Background
Ovarian cancer is the deadliest gynaecological cancer in women with the development of chemotherapeutic drug resistance being the major obstacle to successful treatment. Recent data suggests that the extracellular matrix (ECM) can directly modulate cell sensitivity to both platinum- and taxane-based drug treatment therapies [
1‐
4]. Also, as the ECM regulates other key aspects of cell behaviour including growth control, cell migration, survival, and gene expression [
5], it represents an important target in designing treatment therapies.
We have shown that the secreted extracellular matrix protein, TGFBI (transforming growth factor beta induced), is a critical component of the ovarian cancer tumor microenvironment that sensitizes cells to paclitaxel-induced cell death by stabilizing microtubules via integrin-mediated activation of focal adhesion kinase (FAK) and the Rho family GTPase RhoA [
1]. TGFBI has been suggested to have both tumor suppressor and tumor promoting properties, depending on the cancer of origin [
6]. Specifically, TGFBI has been shown to be underexpressed in breast [
7], ovarian, and lung cancer [
8]; and overexpressed in clear cell renal carcinoma [
9], pancreatic cancer [
10], and colorectal cancer [
11]. In addition, mice lacking
Tgfbi show spontaneous tumor formation, further supporting a potential tumor suppressor function [
12]. Interestingly, loss of TGFBI expression is associated with centrosome duplication and chromosomal instability, both causal factors associated with carcinogenesis and drug resistant phenotypes [
1,
12,
13]. However, the mechanism by which extracellular TGFBI mediates these effects is unclear.
Structurally, TGFBI contains an amino-terminal signal peptide sequence necessary for secretion into the extracellular environment, a cysteine-rich EMI domain similar to regions found in proteins of the EMILIN family, along with four highly conserved fasciclin I (FAS I) domains and a carboxy-terminal Arginine-Glycine-Aspartic Acid (RGD) motif. Various heterodimeric integrin receptor combinations mediate interactions with TGFBI and its RGD and FAS I domains [
14‐
16]. Specifically, corneal epithelial cell adhesion to TGFBI is predominantly mediated by the α3β1 integrin heterodimer [
14], while in endothelial cells the αvβ3 integrin heterodimer is dominant [
15]. Furthermore, TGFBI can bind many ECM proteins including Collagen type I, II, IV, and VI [
17‐
19], fibronectin [
20], periostin [
21], laminin [
18], as well as the proteoglycans biglycan and decorin [
22]. The FAS domains are highly conserved and three human proteins, TGFBI, periostin, and stabilin, contain these motifs [
23].
Periostin is a paralogue of TGFBI and is also a TGFβ1-inducible secreted protein. Both TGFBI and periostin have been implicated in ovarian cancer [
1,
24]. Periostin is secreted by ovarian cancer, similar to TGFBI, and promotes integrin-mediated cell motility [
24]. However, although they have similar domain structure, very little is known as to whether their function is complementary or antagonistic. Periostin shares with TGFBI an EMI domain and four highly conserved FAS I domains. However, it differs in having an extended carboxy-terminus, which does not contain the RGD motif [
25,
26]. Interestingly, recent data suggests periostin and TGFBI interact through their amino-terminal EMI domains and may have a proactive role in the pathogenesis of corneal dystrophy [
21]. Additionally, periostin contributes to metastasis in both pancreatic and colon cancer due to augmentation of PI3K/Akt signaling [
27,
28] and it has been suggested to be a critical component of metastatic colonization [
29]. Therefore, evaluating the mechanism of TGFBI and periostin function in ovarian cancer cells may shed light on their relationship and function during ovarian carcinogenesis.
Although TGFBI has been shown to signal through multiple integrin heterodimeric receptors, the predominant signaling pathways and the relationship to other ECM components in ovarian cancer is unknown. It has been shown that fibronectin-integrin signaling could protect breast cancer cells against paclitaxel-induced cell death [
30]. Since this contrasts to the function of TGFBI in ovarian cancer [
1], there lacks a clear understanding of the differential signaling that occurs upon engagement of the cell surface with various ECM components. Importantly, previous reports have suggested that cross-talk between different integrin receptors can modulate the response to their respective ECM ligand [
31‐
33].
To understand the function of TGFBI in ovarian cancer and the role of TGFBI-integrin interactions in mediating paclitaxel sensitivity, we therefore delineated the primary domains of TGFBI that are important in mediating the interaction with ovarian cancer cells and the key receptors necessary for this process.
Methods
Antibodies and reagents
Paclitaxel was purchased from Sigma-Aldrich, cat. no. T7402 (Dorset, UK). The GRGDSP peptide was purchased from Merck Chemicals Ltd. (Nottinghamshire, UK) and the ERGDEL peptide was custom produced by Sigma Genosys (Haverhill, UK). Human plasma fibronectin was purchased from Millipore (Watford, UK) and human vitronectin was purchased from R&D systems Europe Ltd. (Abingdon, UK). Affinity purified polyclonal antibody directed against TGFBI was produced by immunizing rabbits with a C-terminal peptide of human TGFBI (aa 498–683). All antibody production was performed in collaboration with Cambridge Research Biochemicals (Cleveland, UK). TGFBI polyclonal antiserum was a kind gift from Dr. Ching Yuan (University of Minnesota, Minnesota, USA). Alpha-tubulin antibody was purchased from Sigma-Aldrich. The periostin polyclonal antibody was purchased from BioVendor Laboratory Medicine Inc. (Czech Republic) and the periostin monoclonal antibody (clone 345613) from R&D Systems Europe Ltd. Akt phospho-S473 and pan-Akt polyclonal antibodies were purchased from Cell Signaling. Fibronectin, ILK, and FAK phospho-Y397 monoclonal antibodies were purchased from BD Biosciences (Oxford Science Park, Oxford, UK). Alexa Fluor 568-phalloidin was purchased from Invitrogen (Inchinnan Business Park, Paisley, UK). β3 integrin polyclonal antibody was purchased from Santa Cruz (Santa Cruz, California) and β1 integrin polyclonal antibody was purchased from Cambridge Bioscience (Cambridge, UK). Integrin blocking antibodies against β1 integrin (clone P5D2), αvβ3 (clone LM609), and αvβ5 (clone P1F6) were purchased from Millipore. Syndecan-1 monoclonal antibody was purchased from Serotec (Oxford, UK) and Syndecan-4 polyclonal antibody was purchased from R&D Systems Europe Ltd.
Cell culture
The ovarian cancer SKOV3 cell line was maintained in RPMI media supplemented with 10% (v/v) heat-inactivated FBS, 50 units/ml penicillin, and 50 μg/ml streptomycin. The ovarian cancer PEO1 cell line was maintained in DMEM/F12 (50:50) supplemented with 10% (v/v) heat inactivated FBS, 50 units/ml penicillin, and 50 μg/ml streptomycin. NIH 3 T3 cells were maintained in DMEM supplemented with 10% (v/v) heat inactivated FBS, 50 units/ml penicillin, and 50 μg/ml streptomycin. All cell lines were verified by short tandem repeat genotyping. Lentivirus expressing individual shRNA targeted against β1 integrin, β3 integrin, TGFBI, and fibronectin were purchased from Sigma-Aldrich Mission® shRNA library. Cells were infected at an MOI of 10 and subsequently stable pools of cells were selected in Puromycin. Syndecan-1 siGenome SMARTpool® siRNA, syndecan-4 siGenome SMARTpool®, β1 integrin ON-TARGETplus® pool, β3 integrin ON-TARGETplus® pool, and siGenome® non-target control #2 siRNA were purchased from Perbio (Northumberland, UK). siRNA transfections were performed using Lipofectamine 2000 (Invitrogen, Inchinnan Business Park, Paisley, UK) according to manufacturer’s instructions.
Western blot
Cell lysates were harvested in RIPA buffer (1% Triton X-100, 0.1% SDS, 1% DOC, 10 mM Tris–HCl pH 7.4, 150 mM NaCl, 5 mM EDTA, 10 μg/ml leupeptin, and 1 mM Na3VO4). Lysates were cleared by centrifugation at 14,000xg at 4°C. Protein content was quantified by the BioRad Dc Protein Assay (Hertfordshire, UK). Following the addition of 2X SDS-sample buffer and boiling, samples were loaded onto 7.5-10% SDS-PAGE gels and transferred to PVDF (Fisher Scientific UK, Leicestershire, UK). Membranes were blocked with either 5% non-fat dry milk or 3% BSA, probed with the indicated antibodies, and visualized following the addition of HRP conjugated secondary antibodies (Dako UK Ltd., Cambridgeshire, UK) and incubation with enhanced chemiluminescence (GE Healthcare UK Ltd., Buckinghamshire, UK). Western blots were either directly reprobed or parallel Western blots were performed on the same cell lysates for alpha-tubulin loading controls.
Cell surface biotinylation
Cells were washed in cold PBS, incubated 30 minutes with 0.2 mg/ml EZ-Link Sulfo-NHS-Biotin (Fisher Scientific UK Limited, Loughborough, UK) on ice. After 30 minutes, cells were washed two times in cold PBS and lysed in immunoprecipitation buffer (200 mM NaCl, 75 mM Tris pH7.4, 7.5 mM EDTA, 7.5 mM EGTA, 1.5% Triton-X 100 and 0.7% NP-40 with protease inhibitor cocktail). Lysate was cleared at 14,000 xg for 10 minutes at 4°C and the resulting supernatant was incubated with anti-β3 integrin/CD61 antibody (clone VI-PL2; BD Biosciences) overnight at 4°C, followed by addition of 15 μL of 50 mg/ml Protein A-sepharose and incubation at 4°C for 1 hour. Beads were washed four times in lysis buffer, followed by addition of 2X SSB, and samples were run under non-reducing conditions on 7.5% SDS-PAGE. Western blot analysis was performed with HRP-conjugated streptavidin (Fisher Scientific UK Limited).
Recombinant protein production
The pET27 TGFBI plasmid utilized for recombinant protein production in bacteria was a kind gift from Dr. Ching Yuan (University of Minnesota, Minnesota, USA). Both recombinant TGFBI and periostin were engineered with a carboxy-terminal His-tag. Periostin cDNA was a kind gift from Dr. Nick Lemoine (Barts, London, UK). Periostin cDNA, lacking the amino-terminal signal peptide, was cloned into the pET27 vector for subsequent production of bacterial expressed recombinant protein. Deletion constructs were made by PCR addition of NheI and NdeI unique restriction sites for subsequent cloning into the pET27 vector. Site-directed mutagenesis was performed on pET27 TGFBI to produce an amino acid RGD to RAE substitution using the oligonucleotide primer 5’-agacctcaggaaagagcggaggaacttgcagactctg-3’ and an amino acid YH to SR substitution using the oligonucleotide primer 5’-gaacttgccaacatcctgaaagccgccattggtgatgaaatcctgg-3’. All constructs were verified by sequencing. All recombinant proteins were produced in Rosetta BL21 (DE3) E.coli (Merck, Nottingham, UK) and either purified from an insoluble fraction [
34] for full-length TGFBI and periostin or from a soluble fraction [
35] utilizing Ni-NTA agarose beads (Merck). Refolding of purified full-length TGFBI and periostin was performed by buffer exchange through a PD10 Desalting Column (GE Healthcare, Buckinghamshire, UK) into 10 mM Tris–HCl pH 7.4, 0.5 M Arginine-HCl, and 10% Glycerol solution.
Adhesion assay
96-well or 24-well tissue culture treated plastic dishes were incubated overnight at 37°C with 20 μg/ml of recombinant protein diluted in PBS. Dishes were subsequently washed with PBS, blocked with 3% BSA for 1 hour at 37°C, followed by washing with PBS and SF media containing 0.1% BSA. Cells were collected, washed once with growth media, washed twice with serum-free media containing 0.1% BSA, and incubated in serum-free media containing 0.1% BSA for 1 hour at 37°C in suspension. Cells were plated on uncoated, poly-L-lysine, or matrix-coated dishes for indicated time periods. Adherent cells were subsequently washed once with PBS, fixed in methanol, and stained with Giemsa (Fisher Scientific UK). Stain was eluted with 10% acetic acid and an absorbance reading was obtained at 540 nm. To account for non-specific adhesion, values from uncoated wells were subtracted from all experimental values. All experiments were performed in triplicate. Due to technical variability of raw values between replicate experiments, data were represented as percent adhesion to control. All statistics were performed in GraphPad Prism® using either one- or two-way Anova along with Bonferroni’s multiple comparison test when appropriate. Error bars represent standard deviation. Bright-field images were taken with a DS-Fi1 CCD camera and processed with Adobe Photoshop® CS2.
Apoptosis and viability assays
For apoptosis analysis, cells plated on uncoated tissue culture dishes were treated with varied concentrations of Paclitaxel or DMSO vehicle control diluted in complete growth media. Following incubation at 37°C for 24 hours, both adherent and floating cells were harvested and washed in cold PBS. The TACS Annexin-V Apoptosis kit (R&D systems Europe Ltd.) was performed according to manufacturer’s instructions. 10,000 cell events were recorded on a BD FACS Calibur and data was analyzed with FlowJo 8.8.4 flow cytometry analysis software (Tree Star Inc., Ashland, Oregon, USA). Results are represented as the percentage of early apoptotic events (Annexin-V positive, propidium iodide negative) compared to total events and error bars represent standard deviation. For cell viability analysis, cells were transiently transfected with siRNA prior to replating on white 96-well tissue culture dishes. Cells were treated with vehicle (DMSO) or increasing concentrations of Paclitaxel for 48 hours prior to administration of the Cell Titre Glo® Luminescent cell viability reagent as per manufacturer’s instructions (Promega UK, Southampton, UK). Results were normalized to a DMSO treated control and the experiment was performed in triplicate. Error bars represent standard deviation and a one-way anova along with a Bonferroni multiple comparison test was performed.
Immunofluorescence microscopy
Cells were fixed in 3.7% Formaldehyde in PBS for 8 minutes and permeabilized with 0.2% Triton X-100 for 2 minutes. Fixed cells were incubated with primary antibody in TBS containing 1% BSA at 37°C for 1.5 hours, washed in TBS, incubated with either Alexa Fluor® 488 or 568 secondary antibodies (Invitrogen) in TBS containing 1% BSA, washed in TBS, and mounted in Fluorsave (Calbiochem). For live cell immunostaining with anti-αvβ3 integrin antibody (clone LM609), cells were first washed into CO2-independent medium supplemented with 2% FBS, next incubated in primary antibody for 20 minutes, followed by incubation with Alexa Fluor 488 antibody for 20 minutes. Cells were washed in PBS and fixed for 2 minutes in ice-cold methanol. Nuclei were stained with Hoechst and coverslips were mounted in Fluorsave. Images were captured on a Leica Tandem SP5 confocal microscope (Leica-Microsystems, Milton Keynes, UK) or a Zeiss Axioplan epifluorescence microscope equipped with a Hamamatsu ORCA-R2 CCD camera driven by Simple PCI software (Carl Zeiss MicroImaging Inc.) and processed with Adobe Photoshop® CS2. Image analysis of cell surface integrin immunostaining was performed using ImageJ software. Briefly, the integrated intensity of integrin immunostaining was calculated and due to technical variability between replicate experiments, values were normalized to control and represented as the percent change in fluorescence intensity. The data represents at least 100 individual cells taken from two independent experiments.
Bright-field time lapse video microscopy was performed using a Nikon TE2000 PFS microscope equipped with a DS-Fi1 CCD camera. Cells were plated on a matrix-coated ibidi 35 mm μ-dish, low (Thistle Scientific, Glasgow, UK) and images were acquired using a 10X objective every 2 minutes for 6 hours using NIS elements software (Nikon Instruments Europe) in a temperature controlled and 5% CO2 maintained environment.
Discussion
TGFBI is a multifunctional protein implicated in a variety of physiological processes including cell growth, wound healing, inflammation, and developmental morphogenesis [
6]. However, its dysregulation can lead to the pathogenesis of a variety of diseases, including cancer [
6]. More specifically, recent evidence suggests that TGFBI is dysregulated in ovarian cancer and its expression level may influence cancer response to the chemotherapeutic agent paclitaxel [
1]. In addition, extracellular TGFBI increases the motility and invasiveness of ovarian cancer cells and stimulates a peritoneal cell interaction [
47]. Therefore, we sought to understand the molecular mechanisms that influence TGFBI function and its interrelationship with other ECM components known to be present in the tumor microenvironment in order to better determine potential therapeutic targets and indicators of treatment response.
In ovarian cancer cells, which express both the β1 and β3 integrin subunits, TGFBI preferentially interacts with cells through an αvβ3 integrin-mediated mechanism. This is in contrast to the predominant β1 integrin-mediated mechanism elicited by fibronectin and periostin (Figure
2C). Although this contradicts recent evidence that suggests periostin primarily interacts with ovarian cancer cells via an αvβ3 integrin-dependent mechanism [
24], it also suggests a delicate balance may exist between different integrin receptors on the cell surface that dictate specificity to the ECM. This is further supported by our data showing that loss of β1 integrin in SKOV3 cells increases adhesion to rTGFBI, but not to fibronectin or periostin, in an αvβ3 integrin dependent manner (Figure
3; Additional file
5: Figure S3).
Additionally, integrin cross-talk may play a major role in the diversity seen within different cell systems and within different tumor types that have varying integrin subunit expression profiles. For example, divergent signaling through β1 and β3 integrins has major impacts on downstream Rho GTPase signaling, which may subsequently result in contrasting effects on cell adhesion and migration [
48]. In addition, distinct β1 and β3 integrin expression along with oncogene expression, such as oncogenic Src, may differentially influence chemosensitivity [
49]. Our data supports this notion as suppression of β1 integrin expression stimulates a TGFBI-β3 integrin-mediated adhesion response (Figure
3). Although our data suggests an increased cell surface expression of the αvβ3 integrin heterodimer following suppression of β1 integrin expression (Figure
3E
4C), there likely also exists cross-talk between downstream signaling complexes associated with the activation of different integrin receptors. Furthermore, our data indicate that in ovarian cancer cells the loss of β3 integrin expression partially induces a paclitaxel-resistant phenotype, while loss of β1 integrin expression leads to a potential paclitaxel-sensitive phenotype. With regards to integrin receptor cross-talk, it has been previously reported that forced expression of α5β1 integrin negatively regulates αvβ3 integrin function in Chinese hamster ovary cells [
31]. In addition, it has been shown, with regards to the α3 noncollagenous domain of collagen IV, that the α3β1 heterodimer can modulate αvβ3-mediated cell adhesion [
32]. Lastly, β1 integrin activation negatively effects αvβ3 activation via activation of PKA and inhibition of PP1 activity [
33]. Since β3 integrin expression has been suggested to be a potential prognostic biomarker in ovarian cancer [
50], it will be important to delineate the specific β3 integrin-dependent signals and determine their impact on ovarian carcinogenesis and chemotherapy response.
Attachment of ovarian cancer to the mesothelium and its associated ECM, which lines the peritoneal cavity, may not be exclusively integrin-mediated [
38,
51]. Therefore, other integrin-independent cell-ECM receptors may be involved in mediating adhesion in the tumor microenvironment. The primary co-receptor family involved in cell-ECM adhesion, which synergizes with integrin engagement to mediate a complete cellular response, is the Syndecan family of receptors [
42]. For αvβ3 adhesion, Syndecan-1 is the predominant co-receptor that mediates this process [
52]. It is important to note that although Syndecan-1 expression is absent in the normal ovary, it is upregulated in ovarian cancer as well as in tumor stroma [
53]. Our data suggests that the loss of Syndecan-1 cooperates with the loss of β1 integrin to stimulate adhesion to TGFBI and is therefore dispensable for TGFBI adhesion. Thus, for ovarian cancer cells, it appears neither Syndecan-1 nor Syndecan-4 is necessary for adhesion to TGFBI, nor does the loss of Syndecan-4 synergize with αvβ3 to stimulate adhesion to TGFBI. In contrast, for periostin, loss of both Syndecan-1 and Syndecan-4 negatively affects ovarian cancer cell adhesion, which supports the notion that periostin utilizes a distinct mode of cellular interaction.
Previous literature has attempted to dissect the specific domains and motifs within TGFBI that are critical for its interactions with the cell surface. Since these results seem to be cell-type and system specific, we attempted to extend a similar analysis with respect to ovarian cancer cells, including the comparison to its paralogue, periostin, which has been suggested to have a proactive role in ovarian cancer migration [
24]. Importantly, unlike TGFBI, the periostin carboxy-terminus, which lacks an RGD motif, is unable to support adhesion. This suggests that the specificity of TGFBI and periostin for their respective cell surface receptors is partially dictated by differences in this region.
The function of the TGFBI-derived RGD appears to vary depending on the cell-based system and the primary integrin heterodimeric receptor. Initially, it was suggested that the carboxy-terminus underwent proteolytic processing resulting in loss of the RGD motif [
54,
55]. In addition, this carboxy-terminal released peptide induced apoptosis in CHO cells is dependent on an intact RGD motif [
55]. However,
in vitro biochemical analysis suggested the potential carboxy-terminal cleavage site on human TGFBI lies downstream of the RGD motif suggesting it is not cleaved from the full-length protein [
56]. Therefore, a mature TGFBI protein that contains the RGD motif is likely functional in a biological context. This is supported by recent data that suggests the RGD motif of TGFBI is necessary for promoting extravasation of metastatic colon cancer cells [
16]. Our results suggest that the RGD motif of full-length TGFBI is necessary, but not sufficient, for ovarian cancer cell adhesion, thus indicating it may cooperate with flanking residues or other motifs, potentially present within the fourth Fasciclin I domain to mediate this process. Importantly, we found that the TGFBI derived RGD peptide (ERGDEL) was unable to competitively inhibit SKOV3 adhesion to rTGFBI, suggesting its use as a therapeutic agent to inhibit TGFBI function may depend on the cellular context.
Conclusions
Ovarian cancer is a complex disease where the tumor microenvironment plays an active role in the dissemination of the disease and influences the response to chemotherapy. Since it has previously been shown that fibronectin-mediated β1 integrin signaling represses paclitaxel-induced cell death [
30], specific ECM-receptor pathways may be important in differentially modulating chemotherapeutic response. This is confirmed by our data which showed suppression of fibronectin expression sensitizes cells to paclitaxel-induced death, while suppression of TGFBI leads to a resistant phenotype (Figure
7) [
1]. This is further supported by recent data in non-small cell lung cancer showing that TGFBI-mediated induction of apoptosis in response to chemotherapy requires the αvβ3 integrin receptor [
57]. In addition, distinct ECM-integrin receptor engagement may trigger intracellular cues that stabilize the microtubule cytoskeleton, which has been suggested to be a mechanism to enhance the cytotoxicity of paclitaxel [
58]. Therefore, it will be crucial in a clinical context to define the relationship between discrete integrin heterodimers and their respective extracellular binding partners in order to understand the intracellular signaling pathways that occur. Importantly, further characterization of the differential signaling downstream of TGFBI-β3 integrin engagement in comparison to other ECM-receptor mediated pathways will be needed to identify distinct mechanisms of chemotherapeutic response.
Competing interests
The authors declare they have no competing interests.
Authors’ contributions
DAT designed and performed experiments, analysed the data, and wrote the manuscript. JT performed the cell titre glo cell viability assay. JDB conceived of the study and wrote the manuscript. All authors read and approved the final manuscript.