The A818–6 system as an in-vitro model for studying the role of the transportome in pancreatic cancer

The cell line A818 was originally isolated from the ascites fluid of a 75-years old female patient suffering from pancreatic adenocarcinoma. A dilution series was previously performed in our Institute and clone number 6 (A818–6) that was able to form 3D hollow spheres when seeded on agar-plated wells was isolated [29]. In the present study, A818–6 cell line was cultured in RPMI Medium 1640 (Gibco, Life Technologies) with 1% Glutamax (Gibco, Life Technologies), 1% Sodium Pyruvate 100 M (Gibco, Life Technologies) and 10% foetal bovine serum (PAN, BIOTECH GmbH). The cells were incubated at 37 °C under 5% CO₂ humid atmosphere and they were grown both as a 2D ML via direct seeding of the cells (3 × 10⁵/ml) in 6-well culture plates or as HS. To create hollow spheres, cells were seeded at a density of 1 × 10⁵ per well in 3 mL media on a culture plate pre-coated with 3.1% agarose (6-well format). Following HS formation, which took 8–10 days, the HS were transferred to normal cell culture plates (6-well format) for further maturation.

Whole cell lysates were prepared using RIPA buffer and analysed via immunoblotting. Antibodies were purchased from: Santa Cruz Biotechnology, Inc., Heidelberg, Germany (anti-Vimentin, anti-LDHA and anti-LDHB); BD transduction laboratories, Heidelberg, Germany (ant-E-cadherin); Sigma Aldrich, Taufkirchen, Germany (anti-β-actin), Cell Signalling Technologies; Frankfurt am Main, Germany (anti-c-myc, anti-HMGA2, anti-p27 and HRP-conjugated anti-mouse and anti-rabbit).

The cell microarray experiment was performed to analyse the changes in gene expression in A818–6 cells grown as a ML or as HS. RNA was extracted using Qiagen RNeasy mini kit (Qiagen, Germany). The amount and purity of RNA was measured by Nanodrop (Thermo Scientific) and the Agilent 2100 bioanalyser system. The experiment was performed using the Agilent technology Sureprint G3 Human GE 8 × 60 K (Agilent, Santa Clara, CA, USA) and it was analysed by imaGenes (Agilent Expression Profiling service, Berlin). Regarding the tissue microarray, the mRNA expression levels were investigated using U133 A/B Affymetrix GeneChip, the detailed methodology was previously reported [31, 32]. The patients gave consents and the ethical committee approved the original study [31], which performed a whole genome expression analysis from which we have only taken a subset for further interpretation. The ethical committee of medical faculty of Christian Albrechts university of Kiel approved the study under the number A110/99. The gene expression database used in this study including the patients’ consents to participate was previously published [31, 32].

Nanostring (nCounter assay) is an ultrasensitive technology that tests the gene expression via the molecular barcodes of the genes of interest that are directly counted with high accuracy. The reaction includes a reporter tag, capture tag, target-specific probes (transportome genes), and target molecules that hybridize to one another. The reporter tag carries a signal and the capture tag contains biotin that interacts accordingly with streptavidin. This reaction does not entail any amplification; it directly counts the already present mRNA copies. Here, the PDAC-relevant transportome genes (n = 101) were investigated in both HS and ML phenotypes of A818–6 cell line using nCounter assay (Supplementary Table 1). Prof Ivana Novak kindly helped choosing the PDAC-relevant transportome genes from the cell microarray’s significantly regulated genes and literature. Messenger RNA from the respective cell lines was isolated via Qiagen RNeasy kit. RNA was set to the concentration of 100 ng of purified total RNA in 30 μL reaction volume. nCounter analysis used 8 negative controls were the mean –in addition to the value of (2) as a standard deviation – were subtracted from samples. The samples were also normalized to the geometric mean of 6 positive controls in addition to 6 different housekeeping genes. The nSolver™ software (Nanostring Technologies) was used for analysis. Since precision of the analysis increases with expression level (counts), all the genes with an mRNA level below 30 counts were excluded from further analysis and a fold change cut-off of ≥1.5 or ≤ − 1.5 was also implemented.

For the cell microarray, HS versus ML comparison results were filtered first by fold change then by p-value (done by T-Test with unequal variance, unpaired) and corrected via Benjamini-Hochberg method. The significant differential expression values between the HS and ML together with the gene annotation were loaded in Microsoft Access Engine 2010 in order to create a database. The gene expression profiles of both phenotypes were compared to one another, where the HS expression was correlated to ML as control. After normalization, a number of 10,080 altered genes were detected in HS in correlation to ML. Later, a fold change of ≥4 or ≤ − 4 was applied and a p-value cut-off of ≤0.05 and finally Benjamini-Hochberg correction of ≤0.05 was applied. The whole-genome gene expression database was then screened for a gene list comprising 838 Transportome genes (Supplementary Table 2). This list was defined according to IUPHAR-DB [33], “Guide to Receptor and Channels” [34], and HUGO Gene Nomenclature Committee [35]. The description of the transportome gene list was previously described [32]. The differential expression values extracted for the 838 transportome genes were further validated by setting an adjusted p-value of ≤0.05 (Benjamini, and Hochberg correction) and a fold change (FC) ≤ − 2 or ≥ 2 for the both the cell and tissue microarrays. For the tissue microarray the statistics was computed using Limma R/Biocondoctor package [36] by applying a linear model as a statistical methodology (Supplementary Table 3) [37].

The differentially regulated genes were profiled using several online freely- available bioinformatics tools. Primarily, the gene lists were compared using WebGestalt (http://​bioinfo.​vanderbilt.​edu/​webgestalt) [38, 39] and a multiple gene list feature enrichment analyser ToppCluster (https://​toppcluster.​cchmc.​org/​) [40]. To investigate EMT features we used dbEMT (http://​dbemt.​bioinfo-minzhao.​org/​) gene resource [41]. Moreover, Venny 2.1 software was used to find the overlapping genes between the gene lists [42]. Finally, some of the predicted analyses were validated via pancreas expression database, PED [43‐46] where the inquired gene lists were compared to a database of previously conducted experiments between PDAC patients versus healthy donors. Furthermore, SPEED [(S)ignaling (P)athway (E)nrichment using (E)xperimental (D)atasets] enrichment algorithm (http://​speed.​sys-bio.​net) was used to specifically investigate the involvement of JAK-STAT, MAPK-PI3K, MAPK-only, TGFβ and TNFα in both A818–6 forms [47]. Also, KEGG mapper (https://​www.​genome.​jp/​kegg/​mapper.​html) was used to identify the involved pathways. Gene set enrichment analysis (GSEA) was used to interpret the microarray data by assigning each gene to its specific biological function and distinct pathways [48]. The analysis is based on Gene Ontology (GO) annotation system [49]. This method implements the hypergeometric distribution to calculate the probabilities that a biological attribute is overrepresented in a gene data set.

Though cellular plasticity enables normal cells to maintain homeostasis, it is also responsible for the capability of epithelial tumour cells to invade and metastasize [50]. It has been proposed that cancer cells, via activation of the epithelial-to-mesenchymal transition (EMT), gain the ability to migrate and invade distant organs. Contrariwise, mesenchymal-to-epithelial transition (MET) must occur in these cells for successful colonization of the new tissue [51]. Fittingly, the HS/ML in vitro system of the A818–6 PDAC cell line represents a unique system to study these transitions. The HS/ML system was presented as a model for studying the differences between a malignant/undifferentiated (ML) and a quasi-normal/differentiated pancreatic (HS) epithelium [27‐29] (Fig. 1a). To validate the differentiation/dedifferentiation status, the protein levels of two markers were compared in both forms. Consistent with the more differentiated, epithelial character of cells growing as HS, Western blot analyses showed clearly higher level of E-cadherin than in ML cells, whereas the expression of the mesenchymal marker vimentin was restricted to ML cells (Fig. 1b).

In agreement with the previously described lower proliferative activity of HS, the negative cell cycle regulator p27 was strongly increased in HS. Also, consistent with the malignant phenotype, proteins enhancing cell proliferation like HMGA2 and c-myc were strongly decreased in HS in comparison to ML cells (Fig. 1c). In addition, protein levels of lactate dehydrogenase protein isoforms A and B (LDHA, LDHB), key enzymes participating in glucose metabolism and overexpressed in pancreatic cancer were highly expressed in ML and severely down regulated in HS (Fig. 1d).

In an attempt to explore the possible molecular drivers for (de)differentiation in these cells, we have performed an automated genome wide gene expression microarray (cell microarray) analysis using mRNA extracts from both A818–6 cell phenotypes. This analysis resulted in a total of 424 significantly regulated genes of which 187 and 237 genes were upregulated in HS and ML, respectively (Supplementary Table 4). Initially, the evaluation was enriched by correlating the differentially regulated genes with Pancreatic Expression Database [PED [43‐46];] as a platform for gene expression studies of the pancreatic cancer (Supplementary Tables 5, 6). PED analyses confirmed the association of a higher number of genes upregulated in ML (n = 26) with PDAC development in comparison to healthy epithelium, thus denoting ML’s more malignant nature in comparison to its HS (n = 7) counterpart. Moreover, consistent with lower proliferation rate of cells growing as a HS, both PED and WebGestalt revealed the association of multiple genes overexpressed in ML in cell cycle and proliferation in PDAC (Fig. 2a), conversely, much less number of genes overexpressed in HS were involved in these processes (Fig. 2b & Supplementary Tables 5, 6, 7 and 8). Additionally, Toppcluster revealed a correlation between the genes which were upregulated in ML not only with cell cycle, growth and division but also with cell motility (Figs. 3 and 4). Similar results were obtained from Reactome (via WebGestalt, Supplementary Tables 7, 8). Obviously there were no predicted pathways or biological processes for HS altered genes in Toppcluster. However, many pathways were affected by the ML dysregulated genes. Interestingly, Toppcluster showed that 22 genes of the upregulated group in HS were all targets of microRNA miR-9 (Fig. 5). Additionally, and in agreement with the more differentiated, more benign phenotype of HS, the cell microarray displayed a number of 9 EMT-related genes which were upregulated in ML as compared to 6 in HS. These results were obtained by blotting the cell microarray results against the genes known to be involved in EMT as published in dbEMT database (Fig. 6). Altogether, and in line with our previous data [27‐29], these molecular differences between HS and ML confirmed that A818–6 cells are able to switch between differentiated/quasi benign state and undifferentiated/malignant state thus providing a good system that allows the analysis of the genetic mechanisms driving and sustaining this process.

As formerly described, the epithelial cellular polarity plays an important role in maintaining cellular volume and acid/base transport in pancreatic ductal cells [15], and the involvement of the transportome in PDAC development was also previously reported [17, 23, 52‐54]. Noteworthy, the 3D orientation of HS was previously shown to allow the A818–6 cells to regain relatively normal epithelial polarity which was in turn lost when the HS were disrupted and ML reformed and vice versa [27, 29]. Earlier, mucin− 1, a marker of the apical border of the normal pancreatic epithelial cells, and β-catenin a basolateral marker were utilized to stain HS and ML cells. It was shown that HS regained an inverted polarity where HS displayed an outer apical border and an inner basolateral border stained with mucin− 1 and β-catenin, respectively. HS showed also positive staining of the differentiation marker carbonic anhydrase II (CA II) and secreted carcinoembryonic antigen-related cell adhesion molecules (CEACAMs) into the supernatant around them, both of which denote the polarization of HS in comparison to ML [27, 29]. Collectively, the HS/ML in vitro system could provide a distinctive insight into the involvement of the transportome during the differentiation/dedifferentiation process of A818–6 cells.

In order to study whether the changes in differentiation status between HS and ML correlate with changes in the expression level of transportome genes, two different approaches were implemented. Firstly, the above-mentioned cell microarray dataset was screened for 840 known transportome genes and additionally, a Nanostring nCounter analysis of HS/ML cells was performed. In the cell microarray, among the differentially regulated genes, we found 126 transportome genes; 68 of which were overexpressed and 58 down-regulated in HS compared to ML (Supplementary Table 9). In addition, a custom-made Nanostring nCounter array screening for the important and significantly altered PDAC-relevant expression of transportome genes was performed. The nCounter array included a number of 101 PDAC-related transportome genes, which were identified from both the cell microarray and literature. The final nCounter analysis of the transportome gene expression included 25 up- and 22 down-regulated genes in HS compared to ML (Table 1). From both, the cell microarray and the nCounter assay a number of consistently upregulated (n = 11) as well down-regulated (n = 15) transportome genes were found in HS when compared with its corresponding ML (Table 2). Thus, both approaches underline the involvement of transportome in this (de)differentiation model.

This transportome gene list was uploaded to Toppcluster and a comparison was created between the up and down-regulated genes in terms of pathways, biological processes, molecular functions and the possible miRNA regulators of those genes (Figs. 7, 8 and 9). Regarding the predicted pathways, only one pathway was commonly altered with both the up- and down-regulated genes in HS cells (SLC-mediated transmembrane transport). Specifically, potassium channel genes, genes related to amino acid, oligopeptide transport as well as transport of inorganic cations/anions were enriched in HS in comparison to ML. Conversely, gap junction assembly, metabolism, membrane trafficking and vesicle-mediated signalling pathways were upregulated in ML in comparison to HS. Among the biological processes influenced by the upregulated genes in HS were several ion transport proteins (sodium, potassium, metal) whereas, the down-regulated genes were associated with cell-cell junction assembly, lactate biosynthetic processes and anion transport. Furthermore, the overexpressed transportome genes in HS were involved in molecular functions including; voltage-gated ion channel activity, substrate specific channel activity and antiporter activities among others. On the contrary, the down-regulated transportome genes are those involved in the control of other molecular functions as wide-pore channel activity, symporter activity and lactate dehydrogenase activity. Finally, both the nCounter and cell microarray analyses confirmed the higher expression levels of both metabolic enzymes, LDHA and LDHB, in ML cells compared to HS. Worth mentioning, though Toppcluster is a valuable tool that allows the comparison between different gene groups, it only lists all the possible functions/pathways shared by the investigated genes and it does not specify which is more relevant to the current gene clusters.

To further authenticate our hypothesis that the HS/ML system is a valid system to study the role of transportome in the de-differentiation process in PDAC, we compared the above described results with our recently reported transportome analysis in PDAC patients’ tissues [31, 32]. Previously, we compared the gene expression profile of 19 PDAC patients’ malignant pancreatic microdissected tissue specimens (tumour epithelium) with 13 different normal pancreatic epithelial specimens (normal epithelium) that were performed using a 133 A/B Affymetrix microarray technology (tissue microarray). Among the 616 significantly altered genes, 63 were transportome genes of which genes were overexpressed and 45 suppressed in PDAC. The dysregulated genes were not only involved in pH regulation, the control of the cellular volume, secretion and polarity but also in cellular differentiation [32]. In the current study, we compared these tissue microarray results with our data from both the cell microarray and the nCounter array. This comparison confirmed the differential expression of numerous transportome genes in PDAC and ML in relation to normal epithelium and HS, respectively (Table 3). This comparison identified a number of 3 upregulated genes (GJB2, GJB5 and SLC38A6) in PDAC and ML, and another 3 transportome genes (KCNQ1, SLC4A4 and TRPV6) which were correlated with normal pancreatic epithelium and HS. To predict the role of these differentially regulated transportome genes, a bioinformatics analysis was performed using publicly available open source databases. This part of the analysis showed the enrichment of the pancreatic secretion (KCNQ1, SLC4A4), the epithelial fluid secretion (TRPV6) and the maintenance of bicarbonate transport (SLC4A4) in the normal pancreatic epithelium and HS. Additionally, it was found that some PDAC and ML-related transportome genes were involved in metabolic transport (SLC38A6) and intercellular communication and cell adhesion (GBJ2, GJB5). Correspondingly, the transportome genes in normal pancreatic epithelium and HS were functionally related to transepithelial ion and fluid secretion, where bicarbonate and/or chloride transport were enriched via SLC4A4, transepithelial transport and setting membrane potential via KCNQ1, cAMP and intracellular calcium signalling via TRPV6. Although the current study analysed the transportome gene expression profiling, it did not investigate the overall functions of ion channels. All in all, this study introduces the HS/ML model as a valid system to study the differential regulation of transportome genes in both differentiation and PDAC development as well as during this transition.