Integrated Analysis of Prognostic and Immune Associated Integrin Family in Ovarian Cancer

Human integrin receptors are important for cell-cell and cell-matrix adhesion in normal epithelial cells. Emerging evidences have indicated integrin members are involved in cancer development and progression as well. However, the expression patterns and clinical significance of the whole integrin family in ovarian cancer (OC) have not yet been well understood. In the present study, we utilized the public datasets including GEPIA, GEO, ONCOMINE, cBioPortal, Kaplan-Meier Plotter, TIMER databases, to analyze the expression and prognostic value of integrin members in OC. We found ITGA3/B4/B6/B7/B8 were abnormally overexpressed in OC; ITGA6 was good prognosis predictor in OC; ITGA3/ B4/B8 were poor prognosis predictor specially in advanced OC patients; elevated ITGA3/B4 might promote metastasis and elevated ITGA3/B8 might promote platinum resistance of OC; ITGA3 and ITGB4 might synergistically or independently regulate cell adhesion and proliferation; ITGA4/AL/AM/AX/B2/B7 showed strong correlations with various tumor immune infiltrates (TILs), especially with pro-tumor immunes cell types like monocyte, M2 macrophage and exhaustion T cells infiltration; ITGAL/AM/B2/B7 and residing memory CD8+ T cells marker ITGAE were specially associated with early OC patients outcome. Our results implied that ITGA3/B4 were important prognostic markers of advanced OC, ITGAL/AM/ B2/B7 were immune associated prognosis markers of early OC, together they might render important therapeutic targets for OC.


INTRODUCTION
Ovarian cancer (OC) as the most lethal gynecological cancer, its mortality rate ranks fifth in female's death associated with cancer (Siegel et al., 2019). Patients diagnosed with OC are at a median age of around 60 years, and majority of them are diagnosed at advanced stages, with companion of metastasis. Lacking of effective early stage detection greatly contributes the poor survival rate of OC. Surgical debulking and platinum-based chemotherapy are the first-line therapy strategy for OC treatment. Although most patients show good responses to chemotherapy for the first time, recurrence probably develop within 18 months, and eventually result in succumbing to cancer (Jayson et al., 2014;Torre et al., 2018;Lheureux et al., 2019). Thus, effective indicators for early diagnostic, strategies for solving metastasis and chemotherapy resistance, or other novel treatments like immunotherapy are urgently needed to improve the prognosis of OC patients (Bast, 2003).
ITGA5, ITGAV, ITGB1, and ITGB3 are the most popular studied integrin members in OC, which can form α5β1, αvβ1, αvβ3 heterodimeric receptors. Most literature support they function as oncogenic genes as promoting adhesion, migration and proliferation in OC (Cruet-Hennequart et al., 2003;Sawada et al., 2008;Ruseva et al., 2009;Mitra et al., 2011;Ohyagi-Hara et al., 2013;Xue et al., 2013;Gong et al., 2016;Shinderman-Maman et al., 2016;Villegas-Pineda et al., 2017). However, some studies have reported higher expression of ITGB3 or ITGAV negatively associate with prognosis (Partheen et al., 2008(Partheen et al., , 2009 and invasion (Kim et al., 2008;Chen et al., 2009Chen et al., , 2016Kaur et al., 2009), suggesting they may be tumor suppressors in OC. Meantime, studies on other integrin members like ITGB2/7 and most of ITGAs remain sparse. It's important to figure out which integrin members are crucial and how they function in OC. Thus, we analyzed the expression of each integrin member in OC, and their relationship with prognosis, genetic alteration, metastasis, drug resistance, and immune infiltration, combined with exploring of the possible mechanism of certain crucial integrins, we provide an insight into the potential biological functions and prognostic value of integrins in OC.

GEPIA2 (Gene Expression Profiling Analysis) Database
GEPIA2 is an interactive web server developed by Zhang's lab, helped biologists to explore the large TCGA and GTEx datasets. GEPIA2 provides multiple functions including gene expression comparison with tumor/ normal, various cancer type or pathological stage, correlation analysis between genes or signatures, survival analysis, isoform analysis, similar genes detection and dimensionality reduction (Tang et al., 2019). For the Expression in Box Plot, the threshold of |log2FC| was 1, pvalue cutoff was 0.05, columns in red color represented tumor, columns in blue color represented normal. For the Correlation analysis, the correlation coefficient calculation method was Spearman correlation. Statistical significance with p < 0.05 was indicated with * p < 0.01 was indicated with * * p < 0.001 were indicated with * * * and p < 0.0001 were indicated with * * * * .

GEO (Gene Expression Omnibus) Datasets
GEO is a public functional genomic data repository contains numerous array and sequence-based study data. After searching the GEO database associated with ovarian cancer, GSE26712 (Bonome et al., 2008) was selected to validate GEPIA tumor/normal comparison results; GSE30587 (Brodsky et al., 2014) and GSE131978 GPL96 (Tassi et al., 2019) were selected to analyze the association of integrin members with ovarian cancer metastasis; GSE30161 (Ferriss et al., 2012), GSE51373 (Koti et al., 2013) and GSE131978 GPL570 (Tassi et al., 2019) were selected to analyze the association of integrin members with ovarian cancer chemoresistance. The data of selected GEO datasets were downloaded and reanalyzed with Graphpad Prism 8.0.1. Data from GSE30587 (Brodsky et al., 2014) which contained paired metastasis and primary samples were analyzed with paired t-test, others were analyzed with unpaired t-test, statistical significance with p < 0.05 was indicated with * p < 0.01 was indicated with * * p < 0.001 were indicated with * * * and p <0.0001 were indicated with * * * * . Gentric et al., 2019;Kieffer et al., 2020) and GSE14764 (Denkert et al., 2009) were chosen for external validation datasets. The ovarian cancer patients were split into high and low expression groups by mRNA expression of auto-selected best cutoff. The Kaplan-Meier survival plot showed Hazard ratio (HR), 95% confidence intervals (CI) log-rank p-value for each corresponding gene, and the number at risk below each plot. Cutoff of p < 0.05 was considered as statistical significance.

TIMER (Tumor Immune Estimation Resource)
TIMER (https://cistrome.shinyapps.io/timer/) is an interactive web server that can systematically analyze immune infiltrates in cancer with its own algorithm (Li et al., , 2017. We analyzed the correlation of each integrin member with the abundance of immune infiltrates in OC, including B cells, CD4+ T cells, CD8+ T cells, Neutrophils, Macrophages, and Dendritic cells, by gene modules. In addition, we also analyzed the association of each integrin member and gene markers of tumor-infiltrating immune cells by correlation modules. The gene markers of tumor-infiltrating immune cells listed in Table 4 were reported in previous studies (Sousa and Määttä, 2016;Danaher et al., 2017;Siemers et al., 2017). The correlation analysis is the Spearman's correlation, statistical significance with p < 0.05 was indicated with * p < 0.01 was indicated with * * p < 0.001 were indicated with * * * and p <0.0001 were indicated with * * * * .

Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) Enrichment
The Database for Annotation, Visualization and Integrated Discovery (DAVID) (https://david.ncifcrf.gov/) is a powerful tool to investigate functional annotation of a certain gene list (Huang da et al., 2009a,b). The genes showed correlation coefficient (r) >0.3 with ITGB4/8 were analyzed for GO and KEGG enrichment by DAVID. GO analysis included biological processes (BP), cellular components (CC), and molecular functions (MF). Top 10 significant enrichments of BP, CC, MF (p < 0.05 and rank by p-value) and all significant KEGG enrichments (p < 0.05) were visualized with ImageGP (http://www.ehbio.com/ImageGP/).

UniHI
UniHI 7 (http: //193.136.227.168/UniHI/pages/unihiSearch. jsf) integrates both protein-protein physical interaction and regulatory transcriptional interaction data from various data resources, such as HPRD, BIOGRID, INTACT, REACTOME, miRTarBase and TRANSFAC (Kalathur et al., 2014). In addition, UniHI 7 also integrates 4,203 different drugs and their corresponding targets (2139) information from DrugBank, so it can directly highlight the known drug target in the network contains PPI and regulatory interaction. The network of ITGA3 and ITGB4 was constructed by UniHI, and the interactions between integrin alpha subunit and beta subunit was also retrieved from UniHI.

The Aberrant Expression of Integrin Genes in Ovarian Cancer Patients
Malignant tumor development requires various oncogene activation and suppressor inactivation. To understand the role of integrin genes played in ovarian cancer, we started with a comparison of integrin genes transcriptional expression between OC samples from TCGA (The Cancer Genome Atlas) and autopsy normal ovarian sample from GTEx (Genotype Tissue Expression) by GEPIA2 (Gene expression Profiling Analysis). The results indicated that ITGA3 and ITGB2/B4/B5/B6/B7/B8 transcriptional levels were significantly higher in OC samples than in normal ovary tissue, while ITGA5/A7/A8/A10 expression levels were significantly lower and other integrins showed no statistic differences ( Figure 1A).
In addition, we also explored the expression of integrin genes in various cancers compared to their normal tissues with the ONCOMINE database which compiled numerous independent studies with different patient cohorts ( Figure 1C). One or more datasets showed significantly higher mRNA expressions of ITGA3 and ITGB2/B4/B6/B7/B8 in OC tissue, while ITGA8/A11/AE/AL/AM/AV showed significantly lower expression in OC tissue. However, some studies showed ITGA1/A5/A6/A7/A9 and ITGB1/B5 were statistically significantly higher expressed in OC but others showed the opposite. The specific results were shown in Table 1.
Taken together, it was more reliable to conclude the ITGA3 and ITGB4/B6/B7/B8 abnormally overexpressed in OC patients.

Genetic Alteration of Integrin Genes in Ovarian Cancer
Genetic alteration and transcriptional regulation result in mRNA expression alteration. Genetic alteration of integrin genes in OC were analyzed with cBioportal (TCGA, Firehose Legacy). Integrin genes altered in 343 samples from 594 ovarian cancer patients (58%), of which ITGA10 genes accounted for the highest alteration frequency (12%; Figure 2A). We found that genetic missense mutation was rare, while amplification and mRNA alteration more frequently occurred in OC patients. Furthermore, we found ITGAE/AV and ITGB1/B8 mRNA expression were moderately correlated with their relative linear copy number alteration (CNA) value, ITGA2/A3/A6/A7/A9/A2B and ITGB3/B4 mRNA expression showed weak correlations with their relative linear CNA value ( Figure 2B). Within the OC sample analyzed, most of integrin genes protein expression were in good correlation with their   level, protein level of ITGA8/A10/AD/AE and ITGB7 were not available (Supplementary Figure 1).
These results indicate integrin genes occasionally occurred with amplification and mRNA alteration in OC patients. The genetic alteration such as CNA partially led to the expression alteration of integrin genes, and the effect of post-transcriptional modification in most of integrins was minor.

Prognostic Value of Integrin Genes in Ovarian Cancer
To explore the potential prognostic value of integrin genes in ovarian cancer, the Kaplan-Meier plotter tool was used to perform the survival analysis in 565 TCGA ovarian cancer samples with complete survival and integrin gene transcriptional expression data. Among all analyzed integrin genes, we found increased ITGA3/A5/A10/A2B and ITGB3/B4/B8 expressions were significantly associated with poor OS, while decreased ITGA6/A7/AE and ITGB7 expression were significantly associated with poor OS in ovarian cancer patients ( Figures 3A-K). Meanwhile, higher mRNA expression of ITGA3 and ITGB4/8 were significantly associated with poor PFS, lower mRNA expression of ITGA4/A6/A7/A10/AX and ITGB1 were significantly associated with poor PFS in ovarian cancer patients (Figures 3L-T). To avoid the potential clinical outcome deviation caused by treatment, we further analyzed the prognostic value of integrin after excluding the minority patients without chemotherapy. After all, our results parallel showed ITGA5/B3/B4/B8 were significant risk factors for OS, ITGA6/AE were significant protective factors for OS; ITGA3/B4/B8 and additional ITGB6 were significant risk factors for PFS, ITGA4/A6/A7/A10/B1 were significant protective factors for PFS in OC patients received chemotherapy (Supplementary Table 1). Additionally, GSE26193 (Mateescu et al., 2011;Gentric et al., 2019;Kieffer et al., 2020), GSE14764 (Denkert et al., 2009), and GSE30161 (Ferriss et al., 2012) were selected for external validation of TCGA results. Among fourteen GEO ovarian datasets available on Kaplan-Meier plotter, they were chosen for (1) cohort lager than fifty, (2) major histology component and median OS/PFS was close to TCGA OC cohort. External dataset validated ITGA3/A10/A2B/B4/B8 were significantly associated with shorter OS, ITGA6 was significantly associated with longer OS; ITGA3/B6/B8 were significantly associated with worse PFS, and ITGA6/A7 were significantly associated with better PFS (Supplementary Table 1).
In addition, we also evaluated the prognostic value of integrin family members in different clinical stages and pathological grades through Kaplan-Meier plotter. As shown in Table 2, ITGA2/A6/A7/AE/ITGB5 mRNA expression significantly correlated with better OS with HR<1 and ITGA9/ITGB3 mRNA expression significantly correlated with worse OS with HR>1 in stage I and stage II OC patients. Similarly, ITGA7/B7 mRNA expression significantly correlated with favorable OS and ITGAV/B3/B4/B6 mRNA expression significantly correlated with unfavorable OS in stage III OC; transcriptional expression of ITGA9/AV/B6 significantly correlated with longer OS, and transcriptional expression of ITGA3/A2B/B8 significantly correlated with shorter OS in OC of stage IV. Meanwhile, ITGA6 mRNA expression significantly correlated with longer OS, and ITGA2/A3/A4/A7/A10/AL/AM/B2/B6 mRNA expression significantly correlated with shorter OS in pathological grade 1 and grade 2 OC patients. ITGAE/B7 mRNA expression significantly correlated with good OS, and ITGA2B/B3/B4 mRNA expression significantly correlated with poor OS in grade 3 OC patients.
Taken together, these results indicated that ITGA6 was a potential good prognosis factor in OC, especially valuable for the OS prediction in OC patients diagnosed with early stage or low grade, and valuable for PFS prediction in OC patients with all stages or grades. Our study also supports that ITGA3/ B4 might serve as primary poor prognosis factors in OC across multiple stages and grades; while ITGB8 mainly valuable for poor prognosis of advanced stage OS and high grade PFS in OC. Complementary, a set of integrins included favorable ITGAE/B7 and unfavorable ITGAL/AM/B2, mainly focused on early stage or grade OC predictions. However, these stage/grade-specific prognostic factors may need further validation.

Association With Clinical Parameters and Prognostic Independence of Integrin Genes
Next, we retrieved the clinical data and mRNA expression of TCGA ovarian cancer from the GDC data portal. There were 375 cases available with files of mRNA expression and matched clinical data. This set was used to analyze the association of particular integrin genes with clinical parameters and the Cox regression analysis. As primary prognostic factors for both OS and PFS, ITGA3/A6/B4/B8 were further analyzed. As indicated in the Supplementary Table 2, we found ITGA3 significantly associated with primary therapy outcome (p < 0.0001), ITGA6 significantly associated with OS status (p = 0.026), ITGB4 significantly associated with cancer status (p = 0.042) and tumor residual (p = 0.024), ITGB8 significantly associated with chemotherapy (p = 0.020). No significant association of specific integrin genes with age at diagnosed, race, ethnicity, stage, grade, PFS status, hormone therapy, immune therapy, and target therapy had been found.
Nevertheless, in the following univariant cox regression analysis, clinical features including chemotherapy, cancer status, tumor residual, primary therapy outcome were significantly associated with OS and PFS; while age at diagnosed, race only significantly associated with OS, and ethnicity only significantly associated with PFS. Meantime, high expression of ITGA3/B4/B8 were also significantly associated with poor OS, and high expression of ITGB4/B8 significantly associated with poor PFS      ( Table 4). However, clinical features including race, ethnicity, grade, cancer status, tumor residual, primary therapy outcome were not included for further multivariant cox regression analysis, due to considerable records missing. Multivariant cox regression analysis showed high expression of B4/B8 were independently poor prognostic factors of both OS and PFS, and high expression of ITGA3 were independently poor prognostic factor of OS in OC ( Table 5).

Association of Integrin Genes With Ovarian Cancer Metastasis and Drug-Resistance
Due to the difficulty in early diagnosis, about 2/3 ovarian cancer patients diagnosed at a late stage with extensive peritoneum metastasis, made the ovarian cancer so-called "silent killer." Integrin members played an important role in cellmatrix attachment in normal epithelial cells. It was possible some of them participated in the ovarian cancer metastasis process, such as colonization at distal. Here, we analyzed two distinct GEO datasets which compared the metastasis ovarian cancer to primary ovarian cancer. In the GSE131978 (Tassi et al., 2019) dataset, a group of primary OC samples were compared to a group of omental metastasis OC samples. The result showed ITGA3/7 and ITGB4 were significantly higher expressed in metastasis OC samples, ITGAV and ITGB1 were significantly higher expressed in primary OC samples ( Figure 4A). Furthermore, the GSE30587 (Brodsky et al., 2014) dataset compared the primary OC samples and metastasis OC samples from the same patients. Paired t-test analysis showed the mRNA expression of ITGA3/A5/A7/A11/AL/AM/AX and ITGB2/B4 were significantly higher in the metastasis OC samples ( Figure 4B). Combinatory, ITGA3/A7 and ITGB4 might be important for OC metastasis.
Most OC patients received surgery and chemotherapy after the initial diagnosis. The standard first-line chemotherapy was platinum-based chemotherapy. Platinum response was important for the prognosis of OC patients. The platinumresistant generally led to a poor outcome and remains a major impediment in the treatment. To explore whether integrin genes were associated with intrinsic chemotherapy resistance, we retrieved the transcription data from GEO with ovarian cancer chemotherapy response study. The GSE131978 (Tassi et al., 2019) defined patients with tumor progression time <6 months as relatively resistant, and patients with tumor progression time more than 12 months as relative sensitive. The analysis showed ITGA3 and ITGB4/8 significantly overexpressed in the platinum-resistant group ( Figure 4C). Another GEO dataset GSE30161 (Ferriss et al., 2012), compared to the responders and non-responders post-platinum-based chemotherapy, showed ITGA3/B8 were significantly higher in non-responders ( Figure 4D). However, in the GSE51373 (Koti et al., 2013) which compared resistant patients with <8 months PFS to sensitive patients with more than 18 months PFS, showed no significant difference in integrins expression (data didn't show). These results indicated ITGA3 and ITGB8 might be important for platinum-resistance.

Correlation Between Integrin Genes Expression and the Infiltration of Immune Cells in Ovarian Cancer
Despite the traditional surgery and chemotherapy, immunotherapy is a novel and promising treatment strategy. Ovarian cancer is immunogenic, about 50% of patients had spontaneous tumor immune responses with an abundance of immune cell infiltration. T cell-rich ovarian cancer was associated with improved OS and PFS (Zhang et al., 2003;Goode et al., 2017;Garsed et al., 2018;Marth et al., 2019), it indicated that stronger antitumor immune response might mean longer survival of OC patients. Meanwhile, T cell exhaustion and markers like PD-L1 might affect immune therapy efficacy (Abiko et al., 2013;Duraiswamy et al., 2013;Webb et al., 2015). Therefore, we investigate the relationship between integrin genes and immune infiltration levels in OC from TIMER. We observed the ITGA1/A2/A4/A5/A7/A8/A11/A2B/AL/ AM/AV/AX and ITGB1/B2/B3/B5/B6/B7 showed significantly obvious correlation with tumor purity in OC. Among them, ITGB2 showed highest correlation ratio (r = −0.63). Detail correlation between immune cell type and each integrin member were showed in Figure 5   ITGA4/AL/AM/AX and ITGB2 were more closely associated with neutrophil and dendritic cell infiltration; ITGA4/AL were more closely associated with CD8+ T cell infiltration; while ITGAX and ITGB2 were more closely associated with CD4+ T cell infiltration.

mRNA Correlation Analysis Between Integrin Genes and Tumor Infiltrate Markers
To further explore the correlation between integrin gens and immune infiltrates in ovarian cancer, we analyzed the mRNA correlation between integrin genes and markers of diverse immune infiltrates by TIMER database. Typical markers of (Sousa and Määttä, 2016;Danaher et al., 2017;Siemers et al., 2017).
We observed prominent correlation in ITGA4/AL/AM/AX and ITGB2/B7 with majority of TIL markers in OC, except B cell, M1 macrophage and natural killer cell markers ( Table 6). Among them, ITGA4/AL/AM/AX and ITGB2/B7 showed especially high correlation with monocyte, M2 macrophage and T cell exhaustion markers. ITGA4/AL/AM/AX and ITGB2/B7 showed relatively moderate correlation with T cell general, CD8+ T cell, neutrophil, dendritic cell Th1 and Treg markers. ITGA4/AL/AM/AX and ITGB2/B7 also showed weak correlation with Th2 and Tfh markers, and less correlation with M1 macrophage, Th17, B cell, natural killer cell markers.
Furthermore, we noticed the integrin transcriptional expression data from TCGA RNA seq, automatically clustered into 4 groups. One of the groups contained all the TIL highly associated integrin members, ITGA4/AL/AM/AX/B2/B7, which lacked of significantly association with drug resistance or metastasis, but major of them showed significantly associated with prognosis in early stage or grade. Another group contained integrin members ITGA3/B4/B8, which were significantly associated with drug resistance, metastasis and unfavored prognosis in advanced stage or grade (Supplementary Figure 3A). These results indicate the existence of TIL associated integrins in OC.
Studies proved malignant cancer cells and other cells recruit monocyte to OC microenvironment and differentiate into TAM (tumor-associated macrophage; Gupta et al., 2018). Most of macrophage and monocyte polarize into M2 macrophage and promote tumor progression and suppress immune response Carroll et al., 2016). Exhausted T cells might associate with poor prognosis by immunosuppressive mechanism and limiting antitumor response (Abiko et al., 2013;Webb et al., 2015;Alsaab et al., 2017). Thus, ITGA4/AL/AM/AX and ITGB2/7 might also render a promoting role in OC. Nonetheless, it's worth noting that some of the integrins might express on both immune cells and epithelial cells, unraveling the complicated interaction between TIL and integrins in OC still need further research.

Possible Mechanisms of ITGA3 and ITGB4 Involved in Ovarian Cancer
Our previous analysis revealed ITGA3/B4/B8 were associated with both poor OS and PFS, their prognostic role in OS were confirmed by external GEO datasets and independent as revealed by multivariant cox regression analysis. Other integrins like ITGA5/A2B/B3 were potential risk factors for OS, ITGB6 was a potential risk factor for PFS, ITGAE/B7 were potential protective factors for OS, ITGA4/AX/B1 were potential protective factors for PFS, and ITGA6/A7 were protective factors for both OS and PFS, yet not independent. Besides, ITGA3/B4/B6/B7/B8 were highly expressed in OC compared to normal ovarian tissue, ITGA3/A7/B4 were highly expressed in metastasis site compared to primary ovarian cancer, ITGA3/B8 were highly expressed in platinum-based chemotherapy resistance/non-response patients compared to sensitive/response patients. These highlighted the integrins ITGA3/B4/B8 and indicated they may be important in promoting ovarian cancer progression (Supplementary Figure 3B). It was interesting to notice that, the heterodimeric partners ITGA6 and ITGB4 were contrarily predicting prognosis of OS and PFS in OC, the other similar heterodimeric partners were ITGA5 and ITGB1, indicating possible independent functions of single integrin or novel integrin heterodimers under the pathological context of OC, like PPI indicated ITGA3 and ITGB4. Next, to explore the potential function of ITGA3 and ITGB4 in OC, their co-expressed genes (r > 0.3) were analyzed by GO (Gene Ontology) and KEGG (Kyoto Encyclopedia of Genes and Genomes) in DAVID (Database for Annotation, Visualization, and Integrated Discovery) database (Figures 6A,B). There were the same GOTERM and KEGG PATHWAY appeared in both ITGA3 and ITGB4 top enrichment lists, such as GO:0005515 (protein binding), GO:0098641 (cadherin binding involved in cell-cell adhesion) and GO:0005200 (structural constituent of cytoskeleton) in molecular function enrichment; GO:0005654 (nucleoplasm) and GO:0005913 (cell-cell adherens junction) in cellular components enrichment; hsa03030 (DNA replication), hsa03040 (Spliceosome), hsa05166 (HTLV-I infection), hsa03430 (Mismatch repair), hsa05222 (Small cell lung cancer), and hsa05200 (Pathways in cancer) in KEGG enrichment. In summary, ITGA3 and ITGB4 might synergistically promote OC through affecting mismatch repair, DNA replication, spliceosome and cell adhesion. It worth to mention ITGA3 associated genes uniquely enriched for hsa04512 (ECM-receptor interaction), hsa04510 (Focal adhesion), hsa04151 (PI3K-Akt signaling pathway) in KEGG (Figure 6A), while ITGB4 associated genes uniquely enriched for has04110 (Cell cycle) in KEGG; GO:0051301 (cell division), GO:0007067 (mitotic nuclear division), GO:0006260 (DNA replication), GO:0007093 (mitotic cell cycle checkpoint) and GO:0006271 (DNA strand elongation involved in DNA replication) in BP; GO:0003678 (DNA helicase activity), GO:0017137 (Rab GTPase binding) and GO:0005096 (GTPase activator activity) in MF (Figure 6B). ITGA3 coexpressed genes mainly enriched for extracellular matrix, but ITGB4 co-expressed genes mainly enriched for cytoplasm and nucleus. These results indicate ITGA3 might be more involved in cell-matrix interaction and trigger extracellular signaling delivery to intracellular; ITGB4 might be important for adhesion and proliferation regulation.
Although ITGA3 and ITGB4 might play important roles in the OC process and chemoresistance, the drug directly targeting ITGA3 or ITGB4 were rare. The only ITGB4 directly targeted  drug R1295 was still under investigation for the treatment of rheumatoid arthritis. Basing on text-mining, approved drug tyrosine for antidepressant and investigational drug serine for the natural moisturizing could potentially agonist both ITGA3 and ITGB4. Other drugs like approved drug calcipotriol for the plaque psoriasis, approved drug nitric oxide for the hypoxia respiratory failure, approved drug oxygen for the hypoxemia, approved drug calcium for the nutraceutical and investigational drug for the breakdown of muscle proteins could potentially agonist ITGB4. We also analyzed if they were indirectly druggable through their interaction partner proteins or their regulator microRNA and transcriptional factors by Unihi database (Figure 6C). The nodes in red showed drug targeting interaction partner of ITGA3 and ITGB4, which included GRB2, YES1, VCP, FYN, PRKCA, PRKCD, ITGAL, EGFR, ERBB2, and SMAD2 from ITGB4 partner; ITGB1, HGF, TGM2, ITGB3, CALR, EPHA2, and CD44 from ITGA3 partner; PTK2, ITGB2, MET, and UBC from both of them. The drugs targeted each interaction partners were illuminated in Table 7.
It should be noted that the interaction of ITGB2/ITGAL with ITGB4 and ITGB2/ITGB3/CD44 with ITGA3 were generated using a computational text-mining approach. However, proteinprotein interaction might be affected by a certain context in the cell, which interaction was accused of the OC process still needed further investigation, and effort of developing drugs directly target ITGA3 and ITGB4 should be taken into account.

Imbalance of Integrin Heterodimers in Ovarian Cancer
In the previous analysis, we noticed elevated ITGB4 mRNA showed significant association with poor prognosis, metastasis and platinum resistance. However, its heterodimeric partner ITGA6 showed a positive correlation with prognosis and lack significant correlation with metastasis and platinum resistance. Given that the heterodimeric integrin receptor is important for cell-cell and cell-matrix interaction, the disruption of heterodimeric structure in cancer might lead to metastasis. So, we analyzed the correlation of each integrin members in normal ovarian and ovarian cancer. It's well known that integrin formed heterodimeric receptors by interaction between ITGB1 and ITGA1/A2/A3/A4/A5/A6/A7/A8/A9/A10/A11/AV, ITGB2 and ITGAD/AL/AM/AX, ITGB3 and ITGAV/A2B, ITGB7 and ITGA4/AE, ITGB5/B6/B8 and ITGAV, ITGB4 and ITGA6 (Supplementary Figure 3B; Kobayashi et al., 2017). About half of the reported partner mRNA expression correlations obviously changed between normal and malignant ovarian tissue (Figure 7A), and majority of those altered mRNA expression correlation still established in protein level in OC (Supplementary Figure 3C). These changes also occurred in the potential important integrin members identified by our previous analysis. ITGB2 showed high correlations with ITGAD/AL/AM/AX in OC, while ITGB2 showed very low correlations with ITGAD/AL in the normal ovary. The correlation ratio between ITGB2 and ITGAL mRNA expression raised from 0.27 in normal ovary to 0.85 in OC, ITGB2 protein level was also significantly associated ITGAL protein level with a ratio of 0.67 in OC ( Figure 7B). In addition, ITGB2 showed tendency toward co-occurrency with ITGAX (Log2 Odds Ratio>3, p < 0.001), ITGAM (Log2 Odds Ratio>3, p < 0.001), ITGA4 (Log2 Odds Ratio = 2.756, p = 0.003) and ITGAL (Log2 Odds Ratio = 2.853, p = 0.010) in OC (cBioPortal-TCGA, Firehose Legacy) by mutual exclusivity analysis. This result supported our previous TIL analysis, indicated they were all TILs associated integrins and might function as heterodimeric receptors.
Opposite to TIL associated integrins, we found the mRNA expression correlation ratio between ITGB4 and ITGA6 decreased from 0.5 in the normal ovary to 0.087 in the OC (Figure 7C), which indicated a potential independent function in OC. Accidentally, we found the other integrin alpha subunit ITGA3 which showed elevated mRNA expression in OC and correlated with poor prognosis, metastasis and platinum resistance as well as ITGB4, had a physical protein interaction with ITGB4 as revealed by previous PPI network ( Figure 6C). More importantly, the correlation ration between ITGA3 FIGURE 6 | Possible mechanism of ITGA3 and ITGB4 involved in OC. Top10 significantly enriched GO TERM and all significant enriched KEGG pathways for (A) ITGA3 and (B) ITGB4. (C) Regulatory network of ITGA3 and ITGB4, nodes in orange and edges with an arrow represents upstream transcriptional regulator, nodes with gray edges showed partners with physical protein interaction, gray edges with red nodes showed interaction partners with available drug basing on drugbank database. and ITGB4, raised from 0.37 in normal ovary to 0.56 in OC ( Figure 7D). Although ITGA6 and ITGA3 showed similar gene alteration percentage (9%), and similar correlation ratio with CNA (0.33 and 0.35 respectively) and protein level (0.74 and 0.69 respectively) in OC (Figure 2B and Supplementary Figure 1). ITGB4 and ITGA3 showed a tendency toward co-occurrency   (Log2 Odds Ratio>3, p = 0.019), ITGB4 and ITGA6 showed no association (Log2 Odds Ratio = −0.046, p = 0.726) in OC (cBioPortal-TCGA, Firehose Legacy) by mutual exclusivity analysis. Besides, the ITGB4 protein level showed a lower correlation ratio (r = 0.35) with ITGA6 protein level, and a higher correlation ratio (r = 0.55) with ITGA3 protein level (Figures 7C,D). These results together indicated a potential disruption of alpha6beta4 heterodimer and a potential competitively formation of a novel alpha3beta4 heterodimer. Hence, we surmised a critical role of alpha3beta4 heterodimer or distinctly independent functions of ITGB4 and ITGA3 played in ovarian cancer. However, whether the imbalance or reconstruction of heterodimers affects ovarian cancer development of progression still needs further study.

DISCUSSION
This study demonstrates that the ITGA3 and ITGB4/B6/B7/B8 are highly expressed in OC when compared to normal ovary tissues. Further survival analysis of OC patients indicates increased ITGA6 mRNA positively associated with favored OS and PFS; increased ITGA3/B3/B4/B8 mRNA positively associate with unfavored OS and PFS in OC, especially with advanced OC; higher ITGAE/B7 mRNA associated with better prognosis; while higher ITGAL/AM/B2 mRNA associated with worse prognosis in OC, especially with early OC. ITGA3 and ITGB4 also significantly associated with metastasis and platinum resistance status in OC patients. These may due to their function in regulating cell adhesion and proliferation in OC.
in OC was vacancy. ITGB7 expression in immune cells were usually paired with ITGAE (CD103), especially in the residing memory CD8+ T cells (Le Floc'h et al., 2007;Anz et al., 2011;Franciszkiewicz et al., 2013). ITGAE highly expression accompany with high infiltrating T cells, was strongly associated with favored OS in HGSC (Webb et al., 2014a). Furthermore, CD103(+) CD8 TIL express PD-1 and appear quiescent in the HGSC tumor microenvironment, retain functional competence and demonstrate strong prognostic significance (Webb et al., 2014b). These were consistent with our analysis which showed high ITGAE and ITGB7 mRNA associate with favored OS in OC. Our study has some limitations. First, the present study relies on the public databases, which are constantly supervised and extended, it's possible to affect the outcome of the study. Second, the analysis of the integrin family in OC is limited in the mRNA expression level, while proteins are the actual function executor. Third, this study lacks the in vitro and in vivo experiment to confirm the function of integrins in OC. Hence, further studies are necessary to validate current findings and conjectures.
In summary, our results highlight the neglected integrin members ITGA3, ITGB4 and ITGB8, and speculate that they are important oncogene and independent prognostic factors in OC. This study also calls attention for ITGAL/AM/B2/B7 in immune response of OC, might offer clues to improve immunotherapy of OC.

DATA AVAILABILITY STATEMENT
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/Supplementary Material.

ETHICS STATEMENT
Our study protocol was approved by the Ethics Committee of the Xiangya Medical College of Central South University. All data were retrieved from the online database, it was confirmed that all written informed consent had been obtained.

AUTHOR CONTRIBUTIONS
AW and SZ designed the research. AW, SZ, JL, YH, and WD retrieved and analyzed the data, AW and SZ wrote the paper. GS and GY supervised the analysis and revised the paper. All authors contributed to the article and approved the submitted version.