Data Processing and Risk Score Calculation

737 RBPs examined in the GSE62254 cohort were subjected to Univariate Logistics analysis to select RBPs relevant to the occurrence of PM in GC patients. We selected the top 100 RBPs into Lasso Logistics analysis to acquire the coefficients. Then, four significantly correlated RBPs weighted by their coefficients were recognized to establish the prediction signature. After comparison and combination with TNM staging, a risk score formula for risk of PD in GC was constructed and demonstrated by a nomogram.

Pathway Enrichment Analysis

DAVID (version 6.7) (https://david-d.ncifcrf.gov/), an online bioinformatics analysis tool was used to perform the gene-Gene Ontology (GO) term and the Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis. Metascape (https://metascape.org/gp/index.html#/main/step1) offered complementary annotation.

Patients and Tissue Samples

We went through the pathology database of Renmin Hospital of Wuhan University for GC patients with PM, finding 36 cases in the period from 2014 to 2021. Then we randomly chose 72 GC patients without PM in the year of 2016 as the control group. As a result, a total of 108 formalin-fixed, paraffin-embedded GC tissue samples were obtained. All the patients underwent surgical treatment at Renmin Hospital of Wuhan University and there were none previous chemotherapies, radiotherapies, or other treatments before surgery on these patients. The study was approved by the Ethics Committee of Renmin Hospital of Wuhan University.

Immunohistochemical (IHC)

The paraffin tissues were cut into 4 um-thick sections, dried, dewaxed in xylene, and dehydrated in ascending series of ethanol. Antigen retrieval was conducted by microwave heating with citrate buffer (pH 6.0) for 20 min. Subsequently, paraffin sections were rinsed with PBS (3×5 min) and then blocked with 3% hydrogen peroxide at room temperature for endogenous peroxidase ablation for 25 min. Then the samples were exposed to Bovine Serum Albumin (BSA) at room temperature for 30 min to decrease nonspecific antibody binding after rinsing in PBS. The tissue sections were incubated overnight at 4°C with the primary antibody (anti-COL14A1, 1:200, ThermoFisher, America; anti-TNS1, 1:200, Abcam, British; anti-NUSAP1, 1:100, Abcam, British; anti-YWHAE, 1:500, Abcam, British);. After rinsing in PBS, the tissue sections were incubated with horseradish peroxidase-labeled anti-rabbit antibodies at room temperature for 30 min. Then, the tissue sections were rinsed with PBS for 4 times and then dripped with freshly prepared 3,3-diaminobenzidine (DAB). Microscopically, the staining was terminated when the tissue sections were brown-yellow or brown. Subsequently, all the tissue sections were restrained with hematoxylin for about 3 min. Finally, the slices were dehydrated with ethanol and toluene and then sealed with neutral gum. PBS was used to replace the primary antibody as a negative control. The slides were viewed via Olympus BX53 (Tokyo, Japan) microscope. IHC staining was evaluated independently by two pathologists under the double-blind condition. The staining intensity was classified as four grades as follows: 0 (no staining), 1 (light yellow), 2 (brown-yellow), and 3 (dark brown). The percentage of positive cells was classified as five grades as follows: 0 (0%), 1 (≤30%), 2 (31-50%), 3 (51-80%), and 4 (≥80%). Five most representative fields of high magnification (400x) were selected to calculate the final score. The final immunohistochemical score was the product of staining intensity and extent, theoretically from 0 to 12. Scores less than 4 were defined as low expression, and scores greater than or equal to 4 were described as high expression.

Cell Lines

AGS cell line and MGC-803 cell line were bought from American Type Culture Collection (ATCC, Manassas, USA). AGS cell line was maintained in Dulbecco’s Modified Eagle Medium (DMEM)/F-12 (Servicebio, Wuhan, China) while MGC-803 cell line was maintained in DMEM-H (Servicebio, Wuhan, China), supplemented with 10% fetal bovine serum (ThermoFisher, America) and 1% antibiotics (Servicebio, Wuhan, China). Cells were maintained in a 37° C incubator with 5% CO2. All cell lines tested negative for mycoplasma.

Cell Transfection

The siRNA vectors against COL14A1 or TNS1 were utilized for knockdown of COL14A1 or TNS1 with scrambled siRNA (siNC) as negative control. Sequences of the siRNAs were as follows: siCOL14A1-1: 5′‐GUGGUGGUAGAUGGAACUGUATT‐3′; siCOL14A1-2: 5′‐CUCAGGUUACCUGAUCCUUUATT‐3′; siTNS1-1: 5′‐CAGGUCUUACUCACCUUAUGATT‐3′; siTNS1-2: 5′‐GCAACUACCUGCUGUUCAATT‐3′. For upregulation of NUSAP1 or YWHAE, the full length of NUSAP1 or YWHAE was inserted into pcDNA3.1 vectors (Invitrogen) and the empty plasmids were served as negative control.

Cells were plated in 6-well plates with DMEM/F-12 medium supplemented with 10% medium FBS for 24 h before transfection. Transfections of siRNAs and indicated plasmids were both performed using Lipofectamine 2000 (ThermoFisher, America) according to the manufacturer’s instruction.

Quantitative Real-Time PCR (qRT-PCR)

Total RNA from cells was extracted by TRIzol reagent (ThermoFisher, America) following the supplier’s instructions. Reverse transcription was conducted with the First Strand cDNA Synthesis Kit (Servicebio, Wuhan, China). PCR was implemented with SYBR Green qPCR Master Mix (Servicebio, Wuhan, China). The conditions for qRT-PCR were as follows: 95°C for 3 min, followed by 40 cycles of 10 s at 95°C, 10 s at 60°C, and 15 s at 70°C, followed by heating from 65°C to 95°C.

The sequences of main primers were as follows: COL14A1 (forward): 5′‐AGTGGGTGAGAAGGCAATGA‐3′, COL14A1 (reverse): 5′‐CTCTCAGGCCTGGAAGTTCA‐3′; TNS1 (forward):5’-TCAAGTGGAAGAACTTGTTTGCTT-3’, TNS1 (reverse): 5′‐CACGACAATATAGTGGAGGCACA‐3′; NUSAP1 (forward): 5′-AGCCCATCAATAAGGGAGGG-3′, NUSAP1 (reverse): 5′‐ACCTGACACCCGTTTTAGCTG‐3′; YWHAE (forward): 5′‐GCTGGATCCATGGATGATCGAGAGGATCTG‐3′, YWHAE (reverse): 5′‐GCTGAATTCTCACTGATTTTCGTCTTCCAC‐3′; GAPDH (forward): 5′‐CACCATTGGCAATGAGCGGTTC‐3′, GAPDH (reverse): 5′‐AGGTCTTTGCGGATGTCCACGT‐3′; GAPDH was utilized as an endogenous control.

Cell Proliferation Assays

For Cell Counting Kit‐8 (CCK‐8) assay, transfected cells were inoculated at a density of 2 × 100 cells/well into 96‐well plates and cultivated for 0, 24, 48 and 72 hours. After different incubation times, each well was added with 10 μL of CCK‐8 reagent (Servicebio, Wuhan, China) and cultured for another hour. Then, the absorbance at 450 nm was recorded with a standard microplate reader (EnSight, Perkin Elmer, America).

Establishment and Validation of a 4-RBP-Based Classifier to Predict the Risk for Peritoneal Metastasis in Gastric Cancer

The heatmaps revealed that COL14A1 and TNS1 were highly expressed in GC patients with PM in the training set, while NUSAP1 and YWHAE were highly expressed in the cases without PM ( Figure 2A ). Consistent results were observed in the validating set ( Figure 2B ). To assess the ability of the 4-RBP-based classifier to forecast the risk of PM in GC patients, we developed a risk score according to the coefficients of the four RBPs in Lasso Logistics: Risk Score = (1.08004578 * expression value of COL14A1) + (0.31114396 * expression value of TNS1) - (0.95036402 * expression value of NUSAP1) - (0.02147674 * expression value of YWHAE). The risk score formula was used to calculate the risk score in the training set, and the cases were divided into high-risk and low-risk groups owing to the cutoff of the median risk score. Kaplan–Meier curves showed that patients in the high-risk group had shorter overall survival than those in the low-risk group (p = 0.002) ( Figure 2C ); a similar result was confirmed in the validating set (p = 0.029) ( Figure 2D ). In addition, the Kaplan–Meier curves of DFS of the two sets (p = 0.004 and 0.048, respectively) were in agreement with previous results ( Figures 2E, F ).

Open in a separate window

Figure 2

Validation of the 4-RBP-based signature in the training set and the validating set. (A, B) The heatmap of the four RBPs expression profiles. (C, D) Kaplan–Meier analysis for overall survival (OS) of GC patients based on the risk stratification. (E, F) Kaplan–Meier analysis for disease-free survival (DFS) of GC patients based on the risk stratification. (G, H) Receiver operating characteristic (ROC) analysis for the risk of PM including the risk score, TNM stage and combination of the two. (I) Nomogram to predict the risk of PM in GC.

Prognostic Value of the RBP-Based Classifier for Prediction of the Risk for Peritoneal Metastasis in Gastric Cancer

The 4-RBP-based signature, gender, and pathological stage were significantly related to peritoneal metastasis in the univariate logistics analysis. After the multivariate logistics regression analysis of the abovementioned factors, the 4-RBP-based signature and pathological stage were retained to be dependable factors for peritoneal metastasis in the training set. Except for gender, similar results were observed in the validating set ( Table 2 ). Our result showed that the 4-RBP-based signature was an independent prognostic factor for peritoneal metastasis in gastric cancer in two sets. ROC curves were then plotted to appraise the competence of the 4-RBP-based signature to successfully predict the risk of PM in GC. The AUCs, in the training set and validating set (0.811 and 0.786, respectively), showed that the RBP-based classifier had similar predictive accuracy compared with the TNM staging (AUCs were 0.782 and 0.821 respectively) ( Figures 2G, H ). However, when we combined the RBP-based risk score and TNM staging to predict the risk of PM, the predictive capability was robustly enhanced. The AUC values of this built-up prediction model were 0.896 and 0.884 respectively in the training and validating set, suggesting better predictive accuracy. Whereafter, the 4-RBP-based risk score and TNM staging were used to structure a nomogram for predicting the risk of PM in GC patients ( Figure 2I ).

Pathway Enrichment Analysis of Top 100 Correlated RBPs

To explore the possible effect of the related genes on GC, DAVID and Metascape were used to perform function enrichment analysis. The results of DAVID revealed that the top 100 related genes were primarily enriched in mRNA splicing and RNA processing in biological processes (BP) ( Figure 3A ). In assessment of cell components (CC), the genes were mainly enriched in nucleolus and nucleoplasm ( Figure 3B ). While for molecular function (MF) and KEGG, the genes were generally enriched in poly(A) RNA binding ( Figure 3C ). The results of Metascape showed that the correlated genes mainly enriched in ribonucleoprotein complex biogenesis, mRNA metabolic process, translation, Nop56p-associated pre-rRNA complex, ribonucleoprotein complex assembly and so on, suggesting that these pathways were correlative with the PM of GC with ( Figures 3D, E ).

Open in a separate window

Figure 3

Pathway enrichment analysis of the related RBPs. (A-C) GO term and KEGG enrichment analysis performed by DAVID in BP, CC, MF and KEGG. (D, E) Pathways associated with the related RBPs were enriched by Metascape.

Expression and Predictive Importance of the Four RBPs in Clinical Samples

We obtained 36 peritoneal metastatic samples of GC and 72 samples without PM. There was no significant difference between the two groups in gender, age, lymph node metastasis, HER-2, Ki67 (%), Lauren’s classification or lymphatic invasion assessed by IHC staining in 108 GC patients, 36 of which with PM ( Table 3 ). However, we found that COL14A1 and TNS1 were over-expressed in peritoneal metastatic lesions compared with primary tumor tissues ( Figures 4A–D ); whereas conversely, NUSAP1 and YWHAE were over-expressed in primary tumor tissues compared with peritoneal metastatic lesions ( Figures 4E–H ), which is in accordance with our findings in the bioinformatic analysis above. Kaplan–Meier curves were separately drawn due to differential expression of the four RBPs in the GC patients ( Figures 4I–L ). It revealed that patients with a high expression of COL14A1 had shorter overall survival than those with low expression (p = 0.047), while there was no significance in the graph divided by the expression of TNS1, NUSAP1 and YWHAE (p = 0.855, 0.255 and 0.053, respectively). Nevertheless, according to the Kaplan–Meier curves, patients with high expression of NUSAP1 and YWHAE tended to have longer overall survival compared to those with low expression, which was consistent with our previous findings. Later, the patients were divided into high- score and low- score groups according to their 4-RBP-based signature. Kaplan–Meier curves further revealed that patients in the high-score group had shorter overall survival than those in the low-score group (p = 0.02) ( Figure 4M ).

Open in a separate window

Figure 4

Expression and validation of the 4 RBPs in clinical samples. (A–D) IHC staining of COL14A1, TNS1, NUSAP1 and YWHAE in primary GC lesions. (E–H) IHC staining of COL14A1, TNS1, NUSAP1 and YWHAE in peritoneal metastasis lesions. Scale bar 1000μm. (I–L) Kaplan–Meier analysis for overall survival (OS) of GC patients based on different expression of COL14A1, TNS1, NUSAP1 and YWHAE. (M) Kaplan–Meier analysis for OS based on diverse scores.