CUL3 and NRF2 Mutations Confer an NRF2 Activation Phenotype in a Sporadic Form of Papillary Renal Cell Carcinoma

Nuclear factor [erythroid-derived 2]-like 2 (NRF2) transcription factor regulates the transcription of a large number of genes that constitute biochemical pathways including xenobiotic metabolism, energy metabolism, and cell-cycle regulation. Its activity is largely regulated by Kelch-like erythroid-derived Cap-n-Collar Homology (ECH)-associated protein 1 (KEAP1), which acts as an adaptor protein that facilitates NRF2 ubiquitination and subsequent degradation. Gain-of-function mutation to NRF2 and loss-of-function mutation to KEAP1 have been found in a large number of tumors (1–3). Because both of these mutations cause sustained NRF2 activation, it is likely that such activation is important to tumor progression.

Type 2 papillary renal cell carcinoma (PRCC2) is a kidney cancer subtype that has a high mortality rate. Individuals who inherit a germline mutation in the fumarate hydratase (FH) gene are predisposed to developing PRCC2 (4). Sporadic cases of PRCC2 rarely harbor an FH mutation, and the driver mutations in these tumors remain unknown (5). Despite differences in genetic etiology, both forms of PRCC2 share many clinical and morphologic phenotypes. The most prominent shared biochemical feature is sustained activation of the NRF2 (6). NRF2 activation in hereditary PRCC2 is caused by the accumulation of intracellular fumarate, which is a result of FH inactivation (6); the mechanism of NRF2 activation in sporadic PRCC2 has yet to be determined. We hypothesized that NRF2 activation in sporadic PRCC2 is caused by somatic mutation to genes that directly regulate NRF2 transcription activity.

Primary human normal renal proximal tubule epithelial cells, PCS-400, were purchased from American Type Culture Collection (ATCC) in year 2012. Each batch of the primary cell line was authenticated by ATCC for the presence of γ-glutamyl transferase activity, a marker for renal proximal tubule epithelial cells. Cells also tested negative for hepatitis C, HIV, bacteria, and mycoplasma. Cells from the third passage were used in this study.

Four of the 5 clinical samples were acquired through the Cooperative Human Tissue Network, and one was acquired from Spectrum Health Hospital (Grand Rapids, MI). The Van Andel Institute and Spectrum Health Institutional Review Boards approved the study protocols, and all patients signed a written consent form.

Total genomic DNA was isolated from cryosections of fresh-frozen tissues using a phenol/chloroform/isopropanol method. The coding exons of the indicated samples were isolated using the Agilent SureSelect All Exon 50 Mb Kit, following the manufacturer's recommended protocols. DNA sequencing was conducted using an Illumina HiSEQ. Raw sequencing reads were filtered to remove duplicate reads using Picard Tools 1.29 and were aligned to the reference genome (hg18) using BWA v0.5.8a (7, 8). Variants that differed between tumor and normal samples were identified using SomaticSniper 1.0.0 and annotated using SeattleSeq Annotation server (9). Total RNA was isolated using TRIzol (Invitrogen) reagent and sequenced using the Illumina HiSEQ following manufacturers' protocols. The raw sequencing reads were mapped to the reference genome (hg18) using Tophat v1.3.3 (10). Location-specific sequence extraction and examination were done using SAMTools (11). Small insertions and deletions (indels) that (i) occurred within coding exons, (ii) had associated quality scores greater than 900, (iii) had read depths greater than 100, and (iv) were not present in associated normal samples using the same filtering criteria were identified using the GATK unified genotyper (GenomeAnalysisTK-1.2-58-gc1329c4) and vcftools_0.1.9 (12). The mapped reads for both exome and transcriptome sequencing have been deposited in the Gene Expression Omnibus database.

Statistical analyses were conducted using the R statistical environment (Version 2.14.0, ×86_64-apple-darwin9.8.0/×86_64 [64-bit]; ref. 13). Wherever applicable, computation was accelerated using the “Multicore” package (14). Differential gene expression analysis on transcriptome sequencing data was conducted using the DESeq package (15). Total read counts covering each exon were extracted using SAMTools and were used in the differential gene expression analysis. Gene-specific fragments per kilobase of exon per million (FPKM) fragments mapped values were calculated using Cufflinks (16). Expression data were prefiltered to exclude noninformative features before statistical analysis; features having values less than the first quartile of the overall data in greater than 75% of the samples were deem noninformative. All P-values from statistical significance tests were adjusted for false discovery rate using methods described by Benjamini and Hochberg (17).

Expression plasmid constructs carrying variants of the wild-type genes in this study have been published by others and made available through Addgene. These plasmids were pcDNA3-HA2-KEAP1(18), pCDNA3-Myc3-Nrf2 (18), and pECE-Flag-SIRT1 (19). Mutant variants were created through PCR-mediated site-directed mutagenesis. The reporter constructs pRL-CMV and pGL4-Luc2P-NRF2-RE-Hygro were purchased from Promega. Reporter assays using these clones were conducted using the DualGlo-Luciferase Reporter Assay System (Promega) according to the manufacturer's recommended protocols. RNA interference-mediated knockdown of CUL3 was conducted using 2 species of siRNA with sequences GGAUCGCAAAGUAUACACAUAUGUA and UACAUAUGUGUAUACUUUGCGAUCC (Invitrogen). The 3 × flag-tagged versions of wild-type NRF2, the E82G mutant, and the V36 deletion mutant constructs were also created on the pIRESPuro plasmid backbone (Clonetech).

A cycloheximide chase experiment was conducted by transfecting PCS-400 cells with the 3 × flag-tagged version of wild-type NRF2, of the E82G mutant, or of the V36 deletion mutant. At 24 hours posttransfection, cells were split into 6 separate 6-well cell culture dishes and maintained for another 24 hours. After that, cell culture medium in all cell culture dishes were replaced with fresh warm medium containing 100 μg/mL of cycloheximide. Protein lysate was isolated at a 25-minute interval for 6 continuous time points. NRF2 protein levels were detected by Western blot analysis using an anti-DYKDDDDK tag antibody (Cell Signaling Technology). Relative protein abundance was estimated using a densitometry method (ImageJ). Loading variation was cancelled out by dividing the estimated NRF2 protein density with its corresponding actin protein density. Dividing protein density of each time point by the relative density of time zero give the sample-specific normalization. For NRF2-WT and NRF2-E82G, which follow regular exponential decay, half-life (t_1/2) was estimated using the equation of exponential decay: ln2/λ = t_1/2. Constant λ was estimated by fitting the data onto exponential decay function: N = N₀e^−λt, where N₀ is protein abundance at time zero and N is apparent protein abundance. For NRF2-E82G, which follows a logistic decay function, inflection point (t_i) was estimated according to |$N{\rm \,= \,\,}a{\rm/}\left({{\rm 1} + e^{{\rm ((}ti{\rm -}t{\rm)/}b{\rm)}}} \right)$|⁠, where a is the asymptote, b is a numerical scale constant, and e is the Euler number.

The NRF2 protein level in the formalin-fixed paraffin-embedded tissue samples was detected by immunohistochemical staining using an anti-human-NRF2 rabbit polyclonal antibody (ab31163, Abcam) at a dilution of 1:500. The staining was conducted using EnVision + staining kit (Dako) according to the manufacturer's recommended protocol. Antigen retrieval was conducted by autoclaving the slides (121°C, 15–16 psi, 15 minutes) in Target Retrieval Solution (Dako).

The exome sequencing data generated in this study have been deposited into the Sequence Read Archive of the European Bioinformatics Institute. The data accession number is ERP002035. The transcriptome sequencing data generated in this study have been deposited into the Sequence Read Archive of the National Center for Biotechnology Information. The data accession number is SRP015003.

Many of the current efforts in identifying driver mutations rely on the sequencing of a large number of cancers of a defined subtype to identify somatic mutations that occur more frequently than expected by chance alone (20). The incidence of PRCC2 in the general population is relatively low; consequently collecting large sample size translates into increased cancer-related deaths. To overcome this, we devised a knowledge-guided strategy of driver mutation discovery. We considered that some of the driver mutations in sporadic PRCC2 cause sustained NRF2 activation, because both hereditary and sporadic PRCC2 converge at this biochemical phenotype. Consequently, assessing the effect of mutants on NRF2 activity could identify driver mutations. Specifically, we mined scientific literature databases to construct an NRF2 regulatory pathway (Supplementary Fig. S1), which we subsequently used to screen nonsynonymous mutation calls from exome-captured massively parallel sequencing (exome sequencing).

We conducted exome sequencing on matched tumor versus normal samples from 5 cases of confirmed sporadic PRCC2, that identified 3,318 somatic mutations that involved 1,614 known genes (Supplementary Table S1). The initial screening identified mutations in 3 different genes. These gene mutations could account for NRF2 activation in 4 of the 5 cases of PRCC2. Potential gain-of-function NRF2 mutations were found in 2 cases: Case-831 contained a c.245A>G transition resulting in a change of glutamate 82 to glycine (E82G), and Case-666 contained a c.105_107delGTA deletion causing loss of valine 36 (Fig. 1A and B). Case-666 also had a SIRT1 c.808T>C transition, which changed isoleucine 270 to threonine (I270T; Fig. 1B). Two of the remaining cases contained CUL3 mutations, Case-337 with a c.1168delT deletion and Case-790 with a c.2125C>T transition, which result, respectively, in a frame shift (L390fs) and a change of arginine 708 to tryptophan (R708W; Fig. 1C and D). These somatic mutations were confirmed by dideoxynucleotide sequencing (Supplementary Fig. S2).

The functional consequences of the mutations we identified could be deduced from prior knowledge of how protein levels and transcriptional activity of NRF2 are regulated. NRF2 consists of 6 highly conserved Nrf2-ECH homology (Neh) domains (21). Contained within the Neh2 domain are 2 conserved motifs, DIDLID and ETGE, which are required for efficient binding with KEAP1 through interaction with its Kelch-repeat domain (21, 22). Under homeostatic conditions, NRF2 binds to KEAP1, which is tethered to the cytosol and brings NRF2 into close proximity with CUL3 ubiquitin ligase (23). NRF2 is then ubiquitinated by the KEAP1-CUL3 complex and degraded through a proteasome-mediated degradation pathway.

KEAP1 acts as cellular oxidative sensor: thiol-reactive compounds can react with its surface cysteine residues, causing it to undergo a conformational change that renders it unable to bind NRF2. Loss of the KEAP1–CUL3–NRF2 interaction results in NRF2 accumulation and thus increased expression of NRF2 transcriptional targets (6, 24). The c.245A>G mutation that results in the E82G substitution directly affects the ETGE motif within the Neh2 domain, altering the sequence to ETGG. The c.105_107delGTA deletion removes valine 36, which resides within the Neh2 domain between the DIDLID and ETGE motifs. This mutation does not change the DIDLID or ETGE motifs but shortens the distance between them.

To determine whether the NRF2 mutations resulted in NRF2 transcriptional activity that is insensitive to KEAP1-mediated regulation, we ectopically expressed the wild-type (WT) and the E82G forms of NRF2 in a primary kidney tubular epithelial cell line (PCS-400), with and without the present of ectopically expressed, HA-tagged KEAP1. NRF2 transcriptional activity was followed using a firefly luciferase reporter controlled by a 3-tandem antioxidant response element (ARE) enhancer sequence that was cloned in front of a basic promoter on a pGL4-based vector (Promega; Supplementary Fig. S3). As expected, the E82G mutation produced NRF2 mutant proteins more stable toward KEAP1-mediated degradation, although the transcriptional activity remained unchanged even with coexpression of exogenous KEAP1 (Fig. 2A). These results suggest that the mutation stabilizes NRF2 by disrupting its ability to bind to KEAP1. We conducted similar assays on the transcriptional activity of the NRF2 c.105_107delGTA deletion mutant (denoted as NRF2 V36del). Our results suggest that the mutation produced an NRF2 molecule that was stable even with ectopic expression of KEAP1 (Fig. 2B). A subsequent cycloheximide chase experiment confirmed that both NRF2 E82G and NRF2 V36del are more stable than the wild-type NRF2 (Supplementary Fig. S4).

NRF2 is also regulated by an acetylation/deacetylation cycle mediated by the cyclic AMP response element-binding protein [chronic benzene poisoning (CBP)] and SIRT1 (25). Acetylation of NRF2 by CBP increases its nuclear localization and transcriptional activity, and this modification is reversed by SIRT1-mediated deacetylation. Mutations have been identified in SIRT1 that function as dominant negatives by disrupting the deacetylase activity. Specifically, expression of SIRT1 protein that contains an H355A mutation increases the transcriptional activity of NRF2 (25). We found the I270T SIRT1 mutation in the same sample as the NRF2 c.105_107delGTA mutation. Expression of the I270T mutant has a small effect on the transcriptional activity of NRF2 c.105_107delGTA mutant gene product. Specifically, it enhanced the transcriptional activity of NRF2 to the level of the E82G mutant (Fig. 2A and B). Because CUL3 is the ubiquitin ligase that marks NRF2 for proteasomal degradation, a loss of CUL3 function could increase NRF2 activity, which has been previously shown by Liognon and coworkers in breast cancer tissues (26). One of the 2 CUL3 mutations was a deletion that causes a frame shift and complete loss of CUL3 function. We used siRNA-mediated knockdown of endogenous CUL3 to approximate the frameshift event. As expected, the transient knockdown of CUL3 resulted in higher NRF2 activity (Fig. 2C), supporting our hypothesis that the CUL3 mutations are the driver of NRF2 activation phenotype in 2 of the 5 cases of sporadic PRCC2. We were not able to identify a mutation that could drive NRF2 activation in 1 of the 5 samples sequenced. This suggests either a knowledge gap in the regulation of NRF2 transcriptional activity or an epigenetic event that was not captured by the exome sequencing.

We conducted transcriptome sequencing on the same 5 cases of sporadic PRCC2 to investigate the effect of NRF2 activation and the expression of the mutant alleles that disrupted NRF2 regulation. As expected, the transcript levels of many known transcription targets of NRF2—including AKR1C3 (aldo-keto reductase family 1 member C3; ref. 27), AKR1C1 (aldo-keto reductase family 1 member C1; ref. 27), AKR1B10 (aldo-keto reductase family 1 member B10; ref. 6, 27), NQO1 (NAD[P]H:quinone oxidoreductase; ref. 28), and TXNRD1 (thioredoxin reductase 1; ref. 29)—were significantly higher in the tumor samples than in matched adjacent control samples, indicating the sequenced tumor samples were indeed exhibiting an NRF2 activation phenotype (Fig. 3A and Supplementary Table S2). Subsequent immunohistochemical staining for NRF2 protein in those samples also showed intense positive nuclear staining (Fig. 3B). At the same time, normal kidney and clear cell renal cell carcinoma tissues showed negative staining (Fig. 3C and D). Hence, the results support our deduction that NRF2 was indeed activated in these tissues. Also, the nonframeshift mutant alleles were transcribed in the corresponding samples (Fig. 4A–C). However, the transcript of the CUL3 c.1168delT mutation was not present in the transcriptome sequencing data (Fig. 4D). The absence of the frameshift transcript suggests possible nonsense-mediated decay.

Despite being an aggressive cancer, PRCC2 remains largely understudied. Much of our understanding of PRCC2 comes from studying the hereditary form of the tumor, which arises in patients with hereditary leiomyomatosis and renal cell cancer (HLRCC) syndrome. Ever since the discovery of the association between FH mutation and HLRCC, studies on the hereditary form of PRCC2 have been focusing on the activation of the hypoxia-inducible factor-1α (HIF-1α). We recently discovered that in NRF2 is activated in hereditary PRCC2 tissues, and it is a direct consequence of FH inactivation (6). Specifically, fumarate, which accumulates as a result of FH inactivation, covalently modifies KEAP1 and renders it unable to bind NRF2, leading to NRF2 activation (6). A separate study using an Fh-Hif1a double knockout mouse showed that the knocking out Hif1a against the background of the Fh knockout exacerbated the disease phenotype, providing evidence that the Hif1a could be a compensatory event (30). Furthermore, the expression profile of the diseased tissue derived from the Fh knockout and the Fh-Hif1a double knockout were very similar. Notably, the overexpression of pro-glycolytic genes—such as Slc2A1 (glucose transporter 1) and PKM2 (pyruvate kinase)—that was thought to be the result of Hif1a activation, remained unchanged in the Fh-Hif1a double knockout (31).

Results from this study and others have shown that NRF2 activation is positively selected in tumors, suggesting that it contributes to tumorigenesis. For instance, NRF2 activation is a known feature of a large fraction of lung cancers. The NRF2 activation phenotype is known to result from NRF2 gain-of-function and KEAP1 loss-of-function mutations (1). We also found that a subset of lung cancers harbor CUL3 mutations, and the transcription profiles from these cases mirror those with NRF2 activation (results not shown). NRF2 activation has been reported to be a consequence of oncogene activation, suggesting that NRF2 could be the downstream effector that is responsible for tumorigenesis (32). Ectopic expression of constitutively active NRF2 has been shown to confer tumorigenic potential (33), although the mechanism for NRF2 conferring such potential is unknown.

Paradoxically, NRF2 modulates the expression of many genes involved in xenobiotic metabolism and neutralization of reactive oxygen species. NRF2 activation is believed to be the mechanism of tumor suppressive effects of many compounds, as well as caloric restriction (34, 35). These contradictory effects of NRF2 activation could be partly explained by the variability that exists in the regulatory effect of NRF2 on its transcription targets. The availability of NRF2 targets can be affected by the activity of chromatin modifiers such as BRG1 (SWI/SNF-related, matrix-associated, actin-dependent regulator of chromatin, subfamily a, member 4) and other regulatory proteins such as BACH1 (BTB and CNC homology 1, basic leucine zipper transcription factor 1; refs. 36, 37). Indeed, chromatin modifier genes such as BRD7 (bromodomain containing 7), MLL2, and MLL3 (myeloid/lymphoid or mixed-lineage leukemia 2 and 3 were mutated in the PRCC2 samples; Supplementary Table S1). This means that NRF2 activation in tumor cells or precancerous cells with malignant potential may lead to the activation of a different combination of genes than in normal cells, thus allowing identification of therapeutic targets unique to such cells.

No potential conflicts of interest were disclosed.

Conception and design: A. Ooi, K.A. Furge

Development of methodology: A. Ooi, J. Carpten

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Ooi, K. Dykema, A. Ansari, D. Petillo, J. Snider, R. Kahnoski, J. Anema, J. Carpten, B.T. Teh, K.A. Furge

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Ooi, K. Dykema, D. Craig, J. Carpten

Writing, review, and/or revision of the manuscript: A. Ooi, K.A. Furge

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Ooi, K.A. Furge

Study supervision: K.A. Furge, A. Ooi

The authors thank Sabrina Noyes for administrative support and David Nadziejka for technical editing.

This work was funded by the Van Andel Research Institute.

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.