectable regardless of the presence or absence of glucose (Fig. 4C). Thus, glucose is a critical metabolic factor for DR4 O-GlcNAcylation as well.

In general, nutrient deprivation, such as glucose starvation, is a stress that eventually triggers cell death. Thus, sustained exposure to a low concentration of glucose or the inhibition of glucose metabolism augments TRAIL cytotoxicity (38). Our experimental results from maintaining cells under glucose deprivation for a long duration coincided with those from the previous studies. When DLD-1 cells were exposed to TRAIL under a glucose-deprived condition for more than 12 hours, the cells became more vulnerable to TRAIL-induced cell death (Supplementary Fig. S4A). Unlike this long-lasting effect of glucose deprivation, however, when DLD-1 cells were incubated under glucose-free conditions for a short time, cell death triggered by TRAIL was markedly mitigated at such early time points (Fig. 4D). In contrast, there was no significant difference in the incidence of cell death triggered by other cytotoxic stimuli, such as TNFα plus cycloheximide, an extrinsic death signal, or a calcium ionophore A23187 that causes intrinsic death (Fig. 4D). Accordingly, subjecting cells to the short-term deprivation of glucose impaired the formation of DISC and subsequent caspase-8 activation under TRAIL stimulation (Fig. 4E; Supplementary Fig. S4B). To further evaluate the effect of glucose concentration on TRAIL efficacy in detail, we generated tumor cells adapted to a low-glucose medium. DLD-1 cells were cultured with a complete (25 mmol/L; DLD-1/Con-Glc) or a low-glucose (5.5 mmol/L; DLD-1/Low-Glc) medium for 1 month and tested for their TRAIL sensitivity. Although membrane expressions of DR4 in these cell lines were equivalent (Fig. 4F), DR4 O-GlcNAcylation triggered by TRAIL (Fig. 4G) and sensitivity to TRAIL (Fig. 4H) were significantly reduced in DLD-1/Low-Glc cells as compared with DLD-1/Con-Glc cells. This evidence suggests that cancer cells become less susceptible to TRAIL-mediated cell death under glucose-deficient conditions owing to the impairment of DR4 O-GlcNAcylation.

In cancer cells mainly using DR4 for the transduction of TRAIL signaling, TRAIL efficacy was highly associated with glucose concentrations and ultimately with DR4 O-GlcNAcylation. To inquire whether impaired O-GlcNAcylation could be a novel causative factor of TRAIL resistance in cancer cells, we sought to determine whether TRAIL sensitivity correlated with DR4 O-GlcNAcylation in various cancer cell lines. Total 15 cancer cell lines were assessed for their receptor expressions and sensitivity to TRAIL (Fig. 5A; Supplementary Fig. S5A–S5C). We classified them into TRAIL-sensitive (TRAIL-S) and -resistant (TRAIL-R; death rates ≤ 20%) groups and further classified the DR4-positive (DR4⁺)/TRAIL-sensitive cell lines into DR4-dominant (DR4*) or DR5-dominant (DR5*) groups according to their preferential receptor engagement. Receptor preference was classified by analyzing cell death after treatment with TRAIL and αDR5 (Fig. 5A; Supplementary Fig. S5B). Then, the status of DR4 O-GlcNAcylation in DR4-positive cancer cells was examined with O-GlcNAcylation assays using either an anti-DR4 antibody or sWGA. Surprisingly, DR4 O-GlcNAcylation was markedly promoted by TRAIL stimulation only in the TRAIL-sensitive cell lines, but not observed or not increased upon TRAIL treatment in all of the TRAIL-resistant cell lines we tested (Fig. 5B; Supplementary Fig. S5D). Thus, we suggest that the impairment of DR4 O-GlcNAcylation after TRAIL ligation would lead to TRAIL resistance in cancer cells.

To verify a direct correlation and causality between DR4 O-GlcNAcylation and TRAIL susceptibility, we attempted to explore this association further focusing on the drug-tolerant “persistent state,” which is a recently proposed aspect of drug resistance (7–9). First, we obtained a small fraction of TRAIL-tolerant persisters (TTP) from TRAIL-sensitive DLD-1 cells (DLD-1^Naïve) after exposure to TRAIL and further established expanded persistent populations (DLD-1^PER) that grew in the continuous presence of TRAIL (Fig. 5C and D). We found that both DR4 O-GlcNAcylation and its interaction with OGT were abrogated in DLD-1^PER cells showing acquired resistance to TRAIL, whereas the expression levels of OGT and death receptors were unchanged as compared with DLD-1^Naïve cells (Fig. 5E; Supplementary Fig. S5E). Next, we abolished the TRAIL-tolerant state of DLD-1^PER cells by propagating the cells in the absence of TRAIL (DLD-1^PER-S) to resensitize them to TRAIL (Fig. 5C and D). Interestingly, we found that DR4 O-GlcNAcylation was restored in the DLD-1^PER-S cells (Supplementary Fig. S5F). Consistently, TRAIL-DISC was not assembled in DLD-1^PER cells, but was restored in DLD-1^PER-S cells (Fig. 5F), suggesting that the deficiency of DR4 O-GlcNAcylation also renders TRAIL-sensitive cancer cells persistent to TRAIL.

Considering the significant influence of cellular glucose concentration on O-GlcNAcylation and the subsequent cytotoxicity of DR4, we hypothesized that increasing the glucose concentration could render TRAIL-resistant cancer cells susceptible to TRAIL through enhancing DR4 O-GlcNAcylation. To test this hypothesis, cancer cells were exposed to TRAIL together with a high concentration of glucose (high-Glc, 100 mmol/L; TRAIL+high-Glc) or a glucose analogue 2-deoxy-d-glucose (2-DG; TRAIL+2-DG). It is well-established that 2-DG increases TRAIL sensitivity in clinical studies (39) and also leads to the hyper-O-GlcNAcylation of certain proteins despite its inhibitory effect on glycolysis (40). Notably, TRAIL-resistant SNU-638 and SNU-719 human stomach adenocarcinoma cells became significantly vulnerable to TRAIL upon incubation with high-Glc or 2-DG (Fig. 6A and B; Supplementary Fig. S6A).

Because upregulation and self-aggregation of death receptors by 2-DG and/or ER stress have been reported (38, 39, 41, 42), we examined receptor levels after incubating SNU-638 cells with TRAIL+2-DG or TRAIL+high-Glc. When SNU-638 cells were stressed by treatment with TRAIL+2-DG, DR4 with a lower molecular weight was detected and DR5 was upregulated by 2-fold (Supplementary Fig. S6B and S6C). On the other hand, induction of DR5 expression and ER stress was not observed by treatment with TRAIL+high-Glc (Supplementary Fig. S6B, right). We found that 4PBA, an ER stress inhibitor, potently suppressed the increase of DR5 triggered by TRAIL+2-DG (Supplementary Fig. S6D). We then analyzed the effects of DR4-O-GlcNAcylation on TRAIL sensitivity in SNU-638 cells using both αDR5 and 4PBA. We found that TRAIL+2-DG-induced cell death via DR4 was potentiated by Thiamet-G but was alleviated by ST045849 in SNU-638 cells (Fig. 6A, white bars). The same results were observed when 2-DG–induced ER stress was inhibited by 4PBA (Fig. 6A, dark gray bars) and when SNU-638 cells were incubated in a high-Glc–containing culture medium (Fig. 6B), indicating that the sensitization of TRAIL-resistant SNU-638 cells to TRAIL by treatment with 2-DG and high-Glc is indeed due to DR4 O-GlcNAcylation.

Confirmatively, we found that DR4 O-GlcNAcylation by the treatment with TRAIL+2-DG or TRAIL+high-Glc was promoted in the presence of Thiamet-G, while this modification was not observed in the presence of ST045849 (Fig. 6C and D). Moreover, nongenetically TRAIL-resistant DLD-1^PER cells were also resensitized to TRAIL in the presence of high-Glc or 2-DG (Fig. 6E), and DR4 O-GlcNAcylation in these DLD-1^PER cells was recovered as well (Fig. 6F). Thus, enhancing DR4 O-GlcNAcylation improves TRAIL sensitivity in TRAIL-resistant cancer cells. Taken together, we propose again that DR4-Ser424 is the essential residue for the cytotoxic functioning of DR4 via O-GlcNAcylation in TRAIL signaling, providing clues for increasing the responsiveness to TRAIL therapy in patients with cancer.

Recent studies have highlighted the importance of sugar modifications, such as DR4 N-glycosylation (18) and DR5 O-glycosylation (19). Most of the reported modifications of TRAIL receptors, however, have focused only on the apoptotic elimination of cancer cells. In this study, we demonstrate that DR4 O-GlcNAcylation is important for the induction of not only apoptosis but also necrosis by TRAIL. Although necrosis is highly associated with inflammation and accelerates the pathogenesis of human diseases, including inflammatory diseases (43) and tumorigenesis (44), programmed necrosis works as a second barrier against tumor cells that avoid apoptosis (16). Notably, necrosis is the crucial step for the killing and elimination of stem-like cells (45) or cancer stem cells (46). Tumors that can evade both apoptotic and necrotic defenses tend to be aggressive and metastatic. Because O-GlcNAcylation is responsible for DR4-mediated necrosis as well as apoptosis, tumor cells harboring mutations at DR4-Ser424 have a high risk of becoming more tolerant against TRAIL therapy.

Tumor cells frequently display altered glucose metabolism, which is one of the hallmarks of cancer (47). O-GlcNAcylation is highly dependent on glucose and this modification generally becomes more abundant in various metabolic pathways in cancer cells (48). It appears that DR4 O-GlcNAcylation is a nutrient-dependent checkpoint that licenses DR4 to transmit death signals upon TRAIL stimulation. A low basal level of DR4 O-GlcNAcylation was observed in our experiments, but this may not reach the threshold to trigger cell death in nontransformed or unstimulated cells. In tumor cells, however, glucose uptake and HBP flux become greatly increased so that upregulated DR4 O-GlcNAcylation, as an ignition switch, may help TRAIL to selectively kill cancer cells. In addition, a sustained scarcity of glucose can create a favorable environment for malignant tumor cells (49) and consistently, an absence of DR4 O-GlcNAcylation under conditions of glucose shortage renders tumor cells more resistant to TRAIL. In short, the O-GlcNAcylation status of DR4 may determine the sensitivity of DR4-preferred cancer cells to TRAIL.

We also found that TRAIL efficacy was enhanced in TRAIL-resistant cancer cells, even in TRAIL-tolerant cancer persister cells, when we experimentally forced the O-GlcNAcylation of DR4 (Figs. 6 and 7). Our study addresses the mechanism of combined TRAIL chemotherapies in that treatment with a high concentration of glucose or 2-DG together with TRAIL encourages DR4 O-GlcNAcylation in TRAIL-resistant cancer cells, which overcomes their tolerance and restores the effectiveness of TRAIL. That is, we have demonstrated the importance of DR4 participation in the synergistic induction of cell death by TRAIL in combination with 2-DG, which is well-established to augment TRAIL sensitivity in TRAIL-resistant cancers (39). Therefore, we propose that manipulating the intracellular glucose level can provide a new strategy for chemotherapy against TRAIL-resistant cancer cells, although the safety of altering the glucose concentration must be carefully considered for clinical application in practice. In addition, the fast-migrating band on DR4 blot that we observed after 2-DG treatment (Fig. 6D; Supplementary Fig. S6B) remains to be characterized. It might be associated with the effects of 2-DG on ER stress and N-glycosylation (50) because it was similar with the size of unmodified DR4 deficient in N-glycosylation (18).

Compared with the loss-of-function screen using siRNA library, our screen was effective to identify regulators displaying a gain-of-function in a signaling pathway. We discovered DR4-Ser424 mutations as a potent suppressor of TRAIL-induced death and characterized that DR4-Ser424 is O-GlcNAcylated to initiate TRAIL signaling. DR5 was not modified by O-GlcNAc, enabling us to discriminate a novel difference between DR4 and DR5. Despite the low frequency in TCGA data, it is interesting to note the power of DR4-Ser424 mutations because the impairment of DR4 O-GlcNAcylation even has an adverse effect not only on DR4 signaling but also on DR5-mediated cytotoxicity (Supplementary Fig. S7A and S7B). Given that death receptors can form a hetero-trimer upon TRAIL ligation (Supplementary Fig. S3B; ref. 23), the recruitment of the DR4-Ser424 mutant to DR5 can undermine DR5-mediated TRAIL cytotoxic signaling in a dominant-negative manner, which might contribute to tumorigenesis by enabling cancer cells to escape cell death. Thus, we propose that the DR4-Ser424 residue deserves careful consideration as a potential diagnostic marker for the genetic test of patients with cancer. However, we have to consider that DR4 O-GlcNAcylation is not totally responsible for TRAIL death signaling because the impairment of this modification by OGT knockdown and inhibition suppressed TRAIL death considerably but not entirely to the control level. Although further exploration is needed to determine whether DR4 O-GlcNAcylation has other roles in the immune system or NF-κB signaling and how TRAIL enhances DR4 O-GlcNAcylation specifically, our findings may be beneficial for the design of specific TRAIL treatments, bringing us closer to personalized medicine.

No potential conflicts of interest were disclosed.

Conception and design: H. Lee, Y.-K. Jung

Development of methodology: H. Lee, Y. Oh, Y.-J. Jeon, H. Kim, Y.-K. Jung

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Lee, Y. Oh, Y.-K. Jung

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Lee, Y. Oh, Y.-J. Jeon, Y.-K. Jung

Writing, review, and/or revision of the manuscript: H. Lee, H.-J. Lee, Y.-K. Jung

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Lee, Y.-J. Jeon, S.-Y. Lee, H. Kim, Y.-K. Jung Study supervision: Y.-K. Jung

This work was supported by a Global Research Laboratory grant (NRF-2010-00341 to Y.-K. Jung) and by the CRI grant (NRF-2016R1A2A1A05005304 to Y.-K. Jung) funded by the Ministry of Education, Science, and Technology. We thank Prof. J.-W. Cho (Yonsei University, Korea) for his kind discussion and for providing the human OGT plasmid and an anti-O-GlcNAc antibody that were used for our preliminary experiments.

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.