Identification of KIF23 as a prognostic signature for ovarian cancer based on large scale samples and clinical validation

Background: Ovarian cancer is one of the common malignant tumors in gynecology. Although the treatment strategy for ovarian cancer has been greatly improved in recent years, due to the metastasis, recurrence and drug resistance, the 5-year overall survival rate of patients is still less than 47%. However, at present, there is no specific markers for clinical application. The purpose of this study is to verify the expression and clinical significance of KIF23 in ovarian cancer and identify potential targets for the clinical treatment of ovarian cancer. Methods: The expression of KIF23 in ovarian cancer tissues and its relationship between survival prognosis and clinical pathological parameters were analyzed in Oncomine, GEO, and TCGA databases. KIF23 expression was analyzed by Kaplan-Meier plotter database and its relationship with chemo-resistance was studied. The molecular mechanism involved in KIF23 was analyzed from the perspective of gene mutation, copy number variation and other genomics. Finally, immunohistochemistry experiment was used to verify the expression of KIF2, and its relationship between the clinical pathological parameters and prognosis of ovarian cancer patients was analyzed by single factor and multivariate Cox regression models. Results: Bioinformatic and experimental results have demonstrated that KIF23 is highly expressed in ovarian cancer, and its high expression is positively correlated with poor prognosis. Overexpression of KIF23 can cause chemotherapy resistance in ovarian cancer and affect the overall survival of patients. Genomics analysis showed that KIF23 expression was associated with mutations such as FLG2 and TTN, and it was significantly enriched in tumor signaling pathways such as DNA replication and cell cycle. Conclusions: KIF23 can not only be used as a biomarker of poor prognosis in patients with various stages of ovarian cancer, but also be used as a molecular targeted drug and an

4 independent prognostic biomarker for the treatment of ovarian cancer patients.

Background
Ovarian cancer has the highest fatality rate among gynecological malignant tumors [1].
Due to its unclear pathogenesis and lack of sensitive screening methods, 70 to 80% of patients are diagnosed with advanced-stage disease and have a 5-year survival rate of less than 45%. Therefore, exploring the mechanisms underlying the development and progression of ovarian cancer and identifying useful biomarkers for this disease are critical needs [2].
The kinesin protein superfamily (kinesin family, KIF) belongs to the class of molecular motors. Its globular head has ATPase activity, which can obtain energy by hydrolyzing ATP and changing its configuration [3]. Kinesins are involved in the transport of vesicles, organelles, chromosomes, and RNA-binding proteins in cells, the formation of spindles and intermediates, and the separation of chromosomes. Abnormal expression of KIF family members plays an important role in tumor development [4][5]. Kinesin family member 23 (KIF23), which belongs to the kinesin 6 family, is localized to the mitotic spindle region. It plays an important role in mitotic cytoplasmic separation [6][7]. High KIF23 expression levels affect normal cytokinesis and centrosome formation, which in turn leads to the arrest of cell division or abnormal division, resulting in aneuploid cells that cause tumorigenesis. High KIF23 expression can also regulate AKT activity, levels of phosphorylation and proteasome degradation pathway, leading to tumor invasion [8][9].
KIF23 is highly expressed in a variety of tumors, such as breast cancer, gastric cancer, and lung cancer. KIF23 overexpression is significantly associated with tumor grade, invasion, and prognosis in breast cancer [5]. The high expression of KIF23 in glioma cells may be related to transcriptional activation. KIF23 knockdown can significantly inhibit glioma cell proliferation in vitro and in vivo [10]. Zhao C et al found that KIF23 expression is significantly elevated in glioma samples. MiR-424 acts as a tumor suppressor and inhibits cell migration and EMT by targeting KIF23 in gliomas [11]. Murakami et al. [12] found that KIF23 expression levels are significantly elevated in gastric cancer and associated with poor prognosis. Kato et al. [13] found that KIF23 expression is significantly increased in NSCLC tissues, especially in adenocarcinoma tissues, and patients with high expression typically have a poor prognosis.
Previously, we found that KIF23 was a poor prognostic indicator for endometrial cancer [14]. In this study, we analyzed KIF23 expression in multiple databases and studied the correlation between its expression and clinical stage, tumor grade, and patient prognosis in ovarian cancer. Immunohistochemistry experiment was conducted in order to verify the conclusion. Finally, we explored the molecular pathways and functions of KIF23 involved in the development of ovarian cancer. Our results provide clues for understanding the roles and mechanisms of KIF23 in the development of ovarian cancer.

Oncomine analysis
The Oncomine database is a database of gene chips with an integrated data mining platform. In this database, the conditions for filtering and mining data can be set to accommodate specific needs. The screening conditions for the current study were as follows: ① "Cancer Type: Ovary cancer"; ② "Gene: KIF23"; ③ "Analysis Type: Cancer vs Normal Analysis"; ④ threshold setting conditions of P < 0.01, fold-change > 2, and gene rank is top 10%.

GEO analysis
Three ovarian cancer datasets were analyzed from the GEO database (GSE14407 [15], ([HG-U133_Plus_2] Affymetrix Human Genome U133 + 2.0 Array). GSE14407 contained 12 samples each of serous papillary ovarian cancer and normal ovarian epithelium. GSE18520 contained 53 high-grade serous papillary carcinoma samples and ten paracancerous samples. GSE54388 contained 16 well or moderately differentiated ovarian cancer samples and six normal ovarian epithelial samples. The datasets were processed, calibrated, standardized, and log2-converted using the R package. The KIF23 expression was extracted from the three datasets, and differential expression box plots were drawn using "ggpubr" in the R package.
Subgroup conditions including clinical stage, tissue classification, and mutation status were defined in the database. The prognosis of ovarian cancer patients with different stages, tumor grades, and mutations was analyzed based on KIF23 expression levels. The hazard ratios (HRs) with 95% confidence intervals (CIs) and log rank p-values were generated.

TCGA analysis
An ovarian cancer dataset consisting of 374 tumor tissue samples was downloaded and pre-processed from the TCGA database (https://tcga-data.nci.nih.gov/tcga/). KIF23 expression was ranked from low to high, and the samples were divided into four equal parts. The first and last 25% of the samples were selected as the low expression and high expression groups, respectively.

Function and Pathway Enrichment Analysis
Proteins that interact with KIF23 were found in the cBioportal (http://www.cbioportal.org/) [19] database. Based on the p-values, the top 50 significantly related molecules were screened using DAVID Bioinformatics Resources (http://david.abcc. Ncifrrf.gov/) [20], Gene Ontology (GO) functional analysis was performed on the genes corresponding to the above proteins, and the significance of the enrichment to the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways was calculated using hypergeometric test according to the formula.
In formula (1), N is the number of genome-wide genes, M is the number of genes annotated to a given pathway in the whole genome, n is the number of genes in the network, and m is the number of genes annotated to a given pathway. In addition, there were 28 cases of pelvic or para-aortic lymph node metastasis, while 87 had no metastasis.

Immunohistochemistry
Ovarian tissues were fixed in 10% formalin and processed into 5-∝m thick paraffin sections. The samples were dewaxed with discontinuous concentrations of ethanol and blocked to inhibit endogenous peroxidase. The sections were heated in a microwave for antigen retrieval, cooled to room temperature, and blocked by incubation in goat serum for 30 minutes at 37ºC. Samples were incubated in rabbit anti-KIF23 (Abcam, 1:200 dilution) overnight at 4ºC, followed by incubation with horseradish peroxidase-coupled goat anti-rabbit secondary antibody at 37ºC for 30 minutes. Nuclei were stained blue by hematoxylin. Sections were then dehydrated, cleared by xylene, and mounted. The SP kit was used according to the manufacturer's instructions. Samples were deemed KIF23positive when strong granular staining in the cell nucleus and cytoplasm was present.

Statistical analysis
Data were analyzed using SPSS 22.0 software (IBM Corporation, Armonk, NY, USA). Chisquared and Fisher's exact tests were used to analyze counting data, and the t-test was used to analyze measurement data. Survival curves were analyzed by KM and log-rank tests. The Cox regression model was used to analyze the relationships between KIF23 expression and clinical data. P < 0.05 indicated statistically significant differences.

KIF23 expression in Oncomine and GEO databases
A total of 447 KIF23 studies in different types of cancer were collected from the Oncomine database (Fig. 1A). Of these studies, 67 studies showed statistically significant differences in the expression of KIF23 (64 studies showed significant increases; three studies showed significant decreases). Furthermore, analysis of KIF23 expression in independent ovarian cancer datasets [21][22][23], which contained 657 ovarian cancer samples and 22 normal samples, showed that KIF23 expression levels in all ovarian cancer groups were significantly higher than in the normal group (P < 0.01) (Fig. 1B-D). These results were verified using three independent ovarian cancer microarrays (GSE14407, GSE18520, and GSE54388) from the GEO database. Together, these results indicated that KIF23 expression in ovarian cancer tissues was significantly higher than that of adjacent noncancerous tissues (P < 0.001) ( Fig. 1E-G).
Correlation between KIF23 mRNA expression and clinical pathological parameters in ovarian cancer using TCGA database To investigate the relationship between KIF23 expression and clinical-pathological parameters in ovarian cancer, we used the pathological data for ovarian cancer from the TCGA database, which contained the complete clinical data for 360 cases, including clinical stage, tumor grade, and age. Statistical analysis showed that KIF23 high expression was significantly correlated with poor differentiation (P < 0.05) ( Table 1).
However, KIF23 expression did not significantly correlate with the FIGO stage or age. Box plots of KIF23 expression and the clinical-pathological parameters are presented in Fig. 2A  Relationship between KIF23 mRNA expression and ovarian cancer prognosis The censored data analyzed for overall survival (OS) consisted of 1,656 ovarian cancer cases. The overall survival of patients with high KIF23 expression was significantly worse than that of patients with low KIF23 expression (HR 1.31; CI 1.13 to 1.53; log-rank P = 0.00032) (Fig. 3A). Analysis of 587 ovarian cancer cases (censored data) for the relationship between progression-free survival (PFS) and KIF23 expression yielded similar results (HR 1.27; CI 1.12 to 1.45; log-rank P = 0.00026) (Fig. 3B). Thus, the disease-free survival time of patients with high KIF23 expression was significantly shorter than that of patients with low KIF23 expression. The analysis also demonstrated that patients with low KIF23 expression had a better prognosis for mutant TP53 ovarian cancer (Fig. 3C), while low KIF23 expression did not affect the prognosis of wild-type TP53 ovarian cancer patients (Fig. 3D).
Relationship between KIF23 mRNA levels and prognosis of patients with different stages of ovarian cancer For PFS, higher KIF23 expression levels in early-stage ovarian cancer (stages I/II/I + II) were associated with a worse prognosis than low expression levels ( log-rank P = 0.000) for stage I, 6.15 (CI 1.83 to 20.7; log-rank P = 0.001) for stage I + II, and 1.29 (CI 1.11 to 1.49; log-rank P = 0.001) for stage III + IV ( Table 2, Fig. 4G, H). Based on these results, KIF23 could be used as an indicator of poor overall prognosis for all stages of ovarian cancer.  Fig. 5E, F). There were no significant differences between grades 13 for grade 4.  (Fig. 6D), suggesting that overexpression of KIF23 may cause resistance to platinum-based chemotherapy. However, KIF23 had no significant effect on the PFS of ovarian cancer patients treated with paclitaxel chemotherapy alone or platinum-paclitaxel combination therapy (Fig. 6E, F showed that KIF23 expression was significantly upregulated in epithelial ovarian cancer (Fig. 7). The positive expression and high positive expression rates in the epithelial ovarian cancer group were 94.78% and 87.83%, respectively, which was significantly higher than that of the borderline (40% and 10%, respectively), benign (30% and 5%, respectively), and normal (25% and 0%) groups (P < 0.05 for all comparisons) ( Table 4).  (Table 6). The 5-year survival time of the KIF23 high expression group was significantly lower than that of the KIF23 low expression group (Fig. 8A). Patients with stage III-IV disease had a shorter survival time than those with stage I-II (Fig. 8B). Patients with poorly differentiated tumors had shorter survival times than those with well or moderately differentiated tumors (Fig. 8C). Patients with lymph node metastasis had shorter survival times than patients with no metastasis (Fig. 8D) (Fig. 9A, B). However, the role of these mutations in ovarian cancer has not been thoroughly investigated. Using KEGG pathway analysis, we found that the high KIF23 expression mutation group was significantly enriched for dopaminergic synapse, beta-alanine metabolism, and glycosaminoglycan degradation (Fig. 9C). We hypothesize that KIF23 regulates these pathways to influence the occurrence and development of ovarian cancer mutations.
Ovarian cancer copy number variation data obtained from XENA (https://xenabrowser.net/datapages/) was divided into deep deletion, shallow deletion, diploid, gain, and amplification groups and compared to KIF23 expression levels. Our analysis demonstrated that KIF23 expression was significantly increased as the copy number was amplified (Fig. 9H).
GSEA gene enrichment analysis revealed that high KIF23 expression samples were significantly enriched for DNA replication and cell cycle gene sets (Fig. 9D, E). Of the identified genes, 160 genes were closely related to KIF23 expression (Pearson |R| > 0.5 using the TCGA ovarian cancer gene expression profile data). GO functional enrichment analysis was performed to explore the biological behavior and molecular function of these genes. We found that gene expression associated with KIF23 expression was mainly involved in the regulation of cell cycle phase transition, cell cycle checkpoint, G2/M transition of the cell cycle, and DNA replication (Fig. 9F). The heat map based on the expression of the KIF23 interacting genes, GO function, and tissue differentiation is presented in Fig. 9G. The data showed that the genes associated with KIF23 expression were highly correlated with these functional sets, and, like KIF23, these genes were expressed more strongly in poor differentiated ovarian cancer samples compared to well or moderately differentiated ovarian tumors.

Discussion
The kinesin superfamily (KIF) is involved in a variety of normal cellular biological activities, such as cell mitosis and intracellular transport of vesicles and organelles [24].
Overexpression of certain kinesins, such as Eg5, can induce excessive spindle separation, causing uneven distribution of genetic material, thereby forming aneuploid progeny cells, which are involved in cancer invasion and metastasis [25][26]. Downregulation of certain kinesins, such as KIF20B, can cause mitotic arrest or cytokinesis defects, triggering apoptosis through p53 or other signaling pathways [25,27]. Studies have shown that KIF23, as a member of the kinesin superfamily 6, is highly expressed in gastric cancer and is positively correlated with pTNM stage and poor prognosis. Knocking down KIF23 can inhibit the proliferation of gastric cancer cells [28]. Sun [29]  It is reported that TP53 mutations exist in more than 50% of the advanced epithelial serous ovarian cancers, and the frequency of TP53 mutations can be as high as 80% when using purified tumor samples for sequence analysis [33][34][35][36]. TP53 mutation is involved in the occurrence and development of epithelial serous ovarian cancer [37]. Therefore, we  [41]. OBSCN mutation is closely related to breast cancer and colorectal cancer [42][43]. However, the role of these mutations in ovarian cancer is rarely discussed in existing studies. Based on these findings, we further discussed the KEGG pathway of gene mutation, and we found that the high expression mutant group of KIF23 gene was significantly enriched in Dopaminergic synapse, beta-Alanine metabolism, Glycosaminoglycan degradation and other pathways.
Beta-Alanine plays an anti-tumor role by inhibiting the migration of cervical cancer and kidney tumor cells [44]. Glycosaminoglycan is involved in multiple signal cascades required for angiogenesis, invasion and metastasis [45]. Therefore, we suggest that KIF23 can affect the occurrence and development of ovarian cancer by regulating these pathways.
Copy number variation ((CNV)) is closely related to genetic and phenotypic diversity of cancer [46]. By identifying the copy number variation of the whole genome of ovarian cancer, the regions with frequent increase and decrease of copy number have been identified. In addition, high-level amplification of CCNE1, RB1, MYC, ERBB2, PIK3CA, EVI1, AKT2, NOTCH3 and FGFR1 genes can be used as a predictive marker of ovarian cancer [47][48][49]. However, the copy number of KIF23 in ovarian cancer has not been studied.
Therefore, we analyzed the KIF23 copy number, and the results showed that KIF23 gene expression increased significantly with the amplification of sample copy number. Our results showed that the high expression of KIF23 in ovarian cancer is partly caused by copy number amplification. GSEA gene enrichment analysis showed that KIF23 highly

Conclusions
In conclusion, KIF23 is highly expressed in epithelial ovarian cancer, and its high expression indicates a poor prognosis. KIF23 can be used as an independent marker to predict the prognosis of patients with ovarian cancer.     Relationship between prognosis and KIF23 mRNA expression in patients with different stages of ovarian cancer   Molecular mechanism of KIF23 in the development of ovarian cancer