Papillary RCC with MYC activation has a worse prognosis

Prior work has shown that a subset of pRCC, primarily Type 2 pRCC, is enriched for gene signatures of MYC activation and that tumours with MYC activation have a worse overall survival²⁴. We confirmed and extended these findings using multiple gene expression datasets, which revealed that pRCC tumours were enriched for MYC activation gene signatures when compared to normal kidney (Fig. 1c and Supplementary Fig. 1B)²⁵. Finally, to determine the impact of MYC activation on prognosis, we classified TCGA KIRP tumours as MYC activated (n=23, defined as tumours with a Z score of the PID_MYC_ACTIV_PATHWAY greater than 1.0 s.d. above the mean) or not MYC-activated (n=153). Consistent with previous work, patients with MYC activated pRCC tumours had a significantly worse overall survival (Fig. 1d, P=1.46e–9)²⁴.

Cell lines from MYC mice are dependent upon MYC expression

To better characterize the phenotype of the kidney tumours from MYC mice we generated two cell lines from separate MYC tumours (MYC-2927 and MYC-2983). After verifying the Dox responsiveness of the Tet-O-Myc transgene in these MYC cells (Fig. 1e), we examined the effect of MYC expression on cell proliferation in vitro. MYC cells grown in the presence of Dox (activated MYC) had increased proliferation at 4 days (Fig. 1f), as well as a visible increase in cell number (Fig. 1g). In addition, while MYC cells without Dox were unable to form colonies in soft agar, MYC cells grown in Dox formed colonies robustly in an anchorage independent manner (Fig. 1h). Finally, we assessed the ability of MYC-2983 cells to form tumours in vivo and their in vivo growth dependence on MYC expression. To this end, 5 × 10⁶ MYC-2983 cells were injected subcutaneously and monitored for growth in mice fed Dox chow. Once tumours reached ~300mm³, mice were either continued on or withdrawn from Dox (Fig. 1i). As expected, tumours of mice that remained on Dox continued to have rapid growth, while tumours in mice withdrawn from Dox remained dormant. Therefore, MYC-induced papillary RCCs are dependent upon MYC expression for in vitro and in vivo growth.

Vhl loss with MYC activation promotes clear cell changes

Focal amplification of 8q24 is found in 15% of ccRCC in the TCGA KIRC data set⁵. To ensure that MYC was located in the minimal common region (MCR) of amplification, we used GISTIC 2.0 analysis to identify a broad statistically significant (q value<0.25) region of amplification on chr8. This MCR contains MYC. A plot of mean gene-level DNA copy number measurement by genomic position (Supplementary Data 1 and Supplementary Fig. 2A) shows that mean MYC copy number is larger than the 70% of the mean gene-level DNA copy values. Therefore, while these data suggest that MYC is an important target of amplification in this region there are still ~100 genes with mean copy number values that are higher. Similarly, focal loss of 9p21 was found in 32% of ccRCC in the TCGA KIRC data set⁵. GISTIC 2.0 analysis demonstrated the presence of only three genes in the minimal common region of copy number loss, CDKN2A, CDKN2B and C9orf53 (Supplementary Fig. 2B). These results suggest that both MYC and CDKN2A may be involved in the development or progression of ccRCC.

To model the interplay of genomic events observed in human RCC, we next examined the phenotypes of kidney specific Vhl inactivation in combination with MYC overexpression or combined MYC overexpression and Ink/Arf deletion. Specifically, MYC mice were crossed to conditional Vhl knock-out mice (Vhl^F/F)²⁶, Cdkn2a (Ink4a/Arf^−/−) germline knock-out mice²⁷, as well as mice expressing tamoxifen-inducible Cre recombinase under control of the Ksp cadherin promoter (KspCre^ERT2) to generate cohorts of KspCre^ERT2; Vhl^F/F control mice (conditional inactivation of Vhl, hereafter called V), KspCre^ERT2; Vhl^F/F; Ksp-rtTA; Tet-O-Myc (conditional inactivation of Vhl and Dox-inducible MYC overexpression, hereafter called VM), and KspCre^ERT2; Vhl^F/F; Ink/Arf^−/−; Ksp-rtTA; Tet-O-Myc (conditional inactivation of Vhl, germline Ink/Arf^−/−, and Dox inducible MYC overexpression, hereafter called VIM) (see Supplementary Table 1).

Cohorts of V, VM, and VIM mice were followed for survival (Fig. 2a). VM mice had a significantly worse survival than V mice (P=0.007) and VIM mice had a significantly worse survival than both V and VM mice (both P<0.001, Fig. 2a). Histologic examination showed that the kidneys of V mice were essentially normal (Fig. 2b). In contrast, kidneys from VM and VIM mice had a high incidence of tumour formation (67% and 100% respectively, Table 1). Kidney tumours of VM mice had either a tubulo-papillary or solid appearance with intervening vessels. The tumour cells were hyperchromatic compared to the normal renal tubular epithelium. Occasional clear cell features were identified (Fig. 2b; Supplementary Fig. 3). Small tumours were generally well demarcated, while larger tumours showed a multilobulated growth pattern with frequent necrosis, haemorrhage and dystrophic calcifications (Supplementary Fig. 4A).

Open in a separate window

Figure 2

MYC activation combined with Vhl and Ink4a/Arf loss results in histopathological changes in the kidney resembling human clear cell renal carcinomas.

(a) Kaplan-Meier survival curve comparing survival rates between V (n=8), VM (n=21) and VIM (n=12) mice. Log rank V versus VM P=0.007, V versus VIM P<0.0001, VM versus VIM P<0.001. (b) Representative gross and H&E images of kidney sections from V (33 weeks), VM (30 weeks) and VIM (19 weeks) mice at the indicated times post Dox treatment. (scale bars, 200 and 100μm). (c) cBioPortal OncoPrint plot showing the distribution of MYC, VHL and CDKN2A copy number alterations in the TCGA KIRC data set. (d) Kaplan–Meier survival curve comparing human kidney renal clear cell carcinomas with V, VM and VIM alterations. Log rank V versus VIM P=5.08e-9, V versus VM P=0.005, VM versus VIM P=0.200. (e) Bar graph showing the correlation of V, VM and VIM tumours from the TCGA KIRC dataset with stage and (f) metastasis. (g) Bar graphs showing percentage of TCGA KIRC tumours with alterations in VHL, CDKN2A, and MYC by TNM Stage. (e–g) *P<0.05; **P<0.01, ***P<0.001. P values obtained from Chi-square test.

Table 1

Comparison of VM and VIM mice.

	VM	VIM
Median survival (weeks after Tam/Dox)	57	29.5
Cysts present	9/9 (100%)	6/6 (100%)
Kidney tumour present	6/9 (67%)	6/6 (100%)
Large kidney tumour (>3mm)	2/9 (22%)	4/6 (67%)
Low grade	3/9 (33%)	1/6 (17%)
High grade	3/9 (33%)	5/6 (83%)
Necrosis present	1/9 (11%)	4/6 (67%)
Bona fide clear cell histology	0/9 (0%)	5/6 (83%)
Liver metastases	0/9 (0%)	2/6 (33%)

Open in a separate window

While tumours similar to those seen in the VM mice were observed in VIM mice, the kidneys of VIM mice also harboured tumours with a solid or tightly packed tubulo-papillary appearance. The cells were small to intermediate in size with increased nuclear to cytoplasmic ratio with nuclei that were round to oval with prominent single nucleoli (Supplementary Fig. 3). Clear cell changes within the larger tumours were also present with areas within some tumours showing what appeared to be histology strikingly similar to human ccRCC (Fig. 2b). These areas were typified by nests of tumour cells with an increased amount of clear to granular cytoplasm, separated by thin vascular channels. Therefore, while Vhl loss combined with MYC activation is sufficient to induce modest clear cell changes, Vhl loss combined with MYC activation and Ink/Arf deletion induces bona fide clear cell RCC (Table 1).

ccRCC with VM and VIM alterations have worse prognosis

We next examined whether human ccRCC tumours with similar genomic characteristics (VHL inactivation; VHL inactivation and MYC activation; and VHL inactivation, CDKN2A loss and MYC activation) have similar outcomes as those seen in our mouse models. Of TCGA KIRC tumours (n=525), 87, 13 and 31% had VHL inactivation, MYC activation and CDKN2A deletion, respectively (Fig. 2c). We further classified these patients as V (n=286, 54%), VM (n=26, 5%) and VIM (n=32, 6.1%) and noted that VM and VIM patients had a significantly decreased survival relative to V patients (P=0.005 and P=5.086 × 10⁻⁹, respectively (Fig. 2d), but that VM and VIM patients had a similar overall survival. In keeping with the notion that VIM tumours are more clinically aggressive, we also noted significant enrichment of higher stage and presence of metastases in patients with VIM tumours relative to patients with V tumours (Fig. 2e,f). When limiting our analysis to stage IV tumours, we did not see significant survival differences by genotype (V, VM and VIM) (Supplementary Fig. 5). Because of the very limited patient numbers (V=28, VM=7 and VIM=11), we examined whether VIM genotype was prognostic when controlled for both TNM stage and Furhman grade. Cox proportional hazards (Cox PH) modelling using the V genotype as the reference demonstrated that the VIM genotype was still prognostic (Cox PH=1.79, P=0.034). Therefore, VIM tumours appear to have worse clinical characteristics and outcome in human ccRCC patients even when controlling for stage and grade.

Finally, we attempted to assess whether VHL loss, CDKN2A loss and MYC activation are dysregulated at tumour initiation or tumour progression, by analysing their relative frequency by TNM stage in the TCGA KIRC data set (Fig. 2g). We found that the rate of VHL inactivation was relatively similar across all stages, while the rate of CDKN2A loss and MYC activation appeared to increase with TNM stage. This data are consistent with the known role of VHL as a gatekeeper tumour suppressor gene and suggests that CDKN2A loss and MYC activation are events involved in the progression of a subset of ccRCC tumours.

VM and VIM cell lines are dependent upon MYC expression

We generated cell lines from VM and VIM tumours to assess their dependency on MYC expression. We first confirmed the Dox inducibility of the MYC transgene in these VM (2849 and 3055) and VIM (3039 and 3131) cells (Fig. 3a), as well as the fact that they had undergone Cre mediated recombination of the Vhl locus (Supplementary Fig. 4B). Furthermore, we set out to confirm that VM-3055 and VIM-3039 cell lines are epithelial in nature. We performed RNAseq and co-clustered mouse 3T3 fibroblasts with VM-3055 and VIM-3039 cell lines using the top 10% of the most differentially expressed genes (Supplementary Fig. 4C). 3T3 fibroblasts show a distinctly different expression pattern from both VM-3055 and VIM-3039 cell lines showing that our VM and VIM cell lines are not fibroblastic in nature. After confirming that the VM and VIM cells were derived from renal epithelium, we assessed their growth in in vitro assays. Similar to MYC cells, the proliferation and anchorage independent growth of VM and VIM cells were dependent on MYC expression (Fig. 3b–d).

Open in a separate window

Figure 3

VM and VIM tumours are dependent on MYC expression.

(a) Immunoblot of whole cell lysates from VM (VM-2849 and VM-3055) and VIM (VIM-3039 and VIM-3131) tumour-derived cell lines shows expression of MYC is Dox dependent. (b) Proliferation of VM and VIM cells is significantly reduced with removal of Dox. Cells grown on Dox or vehicle were analysed in replicates of n=8 each day. (c) Bright field image of VM and VIM cells three days after Dox removal show a decrease in cell number (scale bar, 100μM). (d) Soft agar assays show anchorage independent growth of VM and VIM cells is reduced upon Dox removal. Images are representative of each condition performed in triplicate. (e) Venn diagram showing overlapping and distinct sets of genes altered upon Dox removal in VM-3055 and VIM-3039 cells. (f) Venn diagram showing Dox dependent gene expression in VM-3055 and VIM-3039 activates distinct gene sets. (b,d) ***P<0.0001. P values obtained from student t-test. Data are presented as mean ±s.e.m.

To understand the MYC dependent changes on the transcriptomes of VM-3055 and VIM-3039 cells, we performed RNAseq on these cells grown in tissue culture in the presence or absence of Dox. There were a large number of genes that were differentially regulated (increased or decreased greater than two-fold; t-test FDR<0.05) by MYC activation (Dox) in both VM-3055 (799 up and 2491 down) and VIM-3039 (1017 up and 2159 down) cells (Fig. 3e). While there was overlap among genes that were upregulated by Dox in both VM-3055 (634 of 799) and VIM-3039 cells (634 of 1017) (Fig. 3e), nearly a third of Dox inducible genes in VIM-3039 cells (378 of 1017) were not upregulated in VM-3055 cells. A similar percentage of genes repressed by MYC activation (Dox) in VIM-3039 cells were not repressed in VM-3055 cells either, suggesting that MYC may regulate both an overlapping and a unique spectrum of genes in VM-3055 and VIM-3039 cells.

We next performed gene set analysis. Gene sets that were enriched in VM-3055 cells cultured in Dox relative to those cultured without Dox included a number of gene signatures characteristic of MYC activation including those related to cell cycle progression and ribosome biogenesis (Supplementary Data 2). A parallel analysis in VIM-3039 cells also demonstrated high enrichment in proliferation and ribosome biogenesis gene signatures (Supplementary Data 3), but also included gene signatures related to DNA methylation, as well as RNA binding. Indeed, the top gene signatures significantly enriched in VM-3055 and VIM-3039 cells (an area under the curve (AUC) greater than 0.25) were relatively non-overlapping (Fig. 3f). While these findings are in keeping with the notion that MYC activation induces many unique gene expression changes specific to VM-3055 and VIM-3039 cells given the analysis is on a single-cell line, they will need to be validated in future studies.

GEM tumours correlate with human RCC

Given that MYC tumours resemble human papillary tumours histologically, while VM and VIM tumours resemble ccRCC, we wanted to determine if the similarities would also be apparent at the gene expression level. Pearson correlations on whole transcriptome centroids between MYC (n=4), VM (n=4) and VIM (n=4) mouse tumours and the TCGA KIRC and KIRP data revealed that VM and VIM tumours correlated most highly with TCGA KIRC tumours, while MYC tumours correlated more highly with TCGA KIRP tumours (Fig. 4a). Further verifying the importance of MYC activity in pRCC Type 2, a cross species analysis of mouse MYC tumours with the TCGA KIRP data set demonstrated co-clustering of MYC tumours with human papillary Type 2 tumours (Fig. 4b). Moreover, transcriptome wide Pearson correlations revealed that MYC tumours more closely resemble TCGA KIRP tumours verified to be Type 2 pRCC (Fig. 4c). Collectively, these analyses support the clinical relevance of the MYC, VM and VIM mouse models with human papillary and clear cell renal cancers.

Open in a separate window

Figure 4

Renal carcinoma mouse models reflect the transcriptomic landscape of human renal carcinoma.

(a) Pearson correlation of whole-transcriptome centroids of M/VM/VIM mouse models and papillary renal cell carcinoma (KIRP)/clear cell renal cell carcinoma (KIRC) TCGA tumor samples. (b) Heatmap clustering of M mouse model samples with TCGA KIRP samples identified by subtype. (c) Pearson correlation of whole-transcriptome centroids of the M mouse model with KIRP TCGA Type 1/Type 2 samples.

Ink4a/Arf loss in VM mice promotes metastases

During necropsies of VIM mice, we noted that a number of mice had what appeared to be macroscopic metastases to the liver at a low frequency (Fig. 5a,b, and Table 1). Histologically, the liver metastases were either tubulo-solid or clear cell with cytologic features consistent with the primary renal tumour. Metastases were identified in the subcapsular area and within the parenchyma. Notably, vascular spaces were involved by tumour in some of the liver specimens (Fig. 5a). We did not see any macroscopic metastases to other organs such as the lung.

Open in a separate window

Figure 5

Combinatorial loss of Vhl and Ink4a/Arf with MYC activation promotes metastasis and activation of EMT genes.

(a) A representative H&E stained liver sections from a VIM mouse shows metastasis to the liver parenchyma as well as intravascular mets. Higher power magnification reveals features resembling renal clear cell carcinomas within the metastatic lesions. (Scale bar, 100μm). (b) Bar graph showing percentage of mice with metastases. (c) Heat map showing the expression patterns of EMT associated genes in VM-3055 and VIM-3039 cells. (d) Higher expression of genes involved in invasion and metastasis is observed in VIM-3039 cells compared to VM-3055 cells. (e) Matrigel invasion assays show increased invasion from VIM-3039 cells compared to VM-3055 cells. ****P<0.0001. (d,e) P values obtained from student t-test. Data are presented as mean ±s.e.m.

Given the metastatic phenotype seen in the VIM mice but not in the VM mice, we examined a panel of genes that characterize epithelial to mesenchymal transition (EMT) (Cdh1 (E-cadherin), Cdh2 (N-cadherin), Vim (vimentin)), as well as a panel of transcription factors well known to regulate EMT (Snai1, Snai2, Zeb1, Zeb2, Twist1) in the RNAseq data from our cell lines. VIM-3039 cells demonstrated gene expression changes that were consistent with a state of EMT with decreased expression of E-cadherin (Cdh1) and increased expression of N-cadherin (Cdh2) and Vimentin (Vim) as well as upregulated expression of Snai1, Snai2, Twist1, Zeb1 and Zeb2 (Fig. 5c). In addition, we noted that a number of the most differentially expressed genes (Supplementary Data 4) are implicated in invasion and metastasis through remodelling of the extracellular matrix such as the lysyl oxidase family members Lox, Loxl1, Loxl2 and hyaluronan synthase 2 (Has2) (Fig. 5d)²⁸. Given these differences in gene expression, we examined the invasiveness of our VM-3055 and VIM-3039 cell lines in vitro in matrigel invasion assays. As predicted, VIM-3039 cells had significantly increased matrigel invasion (Fig. 5e). Therefore, Ink4a/Arf inactivation appears to facilitate EMT, invasion and metastases and the metastatic phenotype seen in vivo is recapitulated in vitro.

Reagents for in vivo studies

Conditional activation of cre recombinase was induced by oral administration of tamoxifen (Sigma T5648). Per ml of in vivo oral delivery, tamoxifen was formulated as: 100mg of tamoxifen dissolved in 100μl 100% EtOH, followed by the addition of 1ml of sunflower oil to achieve 100mgml⁻¹. The solution was sonicated in a water bath sonicator located at 4°C for 10-s pulses. The solution was then placed in a water bath at 50°C for 1min until clear. Sonication and water bath steps were repeated until no particulates were noticeable. Animals were given 50μl (5mg per day) for three consecutive days by oral gavage. Tamoxifen oral solution was aliquoted and stored at −20°C and brought to administrable solution by 50°C water bath for no longer than 5min. The Tet-ON conditional system was activated with doxycycline chow (Research Diets, Inc. C11300–2000 with 2,000p.p.m. doxycycline) 7 days post oral tamoxifen administration.

Primary tumour cell line generation and culture conditions

Primary renal GEMM tumours were excised and washed in a solution of Pen-Strep, PBS solution (1:1). In sterile conditions, primary tumours were cut into 2 × 2mm fragments and dissociated in a gentleMACS C-Tube (Miltenyi Biotec) using the gentleMACs Dissociator (program: m_imp Tumor_02) in 5ml of 1XDMEM, 10% FBS, 1% PenStrep. 100μl of collagenase D/dispase II (Roche: 40mgml⁻¹) was added to the tumour fragments and continuously inverted for 30min at 37°C. Fragments were then subjected to a second round of dissociation using the gentleMACS Dissociator (program: m_imp Tumor_03). A total of 5ml of protein extraction buffer (PEB: buffer 0.5% FBS, 2mM EDTA in PBS) was added to the dissociated fragments and resuspended by pipetting. The cell suspension was transferred to a 50ml conical tube through a 40μm nylon mesh sterile cell strainer (Fisher). An additional 20ml of PEB buffer was added to the cell suspension and then centrifuged at 300g for 5min. Supernatant was removed and cell pellet was resuspended in 6ml of ‘conditioned media'⁵⁹ containing 2μgml⁻¹ doxycycline (Sigma) and placed in a 6cm sterile cell culture plate.On reaching confluency, cells were split and cultured in 1XDMEM, 10% FBS, 1%PS containing 2μgml⁻¹ doxycycline. In vitro doxycyline was dissolved in DMSO.

Xenograft

Primary renal GEMM tumour cell lines were implanted subcutaneously into SCID mice flanks at 5 × 10⁶ cells in 200μl of PBS containing 2μgml⁻¹ doxycycline. To maintain expression of c-MYC, SCID mice harbouring xenografts were placed on doxycycline chow and withdrawn from doxycycline at the indicated times.

Immunoblotting conditions

Cells were lysed in RIPA buffer complemented with Set I and Set II phosphatase inhibitors at 1 × (Calbiochem), and protease inhibitors at 1 × (Roche). Whole cell lysate concentration was determined with Bio-Rad Protein Assay Dye Reagent Concentrate (Bio-Rad). Proteins were resolved on SDS–PAGE gels and electrotransferred to nitrocellulose membranes, 0.2μm (Bio-Rad). For detection of c-MYC protein, proteins were electrotransferred onto PVDF membranes (Millipore). Primary antibodies, c-Myc (1:1,000; SC-42), VHL (1:500; SC-5575), Ku80 (1:1,000; Cell Signaling #2,180), Vinculin-HRP (1:1,000; Cell Signaling #18799); SC: Santa Cruz Inc. (Supplementary Fig. 6).

Cell viability assay

Cell viability in the context of the various culture conditions was measured by CellTiter-Glo Luminescent Cell Viability Assay (Promega) per manufacture's protocol. Cells were counted and in a 96 well opaque side/clear bottom cell culture plates (Corning). Luminescence measurements were captured using a Biotek Synergy 2 plate reader. Statistical significance was measured by student t-test.

Soft agar assay

Anchorage independent growth was assayed following a standard soft agar assay protocol (bottom layer, 0.6% Difco Noble Agar (BD Biosciences) and top layer 0.4% Noble agar (Difco)). 200μl of control media or media containing 2μgml⁻¹ doxycycline was added to hydrate top agar every 72h. A total of 5,000 cells were plated in each well of a tissue culture treated 6 well plate (Corning). Colonies were stained with crystal violet solution for 4h and destained with H₂O. Pictures were taken using a digital camera (Cannon). Treatment conditions were performed in triplicate. Statistical significance was measured by t-test.

Mouse RNAseq

An equal number of cells (2 × 10⁵ cells per dish) were plated with or without Dox and collected for RNA 24h later. RNA was isolated and purified for RNAseq analysis using the RNeasy Mini Kit (Qiagen) following manufacturer's protocol and eluted in water. A total of 200–1,000ng of total RNA was used to prepare libraries with the TruSeq Stranded mRNA Sample Prep Kit (Illumina). Around 75b paired-end reads were sequenced on a NextSeq 500 Desktop Sequencer using a high output flow cell kit (Illumina), yielding an average of over 28M reads per sample. QC-passed reads were aligned to the mouse reference genome (mm9) using MapSplice⁶⁰. The alignment profile was determined by Picard Tools v1.64 (http://broadinstitute.github.io/picard/). Aligned reads were sorted and indexed using SAMtools and translated to transcriptome coordinates then filtered for indels, large inserts and zero mapping quality using UBU v1.0 (https://github.com/mozack/ubu). Transcript abundance estimates for each sample were performed using RSEM, an expectation-maximization algorithm (Li and Dewey, 2011) using the UCSC knownGene transcript and gene definitions⁶¹. Raw RSEM read counts for all RNAseq samples were normalized to the overall upper quartile⁶².

Gene set analysis was performed between two groups of samples by first ranking all genes by t-statistic between groups. A K–S test was performed to determine if the genes in each gene set from MSigDB were uniformly distributed among the ranked list. K–S test P values were corrected via B–H procedure with an overall FDR of 5%. ROC curves were then generated for each gene set using the ranked list of t-statistics, and the AUC for the top 10% of ranked genes was used to rank gene sets among those passing the K–S test.

Analysis of human renal gene expression data from TCGA

GSE11151 data set was downloaded on Jan 27, 2015 from Gene Expression Omnibus (GEO) website. RNA Expression dataset was log2-transformed/median centred across pRCC and normal tissue samples. RNA expression dataset was analysed with the Gene Set Enrichment Analysis (GSEA) desktop platform.

TCGA KIRP dataset was downloaded on Feb 1, 2015 from Broad TCGA website. RNA Expression dataset was log2-transformed/median centred. RNA expression data set was analysed with the GSEA desktop platform.

TCGA KIRC data sets were downloaded on Feb 15, 2015 from Broad TCGA website. RNA Expression dataset was log2-transformed/median centred. RNA expression dataset was analysed with the cBioPortal Oncoprint function and on R platforms.

Analysis of human copy number data from TCGA

Gene-level segmented DNA copy values for the TCGA KIRC cohort (n=528) were downloaded from the Broad Institute GDAC (https://gdac.broadinstitute.org/). The GISTIC2 analysis (Mermel, 2012) identified a statistically significant (q<0.25) region of amplification in chr8 (69506698-146364022). Mean gene-level DNA copy number measurements were computed and plotted by genomic position.

Correlation analyses

The TCGA clear cell renal cell carcinoma (KIRC) and papillary renal cell carcinoma (KIRP) RNASeq expression datasets were downloaded through the Broad Institure pipeline (Broad Institute TCGA Genome Data Analysis Center (2016): Aggregate Analysis Features. Broad Institute of MIT and Harvard). Human homologues were identified for mouse genes from the Jackson Laboratory Mouse Genome Informatics database (http://www.informatics.jax.org/homology.shtml). When combining datasets, batch effects were adjusted for using the SVA (surrogate variable analysis) R package version 3.12.0. Centroids were derived by taking the median expression of each gene across samples in a designated sample cohort. Centroid similarity metrics were derived by calculating the Pearson correlation between centroids. Heatmap clustering was done using centred average linkage clustering on all expressed genes in the data set.

Matrigel invasion assay

RCC GEMM cell lines were plated in serum-free medium for 24h. A total of 50,000 cells were seeded into an 8-μm matrigel chamber (BD Biosciences) and placed in wells containing medium containing 10% FBS. Cells were allowed to invade for 24h. Cells that had invaded through the matrigel were then stained with the Siemens Staining Kit according to manufacturer's directions. Matrigel wells were then placed on a glass slide and analysed and photographed using an Olympus IX51 microscope. Representative pictures were taken of each matrigel well quadrant, and the number of cells that had invaded into the matrigel was determined by counting the number of stained nuclei using ImageJ.

We acknowledge the members of the Kim lab for useful discussions, the UNC Mouse Phase 1 Unit (MP1U), UNC Animal Studies Core and CGIBD for their technical assistance. This work was supported by the NIH R01-CA202053 (W.Y.K.), AACR Kure It (W.Y.K.), DoD CA120297 (B.K.), NIH T32 5-T32GM007092 (S.T.B.), NIH F31 CA159897 (S.T.B.), NIH F31 CA213985, NIH 2R25GM055336-15 (A.M.S.), UNC Lineberger Comprehensive Cancer Center Developmental Research Award (J.Y.L.), R21-CA194987 (J.Y.L.). W.Y.K. is a Damon Runyon Merck Clinical Investigator. Cancer Prevention and Research Institute of Texas-RP160340 and 1P50CA196516 (T.J.C.), UTSW O'Brien Center-5P30DK079328 (T.J.C).