Comprehensive analysis of circular RNA-associated competing endogenous RNA networks and immune infiltration in gastric cancer.

BACKGROUND
Cyclic RNA (circRNA) has been proved to be an important regulator of gastric cancer (GC). However, the role and regulatory mechanism of circrna related competitive endogenous RNA (ceRNA) in GC have not been established.


METHODS
CircRNA data and clinical data were obtained from the GEO and TCGA databases. The ceRNA networks were constructed and a function enrichment analysis was completed. Additionally, correlations between hub genes expression, immune cell infiltration, and clinical phenotypes were determined. The differentially expressed circRNAs and their downstream microRNAs (miRNAs) were validated by quantitative real-time polymerase chain reaction, and the hub genes were validated by western blot analysis. The migration and invasion ability of overexpressed hsa_circ_0002504 was determined by a transwell assay.


RESULTS
The ceRNA network contained 2 circRNAs, 3 miRNAs, and 55 messenger RNAs (mRNAs). 323 biological processes terms, 53 cellular components terms, 51 molecular functions terms, and 4 signaling pathways were revealed by the function enrichment analysis. The GSEA analysis revealed that the hub genes were positively correlated with the axon guidance and adhesion molecules pathways. The correlation analysis revealed that overexpressed EPHA4 and KCNA1 indicated poor tissue differentiation and were associated with clinically advanced stages of GC. The in vitro experiments showed that hsa_circ_0002504 was significantly down-regulated in GC cell lines. In addition, the overexpression of hsa_circ_0002504 led to a significant downregulation of hsa-miR-615-5p and hsa-miR-767-5p, as well as an upregulation of EPHA4, KCNA1, and NCAM1. Furthermore, it suppressed the migration and invasion ability of GC cells.


CONCLUSIONS
Hsa_circ_0002504 is a potential diagnostic biomarker for GC. High expression of EPHA4 and KCNA1 may indicate poor prognosis.


ms of circRNA
in GC and to identify better biomarkers for diagnosis or treatment, we predicted target genes by screening differentially expressed circRNAs (DEcircRNAs) and constructing a circRNA-miRNA-mRNA interaction network across multiple databases.Clinical values of these hub-genes were determined through correlation analysis and immune in ltration analysis.

It has been reported that circRNAs can be used as ceRNAs to retain endogenous RNA and are also regulators of related genes expression to intervene in the occurrence and development of many kinds of carcinomas (Li et al. 2019b;Liu et al. 2017;Su et al. 2019).Han et al. evaluated the role of circRNAs in GC and its regulatory mechanisms by constructing a ceRNA interaction network, and screened out more than 10 genes, such as SERPINE1, which were identi ed as possible diagnostic and prognostic indicators for GC (Han et al. 2021).Deng et al. found that CircRHOBTB3 is a ceRNA of miR-654-3p, which inhibits GC progression by activating the p21 signaling pathway (Deng et al. 2020).Li reported that hsa_circ_0017639 promotes GC proliferation and metastasis by sponging miR-22 -5p and up-regulating USP3 (Li et al. 2020).Moreover, circPSMC3, had been shown to regulate PTEN expression by sponging miR-296-5p in GC (Rong et al. 2020).

However, since studies on circRNAs is still in their infancy, various limitations exist.For instance, many results are based on bioinformatics analysis, whose research scope is not comprehensive enough and they lack further investigations to con rm their conclusions and exact mechanisms of circRNA in tumorigenes s (Han et al. 2021).Then, small numbers of tissue samples are mostly sourced from the same hospital in most studies, as they should preferably be sourced from different hospitals or even different regions.Moreover, many studies on GC, such as analysis of the correlation between targeted genes and clinical features and immune in ltration are not in-depth.

In this study, we constructed a ceRNA regulatory network composed of 2 circRNA, 3 miRNAs and 55 mRNAs through bioinformatics analysis, studied the correlativity between expression levels of the top 10 hub genes and in ltration levels of immune cells, as well as histopathological and clinical stages.Our ndings illuminate the potential regulatory mechanisms of circRNAs in GC development at molecular level, and indicate that circRNAs have the potential as diagnostic and therapeutic biomarkers for GC.


Materials And Methods


Data pretreatment

A total of two datasets of cir RNA microarray data, GSE83521 and GSE100170, were downloaded from the GEO database.Clinical (phenotype) features, RNAseq data and miRNA data of GC were obtained from the TCGA database using the UCSC Xena browser.To connect different data, we matched, selected and deleted different data for various samples.Samples were grouped according to cancer and the adjacent non-cancerous group, mRNAs with more than 80% zero expression values were deleted, as well as miRNAs with all zero expression values.In total, mRNA sequencing data for 407 samples (375 cancerous vs 32 adjacent non-cancerous) and miRNA data for 477 samples (436 cancerous vs 41 adjacent non-cancerous) were obt ined.


Differentiation analysis

By comparing the cancer and the adjacent non-cancerous groups, we determined differentially expressed genes (DEGs), including DEcircRNAs, DEmiRNAs and DEmRNAs with thresholds p < 0.05 and |log2FC| > 1 using R (Version 3.6.3)package limma (Version 3.42.2) (Ritchie et al. 2015).Then, expression values of DEGs were clustered based on Euclidean distance by pheatmap package (Version 1.0.12)(Tian et al. 2021), and were displayed using heatmaps, volcano plots and PCA.


Co

I networks

The DE
ircRNAs were obtained from the intersection of GSE83521 and GSE100170 datasets.Potential miRNA targets for DEcircRNAs were predicted using the CircInteractome online tool (https://circinteractome.irp.nia.nih.gov/mirna_target_sites.html) (Dudekula et al. 2016) and compared with the screened-out DEmiRNAs downloaded from the TCGA.Then, downstream DEmiRNAs regulated by DEcircRNAs were ltered out.Subsequently, target mRNAs that are regulated by DEmiRNAs were predicted using the mirwalk online tool (http://mirwalk.umm.uniheidelberg.de/)(Sticht et al. 2018) and compared with the DEmRNAs downloaded from the TCGA.Then, downstream DEmRNAs that are regulated by DEmiRNAs were ltered out.

Ultimately, intersection resul

EmRNAs wer
presented using a Venn diagram by R package VennDiagram (1.6.20)(Lam et al. 2016).

The PPI analysis on DEmRNAs in the ceRNA network was constructed by the STRING database (version: 11.0, http://www.string-db.org/)(Szklarczyk et al. 2019) with a con dence (combined score) of > 0.15 as the threshold.

The relevant tsv les were downloaded and the PPI network was visualized using the Cytoscape software (Doncheva et al. 2019).

Regulatory networks for un ltered circRN

miRNA and ltered miRNA-mRNA were constr
cted and visualized using cytoscape.Ultimately, the ceRNA network, constructed with DecircRNAs-DEmiRNAs-DEmRNAs, was then visualized.


GO and KEGG Analyses

We performed the GO and KEGG pathway enrichment analyses of the DEmRNAs using R package ClusterPro ler (Yu et al. 2012) to identify the signaling pathways and biological functions that might be affected by the mRNAs involved in the ceRNA network.The signi cance threshold was set at p ≤ 0.05, and results were visualized using a bubble diagram.


GSEA enrichment analysis

We performed GSEA enrichment analysis on the top 10 hub genes in the PPI network using the R-package clusterPro ler.The GSEA statistical processes included calculatin the enrichment fraction, estimating the signi cance of enrichment fraction and correcting multiple hypothesis testing.


Correlation analysis of immune in ltration

We used the TIMER (https://cistrome.shinyapps.io/timer/)(Li et al. 2017), a online tool, to analyze and visualize correlations between expression levels of the top 10 hub genes and immune in ltrate levels in C.By inputting genes and cancer types, the scatter diagram showing the purity-corrected Spelman Rau value and statistical sig i cance was generated.


Correlations between hub genes and clinical features

The GC patients were grouped based on clinical features, including age, cancer stage and histological grade among others.Differential expr

sions of the top 10 g
nes between each pair of groups were presented using a box diagram.


Valiation of the regulatory role of the ceRNA network in vitro assays

Human gastric epithelial cell line GES-1 and GC cell lines HGC-27 were obtained from Beyotime (ShangHai, China), GC cell lines AGS were derived from American Tissue Culture Collection (ATCC, Rockville, MD,

al RNA extraction kit (Takara, Japan) and r
verse transcribed into cDNA using the iScript cDNA synthesis kit (Bio-Rad, Hercules, CA, USA).CircRNA expression level was measured by qRT-PCR using SYBR Green Premix Ex TaqTM (Takara, Japan) on ABI 7500 real-time uorescence quantitative PCR instrument (Applied Biosystems Inc, USA).Primers for circRNAs and GAPDH were as follows: hsa_circ_0001998: F-CCTGGCGTT

ATTATGCTC and R-AAGCAGCAACCGGAGAGATT; hsa_circ_000250
: F-TTTCGTTTTGATGGCTGGCT and R-TCATCTTTGATGCTGTTGTGCT; GAPDH: F-GCACCGTCAAGGCTGAGAAC and R-TGGTGAAGACGCCAGTGGA.


Statistical analysis

All bioinformatic analysis was conducted using R Version 3.6.3software.All data related to qRT-PCR were analyzed with Graphpad 7.0 and indicated as mean ± SD.The statistical differences between groups were tested by one-way ANOVA and Tukey's test.P-value < 0.05 was considered as statistically signi cant.


Results


Identi cation of DEGs

A total of 150 DEcircRNAs in the GSE83521 data set (80 up-regulated vs 70 down-regulated; Fig. 1A-C), and 1477 DEcircRNAs in the GSE100170 data set (497 up-regulated vs 980 down-regulated; ig. 1D-F) were obtained from the GEO database (supplementary les S1).

Baseline characteristics of GC patients from the TCGA database are presented in Table 1.A total of 29579 exp

ing sites (Fig. 2D), which were then intersected
with 108 DEmiRNAs from the TCGA database.Three DEmiRNAs were screened out (Fig. 2B).Based on the mirwalk database, the 3 DEmiRNAs predicted 254 target mRNAs, which were then intersected with 6516 DEmRNAs from the TCGA database.Only 55 DEmRNAs were screened (Fig. 2C), and were then used to construct the DEmiRNA-DEmRNA network (Fig. 2E).Finally, the ceRNA interaction network consisting of the 2 DEcircRNAs, 3 DEmiRNAs and 55 DEmRNAs was constructed (Fig. 2F).Additionally, we constructed the PPI network using the 55 DEmRNAs through the STRING database (Fig. 2G-I) (supplementary les S3).


GO\KEGG analysis

In order to investigate the f nctions of DEGs in the ceRNA network, we performed GO and KEGG analysis.A total of 323 BP terms, 53 CC terms, 51 MF terms were enriched in GO analysis.Signi cantly enriched BP terms

re correlated wit
central nervous system projection neuron axonogenesis, central nervous system neuron axonogenesis, regulation of cation transmembrane transport, regulation of ion transmembrane transport, and central nervous system neuron development.Signi cantly enriched CC terms were voltage-gated potassium channel complex, potassium channel complex, ion channel complex, synaptic membrane, and transmembrane transporter complex.In the aspect of molecular function (MF), the most enriched terms were voltage-gated ion channel activity, voltage-gated channel activity, ion gated channel activity, potassium channel activity, and gated channel activity (Fig. 3A-C; Table 2; supplementary les S4).KEGG analysis revealed that those DEGs were notably enriched in glycosaminoglycan biosynthesis-chondroitin sulfate/dermatan sulfate, axon guidance, GABAergic synapse and salmonella infection (Fig. 3D; Table 3; supplementary les

4).


GSEA ana
ysis

We performed GSEA analysis for top 10 hub genes to identify their correlation with with signaling pathways i GC.

According to the expression correlation between EPHA4, NCAM1 and NRXN1 and other genes, the most differentially enriched pathway in EPHA4 was axon guidance, while in both NCAM1 and NRXN1, adhesion molecules pathways were enr

hed (Fig. 3E-G; Table 4; sup
lementary les S5).


Immune in ltration analysis

To evaluate the clinical value of these hubgenes in GC, we analyzed the correlation between the top10 hub genes expression and immune in ltrates, including B cells, CD8+ T cells, CD4+ T cells, macrophages, neutrophils and dendritic cells.As shown in Figure 9, the far left represents the xpression levels of genes showing tumor purity.

Half of the 10 hubgenes were inversely correlated with tumor purity, while the rest were not.DCLK1, DLG2, EPHA4, NCAM1, NRXN1 and SCN2B observably have positive correlati n with all the above mentioned 6 immune cells; KCNA1 was positively correlated with CD8+ T cells, CD4+ T cells, macrophages, neutrophils and dendritic cells; KCNIP2 was positively associated with B cells, CD4+ T cells, macrophages and dendritic cells; UNC13A was positively associated with CD4+ T cells, macrophages and dendritic cells; and EPHB2 were negatively associated with CD8+ T cells, neutrophils and dendritic cells (Fig. 4A-J).Among them, EPHA4 and KCNA1 were positively correlated with both macrophage (R=0.417,p=5.53E-17 and R=0.266, p=1.93E-07, respectively) and dendritic cell in ltrations (R=0.559,p=8.24E-32 and R=0.376, p=6.4E-14, respectively).These ndings imply that hubgenes involved in our PPI network were st

associated with poor prognosis.
Validation of circRNA expression

Since has_circ_0002504 was down-regulated in both GSE83521 and GSE100170 databases showing higher consistency(Fig.6A-B), it was selected for further validation by qRT-PCR in 4 gastric cell lines (GES-1, AGS, MKN-45 and HGC-27).The results suggested that the expression of has_circ_0002504 was signi cantly down-regulated in GC cells (Fig. 6C) which was consistent

ith the res
lts of our bioinformatics analysis.


Discussion

Globally, GC is one of the most common malignancies.Its symptoms usually appear in the late stage, which inhibits early diagnosis, and most patients are diagnosed at the advanced stage of the disease (Allemani et al. 2015;Jiang et al. 2018).In various global regions, the case fatality rate for GC is still as high as 75%, which is a leading contributor to the global burden of disability-adjusted life-year (Thrift & El-Serag 2020).Due to late diagnosis, most GC patients often miss the optimal treatment window.Early effective detection has particularly importantance on reducing the mortality associated with GC.Researches have reported that circRNAs play a considerable role in GC occurrence and progression and are potentia biomarkers and therapeutic targets for GC.

In this study, we constructed the ceRNA and PPI interaction network using differentially expressed genes (DEGs) and screened out the hub genes.Then, we performed GO and KEGG analysis, as well as GSEA analysis to elucidate on the enrichment pathways and functions of DEGs and hub genes.We also evaluated the relevance of the top 10 hub genes expression levels and immune in ltrates, and explored the correlations between hub genes and histopathological grade, clinical stage a d other clinical features of GC patients.

In this study, we constructed the ceRNA interaction network using 2 key circRNAs, 3 key miRNAs, and 55 key mRNAs.In addition, we used the STRING database to establish the PPI interaction network.We found that one of the 2 key circRNAs, has_circ_0001998, function as miR-490-5p sponge to act on the target genes (EPHA4, KCNA1, etc.; Fig. 2D).Yuan reported that expression levels of has_circ_0001998 are up-regulated in colorectal cancer patients, where they play an anti-tumor role by competitively inhib ting the expression of has-miR-490-5p.

Moreover, there was no obvious correlation between the expression of has_circ_0001998 and the host gene, while its mRNA expressions were markedly relevant to survival time of colorectal cancer patients.These ndings suggest that has_circ_0001998 is highly stable and has a prognostic value (Yuan et al. 2019).In addition, it has been reported that has-miR-490-5p expressed at a low level in various tumor tissues while its overexpression participates in regulating tumor proliferation and invasion as a tumor suppressor in GC (Zhang et al. 2019a), bladder cancer (Wu et al. 2019), esophageal squamous cell carcinoma (Li & Zang 2019), liver cancer (Fang et al. 2018;Yu et al. 2019), pharyngolaryngeal cancer (Abdeyrim et al. 2019), and renal cell carcinoma (Chen et al. 2016) among others.In GC, has-miR-490-5p may be down-regulated by lncRNA LINC02532 (Zhang et al. 2019a) and circNRIP1 (Zhang et al. 2019b) sponge, promoting GC proliferation, migration and invasion.This result is consistent with the ndings of our analysis of the GSE83521 database.Has_ci c_0002504 had not been previously reported, but our ndings revealed that the expression of this circRNA in both data sets was consistent (supplementary les S9) and was valeted in vitro, suggesting that it may be a relatively new and more speci c biomarker.

To further evaluate the 55 target mRNAs that are regulated by the 2 DEcircRNAs in the ceRNA network, GO and KEGG were performed for enrichment analysis.GO analysis revealed that among the BP categories, top ve enrichment categories were central nervous system projection neuron axonogenesis, central nervous system neuron axonogenesis, regulation of cation transmembrane transport, regulation of ion transmembrane transport, and central nervous system neuron development.Among the CC categories, the top ve categories were voltagegated potassium channel complex, potassium channel complex, ion ch nnel complex, synaptic membrane, transmembrane transporter complex.The top ve in the MF category were voltage-gated ion channel activity, voltage-gated channel activity, ion gated channel activity, potassium channel activity, and gated channel activity.

Various pathways were enriched in ion channels in GO analysis.Previous studies reported that GC occurrence is correlated with abnormal functions of cross-cell membrane ion channels.Golgi anti-apoptotic protein (GAAP), a multi-transmembrane protein, can form cation-selective ion channels and mediate cell movement as well as adhesion, while the typical signs of malignant differentiation are the changes in migration and adhesion abilities of cancer cells to other cells or extracellular matrix (Carrara et al. 2017).Currently, overexpression of human GAAP (hGAAP) has been reported in some human malignant tissues.The expression of hGAAP between lung cancer tissues and para cancer tissues was signi cantly different, suggesting that it may be involved in lung carcinogenesis, and is therefore, a new prognostic biomarker for lung cancer patients who never smoked (Wu et al. 2013).It has also been reported that overexpressed hGAAP mRNA in glioblastoma multiforme tumors is related to poor outcomes (Chen et al. 2014).In osteosarcoma and cervical cancer cells, enhanced cell adhesion and migration could regulate local activation of calcium-dependent calpain (Carrara et al. 2015;Saraiva et al. 2013).KCNE2 is a potassium channel protein which is mostly expressed in the cytoplasm of gastric parietal cells.

In animal models, suppressed KCNE2 levels were associated signi cantly low proton secretion, abnormal parietal cell morphologies, acidosis and hypergastrinemia.Functionally, KCNE2 also inhibits cell growth by downregulating CyclinD1, and has an anti-proliferative effect on GC cells (Kundu et al. 2008).Elucidation of these pathways may help clarify GC pathogenesis and hopefully provide prognostic indicators and therapeutic targets for GC.In KEGG analysis, the major enriched pathways were Glycosaminoglycan biosynthesis-chondroitin sulfate/dermatan sulfate, axon guidance, GABAergic synapse, and salmonella infection.Nerve invasion, which is correlated with postoperative tumor recurrence, cancer pain and poor prognosis, is one of the important ways through which GC locally spreads (Bilici et al. 2010).Netrin-1 (NTN1), an axon guiding molecule (Lai Wing Sun et al. 2011), has been shown to be highly expressed in GC, and its expression is associated with nerve invasion and lymphatic metastasis.It had been reported that NTN1 plays a key vital in regulating GC invasion in vivo and in vitro (Yin et al. 2019).Another axon guiding molecule, semaphoring 5A, is involved in GC metastasis (Pan et al. 2009), the speci c mechanisms may be through MEK/ ERKS pathway activations and the subsequent MMP9trans-activation (Pan et al. 2013).Salmonella infection, through its role in 'i ammatory stimulation' is associated with gallbladder cancer (Iyer et al. 2016).Since in ammatory stimuli are associated with tumor development (Sheweita & Alsamghan 2020), we found that salmonella infection is involved in GC occurrence.

GSEA analysis of hub genes revealed that EPHA4 was positively correlated with axon guidan pathway, while both NCAM1 and NRXN1 were positively correlated with adhesion molecules cams pathway.Semaphorin proteins are a class of axon guiding molecules that play key roles in the development of central nervous system.Their expression is closely correlated with invasion and metastasis for certain human carcinomas (Muratori & Tamagnone 2012).In oral squamous cell carcinoma, semaphorin 4D can cooperate with Vascular endothelial growth factor (VEGF) to induce cancer growth and angiogenesis (Zhou et al. 2012).Semaphorin 6B has been associated with GC differentiation and metastasis in vivo, as well as tumor cells migration, adhesion and invasion in vitro.These ndings suggest that semaphorin 6B may play a part in GC occurrence and progression and is a potential biomarker for GC diagnosis, assessment and targeting gene therapy (Ge et al. 2013).Studies have also shown that ectopic expression of semaphorin 3E inhibits proliferation and colony formation of GC in vitro and of xenografts in vivo.However, its high expression lever inhibited migration and invasion in vitro, implying that downregulated semaphorin 3E is involved in GC pathogenesis (Chen et al. 2015).Cell adhesion molecules are important in carcinogenesis.L1 cell adhesion molecule (L1CAM) was over expressed in GC tissue and was signi cantly relevant to poor outcome.Overexpressed L1CAM is an independent prognostic factor for overall and event-free survival, as well as an independent risk factor for distant metastasis of GC, while low levels of L1CAM inhibit proliferation, cell cycle progression, invasion, migration and antinestin apoptosis of GC cells, and suppresses peritoneal metastasis (Ichikawa et al. 2019).Overexpression of the epithelial cell adhesion molecule (EpC M) is corelated with larger size of tumor, lymphatic metastasis and worse outcomes in GC and might be helpful in prognostic assessment (Dai et al. 2017).Taken together, both pathways are highly associated with GC progression.

Immune cells in the tumor microenvironment are involved in maintaining the homeostasis and tolerance of immune (Li et al. 2019a;Liu et al. 2018;Zeng et al. 2019).Therefore, we evaluated the correlation between the top10 hub genes and immune in ltrates.Most of the 10 hub genes were positively associated with immune in ltrates levels (Fig. 4).Among them, EPHA4 and KCNA1 were positively correlated with macrophage and dendritic cell in ltration.Macrophages in ltrating into tumor tissues, also known as tumor-associated macrophages (TAMs), promote tumor growth by secreting various factors, interacting with other matrix components, promoting angiogenesis and rebuilding the tumor microenvironment (Lee et al. 2014;Serizawa et al. 2016;Song et al. 2019;Veremeyko et al. 2018;Zhao et al. 2016).The degree of macrophage in ltration was positively associated with poor outcome of various carcinomas (Oya et al. 2020).Additionally, M2 polarized TAMs are potential independent predictors for GC prognosis (Zhang et al. 2014), therefore, targeted phenotypic or function speci c macrophage subsets may be promising targets for postoperative adjuvant therapy in GC patients (Lin et al. 2019;Zhang et al. 2014).Dendritic cells (DCs) in ltrate various tumor tissues, and the presence of in ltrating DCs in ovarian cancer (Labidi-Galy et al. 2012), breast cancer (Faget et al. 2013) and GC (Liu et al. 2020) indicates a poor prognosis.Li et al. reported that malignant tumor cells and tumor-related pDC secrete indoleamine 2-dioxygenase, a powerful promoter activated by Treg, which ay lead to non-responses and thus allow malignant cells to escape immune surveillance (Li et al. 2019a).These results suggest that EPHA4 and KCNA1 might play an considerable role in GC immune in ltration and indicate a poor outcome.

To identify the clinical values of these hub genes, we performed a clinical correlation st dy.Overexpressions of NCAM1, SCN2B and NRXN1 were correlated with advanced ages of GC patients.Overexpressions of NCAM1, SCN2B and NRXN1 were correlated with higher histological grades, implying worse histological di ferentiation.Suppressed expression levels of EPHB2 were associated with worse histological differentiation.Furthermore, elevated expressions of EPHA4, UNC13A, NRXN1, SCN2B and KCNA1 were correlated with late GC clinical stages.

Elevated expression levels of EPHA4 and KCNA1 indicated poor tissue differentiation and were correlated with clinically advanced GC stages.These two genes are potential biomarkers for poor prognosis of GC.MarikoOki et al. found that EphA4 mRNA was highly expressed in GC cells and tissues, and at the protein level, overexpression of EphA4 was markablely correlated with tumor invasion and relapse.Moreover, overall survival time of EphA4 positive cancer patients was observably shorter than that of EphA4 negative cancer patients (Oki et al. 2008).It has also been reported that overexpressed EphA4 is signi cantly associated with GC recurrence; its high expression level is an independent exist as a predictor for recurrence-free survival and is associated with low survival outcomes (Inokuchi et al. 2018).These conclusions support our postulate, implying that EphA4 plays an important part in GC progression and is a predictor for poor prognosis.Moreover, the mechanisms of KCNA1 in tumors have not been clearly de ned.Some studies report that KCNA1 is an overexpressed carcinogene which promotes tumor progression by regulating Wnt, Hhg and Notch signaling pathways as well as mitochondrial ability in cervical cancer (Liu et al. 2018).Lallet-Daher et al. reported that KCNA1, as a voltage-gated delayed potassium channel, can block the transformation induced by oncogenes in NIHT3T3 cells without inducing cell senescence.In addition, they found that down-regulation of KCNA1 enhances the invasiveness of primary breast tumors, suggesting that KCNA1 has tumor inhibitory functions (Lallet-Daher et al. 2013).We proposed for the rst time that KCNA1 may be a potential ene in promoting the progression of GC, suggesting a poor prognosis.However, the role of KCNA1 in GC has not been reported yet, and our ndings should be further con rmed by more in-depth in vitro and in vivo studies.

This study is corelated with some limitations.First, although microarray-based bioinformatics analysis is a powerful tool for effective evaluation of molecular mechanisms as well as for identifying potential GC biomarkers, a whole loss-of-function and gain-on-function study in vivo and in vitro is needed to prove the expression of DEGs and Hub genes and clarify their functions in GC.Second, we used two microarray analyses, and the data sets were relatively small, which may lead to high false positive rates and ex parte results.To determine the diagnostic accuracy of GC-related circRNAs, multiple data sets should be integrated and the sample size increased.Finally, to con rm the correlations between hub genes and clinical phenotypes of GC, more GC patients should be included in subgroup analysis.

In summary, we constructed the ceRNA interaction and PPI network of GC circRNAs, and analyzed pathway enrichments of DEGs and hub genes.We also determined the correlation between expressions of hub genes and immune in ltrates in GC, as well as its correlation with clinical characteristics.Besides, we valeted the expression of has_circ_0002504 in vitro.Our ndings elucidate on the regulatory mechanisms of GC from the perspective of circRNAs.Two circRNAs were found as candidate biomarkers for diagnostic or possible therapeutic targets, while 2 hub genes were found to be novel prognostic factors.It will provide bene cial insights to study on circRNAs and generate novel hypotheses with regards to cancer pathogenesis.However, since our results are base

on bioinformatics analys
s and in vitro experiments, more studies should con rm these conclusions.Immune in ltration analysis of hub genes.



Furthermore, overexpression of DCLK1 (stage I vs stage II, p=