Arsenic Exposure Perturbs Epithelial-Mesenchymal Cell Transition and Gene Expression In a Collagen Gel Assay

Abstract

Arsenic is a naturally occurring metalloid and environmental contaminant. Arsenic exposure in drinking water is reported to cause cancer of the liver, kidneys, lung, bladder, and skin as well as birth defects, including neural tube, facial, and vasculogenic defects. The early embryonic period most sensitive to arsenic includes a variety of cellular processes. One key cellular process is epithelial-mesenchymal transition (EMT) where epithelial sheets develop into three-dimensional structures. An embryonic prototype of EMT is found in the atrioventricular (AV) canal of the developing heart, where endothelia differentiate to form heart valves. Effects of arsenic on this cellular process were examined by collagen gel invasion assay (EMT assay) using explanted AV canals from chicken embryo hearts. AV canals treated with 12.5–500 ppb arsenic showed a loss of mesenchyme at 12.5 ppb, and mesenchyme formation was completely inhibited at 500 ppb. Altered gene expression in arsenic-treated explants was investigated by microarray analysis. Genes whose expression was altered consistently at exposure levels of 10, 25, and 100 ppb were identified, and results showed that 25 ppb in vitro was particularly effective. Three hundred and eighty two genes were significantly altered at this exposure level. Cytoscape analysis of the microarray data using the chicken interactome identified four clusters of altered genes based on published relationships and pathways. This analysis identified cytoskeleton and cell adhesion–related genes whose disruption is consistent with an altered ability to undergo EMT. These studies show that EMT is sensitive to arsenic and that an interactome-based approach can be useful in identifying targets.

Arsenic is a naturally occurring metalloid and environmental contaminant. Exposure to arsenic in drinking water was shown to cause cancer of the liver, kidneys, lung, bladder, and skin (Bates et al., 1992; Chen et al., 1992; Smith et al., 1992). The detrimental effects of arsenic exposure also include birth defects. Birth defects reported involve neural tube defects, coarction of the aorta, and midfacial cleft (Aggarwal et al., 2007; Hill et al., 2008). At a cellular level, arsenic can disturb cellular differentiation and migration (Li and Loch-Caruso, 2007; Trouba et al., 2000; Yancy et al., 2005).

One important cellular differentiation process that takes place early in development is epithelila-mesenchymal transition (EMT). EMT is a mechanism where a three-dimensional structure is formed from epithelial cell layers. The first EMT takes place at gastrulation in amniotes, but the process is also important during other embryonic events, including neural crest migration, valve formation of the heart, and facial development. As shown in the developing heart, EMT is regulated at the molecular level by factors found in the extracellular matrix, including transforming growth factor β (TGFβ) isoforms (Krug et al., 1987; Mjaatvedt et al., 1989; Potts and Runyan, 1989). Hays et al. (2008) found that arsenic exposure affects the normal production of extracellular matrix by lung cells. We considered the possibility that arsenic would disrupt a cellular process dependent on signaling through the extracellular matrix in the heart. A pilot experiment showed that low levels of arsenic exposure altered the EMT process in an in vitro assay of cardiac development.

EMT in embryonic heart valve development is a fairly well-described developmental process that has emerged from in vitro studies of mesenchymal cell formation and invasion of collagen gels (Person et al., 2005; Potts and Runyan, 1989). Atrioventricular (AV) canal explants undergo a temporally and spatially defined EMT in response to a stimulus from the myocardium involving TGFβ family members and other factors in the extracellular matrix (Brown et al., 1996; Potts and Runyan, 1989; Tavares et al., 2006). The EMT assay recapitulates the EMT process seen in vivo where progenitors of mitral and tricuspid valves are formed from mesenchymal precursors derived from the AV canal endothelium. This process only takes place in the AV canal, while the adjacent ventricular endothelium remains endothelial (Markwald et al., 1977). Experiments in this study used chicken embryos, while the EMT assay can use AV canals dissected from chicken or mouse embryos; the principle advantage of chicken embryo is that explants can be isolated from the embryo prior to onset of EMT. Through this the separation, activation, and invasion steps of EMT can be observed in vitro. The molecular mechanisms of TGFβ signaling and utilization of zinc finger transcription factors as transcriptional regulators of the process suggests that this EMT is fairly representative of the mechanisms used for both normal and pathologic EMT (Yang and Weinberg, 2008).

The present study reports on chicken embryo AV canal explants treated with 12.5–500 ppb (0.16–6.6μM) sodium arsenite (As III) and evaluated for EMT. As the arsenite concentration was increased to 500 ppb (6.6μM), the number of migratory cells, produced by EMT, decreased drastically, illustrating the ability of arsenic to block this process. This loss of mesenchymal cells was neither due to increased apoptosis nor due to a change in mesenchymal cell proliferation. To explore potential mechanisms, treated explants were collected for microarray analysis to obtain a profile of transcriptional response to arsenic treatment. Bioinformatic analysis identified several gene clusters representing perturbed cellular functions. The gene clusters identified cell cytoskeleton molecules, cell adhesion, and xenobiotic-metabolizing enzymes. To explore the validity of the clustering approach, genes within a cluster, but not found in the microarray, were tested and shown to be perturbed using real-time reverse transcriptase-PCR (RT-PCR). These studies identify a number of transcriptional changes associated with arsenic toxicity and point to EMT as a specific cell biological process affected by arsenic exposure.

As EMT is a fundamental process in early development, the reported relationship of Arsenic to high rates of spontaneous abortion (He et al., 2007; Milton et al., 2005; Rahman et al., 2007) and elevated levels of congenital heart disease (Engel and Smith, 1994) may be related to this mechanism. Spontaneous abortions can be produced by failure of EMT at gastrulation, neural crest formation, or heart development as each of these contributes to the production of a viable embryo. Therefore, an understanding of the cellular and molecular events in a model system of EMT may be relevant to understanding the pathology of arsenic exposure.

MATERIALS AND METHODS

Chicken embryo isolation.

Fertilized white leghorn chicken eggs were obtained from MacIntire Eggs (San Diego, CA). Eggs were incubated at 37°C for 48–72 h to obtain embryos at Hamburger and Hamilton stage 16. Using a dissecting stereo microscope, embryos were dissected free of embryonic membranes in 4°C sterile 1× Tyrode’s salt buffer and after rinsing in fresh sterile Tyrode’s buffer were once again dissected by pinching off the AV canal segment of the heart with #5 forceps. A video showing the dissection and subsequent assay is available from R.B.R. on request.

Collagen gel invasion assay.

Collagen gels were prepared as previously reported (Boyer et al., 1999a; Potts et al., 1992) and pretreated with 12.5–500 ppb (0.16–6.6μM) sodium arsenite (As III) (Cat 71287; Fluka). The low-dose arsenite concentrations of 12.5–100 ppb (0.16–1.3μM) are consistent with arsenic exposures of chicken chorioallantoic membranes (CAMs) that alter angiogenesis (Mousa et al., 2007; Soucy et al., 2003). The high arsenic concentration of 500 ppb (6.6μM) is comparable to exposure of both rat epithelial lung cells and rat embryos in an examination of lung development (Petrick et al., 2009). The physiological relevance of the arsenic concentration range is noted by starting just above the maximum contaminant level for drinking water in the United States (10 ppb) to levels observed in water associated with skin lesions and pathologies in other areas (Chen et al., 1985; Wu et al., 1989). Embryonic AV canals were located, dissected, and explanted onto the pretreated collagen. AV canal explants were incubated on the collagen gel for 24–36 h. One set of explants was collected after 36 h and fixed in 4% paraformaldehyde (PFA) for 30 min. These explants were observed on an inverted microscope equipped with Hoffman Modulation Optics to collect mesenchymal cell counts or prepared for TUNEL and cell proliferation assays. A second set of explants was extracted at 24 h to collect messenger RNA for microarray gene profiling and real-time RT-PCR. Cell counts and pictures were obtained from 36-h cultures in order to more clearly demonstrate morphological differences between cultures. RNA was collected after 24 h of treatment and compared on whole-genome microarrays to represent a more acute treatment corresponding to the time of EMT.

TUNEL assay.

Damaged and fragmented DNA was detected by TdT (terminal deoxynucleotidyl transferase)-mediated dUTP nick-end labeling (TUNEL) assay using the DeadEnd Fluorometric TUNEL System kit (Promega, WI). Explants on collagen gels were treated with 25 ppb sodium arsenite for 36 h. The collagen and AV explants were fixed with 4% PFA and permeabilized with 0.1% Triton X-100. The AV explants were labeled by incubation in 50-μl TUNEL reaction mixture containing the mixture of terminal deoxynucleotidyl transferase at dUTP. The labeling of AV explants was done in the dark at 37°C for 1 h. Cellular nuclei were stained using 4′,6-diamidino-2-phenylindole (DAPI) 1:10,000 (Sigma, MO) fluorescent stain. Eight AV canal explants for control and treated including endothelial outgrowth were randomly selected for analysis. Approximately 100 cells were counted from each explant. Analysis was performed using a fluorescent microscope, and cells were categorized as apoptotic (intense green nuclear fluorescence) or normal (no fluorescence).

Proliferation/immunostaining assay.

Proliferating cells were detected by immunostaining for Phospho-Histone H3 (ser 10). The collagen gels with cultured AV explants were treated with 25 ppb (0.33μM) sodium arsenite (As III) for 36 h. The AV explants were then fixed in 4% PFA and permeabilized with 0.1% Triton X-100. Polyclonal rabbit antibody for phospho-histone H3 was used as the primary antibody at a 1:300 dilution (Sigma). After washing with 1× PBS, Alexa 546 goat anti-rabbit IgG in a 1:400 dilution was used as the secondary antibody (Molecular Probes, CA). PBS (1×) was used to wash excess secondary antibody, and DAPI 1:10,000 (Sigma) was used to counterstain cell nuclei. Eight AV canal explants from control and treated including epithelial outgrowth were randomly selected for analysis, and ∼100 cells were counted from each explant. Analysis was performed using a fluorescent microscope, and cells were categorized as either proliferating (red nuclear fluorescence) or normal (no red fluorescence).

Avian microarray.

The collagen gel invasion assay was performed at three arsenic (As III) concentrations (10, 25, and 100 ppb or 0.13, 0.33, and 1.3μM, respectively). Avian complementary DNA (cDNA) arrays consisting of ∼20,000 oligo-DNA elements were printed in duplicate by the Genomic Research Laboratory in the Steele Children’s Research Center at the University of Arizona. The arrays were utilized as described by Konieczka et al., (2009). Total RNA was extracted from treated AV canal cultures using the RNeasy Mini Kit (Qiagen, MD). RNA was concentrated using RNeasy Mini Elute Cleanup Kit (50) (Qiagen). Sense amplification was performed using SenseAMP Amplification Kit (Genisphere, PA). The cDNA was then labeled using the SuperScript Indirect cDNA Labeling System (Invitrogen, CA). The labeled cDNA was hybridized to the array slides using the Slide Hyb Hybridization Buffer #1 (Applied Biosystems, CA). The hybridization process was done at a 50:50 ratio at 42°C for 16 h. The slides were then scanned using the Applied Precisions Array Worxe (white light scanner). Spot finding was then done with the Soft/Worxe Tracker 2.8.

A standard wheel design was used for comparing four samples, in which each sample was compared with the others. A dye swap was used for each comparison, making a total of eight microarray chips. To normalize the results, within-chip normalization was performed using the R package OLIN (Optimized Location- and Intensity-dependent Normalization) (Futschik and Crompton, 2005). The false discovery rate was computed for each spot based on intensity- and location-dependent bias. Standard libraries in the R BioConductor package were used to normalize between array chips (Bolstad et al., 2003). Finally, linear models were fit to the normalized gene expression using the limma library, which computes log₂ fold change, indicating the quantity and directional change of gene expression, T- and B-statistics, and adjusted p value that takes into account the false discovery rate (Smyth et al., 2003). Genes differentially expressed by a p < 0.8 value were selected to carry out further analysis using Cytoscape/BioNetBuilder2.0/jActiveModules software. While the p value selected is higher than the often-used 0.05 value, it is important to note that the 0.08 is the adjusted p value that includes false discovery rate calculations for a more stringent selection. Previous microarray work looking at known components of EMT signaling pathways during chick gastrulation showed that 0.08 appeared to be a better fit than 0.05. The better fit was conveyed by inclusion of known molecular networks under 0.08 p value (Konieczka et al., 2009).

Quantitative real-time RT-PCR.

Total RNA was extracted from AV canal explants cultured on the collagen gel by using TRIZOL reagent (Gibco BRL). RNA was DNase treated with TURBO-DNA free kit (Ambion). cDNA was transcribed using the iScript cDNA Synthesis Kit (Bio-Rad, Richmond, CA). Data normalization against the specific housekeeping genes (GAPDH, LDH, β-actin, and mitochondrial DNA) was problematic due to either the substantial phenotypic change during EMT or the effects of the treatment. Therefore, total cDNA was used to normalize the quantitative PCR reactions. To sensitively measure cDNA in each reaction, Quanti-iT Oligreen ssDNA reagent (Molecular Probes) was used to measure and aliquot single-stranded cDNA in combination with a fluorometer (Turner Biosystems). Each reaction was performed in triplicate. Primer sequences for the selected genes are listed in Table 1.

TABLE 1.

Gene Primer Sequences Used for Real-Time RT-PCR in Figure 7. Genes Selected were Identified in the Arsenic Network as Blank Nodes. The Uid Number Provides Access to Gene Information Via the Kyoto Encyclopedia of Genes and Genome.

Gene name	Uid	Primer sequence
Actinin, alpha 2 (ACTN2)	396263	5′-TTGACCCCAATGGACAAGGAAC-3′
		5′-AAGCCAGGATTCTGAAGGAAGC-3′
Rho guanine nucleotide exchange factor (GEF) 7 (ARHGEF7)	418761	5′-ACCTCTGGAACCTCCTAAAACACC-3′
		5′-TGTAACAAAGTGCTGCTGAAGGC-3′
Neuroblastoma RAS viral (v-ras) oncogene homolog (NRAS)	419885	5′-AAGTGCGATTTGCCAACAAGGAC-3′
		5′-TGATGTCTCTATGAAGGGAATGCCG-3′
Forkhead box A2 (FOXA2)	395539	5′-TTCAACCACCCCTTCTCCATC-3′
		5′-TGAGGTCCATTTTGTGCGAGG-3′
Myosin, heavy chain 11 (MYH11)	396211	5′-AACCAGTATCTTTTGCCCCACC-3′
		5′-TTCTTCTTCACCTCCATGTGTTGC-3′
Cluster of differentiation 81 (CD81)	374256	5′-AAGCAGTTCTACGACCAAGCGTTC-3′
		5′-TGGAAAGTCTTCACCACAGCCTTC-3′
Colony stimulating factor 2 receptor beta (CSF2RB)	771315	5′TTTTCCTTTGCCAGGTTGGTGAC-3′
		5′-TTTGATACATCCGCTTCCTTGC-3′
Growth arrest-specific 7 (GAS7)	417305	5′-TGCCAGAGCAGCAGTTGTTGAAAC-3′
		5′-ATCCTGACACCGTATTGTTCCCCTG-3′
Tubulin, beta 3 (TUBB3)	431043	5′-TCCCCAACAATGTCAAGGTGGC-3′
		5′-AACTCCATCTCGTCCATCCCCTCTC-3′
Myosin, heavy chain 1 (MYH1)	417309	5′-TACCCAGATACACTTGGACGATGC-3′
		5′-TCCTGCTCAGCCACTTTTCTTG-3′

Cytoscape analysis.

To visualize and characterize data obtained with the avian microarray, we used the chicken interactome downloaded via BioNetBuilder2.0 (Konieczka et al., 2009) and Cytoscape 2.6.6 (Killcoyne et al., 2009). The chicken interactome is a genome-wide set of molecular interactions transferred and integrated from interaction data in diverse eukaryotic species. As described by Konieczka et al. (2009), a variety of databases showing both hierarchical regulation and molecular interactions were mined to develop the interactome. Nodes (represented as circles) in the chicken interactome correspond to genes, and edges (connecting lines) represent documented interactions. The Cytoscape plug-in jActiveModules was used with microarray data to identify clusters of genes in the network most representative of arsenic treatment (Campanaro et al., 2007; Cline et al., 2007; Yeung et al., 2008). Cytoscape version 2.6.2 and the plug-ins were downloaded from www.cytoscape.org.

GOminer analysis.

GOminer uses gene ontology annotation of molecules to generate categories of cellular processes. This is another free open-source software commonly used to categorize microarray data into cellular processes (Kestler et al., 2008; Lohrig et al., 2009; Morandi et al., 2008). This software was obtained from http://discover.nci.nih.gov/gominer/. In this study, the Cytoscape network, described above, was used to generate the gene list to be uploaded into the GOminer software. The mechanistic categories assigned by GOminer were then used to identify biological responses seen with arsenic treatment.

RESULTS

Arsenite Blocks EMT

Initial experiments with the collagen gel invasion assay were performed with AV canal segments from stage 16 embryos. The AV explants were placed onto collagen gels and treated with either 12.5 ppb (0.16μM) sodium arsenite (As III) or buffer vehicle for 36 h. Control explants (Fig. 1, panel A) show both endothelial cell outgrowth and elongated mesenchymal cells (arrows) on and within the collagen gel. The normal response to the inductive signal by the myocardium is an activation and separation of the endothelial cells with invasion by up to 10% of cells after 48 h. After As (III) treatment (Fig. 1, panel B), endothelial cells growing from the AV canal explants were unable to separate and transform into invasive mesenchymal cells. As (III)-treated AV canals showed fewer mesenchymal cells (arrowhead), indicating an inhibition of EMT.

FIG. 1.

Collagen gel invasion assay performed with arsenic exposure (12.5 ppb) and chicken AV canal explants. Treatment was carried out for 36 h. Arrows on the control panel (panel A) indicate mesenchymal cells. The arrowhead in the sodium arsenite (As III) panel (panel B) indicates a mesenchymal cell. Sodium arsenite (As III) treatment decreased the presence of mesenchymal cells.

With fewer mesenchymal cells in the As (III)-treated AV canals, a dose-response series was performed to investigate effects of As (III) in the range of 12.5–500 ppb (0.16–6.6μM). The number of mesenchymal cells produced by each explant was counted for each arsenic treatment (Fig. 2). As the concentration of As (III) increased, the number of mesenchymal cells decreased to produce a complete EMT inhibition at 500 ppb (6.6μM) As (III) treatment. The decrease of mesenchymal cells was considerable even at the lowest arsenite dose (12.5 ppb or 0.16μM) where it showed a substantial decrease from control cultures.

FIG. 2.

Collagen gel invasion assays were performed at various concentrations of sodium arsenite (0, 12.5, 62.5, 125, and 500 ppb). The sodium arsenite treatment was carried out for 36 h. Increasing concentrations of sodium arsenite produced a loss of mesenchymal cells.

Loss of mesenchymal cells in the collagen gel culture can reflect inhibition of EMT, a reduction in normal cell proliferation, or increased apoptosis by either mesenchymal or progenitor endothelial cells (Person et al., 2005). To explore whether alternative explanations to EMT were viable, collagen gel cultures treated with 25 ppb (0.33μM) As (III) were fixed after 36 h and examined for evidence of altered cell proliferation or cell death in endothelial and mesenchymal cells compared to control cultures. Cell proliferation was evaluated by staining fixed gels for phospho-histone H3, as an index of cell replication. Figure 3 panels A and B illustrate a low and comparable number of cells dividing under control and As (III) treatment. The percentage of stained cells in the mixed mesenchymal and endothelial population was unchanged by treatment (Fig. 3C). Cell count percent reported was obtained from eight explants in each group. To measure cell death in treated cultures, a TUNEL assay was performed. No difference was observed between arsenic-exposed and control populations (Figs. 4A and 4B). The percentage of apoptotic cells in the mixed mesenchymal and endothelial population was unchanged by treatment (Fig. 4C). Cell count percent reported was obtained from eight explants in each group.

FIG. 3.

Representative picture of epithelial cells from AV canal stained for proliferation using of phospho-Histone H3 (ser 10) and nuclear DAPI stain. Panel A illustrates phospho-Histone H3 stating of one cell indicating proliferation under control conditions. Panel B illustrates proliferation of one cell under 25-ppb arsenic treatment for 36 h. The pictures shown are a representation of the low number of proliferating cells compared to nonproliferating epithelial and mesenchymal cells. Figure 3C illustrates the percentage of proliferating cells (proliferating cell/epithelial cells) in control samples versus 25 ppb arsenic treated (As). The percentage of proliferating cells and the SEM was calculated from n = 8 (eight AV canals from different embryos).

FIG. 4.

Representative picture of epithelial cells from AV canal stained for apoptotic DNA fragmentation. The nuclei are counterstained with DAPI. Panel A illustrates control epithelial cells with a couple apoptotic cells. Panel B illustrates epithelial cells from AV canal treated with 25 ppb for 36 h. Apoptotic bodies seen under arsenic treatment are minor and comparable to control. Figure 4C illustrates the percentage of apoptotic cells (apoptotic cell/epithelial cells) in control samples versus 25 ppb arsenic treated (As). The percentage of apoptotic cells and the SEM was calculated from n = 8 (eight AV canals from different embryos).

Previous work in this laboratory has indicated a critical role of TGFβ isoforms in the process of EMT (Boyer et al., 1999b; Mercado-Pimentel and Runyan, 2007; Potts et al., 1991). To evaluate whether loss of EMT was due to a loss of growth factor expression in the explants, we extracted RNA from 10 explants at each arsenite concentration and measured TGFβ2 and TGFβ3 RNAs by real-time PCR. Data indicated that TGFβ2 expression was unchanged at any dose and TGFβ3 expression was only marginally reduced by 20% at 25 ppb (data not shown).

Microarray and Bioinformatics Analysis

To examine how arsenite exposure might perturb EMT, microarray analysis was carried out. Three As (III) concentrations (10, 25, and 100 ppb or 0.13, 0.33, and 1.3μM, respectively) were used to provide a dose-response range that included the lowest critical dose of mesenchymal cell loss (Fig. 1). Table 2 reports gene changes seen at each arsenic concentration. At an adjusted p value of 0.08, the 10-ppb dose showed 413 genes with altered expression, while the 25-ppb (0.33μM) treatment showed 382 changed genes and the 100-ppb treatment showed 263 changed genes (Table 2). An examination of both fold change and total affected gene numbers suggested that 10 and 25 ppb (0.13 and 0.33μM) produced substantial effects (Supplementary figure 1). The 25 ppb (0.33μM) dose also appeared to be close to the half maximal effect on EMT (Figs. 1 and 2); therefore, bioinformatics analysis focused on the 25- and 10-ppb doses (0.33 and 0.13μM). The 100-ppb dose was not included into the bioinformatic analysis as it added little new information and it seemed less physiologically relevant when compared to 10 and 25 ppb (microarray data, including all doses, are provided as Supplementary data).

TABLE 2.

Number of Affected Genes Under 12.5, 25, and 100 ppb Sodium Arsenite Acute Exposure of 24 h. Changed Gene Expression was had a Fold Change of ± 0.58 and an Adjusted p Value of 0.08 or Less.

Arsenite treatment (ppb)	Downregulated	Upregulated
10 ppb (413 affected genes)	214	199
25 ppb (382 affected genes)	133	249
100 ppb (263 affected genes)	108	155

Cytoscape version 2.62 and the plug-in jActiveModules were used to identify relevant clusters of interacting molecules within the chicken interactome (Konieczka et al., 2009). As described, the chicken interactome is a cross-species integrated network of functional interactions among genes and proteins projected onto the chicken genome. This database is a compilation of seven databases, including protein-protein interaction and gene regulatory networks. Genes in the chicken interactome are represented by nodes (circles) and known interactions with other molecules are represented by edges (lines). The arsenic microarray data were applied to the entire chicken interactome, and jActiveModules was used to identify network clusters germane to As (III) treatment (Fig. 5). Three hundred and eighty two genes were identified as altered by arsenic exposure; however, only 142 of these genes were present in the chicken interactome. The 142 genes were organized and visualized into four clusters. The genes in this network were color coded (red: upregulation and green: downregulation) when significantly affected by As (III) at either 10 or 25 ppb (0.13 or 0.33μM) or at both concentrations. Genes within each cluster showed a mixture of upregulated, downregulated, and unchanged RNA expression among the interacting molecules. Cluster A showed the majority of genes to be upregulated with some downregulated genes. A few genes were unaffected by arsenic and are illustrated as gray nodes. Cluster A included the cytoskeleton-related genes, cofilin 2, actin alpha 2, neurofilament 1, actin-related 2/3, and myozenin 2 (Fig. 6). There is a common element of upregulation of these genes. In contrast, diaph1, an actin-regulating molecule, is downregulated along with actin 1. This cluster also includes a number of Guanosine-5′-triphosphatases (GTPases) and related exchange factors, including NF1, N-Ras, GAS7, ARHGEF7, ARHGEF12, and ARHGEF6. Interestingly, the homeotic gene, HoxB5, is upregulated in this cluster and is associated with endothelial phenotype maintenance (Wu et al., 2003). Another gene in this cluster, TSC1, is a tumor suppressor. Together, the genes in cluster A appear to be consistent with the maintenance of endothelial phenotype in cultures and inhibition of change in the cytoskeleton associated with EMT.

FIG. 5.

The chicken interactome and Cytoscape (version 2.6.2)/BioNetBuilder2.0/jActiveModules were used to interpret microarray gene expression data. The network shown here represents gene expression from 10- and 25-ppb sodium arsenite treatment for 24 h. Red, green, and gray colored nodes were used to illustrate the gene expression obtained by microarray. Red nodes represent upregulation, green represents downregulation, and gray represents no change in expression. The connections between molecules show molecular interactions identified in the chicken interactome. Gene expression illustrated in colored nodes was selected with a p < 0.8 value. Four gene clusters were found and arbitrarily labeled as A, B, C and D. Tables identifying the genes in each cluster are provided in supplemental data.

FIG. 6.

This figure illustrates a closer look at gene cluster A within the arsenic network. This gene cluster includes actin-related molecule 2/3, neurofilament, neurofibromin, actin 2, and myozenin 2, which have common biological roles in the cell cytoskeleton. Red nodes are upregulated, green nodes are downregulated, and gray nodes show no change. Blank nodes are not present in the microarray but are included in the network, as they have known connections to molecules found in the array data.

Additional clusters were labeled as B, C, and D (cluster members are displayed in Supplementary tables). Cluster B is a group of 46 interacting genes but only 20 of these genes display changes due to arsenic exposure and a majority of these are upregulated (15 up vs. 7 down). Genes in cluster B showed no obvious relation to the mechanisms of EMT. Cluster C comprises 26 genes with 10 significantly upregulated and 5 downregulated. Downregulated genes in relation to EMT include tetraspanin 4, a molecule that can interact with integrins. Tetraspanin 4 is also a member of a family of proteins associated with metastasis (Zhou et al., 2008) and ADAMTS1, a protease involved in cell invasion (Su et al., 2008). Upregulated genes in cluster C also include V-Cadherin and VCAM, molecules associated with retention of the endothelial phenotype. The last cluster (D) displayed 24 genes of which 7 were downregulated and 4 upregulated. The cluster appears to be fairly heterogeneous but includes a Notch ligand, DLL4, that is downregulated and another DLL1 that is essentially unchanged. Notch signaling has been correlated with EMT (Timmerman et al., 2003).

To validate the clustering approach, we explored the regulation of genes that interacted with members of the cluster but were not found in the microarray (represented as blank or uncolored nodes in Fig. 6). As the blank genes should interact with genes present in the clusters and may be also regulated by As (III) exposure, real-time RT-PCR analysis was performed to examine their expression. The missing (blank) genes used for real-time RT-PCR were actin α 2, Rho GEF 7, neuroblastoma Ras viral oncogene, forkhead box 2, myosin heavy chain 11, CD81, colony stimulating factor 2 receptor, growth arrest-specific 7, tubulin β3, and myosin heavy chain 1 (Fig. 7). Real-time RT-PCR showed that all these genes were affected by As (III) exposure, and the data relevant to cluster A were added to Figure 6 to generate a new figure with the same color code for up and downregulation (Fig. 8). Comparison of gene expression, by the missing genes, showed that the genes are differentially expressed in As (III) treatment, as might be predicted by their presence in the significant cluster identified by jActiveModules. The direction of regulation was largely predictable as it was similar to the regulation of the nearest neighbors in the cluster distribution. This argues that the clustering process is relevant to the effects of arsenic on EMT.

FIG. 7.

Blank (white) genes found in the arsenic network were selected for real-time RT-PCR analysis. Collagen gel cultures were prepared as before with 25 ppb sodium arsenite and compared to control cultures. Gene expression is normalized to total cDNA to control the amount loaded on each reaction. The Quanti-iT Oligreen ssDNA reagent was used to measure cDNA.

FIG. 8.

This figure incorporates the previously blank genes (shown in Fig. 6) using expression changes obtained with real-time RT-PCR (shown in Fig. 7), red upregulated and green downregulated. Inserted data are shown with hatched patterns. Expression patterns correlate with neighboring molecules and validate the clustering process.

GOminer

To further explore the microarray data, GOminer software was used to categorize genes found in the clusters generated by Cytoscape and plug-in software. The GOminer software identified biological processes related to genes in each cluster. Among the most interesting categories found were cell cytoskeleton (Supplementary table A), cell adhesion (Supplemental table C), and heterogeneous and undefined clusters B and D (Supplementary tables B and D). GOminer also facilitated the search of signaling pathways that included molecules with changed expression under arsenic treatment. Important signaling pathways that help explain the arsenic inhibition of EMT were appear to include cell adhesion and matrix molecules, cytoskeletal components, small GTPases, and their regulators (gene lists under each cluster are found in the Supplementary tables for this paper).

DISCUSSION

In the present study, we found that sodium arsenite (As III) inhibited cardiac endothelial EMT, even at doses as low as 12.5 ppb (0.16μM). EMT is a central cellular process used in embryo development. Among the most widely studied embryonic EMTs are gastrulation, neural crest cells formation, and heart valve formation (McKeown et al., 2005; Nakaya and Sheng, 2008; Runyan and Markwald, 1983). The conservation of embryonic transcription factors and common utilization of proteases and cytoskeletal elements in metastatic cancer and tissue fibrosis argues that EMT is reiterated in adult pathologies (Klymkowsky and Savagner, 2009; Yang and Weinberg, 2008). In this study, we used the collagen gel invasion assay, a well-described model of EMT. Chicken AV canal explants at stage 16 were placed on a collagen gel to explore the effects of As (III) in vitro. Explants were grown on collagen gels and exposed to As (III) at different concentrations (0, 12.5, 62.5, 100, and 500 ppb or 0, 0.16, 0.83, 1.3, and 6.6μM, respectively). Treatment time was kept as an acute exposure of 36 h (24 h for microarray data). The data show that mesenchymal cell numbers are profoundly inhibited at low doses of arsenic and that the normal endothelial separation that accompanies EMT was inhibited.

Though comparison of environmental exposure and the direct acute exposure examined here in a tissue culture system is difficult, the concentrations tested here appear to be relevant to environmental exposure. Arsenic is readily transferred across the placenta as shown by observations that cord blood concentrations of arsenic can reach 80% of the maternal blood concentration (Rudge et al., 2009). Data from Valentine et al. (1979) suggest that blood concentrations of arsenic average 4–5% of local water levels. Extrapolation of these numbers suggests that this tissue culture exposure is roughly equivalent to a drinking water range of 200–500 ppb of sodium arsenite, a range that can be found in drinking water supplies in several parts of the world (Chen et al., 1985; Wu et al., 1989). Though there are clearly differences in acute and chronic exposure to arsenic (Olsen et al., 2008), EMT events in the early embryo are temporally defined and could be considered acute events even in the background of chronic exposures in the mother.

The finding that low doses of arsenite inhibit EMT by cardiac endothelia is striking. Similar doses of arsenite were used with the CAM assay, and it is clear that concentrations lower than 1μM stimulate angiogenesis by avian endothelia (Mousa et al., 2007; Soucy et al., 2003). This effect appears contradictory as angiogenesis by endothelial cells involves a similar process of endothelial cell invasion into the extracellular matrix. Molecules associated with EMT in the heart such as Sphingosine-1-P (Wendler and Rivkees, 2006) and matrix metalloproteinases (Alexander et al., 1997) are similarly involved in angiogenesis (Iwasaka et al., 1996; Straub et al., 2009). One notable difference between the CAM and cardiac EMT assay described here appears to be one of the effects of vascular endothelial growth factor (VEGF). VEGF is quickly upregulated by As (III) in endothelial cells such as Human Umbilical Vein Endothelial Cells and corresponds to angiogenesis in the CAM assay (Kao et al., 2003; Soucy et al., 2003). While the literature on VEGF in valve development is confusing (Chang et al., 2004; Dor et al., 2003; Enciso et al., 2003), the data are most consistent with an inhibitory role of VEGF on cardiac EMT. Although differential VEGF responses might explain the present data, VEGF was not detected in the microarray data at this stage of development and this is consistent with the literature. We note that similar doses of sodium arsenite (10–50 ppb or 0.13–0.66μM) prevent normal wound repair by lung epithelial cells in a scrape assay (Olsen et al., 2008). While this assay does not measure cell invasion, the change in epithelial phenotype associated with wound repair is an EMT (Arnoux et al., 2005).

The invasion assay is based on the quantitation of mesenchymal cells found within the matrix of the collagen gel. Explants are composed of myocardial cells, endothelial cells, and mesenchymal cells emerging by EMT from the endothelia (Person et al., 2005). Mesenchymal cell numbers largely reflect the process of EMT but can be altered by proliferation or cell death of mesenchyme within the gel. Another way mesenchymal cell number changes is by changing endothelial progenitors. The cell proliferation and cell death assays performed here argue against these alternative explanations and point to the perturbation of the EMT process. There is the possibility that EMT in the system might be altered if As (III) alters normal differentiation of endothelial cells. The upregulation of HoxB5 seen in our data is consistent with enhanced differentiation of the endothelia. However, the differentiation status of endothelial cells and EMT does not appear to be an issue as fully differentiated endothelial cells in the adult mouse heart remain susceptible to TGFβ-induced EMT (Zeisberg et al., 2007).

The most widely recognized outcome of As (III) exposure in drinking water is skin cancer (Fumal et al., 2005; Graham-Evans et al., 2004; Yu et al., 2006). Metastatic lesions require fairly high levels of chronic arsenic exposure (Wang and Rossman, 1995), but it is reported that many mild to moderate arsenic-induced skin lesions revert upon cessation of arsenic exposure (Smith et al., 2000). The noninvasive nature of skin lesions produced under arsenic exposure suggests that they are slow to convert to a metastatic phenotype. Arsenic trioxide, a form of As (III), has been tested as an antitumor agent (Murgo et al., 2000). As metastasis is mediated by the cellular process of EMT (Yang and Weinberg, 2008), we suggest that a negative role of arsenic on EMT may be related to the reversion of skin lesions and effectiveness against tumors (Mani et al., 2008; Sarkar et al., 2009). As EMT requires an element of cell migration, the ability of arsenic to inhibit cell migration in HeLa cells and lung tissue in concentrations lower than 1μM may also be related to the gene expression changes reported here (Olsen et al., 2008; Wei et al., 2005). This role is not unique as other reports suggest that arsenic (4–10μM) inhibits tumor formation and metastasis by increasing stress proteins, thermotolerance, and inducing apoptosis, mechanisms that do not necessarily relate to EMT (Ivanova and Heia, 2006; Namgung and Xia, 2000; Tomasavic and Welch, 1986). In this light, the ability of As (III) to induce angiogenesis under low-dose (< 1μM) exposure does not appear consistent (Mousa et al., 2007; Soucy et al., 2003). The conflicting results indicate that arsenic induces a dose-, cell-, and context-specific effect as also noted by Soucy et al., (2003).

The collagen gel assay has been widely used to demonstrate the inductive effects of TGFβ isoforms as part of the myocardial stimulus for EMT (Boyer et al., 1999c; Potts and Runyan, 1989; Tavares et al., 2006). We first examined the effect of arsenic on TGFβ expression but found no evidence that either isoform present in the chicken heart was sufficiently altered to prevent normal EMT. To explore the basis for this inhibition of EMT in the assay, we turned to a microarray survey of altered gene transcription with the expectation that missing or altered expression of molecules would identify the likely basis for the inhibition of EMT.

The effects of As (III) on the explanted tissue were remarkably limited. Microarray analysis of this acute assay showed that less than 2% of the elements of the array were perturbed. The low number of affected genes is consistent with previous microarray studies performed on human keratinocytes under arsenic exposure (Rea et al., 2003). In contrast, direct perturbations of TGFβ, Notch pathways, or treatment with a known cardiac teratogen (trichloroethylene) in this tissue perturbs expression of more than 1500 genes on the array (unpublished). We found similar effects of As (III) at 10, 25, and 100 ppb (0.13, 0.33, and 1.3μM) but the 10- and 25-ppb exposures overlapped to a greater extent, and we concentrated on results from these exposures. As the process of EMT has become more widely investigated in normal and pathological tissues, it is clear that EMT is a transcriptionally regulated process utilizing an overlapping set of transcription factors, including gene families of snail, zeb, twist, and several other transcription factors (Yang and Weinberg, 2008). Surprisingly, we found few changes in transcription factor expression. One transcription factor change observed was the upregulation of HoxB5, a gene associated with maintenance of endothelial cell phenotype and consistent with the lack of EMT (Wu et al. 2003).

We approached the bioinformatic analysis of the data by organizing the results in the context of the chicken interactome. While optimized for the chicken, the interactome is a compendium of data obtained from all vertebrate species and is based on data collected from seven molecular interaction databases designed to describe the biological interactions that regulate complex development (Konieczka et al., 2009). The software for this analysis is freely available and is based on the Cytoscape imaging platform using the BionetBuilder2.0 plug-in to incorporate the chicken interactome and the jActiveModules plug-in to evaluate and select microarray data. The results of our analysis showed that the molecules identified by the software fell into four clusters. The clusters appeared to be heterogeneous and included microarray elements that were both up and downregulated (red and green colored, respectively). The clusters also included known interactions that were unaffected by As (III) exposure. The functional relationship between members of a cluster was not always obvious, but we found the following affected genes in cluster A: actin-related 2/3 subunit 5, myozenin 2, actin 2, neurofilament, and cofilin 2. We also found several GTPases and GTP exchange factors in this cluster. A common cellular function these molecules share is a role in the organization and maintenance of the cell cytoskeleton. Arsenic exposure (Fig. 1B) blocks the ability of the endothelial cells to separate and invade. The increased expression of these molecules under As (III) is consistent with the lack of change in cell shape and the maintenance of the endothelial phenotype. The molecules, diaph1 and associated actins, which play a role in reorganization of the cytoskeleton were downregulated under As (III).

The other three clusters do not show obvious correlations that we understand either in relation to EMT or in relation to the clusters themselves. Individual molecules can be selected from each group. For example, cluster C includes a cell adhesion molecule that is upregulated (VCAM1) and is consistent with retained cell adhesion. Secreted extracellular matrix molecules such as laminin 3 and versican, also found in cluster C, are consistent with endothelial differentiation and retention of phenotype. A protease associated with EMT (ADAMTS1) is downregulated. Cluster C also includes microsomal glutathione-s-transferase and a detoxification/stress response subset. Genes in cluster D also form a diverse group of cellular processes. Among the themes in this cluster are array elements related to microtubules assembly and cell secretion.

The heterogeneous nature of the clusters raises the question as to whether this analysis is useful in relation to the As (III) inhibition of EMT. The relationships within the clusters are derived from a variety of data and cell types and may or may not be relevant to the present treatment. However, the predictive aspects of the clustering support the utility of the analysis. The data analysis identified a number of genes that were not found in our microarray dataset but were identified by the interactome as directly connecting two or more members of the cluster. We obtained primers for 10 of these molecules and performed real-time RT-PCR on arsenic-treated and control tissue. As seen in Figures 7 and 8, all 10 molecules showed a significant change in expression (albeit changes by NRAS and ACTN2 were modest) with 25 ppb (0.33μM) As (III) exposure. The changes seen were consistent with direction of regulation of their interacting neighbors in the cell cytoskeleton cluster. This test of clustering suggests that this approach to bioinformatic analysis of microarray data can identify relevant target areas for analysis. As the interactome is further developed, the quality of the analysis should improve and additional genes identified in the microarray can be included in subnetwork analysis.

In summary, our findings suggest that the basic cellular process of EMT is sensitive to low levels of As (III) exposure. Low exposure levels are consistent with high spontaneous abortions, where early embryonic processes such as gastrulation, vasculogenesis, neural crest formation, and heart development are strongly inhibited. The observation is also relevant to the pathology of skin lesions produced by As (III) in exposed human populations. This suggests that low levels of As (III) may be inhibitory to metastatic forms of cancer and impair normal wound healing and angiogenic processes that use part of the EMT program. Microarray analysis of altered gene expression suggests that this inhibition is produced, for the most part, by acute effects on a discrete set of cytoskeletal and cell adhesion genes and not by inhibition of known transcriptional regulators of EMT.

SUPPLEMENTARY DATA

Supplementary data are available online at http://toxsci.oxfordjournals.org/.

FUNDING

National Heart, Lung and Blood Institute (R01 HL82851); the Initiative for Maximizing Student Diversity (National Institutes of Health); More Graduate Education at Mountain State Alliance (MGE@MSA).