This study followed worldwide standard animal care condition via Institutional Animal Care and Use Committee (IACUC). The research proposal was approved by Yonsei University IACUC (2014-0249). All experimental mice were housed in colony cages and maintained on a 12-light:12 hours dark cycle in the Association for Assessment and Accreditation of Laboratory Animal Care International-certified specific pathogen-free facility. Conditional EML4-ALK transgenic mice were generated as described previously.19 Genotypes of EML4-ALK transgenic mice were confirmed by PCR with the following three primers: pCB-F: 5’-TGT CTG GAT CCC CAT CAA GC-3’, mEML4-intron-F-5’-TTA CCT GCT GTG CCA TCC TG-3’, EML4-R_common: 5’-GAA CTC GTG ACT CAA GAG CTG-3’. For tumorigenesis, mice were treated with tamoxifen twice a week. For tumorigenesis, treatment with tamoxifen is needed for activation of Cre-ERT2. Tamoxifen (Sigma Chemical, St. Louis, Missouri, USA) was given intraperitoneally twice at a 3-day interval, and all mice developed lung tumors after 1 week.

After tumor formation was confirmed on MRI, mice were treated with ALK inhibitor, ceritinib. Ceritinib was provided by Novartis Pharmaceuticals, and stored in −20ºC until use. Before treatment, ceritinib was diluted with 20% PEG400, 3% Tween-80, based on deionized water. The diluted ceritinib was homogenized and vortexed vigorously, and used within 24 hours. Anti-PD-1 (BioXcell, West Lebanon, New Hampshire) used was the RMP1-14 monoclonal antibody which reacts with mouse PD-1 also known as CD279. The drug was stored 4ºC until use and diluted with phosphate-buffered saline. Anti-PD-1 treatment was injected intraperitoneally twice a week at the dose of 200 µg per mouse.

The optimization of MRI was completed before the beginning of the study. For MRI, the mice were anesthetized with isoflurane in 100% oxygen. MRI protocols were optimized for assessing lung parenchyma at 9.4 T (BioSpec 94/20 USR MRI system (Bruker, Billeria, Massachusetts)). All tumors were analyzed on T2-weighted image. One week after intraperitoneal treatment with tamoxifen, tumor size was measured on MRI. Then TKIs were treated after the first MRI. The tumor measurement and response evaluation criteria are as following: (1) complete response (CR) is defined as complete disappearance of all tumor lesions, (2) partial response is defined as >30% shrinkage in the sum of tumor diameter from baseline, (3) progressive disease (PD) is defined as the growth of tumor size of >20% from the baseline (online supplementary figure S1). PFS was defined as the start of the drug treatment until the date of disease progression.

Tumor tissue was dissociated by collagenase, and then isolated cells were preserved in liquid nitrogen tank with frozen media until use. Samples were stained with the following antibodies: mCD3e (145-2C11; eBioscience), mCD4 (RM4-5; eBioscience), mCD8a (53-6.7; eBioscience), mCD25 (PC61.5; eBioscience), mCD44 (IM7; eBioscience), mGranzyme B (NGZB; eBioscience), mFoxP3 (FJK-16s; eBioscience), mF4/80 (BM8; eBioscience), mCD11c (N418; eBioscience), mCD11b (M1/70; eBioscience), mLy-6G (1A8-Ly6G; eBioscience). All samples were run on LSRFortessa (BD biosciences) and analyzed with FlowJo software (Tree Star).

Measurement of interferon (IFN)-γ and interleukin (IL)-12p40 in the serum of transgenic mice was performed using ELISA kit (Biolegend, San Diego, California, USA). Optical density reading at 450 nm were measured using a VERSAmax microplate reader (Molecular Devices). The cytokine concentration was normalized by protein concentration which has been measured by BCA kit (Thermo Fisher, USA).

Immunohistochemical protocols were performed using 4 mm thick sections, mCD8 antibody (Cell Signaling) as primary antibody. All slides were stained by Bond RX fully automated staining device (LEICA), scanned by Vectra Polaris (Perkin Elmer), and analyzed by Phenochart and InForm (Perkin Elmer) according to the manufacturer’s protocol. The anti-mPD-L1, anti-Pan-CK, anti-mCD8, anti-mFoxp3, anti-mPD-1 and anti-mGranzyme B antibodies were applied for multiplex immunohistochemistry. The InForm software was used for analysis tissue and cell segmentation. The individual cells including target expression were integrated and converted by R (V.3.6.1). The converted matrix files were analyzed for deconvoluted results by FlowJo software (V.10.6.2).

Total RNA was extracted from vehicle and ceritinib-resistant tumors using TRIzol RNA isolation reagents (Life Technologies, Carlsbad, California, USA). RNA integrity was analyzed using a bioanalyzer with the Agilent RNA 6000 pico kit (Agilent Technologies, Santa Clara, California, USA). The extracted total RNA was processed to prepare the mRNA-sequencing library using the TruSeq stranded mRNA LT sample preparation kit (Illumina, San Diego, California, USA) according to manufacturer’s instructions. All samples were sequenced on an Illumina HiSeq2500 sequencer using paired-end 100 bp reads. The raw image data were transformed by base-calling into sequence data and stored in FASTQ format. Paired-end reads of the five independent samples were trimmed for both PCR and sequencing adapters with Cutadapt (http://cutadapt.readthedocs.io/en/stable/; V.1.16). Trimmed reads were aligned to the GRCm38 mouse reference genome using STAR (V.2.6.0a)20 and gene-level read counts were generated using feature Counts function from the Subread package (V.1.6.2).21 The presence of significant differential expression was determined by DESeq2 package (V.1.26)22 at the gene level. All gene sets in V.7.0 of the Molecular Signatures Database were analyzed by V.4.0.3 of gene set enrichment analysis (GSEA) and corrected for multiple hypothesis testing. The p value threshold was set at 0.05.23 Heatmaps of differentially expressing genes was generated using the prism (V.8.0). To analyze TCR repertoire, the high-depth RNA sequencing was performed. TCR sequences were extracted by V.3.0 of MiXCR, and analyzed by TcR packages (V.2.2.4.1).

SureSelect (target enrichment system) was applied for exome capture sequencing process. The hybridized genome sample with biotinylated RNA library ‘BAITS’ was mixed with streptavidin-coated magnetic beads. The bead captured genomes were washed and RNA digest. The collected genome particles were analyzed by next-generation sequencing. The FASTQ file was generated by RTA V.1.12.4/CASAVA V.1.8.2. The FASTQ files were aligned on mm10 genome reference, then generated to BAM file. Extract mappable read, on-target reads, SNP and Indels were generated by analysis pipe lines. The read mapping was performed by BWA. The mark duplicates filter (Picard), Indel realignment (GATK) and base recalibration were performed for variant calling. The GATK was used for variant calling, variant filtering.

In sections stained with Masson’s trichrome stain, collagen staining was separately quantitated in the liver. Gomori’s silver impregnation methods for reticulin fibers were carried out on formalin-fixed, paraffin-embedded sections mounted on glass slides. Three slides per group were analyzed. The proportional area of blue stain in each section was recorded by using analytical imaging software, and the average from five fields was calculated to yield a relative value of collagen staining per field.

Survival analyses were estimated using Kaplan-Meier method and associated 95% CIs. All statistical analyses and graphs were performed using SPSS V.23.0 software (IBM, Chicago, Illinois, USA) and GraphPad Prism (V.8.0).

We designed a three-arm trial consisting of anti-PD-1 and ceritinib to evaluate the efficacy and toxicity of each treatment. The schematic design of this study is shown in figure 1A. Mice in each arm were allocated to the following treatment arms. arm A: anti-PD-1 followed by ceritinib (n=9), arm B: ceritinib followed by anti-PD-1 (n=11) and arm C: anti-PD-1 and ceritinib combination (n=6). When tumors showed progression in arm A and arm B (defined as PD¹), second-line therapy was designated as the following: anti-PD-1 was switched to ceritinib and was continued until progression (PD²), ceritinib was switched to anti-PD-1 and was continued until progression (PD²). During the drug treatment, MRI was performed every week to measure tumor size and objective response was determined as described in online supplementary figure 1.

The swimmer’s plot in figure 1B shows the treatment duration of three arms. Upfront ceritinib (arm B) or ceritinib and anti-PD-1 combination (arm C) clearly produced more durable responses compared with anti-PD-1 alone upfront. Representative MRI images of mouse tumors in each arm are shown in figure 1C. Next, we analyzed the survival outcome of each treatment in terms of PFS and OS (figure 1D). The median PFS¹ was 13, 132 and 129 days for arms A, B and C, respectively. As expected, arm B and arm C showed significantly prolonged PFS compared with arm A. The PFS¹ of arm B and arm C were not significantly different, suggesting that combination treatment did not prolong PFS in the frontline. The median PFS² was 99 days for arm A and 7 days for arm B, suggesting that anti-PD-1 treatment after ceritinib resistance was not effective. The median PFS¹+PFS² was 112, 139 and 129 days, respectively, confirming that upfront ceritinib was the most effective treatment regimen while combination of ceritinib and anti-PD-1 did not add any improvement in terms of PFS. These data were in concordance with the data from clinical trials summarized previously.10–12 In addition, we also observed significant hepatotoxicity resulting from ceritinib and anti-PD-1 combination treatment, characterized by fibrosis and infiltration of lymphoplasmacytic cells (online supplementary figure 2A‒D). There was also a significant reduction in body weight after day 20 in the combination-treated group compared with ceritinib group (online supplementary figure 2E). Next, we questioned whether combination treatment after failure on sequential treatment could overcome resistance to either sequential treatment (online supplementary figure 3). Anti-PD-1 was added on after progressing on ceritinib in arm A (n=5), ceritinib or anti-PD-1 was rechallenged in mice progressing on arm B and PFS³ was compared between these groups. There was no significant difference between PFS³ among three groups, and all regimens were not efficacious. Still, partial responses were observed in all mice which may have resulted from remaining ALK-dependent clones reactive to TKI. Altogether, anti-PD-1 treatment was ineffective in ALK-positive lung cancer, either as a single agent, or in combination with ALK TKI in delaying lung cancer progression.

To further investigate immunodynamics following treatment, we examined changes in the tumor immune microenvironment by performing flow cytometry of tumors and bronchoalveolar lavage (BAL) fluid. We compared three subsets of immune cells (T lymphocytes, antigen-presenting cells, myeloid-derived suppressor cells (MDSCs)) at baseline, and at the time of progression. Gating strategies are described in online supplementary figures 4 and 5. Overall, there were no notable changes among T lymphocytes, except regulatory T (Treg) cells (CD3⁺CD4⁺CD25⁺Foxp3⁺), which showed a significant increase after ceritinib resistance and after ceritinib followed by anti-PD-1 in arm B. Of note, granzyme B production did not alter, implying lack of cytolytic function by CD8⁺ T cells24 (figure 2A). Among antigen-presenting cells, macrophages (F4/80⁺) and dendritic cells (CD11c⁺/F4/80⁻) increased after anti-PD-1 or ceritinib or combination treatment (figure 2B). Among MDSCs, monocytic MDSCs (CD11b⁺Ly-6G⁻) were increased after anti-PD-1, and granulocytic MDSCs (CD11b⁺Ly-6G⁺) were increased after ceritinib followed by anti-PD-1 (figure 2C). We also noted changes in the cytokines obtained from BAL fluid (figure 2D), that IL-12p40 which upregulates IFN-γ secretion,25 was increased after anti-PD-1 followed by ceritinib. IFN-γ was increased after anti-PD-1 and ceritinib, which may result from immunogenic cell deaths following treatment26 or may be produced by increased antigen-presenting cells.27 Altogether, we presume that upregulation of IFN-γ following treatment may have stimulated development and recruitment of Tregs, MDSCs, tumor-associated macrophages and dendritic cells in tumor.27 28 To better compare the tumor immune microenvironment before and after treatment, we focused on pretreatment, postceritinib resistance, postcombination resistance samples for further experiments involving multispectral imaging, whole exome sequencing and RNA sequencing.

Next, we compared tumor immune microenvironment in the treatment-naïve and ceritinib-resistant and combination-resistant tumors to examine changes in the immune microenvironment. On multispectral imaging, treatment-naïve tumors showed paucity of tumor infiltrating T cells, characterizing the so-called ‘cold tumors’.29 Ceritinib-resistant and combination-resistant tumors showed dramatically increased PD-L1 staining cells, while changes in the expression of CD8, FoxP3, granzyme B and PD-1 staining cells were modest. A representative image is seen in figure 3. For quantification, we further examined the expression of CD8, Foxp3, PD-1, granzyme B, PD-L1-positive cells by deconvolution analysis obtained from multispectral imaging (figure 4A). When we compared expression levels of PD-L1, PD-1 and Foxp3 at pretreatment, postceritinib resistance and postcombination resistance, PD-L1 and Foxp3 were again dramatically increased after ceritinib and combination resistance, but not PD-1 expression (figure 4B). Phenotypic analyses revealed that PD-L1-expressing cancer cells were increased after ceritinib and combination treatment, while there were no significant changes in PD-1-expressing CD8⁺ T cells and granzyme B-expressing CD8⁺ T cells, suggesting that these CD8⁺ T cells may not be cytotoxic (figure 4C). Segmentation of tumors from tumor nest and stroma showed upregulation of PD-L1 staining cells both in the tumor nest and stroma after ceritinib resistance (figure 4D). A close up image of ceritinib-resistant tumor showed staining of CD8⁺ T cells surrounded by PD-L1-staining cells and Foxp3-positive Treg cells (figure 4E,F). Altogether, we assumed that resistance to ceritinib or combination treatment may lead to increased number of CD8⁺ T cells which are just bystanders.30

We have previously observed that ALK-positive tumors are resistant to anti-PD-1 treatment either upfront or after ceritinib resistance (figure 1). Treatment-naïve ALK-positive tumors were also devoid of CD8⁺ T cells (figure 3), suggesting that these tumors are non-immunogenic. To further explore immunological landscape of primary resistance to ICIs, we performed whole exome sequencing of treatment-naïve tumors and ceritinib-resistant tumors. ALK kinase domain of the ALK gene was directly sequenced to identify known resistant ALK mutations, but no known ALK mutation was found (data not shown). When the TMB, as calculated by mutations per mega-base, was compared, there was no significant difference between the two groups (figure 5A). As tumor antigen-specific CD8⁺ T cells expand clonally during immune response, we analyzed T-cell receptor (TCR) sequences to identify clonally expanded cells as an indicator of tumor specificity. We also compared the changes in CDR3 length which could affect TRB (which encodes TCRβ) sequences. TRB analysis revealed that relative usage frequency of TRBV and TRBJ gene was not significantly different between pretreatment and postceritinib tumor (online supplementary figure 6). We noted that variations in CDR3 length did not affect TCR repertoire (figure 5B, online supplementary figure 7). In addition, there was no significant change in clonality between pretreatment and postceritinib tumor (online supplementary figure 8). Principal component analyses of treatment-naïve and ceritinib-resistant tumors showed a large variation between the two groups (figure 5C). These findings suggest a lack of antigen specificity among CD8⁺ T cells in ceritinib resistance.

Furthermore, RNA sequencing analysis of pretreatment and ceritinib-resistant tumor was performed to gain a more in-depth understanding of the immune phenotype. Volcano plot comparisons between the pretreatment and postceritinib tumor identified 197 transcripts (145 genes up, 52 genes down) that were differentially expressed between the two cell populations (figure 6A). Network analysis showed that upregulated genes showed a higher clustering coefficient than downregulated genes (online supplementary figure 9). Among upregulated gene set, genes that respond to IFN-γ, as well as negative regulation of immune system process inflammatory response and monocyte chemotaxis were noted by GO term analysis (figure 6B,C). GSEA also revealed that gene transcripts associated with Treg cell differentiation (GO), classically activated macrophage versus type 2 activated macrophage up (GSE4811), and IL-4 versus IL-4 and dexamethasone-treated macrophage up (GSE7568) were significantly enriched in ceritinib-resistant tumors (figure 6D). Dexamethasone and IL-4-treated macrophages help conversion of M2 macrophage to M2c subtype, which plays a role of strong immune suppression and decreased antigen presentation.31 Of note, among the core enrichment genes described in figure 6E, genes involved in Treg cell function such as suppressor of cytokine signaling (SOCS1)32 showed a significant enrichment in ceritinib resistance. These data are concordant with the previous data in figures 2–4 that Treg cells are increased after ceritinib resistance.