An integrative pharmacogenomics analysis identifies therapeutic targets in KRAS-mutant lung cancer

Abstract

Background

KRAS mutations are the most frequent oncogenic aberration in lung adenocarcinoma. KRAS mutant isoforms differentially shape tumour biology and influence drug responses. This heterogeneity challenges the development of effective therapies for patients with KRAS-driven non-small cell lung cancer (NSCLC).

Methods

We developed an integrative pharmacogenomics analysis to identify potential drug targets to overcome MEK/ERK inhibitor resistance in lung cancer cell lines with KRAS(G12C) mutation (n = 12). We validated our predictive in silico results with in vitro models using gene knockdown, pharmacological target inhibition and reporter assays.

Findings

Our computational analysis identifies casein kinase 2A1 (CSNK2A1) as a mediator of MEK/ERK inhibitor resistance in KRAS(G12C) mutant lung cancer cells. CSNK2A1 knockdown reduces cell proliferation, inhibits Wnt/β-catenin signalling and increases the anti-proliferative effect of MEK inhibition selectively in KRAS(G12C) mutant lung cancer cells. The specific CK2-inhibitor silmitasertib phenocopies the CSNK2A1 knockdown effect and sensitizes KRAS(G12C) mutant cells to MEK inhibition.

Interpretation

Our study supports the importance of accurate patient stratification and rational drug combinations to gain benefit from MEK inhibition in patients with KRAS mutant NSCLC. We develop a genotype-based strategy that identifies CK2 as a promising co-target in KRAS(G12C) mutant NSCLC by using available pharmacogenomics gene expression datasets. This approach is applicable to other oncogene driven cancers.

Fund

This work was supported by grants from the National Natural Science Foundation of China, the National Key Research and Development Program of China, the Lung Cancer Research Foundation and a Mildred-Scheel postdoctoral fellowship from the German Cancer Aid Foundation.

1. Introduction

The Kirsten rat sarcoma oncogene (KRAS) encodes for a small GTPase that couples growth factor signalling to various downstream signalling pathways among them the MAPK pathway. Despite being an oncogene with a prevalence of 30% in non-small cell lung cancer (NSCLC), the development of KRAS targeted therapies has been largely unsuccessful in the past. The major reason has been the difficulty to interfere with active GTP-loaded KRAS due to the protein's high affinity for intracellular GTP [1], [2]. Recently, pharmacokinetic and pharmacodynamic improvements of direct G12C inhibitors (ARS1620, AMG510) have raised great excitement [3] and ultimately led to two currently ongoing clinical trials (https://clinicaltrials.gov/ct2/results?cond=G12C&term=&cntry=&state=&city=&dist=). However, pre-clinical studies indicate that “K-Ras addiction” is reduced in mesenchymal cancer cells implicating that direct KRAS inhibition may not be efficacious in all patients [4]. Alternatively, inhibitors targeting kinases downstream of KRAS, such as BRAF and MEK, have shown promising activity in metastatic melanoma but were largely ineffective in KRAS mutant NSCLC in combination with chemotherapy [5], [6]. Apart from intrinsic resistance e.g. due to mesenchymal cancer cell differentiation [7], efficacy of MEK inhibitors is limited by the development of acquired resistance [8], [9], [10], [11], [12], [13], [14], [15], [16], [17]. Another factor contributing to the difficulty to treat NSCLC is the heterogeneity of different KRAS mutations which are defined by the respective amino acid substitutions. These change the protein structure and GTPase activity of KRAS and substantially affect the tumour biology and response to chemotherapy [18], [19], [20], [21], [22]. Hence, an unmet need remains to develop more efficacious targeted treatment strategies for patients with KRAS mutant lung cancer.

Research in context.

Evidence before this study

In non-small cell lung cancer (NSCLC), different KRAS mutations are defined by respective amino acid substitutions which substantially affect tumour biology and response to chemotherapy. Despite aberrant MAPK signaling as a consequence of KRAS mutations, MEK inhibition without further stratification is ineffective in patients with KRAS mutant NSCLC.

Added value of this study

We systematically interrogated publicly available pharmacogenomics datasets of the Cancer Genome Project (CGP) and found that cancer cells with different KRAS mutant isoforms differ in their sensitivities to MEK/ERK inhibition. We further more developed a computational pipeline to identify novel therapeutic co-targets and identify and subsequently experimentally validate casein kinase 2A1 (CSNK2A1) as a mediator of MEK inhibitor resistance in lung cancer cell lines with the most frequent KRAS(G12C) mutation.

Implications of all the available evidence

Predicting innovative therapeutic co-targets with respect to the mutational heterogeneity of cancer histotypes will help to guide therapeutic decision-making and ultimately improve treatment outcomes. Our pipeline can be extended to tumours with different KRAS mutations and other oncogene-driven cancers given that adequate pharmacogenomic data sets are available for statistical analysis.

Alt-text: Unlabelled box

Therefore, we performed a Cancer Genome Project (CGP)-based pan-cancer analysis and systematically investigated the impact of different KRAS mutations on MEK and ERK inhibitor efficacy. An integrative pharmacogenomics analysis pipeline was then developed to identify genes (“gR”) which encode for potential mediators of MEK/ERK inhibitor resistance in lung cancer with KRAS(G12C) mutation, the most frequent mutation (>40%) in patients with primary or metastatic KRAS mutant lung adenocarcinoma (LUAD) [20]. The most promising target predicted by this pipeline is the casein kinase 2 subunit alpha (CK2 alpha) encoded by CSNK2A1. CK2 alpha is a serine/threonine kinase that phosphorylates acidic proteins such as casein and influences cellular processes including apoptosis, cell cycle [23] and Wnt/β-catenin signalling [24]. Although there is strong evidence that CK2 is important for cancer pathogenesis [25], [26], [27] and several CK2 inhibitors have entered clinical trials, the role of CK2 as a therapeutic target in lung cancer in the context of different KRAS mutations remained unknown. In summary, our study links CK2 (α-subunit encoded by CSNK2A1) and Wnt/β-catenin signalling to MEK and ERK inhibitor resistance in lung cancer preferentially in the context of KRAS(G12C) mutations and explores its potential as a therapeutic target in combination treatment approaches.

2. Materials & methods

2.1. Key resources table

KRAS(G12C) mutant cell lines used for the integrative pharmacogenomics analysis (Table 1)

Table 1.

KRAS(G12C) mutant cell lines used for the integrative pharmacogenomics analysis.

Cell line	KRAS mutation	Histology
LU-65	G12C	NSCLC_large cell
NCI-H2030	G12C	NSCLC_adenocarcinoma
NCI-H2122	G12C	NSCLC_adenocarcinoma
LU-99A	G12C	NSCLC_large cell
NCI-H1792	G12C	NSCLC_adenocarcinoma
HCC-44	G12C	NSCLC_adenocarcinoma
NCI-H23	G12C	NSCLC_adenocarcinoma
NCI-H2291	G12C; G12V	NSCLC_adenocarcinoma
NCI-H358	G12C	NSCLC_adenocarcinoma
SW1573	G12C	NSCLC_adenocarcinoma
IA-LM	G12C	NSCLC_large cell
HOP-62	G12C	NSCLC_adenocarcinoma

KRAS mutant cell lines used for the in vitro assays (Table 2)

Table 2.

KRAS mutant cell lines used for the in vitro assays.

Cell line	KRAS mutation	Histology
Calu1	G12C	NSCLC_ adenocarcinoma
NCI-H2030	G12C	NSCLC_adenocarcinoma
A549	G12S	NSCLC_adenocarcinoma
NCI-H2009	G12A	NSCLC_adenocarcinoma

2.2. Pharmacogenomics analysis to identify potential targets in KRAS(G12C) mutant lung cancer

The Cancer Genome Project (CGP) at the Wellcome Trust Sanger Institute has developed a large-scale, high-throughput pharmacogenomic dataset for 1001 human cancer cell lines which includes the mutation status of 19,100 genes, genome-wide DNA copy number variations (CNV), mRNA expression profiling of 17,419 genes, and pharmacological profiling for 267 anti-cancer drugs (dataset version 2016/2017). In this dataset, drug responses are represented as the natural logarithm of the IC50 value, which corresponds to the half maximal inhibitory concentration of a given anti-cancer drug. This dataset includes five MEK inhibitors (PD0325901, selumetinib, CI-1040, trametinib, refametinib) and two ERK inhibitors (FR180204 and VX-11e). Thirty-five of the 137 cancer cell lines harbouring KRAS mutations are derived from lung cancer.

We developed a computational pipeline to identify novel therapeutic targets for KRAS mutant lung cancer (Fig. 1). The first step of the analysis was based on the CGP dataset. Each of the 17,419 genes with gene expression profiles was considered a potential target gene “gR”. Due to the heterogeneity of KRAS mutant LUAD, in this study, we only focused on cancer cell lines with the most frequent KRAS(G12C) mutation. We included twelve cell lines in our analysis (Table 1), five different MEK inhibitors (PD0325901, selumetinib, CI-1040, trametinib, and refametinib) and two ERK inhibitors (FR180204, VX-11e). Two replicate experiments for selumetinib and refametinib were considered independent experiments. The expression of 17,419 genes was used to individually calculate their correlation with drug sensitivities (Spearman Correlation) (Step 2). Genes for which we observed a correlation between expression and resistance to two and more MAPK pathway inhibitors were further considered as potential target genes. Next, expression of “gR” was analysed in lung adenocarcinoma (LUAD) based on the TCGA dataset, which includes tumours of 517 patients, 59 matched normal lung tissue samples and 36 tumours with KRAS(G12C) mutation (Step 3). Only those genes were further considered which were upregulated in KRAS(G12C) mutant LUAD in comparison with normal lung tissue. An optional criterion for inclusion of a gene was a worse clinical outcome of LUAD patients whose tumours exhibit higher expression of “gR”. We further filtered the genes by requiring them to be part of cancer core pathways selected from gene set enrichment analysis (GSEA) data sets (Step 4) and finally, the respective encoded proteins had to be known drug targets (Step 5). By using this integrative analytical algorithm requiring all of the above-mentioned criteria, we narrowed down the number of target genes “gR” and categorized them into four Tier categories based on the number of MEK/ERK inhibitors for which we observed a positive correlation between “gR” expression and drug resistance (Tier 1 = resistance to all 5 MEK inhibitors) (Table in Fig. 1).

Fig. 1.

Algorithm of the applied pharmacogenomics analyses to identify potential therapeutic co-targets in KRAS(G12C) mutant lung cancer cells. Pharmacogenomics datasets of the Cancer Genome Project (CGP) including 17,419 genes and drug sensitivity data (Step 1) were used to identify potential target genes “gR” with correlation between “gR” expression and MEK/ERK inhibitor resistance in 12 KRAS(G12C) mutant lung cancer cell lines (Step 2). TCGA data were analysed to evaluate the expression of “gR”in lung adenocarcinoma (LUAD) and their correlation with clinical outcome (Step 3). Target genes “gR”had to be part of cancer core pathways to enrich for biologically relevant genes (Step 4). DrugBank information was used to identify druggable targets (Step 5). Potential target genes “gR” meeting all step 1 to 5 criteria were sorted into four Tier groups based on the number of MEK/ERK inhibitors with positive correlation between“gR”expression and drug IC50s. Tier 1 indicates a correlation with 5 MEK inhibitors including two replicate experiments for selumetinib and refametinib (Table S1).

2.3. TCGA data analysis

RNA-seq and clinical data of LUAD patients were downloaded from TCGA cBioPortal (http://www.cbioportal.org/index.do). The expression of each gene was calculated as RSEM value by the statistical RSEM method (RNA-Seq by Expectation Maximization). RSEM uses a generative model of RNA-seq reads and the Expectation-Maximization (EM) algorithm, and takes read mapping uncertainty into account to achieve the most accurate abundance estimates [28]. The statistical analysis of differentially expressed genes between lung cancer and normal lung tissue samples was performed using DESeq2 [29]. LUAD patients were divided into a CSNK2A1 high and low expressing group, based on the median value of gene expression across all patients. The Kaplan-Meier test was used to compare the overall and relapse free survival between both groups. Deseq2 was applied to call differentially expressed genes between the two groups. Gene set enrichment analysis (GSEA) [30] was used to determine those pathways enriched by a pre-ranked list of all genes, which were sorted by the statistical significance of differential expression defined by the Deseq2 analysis.

2.4. Cell lines and reagents

The human lung cancer cell lines A549, H2030, H2009 and Calu1 (Table 2) were purchased from ATCC and grown at 37 °C in RPMI medium supplemented with 10% fetal bovine serum (FBS), 100 μg/ml penicillin and 100 units/ml streptomycin (complete medium). Cell lines were authenticated at the RTSF Genomics Core of Michigan State University using the Promega GenePrint 10 System. All cell lines included in this study were negative for Mycoplasma as regularly tested with the Mycoplasma Plus PCR Primer Set (Agilent). Selumetinib (Cat#S1008), trametinib (Cat#S2673) and silmitasertib (Cat#S2248) were purchased from SelleckChem.

2.5. Quantification of cell proliferation

One thousand cells were seeded in 96-well plates in 100 μl RPMI media supplemented with 10% FBS (Sigma-Aldrich Cat#F2442) and penicillin/streptomycin (Gibco Cat#15140122). From the following day onwards, plates were imaged with three fields/per well under 10x magnification every two hours for a total of 120 h in the IncuCyte ZOOM™ (Essen BioScience). Resulting data were analysed with the IncuCyte Confluence software (Version 1.5) (Essen BioScience), which quantifies confluence via cell surface area coverage. IncuCyte experiments were performed in triplicate and a representative growth curve is shown for each condition. Proliferation endpoint analyses were measured by CellTiter Glo® Assay (Promega Cat#G7570).

2.6. Western blot analysis

Cells were lysed in RIPA lysis buffer (Thermo Fisher Cat#89900) supplemented with protease and phosphatase inhibitor cocktail tablets (Roche Cat#11836170001 and 04906837001). The antibodies used for western blotting included those against: phosphorylated Akt (Ser473) (Cell Signaling Cat#4060), Akt (Cell Signaling Cat#9272), phosphorylated S6 (Ser235/236) (Cell Signaling Cat#4858), S6 ribosomal protein (Cell Signaling Cat#2217), phosphorylated ERK1/2 (Cell Signaling Cat#4370), ERK1/2 (Cell Signaling Cat#4695), E-Cadherin (Cell Signaling Cat#3195), MEK (Cell Signaling Cat#8727), β-catenin (Cell Signaling Cat#8480), phospho-β-catenin (Ser552) (Cell Signaling Cat#9566), Zeb1 (Bethyl Cat#A301-922A), Vimentin (Cell Signaling Cat#3932), p27 (Cell Signaling Cat#3688), cMyc (Cell Signaling Cat#2276), HSP90 (Santa Cruz Biotech Cat#sc-7947), Lamin (Santa Cruz Cat#sc-6216), HRP-linked anti-rabbit IgG secondary antibody (Cell Signaling Cat#7074P2), HRP-linked ECL Sheep anti-Mouse IgG secondary antibody (GE Healthcare Cat#NA931V), HRP-linked Donkey anti-Goat IgG secondary antibody (Santa Cruz Cat#sc-2020), HRP-linked ECL Donkey anti-Rabbit IgG secondary antibody (GE Healthcare Cat#NA934V). Western blots depicted in this manuscript are representative of at least three independent experiments.

2.7. SiRNA-mediated gene knockdown

Cells (1.5× 10⁶) were seeded into a 10 cm plate and incubated overnight at 37 °C. On the next day, media was replaced by antibiotic free full media. The mixture of non-targeting or anti-CSNK2A1 SMART-Pool ON-TARGET plus siRNA (Dharmacon Cat#D-0018101005 and L-003475000005L) at a final concentration of 20 nM was added together with DharmaFECT 1 (Dharmacon, Cat#T-2001–03) after allowing 30 min of complex formation in serum-free media. Knockdown efficacy was assessed by Western blot or quantitative reverse transcriptase (RT)-PCR after 48 hrs of transfection. For subsequent drug treatment, cells were harvested and re-seeded after 48 hrs of siRNA treatment and then treated with selumetinib or trametinib for another 24 to 96 hrs.

2.8. TOPFlash reporter assay

Cells (1.5 × 10⁶) were seeded in a 10 cm plate and incubated overnight at 37 °C. On the next day, cells were transiently transfected with 1 μg of M50 Super 8x TOPFlash reporter plasmid, 100 ng of a pRL Renilla Luciferase control vector (Promega Cat#E2231) and FuGENE® HD (Promega Cat#PRE2311). M50 Super 8x TOPFlash was a gift from Randall Moon (Addgene plasmid #12456). After 24 hrs, cells were washed with PBS and full media was added for another 24 hrs without or with MEK inhibitor (selumetinib 1 μm, trametinib 100 nM). Luciferase activity was measured with the Dual Luciferase reporter assay (Promega Cat#E2920) on a POLARstar Omega microplate reader (BMG Labtech).

2.9. Cytoplasmic-nuclear protein fractionation

Nuclear and cytoplasmic proteins were separated with the NE-PER™ Nuclear and Cytoplasmic Extraction Kit (Thermo Scientific Cat#78833) according to the manufacturer's protocol. Lamin (nuclear protein fraction) and MEK (cytoplasmic protein fraction) were used as loading control.

2.10. Statistical analyses

The Kruskal-Wallis H-test was used to compare drug sensitivity values Ln(IC50) between multiple groups and the Wilcoxon rank-sum test to compare drug sensitivities between two groups. The correlation between expression of “gR” and drug sensitivities was evaluated using the Spearman Correlation. Wnt signaling pathway activity scores of LUAD patients in the TCGA dataset were derived from single sample gene set enrichment analysis (ssGSEA) [31]. The correlation between Wnt signalling pathway activity score and gene expression in LUAD patients was evaluated using the Pearson Correlation. All statistical analyses were executed in Python, SciPy function or in R.

3. Results

3.1. Cancer cells with different KRAS mutant isoforms differ in their MEK/ERK inhibitor sensitivities

We first interrogated the publicly available pharmacogenomics Cancer Genome Project (CGP) dataset, which includes mutational and pharmacological profiles of >1000 human cancer cell lines treated with 267 anti-cancer drugs [32] (Fig. 1). To investigate MEK/ERK inhibitor sensitivities for different KRAS mutant isoforms across cancer histotypes, we grouped cancer cell lines based on their KRAS mutation status into 12 groups (A146T, G12A, G12C, G12D, G12R, G12S, G12V, G13C, G13D, K117N, Q61H, Q61L). We found that sensitivities for MEK (CI-1040, refametinib (RDEA119), PD0325901, selumetinib, trametinib) and ERK inhibitors (VX-11e) vary in cell lines with different KRAS mutations (Fig. 2a-g). Overall, cell lines with G12R mutation were more sensitive to MEK inhibitors in comparison with other types of KRAS mutations and cell lines with G12C mutations exhibited relative drug resistance (Fig. 2a-f). The relative drug sensitivity profiles for VX-11e (ERK inhibitor) were different from those for MEK inhibitors, but again, cell lines with KRAS(G12C) mutation exhibited relative resistance compared to other KRAS mutations (Fig. 2g). To address the question if the tissue of origin influences response to MAPK pathway inhibition, we furthermore investigated the effect of different KRAS mutations on drug sensitivities in the two major cancer histotypes of lung and pancreatic cancer. Differences in MEK/ERK inhibitor sensitivities across different KRAS mutations were observed in both cancer types. Overall, pancreatic cancer cells with G12R mutation (Fig. S1a-c) and lung cancer cells with G12A mutation (Fig. S1d-e) were most sensitive to MEK inhibition, respectively. However, the numbers of cancer cell lines with these mutations were low.

Fig. 2.

Pan-cancer analysis of MEK/ERK inhibitor sensitivities in cancer cell lines with different types of KRAS mutations. Cell lines were grouped according to KRAS mutational subtypes and drug responses to MEK (a-f) or ERK (g) inhibitors are represented as the natural logarithm of the IC50 value. The p-value of multiple-group comparisons is indicated (Kruskal-Wallis H-test). “*” denotes a pairwise comparison with p < 0.05.

3.2. KRAS(G12C) is the dominant mutation in primary and metastatic LUAD

Next, we analysed the distribution of different KRAS mutations in primary (TCGA dataset) and metastatic (MSK-IMPACT dataset) LUAD [33] (Fig. 3). 33% of patients with primary and 27% of patients with metastatic LUAD harbour KRAS mutations, respectively. In primary LUAD, we observed ten different types of KRAS mutations (G12C, G12D, G12A, G12F, G12R, G12S, G12V, G12Y, Q61L, D33E) (Fig. 3a), whereas patients with metastatic LUAD exhibited a more complex mutational pattern - among 19 types of KRAS mutations, 11 were exclusively found in patients with metastatic LUAD (A146T, A146V, A59T, AG59GV, G13C, G13D, G13E, G13R, G13V, Q61R, T58I) (Fig. 3b). In both groups, KRAS(G12C) was the dominant mutation (primary LUAD: 48%, metastatic LUAD ∼43%), which confirms previously published analyses [34].

Fig. 3.

Frequencies of different KRAS mutations in LUAD.

Distribution of different KRAS mutations were analysed in tumour tissue of patients with primary (TCGA dataset, n = 75) (a) or metastatic LUAD (MSK-IMPACT dataset, n = 241) (b). In both groups, KRAS(G12C) is the most common mutation (>40 %). Patients with metastatic lung adenocarcinoma exhibit a more complex mutational pattern - among 19 types of KRAS mutations, 11 were exclusively found in patients with metastatic LUAD.

3.3. A pharmacogenomics analysis identifies potential co-targets in KRAS(G12C) mutant lung cancer

Our analysis suggests that across cancer histotypes cancer cell lines with different KRAS mutations exhibit different sensitivities to MAPK pathway inhibition. Due to the high frequency (Fig. 3), and the relative resistance against MEK and ERK inhibitors (Fig. 2), we next focused on lung cancer cell lines with KRAS(G12C) mutation and developed a computational pipeline to identify potential therapeutic targets to overcome MEK/ERK inhibitor resistance (Fig. 1). Initially, 1212 genes with positive correlation between gene expression and resistance to two or more (Fig. 1) MAPK pathway inhibitors were identified as potential target genes “gR” (Step 2). Out of these, 494 genes were identified to be upregulated in LUAD and associated with poor survival (TCGA dataset) (Step 3). We finally narrowed down the number of genes by requiring “gR” to be part of cancer core pathways (Step 4) as well as to be known drug targets according to the DrugBank database (Step 5). This algorithm led to the identification of 14 genes which encode for potential therapeutic targets to overcome MEK/ERK inhibitor resistance in KRAS(G12C) mutant lung cancer (CSNK2A1, CARS, EPRS, RPL8, YARS, AARS2, ALKBH2, CDK8, COMP, DARS, HDAC1, IARS2, MAPK8, PARS2) (gene list in Fig. 1 and Table S1). Only the expression of CSNK2A1 and CARS correlated with resistance to the maximum number of MEK inhibitors (Tier 1), for CSNK2A1 5 MEK inhibitors including two replicate experiments for selumetinib and refametinib which were considered as independent experiments (Fig. 4a). There was also a trend (p = 0.112, Permutation test) between CSNK2A1 expression and Ln(IC50) for CI-1040 (Fig. 4a). Importantly, we found that expression of CSNK2A1 was increased in LUAD tumour tissue and those tumours with KRAS(G12C) mutation compared to matched normal lung tissue (Fig. 4b, p = 1.35e-18, Wilcoxon rank-sum test), but also in other tumour entities (Fig. S2). Furthermore, LUAD patients with high intratumoral CSNK2A1 expression had a trend towards poorer survival (Fig. 4c, p = 0.07, Kaplan-Meier Test). Significant differences in overall and progression-free survival were observed for patients with pancreatic adenocarcinoma (PAAD) (Fig. S3).

Fig. 4.

A pharmacogenomics analytical approach identifies CSNK2A1 as a potential therapeutic co-target in KRAS(G12C) mutant lung cancer cells. Pharmacogenomics data from the Cancer Genome project (CGP) were used to identify target genes “gR” which show a correlation between expression and MEK/ERK inhibitor resistance. (a) Among 14 potential target genes (Table in Fig. 1), CSNK2A1 expression was correlated with resistance to the highest number of MEK inhibitors (Tier 1 indicates positive correlation for 5 MEK inhibitors, including two replicate experiments for selumetinib and refametinib). There was also a trend (p = 0.112, Permutation test) towards correlation between CSNK2A1 expression and CI-1040 sensitivity. (b) Tumours of LUAD patients including those with KRAS(G12C) mutation exhibit significantly higher CSNK2A1 expression than matched normal lung tissue (p = 1.35e⁻¹⁸, Wilcoxon rank-sum test). (c) LUAD patients with high intratumoral CSNK2A1 expression have a trend towards inferior overall survival (HR=1.305 (95% CI, 0.976 - 1.746), p = 0.07, Kaplan-Meier test). Numbers at risk at each time point are included.

3.4. Correlation between CSNK2A1 expression and MEK inhibitor resistance is specific for KRAS(G12C) mutant lung cancer cells

We next investigated if the correlation between CSNK2A1 expression and MEK inhibitor resistance is specific to KRAS(G12C) mutant lung cancer cells or if it can also be observed in cells with other KRAS mutational subtypes. No correlation was found between CSNK2A1 expression and sensitivity to 7 MEK inhibitors in lung cancer cell lines with the second most frequent KRAS(G12V) mutation (n = 9) (Fig. S4a), nor in the pooled group of cell lines with other non-KRAS(G12C) mutations (n = 23) (Fig. S4b). Furthermore, there was neither a correlation for KRAS(G12V) and KRAS(G12D) mutant pancreatic cancer cell lines (Fig. S4c, d) nor for lung cancer cell lines harbouring other oncogenic mutations affecting MAPK signalling (BRAF, EGFR, NRAS) (Fig. S5a-c). All cell lines included in this analysis are listed in Table S2.

3.5. CSNK2A1 increases resistance to MEK inhibition in KRAS(G12C) mutant lung cancer cells

To validate our in silico prediction results, we selected two lung cancer cell lines with KRAS(G12C) mutation (Calu1 and H2030) and two with non-KRAS(G12C) mutations (A549 (G12S) and H2009 (G12A)) (Table 2). CSNK2A1 knockdown alone dramatically decreased proliferation of Calu1 and H2030 cells and increased the anti-proliferative activity of simultaneous MEK inhibition with 1 μM of selumetinib (Fig. 5a). In contrast, these effects were not observed in non-KRAS(G12C) mutant lung cancer cell lines A549 and H2009 (Fig. 5b). We furthermore treated Calu1 and A549 cells with the specific CK2 inhibitor silmitasertib (CX-4945, 6 μM) alone or in combination with MEK inhibitor (10 nM trametinib) (Fig. 5c). Whereas A549 (KRAS(G12S)) cells remained basically unaffected, MAPK (pERK) and PI3 kinase (pAKT, pS6) signalling as well as cell cycle promoting proteins cMyc and Cyclin D1 were strongly suppressed in Calu1 cells with KRAS(G12C) mutation upon combined MEK and CK2 inhibition compared to MEK inhibition alone. This translated into a greater sensitization of Calu1 cells to MEK inhibition compared to A549 cells (Fig. 5d). In both approaches - genetic CSNK2A1 knockdown and pharmacological CK2 inhibition plus MEK inhibitor treatment - no significant PARP cleavage (Fig. S6, Fig. 5c) or caspase-3 activity were detectable (Incucyte experiments, data not shown). This indicates that CSNK2A1 loss or CK2 inhibition plus MEK inhibition exert anti-proliferative but not pro-apoptotic effects.

Fig. 5.

CSNK2A1 promotes proliferation, mitogenic signalling and MEK inhibitor resistance in KRAS(G12C) mutant lung cancer cells. (a) siRNA-induced CSNK2A1 knockdown significantly reduced proliferation of KRAS(G12C) mutant Calu1 and H2030 cell lines and increased the anti-proliferative effect of simultaneous MEK inhibition (1 μM selumetinib). (b) CSNK2A1 knockdown in non-KRAS(G12C) cell lines A549 (KRAS(G12S)) and H2009 (KRAS(G12A)) did not significantly affect cell proliferation or MEK inhibitor sensitivity. (c) Combined MEK (100 nM trametinib) and CK2 inhibition (6 μM silmitasertib) suppresses mitogenic signalling in Calu1 cells (G12C) but not in A549 cells (G12S) and (d) translates into higher relative MEK inhibitor efficacy after 120 hrs in the context of a KRAS(G12C) mutation.

3.6. CSNK2A1 increases Wnt/β-catenin pathway activity in KRAS(G12C) mutant lung cancer cells

To gain more insight into the molecular mechanisms of CSNK2A1-mediated MEK/ERK inhibitor resistance, we performed GSEA between CSNK2A1 high- and low-expressing KRAS mutant lung cancer cell lines and human LUAD tumors. Genes within the Wnt signaling pathway were significantly enriched in the CSNK2A1 high-expressing group in CCLE (p = 0.008, Permutation test) [35] and TCGA (p = 0.014, Permutation test) (Fig. 6a-b). We further applied single sample gene set enrichment analysis (ssGSEA) to generate Wnt pathway activity scores for LUAD patients (TCGA) and correlated these scores with CSNK2A1 expression levels (Fig. S7). CSNK2A1 expression was stronger correlated with increased Wnt pathway activity in KRAS(G12C) mutant tumours (corr=0.268, p = 0.114, Permutation test) compared to non-KRAS(G12C) mutant LUAD (corr=0.168, p = 0.308, Permutation test) (Fig. S7). However, due to the limited number of available KRAS(G12C) mutant tumours, this correlation did not reach statistical significance. To experimentally validate our computational findings of preferential CSNK2A1-dependent Wnt pathway activation in the context of G12C mutations, in the next step, we knocked down CSNK2A1 with siRNA and then transiently transfected a luciferase reporter plasmid (8xTOPFlash) into Calu1 and A549 cells which detects Wnt/β-catenin/T-cell factor (TCF) transcriptional activity. Loss of CSNK2A1 alone decreased reporter activity in Calu1 (KRAS(G12C)) but not in A549 (KRAS(G12S)) cells and the MEK inhibitor-induced increase in TOPFlash reporter activity after 24 hrs was partially reversed upon simultaneous CSNK2A1 knockdown in Calu1 cells (trametinib 100 nM: Fig. 6c, selumetinib 1 μM: Fig. S8). We furthermore separated nuclear and cytoplasmic protein fractions in both cell lines after 24 hrs of trametinib (100 nM) treatment. Nuclear translocation of β-catenin is a hallmark of Wnt pathway activation and crucial for Wnt-dependent target gene expression (e.g. cMyc, Cyclin D1). We observed a reduction of total and transcriptionally active full length and low-molecular weight (LMW) (Ser552-phosphorylated) β-catenin [36] in the nucleus upon CSNK2A1 reduction in Calu1 (G12C) but not in A549 (G12S) cells. Wnt-target proteins cMyc (nucleus) and cyclin D1 (cytoplasm) were also exclusively reduced in Calu1 cells with combined CSNK2A1 siRNA plus trametinib treatment (Fig. 6d).

Fig. 6.

CSNK2A1 promotes Wnt/β-catenin signalling in KRAS(G12C) mutant lung cancer cells.

(a) GSEA analysis showed enrichment for Wnt pathway activation in CSNK2A1-high expressing lung cancer cell lines (CCLE dataset, n = 87, p = 0.008, Permutation test) and tumours from patients with LUAD (TCGA dataset, n = 517, p = 0.014, Permutation test). (b) Schematic depiction of the Wnt signalling pathway which can be influenced by casein kinase 2 (CK2) at different levels. Cytoplasmic-nuclear shuttling of active β-catenin is a hallmark of this pathway and crucial for Wnt-dependent target gene expression such as c-myc and Cyclin D1. (c) Transiently transfected 8xTOPFlash reporter (RLU – relative luminescence units) indicates reduction of basal and MEK inhibitor-induced (100 nM trametinib for 24 hrs) Wnt pathway activity in Calu1 KRAS(G12C) but not in A549 KRAS(G12S) cells upon siRNA-induced CSNK2A1 knockdown. A representative experiment of three independent experiments is shown (n = 3 biological replicates each). d) Protein fractionation indicates reduced levels of Wnt target proteins cMyc, cyclin D1 and Zeb1 as well as reduced nuclear translocation of total and transcriptionally active full length (92 kDa) and low molecular weight (LMW, 56 kDa) phospho-Ser552-β-catenin upon CSNK2A1 knockdown and 24 hrs of trametinib treatment (100 nM) in Calu1 but not A549 cells. “**” and “****” indicate p-values of < 0.01 and 0.0001, respectively (Student t-test).

4. Discussion

In recent years, treatment paradigms for malignancies have shifted from histology-based to genotype-based approaches. The discovery of driver mutations in lung adenocarcinoma (LUAD) such as EGFR [37] or EML4-ALK [38] paved the way for the development of targeted therapies. Unfortunately, these are still unavailable for KRAS-driven tumours with aberrant MAPK signaling [39], for which MEK inhibitors failed to prove benefit in combination with chemotherapy [6]. Furthermore, the impact of KRAS mutation subtypes on clinical response to MEK inhibition remains under debate [40]. Despite the very recent development of direct KRAS(G12C) inhibitors (ARS1620, AMG510) [3, 41] currently under evaluation in clinical trials, primary treatment resistance will likely occur and better patient stratification for drug combination approaches is required [42].

A plethora of KRAS mutations with different protein structures and GTPase activities [18], [19], [20], [21], [22] across cancer histotypes renders a uniform treatment strategy basically impossible. It is reasonable to think though that rationally designed MEK- or ERK- inhibitor-based drug combinations may lead to better treatment outcomes provided they are tolerated by the patient [34]. In non-small cell lung cancer (NSCLC), KRAS mutations represent the dominant oncogenic event (∼30% of patients) and KRAS(G12C) is the most frequent mutational subtype in primary (48%) and metastatic (∼43%) LUAD (Fig. 3) [34].

To investigate, which drug combinations could overcome MEK/ERK inhibitor resistance in specific KRAS mutational subsets of lung cancer, in the present study, we used publicly available pharmacogenomics datasets from the Cancer Genome Project (CGP) and systematically investigated MEK/ERK inhibitor sensitivities across cancer histotypes in the context of different KRAS mutations. We find, that cancer cell lines (n = 137) grouped by KRAS mutation differ substantially in their sensitivities to MAPK pathway inhibition and that cell lines with KRAS(G12C) mutation exhibit relative resistance to MEK and ERK inhibition compared to other less frequent mutations (Fig. 2). We subsequently decipher potential mutation specific vulnerabilities to overcome resistance in KRAS(G12C) mutant lung cancer cell lines (Fig. 1) and identify 14 potential co-targets for this subgroup of lung cancer (Table in Fig. 1). We required potential target proteins to be druggable with currently available inhibitors and thereby sought to provide a platform which can translate our in silico and in vitro findings relatively easy into potential clinical applications without requiring time-consuming and cost-intensive drug development steps.

Among the 14 potential target genes, CSNK2A1 expression was correlated with resistance to the highest number of MEK inhibitors (5 MEK inhibitors including two replicate experiments for selumetinib and refametinib as well as one experiment for trametinib) (Fig. 4a). CSNK2A1 encodes for the catalytic α-subunit of casein kinase 2 (CK2), a heterotetramer, which is composed of two catalytic and two noncatalytic β-subunits. CK2 is involved in the regulation of cell cycle progression, apoptosis, transcription as well as Wnt and PI3K signalling and therefore plays an important role in the pathogenesis of solid and hematological malignancies [26, 27, [43], [44], [45], [46], [47]]. We find that CSNK2A1 is overexpressed in KRAS(G12C) mutant LUAD (Fig. 4b) and several other cancers (Fig. S2) and - importantly - is associated with an inferior prognosis (Fig 4c, Fig. S3). We considered CSNK2A1 also as an interesting candidate gene, because specific CK2 inhibitors like silmitasertib (CX-4945) are currently under clinical investigation (https://clinicaltrials.gov/ct2/results?cond=&term=CX-4945&cntry=&state=&city=&dist=).

Intriguingly, the correlation between CSNK2A1 expression and MEK inhibitor resistance exclusively occurred in lung cancer cell lines with KRAS(G12C) mutation, but not in cell lines with other KRAS mutations or other oncogenic events such as mutations in EGFR, BRAF, or NRAS (Fig. S4 and S5). CSNK2A1 knockdown experiments confirmed our in silico predictions and significantly reduced cell proliferation and mitogenic signalling (Fig. S6) and increased MEK inhibitor sensitivity (Fig. 5a) in KRAS(G12C) mutant lung cancer cells, but not in non-KRAS(G12C) mutant cells (Fig. 5b). Furthermore, pharmacological inhibition of CK2 with silmitasertib in combination with MEK inhibition strongly inhibited mitogenic signalling in the KRAS(G12C) but not in the non-KRAS(G12C) mutant context (Fig. 5c). In Calu1 (KRAS(G12C)) cells, silmitasertib was equipotent to trametinib and sensitized the cells relatively more to combined MEK inhibition compared to cells with non-KRAS(G12C) mutation (Fig. 5d).

Gene-set enrichment analyses indicated enrichment for Wnt pathway activation (pathway depicted in Fig. 6b) in CSNK2A1 high-expressing lung cancer cell lines and LUAD (Fig. 6a) which confirms previously published data on the role of the CK2 heterotetramer [26]. CSNK2A1 knockdown reduced basal and MEK inhibitor-induced Wnt pathway activity (Fig. 6c and S8) and led to a reduced nuclear translocation of total and transcriptionally active full length and low molecular weight (LMW) phospho-S552-β-catenin [36] in the KRAS(G12C) mutant but not in the non-KRAS(G12C) mutant context (Fig. 6d). Wnt-dependent gene expression (cMyc, Cyclin D1 and Zeb1) [48] was also exclusively reduced in cells with KRAS(G12C) mutation (Fig. 6d).

These results suggest that CSNK2A1-dependent Wnt/β-catenin pathway activation and associated MEK inhibitor resistance preferentially occur in lung cancer cell lines in the context of a KRAS(G12C) mutation. Supporting this, KRAS(G12C) mutant lung tumours showed a stronger correlation between CSNK2A1 expression and Wnt pathway activity scores [31] (n = 36, corr=0.268, p = 0.114, Permutation test) than non-KRAS(G12C) mutant tumors (n = 39, corr=0.168, p = 0.308, Permutation test) (Fig. S7). However, these trends did not reach statistical significance and need to be validated in larger patient cohorts. Contrariwise, previous studies have shown that Wnt signalling is also important for lung tumorigenesis in the context of other KRAS mutations [49]. Therefore, we speculate, that non-KRAS(G12C) mutant cancer cells preferentially depend on CK2-independent mechanisms of Wnt/β-catenin pathway activation. Supporting this, CSNK2A1 reduction in KRAS(G12S) mutant A549 cells induced a strong increase in Wnt reporter activity (Fig. 6c) potentially indicating greater dependency on other mechanisms of Wnt pathway activation [50], [51], [52], [53]. However, CSNK2A1 knockdown (Fig. 5b) and silmitasertib treatment (Fig. 5d) also had a slight anti-proliferative effect in A549 cells suggesting that non-KRAS(G12C) mutant cancer cells retain some but a reduced dependency on CK2-mediated Wnt pathway activation. Also, CSNK2A1 expression levels may not necessarily be a direct consequence of a given RAS mutation, but rather represent a molecular mechanism to withstand e.g. cellular stress imposed by a specific mutant KRAS protein [54]. To date, the impact of different KRAS mutations on activation of downstream signalling pathways still remains unclear and is subject to ongoing pre-clinical and clinical studies (Ambrogio et al., manuscript in preparation) [34], [55], [56].

In an era, in which direct KRAS(G12C) inhibitors such as ARS1620 or AMG510 finally enter clinical trials, we consider our approach as complementary to the concept of direct KRAS(G12C) inhibition. Previous studies revealed, that sensitivity profiles for KRAS(G12C) inhibitors not necessarily overlap with those for MEK inhibitors [57] and therefore, combined CK2 plus MEK inhibition could be relevant for patients whose tumors are resistant to KRAS(G12C) inhibitors. The dependency of cancer cells on Ras is context-dependent and decreases e.g. during epithelial-to-mesenchymal transition (EMT) [4] which frequently occurs during cancer progression [58], [59], [60]. Interestingly, CSNK2A1 high-expressing cancer cells showed an enrichment for mesenchymal genes (hallmark “Epithelial_to_mesenchymal_transition” (EMT)) in GSEA (p = 0.006, Permutation test, Fig. S9). Furthermore, CSNK2A1 reduction reverted the mesenchymal phenotype (indicated by reduced expression of the mesenchymal marker protein Axl) and partially re-sensitized mesenchymal Calu1 cells to ARS1620. This effect was not observed in epithelial H358 cells which exhibit baseline ARS1620 sensitivity (data not shown). Hence, reversal of EMT not only re-sensitizes mesenchymal cancer cells to MEK inhibition [61], mesenchymal-to-epithelial transition (MET) could also have the potential to increase efficacy of direct KRAS inhibitors. However, this will need further experimental validation in other tumour models.

In summary, the present study identifies CK2 (catalytical subunit encoded by CSNK2A1) as a promising co-target to overcome MEK/ERK inhibitor resistance in KRAS(G12C) mutant LUAD. It also reinforces the notion, that accurate patient stratification is crucial for the development of genotype-based precision treatment strategies. Utmost, we consider this a proof of principle study applicable to any oncogene-driven cancer in the future, provided sufficient pharmacogenomics data are available.

Declaration of Competing Interest

All authors declare no conflicts of interest.