EGFR inhibitors identified as a potential treatment for chordoma in a focused compound screen

Abstract Chordoma is a rare malignant bone tumour with a poor prognosis and limited therapeutic options. We undertook a focused compound screen (FCS) against 1097 compounds on three well‐characterized chordoma cell lines; 154 compounds were selected from the single concentration screen (1 µm), based on their growth‐inhibitory effect. Their half‐maximal effective concentration (EC50) values were determined in chordoma cells and normal fibroblasts. Twenty‐seven of these compounds displayed chordoma selective cell kill and 21/27 (78%) were found to be EGFR/ERBB family inhibitors. EGFR inhibitors in clinical development were then studied on an extended cell line panel of seven chordoma cell lines, four of which were sensitive to EGFR inhibition. Sapitinib (AstraZeneca) emerged as the lead compound, followed by gefitinib (AstraZeneca) and erlotinib (Roche/Genentech). The compounds were shown to induce apoptosis in the sensitive cell lines and suppressed phospho‐EGFR and its downstream pathways in a dose‐dependent manner. Analysis of substituent patterns suggested that EGFR‐inhibitors with small aniline substituents in the 4‐position of the quinazoline ring were more effective than inhibitors with large substituents in that position. Sapitinib showed significantly reduced tumour growth in two xenograft mouse models (U‐CH1 xenograft and a patient‐derived xenograft, SF8894). One of the resistant cell lines (U‐CH2) was shown to express high levels of phospho‐MET, a known bypass signalling pathway to EGFR. Neither amplifications (EGFR, ERBB2, MET) nor mutations in EGFR, ERBB2, ERBB4, PIK3CA, BRAF, NRAS, KRAS, PTEN, MET or other cancer gene hotspots were detected in the cell lines. Our findings are consistent with the reported (p‐)EGFR expression in the majority of clinical samples, and provide evidence for exploring the efficacy of EGFR inhibitors in the treatment of patients with chordoma and studying possible resistance mechanisms to these compounds in vitro and in vivo. © 2016 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of Pathological Society of Great Britain and Ireland.


Introduction
Chordoma is a rare primary malignant bone tumour showing notochordal differentiation and develops for the most part in the bones of the base of the skull, the vertebral bodies and the sacro-coccygeal region [1][2][3].
There are occasional reports of extra-axial and soft tissue lesions [1,4]. The median survival for patients with chordoma is 7 years [2,3]. Advances in radiation technology with either particles or photons have allowed delivery of higher doses of radiation [2,3] and can be beneficial for local disease control. However, 30-40% of chordomas metastasise and there are no approved agents for the treatment of patients with inoperable and metastatic chordoma [3]. Cytotoxic chemotherapy is not active in this tumour type [3,5]. Imatinib, an inhibitor of platelet-derived growth factor receptor (PDGFR), has demonstrated limited activity in a phase II study and when used in a compassionate programme [6,7]. However, there are encouraging results, in the form of anecdotal reports, on the response of chordoma to epidermal growth factor receptor (EGFR) [8][9][10][11][12][13] and vascular endothelial growth factor (VEGF) inhibitors [5,[12][13][14], although data from prospective randomized clinical trials are lacking [5,14].
Chordoma is characterised by the expression of the transcription factor T (brachyury) [15] and there is a body of evidence supporting its critical role in this disease [16]. Specifically, study of the T regulatory network revealed that epidermal growth factor (EGF), transforming growth factor-α (TGFα) and fibroblast growth factor 1 (FGF1) ligands, amongst others, are direct products of T-mediated transcription [17]. These findings are supported by strong immuno-expression of the phosphorylated proteins in chordoma [18][19][20][21].
Genotype-directed therapy represents a major strategy for planning new cancer treatments, and this has demonstrated success by improving outcome in close to 70% of patients with non-small cell lung cancer harbouring EFGR mutations, albeit for 1-2 years before developing resistance [22]. However, despite chordomas being immunoreactive for the activated form of EGFR (p-EGFR), they do not harbour EGFR mutations and only infrequently other currently potentially tractable targets, such as PIK3CA mutations [18,[23][24][25]. In view of the unmet need for effective treatment of patients with chordoma, we undertook a large-scale compound screen on three chordoma cell lines and validated the key target in an extended panel of seven cell lines, with the aim of finding therapies and understanding the mechanism by which this disease develops. Such approaches are reported to be more successful than target-based approaches in identifying drug candidates with clinically relevant mechanisms of action [26][27][28].

Cells and cell lines used in the screen
Seven human chordoma cell lines, U-CH1, U-CH2, U-CH7, U-CH10, MUG-Chor1, JHC7 and UM-Chor1, were studied (see supplementary material) and quality controlled by short-tandem-repeat (STR) analysis (DNA Diagnostic Centre, London, UK) (see supplementary material, Table S1) and regular Mycoplasma testing [29][30][31][32][33]. In the absence of the availability of notochordal tissue, a transient embryonic structure considered to represent the origin of chordoma [15], human dermal fibroblasts (ATCC ® PCS-201-012™) were used as a non-neoplastic control cell population. NCI-N87 (ATCC ® CRL-5822™), a gastric cancer cell line which strongly expresses EGFR and ERBB2 in the absence of downstream mutations [34,35], served as a positive control to EGFR inhibitors. Cells were cultured according to ATCC guidelines (see supplementary material). All chordoma lines included derive from sacral tumours other than UM-Chor1 which derives from a clival neoplasm (http://www.chordomafoundation.org/) [29][30][31][32][33]. Cell pellets from the chordoma cell lines were formalin-fixed and paraffin-embedded and 3 μm sections cut for immunohistochemistry and FISH [18]. Ethical approval was obtained from the Cambridgeshire 2 Research Ethics Service (reference 09/H0308/165) and the UCL Biobank for Health and Disease Ethics Committee.

Focused compound screen (Figure 1)
Compounds were tested on three chordoma cell lines (U-CH1, U-CH2 and MUG-Chor1) using a non-randomised plate layout in a 96-well plate format (80 compounds/plate) at a single concentration of 1 μM (n = 3 minimum). Cells were seeded in medium (90 μl/well) using a Multidrop Combi (MDC; Thermo Fisher Scientific, Loughborough, UK) and cultured for 24 h before the compounds were added. The compounds were diluted from 10 mM stocks using an ECHO 550 (Labcyte, CA, USA) to create 10× compound plates and added (10 μl/well) using a Biomek FX (Beckman Coulter, Brea, CA, USA). Cell survival was assessed following 96 h of compound treatment using the water-soluble tetrazolium salt (WST1) assay (Roche Diagnostics, Burgess Hill, UK) according to the manufacturer's recommendations.

Hit selection
Hit selection thresholds were calculated independently for each cell line. Percentage inhibition for each compound at 1 μM was calculated from raw data relative to the controls on each plate. Thresholds were calculated by combining the results of the different libraries to give a mean percentage inhibition, or by analysing data obtained from the libraries independently (PKIS,

Hit confirmation
The 'hits' of the single concentration screen and their potencies were generated using the half-maximal effective concentration (EC 50 ) in a 10-point dose-response format, with the highest concentration at 30 μM. Maximum percentage inhibition (MI) was recorded at the highest concentration of each compound. Each compound was tested in three independent experiments with two replicates/experiment. Staurosporine (SRPN; Sigma-Aldrich, St. Louis, MO, USA) was used as a positive control, with EC 50 values monitored to ensure reproducibility between each run. An arbitrary threshold of EC 50 < 5 μM in chordoma cell lines and > 10 μM in dermal fibroblasts was applied to select compounds which selectively killed chordoma cells. Selectivity was defined as the fold difference between a compound's EC 50 in fibroblasts and chordoma cells.

Analysis of cell death
The Caspase-Glo ® 3/7 Assay (Promega, Southampton, UK) and the CellTiter-Glo ® Luminescent Cell Viability Assay (Promega) were used on separate assay plates to monitor cell viability and to determine induction of apoptosis. Profiling was conducted from the highest concentration (20 μM) in a dose-dependent manner with a 1:3 serial dilution (minimum two independent experiments/compound). Data analyses were performed using XLfit v. 5.0 (IDBS, Guildford, UK).

Protein extraction and western blot (WB) analysis
Details of these methods are described in Supplementary materials and methods and primary and secondary antibodies are listed in Table S4 (see supplementary material for both).

Combination study of sapitinib with the MET inhibitor crizotinib
The MET inhibitor crizotinib (Xalkori ® , Pfizer, NY, USA) was tested in combination with sapitinib using a non-randomized plate layout in a 384-well format. Details of this experiment are described in Supplementary materials and methods (see supplementary material).

Immunohistochemistry
Immunohistochemistry was performed on the Leica Bond-III detection platform, using the Bond Polymer Refine Detection system (Leica). Expression was evaluated as reported previously (see supplementary material) [18].

Analysis for mutations in cancer gene hotspots of chordoma cell lines
The hotspots in 22 cancer-related genes were analysed for mutations using the Ion AmpliSeq™ Colon and Lung Cancer Research Panel v. 2 (Thermo Scientific) (see supplementary material).

In vivo studies
Sapitinib (AstraZeneca, Cambridge, UK) was tested at South Texas Accelerated Research Therapeutics (START) on two chordoma mouse models: one model was a cell line-derived xenograft (U-CH1) [32], the other a patient-derived xenograft (SF8894) [43] (see supplementary material).

A focused compound screen showed that EGFR/ERBB family inhibitors targeted chordoma cells selectively
Of the 1097 compounds screened at a single concentration in three human chordoma cell lines (U-CH1, U-CH2 and MUG-Chor1; see supplementary material,  Figure 2), of which five also represented BRAF inhibitors. The results were reproducible using two different batches of compounds across all cell lines (data not shown). These 21 EGFR/ERBB inhibitors exerted the highest potency and maximum effects on U-CH1, whereas negligible activity was observed on U-CH2 (Table 1).

Chemical substituents correlated with EGFR/ERBB inhibitor activity in chordoma cell lines
Twenty-one EGFR/ERBB inhibitors were selected by GSK for structural analysis on the basis of their phenotypic (viability) potencies (Table 2). A number of these compounds are currently under clinical development. These compounds were found to represent two chemotypes, pyrimidines and two subtypes of quinazolines. The quinazolines were characterised by either large substituents off the aniline group in the 4-position of the quinazoline ring (hereafter referred to as 'quinazolines large'), or small substituents on the aniline ring in this position ('quinazolines small'). 'Quinazoline small' compounds, including erlotinib, sapitinib and gefitinib, exerted activity on U-CH1, U-CH7 and MUG-Chor1 and showed a trend to greater potency than 'quinazoline large' compounds (Table 2). Neither subtype was active on U-CH2. Biochemical assessment of the IC 50 on a selection of these compounds (n = 14) on EGFR, ERBB2 and ERBB4 (ERBB3 was not tested) showed that both the size of the substituent group on the aniline ring and the structure of the tail portion of the molecule, which extends towards the solvent front of the kinase, had an impact on the potencies of the three targets (see supplementary material, Table S3).

Commercially available EGFR/ERBB family inhibitors exerted good potency and high selectivity in chordoma cell lines
We next tested a set of six commercially available EGFR/ERBB inhibitors consisting of four reversible (erlotinib, sapitinib, gefitinib and lapatinib) and two irreversible (afatinib and poziotinib) compounds in an extended panel of seven chordoma cell lines, including the three previously tested ones (Table 3). Controls included human dermal fibroblasts (ATCC ® PCS-201-012™) and the gastric cancer cell line NCI-N87 (Table 3). Four of the chordoma cell lines, U-CH1, U-CH7, UM-Chor1 and MUG-Chor1, responded to EGFR inhibition with EC 50 concentrations < 1 μM, whereas U-CH2, U-CH10 and JHC7 were largely resistant ( Table 3). Three of four reversible agents, erlotinib, gefitinib and sapitinib, were highly potent, exerting an effect in the nanomolar range in the four 'responsive' cell lines. In contrast, lapatinib was potent on UM-Chor1 (EC 50 320 nM and 98% MI), showed only moderate activity (EC 50 ≥ 1 μM ≤ 3 μM) on MUG-Chor1 and no activity on the other cell lines (EC 50 ≥ 3 μM) ( Table 3). Of the two irreversible EGFR/ERBB inhibitors, afatinib and poziotinib, the former displayed a significant kill effect on UM-Chor1 (EC 50 of 26 nM and 89% MI) and a cytostatic profile on U-CH1 and MUG-Chor1. Poziotinib displayed a cytostatic profile on U-CH1, MUG-Chor1 and UM-Chor1: U-CH7 was resistant to both (Table 3). Sapitinib, a 'quinazoline small' compound, showed the most exciting results, with EC 50 concentrations in the nanomolar range in the four 'responsive' lines comparable to those observed in non-small cell lung cancer (NSCLC) and head and neck cancer cell lines defined as being sensitive to EGFR [51]. In these other cancer models, responses to the drug were also observed in related mouse xenograft models. The other active compounds were two FDA-approved 'quinazoline small' compounds, gefitinib and erlotinib (Table 3).

EGFR/ERBB family inhibitors suppressed p-EGFR and downstream effectors in chordoma cell lines and induced apoptosis in sensitive cell lines
ELISA and western blot data showed that our three most promising compounds (sapitinib > gefitinib > erlotinib) induced a dose-dependent suppression of the biomarker       Tables 1, 2), was studied by western blot on three chordoma cell lines (U-CH1, U-CH2, MUG-Chor1). Cells were serum-starved overnight before being treated with EGFR inhibitors (250 nM) for 4 h and then being exposed to EGF (50 ng/ml) for 15 min.
Quinazoline large: large substituent on the aniline ring in the 4-position of the quinazoline ring system. Quinazoline small, small substituent on the aniline ring in the 4-position of the quinazoline ring system; NAHDF, normal adult human dermal fibroblasts; CS, cytostatic, no potency calculated; NT, not tested. EC 50 was extrapolated in cases of low percentage inhibition where no plateau was observed, indicating inactivity of these drugs (Geomean EC 50 ≥ 30 μM). Poziotinib (Selleckchem) was replaced by the irreversible EGFR TKI canertinib (Selleckchem) in this analysis. p-EGFR at two different phosphorylation sites (Tyr1068 and Tyr1173) in all cell lines, confirming that the drugs hit their key target. Similar results were shown in EGF spiked/serum-starved ( Figure 3; see also supplementary material, Figure S2) and non-spiked/non-starved experiments (see supplementary material, Figures S3, S4). Key effectors of EGFR signalling, including p-AKT (PI3K-AKT-mTOR) and p-ERK1/2 (Ras-Raf-MAPK-ERK1/2) were also dose-dependently suppressed in response to these treatments. We did not observe significant effects on p-STAT3 (see supplementary material, Figures S3, S4). Inhibition of p-EGFR and downstream targets occurred at lower doses with sapitinib than with the other EGFR inhibitors tested. The endogenous/baseline status for all of the markers investigated is shown in Figure S5 (see supplementary material). The EGFR/ERBB inhibitors induced apoptosis in a dose-dependent manner. This resulted in a decrease in cell viability from 24 h onwards (Figure 4; see also supplementary material, Figure S6). Where a cytostatic profile was generated in response to EGFR inhibitors, such as with lapatinib in the U-CH1 cell line, minimal caspase 3/7 activity (∼30%) was observed in support of these phenotypic data (Figure 4; see also supplementary material, Figure S6).

Sapitinib significantly reduced tumour growth in chordoma mouse models
Our most promising compound, sapitinib, significantly reduced tumour growth in two chordoma xenograft models ( Figure 3C, D).

Study of resistance mechanisms in U-CH2, JHC7 and U-CH10
Next we sought reasons why our best compounds were not effective in the unresponsive chordoma cell lines. We addressed the issue of EGFR tyrosine kinase inhibitor resistance by investigating known mechanisms of resistance from work in other cancer types. Next-generation sequencing failed to detect mutations covering the hotspots in 22 tumour-related genes, which included EGFR, ERBB2 and ERBB4 and their downstream effectors (KRAS, BRAF, PIK3CA, AKT1, PTEN, NRAS and MAPK). Thus, we were not able to detect obvious genetic explanations for sensitivity and/or resistance in our cell line panel.
Activation of ERBB2 and MET can result in bypass resistance pathways. Together with EGFR, no amplifications in ERBB2 and MET were identified by FISH in the cell lines (see supplementary material, Table S7). Moreover, no mutations were identified in MET. Furthermore, MET amplification, assessed by FISH, was only seen in two (non-clival) of 114 clinical chordomas, 66 of which were located along the spinal axis and 48 in the clivus. However, western blots and immunohistochemistry for p-MET revealed strong expression in U-CH2 (Figures 5A, 6B). To test whether this contributed to the resistance observed, we treated U-CH2 with sapitinib in combination with the MET-inhibitor crizotinib and observed a significant synergistic effect ( Figures 5B, 6A).
The loss and decreased protein expression of PTEN, a tumour suppressor gene, has also been linked to EGFR tyrosine kinase inhibitor (TKI) resistance [52,53]. All cell lines apart from U-CH1 showed variable immunoreactivity with a PTEN antibody, strong in U-CH2 and U-CH7, moderate in MUG-Chor1, JHC7 and U-CH10 and weak in UM-Chor1. These data were confirmed by western blots for PTEN ( Figure 6C and 6E). Resistance has also been associated with loss of E-Cadherin expression: it has been proposed that this results in epithelial-mesenchymal transition (EMT) and a related increase in motility [52,54]. Immunohistochemistry for E-Cadherin showed weak and only focal expression in U-CH7 and MUG-Chor1 and was negative in the remaining five cell lines ( Figure 6D). The Hippo downstream effector Yes-associated protein (YAP) [55,56] has also been reported to confer resistance to EGFR inhibitors [57][58][59]. Even though YAP was highly expressed in the resistant cell line U-CH10, strong expression levels were also observed in MUG-Chor1, a cell line responsive to EGFR TKIs (see supplementary material, Figure S5). Hence, PTEN, E-Cadherin, and YAP protein expression in the cell lines appear not to show an obvious correlation with the presence and/or absence of resistance to EGFR inhibitors.

Discussion
In the light of the absence of recurrent genetic alterations in chordoma, we chose a phenotypic screening approach to identify the mechanism(s) by which this disease is driven and/or targets that could potentially be translated into clinical practice [28]. We screened three chordoma cell lines (U-CH1, U-CH2 and MUG-Chor1) against 1097 compounds, 1046 of which were small molecule kinase inhibitors. Twenty-seven compounds remained of interest following exclusion of compounds that failed to mediate a chordoma-selective cell kill effect. The majority (21/27) of these compounds targeted the EGFR/ERBB family. Gene enrichment analysis for non-EGFR hit compounds revealed that VEGFR1/2 signalling covers most of their target genes, which is in line with isolated case reports showing activity of VEGF inhibitors in patients with chordoma [12,13]. However, since most of the non-EGFR hits were multikinase inhibitors and their targets ill-defined, and/or the compounds not phenotypically as potent as the EGFR/ERBB inhibitors, we focused on EGFR inhibitors in this study. As the majority of these 21 EGFR inhibitors remain under development, we tested six other EGFR inhibitors [erlotinib (Roche/Genentech); gefitinib (AstraZeneca); sapitinib (AstraZeneca); lapatinib (GSK); afatinib (Boehringer Ingelheim, Germany); and poziotinib (Spectrum Pharmaceuticals, Irvine, CA, USA)] that were either FDA-approved or have been in clinical trials [39,40,42,60]. Of the seven chordoma cell lines tested, we demonstrated that four (U-CH1, U-CH7, MUG-Chor1 and UM-Chor1) were sensitive and three (U-CH2, U-CH10 and JHC7) were resistant to EGFR inhibition. The limitations of all in vitro screens apply to our study, but we imposed a high level of quality control measurements to ensure the generation of robust data [26][27][28].
EGFR was the first tyrosine kinase receptor to be linked to tumourigenesis [52,[61][62][63] and therapeutic inhibition of this pathway has yielded varying success in the treatment of malignant disease [39,52,62]. A common cause for EGFR activation is the presence of mutations and gene amplification, as seen in non-small cell lung cancer and glioblastoma [22,39,64,65]. However, this is not the case for the cell lines in our study, which is consistent with published EGFR genetic profiling reports [19,23] and the unpublished data from ∼30 whole genomes/whole exomes from our laboratory. The absence of EGFR mutations in chordomas is shared with other cancers, such as head and neck squamous cell carcinoma and colorectal and pancreatic cancers, which are known to respond to anti-EGFR therapy to varying degrees [52,[66][67][68][69][70][71][72]. Similar to these tumours, chordoma cell lines express the activated form of the receptor and show suppression of the downstream EGFR signalling pathways following treatment with EGFR inhibitors. The clinical relevance of our in vitro studies is supported by the documented expression of these markers in patients' samples [18][19][20][21][73][74][75]. Specifically, up to 52% of 170 chordoma samples have been reported to express p-EGFR [18][19][20][21]76], although as phosphorylated protein is unstable, this is likely to be an underestimate [19,21,73,74]. The significance of the finding that erlotinib had a significant kill effect on four of seven cell lines tested is supported by the response to erlotinib seen in a well-characterised, patient-derived chordoma xenograft mouse model [77], which is consistent with the significant growth reduction we observed for sapitinib in two xenograft mouse models. There are also a number of well-documented reports of patients with chordoma showing partial regression and/or clinical improvement following EGFR TKI treatment [8][9][10][11][12][13]78]. In contrast, a small non-randomised phase II clinical trial with the FDA-approved dual EGFR/ERBB2 inhibitor lapatinib (Tykerb ® /Tyverb ® ), involving 18 patients with advanced chordoma, showed only a modest clinical success with partial intratumoural response in six of 18 patients, according to Choi [76]. This finding is consistent with the failure of lapatinib to exert a kill effect at therapeutic concentrations in all but one of our chordoma cell lines. In an attempt to address differences in the phenotypic kill effect observed in response to different EGFR TKIs in our large compound panel, we undertook analysis of their chemical structures and substituents. The overall finding was that S Scheipl et al compounds such as sapitinib, with small substituents appended to the aniline ring in the 4-position of the quinazoline ring system, were more effective than compounds such as lapatinib, with large substituents, albeit these findings derive from a small sample size. However, the complexities of this potency, selectivity and phenotypic response relationship demand further study so that activity can be optimized for chordoma patients.
The expression of p-EGFR in chordoma may be explained on the basis that epidermal growth factor (EGF) ligands are direct targets of T [17], the expression of which is considered to be critical in the growth of this tumour. This could also explain why T expression is not suppressed on western blot in response to EGFR inhibitors. It is therefore interesting that early-phase clinical trials involving vaccines against T in patients with lung cancer and chordoma are showing some evidence of clinical activity [79][80][81], and it would be of interest to know whether p-EGFR is suppressed in the clinical samples from these patients. The combination of one these vaccines and EGFR inhibition may be more effective than a monotherapy.
In an attempt to understand why three of seven cell lines were resistant to EGFR inhibitors, we studied the common mechanisms by which EGFR TKI resistance occurs [22,52,54]. We were unable to detect downstream mutations in PIK3CA, BRAF, KRAS, MAPK1 and others [22,52,54]. Nevertheless, as PIK3CA alterations have been reported in a minority of chordomas, these could be used to stratify patients for future EGFR inhibitor clinical trials [23]. ERBB2 amplification, another reported resistance mechanism [22,52], was not identified in the cell lines. Loss of heterozygosity for PTEN has been reported frequently in chordoma [19,23,24,52,53,[82][83][84]; however, based on our in vitro results, it was not possible to predict response to EGFR inhibition based on PTEN expression. Other major causes of EGFR resistance may be explained by the activation of bypass signalling pathways such as MET [22,52]. It is therefore noteworthy that MET signalling was activated in the most resistant cell line, U-CH2, and that a combination of the MET-inhibitor crizotinib and the EGFR-inhibitor sapitinib exhibited a synergistic effect on cell kill in this cell line. The absence of gene amplification and a MET mutation in this cell line leaves the mechanism of activation unanswered, although this has not been studied exhaustively, as MET can be activated by various other mechanisms, such as crosstalk with other receptor tyrosine kinases [85]. As there are numerous resistance mechanisms to EGFR TKIs, many of which remain unexplained even in common cancers [22], it was beyond the scope of this project to pursue this further.
The collective data from this study show that EGFR inhibitors represent the group of compounds within our extensive screen that were most effective against chordoma cell growth. There have been reports that other therapeutic agents have been found to be active against chordoma but, ultimately, whether some patients with chordoma benefit from EGFR inhibitors alone or in combination with these other agents is likely only to be resolved in a clinical trial [3,14,29]. We propose that such a study should involve in-depth biological studies of the tumour samples pre-and post-treatment, with the aim of explaining the mechanism by which some chordomas are primarily resistant or develop secondary resistance to EGFR inhibitors. the OeAD (2013). Mette Jorgensen was supported by a Cancer Research UK clinical research fellowship. Nischalan Pillay is a recipient of a clinician scientist fellowship from Cancer Research UK. Support was provided to Adrienne Flanagan and Sandra Strauss by the NIHR, the University College London Hospitals Biomedical Research Centre, and the Cancer Research UK University College London Experimental Cancer Medicine Centre. The work could not have been done without the existence of the Stanmore biobank, a satellite of the UCL Biobank for Health and Disease which is supported by the Research and Development Department of the Royal National Orthopaedic Hospital.
The Chordoma Foundation (CF) provided support with administration and cell line validation. Preclinical studies were conducted through the Chordoma Foundation Drug Screening Pipeline. We are grateful to GlaxoSmithKline (GSK) for the donation of the Published Kinase Inhibitor Set (PKIS) compounds used in these studies. We would like to thank the scientists at START for their help in conducting the in vivo experiments and Patty Cogswell (CF) for providing a statement on the methods employed. Staff members, particularly Mathew Rushbrook (Cancer Research Technology Discovery Laboratories) and Ariadna Mendoza-Naranjo (UCL Cancer Institute), provided technical support and reagents. Franz Quehenberger and Beate Rinner (Medical University of Graz) advised on statistical analysis and provided the cell line MUG-Chor1, respectively. David Allen (UCL Advanced Diagnostics) helped with the creation of the IHC images. Paolo Salomoni and Aikaterini Lampada (UCL Cancer Institute) provided scientific and technical advice. We are grateful to all clinicians who cared for these patients, and for patients who generously donated their tissue for research.

Author contributions
AMF initiated and supervised the project; SS, MB, LC and MJ conducted the experiments; MB and SS outlined the screening cascade; MB conducted the data analysis; DHD and WJZ provided the GSK compounds and biochemical selectivity data and advised on data analysis and data preparation for publication; FT overlooked and advised on compound screening and data analysis; JAS was responsible for compound plate preparation; HY performed FISH analysis; NG performed immunohistochemistry; APL and NP supported data and NGS mutation analysis; APL conducted the MSigDB enrichment analysis; SjS advised on the clinical relevance of the findings; AL provided organizational support and clinical advice; SB and PM established the U-CH7 primary culture; FA, PM and RT provided clinicopatholog-   Table S1. STR profiles of the cell lines used in the experiments Table S2. Summary of compounds included in the single-concentration focused compound screen (n = 1097) Table S3. Biochemical selectivity data for EGFR, ERBB2 and ERBB4 of selected EGFR/ERBB inhibitors (n = 14) Table S4. Antibodies and conditions used for western blot analysis Table S5. Inhibitory hit rates across all compound libraries included in the single-concentration focused compound screen Table S6. Enrichment in pathways of targets of non-EGFR hit compounds. Table S7. FISH data of the cell lines included in the study