Comprehensive Pharmacogenomic Profiling of Malignant Pleural Mesothelioma Identifies a Subgroup Sensitive to FGFR Inhibition

Malignant pleural mesothelioma (MPM) is a tumor arising from the pleural cavity and is strongly associated with occupational exposure to asbestos. Although strict regulation is in place in more than 50 countries, in parts of the world where there is still widespread usage of asbestos, most notably in South America, Russia, and states of the former Soviet Republic, China, and South-East Asia, the incidence of this disease is rising (1, 2). MPM is highly refractory to conventional anticancer therapies, and the prognosis is poor; most patients die within a year of diagnosis. Surgery with curative intent is only possible in a highly selected group of patients and needs to be combined with chemotherapy. The only approved treatment, a combination of the cytotoxic agents cisplatin and pemetrexed, yields at best modest improvements in survival (3, 4). Despite many clinical studies utilizing novel biological therapies, there are as yet no effective targeted therapies for this cancer (5, 6).

A recent comprehensive genomic analysis of 216 MPM samples found BAP1, NF2, TP53, SETD2, and CDKN2A to be recurrently mutated or structurally rearranged (7). The landscape is thus one of mutated tumor suppressor genes and alterations in pathways as diverse as Hippo, mTOR, and TP53, as well as histone methylation. Such loss-of-function oncogenic events are typically considered “undruggable,” but downstream programs of genes, activated as a consequence of such mutations, may themselves be tractable therapeutic targets. This is illustrated by NF2-deficient tumors with activated focal adhesion kinase (FAK). Defactinib, a FAK inhibitor, demonstrated efficacy in NF2-deficient tumors in vitro (8) but a subsequent clinical trial in mesothelioma was halted due to lack of efficacy. Other drugs tested to date that have failed to improve the outcome in MPM include EGFR inhibitors (9), Bcr–Abl inhibitors (10), thalidomide (11), bortezomib (12), and vorinostat (13). In many of these studies, a subgroup of patients appeared to derive some benefit. However, in MPM, it has been difficult to elucidate reproducible biomarkers that identify these sensitive subgroups. Some research groups have demonstrated coactivation of multiple RTK pathways in MPM tumors, which may provide a rationale for combination therapies with kinase inhibitors (14).

We aimed to utilize high-throughput chemical screening platforms alongside molecular characterization of immortalized and early-passage cell line models of MPM to uncover critical signaling pathways that may be amenable to therapeutic interrogation.

Cells are grown and maintained in either RPMI or DMEM F/12 supplemented with 10% FBS and 1% penicillin/streptomycin. Cell lines were maintained at 37°C at 5% CO₂. All cell lines have been verified by genotyping using short tandem repeat (STRs) profiling and Sequenom profiling of a panel of 92 single-nucleotide polymorphisms.

Cells are trypsinized and counted before seeding at the optimal density for the well size (either 96- or 384-well plates were used) and duration of the assay. Seeding density was optimized by titration of the cells such that upon visual inspection of the control wells at the end of the assay, a confluency of 70% to 90% was observed allowing cells to grow in a linear phase. Adherent cell lines were seeded 24 hours before drug addition. The high-throughput chemical inhibitor screen was carried out using 384-well plates, and viability was measured 72 hours after drug addition with a 5-point serial fourfold concentration range of 265 compounds. All other viability assays were carried out using 96-well plates and a 9-point twofold dilution of the drugs. Drugs were all dissolved in DMSO, and DMSO was used only as a control condition. At the end of the experiment, cells were fixed with 4% paraformaldehyde. Following two washes with dH₂O, 100 μL of Syto60 nucleic acid stain (Invitrogen) was added to a final concentration of 1 μmol/L (a 1/5,000 stock dilution), and plates were fixed for 1 hour at room temperature. Quantification of fluorescent signal was achieved using a Paradigm (BD) plate reader using excitation/emission wavelengths of 630/695 nm. Data were analyzed by adjusting for background signals and normalizing each well to the DMSO-treated control.

Compounds were acquired from academic collaborators or commercial vendors. Each compound, its therapeutically relevant target substrate and pathway, and the minimum and maximum screening concentrations are listed in Supplementary Table S1. Compounds were stored as 10 μmol/L aliquots at −80°C and were subjected to a maximum of 5 freeze–thaw cycles. Each of the agents was screened at a 5-point serial fourfold dilution to provide a 256-fold range from the lowest to highest concentration. The concentrations selected for each compound were based on in vitro data to cover the range of concentrations known to inhibit relevant kinase activity and cell viability.

Cells were seeded in a flat-bottom 384 wells plate at optimal cell density. After 24 hours, PD173074 and AZD 4547 in a concentration range between 0.007813 and 1 μmol/L were added using a Tecan HP D300 Digital Dispenser. Five replicate wells were assayed for each condition. Phenylarsine oxide (20 μmol/L) was used as positive control condition. To assess apoptosis, 5 μmol/L of IncuCyte caspase-3/7 green apoptosis assay reagent was added to the cells. Confluence and apoptosis levels were quantified by IncuCyte Zoom live-cell imaging systems from Essen Bioscience. Relative apoptosis was calculated by dividing the confluence of fluorescent apoptotic cells by total confluence and normalized to the positive control condition.

Cell monolayers were lysed on ice in NP40 Cell Lysis Buffer (Invitrogen) containing fresh protease and phosphatase inhibitors (Roche). Lysates were centrifuged at 13,000 rpm for 10 minutes and the supernatant used for analyses. Protein concentration was calculated from a standard curve of BSA using the BCA assay (calbiotech) according to the manufacturer's instructions. Equal protein concentrations were loaded on pre-cast 4% to 12% Bis–Tris SDS-PAGE Gels (Invitrogen), run at 200 V for 1 hour. Proteins were transferred onto a methanol activated PVDF membrane at 100 V for 1 hour or overnight at 30 V. Membranes were blocked in 5% milk for 1 hour before the addition of primary antibody at a concentration recommended. After overnight incubation with the primary antibody at 4°C, the membrane was washed three times in 0.1% TBS-T followed by incubation with the secondary antibody according to the supplier's description at 1/2,500 dilution). Immunoblots were imaged using Pierce Supersignal Plus chemiluminescent kit on a gel imager (Syngene). Antibodies against BAP1, pERK, ERK, pFGFR (total), and pFGFR1 (all from Cell Signaling Technologies) and the polyclonal p-FGFR3 antibody sc-33041 (Santa Cruz Biotechnology) were used. Beta Tubulin was used as a loading control for Western blots. Phospho-RTK arrays (RD Systems) and caspase-Glo 3/7 assay were used according to the manufacturer's instructions.

All patients whose materials were used provided written informed consent for the use and storage of pleural fluid, tumor biopsies, and germline DNA. Diagnosis was made on tumor biopsies according to local IHC protocols and confirmed by the Dutch mesothelioma panel, a national expert panel of certified pathologists that evaluate all suspected mesothelioma patient samples. Early-passage primary mesothelioma cultures were generated from tumor cells isolated from pleural fluid of patients at the Netherlands Cancer Institute. The pleural fluid was centrifuged at 1,500 rpm for 5 minutes at room temperature. Erythrocyte lysis buffer was used to remove erythrocytes if many were present. Cells were resuspended in Dulbecco's Modified Eagle Medium (DMEM, Gibco) supplemented with peniciline/streptomycin and 8% fetal calf serum. The cells were seeded in T75 flasks at a density of 1 × 10⁶ cells/mL and incubated at 37°C at a humidified 5% CO₂ atmosphere. Medium was refreshed depending on cell growth, usually twice a week. At seeding and during the first two passages, cytospins were made and stained with HE and reviewed by our pathologist to determine the percentage of tumor cells. If the tumor percentage was over 70%, usually reached after one passage, living cell cultures were transported to the Wellcome Trust Sanger Institute within 6 hours for drug screening and genetic analysis. Cells were cultured for a maximum period of 4 weeks.

Lipofectamine RNAiMAX (Thermofisher) was used according to product guidelines for transfection with siRNA against FGFR3 (Thermo Fisher Silencer Select s5167 and s5169) or BAP1 (s15822) utilizing the protocol “forward transfection of mammalian cell lines.” KIF11 siRNA (s7902) was used as a transfection (positive) control. Viability or protein expression was assayed as described above, at specified time points. H226 cell expressing a BAP1 stable construct, and BAP1 C91A mutant lines were a kind gift from K Kolluri (UCL, London).

Microarray data were generated on the Human Genome U219 96-Array Plate using the Gene Titan MC instrument (Affymetrix). The robust multi-array analysis (RMA) algorithm (15) was used to establish intensity values for each of 18562 loci (BrainArray v.10). We discarded transcripts with low sample variance and consolidated duplicated genes by averaging their expression values across duplicates. The resulting data were subsequently normalized (μ = 0; σ = 1) sample-wise and gene-median centered. Raw data were deposited in ArrayExpress (accession E-MTAB-3610).

The RMA processed dataset is available at www.cancerrxgene.org/gdsc1000/GDSC1000_WebResources/Home.html.

The expression-level signal of each gene was normalized using a nonparametric kernel estimation of its cumulative density function as described in ref. 16. Additionally, the normalized expression values were further tissue-centered using as grouping factors the cell line tissue labels of ref. 17.

All animal experiments were conducted according to institutional guidelines under protocol approved by the animal ethics committee of the Netherlands Cancer Institute. To establish xenografts, 3 million human mesothelioma cells (H2731 and MSTO211H) were implanted subcutaneously into the right dorsal flank of 6- to 7-week-old female nude SCID mice. Mice were randomized into vehicle and drugs treatment groups, and treatment was initiated once the tumor volumes reached approximately 200 mm³. Tumor size was measured with calipers twice a week, and tumor volume was determined as a × b² × 0.5, where a and b were the large and small diameters, respectively.

A panel of 889 cancer cell lines was screened with 265 compounds that included targeted and cytotoxic compounds (for detail see http://www.cancerrxgene.org/). It was observed that three of 19 MPM lines (H2795, H2591, and MSTO-211H) had IC₅₀ values among the top 5% of cell lines showing highest sensitivity to the compound PD-173074, an FGFR1 and FGFR3 kinase inhibitor (Fig. 1A; ref. 15). These three cell lines, together with two additional MPM lines (NCI-H28, resistant; MPP-89, partially sensitive) and an FGFR-dependent lung cancer cell line harboring amplification of FGFR1 (NCI-H1581), were rescreened with PD-173074 and were as sensitive to PD-173074 as the FGFR1-dependent lung cancer line NCI-1581 (Fig. 1B). Furthermore, this sensitivity was also seen with two more selective FGFR inhibitors, NVP-BGJ398 and AZD4547 (Supplementary Fig. S1). Sensitivity to PD-173074 in the MPM cell lines was confirmed by clonogenic survival assays (Fig. 1C). Although some sensitive lines died by apoptosis, as is shown by activated caspase activity with both PD-173074 and the multi–FGFR-targeted inhibitor AZD4547 (Fig. 1D and E), not all sensitive lines showed a dose incremental increase in this marker.

These data confirm previous findings (18) that a subset of MPM cell lines require FGF pathway activation for growth and survival, and that targeting this pathway could be a critical step in the control of these tumors.

To test whether these observations could be reproduced in an independent cohort of primary mesothelioma cell lines, a panel of 11 pleural fluid-derived early-passage cultures from patients with MPM tumors were obtained and screened for viability using a panel of 48 small molecule inhibitors including PD-173074. Most of the early-passage cultures were resistant to virtually all agents (Supplementary Fig. S2). However, one MPM early-passage culture (NKI04) did demonstrate marked sensitivity to PD-173074. The sensitivity of NKI04 to FGFR inhibition was confirmed in a longer duration clonogenic survival assay, and the effect on cell viability was comparable with that seen in the FGFR1-amplified NCI-H1581 lung cancer cell line (Fig. 2A–C).

In order to understand the basis for the observed sensitivity to FGFR inhibition, we analyzed whole-exome sequence and copy number array data for 21 MPM lines (http://cancer.sanger.ac.uk/cell_lines). There was no evidence of activating mutations or whole gene amplifications in any FGFR family member. RNA sequencing has been undertaken and shows no evidence of a fusion transcript involving any member of the FGFR family in any of the MPM cell lines (personal communication, M. Garnett). We then analyzed the corresponding gene expression data and focused on differential expression of FGFR and FGF family members in PD-173074-sensitive and -resistant MPM cell lines. Normalized expression of each of the FGF and FGFR family genes was correlated with sensitivity to PD-173074 to explore whether the variation in any single family member, either ligand or receptor, was associated with response to FGFR inhibition. We found a statistically significant correlation between elevated FGF9 mRNA expression and response to PD-173074 (P = 0.0148) and AZD4547 treatment (P0.0098; Fig. 3A). FGF9 is a secreted, high-affinity ligand for the FGFR3 receptor, with low affinity for the FGFR1 and FGFR2 receptors (19). To determine whether a subset of MPM exhibits elevated expression of the FGF9 ligand in patients, we analyzed gene expression from a panel of 53 assorted MPM and matched normal lung clinical samples (Fig. 3B; ref. 20). Overall, we observed significantly higher FGF9 transcript levels in MPM tumors compared with pleura and lung normal tissue (P < 0.0001). Therefore, similar to our observation in the MPM cell lines, a subset of patient samples also demonstrates high levels of FGF9 expression.

A possible premise for the observed sensitivity of MPM lines that express high levels of FGF9 would be activation of the FGFR3 receptor kinase in an autocrine loop and subsequent engagement of prosurvival downstream signaling pathways. Indeed, a comparison of phosphorylation status of 42 receptor tyrosine kinases between a small sample of MPM cell lines demonstrated increased phosphorylation of FGFR3 in the sensitive line H2795 but not in resistant lines Met-5A and NCI-H28 (Fig. 3C).

To further confirm a critical role for FGFR3, this transcript was silenced by siRNA in a panel of MPM cell lines and the direct effect on cell viability was measured. Transient siRNA-mediated silencing of the FGFR3 transcript reduced cell viability in all 3 FGFR inhibitor–sensitive cell lines, but not in the FGFR inhibitor–resistant lines. This indicates a dependency on FGFR3 mediated signaling of the FGFR inhibitor–sensitive lines (Fig. 3D). As would be expected, inhibition of FGFR3 by the specific inhibitors AZD4547 and BJG398 decreased pERK levels (Fig. 3E), and this was also seen following siRNA-mediated silencing of FGFR3 in H2795 and MSTO-211H (Fig. 3F). The addition of the FGF9 ligand to MPM cells lacking baseline FGFR3 activation was able to induce phosphorylation of FGFR3 and a change in the growth kinetics of this cell line in a dose-dependent fashion (Supplementary Fig. S5).

Although we failed to identify genomic alterations in any member of the FGFR family that might explain the sensitivity to FGFR inhibition, we reasoned that this dependency might also be the consequence of other gene aberrations up- or downstream of FGFR3 signaling. We evaluated the gene expression and mutation database for other statistical associations explaining sensitivity to the FGFR inhibitor AZD4547 in the panel of MPM cell lines. We focused on driver mutations or copy-number alterations in three of the most frequently mutated genes in MPM, namely, BAP1, NF2, and CDKN2A (7). We detected a weak but nonsignificant association between AZD4547 sensitivity and BAP1 mutations in the sensitive cell lines (Fig. 4A). Given that loss of BAP1 protein expression might also occur through nonmutational mechanisms as previously described (21), we additionally characterized BAP1 protein status in these lines by Western blot analysis (Supplementary Figs. S3 and S4). When sensitivity to the AZD4547 was correlated with BAP1 protein expression (low/absent vs. expressed), there was a significant correlation between loss of BAP1 expression and sensitivity (P = 0.0208; Fig. 4B).

Because silencing FGFR3 reduced cell viability in a subset of MPM lines, we next investigated whether this dependency on FGFR signaling was regulated by BAP1. BAP1 is a nuclear deubiquitinating enzyme with many unelucidated functions that might include modulation of the FGFR pathway. Silencing of BAP1 expression resulted in increased phosphorylation of FGFR3 (Fig. 4C). Conversely, restoring BAP1 expression in the BAP1-null MPM line (Fig. 4D) H226 resulted in a decrease in pFGFR and a modest increase in resistance to the FGFR inhibitor AZD4547 (Fig. 4E).

We observed increased expression at the protein level in the BAP1 mutant cell lines of other RTK receptor genes and their appropriate ligands also known to be important in cell survival signaling in MPM such as PDGFRB, IGF1-R, and MET (22) using phospho-RTK arrays (Supplementary Fig. S4A and S4B). The H226-null MPM cell line was transfected with a wild-type BAP1 construct and a functionally inactive C91A-mutant BAP1 construct. Gene expression analysis on these two lines was performed and Signaling Pathway Impact Analysis (SPIA) of the data (Supplementary Table S) demonstrated that among the most significantly activated pathways in BAP1-inactive cells is the “Bladder Cancer” pathway including FGFR3 (arrow, Supplementary Fig. S6A) illustrated in Supplementary Figure S6B (23). In summary, the gene expression analysis demonstrates that BAP1 loss of function is associated with a transcriptional response upregulating not only FGFR signaling but also other RTKs such as PDGFRB, CMET, and IGF1R, that may be important mediators of cell growth and survival. However, only FGFR inhibitors showed a significant viability effect as single agents. We analyzed gene expression data from a study of 51 mesothelioma tumor samples to see if a similar effect on the FGFR pathway was seen in vivo (40 BAP1 wild-type and 11 mutant; GEO GSE29354; ref. 24). Amongst members of the FGFR signaling family, BAP1-mutant tumors did indeed demonstrate increased expression of FGF18, FGFR2, and FGFR3 relative to BAP1 wild-type tumors (Supplementary Table). To explore this association further in human tumors, we analyzed the available TCGA data and looked for the incidence of genetic and mRNA alterations of these genes in MPM tumors by BAP1 status (Fig. 4F). This showed the majority of dysregulation (10 of 14) events in FGF9, FGF18, and FGFR3 occurred in the context of BAP1 gene or mRNA dysregulation.

To assess the in vivo efficacy of targeting FGFR in MPM, we established a xenograft model using the FGFR inhibitor–sensitive MPM lines H2795 and MSTO-211H. Mice were treated with AZD4547, a selective inhibitor of FGFR1/2/3, which is currently being evaluated in clinical trials. We observed that treatment with AZD4547 resulted in significant growth inhibition in the H2795- and MSTO-211H–derived tumors (Fig. 5A). Furthermore, AZD45457 treated tumors showed a reduction in pERK signaling by immunohistochemistry compared with vehicle control-treated tumors (Fig. 5B), indicating target engagement by the drug in this model. Caspase activation was also seen in drug-treated tumors suggesting apoptosis (Supplementary Fig. S7).

As the single-agent efficacy of FGFR inhibition was seen only in a subset of MPM cell lines, and because persistent pAKT pathway activation was seen in cell lines not responsive to FGFR inhibition, we hypothesized that a combination screen utilizing a PI3 kinase inhibitor may reveal useful synergies. We undertook an anchor-based combination screen in 15 MPM cell lines using 95 small-molecule inhibitors (see Supplementary Table for details) selected to target many critical pathways in cancer, both as single agents and in combination with a fixed dose of the PI3 Kinase inhibitor AZD6482. The resulting difference in area under the curve (AUC) between single agent small-molecule inhibitor and the combination with AZD6482 was used to calculate synergy. The most recurrent synergistic interactions were seen with IGF1R inhibitor BMS-536924 and FGFR inhibitor PD-173074 (Supplementary Fig. S8A) with synergy observed in seven and six of 15 lines, respectively. Supplementary Fig. S8B shows a validation dose–response curve of the FGFRi-resistant NCI H28 cell lines showing minimal effect of BMS 536824 or AZD6482 alone, but reduced viability and pAKT reduction with the combination. This cytotoxicity is not seen in the mesothelial control cell line Met5a, suggesting that the synergy is not generic but cell line specific.

Because MPM is a rare and heterogeneous tumor, it is notoriously difficult to identify and characterize responding subgroups in clinical trials. Our work illustrates the application and possibilities of comprehensive pharmacogenomic profiling approaches in intractable cancers such as MPM. The finding of FGFR inhibitor sensitivity in a subgroup of immortalized MPM cell lines represents a potentially novel therapeutic approach for this tumor type. As immortalized cell lines may undergo genetic drift, we also confirmed our findings in primary mesothelioma early-passage lines.

Dysregulation of the FGFR pathway has been described in many cancer types (25, 26). FGF9 signaling through FGFR3 has been shown to have a role in the development and progression of tumor cells in mouse models for NSCLC and prostate cancer (27). In MPM cell line models, we observed that high levels of the ligand FGF9 were strongly correlated with sensitivity to the FGFR inhibitor PD-173074 and AZD4547. We hypothesize that the effects of FGF9 are mediated through FGFR3 signaling, as illustrated by modulation of downstream ERK phosphorylation upon chemical inhibition with small-molecule inhibitors of FGFR3 and knockdown of FGFR3. FGFR3 is conversely not phosphorylated in cell lines insensitive to FGFRi, and this phosphorylation can be induced by the addition of synthetic FGF9 ligand. Interestingly, there was variability in FGF9 mRNA expression levels among the MPM cell lines, similar to what is observed in tumors in previously published studies. Recently, other groups demonstrated efficacy of FGFR inhibition in preclinical models of MPM mediated by other FGF-pathway members such as FGFR1 (18, 28, 29). We confirm the efficacy of a clinically utilized FGFR inhibitors including AZD4547 in vivo in MPM xenograft models. Furthermore, since undertaking these studies, early-phase clinical work with pharmacokinetic data has been published (30, 31) on AZD4547 and BGJ398. These have confirmed that the doses used in the in vitro work (100 nmol/L to 1 μnmol/L) here are achievable in plasma in vivo and are able to modulate the target, with pharmacodynamic end points of target engagement with FRS2 downregulation and changes in serum phosphate levels seen.

FGF receptors and ligands are being targeted in clinical trials by both selective and nonselective FGFR TKI's and monoclonal antibodies (32) and AZD4547 has shown modest clinical activity in tumors with FGFR pathway aberrant activation (33). In MPM dovitinib, a multitargeting kinase inhibitor with activity against FGFR has been trialed and has failed in small cohort of patients with MPM (34). Because the data across tumor types demonstrate only a small group of patients responds to FGFR inhibition, it is crucial to find biomarkers that predict response to FGFR inhibition. Guagnano et al. integrated genomic and transcriptomic data of about 500 tumor cell lines with drug-sensitivity data to find predictive biomarkers for response to FGFR inhibitor NVP-BGJ398. A genetic alteration in one of the four FGF receptors was found in 7% of cell lines, but only about half of the cell lines with such an alteration was found to be sensitive (35).

We did not find any mutation, amplification, or fusion transcripts of the FGFR family in the inhibitor-sensitive MPM cell lines. The genes that were most recurrently altered in our MPM cell lines include CDKN2A, BAP1, and NF2. The frequency at which these genes were mutated is broadly similar to those previously described in clinical MPM samples (6, 7).

We show that loss of BAP1 expression was associated with sensitivity to FGFR inhibition. This finding was further validated with modulation of pFGFR signaling and dose–response kinetics to FGFR inhibition following siRNA-mediated knockdown and BAP1 overexpression in MPM cell lines. Caveats with this association were also observed: NCI-H28 was one of the most resistant cell lines to FGFR inhibition but carried a BAP1 homozygous deletion, suggesting that BAP1 loss may enrich for FGFR inhibitor–sensitive cell lines but that some heterogeneity of drug response may still be observed. BAP1 (BRCA-associated protein 1) is a nuclear deubiquinating enzyme that controls gene expression by interaction with numerous transcription factors and other complexes, including those of the double strand DNA-break repair machinery (36). BAP1 thus influences cell-cycle progression (37) and double-strand DNA break repair (38). We show here that its loss may also affect gene expression of FGF pathway members, thereby enhancing signaling through this pathway.

The BAP1 gene is inactivated by somatic mutation in 23% to 64% of patients with MPM and between 1% and 47% in other tumor types (24, 39-43). Furthermore, BAP1 protein levels are undetectable in about 25% of MPM with normal BAP1 gene status, likely by epigenetic modification (24). BAP1 loss was observed to enrich for FGFR inhibitor–sensitive MPM lines, and expression of C91 hydrolase inactive mutant versus wild-type BAP1 protein in the H226 cell line induced activation of FGFR3 signaling. We hypothesize that inactivation of BAP1 in MPM, possibly through its function as a ubiquitin hydrolase, induces changes in gene expression of both FGF family ligands and receptors to stimulate cell growth and survival.

We performed a combination drug screen to assess the impact of novel combinations of targeted therapies on MPM cell lines. On the 15 MPM cell lines screened, we found that FGFR and IGF1R inhibitors were the most recurrently synergistic with the PI3-kinase inhibitor AZD6482. This is the first time, to our knowledge, that both a single agent and combination therapeutic screen have been performed, which point to the primacy of the FGFR signaling pathway in MPM. Interestingly, one of the most resistant cell lines to FGFR inhibition was amenable to treatment with AZD6482 plus IGF1R inhibition with evidence of ablation of pAKT with the combination of drugs but not with either alone, implying true synergy. Previous studies have identified that multiple RTK's are active in MPM (14), and this has provided some rationale to consider combination therapies to overcome innate resistance to targeted therapies. It is also interesting to speculate as to whether IGF1R plus Pi3K inhibition would be of use in acquired resistance to FGFR inhibitors.

High-throughput drug screening revealed a subset of both immortalized and primary mesothelioma cell lines to be highly sensitive to FGFR inhibition. This sensitivity was mediated through FGFR3 and was associated with loss of BAP1 protein expression. The high incidence of BAP1 protein loss in MPM tumors implies potential benefit from FGFR inhibition for a substantial subset of this patient group. In addition, our anchor-based screens revealed synergistic combinations that helped to overcome innate resistance to FGFR inhibition.

No potential conflicts of interest were disclosed.

Conception and design: J.M. Quispel-Janssen, A. Berns, P. Baas, U. McDermott, J. Neefjes, C. Alifrangis

Development of methodology: J.M. Quispel-Janssen, J. Badhai, P. Baas, J. Neefjes, C. Alifrangis

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.M. Quispel-Janssen, J. Badhai, L. Schunselaar, S. Price, J. Brammeld, K. Kolluri, M. Garnett, P. Baas, U. McDermott, C. Alifrangis

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.M. Quispel-Janssen, J. Badhai, L. Schunselaar, F. Iorio, U. McDermott, C. Alifrangis

Writing, review, and/or revision of the manuscript: J.M. Quispel-Janssen, J. Badhai, L. Schunselaar, P. Baas, U. McDermott, J. Neefjes, C. Alifrangis

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Schunselaar, M. Garnett, C. Alifrangis

Study supervision: A. Berns, P. Baas, J. Neefjes, C. Alifrangis

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.