O-GlcNAcylation Signal Mediates Proteasome Inhibitor Resistance in Cancer Cells by Stabilizing NRF1

Cancer cells often heavily depend on the ubiquitin-proteasome system (UPS) for their growth and survival. Irrespective of their strong dependence on the proteasome activity, cancer cells, except for multiple myeloma, are mostly resistant to proteasome inhibitors.

suppression of UPS by proteasome inhibitors does not produce any satisfactory outcomes in the treatment of cancers except in the case of multiple myeloma (4). A major obstacle is the development of resistance to proteasome inhibitors, resulting from the so-called proteasome bounce-back response, which is a cellular response that restores proteasome activity by transcriptionally activating proteasome subunit genes (2,4,7).
Recent studies demonstrated that nuclear factor erythroid 2-related factor 1 (NRF1) is a key transcription factor that is responsible for the induction of proteasome subunit genes by proteasome inhibition and acts as a rheostat of proteasome activity in vivo (8)(9)(10). NRF1 belongs to the cap'n'collar (CNC) family of transcription factors, possessing a well-conserved basic-region leucine zipper (bZip) motif, and it heterodimerizes with small MAF (sMAF) proteins, which are members of another bZip transcription factor family (11,12). Neural tissue-specific Nrf1 knockout mice exhibit abnormal accumulation of polyubiquitinated proteins in the brain, supporting an essential role of NRF1 in the maintenance of proteasome function (13,14). NRF1 is initially synthesized as an endoplasmic reticulum (ER) transmembrane protein possessing a long C-terminal portion with N-linked glycosylation in the ER lumen and a short N-terminal portion in the cytoplasm (15,16). Under normal conditions, NRF1 is subjected to ER-associated degradation (ERAD); the luminal portion of NRF1 is retrotranslocated to the cytoplasm by p97/VCP, followed by its deglycosylation and ubiquitination for degradation (15)(16)(17)(18)(19)(20)(21). When cells are exposed to proteasome inhibitors, NRF1 is stabilized and cleaved by DDI-2 protease, resulting in a release of processed NRF1 from the ER into the nucleus and transcriptional activation of proteasome subunit genes (22)(23)(24). Thus, ERAD is recognized as a critical node in the regulation of NRF1 activity. In contrast, a post-ER mechanism of NRF1 regulation has been described as a stability control by Fbw7-or ␤-TrCP-dependent UPS (25,26).
O-linked N-acetylglucosamine (O-GlcNAc) is a posttranslational modification at serine or threonine residues of target proteins in the nucleus and cytoplasm (27). As serine and threonine residues are also target sites of phosphorylation, O-GlcNAcylation is considered to compete with phosphorylation for certain Ser/Thr target sites (28). O-linked N-acetylglucosamine transferase (OGT) is an enzyme that catalyzes the addition of a GlcNAc moiety by using UDP-GlcNAc as a substrate. UDP-GlcNAc is synthesized through the hexosamine biosynthesis pathway (HBP), and its amount is influenced by nutritional conditions. For example, increased availability of glucose elevates cellular UDP-GlcNAc levels and consequently promotes protein O-GlcNAcylation. Recent studies have described enhancement of protein O-GlcNAcylation in a wide range of cancers, such as breast and colorectal cancers, and O-GlcNAcylation has attracted attention as a new target in cancer treatments (29). However, relevant targets of O-GlcNAcylation in cancers have yet to be fully described.
To obtain a clue to overcoming the resistance to proteasome inhibitors acquired by cancer cells, we explored the possible existence of a new putative pathway that controls the magnitude or threshold of the proteasome bounce-back response by investigating NRF1-interacting proteins in the nucleus, hoping to clarify a post-ER regulation of NRF1 activity. We identified OGT/host cell factor C1 (HCF-1) complex, which turned out to be essential for the proteasome inhibitor-induced induction of proteasome subunit genes through increasing the protein level of NRF1. Meta-analysis of The Cancer Genome Atlas (TCGA) data sets of breast and colorectal cancers revealed positive correlations in protein abundance between OGT and proteasome subunits. OGT knockdown enhanced the anticancer effect of proteasome inhibitor in both culture cells and a xenograft mouse model. This study has revealed a critical contribution of O-GlcNAcylation in cancer cells to maintaining proteasome activity via NRF1 protein stabilization.

RESULTS
The OGT/HCF-1 complex interacts with NRF1. To explore the possible existence of a pathway involved in the proteasome inhibitor-mediated bounce-back response, we explored factors participating in a post-ER regulation of NRF1 activity by identifying proteins that interact with NRF1 in the nucleus. 293F cells that stably express 3ϫFLAG-tagged NRF1 (NRF1-3ϫFLAG) were established, and NRF1-containing protein complexes were purified from their nuclear extracts with anti-FLAG antibody (Fig. 1A). The purified proteins were subjected to liquid chromatography-tandem MS (LC-MS/MS) analysis. Among them, we identified OGT, an enzyme that catalyzes conjugation of O-GlcNAc to target proteins, and HCF-1, which is known as an OGT-binding partner and a chromatin-binding regulator (30)(31)(32), as novel binding proteins of NRF1. Immunoblot analysis confirmed the interaction of OGT and HCF-1 with NRF1 (Fig. 1B). Previous studies showed that OGT regulates many biological processes through O-GlcNAc modification of target proteins (27,33) and that active O-GlcNAcylation is closely related to cancer malignancy (34)(35)(36). Expecting that O-GlcNAcylation modulated the proteasome bounce-back response by regulating NRF1 activity, we focused on the contribution of the OGT/HCF-1 complex to NRF1 function. NRF1 Neh6L domain directly interacts with the C-terminal region of HCF-1. To characterize the interaction between NRF1 and the OGT/HCF-1 complex, we determined which domain of NRF1 mediated the interaction with the OGT/HCF-1 complex. Several functional domains and motifs have been defined in NRF1, as shown in Fig. 1C (20). We generated several deletion mutant molecules of NRF1 with FLAG tags (Fig. 1C) and expressed them in 293F cells. Immunoprecipitation assays using the anti-FLAG antibody showed that NRF1 M1 and M2 mutants, both lacking the Neh6L domain, failed to interact with OGT or HCF-1 proteins (Fig. 1D). The NRF1 ΔbZip mutant interacted with both OGT and HCF-1, indicating that CNC/bZip and Neh3L domains are dispens- 3ϫFLAG-NRF1 WT (1-741), 3ϫFLAG-NRF1 ΔbZip (1-592), 3ϫFLAG-NRF1 M1 (Δ464 -580), and 3ϫFLAG-NRF1 M2 (Δ243-580) were expressed in 293F cells and immunoprecipitated. (D) Detection of OGT and HCF-1 proteins interacting with NRF1 and its mutant molecules. Nuclear extracts of 293F cells expressing NRF1 and its mutant molecules were pulled down with an anti-FLAG antibody. Immunoprecipitated samples were subjected to immunoblot analysis with antibodies against OGT, HCF-1, and the FLAG tag. (E) Constructs of His 6 -tagged proteins of NRF1 mutants and GST fusion proteins of OGT and the C-terminal half of HCF-1 (HCF-1-C). Also shown are NRF1-Neh5L/AD2-His 6 (Fr. 1; 243-430) and NRF1-Neh6L-His 6 (Fr. 2; 431-580). (F) Coomassie brilliant blue staining of GST fusion proteins and His 6 -tag proteins. GST-OGT, GST-HCF-1-C, NRF1-Neh5L/AD2-His 6 (Fr. 1; 243-430), and NRF1-Neh6L-His 6 (Fr. 2; 431-580) were bacterially expressed and purified. Arrowheads indicate purified recombinant fusion proteins. (G) GST pulldown assay of NRF1-His 6 fragments using GST-OGT and GST-HCF-1-C. NRF1 fragments were detected using an anti-His 6 antibody, and OGT and HCF-1 were detected using an anti-GST antibody.
OGT-NRF1 Axis Activates Proteasome Genes Molecular and Cellular Biology able for the interaction. Thus, we concluded that the OGT/HCF-1 complex interacts with NRF1 via the Neh6L domain.
The OGT/HCF-1 complex is essential for the proteasome inhibitor-mediated bounce-back response. To examine whether the OGT/HCF-1 complex is involved in the proteasome bounce-back response by regulating NRF1 activity, we first verified a contribution of NRF1 to the expression of proteasome subunit genes in our experimental setting. Treatment with the proteasome inhibitor MG132 increased the expression levels of proteasome subunit genes in HeLa cells with control short interfering RNA (siRNA), whereas the MG132-induced upregulation was completely abrogated by NRF1 knockdown ( Fig. 2A and 3A). We then examined the contributions of OGT and HCF-1 to the bounce-back response by knocking down each factor (Fig. 2B to D). Knocking down OGT attenuated the upregulation of the proteasome subunit genes in response to MG132 (Fig. 3B). Similar results were obtained in HCF-1 knockdown cells (Fig. 3C). These results indicate that the OGT/HCF-1 complex is required for the proteasome bounce-back response and suggest that the OGT/HCF-1 complex supports the NRF1 activity.
We next examined whether recruitment of the OGT/HCF-1 complex to NRF1 was important for NRF1-mediated transcriptional activation of proteasome subunit genes  OGT-NRF1 Axis Activates Proteasome Genes Molecular and Cellular Biology by utilizing the NRF1 M1 mutant that was incapable of interacting with the OGT/HCF-1 complex ( Fig. 1C and D). Proteasome subunit genes were upregulated by exogenous wild-type NRF1; however, the upregulation was not obvious in the case of the NRF1 M1 mutant (Fig. 3D), indicating that interaction of NRF1 with the OGT/HCF-1 complex is necessary for NRF1-mediated transcriptional activation. HCF-1 is required for chromatin binding to NRF1 at promoter regions of proteasome subunit genes. NRF1 has been shown to activate proteasome subunit genes by binding to their promoter regions (8,9,37). To comprehensively assess the role of NRF1 in transcriptional regulation, we conducted chromatin immunoprecipitation sequencing (ChIP-seq) analysis in HeLa cells that were treated with MG132 by using NRF1 antibody. Consistent with previous reports, NRF1 was localized at promoter regions of almost all proteasome subunit genes (see Fig. S1A and B in the supplemental material).
To understand how the OGT/HCF-1 complex regulates the intranuclear function of NRF1, we knocked down the expression of HCF-1, which directly interacts with NRF1 and is known to be a chromatin-binding regulator (32), and examined NRF1 binding to the promoters of the proteasome subunit genes. In HeLa cells transfected with control siRNA, as observed in the ChIP-seq analysis, MG132 treatment induced robust binding of NRF1 to the promoter regions of the representative proteasome subunit genes PSMA5, PSMD11, and PSMD14 but not to a negative-control locus, GATA1 (Fig. 4). In contrast, RNA interference of HCF-1 reduced the MG132-induced chromatin binding of NRF1 (Fig. 4), suggesting that the OGT/HCF-1 complex is essential for chromatin binding of NRF1 at proteasome subunit gene loci. The OGT/HCF-1 complex is required for NRF1 protein accumulation in response to proteasome inhibition. To understand how the OGT/HCF-1 complex contributes to the chromatin binding of NRF1, we examined the accumulation of endogenous NRF1 protein in response to proteasome inhibition after knocking down OGT or HCF-1. We found that OGT knockdown in HeLa cells abrogated NRF1 protein accumulation that was induced by MG132 ( Fig. 5A and B). HCF-1 or OGT knockdown in a breast cancer cell line, MDA-MB231 cells, also abolished the MG132-induced elevation of NRF1 protein levels ( Fig. 5C and D), which were not necessarily correlated with NRF1 mRNA levels. Abundance of NRF1 mRNA was not altered by HCF-1 knockdown but was reduced by OGT knockdown (Fig. 5E and F). The results indicate that HCF-1, which recruits OGT to NRF1, is involved in the posttranscriptional regulation of NRF1 and mediates NRF1 protein accumulation in response to proteasome inhibition. OGT might be involved in the transcriptional regulation of NRF1 in an HCF-1-independent manner.
To examine the significance of the enzymatic activity of OGT, we inhibited O-GlcNAcylation by suppressing the hexosamine biosynthesis pathway (HBP) and limiting the availability of UDP-GlcNAc, a substrate of O-GlcNAcylation. An HBP-

Thus, O-GlcNAcylation catalyzed by OGT is required for the NRF1 accumulation that is induced by proteasome inhibition.
A previous study demonstrated that OGT inhibits proteasome activity by O-GlcNAcylating a subunit of the 19S proteasome subcomplex (38), implying that proteasome activity, especially that of the 19S proteasome, was increased in the OGT knockdown cells. To evaluate the possibility that OGT knockdown enhanced 19S proteasome activity and subsequently inhibited NRF1 accumulation, we used the proteasome inhibitor b-AP15, which specifically targets the 19S proteasome subunit (39). OGT knockdown and DON treatment both abolished b-AP15-induced accumulation of NRF1 in a manner similar to that induced by MG132, which targets the 20S subunit ( Fig. 6C and D). These results suggest that decreased O-GlcNAcylation limits NRF1 accumulation independently of influencing proteasomal activity. To further exclude the possibility that a decrease in the amount of the OGT/HCF-1 complex potentiated proteasome activity and nonspecifically inhibited NRF1 accumulation, we investigated the protein level of NRF2, which is predominantly regulated by the UPS (40), in MDA-MB231 cells by HCF-1 or OGT knockdown. Knocking down HCF-1 or OGT did not affect NRF2 protein accumulation in response to MG132 ( Fig. 5C and D). These results suggest that O-GlcNAcylation mediated by the OGT/HCF-1 complex is specifically critical to NRF1 accumulation in response to proteasome inhibition. We thus concluded that the OGT/HCF-1 complex supports the activation of proteasome subunit genes by enhancing the accumulation and subsequent chromatin binding of NRF1 when proteasome activity is compromised.
Enhanced cellular O-GlcNAcylation promotes NRF1 protein accumulation. We next investigated whether enhancement of cellular O-GlcNAcylation increased the endogenous abundance of NRF1 protein, because metabolic reprogramming of cancer cells often supports high levels of O-GlcNAcylation (33)(34)(35)(36)41). Cellular O-GlcNAcylation was facilitated by increasing the availability of extracellular glucose, from which UDP-GlcNAc is generated through HBP, or by inhibiting OGA, which catalyzes the removal of GlcNAc from proteins. We treated Hep3B cells with either high-glucose medium or the OGA inhibitor PugNAc to induce cellular protein O-GlcNAcylation. Both treatments robustly increased NRF1 protein accumulation (Fig. 7A). Suppression of HBP by the specific inhibitors DON and azaserine (AZA) attenuated the effect of high glucose on NRF1 accumulation (Fig. 7B). Overexpression of OGA inhibited the high-glucose- To verify whether elevated O-GlcNAcylation increased NRF1 binding to promoters of proteasome subunit genes, we performed the ChIP assay. In HeLa cells, high-glucose treatment facilitated chromatin binding of NRF1 to the promoter regions of the representative proteasome subunit genes PSMA5, PSMD11, and PSMD14 but not to the negative-control locus GATA1 (Fig. 7D), suggesting that NRF1 contributes to the activation of the proteasome subunit genes when it is accumulated in response to enhanced cellular O-GlcNAcylation.
To clarify whether the processed form of NRF1 responded to the facilitation of O-GlcNAcylation, we used 293F cells that expressed the NRF1 P1 mutant, which lacks an ER retention sequence and mainly exists in the nucleus (Fig. 7E). The level of NRF1 P1 protein was robustly increased in response to high-glucose conditions (Fig. 7F) as well as PugNAc Whole-cell extracts were prepared from Hep3B cells that were cultured in medium containing 1.5 g/ml glucose (low), 4.5 g/ml glucose (high), or 1.5 g/ml glucose with 100 M PugNAc (Low-PugNAc) for 24 h and subjected to immunoblot analysis with antibodies against NRF1, O-GlcNAc peptides, and tubulin. Hep3B cells that were treated with 10 M MG132 were used as a positive control in NRF1 detection. (B) Effect of HBP inhibitors on the protein levels of endogenous NRF1. Whole-cell extracts were prepared from 293T cells expressing lacZ shRNA or NRF1 shRNA. Cells were cultured in low-or high-glucose medium with or without HBP inhibitors, 100 M DON, or 100 M AZA for 24 h before harvest. The protein extracts were subjected to immunoblot analysis with antibodies against NRF1 and tubulin. (C) Effect of OGA expression on the protein levels of endogenous NRF1. 293T cells were transfected with an empty or a FLAG-OGA expression vector. At 24 h after transfection, the cells were cultured in high-glucose medium for another 24 h and then harvested. Whole-cell extracts were prepared and subjected to immunoblot analysis with antibodies against NRF1, O-GlcNAc peptides, the FLAG tag, and tubulin. (D) Quantitative ChIP assay at the PSMA5, PSMD11, PSMD14, and GATA-1 gene loci in HeLa cells. Chromatin localization of NRF1 was examined in each sample that was treated with low or high glucose for 24 h. The values show the enrichment of immunoprecipitated DNA relative to input DNA. Averages and SD were calculated from triplicate samples. *, P Ͻ 0.05; **, P Ͻ 0.01. ns, not significant. (E) Constructs of 3ϫFLAG fusion proteins of NRF1 WT and P1 mutant (Δ30). (F and G) Accumulation of nuclear NRF1 (NRF1 P1) by enhancing cellular O-GlcNAcylation. 293F cells expressing 3ϫFLAG-NRF1 P1 were cultured in the medium containing low or high glucose (F) and treated with 100 M PugNAc or left untreated (G). After 24 h, whole-cell extracts were prepared and subjected to immunoblot analysis with antibodies against the FLAG tag and tubulin. treatment (Fig. 7G). Thus, O-GlcNAcylation is able to act on processed NRF1 to promote its accumulation. Based on this result and intracellular localization of OGT/HCF-1 in cytoplasm and nucleus but not in ER, we surmise that NRF1 is O-GlcNAcylated in the nucleus or cytoplasm after being processed from ER. Namely, O-GlcNAcylation contributes to post-ER regulatory mechanisms of NRF1 activity.
NRF1 is O-GlcNAcylated at serine residues that are critical for phosphorylation and interaction with ␤-TrCP. Recent reports have described that many transcription factors are O-GlcNAcylated and functionally modified (42). Because transcriptional activation of proteasome subunit genes by NRF1 requires the OGT/HCF-1 complex to be recruited to the CNC-bZip transcription factor via its Neh6L domain (Fig. 3D), we hypothesized that NRF1 is O-GlcNAcylated by OGT and that this modification enables NRF1 to activate the target genes. To detect O-GlcNAcylation of NRF1, we introduced NRF1-3ϫFLAG in 293T cells, and a soluble nuclear fraction was prepared. FLAG-tagged NRF1 was O-GlcNAcylated sufficiently by endogenous OGT (Fig. 8A).
We moved on to identify amino acids in NRF1 that are the targets for O-GlcNAcylation. Previous reports stated that NRF1 interacts with Fbw7 and ␤-TrCP ubiquitin E3 ligases, resulting in the ubiquitination and degradation of NRF1 (25,26). Fbw7 and ␤-TrCP have been shown to recognize phosphorylated serine residues of NRF1, S350 and S448/S451, respectively ( Fig. 8B and C) (25,26). As O-GlcNAcylation is considered to compete with phosphorylation for Ser/Thr target sites (27), we hypothesized that O-GlcNAcylation of these serine residues stabilizes NRF1 by antagonizing their phosphorylation and consequently protecting NRF1 from the Fbw7-and ␤-TrCP-mediated degradation. To prove this hypothesis, S350 and S448/S451 were replaced with alanine, and three NRF1 mutant molecules, A-SS, S-AA, and A-AA, were constructed ( Fig. 8B and C). FLAG-tagged NRF1 (WT) and its mutant molecules (A-SS, S-AA and A-AA) were stably expressed in 293T cells.
To examine the importance of the serine residues in O-GlcNAcylation-induced stabilization of NRF1, we treated these 293T cells with high glucose or PugNAc, both of which enhance O-GlcNAcylation. The NRF1 WT and A-SS proteins were detected in nuclear extracts of the cells treated with high glucose but not in those treated with low glucose (Fig. 8D). In contrast, S-AA and A-AA mutants were accumulated even under the low-glucose condition, and the high-glucose treatment did not increase these protein levels any further (Fig. 8D). Similarly, the NRF1 WT and A-SS protein levels were increased in response to PugNAc treatment, but the S-AA and A-AA mutant proteins did not show any further increase (Fig. 8E). These results suggest that S448/S451 are required for the destabilization of NRF1 under low O-GlcNAcylation activity and the stabilization of NRF1 in response to the enhancement of cellular O-GlcNAcylation.
Conversely, decreased cellular O-GlcNAcylation by overexpression of OGA inhibited the high-glucose-induced accumulation of the NRF1 WT and A-SS proteins (Fig. 8F). Alanine substitution of S448/S451 abrogated the response to OGA, suggesting again that S448/S451 are required for NRF1 destabilization under low O-GlcNAcylation activity. Indeed, O-GlcNAcylation was dramatically suppressed in the NRF1 S-AA mutant compared with that of wild-type NRF1 (Fig. 8G), indicating that S448/S451 are essential for NRF1 O-GlcNAcylation and suggesting that S448/ S451 are O-GlcNAcylation target sites of NRF1.
O-GlcNAcylation of NRF1 interferes with the interaction between NRF1 and ␤-TrCP ubiquitin E3-ligase. Because ␤-TrCP interacts with NRF1 by recognizing phosphorylated S448/S451, it was hypothesized that O-GlcNAcylation of S448/S451 disrupts the interaction between NRF1 and ␤-TrCP. To test this hypothesis, we conducted an immunoprecipitation assay using nuclear extracts of 293T cells expressing the NRF1 WT or A-SS proteins. The cells were treated with 1 M MG132 for 4 h in the presence or absence of PugNAc before harvesting. When the cells expressing wild-type NRF1 were used, ␤-TrCP was detected as an NRF1-interacting protein under the control condition together with OGT and HCF-1, whereas PugNAc treatment remarkably decreased the amount of ␤-TrCP in the NRF1 complex (Fig. 8H). Thus, we infer that O-GlcNAcylation at S448/S451 induced by PugNAc disrupts the association between ␤-TrCP and NRF1.
When cells expressing NRF1 S-AA were used, ␤-TrCP was not detected in the NRF1 S-AA complex, irrespective of the presence or absence of PugNAc. The alanine substitution can be interpreted as mimicking the effects of O-GlcNAcylation by preventing phosphorylation of S448/S451. In agreement with the decreased association between ␤-TrCP and NRF1 in the presence of PugNAc, polyubiquitination of NRF1 was suppressed in PugNAc-treated cells (Fig. 8I). 293T cells that were stably transfected with an empty vector or an expression vector of the WT or S-AA mutant of NRF1-3ϫFLAG were pretreated with 100 M PugNAc or left untreated for 24 h before nuclear extracts were prepared for immunoprecipitation with an anti-FLAG antibody. Immunoprecipitated proteins were subjected to immunoblot analysis with antibodies against ␤-TrCP, OGT, HCF-1, and the FLAG tag. (I) Effects of PugNAc on ubiquitination of NRF1. 293F cells that were stably transfected with an empty vector or an expression vector of NRF1-3ϫFLAG were pretreated with or without 100 M PugNAc for 24 h and incubated with 10 M MG132 for another 4 h before whole-cell extracts were prepared for immunoprecipitation with an anti-FLAG antibody. Immunoprecipitated proteins were subjected to immunoblot analysis with antibodies against NRF1 and the FLAG tag.
The data obtained so far indicated that OGT associates with NRF1 through HCF-1 and catalyzes O-GlcNAcylation of NRF1, resulting in the release of ␤-TrCP from NRF1 and the consequent stabilization of NRF1.
Protein abundances of OGT and proteasome subunits are positively correlated in breast invasive carcinoma and colorectal adenocarcinoma. Elevated expression of OGT in cancers has been shown to be advantageous for their tumorigenesis and metastasis (34,43). Another line of evidence has indicated that certain types of cancers depend heavily on proteasome activity for their aggressive growth (5,6). These studies suggest that enhanced activities of OGT and proteasome both contribute to cancer malignancy. Because our finding indicates that the increased activity of OGT elevates expression levels of proteasome subunit genes through NRF1 stabilization, we examined whether a correlation exists between the abundance of OGT protein and that of proteasome subunits using the cBioPortal to examine clinical cancer cases that were enrolled in the TCGA database (44,45).
We analyzed proteomic data sets that were constructed from 102 breast invasive carcinoma cases and found positive correlations between OGT and the majority of proteasome subunits ( Fig. 9A and B, left, and Table S1). In 62 cases of colorectal adenocarcinoma that expressed a detectable amount of OGT protein, positive correlation was obtained between OGT and almost all proteasome subunits that were examined ( Fig. 9A and B, right, and Table S2). These results strongly suggest that OGT controls proteasome subunit expression in cancer cells.
Suppression of the OGT/HCF-1 complex improves the efficacy of a proteasome inhibitor as an anticancer drug. Proteasome inhibitors, including bortezomib, have been used as effective anticancer drugs for certain types of cancers that possess high proteasome activity (4). Previous reports that NRF1 confers resistance to proteasome inhibitors by mediating the proteasome bounce-back response (7-10), in combination with the results that we have obtained so far, strongly suggest that OGT inhibition augments the anticancer effects of proteasome inhibitors.
To prove this hypothesis, we knocked down OGT in two different cancer cell lines, MDA-MB-231 and NCI-H460, and examined the amount of bortezomibinduced cell death. Increasing concentrations of bortezomib were applied to control and OGT knockdown cells. As we expected, OGT knockdown cells were more sensitive to bortezomib than control cells (Fig. 10A to C). HCF-1 knockdown similarly sensitized MDA-MB-231 cells to bortezomib ( Fig. 10D and E). To verify that this sensitization effect was due to the insufficient stabilization of NRF1, NRF1 was exogenously overexpressed in OGT knockdown cells. The NRF1-overexpressing cells turned out to be resistant to bortezomib irrespective of the OGT knockdown (Fig.  10F), indicating that excessive NRF1 canceled the OGT knockdown-mediated sensitization to bortezomib.
Finally, we examined whether OGT suppression improved efficacy of proteasome inhibitors in a xenograft mouse model. To this end, we established NCI-H460 cells expressing control or OGT short hairpin RNAs (shRNAs) in a doxycycline-inducible manner (Fig. 11A). The NCI-H460 cells were subcutaneously transplanted to nude mice, and the shRNAs were induced by treating the mice with doxycycline. After 15 days of transplantation, bortezomib treatment was initiated for tumors whose sizes ranged from 100 to 300 mm 3 . The tumor size of OGT knockdown cells was significantly reduced by bortezomib treatment on day 6, whereas that of control cells was not affected (Fig.  11B). When fold changes in the tumor volume on days 3 and 6 against the one on day 0 were calculated, significant suppressive effects of bortezomib were also apparent for the OGT knockdown cells but not for the control cells (Fig. 11C). Thus, OGT suppression enhanced the antitumorigenic effect of bortezomib.
From these results, we concluded that O-GlcNAcylation is required for the NRF1-dependent proteasome bounce-back response and that OGT inhibition sensitizes cancer cells to proteasome inhibitors by antagonizing NRF1 stabilization.

DISCUSSION
Our study has revealed that an O-GlcNAcylation signal makes a critical contribution to the proteasome bounce-back response via increasing NRF1 stability (Fig. 12). Protein O-GlcNAcylation status is determined by the balance between OGT and OGA activities as well as the availability of the substrate UDP-GlcNAc. When OGT activity predominates with a sufficient supply of UDP-GlcNAc, O-GlcNAcylated NRF1 releases ␤-TrCP ubiquitin E3-ligase from the NRF1 complex, resulting in the stabilization of NRF1 and upregulation of proteasome subunit genes. Thus, the O-GlcNAcylation level is critical for the maintenance of proteasome activity by regulating NRF1 stability. Our discovery of the interaction between NRF1 and the OGT/HCF-1 complex has added a new regulatory axis to the proteasome bounce-back response and identified OGT as a new therapeutic target for sensitizing cancer cells to proteasome inhibitors.
From the viewpoint of cancer metabolism, facilitation of glycolysis and glutamin- olysis in cancer cells provides a high availability of UDP-GlcNAc, resulting in an advantageous metabolic environment for O-GlcNAcylation (41). Thus, in cancer cells, OGT is likely to exert its full activity without being restricted by substrate limitation, which would positively correlate the OGT protein level with the O-GlcNAcylation activity, the NRF1 protein level, and the expression levels of proteasome subunit genes. This scenario may explain the strong positive correlations between protein abundances of OGT and most of the proteasome subunits in breast and colorectal cancers (Fig. 9). We propose that metabolic reprogramming of cancers allows highly expressed OGT to maximally contribute to O-GlcNAcylation and to eventually develop dependence on the UPS and resistance to proteasome inhibitors by enhancing NRF1-mediated transcriptional activation of proteasome subunit genes. A previous study reported that NRF1 is O-GlcNAcylated and destabilized by this modification (46), which contrasts with our observation. One possible explanation of Averages and SD were calculated from quadruplicate samples. Relative absorbances of samples that were treated with 10 Ϫ11 M bortezomib were set to 1. **, P Ͻ 0.01. (C and E) Endogenous NRF1 accumulation in response to bortezomib treatment in MDA-MB-231 cells. MDA-MB-231 cells that were transfected with control siRNA, OGT siRNAs (C), or HCF-1 siRNAs (E) were treated with 10 nM bortezomib at 72 h after the transfection. After 4 h, nuclear extracts were prepared. Lamin B was used as a loading control. (F) Effects of NRF1 overexpression on OGT knockdown-induced sensitization to bortezomib. 293F cells expressing NRF1-3ϫFLAG or containing an empty vector (mock) were transfected with control siRNA or OGT siRNAs. At 24 h after the transfection, the cells were reseeded in 96-well plates. The next day, the cells were treated with the indicated concentrations of bortezomib for 48 h. Cell viability was assessed using a trypan blue exclusion test. Averages and SDs were calculated from the results of three independent experiments. Viable cell numbers for samples that were treated with 10 Ϫ11 M bortezomib were set to 100%. **, P Ͻ 0.01. this discrepancy would be a different antibody used in each study. The antibody used in their study for detecting O-GlcNAcylated protein (CTD-110.6) has been reported to cross-react with N-linked GlcNAcylation (47), while the one used in our study (RL-2) has higher specificity. NRF1, in its inactive state, is N-glycosylated and anchored in the ER membrane, resulting in the ERAD-mediated degradation (16,19). De-N-glycosylation generates an active form of NRF1 in the cytoplasm and nucleus, allowing transcriptional activation of its target genes. Thus, if N-linked GlcNAcylation had been detected by CTD-110.6 antibody instead of O-GlcNAcylation, the modification would have been linked to destabilization of NRF1.
In contrast, a recent paper describes that O-GlcNAcylation of NRF1 by OGT increases NRF1 stability (48), which is consistent with our results. This paper suggested that HCF-1 directly interacts with NRF1 and intercalates between OGT and NRF1 for O-GlcNAcylation of NRF1, and our study further verified the direct interaction of NRF1 and HCF-1 using recombinant proteins (Fig. 1G). The positive impact of OGT on the activity of NRF1 has been well supported by two other independent reports that exploited siRNA screenings. OGT was included in the list of factors that confer resistance to bortezomib (49) and identified as one of the factors that promote translocation of NRF1 to the nucleus (22); both of these behaviors are consistent with the notion that we propose in this study, namely, that OGT confers bortezomib resistance on cancers by increasing NRF1 stability and thereby elevating the expression of proteasome subunit genes.
Based on the observation that the ER retention sequence of NRF1 is dispensable for the accumulation of NRF1 protein in response to the facilitation of O-GlcNAcylation (Fig. 7F and G), we consider that O-GlcNAcylation of NRF1 leads to its activation through the post-ER regulation of NRF1 activity, although we cannot completely exclude the possibility that the OGT-HCF1 complex also modulates the NRF1 activation machinery coupled with ERAD. In contrast, N-glycosylation of NRF1 is found to occur in the ER and has a negative impact on NRF1 activity. De-N-glycosylation of NRF1 constitutes the activation process of NRF1 at the ER membrane (15-17, 19-21), raising the possibility that lowering glucose availability is beneficial for NRF1 activation by limiting the amount of substrate for N-glycosylation. Indeed, de-N-glycosylation of NRF1 under no-glucose conditions has been observed to increase NRF1 activity (21). Considering our result that O-GlcNAcylation of NRF1 under high-glucose conditions also increases NRF1 activity, an integrated understanding of N-glycosylation and O-GlcNAcylation of NRF1 in relation to glucose availability must wait for further investigation.
Our study strongly suggests that OGT inhibition provides a promising clue to anticancer therapy by suppressing the NRF1-mediated proteasome bounce-back response. In addition, wide-ranging roles of OGT in cancer development and progression have been described (50,51), providing a rationale for targeting OGT in anticancer therapies. Currently, no clinical data are available regarding correlations between OGT activity and therapeutic efficacy of proteasome inhibitors. It is worthwhile to examine O-GlcNAcylated protein levels in multiple myeloma and other cancer types in relation to their sensitivities to proteasome inhibitors.
In terms of the suppression of the proteasome bounce-back response, an alternative target of OGT has been implicated. Sharing the same DNA recognition sequence, the antioxidant response element (ARE) with NRF1, NRF2 was reported to be another regulator of proteasome subunit genes (52). A recent study has reported that a gain-of-function mutant of p53 serves as a coactivator of NRF2 and makes NRF2 coordinately activate the proteasome subunit genes through the ARE in breast cancer cells (53). Because O-GlcNAcylation has been shown to stabilize p53 by interfering with its interaction with the Mdm2 ubiquitin E3-ligase (54), inhibition of OGT would disrupt the p53-NRF2 pathway by destabilizing p53 and reducing the expression of proteasome subunit genes. Thus, OGT inhibition is likely to be effective in repressing the proteasome bounce-back response by suppressing NRF2 activity as well as that of NRF1.
Cellular O-GlcNAcylation is regulated by balancing the functions of OGT and OGA. Dysregulation of O-GlcNAc cycling is often found under various pathological conditions other than cancers, such as diabetes and neurodegenerative diseases (55,56). Thiamet-G, a newly developed OGA inhibitor, has been shown to slow progression of Alzheimer's disease by an increase in tau O-GlcNAcylation and the reciprocal decrease in tau phosphorylation and aggregation (57). Nevertheless, it is also plausible that O-GlcNAcylation of NRF1 improves proteostatic failure by enhancing proteasome activity and contributing to the suppression of neurodegeneration. Our study has revealed that NRF1 transduces O-GlcNAcylation status into proteasome activity and suggested that O-GlcNAc cycling is a promising therapeutic target for diseases that feature proteostatic disturbances via the regulation of NRF1 activity.
Retroviral infection. pQC-mNRF1 (WT, S-AA, A-SS, and A-AA)-3xFLAG and pQCXIP (an empty control vector) were transfected into Plat-A cells. At 24 h after the transfection, the medium was replaced with fresh medium and cultured for another 24 h. The medium was harvested and used as the medium that contained retrovirus particles.
Immunoprecipitation of NRF1 and its mutant molecules. Soluble nuclear extracts were prepared from 293F cells that expressed mNRF1-WT, mNRF1-ΔbZip, mNRF1-M1, mNRF1-M2, and mNRF1-P1 with 3ϫFLAG tag at their N termini and from 293T cells that expressed mNRF1-WT, mNRF1-A-SS, mNRF1-S-AA, and mNRF1-A-AA with 3ϫFLAG tag at their C termini. The nuclear extracts were subjected to immunoprecipitation with anti-FLAG antibody. The immunoprecipitated samples were analyzed using immunoblot analysis.
GST pulldown assay. GST pulldown assays were performed as described previously (60). To detect the interaction of OGT and HCF-1 with NRF1 fragments (Fr. 1 and Fr. 2), glutathione-Sepharoseimmobilized GST-OGT and GST-HCF-1-C were incubated with recombinant NRF1 fragments Fr. 1 and Fr. 2 and washed extensively with PBS-T. Subsequently, pulldown proteins were eluted in Laemmli sample buffer at 94°C. Eluates were resolved by SDS-PAGE and analyzed by immunoblot assay.
Identification of NRF1-interacting proteins. Nuclear extract from mNRF1-3ϫFLAG-expressing 293F cells was prepared. The nuclear extract was subjected to anti-FLAG affinity purification. The mNRF1-3ϫFLAG complex was eluted using a 3ϫFLAG peptide according to the manufacturer's protocol (Sigma).
The eluate was subjected to gel-based high-performance LC-MS/MS analysis, and NRF1-associated proteins were identified by searching the protein sequence database.
Gel-based LC-MS/MS analysis and protein sequence database searches. The detailed protocol for gel-based LC-MS/MS analysis and protein sequence database searches was published previously (61,62). Briefly, after SDS-PAGE using a 5 to 20% polyacrylamide gradient gel (Oriental Instruments) and Coomassie brilliant blue staining (63), each lane in the gel was divided into 18 sections. The resulting gel blocks were treated with dithiothreitol (DTT) and acrylamide to reduce and alkylate the sulfhydryl groups. After overnight tryptic digestion, the resulting peptides in each gel block were extracted, and half of each sample was subjected to LC-MS/MS using an LTQ Orbitrap Velos mass spectrometer (Thermo Scientific). The data acquisition of every sample was performed for 60 min after a 50-min LC gradient was started; MS 1 scans from m/z ϭ 321 to 1,600 were carried out in the Orbitrap with the resolution set at 60,000 and the lockmass at m/z ϭ 445.120025, followed by top-15 MS 2 acquisition by collision-induced dissociation (CID) in the ion trap in normal-resolution mode. The settings for the MS 2 scans were the following: minimal signal intensity required, 500; AGC target, 5,000; maximum ion injection time, 50 ms (64). The raw data files that were derived from samples in the same SDS-PAGE lane were converted together into a single MASCOT generic format file and were used for the database search by MASCOT (version 2.5.1; Matrix Science) against the mouse proteins in Swiss-Prot (July 2016) and a custom database that included mouse NRF1 protein and contaminant proteins. The peptide expectation value cutoff was set at 0.05. Protein N-terminal acetylation (ϩ42.0106) and oxidation of methionine (ϩ15.9949) were considered variable modifications, and propionamidated cysteine (ϩ71.0371) was set as a fixed modification. The false discovery rates (FDR) were automatically adjusted to 1% by the MASCOT Percolator in every search.
ChIP-seq analysis. Sequencing libraries were prepared from 1.0 ng of DNA subjected to ChIP and input samples by a Mondrian SPϩ system (Nugen) with an Ovation SP ultralow DR multiplex system (Nugen). The libraries were further purified and size-selected using an AMPure XP kit (Beckman Coulter) and were quantified by a quantitative MiSeq (qMiSeq) method (65). The samples were sequenced on a HiSeq 2500 (Illumina) that generated 101-base reads. Sequencing data were aligned with the hg19 reference genome with Bowtie2 (66), and peaks were called with MACS2 (67). ChIP-seq peak visualization was done with Integrative Genomic Viewer (68).
ChIP assay. ChIP assays were performed with control, HCF-1, or NRF1 siRNA-treated or nontreated HeLa cells using anti-NRF1 antibody (D5B10; Cell Signaling). The cells were treated with dimethyl sulfoxide (DMSO) or 1 M MG132 for 4 h or with low/high glucose for 24 h and cross-linked with 1.5 mM ethylene glycol bis(succinimidyl succinate) (Thermo Scientific) for 20 min, followed by treatment with 1% formaldehyde for 10 min. The samples were then lysed and digested with micrococcal nuclease (New England BioLabs) to shear DNA. The nuclear lysis solution was incubated overnight with anti-NRF1 antibody, followed by incubation with an equal amount of Dynabeads anti-rabbit IgG (Life Technologies). Precipitated DNA was analyzed by real-time PCR using the primer sets described in Table 1.
RNA purification and quantitative reverse transcription-PCR. Total RNA samples were prepared from cells using Isogen (Nippon Gene) as previously described (60). First-strand cDNA was synthesized from 0.5 g of total RNA using ReverTra Ace quantitative PCR reverse transcription master mix with genomic DNA remover (Toyobo). Real-time PCR was performed in triplicate for each sample with the StepOnePlus real-time PCR systems (Applied Biosystems) using primers listed in Table 2. Expression levels of rRNA were used as internal controls for normalization. siRNA transfection. HeLa, MDA-MB-231, NCI-H460, and 293F cells were transfected with 50 nmol of siRNA using RNAiMAX (Invitrogen) according to the manufacturer's protocol. Cells were treated with MG132, bAP-15, or DMEM at 72 h after the transfection unless otherwise stated. Mission predesigned siRNAs targeting NRF1 (SASI_Hs01_00082559 and SASI_Hs01_00082561), OGT (SASI_Hs01_00141132 and SASI_Hs01_00141134), and HCF-1 (SASI_Hs01_00053487 and SASI_Hs01_00053490) were purchased from Sigma-Aldrich.
Meta-analysis of breast and colorectal cancer cases. TCGA data sets (TCGA; Provisional) of breast invasive carcinoma and colorectal adenocarcinoma were used for analyzing the correlation of protein abundance in the frame of cBioPortal (44,45).
Cell viability study. The cell viability of MDA-MB-231 and NCIH460 cells after bortezomib treatment was determined by using a cell counting kit 8 (Nacalai Tesque) according to the manufacturer's protocol or the trypan blue exclusion test using a hemocytometer.
Xenograft experiment. A total of 1 ϫ 10 6 NCI-H460 cells expressing control or OGT shRNA in a doxycycline-inducible manner were subcutaneously transplanted with Matrigel to flanks of 4-week-old GCTGCTGTTGCCTCTGTCTT GCCTGCCTTCTGGGTCTTAC GATA1 GCCTCAACTGTGTGTCCCAC GAAGGTACTGGAAAAGTCAG male BALB/c nu/nu mice. Twenty-four hours after the transplantation, the mice were administered 1 mg/ml doxycycline in 5% sucrose as drinking water. After 15 days of the transplantation (day 0 in Fig. 11B), 20 g bortezomib in 100 l phosphate-buffered saline (PBS) or 100 l PBS as vehicle was directly injected into tumors whose volumes ranged from 100 to 300 mm 3 . The second injection of bortezomib was conducted after 3 days of the initial treatment (day 3 in Fig. 11B and C). Tumor volumes were measured 3 and 6 days after the initial treatment with bortezomib. All animals were housed under specific-pathogen-free conditions according to the regulations of the standards for human care and use of laboratory animals of Tohoku University and the guidelines for proper conduct of animal experiments of the Ministry of Education, Culture, Sports, Science, and Technology of Japan (http://www.mext.go.jp /b_menu/hakusho/nc/06060904.htm). Statistical analysis. The quantitative data are presented as the means Ϯ standard deviations (SD). To evaluate statistical significance, Student's t test was used for experiments depicted in Fig. 3, 4, and 7, and analysis of variance with Tukey-Kramer's test was used for Fig. 10 and 11.
Accession number(s). The data have been deposited at GEO-NCBI under the accession number GSE108856.

SUPPLEMENTAL MATERIAL
Supplemental material for this article may be found at https://doi.org/10.1128/MCB .00252-18.