Abstract
Purpose: Introduced in 1987, platinum-based chemotherapy remains standard of care for small cell lung cancer (SCLC), a most aggressive, recalcitrant tumor. Prominent barriers to progress are paucity of tumor tissue to identify drug targets and patient-relevant models to interrogate novel therapies. Following our development of circulating tumor cell patient–derived explants (CDX) as models that faithfully mirror patient disease, here we exploit CDX to examine new therapeutic options for SCLC.
Experimental Design: We investigated the efficacy of the PARP inhibitor olaparib alone or in combination with the WEE1 kinase inhibitor AZD1775 in 10 phenotypically distinct SCLC CDX in vivo and/or ex vivo. These CDX represent chemosensitive and chemorefractory disease including the first reported paired CDX generated longitudinally before treatment and upon disease progression.
Results: There was a heterogeneous depth and duration of response to olaparib/AZD1775 that diminished when tested at disease progression. However, efficacy of this combination consistently exceeded that of cisplatin/etoposide, with cures in one CDX model. Genomic and protein analyses revealed defects in homologous recombination repair genes and oncogenes that induce replication stress (such as MYC family members), predisposed CDX to combined olaparib/AZD1775 sensitivity, although universal predictors of response were not noted.
Conclusions: These preclinical data provide a strong rationale to trial this combination in the clinic informed by prevalent, readily accessed circulating tumor cell–based biomarkers. New therapies will be evaluated in SCLC patients after first-line chemotherapy, and our data suggest that the combination of olaparib/AZD1775 should be used as early as possible and before disease relapse. Clin Cancer Res; 24(20); 5153–64. ©2018 AACR.
Translational Relevance
The clinical management of small cell lung cancer (SCLC) has remained largely unchanged for over 30 years due in part to the lack of model systems that predict clinical outcomes. SCLC prognosis remains dismal. Our recent introduction of SCLC circulating tumor cell–derived explant models (CDX) that can be generated from a simple patient blood draw before treatment and again after treatment upon disease relapse offers research opportunities to test new therapies and explore predictive biomarkers. Here, we report promising data in multiple CDX models that support a combination of PARP and WEE1 inhibitors in SCLC that is superior in efficacy to standard-of-care chemotherapy. We also identify a “super-responder” CDX model that provides molecular insights into strategies for patient selection for this drug combination.
Introduction
Small cell lung cancer (SCLC) is an aggressive and recalcitrant neuroendocrine tumor which metastasizes early (1). The 5-year survival rate for SCLC has been less than 5% for over 30 years, despite multiple phase I clinical trials with a variety of molecularly targeted agents. Although platinum-based therapies yield objective response rates in the majority of patients, this benefit is usually transient with relapse frequently occurring in less than 1 year from initial diagnosis (2). Recent large-scale genome sequencing efforts have failed to identify recurrent actionable mutations in SCLC, and the failure to improve survival rates has largely been attributed to the fact that few patients undergo surgery, and thus the amount of tissue available for research is limited, and tumor biopsies at disease relapse after chemotherapy are rarely obtained (1).
We recently reported the generation of circulating tumor cell (CTC)-derived explant (CDX) models in which CTCs from extensive stage SCLC patients are grown subcutaneously in immunocompromised mice (3). CDX models provide additional material to examine the biology of SCLC and serve as a robust and relevant preclinical pharmacology platform to examine the efficacy of new potential therapies. Importantly, the in vivo response to standard-of-care chemotherapy in CDX models closely resembles that of the corresponding donor patient, suggesting that preclinical results for experimental therapies may be more robust for translation to the clinic than conventional cell line–based xenografts.
Proteomic profiling in lung cancer cell lines revealed elevated levels of PARP specifically in SCLC (4). PARP is recruited to DNA single-strand breaks (SSB) where it PARylates multiple substrates including histones to facilitate the relaxation of chromatin, as well as itself. The automodification of PARP is required for its dissociation from DNA and subsequent repair. The mechanism of action for PARP inhibitors with single-agent activity has been proposed to involve the trapping of PARP onto the DNA SSBs preventing their repair and generating a potential block for cellular DNA replication (5, 6). An important consequence of this and the proposed basis of monotherapy activity is the generation of DNA double-strand breaks (DSB) that would normally be repaired by the homologous recombination repair (HRR) pathway. In cancers with HRR deficiency, PARP trapping will result in significant genomic instability until it is no longer sustainable and tumor cell death results (7). Although deficiencies in the breast and ovarian cancer–associated genes BRCA1 and BRCA2 are the most investigated examples of defective components of HRR that lead to increased PARP inhibitor activity, mutations in other HRR pathways genes, including ATM, ATR, and PALB2, can also convey sensitivity to PARP inhibitors (8).
WEE1 is a kinase that regulates both S-phase and G2–M progression by phosphorylating and inhibiting the cell-cycle–dependent kinases CDK2 and CDK1, respectively. WEE1 control of CDK2 activity in S-phase is an important component of the replication stress response, whereas the control of CDK1 and the G2–M checkpoint is critical for ensuring proper DNA repair before initiating cell division (9). Initially, it was hypothesized that cells deficient for the tumor suppressor p53 were hyper-reliant on WEE1 function and the G2–M checkpoint, and that efficacy of the WEE1 inhibitors could be linked to TP53 status (10). Subsequent experiments revealed that this correlation was not universally true, although WEE1 inhibitor activity is still likely linked to G1–S checkpoint deficiencies in one form or another (11). Recent reports suggest that cells overexpressing potent oncogenes, such as MYC or CCNE1, that induce replication stress are also more sensitive to WEE1 inhibition, but again does not appear to be a universal prerequisite for response (9).
Given the potential for PARP inhibitors to induce S-phase damage and a subsequent dependency on Wee1 S-phase and G2–M checkpoint activities, coupled with the high mutagenic burden and proliferative rate of SCLC, we examined the efficacy of the PARP inhibitor olaparib and the WEE1 inhibitor AZD1775 in our SCLC CDX models. The most important advantage of the CDX approach is the ability to assess biology and therapy responses not only in patient-derived models generated at baseline prior to any treatment, but also in CDX derived from the same patient at disease progression after treatment. Here, we examine the novel combination of olaparib and AZD1775 for the first time in “paired” CDX, providing a clinically relevant test-bed for the introduction of new therapeutic options following first-line chemotherapy.
Our studies reveal that SCLC CDX tumors exhibit a range of sensitivities to drugs targeting PARP and WEE1 as single agents, but with clear indications of enhanced efficacy when given in combination. Moreover, these studies have revealed the presence of previously unappreciated DNA damage response deficiencies in a subset of SCLC tumors. These results therefore provide a strong rationale for the clinical development of AZD1775 in combination with olaparib or other DNA-damaging agents, as well as candidate biomarkers to explore in tandem toward a precision medicine approach for this disease.
Materials and Methods
Study design
The primary goal of this study was to assess the efficacy of olaparib and AZD1775 monotherapies, as well as the combination, in mouse models of SCLC. Sample sizes (n = 7–11 mice per cohort) were determined to detect statistically a 2-fold difference in tumor response between the various treatments. Once tumors reached enrollment size, animals were allocated to different cohorts via stratified randomization in which attempts were made to evenly distribute initial tumor volume sizes. Data collection was stopped at ethical endpoints, including deterioration in health or when tumors reached ≥4 times the initial tumor volume at enrollment. If animals became ill within 3 days of enrollment, they were excluded from the experiment, regardless of which treatment they received. No other data were excluded.
CDX establishment
CDX models were generated as previously described (3). Briefly, 10 mL of peripheral blood was collected from SCLC patients into EDTA tubes in accordance with the ethically approved CHEMORES protocol (07/H1014/96), processed via RosetteSep CTC Enrichment Cocktail (15167; Stemcell Technologies), and resuspended in 100 μL of a 1:1 solution of HITES medium and Matrigel (354234, BD Biosciences). Cells were then subcutaneously implanted into flanks of 8- to 16-week-old female NOD.Cg-PrkdcscidIl2rgtm1Wjl/SzJ (NSG) mice (Charles River). All procedures were carried out in accordance with Home Office Regulations (UK) and the UK Coordinating Committee on Cancer Research guidelines and by approved protocols (Home Office Project license nos. 40–3306/70-8252 and Cancer Research UK Manchester Institute Animal Welfare and Ethical Review Advisory Board). In a parallel patient blood sample, CTC burden was quantified via CellSearch according to the manufacturer's instructions.
In vivo pharmacology studies
Therapeutic studies were carried out essentially as previously described (3). A total of 100,000 viable CDX cells in 100 μL 1:1 RPMI:matrigel were injected subcutaneously into the right flank of procedure/treatment-naïve, 8- to 10-week-old (20–25 g) female C.B-17/lcrHsd-PrkdcscidLystbg-J mice (Envigo). When tumors reached 150 to 250 mm3, they were allocated via stratified randomization into cohorts to be treated each morning with vehicle(s), olaparib (50 mg kg−1 unless indicated otherwise), AZD1775 (120 mg kg−1), the combination, or topotecan using tolerated dosing regimens previously shown to optimally achieve target inhibition. Olaparib (AstraZeneca) was formulated at 10 mg mL−1 in 10% DMSO/30% KLEPTOSE HPβCD, and AZD1775 (AstraZeneca) was formulated at 12 mg mL−1 in 0.5% methylcellulose. Compounds were orally dosed at 5 mL kg−1 daily (olaparib) or 10 mL kg−1 on a 5-on, 2-off (AZD1775) schedule for 21 days. On days when both drugs were administered, olaparib was administered 1 hour after AZD1775. Topotecan (The Christie NHS Foundation Trust) was administered at 3 mg kg−1 daily for 3 days via intraperitoneal injection. Cisplatin and etoposide (The Christie NHS Foundation Trust) were administered as previously described (3), and these studies were performed separately from studies on olaparib and AZD1775. Mice were observed for a period of time after dose to ensure no adverse effects were seen. Tumors were measured twice a week by caliper until they reached 4 times initial tumor volume (4 × ITV), as determined by the formula V = (L × W2)/2, or until animal health deteriorated. For pharmacokinetic and pharmacodynamic studies, mice were sacrificed 2 or 24 hours after a single dose of indicated drugs was administered and relevant tissues were collected. Statistical assessment of tumor response was performed as described (12). Tumor growth delay (TGD) measures the normalized time to quadrupling of initial tumor growth, and therefore a higher TGD value is indicative of more robust tumor growth inhibition.
Pharmacokinetic analysis
Note that 25 μL plasma was prepared using an appropriate dilution factor, and compared against an 11-point standard calibration curve (1–10,000 nmol/L) prepared in DMSO and spiked into blank plasma. Acetonitrile (100 μL) was added with the internal standard, followed by centrifugation, and supernatants were then diluted 1:7 in water and analyzed via UPLC-MS/MS using a Xevo TQ-S mass spectrometer (Waters) and the system and optimization parameters listed (Supplementary Tables S1 and S2).
Ex vivo studies
CDX tumors were dissociated as previously described (13), and cells were grown in HITES medium supplemented with 5 μmol/L Y-27632 (Selleckchem) and 2.5% FBS. CDX-derived cells were seeded into 96-well plates, incubated for 48 hours, and treated with increasing concentrations of the compound(s) of interest. Seven days after treatment, cell viability was assessed via Cell-Titer Glo 3D assay (Promega). Single dose response curves were computed with the open-source CRUK software Combenefit (14). Combinatorial activity was assessed using the Loewe additivity model (15) and computed using the R package synergyfinder (16). For radiosensitization assays, cells were irradiated with indicated doses using a Gulmay Bipolar 320 kV X-ray source. After irradiation, cells were plated into fresh media, and viability was measured after 5 days.
RAD51 focus assay
CDX-derived cells were irradiated with 10 Gy and incubated for the indicated times. After incubation, cells were cytospun at a concentration of 150,000 cells per slides, fixed in 4% PFA for 20 minutes at room temperature, permeabilized with 0.2% triton X-100 (Sigma), and washed with 10% FBS 0.2% Tween20 in PBS. Samples were then blocked with avidin/biotin-blocking solution (VECTOR laboratories, SP-2001) and 10% goat serum (Dako). Sections were then incubated with primary antibodies (Supplementary Table S3) overnight at 4°C overnight, washed, and incubated with secondary antibodies for 1 hour at room temperature. Slides were mounted in ProLong DAPI (Life Technologies) and left dry in the dark. Images were acquired with a Leica gated Stimulated Emission Depletion microscope, and cells were scored as HR-proficient if there were ≥5 foci per nucleus.
Immunohistochemistry
Formalin-fixed paraffin-embedded tumor sections (4 μm) were stained for the indicated antigens using the BondMax autostainer and Bond polymer refine detection system (Leica Biosystems), Venatana Discovery Ultra, or I6000 BioGenex autostainer and Dako EnVision system HRP (Dako; Supplementary Table S3). For these antigens, digital images of whole tissue sections were acquired using a Leica SCN400 histology scanner (Leica Microsystems). Images were analyzed using Definiens Developer XD and the Tissue Studio Portal version 4.4 (Definiens AG).
Immunoblot analyses
Flash-frozen CDX tumors were homogenized in Fastprep tubes with matrix A using a TissueLyser LT (Qiagen) in ice-cold lysis buffer (10 mmol/L HEPES, pH 7.1, 50 mmol/L NaCl, 0.3 mol/L sucrose, 0.1 mmol/L EDTA, 0.5% Triton X-100, 1 mmol/L DTT; or 20 mmol/L Tris, 137 mmol/L NaCl, 10% Glycerol, 50 mmol/L sodium fluoride, 1 mmol/L activated sodium orthovanadate, 1% SDS, 1% NP40 substitute) in the presence of Protease Inhibitor Cocktail and Phosphatase Inhibitor Cocktail I and III (Sigma-Aldrich). Protein concentration was determined by BCA protein assay reagent kit (Thermo Scientific). Protein extracts were resuspended in NuPAGE LDS sample buffer 4x (Invitrogen) and 10x NuPAGE sample reducing agent (Invitrogen). Total protein was resolved on a NuPAGE 4%–12% Bis-Tris 1.0 mm gel (Invitrogen). After transfer onto PVDF or nitrocellulose membranes, blots were incubated with the appropriate primary antibody (Supplementary Table S3) and the corresponding horseradish peroxidase–coupled secondary IgG. Detection was obtained with the Supersignal West Dura extended duration substrate (ThermoFisher) on a ChemiGenius Imaging System (Syngene) or film.
PAR ELISA
Flash-frozen CDX tumor tissue was homogenized for 3 × 30 seconds at speed 6.0 in Fastprep tubes with matrix A using ice-cold lysis buffer (20 mmol/L Tris, 137 mmol/L NaCl, 10% Glycerol, 50 mmol/L sodium, 1 mmol/L activated sodium orthovanadate, 1% SDS, 1% NP40 substitute) in the presence of protease inhibitor cocktail (Roche) and phosphatase inhibitor cocktails 2 and 3 (Sigma-Aldrich). Lysates were sonicated, incubated on ice for 30 minutes, centrifuged, and supernatants were boiled for 5 minutes. They were then cooled on ice for 1 minute, centrifuged, and protein concentration was estimated using a Pierce BCA assay. Lysate (2 μg) was used to quantify total PAR levels according to the manufacturer's recommendations (Trevigen). Samples were quantified on a Tecan Safire II microplate reader (Tecan), and absolute values were determined via interpolation from a standard curve.
RNA-seq
Total RNA was extracted from three independent RNAlater (Qiagen)-treated tumors from each model with RNeasy mini kits (Qiagen) and quantified by Qubit (Thermo Fisher). RNA with an RNA integrity number > 8 (Agilent 2100 Bioanalyzer) was used to generate libraries with SureSelect poly A samples (Agilent) and sequenced on the NextSeq 500 using 75-bp paired-end reads. RNA-seq data were aligned to Human GRCh38 and Mouse GRCm38 assembly using Mapsplice (version 2.1.6). Xenograft data were filtered via a novel algorithm to distinguish human and mouse reads (17). The filtered reads are then used to generate counts data using Rsubread package (version 1.16.1) with Ensembl 77 GTF file in R. The counts were converted into RPKM values using edgeR (version 3.10.5).
Whole-exome sequencing
Genomic DNA was extracted from a representative flash-frozen CDX tumor from each model with DNeasy Blood and Tissue kits (Qiagen) and quantified by Qubit. Note that 150 bp paired-end reads were obtained from samples assessed on a NextSeq 500 High run (Illumina). The whole-exome sequencing (WES) data were aligned to Human GRCh37 and Mouse GRCm37 assembly with bwa-mem (version 0.7.12). Deduplication, realignment, and recalibration were performed on the aligned data using Picard (version 1.96) and GATK tools (version 3.3), as stated in the GATK best practices. The reads aligning to mouse genome were removed using an in-house developed algorithm (17). Somatic mutation calling was done on the filtered reads using Mutect (version 1.1.7), and the mutation calls were annotated using Ensembl Variant Effect Predictor.
Statistical analysis
All statistical analyses were performed on Prism v7.04 (GraphPad) using one-way ANOVA or Student t test for multiple or single pairwise comparisons, respectively. Single and double asterisks denote P < 0.05 and P < 0.005, respectively. The absence of an asterisk means that there is no statistically significant difference between groups.
Results
To examine the efficacy of olaparib and AZD1775 in SCLC, we first treated two previously characterized CDX models representative of chemosensitive and chemorefractory patient donors (3). The chemosensitive CDX3 and chemoresistant CDX4 models were generated from baseline blood samples taken from a patient that lived for 9.7 months and 0.9 months, respectively, following treatment with platinum-based therapies. Treatment of mice bearing CDX3 tumors with either olaparib or AZD1775 monotherapy led to complete or modest regression, with tumors rapidly recurring after treatments ended (Fig. 1; Supplementary Fig. S1A). Remarkably, combination therapy in CDX3 promoted complete and durable regressions with 5 of 7 mice remaining tumor-free for over 1 year. Conversely, CDX4, a model completely resistant to cisplatin/etoposide, showed no response to olaparib, whereas treatment with either AZD1775 monotherapy or the combination led to disease stabilization during the treatment period (Fig. 1; Supplementary Fig. S1B).
Combination treatment with olaparib/AZD1775 is superior to cisplatin/etoposide. A, The cohort average best response (±SD) is shown for cisplatin/etoposide (white), AZD1775 (red), olaparib (green), and AZD1775/olaparib (blue). B, The cohort average benefit (±SD) of each treatment over vehicle is shown for each treatment. Cohort sizes are ≥7. *, P < 0.05; **, P < 0.005 vs. cisplatin/etoposide treatment.
Combination treatment with olaparib/AZD1775 is superior to cisplatin/etoposide. A, The cohort average best response (±SD) is shown for cisplatin/etoposide (white), AZD1775 (red), olaparib (green), and AZD1775/olaparib (blue). B, The cohort average benefit (±SD) of each treatment over vehicle is shown for each treatment. Cohort sizes are ≥7. *, P < 0.05; **, P < 0.005 vs. cisplatin/etoposide treatment.
Topotecan, a DNA-damaging agent that is the only approved second-line therapy for SCLC, was more effective than cisplatin/etoposide against CDX4, eliciting a median 66% tumor regression (Supplementary Fig. S1E). Furthermore, combination of topotecan with AZD1775 yielded impressive tumor regression in this model, with median 83% tumor regression. These data indicate that AZD1775 can cooperate with an effective DNA-damaging agent even in a platinum-resistant model.
The majority of SCLC patients present with chemosensitive disease followed by rapid acquisition of chemoresistance and disease progression. To examine the efficacy of the olaparib/AZD1775 combination in patients with acquired chemoresistance, we utilized a pair of CDX models, CDX8 and CDX8p, which were derived from blood samples from the same patient taken at baseline and upon disease progression after platinum treatment (Supplementary Fig. S2A). Treatment of CDX8 tumors with cisplatin/etoposide induced a partial response (median 47% regression) and rapid relapse (median TGD = 1.9), whereas mice bearing CDX8p received little benefit from cisplatin/etoposide treatment (median 15% growth and TGD = 1.3), consistent with the outcome of chemotherapy in the donor patient (Fig. 1; Supplementary Fig. S2B and S2C). When these models were treated with olaparib or AZD1775, CDX8 exhibited relatively stable disease during the treatment period, whereas CDX8p tumors grew at the same rate as the vehicle control cohort (Fig. 1; Supplementary Fig. S1C and S1D). Similar to that seen in CDX3, the combination exhibited synergistic activity against CDX8 with many complete regressions. Although CDX8 tumors recurred more quickly than CDX3 tumors (Fig. 1B), olaparib/AZD1775 yielded a more durable response than cisplatin/etoposide (median TGD = 1.9; Fig. 1). Consistent with reports of reduced olaparib efficacy against platinum-resistant ovarian cancer (18), treatment with olaparib alone or in combination with AZD1775 achieved only TGD rather than regression in CDX8p (Fig. 1; Supplementary Fig. S1D). Taken together, these results indicate that although the efficacy of olaparib/AZD1775 was model dependent, the activity of the combination is consistently superior to that achieved with conventional cisplatin and etoposide in chemosensitive models. Furthermore, the combination of AZD1775 and an active DNA-damaging agent has activity in a chemoresistant model.
As platinum-based regimens are initially quite effective in treating SCLC, it is unlikely that olaparib/AZD1775 or any other new agent will be used clinically in treatment-naïve patients. We therefore sought to examine the efficacy of olaparib/AZD1775 in a second-line setting following initial treatment with cisplatin/etoposide. Toward this end, we treated mice bearing CDX3 tumors with a single cycle of cisplatin/etoposide to induce an initial tumor response. After a 3-week drug holiday during which tumors began to regrow, mice were randomized to receive vehicle, olaparib, AZD1775, or the combination. In this setting, only 1 of 12 combination-treated mice remained tumor-free, and the median time to initial tumor volume was 101 days following completion of dosing (Supplementary Fig. S3), suggesting that this combination is still active after initial treatment with cisplatin/etoposide but may prove most beneficial as a first-line option if the proper patient cohort could be identified.
In vivo pharmacology experiments are both time consuming and expensive, so we sought to use an in vitro platform to expedite drug screening as previously reported for colorectal and breast cancers (19, 20). Using an ex vivo SCLC CDX culture system that faithfully recapitulates in vivo drug response in CDX models (21), we investigated the in vitro efficacy of olaparib, AZD1775, and the combination. Both monotherapies as well as the combination correlated with in vivo response, with evidence of synergy in CDX3 and CDX8, but not CDX4 or CDX8p (Fig. 2A–D; Supplementary Fig. S4). We therefore decided to screen additional CDX models with varying sensitivity to platinum, derived from baseline and relapse patients (Table 1). Examination of ex vivo cultures derived from CDX2, CDX3p, CDX10, CDX14p, CDX15p, and CDX15p2 revealed a range of sensitivity to cisplatin and olaparib monotherapies, whereas most models were relatively sensitive to AZD1775 with submicromolar GI50s. Similarly, the majority of models were sensitive to olaparib/AZD1775, although CDX4 and CDX8p were among the most resistant in vitro (Fig. 2E). These results from a broader panel of CDX models tested in vitro support our findings that olaparib/AZD1775 is an effective combination and may represent a promising treatment for SCLC.
AZD1775 and olaparib are synergistic in vitro. CDX3 (A), CDX4 (B), CDX8 (C), or CDX8p (D) ex vivo cultures were treated with AZD1775 and olaparib for 1 week, and relative cell viability was assessed. One representative experiment of at least three is shown. For the combinatorial index, red indicates synergy, and green indicates antagonism between drugs. E, GI50 values for various CDX ex vivo cultures treated with cisplatin, etoposide, AZD1775, or olaparib are shown after 1 week of drug treatment. For the combination, we have shown the GI50 values for olaparib at 150 nmol/L AZD1775.
AZD1775 and olaparib are synergistic in vitro. CDX3 (A), CDX4 (B), CDX8 (C), or CDX8p (D) ex vivo cultures were treated with AZD1775 and olaparib for 1 week, and relative cell viability was assessed. One representative experiment of at least three is shown. For the combinatorial index, red indicates synergy, and green indicates antagonism between drugs. E, GI50 values for various CDX ex vivo cultures treated with cisplatin, etoposide, AZD1775, or olaparib are shown after 1 week of drug treatment. For the combination, we have shown the GI50 values for olaparib at 150 nmol/L AZD1775.
Response to standard-of-care for each patient and derived CDX, with the corresponding CTC count at baseline
Model . | Pt treatment . | Tumor responsea . | Chemosensitivity . | Pt survival (days)a . | CellSearch countb . |
---|---|---|---|---|---|
2 | Carbo/etop | PR | Refractory | 108 | 1,625 |
3 | Carbo/etop | PR | Sensitive | 295 | 507 |
3p | Relapse | 295 | 463 | ||
4 | Carboplatin | PD | Refractory | 27 | 1,375 |
8 | Carboplatin | PR | Refractory | 181 | 388 |
8p | Relapse | 181 | 1,796 | ||
10 | Carbo/etop | PR | Sensitive | 317 | 160 |
14p | Relapse | 227 | 362 | ||
15p | Relapse | 556 | 361 | ||
15p2 | Relapse | 556 | 165 |
Model . | Pt treatment . | Tumor responsea . | Chemosensitivity . | Pt survival (days)a . | CellSearch countb . |
---|---|---|---|---|---|
2 | Carbo/etop | PR | Refractory | 108 | 1,625 |
3 | Carbo/etop | PR | Sensitive | 295 | 507 |
3p | Relapse | 295 | 463 | ||
4 | Carboplatin | PD | Refractory | 27 | 1,375 |
8 | Carboplatin | PR | Refractory | 181 | 388 |
8p | Relapse | 181 | 1,796 | ||
10 | Carbo/etop | PR | Sensitive | 317 | 160 |
14p | Relapse | 227 | 362 | ||
15p | Relapse | 556 | 361 | ||
15p2 | Relapse | 556 | 165 |
aSurvival is defined as the time from diagnosis until death.
bThe number of EpCAM+/CK+ CTCs in a parallel 7.5 mL sample taken from the same patient blood that yielded the corresponding CDX.
It has previously been reported that olaparib is a CYP3A substrate and that AZD1775 is a CYP3A4 inhibitor (22); therefore, it was possible that the combination effect seen in some CDX models could be attributed to a drug–drug interaction in which elevated levels of olaparib result from inhibition of CYP3A by AZD1775. To examine this possibility, we administered 50 mg kg−1 olaparib (olaparib low) both alone and in combination with AZD1775 as previously described, or 100 mg kg−1 olaparib alone (olaparib high) to CDX3-bearing mice. Increasing the dose of olaparib to 100 mg kg−1 had no significant effect on antitumor efficacy against CDX3 (Fig. 3A), and pharmacokinetic analysis of olaparib and AZD1775 indicates that mice dosed with the combination are not exposed to plasma drug concentrations significantly different to mice receiving monotherapies (Fig. 3B and C). Therefore, the efficacy of the olaparib/AZD1775 combination is unlikely to be due to altered olaparib exposure.
Pharmacokinetic analyses. A, Mice were treated with 100 mg kg−1 (dark green) or 50 mg kg−1 (dashed green) olaparib, or with a combination of low-dose olaparib and AZD1775 (blue) for 21 days, and tumor volumes were measured biweekly. Individual relative tumor volumes are depicted. Tumor lysates from cohorts of mice bearing different CDX models treated with AZD1775 (red), olaparib (green), or the combination (blue) were analyzed and quantified for AZD1775 (B) or olaparib (C) at 2 hours after a single dose. Dotted lines represent the reliable limits of detection. Cohorts of ≥4 tumors were used.
Pharmacokinetic analyses. A, Mice were treated with 100 mg kg−1 (dark green) or 50 mg kg−1 (dashed green) olaparib, or with a combination of low-dose olaparib and AZD1775 (blue) for 21 days, and tumor volumes were measured biweekly. Individual relative tumor volumes are depicted. Tumor lysates from cohorts of mice bearing different CDX models treated with AZD1775 (red), olaparib (green), or the combination (blue) were analyzed and quantified for AZD1775 (B) or olaparib (C) at 2 hours after a single dose. Dotted lines represent the reliable limits of detection. Cohorts of ≥4 tumors were used.
To examine the molecular mechanism of olaparib and AZD1775 efficacy, we treated parallel cohorts of mice with a single dose of drug and examined intratumor signal transduction pathways and cellular behavior. WEE1 kinase regulates G2–M progression by directly phosphorylating and inhibiting the cell-cycle kinase CDK1, and this has been used both preclinically and clinically as a pharmacodynamic biomarker of WEE1 inhibition (9). Consistent with WEE1 inhibition, phospho-CDK1 levels were reduced, and the number of mitotic cells was increased in cohorts treated with AZD1775 either alone or in combination with olaparib (Fig. 4A and C; Supplementary Fig. S5). Similarly, global PAR levels were decreased in all cohorts treated with olaparib (Fig. 4B). Together, these data indicate that the different levels of efficacy in different CDX models cannot be attributed to differential target inhibition. Examination of cleaved caspase 3 levels revealed that increased levels of apoptosis correlated with antitumor efficacy (Fig. 4D; Supplementary Fig. S5). Because treatment with both olaparib and AZD1775 promoted increased DNA damage, we investigated several pathways associated with the DNA damage response.
Pharmacodynamic analyses. Tumor lysates from cohorts of mice bearing different CDX models treated with a single dose of vehicle (white), AZD1775 (red), olaparib (green), or the combination (blue) were analyzed and quantified. Cohorts of ≥ 4 tumors were used. A, pCDK1 levels were measured by immunoblotting 2 hours after a single dose and normalized to both total CDK1 and vinculin in each sample. B, Total PAR levels were measured by ELISA 2 hours after a single dose. The dashed line indicates the limit of detection (2 pg/mL). Proof-of-concept and proof-of-mechanism biomarkers were measured by quantitative immunohistochemistry for pHH3 at 2 hours (C), CC3 at 24 hours (D), pCHK1 at 2 hours (E), and γH2AX at 24 hours (F) after dosing. *, P < 0.05 and **, P < 0.005.
Pharmacodynamic analyses. Tumor lysates from cohorts of mice bearing different CDX models treated with a single dose of vehicle (white), AZD1775 (red), olaparib (green), or the combination (blue) were analyzed and quantified. Cohorts of ≥ 4 tumors were used. A, pCDK1 levels were measured by immunoblotting 2 hours after a single dose and normalized to both total CDK1 and vinculin in each sample. B, Total PAR levels were measured by ELISA 2 hours after a single dose. The dashed line indicates the limit of detection (2 pg/mL). Proof-of-concept and proof-of-mechanism biomarkers were measured by quantitative immunohistochemistry for pHH3 at 2 hours (C), CC3 at 24 hours (D), pCHK1 at 2 hours (E), and γH2AX at 24 hours (F) after dosing. *, P < 0.05 and **, P < 0.005.
CHK1 phosphorylation by ATR is induced by the formation of regions of single-strand DNA (ssDNA) either resulting from the processing of DNA DSBs or stalled DNA polymerase during DNA synthesis due to replicative stress (9). The extended ssDNA is coated by replication protein A (RPA) which binds ATR and associated proteins that initiate the replication stress response that includes the phosphorylation and activation of CHK1. The inability to resolve replication stress–induced fork stalling can lead to replication fork collapse, the generation of a single-ended DSBs, and ATM-induced H2AX phosphorylation. Consistent with previous reports, treatment with AZD1775 promoted potent CHK1 phosphorylation, indicative of WEE1 inhibition increasing replication stress (23). However, this effect was similar in all models examined and did not correlate with efficacy (Fig. 4E; Supplementary Fig. S1). Conversely, induction of γH2AX more closely correlated with in vivo efficacy, and highest levels were seen when AZD1775 was combined with olaparib in CDX3 and CDX8, the two most sensitive models (Fig. 4F; Supplementary Fig. S1).
The BRCA1 and BRCA2 gene products are key DDR proteins that play different but pivotal roles in DNA DSB repair (DSBR). Both proteins have been shown to play key roles in the replication stress response as well through replication fork stabilization and restart, and deficiencies in either lead to PARP inhibitor sensitivity (24). Given this association, we were interested to know the DSBR status in our CDX models. BRCA and broader HRR deficiencies have been identified via a variety of mechanisms including transcriptional or mutational impairment of genes that regulate HRR, the accumulation of “genomic scars,” or functional assays such as DNA damage–induced RAD51 focus formation (24). Analysis of RNA-seq data on CDX models did not reveal an obvious HRR mutation signature that correlated with sensitivity, nor did WES reveal an elevated mutation burden associated with impaired DSBR (Supplementary Fig. S6A and S6B). Despite the absence of an HRR mutation signature, CDX3 is exquisitely radiosensitive compared with CDX4, indicating that CDX3 is potentially deficient in DSBR (Fig. 5A). Functional assessment of HRR status via examination of ionizing radiation–induced RAD51 focus formation revealed that CDX3 displayed signs of HRR deficiency, suggesting this model may harbor a mutation in a gene that mediates HRR (Fig. 5B). PALB2 is a BRCA2-associated DDR protein required for effective HRR (9). WES of CDX3 revealed a somatic mutation in PALB2 that resulted in a stop-gain at E178 leading to premature truncation of the protein (Fig. 5C and D). Analysis of RNA-seq data indicates that all PALB2 reads harbor this mutation, suggesting a complete loss of the wild-type allele (Supplementary Fig. S6C). Therefore, it is highly likely that the exquisite sensitivity of CDX3 to both platinum and olaparib can be attributed to the PALB2 mutation.
Deficient DSB repair in CDX3. A, CDX3, CDX4, CDX8, and CDX8p ex vivo cultures were irradiated at indicated doses in triplicate, and relative cell viability was assessed 5 days later. One of three representative experiments is shown. B, At indicated times following irradiation, CDX3 and CDX4 cultures were assessed for RAD51 focus formation. Geminin and DAPI staining was used to identify cells in G2, and these results were quantified by a semiautomated computer algorithm. The dotted line indicates a threshold above which cells are considered positive for damage-induced focus formation. **, P < 0.005. C, Sanger sequencing identified the E178* PALB2 mutation in CDX3, but not CDX4. D, Schematic of PALB2 with protein–protein interaction domains shown in gray and the E178* mutation shown in red.
Deficient DSB repair in CDX3. A, CDX3, CDX4, CDX8, and CDX8p ex vivo cultures were irradiated at indicated doses in triplicate, and relative cell viability was assessed 5 days later. One of three representative experiments is shown. B, At indicated times following irradiation, CDX3 and CDX4 cultures were assessed for RAD51 focus formation. Geminin and DAPI staining was used to identify cells in G2, and these results were quantified by a semiautomated computer algorithm. The dotted line indicates a threshold above which cells are considered positive for damage-induced focus formation. **, P < 0.005. C, Sanger sequencing identified the E178* PALB2 mutation in CDX3, but not CDX4. D, Schematic of PALB2 with protein–protein interaction domains shown in gray and the E178* mutation shown in red.
Several additional mechanisms of acquired resistance to olaparib have been reported, including activation of PI3K or MET signaling (25, 26), decreased expression of PARP1 or SLFN11 (27–30), and restoration of DNA end resection (31). We did not find any consistent differences in MET or PI3K signaling, suggesting that activation of these kinases is not a recurrent mechanism of olaparib resistance in SCLC (Supplementary Fig. S7A). PARP-trapping is now recognized as the primary mechanism of PARP inhibitor–mediated cytotoxicity, and loss of expression of PARP1 is thus associated with resistance. However, expression analysis did not reveal any alteration in PARP1 expression in either CDX4 or CDX8p (Supplementary Fig. S7B). SLFN11 binds to RPA, and it has been postulated that high SLFN11 levels inhibit HRR by promoting the destabilization of the interaction between RPA and ssDNA (32). Although CDX4, the most olaparib-resistant model, exhibits the lowest SLFN11 levels, this correlation does not extend to CDX8p (Supplementary Fig. S7C). End resection is the process of creating stretches of ssDNA necessary for RAD51-induced strand invasion and subsequent repair via homologous recombination. In the context of functionally relevant BRCA1 mutations, decreased expression of the nonhomologous end joining proteins 53BP1 and REV7 has been reported to result in restoration of end resection activation of HRR and consequently resistance to PARP inhibitors (33, 34). However, neither CDX4 nor CDX8p has decreased expression of either protein compared with CDX3 or CDX8 (Supplementary Fig. S7C). Although CDX8 harbors a C342F mutation in FANCD2, a key player in the Fanconi anemia pathway that includes BRCA, this mutation is still present in CDX8p and, in the absence of any TP53BP1-associated mutation, is unlikely to explain the acquired chemoresistance manifest in CDX8p. Neither WES nor RNA-seq has identified likely candidates of resistance, and we are actively pursuing additional strategies to discover the mechanism of acquired drug resistance in CDX8p.
One proposed predictive biomarker of AZD1775 sensitivity is the absence of a G1–S checkpoint (35). It has been hypothesized that cells with a compromised G1–S checkpoint are more dependent on the G2–M checkpoint, and therefore sensitive to WEE1 inhibition. Given the role for p53 in regulating the G1–S checkpoint, multiple studies have investigated whether p53 mutations are stand-alone biomarkers for AZD1775 sensitivity (10, 36). Although all SCLC models tested harbor deleterious mutations in TP53, chemoresistant models tend to retain relatively high levels of the cyclin-dependent kinase inhibitor p21CIP1, a primary mediator of p53-dependent cell-cycle arrest (Fig. 6A and B). Although treatment does not induce expression of p21CIP1 (Supplementary Fig. S8), p53-independent activation of p21CIP1 may partially mediate resistance to AZD1775 by maintaining the G1–S checkpoint in p53-deficient cells. The majority of SCLC cells also possess mutations in RB1, which regulates entry into S-phase. We therefore investigated whether RB1 status correlates with AZD1775 sensitivity. WES and RNA-seq revealed RB1 mutations in CDX3 (splice mutation at 5′ of exon 17, c.1499-1G>T), CDX4 (87 bp deletion in exon 1 and intron 1, c.99_137+48del), and CDX8/8p (G509*), and none of the models used in this study express RB1 as determined by Western blotting (Fig. 6A). Moreover, there is no consistent correlation between cyclin/CDK expression and sensitivity to AZD1775. This does not rule out the potential importance of deregulation of the G1–S checkpoint but does suggest that, at least in SCLC, regulators of G1–S CDKs alone do not appear to be predictive biomarkers of AZD1775 sensitivity.
Sensitivity to AZD1775 correlates with markers of replication stress. A, Immunoblot analysis of duplicate tumor lysates for indicated proteins. The p53 mutation associated with a given model is indicated. The positive control for RB is the H1694 cell line. B, Representative images and quantification of p21CIP1 IHC from CDX3, CDX4, CDX8, and CDX8p. p21CIP1 levels are plotted against tumor regression values for each model. C, Immunoblot analysis of duplicate tumor lysates for indicated proteins. Corresponding average RPKM values for MYCL and MYC in each model are indicated below immunoblots. MYCN expression was not investigated; however, RPKM values below 5 suggest it is not expressed in the models investigated. D, Immunoblot analysis of duplicate lysates from CDX3 and CDX4 tumors treated with vehicle or AZD1775 for 24 hours. The positive control for MYCL is the H1694 cell line.
Sensitivity to AZD1775 correlates with markers of replication stress. A, Immunoblot analysis of duplicate tumor lysates for indicated proteins. The p53 mutation associated with a given model is indicated. The positive control for RB is the H1694 cell line. B, Representative images and quantification of p21CIP1 IHC from CDX3, CDX4, CDX8, and CDX8p. p21CIP1 levels are plotted against tumor regression values for each model. C, Immunoblot analysis of duplicate tumor lysates for indicated proteins. Corresponding average RPKM values for MYCL and MYC in each model are indicated below immunoblots. MYCN expression was not investigated; however, RPKM values below 5 suggest it is not expressed in the models investigated. D, Immunoblot analysis of duplicate lysates from CDX3 and CDX4 tumors treated with vehicle or AZD1775 for 24 hours. The positive control for MYCL is the H1694 cell line.
Low expression of PKMYT1, a cell-cycle kinase that is functionally redundant to WEE1, can predispose cancer cells to AZD1775 sensitivity by lowering the threshold of kinase inhibition (37). However, although the most sensitive models do have the lowest levels of PKMYT1 RNA, there is no correlation between PKMYT1 levels and CDK1 phosphorylation, suggesting this is unlikely to be a major factor governing AZD1775 sensitivity (Fig. 4A; Supplementary Fig. S9A).
Several studies have reported a link between sensitivity to WEE1 inhibition and replication stress (38). Replication stress is a general term used to describe situations that result in stalled replication fork progression and the generation of extended regions of ssDNA (39). Replication stress can be induced by overexpression of oncogenes such as MYC family members or cyclin E1, which can both deplete nucleotides as a result of deregulated origin firing and promote collisions between the transcription and replication machinery (40). WEE1 plays a key role in the replication stress response by regulating CDK2, and markers of replication stress such as pRPA have been shown to increase upon treatment with a WEE1 inhibitor (41). Immunoblot analysis revealed that CDX3, the model that is most sensitive to AZD1775, expressed high levels of MYCL, cyclin E1, and phospho-RPA (Fig. 6C). Consistent with a recent report that tumors deficient for the histone methyltransferase SETD2 and ribonucleotide reductase subunit RRM2 are synthetic lethal with AZD1775 treatment (42), CDX3 expressed approximately half as much SETD2 and RRM2 compared with CDX4, and treatment with AZD1775 either alone or in combination with olaparib led to modest reductions in RRM2 levels (Supplementary Fig. S9A and S9B) Moreover, AZD1775 treatment further increased pRPA levels in CDX3 but not CDX4 (Fig. 6D). Together, these data suggest that both baseline and treatment-induced replication stress may predict sensitivity to AZD1775.
Discussion
SCLC is a cancer of high unmet need and, despite promising preclinical (43) and early clinical results (44) utilizing a DLL3-targeted antibody–drug conjugate, there have been no improvements in systemic therapy in three decades. Although approximately 100 SCLC trials have been registered on clinicaltrials.gov since 2007, strikingly few involved more than 100 patients, many are single-arm studies, and correlative biomarker assessment is rarely obtained largely due to the difficulties in obtaining serial tumor biopsies. The statement in the US Recalcitrant Cancer Act 2012 that “more knowledge of the biology at different phases of SCLC is needed” can partially be attributed to the lack of relevant models. Indeed, the recently reported “patient-derived tumor encyclopedia” that contains over 1,000 patient-derived xenografts does not contain a single SCLC model (45), and 21 of 26 SCLC patient-derived xenografts that have been published were derived prior to treatment of the donor patient with chemotherapy (43, 46, 47). Our CDX models and notably those obtained both at baseline and at progression after chemotherapy and reported here for the first time offer a new approach to study acquired treatment resistance and test novel therapies in both first- and second-line settings. Utilizing these patient-relevant models derived from liquid biopsy, we demonstrated here that the combination of olaparib and AZD1775 exerted significant activity against a panel of SCLC CDX models that span the spectrum of responses to standard-of-care chemotherapy, with some demonstrating a profound and extended response to this novel combination.
The PARP inhibitor olaparib is currently being investigated in clinical trials; however, the genetic determinants of drug sensitivity remain incompletely defined. Our CDX models offer the opportunity to investigate both predictive biomarkers and mechanisms of drug resistance. We observed the most striking combination efficacy in models where both monotherapies had some effect. Olaparib was most efficacious against CDX3, a model that failed to express HRR protein PALB2. Although damaging PALB2 mutations are reported to account for outlier responses to PARP inhibitors in other cancer types (48, 49), targeted therapies that work in one cancer type do not necessarily work in another cancer type, even if the genetics are similar (e.g., vemurafenib in BRAF-mutant melanoma vs. colorectal cancer; ref. 50). To our knowledge, this is the first report correlating a PALB2 mutation to combined PARP/WEE1 inhibition in SCLC. Moreover, SCLC tumors are not typically associated with defects in HRR, a known predictive biomarker for PARP inhibitor sensitivity, and our data suggest further research into this area is warranted.
Multiple studies have shown that olaparib efficacy is significantly reduced against tumors pretreated with chemotherapy. We sought to address this by examining the efficacy of olaparib/AZD1775 against a model generated from a patient who progressed on a platinum-based therapy. Consistent with patient progression, CDX8p was resistant to treatment with platinum/etoposide. Although CDX8p was also relatively insensitive to olaparib/AZD1775, this combination elicited a more durable response than platinum/etoposide. Similarly, pretreatment of CDX3 with platinum/etoposide blunted the response of chemonaive CDX3 to olaparib/AZD1775. Given the deep, though short-lived response to patient standard-of-care platinum-based chemotherapy, it may be challenging to administer an olaparib/AZD1775 combination to treatment-naïve SCLC patients. However, consideration should be given to minimizing the number of chemotherapy cycles given to debulk the disease prior to testing this novel combination.
The exquisite sensitivity of CDX3 to combination therapy is likely due to the fact that both olaparib and AZD1775 exhibited single-agent efficacy against this model. AZD1775 can both compromise the ability to restrain mitotic commitment in response to DNA damage and promote catastrophic DNA damage in G1–S via inappropriate activation of the MUS81-SLX4 nuclease (51). Thus, combinatorial activity of olaparib/AZD1775 may stem from AZD1775-induced inappropriate cleavage of stalled replication forks stemming from PARP trapping. However, our understanding of the biology of inappropriate induction of these pathways in S phase remains in its infancy, and a detailed mechanistic understanding of what role PARP may play in this process remains beyond the scope of this study.
In addition to affecting sensitivity to olaparib, the PALB2 mutation could potentially explain the response of CDX3 to AZD1775. BRCA2 and PALB2 function has been linked to replication fork stabilization/restart as part of the replication stress response. CDX3, the model that is most sensitive to AZD1775 single treatment, already has significantly elevated levels of replicative stress as indicated by elevated levels of MYCL, cyclin E1, and pRPA compared with the lesser responsive models. Moreover, we have demonstrated that following AZD1775 treatment, there is a further increase in pRPA. Phosphorylated RPA can recruit PALB2 and BRCA2 to the stalled fork, and PALB2 loss has been associated with defects in the recovery of the stalled fork and therefore increased DNA damage (52). In addition, BRCA2 and PALB2 are key regulators of the G2–M cell-cycle checkpoint upon DNA damage (53). Therefore, the absence of PALB2 in combination with the inhibition of WEE1 could increase premature mitotic entry and therefore genomic instability in CDX3 cells. Together, these data suggest the efficacy of the olaparib/AZD1775 combination is likely to involve both an increase in replication stress response in S-phase and carryover of S-phase–induced damage into mitosis.
Combination trials using PARP inhibitors are already in early phase development, including a second-line study comparing temozolomide and veliparib with temozolomide and placebo where the combination reported increased response rates but no significant difference in 4-month progression-free survival (54). The ongoing phase I/II trial of olaparib and temozolomide in patients progressing after chemotherapy also reported a 46% response rate for the combination compared with 14% for temozolomide alone (55). Olaparib/AZD1775 has demonstrated preclinical activity against AML (56), and a phase I clinical trial utilizing olaparib and AZD1775 has been initiated in patients with refractory solid tumors (https://ClinicalTrials.gov/show/NCT02511795), including SCLC, and an extended cohort in SCLC is planned.
Conversion of improved clinical response rates to lengthened progression-free and overall survival will likely require development of robust predictive biomarkers. Our preclinical CDX data suggest that predicting the depth and duration of responses to olaparib and AZD1775 in SCLC patients is unlikely to be straightforward; the speed with which acquired chemotherapy resistance occurs, the genomic instability, and heterogeneity of this devastating disease all point to the need for development of broad biomarker panels for the optimal use of this combination therapy, where different patients’ tumors may acquire resistance via different and multiple molecular mechanisms. Further characterization of the DNA damage response status across a wide range of SCLC tumors, as well as functional validation of potential biomarkers is ongoing. Our results raise the possibility that a subset of SCLC may have defects in DNA damage response pathways that could be exploited by the combination of PARP and WEE1 inhibitors. Moreover, replication stress is often associated with oncogene overexpression and MYC family members, which are known to be amplified in approximately 20% SCLC, and the role of MYC as a predictive biomarker of response to olaparib and AZD1775 can now be studied further in our burgeoning panel of 45 CDX models (57–59). The CDX3 cellular context represents a “super responder” to the olaparib/AZD1775 combination and has revealed important molecular insights that we can now take forward in an assessment in the larger CDX panel.
Surgical specimens and archival biopsies are unlikely to reflect the disease after first- or second-line treatments. We previously reported the prevalence of CTCs in SCLC (60), their prognostic significance, and their potential as a readily obtained and repeatable source of pharmacodynamic and predictive biomarkers. Moreover, mutations in a panel of genes associated with DNA damage response measured in circulating tumor DNA may also assist patient selection. These liquid biopsy approaches should now be fully exploited in upcoming trials of novel therapies in SCLC where promising data in patient-relevant preclinical models are forthcoming.
Disclosure of Potential Conflicts of Interest
C. Dive reports receiving commercial research grants from AstraZeneca. C.J. Morrow is an employee of AstraZeneca. F. Blackhall reports receiving other commercial research support from and is a consultant/advisory board member for AstraZeneca. No potential conflicts of interest were disclosed by the other authors.
Authors' Contributions
Conception and design: K.K. Frese, C.J. Morrow, F. Blackhall, M.J. O'Connor, C. Dive
Development of methodology: A. Lallo, C.J. Morrow, C.L. Hodgkinson, M.J. O'Connor
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Lallo, R. Sloane, M.W. Schenk, F. Trapani, M. Galvin, S. Brown, C.L. Hodgkinson, L. Priest, A. Hughes, C. Miller
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Lallo, K.K. Frese, C.J. Morrow, S. Gulati, F. Trapani, N. Simms, M. Galvin, A. Hughes, Z. Lai, E. Cadogan, G. Khandelwal, K.L. Simpson, C. Miller, F. Blackhall, M.J. O'Connor, C. Dive
Writing, review, and/or revision of the manuscript: A. Lallo, K.K. Frese, R. Sloane, F. Trapani, Z. Lai, E. Cadogan, G. Khandelwal, K.L. Simpson, C. Miller, F. Blackhall, M.J. O'Connor, C. Dive
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Galvin, C.L. Hodgkinson, E. Cadogan
Study supervision: K.K. Frese, C.J. Morrow, M.J. O'Connor, C. Dive
Acknowledgments
The authors thank Matt Carter, Matt Lancashire, the CEP in vivo team, CRUK Manchester Institute core facilities, and the patients and their families. They also thank Ian Hagan for his constructive comments on the article.
C. Dive receives funding from the Cancer Research UK Manchester Institute core (C5759/A27412), the Cancer Research UK Manchester Centre Award (C5759/A25254), the Cancer Research UK Lung Cancer Centre of Excellence Award (C5759/A20465), and the Cancer Research UK/AstraZeneca biomarkers alliance (10001080/AgrID486).
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.