A centrosome clustering protein, KIFC1, predicts aggressive disease course in serous ovarian adenocarcinomas

We first examined whether KIFC1 is upregulated in human ovarian cancers by analyzing KIFC1 overexpression in EOC clinical samples. To this end, we immunostained paraffin-embedded formalin-fixed tissue microarrays of EOC (n = 120) and normal ovarian epithelial tissue (n = 13) for KIFC1. The staining intensity was scored as 0 = none, 1 = low, 2 = moderate, or 3 = high, and percentage of positive cells (i.e., with 1+ staining intensity) from randomly selected fields (~500 cells) was determined [18]. The product of the staining intensity and the percent of positive cells constituted the Weighted Index (WI). Descriptive statistics regarding patient and clinicopathological characteristics is given in Tables 1, 2 and 3. In consonance with our previously published study [24], our immunohistochemical analysis showed overexpression of KIFC1 in EOC tissues with negligible expression in normal ovarian epithelial tissue (Fig. 1A). We found that the number of positively-stained nuclei per field in high-grade ovarian cancers (Fig. 1A) was significantly higher compared to low-grade ones. We then compared the nuclear KIFC1 WI values for normal and tumor samples and also across grades for tumor samples. Interestingly, we observed that nuclear KIFC1 WI was significantly higher in EOC tissues when compared to normal tissues (p < 0.01). Also, the nuclear KIFC1 WI increased with increasing tumor grade (Figure 1Bii) (p < 0.05). Among subtypes, we noticed that the number of positively-stained nuclei per field in high-grade serous ovarian cancers (Additional file 1: Figure S1A) was significantly higher compared to low-grade serous ovarian cancers (p < 0.05). Collectively, these observations indicate robust KIFC1 overexpression in human ovarian adenocarcinoma and strong association of KIFC1 expression levels with clinical progression of the disease. These data suggest that KIFC1 might play an active role in driving the progression of tumors into more malignant and aggressive forms.

Having established a significant correlation between KIFC1 expression and tumor differentiation, we next wanted to determine if there is any association between KIFC1 gene expression and clinical outcomes (overall survival (OS)) for ovarian cancer patients. To this end, we examined single channel microarray data from GEO (GSE9899) [25] to compare the expression levels of KIFC1 among different subtypes. Interestingly, we found that the gene expression levels of KIFC1 were significantly higher in serous ovarian adenocarcinoma (n = 154) when compared to all other subtypes (Borderline serous adenocarcinoma, n = 18 and Peritoneal serous adenocarcinoma, n = 22) (Fig. 2A). Further, we examined grade-wise trends in KIFC1 expression in serous ovarian adenocarcinoma. We observed a significant increase in KIFC1 expression levels with increasing grade (Fig. 2B). OS was calculated as the time interval (in months) from the date of histological diagnosis to date of death from any cause. We then carried out a survival analysis wherein patients were stratified into high- and low-KIFC1 expressing subgroups using the optimal KIFC1 expression cut-point (based on the log-rank test). Irrespective of the histological subtypes (n = 284), those with higher KIFC1 expression had shorter OS (p < 0.067) than patients with lower KIFC1 (Additional file 1: Figure S2A). To investigate in-depth, we performed a similar survival analysis by stratifying serous ovarian adenocarcinoma patients (n = 201) on the basis of site (primary, n = 154 and metastatic, n = 47) of sample collection. Univariate regression revealed high KIFC1 gene expression correlated significantly (HR = 2.14, p = 0.024) with poor OS in primary tumors only (Fig. 2C) but not in metastatic ones (data not shown). This association stayed significant (HR = 2.6, p = 0.006) during multivariate analysis when potentially confounding factors like grade and tumor stage were added (Additional file 1: Figure S2B). In sum, enhanced gene expression level of KIFC1 in primary tumors is strongly associated with poor clinical outcomes.

Next, we sought to examine the correlation of KIFC1 and genes driving CA. We analyzed expression levels of genes including CCNA2, CDK1, NEK2, AURKA, MYCN, CCNE2, STIL, LMO4, PLK4, MDM2, CEP63, E2F1, E2F2, E2F3, CEP152, PIM1, PIN1 andCCND1, whose deregulation is known to drive CA [18, 26‐28]. Specifically, we tested the associations between Robust Multi-array Average-normalized expression levels of these genes in primary SOC from 154 patients using Gene Expression Omnibus (GEO) series GSE9899. Higher expression of KIFC1 was significantly correlated with high expression of CCNA2, CDK1, NEK2, AURKA, E2F2, MYCN, STIL, CCNE2, E2F3, LMO4, PLK4, PIN1 and E2F3 (Table 4). These results suggest that KIFC1 upregulation and enhanced centrosome clustering in the serous ovarian adenocarcinomas may enable tumor cells to manage their increased centrosomal load, avert mitotic catastrophe and promote survival.

Next, we identified the biological processes which are deregulated in the KIFC1 high risk group. To this end, we probed the publicly-available microarray dataset (GSE9899) and stratified the 154 serous ovarian adenocarcinoma patients from the dataset into KIFC1-high and KIFC1-low groups. We then identified the gene ontologies of significantly overexpressed genes associated with the KIFC1-high group utilizing the PANTHER classification system. When pathway analysis was performed we found that majority of the genes overexpressed were associated to cellular processes like cell communication, cell cycling, cytokinesis and cell proliferation (Fig. 3Ai, ii). We then validated these results by performing the Gene set enrichment analysis (26). We found that KIFC1 high group was significantly (Fdr <0.25 and ES p < 0.05) enriched in centrosome and cell cycle gene sets (Fig. 3Bi,ii and Additional file 1: Figure S3A) (see Additional file 1: Tables S1, S2 and S3 for these and all other enriched gene ontologies). The results from GSEA showed that the top 100 gene sets enriched in KIFC1 high group were among the ones which plays key roles in, driving CA (NEK2, PLK1,CCNA2), clustering centrosomes (PRC1), microtubule spindle (KIF11, NUSAP1, NUMA1) etc. Altogether our data shows that the KIFC1-high group had a preponderance of genes representing all four important mitotic kinases –namely Polo-like kinases (PLK1), Aurora kinases (AURKA, AURK B), cyclin dependent kinases (CDK1) and NIMA related kinases (NEK1, NEK2). The coordination of progression through mitosis is mainly orchestrated by protein phosphorylation ensured by these kinases. Thus, it is reasonable to speculate that overexpression of these kinases results in deregulation of the cell cycle resulting in abnormal mitosis that generates cells with aberrant centrosomes and abnormal chromosomal content.

Having confirmed the association between upregulation of KIFC1 gene and CA genes in HGSOC, we wanted to investigate the CA profile in well-established in vitro cell lines that mimic HGSOC. To this end, we first screened four well-established cancer cell lines (namely, KURAMOCHI, OVCAR3, OVSAHO and SKOV3 by immunostaining centrosomes (γ-tubulin, green) and microtubules (α-tubulin, red) and counterstaining nuclei with DAPI (blue) Fig. 4a. Employing confocal microscopy we imaged 10 areas of interest (at least 500 cells were counted per cell type). Cells with abnormal number (more than two) of gamma tubulin spots were considered as cells with amplified centrosomes. We found that KURAMOCHI exhibited the highest percentage of cells with amplified centrosomes (~38 %) followed by OVSAHO (~24 %), OVCAR3 (~15 %) and SKOV3 (~9 %) (Fig. 4b). In a recent molecular profiling study by Domcke et al., KURAMOCHI and OVSAHO were selected as the representative cell lines for HGSOC [29]. Thus, our findings here parallel previous studies that recognize CA as a biomarker of aggressive tumors. Furthermore, we validated our results by evaluating the expression levels of centrosome-related proteins by performing immunoblotting assays. We found that the cell lines with high CA (KURAMOCHI and OVSAHO) expressed higher levels of centrosome structural proteins (densitometry values for centrin-2 relative to loading control β actin (KURAMOCHI - 0.291445, OVSAHO - 0.432561) and proteins whose dysregulation is known to drive CA (for Cyclin-E and Aurora A, KURAMOCHI- 0.194213 and 0.256828, OVSAHO- 0.428814, 1.664283 respectively) (Fig. 4c). Our next step was to investigate if aberrations in centrosome numbers among the different cell lines had any bearing on the mitotic spindle geometry. Interestingly, we found that the percentage of multipolar mitotic cells in three cell lines (OVSAHO, SKOV3 and OVCAR3) was lower (by ~2 fold) in comparison with the proportion of cells with supernumerary centrosomes (Fig. 4b). This difference in the proportion of cells with CA and multipolar spindles clearly supports the hypothesis that ovarian cancer cells cluster supernumerary centrosomes to form pseudobipolar poles. But as the results here indicate that KURAMOCHI showed significantly more multipolar mitoses when compared to the other ovarian cancer cell lines we tested, we evaluated if there existed variability in the level of clustering molecules that help cancer cells to deal with supernumerary centrosomes by corralling them to form pseudobipolar spindles [24]. To this end, we performed immunoblotting to evaluate expression level of centrosome clustering protein KIFC1 in cell lysates obtained from the ovarian adenocarcinoma cells (KURAMOCHI, OVCAR3, OVSAHO and SKOV3). We found that all the three cell lines with pronounced centrosomal clustering expressed higher levels of KIFC1 (SKOV3- 0.342396, OVCAR3- 0.204796 and OVSAHO- 0.452534) whereas negligible KIFCI expression was noted in KURAMOCHI (0.145452). It is noteworthy to mention that a recent report shows that KURAMOCHI is the only cell line that did not induce tumorigenesis in vivo [11]. This finding resonates with our notion that centrosome clustering is essential for the viability of cancer cells with extra centrosomes and therefore determines their tumorigenicity.

The in vitro findings were validated in silico by probing publically-available microarray dataset using Gene set. We interrogated publically-available microarray dataset of ovarian cancer cell lines (GSM133614, GSM133609, GSM887467 and GSM887488). We calculated a cumulative gene expression-based centrosome amplification index (CAI) by adding log-transformed, normalized gene expression for both structural centrosomal proteins (CETN2 (centrin-2), TUBG1 (γ-tubulin), PCNT2 (pericentrin)), and genes implicated in centrosome amplification (PLK4 (polo-like kinase 4) and CCNE1 (cyclin E) genes) (Additional file 1: Fig. S4 Ai). The analysis showed that CAI genes are expressed in all cell lines but is highest in OVSAHO. In addition, we evaluated gene expression levels of KIFC1 and found that the gene expression levels of KIFC1 were higher in cancer cell lines in comparison to normal ovarian surface epithelial cells (Additional file 1: Figure S4 Aii). Taken together, our results indicated that CA and KIFC1 levels are associated with HGSOC cell lines.