Combinatorial interventions inhibit TGFβ-driven epithelial-to-mesenchymal transition and support hybrid cellular phenotypes

Abstract

Epithelial-to-mesenchymal transition (EMT) is a developmental process hijacked by cancer cells to leave the primary tumor site, invade surrounding tissue and establish distant metastases. A hallmark of EMT is the loss of E-cadherin expression, and one major signal for the induction of EMT is transforming growth factor beta (TGFβ), which is dysregulated in up to 40% of hepatocellular carcinoma (HCC). We aim to identify network perturbations that suppress TGFβ-driven EMT, with the goal of suppressing invasive properties of cancer cells. We use a systems-level Boolean dynamic model of EMT to systematically screen individual and combination perturbations (inhibition or constitutive activation of up to four nodes). We use a recently developed network control approach to understand the mechanism through which the combinatorial interventions suppress EMT. We test the results of our in silico analysis using siRNA. Our model predicts that targeting key elements of feedback loops in combination with the SMAD complex is more effective than suppressing the SMAD complex alone. We demonstrate experimentally that expression of a majority of these elements is enriched in mesenchymal relative to epithelial phenotype HCC cell lines. An siRNA screen of the predicted combinations confirms that many targeting strategies suppress TGFβ-driven EMT measured by E-cadherin expression and cell migration. Our analysis reveals that some perturbations give rise to hybrid states intermediate to the epithelial and mesenchymal states. Our results indicate that EMT is driven by an interconnected signaling network and many apparently successful single interventions may lead to steady states that are in-between epithelial and mesenchymal states. As these putative hybrid or partial EMT states may retain invasive properties, our results suggest that combinatorial therapies are necessary to fully suppress invasive properties of tumor cells.

Figure 1

An in silico combinatorial knockout screen in the EMT network model reveals specific node combinations that can suppress the TGFβ-driven EMT. A previously constructed network model of EMT was used to identify nodes whose knockout (sustained OFF state) blocked TGFβ-driven EMT. (a) Schematic demonstrating that epithelial and mesenchymal cellular states are stable unless an external signal (e.g., TGFβ) or perturbation is applied. Our goal is to identify inhibitory perturbations that block the transition between the epithelial and mesenchymal states even in the presence of TGFβ. (b) The effect of knocking out nodes individually and in combinations and their effect on the percentage of EMT in 1,000 simulations per knockout combination. The EMT Percentage is given by the percentage of simulations at the end of which the EMT node is found in the ON state. (c) A network control approach was employed to identify epithelial control sets, i.e., set of nodes that when their states are controlled lead to the epithelial steady state. The schematic illustrates the effect of applying the epithelial control set on the network states. (d) The stable motif associated with the epithelial steady state. The nodes and edges that are part of the epithelial stable motif have thick lines, while the nodes and edges in the EMT network that are not part of the epithelial stable motif have dashed lines. The epithelial control sets contain one node from each connected group of nodes highlighted by a yellow background. All nodes of the predicted knockout combinations (blue shadow) are part of the epithelial stable motif, and are contained within an epithelial control set or form a path of nodes directly upstream of an epithelial control set. Black or white background on the node symbol indicates that the node is OFF or ON in the epithelial stable motif, respectively. EMT, Epithelial-to-mesenchymal transition, TGFβ, transforming growth factor beta.

Figure 2

Mesenchymal phenotype cells are enriched for combinatorial intervention target nodes predicted by the EMT network model. (a) Expression of epithelial (E-cadherin) and mesenchymal (vimentin) markers in HCC cell lines. (b) Immunofluorescent staining of EMT markers E-cadherin (green), vimentin (red), and nuclear stain DAPI (blue) in HCC cell lines Huh7, PLC/PRF/5 (Alexander), HepG2, and HLF. Expression of mRNA (c) by qRT-PCR and protein expression by immunoblot (d) of nodes whose paired knockout with SMAD is predicted by the EMT network model to inhibit TGFβ-driven EMT. (e) Epithelial-like HCC cells Huh7 were treated with TGFβ in serum free media for 48 h (1 ng/ml and 5 ng/ml), then mRNA expression of nodes whose paired knockout is predicted by the EMT network model to inhibit TGFβ-driven EMT was measured by qRT-PCR. EMT, Epithelial-to-mesenchymal transition, HCC, hepatocellular carcinoma, TGFβ, transforming growth factor beta, qRT-PCR, quantitative real-time PCR.

Figure 3

A multi-faceted siRNA screen to test predicted node knockdown combinations in vitro. (a) Schematic of the experimental design of the screen to test node knockdown combinations that are predicted to inhibit TGFβ-driven EMT. At 0 h, epithelial-like Huh7 cells were transfected with siRNA combinations or a scrambled siRNA control (2 nM per siRNA), and then plated for harvesting of protein, mRNA, and cell migration. At 45 h post transfection, cells were serum starved for 3 h. At 48 h, cells were treated with TGFβ (5 ng/ml) or a vehicle control for an additional 48 h. At 96 h, cells were harvested for mRNA and protein expression. An aliquot of siRNA-transfected cells were plated at 0 h in 96-well plates for assessment of migration with the Oris Cell Migration Assay (see Materials and Methods for details). These cells were treated with either TGFβ or a vehicle control at 48 h. Simultaneously, migration stoppers were removed at 48 h post transfection and analysis of cell migration was performed at 48, 72, and 96 h post transfection. (b) The effect of model-predicted node knockout combinations on E-cadherin expression in TGFβ-treated cells relative to vehicle control treated cells, as measured by quantitative immunoblotting. (c) The effect of model-predicted node knockout combinations on in vitro cell migration in TGFβ-treated cells relative to vehicle control treated cells. The percent change in TGFβ-driven migration (top) and heat-map of TGFβ-driven migration (bottom) relative to a scrambled siRNA control 24 and 48 h after TGFβ treatment. EMT, Epithelial-to-mesenchymal transition, mRNA, messenger RNA; siRNA, small interfering RNA, TGFβ, transforming growth factor beta.

Figure 4

Analysis of network motifs that arise after SMAD inhibition in the EMT network reveal an attractor landscape that is distinct from the unperturbed EMT network. The TGFβ-driven SMAD-perturbed EMT network (the EMT network with fixed TGFβ=ON and SMAD=OFF) has an almost identical mesenchymal state as the unperturbed network, as well as epithelial and mesenchymal states that are relatively different from the epithelial state of the unperturbed EMT model. We identify these relatively different states as hybrid epithelial-mesenchymal states. (a) The stable motifs associated with the hybrid epithelial–mesenchymal steady states and the sequence in which they stabilize after the TGFβ signal and SMAD suppression. The color of the node background indicates the node state; black corresponds to OFF and white corresponds to ON. The stable motif in the middle contains subsets of the stable motifs associated to the mesenchymal steady state (shown in b) but with the opposite states. These motifs are the Wnt/β-catenin stable motif (blue background), the AKT stable motif (red background), and the SHH stable motif (yellow background). (b) The stable motifs associated with the mesenchymal steady state in the TGFβ-driven SMAD-perturbed (TGFβ=ON; SMAD=OFF) EMT network and in the unperturbed EMT network. The color of the box that contains the stable motifs is the same as the background color used in a. EMT, Epithelial-to-mesenchymal transition, TGFβ, transforming growth factor beta.

Figure 5

Quantitative analysis of the steady states associated to single-node perturbations in the EMT network model supports the existence of an EMT spectrum. Single-node knockouts and constitutive activations were performed in the TGFβ-driven EMT network model. The steady states from each of these perturbed networks were projected onto the epithelial and mesenchymal steady states from the unperturbed EMT network (a), and onto the first and second principal components obtained by principal component analysis (b). Each steady state is represented by a square whose color denotes the presence (yellow) or absence (blue) of E-cadherin. Steady states that have the same coordinates when projected in the epithelial/mesenchymal plane but differ in the node state of E-cadherin are represented by a square with both yellow and blue color. The background color of the marked groups denotes the type of steady state (epithelial-like, yellow; hybrid-like, green; or mesenchymal-like, blue). Groups of states are labeled by the node whose knockout or overexpression lead to this state; the abbreviation ‘TFs’ in the group labels represents any of the transcription factors FOXC2, HEY1, TWIST1, SNAI1, SNAI2, ZEB1, or ZEB2. (a) In the epithelial/mesenchymal plane, the epithelial and mesenchymal states corresponds to (1, 0) and (0, 1) in the x–y plane, respectively. Quantitative analysis of the steady states in single-node perturbed models reveals a spectrum of steady states, many of which are intermediate to the unperturbed epithelial and mesenchymal steady states. (b) In the principal component plane, the epithelial steady states from the unperturbed EMT network model cluster in the bottom right corner of the plot and the mesenchymal steady states, along with numerous other perturbed states, cluster in the bottom left corner of this plot. Numerous other steady states exist in distinct and intermediate clusters between the epithelial and mesenchymal clusters. EMT, Epithelial-to-mesenchymal transition, TGFβ, transforming growth factor beta.