Modeling Renal Cell Carcinoma in Mice: Bap1 and Pbrm1 Inactivation Drive Tumor Grade

Kidney cancer is one of the top 10 most common cancers in the United States, with approximately 60,000 new cases diagnosed annually (1). Most cases of renal cell carcinoma (RCC) are sporadic (>95%), and over 70% are clear cell renal cell carcinoma (ccRCC; ref. 2). Inactivation of the von Hippel–Lindau (VHL) gene by either mutation or methylation is observed in over 80% of ccRCC (3–6). Germline loss-of-function mutations in VHL lead to VHL syndrome and a high incidence of ccRCC (7). The VHL gene, which encodes an E3 ubiquitin ligase that negatively regulates hypoxia-inducible factor (HIF) α subunits, is a typical two-hit tumor suppressor gene, in which both alleles are inactivated for tumor development. Typically, one allele is inactivated by an intragenic mutation, and the second allele is lost as part of a deletion event resulting in loss of heterozygosity (LOH; refs. 3, 8), often involving the whole chromosome 3p arm where the VHL gene is located. Intragenic mutation in one copy of VHL and 3p LOH are early events in ccRCC tumorigenesis, in fact the only consistent “truncal” events (9, 10).

Recurrent mutations in other tumor suppressor genes have also been identified in ccRCC with varying frequencies. Mutations in the Polybromo-1 (PBRM1) gene, encoding BRG1-associated factor 180 (BAF180), a component of the SWI/SNF-B (PBAF) chromatin-remodeling complex, are observed in roughly 50% of ccRCC (11, 12). BRCA1 associated protein-1 (BAP1), encoding a deubiquitinating enzyme of the ubiquitin carboxyl-terminal hydrolase (UCH) family, and SET domain-containing 2 (SETD2), encoding a histone methyltransferase for lysine 36 of histone H3, are mutated in approximately 10% to 15% of ccRCC (13–15). Like VHL, all three of these genes are located on chromosome 3p, within a 50-Mb region, and one copy is deleted along with VHL when 3p is lost (16).

Remarkably, PBRM1 mutation tends to anticorrelate with BAP1 mutation in ccRCC, and tumors with BAP1 versus PBRM1 mutation exhibit distinct biology with markedly different outcomes (4, 17). Tumors with BAP1 mutation are higher grade and more aggressive compared with PBRM1-deficient tumors, and patients have significantly shorter survival rates (13, 17). In addition, whereas BAP1-deficient tumors exhibit higher levels of mTORC1 activation, levels are reduced in PBRM1-deficient tumors (13, 17). These differences led to the first molecular genetic classification of ccRCC (4). However, whether the difference in tumor grade and patient survival is actually driven by BAP1 and PBRM1 loss is not known.

An important limitation of the field for many years has been the lack of a genetically engineered mouse model (GEMM) reproducing genetic events in human ccRCC. Despite the discovery of the VHL gene in 1993 (18), VHL targeting failed to generate such a model. The discovery that BAP1 and PBRM1, two other frequently mutated ccRCC tumor suppressor genes, reside on 3p, together with the finding that this region is lost in the majority of ccRCCs, led us to hypothesize that RCC tumorigenesis requires the simultaneous loss of VHL and BAP1 (or PBRM1; ref. 4). Surprisingly, although VHL heterozygous patients are predisposed to developing ccRCC, Vhl heterozygous mice do not develop ccRCC (19–23). A possible explanation for this may be that in the mouse, Vhl is located on chromosome 6 whereas Pbrm1 and Bap1 are on chromosome 14, and thus LOH in the Vhl region of chromosome 6 would still leave two copies of Pbrm1 and Bap1 intact (4). To evaluate this possibility, we simultaneously targeted Vhl and Bap1 in nephron progenitor cells (NPC). However, Six2-Cre;Vhl^F/F;Bap1^F/F mice died soon after birth. Nevertheless, Six2-Cre;Vhl^F/F;Bap1^F/+ mice developed ccRCC and the kidneys bore a resemblance to the kidneys of patients with VHL syndrome (24). Unfortunately, however, the mice died from renal failure at around 8 months of age (as did Six2-Cre;Vhl^F/F mice), and these tumors remained uniformly small.

Additionally, choosing an appropriate Cre driver is confounded by limited knowledge regarding the cell type of origin. On the basis of previous IHC evidence, it has been proposed that ccRCC arises from the epithelium of proximal convoluted tubule cells (25, 26). Subsequent analyses of gene expression data from normal tissue microdissection from various regions of the nephron (27, 28) supported this idea (29, 30).

We speculated that the paired box 8 (PAX8) gene, which encodes a transcription factor that plays a critical role in kidney development (31, 32), would be a good driver. PAX8 is expressed in the mesonephros, metanephros, nephric duct, and ureteric bud (33, 34). PAX8 is also a sensitive and specific diagnostic marker for renal tumors routinely used in clinical practice (35–37). Here, we report a novel RCC GEMM based on Pax8-Cre (34) deletion of Vhl together with either Bap1 or Pbrm1 and show that Bap1 and Pbrm1 are not only drivers of RCC, but also determinants of tumor grade.

To optimally model ccRCC in the mouse, we sought to identify a driver that could be used to broadly target genes implicated in RCC pathogenesis. PAX8, a known nephric lineage transcription factor, is activated later in development than SIX2, and was particularly attractive as it is routinely used in clinical practice for diagnosis of primary and metastatic RCC (36, 37). IHC staining for PAX8 yields a robust nuclear signal in normal tubular epithelial cells, and in both primary and metastatic ccRCC (Fig. 1A and not shown). Tissue microarray (TMA) analyses of 123 molecularly annotated UT Southwestern ccRCCs showed positive staining for PAX8 in normal renal tubules and 97% (119/123) of ccRCCs. Importantly, nearly all BAP1- and PBRM1-deficient ccRCCs expressed PAX8 (Fig. 1B and C). In normal human kidney, PAX8 is expressed in most epithelial cells of proximal and distal renal tubules, loops of Henle, collecting ducts, and the parietal epithelial cells of Bowman capsule. Interestingly, and as shown in Fig. 1A, PAX8 staining is stronger in the parietal cells of Bowman capsule, distal tubules, and collecting ducts compared to the proximal tubules, in which weak cytoplasmic staining is also seen. A similar PAX8 pattern of expression was seen in normal adult mouse kidney (Fig. 1A).

To explore Pax8 as a driver, we evaluated the overlap between the Pax8 lineage and cells expressing Bap1 and Pbrm1 in the kidney. We crossed Pax8-Cre mice with a Rosa26-CAG-loxP-stop-loxP-tdTomato reporter line to generate Pax8-Cre-tdT mice. Expression of Pax8-Cre recombinase in these mice resulted in indelible labeling of the cells and their progeny with tdTomato (tdT) fluorescence (Fig. 1D). Lineage-tracing experiments showed that Bowman capsule and renal tubular cells were broadly derived from Pax8-Cre–expressing cells (Fig. 1D). Coimmunostaining with Lotus tetragonolobus lectin (LTL), a proximal tubule marker, revealed that Pax8-lineage cells included proximal tubules (Fig. 1D).

Next, we determined the overlap between Bap1- and Pbrm1-expressing cells and Pax8-lineage cells. We crossed Pax8-Cre-tdT mice with knock-in Bap1-LacZ and Pbrm1-LacZ mice harboring a β-galactosidase gene trap in the respective loci. As assessed by β-galactosidase immunostaining, Bap1 was widely expressed in the kidney, and we observed that most Bap1-expressing cells were labeled with tdTomato, indicating that a Pax8 driver would disrupt Bap1 expression in most Bap1-expressing cells (Fig. 1E). Similar results were observed for Pbrm1-expressing cells (Fig. 1F). These data show that Pax8 is a suitable driver to disrupt normal Bap1 and Pbrm1 function in the kidney.

Mouse cohorts harboring Pax8-Cre–driven deletion of Vhl and Bap1 were generated. As shown by PCR genotyping analysis, recombination of LoxP sites occurred primarily in the kidney (Supplementary Fig. S1A). As further confirmation, Western blot analyses demonstrated that VHL and BAP1 protein levels were greatly reduced in the kidneys of Pax8-Cre;Vhl^F/F;Bap1^F/F mice (Fig. 2A). As established by the evaluation of downstream targets, deletion of both Vhl and Bap1 resulted in activation of the corresponding effector pathways (Fig. 2A and B). Deletion of Vhl leads to HIF2 and HIF1 activation and increased expression of target genes, including Ccnd1 (encoding cyclin D1), Pai1, Vegf, Glut1, Igfbp3, Pgk1, and Tgfa (Fig. 2B). A HIF target gene overexpressed in ccRCC is carbonic anhydrase IX (CAIX; refs. 38, 39), which was also markedly upregulated (Fig. 2A). BAP1 deubiquitinates H2A (lysine 119 in both human and mouse; refs. 40–42) and the level of ubiquitinated histone H2A protein (Ub-H2A) was dramatically increased (Fig. 2A). These data demonstrate that deletion of Vhl and Bap1 with a Pax8 driver leads to activation of expected downstream pathways.

Pax8-Cre;Vhl^F/F;Bap1^F/F mice were born at expected Mendelian ratios, and lived longer than Six2-Cre;Vhl^F/F;Bap1^F/F mice, which die shortly after birth (24). However, pups were small and sickly, and mice died at around 2–3 months of age (Fig. 2C and D). MRI scans on moribund mice showed small kidney lesions (Fig. 2E). On gross examination, the kidneys were enlarged and pale with macroscopic cortical cystic lesions that varied in size from 0.1 to 1.25 mm (Fig. 2F). Histologically, there was architectural distortion with multiple large cysts and dilatation of the tubules, although the glomeruli appeared normal. Cystic changes, which were seen in all 16 mice (2–4 months of age; Table 1), were more pronounced after age 2.5 months (Fig. 2G). Blood urea nitrogen (BUN) and creatinine were significantly increased, indicating kidney dysfunction as the probable cause of death (Fig. 2H).

Even though the mice died by 3 months, by that time small RCCs and atypical cystic lesions were observed in all mice. The number and size of lesions increased with the age of the mice. We observed a spectrum of premalignant and malignant tumors. Throughout the cortex, numerous simple cysts lined by a single layer of plump columnar epithelial cells with varying degree of cytoplasmic clearing and nuclear atypia were seen (Fig. 2Gi, ii). Some of the cysts showed varying degrees of epithelial proliferation from focal papillation to prominent intracystic growth (Fig. 2Giii–v). Solid tumor masses were infrequent, but also observed (Fig. 2Gvi). The cytologic characteristics of the cells across these lesions were similar and most consistent with neoplastic cells displaying varying degree of nuclear enlargement, pleomorphism, hyperchromasia, nuclear membrane irregularities, and prominent nucleoli. Mitotic figures and apoptotic bodies were frequently identifiable (Fig. 2Gvii and viii). The cell size was larger than that of normal renal tubular epithelium and the cytoplasm varied from eosinophilic to clear and had prominent intracytoplasmic eosinophilic inclusions (Fig. 2Gvi–viii). The morphologic features were indistinguishable from those observed in the Six2-Cre;Vhl^F/F;Bap1^F/+ mice reported previously (24), suggesting that the cell of origin is retained in the narrower Pax8 lineage.

Increased expression of the cell proliferation marker Ki-67 was seen in atypical cysts and neoplastic nodules (Fig. 2I). CAIX, a HIF target and marker of ccRCC, along with CD10, another routinely used ccRCC marker, were positive in cysts and neoplastic nodules in the mutants (Fig. 2I). PAX8 staining was consistently positive in neoplastic cells in the lesions of Pax8-Cre;Vhl^F/F;Bap1^F/F mutants (Fig. 2I).

Pax8-Cre;Vhl^F/F;Bap1^F/+ mice survived much longer, with median survival of around 14.5 months (Fig. 2D). Renal tumors developed in these mice (Fig. 3A) and began to be observed at around 11 months of age (Table 1). Small atypical cysts could be seen starting at around 10 months of age (no lesions were seen in mice less than 7.5 months old). Although the spectrum of cysts and tumors was similar to the Pax8-Cre;Vhl^F/F; Bap1^F/F mice, these lesions were both more numerous and larger in size. The tumors consisted of expansile cystic masses (0.7–2.4 mm in size) with epithelial proliferation and occasional central cavitation, and included in some cases serous fluid (Fig. 3Bi, ii, iv). Scattered solid cystic masses, occasionally exhibiting a nested pattern and vascular network characteristic of human ccRCC, were observed (Fig. 3Bii, iii, vi, vii). Cytologic features were similar to Pax8-Cre;Vhl^F/F;Bap1^F/F cells, but were more pronounced, including pleomorphism, nucleolar prominence, atypia, and mitosis (Fig. 3B). Focal lymphovascular invasion was also identified (Fig. 3Bv). As observed in human ccRCC, some of these tumors accumulated neutral lipids and glycogen (Fig. 3C). Similar to the Pax8-Cre;Vhl^F/F;Bap1^F/F mice and as observed in human ccRCC, the expression of Ki-67, CAIX, CD10, and vimentin was increased (Fig. 3D). BAP1 protein expression by IHC was reduced across lesions and absent in approximately 30% (Fig. 3D). As expected, HIF2 and HIF1 target genes were upregulated (Fig. 3E and F). Finally, BUN and creatinine levels were also significantly elevated in moribund Pax8-Cre;Vhl^F/F;Bap1^F/+ mice, likely explaining their deaths (Fig. 3G).

As a reference, Pax8-Cre;Vhl^F/F mice survived 16–18 months and, histologically, the kidneys appeared normal with occasional manifestations of chronic interstitial inflammation, mild tubular dilatation, or rare simple cysts in older mice. However, no proliferative lesions were observed (Supplementary Fig. S1B), despite evidence of functional Vhl inactivation (Supplementary Fig. S1B). These data are consistent with the notion that loss of Vhl is insufficient for tumorigenesis.

We previously showed increased staining of phosphorylated S6 ribosomal protein [pS6 (Ser240/244)], a marker of mammalian target of rapamycin complex 1 (mTORC1) activation, in BAP1-deficient human ccRCC (13). Provocatively, this was also observed in Vhl;Bap1-deficient neoplastic cells (Figs. 2I and 3D), but was not seen in Pax8-Cre;Vhl^F/F mice (Supplementary Fig. S1B).

Overall, these data show that the lesions that arise in the kidney of Pax8-Cre;Vhl^F/F;Bap1^F/F and Pax8-Cre;Vhl^F/F;Bap1^F/+ mice mimic the morphologic and immunohistochemical characteristics of the corresponding tumors in humans, including increased pS6 staining.

We deleted Pbrm1 sequences using the same Pax8-Cre driver (Supplementary Fig. S2A). As demonstrated by Western blot analysis, VHL and PBRM1 protein levels were greatly reduced in the kidneys of Vhl;Pbrm1-deficient mice (Fig. 4A).

Median survival of Pax8-Cre;Vhl^F/F;Pbrm1^F/+ mice was around 15 months (Fig. 4B), and the kidneys looked similar to wild-type mice (not shown). Unlike the Pax8-Cre;Vhl^F/F; Bap1^F/+ mice, no lesions developed in the kidneys of Pax8-Cre;Vhl^F/F;Pbrm1^F/+ mice (Supplementary Fig. S2B). The kidneys also appeared normal histologically, with occasional manifestations of inflammation or cysts in older mice. Similarly, Pax8-Cre;Vhl^F/+;Pbrm1^F/F mice looked indistinguishable from wild-type mice, surviving typically for >24 months (Supplementary Fig. S2B and Fig. 4B).

In contrast, Pax8-Cre;Vhl^F/F;Pbrm1^F/F mice were born at expected Mendelian ratios, but were runty and smaller over the course of their lifetime compared with littermate controls (Fig. 4C). Median survival was approximately 12 months (Fig. 4B). MRI analyses revealed the presence of extensive large bilateral tumors (Fig. 4D). However, despite their large size, tumors appeared quite homogeneous with low signal intensity on T2-weighted images, and lacked central necrosis.

Grossly, the kidneys of moribund mice were enlarged and firm, with a nodular capsular surface. The cut surface showed numerous variably sized tan masses (Fig. 4E and F). Histologically, the renal architecture of young Pax8-Cre;Vhl^F/F;Pbrm1^F/F mice looked normal until about 7 months of age. Microscopic tumors could be detected in mice at around 9 months of age (Fig. 4G and Table 1). These tumorlets ranged in number from 1 to 26 and were observed in all but one mouse (6/7) analyzed between the ages of 9 and 13 months. However, these tumors were uniformly small. The background kidney also showed small, cystically dilated tubules, dilated Bowman space, simple cysts, and rare scattered atypical cysts. All mice older than 13 months of age showed multiple tumors, with tumor size increasing with the age of the mice, and tumor nodules almost completely replacing the entire kidney after 16 months.

These solid tumors were very different from those seen in Vhl;Bap1-floxed mice. They consisted of solid, circumscribed masses (0.5–12 mm in size) with pushing borders (Fig. 4Gi). Cells were arranged in small nested architecture with a delicate interconnected vascular network (highlighted by CD31 IHC), a pattern prototypic of low-grade human ccRCC (Fig. 4Gi, iii). Cells had a moderate amount of cytoplasm varying from predominantly (or exclusively) eosinophilic to moderately clear (Fig. 4Gi, iii). Nuclei were monomorphic and without prominent nucleoli (Fig. 4Giii). Cystic dilatation of tubules and glomerular capsule was observed to varying degrees in the background.

Tumors showed accumulation of lipid and glycogen (Fig. 4H) and an immunohistochemical staining pattern similar to low-grade human ccRCC, as depicted by positive membranous CAIX, CD10, moderate Ki-67, and strong vimentin expression (Fig. 4I). As expected, PAX8 expression was observed in Vhl-Pbrm1–deficient lesions (Fig. 4I). PBRM1 showed loss of nuclear reactivity in the lesions and most of the normal tubules (Fig. 4I). Notably, in contrast to the RCC arising in Bap1-floxed kidneys, these solid tumors exhibited weak to largely negative pS6 expression (Fig. 4I). As expected, there was increased expression of HIF target genes (Fig. 4A, I, and J).

A significant increase in BUN and creatinine levels was observed in older mice, suggesting that these mice had impaired kidney function (Fig. 4K). However, in contrast to Vhl;Bap1–deficient kidneys, there were fewer cystic changes of the renal parenchyma, with relatively preserved background renal parenchyma, therefore suggesting that the cause of renal failure may be tumor replacement.

We generated a cell line from one tumor, which was confirmed to be Vhl- and Pbrm1-deficient by PCR (Supplementary Fig. S2C–S2E). Immunoblotting confirmed the loss of VHL and PBRM1 proteins (Supplementary Fig. S2F). Consistent with their origin, these cells showed significant induction of both HIF1α and HIF2α, as well as the HIF target CAIX, akin to what is often observed in human ccRCC (Supplementary Fig. S2F).

Spontaneous renal tumor development in Pax8-Cre;Vhl^F/F;Pbrm1^F/F mice, but not other genetic configurations, indicates that Pax8-Cre–driven inactivation of both alleles of Vhl and Pbrm1 is required for ccRCC tumorigenesis. The ccRCCs observed in Pax8-Cre;Vhl^F/F;Pbrm1^F/F kidneys are reminiscent of VHL;PBRM1–deficient human tumors in their morphology and immunohistochemical features. In addition to the solid, low-grade tumors that were predominantly observed, rare, high-grade cystic tumors similar to those seen in the Vhl;Bap1–floxed mice were found in 20% of the mice (Fig. 4L). We observed 11 high-grade cystic Bap1-like tumors and 242 low-grade solid tumors. Notably, the cystic, high-grade tumors showed significantly stronger pS6 expression (Fig. 4L), similar to tumors in Pax8-Cre;Vhl^F/F;Bap1^F/F mice, suggesting that mTORC1 may contribute to higher grade.

To assess the role of mTORC1 in ccRCC tumor grade, we disrupted a single allele of the mTORC1-negative regulator Tsc1. TSC1 forms a complex with TSC2 that functions as a GAP toward the obligatory activator of mTORC1, RHEB. Reduced TSC1/TSC2 complexes increase RHEB-GTP leading to mTORC1 activation (43, 44). Interestingly, loss of one copy of Tsc1 in Pax8-Cre;Vhl^F/F;Pbrm1^F/F;Tsc1^F/+ mice led to the development of higher-grade tumors in 50% of the mice analyzed (Fig. 5A; 5/10 mice). Although the majority of ccRCCs were still of low grade (12 high grade vs. 112 low grade), the high-grade ccRCCs were solid and their morphology suggested that they arose from the lower-grade Pax8-Cre;Vhl^F/F;Pbrm1^F/F tumors. These higher-grade ccRCC had increased nuclear pleomorphism, large nuclei, prominent nucleoli, more abundant cytoplasm with focal prominent clearing, and tumor necrosis (Fig. 5B). In contrast, no high-grade solid tumors were observed among the 242 solid tumors evaluated in Pax8-Cre;Vhl^F/F;Pbrm1^F/F mice. Consistent with our hypothesis, strong pS6 was seen in the high-grade tumors, whereas low-grade tumors in the same mice had weak (or absent) pS6 expression (Fig. 5C). Overall, these data suggest that loss of one Tsc1 allele promotes the progression of lower-grade PBRM1-deficient tumors into higher-grade ccRCC, albeit at low frequency. In addition, in Pax8-Cre;Vhl^F/F;Pbrm1^F/F;Tsc1^F/+ mice tumor development appeared to be accelerated by approximately 3 months (compare Fig. 5A with 4F). Finally, although TSC1 mutations are infrequent in ccRCC, an analysis of COSMIC identifies several human ccRCCs with mutations in VHL, PBRM1, and TSC1, establishing a proof of principle for this gene combination (Supplementary Fig. S2G).

Careful histologic analyses in the Pax8-Cre;Vhl^F/F;Pbrm1^F/F kidneys in particular, but also scattered in other genotypes, showed alterations of the lining parietal epithelial cells of Bowman capsule, suggesting that these cells may give rise to ccRCC. These alterations were of different magnitude and could be arranged in what appeared to be a series of progressive steps with increased abnormalities (Fig. 6A). The lining parietal epithelial cells showed nuclear enlargement and variable cytoplasmic clearing (Fig. 6Ai–ii). Some of these dilated formerly Bowman capsule structures showed only a minute rudimentary glomerular tuft, suggesting that they are the source of cysts (Fig. 6Aiii). Occasionally, parietal epithelial cell proliferation was observed (Fig. 6Aiv-vi; arrow points to entrapped glomerular tuft). Interestingly, similar features were also observed in uninvolved renal parenchyma adjacent to human ccRCC (Fig. 6Bi-vi). These data suggest that the parietal epithelial cell of the Bowman capsule, rather than the proximal convoluted tubule cell, may be the likely cell of origin of these tumors.

Further insight was provided by targeting Vhl together with either Bap1 or Pbrm1 using Cre drivers expressed in proximal tubules, and not in Bowman capsule: Villin-Cre and Sglt2-Cre (45–47). As shown in Fig. 6C and D, Villin- and Sglt2-lineage cells are mainly distributed in proximal tubules. Interestingly, murine cohorts harboring Villin-Cre– or Sglt2-Cre–driven deletion of both alleles of Vhl together with both alleles of either Bap1 or Pbrm1 (n ≥ 14 for each genotype) failed to develop tumors even at an old age (>24 months; Supplementary Fig. S3A). One important caveat is that significantly fewer cells are targeted by Villin and Sglt2 drivers than with Pax8 (Fig. 6C and D). However, substantial recombination rates were observed (Supplementary Fig. S3B and S3C) and the recombination events were associated with loss of BAP1 and PBRM1 (Supplementary Fig. S3A) and activation of downstream VHL targets such as CAIX (Supplementary Fig. S3A). Although we cannot exclude that failure to develop tumors with Villin and Sglt2 drivers results from a reduced number of targeted cells, no tumors were observed even in quite old mice, and our data suggest that Villin- and Sglt2-lineage cells are not transformed by the combined loss of VHL and either BAP1 or PBRM1 and that these cells are not the likely source for ccRCC. Finally, an examination of Villin (VIL1) and SGLT2 (SLC5A2) in ccRCCs from The Cancer Genome Atlas and a University of Texas Southwestern Medical Center (UTSW) cohort reveals expression levels that are comparable to or below those found in normal kidneys (Supplementary Fig. S3D). Lower expression of these proteins in ccRCC adds to the notion that kidney cells expressing these proteins in proximal tubules are not the source of ccRCC.

Genomic analyses of ccRCC by us and others identified frequent mutation of the PBRM1 and BAP1 genes in 50% and 15% of tumors, respectively (11, 13). However, whether loss of BAP1 and PBRM1 causes ccRCC was unclear. Germline mutations in BAP1 and PBRM1 have been associated with ccRCC, but these mutations are rare and BAP1 germline mutations are primarily associated with other tumor types such as melanoma and mesothelioma (48–51). Here we show that BAP1 and PBRM1 mutations are necessary for ccRCC development and that VHL loss is insufficient.We previously showed that combined targeting of Bap1 and Vhl in the mouse kidney using a Six2 driver results in ccRCC development indicating that BAP1 and VHL cooperate in tumor development (24). However, even in the conditional setting, Bap1 loss caused perinatal lethality, and mice with loss of Vhl and one copy of Bap1 in the kidneys survived only for approximately 8 months (24). In addition, ccRCCs were small. Here, we used PAX8, a more restricted lineage-specific transcription factor than SIX2 (52), and a commonly used RCC clinical diagnostic marker. We show that there is substantial overlap between BAP1 expression and the Pax8 lineage, and that by targeting BAP1 in this compartment, the survival of mice can be extended, allowing for greater time to observe tumor development. Tumors developing in these mice are indistinguishable from those developing in the Six2 lineage, but they are more advanced and larger. These data also support that the narrower Pax8 lineage contains the cell of origin of Bap1-deficient ccRCC.

We previously reported that BAP1 and PBRM1 mutations in ccRCC tend to anticorrelate (13, 16). One potential explanation is that BAP1- and PBRM1-deficient tumors arise from different cell types. Here, we show that targeting Pbrm1 (along with Vhl) to the Pax8 lineage also causes ccRCC. These data show that the Pax8 lineage contains cells capable of transformation by the loss of both Bap1 and Pbrm1. In contrast, similar gene targeting of the Sglt2 and Villin lineages did not lead to ccRCC. However, effective gene targeting with both Sglt2 and Villin drivers in proximal convoluted cells was demonstrated by lineage-tracing experiments, PCR, and IHC analysis showing protein loss and downstream effector pathway activation. These data suggest that gene targeting in proximal tubules is not sufficient to give rise to tumors, challenging the notion that ccRCC arises from this compartment. One caveat is that not all proximal convoluted tubule cells were targeted by Sglt2 and Villin. Our morphologic analyses suggest that ccRCC arises from parietal cells of the Bowman capsule, which are derived from the Pax8 lineage but not the Sglt2 and Villin lineages. Similar observations in kidneys from patients with ccRCC suggest that they are relevant to humans. However, whether it is the same cell or different cells that lead to Bap1- and Pbrm1-deficient tumors remains to be determined.

We observed a spectrum of cystic and solid lesions in the Bap1- and Pbrm1-deficient kidneys that recapitulated the morphology and biological behavior of human ccRCC. Our mouse models reproduce the features of patient tumors, including CD10, vimentin, CAIX, and PAX8 expression. In humans, BAP1-deficient tumors tend to be of high grade and PBRM1-deficient tumors of low grade. In addition, BAP1-deficient tumors are associated with significantly worse patient survival than PBRM1-deficient tumors (17, 30, 53–56). Whether BAP1 and PBRM1 are responsible for these differences, however, was unclear. Here we show that, as in humans, Bap1-deficient ccRCCs in the mouse are different from Pbrm1-deficient tumors, and that they are of high grade. In contrast, Pbrm1-deficient tumors arise later and tend to be homogeneous and of low grade. The grade of Bap1-deficient ccRCC corresponds to Fuhrman nuclear grade 3, versus Fuhrman nuclear grade 1 or 2 for Pbrm1-deficient ccRCC. Overall, our data show that BAP1 and PBRM1 are determinants of tumor grade and likely the drivers of the differences in patient survival.

Interestingly, a minority of Pbrm1-deficient tumors (∼4%) were similar to those observed following Bap1 loss—cystic tumors of high grade. Unlike lower-grade tumors, these tumors displayed strong pS6 staining indicative of mTORC1 activation. This led us to hypothesize that mTORC1 may be implicated in tumor grade. To test the role of mTORC1, we inactivated one allele of the mTORC1 negative regulator TSC1. Targeting Tsc1 in kidneys along with Vhl and Pbrm1 led to the development of solid tumors that were similar to those observed in kidneys with intact TSC1, but of higher grade. These tumors exhibited focal pleomorphism, large nuclei, prominent nucleoli, more abundant cytoplasm with prominent clearing, and tumor necrosis. In contrast, no high-grade solid tumors were observed in Vhl/Pbrm1-deficient kidneys. As expected, strong mTORC1 activation was observed in the higher-grade, but not lower-grade, tumors. Overall, these data suggest that inactivation of one Tsc1 allele promotes the progression of lower-grade PBRM1-deficient solid tumors to high grade. These data are consistent with the notion that mTORC1 is a driver of the transformation of PBRM1-deficient tumors from low to high grade. Interestingly, although somatic mutations in TSC1 are infrequent in human ccRCC, several human ccRCCs with mutations in VHL, PBRM1, and TSC1 have been reported. It is worth noting, in this context, that human BAP1-deficient ccRCCs, which are of high grade, also tend to exhibit higher levels of mTORC1 activity (13). These data suggest that mTORC1 plays a broad role in tumor grade determination. There is a biological basis for this observation, as tumor grade is heavily influenced by nuclear and nucleolar size, which are regulated by mTORC1 (57–59).

While this article was under review, an article was published reporting the generation of mice with targeted disruption of Vhl and Pbrm1 in the mouse kidney (60). There are several important differences between the studies. First, Nargund and colleagues targeted Vhl and Pbrm1 using the traditional Ksp-Cre driver, and the phenotype was dominated by large cysts sometimes taking up half of the kidney (60). In contrast, cysts are infrequently observed in mice with targeted loss of Vhl and Pbrm1 using a PAX8 driver, and the mice survive significantly longer. Second, although the authors were able to detect occasional tumors, they were found in only 30% of the mice, and tumors were uniformly small. In contrast, in our model, tumors were observed with 100% penetrance, were often multiple, and reached sizes as large as one third of the kidney. Third, the small size of the tumors in the study by Nargund precluded studies of tumor progression, which required transplantation into immunocompromised mice (60). Fourth, levels of mTORC1 activation higher than the normal kidney led Nargund and colleagues to conclude that mTORC1 is a dominant feature of PBRM1-deficient tumors. In contrast, our extensive analysis of tumors arising from loss of BAP1 and PBRM1 and through TSC1 targeting led us to the conclusion that mTORC1 activity characterizes higher-grade tumors. This is also consistent with the role of mTORC1 in the regulation of cell and nuclear size. In addition, by targeting both BAP1 and PBRM1 in the same compartment, we could draw conclusions about their role in driving tumor grade, which are otherwise not possible. Finally, our careful histologic studies point to the parietal cell of the Bowman capsule as a cell of origin for ccRCC. Overall, our data implicate BAP1 and PBRM1 loss as drivers not only of ccRCC, but also of tumor grade, and establish mTORC1 as a nuclear grade rheostat.

Kidneys were removed and fixed in 10% (v/v) neutral-buffered formalin for ∼24 hours, and then routinely processed, embedded in paraffin, sectioned at 5 μm, and stained with hematoxylin and eosin (H&E). Stained sections were evaluated by a board-certified pathologist (P. Kapur). H&E sections of kidneys from mice that were found dead and showed significant autolysis precluding accurate morphologic assessment were not included in the morphologic analysis. The number of tumors was counted on a single representative H&E slide from each mouse kidney, which might underestimate the numbers owing to unidentified small tumors.

Blood was collected in heparinized tubes before mice were sacrificed and plasma was obtained by spinning down samples at 3,000 rpm for 10 minutes within 30 minutes of collection. Plasma samples were kept in a −20°C freezer until submission to the mouse metabolic phenotyping core for measurement of BUN and creatinine levels by VITROS MicroSlideTM technology (University of Texas Southwestern Medical Center).

For immunofluorescence, kidneys were fixed in 4% (w/v) paraformaldehyde, embedded in OCT compound, kept in a −80°C freezer, and sectioned on a cryostat. Frozen sections were washed with PBS, treated with 0.2% (vol/vol) Triton X-100, blocked with 5% (wt/vol) BSA/PBS for 1 hour at room temperature, incubated with a primary antibody to anti–β-galactosidase (ab9361; Abcam), 1:1,000 dilution, at 4°C overnight, and detected by secondary antibodies with Cy2 (Jackson ImmunoResearch Laboratories). Sections counterstained with LTL were incubated with biotinylated Lotus tetragonolobus lectin (FL-1321; Vector Laboratories) at room temperature for 30 minutes before washing. Nuclei were stained by DAPI (Thermo Fisher Scientific Inc.). For immunohistochemistry, paraffin-embedded tissue was sectioned at 4 μm, and staining was performed as described (13, 24). Immunohistochemistry for BAP1, PBRM1, CD10, CD31, pS6 ribosomal protein (Ser240/244), Ki-67, CAIX, and PAX8 were performed using an AutostainerLink 48 (Dako, Agilent Technologies). Briefly, kidneys were formalin-fixed, paraffin-embedded, and tissue was cut in 3- to 4-μm sections and air-dried overnight. The sections were deparaffinized, rehydrated, and subjected to heat-induced epitope retrieval using low pH target retrieval solution for 15 minutes (for CD31, pS6, and Ki-67 IHC) and high pH for target retrieval solution (for CD10, BAP1, PBRM1, CAIX, PAX8, and vimentin; Envision FLEX Target Retrieval Solution, High and Low pH; Dako). Sections were incubated with primary antibodies from Cell Signaling Technology: pS6 ribosomal protein–pS6 (Ser240/244), 1:100 dilution, #5364; Ki-67, 1:100 dilution, #12202; vimentin, 1:100 dilution, #5741; Thermo Fisher Scientific: CAIX, 1:400 dilution, PA1-16592; Proteintech: PAX8, 1:200 dilution, 10336-1-AP; Bethyl Laboratories: PBRM1, 1:2,000 dilution, A301-591A; Dako, Agilent Technologies: CD10, 1:40 dilution, M7308; CD31, 1:50 dilution, M0823. BAP1 antibody was provided by Genentech at 1:1,000 dilution. For signal detection, the Envision FLEX System (Dako, Agilent Technologies) was used according to the manufacturer's protocols for all antibodies except for Ki-67, for which the avidin/biotin detection system (Vector Laboratories) was used. Slides were developed using 3,3′-diaminobenzidine chromogen and counterstained with hematoxylin. Appropriate positive and negative controls were used for each run of immunostaining.

Representative kidneys were embedded in OCT media, and 5-μm-thick cryosections were cut. Oil Red O stains to evaluate neutral lipids and triglycerides were performed according to the manufacturer's instructions (Poly Scientific R&D Corp.). Periodic acid-Schiff staining without (PAS) and with (PAS-D) diastase were performed on paraffin-embedded sections in accordance with the manufacturer's instructions (Poly Scientific R&D Corp.).

All RCC patient samples were from the UTSW tissue bank. PAX8 IHC staining analysis on patient tissue was performed by double-blind experiment. The study was performed in accordance with the protocol approved by the Institutional Review Board of The University of Texas Southwestern Medical Center. Approval numbers are 012011-190 and STU-22013-052.

Bap1^{tm2a(EUCOMM)hmgu}, Pbrm1^{tm1a(EUCOMM)Wtsi} ES cells were purchased from the European Conditional Mouse Mutagenesis Program and injected into the cavity of day 3.5 blastocysts from C57Bl/6N mice at The University of Texas Southwestern Medical Center Transgenic Core to evaluate Bap1 and Pbrm1 expression in the kidney. Male chimeras were mated with C57BL/6NTac female mice to generate germline transmission. Mice heterozygous were used for β-galactosidase staining to determine Bap1/Pbrm1 expression. Pbrm1-LacZ/+ mice were mated with FLP mice (The Jackson Laboratory) to introduce FRT recombinase for excision of the LacZ element to generate floxed Pbrm1 mice. Floxed Vhl^F/F mice were kindly provided by Dr. Volker H. Haase (Vanderbilt University Medical Center, Nashville, TN; ref. 19). Bap1^F/F mice, which were used herein for tumor modeling, were kindly provided by Dr. Vishva M. Dixit (Genentech; ref. 61). Sglt2-Cre mice were kindly provided by Dr. Michel Tauc (University of Nice-Sophia Antipolis, France; ref. 47). Villin-Cre mice (The Jackson Laboratory; JAX 004586) were kindly provided by Dr. Joshua Mendell (The University of Texas Southwestern Medical Center). Pax8-Cre mice were kindly provided by Dr. Meinrad Busslinger (The Research Institute of Molecular Pathology, Vienna, Austria). Rosa26-CAG-loxP-stop-loxP-tdTomato mice were from The Jackson Laboratory. Mouse protocols were approved by the Institutional Animal Care and Use Committee (APN#2015-100932) at The University of Texas Southwestern Medical Center.

The tumor cell line was generated as described by Tran and colleagues (62). Briefly, tumor tissue was dissected from the kidney in cold PBS, placed in MEM with Earle's salts, 2 mmol/L glutamine, 50 μg/mL of gentamicin (Life Technologies), 1% penicillin/streptomycin, and 1% amphotericin B (Life Technologies), and minced into small pieces. Subsequently, 15 mL of tissue digestion medium [transport medium with 0.1 mg/mL collagenase (Sigma), 0.1 mg/mL hyaluronidase (Sigma), and 20 μg/mL DNase I (Sigma; D5025)] was added. Dissociated cells were plated in primary RCC cell medium (MEM with Earle's salts, 2 mmol/L glutamine, 10 ng/mL EGF (Life Technologies), 1% MEM nonessential amino acid (CellGro), 0.4 μg/mL hydrocortisone (Sigma), 1% penicillin/streptomycin, and 10% FBS), and cells were maintained in a 37°C incubator with 5% CO₂. Cell medium was changed every 3 days and cells were split 1:2–3 until confluent. Cells were not tested for Mycoplasma.

HEK293T cells [ATCC, negative for Mycoplasma on August 17, 2016, using the MycoAlert Mycoplasma Detection Kit (Lonza, LT07-118)] were grown in DMEM with 1% penicillin/streptomycin and 10% FBS, and maintained at 37°C with 5% CO₂. Plasmids for mouse HIF1α and HIF2α (Addgene #44028 and #44027, respectively; laboratory plasmid database #966 and #967, respectively) were transfected using Lipofectamine 2000 (Thermo Fisher Scientific) into HEK293T cells. After 48 hours, transfected cells were harvested as positive controls for expression of mouse HIF1α and HIF2α.

Genotyping primer sequences available upon request.

For qRT-PCR, frozen tissue was homogenized, and RNA was extracted as previously described (63). cDNA was synthesized from 1 to 4 μg of total RNA (RNA was isolated from 3 independent mice) using random primers (Invitrogen) and Moloney MLV reverse transcriptase (Invitrogen). PCR was performed in triplicate. Primer sequences available upon request.

Expression data for The Cancer Genome Atlas Kidney Renal Clear Cell Carcinoma patients were downloaded and combined with a UTSW cohort, inclusive of previously reported tumors (64). Quantile normalization was carried out on log-transformed gene expression matrix. A two-tailed t test was used to test for differences in gene expression values on the log scale.

Tissues were homogenized in lysis buffer (50 mmol/L Tris–HCl pH 7.4, 250 mmol/L NaCl, 0.5% (vol/vol) Igepal with complete protease and phosphatase inhibitors [protease inhibitors: 0.1-μm aprotonin (USB), 0.02 mmol/L leupeptin (USB), 0.01 mmol/L pepstatin (USB), 0.5 mmol/L PMSF (Sigma); phosphatase inhibitors: 2 mmol/L imidazole (Sigma), 1.15 mmol/L sodium molybdate (Sigma), 1 mmol/L sodium orthovanadate (Sigma), 5 nmol/L microcystin (Calbiochem)]. Western blotting was performed with the following antibodies from Cell Signaling Technology: BAP1 (#13271); Bethyl Laboratories: PBRM1 (A301-591A), HIF1α (A300-286A); Novus Biologicals: CAIX (AF2344), HIF2α (AF2997); Santa Cruz Biotechnology: VHL (sc-1534); Millipore: H2A (07-146), Ubiquityl-Histone H2A (05-678); Sigma: Tubulin (T5168); Thermo Fisher Scientific: HRP-conjugated goat anti-mouse IgG (#31430) and HRP-conjugated goat anti-rabbit IgG (#31460); Novus Biologicals: Donkey anti-Goat IgG Secondary Antibody (HAF109).

All imaging was performed on a 1T Desktop MR scanner (M2 Compact, Aspect Imaging), using a mouse volume coil. The general T1-weighted and T2-weighted imaging were performed with a spin echo (SE; TR/TE = 326/13 ms) and a fast spin echo (FSE; TR/TE = 2500/80 ms) sequence, respectively (prone position).

J. Brugarolas is a consultant/advisory board member for Bethyl Laboratories. No potential conflicts of interest were disclosed by the other authors.

Conception and design: Y.-F. Gu, J. Brugarolas

Development of methodology: Y.-F. Gu, M. Busslinger, J. Brugarolas

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y.-F. Gu, S. Cohn, N. Wolff, Q.N. Do, A.J. Madhuranthakam, A. Dey, R.E. Hammer, P. Kapur

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y.-F. Gu, A. Christie, Q.N. Do, A.J. Madhuranthakam, I. Pedrosa, T. Wang, X.-J. Xie, R.M. McKay, P. Kapur, J. Brugarolas

Writing, review, and/or revision of the manuscript: Y.-F. Gu, A. Christie, N. Wolff, Q.N. Do, I. Pedrosa, T. Wang, A. Dey, R.M. McKay, P. Kapur, J. Brugarolas

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y.-F. Gu, S. Cohn, T. McKenzie

Study supervision: P. Kapur, J. Brugarolas

Other (performed IHC stains): T. McKenzie

We wish to thank the patients who generously provided tissues and participated in our studies. We thank Dr. Volker Haase for the floxed Vhl^F/F mice, Dr. Vishva Dixit for the floxed Bap1^F/F mice, Dr. Michel Tauc for the Sglt2-Cre mice, Dr. Meinrad Busslinger for the Pax8-Cre mice, and Dr. Joshua Mendell for the Villin-Cre mice. We thank members of the Brugarolas Lab for assistance and helpful discussions.

This work was supported by Cancer Prevention and Research Institute of Texas grant RP130603 (J. Brugarolas), and NIH grants R01CA175754 (J. Brugarolas), R01CA154475 (I. Pedrosa), and P50CA196516 (J. Brugarolas, P. Kapur, I. Pedrosa, and T. Wang). The authors would like to acknowledge the assistance of the UT Southwestern Small Animal Imaging Resource, which is supported in part through an NCI Cancer Center Support Grant, 1P30 CA142543.

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.