SQSTM1 is a Pathogenic Target of 5q Copy Number Gains in Kidney Cancer

INTRODUCTION

Kidney cancer is a common cancer in the developed world. The most common form of kidney cancer is clear cell renal cell carcinoma (ccRCC), which is usually linked to biallelic inactivation of the VHL tumor suppressor gene located on chromosome 3p25 (Shen and Kaelin, 2012). Loss of chromosome 3p, which also harbors the renal cancer suppressors PBRM1, BAP1, and SETD2, is found in virtually all ccRCCs (Cancer Genome Atlas Research, 2013; Dalgliesh et al., 2010; Guo et al., 2011; Pena-Llopis et al., 2012; Sato et al., 2013; Varela et al., 2011). The VHL gene product (pVHL) targets the HIFα transcription factors for proteasomal degradation. Increased HIF2α promotes pVHL-defective tumorigenesis (Kondo et al., 2003; Raval et al., 2005; Shen and Kaelin, 2012).

In addition to 3p loss, ccRCCs often harbor large deletions of chromosome 14q (~40% of cases) and copy number gains of chromosome 5q (~70% of cases) (Beroukhim et al., 2009; Cancer Genome Atlas Research, 2013; Chen et al., 2009; Dondeti et al., 2012; Hagenkord et al., 2011; Krill-Burger et al., 2012; Sato et al., 2013; Shen et al., 2011). Loss of chromosome 14q and gain of chromosome 5q, considered either individually or together, are seen more often in kidney cancer than in other cancers and presumably reflect selection pressure to silence one or more 14q kidney cancer suppressor genes and to increase the expression of one or more 5q kidney cancer oncogenes (Shen et al., 2011). Interestingly, unbalanced translocations involving chromosomes 3p and 5q, including constitutional translocations, have been reported in kidney cancer that result in loss of chromosome 3p and gain of 5q (Bos et al., 1998; Iqbal et al., 1996; Kenck et al., 1997; Kovacs et al., 1991; Kovacs and Frisch, 1989; Kovacs and Kung, 1991; Kovacs et al., 1987; Presti et al., 1991).

HIF1α, which antagonizes HIF2α in certain settings, appears to be a target of the 14q deletions (Shen et al., 2011), while the relevant chromosome 5q gene(s) is/are unknown. We recently used high-density SNP arrays to measure copy number changes in 90 ccRCCs and 21 ccRCC cell lines (Beroukhim et al., 2009). Approximately 70% of the samples exhibited increased copies of a region of 5q with a peak at 5q35.3 that, allowing for possible passenger events, contained about 61 genes (Beroukhim et al., 2009). Notably, at least 12 of these genes were overexpressed relative to non-amplified tumors (p<0.05) including GNB2L1, MGAT1, RUFY1, RNF130, MAPK9, CANX, CNOT6, SQSTM1, LTC4S, TBC1D9B, HNRPH1, and FLT4 (Beroukhim et al., 2009). In this study we sought to identify the 5q amplicon gene underlying the selection pressure for 5q copy number gains in ccRCC.

RESULTS

We interrogated the copy number changes in 16 ccRCC cell lines using multiplex ligation-dependent probe amplification (MLPA) of 12 randomly selected genes spanning the 5q amplicon (2–4 exons/gene) as well as 3 randomly selected control exons located elsewhere in the genome. Two cell lines (A498 and SLR21) did not exhibit 5q gains (MLPA score 1) relative to HK-2 immortalized renal epithelial cells (Ryan et al., 1994) while the remaining 14 lines exhibited low level copy gains suggestive of 3–6 copies (MLPA score 1.5–3) of 5q relative to a diploid cell with 2 copies (MLPA score 1)(Figures 1 and S1). These 5q copy numbers are consistent with earlier cytogenetic and Southern blot studies of ccRCCs (Kenck et al., 1997; Kovacs et al., 1991; Kovacs and Frisch, 1989; Kovacs and Kung, 1991). Interestingly, the 5q gain in one cell line (UMRC-6) appeared to be restricted to the telomeric genes MGAT4B, SQSTM1, MAPK9 and GNB2L1 while the other cell lines appeared, given the variability of the assay, to have sustained broader amplifications that encompassed all 12 genes (Figures 1 and S1).

In parallel, we measured the mRNA levels of the 61 recurrently amplified 5q genes by quantitative real-time PCR in 4 of the 5q amplified cancer cell lines analyzed above as well as in HK-2 cells. As expected, multiple mRNAs were increased in the 4 5q amplified lines relative to HK-2 cells (Figure 2A and Table S1). Three mRNAs were induced greater than 12-fold but were of very low abundance (ADAMTS2, MSX2 and PCDH24). We elected to study SQSTM1 further because SQSTM1 overexpression was conspicuous when taking into account both fold-induction and absolute levels of the mRNA expression of the 61 genes, and because SQSTM1 is known or suspected of being an oncogene in other settings (Duran et al., 2008; Inami et al., 2011; Mathew et al., 2009; Nezis and Stenmark, 2012; Puissant et al., 2012).

SQSTM1 copy number was increased in every 5q amplified cell line (Figures 2B and S2A) and was associated with increased SQSTM1 mRNA levels, as determined by real-time PCR (Figure 2C) and Northern blot analysis (Figure S2B), and protein (p62) levels (Figure 2D), relative to HK-2 cells. Notably, SQSTM1 mRNA levels, and particularly p62 protein levels, were also increased in A498 cells, which are not 5q-amplified (Figures 2C, 2D, and S2B). This suggests that A498 cells also upregulate p62 but through an alternative mechanism (see also below). Conversely, in 786-O and RCC4 cells, SQSTM1 mRNA and protein levels reverted toward those seen in HK-2 cells, possibly due to prolonged passage in culture (Figures 2C, 2D and S2B). Importantly, SQSTM1 mRNA levels are increased in 5q amplified kidney tumors compared to non-amplified kidney tumors (or normal kidney) [(Beroukhim et al., 2009; Cancer Genome Atlas Research, 2013) and Figure 2E], indicating that SQSTM1 mRNA expression level is regulated at the gene copy number level. However, SQSTM1 mRNA expression levels also showed considerable variation that cannot be attributed to gene copy number aberrations alone. Notably, increased levels of p62 are associated with increased tumor grade in ccRCCs (Figures 2F, 2G, and S2C–S2F).

p62 inactivates KEAP1, which is a negative regulator of NRF2 (Nezis and Stenmark, 2012; Puissant et al., 2012). Conversely, NRF2 can activate expression of SQSTM1 (Jain et al., 2010). Interestingly, deregulation of NRF2, which is a master regulator of the antioxidant response, plays an important role in papillary renal carcinoma as well as other tumors (Adam et al., 2011; Hayes and McMahon, 2009; Ooi et al., 2011). Moreover, NRF2 (official name NFE2L2) and KEAP1 mutations have been detected, albeit infrequently, in ccRCC (Figure 3A) (Cancer Genome Atlas Research, 2013; Sato et al., 2013). The NRF2 gene had inferred somatic mutations in 5 out of 368 cases and high-level DNA amplification in 2 additional cases. The KEAP1 gene had somatic missense mutations in 3 cases and a homozygous deletion in one additional case (Figure 3A). Notably, the NRF2 missense mutations would be predicted to disrupt KEAP1-binding and to thereby activate NRF2. Of note, NRF2 mutations, KEAP1 mutations, and high level SQSTM1 amplification (gain of 2 or more copies) appear to be mutually exclusive.

We did not observe a strong NRF2 transcriptional signature in 5q amplified tumors relative to non-amplified tumors by gene set enrichment analysis (GSEA, data not shown), possibly because changes in SQSTM1 mRNA levels cannot be attributed to copy number aberrations alone or because non-amplified tumors can activate NRF2 by other means, such as described above. With respect to the former possibility, we reasoned that we could more directly probe the function of SQSTM1 in the TCGA data by correlating gene set expression with SQSTM1 mRNA expression levels. Indeed, we found that human ccRCCs with high levels of SQSTM1 mRNA exhibit transcriptional profiles that are consistent with activation of NRF2 as determined by GSEA using gene sets consisting of previously annotated NRF2 target genes (Figure 3B) and gene sets linked to NRF2-regulated processes, such as response to oxidative stress and mitochondrial biosynthesis (Figures 3C and 3D).

The sensitivity of ccRCC cell lines to oxidative stress mirrored their basal p62 levels and particularly their NRF2 levels (Figures 4A and 4B). A498 cells displayed high basal levels of NRF2 and modestly elevated p62 levels, suggesting that these cells harbor a mutation that drives the production of NRF2, which secondarily induces p62. SLR21 cells also displayed high NRF2 levels, possibly due to low levels of KEAP1. NRF2 and KEAP1 themselves, however, appear to be wild-type in A498 and SLR21 cells based on cDNA sequencing (data not shown). Exogenous expression of p62 in cells with low basal p62 levels induced NRF2 and reduced sensitivity to oxidative stress (Figures 4C, 4D and S3A). Conversely, knockdown of p62 with 3 distinct shRNAs reduced NRF2 levels and increased reactive oxygen species (ROS) sensitivity (Figures 4E, 4F, S3B and S3C). Collectively, these results support that 5q copy number gains in ccRCC increase p62 levels, which activate NRF2 and confer resistance to oxidative stress.

Acute VHL loss can induce senescence mediated, at least partly, by oxidative stress (Welford et al., 2010; Young et al., 2008), suggesting that p62 overexpression and VHL loss might cooperate with one another to promote renal carcinogenesis. In keeping with this view, we found that three independent SQSTM1 shRNAs, but not a control shRNA, impaired the ability of the 5q amplified VHL^−/− ccRCC cell line UMRC-2 to grow in soft agar (Figures 5A and S4A). The same SQSTM1 shRNAs had only minimal effects on soft agar growth by 293 human embyronic kidney cells, which have low basal levels of p62 (Figures S4B–S4D). Induction of a SQSTM1 shRNA, but not a control shRNA, also decreased tumor formation by UMRC-2 cells (Figures 5B–5D) and UMRC-6 cells (Figures S4E and S4F). These phenotypic effects were specifically due to loss of p62 because they were reversed with a lentivirus encoding an shRNA-resistant SQSTM1 mRNA (Figures 5E–5H).

The finding that p62 downregulation impaired ccRCC growth could simply indicate that SQSTM1 is a housekeeping gene and does not necessarily imply that supraphysiological levels of p62 promote tumor growth. Indeed, in parallel studies we infected multiple ccRCC cell lines with a custom shRNA library targeting all 61 5q amplicon genes (3–10 shRNAs/gene), monitored shRNA abundance over time by deep sequencing, and scored shRNA depletion/enrichment for each gene using the RIGER algorithm (Luo et al., 2008). We also evaluated the effects of shRNA-mediated loss of the 61 individual 5q genes on cellular fitness in a different experimental format wherein cells expressing red fluorescent protein and an effective shRNA targeting the 5q gene of interest were cocultured with cells expressing venus fluorescent protein and a control shRNA and monitored by FACS. shRNA-mediated knockdown of many other genes, in addition to SQSTM1, decreased the fitness of ccRCC cell lines (Figures S4G–S4J). Although some of these genes might be oncogenes, it is more likely that many of them are passenger housekeeping genes.

We therefore asked whether increased expression of SQSTM1 promotes ccRCC growth. Lentiviral expression of SQSTM1 in A498 and 786-O cells, did not affect proliferation under standard cell culture conditions (Figures 6A–6D, S5A and S5B) but promoted soft agar growth (Figures 6E–6G and S5C) and tumor formation in vivo (Figure 6H). These results were specific because the same vector did not affect UMRC-2 cells, which have high endogenous levels of p62 (Figures S5D and S5E). In addition, the SQSTM1 shRNAs that we documented impaired ccRCC fitness and tumor growth (Figures 4, 5 and S4) did so without lowering p62 levels below those seen in HK-2, 786-O, and SLR21 cells (Figures 4E, S3C, S4E and 5B). Collectively these results argue that supraphysiological p62 levels promote renal tumorigenesis.

p62 is a multifunctional protein with multiple subdomains that interact with a variety of downstream effector molecules (Figure 7A) (Geetha and Wooten, 2002; Moscat and Diaz-Meco, 2009; Nezis and Stenmark, 2012; Puissant et al., 2012). We found that a previously described p62 mutant (“CT”) (Duran et al., 2011) lacking amino acids 1-229, and hence lacking both its PB1 (Phox/Bem1p) and ZZ (central zinc finger) domains as well as part of its TBS (TRAF6-Binding) domain, failed to promote soft agar growth in A498 cells (Figures 7A–7C) and did not rescue soft agar growth in UMRC-2 cells that were depleted of endogenous p62 with an inducible shRNA, despite retaining the ability to induce NRF2 (Figures 7D–7F and S6). Therefore promotion of soft agar growth by p62 requires certain functions mediated by its N-terminus. On the other hand, a p62 variant (“m6A”) bearing an inactivating mutation within its KEAP1-binding domain, and hence unable to induce NRF2, retained the ability to promote soft agar growth in A498 cells and could rescue UMRC-2 cells lacking endogenous p62 (Komatsu et al., 2010)(Figures 7A–7F and S6). Moreover, NRF2 levels in A498 cells are relatively high even in the absence of exogenous p62, as noted above. These two observations suggested that promotion of soft agar growth by p62 involves at least one effector other than NRF2.

It has been reported that wild-type p62, but not p62 CT, can activate mTOR (Duran et al., 2011). In addition, we found that activation of mTOR by Rheb2 or a constitutively active RagB mutant (RagB^GTP), like overexpression of p62 itself, enhances the soft agar growth of A498 cells (Figures S7A–S7C). Moreover, some ccRCCs harbor mutations, such as TSC2 or MTOR mutations, that activate the mTOR pathway and such mutations, like NRF2 pathway mutations, are enriched in tumors that do not have high level chromosome 5q copy number gains, suggesting that p62 provides an alternative mechanism to activate mTOR (Cancer Genome Atlas Research, 2013). Although overexpression of p62 in ccRCC cells did not consistently activate mTOR under any condition tested (data not shown), we did observe that acute siRNA-mediated knockdown of p62 attenuated mTOR signaling (Figure S7D). Moreover, activation of mTOR with Rheb2 restored soft agar growth in UMRC-2 cells depleted of p62 (Figures 8A and 8B). Conversely, pharmacological blockade of mTOR with 1 nM rapamycin prevented soft agar growth driven by p62 or Rheb2 (Figure 8C). Therefore mTOR acts downstream or parallel of p62.

DISCUSSION

Chromosome 5q is the region of the genome most frequently amplified in ccRCC and copy number changes involving chromosome 5q are more common in ccRCC than in any other cancer (Beroukhim et al., 2009; Chen et al., 2009; Dondeti et al., 2012; Hagenkord et al., 2011; Krill-Burger et al., 2012; Moore et al., 2012; Shen et al., 2011). Indeed, the triad of chromosome 3p loss, chromosome 14q loss, and chromosome 5q gain is a signature abnormality in ccRCC. Our studies strongly suggest that SQSTM1 is one of the genes underlying the recurrent 5q copy number gains in ccRCC.

SQSTM1 has been implicated as a potential oncogene in other settings, including human hepatocelluar carcinomas, lung carcinomas, pancreatic carcinomas, breast carcinomas, and in immortalized baby mouse kidney cells (Duran et al., 2008; Inami et al., 2011; Inoue et al., 2012; Mathew et al., 2009; Rolland et al., 2007; Thompson et al., 2003). The SQSTM1 gene product, p62, is a multifunctional protein that serves as an adaptor molecule to facilitate the degradation of specific proteins by autophagy (Geetha and Wooten, 2002; Moscat and Diaz-Meco, 2009; Nezis and Stenmark, 2012; Puissant et al., 2012). Of note, VHL^−/− ccRCC have high basal rates of autophagy (Bray et al, 2012). The p62 protein also physically interacts with a number of signaling molecules to enhance the activity of downstream effectors such as NRF2, NF-κB, and mTOR. For example, p62 can bind to KEAP1 and prevent it from targeting NRF2 for degradation. There is increasing genetic and functional evidence that NRF2, which activates a transcriptional program that promotes resistance to redox stress, can promote tumor growth in some settings (DeNicola et al., 2011; Hayes and McMahon, 2009; Sporn and Liby, 2012). Acute VHL loss can, at least in some cells, cause senescence due, at least partly, to oxidative stress (Welford et al., 2010; Young et al., 2008). p62 might mitigate this effect as well as contribute to the resistance of ccRCCs to conventional cytotoxic therapy. Of note, SQSTM1 is itself an NRF2 target gene and hence potentially capable of participating in a positive feedback loop (Jain et al., 2010). This might contribute to the disproportionate increase in SQSTM1 mRNA levels relative to the other 5q amplicon genes observed in ccRCCs that have amplified chromosome 5q.

NRF2 pathway mutations occur, albeit rarely, in ccRCC (Cancer Genome Atlas Research, 2013; Sato et al., 2013) and, as shown here, are mutually exclusive with higher level SQSTM1 amplification (gain of 2 or more copies), further suggesting that NRF2 is likely an important target of p62 in ccRCC. On the other hand, we found that the KEAP1-binding domain of p62 was neither necessary nor sufficient for p62 to promote renal transformation, as measured in soft agar assays, indicating that additional p62 targets play roles in renal carcinogenesis. p62 also binds to several proteins upstream of NF-κB, including atypical PKC members and TRAF6, leading to enhanced NF-κB activity in certain settings (Nakamura et al., 2010; Nezis and Stenmark, 2012; Puissant et al., 2012). Interestingly, pVHL binds to and inhibits atypical PKC members and, through several mechanisms, suppresses NF-κB activity (An et al., 2005; An and Rettig, 2005; Datta et al., 2000; Datta et al., 2001; Lee et al., 2005; Okuda et al., 2001; Pantuck et al., 2010; Schermer et al., 2006; Yang et al., 2007) (Okuda et al., 1999). NF-κB promotes VHL^−/− ccRCC growth in vivo (An et al., 2005; An and Rettig, 2005; Pantuck et al., 2010; Yang et al., 2007). Moreover, a recent report suggested that p62 binds to the Raptor complex and accentuates mTOR signaling under some experimental conditions (Duran et al., 2011). A significant fraction of ccRCCs have mutations that directly or indirectly activate mTOR and we found that forced activation of mTOR rescues soft agar growth in p62-depleted ccRCC cells, indicating that mTOR acts downstream or parallel of p62 (Cancer Genome Atlas Research, 2013).

As was true for NRF2, mTOR pathway mutations are mutually exclusive of high-level chromosome 5q gains in ccRCC (Cancer Genome Atlas Research, 2013). Several non-mutually exclusive explanations might account for this observation. First, as indicated above, p62 might enhance mTOR activity and decrease the selection pressure to mutate the mTOR pathway. Second, mTOR activation inhibits autophagy. This promotes the accumulation of p62 by preventing p62 from being destroyed together with its autophagocytic cargo. Therefore mTOR activation might decrease the selection pressure to increase SQSTM1 copy number. Finally, it is possible that p62 and mTOR operate on parallel, but redundant, pathways to promote renal carcinogenesis. Interestingly, a recent study showed that NRF2 activation in lung cancer cells renders them mTOR-dependent (Shibata et al., 2010).

mTOR is a validated therapeutic target in ccRCC. Notably, VHL^−/− ccRCC cells are more sensitive than their pVHL-proficient counterparts to drugs, such as mTOR inhibitors, that promote autophagy (Hudes et al., 2007; Thomas et al., 2006; Turcotte et al., 2008). Moreover, a recent study showed that murine tumors linked to Tsc2 mutations, and hence mTOR activation, are p62-dependent (Parkhitko et al., 2011). The relationship between mTOR inhibition, induction of autophagy, and suppression of ccRCC is likely to be complex, however. In some preclinical models blocking autophagy actually enhances the activity of mTOR inhibitors by increasing mitochondrial ROS production in the face of impaired NRF2 activity stemming from the effects of mTOR on NRF2 nuclear localization (Bray et al., 2012). It will be important to ask whether the therapeutic efficacy of mTOR inhibitors is due, at least partly, to downregulation of p62 and, if so, whether p62 levels can be used as predictive or pharmocodynamic biomarkers.

Although SQSTM1 appears to be an important driver on chromosome 5q, the chromosomal gains and losses in renal cancer, including 5q gains, are non-focal, suggesting that they are driven by more than one gene. Indeed, 3p loss targets several ccRCC suppressor genes. In this regard, the 5q amplicon contains a number of other genes that could plausibly act as renal cancer oncogenes including MAML1, FGFR4 and MAPK9. These genes, however, did not consistently score in the oncogene assays described here. For example, FGFR4 shRNAs impaired cell fitness in vitro (Figure S4J) but promoted orthotopic tumor growth (data not shown). GNB2L1/RACK1 was, together with SQSTM1, recently highlighted as a potential ccRCC 5q oncogene because its protein product, like p62, activates mTOR (Duran et al., 2011; He et al., 2011) and because high-level 5q amplification is, as stated above, mutually exclusive with mTOR pathway mutations in ccRCC (Cancer Genome Atlas Research, 2013). Nonetheless, GNB2L1/RACK1 did not score as an oncogene in our assays (Figures S4G and S4I). It remains possible, however, that some of these genes confer a growth advantage under conditions that are not replicated by our assays. Knockdown of the 5q amplicon gene STC2 impaired ccRCC fitness in our hands (Figure S4J) and in a recent report (Dondeti et al., 2012). Whether STC2 is a housekeeping gene or an oncogene warrants further study.

In summary, enhanced expression of SQSTM1 promotes renal carcinogenesis in vitro and in vivo. The SQSTM1 gene product, p62, regulates NRF2, NF-κB, and mTOR, all 3 of which network with known renal carcinoma suppressor genes such as VHL, FH, TSC1, and TSC2 (Figure 8D). It will be important to determine which p62 effector pathways contribute to renal carcinogenesis and whether they can be pharmacologically attacked.

EXPERIMENTAL PROCEDURES

Supplementary Material