Tumor-derived Mutations in the Gene Associated with Retinoid Interferon-induced Mortality (GRIM-19) Disrupt Its Anti-signal Transducer and Activator of Transcription 3 (STAT3) Activity and Promote Oncogenesis*

Background: Aberrantly active STAT3 promotes tumorigenesis. GRIM-19 binds to STAT3 and inhibits its growth promotion. Results: We identified three mutations in the GRIM-19 gene that failed to block STAT3-dependent gene expression and tumor development. Conclusion: GRIM-19 mutations unleash STAT3 activity to promote tumor growth. Significance: This study identifies a new mechanism by which normal cells acquire cancerous properties. The signal transducer and activator of transcription 3 (STAT3) protein is critical for multiple cytokine and growth factor-induced biological responses in vivo. Its transcriptional activity is controlled by a transient phosphorylation of a critical tyrosine. Constitutive activation of STAT3 imparts resistance to apoptosis, promotes cell proliferation, and induces de novo micro-angiogenesis, three of the six cardinal hallmarks of a typical cancer cell. Earlier we reported the isolation of GRIM-19 as a growth suppressor using a genome-wide expression knockdown strategy. GRIM-19 binds to STAT3 and suppresses its transcriptional activity. To understand the pathological relevance of GRIM-19, we screened a set of primary head and neck tumors and identified three somatic mutations in GRIM-19. Wild-type GRIM-19 suppressed cellular transformation by a constitutively active form of STAT3, whereas tumor-derived mutants L71P, L91P and A95T significantly lost their ability to associate with STAT3, block gene expression, and suppress cellular transformation and tumor growth in vivo. Additionally, these mutants lost their capacity to prevent metastasis. These mutations define a mechanism by which STAT3 activity is deregulated in certain human head and neck tumors.

Aberrant growth promotion occurs because of the inactivation of tumor suppressors and activation of oncogenes. Classical tumor suppressors such as p53, pRB, and PTEN are frequently mutated in many human tumors. The Hanahan-Weinberg model (1) suggests that at least 10 different genetic and physiologic alterations occur in a mammalian cell before reaching the full-blown malignant state. These include resistance to apoptosis, enhanced motility, development of neovasculature, activation of tumor-proliferating inflammation, and loss of anti-tumor immunity. Several of these processes are dependent on cytokines and/or other secretory factors.
The interferon (IFN) family of cytokines is a critical player in promoting host defenses against pathogens and neoplastic cells (2). Unlike many other cytokines, which primarily attack the tumor through activation of immune cells, IFNs not only activate direct growth-suppressive gene expression in the tumor cells but also activate immune cells (3). Their direct anti-tumor effects include the inhibition of cell cycle and activation of apoptosis. A number of tumor suppressor genes such as STAT1 (4), IRF1 (5), IRF7 (6), IRF8 (7), and DAPK1 (8) were discovered originally as critical players in IFN action. That endogenous IFNs act as sentinels against tumors (9), many cell types are capable of producing and responding to IFNs; and the fact that most of the above proteins are transcription factors it is likely that we have not fully uncovered the tumor-suppressive gene spectrum controlled by these cytokines. In addition, we and others have shown that IFN in association with other biological response modifiers, such as retinoic acid, exhibit potent tumor suppression in several clinical and preclinical models (2). To investigate the underlying mechanisms of tumor growth suppression, we employed a genome-wide knockdown strategy and identified few potent growth suppressors. One such growth inhibitor was GRIM-19, a novel protein whose depletion promoted and overexpression suppressed tumor growth (10,11). GRIM-19 targets STAT3 to inhibition via a direct interaction (12,13).
STAT3 was originally identified as a cytokine-inducible transcription factor whose activity was critical for embryonic development and differentiation (14). In normal resting cells, STAT3 is transiently activated via tyrosyl phosphorylation by the JAK family of protein-tyrosine kinases upon specific ligand-receptor engagement, which causes STAT3 homo-dimerization, nuclear migration, and transcriptional induction of target genes (14). Excessive STAT3 phosphorylation is prevented by cessation of JAK activity and nuclear export of active STAT3 (14). In contrast, STAT3 is highly active and is primarily found in the nucleus of a number of human tumors. Consistent with these observations, an experimentally generated spontaneously dimerizing STAT3 transforms cells to an oncogenic state (15). STAT3 meets at least 5 of the 10 criteria of the Hanahan-Weinberg model of tumorigenesis such as promotion of proliferation and neoangiogenesis, resistance to apoptosis, evasion of immune response, and development of tumor-promoting inflammatory response (16). Although STAT3 tyrosyl phosphorylation is one important step in growth promotion, mechanisms that deregulate STAT3 activity are far from clear, as mutations in STAT3 are not reported in human tumors to date. Therefore, we hypothesized that inactivation of STAT3 inhibitors is an important event in promoting tumor growth.
Earlier we showed that expression of GRIM-19 is suppressed in a number of primary renal cell carcinomas (10). Whether GRIM-19 mutations occur in human tumors is unclear at this stage. Here we show that the GRIM-19 gene is mutated in primary human head and neck cancers. We also show that the tumor-derived mutants fail to counteract the growth-promoting effects of STAT3 and are ineffective at suppressing metastasis due to their weak binding to STAT3. These studies not only show a new way tumor cells lose control over STAT3 but also highlight the importance of GRIM-19 mutations in tumor development.

EXPERIMENTAL PROCEDURES
Antibodies-Antibodies specific for FLAG (Sigma) and Myc (Cell signaling technology) epitopes, actin (Sigma), and cyclin D1, Bcl2, and MMP-9 (Santa Cruz Biotechnology) were used in these studies. Alexa fluor 700-coupled anti-rabbit IgG and Alexa fluor 750-coupled anti-mouse IgG (Invitrogen) were used to capture specific signals using the Odyssey Infrared system (LI-COR), and band intensities were quantified using the manufacturer's software.
Tumor Tissues-Tumors were collected with informed consent of the patients undergoing surgical excision at the Cleveland Clinic under an institutionally approved protocol. Adjacent normal tissue was used as a control in each case. Samples were snap-frozen in liquid N 2 and then stored at Ϫ80°C until use. Total RNA and genomic DNA were isolated from these samples using a commercially available Allprep DNA/RNA/ Protein mini kit (Qiagen, Inc.). Total RNA was converted to cDNA using SuperScript III reverse transcriptase (Invitrogen) and used templates for qualitative and quantitative PCR using gene-specific primers (supplemental Table S1). GRIM-19 ORF from tumor and normal tissue was PCR-amplified and sequenced at the Biopolymer/Genomics core facility, University of Maryland School of Medicine, Baltimore, MD. Base changes, if any, were verified using matched genomic DNA as the background. The presence or absence of such base changes in the matched normal tissue was used to score changes as germ line or somatic origin, respectively.
Expression Vectors-A FLAG-tagged spontaneously dimerizing STAT3 variant (S3C) has been described (15), and transfected cells were selected with G418. Tumor-derived GRIM-19 mutant ORFs were expressed as C-terminal myc-tagged protein using the pLVX-Puro expression vector (Clontech, Inc). Preparation of lentiviral particles were as per our earlier reports. Briefly, medium from the cultures were collected for 3 consecutive days, pooled, passed through a 0.45-m filter, and used to transduce target cell lines.
Establishment of Cell Lines-3Y1, a non-oncogenic rat fibroblast cell line (JCRB0734, Japanese Collection of Research Bioresources, Osaka, Japan) (17), was grown in DMEM supplemented with 5% FBS, 100 units/ml penicillin, and 100 mg/ml streptomycin. This cell has two point mutations (K130T and A136T) in one of the p53 alleles (18). These mutations do not belong to the category of "hotspot mutations" observed in human tumors that destroy its DNA binding capacity (19). HaCaT (non-oncogenic keratinocytes) was grown in DMEM supplemented with 10% FBS. HSC3 cells (human oral squamous carcinoma cell line) was grown in DMEM supplemented with 10% FBS; this cell line harbors mutant the p53 allele(s). PC3 cells (prostate cancer cell line) were grown in minimum Eagle's medium supplemented with 10% FBS, glutamine, and sodium pyruvate. Cells were electroporated with the S3C expression plasmid using the Nucleofector technology (Amaxa, Inc.). After transfection with the genes of interest, cells were selected with G418 (1500 g/ml) for 10 -12 days. Drugresistant colony pools (n Ͼ 50) were expanded to avoid a clonal bias and transduced with lentiviral GRIM-19 and selected with puromycin. Expression-positive population was used in all experiments. A lentiviral luciferase reporter (pCCL-c-MNDU3c-Luc) (20) was used to transduce cells and monitor metastasis in real time.
Gene Expression Analyses-Total RNA (5 g) was reversetranscribed using a commercially available ImProm-II enzyme (Promega). Quantitative RT-PCR was performed using Jump-Start SYBR Green master mix (Sigma) on Stratagene Mx3005P real-time PCR machine. Input cDNA was 10 ng per reaction, and relative transcript abundance was calculated using RPL32 as the internal control by the ⌬⌬Ct method. For all cell linebased experiments, three separate batches of RNA (n ϭ 3) were employed for a sample. Multiple technical replicates per sample were used for calculating transcript levels, and the statistical significance of differences between various groups was assessed using Student's t test.
Soft Agar Colony Formation, in Vitro Wound Healing, and Cell Migration Assays-These were performed as described in our previous publications (21). Chromatin immunoprecipitation (ChIP) analyses were performed as described earlier (21).
Tumorigenic Assays-Three-to four-week-old athymic nude (nu/nu) NCr mice (Taconic) were used for these studies as described earlier (22). Procedures involving animals were conducted in conformity with an institutionally approved protocol compliant with United States National Institutes of Health policies. Each experimental group contained 10 mice, with 1 tumor injection site per mouse. Cells (1 ϫ 10 6 ) were inoculated into flanks, and tumor development was monitored for several weeks. Every week, tumor volume was calculated using caliper measurements and the formula V ϭ (4/3)a 2 b, where 2a ϭ minor axis, and 2b ϭ major axis of prolate spheroid. Student's two-tailed t test was used to assess the statistical significance of difference between pairs of means of tumor volume. For metastasis experiments, after laparotomy, 5 ϫ 10 5 PC3 cells in 10 l were inoculated into the ventral lobe of the prostate gland using a 27-gauge needle. These cells also carried a stably integrated firefly luciferase gene, which permits live imaging. Animals were monitored for metastasis after 45 days using the Xenogen IVS100 live imaging system. Mice were anesthetized with inhaled isoflurane and administered 3 mg of luciferin intraperitoneally 5 min before imaging. Image acquisition settings were: Bin 8, FOV25, f1, 30-s exposure, background subtracted, flatfielded, cosmic.
Statistical Analyses-Data obtained with each mutant were subjected to Student's t test with respect to wild-type GRIM-19. Effects of wild-type GRIM-19 on S3C were compared with corresponding data from S3C/EV cells. A p value Ͻ0.05 was considered significant in all experiments.

Identification of Mutations in GRIM-19 in Human Oral
Squamous Cell Carcinomas-We recently obtained several primary human oral squamous cell carcinomas (n ϭ 12) and adjacent normal tissues from patients who were tobacco chewers. Total RNA and genomic DNA were isolated from pathologistcertified surgically isolated tumors and the adjacent normal tissues. RT-PCR analysis showed GRIM-19 mRNA levels in some of these tumors was lower compared with their matched normal tissue, whereas there was no significant difference in others. Three of these tumors (from different patients and arbitrarily named as sample IDs 1, 2, and 3) selected for this study had higher GRIM-19 than their normal counterparts (Fig. 1A, top blot). Based on our previous studies that showed a loss of GRIM-19 correlates with elevated STAT3 levels in clear cell renal cell carcinomas (10), we expected lower STAT3 mRNA levels in these tumors compared with their control. Contrary to this notion, STAT3 mRNA levels were very high compared with their matched normal tissue (Fig. 1A, middle blot). Real-time PCR analysis (Fig. 1B) of two STAT3-regulated genes, BCL2-L1 (BCL-XL) and CCND1 (cyclin D1), showed increased expression, which was consistent with elevated STAT3 levels. Based on this confounding observation that high GRIM-19 levels in a tumor corresponded to an increased STAT3 activity, we hypothesized that the GRIM-19 in these tumors may have suffered mutation and consequently lost its anti-STAT3 activity. Therefore, we sequenced the GRIM-19 cDNA and genomic DNA from the selected tumors and their matched normal tissues. These studies identified three separate point mutations in the tumor GRIM-19 gene but not in those of normal cells (supplemental Fig. S1). Because these mutations were absent in the matching normal tissues, they truly appear to be tumor-derived and somatic in nature. Each mutation was identified from a tumor derived from one patient. Within the given tumor, the mutant GRIM-19 ORF appears to be the dominant form. No other mutations were observed in the same tumor. These mutations resulted in the following amino acid substitutions in the coding region: L71P ( Fig. 1 sample ID 2), L91P ( Fig. 1 sample ID 1), and A95T ( Fig. 1 sample ID 3). Because the available patient sample size is smaller, we could not determine the frequency of their occurrence at this time.
Tumor-derived GRIM-19 Mutants Fail to Suppress STAT3induced Cell Growth-We first determined if the GRIM-19 mutants had an effect on the growth of non-oncogenic cells. We infected 3Y1, a non-oncogenic rodent fibroblast cell line, with lentiviral particles expressing wild-type GRIM-19 and the tumor-derived mutants, and stable cell lines were isolated by selecting with puromycin. The growth profiles of cells expressing mutants were compared with those expressing wild-type GRIM-19 (supplemental Fig. S2). The mutants did not significantly stimulate cell growth compared with wild-type GRIM-19, except for A95T, which seemed to slow down cell growth slightly. Thus, mutants seem not to differ significantly from wild-type GRIM-19 in terms of their effects on normal cell growth.
Because GRIM-19 is an inhibitor of STAT3, we next determined the impact of these mutations on STAT3-induced cell growth. A constitutively active form of STAT3 (S3C) was stably transfected into 3Y1 cells, which is known to cause oncogenic transformation. Because S3C is the only oncogenic product under these conditions, the anti-STAT3 effects of GRIM-19 and its mutants can be readily discerned using this model. Stable cell pools expressing S3C were generated after selecting cells with G418 followed by GRIM-19 transduction via lentiviral particles and selection with puromycin. First, we wanted to ascertain whether the mutants differ from wild-type GRIM-19 in regulating cell growth in the presence of the S3C oncogene. S3C alone stimulated cell growth, which was strongly suppressed by wild-type GRIM-19 (p Ͻ 0.0001) ( Fig. 2A). In contrast, all three mutants significantly lost their ability to suppress S3C-induced growth compared with wild-type GRIM-19 under the same conditions. These differences in growth characteristics are not due to differences in expression levels of the proteins (Fig. 2B).

Effect of GRIM-19 Mutants on S3C-induced Transformation-
Because growth stimulation occurred in the presence of GRIM-19 mutants, we next determined their effects on oncogenic phenotype using two models. In the first model, 3Y1 cells expressing S3C alone or in combination with GRIM-19 mutants were used for anchorage-independent growth on soft agar media. As expected, S3C-expressing cells formed a large number of soft-agar colonies, unlike the empty vector or GRIM-19 singly transfected cells (Fig. 3A). S3C-dependent soft agar colony formation was robustly suppressed by GRIM-19. We then quantified and compared the differences between wild-type and mutant GRIM-19 to measure their impact on S3C-induced transformation (Fig. 3B). The mutants, L71P, L91P, and A95T significantly lost (p Ͻ 0.001, Ͻ0.008, and Ͻ0.0001, respectively) their ability to suppress S3C-induced transformation compared with wild type. The mutants alone did not have any transforming effect in these cells. In a second model, we employed a human oral squamous cell carcinoma cell line, HSC3, which expresses a very low level of GRIM-19 (23) and high basal STAT3 activity, to determine the impact of mutants on colony formation. We noted similar effects of GRIM-19 mutants on colony formation in HSC3 cells (Fig. 3C). However, A95T mutant was least inhibitory in 3Y1 cells, whereas the same mutant was strongest in HSC3 cells. The expression of mutants is shown in Fig. 3D. The A95T mutant had a lower expression than the wild-type protein in these cells. Overall, in S3C-transformed rat fibroblasts and human oral squamous carcinomas cells, GRIM-19 mutants were significantly weaker than the wild type at suppressing transformed phenotype.
Effect of GRIM-19 Mutants on S3C-induced Cell Motility-One of the major phenotypes of transformed cells is the acquisition of rapid motility. In the next experiments we examined the effects of GRIM-19 on S3C-dependent cell motility. Because HSC3 cells may have more than one genetic alteration that contributes to its transformed phenotype, it is difficult to precisely say whether all oncogenic effects are STAT3-driven. Therefore, we employed 3Y1 and HaCaT (a non-oncogenic keratinocyte) cells to study the impact of GRIM-19 mutants on S3C-dependent motility. In the initial assays, we employed 3Y1 cells expressing S3C in the presence or absence of GRIM-19 mutants. A confluent monolayer of each of these cell lines was scratched with a sterile pipette tip, and cell motility into the denuded area was monitored over time. As shown in Fig. 4A, very few cells moved into the denuded area in empty vector (EV) 2 -transfected cells. Several S3C-expressing cells migrated into the denuded area within 4 h. Wild-type GRIM-19 blunted S3C-induced motility. In contrast to the wild type, the L71P, L91P, and A95T mutants significantly lost their capacity to block S3C-induced motility, as evidenced by migration of cells into the denuded area. Fig. 4B shows quantified data from a number of plates (n ϭ 6; three replicates per experiment). HaCaT cells were stably transfected with a modified S3C expression vector. The original pRc/CMV-S3C with BGHpoly(A) signals did not express in keratinocytes; hence, we replaced this sequence with a poly(A) signal sequence from KRT-14. After establishing stable cell lines that expressed S3C (Fig. 4C), they were infected with lentiviral particles coding for wild-type or mutant versions of GRIM-19. In HaCaT cells, L71P mutant expressed to a lower extent than wild type or L91P, whereas A95T mutant expressed very weakly.
In the next set of experiments migration was scored using Transwell migration assays. 3Y1 and HaCaT cells expressing the indicated gene products were employed for this study (Fig.  4D). Relative to S3C-expressing cells, fewer GRIM-19-expressing cells moved into the lower chamber. In contrast, significantly higher numbers of mutant-expressing cells migrated into the lower chambers in both cell types. More importantly, both these appear to be resulting from a failure to antagonize S3C. Lastly, we investigated the impact of these mutants on IL-6-induced cell migration using HSC3 cells expressing GRIM-19 mutants in Transwell chambers (Fig.  4E) given recent reports that IL-6 is a potential growth factor for many tumors (24). IL-6-stimulated cell migration was high in the EV transfectants, which was strongly suppressed by wild-type GRIM-19 (p Ͻ 0.0001). The mutants when compared with wild-type GRIM-19 significantly (p Ͻ 0.002) lost the capacity to block IL-6-induced migration.  MARCH 15, 2013 • VOLUME 288 • NUMBER 11

JOURNAL OF BIOLOGICAL CHEMISTRY 7933
Anti-tumor Effects of GRIM-19 Mutants-To determine the effects of GRIM-19 mutants on tumor growth, we subcutaneously transplanted HSC3 and PC3 cells expressing these mutants into athymic nude mice (Figs. 5, A and B). We showed earlier that these cells express an extremely reduced level of endogenous GRIM-19 and high STAT3 activity (23,25). Hence, the effects of mutants on cell growth can be readily distinguished in these cells. HSC3 cells formed relatively smallersized tumors unlike PC3 cells. Nonetheless, in both cell types wild-type GRIM-19 suppressed tumor formation robustly (p Ͻ 0.001), whereas the mutants (L71P, L91P, and A95T) were relatively weaker (p Ͻ 0.01) at suppressing tumor growth. Among these, L71P and L91P were the weakest growth suppressors. Relative to these mutants, A95T retained a significantly higher growth inhibitory capacity (p Ͻ 0.05). Thus, patient-derived mutant GRIM-19 proteins failed to suppress tumor growth.
The above experiments did not unequivocally prove whether the anti-tumor effects are directly a result of the effects of GRIM-19 on a deregulated STAT3. Because S3C-transformed HaCaT cells did not form tumors in mice consistently, we did not use them for these studies. Hence, we subcutaneously administered S3C-expressing 3Y1 cell line derivatives into nude mice. As shown in Fig. 5C, unlike the parent cells (which did not form any noticeable tumors), S3C-expressing 3Y1 cells formed large tumors. Wild-type GRIM-19 completely suppressed S3C-dependent tumor formation (p Ͻ 0.0001). The mutants, unlike the wild type, lost their ability to suppress S3Cdependent tumor formation. Specifically, L71P and L91P were the weakest of three mutants (p Ͻ 0.005). Although A95T mutant suppressed tumor growth, it was significantly (p Ͻ 0.02) weaker than wild-type GRIM-19.

Anti-metastatic Effects of GRIM-19
Mutants-We next investigated if these GRIM-19 mutants had any effect on tumor metastases. Because HSC3 cells were not metastatic and there was no other metastatic oral cancer cell line available to us at this time, to determine the anti-metastatic effects, we transduced lentiviral particles coding for GRIM-19 mutants into a metastatic PC3 cell line expressing firefly luciferase gene, so the migration of tumor cells could be monitored in real time. The tumor cells expressing different GRIM-19 mutants were orthotopically transplanted into the prostate gland of nude mice and were imaged after 45 days. Mice transplanted with empty vector-expressing PC3 cells showed high metastatic behavior as seen by the migration of cells from prostate to other parts of the body (Fig. 6A). In contrast, wild-type GRIM-19-expressing cells showed an extremely weak metastatic behavior. All three mutants significantly (p Ͻ 0.02) lost their capacity to block metastatic spread when compared with the wild type (Fig. 6A). A quantitative representation of luciferase signals from metastasized regions is presented in Fig. 6B. The expression of GRIM-19 and its mutants in these cells was verified using a Western blot analysis (Fig. 6C).
The GRIM-19 Mutants Fail to Suppress S3C-induced Cellular Resistance to Chemotherapeutics-Constitutive STAT3 activity causes chemoresistance (26) in the tumors. We next examined if these GRIM-19 mutations have any effect on drug sensitivity of cells. Equal numbers of 3Y1 cells expressing S3C and GRIM-19 mutants were treated with Adriamycin (100 ng/ml) for 24 h, and cell survival was monitored compared with the corresponding untreated cells (Fig. 7). The presence of S3C caused an ϳ60% survival of cells compared with the control EV cells. In the presence of wild type GRIM-19, cells became as drug-sensitive as the EV cells despite the presence of S3C. All three mutants significantly lost their capacity to enhance chemosensitivity of cells compared with the wild-type GRIM-19. Thus, the loss of GRIM-19 activity (due to mutations) in the context of an oncogenic STAT3 may decrease cellular sensitivity to chemotherapeutics.

GRIM-19 Mutants Fail to Suppress STAT3-induced Gene
Expression-To understand the underlying mechanisms by which GRIM-19 mutants lost their anti-STAT3 activity, we first measured the expression of STAT3-inducible genes. We performed these experiments in 3Y1 and HaCaT cell lines expressing S3C and GRIM-19 mutants. Because these cell lines represent two different species, we could only perform the quantitative PCR analysis for genes that showed STAT3 responsiveness in these cell lines. As expected, BCL2-L1, DP1 (the binding partner for cell cycle regulating transcription factor E2F1), and cytoskeletal remodeling factor CTTN1 (Cortactin 1) genes were strongly induced in cells expressing S3C in HaCaT cells (Fig. 8A). Wild-type GRIM-19 suppressed the expression of all three genes, comparable with levels found in the EV-transfected cells. The mutants were still able to suppress S3C-responsive gene expression, albeit not as well as the wild type. The L91P mutant was able to suppress BCL2-L1 levels better than wild-type protein and the best mutant capable of suppressing S3C-responsive genes in HaCaT cells.
In 3Y1 cells, transcripts coding for the anti-apoptotic Bcl2-l1 and Ccnb1 (cyclin B1) were induced strongly by S3C, which were suppressed by wild-type GRIM-19 (Fig. 8B). All three mutants significantly lost their capacity to suppress S3C-inducible gene expression relative to wild-type GRIM-19. Interestingly, the L71P mutant was the only mutant that repressed Ccnb1 better than the wild type in these cells. The basis for the repressive effect of L91P (in HaCaT) on BCL2-L1 and L71P (in 3Y1) on Ccnb1 is unclear at this stage. The differential effects of these mutants may be due to a post-transcriptional effect or a different chromatin niche or the regulatory elements controlling them and not due to variations in their cellular localization (supplemental Fig. S3). The wild-type and mutant proteins distributed similarly in the cytoplasmic and nuclear compartments.  MARCH 15, 2013 • VOLUME 288 • NUMBER 11

JOURNAL OF BIOLOGICAL CHEMISTRY 7935
We also determined the impact of GRIM-19 mutants on the expression of certain STAT3-regulated genes in HSC3 and PC3 cells using Western blots. In HSC3 cells wild-type GRIM-19

GRIM-19 Mutations in Head and Neck Cancers
suppressed the expression of BCL2 protein, but its expression increased when GRIM-19 mutants were present (Fig. 8C). In PC3 cells, expression of cyclin D1, BCL2, and metastasis-associated protease MMP-9 was suppressed by wild-type GRIM-19, but the mutants were unable to block the expression of these proteins (Fig. 8D). Taken together, these data show a global loss of control by the tumor-derived GRIM-19 mutants over STAT3.
Defective Binding of Mutant GRIM-19 Proteins to STAT3-We next checked if the GRIM-19 mutants were capable of binding to STAT3 by co-immunoprecipitation (co-IP). Mutant and wild-type GRIM-19 proteins from 3Y1 cells were immunoprecipitated with myc-tag-specific antibodies. The IP products were probed for the presence of endogenous STAT3 using Western blotting. Indeed, all GRIM-19 mutants significantly lost their capacity to bind STAT3, unlike the wild type (Figs. 9,  A and B). Similar results were obtained in other cells. Recent studies showed that STAT3 can drive transcription in the absence of phosphorylation at Tyr-705; hence, we asked if the mutants were also defective at binding to such a STAT3. To this end, we examined the interactions of mutant GRIM-19 proteins with a FLAG-tagged STAT3-Y705F mutant. Although wild-type GRIM-19 protein interacted with STAT3-Y705F mutant, GRIM-19 mutants were significantly defective (Fig.  9C).
Because this assay simply shows a loss of binding, it does not reflect if the mutations have functional consequences. Because GRIM-19 binds to the transactivation domain of STAT3, its inhibitory effect on transcription should involve recruitment to the enhancers of STAT3-regulated genes. We have previously shown that GRIM-19 is recruited to the promoters of STAT3regulated genes like Bcl2, Bcl2-l1, and Ccnd1 (21) to exert a repressive effect. Therefore, we examined if mutant GRIM-19 proteins were capable of associating with chromatin. HSC3 cells were subjected to ChIP assays with STAT3-and GRIM-19-specific antibodies (Fig. 9D). As an example, we used BCL2 promoter, as it is repressed by GRIM-19 in all cell types tested and is overexpressed in primary tumors. After preparation of soluble chromatin, IP with the indicated antibodies was conducted, and DNA recovered from the IP products was subjected to PCR with BCL2 promoter-specific primers. In EV-transfected cells, only STAT3 was present at the promoter, whereas wild-type GRIM-19 was found when STAT3 was present. In contrast, the mutant proteins weakly associated with BCL2 promoter. To demonstrate that GRIM-19 and STAT3 are with the same promoter fragment, chromatin was chromatin-immunoprecipitated first with GRIM-19-specific (myc tag) antibodies, and the products were re-chromatin-immunoprecipitated with a STAT3-specific or a control IgG. Although the control IgG could not IP BCL2 promoter region, the STAT3-specific antibodies could IP the BCL2 promoter region. Although a strong signal was seen with wild-type GRIM-19, a reduced signal was observed with mutants. Altogether, these data show a severely diminished interaction of tumor-derived GRIM-19 mutant with STAT3 compromises their growth-suppressive actions.

DISCUSSION
STAT3 is a major oncogene (15) functioning downstream of many activated tyrosine kinase oncogenes, mutated, and/or overexpressed growth factor receptors in different human tumors (16). Its constitutive activity is linked to the induction of cell growth promoters like M-Ras, c-Myc, cyclin D1, cyclin B1, and DP1 and anti-apoptotic regulators like BCL2, BCL2-L1, MCL1, metastasis-promoting members of MMP family, all of which are associated with tumor initiation and progression (27). High STAT3 activity is also associated with chemoresistance (26). Although it was thought initially that only the tyrosyl-phosphorylated dimeric STAT3 was capable of driving up oncogenic gene expression (16), recent studies point out that un(tyrosyl)-phosphorylated STAT3 (U-STAT3) can also do so (28). Despite the extensive knowledge of the oncogenic mechanisms downstream of STAT3, it is unclear what leads to its chronic activity in cells. STAT3 mutations have been reported in certain forms of immune disorders (29 -33) but not in human tumors. This leaves one possibility for constitutive STAT3 activation, i.e. the loss or inactivation of STAT3 inhibitors. Among these proteins are SOCS3 (which inhibit JAKs that activate STAT3) and the proteinaceous inhibitor of activated STAT3 (PIAS3). SOCS proteins are not known to block non-JAK-tyrosine kinases; hence, they cannot provide a basis for the loss of control over STAT3. In fact, SOCS3 cannot block oncogene-induced STAT3 activation (34). Furthermore, SOCS3 expression itself is STAT3-dependent and is used as marker of STAT3 activity (35). The PIAS proteins are SUMO ligases that primarily target steroid and other transcription factors (36 -39). Their direct relevance to constitutively active STAT3 is largely unknown. The tumor suppressor LKB1 (STK11) is a cytoplasmic/nuclear serine/threonine kinase, defects in which cause Peutz-Jeghers syndrome in humans and animals (40). Recent studies showed that loss of function of LKB1 is associated with sporadic forms of lung, pancreatic, and ovarian cancers (41,42). LKB1 is inactivated by two different mechanisms: (a) mutations in its central kinase domain or (b) complete loss of its expression. Inactivation of LKB1 was associated with progression of Peutz-Jeghers syndrome and transformation of benign polyps into malignant tumors. We reported earlier that LKB1 inhibited rearranged in transformation/papillary thyroid carcinoma -dependent activation of STAT3 (phosphorylation of Tyr-705) (43). The kinase domain of LKB1, but not its activity, was critical for this effect (43). In contrast to these observations, a latest study showed that inactivation of STAT3 promoted tumor growth in certain forms of human papillary thyroid cancers and those driven by the BRAF V600E (44). These latter observations suggest a tissue and/or cell type-specific oncogenic/tumor suppressive effects of STAT3. One possible explanation for these differential effects is the tissue-specific post-translational modifications such as acetylation, methylation, and SUMOylation the STAT proteins undergo (14). Because RET/ papillary thyroid carcinoma is not a common alteration in most human tumors and LKB1 is not expressed in all tissues, inactivation of other potential inhibitors of STAT3 may contribute to unregulated STAT3 activity and consequent ontogenesis. In this connection, GRIM-19 appears to be such a candidate, as it is expressed in all cell types.
Overexpression of GRIM-19 caused growth inhibition and apoptosis, whereas depletion of GRIM-19 levels promoted growth (11). Unlike the other known STAT3 inhibitors, basal levels of GRIM-19 are found in all mammalian cells, which further support our hypothesis that it may act as an anti-oncogene in many cell types. Consistent with this, our earliest observations showed that the vIRF1 oncoprotein of human herpesvirus 8 (KSHV) inactivates anti-tumor activity of GRIM-19 by a direct interaction (45). In addition, the SV40 T antigen and HPV-E6 oncoprotein also bind to . The E6/E6AP complex has been shown to target p53 for degradation in cervical cancers (46). We have recently shown that GRIM-19 disrupts the E6/E6AP complex formation to protect tumor suppressor p53 (47). Thus, GRIM-19 collaborates with other tumor-suppressive pathways to enforce normal cell division. The subsequent protein-protein interaction studies by us and others identified several growth regulatory proteins as its targets, foremost of which is STAT3 (12,13). Binding of GRIM-19 inactivated the transcriptional activity of STAT3. The Ser-727 residue located in the transactivation domain of STAT3 was necessary for exerting its anti-STAT3 effects (12). The impor- tance of GRIM-19 to oncogenesis was further attested by our subsequent report that expression of GRIM-19 is inactivated in primary human renal cell carcinomas, which accompanied a high STAT3 transcriptional activity (10). The direct anti-oncogenic effects of GRIM-19 were shown by an inactivation of S3Cinduced oncogenic transformation (21). In addition, many studies by us and others have shown a loss of GRIM-19 expression in a variety of primary human tumors (25, 48 -55).
The current study identified mutations in the primary head and neck tumors of the oral cavity. This is the first documentation of functionally inactivating oncogenic mutations in GRIM-19 from solid tumors. Two GRIM-19 mutations, different from the ones shown here, were reported in Hürthle cell tumor of the thyroid (56), whose functional impact is unclear given the unexpected tumor-suppressive effect of STAT3 in thyroid tumors (44). The mutants identified here have clearly lost their ability to suppress S3C-induced tumor growth. The loss of anti-STAT3 effect appears to be global given similar data were obtained with the 3Y1 and HaCaT, both non-oncogenic cells. The validity of these data was further ascertained with a human oral cancer cell line HSC3, which expresses an extremely low level of GRIM-19 (23). In this model restoration of GRIM-19 robustly suppressed growth, whereas the mutants did not. It is important to note that these mutants themselves did not significantly alter normal cell growth in the absence of S3C. One of the mutations, L71P, is located in a region of GRIM-19 (amino acids 36 -72) that was reported to bind STAT3 (13), confirming the previous observations. It should be noted that the large deletion that removed one-fourth of total protein (13) simply showed a required domain but did not prove whether that domain itself was sufficient for GRIM-19 binding to STAT3. However, the other two mutations, L91P and A95T, which also weakly interacted with STAT3, indicated STAT3 and GRIM-19 proteins interact using more than one contact sites. Thus, our conclusion based on point mutants was more robust than those arrived at by Lufei et al. (13). Having said, we do not know either the crystal or NMR structure of GRIM-19 to define how these mutations affect protein-protein interactions. Future structural analyses are clearly required to resolve these issues.
In addition, unlike wild-type GRIM-19, the mutants were unable to contain S3C-driven cell migration in vitro. This clearly appears to be independent of cell division, given these effects are observed within few hours. In agreement with these data, these mutants failed to suppress tumor metastasis in vivo. An observation that testifies to the enhanced metastasis in presence of GRIM-19 mutants (Fig. 6) is the overexpression of cortactin (57), a cytoskeletal actin remodeling protein (Fig. 8A). Phosphorylation of cortactin by Src family kinases initiates a sequential cellular dynamic process; 1) neutralizes cross-linked actin network (58), 2) enhances actin assembly in vivo (59), and 3) plays an essential role in the formation of podosome (59). This protein has been previously reported as a target of STAT3 (60) and is overexpressed in a number of metastatic tumors (61)(62)(63)(64)(65). Similarly, MMP-9, a protease involved tumor invasion, is also up-regulated in these cells (Fig. 8D). Wild-type GRIM-19 suppressed S3C-dependent cell survival in the presence of the drug, whereas GRIM-19 mutants were unable to block chemotherapeutic-induced cell death (Fig. 7). In the presence of wild-type protein, cell death was significantly enhanced. These observations suggest the importance of a functional GRIM-19 for enhancing tumor therapeutic response. These mutants also have significantly lost their ability to promote IFN-␤-induced growth suppression (supplemental Fig. S4).
Mechanistically, GRIM-19 mutants have significantly lost their capacity to interact with STAT3 and as a result were incapable of inhibiting STAT3-driven gene expression (Fig. 8). Whereas wild-type GRIM-19 was able to associate with enhancer-bound STAT3, mutants significantly lost such capacity (Fig. 9D). Although multiple STAT3-inducible genes have been reported, S3C appears to induce them in a tissue-and cell-specific manner. Recent studies reported that U-STAT3 in association with the NF-B/RelA subunit promoted oncogenic gene expression (35). Interestingly, many genes like CCNB1 and DP1 are induced by U-STAT3 (28), indicating that the inhibitory effects of GRIM-19 may extend to U-STAT3. Indeed, the mutants were also defective at binding U-STAT3 (Fig. 9C). One interesting aspect of the A95T is it appears to retain a slightly better ability to suppress metastasis than the others, even though it was as poor as the other mutants with respect to binding STAT3. This difference could not be due to differential cellular localization of this mutant (supplemental Fig. S3). Although it has lost its interaction with STAT3 like the others, it is quite likely A95T activity may target other factors, or its activity is controlled by other factors.
That said, some of the oncogenic effects of STAT3 are also non-transcriptional. For example, STAT3 has been shown to modulate cytoskeletal organization through stathmin (66), a protein that modulates microtubule polymerization. Other studies have shown that STAT3 is present in the mitochondrion of Pro-B cells and modulates oxidative phosphorylation (67). In this connection, the H-ras oncogene, a non-tyrosine kinase, has been shown to promote mitochondrial localization of STAT3 and modulation of oxidative phosphorylation (68). A sizable fraction of cellular GRIM-19 is also present in mitochondrion as part of electron transport chain complex I (69). Grim-19 deletion causes early embryonic lethality due to severe dysfunction of electron transport chain complexes I and IV (70). In contrast to GRIM-19, STAT3 associated with electron transport chain complexes II and III in non-stoichiometric ratios (68). The exact effects of mitochondrial GRIM-19 and STAT3 interaction are unclear at present. Future studies should resolve these aspects. However, a recent study reported that during TNF-␣-induced necroptosis RIPK1 phosphorylates Ser-727-STAT3 leading to recruitment of GRIM-19 and the transportation of STAT3 to mitochondrion in murine L929 cells (71). In this model STAT3 migration to the mitochondrion causes increased ROS production via enhanced electron transport chain activity and cell death. Thus, GRIM-19, although discovered in an IFN-regulated pathway, extends into other cytokine-induced biological responses. Consequently, GRIM-19 mutations described here may also affect several cellular functions to bring about oncogenic changes. Last but not the least, these mutations may help in predicting patient populations that might respond to chemotherapeutics in association with STAT3 inhibitors.