The antitumor activity of osimertinib plus palbociclib in non-small cell lung cancer patient-derived xenograft (PDX)/2D/3D culture models harboring EGFR amplification and CDKN2A/2B homozygous deletions

Highlights

• •We created PDX/2D/3D models of non-small cell lung cancer without EGFR and ALK/ROS1 aberrations.
• •Next-generation sequencing data showed a high frequency of wildtype EGFR amplification and CDKN2A/2B homozygous deletion in the PDX models.
• •Data indicated that PDX/2D/3D models with wildtype EGFR amplification and CDKN2A/2B homozygous deletion are highly sensitive to osimertinib plus palbociclib in vivo and in vitro.
• •Osimertinib's antitumor activity decreased in oncogenic KRAS mutation model.
• •Amplification of the wildtype EGFR leads to increased osimertinib sensitivity.

Abstract

Non-small cell lung cancer (NSCLC) patients without targetable driver mutation have limited treatment options. In this study, we aimed to explore a new therapeutic strategy by using established nine patient-derived xenograft (PDX) and two-dimensional (2D) /3D culture models with specific genetic alternations. The gene mutations and copy number aberrations were detected by next-generation sequencing and confirmed using polymerase chain reaction (PCR) followed by DNA sequencing, and genomic DNA quantitative PCR. Protein expression was evaluated by immunohistochemistry. Drug sensitivities of PDX/2D/3D models were evaluated by in vivo and in vitro antitumor assays. RNA interference was performed to silence gene expression. Our study found that 44.4 % (4/9) of cases had CDKN2A homozygous deletion (homdel), while 33.3 % (3/9) had CDKN2B homdel. Additionally, 22.2 % (2/9) had amplification (amp) in wildtype CDK4, 44.4 % (4/9) in CDK6, and 44.4 % (4/9) in EGFR. Among the cases, 77.8 % (7/9) lacked CDKN2A, and 33.3 % (3/9) had high CDK4, CDK6, and EGFR had high protein expression. Moreover, 33.3 % (3/9) had KRAS mutations, and 66.7 % (6/9) had TP53 mutations. Antitumor activity of osimertinib plus palbociclib was assessed in four PDX/2D/3D models, two of which had simultaneous EGFR amp and CDKN2A/2B homdel. The data showed that NSCLC with EGFR amp and CDKN2A/2B homdel were sensitive to combined drugs. Additional oncogenic KRAS mutation reduced the drug's antitumor effect. EGFR amp is responsible for osimertinib sensitivity. Osimertinib plus palbociclib effectively treat NSCLC with wildtype EGFR and CDK6 amp and CDKN2A/2B homdel in the absence of oncogenic KRAS mutation.

Graphical abstract

PDX/2D/3D models were generated from non-small cell lung cancer without EGFR and ALK/ROS1 aberrations. We found that these PDX models have high frequencies of wildtype EGFR and CDK4/6 amplification and CDKN2A/2B homozygous deletion. Anti-cancer activity assays showed that PDX/2D/3D models with wildtype EGFR and CDK4/6 amplification and CDKN2A/2B homozygous deletion were highly sensitive to osimertinib plus palbociclib.

Introduction

Lung cancer is the most diagnosed cancer worldwide (12.4 %) and the leading cause of cancer-related deaths (18.7 %) globally [1]. Lung cancer can be histologically categorized into two main types: small cell lung cancer, which accounts for about ∼15 % of lung cancer cases, and non-small cell lung cancer (NSCLC), which accounts for ∼85 % of lung cancer cases. Recent advances in the early detection and approval of targeted therapies and immunotherapies have significantly improved the outcome of NSCLC [2]. In the United States, the life expectancy for a 65-year-old male lung cancer patient was 1.9 remaining life-years in 1980, and it increased to 3.0 remaining life-years in 2010 [3]. The 5-year relative survival rates for all stages of lung cancer increased from 12 % (1975–1977) to 25 % (2013–2019) in the United States [2]. Nonetheless, NSCLC patients without actionable driver mutations have limited treatment options (primarily chemotherapy with or without immunotherapy). NSCLC patients with advanced stages without EGFR mutations or ALK fusions had a median overall survival (OS) of 14.9 m when treated with chemotherapy alone for cases with PD-L1 expression ≧ 1 % and 12.2 m for cases with PD-L1 expression ≦ 1 %; the five-year OS rates were 14 % and 7 %, respectively [4]. Patients treated with dual immunotherapy (nivolumab plus ipilimumab) or nivolumab in combination with chemotherapy in cases with PD-L1 expression ≦ 1 % had slightly longer median OS at 17 m and 15 m, respectively [4]. The poor OS of these patients underscores the urgent need for novel treatments to improve their survival.

A dataset of 1,668 NSCLC molecular profiles deposited on the cBioPortal platform [5] shows that CDKN2A/2B homozygous deletion (homdel) and EGFR amplification (amp) are the three most common copy number aberrations, with incidence rates of 8.2 %, 7.9 % and 7.6 %, respectively. The proteins P16 and P15, which are encoded by CDKN2A and CDKN2B, respectively, inhibit the binding between cyclin-dependent kinases (CDK) 4/6 and cyclin D (CCND). CDK4/6 combines with CCND, together they phosphorylate and inactivate retinoblastoma (RB1) protein, thereby releasing the E2F transcription factors from RB1 and translocation of E2F into nucleus to activate genes promoting cell cycle progression from G1 to S phase. The loss of P16 and P15 proteins due to CDKN2A/2B homdel leads to uncontrolled cell cycle progression and tumor formation. This dysregulation has been found in various cancers, such as lung and breast cancer [6]. The FDA has approved inhibitors targeting CDK4/6, namely palbociclib, ribociclib, and abemaciclib, to treat human epidermal growth factor receptor 2 (HER2)-negative advanced breast cancer with dysregulated CDK4/6. Genetic variants in the TP53-CCND-CDK4/6-CDKN2A/2B-RB1 pathway, which is involved in cell cycle progression from G1 to S phase, have been detected in NSCLC [5]. Several preclinical and clinical studies have evaluated the effectiveness of CDK4/6 inhibitors in combination with other inhibitors, chemotherapy agents, radiotherapy, and immunotherapy for treating NSCLC, and promising results have been obtained [7]. However, CDK4/6 inhibitors are not approved for treating NSCLC. The FDA has approved first- to third-generation EGFR tyrosine kinase inhibitors, such as gefitinib, erlotinib, afatinib, and osimertinb, for the treatment of NSCLC with EGFR mutations, including L858R, exon 19 deletions, T790M, etc. The antitumor activity of osimertinib has not been reported in NSCLC patients harboring wildtype EGFR amp and, therefore, is not approved for treating this type of NSCLC patients. Gregorc and co-authors reported a clinical trial using erlotinib to treat wildtype EGFR NSCLC and found that it was more effective than chemotherapy in a subpopulation with a poor prognostic signature [8]. Toffalorio and co-authors analyzed the effect of erlotinib in NSCLC with wildtype EGFR amp, chromosome 7 trisomy or polysomy (EGFR is located on chromosome 7), and KRAS mutations. The results showed that NSCLC with chromosome 7 polysomy has a higher median PFS [9]. However, due to the small number of cases, more experiments are still needed to confirm the treatment effect of EGFR tyrosine kinase inhibitors on wildtype EGFR amp+ NSCLC and those with coexisting genetic abnormalities (such as KRAS mutations).

Tumor cell lines cultured in a two-dimensional (2D) format have traditionally assessed drug antitumor activity. However, despite its cost-effectiveness and efficiency, nearly 90 % of drugs showing positive antitumor activity in 2D cell cultures fail in clinical trials [10]. Patient-derived xenograft (PDX) models more accurately mimic human biological and pathological microenvironments [11] but are time-consuming and costly. Recently, three-dimensional (3D) cell culture models have emerged as a bridge between 2D and PDX models [12]. However, the fidelity of 3D culture models in assessing drug antitumor activity still requires more research.

This study established PDX/2D/3D cell models from EGFR and ALK/ROS1 mutation-negative NSCLC. These models allowed us to evaluate novel drug combinations. Next-generation sequencing (NGS) identified gene mutations, copy number variants (CNVs), and structural variants in our NSCLC models. By analyzing genetic variants through the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, we found enriched CNVs and single nucleotide variants (SNVs) related to cell cycle control and EGFR-KRAS signaling in our NSCLC PDX models. We evaluated the in vivo and in vitro antitumor activities of palbociclib and osimertinib using our PDX/2D/3D models. Finally, we performed RNA interference in cell lines to demonstrate the relationship between wildtype EGFR expression levels and osimertinib's antitumor activity.

Materials and methods

Patients

Between November 2015 and November 2022, we enrolled 60 histologically diagnosed advanced NSCLC patients (stages IIIb ∼ IVc) who did not have EGFR or ALK/ROS1 mutations. Human sample collection was performed in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board of Chang Gung Memorial Hospital. IRB No.: 201509972A3 (20150301∼20170630), 1812180001 (20190801∼20210731), 2006120001 (20201101∼20231031), 201801935A3 (20190101∼20211231), 1912170058 (20200801∼20210731). Written informed consent was obtained from all patients.

PDXs

Tumor specimens were cut into small pieces (3–5 mm³ cubes) and implanted subcutaneously into the flank tissue of 6- to 8-week-old male NOD/SCID (NOD.CB17-Prkdc^scid/NcrCrl; BioLASCO Taiwan Co., Ltd) or NPG (NOD.Cg-Prkdc^scid Il2rg^tm1Vst/Vst; BioLASCO) mice. Tumor size was measured weekly using calipers. The mice were euthanized when tumors reached 10–15 mm in diameter. Tumor masses were harvested and subsequently transplanted into new mice. PDX models passaged at least three times were considered to be successfully established.

2D and 3D cell cultures

2D cell cultures were generated from tumors at xenografts' third to fourth passages. Tumor samples were finely minced and cultured in ACL-4 medium [13,14] with some modifications and supplemented with 10 % fetal bovine serum. Modifications included replacing RPMI 1640 medium with DMEM:F12 and adding epidermal growth factor (50 ng/mL), basic fibroblast growth factor (20 ng/mL), and B-27 (1 ×) (Thermo Fisher Scientific, MA, USA). Tumor cells were cultured at 37°C and 5 % CO₂ on collagen I-coated culture dishes (Corning BioCoat, Corning, ME, USA). According to the manufacturer's instructions, 2D cell cultures were tested for mycoplasma contamination using MycoAlert™ PLUS Mycoplasma Detection Kit (Lonza, Basel, Switzerland). A ratio < 1 was considered to be free of contamination.

We used the scaffold-based method to generate 3D cultures. Briefly, 100 tumor cells in 50 μL modified ACL-4 medium were mixed with 50 μL Geltrex^TM (LDEV-Free Reduced Growth Factor Basement Membrane Matrix, Gibco, Paisley, UK) and seeded into ultra-low attachment flat-bottom 96-well plates (Costar, Corning). The plates were incubated at 37°C and 5 % CO₂ for 10 min until matrix solidification, then 200 μL modified ACL-4 medium was added on the top of the matrix. The liquid medium was replaced with fresh medium every two days until 3D cell aggregates formed.

Genomic DNA and total RNA preparation

Genomic DNA was prepared from the freshly excised tumor mass, peripheral blood mononuclear cells, and tumor cell line using QIAamp DNA Mini Kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions. Total RNA was extracted from tumor mass using TRIzol Reagent (Thermo Fisher) according to the manufacturer's instructions, except that 1-bromo-3-chloropropane was substituted for chloroform during phase separation. RNA in the upper aqueous layer was precipitated by ethanol. The concentration and purity of genomic DNA and total RNA were determined using Qubit dsDNA HS Assay Kit and Qubit RNA HS Assay Kit (Invitrogen, Carlsbad, CA, USA), respectively, or using a NanoDrop spectrophotometer ND-1000 (Thermo Fisher).

Whole exome sequencing (WES)

Genomic DNA (50 ng) was subjected to enzymatic fragmentation and examined using a fragment analyzer Agilent 5300 (Agilent Technologies, Palo Alto, CA, USA). The Twist Exome 2.0 kit (TWIST Biosciences, San Francisco, CA, USA) captured the whole exome for library construction. Libraries were sequenced on an Illumina NovaSeq 6000 System (Illumina, San Diego, CA, USA). Sequence reads from PDX (fastq format) were filtered to remove any originating from the mouse using the Xenome tool (Conway 2012). The remaining human sequence reads were analyzed using the Dragen Bio-IT Platform v4.0-DNA pipeline (Illumina) against the Illumina DRAGEN Multigenome Graph Reference-hg38. We compared variants in PDX/human tumors and normal peripheral blood mononuclear cells to identify SNVs, insertions/deletions (InDels), and CNVs.

RNA-sequencing

Total RNA (100 ng) was used to prepare libraries with the KAPA RNA HyperPrep kit (Roche Molecular Diagnostics, Basel, Switzerland) and then sequenced on the NovaSeq 6000 system (Illumina). Sequence reads from PDX tumor masses were filtered using the Xenome tool to exclude any reading from the mouse. The remaining reads were processed by the Dragen Bio-IT Platform v4.0-RNA pipeline (Illumina) and aligned to the Illumina DRAGEN Multigenome Graph Reference-hg38 for fusion calling.

Gene functional annotation analysis

Genes with mutation, copy number aberration (CN ≧ 4 or = 0), and fusion in PDX detected by WES and RNA-sequencing analyses were functionally annotated through KEGG pathways analysis using online Database for Annotation, Visualization and Integrated Discovery (DAVID) v6.8 annotation tools (https://david.ncifcrf.gov/). The KEGG pathways with p < 0.05 were considered significantly enriched.

Genomic DNA-quantitative polymerase chain reaction (gDNA-qPCR), Reverse-transcriptase qPCR (RT-qPCR), PCR followed DNA-sequencing, and immunohistochemistry (IHC)

Standard protocols were used to perform these assays. SYBR-Green PCR Master Mix (Applied Biosystem; Thermo Fisher) was used for qPCR, and the product was analyzed by the ABI Prism 7900 system (Applied Biosystem). Primer sets for gDNA-qPCR was as follows: CCND1, 5′-CTG CGA AGT GGA AAC CAT CC-3′ and 5′-CTT AAG AGA GCC GCC CGA AG-3′; CCND3, 5′-CAT CTA CAC CGA CCA CGC-3′ and 5′-ATC ACA CCA CAA TGC CCC AT-3′; CDK4, 5′-CCC GAA GTT CTT CTG CAG TCC-3′ and 5′-GCG ATT TGG GGA ATT CAA GGT-3′; CDKN2B 5′-AAC GGA GTC AAC CGT TTC GG-3′ and 5′-CGT GGA ATG CAC ACC TCC G-3′; CCNE1, 5′-AGC GGT AAG AAG CAG AGC AG-3′ and 5′-CGC TGC AAC AGA CAG AAG AG-3′; CDK6, 5′-CAG CAG TAC GAA TGC GTG G-3′ and 5′-CGC TTC AAC GCC ACG AAA C-3′; CDKN2A, 5′-CTC TCA CCC GAC CCG TG-3′ and 5′-TAT GCG GGC ATG GTT ACT G-3′; EGFR, 5′-TGG TGA AAA CAC CGC AGC AT-3′ and 5′-CTC CTT CTG CAT GGT ATT CTT TC-3′. Primer set for RT-qPCR was as follows: EGFR, 5′-TGG TGA AAA CAC CGC AGC AT-3′ and 5′-CTC CTT CTG CAT GGT ATT CTT TC-3′. The primer sets serving as internal control for gDNA-qPCR were BCAS3, 5′-CCA AGA AGA CCC AGT CGT TGT-3′ and 5′-TTC ATC CCA TAA AGC CAA CAG TTA-3′; for RT-qPCR was: ABL, 5′‑TGG AGA TAA CAC TCT AAG CAT AACTAA AGG‑3′ and 5′‑GAT GTA GTT GCT TGG GAC CCA‑3′. Primer sets for RT-PCR followed by DNA sequencing were: KRAS, 5′-ATG ACT GAA TAT AAA CTT GTG-3′ and 5′-AGG CAT CAT CAA CAC CCT GTC TTG T-3′; TP53, 5′-GTA CTC CCC TGC CCT CAA C-3′ and 5′-AGA AGT GGA GAA TGT CAG TC-3′. DNA sequencing primers were underlined. The primary antibodies employed in the IHC were: CDKN2A/p16INK4a mouse monoclonal Ab, JC2 (1:50; GeneTex, Texas, USA); CDK4 rabbit monoclonal Ab, D9G3E (1:200; Cell signaling, Beverly, MA, USA); CDK6 rabbit monoclonal Ab, EPR4515 (1:100; Abcam, Cambridge, MA, USA); EGFR rabbit monoclonal Ab, EP22 (1:50; Cell Marque, Roklin, CA, USA); TTF1/NKX2-1 rabbit monoclonal Ab, A3292 (1:100; ABclonal, Boston, MA, USA); CK7 rabbit polyclonal Ab, A2574 (1:200; ABclonal). According to the manufacturer's instructions, primary antibody binding to tissue sections was visualized using the BOND Polymer Refine Detection System (DS9800, Leica Biosystems, Vista, CA, USA).

Drugs and anticancer activity assay

For in vivo anticancer activity assay, osimertinib (80 mg/tab, Tagrisso; AstraZeneca, Cambridge, UK) and palbociclib (125 mg/cap, Ibrance; Pfizer, New York, NY, USA) were administered to mice. Briefly, when the tumor volume of the sixth-generation PDX reached approximately 150 mm³, the mice were randomly divided into four groups (n = 5∼7 for each group). Vehicle (sterile water), palbociclib (100 mg Kg^-1), osimertinib (5 mg Kg^-1), and palbociclib plus osimertinib were administered via oral gavage TIW (3 times a week). Mouse body weight and tumor volume were monitored TIW before oral gavage. Tumor volume was calculated using the formula: length × width² × 0.5 (mm³) [15,16]. The data presented was mean ± standard deviation (SD). Statistically significant differences were calculated by Student t-tests. p < 0.05 was considered statistically significant.

For in vitro antitumor assay, 2D and 3D cells were treated with osimertinib (HY-15772; MedChemExpress, NJ, USA) and/or palbociclib (HY-50767; MedChemExpress) and the cell viability was determined by Cell Counting Kit-8 (Dojindo Molecular Technologies, Kumamoto, Japan) according to the manufacturer's instruction. Briefly, 2D cell culture (5 × 10³ cells/well) was cultured in the 50 μL modified ACL-4 medium containing 10 % fetal bovine serum overnight until fully adhered to the plate well. Then, 50 μL fresh medium was added containing gradient concentrations of palbociclib (i.e., 0, 10, 20, 30, 40, 50, 60, 70 μM), osimertinib (i.e., 0, 10, 20, 30, 40, 50, 60, 70 μM), or palbociclib (24 μM) plus osimertinib (8 μM) and incubated at 37°C and 5 % CO₂ for 48 h. To determine the cytotoxicity of 3D cell aggregates produced in the scaffold-based method, the suspension medium was replaced by100 μL of modified ACL-4 medium containing palbociclib (24 μM) and/or osimertinib (8 μM) and incubated at 37°C and 5 % CO₂ for an additional 48 h. Afterward, Cell Counting Kit-8 reagent was added to each well and incubated for 2 h (2D) or 3 h (3D). OD₄₅₀ and OD₆₅₀ of the solution were read in the Multiskan^TM GO Microplate Spectrophotometer (Thermo Scientific). Readings obtained from cells incubated in drug-free media represented 100 % survival. Conversely, readings from cell-free media were considered to represent 0 % survival. After Cell Counting Kit-8 assay, eight to 14 3D cell aggregate images were taken using a Nexcope INVERTED Microscope (Ningbo, China).

RNA interference

To generate stable EGFR knockdown cell lines, H9 and A18 cell lines were transfected with plasmid pLKO.1 expressing shRNA against EGFR (The RNAi Consortium Number (TRCN) 0000039634 and TRCN0000121068) and selected in modified ACL-4 medium containing puromycin (2.5 μg/ml) for a total of 2 weeks. Cell lines stably transfected with blank lentiviral vector pLKO.1 were used as negative controls. All the lentiviral vectors with/without shRNA coding DNA were obtained from the National RNAi Core Facility at the Institute of Molecular Biology/Genomic Research Center, Academia Sinica (Taiwan).

Ethical approval

Animal care was conducted in accordance with the recommendations of the Guide for the Care and Use of Laboratory Animals published by the National Institute of Health, USA. The Animal Research Committee of Chang Gung Memorial Hospital approved all experimental procedures involving laboratory animals.

Data availability/Availability of data and materials

The data presented in this study are available on request from the corresponding author.

Results

Establishment of PDX/2D/3D models from NSCLC without EGFR and ALK/ROS1 aberrations

A total of 60 NSCLC patients were enrolled in this study. The tumor locations were lymph nodes in two cases, kidney in one case, pleura in 27 cases, lungs in nine cases, pleural effusions in five cases, brain in 14 cases, bone marrow in one case, and bone in one case. We established nine PDX models from these patients’ tumors: one from the lymph node, one from the kidney, three from the pleura, one from the lung, and three from the brain (Table 1 and Supplementary Fig. S1A). The rate of successful generation of PDX from NSCLC tumors was 15 %. The gender, age, smoking status, survival, histology, stage, TNM classification, and treatment prescription before and after tumor resection of these nine cases are listed in Table 1. In summary, six cases were female, and three were male; six cases were younger than 64 years old, and three cases were older than 65 years old; five cases were smokers and four non-smokers; the survival time ranged from 26.5 m to 1.7 m (median 3.9 m); five cases were diagnosed as adenocarcinoma; two cases were from NSCLC stage IIIc, 3 were from stage IVa, and four were from stage IVc. Before tumor resection, five cases received cisplatin-based doublet chemotherapeutic regimens, of which two received concurrent chemoradiotherapy. Pemetrexed alone was used in one case, afatinib in another case, and no treatment was given in two cases. After tumor resection, three cases received cisplatin-based doublet chemotherapy, of which two added Avastin; EGFR tyrosine kinase inhibitors (erlotinib or osimertinib) were used in two cases; one case was treated with immunotherapy pembrolizumab.

Table 1.

Clinical data of NSCLC patients who successfully established PDX models.

Code	Gender	Age	Smoker	Survival (m)	Histology	Stage	TNM	(Potential) driver mutation	Tumor site	Therapy before sampling for PDX	Therapy after sampling for PDX
H9	female	38	no	26.5	Adenosquamous cell carcinoma	IVc	T1N3M1c	EGFR(4), CDK6(4), CDKN2A/2B(0)	kidney	bevacizumab + cisplatin + pemetrexed 9.6 m	erlotinib 8.8 m; pemetrexed + carboplatin + avastin x4, avastin x5, erlotinib 8.1 m
H10	female	51	no	1.7	Adenocarcinoma	IVc	T2N3M1c	KIF5B-RET, CDK4(6)	Lymphnode	-	cisplatin + docetaxel 0.9 m
A14	female	41	yes	3.9	Large cell neuroendocrine	IVc	T4N3M1c	BRAF(4), NTRK1(5), TP53-G154V	brain	cisplatin + pemetrexed 2 m; cisplatin + docetaxel 1.9 m; gemcitabine 1 m	gemcitabine 2.9 m
A18	female	61	no	17.5	Adenocarcinoma	IVa	T4N3M1a	EGFR(4), KRAS-G12D, CDKN2A/2B(0)	lung	afatinib 15.1 m	osimertinib 1.7 m
A38	female	68	yes	2.9	Non small cell carcinoma	IIIc	T4N3M0	KRAS-G12C, CDKN2A(0)	pleura	cisplatin + pemetrexed + pembrolizumab 1 time	-
A41	female	73	no	3.0	Adenocarcinoma	IVc	T4N3M1c	KRAS-Q61K, CDKN2A/2B(0)	pleura	-	pembrolizumab 1 time
A47	male	39	yes	6.5	Non small cell carcinoma	IIIc	T4N3M0	NTRK3-A591E & -G763H, ERBB2-V314A, CDKN2A-G74R, TP53-V225F, CDK6(5)	pleura	radiation + cisplatin + pemetrexed 4.9 m	-
A50	male	67	yes	3.0	Adenocarcinoma	IVa	T4N2M0	EGFR(4), BRAF(4), MET(5), CDK6(5), CDKN2A(0)	brain	pemetrexed 6 d	pemetrexed + cisplatin + avastin 1 m
A58	male	61	yes	7.7	Adenocarcinoma	IVa	T4N3M1a	MET(21), CDK6(9)	brain	radiation + cisplatin + docetaxel 2.2 m	-

m, month

Using tumors from PDX passages 2‒4, we successfully established 2D cell lines for all PDXs (Supplementary Fig. S1B). All 2D cell lines except for A47 generated 3D cell cultures (aggregates) (Supplementary Fig. S1C).

NGS analysis and KEGG pathway enrichment analysis of PDXs

We performed RNA-sequencing and WES on all PDXs and selected patient tumors and cell lines. H10 PDX is the only one that carries an oncogenic fusion gene—KIF5B-RET. The number of SNVs and InDels detected in these PDXs ranged from 223 ‒ 920. The number of chromosome segments with CNV ranged from 130 ‒ 1,564. We compared CNVs between paired PDXs and cell lines or paired human tumors and PDXs. The data showed slight differences between paired samples (Supplementary Fig. S1D), suggesting limited clonal selection during PDX and cell line establishment.

The (potential) CNVs, SNVs and fusion genes involved in NSCLC development in our PDXs are shown in Table 1. To more comprehensively annotate the functional pathways of genetic variants in each PDX, we used DAVID Bioinformatics Resources v6.8 tools to analyze all CNVs, SNVs, and fusion genes. All KEGG pathways with a p-value < 0.05 in the nine PDXs are listed (Supplementary Table S1). The 10 most common KEGG pathways among these nine PDXs are shown in Fig. 1A. The most common KEGG pathway is “hsa04510:Focal adhesion”, enriched in five of the nine PDXs. We found that five of the 10 common KEGG pathways—“hsa04510:Focal adhesion”, “hsa04151:PI3K-Akt signaling pathway”, “hsa05165:Human papillomavirus infection”, “hsa05200:Pathways in cancer”, and “hsa04810:Regulation of actin cytoskeleton” (Fig. 1A) contained genes involved in cell cycle control, primarily in the progression from G1 to S phase, and (receptor) tyrosine kinase signaling pathways, including RET, EGFR, and KRAS (Supplementary Table S1). The SNVs and CNVs of cell cycle and signaling-related genes in PDXs are shown in Table 2.

Fig. 1.

KEGG pathway enrichment analysis of genetic variants in PDXs and confirmation of variants. (A) Ten common KEGG pathways are significantly enriched (p <0.05) in nine PDXs. Five KEGG pathways containing genes related to cell cycle progression and EGFR-KRAS signaling are marked in blue. (B ‒ E) Relative copy numbers of CDKN2A (B), CDKN2B (C), CDK6 (D), and EGFR (E) in nine PDXs determined by gDNA-qPCR. gDNA-qPCR experiment for each gene was performed in triplicate. NC, normal control.

Table 2.

Genetic aberrations involved in cell cycle control and EGFR signaling pathway in nine PDXs.

Gene	Aberration	%	H9-PDX	H10-PDX	A14-PDX	A18-PDX	A38-PDX	A41-PDX	A47-PDX	A50-PDX	A58-PDX
CDKN2A	SNV	7.2	-	-	-	-	-	-	G74R	-	-
CDKN2B	SNV	0.3	-	-	-	-	-	-	-	-	-
CDK4	SNV	0.7	-	-	-	-	-	-	-	-	-
CDK6	SNV	0.3	-	-	-	-	-	-	-	-	-
CCND1	SNV	0.5	-	-	-	-	-	-	-	-	-
CCND3	SNV	0.5	-	-	-	-	Q179X	-	-	-	-
CCNE1	SNV	0.8	-	-	-	-	-	-	-	-	-
TP53	SNV	55	-	-	G154V	R158L+G154V	E271Q	-	V225 splice	P278R	R158L
KRAS	SNV	26.9	-	-	-	G12D	G12C	Q61K	-	-	-
EGFR	SNV	24.5	-	-	-	-	-	-	-	-	-
CDKN2A	HOMDEL	8.2	0 (0)	1 (1)	2 (2)	1 (0)	1 (0.15)	2 (0)	2 (2)	1 (1)	1 (1.5)
CDKN2B	HOMDEL	7.9	0 (0)	1 (1)	2 (2)	1 (0)	1 (1)	2 (0)	2 (2)	1 (1)	1 (1)
CDK4	AMP	4	2 (4)	14 (20)	2 (3)	2 (4)	2 (6)	3 (3)	2 (3)	2 (2)	3 (3.6)
CDK6	AMP	0.3	4 (4)	3 (3)	3 (2)	3 (3)	2 (2)	2 (2)	5 (5)	5 (4)	9/5 (6)
CCND1	AMP	2.5	1 (3)	2 (4)	2 (2)	3 (3)	2 (2)	9 (1.5)	3 (2.5)	3 (3)	2 (2)
CCND3	AMP	0.9	5 (5.4)	2 (1.4)	3 (2)	2 (1.5)	2 (2)	2 (2)	3 (4)	2 (1.6)	2 (2)
CCNE1	AMP	2.3	2 (2)	4 (5)	3 (2)	1 (2)	2 (2)	2 (2)	7 (8)	1 (3)	3 (4)
TP53	HOMDEL	0.4	2	1	3	3	2	3	2	1	1
EGFR	AMP	7.6	4 (8)	2 (3)	2 (3)	4 (5)	2 (2)	4 (2)	3 (3)	4 (2)	1 (1)
KRAS	AMP	1.6	2	2	2	2	4	5	2	2	2
KIF5B-RET	FUS	1.2		+

Bold type indicates mutation is 100 % presence in PDX

Blue indicates genes involved in cell cycle progression from G1 to S phase

Red indicates genes involved in the (receptor) tyrosine kinase signaling pathway

() represents the relative copy number determined by gDNA-qPCR

SNV, single nucleotide polymorphism

HOMDEL, homozygous deletion

AMP, amplification

FUS, fusion gene

After verification with gDNA-qPCR, we found that in addition to H9, CDKN2A/2B homdel was also detected in A18 and A41 PDXs; A38 PDX had a low copy number (0.15) of CDKN2A (Fig. 1B ‒ 1C). H10 PDX had ∼20 copies of CDK4, and A38 PDX also had six copies of CDK4 (Supplementary Fig. S2). H9, A47, A50, and A58 PDXs had four, five, four, and six copies of CDK6, respectively (Fig. 1D). H9 PDX had five copies of CCND3; A47 PDX had eight copies of CCNE1 (Supplementary Fig. S2A). H9 and A18 PDXs had eight and five copies of wildtype EGFR, respectively (Fig. 1E). A41 and A50 PDXs showed conflicting data, WES showed four copies of EGFR, but gDNA-qPCR showed a normal copy number (Table 2 and Fig. 1E). Using RT-PCR followed by DNA sequencing, we also confirmed several TP53 and all KRAS gene mutations in PDXs (Supplementary Fig. S2B and S2C). Our data showed that CNVs detected by WES are not always consistent with gDNA-qPCR. On the other hand, compared to NSCLC profiles in the cBioPortal platform, our PDXs (without EGFR and ALK/ROS1 aberrations) showed exceptionally high frequencies of CDKN2A and CDKN2B homdel and wildtype EGFR and CDK6 amp—8.2 % vs. 44.4 %, 7.9 % vs. 33.3 %, 7.6 % vs. 22.2 %, and 0.3 % vs. 44.4 %, respectively (Table 2).

Protein levels of cell cycle control genes and EGFR

To further confirm the expression levels of CNVs in PDXs, we performed IHC to determine protein levels of the cell cycle control and EGFR genes. The results showed that consistent with CNV or gDNA-qPCR data, H9, A18, A38, and A41 PDXs lacked CDKN2A protein expression in IHC analysis; Surprisingly, A47, A50, and A58 also lacked CDKN2A protein expression (Fig. 2). H10 and A38 PDXs had high protein levels of CDK4 expression; A14 also showed high CD4 protein level (Fig. 2). H9 and A50 (but not A47 and A58) PDXs had high CDK6 expression; additionally, A14 had high CDK6 expression (Fig. 2). H9, A18, and A50 (but not A41) PDXs had high wildtype EGFR expression (Fig. 2). The expression of CK7 and TTF-1 prove that these PDXs originate from the lung (Fig. 2).

Fig. 2.

Measuring gene expression levels with IHC analysis. Protein amounts of CDKN2A, CDK4, CDK6, EGFR, CK7, and TTF-1 were determined by IHC analysis of 4 μM thick formalin-fixed, paraffin-embedded xenograft tumor sections.

In vivo and in vitro anticancer activity of palbociclib and osimertinib in PDX/2D/3D models

Our data showed that H9 and A18 PDXs lacked CDKN2A/2B and had high wildtype EGFR expression. Furthermore, H9 PDX had high CDK6 expression, whereas A18 PDX had oncogenic TP53 and KRAS mutations. Meanwhile, H10 PDX with KIF5B-RET and high CDK4 expression and A14 PDX with high CDK4 and CDK6 expression and oncogenic TP53 mutation were used as controls. We evaluated whether the CDK4/6 inhibitor palbociclib and the third-generation EGFR inhibitor osimertinib could inhibit tumor growth in vivo. Our results showed that H9 PDX mice treated with palbociclib, osimertinib, and combined drugs experienced a significant reduction in tumor size (all with p < 0.001); in particular, tumor size in mice treated with the combined drugs did not increase at all during the experimental period (Fig. 3A-i). H10 PDX mice treated with either palbociclib, osimertinib, or a combination of drugs failed to reduce tumor growth (Fig. 3A-ii). Treatment of A14 PDX mice with palbociclib significantly reduced tumor growth (p < 0.05). Although osimertinib alone did not significantly reduce A14 tumor growth, palbociclib plus osimertinib further reduced tumor growth (p < 0.005) (Fig. 3A-iii). In A18 PDX mice, only combined drug treatment shows a significant reduction in tumor growth (p < 0.05), with palbociclib and osimertinib alone showing efficacy in reducing tumor growth but not reaching statistical significance (Fig. 3A-iv). The size of the resected tumor mass further supports the in vivo tumor size data (Fig. 3B). In summary, PDX mice with CDKN2A/2B- and high CDK6 are highly sensitive to palbociclib; PDX mice with high CDK4 and CDK6 are sensitive to palbociclib despite TP53 mutation; PDX mice with CDKN2A/2B- and TP53 mutation have reduced sensitivity to palbociclib. PDX mice with high EGFR expression are highly sensitive to osimertinib; however, PDX mice with high EGFR expression and oncogenic KRAS mutation have reduced sensitivity to osimertinib. PDX mice with KIF5B-RET are insensitive to palbociclib despite high CDK4 expression.

Fig. 3.

In vivo and in vitro anticancer activity analyses using PDX/2D/3D models. (A) H9- (i), H10- (ii), A14- (iii), and A18- (iv) PDX mice were administered palbociclib (Pal, 100 mg kg^-1) alone, osimertinib (Osi, 5 mg kg^-1) alone, combined drugs (Pal+Osi), or distilled water (NC, without drug) by oral gavage three days a week until the tumor sizes of NC-treated PDX mice reached 1500 mm³∼2000 mm³. Tumor volume was calculated using the formula: length × width [2] × 0.5 (mm³). The data presented were mean ± SD. Statistical significance between the final data was calculated using a Student's t-test. *, p < 0.05; **, p < 0.01; ***, p < 0.005. (B) The PDX mice's tumors were removed and photographed at the end of the experiment. (C) In vitro drug sensitivity of H9-, H10-, A14-, and A18-PDX-derived 2D cell lines treated with Pal (i) or Osi (ii) at the indicated concentrations (0∼35 μM) and Pal (12 μM) plus Osi (4 μM) (iii) was measured using Cell Counting Kit-8 assay. Assays were performed in triplicate, and the data are representative of two independent experiments. Error bars represent SD. LC₅₀ values for cell lines treated with Pal or Osi alone were shown. (D) In vitro drug sensitivity of non-drug-treated (NC) and drug-treated H9-, H10-, A14-, and A18-3D cell aggregates was assessed by (i) Cell Counting Kit-8 assay and compared by (ii) morphologic characteristics using an inverted microscope Eclipse TS100 (Nikon, Tokyo, Japan). Cell Counting Kit-8 assays were performed in triplicate, and the data are representative of two independent experiments. 3D cell aggregate images are representative from each cell line cultured in the matrix-containing media. Statistical significance between drug-treated and non-drug-treated (NC) cells was calculated using a Student's t-test.*, p < 0.05; **, p < 0.01; ***, p < 0.005. Pal, palbociclib; Osi, osimertinib.

We also evaluated the drug sensitivity of 2D cell cultures of H9, H10, A14, and A18. The results showed that the sensitivity of the cell lines to palbociclib was H9 > A18 > A14 > H10 according to LC₅₀ (Fig. 3C-i). On the other hand, the sensitivity of cell lines to osimertinib was H9 > A14 > A18 > H10 according to LC₅₀ (Fig. 3C-ii). We performed combined drug sensitivity assays on 2D (and 3D) cell cultures using the LC₅₀ drug concentration of H9—12 μM palbociclib and 4 μM osimertinib. The drug sensitivity of 2D cell lines showed H9 and A14 were significantly sensitive to palbociclib and osimertinib alone and combined drugs (all p < 0.005); A18 was significantly sensitive to palbociclib alone and combined drugs (p < 0.01 and p < 0.005, respectively); H10 cell line did not reach statistical significances for any treatments (Fig. 3C-iii). Our results showed that the responses of 2D cell lines treated with palbociclib and osimertinib alone and combined drugs were quite similar with PDXs, except that A18 cell line was sensitive to palbociclib but not A18 PDX mice and A14 cell line was sensitive to osimertinib but not A14 PDX mice. These results indicate that 2D cell lines are more sensitive to drug treatment compared with PDXs.

The drug sensitivity of 3D cell aggregates showed H9 and A18 were significantly sensitive to palbociclib (12 μM) and osimertinib (4 μM) alone and combined drugs (all p < 0.05); H10 and A14 were significantly sensitive to combined drugs (both p < 0.05) (Fig. 4D-i). Morphological observations of 3D cell aggregates further supported the results of the in vitro drug sensitivity assay (Fig. 4D-ii). Our results showed that the responses of 3D cell lines to drug alone and combined drugs were generally the same as PDXs, except that H10 3D cells were sensitive to combined drugs, A14 3D cells were insensitive to palbociclib, and A18 3D cells were sensitive to palbociclib and osimertinib alone.

Fig. 4.

The relationship between EGFR expression level and osimertinib sensitivity. (A – B) RT-qPCR analyses determine EGFR expression level in EGFR knockdown H9 (H9-shEGFR34, H9-shEGFR68) and control (H9-shV) cell lines (A), or EGFR knockdown A18 (A18-shEGFR34, A18-shEGFR68) and control (A18-shV) cell lines (B). Assays are performed in triplicate, and the data shown are representative of three independent experiments. Error bars indicate SD. (C – D) Cytological morphology of EGFR knockdown and control H9 and A18 2D cells stained with Liu reagent (Handsel Technologies) (oil immersion, × 1000). (E – F) In vitro drug sensitivity of EGFR knockdown and control H9 and A18 2D cell lines treated with osimertinib at the concentrations ranged 0∼5 μM (E) or 0∼15 μM (F). Assays were performed in triplicate, and data are representative of three independent experiments. Error bars represent SD. Statistical significance between drug-treated and non-drug-treated cells was calculated using a Student's t-test. *, p < 0.05; **, p < 0.01; ***, p < 0.005.

Relationship between wildtype EGFR expression level and osimertinib sensitivity in 2D cell

To further confirm that wildtype EGFR amp increased osimertinib sensitivity, we generated stable EGFR knockdown H9 and A18 cell lines by transducing lentiviral-based shRNA targeting EGFR. Our data revealed that, compared to the control cell lines (H9- and A18-shV), the cell lines transduced with TRCN0000039634 shRNA (H9- and A18-shEGFR34) and TRCN0000121068 shRNA (H9- and A18-shEGFR68) reduced EGFR expression levels by 15 % and 57 %, respectively (H9) and 66 % and 80 %, respectively (A18) (Fig. 4A and 4B). Reduction in the EGFR expression levels did not affect 2D cell morphologies of H9 and A18 (Fig. 4C and 4D). H9-shV, H9-shEGFR34, and H9-shEGFR68 cells were treated with a gradient concentration (0∼5 μM) of osimertinib for 48 h. Compared with H9-shV, the cell survival rate of H9-shEGFR68 cells was significantly increased at higher drug concentrations (Fig. 4C-i). Similar results were observed in A18-shV and A18-shEGFR68 cells treated with a gradient concentration (0∼15 μM) of osimertinib for 48 h (Fig. 4C-ii). These results show that the expression level of wildtype EGFR correlates with osimertinib sensitivity.

Discussion

In this study, we created nine NSCLC PDX/2D/3D models without EGFR mutations and ALK/ROS1 fusions. The frequency of these PDX models lacking CDKN2A/2B and having high wildtype EGFR expression is exceptionally high. The PDX/2D/3D model, which lacks CDKN2A/2B and have high CDK6 and wildtype EGFR expression, is very sensitive to the drug combination of palbociclib and osimertinib. Our study provides a novel therapeutic approach for NSCLC with cell cycle aberrations and wildtype EGFR amp.

In the analysis of previous real-world studies, NSCLC patients without EGFR and ALK/ROS1 aberrations were shown to have a median survival of about 12 m [4]. However, among the nine patients that successfully established PDXs in our study, the median survival was 3.9 m even after chemotherapy, and seven patients had overall survival time shorter than 8 m. This result suggested that patients with high malignant potential (with shorter survival) have a higher probability of successfully establishing PDX. The shorter median survival with successful xenografts has also been observed in patients with breast, pancreatic, and colorectal cancers [10,17,18]. These reports further support our observation.

While platinum-based doublet chemotherapy is currently the gold standard for the treatment of advanced NSCLC without actionable driver mutations, the discovery of rare targetable and novel driver mutations could improve patient's outcomes and quality of life. Therefore, NGS data are becoming increasingly important for treatment decisions in NSCLC patients who do not have common actionable mutations. However, our results showed inconsistencies in gene expression between CNVs obtained from WES and gDNA-qPCR and IHC analyses. Therefore, it is recommended that WES be performed in conjunction with other assays, such as IHC, to help improve the accuracy of NGS data. The regulatory mechanisms lead to differences in transcription and translation of the cell cycle-related and EGFR genes and their impact on treatment outcome are currently unclear. Our PDX/2D/3D models are valuable in exploring these issues.

Compared to the molecular profiling of unselected NSCLC [5], our data showed particularly high frequencies of CDKN2A/2B homdel and wildtype EGFR amp in the PDXs derived from NSCLC without EGFR mutations and ALK/ROS1 fusions—8.2 % vs. 44.4 %, 7.9 % vs. 33.3 %, and 7.6 % vs. 33.3 %, respectively. CDKN2A/2B homdel and EGFR amp are the three most frequent CNVs in NSCLC, indicating a considerable number of patients and, therefore, warrant the development of new treatments. Our in vivo and in vitro anticancer assays demonstrated that H9 PDX/2D/3D models with CDKN2A/2B homdel, CDK6 and EGFR amp were highly sensitive to palbociclib plus osimertinib. This result may provide a new treatment modality for NSCLC patients with these CNVs. However, it should be noted that oncogenic KRAS mutation, as seen in A18 PDX/2D models, may reduce the anticancer activity of osimertinib. Pao and colleagues (2005) reported that KRAS mutations in NSCLC are associated with resistance to gefitinib or erlotinib alone, further supporting our observations [19]. Despite this, palbociclib plus osimertinib showed significant antitumor activity in PDX/2D/3D models even with oncogenic KRAS mutation.

Our RNA interference experimental data confirmed that EGFR expression level was related to osimerinib sensitivity. This result was further supported by Cappuzo and colleagues (2005) [20]. They reported that EGFR amp, high polysomy, and high protein expression were statistically significantly associated with better response to gefitinib.

Although H10 PDX/2D/3D model had high CDK4 expression, they were insensitive to palbociclib. Studies have shown that CCDC6-RET fusion downregulated cell cycle regulation-related genes and RET inhibitor-induced cell cycle arrest in a lung adenocarcinoma cell line [21]. It is possible that KIF5B-RET enhances cell cycle progression and circumvents the effect of palbociclib. Therefore, cases with RET fusions may not be suitable for treatment with palbociclib.

Inconsistencies between drug treatment responses in PDX and 2D or 3D models occurred in our experiments, which may be resulted from the tumor cell heterogeneity and drug concentrations used in the experiments. Increased drug concentrations may increase the sensitivity of PDX/2D/3D cell cultures. In this case, to use PDX/2D/3D models as drug evaluation platforms, selecting the appropriate drug concentration is a decisive step. More studies will be needed to validate which system(s) can better mimic the anticancer effect of drug(s) in the human body.

In this study, we established a series of NSCLC PDX/2D/3D models without common driver mutations and with complete NGS data. These animal and cell models can be used for drug evaluation and will help identify rare and novel driver genes and develop new treatments. For instance, Minami and coworkers (2023) have reported CDKN2A deletion remodels lipid metabolism to prime glioblastoma for ferroptosis [22]. Our PDX/2D/3D models can be used to analyze lipidomics and ferroptosis-related gene expression and test the effect of ferroptosis inducers in the treatment of NSCLC with wildtype EGFR and ALK/ROS1.

CRediT authorship contribution statement

Jen-Fen Fu: Writing – review & editing, Writing – original draft, Visualization, Validation, Software, Resources, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Cheng-Lung Hsu: Writing – review & editing, Resources, Methodology, Investigation, Data curation, Conceptualization, Funding acquisition. Ping-Chih Hsu: Writing – review & editing, Resources, Investigation, Funding acquisition, Data curation, Conceptualization.

Declaration of competing interest

The authors declare no conflict-of-interest.

Appendix. Supplementary materials

Fig. S1. Establishment of PDX/2D/3D models from NSCLC tumors and similarity of genome-wide copy number variations in PDX, tumor and cell line. The first four successfully established (A) PDXs, (B) 2D cell lines and (C) 3D cell aggregates models. (D) Genome-wide copy number variations (CNVs) of H9-PDX versus corresponding 2D cell line (H9 cell); CNVs of A14 tumor from patient versus corresponding PDX (A14-PDX). Arrows indicated EGFR CNV gain (red) and CDKN2A CNV loss (blue).

Fig. S2. Copy number variations and gene mutations in the PDXs. (A) The relative copy number of cell cycle-related genes was determined by gDNA-qPCR. NC, normal control. Assays were performed in triplicate. Error bars represent SD. (B ‒ C) Gene mutations in TP53 (B) and KRAS (C) in some PDXs were confirmed by RT-PCR followed by DNA sequencing. Amino acid changes caused by DNA mutations are shown at the bottom.

Table S1. KEGG pathways were significantly enriched in nine PDXs by analyzing fusion genes, SNVs and CNVs. CNVs contained only genes with copy number ≧ 4 and = 0.