1. Gene variation of claudins in ovarian cancer
Twenty-four reviewed proteins of claudin family were obtained from the UniProt Knowledgebase (UniProtKB)(https://www.uniprot.org/) (Table 1) (An additional file shows this in more detail [see Table 1]). Firstly, we investigated the genetic variation of claudin family in ovarian cancer through the cBioProtal for Cancer Genomics (https://www.cbioportal.org/). Twenty-four genes were queried in 585 samples of ovarian serous cystadenocarcinoma (TCGA, PanCancer Atlas). Figure 1A showed the alteration frequency of genetic variation in serous ovarian cancer. Figure 1B showed that queried genes were altered in 363 (62%) of queried patients/samples. Among them, the top three gene variation were CLDN11 (24%), CLDN16 (22%) and CLDN1 (16%). Then, overall survival differences between altered group and unaltered group were compared by Kruskal Wallis test. We found that overall survival is reduced in altered group compared to unaltered group (p = 7.981e-3) (Fig. 1C). Previous studies have shown that claudin gene family dysregulated in a variety of tumors and involved in diagnostic, tumorigenesis, and prognosis[15–17]. Thus, this gene family is worthy of further research in ovarian cancer.
2. The expression of claudin family is dysregulated in various cancers
To explored the mRNA expression of claudin gene family, we investigated the expression profile of claudin genes in various cancer via the ONCOMINE. The thresholds were set: p-value of 0.05, fold change of 1.5, and gene rank of all. The significant unique analyses were shown in supplementary Fig. 1 (Those with less than 3 meaningful analyses were not considered). Most of claudins were dysregulated in various cancers. In order to further verify the expression of claudins in ovarian cancer, GEPIA2 were used to analyze the mRNA expression in TCGA samples and the GTEx data. The |Log2FC| cutoff was set 1, and p-value cutoff was set 0.01. As shown in Fig. 2, 8 genes were overexpression between ovarian cancer and normal samples, including CLDN1, CLDN3, CLDN4, CLDN6, CLDN7, CLDN9, CLDN10 and CLDN16; and 3 genes were low expression including CLDN5, CLDN11 and CLDN15.
3. Claudins expression were correlated with the prognosis of ovarian cancer
To identify which of these genes have clinical significance, we studied the relationship between these differentially expressed genes and the prognosis of patients with ovarian cancer using Kaplan-Meier plotter. As shown in Fig. 3, genes overexpression including CLDN3, CLDN4, CLDN6, and CLDN16 were found to be significantly correlated with poor overall survival (OS) (Fig. 3A) and progression free survival (PFS) (Fig. 3B) of patients with ovarian cancer. Besides, high expression of CLDN10 and CLDN15 predicted good prognosis among ovarian cancer (Fig. 3C-D). Surprisingly, CLDN10 is overexpression in cancer, but patients with high expression of CLDN10 predicted good overall survival (OS, HR = 0.73, logrank P = 1.6e-06), progression free survival (PFS, HR = 0.83, logrank P = 0.0067), and post progression survival (PPS, HR = 0.73, logrank P = 0.00029). These results are somewhat counterintuitive. Why does this happen? Further mechanism has to be explored.
4. GSEA of immunologic signature gene sets
To characterize the potential function of claudins, GSEA was performed using the gene expression data of ovarian cancer patients in TCGA. Immunologic signature gene sets were used. As shown in Fig. 4, we found that CLDN6 and CLDN10 were related to effector differentiation of B cell, CD4 T cell, and CD8 T cell.
5. Correlation Analysis between claudins and the tumor microenvironment
To understand the role of claudins in immunity, we downloaded 379 RNA-seq FPKM (Fragments per kilobase per million) data of ovarian cancer from TCGA. Subsequently, the FPKM were converted to TPM (transcripts per million)[18]. ESTIMATE algorithm[19] was used to predict tumor purity based on TCGA ovarian cancer samples. Then, the relationship between claudins expression and tumor microenvironment was explored. As shown in Fig. 5A, a meaningful negative correlation between CLDN6 expression and immune score was observed (spearman correlation = -0.23, p < 0.001). And, there was a positive correlation between CLDN10 expression and immune score (spearman correlation = 0.21, p < 0.001) (Fig. 5B). Neither CLDN6 expression nor CLDN10 expression was correlation with stromal score. Immune score represents the infiltration of immune cells in tumor tissue.
Then, we examined the relationship between immune infiltration and claudins expression. RNA-seq TPM data (n = 379) from TCGA ovarian cancer were used to assess 22 immune cells subtypes concentrations through the CIBERSORT algorithm[20]. They were grouped by the median value of CLDN6 and CLDN10, respectively. Dendritic cells activated were found to be statistically significant different between CLDN6_high and CLDN6_low group. Several cell types were significantly different between the CLDN10_high and CLDN10_low group, including B cells naïve, B cells memory, T cells CD4 naïve, T cells CD4 memory activated, monocytes, M1 macrophage and dendritic cells activated (Fig. 5C).
Besides, the microarray expression values of ovarian cancer were used for calculation the abundances of six immune infiltrates (B cells, CD4 + T cells, CD8 + T cells, Neutrophils, Macrophages, and Dendritic cells) via TIMER algorithm[19]. The gene expression levels correlated with tumor purity were displayed on the left-most panel (Fig. 6A-B). Our results showed the CLDN6 expression was negatively related to B cell infiltration (partial correlation = -0.284, p = 2.21e-10), CD8 + T cell (partial correlation = -0.254, p = 1.64e-08), neutrophil (partial correlation = -0.152, p = 8.29e-04), and dendritic cell (partial correlation = -0.182, p = 6.31e-05) (Fig. 6A). In contrast, there is a small but significant positive correlation between CLDN10 expression and neutrophil (partial correlation = 0.185, p = 4.66e-05), and dendritic cell (partial correlation = 0.153, p = 7.74e-04) (Fig. 6B).
In order to more accurately describe the relationship of gene expression and immune cell infiltration, several methods including TIMER, CIBERSORT, quanTIseq, xCell, MCP-counter and EPIC algorithms were used to assess the immune infiltration of tumor tissue[21]. TIMER2.0 provides a platform for analysis immune infiltrates across diverse cancer types based on available TCGA RNA-seq data[22, 23]. The correlations between claudins (CLDN6 and CLDN10) expression and various immune cells infiltration of ovarian cancer were shown in Table 2. As shown in Fig. 6C, CLDN6 was negative correlated to immune cell infiltration including B cell, CD8 + T cell, CD4 + T cell effector memory, M1 macrophage and myeloid dendritic cell. By contrast, CLDN10 was positive correlated to immune cell infiltration including B cell, CD8 + T cell, CD4 + T cell effector memory, M1 macrophage and myeloid dendritic cell (Fig. 6D). Relevant evidences reported that cancer associated fibroblast (CAF) plays an important role in the progression of ovarian cancer[24, 25]. Interestingly, we also found that CAF has a positive correlation with CLDN6 expression, but a negative correlation with CLDN10 expression. In ovarian cancer, increased infiltration of tumor-infiltrating lymphocytes (TILs) and more specifically CD8 + T cells, has been proven to be associated with improved clinical outcome[26–28]. These results suggest that CLDN6 and CLDN10 may participate in the immune cells infiltration of ovarian cancer, and these mechanisms may be the reasons for poor prognosis of ovarian cancer.
6. Relationship between claudins expression and gene markers of immune cells
To further illustrate the correlations between claudins (CLDN6 and CLDN10) and immune infiltration, we focused on the relationship between claudins (CLDN6 and CLDN10) and gene markers of various immune cells in ovarian cancer through the TIMER 2.0 databases. We analyzed the correlations between claudins (CLDN6 and CLDN10) expression and gene markers of different immune cells, including B cells, T cells (general), CD8 + T cells, macrophages, dendritic cells, neutrophils, monocytes, NK cells and Tregs in ovarian cancer (Table 3). The purity-adjusted correlation heatmaps were shown on supplementary Fig. 2. After the correlation adjustment by purity, the results revealed the CLDN6 expression level was negatively correlated with most gene markers of dendritic cells, M1 macrophages, monocyte, NK cells, and tumor-associated macrophages (TAMs) in ovarian cancer. By contrast, the expression of CLDN10 was positive correlated with gene markers of dendritic cells, T cell (general) and TAMs in ovarian cancer.
Studies have shown that the infiltration of these immune cells in the tumor microenvironment is related to the tumor immunotherapy response[29]. Immune cell-based immunotherapy[30] including NK Cells[31] and dendritic cells[32] play important roles in the treatment of ovarian cancer. Taken these analyses together, our research showed that CLDN6 and CLDN10 may play important roles in immunotherapy in the future.