PI3Kβ inhibition enhances ALK‐inhibitor sensitivity in ALK‐rearranged lung cancer

Treatment with anaplastic lymphoma kinase (ALK) inhibitors significantly improves outcome for non‐small‐cell lung cancer (NSCLC) patients with ALK‐rearranged tumors. However, clinical resistance typically develops over time and, in the majority of cases, resistance mechanisms are ALK‐independent. We generated tumor cell cultures from multiple regions of an ALK‐rearranged clinical tumor specimen and deployed functional drug screens to identify modulators of ALK‐inhibitor response. This identified a role for PI3Kβ and EGFR inhibition in sensitizing the response regulating resistance to ALK inhibition. Inhibition of ALK elicited activation of EGFR, and subsequent MAPK and PI3K‐AKT pathway reactivation. Sensitivity to ALK targeting was enhanced by inhibition or knockdown of PI3Kβ. In ALK‐rearranged primary cultures, the combined inhibition of ALK and PI3Kβ prevented the EGFR‐mediated ALK‐inhibitor resistance, and selectively targeted the cancer cells. The combinatorial effect was seen also in the background of TP53 mutations and in epithelial‐to‐mesenchymal transformed cells. In conclusion, combinatorial ALK‐ and PI3Kβ‐inhibitor treatment carries promise as a treatment for ALK‐rearranged NSCLC.


Introduction
In 3-7% of NSCLCs, ALK gene rearrangements lead to the expression of oncogenic fusion proteins that confer constitutive activity of the ALK kinase domain [1,2]. Aberrant ALK activity, in turn, activates the MAPK and PI3K-AKT oncogenic signaling pathways [3,4]. Treatment of ALK-rearranged NSCLC with first-or second-generation ALK inhibitors associates with favorable initial response in the majority of patients, although the development of resistance and clinical relapse typically occurs within a few years [5]. Five to 10% of patients show progressive disease following alectinib treatment, at present the recommended first-line therapy for ALK-rearranged NSCLC [5,6]. Intrinsic resistance may relate to, for example, the precise ALK fusion variant [7] or TP53 mutations which frequently co-occur with ALK rearrangements [8][9][10]. Among tumors that acquire resistance to ALK inhibitors, 30-50% exhibit ALK-dependent resistance mechanisms, particularly ALK mutations and/or amplification [11,12]. In the remainder, resistance associates with diverse mechanisms, including activation of bypass signaling pathways (e.g., EGFR, KIT, IGF-1R, or GPCRs), epithelial-tomesenchymal transition (EMT), a histopathology switch, or increased activity of drug efflux pumps [3].
For ALK-rearranged NSCLCs exhibiting ALKdependent acquired resistance, second-or thirdgeneration ALK inhibitors that retain activity in the presence of resistance mutations to first-line inhibitors offer optional "sequential" treatment opportunities. However, the mechanistic drivers for resistance, such as bypass signaling, are not diagnostically assessed and likely differ between samples. In addition to pronounced intertumor variability in resistance mechanisms, ALKrearranged lung tumors also exhibit intratumor heterogeneity in ALK inhibitor resistance [3]. To address this clinical problem, patient-derived cultures from ALK inhibitor-resistant tumors can be used to identify treatment-adaptive resistance mechanisms and combinatorial treatment approaches [13][14][15]. Nevertheless, as of yet, no combinatorial treatments to counter bypass signaling are available in the clinic.
Here, we performed multiregional characterization of an aggressive chemoresistant ALK-rearranged lung tumor and corresponding tumor-derived cultures. Drug response profiles were different in cultures derived from a region exhibiting an epithelial-to-mesenchymal transition (EMT) phenotype. Importantly, our study identified PI3Kb as a novel target for improving response to ALK inhibition, and combinatorial ALK and PI3Kb inhibitory response was detected in all cultures, also in the context of EMT and TP53 mutation. As PI3Kb is an effector of multiple tyrosine kinase activities that can mediate ALK inhibitor resistance, including EGFR, the co-targeting of PI3Kb and ALK offers promise as a treatment for ALK-rearranged lung cancer.

Drug sensitivity and resistance testing (DSRT) assay
Drug sensitivity and resistance testing assays for anticancer compounds as a single agent or in combination with other compounds were performed as previously described [17]. For drug screening, 384-well plates (Corning, Corning, NY, USA; 3712) with compounds were prepared in advance by dispensing compounds using an Echo 550 liquid handler (Labcyte, Synnyvale, CA, USA), at five concentrations covering a 10 000-fold concentration range. For storage, the predrugged plates were kept in pressurized StoragePods (Roylan Developments Ltd., Fetcham, UK) under inert nitrogen gas. For drug screening, the predrugged compounds were dissolved in 5 lL of culture medium per well, with (1 : 2000 final volume) or without CellTox Green (Promega, Madison, WI, USA) depending on the experiment, and 20 lL cell suspension per well was seeded at a concentration of 1500 cells per well. After 72-h incubation at 37°C, cell death was assessed by measuring fluorescence signals from Cell-Tox Green (485/520 nm excitation/emission filters). Subsequently, to assess cell viability, 25 lL per well CellTiter-Glo reagent (Promega) was added, and luminescence was recorded. Both fluorescence and luminescence readouts were recorded using a PHERAStar FS plate reader (BMG Labtech, Ortenberg, Germany). To plot dose-response curves for each drug, the Marquardt-Levenberg algorithm was implemented using the inhouse-developed bioinformatic 'Breeze' pipeline [18]. To compare drug responses across samples, Drug Sensitivity Scores (DSSs) were calculated using dose-response curve parameters including the IC50, slope, top, and lower asymptotes, as described [19]. To manually assess the drug responses of single agents or drug combinations, a denser concentration range (nine doses between 0.5/1 and 5000/10 000 nM) of compounds was used. Cells (1500 per well) were seeded in 384-well plates in 20 lL of media. After 24-h incubation at 37°C, cells were treated with vehicle control or drug in 10 lL of media with three technical replicates for each condition. After 72-h incubation at 37°C, cell viability was quantified using CellTiter-Glo reagents. The relative cell viability was calculated using the formula: (cell viability of drug treatment)/(cell viability of vehicle control) 9 100. Values for the selectivity index are calculated using the formula p(IC50 ALKrearranged cells/IC50 ALK wildtype cells).

Immunoblotting
Lysates were prepared from tumor tissue or cultured cells using RIPA buffer supplemented with fresh protease and phosphatase inhibitors (Roche, Penzberg, Germany). Protein quantification was performed using the BCA Protein Assay (G Biosciences, St. Louis, MO, USA; 786-570). Lysates were fractionated using Mini-PROTEAN TGX precast gels (Biorad, Hercules, CA, USA) and transferred to PVDF membranes (Millipore, Burlington, MA, USA; IPFL00010) using the XCell II blot module (Thermo-Scientific). After transfer, membranes were blocked for 30 min at room temperature using Odyssey Blocking Buffer. Two-color immunoblotting was performed using Odyssey Blocking Buffer (LI-COR, Lincoln, NE, USA) and IRDye (800CW/680RD) secondary antibodies (LI-COR) diluted 1:10 000 in Odyssey Blocking Buffer. Membranes were scanned using an Odyssey infrared imager (LI-COR) and quantifications were performed using the Image Studio software (LI-COR). Primary antibodies are listed in the Supplemental Experimental Procedures.

Co-immunoprecipitation
Co-immunoprecipitation (Co-IP) experiments were performed using the Thermo Fisher Scientific Pierce Co-IP Kit (26149), following the manufacturer's protocol. In brief, cells were exposed to drugs when cell confluency reached 60-70%. After 4 or 24 h of drug exposure, cells were lysed with an ice-cold Lysis/Wash Buffer supplemented with fresh protease and phosphatase inhibitors (Roche). Cell lysates were precleared by incubating the lysate with Control Agarose Resin at 4°C for 1 h. To prepare antibody-immobilized AminoLink Plus Coupling Resin, 25 lL Coupling Resin, and 1 lg of rabbit anti-EGFR (Santa Cruz Biotechnology, Dallas, TX, USA; sc-120) or IgG control antibody (Invitrogen, Waltham, MA, USA; 02-6102) were co-incubated in a spin column for 2 h at room temperature. Columns were then washed twice with 19 Coupling Buffer and six times with Wash Solution, and centrifuged after each wash. Subsequently, precleared cell lysate was added to the column containing antibody-coupled resin, and incubated overnight at 4°C with gentle mixing. Proteins were eluted using 60 lL of Elution Buffer, and eluates were analyzed by western blotting with primary antibodies for EGFR, pEGFR, PI3Kb, and tubulin. See Supplementary Information section for antibody details. Table S1 in the Supporting Information section lists the antibodies and dilutions used for Co-IP.

IncuCyte confluency assay
Cells were seeded at a density of 5000-10 000 cells per well in 96-well plates and were exposed to drugs or DMSO on the following day. Immediately after drug treatment, cells were followed by live-cell imaging using an IncuCyte ZOOM microscope (Essen Bioscience, Ann Arbor, MI, USA) by taking pictures every 3 or 4 h for a total of 72 or 120 h. Images were then quantified for cell confluency (cell surface area coverage as confluence values) using the INCUCYTE application software (Essen Bioscience), and cell confluency data were plotted with GraphPad Prism (GraphPad Software, San Diego, CA, USA).

Colony formation assay
Cells were seeded at a density of 10 000-20 000 cells per well in 24-well plates and 50 000 cells per well in six-well plates. On the following day, cells were exposed to drugs or DMSO. Medium and DMSO/drugs were replaced every 72 h for 15 days. Cells were fixed with colony fixation solution [acetic acid/methanol 1 : 7 (vol/vol)] and stained with 0.5% crystal violet. After drying, stained plates were scanned with the Cytation 5 Cell Imaging Multi-Mode Reader (BioTek, Winooski, VT, USA) and the number and size of colonies were quantified with the compatible Gen5 TM Multi-Mode Reader and Imager Software (BioTek). To assess the long-term effect of ALK inhibitors, cells were treated with IC 50 doses of each ALK inhibitor that were calculated based on 3-day drug response measurements.

Autophagy analysis
Cells were seeded at a density of 10 000-20 000 cells per well in 96-well plates and exposed to drugs or DMSO on the following day. Following 24-h incubation, cells were costained with CYTO-ID Green detection reagent and Hoechst 33342 nuclear stain according to the manufacturer's instructions (Enzo Life Science, San Diego, CA, USA). Cells were observed and imaged using Opera Phenix High-Content Screening System (PerkinElmer, Waltham, MA, USA). Quantification of the number of autophagic vacuoles, vacuoles per cell, their size, and the intensity of the vacuoles was carried out using the system's hHARMONY software (PerkinElmer).

Drug treatment with or without EGF
Cells (5000 cells per well) were cultured in 96-well plates in media without EGF. After 24 h, cells were treated with indicated drug concentration or vehicle control in media supplemented with 10 or 100 ngÁmL À1 fresh EGF, or without addition of EGF. Drug and EGF treatments were repeated every 24 h for 3 days. After 72 h of total drug treatment, cell viability was quantified using CellTiter-Glo reagents. The relative cell viability was calculated using the formula: (cell viability of drug treatment)/(cell viability of vehicle control) 9 100.

In vivo treatment experiment
Treatment studies on patient-derived xenograft (PDX) models derived from primary TR3 and TR5 ALKrearranged NSCLC cultures were performed at the Charles River facilities. This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the Society of Laboratory Animals (GV SOLAS) in an AAALAC accredited animal facility. All animal experiments were approved by the Committee on the Ethics of Animal Experiments of the regional council (Regierungspr€ asidium Freiburg, Abt. Landwirtschaft, L€ andlicher Raum, Veterin€ ar-und Lebensmittelwesenpermit # G-18/12). NSG (NOD.Cg-Prkdcscid Il2rgt-m1Wjl/SzJ) animals were used for in vivo treatment experiments, and animals were sourced from Charles River, France. Both male and female NSG mice (6-8 weeks old) were implanted subcutaneously with fragments from either TR3 or TR5 PDX tumors. Once average tumor volume reached 100 mm 3 , animals were randomized to different treatment groups (n = 5 per treatment group) and were administered with vehicle control, ceritinib alone, AZD-8186 alone or combination of ceritinib plus AZD-8186. Ceritinib was administered at 25 mgÁkg À1 body weight p.o. daily for 21 days. AZD-8186 was administered at 2 9 25 mgÁkg À1 body weight p.o. daily for 21 days. Ceritinib (25 mgÁkg À1 ) and AZD-8186 (2 9 25 mgÁkg À1 ) were administered together p.o. daily for 21 days. The control vehicle (0.5% Methylcellulose, 0.5% Tween80) p.o. for 21 days. Tumors were measured using electronic calipers twice a week, and tumor volumes were calculated using the formula length 9 width 2 9 0.52. Body weights were recorded in parallel to the tumor volume measurement. Animals were monitored daily for signs of morbidity and/or mortality.

Animal housing and handling
Animals were housed in individually ventilated cages (TECNIPLAST Sealsafe-IVC System, TECNIPLAST, Hohenpeissenberg, Germany), depending on group size, either in type III or in type II long cages. They were kept under a 14L:10D artificial light cycle. The temperature inside the cages was maintained at 22-26°C with a relative humidity of 45-65% and 60-65 air changesÁh À1 in the cage. Dust-free bedding consisting of aspen wood chips with approximate dimensions of 5 mm 9 5 mm 9 1 mm (ABEDD, LAB & VET Service GmbH, Vienna, Austria; LTE E 001) and additional nesting material were used. The cages including the bedding and the nesting material were changed weekly. The animals were fed autoclaved Teklad Global Extruded 19% Protein Rodent Diet from Envigo RMS SARL (Gannat, France) and had access to sterile filtered and acidified (pH 2.5) tap water that was changed twice weekly. Feed and water were provided ad libitum. All materials were autoclaved prior to use. Animals were routinely monitored at least twice daily on working days and at least once daily on weekends and public holidays. Routine monitoring included inspections for dead animals, assessment of animal welfare and tumor growth by observation, control of feed and water supply and of technical housing conditions. Any observed or suspected impairment of animal welfare was documented.

Statistics and reproducibility
GRAPHPAD PRISM 9 (GraphPad Software Inc) was used to generate all figures presented and to perform statistical analyses of experimental data. Statistical significance was assessed using a Student's t test and nonparametric Mann-Whitney test or Wilcoxon matched-pairs signed rank test. P-values > 0.05 were considered as statistically significant. Error bars indicate standard deviation or standard error of the mean. Pearson's correlation coefficients were used to assess the significance of correlations and displayed in the XY plots. DSRT, NGS, and IHC experiments were performed a single time, with biological replicates (n = 2-6). Immunoblot analyses including coimmunoprecipitation, FISH, and validation of siRNAmediated PIK3CB knockdown were performed two times, with similar results. Follow-up experiments including colony formation assay, IncuCyte confluency assay, CYTO-ID based autophagy assay, and experiments evaluating the effect of PIK3CB knockdown, EGF treatment, GPCR inhibition or autophagy inhibition on responses of ALKi, PI3Kbi or combination of ALKi plus PI3Kbi were performed three times, with similar results. Xenograft experiments were performed once at Charles River by different experimenters.

Immunohistochemistry
Tissue processing and IHC procedures were performed essentially as described previously [17]. Antibodies and stain-specific details are listed in the Supplemental Information section. To acquire whole slide scans of stained tissue sections, the Pannoramic 250 digital slide scanner (3DHISTECH, Budapest, Hungary) was used and the scanned TIFF images are exported using the Pannoramic Viewer (3DHISTECH). Table S1 in the Supporting Information section lists the antibodies and dilutions used for IHC.

Fluorescence in situ hybridization (FISH)
The ALK-rearrangement status of the samples was evaluated by FISH using ALK dual-color break-apart probe according to the manufacturer's protocol (Vysis, Abbott Molecular Inc., Abbott Park, IL, USA). Tumor cell nuclei with split red and green signals were defined as positive for ALK-rearrangement, and at least 100 cells were evaluated for each sample to conclude the ALK-rearrangement status. Samples displaying ALK-rearrangement in more than 10% of the cells were labeled as positive.

Genetic analysis
Genomic DNA was extracted from healthy lung and tumor tissue samples and from corresponding CR cultures using a DNeasy Blood & Tissue kit (Qiagen, Hamburg, Germany). Targeted next-generation sequencing was performed using the NimbleGen Cancer Panel (captures the exons of 578 cancer-related genes; Roche Nimblegen, Madison, WI, USA) and the Illumina HiSeq2500 system in HiSeq high output mode using v4 chemistry or HiSeq Rapid run mode using v2 chemistry (Illumina, San Diego, CA, USA), as described [20].

Intratumor heterogeneity of an ALKrearranged lung tumor
To capture the extent of intratumor heterogeneity and to identify treatments that can enhance sensitivity to ALK inhibition in ALK-rearranged lung cancer, we characterized tumor tissues (n = 6) and cultures (n = 4) established from multiple tumor regions (TRs) of a large surgically resected tumor (9 9 12 9 9 cm) from a 55-year-old never-smoker female patient with ALKrearranged T3N2M0 stage lung adenocarcinoma with diffuse metastasis. Before surgery, the patient was treated with chemotherapy, which did not halt disease progression ( Fig. 1A and Fig. S1A,B). Immunohistochemical characterization of six tumor regions was done for ALK and markers of lung adenocarcinoma (NKX2.1), epithelial (E-cadherin, pan-cytokeratin, and cytokeratin 18) and mesenchymal (vimentin) phenotypes, tumor vasculature (CD31), and basement membrane organization (collagen type IV). All but one tissue region showed a similar phenotype with cancer cells exhibiting positivity for both NKX2.1 and the epithelial marker E-cadherin. The exception was tumor region #5 (TR5), which exclusively showed aggressive lung cancer features, including aberrant basement membrane organization evidenced by enhanced collagen type IV staining, high tumor vasculature, and mesenchymal phenotype cancer cells lacking NKX2.1 expression ( Fig. 1B and Fig. S1C). Importantly, TR5derived cells exhibited predominant expression of vimentin corroborating the parental mesenchymal tissue phenotype of the cancer cells in TR5. Unlike other TRs, TR5 cells expressed a high level of ALK fusion protein and formed multilayered tightly packed colonies in ex vivo culture (Fig. 1C,D and Fig. S1D). The EML4-ALK fusion was shown to be variant 1 via comparative western blotting of the same variant harbored by the commonly used ALK-rearranged lung cancer cell line NCI-H3122 (Fig. 1D). We next asked whether TR5 cancer cells carried unique genetic aberrations to explain their divergent phenotype. Panel sequencing of the TR3 and TR5 tumor tissues and their corresponding cultures, however, showed an identical set of somatic mutations ( Fig. 1E and Table S2), suggesting that the difference between TR5 and other regions was phenotypic rather than mutational. The aggressive nature of the cancer was underscored by multiple mutations in tumor suppressor genes, including TP53, CBFA2T3, PPP2R1A, and SMARCA4. As expected, variant allele frequencies were substantially higher in cultures than in their respective tissues, indicating cancer cell enrichment in cultures. Overall, this demonstrates intratumor phenotypic heterogeneity and a genomic profile that underscores the aggressive nature of this ALK-rearranged lung cancer sample.

Patient-derived cells show moderate sensitivity to ALK inhibition
To understand how the phenotypic heterogeneity of different tumor regions relates to functional differences, we undertook drug profiling of tumorderived cultures from four regions, and compared responses to those in patient-matched normal lung tissue-derived cells and ALK-rearranged H3122 cells. This used a customized panel of 527 anticancer compounds representing both approved and investigational compounds, including > 100 compounds targeting ALK or its effector pathways (Table S3). Matching with the patient's clinical response, tumor-derived cultures were mostly resistant to cisplatin and pemetrexed (Fig. S2A). The drug sensitivity score (DSS) was calculated for each drug response measurement by integrating the IC50 values, area under the curve (AUC),  slope as well as top and lower asymptotes of each dose-response curve, allowing for rational comparisons across a large number of drugs and samples [19]. Selectivity was demonstrated by the finding that only ALK-rearranged cancer cells, but not normal lung epithelial cells, showed sensitivity to ALK inhibitors (ALKi) ( Fig. 2A-C and Fig. S2B,C). However, these ALKi responses were moderate when compared to those measured in ALK-rearranged H3122 cells, but higher than in H2228 cells.  TR5   TR3   TR2   TR1   TR3  TR5  TR3  TR5  TR3  TR5  100  150  200 Colony area (%)

A combinatorial drug screen identifies PI3Kb as a novel target to restore ALKi sensitivity
To assess long-term sensitivity of tumor-derived cultures to ALKi, cells were exposed to IC 50 doses of five ALKi for 15 days. All tumor-derived cultures showed paradoxical increased cell growth as evidenced by increased colony formation in presence of ALKi (Fig. 2D). We selected TR3 and TR5 cells for the remainder of the follow-up experiments, as TR3 resembled the phenotypic features of TR1 and TR2, and TR5 uniquely differed from the other tumorderived cultures. Several ALK inhibitors were utilized in subsequent drug sensitivity experiments, with ceritinib being the most frequently used because it was the only second-generation approved ALKi at the time of our investigation. With regard to the different sensitivity of tumor-derived cultures to ALKi in short-term and long-term assays, we treated TR3 and TR5 cells with ceritinib and alectinib and measured the colonies every 3 days for up to 15 days to determine how long it would take for cancer cells to overcome the cytostatic effect of ALKi. Similar to the short-term viability assay, the colony area of drug-treated cells was lower than that of control cells at the 3-day time point, while at later time points the colony areas of drug-treated cells surpassed control cells. This suggested that 3-6 days are needed to overcome an ALKi-induced halt in cell proliferation (Fig. S2C), eluding to loss of ALKi sensitivity in patient-derived cells.
To identify treatments that may improve sensitivity to ALKi, we implemented a combinatorial drug screen on tumor-derived cells using a drug panel in combination with 200 nM ceritinib. As expected, we identified combinatorial responses of ALKi plus EGFR inhibitors (EGFRi; n = 4) and pan-ERBB inhibitors (n = 12). In addition, ALKi plus PI3Kb inhibitors (PI3Kbi; n = 4) showed strong combinatorial responses ( Fig. 2E and Fig. S3A-E). On the contrary, no combinatorial responses were seen between ALKi and PI3Ka isoform inhibitors or mTOR inhibitors (Fig. S3C). To crossvalidate our discoveries, we performed additional combinatorial screens in combination with gefitinib (an EGFRi) or AZD-8186 (a PI3Kb-selective inhibitor). Both screens identified ALK as the most effective combination target (Fig. 2F,G and Fig. S3F), altogether demonstrating that partial sensitivity to ALKi primary tumor cells was regulated by EGFR and PI3Kb and could be effectively overcome by combined inhibition of ALK and EGFR or PI3Kb.
To understand the molecular mechanism for this limited ALKi sensitivity of tumor-derived cultures, we analyzed signaling events following ceritinib treatment. Evaluation of phosphorylation cascades in TR3, TR5, and H3122 cells treated with different ceritinib concentrations showed a reduction in the level of ALK phosphorylation, and adaptive activation of EGFR and AKT was evident in all treated samples. When compared to TR3 and H3122, TR5 appeared to have higher levels of baseline ERK phosphorylation, while TR3 predominantly exhibited ERK rebound activation upon treatment with ceritinib (Fig. S4A,B). Next, to explore whether EGFR activation reduced sensitivity to ALKi, ALK-rearranged lung cancer cells were cotreated with the EGFR ligand EGF and ceritinib. As expected, EGF treatment decreased the effect of ALKi on cell viability in all tested ALK-rearranged lung cancer cell models (Fig. S4C). Since ALK inhibition led to adaptive activation of EGFR and downstream MAPK and PI3K-AKT signaling, and as EGF perturbation dampened sensitivity to ALKi, the models evaluated in this study appear to exhibit EGFRdriven ALKi resistance, matching findings of other studies [12,22,23].

Normal lung cells are insensitive to combined inhibition of ALK and PI3Kb
To evaluate sensitivities and possible generic toxicities of compounds targeting the EGFR receptor family and the PI3K-AKT-mTOR pathway, we utilized normal lung epithelial cells derived from three individuals. EGFR or pan-ERBB inhibition produced toxic and antiproliferative responses in normal lung cells. Similarly, multiple compounds targeting various nodes of the PI3K-AKT-mTOR pathway showed toxicities in the normal cells. However, compounds selectively targeting PI3Kb, either on their own or in combinations with ALKi, did not detectably affect the normal lung epithelial cells (Fig. 3A-D and Fig. S5A-E). Furthermore, selectivity index analysis confirmed that the combination of ALKi and PI3Kbi provided a significantly wider therapeutic window in comparison to single ALKi or the combination of ALKi and EGFRi (Fig. 3E). In addition, selectivity index analysis demonstrated that next-generation ALK inhibitors such as alectinib, brigatinib, and lorlatinib had superior selectivity when used in combination with PI3Kbi than when used alone (Fig. 3F). Therefore, the combination of PI3Kbi and ALKi acts highly selectively on ALK-rearranged NSCLC cells, while the combination of EGFRi and ALKi appears to act non-specifically, also targeting the normal epithelial cells.  TR3 TR5 TR3 TR5 TR3 TR5 TR3 TR5 TR3 TR5 TR3 TR5 TR3

Combined inhibition of ALK and PI3Kb elicits synergistic responses in ALK-rearranged lung cancer cells
To widen the studies of the combinatorial effect of ALKi and PI3Kbi, drug response measurements were repeated and extended to two ALK-rearranged lung cancer cell lines, H3122 and H2228. ALKi and PI3Kbi combination treatment effectively and synergistically prevented the proliferative and clonogenic potential of both the cell lines and the patient-derived ALK-rearranged cells (Fig. 4A-C). Using a wider concentration range of compounds and additionally by measuring death as a readout of drug responses we explored the synergistic potential of ALKi and PI3Kbi combination treatment. The previously detected viability-based drug sensitivities were confirmed to be synergistic in TR3 and TR5 cells, but not in H3122 and H2228 cells (Fig. S6A-C). However, a cell death-based readout showed a synergistic response of combinatorial ALK and PI3Kb inhibition in H3122 cells (Fig. S6A-C). Furthermore, siRNAmediated gene silencing of PIK3CB in TR3 and TR5 mimicked the combinatorial responses seen with PI3Kbi (Fig. 4D), even though the knockdown of PIK3CB was only partial (Fig. S6D). The ALKi/ PI3Kbi combination did not show an effect in four KRAS mutant lung cancer lines, validating the selectivity of this combination for ALK-rearranged lung cancer (Fig. S6E). Lastly, we assessed the in vivo efficacy of combinatorial ALKi and PI3Kbi treatment, and treated NSG mice (five animals per group) bearing subcutaneous tumors derived from TR3 and TR5 cells with ceritinib, AZD-8186 or a combination of both treatments for 21 days. Both single agent and combination treatments were well-tolerated with minor or no body weight loss over the course of treatment (Fig. S7A,B). Contrary to in vitro findings, both TR3 and TR5 cell-derived tumors displayed strong sensitivity to ceritinib single therapy, and in both single and combination treatment arms, tumors became unmeasurable by the end of the treatment (Day 21; Fig. S7C,D). Within a week of treatment cessation, tumor growth resumed in both treatment arms (Fig. S7E,F), with tumor regrowth being notably slower in the combination treatment compared to the ceritinib arm (Fig. S7G,H).

Combinatorial ALKi plus PI3Kbi response appears independent of GPCR signaling and autophagy
Because autophagy and purinergic G-protein-coupled receptor (GPCR) activity have both been proposed to regulate ALKi sensitivity [23][24][25][26], and that these processes are regulated by PI3Kb, but not PI3Ka [24,25], we investigated their involvement. Consistent with published data [26], ALK inhibition triggered autophagy, and the co-targeting of ALK and PI3Kb reduced autophagic vacuole formation (Fig. S8). We therefore tested whether autophagy inhibitors or autophagy and PI3Ka inhibitors could mimic the response to ALK plus PI3Kb inhibition. However, none of the tested combinatorial treatments resulted in synergistic responses (Fig. S9A,B), suggesting that PI3Kb acts beyond canonical PI3K-AKT signaling and autophagy. Furthermore, while elevated P2Y subfamily of P2 purinergic GPCRs expression was reported to associate with ALKi resistance in both clinical samples and cultured cells [23], inhibition of P2YRs did not enhance ALKi response in our cell systems (Fig. S9C). In conclusion, our data suggest that PI3Kb-mediated inhibition of autophagy or P2YR signaling is not sufficient to induce the combinatorial effect with ALK inhibition in ALKrearranged cells.

Resistance to ALK inhibition is associated with MAPK and PI3K-AKT pathway activation downstream of EGFR-PI3Kb
To understand the consequences of combinatorial ALK plus PI3Kb inhibition at the molecular level, we analyzed MAPK and PI3K-AKT pathway activities in cells treated with vehicle, ceritinib, AZD-8186, or their combination. In comparison with single treatments, cells treated with the combination showed a substantial reduction in both MAPK and PI3K-AKT activities, as well as an increase in cell death evidenced by increased PARP cleavage accompanied by reduced ERK and AKT phosphorylation (Fig. 5A).
Next, we explored how PI3Kb function is regulated upon ALK inhibition. Considering that the ERBB family member ERBB3 was reported to recruit PI3Kb and thereby drive PI3Ka inhibitor resistance in HER2amplified and PIK3CA mutant cancers [27], we hypothesized that ALK inhibition may similarly activate EGFR and thereby regulate PI3Kb function. We therefore immunoprecipitated EGFR from TR3 and TR5 cell lysates treated with DMSO control or 300 nM ceritinib for different times and assessed the phosphorylation of EGFR and PI3Kb binding. EGFR phosphorylation was confirmed to significantly increase following ALK inhibition (Fig. 5B). Interestingly, PI3Kb coprecipitated with EGFR, demonstrating that EGFR physically interacts with PI3Kb in those cells (Fig. 5B). Importantly, treatment of TR3, TR5, H3122, or H2228 cells with EGF elicited increased resistance to ALK inhibition, while the combined inhibition of ALK and PI3Kb rescued the EGFinduced resistance, together suggesting that inhibition of PI3Kb restricts EGFR-mediated bypass resistance to ALK inhibition in ALK-rearranged lung cancer ( Fig. 5C and Fig. S10).   In summary, dissection of the mechanisms underpinning combinatorial ALK and PI3Kb inhibition efficacy suggest that the sensitization of ALK-rearranged cancer cells to ALK inhibition by selective PI3Kb inhibition was, at least in part, mediated through a blockade of EGFR-mediated rebound activation of MAPK and PI3K-AKT signaling, enhancing cell death (Fig. 6).

Discussion
ALK inhibition is an effective treatment option for ALK-rearranged NSCLC. However, resistance to ALKi develops over time, and it is most often driven by ALK-independent mechanisms [3]. We here studied primary cells derived from an aggressive TP53 mutant and chemoresistant ALK-rearranged tumor that exhibited region-specific EMT features to identify novel/effective combination treatments to treat ALKrearranged NSCLC. We identified selective PI3Kb inhibition as a promising novel target to potentiate ALK inhibition responses in ALK-rearranged lung cancer. Importantly, this combinatorial effect was seen with all tested ALK inhibitors, including the most commonly used ALK inhibitor in the clinic, alectinib, and the third-generation ALK inhibitor lorlatinib. Cotargeting of ALK and PI3Kb lead to blockade of both MAPK and PI3K-AKT signaling, attenuated EGFRmediated adaptive resistance, and elicited tumor cellselective toxicity with efficacy even in TP53 mutant ALK-rearranged cells that had undergone EMT. Our study therefore demonstrates the utility of primary patient tumor-derived cultures for pharmacogenomic profiling and dissection of drug resistance mechanisms.
Both the a and b PI3K class IA isoforms are ubiquitously expressed in both normal and malignant tissues, but these subunits are regulated differently, induce different modes of signaling, and have divergent roles in health and disease [28]. The PI3Ka isoform is a well-established oncogenic driver that is frequently oncogenically mutated and constitutes a drug target for several cancer types, including lung cancer [28]. On the contrary, PI3Kb has both kinase-dependent and -independent functions as an effector of both RTKs and GPCRs [24]. Similar to PI3Ka, PI3Kb participates in canonical PI3K-AKT signaling, but also uniquely regulates DNA replication, cell cycle progression, nuclear envelope maintenance and nuclear pore complex assembly, autophagy, and apoptosis [24]. Despite PI3Kb rarely being mutated in cancers, PI3Kb expression correlates with poor prognosis in early-stage NSCLC patients [29]. It has also been proposed as a drug target in PTEN-deficient cancers [30] and in PI3Ka inhibitor-resistant PIK3CA mutant cancers [27]. Our discovery of PI3Kb as a target to enhance ALKi sensitivity in ALK-rearranged lung cancer is unexpected, as we here found combinatorial ALKi/PI3Kbi sensitivity independent of PTEN/PIK3CA mutations or resistance to PI3Ka inhibitors. Despite lack of evidence for a clear role of autophagy or GPCR signaling in the PI3Kb inhibitory responses identified here, PI3Kb is a downstream node for signaling directed by RTKs that are widely implicated in driving resistance to ALKi, particularly IGF-1R [31,32], ERBB2/3 [23,27] and GPCRs [23,33]. The targeting of PI3Kb may therefore counter a variety of bypass/acquired ALKi resistance mechanisms across sample types.   Hence, our findings warrant further testing of the efficacy of PI3Kbi drug combinations on a larger set of patient-derived samples. Due to the low success rate of establishing patient-derived primary cultures, as underlined in our previous study [20], we could use cultures generated from different regions of a single tumor specimen. Our findings in the patient-derived primary cultures were also confirmed in two commonly used ALK-rearranged conventional lung cancer cell lines. Previous research has identified various bypass signaling tracks that can mediate ALKi resistance, fostering the formulation of combinatorial treatment options that may overcome or even prevent drug resistance. The efficacy and tolerability of co-targeting ALK with MEK (NCT03202940, NCT03087448) or mTORC1 (NCT02321501) are currently tested in patients with ALK-rearranged NSCLC. In addition, following promising preclinical findings, the clinical testing of combinations co-targeting ALK with AXL, SRC, KIT, MET or SHP2 have been proposed [3,14]. Notably, the co-inhibition of MEK, mTOR, AXL, SRC, KIT, MET or SHP2 did not enhance ALKi responses in our study. Eventually, only those combinations that exhibit both efficacy and manageable toxicity profiles will be viable treatment strategies. Use of organ-matched healthy epithelial cells may help to exclude treatments that show generic toxicity [20]. For example, we detected sensitivity to HSP90 inhibition in both normal epithelial and cancer cell cultures, corroborating the finding that HSP90 inhibition induces toxicity without tumor-selective response in patients carrying ALK-rearranged NSCLC [34,35]. Similarly, we found that even though EGFR inhibition potentiated the effect of ALK inhibition in the ALKrearranged lung cancer cells, normal lung epithelial cells also showed a generic toxic response to cotargeting of EGFR, which aligns with severe toxicity in NSCLC patients receiving combinatorial treatments with ALKi and EGFRi [36,37]. On the contrary, PI3Kb inhibition, either on its own or combined with ALKi, did not detectably affect the normal lung epithelial cells. Our results suggest that EGFR and PI3Kb share common pathways toward resistance, indicating that co-targeting of ALK and PI3Kb can be a safe strategy to counteract EGFR-mediated adaptive resistance to ALKi. Promisingly, PI3Kb-selective inhibitors are being investigated in seven phase-I/II clinical trials as a single agent or in combination with chemoor immunotherapy (Table S4), following initial reports of satisfactory tolerability and efficacy [38][39][40].
From a heterogeneity standpoint, varied expression of lung adenocarcinoma markers, EMT, and tumor vasculature was observed across six different regions of the same tumor, suggesting that a single tumor biopsy would not provide a holistic assessment of this particular ALK-rearranged cancer. Given that the ALKi plus PI3Kbi combination was effective on cells derived from TP53 mutated tumor regions representing both epithelial and mesenchymal phenotypes, as well as established ALK-rearranged H3122 (TP53 mutant) and H2228 (TP53 and NFE2L2 mutant) cell lines, our findings suggest that this combination carries promise to counteract ALKi resistance associated with EMT, TP53 or NFE2L2 mutations [3,41]. The in vivo treatment evaluation revealed that the combination treatment may provide a long-term benefit, although the data included in the study lacked statistical significance and power of large cohorts. Therefore, to better evaluate the therapeutic potential of combination ALKi and PI3Kbi treatment, a sequential therapeutic approach and the use of a more resilient mouse strain, such as NMRI nu/nu, with a better tolerance for kinase inhibitors, may prove informative going ahead.

Conclusions
In summary, our study proposes PI3Kb as a novel regulator of ALKi sensitivity in ALK-rearranged lung cancer cells, showing efficacy even in aggressive TP53 mutant and mesenchymal cells. Given that multiple salvage signals are activated upon resistance to ALKi, the co-targeting of PI3Kb provides a promising therapy option as PI3Kb acts as a common effector of multiple salvage pathways, including EGFR, autophagy and GPCRs. Our data shows that this is the case for all clinically relevant ALK inhibitors. Importantly, we found that co-targeting of PI3Kb showed minimal toxicity in normal lung epithelial cells when compared with co-targeting of EGFR, possibly providing for a safe clinical direction to treat ALK-rearranged tumors. collection, the FIMM Digital Microscopy and Molecular Pathology Unit for scanning histological slides, the FIMM High-Throughput Biomedicine facility for drug screening resources, and the FIMM High Content Imaging and Analysis unit for imaging and data analysis for autophagy evaluation. We thank Antti Hassinen for

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article.  Fig. S2. ALK-rearranged lung cancer cells exhibit a range of ALKi sensitivities. Fig. S3. Effect of ALK and PI3Kb inhibition is specific for tumor cells. Fig. S4. Involvement of EGFR in dampening ALKi sensitivities. Fig. S5. ALKi and PI3Kbi combination is cancer cellselective. Fig. S6. PI3Kbi AZD-8186 increases ceritinib efficacy in ALK-rearranged lung cancer. Fig. S7. In vivo testing of ceritinib, AZD-8186 or their combination. Fig. S8. ALK inhibition leads to autophagy. Fig. S9. Inhibition of autophagy or P2Y receptors does not improve the response of ceritinib. Fig. S10. Combined inhibition of ALK and PI3Kb overcomes EGFR-mediated resistance in ALK-rearranged lung cancer cells. Table S1. Details of primary antibodies used in immunohistochemistry and western blotting analyses. Table S2. List of somatic mutations identified in tumor tissue and tumor-derived cells. Table S3. Drug library used for Drug Sensitivity and Resistance Testing. Table S4. Clinical trial information for PI3Kb inhibitors.