The National Cancer Center/Cancer Hospital, Chinese Academy of Medical Sciences (NCC, CAMS, Beijing, China) cohort consisted of 475 patients who had undergone R0 surgical resection and were diagnosed with LUAD between June 2006 and June 2014 in the Department of Thoracic Surgery, NCC. The inclusion criteria were as follows: (1) radical surgery with R0 resection; (2) histologically confirmed LUAD. The exclusion criteria were as follows: (1) if the patients received preoperative chemotherapy and (or) radiotherapy; (2) if the patients lacked detailed clinical information; (3) if the patients lost to regular follow-up. Seventy patients were excluded from this study. The flowchart of the enrollment process is shown in Figure 1 . Invalid data (n = 60) were excluded, and finally pathological tissue samples and clinical information, including age, sex, smoking history, tumor differentiation status, T stage, lymph node metastasis, and TNM stage were obtained from 415 patients with LUAD. All LUAD tissues were preserved in the NCC biobank and pathologically confirmed by two experienced pathologists.

This research was conducted in accordance with the Declaration of Helsinki and was approved by the Clinical Research Ethics Committee of the National Cancer Center/Cancer Hospital, CAMS.

Patients were followed up every 3–6 months in the outpatient clinic for the first 2 years after surgery, and then annually. Follow-up comprised recording medical history, survival status, physical examination, and chest computed tomography. Final follow-up was March 4, 2019.

TCGA is a huge cancer genetic data sharing platform containing large amounts of clinical and genomic information to help researchers better understand the genetic basis of cancer and improve their ability to diagnose, treat, and prevent cancer. We downloaded RNA-seq expression profiles and corresponding clinical information for 416 patients with LUAD. RNA expression levels were expressed as reads per kilobase per million mapped reads (RPKM) and converted to transcripts per million (TPM) for comparisons among patients stratified by clinical variables including sex, age, smoking history, T stage, N stage, M stage, TNM stage, ethnicity, and OS data.

The Oncomine database is a website-based cancer-related gene database that contains GEO, TCGA, cancer gene chips, and published research data, and is a central repository storing microarray data, sample data, and analysis results (28). The database contains 715 gene expression datasets, comprising gene expression data from 86,733 cancer and normal tissues. Nine LUAD data sets, including tumors and normal tissues, were extracted for analysis of HNRNPC expression. Gene expression profiling interactive analysis (GEPIA) is a newly-developed interactive web server for analysis of RNA sequencing expression data from 9,736 tumors and 8,587 normal samples from TCGA and GTEx, using a standard processing pipeline (29). GEPIA can be used to analyze tumor/normal differential expression, based on cancer type or pathological stage, as well as for patient survival analysis. In the present study, both Oncomine and GEPIA were used for preliminary investigation of HNRNPC expression. Furthermore, GEPIA was applied for survival analysis of data from TCGA LUAD cohort. Median TPM value was chosen as the cutoff point for grouping.

The online survival analysis tool, Kaplan–Meier plotter, integrates multiple cancer data from the Gene Expression Omnibus (GEO), TCGA, and European Genome-phenome Archive (EGA), including lung cancer, breast cancer, gastric and ovarian cancer, and hepatocellular carcinoma data (30). In this research, K-M plotter was used to analyze the correlation between HNRNPC and OS in patients with LUAD, followed by survival curve construction. According to median mRNA expression values, LUAD patients were divided into two groups, and prognostic value analyzed according to hazard ratio (HR) and log-rank p value. Subsequently, data from the NCC cohort were used to verify the prognostic value of HNRNPC.

The STRING database (http://string-db.org) aims to provide critical assessment and integration of protein-protein interactions, including direct (physical) as well as indirect (functional) associations (31). A PPI network, centered on HNRNPC, was explored using the “Bioconductor clusterProfiler”, “enrichplot”, and “ggplot2” packages in R (version 3.6.3), and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways and Gene Ontology (GO) enrichment analysis of molecular functions (MF), biological processes (BP), and cellular components (CC) of the PPI network conducted. An interaction score > 0.4 was considered significant.

The Cancer Cell Line Encyclopedia (CCLE, www.broadinstitute.org/ccle) is a large-scale public database in collaboration between the Broad Institute and the Novartis Institute of Biomedicine and the Genomics Institute of the Novartis Research Foundation. It contains details of 84,434 genes and 1,457 cell lines. The data will help us in-depth research on the molecular characteristics of cancer phenotypes, DNA methylation and potential targets. The project covers most common cancer types, such as lung cancer, liver cancer, stomach cancer, breast cancer and kidney cancer (32). In order to explore the possible molecular mechanisms of HNRNPC in lung cancer cell lines, we downloaded and compiled RNA-seq data of 188 lung cancer cell lines on the CCLE website. Taking the HNRNPC high expression group as a reference, the RNA-seq of these lung cancer cell lines was grouped for subsequent analysis.

GSEA (http://software.broadinstitute.org/gsea/index.jsp) is a powerful algorithm that can identify different gene sets, sort the predefined genes according to the degree of differential expression in two types of sample, and distinguish biological functions or molecular pathways that are significantly aggregated between two different biological states (33). To determine pathways closely related to HNRNPC expression, data from 515 patients with LUAD from TCGA were subjected to GESA. Median HNRNPC expression in the high and low groups were used as standards. Hallmark in GSEA was used to identify predefined gene sets; 5,000 permutations were performed, according to the gene set, to determine p-values. Pathways with p < 0.01 and false discovery rate (FDR) < 0.25 were considered significant, as listed in the Results section. Finally, we performed GSEA on the RNA-seq data of 188 lung cancer cell lines from CCLE to further explore the signal pathways of HNRNPC that are significantly related in cell lines.

Tissue microarrays (TMAs) were prepared from 415 LUAD patient tissue blocks retrieved from the hospital biobank. TMAs were immune-stained using immunohistochemistry methods. Briefly, representative areas of tumor tissue on Hematoxylin and Eosin (H&E) stained sections were identified and labeled by pathologists. TMAs were deparaffinized and rehydrated in sequence, treated with 2 N HCl for 15 min, and then treated with 100 mM Tris HCl (pH 8.5) for 10 min. Samples were then treated with H₂O₂ (3%) and goat serum at room temperature for 30 min. After confirmation of blocking, rabbit anti-HNRNPC polyclonal antibody (1:50, HPA051075; Sigma-Aldrich, St. Louis, MO) was added and incubated overnight at 4°C. Then, samples were incubated with polyclonal peroxidase-conjugated anti-rabbit IgG (Zhongshanjinqiao, Beijing, China) for 20 min at room temperature, according to the manufacturer’s instructions.

Two experienced pathologists, blinded to the clinical data, independently reviewed TMA IHC staining. The percentage of positive cells and staining intensity of positive cells were used to generate IHC HNRNPC staining scores, calculated based on the percentage of positive cells, as follows: score 0, 0%–25% HNRNPC-positive staining; score 1, 25%–50% HNRNPC-positive staining; and score 2, > 50% HNRNPC-positive staining. Another score was calculated based on the staining intensity of positive cells, as follows: 0, no staining; 1, weak; 2, moderate; and 3, strong. The product of these two scores was the IHC score for each LUAD patient. In the present study, tissues with scores ≤ 1 were classified into the low expression group, and tissues with scores ≥ 2 were classified into the high expression group.

Nonparametric tests were used to assess correlations between expression of HNRNPC and clinicopathological parameters (a Wilcox test for comparisons between two groups of data and the Kruskal test for three or more groups) in R software (version 3.6.3; The R Foundation for Statistical Computing). HNRNPC expression and clinical data were analyzed by χ² text or Fisher’s exact test. Function enrichment analyses were performed using the Bioconductor package in R. Kaplan-Meier survival analysis and the log-rank test were used to evaluate the prognostic value of HNRNPC. Multivariate Cox regression was applied to identify independent prognostic factors. A p value < 0.05 was considered statistically significant, and the median TPM set as the cutoff value.

First, data from the Oncomine database were used to analyze HNRNPC expression in various cancer types. Abnormal HNRNPC expression was retrieved from a total of 441 datasets. Among them, HNRNPC expression levels were significantly increased in tumor tissues in 59 datasets, including brain and central nervous system cancer, breast cancer, cervical cancer, colorectal cancer, esophageal cancer, gastric cancer, head and neck cancer, kidney cancer, lung cancer, myeloma, and many other cancers ( Figure 2A ). In all LUAD datasets, HNRNPC expression was significantly higher in tumor than in normal tissues ( Table 1 ). Analysis of HNRNPC expression in TCGA pan-cancer data using GEPIA showed high expression in a variety of tumors ( Figure 2B ). In TCGA LUAD cohort, HNRNPC expression was higher in tumor than normal tissues ( Figure 2C ). Moreover, Kaplan-Meier curves showed that the prognosis of LUAD patients with high HNRNPC expression was significantly worse than that of those with low HNRNPC expression ( Figure 2D ).

Next, we determined whether HNRNPC expression in LUAD tissues was associated with clinicopathological features of patients in TCGA dataset. As shown in Figures 3A–F , high HNRNPC expression was significantly associated with age (p < 0.001), sex (p < 0.001), smoking history (p < 0.001), ethnicity (p < 0.001), nodal metastasis (p < 0.001), and TNM stage (p < 0.001). In addition, after stratification according to different clinical factors, significant differences were observed among tumor subgroups and the reference/control group (p < 0.05) ( Figures 3A–F ).

To further determine the association between these variables and OS in TCGA datasets, we performed univariate and multivariate Cox logistic regression analyses ( Table 2 ). In the univariate model, pT stage (hazard ratio [HR] = 1.168; 95% confidence interval [CI] = 1.076–1.269; p < 0.001), pN stage (HR = 1.374; 95% CI = 1.184–1.594; p < 0.001), TNM stage (HR = 1.207; 95% CI = 1.128–1.29; p < 0.001), and HNRNPC expression (HR = 1.991; 95% CI = 1.478–2.681, p < 0.001) were significantly correlated with OS. Multivariate Cox regression analysis showed that TNM stage (HR = 1.144; 95% CI = 1.042–1.256; p = 0.005), and HNRNPC expression (HR = 1.691; 95% CI = 1.094–2.613; p = 0.018) were independent prognostic factors in the LUAD cohort from TCGA.

To evaluate HNRNPC expression in LUAD specimens, we performed IHC staining on tissue samples from 415 patients from NCC; 211 and 204 patients with LUAD were classified into the high and low expression groups, respectively. Representative photomicrographs of H&E and HNRNPC immunohistochemical staining are shown in Figure 4 . The Kaplan-Meier plotter database was used to explore the relationship between HNRNPC expression levels and OS of 720 patients with LUAD and HNRNPC probe information. The K-M curve for all patients showed that high HNRNPC levels were significantly correlated with poor OS (p < 1.1e-07) ( Figure 5A ). Subsequently, survival analysis of patients with LUAD from the NCC cohort, by construction of a K-M survival curve, also showed that high HNRNPC expression was significantly correlated with poor OS (p < 0.05) ( Figure 5B ).

We next determined whether HNRNPC expression in LUAD tissue was associated with patient clinicopathological characteristics. As shown in Table 3 , other than smoking history (p < 0.05), HNRNPC expression was not significantly associated with age, sex, tumor diameter, tumor differentiation, T stage, N stage, or pathological TNM stage.

Univariate and multivariate Cox regression analyses were also performed using NCC cohort data, to verify the independent prognostic value of HNRPNC expression ( Table 4 ). In univariate analysis, OS was significantly associated with age, sex, smoking history, tumor size, tumor differentiation, T stage, lymph node metastasis, pathological TNM stage, and HNRNPC expression (p < 0.05). Further, in multivariate analysis, age, tumor size, lymph node metastasis, pathological TNM stage, and HNRNPC expression remained significantly correlated with OS (p < 0.05).

A PPI network was constructed from the STRING database to investigate the relationships among 10 molecules that were significantly related to HNRNPC, namely cell division cycle 5 like (CDC5L), heterogeneous nuclear ribonucleoprotein K (HNRNPK), ELAV like RNA binding protein 1 (ELAVL1), Aly/REF export factor (ALYREF), serine/arginine rich splicing factor 1 (SRSF1), heterogeneous nuclear ribonucleoprotein L (HNRNPL), heterogeneous nuclear ribonucleoprotein H1 (HNRNPH1), heterogeneous nuclear ribonucleoprotein M (HNRNPM), heterogeneous nuclear ribonucleoprotein A2/B1 (HNRNPA2B1), and heterogeneous nuclear ribonucleoprotein A1 (HNRNPA1) ( Figure 6A ). To explore the functions of all proteins involved in the PPI network, we performed GO and KEGG functional enrichment analyses; significantly enriched terms are presented in Figure 6B . Subsequently, we performed GSEA of HNRNPC using Hallmark gene sets. The results showed that, in samples expressing high levels of HNRNPC, many gene sets were positively enriched, with E2F targets, MYC targets, G2M checkpoint, and oxidative phosphorylation the top four most significantly related biological processes ( Figures 7A–D and Table 5 ). The top 100 genes that were significantly positive or negatively correlated with HNRNPC expression are presented in Figure 7E . In addition, the RNA-seq data of 118 lung cancer cell lines from the CCLE database are divided into two groups according to the average expression level of HNRNPC. As shown in Figures 8A–D , GSEA analysis demonstrated that HNRNPC was significantly associated with several KEGG signaling pathways, including pyrimidine metabolism, purine metabolism, spliceosome and cell cycle. The top 10 KEGG pathways enriched by HNRNPC in 118 lung cancer cell lines from CCLE are listed ( Table 6 ).