Hepatocellular carcinoma (HCC) is the sixth most common malignancy worldwide and ranks third in mortality among cancer [1
]. HCC frequently arises from long-term inflammatory conditions and is mostly diagnosed at advanced stages resulting in poor prognosis [3
]. The development of HCC is classified by the Barcelona Clinic Liver Cancer (BCLC) staging with a link to therapeutic options [4
]. At very early (BCLC-0) and early (BCLC-A) stages displaying a uni-nodular disease, tumor resection or liver transplantation provide curative therapeutic options [5
]. At advanced stages of HCC with multiple nodules (BCLC-B) including vessel invasion (BCLC-C) and intra- and extrahepatic metastasis (BCLC-C), however, treatment remains solely palliative with five-year overall survival rates below 10% (Cancer Facts & Figures 2016; Atlanta: American Cancer Society).
One crucial event that strongly worsens the patient’s prognosis is the entry of HCC cells into the vasculature and the subsequent dissemination to intra- or extrahepatic metastatic sites [6
]. While intrahepatic metastasis is frequently observed at late HCC stages, extrahepatic metastasis to e.g., the lung occurs in only 10–20% of all HCC patients. For a successful metastatic spread, cancer cells must proceed through several steps to initiate colonization at distal sites [7
]. In this scenario, neoplastic epithelial hepatocytes are considered to undergo reprogramming by transforming into a migratory, mesenchymal-like phenotype during a process called epithelial-mesenchymal transition (EMT) [8
]. Multiple reports provide evidence that EMT-transformed hepatocytes invade the surrounding stromal microenvironment and move towards endothelial cells to cross the barrier and get entry into the circulation for further dissemination. Thereby, invading hepatocytes interact with endothelial cells and transmigrate through the endothelial layer in a process termed intravasation, which is still poorly understood in HCC.
A solid tumor is a hostile place, which already at a diameter of 2 mm, suffers from low oxygen and nutritional supply [10
]. To overcome this consistent survival pressure, cancer cells produce endogenous growth factors that stimulate angiogenesis and enhance tumor cell survival through the formation of new blood vessels [11
]. It is commonly believed that excessive proliferation of endothelial cells in tumor neo-angiogenesis leads to fragile and leaky vessels which cannot properly maintain their barrier function [12
]. However, recent findings suggest that intravasation does not appear randomly. Instead, intravasation seems to be a well-orchestrated event, in which both cancer and endothelial cells mediate the transmigration process. In vitro studies showed that the interaction between cancer cells and endothelial cells may lead to changes in biomechanical properties of the endothelial barrier, thereby reducing its stiffness [13
]. Interestingly, migrating cancer cells actively force their way through the endothelium as it has been shown for cancer cells with overexpression and activation of the epidermal growth factor receptor, which induces a neutrophil-dependent release of vascular endothelial growth factor (VEGF) to modulate endothelial permeability [14
]. Intravasating cancer cells can further collaborate with macrophages which, upon direct physical contact activate RhoA GTPase, leading to the formation of actin-rich degradative protrusions termed invadopodia [15
]. In endothelial cells, the generation of reactive oxygen species (ROS) causes the disruption of adherens junctions via phosphorylation of VE-cadherin and its dissociation from complexes with β-catenin [16
], allowing cancer cells to transmigrate through endothelial cell layers in a paracellular fashion. Furthermore, physical interaction between cancer cells and endothelial cells can activate endothelial myosin light chain (MLC) kinase which then phosphorylates myosin-II regulatory light chain followed by myosin contraction and formation of a structure called an “invasion ring”. This pore-like structure enables direct migration of cancer cells through individual endothelial cells in a transcellular manner [18
Transforming growth factor (TGF)-β is a multifunctional cytokine that can act as a pro-metastatic by inducing EMT and local cell invasion, supporting immune evasion, stimulating angiogenesis and facilitating organotropism of seeding cancer cells to metastasis-competent distal organs [19
]. Recent findings show a particular role of TGF-β in the transmigration of cancer cells during extravasation, i.e., the exit of disseminating tumor cells from blood vessels at distant sites. For instance, TGF-β/Smad stimulates the expression of angiopoietin-like 4 in estrogen-negative breast cancer cells which causes dissociation of cell-to-cell contacts in lung endothelial cells for successful pulmonary metastatic colonization of departing cells [21
]. Another study showed that TGF-β/Smad signaling induces DOCK4—in lung adenocarcinoma cells, allowing transit of the vasculature by Rac1-mediated formation of cancer cell protrusions [22
]. In pulmonary endothelial cells, TGF-β induces endothelial MLC phosphorylation associated with cytoskeletal reorganization by stress fiber formation and disintegration of barrier integrity that is either dependent or independent of the RhoA/Rho-kinase pathway [23
]. A role of TGF-β was further demonstrated in vascular invasion of HCC cells using the chicken chorio-allantoic membrane assay, where HCC cells intravasate via α5β1 integrins that harbor a TGF-β/Smad-activated cytoplasmic domain of β1 integrin [24
]. Yet, blood vessel invasion is still the least studied process during HCC cell dissemination due to the lack of suitable experimental models.
In this study, we established and exploited an in vitro model of hepatocellular intravasation which reflects aspects of active transmigration of invasive HCC cells through hepatic sinusoidal endothelial cells dependent on TGF-β. Using stable isotope labelling with amino acids (SILAC) of the interacting cell types and the resulting “real-time” monitoring of gene expression during transmigration, we identified molecular alterations in both endothelial as well as HCC cells and validated their relevance in HCC patients.
Here we established and examined a homotypic model of hepato-specific transendothelial migration using EMT-transformed hepatocytes (MIM-RT) and liver sinusoidal endothelial cells (mLSECs). The characterization of both cell types has been recently reported [25
]. In this model, (i) endothelial cells at the bottom of a Transwell membrane show polarization towards invading malignant hepatocytes mimicking an intravasation-like process during HCC metastasis [15
] (Figure 1
A); (ii) hepatocytes move across a tightly polarized endothelial monolayer that is separated from invasive hepatocytes by a semi-permeable membrane allowing the study of the “active” process of transmigration; (iii) alterations in protein expression can be “real-time” detected after labelling of both cell types with SILAC and subsequent analysis by mass spectrometry (Figure S1
); and (iv) experimental data is translated to the HCC patient situation using a TCGA database.
MIM-RT cells represent highly invasive, metastatic hepatocytes which are induced to EMT by TGF-β1 and maintained with this phenotype by an autocrine TGF-β1 signalling loop that must be supported over time by the addition of exogenous TGF-β1 [25
]. These cells escape from tumor-suppressive actions by de-differentiation through dissociation of epithelial cell–cell contacts [25
]. They are able to form pulmonary metastatic colonies after orthotopic liver transplantation by showing intravasation, survival in the blood stream and extravasation (data not shown). For endothelial cells, TGF-β is simultaneously pro-apoptotic and pro-angiogenic [28
], which has been explained via a transient apoptotic state mediated by VEGF/VEGF-R2 [29
]. This transient apoptotic state occurs within 6–9 h upon TGF-β treatment, when 20% of cells evince apoptotic hallmarks. Notably, our analysis revealed that MIM-RT cells efficiently transmigrate through the mLSECs barrier showing functional tight junctions within 5 h upon TGF-β1 treatment (Figure 2
), which is earlier than the suggested transient apoptotic state. In addition, no migration has been observed in MIM-RT cells blocked in TGF-β signalling, suggesting that active TGF-β/Smad signalling is essentially required for transmigration in this experimental model.
Our experimental setting using SILAC and mass spectrometry analysis allows us to distinguish between the protein expression in invasive hepatocytes and sinusoidal endothelial cells particularly during transmigration rather than after completion of the process. Endothelial cells treated with TGF-β1 on its own showed a significantly changed expression of 154 proteins (7 up- and 147 downregulated; Figure 3
and Table S1
), while mLSECs exhibited increased alterations in protein expression during transmigration showing 559 differentially expressed proteins (2 up- and 557 downregulated; Figure 3
and Table S3
). The comparison of both mLSEC expression profiles revealed 68 proteins being commonly regulated in both groups. This distinct set of regulated proteins in mLSECs is considered as a result of either a unique cellular program orchestrated by TGF-β or direct cell–cell contacts with malignant hepatocytes. Expression changes in MIM-RT cells upon TGF-β1 treatment revealed 14 up- and 13 downregulated proteins and a similar number of regulated proteins were detected in MIM-RT during transmigration (3 up- and 33 downregulated; Figure 3
, Tables S2 and S4
). From the number of regulated proteins in mLSECs during transmigration, we conclude that endothelial cells might be actively involved in intravasation, thus challenging the role of the endothelium as a passive barrier. Therapies based on targeting the endothelium or combined therapies could be an efficient way to treat HCC patients.
The Cancer Genome Atlas (TCGA) provides valuable RNA expression data from diseased patients which assists scientists to identify novel molecular targets and cancer biomarkers [30
]. Here, we correlated the changes in protein expression obtained from our experimental model with RNA data of HCC patients and asked for those genes with a significant impact on HCC patient’s overall survival [31
]. We could identify 16 genes during transmigration of mLSECs with a significant impact on patient survival. Most notably, peroxiredoxin (PRDX3; Uniprot: P20108) and epoxide hydrolase (EPHX2; Uniprot: P34914) were found to be downregulated at both protein and mRNA levels (Figure 4
). In accordance with this decrease, low expression of PRDX3 or EPHX2 exhibited a strongly reduced overall survival of HCC patients (Figure 5
A,D). In addition, PRDX3 but not EPHX2 was found to be regulated in the same fashion upon TGF-β treatment of mono-cultured mLSECs. Prdx3 belongs to the family of peroxiredoxin proteins which act as peroxidases by the use of electrons provided by thioredoxin [32
]. In contrast to our findings and those in the TCGA database, Prdx3 was shown to be overexpressed in 39–94% of HCC cases [33
]. A recent study further suggests that Prdx3 is an indispensable scavenger of ROS that protects HCC cells against ROS-induced damage and subsequent apoptosis, allowing favourable conditions for cancer cell proliferation and chemoresistance [35
]. In this scenario, it is conceivable that decreased levels of PRDX3 maintains high levels of ROS that are essentially involved in the dissociation of endothelial cells by inducing phosphorylation of VE-cadherin [16
]. Another study showed that CUL4B transgenic mice exhibit enhanced diethylnitrosamine (DEN)-induced hepatocarcinogenesis which is mediated by decreased levels of Prdx3 and increased oxidative liver damage upon DEN treatment [37
]. In addition, a recent report suggests that PRDX3 is associated with tumor suppressor functions in pancreatic adenocarcinoma as its strong expression correlates with smaller tumor size, reduced invasion and negative nodal status [38
]. With respect to Ephx2, which catalyses the hydrolysis of epoxide of xenobiotics to diols, its overabundance has been reported to be a marker of HCC [39
]. Yet understanding the mechanistic role of Ephx2 in HCC progression is a matter for further investigation.
Notably, a common upregulation of transgelin-2 (TAGLN2; Uniprot: Q9WVA) and collectin 12 (COLEC12; Uniprot: Q8K4Q8) at both protein and mRNA levels was observed in mono-cultured MIM-RT and mLSECs cells. In addition, alterations in Tagln2 levels were displayed in transmigrating mLSECs as well, however, Tagln2 showed an opposite regulation, i.e., downregulated protein and upregulated RNA expression (Figure 4
). Further, the actin-binding protein Tagln2 was suggested as a diagnostic biomarker of HCC as it is overexpressed in 69% of patients and was described to be a target of TGF-β/Smad4 in colon cancer cells [40
]. Little is known about COLEC12 in liver pathophysiology. Yet, COLEC12 is part of the innate immune system and highly expressed in umbilical cord vascular endothelial cells. It works as a pattern recognition molecule that can act as a transmembrane receptor or may lead to opsonophagocytosis in its soluble form [42
].Significantly, COLEC12 binds sialyl Lewis X which interacts with E-selectin during extravasation of leukocytes and cancer cells [43
]. Patients with colorectal cancer show activation of the lectin-complement pathway that works similar to collectins, and concomitantly exhibit increased levels of mannose-binding lectin [44
]. Further studies on COLEC12 are needed to understand its role in hepatocarcinogenesis.
Among 16 targets regulated in mLSECs and 2 targets regulated in MIM-RT cells during intravasation (Figure 4
), we could observe an opposite trend where proteins were downregulated and mRNA level upregulated. It should be noted that TCGA uses a comparison between samples from tumor tissue versus non-tumor tissue from healthy individuals and HCC patients. As opposed to this, we compared HCC cells treated with a TGF-β inhibitor as a reference and migrating HCC cells stimulated with TGF-β as the subject of interest. The fact that the detected expression changes at the protein level in intravasating HCC cells do not correspond to mRNA expression obtained from primary HCC cells could also be explained by similar trends observed in circulating tumor cells, which vary in their expression profiles from residing cancer cells [45
Recent studies emphasize differential dynamics of mRNA and protein expression, and suggest that regulations on the protein level might be more important for phenotypic adaptation than transcriptomic changes [46
]. In addition, a substantial limitation of our study must be taken into consideration. Phosphorylation and dephosphorylation of proteins represent important mechanisms in the regulation of cellular response. However, the detection and quantification of phosphorylated proteins is still far from becoming routine in mass spectrometry [47
Our results suggest a key role of TGF-β during endothelial transmigration and highlight the active contribution of endothelial cells in this process. Endothelial cells undergo multiple expression changes which call their role as a passive barrier into question. Directly targeting the disintegration of the endothelium, either alone or as part of a combined therapy, could have a valuable therapeutic potential in preventing intra- and extrahepatic metastasis of HCC cells. Moreover, intravasating cancer cells undergo expression changes which vary from individual cell migration driven by TGF-β and from solid tumor tissue samples, thus representing a unique cellular program. The validation of the phenotypical impact of these identified proteins and their involvement in cellular and molecular mechanisms requires further experimental evaluation by gain- and loss-of-function studies. Discrepancies between TCGA and proteomics data indicate that a combined-omics approaches at all stages of the metastatic cascade provide deeper insights into the molecular mechanisms of cancer development.
4. Materials and Methods
4.1. Cell Culture
The murine p19ARF
-deficient, EMT-transformed MIM-RT hepatocytes expressing green fluorescent protein (GFP) were cultivated in RPMI 1640 plus 10% fetal calf serum (FCS) and continuously supplied with 1 ng/mL TGF-β1 (Peprotech, Rocky Hill, NJ, USA) as described previously [25
]. The murine p19ARF
-deficient liver sinusoidal endothelial cells, termed mLSECs, were propagated on collagen-coated (Collagen Type I-Rat Tail, BD Biosciences, San Jose, CA, USA, Cat.#354236) petri dishes in Dulbecco’s Modified Eagle’s Medium (DMEM) containing 100 µg/mL Endothelial Cell Growth Supplement (ECGS; Biomedical Technologies, Stoughton, MA, USA, Cat.#BT203), 0.2 µg/mL hydrocortisone (Alfa/Aesar, Karlsruhe, Germany, Cat.#A16292), and 50 µg/mL heparin (AppliChem, Darmstadt, Germany, #3U009511) as outlined recently [27
]. For fluorescent labelling, mLSECs cells were lentivirally transmitted with a vector harbouring red fluorescent protein (RFP), resulting in mLSECs-R cells. Human umbilical vein endothelial cells (HUVECs) were grown in Endothelial Cell Medium (ECM; ScienCell, Carlsbad, CA, USA, Cat.#1001) following the manufacturer’s instruction. All cells were grown at 37 °C and 5% CO2
, and were routinely screened for the absence of mycoplasma.
4.2. Cultivation of Cells on Transwell Membrane
Twenty-four mm Transwell permeable supports (Corning, New York, NY, USA) were placed upside-down, coated with collagen (Collagen Type I-Rat Tail, BD Biosciences, Cat.#354236) on the bottom and allowed to air-dry under sterile conditions. 1 mL of mLSEC suspension containing 5 × 105 cells was seeded upside-down onto the bottom of the Transwell membrane and placed into the humidified incubator for 5 h to allow attachment of the cells. After attachment, the Transwell membranes were inserted into 6-well plates containing endothelial growth medium and cells were allowed to proliferate for 4 days. Subsequently, the endothelial medium was changed and 5 × 105 MIM-RT cells were seeded on top of the Transwells in medium containing 2.5 ng/mL TGF-β1 (Peprotech).
4.3. Transmigration Kinetics
MIM-RT cells stably expressing GFP were added on top of the Transwell membranes (Corning, New York, NY, USA) that was covered with a monolayer of RFP-expressing mLSECs (mLSECs-R) on the bottom of the Transwell. The transmigration was performed in the presence of 2.5 ng/mL TGF-β1 for 1, 2, 3, 4, 5 and 24 h. The Transwell filter inserts were subsequently fixed with 4% phosphate-buffered formalin and analyzed by confocal fluorescence microscopy (Zeiss, Oberkochen, Germany) using the tile scan function to obtain high resolution large field images.
4.4. Transendothelial Electrical Resistance
The transendothelial electrical resistance (TEER) of mLSECs monolayers at the bottom of the Transwell membrane was analyzed as described previously [48
]. The growth medium was changed every second day and the resistance was determined daily with a volt-ohm meter. All TEER values were normalized to background values, i.e., Transwell filter in growth medium only.
4.5. Confocal Immunofluoresecence Microscopy
Cells were seeded on collagen-coated Transwell membranes (Corning) and fixed with 4% formaldehyde. After permeabilization with 0.25% Triton-X 100 and blocking, cells were stained with primary antibody against ZO-1 (Zymed Laboratories, South San Francisco, CA, USA) and collagen IV (Santa Cruz, Dallas, TX, USA) at a concentration of 1:75 and 1:100, respectively, and further incubated with secondary antibody (1:200). Images were obtained by confocal immunofluorescence microscopy (Zeiss).
4.6. Stable Isotope Labelling with Amino Acids (SILAC)
mLSECs and MIM-RT cells were cultivated in corresponding SILAC-media following the manufacturer’s instructions (Thermo Fisher Scientific, Waltham, MA, USA). Briefly, MIM-RT cells were cultivated in SILAC-RPMI medium containing l-lysine–2HCl 13C6 (MIM-RT Heavy) and l-lysine–2HCl 13C615N2 (MIM-RT Super-Heavy), respectively. mLSECs were cultivated in SILAC-DMEM medium containing l-arginine–HCl 13C615N4 (mLSECs Heavy) and SILAC-DMEM without heavy isotopes (mLSECs Light).
4.7. Electrophoresis and In-Gel Digestion
The cell lysate samples corresponding to the same group (3 samples TGF-β1-treated and untreated each, 5 h incubation time) were pooled so that each sample in the group contributed with the same amount of total protein. The group samples were heated to 95 °C for 5 min and cooled on ice prior to loading onto NuPAGE 4–12% Bis-Tris gels (Thermo Fisher Scientific). SDS polyacrylamide gelelectrophoresis (SDS-PAGE) was performed according to the manufacturer’s specifications. Proteins were fixed within the polyacrylamide matrix by incubating the entire gel in 5% acetic acid in 1:1 water:methanol for 30 min. After Coomassie staining, the gel slab was rinsed with water and each lane was excised and cut into small pieces. Subsequently, the proteins were in-gel destained (100 mM ammonium bicarbonate/acetonitrile 1:1), reduced (10 mM DTT), alkylated (50 mm iodoacetamide) and finally trypsin-digested overnight at 37 °C. The generated peptides were collected from the gel pieces which were further subjected to a peptide extraction step with an acidic (1.5% formic acid) acetonitrile (66%) solution. Both peptides containing samples were combined and dried in a vacuum centrifuge.
4.8. Mass Spectrometry
The dried peptides were re-dissolved in 0.1% trifluoroacetic acid and loaded on a C18 precolumn (Acclaim; Dionex, Sunnyvale, CA, USA) using an RSLCnano HPLC system (Dionex). Peptides were then eluted with an aqueous-organic gradient, resolved on a C18 column (Acclaim; Dionex, Sunnyvale, CA, USA) with a flow rate of 300 nL/min and electro sprayed into an LTQ Orbitrap XL mass spectrometer (Thermo Scientific). A Triversa Automate (Advion Biosciences, Ithaca, NY, USA) was used as ion source. Each scan cycle consisted of one FTMS full scan and up to seven ITMS dependent MS/MS scans of the seven most intense ions. Dynamic exclusion (30 s), mass width (10 ppm) and monoisotopic precursor selection were enabled. All analyses were performed in positive ion mode. Extracted MS/MS spectra were searched against the Uniprot/Swissprot database using the PEAKS search engine (Bioinformatics Solutions Inc., Waterloo, ON, Canada) accepting common variable modifications and one missed tryptic cleavage. Peptide tolerance was ±10 ppm and MS/MS tolerance was ±0.5 Da. All protein identification experiments were carried out using the corresponding decoy database and a false discovery rate (FDR) of 1%. The SILAC precursor ion quantification of the proteins was performed with the SILAC quantification tool of the PEAKS Studio Software (Bioinformatics Solutions Inc) using a mass error tolerance of ±0.1 Da and a retention time shift of ±0.5 min.
4.9. Comparison of Differentially Expressed Genes with TCGA Data
Orthologous genes between mice and humans were mapped using the Ensembl genome browser in order to obtain the human Ensembl gene IDs and human gene symbols for comparison with TCGA (The Cancer Genome Atlas) gene expression data. The RNASeq version2 data for HCC were downloaded from the TCGA Data Portal (https://tcga-data.nci.nih.gov/tcga/
, July 2015 release). For RNASeqV2, 410 samples with clinical patient information were available (50 normal solid tissue and 360 primary solid tumor samples). Raw counts from gene level data were used for the analysis of differentially expressed genes using the Bioconductors “edgeR” package. Gene counts were normalized using edgeR’s TMM (trimmed mean of M
values). Differential gene expression between tumor and normal samples of RNASeqV2 sample data, respectively, was assessed using edgeR’s exact test.
4.10. Survival Analysis Using Differentially Expressed Genes
The survival analysis was performed in R using the “survival” package for those proteins which were differentially expressed in mouse MS data as well as in the TCGA RNAseq data set using a cut-off for differentially expressed genes with a minimum of 1.5-fold change in same direction, and an adjusted p-value < 0.05. Therefore, RPKM values were calculated from TMM-normalized counts using edgeR’s RPKM function. In order to split the patients into two groups with different survival probabilities exhibiting higher or lower gene expression, two different approaches were used: (1) For each gene, the patients were split into high- or low-expressing groups according to whether the expression of the candidate gene was greater than the median expression of the candidate gene; (2) For each gene, the patients were split into high- or low-expressing groups using maximally selected rank statistics as implemented in the maxstat R package. The statistical significance of differences in overall survival between the two groups was calculated by the log-rank test, and survival curves were plotted using the Kaplan–Meier method.
The data are expressed as mean ± standard deviation. The statistical significance of differences was evaluated using a paired, non-parametric Student’s t-test. Significant differences between experimental groups were * p < 0.05.