Cancer-Associated Fibroblasts Affect Tumor Metabolism and Immune Microenvironment in Gastric Cancer and Identification of Its Characteristic Genes

Background Cancer-associated fibroblasts (CAFs) have reported widely involved in cancer progression. However, its underlying mechanism in gastric cancer is still not clarified. Methods The data used in this study were all downloaded from the Cancer Genome Atlas database. R software and the R packages were used for all the analyses. Results In our study, we first quantified the CAFs infiltration using the ssGSEA algorithm. The clinical correlation result showed that CAFs were associated with a worse prognosis and clinical features. Pathway enrichment also indicated several oncogenic pathways in GC patients with high CAFs infiltration, including epithelial-mesenchymal transition (EMT), myogenesis, allograft rejection, the inflammatory response, and IL2/STAT5 signaling. Furthermore, FNDC1 and RSPO3 were identified as the characteristic genes of CAFs through two machine learning algorithms, LASSO logistic regression and SVM-RFE. The following analysis showed that FNDC1 and RSPO3 were associated with more progressive clinical features and had a good prediction efficiency of the CAFs infiltration status in GC patients. Pathway enrichment and genomic instability were performed to explore the underlying mechanisms of FNDC1 and RSPO3. Immune infiltration analysis showed that CAFs were positively correlated with M2 macrophages. Moreover, we found that the GC patients with low CAFs infiltration were more sensitive to immunotherapy. Also, the CAFs, FNDC1, and RSPO3 could generate a certain effect on the sensitivity of doxorubicin, mitomycin, and paclitaxel. Conclusions In summary, our study comprehensively investigated the role of CAFs in GC, which might be associated with immunotherapy sensitivity. Meanwhile, FNDC1 and RSPO3 were identified as the underlying targets of GC.


Introduction
Gastric cancer (GC) is the ffth most common cancer around the world, with over one million new cases diagnosed annually [1]. Tere has been a noticeable increase in the incidence of GC worldwide, along with its high mortality and metastasis rate [1]. At present, surgery is still the frst-line therapy option for early-staged GC and can lead to persistent prognosis benefts [2]. Meanwhile, combined therapies, including chemotherapy and targeted therapy, have also prolonged the overall survival (OS) of advanced GC patients [3]. Despite this, however, the fve years survival rate of advanced GC patients is still less than 20% [3]. Terefore, early diagnosis and precise therapy of GC patients remain the focus of research.
Tumor cells are continuously afected by the tumor microenvironment (TME) they exist in, the components of which mainly consist of immune and stromal cells [4]. Cancer-associated fbroblasts (CAFs) are one of the most prominent cell types in TME that can infuence tumor progression in multiple manners [5]. CAFs can secrete specifc biological factors such as EGF, TGF-β, and IL6 to facilitate tumor malignant phenotype, including tumor neovascularization and immune escape, leading to tumor deterioration [6]. Meanwhile, CAFs can regulate tumor metabolism. CAFs can enhance glycolysis and excrete plenty of lactic acid and hydrogen ions, forming an acidic microenvironment to inhibit the activity of immune cells. Also, the metabolites of lactic acid and pyruvate produced by CAFs can be used as nutrients for tumor cells to stimulate their growth [7]. Recently, increasing attention has been paid to the role of CAFs in cancers for its diverse biological functions. For instance, Liubomirski et al. found that in breast cancer, the interactions between cancer cells and CAFs can signifcantly enhance the prometastatic phenotypes of the TME, further resulting in the higher angiogenesis, migratory, and invasive potential of cancer cells [8].
In esophageal squamous cell carcinoma, Jolly et al. revealed that CAFs can secrete IL-6 and exosomal miR-21 to induce the generation of monocytic myeloid-derived suppressor cells, which not only suppressed immune function but also enhanced drug resistance [9]. However, few studies have focused on the role of CAFs in GC, and therefore, it is meaningful to explore the underlying efect of CAFs to guide the treatment of GC.
Advancements in bioinformatic analysis provide a great convenience for researchers in investigating the underlying biological mechanisms of diseases [10]. In our study, we quantifed the CAFs infltration using the ssGSEA algorithm and comprehensively explored its role in GC. CFNDC1 and RSPO3 were identifed as the characteristic genes of CAFs through two machine learning algorithms, LASSO logistic regression and SVM-RFE. Further following analysis showed that FNDC1 and RSPO3 were associated with more progressive clinical features and had a good prediction effciency of the CAFs infltration status in GC patients. Pathway enrichment and genomic instability were performed to explore the underlying mechanisms of FNDC1 and RSPO3. Immune infltration analysis showed that CAFs were positively correlated with M2 macrophages. Moreover, we found that the GC patients with low CAFs infltration were more sensitive to immunotherapy. Also, the CAFs, FNDC1, and RSPO3 could generate a certain efect on the sensitivity of doxorubicin, mitomycin, and paclitaxel.

Available Data Acquisition.
Te public transcription profles and clinical information of GC patients were downloaded from Te Cancer Genome Atlas database-TCGA-STAD project. Te expression profle was in TPM form and was annotated based on the Homo sapiens.GRCh38.107.gtf fle. Clinical information was in a "bcrxml" fle and extracted using the Perl code. Diferentially expressed genes (DEGs) analysis was performed using the limma package with the threshold of |logFC| > 1 and adj.P < 0.05. Te basic information of enrolled patients is shown in Table 1.

Pathway Enrichment
Analysis. Pathway enrichment analysis was performed using the gene set enrichment analysis (GSEA) algorithm, and the analyzed gene set was the Hallmark signature. Te terms with |normalized enrichment score (NES)| > 1 and adj.P < 0.05 were considered statistically signifcant.

Characteristic Gene Identifcation.
Two machine learning algorithms, LASSO logistic regression and support vector machine recursive feature elimination (SVM-RFE), were utilized to identify the characteristic genes of specifc features [12]. Receiver operating characteristic (ROC) curves were used to evaluate the prediction efciency of characteristic genes. Principal component analysis (PCA) was performed using the ade4 package in R environments.

Immune Infltration and Genomic Analyses.
Te quantifcation of the immune microenvironment of GC was conducted using the CIBERSORT algorithm, and 22 types of infltrating immune cells were extracted [13]. Te scores of TMB and MSI were downloaded from the TCGA database. Te tumor stemness index mRNAsi and EREG-mRNAsi were calculated according to the one-class logistic regression  [14].

Immunotherapy and Drug Sensitivity Analyses.
Tumor Immune Dysfunction and Exclusion (TIDE) analysis (https://tide.dfci.harvard.edu/) and submap algorithm were utilized to evaluate the immunotherapy response rate of GC patients. Drug sensitivity analysis was conducted based on the data from the Genomics of Drug Sensitivity in Cancer (GDSC) database (https://www.cancerrxgene.org).

Statistical
Analysis. R software was responsible for all the analysis. Here, the comparison with P value less than 0.05 was considered statistically signifcant. Te ggplot2 package was utilized for most plots [15]. Te correlation of continuous variables was compared using the Spearman method. Te comparison of variables with a normal distribution was performed using the Student's T-test.
Kaplan-Meier (KM) survival curves were used to evaluate the prognosis efect of specifc index.

Quantifcation of CAFs in TCGA Data
. Te fowchart of whole study is shown in Figure S1. First, based on the marker genes mentioned above, the relative infltration of CAFs in GC tissue was quantifed using the ssGSEA algorithm ( Figure 1(a)). KM survival curve showed that the patients with higher CAFs infltration might have a worse overall survival (OS) (Figure 1(b), HR � 1.41, P � 0.041). Furthermore, we explored the CAFs diferences in patients with diferent clinical features. Te result showed that CAFs might be associated with a more progressive grade and T stage (Figures 1(c) and 1(d)). However, no signifcant difference was observed in M and N stages (Figures 1(e) and 1(f )). Pathway enrichment analysis showed that in the patients with higher CAFs infltration, the pathway of epithelial-mesenchymal transition (EMT), myogenesis, allograft rejection, infammatory response, and IL2/STAT5 signaling were remarkably enriched in (Figure 1(g)).

Identifcation of the Characteristic Genes of CAFs.
Ten, we performed the DEGs analysis with the threshold of |logFC| > 1 and adj.P < 0.05. A total of 268 downregulated and 1697 upregulated DEGs were identifed (Figure 1(h)). LASSO logistic regression and the SVM-RFE algorithm were used to identify the characteristic genes of CAFs (Figures 2(a)-2(c)). LASSO logistic regression identifed four genes, including FNDC1, SGCD, FGF7, and RSPO3. Further, among these four genes, the SVM-RFE algorithm screened two genes FNDC1 and RSPOS, as the characteristic genes of CAFs (Figure 2(d)). ROC curves showed that FNDC1 and RSPO3 had great prediction in the CAFs infltration status of GC patients (Figures 2(e) and 2(f ), FNDC1, AUC � 0.890; RSPO3, AUC � 0.885). Ten, logistic regression was performed based on the FNDC1 and RSPO3. Te formula was "score � −6.691 + 0.9797 * FNDC1 + 1.2415 * RSPO3." Te ROC curve showed that the logistic score had an excellent prediction ability of the CAFs infltration of GC patients (Figure 2(g)). PCA analysis indicated that the genes FNDC1 and RSPO3 could efectively distinguish the GC patients with high and low CAFs infltration ( Figure 2(h)).

Prognosis Efect and Clinical
Correlation of FNDC1 and RSPO3. KM survival curves showed that the patients with high FNDC1 and RSPO3 expression might have a worse OS, DSS and PFI (Figures 3(a)-3(f )). Also, we found that the patients with higher CAFs infltration might have a higher FNDC1 and RSPO3 expression (Figure 3(g)). Meanwhile, the young patients (≤65 years old) tend to have a higher RSPO3 expression (Figure 3(h)); the G3 GC patients might have a higher FNDC1 and RSPO3 expression than G1-2 patients (Figure 3(i)); the stage III-IV patients might have an higher RSPO3 expression ( Figure 3(j)); the T3-4 GC patients might have a higher FNDC1 and RSPO3 expression than T1-2 patients (Figure 3(k)); the N1-3 GC patients might have a higher RSPO3 expression than N0 patients ( Figure 3(l)).

Biological Explorations of FNDC1 and RSPO3.
CAFs have been reported to afect tumor metabolism. Pathway correlation analysis indicated that CAFs was negatively correlated with TRN-α metabolism, KREBS cycle metabolism, amino acid metabolism, vitamin metabolism, abnormal metabolism, and vitamin metabolism, yet positively correlated with folate metabolism (Figure 4(a)). We next explored the underlying pathways of FNDC1 and RSPO3. Pathway enrichment analysis of RSPO3 showed that the pathway of the apical junction, infammatory response, KRAS signaling, and EMT were signifcantly enriched in the patients with high RSPO3 expression ( Figure 4(b)). For FNDC1, the pathway of NOTCH signaling, angiogenesis, hedgehog signaling, TGF-β signaling, and IL6/JAK/STAT3 signaling were signifcantly enriched in ( . Immune analysis showed that FNDC1 was positively correlated with NK cells, macrophages, and iDC, while negatively correlated with T17 cells ( Figure S2A); RSPO3 was positively correlated with NK cells, mast cells, and pDC yet negatively correlated with T17 cells and T2 cells ( Figure S2B).

CAFs and Its Characteristic Genes Were Associated with the Sensitivity of Immunotherapy and Chemotherapy.
Immunotherapy is a novel therapeutic option for advanced GC. Tus, we explored the underlying diference in immunotherapy sensibility between high and low CAFs infltration patients. Immune checkpoint correlation analysis showed that CTLA4, HAVCR2, PDCD1LG2, PDCD1, and TIGIT were diferentially expressed in high and low CAFs infltration patients (Figure 7(a)). Te TIDE analysis was then performed, in which the patients with TIDE a score >0 were defned as nonresponders and <0 were defned as responders. Te result showed in low CAFs infltration patients, the proportion of immunotherapy responders was 53.2%. However, in high CAFs infltration patients, the proportion of immunotherapy responders was only 20.9%, indicating that low CAFs infltration GC patients might be more sensitive to immunotherapy (Figure 7(b)). Submap analysis indicated that the patiens with low CAFs infltration might be more sensitive to both PD-1 and CTLA4 therapies ( Figure S3). Considering the signifcant correlation between CAFs and M2 macrophages, we further explored the efect of M2 macrophages on immunotherapy. Results showed a positive correlation between the TIDE score and M2 macrophages ( Figure S4A). Moreover, we found that the patients with high M2 macrophages infltration tend to have   a higher TIDE score, as well as a lower percentage of immunotherapy responders (Figures S4B and S4C). Moreover, the immunotherapy responder had a low CAFs level (Figure 7(c)), as well as a lower FNDC1 and RSPO3 expression (Figures 7(d) and 7(e)). Drug sensitivity analysis showed that CAFs were negatively correlated with the IC 50 of doxorubicin, while positively correlated with the IC 50 of mitomycin and paclitaxel (Figures 8(a)-8(c)); FNDC1 was negatively correlated with the IC 50 of doxorubicin, while positively correlated with the IC 50 of paclitaxel (Figures 8(d)-8(f )); RSPO3 was negatively correlated with the IC 50 of doxorubicin, while positively correlated with the IC 50 of mitomycin and paclitaxel (Figures 8(g)-8(i)).

Discussion
A common cancer, GC poses one of the most serious public health problems [1]. CAFs are an important part of the TME in GC that can signifcantly afect cancer progression. Terefore, a deep investigation of CAFs and their related molecule targets would contribute to understanding the intrinsic biological mechanism of GC. In medical research, the investigation and analysis of the classifcation or prediction of response variables in biomedical research are often challenging due to the data sparsity generated by limited sample sizes and a moderate or very large number of predictors. Bioinformatic analysis can efectively solve this contradiction and is a powerful tool for screening clinical predictors [16].
In our study, we frst quantifed the CAFs infltration using the ssGSEA algorithm. Te clinical correlation result showed that CAFs were associated with a worse prognosis and clinical features. Pathway enrichment also indicated several oncogenic pathways in GC patients with high CAFs infltration. Further, FNDC1 and RSPO3 were identifed as the characteristic genes of CAFs through two machine learning algorithms, LASSO logistic regression and SVM-RFE. Te following analysis showed that FNDC1 and RSPO3 were associated with more progressive clinical features and had a good prediction efciency of the CAFs infltration status in GC patients. Pathway enrichment and genomic instability were performed to explore the underlying mechanisms of FNDC1 and RSPO3. Immune infltration analysis showed that CAFs were positively correlated with M2 macrophages. Moreover, we found that the GC patients with low CAFs infltration were more sensitive to immunotherapy. Also, the CAFs, FNDC1, and RSPO3 could generate a certain efect on the sensitivity of doxorubicin, mitomycin, and paclitaxel.
Generally, in TME, the content of CAF is the most abundant, and it can afect the occurrence and development of cancer through intercellular contact, the release of various regulatory factors, and the remodeling of the extracellular matrix [17]. In colon cancer, Hu et al. indicated that CAFs could secret the exosome miR-92a-3p that was engulfed by colon cancer cells, further activating Wnt/β-catenin pathway and inhibiting mitochondrial apoptosis, leading to metastasis and chemotherapy resistance [18]. Su et al. revealed that CD10 + GPR77 + CAFs could induce cancer formation and chemoresistance through sustaining tumor stemness [19]. Wen et al. indicated that CAFs-derived IL32 could promote breast cancer cell invasion and metastasis through integrin β3-p38 MAPK signaling [20]. Pathway enrichment analysis showed that CAFs could activate the EMT, KRAS, and IL2/STAT5 signaling. In GC, Li et al. found that cancer-associated neutrophils could induce EMT through IL-17a to facilitate the invasion and migration of cancer cells [21]. Also, Wang et al. indicated that the downregulation of miRNA-214 in CAFs could enhance the migration and invasion of GC cells by targeting FGF9 and inducing EMT [22]. Our results were consistent with previous studies, which refect the validity of the analysis.
Trough machine learning algorithms, FNDC1 and RSPO3 were identifed as the characteristic genes of CAFs. FNDC1, whose full name is "fbronectin type III domain containing 1", has been reported to promote GC development. Jiang et al. demonstrated that FNDC1 could facilitate the invasion of GC by regulating the Wnt/β-catenin signaling and is correlated with peritoneal metastasis [23]. RSPO3 has been reported as being widely involved in cancer progression. For example, Chen et al. revealed that RSPO3 could enhance the aggressiveness of bladder cancer through Wnt/β-catenin and Hedgehog signaling pathways [24]. Fischer et al. found that in colon cancer with Wnt mutations, RSPO3 antagonism could hamper the malignant biological behavior of cancer cells [25]. However, virtually no study explored the RSPO3 in GC. Our study comprehensively investigated the underlying role of RSPO3 in GC, which can provide direction for future studies. In clinical practice, detecting the relative expression levels of FNDC1 and RSPO3 could indicate the CAFs infltration level of patients, as well as their response on GC immunotherapy.
Interestingly, immune infltration analysis showed that CAFs were associated with M2 macrophages. Te interaction between diferent cells can signifcantly afect the remodeling efects of TME [26]. Previous studies have shown the underlying crosstalk between CAFs and M2 macrophages. Based on a coculture system, Cho et al. found that cancer-stimulated CAFs could promote M2 macrophage activation through secreting IL6 and GM-CSF [27]. Meanwhile, from a review summarized by Gunaydin, the interaction between CAFs and tumor-associated macrophages in TME can enhance tumorigenesis and immune escape [28]. Notably, our results also showed that in patients with low CAFs infltration, the response rate to immunotherapy is higher (53.2% vs. 23.9%). Immunotherapy has shown a promising efect for specifc advanced GC patients.
Although our research is based on high-quality bioinformatics analysis, some limitations should be noticed. First, the potential race bias is hard to ignore. Most patients enrolled in our study were from Western populations, which might decrease the credibility of our conclusions. Second, detailed laboratory examinations are hard to obtain. If all the 12 Journal of Oncology data from all examinations can be obtained, our conclusion will be more abundant.

Data Availability
Te raw data mentioned in this study can be downloaded from online databases. Te data used to support the fndings of this study are available from the corresponding author upon request.

Conflicts of Interest
All the authors declare that there are no conficts of interest.  Figure S1: the fowchart of whole study. Figure S2: immune correlation analysis of FNDC1 and RSPO3. Notes: A: immune correlation analysis of FNDC1; B: immune correlation analysis of RSPO3. Figure S3: submap analysis was used to indicate patients' sensitivity to PD-1 and CTLA4 therapy. Figure S4: efect of M2 macrophages on GC immunotherapy. Notes: A: correlation of M2 macrophages and TIDE score; B: TIDE score in patients with high and low M2 macrophages infltration; C: the percentage of immunotherapy responders and nonresponders in patients with high and low M2 macrophages infltration. (Supplementary Materials)