The datasets used in the present study were downloaded from The Cancer Genome Atlas (TCGA) database (https://www.cancer.gov/about-nci/organization/ccg/research/structural-genomics/tcga; Project ID, TCGA-LUAD). The transcriptome data of 24 relapsed patients, including 14 men and 10 women, was screened. Additionally, the transcriptome data of paracancerous tissues from 53 patients (33 men and 20 women) was included. The soft connectivity function in the weighted gene co-expression network analysis (WGCNA; v1.69; http://cran.r-project.org/web/packages/WGCNA/index.html) package (23) was used to assess the effects of different power values on the co-expression network and co-expression modules in the scale independence and average connectivity by Pearson's correlation analysis. The ‘randomly selected gene’ parameter was set to 5,000, and all other parameters were set to the default values. The expression values were summarized using the collapse rows function in the WGCNA package, and cluster analysis was subsequently performed using flashClust v1.01-2 (https://cran.r-project.org/web/packages/flashClust/index.html) (24). The interaction/association of each module was visualized using heat maps. In addition, Gene Ontology (GO) analysis of the gene modules of interest was performed using the Database for Annotation, Visualization and Integrated Discovery (DAVID; http://david.ncifcrf.gov/summary.jsp). The differential expression of genes was defined as log(fold-change)|<0.6|; false discovery rate <0.05. Cytoscape v3.11 was used to construct co-expression network (https://cytoscape.org/download.html). Kaplan-Meier survival analysis of the corresponding genes was performed using the Oncomine database (https://www.oncomine.org/resource/login.html) and the P-values were calculated using the log-rank test.

Blood samples were collected from 132 patients with LUAD who underwent excision surgery at the First Affiliated Hospital of the Second Military Medical University (Shanghai, China) between February 2006 and March 2010. All patients received systematic treatment following surgery, including optimal local treatments, such as radiochemotherapy and targeted therapy. Disease recurrence was based on chest X-ray or CT imaging. The present study was performed in accordance with the Declaration of Helsinki and Good Clinical Practice, and was approved by the Ethics Committee of the First Affiliated Hospital of the Second Military Medical University. All patients provided oral informed consent for the use of their blood for scientific purposes.

Blood samples (15 ml) from 105 patients with early stage LUAD of the 132 patients (pathological stage I, II and IIIA) were initially collected 90 days after surgery. The second, third and fourth samples were collected 180 days, 1 year and 2 years after surgery, respectively. Blood samples were no longer collected if disease recurrence was detected during repeated examinations. The samples were collected in 5-ml heparin anticoagulation tubes and immediately used to extract free mRNA according to the manufacturer's protocol (Qiagen, Inc.; cat. no. 72022).

Free mRNA was reverse transcribed using the reverse transcription (RT) kit according to the manufacturer's protocol (Takara Bio, Inc.; cat. no. 639522). A quarter of the RT product was mixed with a pre-formulated 2X SYBR Green PCR mix (Roche Diagnostics; cat. no. 06924204001) containing uroplakin 2 (UPK2) RT-quantitative (q)PCR primers (UPK2 forward, 5′-CACTGAGTCCAGCAGAGAGATC-3′ and reverse, 5′-ACAGAGAGCAGCACCGTGATGA-3′; GAPDH forward, 5′-GTCTCCTCTGACTTCAACAGCG-3′ and reverse, 5′-ACCACCCTGTTGCTGTAGCCAA-3′) and was subsequently supplemented with distilled water to reach a final volume of 20 µl. Amplification was performed to obtain the dissolution curve, using the following thermocycling conditions: 94°C for 5 min, followed by 40 cycles at 94°C for 2 min, 60°C for 1 min and 72°C for 2 min, and a final extension step at 72°C for 2 min. Relative UPK2 expression was calculated using the 2^−ΔΔCq method (25) and normalized to the internal reference gene GAPDH.

The serum expression levels of UPK2 were detected in triplicate. GraphPad Prism 8.0 (GraphPad Software Inc.) was used for statistical analysis. The data are presented as the mean ± SD. The difference between two groups was analyzed by unpaired Student's t-test. The difference between paired samples (before and after relapse) was analyzed by paired Student's t-test. P<0.05 was considered to indicate a statistically significant difference. The receiver operating characteristic (ROC) curve of LUAD recurrence was assessed using the pROC package v1.16.2 (https://cran.r-project.org/web/packages/pROC/index.html) (26) within R software, and the area under the curve (AUC) was calculated. The survival curve was plotted using the ggsurvplot package v0.4.8 (https://cran.r-project.org/web/packages/survminer/readme/README.html) within R language and the log-rank test was used to obtain the P-value.

In order to establish a co-expressed gene network associated with postoperative recurrence of LUAD, WGCNA was used to assess the gene expression profile data from patients with recurrent LUAD within the TCGA database. The transcriptome data of 24 relapsed patients, including 14 men and 10 women, were screened. The association between genes was calculated using WGCNA, and the gene expression data of patients with recurrent LUAD were classified into 39 gene modules using unsupervised average linkage hierarchical clustering and were labelled in a heat map with different colors (Fig. 1A). Gene modules of different colors contained mutually exclusive co-expressed genes. Genes that could not be classified into a specific module were incorporated into gray modules. WGCNA can determine the association between gene modules and a series of phenotypes (23); therefore, this method was used to assess the association between the specific gene modules of patients with postoperative recurrent early stage LUAD and a series of phenotypes, including age, sex, survival, recurrence, recurrence type and pathological stage (27). Without any phenotypic or genetic preferences for module partitioning, the results demonstrated that the purple module was significantly associated with survival and recurrence, with Pearson's correlation coefficients >0.7 (Fig. 1B). Overall, the present results suggested that these genes and their co-expression patterns may be associated with the recurrence of LUAD.

WGCNA classifies co-expressed genes of patient samples into specific modules associated with a series of traits regulated by the same mechanism (23). The purple module was identified as the most relevant to postoperative recurrence. In order to verify the association between the co-expressed genes within the purple module and LUAD, a heat map containing the gene expression levels of 24 recurrent tumor tissues and 53 paracancerous tissues from TCGA was constructed. The results demonstrated a marked difference in the expression pattern of the purple module between paracancerous and recurrent tumor tissues (Fig. 2A). However, the expression pattern of the purple module genes in patients with recurrent tumor tissues was not as consistent compared with paracancerous tissues; the expression pattern in tumor tissues exhibited three expression patterns of light red, light blue and deep red, suggesting different mechanisms of relapse (Fig. 2A). Of the 117 genes in the purple module, 68 genes exhibited significant differences (data not shown) between recurrent tumor tissues and paracancerous tissues [log(fold-change)|<0.6|; false discovery rate <0.05]. These 68 genes were used to construct an expression heat map of the two types of tissues (Fig. 2C), which demonstrated that the expression patterns of the 68 genes were more uniform than that of the heat map constructed using 117 genes. The biological function of the 117 genes in the purple module was further analyzed via GO analysis using DAVID, with the most significant GO term being ‘cytosol’ (P=0.0356; Fig. 2B).

Gene significance is closely associated with gene connectivity, which means that nodes with higher connectivity in the co-expression network serve an important role in the process of performing biological functions (28). Therefore, a co-expression network of genes for LUAD recurrence was constructed, identifying 47 edges and 17 nodes (power=8; Fig. 3A). The results demonstrated that there were four genes [UPK2, kelch domain containing 3 (KLHDC3), galanin receptor 2 (GALR2) and tyrosinase-related protein 1 (TYRP1)] in the co-expression network with more nodes linked appearing in the purple module, which were significantly associated with survival and recurrence (Table I and Fig. 3B-E). Among these genes, high expression levels of UPK2, KLHDC3 and GALR2, and low expression levels of TYRP1 were associated with a poor prognosis (Table I and Fig. 3B-E). The function of these four genes in LUAD was further analyzed using clinical data downloaded from the Oncomine database. The results demonstrated that patients with low expression levels of UPK2, KLHDC3 and GALR2 had significantly improved survival outcomes than those with high expression levels of UPK2, KLHDC3 and GALR2 (P=4.9×10⁻⁵, P=1.7×10⁻⁵ and P=0.0093, respectively; Fig. 3B-D). Conversely, patients with high TYRP1 expression had a significantly improved prognosis than those with low TYRP1 expression (P=2×10⁻⁷; Fig. 3E).

Previous studies have reported that free mRNA in plasma has the potential to act as a tumor marker (29,30). Table II presents the demographic information and clinical characteristics of the 105 patients with early stage LUAD who met the study criteria out of 132 patients. Of these LUAD patients, 58 were men (55%) and 47 were women (45%), with a median age of 58 years (age range, 39–83 years). The pathological stage of most patients was stage I or II (83%), whilst the remaining patients were at stage IIIa (17%). Following surgery, 43 patients received adjuvant therapy (41%), including 35 patients receiving radiotherapy, 4 patients receiving adjuvant chemotherapy and 4 patients receiving both radiotherapy and chemotherapy (Table II).

Patient blood samples were collected and assessed for free UPK2 expression. Imaging examination was performed every three months from the time of the first repeated examination, which was 90 days after surgery. If the imaging examination indicated that the patient had relapsed, they were classified into the relapsed group (35 patients), while 70 patients did not relapse, and the level of UPK2 mRNA detected was recorded. If the patient had no recurrence during the 3-year follow-up period, then the mean value of multiple testing was recorded as the relative expression level of UPK2. The results demonstrated no significant differences in UPK2 expression between patients with LUAD of different ages and sex (Fig. 4A and B). Notably, non-relapsed patients exhibited low UPK2 expression. The mean expression level of UPK2 for relapsed patients was measured after recurrence was detected via imaging. The mean UPK2 expression relative to GADPH in relapsed patients was 0.2763, while the mean UPK2 expression in non-relapsed patients was 0.1623, which was significantly lower than that of relapsed patients (P<0.0001; Fig. 4C). Notably, for relapsed patients, UPK2 expression at relapse was significantly higher than that before relapse (P=0.0351; Fig. 4D). In addition, the ROC curve was plotted and the AUC was calculated to determine whether plasma UPK2 expression may be used to distinguish between relapsed and non-relapsed patients (Fig. 4E). The results demonstrated that when plasma UPK2 expression was used alone as a diagnostic biomarker, the AUC was 0.767 with a 95% CI of 0.675–0.858. Furthermore, patients with LUAD were divided into two groups, namely the high (≥0.1623) and low (<0.1623) UPK2 expression groups, and their survival curves were plotted. The results indicated that patients with high plasma UPK2 mRNA expression had a poorer survival outcome than those with low plasma UPK2 mRNA expression (Fig. 4F).