Discovery of Novel RASGRF2 Fusions as a Therapeutic Target in Lung Adenocarcinoma of Never or Light Smokers

ABSTRACT

Lung adenocarcinomas (LUADs) in never‐smokers exhibit distinct molecular profiles from those of smokers, and their driver mutations are quite divergent. We aimed to evaluate the utility of RNA‐seq for the molecular profiling of LUAD in Japanese never or light smokers. A hybridization capture‐based RNA panel (TOP2‐RNA) was used to confirm the validity of mutational and expression analyses of the panel in 122 Japanese LUAD cases. For the discovery cohort, 270 primary LUADs were molecularly profiled using TOP2‐RNA. Whole transcriptome sequencing (WTS) was conducted for the samples without any oncogenic driver mutations. A risk score was developed using TOP2‐RNA expression data to predict the prognosis of surgically resected LUAD. Driver oncogenes were identified in 180 cases (66.7%) of the discovery cohort. The frequency of MET ex14 skipping was high (12.6%) among cases without EGFR mutations. Actionable novel fusions of RDX‐RASGRF1, PRKCI‐RASGRF2, and OCLN‐RASGRF2 were identified in three never‐smoker cases by WTS. A functional assay identified that the expression of RASGRF fusions transformed the cells through phosphorylation of MEK, which was inhibited by cobimetinib treatment. High‐risk patients defined by the risk score based on the four‐gene signature had significantly worse RFS and OS for all stages and stage I patients in the discovery and validation cohorts. This study identified novel RASGRF1/2 fusions that might be targetable by MEK inhibitors. RNA‐based molecular profiling could identify actionable mutations and assess the prognostic biomarkers for patient stratification to determine the optimal treatment based on the molecular profiling of individual LUAD cases.

This study discovered actionable novel fusions of RASGRF1 and RASGRF2 in lung adenocarcinoma of never‐smoker and identified that those fusions transformed the cells through phosphorylation of MEK, which was inhibited by cobimetinib treatment. A risk score based on the four‐gene signature could predict the prognosis of surgically resected lung adenocarcinoma using RNA expression data.

Abbreviations

LUAD

lung adenocarcinoma

NSCLC

non‐small cell lung cancer

RASGRF

Ras‐specific guanine nucleotide‐releasing factor

WES

whole‐exome sequencing

WTS

whole‐transcriptome sequencing

1. Introduction

Globally, lung cancer is the leading cause of cancer morbidity and mortality, accounting for over 1.8 million deaths annually [1]. The treatment options for patients with advanced lung adenocarcinoma (LUAD), the most common histologic type of lung cancer, have changed dramatically in two decades with the introduction of small molecule‐targeted and immuno‐directed therapies. Molecularly targeted therapies have been developed for patients with non‐small cell lung cancer (NSCLC) harboring mutations in EGFR [2, 3], KRAS [4], BRAF [5], and MET [6, 7], as well as those with fusion genes of ALK [8, 9], RET [10], ROS1 [11, 12], and NTRK1/2/3 [13, 14]. The other targeted therapies directed against activating alterations, such as ERBB2 mutations [15, 16], and NRG1 fusion [17], are currently being studied.

Today, targetable molecular alterations are identified in approximately 60% and 80% of patients with Western and Asian LUAD, respectively [18]. However, some of the more recently recognized drivers, such as MET ex14 (exon 14) skipping or rare fusions, were not consistently evaluated [19, 20].

Although smoking‐related lung cancers continue to account for the majority of diagnoses, smoking rates have been decreasing for several decades, resulting in a proportional increase in the incidence of lung cancer in individuals who have never smoked (LCINS) [21]. Constitutive activation of oncogenes through mutation, rearrangement, and amplification accounts for 80% of LCINS cases and 50% of smoking‐related LUAD cases [22, 23]. Previous studies have shown that never‐smokers have lower rates of mutation in the KRAS and TP53 genes than smokers, whereas never‐smokers are more likely to develop EGFR mutations [22, 24]. These findings suggest that lung cancer in never‐smokers is a distinct entity, emphasizing the need to explore novel genetic factors affecting survival in this population.

Given the increased likelihood of targetable somatic alterations, several diagnostic and management principles can be applied to LCINS. While panel‐based NGS is preferred for broad target coverage, fusion detection via DNA‐based NGS is challenging due to degraded DNA in formalin‐fixed and paraffin‐embedded (FFPE) samples. Targeted DNA panels often miss rarer fusions like NTRK1/2/3, NRG1, and some ROS1 fusions within long intronic sequences [25, 26]. For this reason, RNA‐based NGS has been suggested, as it can identify fusions even in low tumor purity samples due to high expression of fusion oncogenes [27, 28, 29].

Significant research on gene expression‐based prognostic signatures using microarrays has led to the commercialization of two LUAD biomarkers (myPlan Lung Cancer and Pervenio Lung RS), but their accuracy in predicting survival remains limited. RNA‐seq can address some microarray limitations, especially for detecting low‐level transcripts. While a recent study used RNA‐seq for prognostic analysis in LUAD with TCGA data [29, 30], further large‐scale validation is needed.

To assess the importance of RNA‐seq in clinical practice, the present study aimed to evaluate the validity and utility of a targeted RNA panel, TOP2‐RNA, in detecting known oncogenic driver mutations in LUAD of never or light smokers. Gene expression analysis was performed to develop a risk score to predict cancer recurrence after surgery for LUAD in the early stage. Furthermore, whole transcriptome sequencing (WTS) was conducted to explore novel oncogenic alterations among tumors lacking known drivers.

2. Materials and Methods

Detailed information can be found in Doc. S1.

2.1. Study Design and Patient Specimens

The study included 392 LUAD patients who underwent surgery at Juntendo University (2010–2019). The validation cohort (122 patients, 2010–2014) included 122 of 125 cases from a previous whole‐exome sequencing (WES) and RNA‐seq study [31]. The discovery cohort (270 patients, 2015–2019) excluded common EGFR mutations, screened via peptide nucleic acid‐locked nucleic acid PCR clamp method. Histology followed WHO classification [32]. Clinical data are in Table 1.

TABLE 1.

Clinicopathological description of the two cohorts of the present study.

Variable	Stratication	#Cases (%)
		Validation cohort	Discovery cohort
Age	< 40	5 (4.1%)	11 (4.1%)
	40–59	32 (26.2%)	57 (21.1%)
	60–79	76 (62.3%)	170 (63.0%)
	≥ 80	9 (7.4%)	32 (11.9%)
Gender	Female	93 (76.2%)	183 (67.8%)
	Male	29 (23.8%)	87 (32.2%)
Smoking	Never smoker	99 (81.1%)	214 (79.3%)
	Light smoker	23 (18.9%)	56 (20.7%)
Stage	0	0 (0%)	21 (7.8%)
	I	104 (85.2%)	191 (70.7%)
	II	3 (2.5%)	23 (8.5%)
	III	3 (2.5%)	28 (10.4%)
	IV	0 (0%)	5 (1.9%)
	NA	12 (9.8%)	2 (0.7%)

2.2. Data Collection From the Center for Cancer Genomics and Advanced Therapeutics (C‐CAT) and GENIE

C‐CAT is a national cancer genomics database storing clinical and genomic data from comprehensive genomic profiling tests in Japan [33]. The AACR Project GENIE is an international consortium integrating clinical and genomic cancer data [34]. C‐CAT data were accessed on February 29, 2024, via its portal, extracting LUAD cases from 2038 NSCLC patients. GENIE data were accessed on August 16, 2022, through cBioPortal, identifying 1667 LUAD cases.

2.3. WTS and WES for the Validation Cohort

The molecular profiling of the validation cohort was conducted and described elsewhere [31].

2.4. TOP2‐RNA for Mutation Calls

Total RNA was extracted from the fresh frozen samples. cDNA was synthesized from 200 ng RNA. Target enrichment used TOP2‐RNA‐V6 probes and the Twist Library Preparation EF Kit (Twist Bioscience, South San Francisco, CA). Sequencing on a NovaSeq 6000 (Illumina) generated paired‐end reads, aligned to hg38. Mutations were called using an in‐house method based on SAMtools' mpileup and filtered by depth (< 20), allele frequency (< 0.05), or presence in the 1000 Genomes Project or in‐house database.

2.5. TOP2‐RNA for Detecting Fusion Genes and Exon Skipping and Expression Analysis

Fusion genes were detected using STAR‐Fusion (v1.2.0; https://github.com/STAR‐Fusion). The in‐house pipeline mapped hypothetical fusions with BWA (v0.7.12; https://bio‐bwa.sourceforge.net/), requiring ≥ 30‐bp matched reads from breakpoints and ≥ 10 reads. Expression was analyzed as TPM from 1390 genes mapped to hg38 with STAR and calculated using featureCounts (subread v2.0.3; https://subread.sourceforge.net/).

2.6. Targeted Amplicon Sequencing of MET

Genomic DNA was prepared from the frozen samples. MET gene around exon 14 was PCR‐amplified using three primer sets (sequences in Doc. S1). Libraries were prepared, sequenced on MiSeq, and mapped to hg38 with BWA (v0.7.18‐r1243‐dirty). Mutations were called using Mutect2 (v4.3.0.0) and an in‐house pipeline with snpEff (v4.3t), then manually verified.

2.7. WTS for the Discovery Cohort

Total RNA was extracted from fresh frozen samples. A total of 200 ng of RNA was converted to cDNA. Using a paired‐end option, sequencing was conducted using NovaSeq 6000 (Illumina). Gene fusion was analyzed using STAR‐Fusion and Arriba (https://github.com/suhrig/arriba). Exon skipping was analyzed using an in‐house pipeline: RNA‐seq reads were aligned to hg38 and RefSeq (NCBI) with BWA, skipped exons were detected, virtual transcriptome sequences were generated, reads were re‐aligned to these sequences, and candidates were identified based on breakpoint reads.

2.8. Prognostic Signature Generation and Gene Set Enrichment Analysis (GSEA)

Univariate Cox regression identified 15 genes (p < 0.01, HR > 1) from 1390 in TOP2‐RNA, and four linked to prognosis were used for a predictive model. The risk score, calculated from gene expression levels and regression coefficients, is detailed in Doc. S1. Normalized raw counts (22,873 genes) were used for GSEA, performed on high‐risk (HR) and low‐risk (LR) groups using the hallmark gene set. OS and RFS were analyzed with Kaplan–Meier curves and log‐rank tests.

2.9. Cell Lines

HEK293T cells (RRID:CVCL_0063) and 3 T3 (RRID:CVCL_0594) fibroblasts were maintained in Dulbecco's modified Eagle's medium‐F12 (DMEM‐F12) with 10% FBS (both from Thermo Fisher Scientific). Authentication and Mycoplasma testing were not conducted within 6 months.

2.10. Retrovirus Preparation and Cell Transduction

Full‐length and fusion cDNAs of RASGRF1 (NM_001145648) and RASGRF2 (NM_006909) were cloned into pcx6 retroviral vector. Plasmids were generated by Azenta Life Sciences (Burlington, MA) and verified by Sanger sequencing. HEK293T cells produced recombinant retroviruses using packaging plasmids (Takara Bio, Shiga, Japan). 3T3 cells were infected with ecotropic retroviruses in 4 μg/mL polybrene (Sigma‐Aldrich, Burlington, MA) for 24 h.

2.11. Focus Formation Assay

3 T3 cells expressing variants were cultured for 2 weeks, then stained with Giemsa solution.

2.12. PrestoBlue Cell Viability Assay

The transformed 3 T3 cells expressing each RASGRF1/2 variant were cultivated in 96‐well plates with the MEK inhibitor cobimetinib (Selleck Chemicals, Houston, TX) at concentrations ranging from 0.1 nM to 10 μM for 4 days. A total of 10 μL PrestoBlue (Thermo Fisher Scientific) was added to the plates, and the fluorescence was measured.

2.13. Western Blotting

Protein separation and detection were performed using a Simple Western system and Compass software (ProteinSimple, San Jose, CA). Antibody information is provided in Doc. S1.

3. Results

3.1. The Validity of Targeted RNA Sequencing for Molecular Profiling of LUAD

The study overview is shown in Figure 1A. The validation cohort confirmed TOP2‐RNA's ability to screen driver oncogenes (Figure 1B and Table S1). TOP2‐RNA detected 88.7% (55/62) of the short nucleotide variants and indels, 92.3% (12/13) of MET ex14 skipping, and 100% (5/5) of oncogenic fusion genes identified by WTS or WES. Overall, 90% (72/80) of the targetable oncogene mutations identified by WTS or WES were detected by TOP2‐RNA. TOP2‐RNA identified additional EGFR, ERBB2, and BRAF mutations in 6, 1, and 2 samples. TOP2‐RNA also detected one EML4‐ALK fusion and two additional MET ex14 skipping cases.

FIGURE 1.

Study design and the validity of targeted RNA sequencing for the molecular profiling of LUAD. (A) The overview of the study. (B) TOP2‐RNA successfully identified the oncogenic mutations of EGFR, ERBB2, BRAF, and MAP2K1, as well as fusions and MET ex14 skipping in the validation cohort. (C) Comparison of RNA expression (TPM) between those identified by TOP2‐RNA and WTS. The RNA expressions of EGFR, ERBB2, and MET evaluated by TOP2‐RNA were highly concordant with WTS (r = 0.92, 0.98, and 0.87, respectively).

TOP2‐RNA TPM expression correlated with WTS (EGFR: 0.92, ERBB2: 0.98, MET: 0.87) (Figure 1C and Table S2). Overexpression of EGFR, ERBB2, and MET (mean + 3SD) was consistent between methods, except in one sample. Among the other 24 genes, overexpression was identified in 44 cases by TOP2‐RNA and 41 by WTS, with 15 overlapping cases (Figure S1).

3.2. Comparison of Gene Alteration Frequencies in LUAD Between Japan and the United States

Driver gene alteration frequencies in Japanese and US LUAD cohorts were compared to assess Comprehensive genomic profiling (CGP) test utility. While overall frequencies differed, they were similar when stratified by smoking history. EGFR mutations and fusion genes were more frequently identified for never‐smokers, whereas KRAS and BRAF mutations were more common for heavy or current smokers (Figure 2A and Table S3). Logistic regression indicates that smoking negatively correlated with EGFR, ALK, and RET alterations, while KRAS mutation was smoking‐associated (Figure 2A and Table S4). Asian ethnicity was an independent risk factor for EGFR mutation (p = 4.98 × 10⁻⁶, HR = 2.65) and significantly enriched even in never‐smokers (p = 5.0 × 10⁻⁴, Fisher's exact test, HR = 1.45–3.73).

FIGURE 2.

Comparison of the frequency of gene alterations in the discovery cohort to the real‐world data of LUAD in Japan and the United States. (A) Comparison of the frequency of gene alterations between patients LUAD in Japan and those in the United States and the clinical features related to gene alterations. The volcano plots (upper panels) indicate the clinical features correlating with gene alterations. The odds ratio (logistic regression) to acquire the stated gene alterations and p value are shown on the horizontal and vertical axes. All results are described in Table S4. The frequencies of driver gene alterations in the clinical sequencing cohorts in Japan and the United States are presented in the pie charts (middle panels). The case number and percentage (in parentheses) of the total number of the corresponding smoking statuses indicate the development of individual gene alterations. The gene alterations include mutations in EGFR, KRAS, ERBB2, BRAF, MAP2K1, MET ex14 skipping, and fusion genes of ALK, RET, ROS1, FGFR1/2/3, and NTRK1/2/3. Compound mutation means multiple driver mutations in one sample. (B) Driver mutations in the discovery cohort (n = 270). The distribution of driver mutations identified by the TOP2‐RNA panel in the discovery cohort is indicated. (C) The frequencies of oncogenic diver genes among cases without EGFR mutations in the discovery cohort were compared to those of the C‐CAT and GENIE cohorts. (D) Mutational analysis on the MET ex14 splice region. Target amplicon sequencing around the MET ex14 splice site was conducted in 25 cases with MET ex14 skipping transcripts detected by the TOP2‐RNA panel. Altogether, 41 MET mutations that could potentially affect MET ex14 splicing in 22 cases are shown.

C > A single base substitution, linked to smoking, was more frequent in smokers of C‐CAT and GENIE cohorts (21.2% vs. 12.6% for C‐CAT and 39.3% vs. 13.2% for GENIE) (Figure S2). This substitution was lower in heavy smokers in C‐CAT than in current smokers in GENIE (23.7% vs. 42.6%) but similar in never‐smokers (12.6% vs. 13.2%).

3.3. Targeted RNA Sequencing Identifies Frequent MET ex14 Skipping

In the discovery cohort (n = 270), TOP2‐RNA identified oncogenic mutations in 180 cases (Figure 2B and Table S5).

The frequencies of oncogenic mutations in EGFR, KRAS, ERBB2, BRAF, and MAP2K1 were 5.9%, 12.6%, 8.5%, 5.2%, and 3.7%, respectively. MET ex14 skipping was detected in 11.5% of cases. ALK, RET, ROS1, and NRG1 fusions were identified in 4.8%, 2.2%, 2.6%, and 3.0% (Figure S3). Rare fusions (KIF5B‐EGFR, TPM3‐NTRK1, MYB‐NFIB, MYBL1‐NFIB) were detected in 1, 1, 1, and 2 cases. Among the cases without EGFR mutations, mutation rates for KRAS, ERBB2, BRAF, and MAP2K1 were 12.1%, 10.2%, 5.3%, and 4.9%. MET ex14 skipping and ALK, RET, ROS1, and NRG1 fusions were 12.6%, 4.9%, 1.9%, 2.9%, and 3.4% (Figure 2C). The frequency of MET ex14 skipping was more frequent in the discovery cohort than in C‐CAT (12.6% vs. 2.6%, p < 0.0001 Fisher's exact test). As the current popular DNA‐based gene panel test does not target NRG1, NRG1 fusion was not detected in C‐CAT or GENIE.

mRNA overexpression was detected in EGFR (5), FGFR1 (3), FGFR2 (2), FGFR3 (1), MET (7), PDGFRA (5), PDGFRB (4), MDM2 (5), and CDK4 (4) cases (Figure 2A and Figure S4). Tyrosine kinase gene overexpression was observed in 27 (10%) cases. Overall, driver oncogenes were identified in 66.7% of LUAD cases.

3.4. Mutational Analysis on the MET ex14 Splice Region

Targeted amplicon sequencing of 25 MET ex14 skipping cases detected 41 splicing‐related mutations in 22 cases (Figure 2D and Table S6). In 12 cases, 14 mutations had variant allele frequencies (VAFs) > 10%. MET splice region mutation distributions in C‐CAT and GENIE are in Figures S5A and S5B. Canonical splice‐site mutation rates were 10.3% (discovery), 2.6% (C‐CAT), and 5.7% (GENIE) (Figure S5C).

3.5. Discovery of Novel Oncogenic RASGRF2 Fusions

WTS of 114 never/light smokers' samples, including 90 samples without driver oncogenes, identified 30 fusion genes in 16 cases by both STAR‐Fusion and Arriba, among them novel RDX‐RASGRF1 and PRKCI‐RASGRF2 fusions (Figure 3A and Table S7). A re‐evaluation of the validation cohort further revealed OCLN‐RASGRF2. These fusions retained the RAS‐GEF domain, encoded by exons 21–27 of RASGRF1 or exons 18–24 of RASGRF2, along with membrane‐localization domains (FERM, C1, and MARVEL).

FIGURE 3.

Discovery of novel oncogenic RASGRF2 fusions. (A) A schematic diagram of RDX‐RASGRF1, PRKCI‐RASGRF2, OCLN‐RASGRF2. (B) The RASGRF1/2 expression in the samples with RASGRF1/2 fusion. (C) The 3T3 focus formation assay. Compared with the wild type of RASGRF1/2 and negative control of GFP, more focus formations were observed in the 3T3 cells expressing the RASGRF1/2 fusion. All scale bars are 1 cm in length. (D) The sensitivity of RASGRF1/2 fusion to MEK inhibitor. The transformed 3T3 cells expressing each RASGRF1/2 variant were cultivated with MEK inhibitor cobimetinib at concentrations ranging from 0.1 nM to 10 μM for 4 days. (E) The RASGRF1 and RASGRF2 protein expression and MEK phosphorylation were evaluated by antibody‐based quantitative protein analysis. The images were generated using Compass software (ProteinSimple).

RASGRF1/2 fusions were found in women in their 60s–80s. The other clinical information is provided in Table 2, and the pathological images are shown in Figure S6A–C.

TABLE 2.

Clinicopathologic characteristics of patients with RASGRF1 or RASGRF2 fusions.

Sample ID	Fusion	Age	Gender	Smoking index	Primary tumor site	Surgery	Postoperative pathological diagnosis	Proportions of histological subtypes	Pathological stage	IASLC grade	STAS	Adjuvant chemotherapy	Postoperative course
LUAD_267	PRKCI‐RASGRF2	64	Female	0	Left lower lobe	Left lower lobectomy	Papillary adenocarcinoma	Papillary 60% Acinar 20% Micropapillary 20%	IA3 (pT1cN0M0)	Grade 3	+	UFT for 2 years	Alive with no reccurence at 5 years and 9 months.
LUAD_100	OCLN‐RASGRF2	74	Female	0	Right lower lobe	Right lower lobectomy	Papillary adenocarcinoma	Papillary 70% Lepidic 20% Acinar 10%	IIIA (pT1aN2M0)	Grade 2	−	CDDP + VNR (discontinued after one cycle due to AE)	Recurrent at 7 years after the surgery. Died 3 years later.
LUAD_308	RDX‐RASGRF1	81	Female	0	Left lower lobe	Left lower lobectomy	Solid adenocarcinoma	Solid 50% Acinar 20% Papillary 20% Micropapillary 10%	IA3 (pT1cN0M0)	Grade 3	+	−	Recurrent at 29 months after the surgery, and a left total lobecotmy was performed. Died 7 months later.

Abbreviations: AE, adverse events; CDDP, cisplatin; IASLC, the International Association for the Study of Lung Cancer; STAS, a spread through the alveolar spaces; UFT, uracil and tegafur; VNR, vinorelbine; +, present; −, absent.

RASGRF2 expression was the second highest in the PRKCI‐RASGRF2 sample (TPM = 40.4, average 7.1, STD 5.9), while RASGRF1 expression in RDX‐RASGRF1 was also relatively high (TPM = 44.2, average 16.8, STD 23.0) (Figure 3B).

To assess contact inhibition loss, 3 T3 cells expressing RASGRF1/2 fusions were cultured for 2 weeks and stained with Giemsa. They formed more foci than wild‐type RASGRF1/2 or GFP controls (Figure 3C).

Given the expected RAS activity of RASGRF2, MEK inhibitor cobimetinib sensitivity was tested. 3 T3 cells with RDX‐RASGRF1, PRKCI‐RASGRF2, and OCLN‐RASGRF2 showed higher sensitivity than controls (IC₅₀ = 47, 44, 78, and 516 nM, respectively) (Figure 3D). Western blot confirmed RASGRF1 and RASGRF2 protein expression and MEK phosphorylation in 3 T3 cells with RASGRF1/2 fusions (Figure 3E). MEK phosphorylation was concurrent with fusion protein expression and was partially inhibited by cobimetinib treatment.

3.6. Prognostic Biomarker of LUAD

There is a substantial risk for recurrence and death in patients with early‐stage LUAD, even after undergoing complete surgical resection. The use of adjuvant therapy in LUAD at the early stages, particularly at stage I, remains controversial because no consistent survival benefit was demonstrated in previous randomized trials. Reliable prognostic biomarkers are critically needed to identify the patients at high risk for recurrence and who might benefit from additional systemic therapies. Our previous study has not shown a difference in postoperative prognosis based on the presence or absence of driver mutations [32].

We analyzed 1390 genes with a standard deviation of TPM of > 1.0 to ensure adequate variance. Univariate Cox proportional hazards regression analysis in the discovery cohort showed that 15 genes were significantly correlated with RFS (p < 0.01) (Figure S7), although the genes with lower statistical significance may also be important. Among the 15 genes, a prognostic model with four genes, including CA9, GAST, IGF2BP3, and UGT1A1, was generated because the expressions of these genes are prognostic markers in lung cancer [35, 36, 37, 38, 39, 40].

We constructed a risk score with the regression coefficients from this model and determined the optimal cut‐off value to be 2.43 using the surv_cutpoint from the survminer package (Figure 4A). High‐risk patients, as defined by a risk score based on the four‐gene signature, had significantly worse RFS and OS for all stages (p < 1.1 × 10⁻⁸ and p < 7.8 × 10⁻⁸) and for stage I patients (p = 1.4 × 10⁻³ and p = 5.2 × 10⁻⁴) in the discovery cohort, independent of age, gender, smoking index, stage, and gene mutations (Figure 4B).

FIGURE 4.

Four‐gene prognostic signature in LUAD in Japanese patients who have never smoked or only lightly smoked. (A) Four‐gene expression and risk score distribution in the discovery cohort by z‐score (in the lower part). Here, red indicates a higher expression, whereas light blue indicates a lower expression. The risk scores for all patients are plotted in ascending order (in the middle part) and marked as a low risk (blue) or a high risk (red), as divided by the threshold (horizontal black line). The risk score threshold is 2.43. The clinical information and mutational status of the individual samples are shown in the upper part. SI, smoking index; NA, not available. (B) Kaplan–Meier curves of overall survival and recurrence‐free survival for all stages (upper) or for stage I (lower) in the discovery cohort stratified by the four‐gene prognostic signature into those at high and low risk. A univariate Cox analysis was used to calculate the hazard ratio (HR). The HR, 95% confidence interval (CI), p value, and median survival are shown. (C) Statistically significant gene sets were identified through a GSEA analysis to be differentially overexpressed in high‐risk tumors. Table S8 presents the full GSEA results. NA, not available. Two‐sided likelihood ratio test. CI, confidence interval; LR, low risk; HR, high risk.

Using the same risk score cut‐off value, the four‐gene prognostic signature significantly stratified the validation cohort for the RFS and OS for all stages (p = 4.4 × 10⁻³ and p = 3.9 × 10⁻³) and for stage I patients (p = 3.1 × 10⁻² and p = 3.2 × 10⁻²), which was independent of age, gender, stage, and gene mutations (Figure S8).

Finally, in a multivariable Cox analysis that includes the EGFR and ALK alteration status, the risk score was significant (HR = 4.4, p < 7.7 × 10⁻⁵) (Table 3). None of the mutation statuses was statistically significant in the multivariable analysis.

TABLE 3.

Cox proportional hazard models of discovery cohort of Japanese lung adenocarcinoma.

Variables	Categories	No.	Univariate analysis			Multivariate analysis
			Hazard ratio	95% CI	p value	Hazard ratio	95% CI	p value
Gender	Female	183
	Male	87	2.51	1.34–4.72	4.1E‐03	1.63	0.85–3.11	0.14
Age	< 40	11
	40–59	57	0.73	0.08–6.50	0.78
	60–79	170	1.61	0.22–11.9	0.64
	≥ 80	32	3.82	0.48–30.1	0.20
Smoking index	0–100	48
	≥ 101	222	1.15	0.53–2.50	0.73
Stage	I	191
	II	23	2.90	1.08–7.83	0.035	1.95	0.69–5.45	0.21
	III	28	7.03	3.37–14.7	2.1E‐07	4.44	1.94–10.2	4.2E‐04
	IV	5	14.9	4.24–52.2	2.5E‐05	18.6	4.82–72.1	2.3E‐05
Mutations	TP53	62	2.92	1.55–5.51	9.1E‐04	0.99	0.45–2.16	0.98
	MET	38	1.34	0.59–3.03	0.49
	KRAS	38	1.65	0.76–3.59	0.21
	ERBB2	31	1.21	0.47–3.09	0.69
	EGFR	18	0.00	0.00‐Inf	1.00
	BRAF	16	0.40	0.05–2.89	0.36
	ALK	14	0.43	0.06–3.11	0.40
	MAP2K1	12	0.55	0.08–4.02	0.56
	ROS1	9	0.84	0.12–6.15	0.87
	NRG1	8	0.00	0.00‐Inf	1.00
	PIK3CA	7	1.15	0.16–8.37	0.89
	MYC	7	2.04	0.49–8.49	0.33
Risk score	LR	217
	HR	53	5.62	2.99–10.6	7.8E‐08	4.4	2.11–9.18	7.7E‐05

Note: Two‐sided likelihood ratio test.

Abbreviation: CI, confidence interval.

GSEA was performed to understand the biological underpinning of high‐risk tumors. We found significant enrichment for high‐risk tumors for the gene sets related to cancer biology, including the E2F targets, G2M checkpoint, and MYC targets (Figure 4C and Table S8).

4. Discussion

In this study, TOP2‐RNA identified driver mutations in two‐thirds of the patients previously negative for common EGFR mutations. Given that these mutations account for 55.5% of LUAD cases in never and light smokers [31], our data suggested that TOP2‐RNA can detect driver oncogenes in > 85% of this tumor type. This supports the clinical utility of RNA‐based sequencing for identifying driver mutations, including small nucleotide variants and fusion genes, and exon skipping.

Particularly, the detection rate for MET ex14 skipping was higher in our cohort than in two real‐world datasets, likely due to DNA sequencing's limitations in identifying the diverse alterations that cause this event. MET ex14 skipping is caused by not only canonical splice site mutations but also relatively long deletions or mutations in deep introns, which are hard to detect using current exon‐targeted DNA panels with short‐read sequencing. Previous studies have reported the low sensitivity of DNA‐based testing for MET ex14 skipping [41, 42]. Approximately half of the variants identified by the amplicon sequence exhibited a VAF of < 10%, indicating inefficient hybridization or amplification of mutant alleles compared to wild‐type in DNA library prep.

While TOP2‐RNA showed high concordance with WES and WTS for driver detection, some drivers were uniquely identified by either method. This may reflect differences in analysis pipelines or the detection threshold around VAF cutoffs.

A key finding in this study is the identification of the novel RASGRF1/2 fusions. Especially, this study first identified the RASGRF2 fusion in patients with LUAD, although s in melanocytic lesions [43]. Previous studies identified several RASGRF1 fusions, including OCLN‐RASGRF1 and TMEM87A‐RASGRF1 in LUAD, which led to an increased RASGRF1 expression [44, 45]. Moreover, TMEM154‐RASGRF1, SLC4A4‐RASGRF1, and IQGAP1‐RASGRF1 were identified in a patient with relapsed acute myeloid leukemia, a KRAS wild‐type pancreatic adenocarcinoma (PDAC) cell line of PaCaDD137, and a giant cell sarcoma in the TCGA study [44, 46].

A common feature of RASGRF1 and RASGRF2 fusions is the involvement of 5′ partner genes that encode membrane‐associated proteins. This membrane localization likely facilitates anchoring of the RAS‐GEF catalytic domain at the plasma membrane, thereby promoting RAS/MAPK signaling activation. OCLN, a gene encoding tight junction protein localized at the plasma membrane, was a recurrent fusion partner found in previous and present cohorts.

While clinical applications for RASGRF1/2 fusions remain undeveloped, previous studies have suggested that the MEK inhibitor is a potential drug for RASGRF1 fusion [45]. Thus, the treatment could possibly be extended to RASGRF2 fusions and, theoretically, to RASGRF3 and RASGRF4 fusions, as well as other fusions of the RASGEF family. Activating GEF SOS1 mutations in LUAD [47] and ongoing trials of SOS1‐KRAS inhibitor combinations [48, 49] highlight additional GEF‐mediated RAS activation mechanisms. Our findings propose the RAF–MEK–ERK pathway as a potential therapeutic target in RASGRF‐rearranged tumors, alongside other RAS signaling targets like SHP2.

There are some possible reasons why RASGRF2 fusions have not been reported in previous studies. First, RASGRF2 fusion may be distinct in Asian never smokers. Some of the driver oncogenes are related to specific clinical backgrounds. Our data, as well as previous studies, revealed ALK and RET fusions are more frequent in never smokers, while EGFR mutation is more common in Asians. OCLN‐RASGRF1 and TMEM87A‐RASGRF1 fusions in the previous studies were identified in female, never‐smoker patients [44, 45]. Second, the large‐scale cohorts of CGP test use DNA‐based targeted panels that do not adequately capture rare fusion events, as they only cover several common fusion genes. For the detection of novel rare fusion events, whole transcriptome or whole genome sequencing is required. Third, there is currently no universally accepted gold standard for fusion detection, and the whole transcriptome performance is highly dependent on the RNA quality. mRNA enrichment by poly‐A capture is not suitable for the fusion detection using degraded RNA, such as that extracted from FFPE. Further studies will be required to establish the frequency of RASGRF1/2 fusions and other GEF alterations in NSCLC, PDAC, and other malignancies.

While genomic/transcriptomic testing is not yet routine in early‐stage disease, reliable prognostic biomarkers may help identify the patients who are at high risk for recurrence and who might benefit from additional systemic therapies. Our study findings can be easily applied in a clinical setting because the clinical grade test covering the same bait with TOP2‐RNA is available in clinical practice as GenMinTOP (Konica Minolta, Tokyo, Japan), which has been approved by the Pharmaceuticals and Medical Devices Agency (PMDA) and covered by national health insurance of Japan. This RNA‐based assay enables both driver detection and prognostic signature evaluation.

Mutational profiling of LUAD in C‐CAT and GENIE revealed that driver mutation patterns were more strongly associated with smoking status than ethnicity: KRAS was common in smokers, while EGFR, ALK, and RET fusions were enriched in never‐smokers. Considering that the incidence of LUAD in never‐smokers is increasing, it is highly anticipated that the etiology of the disease will be elucidated to reduce the incidence. Epidemiologic studies have identified the risk factors for lung cancer in never‐smokers, including family history of lung cancer, second‐hand tobacco smoke, indoor cooking/heating fumes, occupational exposures, and underlying lung disease [50, 51].

This study has several limitations. First, the C‐CAT and GENIE cohorts are different from our study cohort. Given that gene panel testing is currently approved for patients with advanced cancer who are refractory to the standard treatment in Japan, C‐CAT data may have a selection bias. In fact, the frequency of EGFR mutations in the C‐CAT cohort is lower than that of previous studies of LUAD in chemo‐naïve patients. Second, the four‐gene signature was only assessed at initial surgery and requires validation in recurrent cases. Third, its utility in non‐Japanese populations remains untested. Finally, while WTS uncovered 30 novel fusions in 16 cases, most lack functional characterization.

In conclusion, our study findings can be easily applied in a clinical setting. Moreover, our data indicate the value of applying gene expression profiling for predicting the prognosis after a surgical operation and that identifying actionable mutations is important for optimizing the targeted drugs. RASGRF2 fusions are novel actionable fusions, which several tyrosine kinase inhibitors, including MEK inhibitors, could target. We believe that the transcriptomic analysis, as well as the genomic analysis, highlights the importance of precise tumor profiling to provide the best possible care to patients.

Author Contributions

Yuki Terashima: formal analysis, investigation, visualization, writing – original draft. Soohwan Park: data curation, investigation. Hiroshi Ikeuchi: investigation. Takuo Hayashi: investigation, visualization. Shinya Kojima: funding acquisition, software, visualization. Toshihide Ueno: formal analysis, software, visualization. Masachika Ikegami: formal analysis, software, visualization. Rina Kitada: investigation, visualization. Yoshiyuki Suehara: supervision. Shinya Tanaka: supervision. Kenji Suzuki: supervision. Hiroyuki Mano: funding acquisition, methodology, resources, supervision. Kazuya Takamochi: conceptualization, data curation, funding acquisition, project administration, resources, supervision. Shinji Kohsaka: conceptualization, funding acquisition, investigation, methodology, project administration, supervision, validation, visualization, writing – original draft.

Ethics Statement

This study was approved by the Ethics Committee of the National Cancer Center (2015‐202) and Juntendo University School of Medicine (No. 2014176). This study was conducted in accordance with the Declaration of Helsinki. Informed Consent: all patients provided signed consent forms. Registry and the Registration No. of the study/trial: N/A. Animal Studies: N/A.

Consent

All authors are in agreement with the content of the manuscript.

Conflicts of Interest

S. Kohsaka is an editorial board member of Cancer Science. S. Kohsaka reports research grants from Boehringer Ingelheim, Chordia Therapeutics, Eisai, Konica Minolta, CIMIC, and H.U. Group Research Institute. Other authors declare no conflicts of interest.

Supporting information

Data S1.

Table S1.

Figure S1.

Data Availability Statement

We deposited the raw sequencing data under accession number hum0094.v11 in the Japanese Genotype–Phenotype Archive, which is hosted by the DNA Data Bank of Japan [accession number JGAS000215].