A Mesenchymal Tumor Cell State Confers Increased Dependency on the BCL-X_L Antiapoptotic Protein in Kidney Cancer

This article is featured in Highlights of This Issue, p. 4591

Renal cell carcinoma (RCC), among the top 10 forms of cancer in humans, is expected to cause approximately 14,000 deaths this year in the United States (1). Although early-stage RCC, which is observed in nearly 60% of patients, can be effectively managed with either surgery or surveillance, advanced or metastatic RCC is associated with poorer clinical outcomes. Loss of the von Hippel-Lindau tumor suppressor protein (pVHL) is the hallmark of clear-cell RCCs (ccRCC), which represent approximately 75% of RCCs (2). Therefore, mechanistic studies defining the consequences of pVHL loss have informed the preclinical discovery of actionable dependencies in ccRCC.

pVHL functions in an E3 ligase complex that is, perhaps, best known to mediate oxygen-dependent destruction of the α-subunit of the hypoxia-inducible factor-α (HIFα; refs. 3, 4). Consequently, HIFα accumulates in pVHL-deficient ccRCCs. Among the two transactivation-competent isoforms of HIFα, HIF1α and HIF2α, the chronic activation of HIF2α, in particular, drives ccRCC tumorigenesis (5, 6). Therapeutic strategies that inactivate HIF2α have been developed in the last decade (7, 8). These approaches have shown promise in both preclinical (9, 10) and clinical settings (11). Besides HIF2α, targeting receptor tyrosine kinases in combination with checkpoint blockade, has recently emerged as a viable option for disease management (12). But not all patients respond, and not all responses are durable. Therefore, new treatments, especially to target tumors that are refractory to current therapeutics, are urgently needed.

Recent large-scale screening efforts, such as the Broad Institute's (Cambridge, MA) short hairpin RNA (shRNA) Achilles (13) and CRISPR/Cas9 DepMap screening projects (14), and Novartis’ DRIVE project (15), have enabled the identification of novel oncogenic dependencies. CRISPR/Cas9-based screens routinely yield phenotypes associated with near complete elimination (knockout) of the gene product. In contrast, shRNA screens often score phenotypes associated with partial mRNA reduction (knockdown), and perhaps better model the incomplete blockade of target proteins that is typically achieved with drugs. Importantly, though, the candidates identified through shRNA-based approaches need to be rigorously vetted because of the notorious “off-target” cytotoxicity associated with shRNA use (16).

Here, we reasoned that the genetic lesions acquired during renal oncogenesis and/or the master regulators of lineage specification could be targetable dependencies in ccRCC. Therefore, using genetic dependency databases, we sought to identify kidney lineage-specific dependencies (versus other cells), hoping some of these could be actionable targets in kidney cancer.

A-498, CAKI-2, 769-P (RRID:CVCL_1050), and 786-O (RRID:CVCL_1051) cells were obtained from ATCC. RCC4 (RRID:CVCL_0498) cells were a kind gift from Dr. Peter Ratcliffe's laboratory (Oxford University; Oxford, UK). UMRC-2, UMRC-6, and UOK101 (RRID:CVCL_B076) cells were obtained from Dr. Bert Zbar and Dr. Martson Linehan (NCI). OSRC2 (RCB0735, RRID:CVCL_1626), TUHR4TKB, and TUHR10TKB (RCB1275, RRID:CVCL_5952) cells were obtained from the Riken cell culture collection. The SU-DHL-6 (KCB 2016029YJ-A, RRID:CVCL_2206) cells were obtained from Dr. Alexandru Almasan and Dr. Neetu Gupta (Cleveland Clinic; Cleveland, OH). Cells were maintained in the following media: A-498, RCC4, UMRC-2, UMRC-6, UOK101, and 786-O in DMEM (Life Technologies, 11995073); OSRC2, SU-DHL-6, TUHR4TKB, and 769-P cells in RPMI-1640 (Life Technologies, 11875119); and, CAKI-2 cells in McCoy's 5A medium (Life Technologies, 16600108). All media was supplemented with 10% FBS and 1X penicillin-streptomycin. Lentivirally transduced cells were selected with Puromycin (2 μg/mL) or Blasticidin (10 μg/mL), as appropriate. All cells were grown at 37°C in 5% CO₂.

To find kidney dependencies, we filtered the DepMap 21Q3 sample information file for kidney cell lines (https://depmap.org/portal/download/). Rhabdoid (G401, JMURTK2) and transitional cell lines (BFTC909) were removed, resulting in 32 total kidney cell lines, 26 of which were included in the DEMETER2-Combined RNAi dependency dataset (17). We sought strong gene dependencies (DEMETER2 < −1, the median of positive controls), that scored in at least three kidney cell lines. We removed “pan-dependent” genes that were scored dependencies in over 90% cancer cell lines (18). The remaining 194 strong kidney dependencies were ordered by Likelihood Ratio Test (LRT) score (refs. 15, 19; likelihood ratio test of skewed-t vs. symmetric gaussian fit of each gene dependency distribution across all pan-cancer cell lines), which meant that strong dependencies on higher-ranking genes were less likely to be from the tail of a null distribution. Finally, BCL2L1 gene effects from the DEMETER2-Combined RNAi dataset were compared with the DepMap 21Q3 (Cancer Dependency Map Portal, RRID:SCR_017655) CRISPR/Cas9 Achilles dataset for independent validation.

The shRNA constructs (Sigma) were as follows: BCL-2 (TRCN0000040069, TRCN0000010303, and TRCN00000293675); BCL-X_L (TRCN0000033499, TRCN0000033500, and TRCN0000033503); and Control (SH001 and SH002). BCL-X_L, amplified from a GFP—BCL–X_L fusion (20), was subcloned into pDONR223 (RRID:Addgene_60532) using Gateway cloning (Invitrogen). Rescue constructs (499R and 499/500R) were generated by sequentially introducing silent mutations using site-directed mutagenesis (Quickchange II XL; Agilent 200521), first at the sh499 site and then, using this construct as a template, the sh500 site. Expression vectors were generated by Gateway cloning into the lentiviral pLX304 destination vector (RRID:Addgene_25890). The pLenti—CMV–Puro constructs to express TWIST, was described previously (21).

Lentiviruses were packaged by transfecting 293FT cells (Thermo Fisher Scientific, R70007) with Lipofectamine 2000 (Invitrogen, 11668019), seeded overnight with 1.8 × 10⁶ cells in 6-cm dishes. Transfection mixes had 1.5 μg lentiviral plasmid, 1.5 μg of helper plasmids [psPAX2 (RRID:Addgene_12260):pMD2.G (RRID:Addgene_12259); 3:1 ratio], and 9 μL Lipofectamine 2000, in OptiMEM. 16 hours later, media was changed, and viral supernatant was collected at 48 and 72 hours posttransfection, and combined. The lentivirus-containing supernatant was filtered through a 0.45-μmol/L filter and frozen at −80°C. For infections, 250 μL of lentivirus was added to cells (75,000/well, seeded overnight into 12-well plates, 8 μg/mL polybrene), and the plates were centrifuged at 300 | $ \times g $ | for 40 minutes. Lentivirus was removed after 8 hours and the cells were grown for 24 hours before selection.

All flow cytometry was analyzed with FlowJo v10.4.2 (RRID:SCR_008520). For Annexin V versus propidium iodide (PI) staining, cells (150,000 cells/well, seeded overnight, 6-well plates) were treated with A-1331852 for 16 hours (100 nmol/L for A-498, UMRC-2, and UMRC-6; 10 nmol/L for CAKI2). Floating and adherent cells were harvested and stained using the APC Annexin V Apoptosis Detection Kit with PI (BioLegend, 640932), as per the manufacturer's instructions.

For CD44 measurements, ccRCC cells [typically grown for 3 days and harvested at <90% confluency using the TrypLE reagent (Gibco)], were washed in 1% BSA/PBS, and stained using an allophycocyanin (APC)-conjugated CD44 antibody (Cell Signaling Technology, catalog no. 80813, RRID:AB_2799964). Typically, 1.5 × 10⁶ cells with 1:100 anti-CD44 in 100 μL volume of 1% BSA/PBS, at room temperature for 45 minutes.

Cells were lysed in lysis buffer [50 mmol/L Tris.Cl (pH 7.5), 400 mmol/L NaCl, 1% Nonidet P-40, 1 mmol/L EDTA, and 10% glycerol], freshly supplemented with a protease inhibitor cocktail (cOmplete Mini, Roche). Proteins were resolved by SDS-PAGE using 8%, 10%, or 12% polyacrylamide gels, as relevant, and transferred onto 0.2-μm nitrocellulose membranes. As a loading control, for blots with multiple ccRCC cell lines, membranes were stained with Ponceau S solution (Sigma, P3504), in room temperature for 45 minutes, and imaged, before immunoblotting.

The following antibodies were used: BCL-X_L (Cell Signaling Technology, 2764, 1:1,000), BCL-2 (Cell Signaling Technology, catalog no. 4,223, RRID:AB_1903909, 1:1,000), p53: (Abcam, catalog no. 5,000–1, RRID:AB_514418, 1:1,000), phospho-p53 (Ser15; Cell Signaling Technology, catalog no. 9,284, RRID:AB_331464, 1:1,000), p21 (Santa Cruz Biotechnology, catalog no. sc-397, RRID:AB_632126, 1:1,000), Vinculin (Cell Signaling Technology, catalog no. 13,901, RRID:AB_2728768, 1:1,000), V5 Tag (Cell Signaling Technology, 80076, 1:1000), and horseradish peroxidase (HRP)-conjugated secondary antibodies (Pierce, 1:5,000). Chemiluminescent HRP substrates (Supersignal West Pico PLUS; Thermo Fisher Scientific; Pierce ECL Plus, Thermo Fisher Scientific; or Immobilon ECL Ultra, Millipore) were used.

ccRCC lines were seeded overnight (CAKI-2 and TUHR4TKB: 4,000 cells/well; others: 2,000 cells/well; in 96-well plates) and then treated with the respective inhibitors. Cells were exposed for 72 hours to a six-point, three-fold serial dilution of ABT-263: 1.1 to 10 μmol/L; A-1331852: 0.02 to 2 μmol/L; and ABT-199: 0.02 to 2 μmol/L. Cell viability was analyzed using the XTT assay (Cell Proliferation Kit II, Roche) or CellTiter-Glo (Promega), per manufacturer's instructions. IC₅₀ values were determined though nonlinear regression (curve fitting) using Graphpad Prism (RRID:SCR_000306). For crystal violet staining, cells were washed with 1X PBS and stained with 0.4% crystal violet in 20% methanol for 30 minutes, washed twice with 1X PBS, air-dried overnight, and scanned.

UMRC-2 and OSRC2 cells were seeded at 2,000 cells per well in 96-well plates and treated with compound combinations. Docetaxel and doxorubicin: five-point, three-fold serial dilution, 2 to approximately 222 nmol/L; 5-Flurororuacil: five-point, three-fold serial dilution, 0.07 to 2 μmol/L; A-1331852: two concentrations, 670 nmol/L and 2 μmol/L. After 72 hours, cell viability (relative to DMSO control) was measured by XTT. Synergy was calculated using the SynergyFinder (RRID:SCR_019318) web application (22) with default parameters.

Total RNA was extracted using Trizol (Life Technologies), quantified using the Qubit RNA Assay Kit (Life Tech), and quality was determined on the Bioanalyzer using the RNA Pico Kit (Agilent). cDNA libraries were prepared using the Collibri Stranded RNA Library Prep Kit for Illumina with 50 to 100 ng of total RNA, following the manufacturer's protocol. Prepared DNA libraries were quantified using the Qubit High Sensitivity DNA Kit (Life Tech) and library sizes determined using the Bioanalyzer High Sensitivity Chip Kit (Agilent). Finally, libraries were subjected to qPCR using the Universal Library Quantification Kit for Illumina (Kapa Biosystems) and a 7900HT Fast qPCR (ABI) instrument. Libraries passing all QC criteria were sequenced on the NextSeq 550 Sequencing System (Illumina) at a final concentration of 12 pmol/L on a single-read flowcell with 75 sequencing cycles.

The base-called sequences were obtained using the DRAGEN Bio-IT Platform. Raw sequences (.fastq files) were QC tested using FastQC and had, on average, high-quality metrics (>30 Phred score) and nucleotide distributions. Reads with Phred scores <28 were trimmed using Cutadapt (v1.16). Reads were aligned to the UCSC hg38 build of the human transcriptome using HISAT2 (v2.1.0) and total read counts per gene were measured using featureCounts (v1.6.4). Counts were filtered, normalized (Trimmed Mean of M-values), and differential expression was determined using EdgeR (v3.12).

Gene set enrichment analysis (GSEA) was performed on the EdgeR normalized counts, using software downloaded from http://www.broad.mit.edu/gsea/downloads.jsp. Gene sets with a normalized enrichment scale (NES) > 1.5; FDR < 0.10; and a nominal P < 0.05 were considered significant.

HTSeq counts files for clear cell (TCGA-KIRC) renal cell carcinomas were downloaded from the Genomic Data Commons Data Portal. Counts were filtered and normalized using EdgeR (RRID:SCR_012802). Principal component analysis (PCA) was performed on ccRCC tumors using the differentially expressed genes identified between sensitive (A-498 and CAKI-2) and insensitive (OSRC2 and UMRC-2) RCC cell lines. K-means and t-distributed stochastic neighbor embedding methods were performed using the top principal components, which explained 75% of the variation. Clusters were determined by selecting the optimal within-clusters sum of squares (elbow method) and annotated by their similarity (reciprocal of Euclidean distance) to sensitive or insensitive cell line derived GSEA profiles.

Cells were pretreated for 3 days with all-trans-Retinoic Acid (ATRA; 1 μmol/L) or TGFβ (10 ng/mL), as relevant. After confirmation of cell state changes (using CD44), cells were seeded into 96-well plates in media containing ATRA or TGFβ, as appropriate. The next day, media was aspirated, and cells were cultured for 7 days with A-1331852 (∼14 nmol/L to 10 μmol/L). Viability was evaluated using CellTiter-Glo (Promega). All changes were normalized to the untreated (DMSO) control in each experimental arm to normalize for any inadvertent differences in cell doubling.

Subcutaneous tumors were measured weekly using calipers and volumes were calculated as (L × W²)/2. A-1331852 was synthesized in bulk and purchased from Chem-Space. Oral formulations were prepared, as previously described (23), and animals were dosed twice a day, 25 mg/kg, for up to 4 weeks. Tumor and normal tissues were harvested upon necropsy for histologic analysis. All animal experiments were approved by Cleveland Clinic's Institutional Animal Care and Use Committee (IACUC; protocol no. 0002168).

All the data necessary to evaluate the conclusions of the manuscript are provided in the paper and/or the Supplementary Materials. Gene expression data is deposited into the Gene Expression Omnibus (GEO; RRID:SCR_005012, GSE173618). All reagents are available commercially and published constructs will be deposited in Addgene to ensure public distribution.

To identify dependencies that are selectively enriched in kidney cancer cells (versus all other lineages), we used the Broad Institute/Novartis’ (shRNA) combined dataset. This dataset reports changes in shRNA abundance over 6 to 8 weeks following lentiviral transduction of a pooled genome-wide shRNA library into a large collection of cancer cells, representing various cancer types and lineages (13). To avoid losing critical dependencies due to inherent genetic heterogeneity between cells (i.e., false negatives), we sorted dependencies that showed strong selectivity (LRT > 100) in at least a subset of kidney lineage cells. We excluded “common essential genes”, which were known dependencies in more than 90% of all cancer cells (18). Finally, we confirmed that the scored dependency was observed in a minimum of three independent kidney cell lines. These strategies scored approximately twenty potential dependencies (Supplementary Table S1).

HNF1B and PAX8, both of which have an established role in kidney lineage specification (24, 25), scored among our top 10 “hits”, ranked by LRT (Fig. 1A). Many of the scored candidates were not druggable (e.g., ribosomal subunits); however, BCL2L1, scored as the top gene that encoded a druggable product. Comparing RNAi data with the CRISPR/Cas9 dependency maps (14), we confirmed that several ccRCCs (e.g., CAKI-2, TUHR4TKB, and TUHR10TKB) showed strong BCL2L1 dependence in both datasets (Fig. 1B), independently validating our initial observations.

BCL2L1 encodes the antiapoptotic BCL-X_L protein (26), a member of the BCL-2 protein family (27). To confirm our initial observations, we lentivirally transduced representative human ccRCC cell lines (A-498, CAKI-2, UMRC-2, and UMRC-6) with three shRNAs that targeted either BCL2L1 (499, 500, and 503), BCL2 (069, 303, and 675), or a nontargeting control (shCon). Unlike BCL2 loss, which was well tolerated, BCL2L1 knockdown caused evident cytotoxicity in A-498, CAKI-2, and UMRC-6 within 7 days and in UMRC-2 cells approximately 15 days postinfection (Fig. 1C and D).

Two BCL-X_L targeting shRNAs (499 and 500) also partially downregulated BCL-2. To formally establish that the cytotoxic effects of these BCL2L1 shRNAs were “on-target”, we engineered a BCL-X_L cDNA that harbored silent mutations in the sh499 recognition sequence. Expression of this construct (499R) rescued the cytotoxicity associated with sh499, but not sh500 (Fig. 1E–G). Introducing additional mutations into the 499R backbone to also make it resistant to sh500 (499/500R) rescued the cytotoxicity associated with both sh499 and sh500 (Fig. 1E–G). Together, these studies confirmed that a subset of ccRCCs are highly dependent on BCL-X_L, but less so on BCL-2. Moreover, certain ccRCCs (e.g., UMRC-2) presented with delayed cytotoxicity in response to BCL-X_L loss.

Pharmacologic strategies used to target the BCL-2 family proteins have identified the BCL-X_L/BCL-2 dual inhibitors, ABT-737 (28) and ABT-263 (Navitoclax; ref. 29); the BCL-2-specific inhibitor, ABT-199 (Venetoclax; ref. 30); and more recently the BCL-X_L-specific inhibitor, A-1331852 (31). A-1331852 exhibits a 600-fold specificity for BCL-X_L (versus BCL-2) and readily kills the BCL-X_L–dependent MOLT-4 cells (EC₅₀: 6.3 nmol/L), but not the BCL-2-dependent RS4 cells (EC₅₀: >5000 nmol/L; ref. 31).

To validate our genetic studies, we treated a panel of human ccRCC lines with BCL-2 family inhibitors. We observed that cell lines that were exquisitely sensitive to BCL-X_L loss in genetic studies (e.g., CAKI-2 and TUHR4TKB) were also highly sensitive to acute treatment with either ABT-263 (Supplementary Figs. S1A and S1B) or the BCL-X_L-specific inhibitor A-1331852, with cellular IC₅₀ values in the low-nanometer range (Fig. 2A; Supplementary Fig. S1C). In contrast, cell lines like UMRC-2, which showed delayed response to genetic BCL-X_L loss, were likewise less responsive to 3 days of BCL-X_L blockade (Fig. 2A; Supplementary Fig. S1A–S1C). These analyses also identified the A-498 and SLR23 cells as additional examples of A-1331852 sensitive ccRCCs.

In contrast to the BCL-2–dependent B-cell lymphoma line SU-DHL-6, the ccRCC lines were virtually resistant to the BCL-2 inhibitor ABT-199 (Fig. 2B; Supplementary Fig. S1D). Moreover, as expected from A-1331852's BCL-X_L specificity (31), SU-DHL-6 cells showed virtually no response to A-1331852 at concentrations that were potently cytotoxic for the ccRCCs (Supplementary Fig. S1E).

Relying on structural models (Supplementary Fig. S1F), we noted three critical residues - F97, Y101, and Y195–in the P2 pocket in BCL-X_L. Mutations in all three residues (e.g., F97W, Y101H, and Y195F) retained functionality, as evidenced by their ability to rescue the cytotoxicity associated with the BCL-X_L 499 shRNA (Supplementary Figs. S1G and S1H); however, F97W (and to a lesser extent, Y195F) protected against the cytotoxic effects of A-1331852 (Supplementary Figs. S1I and S1J). Altogether, we concluded that the effects of A-1331852 were largely driven by BCL-X_L blockade and that nearly 40% of the tested ccRCC lines were sensitive (IC₅₀ in the nanometer range) to A-1331852. Perhaps, because of underlying genetic differences, sensitivity in only a subset of cell lines is a common occurrence in preclinical studies. For example, the HIF2α inhibitor, which is now FDA approved for use against ccRCC, efficiently targeted only approximately 30% cell lines in preclinical studies (9, 10).

We next confirmed that the cytotoxicity triggered by A-1331852 treatment occurred because of increased apoptosis. Upon A-1331852 treatment, using Annexin V (AnnV) versus PI staining, we found a significant increase in the apoptotic population in the BCL-X_L–inhibitor sensitive CAKI-2 and A-498 cells, but not the relatively insensitive UMRC-2 and UMRC-6 cells (Fig. 2C–F).

Antiapoptotic BCL-2 family proteins function by binding and sequestering proapoptotic “BH3-only” proteins (e.g., BIM, BID, and PUMA) or the pore-forming proteins (e.g., BAX and BAK). Synthetic, proapoptotic BH3 peptides can mimic or block these interactions and trigger cellular apoptotic programs, which can be probed using hallmarks of apoptosis (e.g., mitochondrial membrane depolarization; ref. 32). Using “BH3 profiling”, we compared ccRCC's dependency on the BCL-2 family proteins. In line with our pharmacologic experiments, we observed that the BCL-X_L–dependent CAKI-2 cells showed significant mitochondrial depolarization (indicated by loss of JC1 fluorescence) with BCL-X_L blockers, such as the Hrk peptide and the BCL-X_L inhibitor, Wehi539, but not with MCL-1 blockers (e.g., MS1 peptide; Fig. 2G). In contrast, the UMRC-2 cells, which were less responsive to acute BCL-X_L inhibition, were nonresponsive to BCL-X_L blockers (Fig. 2H). These results demonstrated that BCL-X_L actively restrains proapoptotic signaling in ccRCCs.

RCCs are known for their resistance to traditional chemotherapeutics (33), which typically function by promoting apoptosis (32). We hypothesized that BCL-X_L function, which drives therapeutic resistance in many cancers, is a physiologic barrier to chemotherapeutic response in ccRCC. To address this hypothesis, we treated the A-1331852–insensitive UMRC-2 and OSRC2 cells with various combinations of A-1331852 and chemotherapeutics such as 5-Fluorouracil (5-FU), docetaxel, and doxorubicin. To measure potential synergy or antagonism, we modeled cell viability using SynergyFinder Plus (22), a visualization package that integrates four drug–drug interaction models (e.g., HSA, Loews, Bliss, and ZIP; ref. 34). This analysis demonstrated synergy at many combinations of the BCL-X_L inhibitor and all three chemotherapy drugs, and using all the models (Supplementary Fig. S2; Supplementary Table S2). Therefore, BCL-X_L inhibition sensitized ccRCCs to several chemotherapeutic agents.

We began our search for potential biomarkers of BCL-X_L dependence in cells by comparing the expression of BCL-X_L and BCL-2. Unfortunately, these studies showed that (i) mRNA levels of BCL2L1 correlated poorly with BCL-X_L DEMETER dependency scores (r² < 0.3, Pearson correlation coefficient; Supplementary Fig. S3A); (ii) neither the protein expression of BCL-X_L or BCL-2 (Supplementary Fig. S3B), nor the relative abundance of BCL-2 and BCL-X_L (Supplementary Fig. S3C–S3E), were sufficient to predict BCL-X_L dependency; and, (iii) there were no discernible differences in the expression of different BCL-X isoforms in ccRCC, with BCL-X_L being the predominant species in both sensitive and insensitive ccRCC lines (Supplementary Fig. S3F).

To identify determinants of BCL-X_L dependency in an unbiased manner, we performed transcriptomics studies. We chose two representative cells that were sensitive (S: A-498 and CAKI-2) and two that were insensitive (I: OSRC-2 and UMRC-2) to acute BCL-X_L blockade. We then transcriptionally profiled, using RNA sequencing (RNA-seq), these cells under (untreated) native conditions and after acute treatment with A-1331852, under experimental conditions that preceded any overt cytotoxicity, which we feared could confound our gene expression studies.

We analyzed the transcriptional signatures using EdgeR and first identified differentially expressed genes (DEG) in the sensitive (S) versus insensitive (I) lines under untreated conditions (Fig. 3A; Supplementary Table S3). We noted that the expression of certain BCL-2–related genes, including BCL2L11 (which encodes BimL), BIK, and BAK1, were elevated in the BCL-X_L–dependent lines (Supplementary Fig. S3G). Moreover, analysis of the DEGs against annotated gene sets described in mSigDB (35), using GSEA (36), showed prominent differences in untreated sensitive cells versus their A-1331852-insensitive counterparts (Fig. 3B and C; Supplementary Table. S4), including differences in apoptosis, p53 response, cytokine response (e.g., IFN response), and cell state [e.g., epithelial–mesenchymal transition (EMT)] pathways (Fig. 3B–D; Supplementary Table S5).

A-1331852–sensitive cells mounted a stronger transcriptional response to acute BCL-X_L blockade, as compared with the insensitive cells (Supplementary Fig. S3H). This transcriptional response further amplified the pathway-level differences that already existed between the sensitive versus insensitive cells, including the apoptotic pathway (Fig. 3D; Supplementary Table S5). We found that nine overlapping gene sets, which predominantly included cytokine response pathways, were induced in response to A-1331852 in both sensitive lines (Supplementary Fig. S3I; Supplementary Table S6). A-1331852 treatment also triggered transcriptional induction of many shared genes between the two sensitive lines (Supplementary Fig. S3J; Supplementary Table S7), including EGR family genes and the Notch-target HEY1. The combinatorial effects of these changes likely drive A-1331852 response.

To identify predictors of BCL-X_L dependence, we focused on the biological pathways that were inherently different in untreated sensitive versus insensitive lines. Activation of p53 triggers cell death via multiple proapoptotic proteins (37, 38), including certain BCL-X_L–dependent pathways (39). We therefore addressed the role of intact p53 in A-1331852 response. We first evaluated p53 mutation status in our panel of ccRCC lines, as described in the Broad Institute's Cancer Cell Line Encyclopedia (CCLE). This analysis showed that wild-type p53 was present in 3 of the 4 BCL-X_L–dependent lines (e.g., CAKI-2, A498, and TUHR4TKB), but also in two of the BCL-X_L independent cell lines (e.g., OSRC2 and UOK101; Supplementary Fig. S4A). We then treated ccRCC lines with doxorubicin and measured the protein levels of total p53, phosphorylated p53, and the p53-target gene CDKN1A, which encodes the cyclin-dependent kinase inhibitor p21. We found that (with the exception of UOK101 cells) the basal levels of p21, and its induction in response to doxorubicin, was lower in many of the intermediate and insensitive ccRCC lines compared with the sensitive cells (Supplementary Fig. S4B). Finally, using isogenic ccRCC cells, where we first inactivated endogenous p53 using CRISPR/Cas9, and then restored expression of single-guide RNA (sgRNA)-resistant versions of either wild-type or mutant p53, we found that changes in p53 activity had no discernible impact on A-1331852 response (Supplementary Fig. S4C–S4E). Altogether, we concluded that intact p53 response was largely insufficient to predict BCL-X_L dependency in ccRCC lines.

We then focused our attention on EMT as a determinant of BCL-X_L dependency. One previous study had suggested this possibility, albeit in cancer cells of other lineages (40). Consistent with our RNA-seq data, using flow cytometry, we noted higher expression levels of the CD44 mesenchymal marker in ccRCC cells that were sensitive to acute BCL-X_L inhibition (Fig. 3E).

Next, we addressed the contribution of EMT in driving apoptosis upon A-1331852 treatment. We began by treating the A-1331852–insensitive, UMRC-2 and OSRC-2, cells for 3 days with TGFβ, a potent inducer of mesenchymal transition (Fig. 4A). Unfortunately, we did not find any changes in CD44 expression in OSRC2 cells and thus substituted these cells with the intermediate-sensitivity UMRC-6 cells. In both UMRC-2 and UMRC-6 cells, TGFβ treatment led to induction of mesenchymal features (e.g., CD44, vimentin, and Zeb1; Fig. 4B and C) and concurrently increased A-1331852 sensitivity (Fig. 4D). We then lentivirally expressed the EMT master transcription factor, TWIST, in UMRC-2 and UMRC-6 cells. Here again, we observed that inducing mesenchymal features (e.g., TWIST and SNAIL; Fig. 4E) increased sensitivity to A-1331852 (Fig. 4F). Importantly, these results were BCL-X_L–specific because mesenchymal transition failed to sensitize ccRCCs to pharmacologic BCL-2 inhibition (Fig. 4G).

Finally, we addressed mesenchymal necessity (Fig. 4H). Treating the sensitive lines, A-498 and CAKI-2, with ATRA for 3 days led to a small, but measurable, decrease in mesenchymal features (e.g., CD44 and Zeb1 decreased; whereas, E-cadherin increased; Fig. 4I and J). Interestingly, this minimal change was sufficient to modestly reduce the sensitivity of both A-498 and CAKI-2 cells to A-1331852 treatment (Fig. 4K). Altogether, EMT was both necessary and sufficient to confer BCL-X_L dependence in ccRCCs.

An examination of the ccRCC cells indicated that recurrent genetic hallmarks of ccRCC, including loss of VHL, PBRM1, BAP1, SETD2, etc (2), were insufficient to predict BCL-X_L dependence (Supplementary Fig. S4F). Next, because pVHL expression can impact EMT in certain cell lines (41, 42), we addressed if pVHL status influences BCL-X_L dependency. Using isogenic pVHL-proficient or -deficient versions of CAKI-2 and A-498 cells (Supplementary Fig. S4G), we found that, surprisingly, reintroduction of pVHL failed to alter their mesenchymal cell state (Supplementary Fig. S4H) and did not impact their response to pharmacologic BCL-X_L inhibition (Supplementary Figs. S4I and S4J).

ccRCCs vary in their HIFα expression patterns, either expressing HIF2α alone (H2) or both HIF1α and HIF2α (H1H2; refs. 43, 44). HIFα status can impact certain biological outcomes, such as response to HIF2α blockade (10). However, we found that A-1331852–sensitive ccRCC cells included both H2 (e.g., A-498) and H1H2 (e.g., CAKI-2) cell lines. Conversely, cell lines that were insensitive to BCL-X_L loss also included H2 (e.g., UOK101) and H1H2 (e.g., UMRC-2, RCC4, etc) cells. Together, these results indicated that (1) the impact of pVHL on EMT was likely dependent on biological context; and (2), neither pVHL nor HIFα status was sufficient to predict BCL-X_L dependence.

Transcriptional differences identified determinants of BCL-X_L dependence in ccRCC cells; therefore, we used DEGs (identified in Fig. 3) to probe for the prevalence of these signatures in human RCC tumor gene expression data from The Cancer Genome Atlas (TCGA; Fig. 5A). PCA segregated RCC tumors into three clusters, each representing one major RCC disease subtype (Fig. 5B; Supplementary Fig. S5). Overlaying the DEGs predicting BCL-X_L dependence onto human ccRCC tumors allowed us to establish that (i) the BCL-X_L dependency signature (red) was observed in approximately 30% of human ccRCCs (Fig. 5C); and, (ii) the EMT signature was among the most differentially expressed gene signatures among these cohorts. Differences in the p53 pathway, albeit statistically significant, showed more subtle differences (Fig. 5D). Finally, as expected from the typically aggressive nature of mesenchymal tumors, the BCL-X_L dependency signature was associated with worse clinical outcomes (Fig. 5E). These studies suggested that mRNA gene expression could be exploited as a biomarker to identify patient cohorts that exhibit features of BCL-X_L dependence.

The BCL-X_L–selective inhibitor, A-1331852, is orally bioavailable in subcutaneous tumor xenografts in NCR^nu/nu mice (23), and could be used to interrogate the therapeutic relevance of BCL-X_L inhibition in ccRCC. Unfortunately, the BCL-X_L-dependent CAKI-2 and A-498 cells, have poor engraftment rates as subcutaneous tumors in NCR^nu/nu mice. The UMRC-2 cells, however, engraft tumors readily. Interestingly, we found that, as seen with our genetic studies, sustained pharmacologic BCL-X_L inhibition (>2 weeks) led to nanometer IC₅₀ in UMRC-2 cells (Supplementary Fig. S6), justifying the use of UMRC-2 cells for in vivo studies, which typically rely on dosing regimens extending to 3 to 4 weeks.

We engrafted UMRC-2 cells into subcutaneous tumors in 7-week–old NCR^nu/nu mice (equal number of males and females). Tumor-bearing animals (>100 mm³) were randomized to receive either 25 mg/kg A-1331852 or sham-vehicle control, twice a day, orally, for up to 4 weeks. Compared with the indistinguishable tumor volumes at the point of enrollment (Fig. 6A), we noted smaller tumor volumes and slower tripling rates in A-1331852–treated mice (Fig. 6B–D). Finally, histologic analysis of the harvested tumors showed zones of dying cells, marked by elevated cleaved caspase-3 staining, consistent with apoptotic cell death, in A-1331852–treated tumors (Fig. 6E and F).

Despite no notable behavioral or body weight changes (Supplementary Fig. S7A), we found profound splenomegaly and a disruption of splenic tissue architecture in A-1331852–treated mice (Fig. 6D; Supplementary Fig. S7B–S7D), but no changes in other major organs (e.g., kidney, liver, lung, and heart; Supplementary Fig. S8). Altogether, these findings were promising. Even with the technical necessity to employ the relatively insensitive UMRC-2 cells, we demonstrated the overall feasibility of BCL-X_L inhibition as a strategy to block ccRCC tumor growth.

Despite recent therapeutic advances, metastatic kidney cancer is ultimately incurable and new therapies that complement existing treatments are sorely needed. To address this need, instead of “synthetic lethal” dependencies that are associated with a specific oncogenic alteration (e.g., truncal genetic lesion), we interrogated kidney lineage-specific dependencies. Second, within the lineage, we sorted for strong dependencies even when restricted to a subset of cells. This strategy enabled the identification of approximately twenty candidate “hits”, including the antiapoptotic BCL-X_L protein.

We focused on BCL-X_L because it already had well characterized pharmacologic inhibitors available, which could facilitate faster adaptation into the clinic. However, other genes in our list (e.g., ITGAV) also encode druggable proteins and warrant further study. Moreover, some otherwise “undruggable” proteins (e.g., HNF1B and PAX8) could be targeted with drugs that promote their proteolysis, such as deubiquitinase inhibitors and/or proteolysis-targeting chimeras (PROTAC). Notably, PROTACs have also been developed for BCL-X_L, in an attempt to reduce drug resistance and to reprogram BCL-X_L inhibition away from platelets, where it leads to undesirable cytotoxic side effects (45).

We rigorously validated BCL-X_L as a dependency in ccRCC. First, we confirmed that ccRCC lines, such as CAKI-2 and TUHR4TKB, which were exceptionally sensitive to BCL2L1 in the shRNA dataset (13), were also identified as BCL-X_L–dependent in the Broad Institute's CRISPR/Cas9 dependency map (21Q3; ref. 14). Next, using three different shRNAs, we confirmed fitness defects in multiple ccRCC cell lines upon loss of BCL-X_L, but not its closely related sibling BCL-2. Finally, we demonstrated that these effects were “on-target” by rescuing the cytotoxicity associated with two BCL-X_L shRNAs.

To develop our genetic studies, we employed rigorous pharmacologic studies. We found remarkable overlap in the BCL-X_L dependence of cell lines in both of these approaches. We also noted that some cell lines (e.g., UMRC-2) showed cytotoxicity only upon chronic BCL-X_L blockade. BCL-2 loss was well tolerated by ccRCCs. Finally, A-1331852's cytotoxicity could be partially reversed using “drug-resistant” BCL-X_L mutants. These findings demonstrate the specificity of BCL-X_L dependence in ccRCC.

Standard chemotherapeutic agents, including topoisomerase inhibitors, taxanes, and nucleoside analogues, have failed in many independent trials against ccRCC (33). The ability to promote apoptosis is central to the efficacy of many traditional chemotherapeutic agents. Our drug–drug interaction studies show that BCL-X_L function presents a physiologic barrier to chemotherapeutic response in kidney cancer.

The importance of antiapoptotic proteins has been previously interrogated in renal cancer; however, these studies focused primarily on the role of BCL-2 (46–48). Our findings demonstrate that ccRCCs exhibit much greater dependence on BCL-X_L than BCL-2. These findings refocus attention on BCL-X_L blockers as potential therapeutic agents in kidney cancer.

Our findings indicate that approximately 30% of human renal tumors, especially those that represent the more aggressive mesenchymal signatures, are likely to be responsive to BCL-X_L inhibition. We also demonstrate the utility of the ex vivo BH3-profiling assay as a faithful predictor of BCL-X_L dependency in renal cells. Importantly, this assay has been optimized for interrogation of apoptotic dependencies in human cancer samples (32, 49), and can be easily adapted for human renal tumors. Together, these assays provide us future opportunities, which were beyond the scope of this initial study, to develop clinically usable biomarkers to identify cohorts of BCL-X_L–dependent renal tumors.

The clinical use of BCL-X_L blockers has been limited due to adverse events (e.g., thrombocytopenia). However, most of these studies were performed using earlier versions of BCL-X_L blockers, which also had “off-target” effects on other BCL-2 family proteins. Specific BCL-X_L inhibitors, such as A-1331852, may have reduced toxicities in vivo. Thrombocytopenia will continue to be a concern given that platelets are highly dependent on BCL-X_L for survival and quickly undergo apoptosis when this protein is inhibited (50). However, novel approaches that degrade BCL-X_L in nucleated cells, but not platelets, reduce thrombocytopenia in animal models, and are now entering clinical trials (45). These pharmacologic advances, combined with assays that predict BCL-X_L dependence, and assign therapy more rationally in patients most likely to respond, could enable efficacious clinical use of BCL-X_L inhibitors in kidney cancer.

L. Stransky reports equity in Blueprint Medicines. K.A. Sarosiek reports grants from NIH, National Institute of Diabetes and Digestive and Kidney Diseases, NCI, Alex's Lemonade Stand Foundation for Childhood Cancers, Making Headway Foundation, and St. Baldrick's Foundation during the conduct of the study. F. Vazquez reports grants from Novo Ventures, Dependency Map Consortium, and Bristol-Myers Squibb during the conduct of the study. No disclosures were reported by the other authors.

T. Grubb: Conceptualization, data curation, formal analysis, investigation, methodology, writing–original draft. S. Maganti: Data curation, formal analysis, investigation, methodology. J.M. Krill-Burger: Data curation, formal analysis, investigation, methodology. C. Fraser: Investigation, methodology. L. Stransky: Conceptualization. T. Radivoyevitch: Formal analysis, methodology, writing–review and editing. K.A. Sarosiek: Conceptualization, resources, data curation, formal analysis, funding acquisition, investigation, methodology, writing–review and editing. F. Vazquez: Data curation, formal analysis. W.G. Kaelin Jr: Conceptualization, resources, data curation, formal analysis, funding acquisition, investigation, writing–review and editing. A.A. Chakraborty: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, writing–review and editing.

A.A. Chakraborty is supported by seed money from the Cleveland Clinic Foundation, the Case Comprehensive Cancer Center Jump Start Award (RES515351), the DoD's Kidney Cancer Research Program Early Career Investigator award (W81XWH-20–1-0804), the Velosano pilot award, and the V Foundation Scholar Award (V2020–011). W.G. Kaelin Jr. is supported by NIH (grant nos. R35CA210068 and P50CA101942) and is an HHMI Investigator. K.A. Sarosiek is supported by Harvard T.H. Chan School of Public Health (Boston, MA) Dean's Fund for Scientific Advancement, Andrew McDonough B+ Foundation, Making Headway Foundation St. Baldrick's Research Grant, NIH/NCI (grant no. R00CA188679), and NIH/NIDDK (grant no. R01DK125263).

We thank Dr. Ruth Keri (Cleveland Clinic) for the GFP-BCL-X_L expression construct, Dr. Mark Jackson (Case Western Reserve University) for the TWIST expression construct, and Drs. Alexandru Almasan and Neetu Gupta (Cleveland Clinic) for the SU-DHL-6 cell line. We thank Dr. Christopher Weight (Cleveland Clinic) for critiques on the clinical relevance of these findings.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.