HIF2α-Dependent Lipid Storage Promotes Endoplasmic Reticulum Homeostasis in Clear-Cell Renal Cell Carcinoma

Clear-cell renal cell carcinoma (ccRCC) is the most common form of kidney cancer worldwide (1, 2). In 90% of ccRCC tumors, pathogenesis is characterized by constitutive activation of the hypoxia-inducible factors (HIF) due to loss of the von Hippel–Lindau (pVHL) tumor suppressor, required for oxygen (O₂)-dependent suppression of HIF signaling (1). HIF-regulated processes are orchestrated by two HIFα subunits (HIF1α and HIF2α), which share several common transcriptional targets, but also exhibit distinct functions (2). This is particularly evident in ccRCC, where HIF1α expression is frequently lost during disease progression, and preclinical data indicate that it can repress tumor growth (3). The central role of HIF2α in ccRCC is supported by findings that (i) all pVHL-null ccRCC maintain HIF2α expression (4); (ii) HIF2α function is required for ccRCC xenograft growth (1, 2); and (iii) polymorphisms in EPAS1/HIF2α are associated with increased ccRCC risk in genome-wide association studies (5). HIF-dependent gene expression contributes directly to enhanced cell proliferation (6) and metabolic alterations that characterize ccRCC (1, 7).

A second hallmark of ccRCC is the presence of intracellular lipid droplets (LD), which consist of a neutral lipid core containing triglycerides and cholesterol-esters surrounded by a phospholipid monolayer and associated LD surface proteins (8). Two well-characterized functions of lipid storage in eukaryotic cells include energy homeostasis and release of lipid species for membrane synthesis during proliferation (8). In addition, LDs are functionally and physically associated with the endoplasmic reticulum (ER), as lipids and the proteins that synthesize/modify them are exchanged between these organelles via transient membrane bridges (9). The PAT (Perilipin, Adipophilin, TIP47) family of LD coat proteins regulates both lipid storage and lipolysis (8). Perilipin (PLIN1) is expressed primarily in adipose and steroidogenic cells, whereas adipophilin/adipose differentiation–related protein (hereafter referred to as perilipin 2, PLIN2) and TIP47 (PLIN3) are ubiquitously expressed and function as the predominant LD coat proteins in nonadipose tissue (10). Interestingly, PLIN2 expression is positively correlated with HIF2α activation in mouse hepatocytes in vivo (11), and our microarray data suggest that HIF2α promotes PLIN2 mRNA expression in ccRCC cells in vitro (12). However, it remains unknown if PLIN2 regulates lipid metabolism and storage downstream of HIF2α or if this phenotype has any significant tumor-promoting functions in ccRCC.

Enhanced lipid storage in ccRCC suggests profoundly altered lipid metabolism. In normal cells, lipid metabolism is carefully regulated to support membrane expansion, organelle homeostasis, signal transduction, and cell viability. Recent work indicates that cellular transformation commits tumors to growth programs that strain ER homeostasis, including dysregulation of protein and lipid metabolism (13). Such ER stress is exacerbated by conditions of nutrient and O₂ deprivation characteristic of solid tumor microenvironments, which further disrupt cellular protein and lipid homeostasis (14). Mammalian cells activate a highly conserved unfolded protein response (UPR) upon elevated misfolded protein load or disruption of ER membrane lipid composition (15). ER stress sensors, including PERK, IRE1α, and ATF6, initiate UPR signaling and adaptive processes, including a generalized reduction in protein synthesis and selective expression of genes encoding lipid synthetic enzymes, protein-folding chaperones, and components of the ER-associated degradation (ERAD) system for enhancing proteasome-dependent proteolysis (15). However, sustained and irremediable ER stress can trigger cell death via a “terminal” UPR (16). Indeed, antitumor activity of the proteasome inhibitor bortezomib in multiple myeloma derives at least partly from elevated misfolded protein levels and induction of a cytotoxic UPR (17).

In this study, we explored mechanisms that regulate lipid storage and its function in ccRCC. Transcriptional profiling of primary ccRCC and normal kidney samples revealed that PLIN2, but not other perilipin family members, is overexpressed in ccRCC and positively correlated with HIF2α activation. HIF2α promoted PLIN2 expression and lipid storage in ccRCC cell lines, and remarkably, PLIN2 activity accounted for a substantial portion of HIF2α's tumor-promoting effects in xenograft assays. Mechanistically, the HIF2α/PLIN2/lipid storage axis was required for ER homeostasis and resistance against cytotoxic ER stress. These findings reveal an unexpected function for the “clear-cell” phenotype and identify enhanced ER stress as a targetable vulnerability created by HIF2α suppression in ccRCC.

To confirm the contribution of neutral lipid storage to the “clear-cell” phenotype in our archived ccRCC tissues, we performed oil red O staining of primary tumor and matched normal kidney samples and observed enhanced neutral lipid staining in ccRCC tumor cells (Fig. 1A). Potential mediators of enhanced lipid storage were identified by analyzing RNA sequencing (RNA-seq) data released by The Cancer Genome Atlas (TCGA) comparing primary ccRCC (n = 480) and normal kidney (n = 69) tissues (18). Focusing on expression of the PAT family of lipid droplet coat proteins, we observed that PLIN2 was overexpressed in tumors across all stages of disease (Fig. 1B). In contrast, mRNA expression of other perilipin family members was not enhanced in ccRCC (Supplementary Fig. S1A). PLIN2 upregulation was confirmed in an independent set of matched ccRCC (n = 10) and normal kidneys (n = 10) from our archived samples (Fig. 1C). Analysis of “Triglyceride Synthesis” and “Cholesterol Synthesis” gene sets within the TCGA RNA-seq dataset, as defined by the Broad Institute Molecular Signatures database, revealed differential expression of many lipogenic genes (Supplementary Tables S1 and S2). Some fatty acid and triglyceride synthesis genes were upregulated within tumor tissues (i.e., ACLY, ELOVL2, LPCAT1), whereas many others were downregulated (i.e., multiple ACSL and AGPAT family members; Supplementary Table S1). On the other hand, the expression of most cholesterol biosynthetic enzymes was decreased in tumor tissue relative to normal kidney (Supplementary Table S2). Although altered expression of triglyceride and cholesterol synthesis enzymes may affect lipid synthesis and storage in ccRCC, we focused our attention on PLIN2 for multiple reasons. First of all, functional studies in murine models of hepatosteatosis have revealed that PLIN2 upregulation is necessary for lipid storage and enhanced de novo lipid synthesis, even if other lipogenic genes were overexpressed (i.e., Elovl3, Fasn, Scd1; refs. 19, 20). Secondly, overexpression of PLIN2 alone was sufficient to drive lipid synthesis and storage in murine liver (19) and mouse embryonic fibroblasts (21). Taken together with the observation that HIF2α activation and PLIN2 expression are positively correlated in various settings (11, 22), these findings suggested a functional connection between two hallmarks of ccRCC—constitutive HIF activity and heightened neutral lipid storage.

Analysis of microarray data profiling gene expression in normal kidney and distinct renal malignancies (23) revealed that PLIN2, along with a panel of established HIF2α target genes, is specifically elevated in ccRCC samples (Supplementary Fig. S1B). Because pVHL loss of function is a defining feature of most ccRCC, we examined whether HIF activation promotes PLIN2 expression. To explore this link in ccRCC patient samples, we performed transcriptome profiling of tumors previously grouped into three categories based on HIFα staining and VHL gene sequencing: VHL wild-type (WT; no HIFα staining), HIF1α and HIF2α expressing (“H1H2”), and HIF2α only (“H2”; ref. 4). This analysis indicated that PLIN2 mRNA was elevated in both H1H2 (9.7-fold overexpression, P = 2.6E–4) and H2 (16.7-fold overexpression, P = 2.5E–6) tumors when compared with VHL WT samples (Fig. 1D). Multiple documented HIF2α targets in ccRCC exhibited a similar expression pattern (Fig. 1D). Collectively, these observations suggest that PLIN2 levels increase early in disease progression and correlate with enhanced HIF2α activity in ccRCC patient samples.

To directly examine the role of HIF2α in PLIN2 regulation in ccRCC, we depleted HIF2A from 786-O (H2) and RCC4 (H1H2) ccRCC cell lines using multiple shRNAs and found that PLIN2 mRNA and protein expression was reduced (Fig. 1E and F). In contrast, HIF1A silencing in RCC4 cells actually enhanced PLIN2 mRNA and protein levels, while decreasing levels of the HIF1α target pyruvate dehydrogenase kinase 1 (PDK1; Fig. 1F). These results are consistent with the PLIN2 expression pattern we observed in primary patient samples (Fig. 1D). We also determined whether PPAR gamma (PPARγ) regulates PLIN2 in ccRCC, as PPARγ is overexpressed in ccRCC tissue and was previously shown to stimulate PLIN2 transcription in the settings of hepatosteatosis and foam cell formation (24). In A498 cells, which exhibit high levels of both PPARγ and PLIN2, PPARG shRNAs reduced PPAR–response element reporter activity, but not PLIN2 levels (Supplementary Fig. S2A and S2B). PPARα (PPARA) also promotes PLIN2 transcription in states of lipid accumulation (10). However, expression of PPARA and many of its target genes is reduced in ccRCC compared with normal kidney tissue (Supplementary Fig. S2C and S2D), decreasing the likelihood that it regulates PLIN2 in ccRCC. These findings suggest that constitutive HIF2α activity, rather than PPARγ, PPARα, or HIF1α, regulates PLIN2 in both ccRCC cell lines and primary patient samples.

PLIN2 is commonly used as a marker of cellular lipid accumulation, but its overexpression is also sufficient to increase lipid synthesis and storage in murine fibroblasts in vitro (21) and liver in vivo (19). Given the numerous tumor-promoting processes downstream of HIF2α, we tested whether PLIN2 restoration alone was sufficient for tumor growth following HIF2α suppression in multiple model systems. Doxycycline-inducible shRNA was used to deplete HIF2A in 786-O cells constitutively expressing either pCDH empty vector (EV) or an exogenous PLIN2 cDNA. Xenografts generated from these two cell lines grew at identical rates before doxycycline exposure (data not shown). After administration of doxycycline chow to inhibit HIF2α, EV xenografts exhibited a steady decline in tumor volume and reduced mass at sacrifice, whereas those expressing exogenous PLIN2 demonstrated partial maintenance of both tumor volume and mass (Fig. 2A, left and center). qRT-PCR analysis confirmed that HIF2α regulates PLIN2 in vivo, and exogenous PLIN2 mRNA levels were maintained following doxycycline-induced HIF2α depletion (Fig. 2A, right). Furthermore, oil red O staining indicated that HIF2α loss reduced neutral lipid staining, whereas exogenous PLIN2 fully restored lipid storage (Fig. 2B and C). These results indicate that PLIN2 is both necessary and sufficient to promote neutral lipid storage in ccRCC xenografts. Further histologic analysis revealed that HIF2α depletion dramatically decreased tumor cell proliferation, as indicated by Ki67 staining, which was partially restored by exogenous PLIN2 (Fig. 2B and C). We also found that HIF2α suppression resulted in large areas of tissue necrosis and induction of apoptosis, based on cleaved caspase 3 accumulation, which were both reduced by PLIN2 expression (Fig. 2B and C).

Despite the complexity of HIF2α-dependent tumorigenesis in vivo, including cell-extrinsic effects such as angiogenesis, PLIN2 restoration alone significantly restored tumor cell proliferation and viability. We reasoned that enhanced lipid storage imparts a cell-intrinsic advantage to tumor cells. To test this hypothesis, we generated in vitro three-dimensional tumor spheroids using 786-O cells, which recapitulate nutrient and O₂ gradients within solid tumors (Supplementary Fig. S3A). In this assay, HIF2α inhibition was associated with loss of neutral lipid staining and enhanced cell death, whereas exogenous PLIN2 expression partially restored lipid storage and reduced cell death (Supplementary Fig. S3B and S3C). Thus, multiple tumor model systems suggest an essential role for PLIN2-dependent lipid storage downstream of HIF2α in ccRCC.

The finding that PLIN2 is required for cell viability in ccRCC xenografts was surprising, as acute PLIN2 depletion experiments have reported no effects on cell viability either in vivo or in vitro (19, 20, 25). To assess the effects of direct PLIN2 depletion in ccRCC, we expressed multiple shRNAs targeting PLIN2 or a scrambled control (SCR) in 786-O and A498 cells. In both cell lines, we observed a dosage-dependent loss of cell viability and proliferation that correlated with degree of PLIN2 depletion (Fig. 3A and B; Supplementary Fig. S4A and S4B). Oil red O staining and boron-dipyrromethene (BODIPY) 493/503 quantification also revealed dosage-dependent decreases in neutral lipid levels (Fig. 3C and D). To assess functional lipid storage capacity, cells were treated with oleic acid, a potent inducer of triglyceride synthesis and neutral lipid storage that is selectively toxic for cells incapable of storing it as triglyceride (26). Consistent with decreased ability to store lipids within LDs, PLIN2-depleted cells were preferentially sensitized to oleic acid–induced cell death (Fig. 3E).

In light of the intimate ER/LD relationship and evidence that altered membrane properties can trigger ER stress (13, 15), we reasoned that decreased lipid storage capacity could disrupt ER homeostasis and trigger the UPR, cell cycle withdrawal, and cell death. PLIN2 ablation in 786-O cells elicited dosage-dependent activation of UPR sensors PERK, IRE1α, and ATF6 and induction of multiple UPR target genes (Fig. 4A). Furthermore, ER Tracker imaging and flow cytometry indicated ER expansion in PLIN2-depleted cells (Fig. 4B), and ultrastructural analysis by transmission electron microscopy (TEM) confirmed the presence of dilated and irregularly shaped rough ER (Fig. 4C), both of which are consistent with ER stress (27). Similarly, PLIN2 depletion in A498 cells elicited morphologic and gene expression changes indicative of ER stress (Supplementary Fig. S4C and S4D). Based on these observations, we quantified UPR target gene expression in HIF2α-deficient xenograft tumors. Indeed, multiple UPR targets, including the “terminal” UPR genes CHOP and TXNIP, were elevated in HIF2α-depleted tumors and decreased by exogenous PLIN2 expression (Fig. 4D).

To determine if UPR activation promotes cell death upon PLIN2 loss in ccRCC, we utilized previously characterized small-molecule PERK and IRE1α inhibitors, along with siRNA against ATF6, to suppress UPR signaling (28, 29). These tools were validated in A498 cells, based on UPR sensor phosphorylation status and target gene expression (Supplementary Fig. S5A–S5D). In 786-O cells, PERK inhibition reduced ATF3 levels (a PERK/ATF4 target) in PLIN2-depleted cells, but enhanced cell death and expression of multiple IRE1α and ATF6 target genes (Fig. 5A; Supplementary Fig. S5B–S5D). Whereas IRE1α and ATF6 suppression each modestly restored cell viability, combined suppression of both yielded enhanced cell survival (Fig. 5B). In A498 cells, ATF6 promoted cell death downstream of PLIN2 inhibition (Supplementary Fig. S6A and S6B). Although the specific UPR sensor(s) mediating cell death varies between cell lines, our results indicate that PLIN2 is required for maintenance of ER homeostasis and prevention of cytotoxic ER stress in ccRCC. Next, we examined the potential sources of ER stress that could explain the enhanced requirement for PLIN2-mediated ER homeostasis in ccRCC.

The observation that ccRCC cells require PLIN2 for proliferation and viability was surprising, because (i) Plin2 knockout mice are viable (30); (ii) Plin2-deficient macrophages do not exhibit enhanced sensitivity to cholesterol loading, which requires lipid storage to alleviate ER stress (31); and (iii) acute Plin2 suppression fails to elicit cell death in steatotic hepatocytes (19, 20), MCF7 breast cancer cells, or U87 glioblastoma cells (25). One potential advantage of lipid storage in ccRCC is the ability to derive energy from fatty-acid breakdown via β-oxidation. However, in contrast with the electron transport chain inhibitor rotenone, PLIN2 depletion did not affect ATP levels in multiple ccRCC cell lines (Supplementary Fig. S7A). Furthermore, transcription factors that promote renal tubular cell β-oxidation, including PPARA and PPARγ coactivator 1-alpha (PPARGC1A), were substantially downregulated in primary ccRCC compared with normal kidney (Supplementary Figs. S2C and S7B). The carnitine/acyl-carnitine transporter (CAT) and mitochondrial β-oxidation enzymes were also underexpressed in ccRCC tissues (Supplementary Fig. S7B and S7C). In agreement with these findings, metabolomic analysis of primary ccRCC and normal kidney tissues revealed elevation of acyl-carnitine levels in tumors (Supplementary Table S3). These features of ccRCC mirror genetic CAT deficiency, which manifests as acyl-carnitine buildup secondary to reduced β-oxidation flux (32). Furthermore, a recent study indicates that HIF2α suppresses peroxisomal β-oxidation via selective autophagy of peroxisomes (33). Thus, alterations to β-oxidation are unlikely to explain the effects of PLIN2 depletion in ccRCC.

Next, we explored whether enhanced tumor anabolic processes, downstream of oncogenic activation, contribute to the requirement for PLIN2-dependent lipid storage in ccRCC. Particularly, increased protein and lipid synthesis via mTORC1 could result in a greater requirement for lipid storage to maintain ER homeostasis, as the ER functions as a “hub” for both protein and lipid production. We focused on mTORC1 because (i) 30% of ccRCC harbor activating mutations in the mTOR pathway (18); (ii) most tumors exhibit elevated mTORC1 activity (pS6K1, p4EBP1 staining; refs. 4, 34); and (iii) mTORC1 can stimulate both protein and lipid synthesis (35, 36). The effects of small-molecule mTOR inhibitors on protein and lipid synthesis were characterized in ccRCC cells. As expected (35), Torin1 was more potent than rapamycin in suppressing protein synthesis and lipogenic enzyme gene expression in ccRCC cells (Fig. 6A and B). In both 786-O and A498 cells, rapamycin and Torin1 treatment enhanced viability and reduced UPR gene expression following PLIN2 depletion, with Torin1 being more potent (Fig. 6C; Supplementary Fig. S8A and S8B). Notably, the magnitude of restored viability upon mTOR suppression was greater than that observed upon UPR inhibition in both 786-O and A498 cells (Figs. 5A and B and 6C; Supplementary Fig. S8B). We reasoned that this was due to amelioration of ER stress, rather than merely UPR signaling itself.

Given that Torin1 suppresses both protein and lipid synthesis, we measured the relative contribution of these activities toward Torin1-dependent effects. Suppression of protein synthesis using the translation inhibitor cycloheximide restored cell viability and ameliorated ER stress in PLIN2-depleted cells (Fig. 6D and E). On the other hand, silencing of sterol regulatory element-binding proteins 1 and 2 (SREBP1/2), mediators of lipid synthesis downstream of mTORC1 (35), selectively reduced viability in PLIN2-depleted cells (Supplementary Fig. S8C and S8D). Consistent with an adaptive function of SREBP1/2 activation in cells experiencing ER stress (16, 37), PLIN2 depletion induced multiple lipid synthesis enzymes in an SREBP1/2–dependent manner (Supplementary Fig. S8C and S8D). Ultimately, cycloheximide treatment restored cell viability in PLIN2-depleted cells, even when SREBP1 and 2 were inhibited, reflecting the aggregate activities of Torin1 (Supplementary Fig. S8C and S8D). These findings suggest that protein synthesis is a prominent source of ER stress and cell death in the setting of PLIN2 depletion.

Given that reduction of ER protein load alleviated ER stress in PLIN2-depleted cells, we determined if such cells would also be more sensitive to agents that increase misfolded protein load. Of note, PLIN2-depleted cells were more sensitive to tunicamycin treatment, which inhibits N-linked glycosylation, compared with controls (Fig. 7A). We performed similar experiments to determine if HIF2α/PLIN2–mediated lipid storage is similarly protective against pharmacologic ER stress. A498 cells expressing HIF2A shRNA exhibited reduced BODIPY staining, whereas exogenous PLIN2 expression was sufficient to restore neutral lipid levels (Fig. 7B). Upon treatment with tunicamycin, HIF2α-depleted cells exhibited a 3-fold enhancement of cell death that was partially ameliorated by restoring PLIN2-dependent lipid storage (Fig. 7C). Treatment with brefeldin A, an ER stress–inducing agent that inhibits ER to Golgi vesicular transport, yielded consistent results (Supplementary Fig. S9A and S9B).

PLIN2 also protected ccRCC cells from the proteasome inhibitor bortezomib (Fig. 7D), an FDA-approved therapy for multiple myeloma that functions partly through ER stress induction (17). Specifically, expression of the IRE1α substrate spliced XBP1 is positively correlated with patient response to bortezomib, and functional studies demonstrate a role for the UPR target gene CHOP in bortezomib-mediated cell death (38, 39). Consistent with a cytotoxic function of ER stress in bortezomib-treated ccRCC cells, the enhanced efficacy of bortezomib in PLIN2-depleted cells was associated with elevated levels of spliced XBP1 and CHOP (Fig. 7E). Furthermore, HIF2α-depleted cells demonstrated enhanced sensitivity to bortezomib that was ameliorated by exogenous PLIN2 (Fig. 7F). Next, we tested whether nutrient and/or O₂ deprivation—ER stress–inducing conditions found within the tumor microenvironment (13)—could further enhance bortezomib-induced cell death. Indeed, previous work indicates that hypoxia can enhance antitumor activity of bortezomib via ER stress induction (40). Growth under conditions of serum and/or O₂ deprivation enhanced bortezomib-induced cell death, most prominently in PLIN2-depleted cells (Fig. 7G). Under each condition tested, expression of the terminal UPR gene CHOP was also positively correlated with degree of cell death (Supplementary Fig. S9C).

Collectively, we suggest a model in which HIF2α/PLIN2–dependent lipid storage promotes ER homeostasis and prevents cytotoxic ER stress in ccRCC cells (Fig. 7H). This phenotype promotes cell viability under multiple conditions that perturb ER homeostasis, including growth under limited nutrient/O₂ delivery within solid tumors, enhanced protein synthesis downstream of oncogenic activation, and exposure to pharmacologic ER stress–inducing drugs.

Despite the long-standing observation that ccRCC tumor cells exhibit abundant intracellular LDs, a clear function for this phenotype had not been identified. In this study, we explored the role of LD coat proteins in ccRCC progression. Analysis of multiple cohorts of primary ccRCC patient samples revealed PLIN2 overexpression in tumor samples and suggested a functional relationship between pVHL loss, constitutive HIF2α activation, and PLIN2 accumulation. Although previous reports indicated that PLIN2 expression and lipid storage correlated with HIF2α activation, it was unknown whether PLIN2 was a driver of this phenotype, or a passenger of broader metabolic changes (11). Our findings indicate that PLIN2 is both necessary and sufficient to promote lipid storage in ccRCC cell lines. Mechanistically, HIF2α-dependent PLIN2 expression and lipid storage are required for maintenance of ER homeostasis and prevention of cytotoxic ER stress.

The significant requirement for PLIN2 in ccRCC cells was intriguing, as Plin2^−/− mice are viable, and acute PLIN2 depletion studies in settings of lipid accumulation (i.e., hepatosteatosis or foam cell formation) have reported no evidence of ER stress or cell death (30, 31, 41). We provide two potential explanations for this observation. First of all, in physiologic scenarios, PPAR family members coordinately enhance expression of PLIN2 and other PAT LD coat proteins (10). In these settings, PLIN2 loss of function is associated with compensatory upregulation of other LD coat proteins (30, 41). However, we determined that PLIN2 is upregulated in ccRCC due to HIF2α activation, rather than by PPARγ or PPARα. Moreover, PLIN2 is overexpressed independently of other PAT LD coat proteins in ccRCC, likely explaining why functional compensation cannot be achieved after PLIN2 depletion. Secondly, our results suggest that an enhanced requirement for PLIN2-dependent lipid storage and ER homeostasis could arise from heightened ER stress downstream of oncogene activation. These include cell-intrinsic stress from enhanced protein synthesis and cell-extrinsic stress due to commitment to a growth rate that outstrips nutrient and O₂ delivery.

Our results fit within an emerging theme in which oncogenic transformation coordinately induces both anabolic processes to increase proliferation and homeostatic pathways that maintain cell viability. These include proteasome activity downstream of mTORC1, autophagy downstream of MYC overexpression (42), and lipid/protein scavenging by RAS-transformed tumors (43, 44). Thus, different oncogenes appear to solve the problem of balancing proliferation and cellular homeostasis in unique ways. In the case of ccRCC, we suggest a model in which HIF2α-dependent lipid storage occurs early in disease progression and functions to buffer tumor cells against cell-intrinsic and cell-extrinsic sources of ER stress. In ccRCC, the heightened proliferation and anabolic metabolism are driven by HIF2α-dependent processes, such as autocrine growth factor signaling via TGFα and VEGFA, mTORC1 stimulation, and cell-cycle progression, and independent oncogenic events that activate mitogenic pathways (6, 18, 45). Although enhanced lipid storage is a hallmark feature of ccRCC, this phenotype is observed in other malignancies, including Burkitt lymphoma, hepatocellular carcinoma, and advanced prostate cancer (46, 47). Although the underlying mechanisms of lipid storage and the function it serves may vary among cancer types, additional studies into the role of lipid storage in cancer are warranted.

Our results indicate that heightened protein synthesis is a prominent source of ER stress in PLIN2-deficient ccRCC cells. Although the initial perturbation to ER homeostasis in such cells is likely due to alterations in ER lipid content, the cumulative level of ER stress likely arises from dysregulation of both protein and lipid metabolism, which independently trigger the UPR (14). In addition, disruption of ER lipid composition can further impair protein-folding capacity and enhance ER stress (48). Mechanistically, PLIN2 promotes neutral lipid content through at least two, non–mutually exclusive, mechanisms: enhancing lipid storage and suppressing lipolysis. Our observation that oleic acid is selectively toxic to PLIN2-depleted cells suggests that at least a portion of the PLIN2-deficient phenotype arises from loss of the ability to package lipids into LDs. This phenotype is also observed in mouse embryonic fibroblasts that are deficient in enzymes (diacylglyceride-acyltransferases) that are required to incorporate oleic acid into triglycerides (26). However, it remains to be tested how PLIN2 affects lipolysis in ccRCC. In addition, future studies that characterize the specific changes in ER lipid composition following PLIN2 depletion may provide additional strategies to enhance tumor cell ER stress.

Lastly, our finding that loss of HIF2α/PLIN2–dependent lipid storage enhances sensitivity to ER stress–inducing agents has implications for ccRCC therapy. A therapeutic index has previously been demonstrated for the proteasome inhibitor bortezomib in multiple myeloma, where heightened immunoglobulin synthesis within ER renders cells more sensitive to pharmacologic ER stress (17). A small phase II clinical trial evaluating bortezomib monotherapy for advanced renal cancer revealed partial responses in only 12% (3/25) of ccRCC patients (49). HIF2α/PLIN2–dependent lipid storage and ER stress resistance could contribute to this limited response rate. Specifically, our observation that multiple components of the ERAD machinery (HERP, HRD1, ERdj4) were induced in HIF2α-depleted tumors provides initial evidence for the rational combination of proteasome inhibitors and HIF2α suppression, especially as HIF2α-specific inhibitors are currently under development for treatment of ccRCC (Clinical Trial No. NCT02293980).

Deidentified fresh-frozen ccRCC and matched normal kidney patient samples were obtained from the Cooperative Human Tissue Network (CHTN), which operates with the review and approval of their local Institutional Review Boards. Samples were obtained from the University of Pennsylvania and Vanderbilt University CHTN divisions.

Authenticated (short tandem repeat profiling) human ccRCC cell lines 786-O, A498, and RCC4 were obtained from the American Type Culture Collection in 2001. Cells were cultured for a maximum of 4 weeks before thawing fresh, early passage cells. All cells were confirmed to be Mycoplasma negative (MycoAlert; tested June 2014) and verified for pVHL and HIFα expression status using Western blot analysis. Cells were cultured in DMEM plus 10% FBS. Cell viability was determined using the FITC–Annexin V, PI Kit (cat. 556547) from BD Biosciences according to the manufacturer's instructions. Flow cytometry was performed using the BD Accuri C6 instrument, and double-negative cells were deemed viable.

Three-dimensional spheroid cultures were generated using the liquid overlay technique. Twenty-four–well plates were coated with 1% agarose in DMEM before plating 100,000 cells per well in DMEM plus 10% FBS. To promote spheroid formation, plates were swirled before incubation. Media were changed every 3 days, and spheroids were harvested after 9 days. Hypoxic cells were labeled by incubating spheroids with 200 μmol/L FITC-conjugated pimonidazole hydrochloride (Hypoxyprobe, cat. HP2) before fixation. For BODIPY 493/503 quantification, spheroids were dissociated with Accutase at 37°C for 30 minutes and stained as described below in the section concerning BODIPY staining.

Oleic acid (cat. O3008), rapamycin (cat. R8781), tunicamycin (cat. T7765), and brefeldin A (cat. B7651) were purchased from Sigma Aldrich. GSK2656157 PERK inhibitor (cat. 5046510001), 4μ8C IRE-1α inhibitor (cat. 412512), Torin1 (cat. 475991), and cycloheximide (cat. 239763) were purchased from Millipore. siRNA pools targeting human ATF6 (cat. L-009917), SREBF1 (cat. L-006891), and SREBF2 (cat. L-009549) were purchased from Dharmacon. Rotenone (cat. 557368) was purchased from EMD Chemicals. Bortezomib was purchased from Cell Signaling Technologies (cat. 2204S).

The lentiviral vector PLKO.1 SCR (plasmid no. 17920) was obtained from Addgene. pLKO.1 vectors expressing shHIF1A_9 (TRCN0000003809), shHIF2A_6 (TRCN0000003806), shHIFA_7 (TRCN0000003807), shPLIN2_1 (TRCN0000136605), shPLIN2_2 (TRCN0000136481), shPPARG_2 (TRCN0000001672), and shPPARG_3 (TRCN0000001673) were obtained from The RNAi Consortium shRNA library at the Broad Institute. The GIPZ vector expressing shHIF1A_52 (V3LMM_441752) was obtained from Dharmacon. The PLIN2 open reading frame was subcloned from Mammalian Gene Collection sequence–verified cDNA (Dharmacon; clone ID: 3844174) into the pCDH-CMV-MCS-EF1-HYGRO mammalian expression vector. The doxycycline-inducible shHIF2A_7 construct was generated using the “Tet-pLKO-puro” plasmid (Addgene; cat. 21915).

Lentivirus was produced by transfecting 293T cells with the indicated expression plasmid, pRSV-Rev, pMDL, and pCMV-VSV-G plasmids using Fugene6 (Promega). The virus was harvested 48 hours after transfection. For viral infection, cells were incubated with medium containing virus and 8 μg/mL polybrene for 16 hours. Cells were allowed to recover for 48 hours before antibiotic selection, and surviving pools were utilized for downstream analyses.

Subcutaneous xenograft experiments were approved by the Animal Care and Use Committee at the University of Pennsylvania. Female NIH-III nude mice (Charles River; 4–6 weeks old) were injected in each flank with 5 million cells in a 1:1 mixture of PBS and Matrigel (BD 356234). Tumor volume was monitored by caliper measurements. After tumors reached 300 mm³, mice were split into cohorts receiving standard chow or doxycycline chow (625 mg/kg; Harlan Labs; cat. TD05125) ad libitum. After 11 days on the indicated chow, animals were sacrificed by CO₂ inhalation, and xenograft tumors were dissected for downstream analyses.

Level 3 RNA-seq data for 480 ccRCC and 69 normal kidney samples were downloaded from the TCGA on April 2, 2013. Differential gene expression analysis of tumor and normal samples was performed using DeSeq (Bioconductor Version 2.12). Box and whisker plots correspond to 1–99th percentiles (bars), 25–75th percentiles (box), and median (line in box). Differentially expressed genes were subjected to gene set enrichment analysis using the Broad Institute Molecular Signature Database.

Classification of primary ccRCC samples into VHL WT, H1H2, and H2 subgroups and microarray analysis is described in Gordan and colleagues (4). Expression data are deposited at the NCBI Gene Expression Omnibus (GEO) under GSE11904. Expression analysis comparing normal kidney, ccRCC, papillary RCC, and chromophobe RCC was described in Jones and colleagues (23). Data were downloaded from GEO (GSE15641).

For frozen patient samples, optimal cutting temperature (OCT)–embedded tissue was cut to 10-μm sections and fixed in 4% paraformaldehyde before staining. For xenograft tumors, samples were fixed in 4% paraformaldehyde, equilibrated in 30% w/v sucrose, and embedded in OCT. Sections (10 μm) were cut for staining. Hematoxylin and eosin staining was performed as previously described (4).

A working oil red O solution was generated by diluting a 3.5 mg/mL stock (in 100% isopropanol) 6:4 with distilled water. This solution was incubated at room temperature for 30 minutes and filtered in Whatman paper before use. Tissue sections were incubated in 60% isopropanol for 5 minutes, dried at room temperature, and incubated in oil red O staining solution for 1 hour at room temperature. Slides were rinsed in distilled water and counterstained with hematoxylin before mounting in Prolong Gold Antifade with 4′,6-diamidino-2-phenylindole (DAPI; Life Technologies; cat. P36935).

Slides were treated in 1% hydrogen peroxide for 30 minutes and blocked in 2% normal goat serum and 4% BSA in Tris buffer with Tween 20. Avidin/biotin blocking was performed, and sections were incubated with primary antibodies overnight at 4°C. Ki67 antibody was used at 1:100 (BD; cat. 550609). Cleaved caspase 3 (Asp175) antibody was used at 1:400 (Cell Signaling Technology; cat. 9661). Slides were incubated in 1:200 dilutions of biotinylated goat anti-mouse (Vector Labs; cat. BA-9200) or anti-rabbit (Vector Labs; cat. BA-1000) secondary antibodies for 1 hour at room temperature. Sections were then processed using the Vectastain Elite ABC Kit (Vector Labs; PK-6100) and DAB peroxidase substrate kit (Vector Labs; cat. SK-4100), dehydrated in a standard ethanol/xylenes series, and mounted in 75% v/v Permount (Fischer; cat. SP15-500) in xylenes.

Slides were incubated in 50 mmol/L ammonium chloride for 10 minutes, permeabilized with 0.25% Triton X-100 for 10 minutes, and blocked in 2% normal goat serum and 4% BSA for 1 hour. Slides were incubated with cleaved caspase 3 (Asp175) antibody at 1:400 (Cell Signaling Technology; cat. 9661) overnight at 4°C. Secondary Alexa Fluor 488 goat anti-rabbit (Life Technologies; cat. A-11008) was used at 1:200 for 1 hour at room temperature. Slides were mounted in Prolong Gold Antifade with DAPI before imaging.

Total RNA was isolated using the RNAeasy purification kit (Qiagen). cDNA was synthesized using the Applied Biosystems High Capacity RNA-to-cDNA master mix. qRT-PCR was performed on a ViiA7 Real-Time PCR system from Applied Biosystems. Predesigned Taqman primers were obtained from Life Technologies for the following genes: TBP (HS01060665_G1), ACTB (HS01060665_G1), VEGFA (HS00900055_M1), PLIN2 (HS00605340_M1), HIF2A/EPAS1 (HS01026149_M1), HIF1A (HS00153153_M1), TGFA (HS00608187_M1), PDK1 (HS01561850_M1), PLIN3 (HS00998416_M1), BiP/HSPA5 (HS00946084_G1), XBP1 (spliced; HS03929085_G1), CHOP/DDIT3 (HS00358796_G1), ERO1A/ERO1L (HS00205880_M1), HERP/HERPUD1 (HS01124269_M1), EDEM1 (HS00976004_M1), ERdj4/DNAJB9 (HS01052402_M1), HRD1/SVN1 (HS00381211_M1), and ATF6 (HS00232586_M1). SYBR-green primers were utilized for human ATF3 (forward: TAGGCTGGAAGAGCCAAAGA, reverse: TTCTCACAGCTGCAAACACC).

BODIPY 493/503 (cat. D3922) was purchased from Life Technologies. Live cells were washed twice in PBS and incubated in 2 μg/mL BODIPY in PBS for 15 minutes at 37°C. After staining, cells were washed twice in PBS and fixed in 2% paraformaldehyde for 15 minutes. Fixed cells were washed and resuspended in PBS, passed through a cell strainer, and analyzed on an Accuri C6 flow cytometer under FL-1. ER-Tracker Red (cat. E34250) was purchased from Life Technologies. Live cells were incubated with 1 μmol/L ER Tracker on DMEM with 10% FBS for 30 minutes. Cells were washed twice in PBS, resuspended in PBS with 5% serum, passed through a cell strainer, and analyzed on an Accuri C6 flow cytometer under FL-3. Data analysis was performed using FlowJo software.

Cells were fixed with 2.5% glutaraldehyde, 2.0% paraformaldehyde in 0.1 mol/L sodium cacodylate buffer, pH 7.4, overnight at 4°C. After subsequent buffer washes, the samples were postfixed in 2.0% osmium tetroxide for 1 hour at room temperature, and then washed again in buffer followed by distilled water. After dehydration through a graded ethanol series, the tissue was infiltrated and embedded in EMbed-812 (Electron Microscopy Sciences). Thin sections were stained with uranyl acetate and lead citrate and examined with a JEOL 1010 electron microscope fitted with a Hamamatsu digital camera and AMT Advantage image capture software.

The PPRE X3-TK-luc plasmid was purchased from Addgene (No. 1015). Cells (30,000) were seeded into 24-well plates and transfected with 1 μg of PPRE X3-TK-luc and 100 ng of Renilla luciferase plasmids using Fugene 6 (Promega). Luciferase activity was measured 2 days after transfection using the Dual Luciferase assay kit (Promega).

The ATP luminescence assay system (cat. 6016941) was purchased from Perkin Elmer. Cells (100,000) were plated into each well of an opaque 96-well plate and analyzed as described by the manufacturer.

Cells were lysed in lysis buffer (40 mmol/L HEPES, 2 mmol/L EDTA, 10 mmol/L pyrophosphate, 10 mmol/L glycerophosphate, 1% Triton X-100) containing Roche complete ultra protease/phosphatase inhibitor (cat. 05892791001). Nuclear and cytoplasmic fractionation was performed using the Thermo Scientific NE-PER Kit (cat. PI-78833). Isolated proteins were resolved by SDS-PAGE, and Western blot analysis was performed. All primary antibodies were diluted at 1:1,000 in 5% w/v nonfat milk, unless otherwise noted. Blots were incubated with primary antibodies overnight at 4°C. HIF2α (cat. NB100-122) and phospho-serine 724 IRE1α (cat. NB-100-2323) were purchased from Novus Biologicals. HIF1α antibody (cat. 610958) was purchased from BD Biosciences. PLIN2 antibody (cat. ab78920) was purchased from Abcam. β-Actin (1:4,000; cat. SC-47778), ATF6 (cat. SC-22799), and ATF4 (1:2,000; cat. SC-200) antibodies were purchased from Santa Cruz Biotechnology. Cleaved caspase 3 (cat. 9661), PERK (1:4,000; cat. 3192), IRE1α (cat. 3294), phospho-threonine 389 S6K1 (cat. 9234), S6K1 (cat. 2708), phosphor-serine 65 4EBP1 (cat. 9451), 4EBP1 (cat. 9452), FASN (cat. 3180), ACC (cat. 3696), HDAC1 (1:4,000; cat. 5365), and PPARG (cat. 2435) antibodies were purchased from Cell Signaling Technology. Rabbit polyclonal phospho-threonine 980 PERK antibody was a gift from Dr. Alan Diehl. Primary antibodies were detected using horseradish peroxidase–conjugated secondary antibodies (Cell Signaling Technologies) followed by exposure to enhanced chemiluminescence substrate (Pierce).

Protein synthesis was measured as described (50). Briefly, cells were pulsed with puromycin (30 minutes, 10 μg/mL) and chased in puromycin-free media (1 hour). Whole-cell lysates were subjected to Western blot analysis using anti-puromycin antibody (Millipore; cat. MABE343) at 1:20,000.

Mass spectrometry–based metabolomics analysis was performed in conjunction with Metabolon, as previously described (7).

No potential conflicts of interest were disclosed.

Conception and design: B. Qiu, D. Ackerman, B. Li, M.C. Simon

Development of methodology: B. Qiu, B. Li

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): B. Qiu, D.J. Sanchez

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B. Qiu, D. Ackerman, J.D. Ochocki, B. Keith, M.C. Simon

Writing, review, and/or revision of the manuscript: B. Qiu, D. Ackerman, D.J. Sanchez, J.D. Ochocki, A. Grazioli, B. Keith, M.C. Simon

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): B. Qiu, J.A. Diehl

Study supervision: M.C. Simon

Other (generation of high-quality antibody for detection of PERK): E. Bobrovnikova-Marjon

The authors thank John Tobias, PhD, for bioinformatics analyses, Ray Meade for assistance in TEM, and Hongwei Yu for histologic preparations.

This work was supported by the Howard Hughes Medical Institute, NIH grant 2-P01-CA104838 (to M.C. Simon), and NIH fellowship 5-F30-CA177106 (to B. Qiu).

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.