Evidence for and against epithelial-to-mesenchymal transition in the liver

Abstract

The outcome of liver injury is determined by the success of repair. Liver repair involves replacement of damaged liver tissue with healthy liver epithelial cells (including both hepatocytes and cholangiocytes) and reconstruction of normal liver structure and function. Current dogma posits that replication of surviving mature hepatocytes and cholangiocytes drives the regeneration of liver epithelium after injury, whereas failure of liver repair commonly leads to fibrosis, a scarring condition in which hepatic stellate cells, the main liver-resident mesenchymal cells, play the major role. The present review discusses other mechanisms that might be responsible for the regeneration of new liver epithelial cells and outgrowth of matrix-producing mesenchymal cells during hepatic injury. This theory proposes that, during liver injury, some epithelial cells undergo epithelial-to-mesenchymal transition (EMT), acquire myofibroblastic phenotypes/features, and contribute to fibrogenesis, whereas certain mesenchymal cells (namely hepatic stellate cells and stellate cell-derived myofibroblasts) undergo mesenchymal-to-epithelial transition (MET), revert to epithelial cells, and ultimately differentiate into either hepatocytes or cholangiocytes. Although this theory is highly controversial, it suggests that the balance between EMT and MET modulates the outcome of liver injury. This review summarizes recent advances that support or refute the concept that certain types of liver cells are capable of phenotype transition (i.e., EMT and MET) during both culture conditions and chronic liver injury.

epithelial-mesenchymal transition (EMT) defines a biological process in which adherent epithelial cells acquire mesenchymal characteristics and become more migratory/invasive (29). The reverse process of EMT is mesenchymal-epithelial transition (MET), which involves a transformation of mesenchymal cells to acquire epithelial traits. Both EMT and MET involve complex molecular and cellular events that ultimately change the gene expression and phenotype of cells. Thus they demonstrate the inherent plasticity of cells that are capable of EMT/MET (11, 46, 67, 68).

Epithelial cells with apical-basolateral polarity usually bind to each other by tight intercellular junctions and form plates. During EMT, epithelial cells eventually lose these key features by changing the expression and distribution of proteins that mediate cell-cell contacts and reorganizing the cytoskeletal elements responsible for normal epithelial polarity. In addition, these epithelial cells start to gain mesenchymal characteristics including the capacity to migrate and invade the extracellular matrix; they also switch to rear-to-front polarity and begin to express distinct mesenchymal markers. This newly formed migratory/invasive phenotype is normally accompanied by induction of mesenchymal filaments, rearrangements of actin cytoskeletal proteins, increased production of enzymes that degrade extracellular matrix, and altered expression of specific microRNAs (15, 29, 39, 94). Such comprehensive changes in cell phenotype do not take place instantaneously. Instead, EMT/MET is composed of a carefully orchestrated series of events modulated at both transcriptional and posttranscriptional levels (77, 78). The term “partial EMT” is used to distinguish cells that have not yet completed EMT from those that have undergone a “full EMT.” Cells in partial EMT are in the intermediate stages of EMT, and as such, continue to express epithelial markers although mesenchymal markers have already been acquired (29). Partial EMT may occur in hepatocellular carcinomas (HCCs) and during liver fibrosis.

EMT has been categorized into three different subtypes based on the biological context in which it occurs: development and organogenesis (type 1); wound healing, tissue, regeneration and organ fibrosis (type 2); and carcinogenesis (type 3) (28, 29, 94). Detailed description of these three subtypes of EMT can be found in many reviews articles (1, 29, 94). In brief, type 1 EMT occurs during implantation, embryo formation, and organ development. It is responsible for the creation of mesoderm, endoderm, and neural crest cells. During embryonic development, the primitive epithelium gives rise to primary mesenchyme through EMT, which can be reinduced to form secondary epithelia by MET and may undergo further differentiation to generate different organs/tissues via subsequent cycles of EMT/MET (29). Type 1 EMT does not induce fibrosis. In contrast, type 2 EMT has been classically associated with fibrosis in adult organs such as kidney, liver, lung, heart, and intestine (29, 38). Type 2 EMT usually occurs during wound healing and organ regeneration. It is typically mediated by inflammatory signals and generally halts when repair is accomplished and inflammation diminishes. However, under persistent injury and inflammation, type 2 EMT can continue to generate myofibroblasts that accumulate and cause progressive fibrosis, which may lead to organ destruction. In recent years, the occurrence of type 2 EMT in vivo has been seriously challenged (33, 38, 60). Type 3 EMT occurs in cancer epithelial cells, which undergo genetic and epigenetic changes and thus are able to invade and metastasize via the circulation, thereby generating distant metastatic tumor. Type 3 EMT resembles type 1 EMT in that generation of epithelia, rather than fibrosis, is typically the ultimate outcome (15).

The focus of this review paper will be type 2 EMT in the context of liver disease, especially liver fibrosis. We will summarize major evidence for or against the concept that certain types of liver cells can undergo EMT/MET in vitro and in vivo and begin by discussing some of the signaling pathways that regulate EMT/MET in the liver.

Signaling Pathways Governing EMT

Many signaling pathways have been documented to induce EMT/MET in embryonic development, normal and transformed cell lines, organ regeneration/fibrosis, and/or cancer progression. These signaling pathways include transforming growth factor-β (TGF-β) superfamily, sonic hedgehog (Hh), Wnt, Notch, epidermal growth factor (EGF), fibroblast growth factor (FGF), platelet-derived growth factor (PDGF), and many others (77, 89). A detailed discussion of these pathways is beyond the scope of this review. Rather we will focus on those signaling pathways that have been studied in liver-associated EMT.

TGF-β signaling is considered as the master regulator of EMT (89, 93) and has been shown to induce EMT in various types of cultured nonmalignant and malignant epithelial cells, including hepatocyte and cholangiocyte (5, 27, 39, 42, 87). Binding of TGF-β1 with type I and II TGF-β receptors induces phosphorylation of Smad2/3. Phosphorylated Smads recruit Smad4 and translocate to the nucleus where Smad complexes control transcription of several target genes, such as SNAI1 (snail), SNAI2 (slug), and twist, to further suppress epithelial genes and induce mesenchymal genes (27, 93). Besides Smad signaling, TGF-β has also been shown to promote EMT through activation of mitogen-activated protein (MAP) kinase, Rho GTPases, and phosphatidylinositol 3 (PI3)-kinase (86). Moreover, a member of the TGF-β superfamily that antagonizes TGF-β signaling, bone morphogenetic protein (BMP)-7, negatively regulates TGF-β-induced EMT in different types of organ injury, including liver fibrosis. Both pharmacological administration and genetic expression of BMP-7 have been shown to inhibit TGF-β-provoked EMT and thus attenuate liver fibrosis in animals with carbon tetrachloride (CCl₄)-induced liver injury (21, 95).

Hh signaling is a key morphogenic pathway that controls fetal liver development and also plays an essential role in many types of adult liver injury (16, 54). Soluble Hh ligands (Sonic Hh, Indian Hh, and Desert Hh) secreted by Hh-producing cells interact with Patched (Ptc) receptor on Hh-responsive cells. This relieves the Ptc coreceptor, Smoothened (Smo), from Ptc-mediated repression. Activated Smo, in turn, transduces intracellular signals that culminate in the nuclear localization of Glioblastoma (Gli) family transcription factors (Gli1, Gli2, and Gli3). Hh signaling promotes EMT during fetal development and during metastasis of several adulthood cancers (40). There is growing evidence that the Hh pathway may also regulate EMT-like transitions in certain types of liver cells during adult liver injury. The most extensive data supporting this concept have been acquired by studies of hepatic stellate cells (HSC). Although HSCs are not generally considered to be epithelial cells, several recent publications demonstrate that at least a subpopulation of HSCs has both mesenchymal and epithelial features (which will be discussed in more detail below) (36, 49). Both pharmacological and genetic inhibition of Hh signaling have been shown to block an EMT-like transition that occurs as HSCs become myofibroblasts, reverting the cells to a more quiescent/epithelial phenotype (17, 49). Similar findings have been reported when Hh pathway activity is manipulated in cholangiocytes and malignant hepatocytes (12, 55). This collective evidence suggests that, like TGF-β, Hh signaling also modulates EMT/MET in cells that are involved in adult liver repair and regeneration.

Notch signaling is another highly conserved cell pathway that becomes activated during both development and adult tissue repair. It is a complex pathway, as evidenced by the fact that multiple Notch receptors (Notch 1–4) and ligands (Jagged and Delta-like) exist. Upon ligand binding, proteolytic cleavage of the Notch receptor occurs. This releases the Notch intracellular domain (NICD) from the cell membrane, and permits its translocation to the nucleus. In the nucleus, the NICD forms a transcription complex with DNA-binding protein RBP-J and induces the transcription of downstream target genes, including genes in the hairy and enhancer of split (Hes) family and hairy/enhancer-of-split related with YRPW (Hey) family (34). A role for Notch in EMT has been documented in embryogenesis, adult tissue repair, and carcinogenesis (7, 8, 30, 64, 79). More recently, evidence for Notch signaling in EMT has also been demonstrated in the liver. Notch inhibition by a γ-secretase inhibitor, DAPT, blocked an EMT-like transition in both primary mouse HSCs and a rat HSC cell line. The suppression of HSC activation ultimately attenuated CCl₄-induced rat liver fibrosis (13, 85). Similarly, treating cholangiocytes with this γ-secretase inhibitor, or with Jagged1-neutralizing antibody, to block Notch signaling also diminished an EMT phenotype (43, 85). Conversely, overexpressing Notch1 induced EMT in a cholangiocarcinoma cell line (100). These recent findings suggest an important role of Notch signaling in EMT during liver injury.

Hypoxia and redox stress are common events during chronic liver disease progression. Indeed, hypoxia-related signaling was recently documented to promote type 3 EMT during the development of liver cancer. Hypoxia-inducible factor (HIF) is the major transcription factor responsible for hypoxic responses. Zhang et al. (97) reported that increased HIF-1α levels correlated with loss of the epithelial marker E-cadherin but increased expressions of the EMT-associated transcriptional factor SNAI1 and the mesenchymal markers N-cadherin and vimentin in human liver cancer samples. They extended these in vivo findings by studying two HCC cell lines and showed that hypoxia-stabilized HIF-1α promoted EMT by increasing SNAI1 transcription (97). Wnt/β-catenin signaling might further enhance hypoxia-induced EMT in HCC cell lines by cross-talking with HIF-1α (98). Hypoxia-induced EMT in HCC cells is also regulated by PI3-kinase/AKT pathway (23, 88). In addition, reactive oxygen species (ROS) produced during hypoxia may play a role in liver cancer-related EMT because attenuation of ROS by antioxidants suppressed TGF-β-induced EMT in HCC cells (31).

Besides the aforementioned pathways, EGF, PDGF, and other pathways have also been reported to regulate EMT in the context of liver disease. More importantly, these pathways cross talk with each other and thus form a complex network to control progression of EMT/MET and determine the cell fate. These signaling pathways usually induce and activate several EMT-related transcription factors such as SNAI1, SNAI2, Twist, zinc finger E-box binding homeobox 1 (ZEB1), Krüppel-Like Factor 8 (KLF8), Goosecoid, and fork-head box protein C2 (FOXC2) (29). Upregulation of these transcription factors can repress expression of E-cadherin and other junction proteins and thus promote EMT and biological functions mediated through EMT. For example, Rowe et al. (62) reported that specific deletion of hepatocyte derived SNAI1 reduced CCl₄-induced liver fibrosis and inflammatory responses. Several groups demonstrated that Twist prompted HCC cell invasion, migration, and vasculogenic mimicry (71, 91).

Challenges in Identifying Cells Undergoing EMT

EMT is a transition process and cells that are in the intermediate stage of EMT are sometimes referred to as “transitioning cells” or cells that have undergone “partial EMT” (29, 94). Coexpression of epithelial and mesenchymal markers is often used as a criterion to identify epithelial cells that are becoming mesenchymal (i.e., undergoing EMT) at a time when they retain some of their epithelial-specific features. Typical epithelial markers usually include E-cadherin and cytokeratin, whereas common mesenchymal markers are N-cadherin, α-smooth muscle actin (α-SMA), collagen 1α1, vimentin, desmin, fibroblast-specific protein 1 (FSP1, also known as S100A4), SNAI1, SNAI2, and twist (94). However, it is important to keep in mind that an epithelial cell that is undergoing EMT may have not yet fully activated expression of mesenchymal genes. In addition, not every mesenchymal marker is expressed concomitantly (for example, quiescent HSCs express large amounts of desmin, but minimal levels of α-SMA and collagen); some mesenchymal markers, such as FSP1, may not be expressed by all types of mesenchymal cells (e.g., liver myofibroblasts) (56); and some mesenchymal markers lack specificity (e.g., vimentin) (24). Moreover, it is difficult to determine colocalization of two markers by immunohistochemistry. All these factors confound our efforts to identify a cell as being (or not being) in the midst of an EMT, especially in whole tissues. An additional challenge that we need to overcome is the fact that EMT and MET are reversible processes. Hence it is possible for a mesenchymal-type cell that was derived from an epithelial cell to regain its epithelial traits (1, 15).

To overcome the limitations of immunostaining, lineage tracing has been recently used to track cells undergoing EMT/MET. This strategy exploits cell-type-specific activation of gene-regulatory elements (usually based on the cre-lox system) to generate permanently expressed markers (e.g., LacZ or a fluorescent-labeled protein) that specifically label all the progeny of that cell type (37). Despite the enormous power of this technique, the utilization of this approach to study EMT/MET has generated conflicting evidence, as will be discussed in more detail below.

Evidence That Specific Liver Cell Types Can Undergo EMT/MET in Culture

Several types of liver cells (including hepatocytes, cholangiocytes, HSCs, and liver progenitor cells) have been shown to undergo EMT/MET or EMT/MET-like transition during culture. Multiple groups demonstrated that treating hepatocytes (either primary cells or cell lines) with TGF-β induced them to downregulate epithelial markers, such as albumin and E-cadherin, and upregulate mesenchymal markers, including SNAI1, α-SMA, collagen, and FSP1, and/or to gain migratory capacity (14, 48, 51, 95, 96, 102). This was further confirmed in primary cultures of human fetal hepatocytes by showing that TGF-β treatment induced SNAI1 and decreased E-cadherin expression, as well reorganized the actin cytoskeleton (9). Besides TGF-β, other proteins or chemicals, such as hepatitis C viral protein and organochlorine pesticides, have also been reported to induce EMT in hepatocytes (2, 103). Meyer et al. (47) recently also reported that hepatocytes can intrinsically undergo dedifferentiation in collagen monolayer culture, as evidenced by change in morphology and upregulation of mesenchymal markers (N-cadherin, vimentin, and collagen 1α1). However, they showed that this phenomenon does not reflect a classical TGF-β-mediated EMT because SNAI1 and E-cadherin were not involved. Hence, in hepatocytes, EMT mechanisms appear to depend on the specific trigger (cytokines, environment, etc.), and this will also determine the markers that can be used to characterize the particular EMT program.

Cholangiocytes (liver ductular cells) are another type of liver epithelial cell that has been shown to undergo EMT in the culture. Omenetti et al. (55) reported that cholangiocytes isolated from rats with biliary fibrosis induced by bile duct ligation (BDL) upregulated expression of FSP1 and downregulated expression of aquaporin-1, cytokeratin 7 (Krt7), and cytokeratin 19 (Krt19). Furthermore, they showed that an immature cholangiocyte line cocultured with myofibroblastic HSCs (MF-HSCs), or treated with conditioned medium from MF-HSCs, was induced to undergo complete EMT (i.e., repression of epithelial gene expression, induction of mesenchymal gene expression, and acquisition of a migratory phenotype). In addition, they demonstrated that blocking Hh signaling by use of a Hh ligand-neutralizing antibody blocked EMT in the cholangiocytes that had been treated with MF-HSC conditioned medium (53, 55). Several other groups also reported that cultured primary human/rodent cholangiocytes induced mesenchymal markers and/or became highly motile when treated with TGF-β (18, 26, 63, 65). Thus, as with hepatocytes, there is also strong evidence that ductular cells are capable of EMT in vitro.

HSCs are liver-resident pericytes that become myofibroblastic during many types of liver injury. As discussed below, published data indicate that HSCs express both epithelial and mesenchymal genes. Moreover, growing evidence supports the concept that HSCs are transitional cells with inherently high plasticity. These findings suggest that HSC might be capable of epithelial-mesenchymal/mesenchymal-epithelial-like transitions. This possibility is supported by the putative origin of HSC. At least three independent groups have reported that HSCs are derived from mesothelial/submesothelial cells during both liver development and adult liver injury (3, 4, 41, 44, 59). During development, mesothelial cells are derived from the epiblast (a tissue that gives rise to all three germ layers). Mesothelial cells express both mesenchymal and epithelial cell markers (50). Therefore, Choi et al. (15) suggested that quiescent HSCs might be transitional cells derived from epithelial cells that have undergone a partial EMT. Consistent with that concept, Kordes et al. (36) showed that a subpopulation of adult primary rat HSCs that expressed the progenitor marker, CD133, could be induced to become either myofibroblastic, or hepatocytic cell under different culture conditions. These observations suggest that some adult HSCs (which are generally considered to be mesenchymal-type cells) can undergo MET and regain epithelial features. During this MET-like process, HSCs-derived hepatocytic cells were demonstrated to express α-fetoprotein (AFP), a marker of immature hepatocytes, and albumin, a marker of both hepatoblasts and hepatocytes. These findings were subsequently confirmed in another laboratory in vivo by lineage tracing studies, which will be discussed subsequently (49, 90). Other published data also support the concept that liver HSCs can transition to become liver epithelial cells. Sicklick et al. (70), for example, reported coexpression of mesenchymal genes and epithelial markers (i.e., AFP and Krt19) in two clonally derived MF-HSCs cell lines. Later on, Choi et al. demonstrated that freshly isolated rat HSCs expressed some epithelial markers, including E-cadherin, Krt7, and Krt19, which were downregulated during HSC activation. Moreover, they reported that Hh inhibition restored the epithelial gene expression in cultured HSCs, thus suggesting a regulatory role for Hh in adult liver EMT (17). More recently, Michelotti et al. (49) used multiple approaches including flow cytometry, immunocytochemistry, in situ hybridization, and PCR analysis to demonstrate coexpression of epithelial markers (such as Krt19, albumin, AFP, Krt18) and mesenchymal markers (such as desmin, α-SMA), as well as some progenitor markers (such as Sox9, Nanog, Oct4), in quiescent and myofibroblastic primary mouse HSCs. They were also able to confirm those findings in two other clonal MF-HSCs cell lines, including LX-2, a widely used human MF-HSC cell line. However, because freshly isolated primary HSC already coexpress epithelial and mesenchymal markers, a complete EMT does not occur during their differentiation/activation to a more myofibroblastic phenotype in culture. Similarly, EMT inhibitors are not able to fully extinguish mesenchymal gene expression in cultured HSC, despite reverting the cells to a more epithelial phenotype. Thus we use the term “EMT/MET-like transition” when describing HSC plasticity to distinguish the process from traditional EMT/MET.

The concept that both cholangiocytes and HSCs are able to undergo an EMT/MET-like transition is intriguing because at least subpopulations of both cell types have been suggested and documented to contribute progenitor populations in the liver (10, 35, 66, 72, 76). Conigliaro et al. (19) recently reported that epithelial and mesenchymal liver cells (hepatocytes and HSCs) may arise from common progenitor isolated from embryonic livers. Their study showed that sca+ (a stem marker) murine progenitor cells coexpressed markers of epithelial and mesenchymal lineages and were able to trans-differentiate into both hepatocytes and HSCs under culture conditions after few passages. They further demonstrated that when these clonally derived progenitor cells were transplanted into normal liver they gave rise to both hepatocytes and HSCs in vivo, confirming that the in vitro findings were not mere artifacts of cell culture. This evidence that HSCs and liver epithelial cells derive from a common, multipotent progenitor is also supported by an earlier study that showed that hepatic progenitor cells (oval cells) coexpressed epithelial and mesenchymal markers and demonstrated that transplantation of these progenitor cells could repopulate injured livers (92).

Thus there seems to be a fairly good amount of evidence showing that mature hepatocytes and cholangiocytes can transiently express markers of mesenchymal cells in vitro and that adult HSCs can also be induced to undergo EMT/MET-like transition under certain culture conditions. Moreover, some liver progenitor cells might be capable of EMT, both in vivo and in vitro, permitting them to repopulate the liver (Fig. 1). However, in vitro conditions do not necessarily always reflect situations that occur in vivo. Thus cell culture data cannot always be extrapolated to the in vivo conditions. Also, because cells that undergo EMT usually generate transitioning cell populations, it has been challenging to prove (or disprove) that any of these cell types actually undergoes EMT (or MET) during chronic liver injury. Consequently, it remains unclear/extremely controversial whether (and how) epithelial-mesenchymal transitions of resident liver cells might be involved in liver repair.

Fig. 1.

Resident liver cells are capable of undergoing epithelial-to-mesenchymal transition (EMT) and mesenchymal-to-epithelial transition (MET) under certain culture conditions. Hepatocytes, cholangiocytes, oval cells, and quiescent hepatic stellate cells (Q-HSCs) express epithelial markers such as E-cadherin, cytokeratin, and tight junction proteins. Some of these cells can undergo an EMT-like process and trans-differentiate into cells with a more mesenchymal phenotype [characterized by expression of N-cadherin, α-smooth muscle actin (α-SMA), collagen 1α1, vimentin, and/or desmin]. Some liver mesenchymal cells (e.g., quiescent and myofibroblastic hepatic stellate cells) can also differentiate into liver epithelial cells by MET. These processes are regulated by signaling pathways that become activated during liver injury, such as TGF-β, Hedgehog, Notch, and hypoxia-inducible factor.

Evidence For and Against the Occurrence of EMT in Adult Liver Injury

As mentioned earlier, it is inherently difficult to determine in situ whether a cell is undergoing EMT/MET. Early studies mostly used immunohistochemistry to assess the role of EMT in adult liver repair. There are several flaws/disadvantages of using this technique (15). First, it is technically challenging to demonstrate expression of several protein markers by any given cell at any point of time. Thus it is almost infeasible to acquire the knowledge of the comprehensive changes in the phenotype of single cells by this method. In addition, it is often difficult to determine by costaining whether one cell expresses both epithelial and mesenchymal markers. It is possible that the apparent coexpression of both markers simply reflects the fact that an adjacent cell expresses one of the markers. Last but not least, EMT/MET is a transitional process (rather than a static one), and even superb immunohistochemistry is incapable of capturing cell transition, which is generally considered as a critical criteria that a cell is undergoing EMT/MET.

Nevertheless, some useful information about EMT in liver injury has been obtained by studying human/animal samples with immunohistochemistry. For example, Rygiel et al. (63) reported that bile ductular cells coexpress ductular markers (Krt7 or Krt19) and FSP1, a mesenchymal marker, in human primary biliary cirrhosis, primary sclerosing cholangitis, and alcoholic liver disease. Similar findings were also reported by other groups who evaluated liver samples from patients with primary biliary cirrhosis and nonalcoholic fatty liver disease (55, 61, 73). Moreover, costaining for albumin (hepatocyte marker) and FSP1 was demonstrated in CCl₄-induced murine liver fibrosis (95). Dooley et al. (21) also observed coexpression of transferrin (specific marker for hepatocyte) and collagen or SNAI1 in hepatitis B virus-infected patients liver samples.

Lineage tracing is becoming a popular and powerful tool to overcome the limitations of immunostaining for identifying EMT. This technique genetically labels cells. Hence, the marker will be present in any progeny of the labeled cells. Zeisberg et al. (95) were the first to utilize this technique to suggest that hepatocyte-mediated EMT might provide an important source of fibrogenic myofibroblasts. These authors generated double transgenic (DTG) mice by crossing Albumin-Cre mice with Rosa26-floxstop-LacZ mice. In Albumin-Cre mice, expression of Cre-recombinase is under the control of albumin promoter. In Rosa36-floxstop-LacZ mice, the LacZ reporter gene is activated only after Cre-mediate excision of the floxed exon. Hence in the DTG mice only albumin(+) cells and their progeny were permanently marked with β-galactosidase. These mice were then examined to determine whether β-galactosidase ever colocalized with FSP1, a putative marker of fibroblasts. In healthy DTG mice, few FSP1-positive cells were detected. However, CCl₄ treatment induced hepatic FSP1 expression and almost half of the FSP1 positive cells coexpressed β-galactosidase. The authors, therefore, concluded that hepatocyte-derived fibroblasts are an additional and significant lineage of mesenchymal cells that contribute to progression of liver fibrosis (95).

The second attempt to use lineage tracing to demonstrate EMT in adult liver injury bred glial fibrillary acidic protein (GFAP)-Cre mice with floxStopRepressorflox green fluorescent protein (GFP) transgenic mice (90). The resultant DTG mice expressed Cre-recombinase exclusively in cells that have activated transcription of GFAP. GFAP is a marker of HSCs and these DTG mice could be used to track the progeny of HSCs to determine whether HSCs undergo MET to generate mature liver epithelial cells. After methionine-choline-deficient, ethionine-supplemented diets, roughly one third of the mature-appearing albumin-positive hepatocytes and almost all of the liver ductular cells of these DTG mice expressed GFP. These data raised the intriguing possibility that hepatocytes, cholangiocytes, and HSCs are derived from common progenitors that are capable of EMT/MET during certain types of liver injury (15).

However, these two reports were seriously challenged by several lineage tracing studies that were published around 2010. Taura and coworkers (75) from Brenner's group bred collagen 1α1-GFP reporter mice with the Albumin-Cre-Rosa26-floxstop-LacZ-expressing reporter mice strain used by Zeisberg to simultaneously identify collagen-producing cells and cells derived from albumin(+) cells in the injured liver. However, unlike Zeisberg, they found no colocalization of FSP1 and β-galactosidase. Moreover, they were unable to detect any hepatocyte-derived collagen-expressing cells at different stages of CCl₄-induced liver fibrosis.

Later, the same group generated multiple DTG mice to study EMT of cholangiocytes, and MET of HSCs, during liver injury (69). First Scholten et al. (69) crossed tamoxifen-inducible Krt19-CreERT mice with Rosa26f/f-YFP mice to trace the fate of cholangiocytes. In both BDL- and CCl₄-induced liver fibrosis, the authors found no evidence of HSC/myofibroblast markers overlapping with the yellow fluorescent protein (YFP)-expressing Krt19⁺ progeny of cholangiocytes. They complemented these data by experiments with animals in which quiescent HSCs or myofibroblastic cells were labeled permanently by crossing mice that expressed Cre-recombinase under the control of GFAP or collagen α2(I) promoter with Rosa26f/f-mT/GFP or Rosa26f/f-YFP reporter mice, but they were unable to detect any colocalization of YFP+ cells with E-cadherin or pan-cytokeratin (a marker for cholangiocyte) either. Thus the authors concluded that EMT in cholangiocytes and MET in HSCs do not contribute to experimental hepatic fibrosis or liver regeneration.

A third study from the same group by Osterreicher et al. (56) demonstrated that FSP1 was not expressed by HSCs or type I collagen-producing fibroblasts. Rather, they found that FSP1-positive liver cells expressed macrophage markers, leading the authors to conclude that classical liver myofibroblasts do not express FSP1. Consistent with that concept, when FSP1-Cre mice were crossed with ROSA26-YFP mice, no colocalization of YFP with desmin or α-SMA was observed in either BDL- or CCl₄-induced liver injury. This work challenged the reliability of FSP1 as a mesenchymal marker for studying EMT in the liver.

The fourth study that provided another piece of strong evidence against EMT during liver injury was performed by Chu et al. (18). These authors generated DTG mice by crossing AFP-Cre mice and Rosa26-YFP mice to tag all the epithelial cells in the liver (including hepatocytes, cholangiocytes, and oval cells) with YFP. Although they were able to detect EMT in cholangiocytes in vitro, they found no evidence that YFP colocalized with various putative mesenchymal markers, including FSP1, vimentin, α-SMA, or procollagen 1α2, in three different hepatic injury models. Hence the authors concluded that EMT by hepatocytes or cholangiocytes did not occur in vivo. Since this study was performed by a different group, Kisseleva and Brenner (33) concluded that EMT does not occur in vivo and has stated that “hepatic epithelial cells do not contribute to experimental liver fibrosis.” Later on, this claim was also supported by the work of Troeger et al. (80) reporting that vimentin-CreER marked myofibroblasts did not undergo MET.

Surprisingly, the battle over EMT has not ended. More recently, Michelotti et al. (49) reported that HSCs were capable of differentiating into hepatocytes and ductular cells by using the same type of lineage tracing techniques described above. In their recent article (49), α-SMA-Cre-ER^T2 or GFAP-Cre-ERTM mice were crossed with Rosa-Stop-flox-YFP mice to trace the fate of HSCs after liver injury. After BDL, three types of YFP-positive cells were observed in fibrotic liver: stromal, hepatocytic, and ductular types. To confirm this finding, hepatocytes were isolated from these DTG mice after BDL and the YFP expression was evaluated directly by flow cytometry without using antibody. Interestingly, they found about 24–34% hepatocytes were YFP positive, consistent with a previous study that also showed a similar percentage of hepatocytes came from HSCs in injured livers (90). Moreover, they analyzed hepatocyte DNA by PCR to examine rearrangement of the Rosa26 locus and demonstrated that Cre-mediated recombination did occur in the tamoxifen-treated group. This piece of direct evidence for transgene rearrangement had not been provided by any of the previous studies that claimed to refute the existence of EMT during liver injury. Evidence for EMT/MET has also been observed recently by the same group using a partial hepatectomy (PHx) model (72a). This study found that, in tamoxifen-treated α-SMA-YFP mice, many progenitors, cholangiocytes, and up to 25% of hepatocytes were YFP+ by 48–72 h after PHx, indicating that liver epithelial cells were derived from α-SMA-YFP+ cells. What is particularly intriguing about these last two studies is that they identify HSCs (pericytes in the liver) as a resident population of inherently plastic cells which can be reprogramed by MET-EMT to replace adult liver epithelial cells, resembling Weinberg's “transitional breast cells” which coexpress epithelial and mesenchymal markers and are postulated to replenish breast cancer stem cell pools (25, 67).

It is difficult to reconcile all of the aforementioned contradictory results for and against the concept that liver cells undergo EMT in vivo (as summarized in Fig. 2). The fact that the conflicting data were generated by studying the same fibrosis models makes interpretation even harder. We will try to explore a few possibilities that may help to explain the discrepancies. First, like all other techniques, lineage tracing also has some pitfalls. The efficiency of Cre-mediated recombination is not perfect. For example, Scholten et al. (69) reported only 40% efficacy of Cre-recombination in their Krt19-Cre/YFP mice, so it is quite possible that EMT might have occurred in the nonlabeled Krt19⁺ cells. In fact, in an early study by Brenner and colleagues (45) that used DTG α-SMA/collagen 1α1 reporter mice to determine the relationship between α-SMA-expressing cells and collagen production following liver injury, around 50% of HSCs did not express either transgene whereas 7% only expressed α-SMA-red fluorescent protein, 14% only expressed collagen-enhanced green fluorescent protein, and 30% expressed both transgenes after 5 days in culture. Second, some of the putative “cell type-specific” markers used to create DTG mice are not really specific. For example, GFAP (a classical HSC marker) has also been reported to be expressed by ductular cells (90). Third, the liver injury models were examined at different, but limited numbers of, time points in different studies. Since EMT/MET are transitional processes, investigators who did not detect EMT/MET may have missed the window when cells were undergoing phenotypic transition, or the cells may have already finished transition and may even have transitioned back to their basal state. Fourth, lineage tracing still heavily relies on double immunohistochemistry to determine whether a cell coexpresses certain markers to decide whether it is undergoing EMT/MET. Consequently, the limitations of immunohistochemistry still apply to this advanced technique. In fact, employing only a few epithelial and mesenchymal markers might not be sufficient to truly determine the existence of EMT. It is also possible that some markers are too weak to be detected by current techniques or they might simply not be expressed in transitioning liver cells that express other kinds of markers. Finally, data in animal models may not exactly reflect the pathophysiological progression that occurs in human liver diseases. As mentioned earlier, immunohistochemistry has generated some compelling evidence supporting the concept that EMT/MET occur during various types of human liver injury. Given this, and all of the caveats about the limitations of currently available animal data, it is premature to conclude that EMT/MET do not occur in human liver injury. To better judge the occurrence of in vivo EMT/MET, we propose a few approaches that may need to be followed in the future lineage tracing studies. First, different types of injury models and different time points during/after injury should be examined in sufficient number of animals. Second, different detection approaches (such as direct immunofluorescence, antibody-mediated immunohistochemistry, isolation of different liver cell types, and examination of Cre recombination and YFP expression by PCR, flow cytometry, etc.) should be used in combination to compensate for each approach's shortcoming. Third, proper controls (both positive and negative) should always be included.

Fig. 2.

Summary of recently published lineage tracing studies that are for or against EMT. Different lineage tracing studies discussed in this review are summarized in chronological order of publication. Genetic labeling markers (promoters), injury models, detection methods, and conclusions for each study are shown. BDL, bile duct ligation; DDC, 3,5-diethoxycarbonyl-1,4-dihydrocollidine; PHx, partial hepatectomy; HNF4α, hepatocyte nuclear factor 4α; MCDE, methionine-choline-deficient, ethionine-supplemented diet; Pan-CK, pan-cytokeratin.

EMT and Liver Carcinogenesis

Unlike type 2 EMT in fibrosis, the concept that EMT occurs in cancer is less controversial. There are fairly good amounts of evidence showing that EMT occurs in HCC and there are several review articles that discuss the role of EMT in liver cancers (52, 81). Here we will only briefly summarize some recent advances in this field. TGF-β was previously demonstrated to induce various types of HCC cell lines to acquire a mesenchymal phenotype with migratory capability (6, 22). Recently Gli1 (a Hh signaling transcription factor) and HAb18G/CD147 were found to mediate TGF-β driven EMT in HCC cells (83, 99), and HAb18G/CD147 expression in patient samples was positively correlated with mesenchymal markers and negatively related to epithelial markers (83). The positive role of Hh signaling in liver cancer cell EMT was also confirmed in a different study (12). Interestingly, the role of microRNAs (miRNAs) in HCC EMT has been rigorously studied as well. miRNA-216a/217 and miRNA-490-3p expression were reported to be upregulated in HCC tissues and also induced EMT in epithelial HCC cells (84, 101). On the other hand, several other miRNAs (such as miRNA-491, miRNA-612, and miRNA-200b) have been reported to inhibit or suppress EMT in HCC cells, and the expression of miRNA-491 and miRNA-612 was inversely associated with liver tumor differentiation and EMT (20, 74, 101). In fact, a tumor suppressor (p53) was demonstrated to suppress liver cancer EMT by upregulating miRNAs, including miR-200 and miR-192 family members (32). More information about the role of miRNAs in HCC can be found in a recent review (82). Despite all this progress that consistently demonstrates an important role for EMT in hepatocarcinogenesis, there are still many questions to be answered to definitely elucidate the role of EMT in the progression of premalignant human liver diseases, such as cirrhosis. Improved knowledge can lead to exploration of potential clinical targets for “EMT pathways” to improve the outcome of hepatocellular carcinoma.

Summary

EMT and MET were originally described as mechanisms for tissue construction during embryogenesis. Subsequently, they were suggested to play important roles during adult tissue remodeling responses, including fibrosis and cancer progression. Under certain cell culture conditions, there is no doubt that several types of resident adult liver cells and some liver progenitor cells are capable of undergoing EMT and/or MET (Fig. 1). These findings raise the possibility that EMT/MET might be involved in liver repair and regeneration. However, despite significant efforts made to determine whether EMT/MET occur during liver injury in vivo, this subject remains highly controversial and many of the data accumulated thus far are contradictory. A new concept called “escape program” or “escape reaction” was proposed to provide an alternative explanation for the complex phenotypical changes observed in EMT/MET (58, 60). This hypothesis was first described in malignant cells and suggested that such cells increase motility to escape dangers like oxidative stress, but do not fully differentiate into myofibroblasts (i.e., partial EMT might be a survival strategy) (57). Pinzani (60) introduced this concept to the liver field and he proposed that hepatocytes might also acquire a migratory phenotype during tissue injury to escape hostile microenvironment. The “escaping” hepatocytes either migrate to a less hostile microenvironment and reacquire their original epithelial phenotype, or undergo apoptosis if the hostile microenvironment persists (60). In any case, given the current level of confusion, further research is warranted to determine the significance of EMT/MET for liver repair. With the aid of more sophisticated techniques to map cell fate, as well advances in EMT research in other organs/tissues, the knowledge we gain might help us design novel diagnostic and therapeutic strategies to prevent and cure liver damage.

GRANTS

This work was supported by National Institutes of Health grant R01-DK077794 (A. M. Diehl).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

G.X. prepared figures; G.X. drafted manuscript; A.M.D. edited and revised manuscript; A.M.D. approved final version of manuscript.