Background
Ovarian cancer is the second most common cause of gynaecologic cancer death in women worldwide. Nearly 295,000 women were estimated to have been diagnosed with ovarian cancer and 185,000 to have died from disease in 2018, with rates varying across the world [
1]. Over 90% of ovarian malignancies are categorized as epithelial ovarian cancers, and currently five main types are identified: high-grade serous (70%), low-grade serous (< 5%), mucinous (3%), endometrioid (10%), and clear-cell (10%) carcinomas. These are heterogeneous diseases with different epidemiological and genetic risk factors, precursor lesions, patterns of spread, molecular events during oncogenesis, response to chemotherapy, and prognosis [
2]. High-Grade Serous Ovarian Cancer (HGSOC) typically presents at advanced stage (III-IV) and, despite the initial response to surgical debulking and first-line therapy with carboplatin and paclitaxel (with or without bevacizumab), most tumours eventually develop drug resistance, with a 5-year survival generally below 30% [
3]. Major improvement in maintenance therapy has been seen by incorporating inhibitors against poly (ADP-ribose) polymerase (PARP), but, until now, no significant improvement in overall survival has been achieved in non-BRCA-mutated patients [
3,
4]. Starting from this background, there is clearly a need for further understanding the mechanisms behind ovarian cancer progression, this allowing the development of new treatment strategies to overcome drug resistance and improve patient survival.
Kruppel-like family (KLFs) comprises 17 members of zinc-finger transcription factors that recognize the GT/GC box or CACCC element sequences in gene promoter and enhancer regions [
5]. KLF family members can be divided into three groups on the basis of functional and structural relationships. Members in group 1 and 3 (KLF3, 8, 12 and KLF9, 10, 11, 14 e 16) act as transcriptional repressor, while members in group 2 serve as transcriptional activators (KLF1, 2, 4, 5, 6 and 7); KLF15 and KLF17 contain no defined protein interaction motifs and are more distantly related [
6]. KLFs play critical roles in a multitude of biological processes like proliferation, differentiation, migration, inflammation and pluripotency [
5,
7]. Due to their ability to control proliferation in a variety of cell types (through transcriptional control of cell cycle regulatory components) several KLFs have been implicated in the onset and development of cancer. Specifically, there is much evidence demonstrating that expression of these transcription factors is altered in a number of human cancers, where they can act as tumour suppressor or oncogenes, in a context-dependent way [
8,
9]. Various recent studies have also identified members of KLF family as novel prognostic biomarkers in cancer [
10‐
12]. With respect to ovarian cancer, only sporadic data are available. Findings come mostly from experimental studies suggesting that KLF2, KLF4, KLF6 and KLF11 act as tumour suppressors in ovarian cancer [
13‐
16] while KLF5, KLF8 and KLF9 might have a potential role in disease development and progression [
17‐
19].
Here, we carried out an extensive meta-analysis of publicly available ovarian cancer transcriptome datasets related to HGSOC patients with advanced stage disease, to explore a possible role of KLF family members as prognostic biomarkers. Univariate and multivariate analyses identified KLF7 as an unfavorable prognostic marker for overall survival in this cohort. In vitro experiments demonstrated that KLF7 is able to promote proliferation, migration, invasion and sphere formation in experimental models representative of HGSOC. Finally, herein we reported results from computational analysis and predictive modeling of discovery of putative small molecule inhibitors of the KLF7-DNA interaction interface.
Material and methods
KLF family expression data from publicly available databases
Further in depth investigation was performed on two out of the six datasets. In particular, the transcriptome series GSE26712 (with clinical features matched to the Affymetrix human U133A microarray data) was downloaded from the Gene Expression Omnibus (GEO) database [
22]. A population cohort of 185 patients with late-stage HGSOC was eligible for the analysis (Table
1). As a second external dataset for KLF7, normalized RNA-Seq results of the TCGA-OV patients’ cohort were downloaded from the National Cancer Institute Genomic Data Commons Data Portal (
https://portal.gdc.cancer.gov/) and matched to the corresponding clinical features, as described by the TCGA group [
23]. A population cohort of 266 patients was selected as late-stage HGSOC, as indicated in Table
1. The prognostic effect of the various parameters on OS probabilities was estimated using the Kaplan–Meier method and survival curves were evaluated using the log-rank test. The best cutoff for KLF7 expression values was chosen based on the results of Cutoff Finder [
24] analyses implemented in R v3.6.1 software environment [
21]. Best cutoff value was used to dichotomize patients into low and high KLF7 expression groups. Only variables with
p-value< 0.1 in the univariate analysis were included in multivariate analysis using the Cox proportional hazards model.
Table 1
Clinicopathological characteristics of GSE26712 and TCGA-OV cohorts
Histological type |
HGSOC | 185 | 266 |
Median Age, years (range) | 63 (26–84)a | 59 (31–87)b |
FIGO Stage |
III | 146 (78.9) | 231 (86.8) |
IV | 36 (19.5) | 35 (13.2) |
Not available | 3 (1.6) | – |
Residual tumor after primary surgery |
≤ 1 cm | 90 (48.6) | 169 (63.5) |
> 1 cm | 95 (51.4) | 69 (25.9) |
Not available | – | 28 (10.6) |
Cell cultures and reagents
The human ovarian carcinoma cell lines PEO1 and COV318 were obtained from the European Collection of Cell Cultures (ECACC, Salisbury, UK), while NIH:OVCAR-3 and OV-90 from the American Type Culture Collection (ATCC, Milan, Italy). Susan Horwitz (Albert Einstein Medical College) donated the HEY cell line. PEO1 is an adherent cell line derived from a malignant effusion from the peritoneal ascites of a chemotherapy-treated patient with a poorly differentiated serous adenocarcinoma [
25]. COV318 is a human ovarian epithelial-serous carcinoma cell line established from a peritoneal ascites [
26]. The NIH:OVCAR-3 line was established from the malignant ascites of a chemotherapy-treated patient with progressive adenocarcinoma of the ovary [
27]. OV-90 cells were derived from a chemotherapy-naive grade 3, stage IIIC, malignant papillary serous adenocarcinoma [
28]. HEY cell line was derived from a human ovarian cancer xenograft originally grown from a peritoneal deposit of a patient with moderately differentiated papillary cystadenocarcinoma of the ovary [
29].
PEO1, NIH:OVCAR-3 and HEY were cultured in RPMI 1640 (Roswell Park Memorial Institute Medium) and COV318 in Dulbeccoʼs modified Eagle′s medium (Sigma-Aldrich, St. Louis, MO, USA). OV-90 cells were cultured in 1:1 mixture of MCDB 105 medium (containing a final concentration of 1.5 g/L sodium bicarbonate) and Medium 199 (containing a final concentration of 2.2 g/L sodium bicarbonate). Medium was supplemented with 10% fetal bovine serum (FBS) for PEO1, COV318, NIH:OVCAR-3 and HEY and 15% FBS for OV-90, plus MEM (Minimum Essential Medium) Non-Essential Amino Acid 1%, glutamine 1 mM and kanamycin 1%. Sodium pyruvate 2 mM was also added to PEO1 medium (Sigma-Aldrich). Cells were grown in a fully humidified atmosphere of 5% CO2/95% air, at 37 °C. Cells were routinely tested for mycoplasma (MycoAlert mycoplasma detection kit, LONZA, Rockland, ME, USA) and validated by STR (Short Tandem Repeat) DNA profiling (BMR Genomics srl, Padua, Italy).
KLF7 silencing
Silencing of KLF7 gene expression in OV-90 and PEO1 cells was obtained by transfection with Transfectin (Bio-Rad, CA, USA) and specific siRNAs (siKLF7) (ON-TARGETplus SMARTpool siRNA KLF7, Dharmacon, Lafayette, CO), while a non-targeting siRNA pool was used as control (siC) (ON-TARGETplus Non-targeting Control Pool, Dharmacon, Lafayette, CO). Sequence information related to ON-TARGETplus SMARTpool siRNA KLF7 (pool of 4 siRNA) are provided in Supplementary File
1 (Table
S1A). Silencing efficiency was verified by RT-qPCR using primers reported in Table
S1B. Cells transiently transfected with siKLF7 or siC were used to perform in vitro assays.
Proliferation assay
For proliferation assay, 1.2 × 106 OV-90 cells and 1.5 × 106 PEO1 cells were seeded in corning flask (25 cm2) in complete culture medium and transfected with siKLF7 or siC. After 24 h, 48 h and 72 h from transfection, cells were harvested by trypsinization and viable cells counted with NucleoCounter NC-200 (Chemometec, Lillerød, Denmark). The mean cell proliferation at Time x (Tx) was expressed as average percentage increase relative to T = 0 h (T0). Experiments were performed at least three times.
Transwell migration and invasion assays
For both Transwell migration and invasion assays, siC and siKLF7 OV-90 (10,000 cells) and PEO1 (100,000 cells) were added into the upper chamber of the insert (8 μm pore size; Corning, NY, USA) after 24 h from transfection and further 24 h of serum starvation. For invasion assays, the upper chamber of the insert was precoated with Matrigel (Corning, NY, USA). In both assays, the lower chamber contained medium with 10% FBS as chemoattractant. Cells were incubated for 24 h at 37° in a 5% CO2 atmosphere, inserts washed with PBS (Phosphate-buffered saline) and cells on the top surface of the insert removed with a cotton swab. Cells adhering to the lower surface were fixed with ethanol, stained with crystal violet, washed with H2O and counted under a microscope. Experiments were performed at least three times. The relative migrated or invaded cell numbers were expressed as the percentage of controls.
Real-time quantitative PCR
Real-time qPCR on mRNAs was performed as previously described [
30] using the primers listed in Supplementary File
1 (Table S2). The geometric mean of SNRPD3 and RPLP0 was taken as reference genes, following GeNorm algorithm [
31]; relative quantification of target mRNA was performed according to the ΔΔCt method [
32]. Experiments were performed at least three times.
Western blotting
Western blot analysis of total cell lysates (30 μg) was performed as previously described [
30], using the following antibodies: anti-KLF7 (sc-101,034, Santa Cruz, Santa Cruz, CA, USA, 1:1000); anti-SNAIL (#3879, Cell Signaling Technology Inc., Danvers, MA, 1:1000); anti-ZEB2 (ab25837, Abcam, Cambridge, UK, 1:1000); anti-VIM monoclonal antibody (sc-73,259, Santa Cruz, 1:1000); anti-CD44 (M7082, Dako Agilent Technologies, Santa Clara, CA, USA, 1:500); anti-GAPDH (ab8245, Abcam, 1:5000). Blots were visualized by enhanced chemiluminescence system (Amersham Biosciences, Buckinghamshire, UK) using a Chemidoc imaging system (Bio-Rad). Experiments were performed at least three times. Proteins were densitometrically quantified using Imager ChemiDoc™ XRS+ Software (Bio-Rad, Version 6.0.1.34). All proteins were normalized to GAPDH loading control.
Zymography
Activities of matrix metalloproteinases 2 and 9 (MMP2 and MMP9) in siC and siKFL7 OV-90 and PEO1 cells were determined by gelatin zymography. Forty-eight hours after transfection, cell culture media was changed to serum-free media and cells were incubated for another 24 h. Culture media were mixed with sample buffer and loaded for SDS-PAGE. Gelatinase zymography was performed in 10% SDS polyacrylamide gel in the presence of 0.1% gelatin under non-reducing conditions. Following electrophoresis the gels were washed three times in 2.5% Triton X-100 for 20 min at room temperature to remove SDS. The gels were then incubated at 37 °C overnight in substrate buffer containing 50 mM Tris-HCl and 5 mM CaCl2, 0.15 M NaCl and 1 μM ZnCl2 and stained with 0.5% Coomassie Blue R250 in 50% methanol and 10% glacial acetic acid for 30 min and destained. Experiments were performed at least three times.
Fluorescence microscopy
Twenty-four hours after transfection, siC and siKFL7 OV-90 and PEO1 cells were seeded in 4-well chamber slides in complete growth medium and incubated for further 48hs. Thereafter, cells were fixed in 4% paraformaldehyde for 20 min at 20 °C and then blocked with 5% v/v goat serum in PBS for 1 h. Immunofluorescence staining was obtained using anti-CD44 (M7082, Dako Agilent Technologies, 1:200), following overnight incubation at + 4 °C. After washing, cells were incubated with secondary antibody anti-mouse Alexa Fluor-488 conjugate (1:200) (Thermo Fisher Scientific, Walthman, MA, USA), in the dark for 30 min at 20 °C. Coverslip was mounted onto slides using an antifade mounting reagent containing DAPI. Slides were observed under a fluorescence microscope (Leica Biosystems, Newcastle, UK) using a 40X objective. Appropriate controls were included to evaluate non-specific staining of the secondary antibody or background autofluorescence.
Spheroid cultures
For spheroid cultures, cells were mixed with a solution of VitroGel 3D-RGD (TheWell, Bioscence, NJ, USA)/0.5X PBS at a 3:1 ratio to obtain the final cell concentration of 2 × 105 cells/mL. Hydrogel/cell mixture was added to well plate and then incubated at room temperature for 20 min. After stabilization, complete cell culture medium was added on the top of the hydrogel. The 3D cell cultures were maintained for 10 days and continuously monitored using a microscope DM IL LED (Leica Microsystems). Spheroids were first recovered from hydrogel by VitroGel Cell Recovery Solution (TheWell, Bioscence), according to the manufacturer’s protocol, and then measured (diameters, μm), using Leica Application Suite (LAS) analysis. At least 20 spheroids were analyzed for each data point. Experiments were performed at least three times. For immunofluorescence analysis, spheroids were fixed in 4% paraformaldehyde for 20 min at room temperature, and permeabilized in 0.5% v/v Triton X-100 in PBS for 10 min, prior to be blocked with 20% v/v serum and 0.1% v/v Triton X-100 in PBS for 1 h. Immunofluorescence staining was obtained using anti-VIM (Clone V9, Ready-to-Use, Dako, Agilent Technologies, Santa Clara, CA), following overnight incubation at 4 °C. After washing, cells were incubated with secondary antibody anti-mouse Alexa Fluor-488 conjugate (Thermo Fisher Scientific) and DAPI in the dark for 30 min and 5 min, respectively, at room temperature. Spheroids were recovered from hydrogel by VitroGel Cell Recovery Solution and mounted onto slides that were observed under a fluorescence microscope (Leica Biosystems, Newcastle, UK) using a 100X oil immersion objective. Appropriate controls were included to evaluate non-specific staining of the secondary antibody or background autofluorescence.
Molecular modelling
Homology modeling was performed using Discovery Studio v.19 (Dassault Systems) software suite. The Kruppel-like factor 7 (KLF7) isoform 1 sequence was subjected to BLAST search (
https://blast.ncbi.nlm.nih.gov/Blast.cgi) for template identification. Due to the low alignment coverage (32–28% for the top template structures) we modelled only the three zinc-fingers, encompassing residues 218–302. Clustal Omega (
https://www.ebi.ac.uk/Tools/msa/clustalo) was used for sequence alignment with the template structure, identified in the crystal structure of the zinc finger domain of KLF4 bound to its target DNA (PDB: 2WBS, 76.47% sequence identity [
33]. The alignment was then submitted to the program Modeller [
34] as implemented in Discovery Studio to generate 20 models of the zinc-finger domain of the KLF structure bound to the DNA, using a procedure we already applied [
35]. The models were optimized by a short simulated annealing refinement protocol, and their consistency was evaluated on the basis of the probability density function (PDF) violations provided by the program. During homology modelling the three zinc-fingers atoms were constrained to the template coordinates. The generated structure was validated using Procheck [
36], Verify3D [
37] and Prosa-Web [
38].
Virtual screening procedure
The homology model of KLF7 was prepared for docking using the Protein Preparation Wizard of Maestro (Schrödinger LLC, Release 2017–4, New York, NY, 2017). The imported structure was submitted to the protein preparation process which included correcting mislabeled elements, adding hydrogen atoms, assigning bond orders and performing restrained energy minimization using the OPLS_2005 force field [
39]. To assess the druggability of KLF7, the site recognition software SiteMap (Maestro, Schrödinger, LLC) was run on the model structure after removal of DNA. The algorithm located potential binding sites evaluating cavity size, exposure to solvent, hydrophobic/hydrophilic balance, and hydrogen bonding. The settings used involved the generation of at least 15 site points per reported site. The SiteScore determines the druggability of the selected pocket and the threshold value for recognition as a drug-binding site is 0.80. The binding sites with highest Sitescore were considered for docking studies. For comparison, the binding pockets of KLF7 were also predicted from the metaPocket server (
http://projects.biotec.tu-dresden.de/metapocket) [
40]. The NCI Diversity Set III (2243 molecules) from ZINC database [
41] and the Maybridge HitFinder collection (14,400 compounds representing the drug-like diversity) were used as libraries for the virtual screening experiment. Specific filters based on drug-like properties [
42] and elimination of pan-assay interference compounds (PAINS) [
43] were applied. The final database of small molecules (13,552 compounds) was prepared for docking using the LigPrep tool (Schrödinger 2017–4) which generates a minimized conformation of each ligand, and assigns likely protonation states at pH 7 ± 2 and tautomers. All ligands were docked into the druggable site identified by SiteMap and metaPocket using inner and outer receptor grid boxes of 10 Å and 16 Å, respectively. Flexible ligand docking was performed using the Glide program [
44] (Schrödinger Release 2017–4) and employing the standard precision mode (SP) followed by the extra precision mode (XP) which uses a more optimized scoring function and a more extensive search of ligand conformations. The MM-GBSA rescoring was performed on the 50% top compounds from XP docking and was used to select virtual hits.
Statistical analysis
In vitro data were analyzed for homogeneity of variance using an F test. If the variances were heterogeneous, log or reciprocal transformations were made in an attempt to stabilize the variances, followed by Student’s t-test. If the variances remained heterogeneous, a non-parametric test such as the Mann–Whitney U test was used. P values were two-sided, with p < 0.05 considered as significant. Statistical Analysis was performed using GraphPad Prism 6 (GraphPad Software, Inc. La Jolla, CA, USA) and StatPlus Version v6 (AnalystSoft Inc., Walnut, CA).
Discussion
Here we present results from a bioinformatic meta-analysis of late-stage HGSOC transcriptome data, focused on KLFs. Univariate and multivariate analyses identified KLF2, KLF5 and KLF7 as prognostic factor for overall survival in this cohort.
Results of bioinformatic analysis on KLF2 and KLF5 were actually not in line with previous literature data indicating that KLF2 may act as tumor suppressor in ovarian cancer, while KLF5 may behave as an oncogene [
13,
17]. With regard to KLF2, previous studies reported a reduction in transcript levels in a series of 23 serous ovarian tumor specimens when compared to eight normal ovaries, and in vitro results suggested a role for KLF2 as a tumor repressor [
13]. However, the overall activity of this transcription factor may be much more complex in patients. Indeed, it is known that KLF2 plays critical roles in the activation of various immune cells [
55]. Relevant to our context is the notion that KLF2 is important for Tregs production, immune cells that in malignancies help cancer cells to evade treatment response [
55]. Interestingly, high Treg infiltration in ovarian cancer has been associated with a metastatic phenotype [
56]. Looking at KLF5, to the best of our knowledge, the only data available refer to in vitro experiments on SKOV3 cells and showed a role for KLF5 in driving stemness and drug resistance in ovarian cancer [
17]. However, SKOV3, although commonly used as models for HGSOC, actually do not closely resemble HGSOC [
57]. Therefore, the reliability of findings from Dong and colleagues [
17] in the context of HGSOC needs to be investigated. Importantly, literature data also suggest context-dependent proliferative or antiproliferative functions for KLF5, even in the same cell types [
5,
58].
Bioinformatic analysis highlighted KLF7 as the most significant prognostic gene in HGSOC, among the 17 family members. Besides, preclinical data showed promise for personalized treatment approach and in silico analysis provided reliable information for drug target interaction prediction.
KLF7, known as the ubiquitous Krüppel-like factor, is widely expressed in numerous human tissues at low levels; alternatively spliced products, resulting in multiple transcript variants, have been also identified and these might have distinct regulatory properties [
59]. Over the past years, the protein has been mainly investigated for its role in neuronal morphogenesis and in the pathogenesis of type 2 diabetes [
60,
61]. More recently, however, new evidence have become available suggesting a role for KLF7 in cancer development and progression. Specifically, the protein has been identified as an independent predictor of poor outcome in paediatric acute lymphoblastic leukaemia, lung and gastric cancer [
12,
62,
63]. Accordingly, functional in vitro experiments demonstrated that KLF7 promotes proliferation, migration and invasion in different cancer cell lines [
12,
63‐
65]. Data from the present study are in line with these previously published findings supporting a uniform oncogenic role for KLF7. However, while other members in the family have been broadly characterized for their functions in various cancer-relevant processes, mechanistic insights relative to KLF7 role in cancer pathogenesis are very limited [
9,
66]. Relevant to this aspect, our data showed that, in HGSOC cells, KLF7 plays a prominent role in inducing EMT. Indeed, we have demonstrated, in two different cellular models, that gene silencing correlates with a significant decrease in Vimentin expression. Vimentin is a well-known EMT actor, promoting migration and invasion of different cell types, through different mechanisms of action [
67,
68]. Notably, the concomitant decrease in SNAIL and/or ZEB2 expression (two factors implicated in the regulation of Vimentin expression [
68]) after KLF7 silencing, actually corroborated our results. More importantly, the role of KLF7 in driving EMT by regulating Vimentin levels, was also confirmed in the spheroid models, these latter more closely resembling the in vivo tumour growth situation. Results from the present study also demonstrated that KLF7 silencing reduces expression and activity of MMP2, a gelatinase shown to be secreted and activated in ovarian cancer and closely correlated with invasion and metastasis of cancer cells and poor survival [
69].
Our data also showed in OV-90 cells a reduced expression of CD44, a putative ovarian cancer stem cells markers, this suggesting a possible role for KFL7 in promoting HGSOC progression also by enhancing cancer stem cell properties. Relevant to this issue, recent data not only support evidence of the co-expression of stemness genes and EMT genes in ovarian cancer, but also that EMT activators can actually induce both EMT and stemness properties, suggesting an intriguing model where EMT arises from ovarian cancer stem cells under the right microenvironment and perhaps vice versa [
52]. In line with our results, KLF7 has been recently identified as a functional regulator of human pluripotency, promoting hPSC (human pluripotent stem cells) self-renewal [
70].
Notably, preliminary analysis using the MatInspector software (Genomatix Software GmbH) [
71] allowed us to identify the presence of putative KLF7 binding site in the SNAIL-, ZEB2-, Vimentin- and CD44-promoter regions (data not shown). These results support the hypothesis that KLF7 may regulate these targets at the transcriptional level, although this assumption requires experimental validation.
Overall, the available literature and our data support the idea that KLF7 oncogenic activity may progress via different pathways, mainly affecting the EMT-program and cancer stemness. Therefore, targeting KLF7 by small compounds may open new possibilities for ovarian cancer treatment. Up to now experimental three-dimensional structural information is not available for KLF7, therefore in the current study we firstly generated a computational model structure of the transcription factor that could serve as starting point for structure-based drug design. In silico homology modelling in fact provides a valuable alternative to generate reasonable three-dimensional model structures for drug discovery [
72]. Thereafter, based on the characteristics of the target druggable site we used a docking-based virtual screening to identify putative KLF7 inhibitors. Molecular docking is indeed a powerful tool for identifying the binding modes of ligands to receptors of biomedical relevance [
73,
74]. However, docking scoring functions employed in high-throughput virtual screening lack accuracy in the approximation of binding affinities of protein-ligand complexes [
75]. Hence we filtered the 10 top ranking poses from docking results on the basis of MMG-BSA method, which generally outperforms the scoring functions of docking algorithms [
76]. Results obtained have finally allowed the identification of initial hit compounds for further medicinal chemistry optimization to improve their potency and/or selectivity as potential candidates for clinical therapy. Besides, future studies are already planned in our Department to test the efficacy of the virtual hits in preclinical experimental models.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.