Prox1 ablation in hepatic progenitors causes defective hepatocyte specification and increases biliary cell commitment

Mouse liver morphogenesis initiates at approximately embryonic day (E) 8.5, with the transition of a region of the hepatic endoderm into columnar epithelium, the onset of expression of Hnf4a, albumin and alpha-fetoprotein (Afp), and the subsequent thickening and bulging of the hepatic epithelium. By E10.0, the basal membrane surrounding the hepatic diverticulum begins to disappear; the level of E-cadherin (Cadherin 1 - Mouse Genome Informatics) expression is downregulated; and the hepatic precursors (or hepatoblasts) start to delaminate and invade the surrounding stromal tissue (Si-Tayeb et al., 2010).

Hepatoblasts are bipotent precursors that develop into either hepatocytes (the main epithelial cells in the liver) or cholangiocytes (the epithelial cells lining the intrahepatic biliary ducts). The formation of hepatocytes and cholangiocytes is temporally and spatially separated, which suggests that localized inducers or repressing mechanisms operate to direct either fate (Zaret, 2002). A network of liver-enriched transcription factors comprising six core regulators [Hnf1α, Hnf1β, FoxA2, Hnf4α, Hnf6 (Onecut1 - Mouse Genome Informatics) and Nr5a2 (also known as Lrh-1)] guides the differentiation of parenchymal hepatoblasts into hepatocytes (Kyrmizi et al., 2006). This transcriptional network is very dynamic; it involves increasing cross-regulatory interactions necessary to establish hepatocyte maturation (Kyrmizi et al., 2006).

Bile duct formation begins at ∼E14.5 in a transient periportal structure called the ductal plate. This process involves several signaling pathways, among which TGFβ/activin signaling affects cholangiocyte and hepatocyte differentiation (Clotman et al., 2005; Lemaigre, 2009). In the ductal plate, hepatoblasts activate the expression of the transcription factor Sox9 and inactivate that of Hnf4α. Asymmetric primitive ductal structures (PDSs) then emerge from the ductal plate; these PDSs comprise a periportal Sox9⁺ cell layer covered with a basal membrane rich in laminin and nidogen proteins (Shiojiri and Sugiyama, 2004) and a parenchymal Hnf4α⁺/Sox9^- cell layer lacking a basal membrane. PDSs gradually remodel to form the intrahepatic bile ducts by acquiring radial symmetry, establishing apicobasal polarity and forming a lumen. The activity of the transcription factors Sox9, Hnf6 and Hnf1β is necessary for proper biliary development, and Notch signaling is important for both bile duct remodeling and cholangiocyte differentiation (Lemaigre, 2009).

The homeobox gene Prox1 is a crucial regulator of cell differentiation and morphogenesis in various tissues (Sosa-Pineda et al., 2000; Lavado et al., 2010; Wigle and Oliver, 1999; Westmoreland et al., 2012), and it is one of the earliest markers of vertebrate hepatic development (Burke and Oliver, 2002). We previously demonstrated that the loss of Prox1 activity disrupts the delamination of hepatoblasts from the hepatic diverticulum, which hampers their migration and causes an early arrest of liver development (Sosa-Pineda et al., 2000). Prox1 is also expressed in the hepatocytes of the adult liver, with recent evidence supporting the theory that in these cells Prox1 negatively regulates the activity of ERRα (Esrra - Mouse Genome Informatics) (Charest-Marcotte et al., 2010), Hnf4α (Song et al., 2006) and Nr5a2 (Qin et al., 2004), three nuclear receptors controlling various hepatic metabolic functions. More recently, we determined that Prox1 function controls ductal cell development in a similar endoderm-derived organ, the pancreas (Westmoreland et al., 2012).

In this study, we explored the hypothesis that Prox1 activity is required for cell differentiation, morphogenesis or both in the fetal liver. By conditionally deleting the gene from hepatoblasts developed beyond the stage when they are released from the liver diverticulum, we found a novel, distinct role of Prox1 in the allocation of epithelial cell types during hepatogenesis.

We previously reported Prox1 expression in the hepatic primordium and the emerging liver bud in E9.0-E10.5 wild-type embryos (Sosa-Pineda et al., 2000). Here, we examined the distribution of Prox1 proteins in fetal and adult mouse livers using immunostaining methods. Prox1 was expressed in all hepatoblasts (Ecadherin⁺ cells) in E10.0-E12.5 livers (Fig. 1A,B). Prox1 was also detected in all hepatocytes in E15.5-E18.5 livers (Hnf4α⁺ cells; Fig. 1C-E; data not shown) and cholangiocytes (Sox9⁺ cells) forming ductal plates (Fig. 1D), PDSs (Fig. 1D,E) or bile ducts (Fig. 1E) in E15.5-E18.5 livers. In adult livers, Prox1 expression was high in hepatocytes (Fig. 1F, arrowheads) and comparatively lower in intrahepatic bile ducts (Fig. 1F, arrows). This last finding is in contrast with a report showing Prox1 expression in hepatocytes, and absence of this protein in cholangiocytes, in the liver of adult rats (Dudas et al., 2004). It is probable that variability in the sensitivity of the immunodetection methods used in each study accounted for the discrepant results.

The persistent expression of Prox1 in hepatic cells suggested that, in addition to hepatoblast delamination (Sosa-Pineda et al., 2000), other aspects of liver development require Prox1 activity. To investigate this possibility, we generated Prox1^loxP/loxP;Foxa3Cre mice (hereafter designated as Prox1^ΔLIV) carrying Prox1-specific ablation in foregut endoderm-derived tissues.

The onset of Prox1 deletion was examined in Prox1^ΔLIV livers using a combination of immunostaining, lineage tracing and qRT-PCR methods. These analyses showed similar distribution and abundance of Prox1⁺ cells between control and Prox1^ΔLIV livers at E10.5 (supplementary material Fig. S1A,B). By contrast, Prox1⁺ cells were numerous in E12.5-E18.5 control livers (supplementary material Fig. S1C and Fig. S2A,C,E) and scarce or absent in livers of Prox1^ΔLIV littermates (supplementary material Fig. S1D and Fig. S2B,D,F, arrows). The qRT-PCR results also showed that Prox1 transcripts began to diminish after E11.5 and were reduced by >90% at E13.5 in Prox1^ΔLIV livers (supplementary material Fig. S1E). Therefore, Prox1 was deleted in hepatoblasts of Prox1^ΔLIV livers after delamination occurred.

Unlike control livers (supplementary material Fig. S2A,C,E), E12.5-E18.5 Prox1-deficient livers contained numerous cellular aggregates covered with a basal membrane rich in laminin (not shown) and nidogen (supplementary material Fig. S2B,D,F). These abnormal structures were first observed at ∼E12.5 close to the hilum, but later (E15.5-E18.5) they were also detected in more peripheral areas of the mutant liver (supplementary material Fig. S2). Interestingly, although some Prox1-expressing cells remained in Prox1^ΔLIV livers, they never localized within the previous epithelial structures (supplementary material Fig. S2F). Thus, the lack of Prox1 activity promotes the formation of a basal membrane in hepatoblasts.

All Prox1^ΔLIV mice died at birth, and their livers were considerably smaller than those of control littermates (Fig. 2A,B). Compared with control livers, E18.5-postnatal day (P) 0 Prox1^ΔLIV livers displayed very unusual tissue architecture (Fig. 2C,D), prominent deposits of fibronectin in the parenchyma (Fig. 2E,F), and excessive mesenchyme (Fig. 2G,H). Microarray analyses also revealed increased expression of transcripts encoding extracellular matrix/basal membrane proteins, cell adhesion proteins and metalloproteases in Prox1-deficient livers at E14.5 (supplementary material Table S2). Thus, the multiple alterations identified in Prox1^ΔLIV livers indicate that hepatic morphogenesis requires constant Prox1 activity.

Abnormal liver morphology combined with defective hepatocyte differentiation could explain the perinatal lethality of Prox1^ΔLIV mice. Therefore, the expression of genes encoding key regulators of hepatocyte development, including the core components of the hepatocyte regulatory network (Kyrmizi et al., 2006), was examined in fetal livers following Prox1 ablation (E12.5). The qRT-PCR data showed relatively normal expression of Hnf1a, Foxa2, Nr5a2, Cebpa, Esrra, Gata6, Hhex and Tbx3 in Prox1^ΔLIV livers (Fig. 3A). By contrast, Prox1 ablation decreased the expression of Hnf4a and increased the expression of Hnf6, Hnf1b and Nr2f2 (also known as COUP-TFII) (Fig. 3A). Combined qRT-PCR and microarray results revealed deficient expression of numerous hepatocyte metabolic transcripts in Prox1-deficient livers at E12.5-E15.5 (Fig. 3C,D; supplementary material Table S3A). These data argue that Prox1 activity is necessary for proper hepatocyte differentiation.

In contrast to its effects on Hnf4a expression, Prox1 deletion significantly increased the expression of the biliary transcripts Sox9, Lamb1 and Krt19 in E12.5-E15.5 Prox1^ΔLIV livers (Fig. 3B; supplementary material Table S3B). Likewise, Cldn7 transcripts, encoding a tight junction protein, which is identified here as a novel biliary marker, were increased in Prox1-depleted livers compared with control livers (Fig. 3B; supplementary material Table S2B).

Opposite effects in hepatocyte versus cholangiocyte gene expression resulting from Prox1 deletion suggested that cholangiocyte formation increases at the expense of hepatocyte differentiation in Prox1^ΔLIV livers. To verify this hypothesis, the distribution of hepatocytes (Hnf4α⁺) and cholangiocytes (Sox9⁺) was compared between control and Prox1^ΔLIV livers using immunostaining methods. Hnf4α⁺ cells were abundant throughout the parenchyma, formed epithelial cords lacking a basal membrane, and were absent around the portal veins (Fig. 4A,C,E) in E12.5-E18.5 control livers. By contrast, Sox9⁺ cells were noticeably less numerous than Hnf4α⁺ cells and were restricted to periportal areas (Fig. 4G,I,K) in E12.5-E18.5 control livers.

Unlike control livers, Prox1^ΔLIV livers showed a gradual decrease in the abundance of Hnf4α⁺ cells (Fig. 4B,D,F). This defect was more pronounced in the parenchymal structures displaying a basal membrane (Fig. 4D,F, arrowheads). We also observed fewer cells expressing the hepatocyte marker C/EBPα in the parenchyma of Prox1-deficient livers (supplementary material Fig. S3A-D). By contrast, Sox9⁺ cells were more numerous in Prox1^ΔLIV livers than in controls at E12.5-E18.5 (Fig. 4H,J,L, arrows). Furthermore, unlike control livers, Prox1-deficient livers had Sox9⁺ cells not only around the portal veins but also in the parenchyma (Fig. 4J,L, arrowheads). These results strongly suggest that the lack of Prox1 activity diverts the specification of parenchymal hepatoblasts from hepatocytes to cholangiocytes.

Sox9⁺ cells were increased in Prox1^ΔLIV livers; therefore, we immunostained the mutant tissues to investigate whether intrahepatic bile duct formation was affected. Cholangiocytes expressing Sox9 were initially detected at ∼E12.5 in periportal areas close to the hilum (Fig. 4G) and later (E15.5-E18.5) around portal veins located in more peripheral areas in control livers (Fig. 4I,K). The few Sox9⁺ cells in periportal areas of E12.5 control livers were scattered (Fig. 4G; Fig. 5A) or started to form ductal plates (data not shown). By E15.5, most cholangiocytes formed ductal plates (Fig. 4I; Fig. 5B, arrow) or asymmetric ductal structures displaying a basal membrane on the periportal side (Fig. 5C, arrow) but not bile ducts in control livers. At this stage, cholangiocytes expressed Sox9 (Fig. 4I), high levels of Hnf1β (Fig. 5B), and claudin 7 (Fig. 5C) but not Hnf4α (Fig. 5B,C). At E18.5, we observed numerous incipient bile ducts around the portal veins of control livers expressing Sox9 (Fig. 4K; Fig. 5G,H) and claudin 7 (Fig. 5I) but not Hnf4α (Fig. 5G,I). Bile ducts of E18.5 control livers displayed apical distribution of osteopontin proteins (Fig. 5H), were surrounded by a nidogen-expressing basolateral membrane (Fig. 5G,I), and had small but well-defined lumens (Fig. 5G-I).

Unlike control livers, all Sox9⁺ cells in periportal areas of E12.5 Prox1^ΔLIV livers formed large epithelial aggregates covered with a nidogen-rich basal lamina (Fig. 4H; Fig. 5D). As early as E15.5, we observed well-developed intrahepatic bile ducts (Fig. 5E,F, arrows) around the portal veins in Prox1^ΔLIV livers. The ducts of Prox1-deficient livers expressed numerous biliary markers, including Sox9 (Fig. 4J), high Hnf1β (Fig. 5E), and claudin 7 (Fig. 5F) but not Hnf4α (Fig. 5E,F), had visible lumens (Fig. 5E,F) and expressed apical osteopontin (Fig. 5K).

In addition to forming prematurely, the intrahepatic bile ducts of Prox1^ΔLIV livers were larger than those of control livers. This defect was first noticed at ∼E15.5 (Fig. 4I,J) and was obvious at E18.5 [compare the size of the bile ducts (arrows) between control (Fig. 5G-I) and Prox1^ΔLIV (Fig. 5J-L) livers]. Quantitative proliferation analyses determined that the mitotic ratios of periportal Sox9⁺ cells did not differ between control and Prox1^ΔLIV fetal livers (data not shown). Thus, increased cholangiocyte commitment of precursors, but not enhanced cell proliferation, probably caused biliary hyperplasia in Prox1^ΔLIV livers.

The epithelial aggregates in the parenchyma of Prox1^ΔLIV livers that were surrounded with a basal membrane (supplementary material Fig. S3C,D,G,H,K,L, arrowheads), had abundant cells expressing high levels of Hnf6 (supplementary material Fig. S3G,H) and Hnf1β (supplementary material Fig. S3K,L) and only a few cells expressing C/EBPα (supplementary material Fig. S3C,D). Some of these parenchymal structures resembled bona fide bile ducts, because they expressed Sox9 (Fig. 4J,L), osteopontin and claudin 7 (Fig. 6G-I), lacked Hnf4a expression (Fig. 6F), displayed apicobasal polarity (Fig. 6H,I), and had prominent lumens (Fig. 6F, asterisks; supplementary material Fig. S3H). Most of these ductal structures were located in the hilar region (Fig. 4L) or in proximity to veins (Fig. 4J) in Prox1^ΔLIV livers. Therefore, Prox1 ablation in hepatic progenitors promotes the formation of ectopic bile ducts in the liver parenchyma.

A different class of epithelial structure was also identified in the parenchyma of E15.5-E18.5 Prox1^ΔLIV livers. These were aggregates covered with a basal membrane, harboring cells that expressed the hepatocyte markers Hnf4α and C/EBPα (Fig. 6B,C) but lacking Prox1 expression (supplementary material Fig. S2D,F). In addition to hepatocytes (i.e. Hnf4α⁺/Sox9^- cells) and cholangiocytes (i.e. Sox9⁺/Hnf4α^- cells), ‘hybrid’ cells (i.e. Sox9⁺/Hnf4α⁺ cells) were identified within the previous parenchymal aggregates of Prox1^ΔLIV livers (Fig. 6D,E). These data argue that not all parenchymal hepatoblasts fully commit to biliary cells in Prox1^ΔLIV livers.

TGFβ signaling promotes biliary differentiation (Lemaigre, 2009), and here we extended this notion by showing that TGFβ and activin A stimulate Sox9, Krt19 and Lamb1 expression in explants from E12.5 wild-type livers (Fig. 7A). Several results also indicated that TGFβ activity increases in Prox1^ΔLIV livers after E12.5. First, microarray analyses showed an increment of the TGFβ targets Sox9, Krt19, Lamb1, Tgfbi (Carey and Chang, 1998) and Gli2 (Dennler et al., 2009) in E14.5 Prox1^ΔLIV livers (supplementary material Table S4). Second, western blot analysis showed increased expression of phospho-Smad2/3 in E13.5-E14.5 Prox1^ΔLIV livers (Fig. 7B). Third, microarray and qRT-PCR analyses showed increased expression of transcripts encoding ligands and receptors of TGFβ signaling in Prox1^ΔLIV livers at E14.5-E15.5 (supplementary material Table S4; Fig. 7C) but not at E12.5 (supplementary material Fig. S4A). By contrast, no evidence was found that expression of ligands or receptors of Notch signaling, another potent inducer of biliary development (Tanimizu and Miyajima, 2004; Zong et al., 2009; Tchorz et al., 2009; Lozier et al., 2008), increase in Prox1^ΔLIV livers (supplementary material Fig. S4B; data not shown). Furthermore, although Hes1 transcripts were slightly more abundant in Prox1^ΔLIV livers than in control livers at E12.5 (supplementary material Fig. S4B), this change was very small (1.4-fold) and was probably a consequence, not a cause, of enhanced cholangiocyte commitment.

The results of microarray, qRT-PCR and in situ hybridization approaches uncovered deficient expression of transcripts encoding the activin inhibitor follistatin (Fst) (Nakamura et al., 1990) and the TGFβ inhibitor alpha-2-HS-glycoprotein (Ahsg) (Szweras et al., 2002) in E12.5-E14.5 Prox1^ΔLIV livers (Fig. 7D; supplementary material Fig. S5B and Table S4). These results suggest that in Prox1^ΔLIV livers, an excessive production of TGFβ ligands combined with deficient expression of inhibitors of this pathway contributed to increasing cholangiocyte commitment in periportal areas and helped to confer biliary cell fate to parenchymal hepatoblasts.

We previously showed that hepatoblast delamination is defective in the liver of Prox1-nullizygous embryos (Sosa-Pineda et al., 2000). This alteration prevented the migration of hepatoblasts towards the liver periphery and retained those cells close to the hilum. We investigated whether biliary development was also affected in the liver of mouse embryos with germline deletion of Prox1 (Prox1^GFP/GFP) (Srinivasan et al., 2010). All hepatoblasts (GFP⁺ cells) in the E11.5 Prox1^GFP/GFP liver (Fig. 8B-D), but not in the control liver (Fig. 8A), formed a structure covered by a nidogen-rich basal membrane, which remained contiguous to the gut epithelium (Fig. 8B,D). At this stage, cells within the mutant hepatic epithelium expressed Hnf4α (Fig. 8C) and negligibly expressed Sox9 (data not shown) or the gall bladder marker Sox17 (Fig. 8D) (Spence et al., 2009; Uemura et al., 2010).

By E14.5, Prox1^GFP/GFP livers were largely devoid of epithelial cells except in the hilar region, where a GFP⁺ epithelium covered with a basal membrane and displaying prominent lumens remained adjoined to the gall bladder (Fig. 8E). At this stage, numerous cells in the Prox1-nullizygous hepatic epithelium were Sox9⁺/Hnf4α^- (Fig. 8F) and expressed the biliary marker Dolichos biflorus agglutinin (data not shown). By contrast, the vast majority of Sox9⁺ cells in the E14.5 wild-type liver formed ductal plates but not ducts (Fig. 8F, inset). These results indicate that the entire Prox1-nullizygous hepatic epithelium evolved into a biliary structure reminiscent of the hyperplastic bile ducts of the Prox1^ΔLIV liver.

We also generated Prox1^f/f;AlbCre mice to investigate whether the formation of hepatocytes and cholangiocytes is affected when Prox1 is deleted in newly committed hepatic cells (supplementary material Fig. S6A,B). The results of qRT-PCR showed negligible changes in Prox1 expression at E13.5 and significant reduction of Prox1 expression at E19.5 (supplementary material Fig. S6E) in Prox1^f/f;AlbCre livers. Immunostaining results also corroborated that very few cells expressed Prox1 in E18.5 Prox1^f/f;AlbCre livers (Fig. 8G,J). However, in contrast to our findings in Prox1^ΔLIV livers, deleting Prox1 in committed hepatic cells (i.e. after E13.5) did not affect the expression of Hnf4α (Fig. 8G,J), nidogen (Fig. 8H,K) or Sox9 (Fig. 8I,L) in the liver. Thus, only the allocation of hepatocytes and cholangiocytes is affected when Prox1 is ablated in hepatoblasts.

Prox1 is an essential component of the regulatory network controlling early hepatic morphogenesis together with Hhex, Tbx3, Hnf6, OC-2 (Onecut2 - Mouse Genome Informatics) and Gata6 (Lemaigre, 2009). Lüdke et al. (Lüdtke et al., 2009) postulated that in this network Tbx3 acts upstream of Prox1 because, similar to our report in Prox1-null mice (Sosa-Pineda et al., 2000), the hepatic precursors of Tbx3-null embryos had reduced proliferation, and these cells did not delaminate from the gut endoderm (Lüdtke et al., 2009; Suzuki et al., 2008). Also, the loss of Tbx3 function did not affect the onset of Prox1 expression in the hepatic endoderm but failed to maintain its expression in the liver bud after E9.5 (Lüdtke et al., 2009). Our finding that Prox1 deletion in hepatoblasts did not affect the expression of Tbx3 also supports the notion that this gene is located upstream of Prox1 in the regulatory network controlling early liver morphogenesis.

In addition to a blockage in hepatoblast delamination, increased expression of genes controlling biliary development (i.e. Hnf6 and Hnf1b) and reduced expression of genes required for hepatocyte development (i.e. Hnf4a and Cebpa) was uncovered in Tbx3-null livers (Lüdtke et al., 2009). Therefore, the inability of Tbx3-null hepatoblasts to delaminate from the gut endoderm was suggested to be a consequence of their failure to initiate hepatocyte differentiation. Likewise, here we showed that Prox1-null hepatoblasts failed to delaminate but instead developed into a ductal structure in which cholangiocytes (Sox9⁺) were more abundant than hepatocytes (Hnf4α⁺). Moreover, when Prox1 was deleted in hepatoblasts post-delamination the cells rapidly deposited basal membrane proteins and formed atypical aggregates in the parenchyma, and these alterations were accompanied by increased expression of Hnf6 and Hnf1β and reduced expression of Hnf4α and C/EBPα. Therefore, our study revealed that Prox1 is a novel key regulator of processes coupling hepatic cell specification and hepatic morphogenesis.

Prox1-depleted hepatoblasts formed aggregates, rather than loose epithelial cords, that rapidly became surrounded with a prominent basal membrane. Currently, we do not know if the unusual deposition of a basal membrane around those mutant hepatoblasts was a direct effect or was secondary to the loss of Prox1 activity. However, signals downstream of TGFβ may have contributed to this phenotypic alteration, because Lamb1 expression increases in fetal liver explants treated with TGFβ.

We observed that some epithelial aggregates in the parenchyma of Prox1^ΔLIV livers contained a mixture of cholangiocytes (i.e. Sox9⁺/Hnf4a^- cells), hepatocytes (i.e. Hnf4a⁺/Sox9^- cells) and hybrid cells (i.e. Hnf4a⁺/Sox9⁺ cells). Parenchymal aggregates expressing cholangiocyte and hepatocyte markers were also reported in the livers of mouse embryos lacking Cebpa, Hnf6, Oc-2 or Hhex (Hunter et al., 2007; Clotman et al., 2005; Yamasaki et al., 2006). However, almost all cells in the epithelial aggregates of Prox1^ΔLIV livers expressed Hnf6, and similarly, the parenchymal aggregates of Hnf6/Oc-2 double knockout livers expressed Prox1 broadly (supplementary material Fig. S7). These paradoxical results underscore the complexity of interactions among the different transcription factors controlling the fate of hepatic cells. By contrast, the expression of C/EBPα rapidly decayed in Prox1^ΔLIV livers after E12.5, becoming almost negligible in numerous parenchymal aggregates at E15.5 (supplementary material Fig. S3C). These results argue that Prox1 activity may be necessary to maintain the expression of C/EBPα in hepatocyte precursors.

One major difference between Prox1^ΔLIV embryos and embryos lacking Cebpa, Hnf6, Oc-2 or Hhex (Hunter et al., 2007; Clotman et al., 2005; Yamasaki et al., 2006) is that parenchymal structures displaying features of bona fide bile ducts (i.e. expressed cholangiocyte markers but not hepatocyte markers, had apicobasal polarity, and had lumens) developed only in the liver of Prox1^ΔLIV embryos. Interestingly, these ectopic biliary structures were especially abundant towards the hilar region and in the vicinity of portal veins, which indicates that specific local cues probably played a role in their formation. We hypothesize that the periportal TGFβ gradient proposed by Clotman et al. (Clotman et al., 2005) contributed to the formation of ectopic bile ducts in Prox1^ΔLIV livers, because this signaling pathway is a known inducer of biliary development (Lemaigre, 2009; Si-Tayeb et al., 2010). Therefore, the presence of both hybrid cholangiocyte-hepatocyte aggregates and bona fide biliary structures in the parenchyma of Prox1-deficient liver may indicate that although cholangiocyte specification is the default fate of hepatic precursors, full activation of a biliary program requires specific inductive cues. In conclusion, our study demonstrates that Prox1 activity is necessary to specify the hepatocyte cell fate of liver precursors.

Prox1^ΔLIV livers have an excess of cells expressing Sox9 around the portal vein branches as early as E12.5. Two major results in our study supported that this alteration was a consequence of enhanced cholangiocyte cell commitment: (1) Prox1 ablation did not increase the proliferation of ductal plate cells or PDS and (2) deleting Prox1 in hepatic cells after E13.5 (i.e. most likely after cholangiocyte cell fate has been specified) did not result in biliary hyperplasia. However, it is unclear why bile ducts formed prematurely in Prox1^ΔLIV livers.

Intrahepatic bile duct morphogenesis is a sequential process involving the formation of a single layer of Sox9⁺/Hnf4α^- cholangiocytes (ductal plate), asymmetric periportal structures expressing Sox9 on the portal side and Hnf4α on the parenchymal side (PDSs), and symmetric tubular structures (Sox9⁺) displaying apicobasal polarity and lumens (bile ducts) (Antoniou et al., 2009). Interestingly, we found that tubular morphogenesis did not follow the same pattern in Prox1-depleted livers. Specifically, clusters of Sox9⁺ cells but not single-layered ductal plates were noticed in the periportal areas of the mutant liver at E12.5. These cellular aggregates contained a mixture of cells expressing Sox9, Hnf4α or both and were surrounded by a basal membrane. By E15.5, when most biliary structures in wild-type liver consist of ductal plates and PDSs (Antoniou et al., 2009), nearly all cholangiocytes in the periportal areas of the Prox1-depleted liver formed large tubular structures resembling more mature bile ducts. The presence of a basal membrane probably helped advance the biliary morphogenetic program, including lumen formation and acquisition of apicobasal polarity in the early periportal epithelial aggregates lacking Prox1 function, because hepatic progenitors maintained in culture in laminin-111-containing gel give rise to ductal structures lined by polarized biliary cells (Tanimizu et al., 2007). Moreover, a recent study (Tanimizu et al., 2012) showed that bile duct morphogenesis requires signals mediated by α1- and α5-containing laminins.

It is possible that newly committed cholangiocytes of Prox1^ΔLIV livers respond better to local TGFβ signals because they lack inhibitors of this pathway. Hence, increased TGFβ responsiveness in combination with other signaling pathways (e.g. Notch) could have triggered biliary morphogenesis prematurely in Prox1^ΔLIV livers. In turn, enhanced bile duct formation probably lead to excessive deposition of extracellular matrix proteins and expansion of mesenchymal cells, because in injured livers the formation of ductular structures is accompanied by deposition of collagens and expansion of fibroblastic cells (Desmet et al., 1995). Finally, increased production of TGFβ ligands probably contributed to those alterations because this cytokine is a known inducer of fibrosis in the adult liver (Bataller and Brenner, 2005).

In summary, Prox1^ΔLIV mice represent a novel animal model of intrahepatic bile duct malformation resulting from excessive commitment of biliary precursor cells (Raynaud et al., 2011; Strazzabosco and Fabris, 2012).

Loss of Prox1 function severely affected the expression of numerous hepatocyte metabolic genes, altered hepatocyte morphogenesis and disrupted hepatocyte architecture. The death of Prox1^ΔLIV mice immediately after birth further supported the hypothesis that Prox1 activity is essential for proper hepatocyte development.

One major finding of our study is that parenchymal hepatoblasts require Prox1 to initiate proper hepatocyte gene expression. Lack of Prox1 activity could alter hepatocyte transcription directly, as suggested by a recent study (Charest-Marcotte et al., 2010) showing that Prox1 binds the promoter region of various hepatocyte genes in adult livers. Interestingly, some of the proposed Prox1 hepatocyte target genes (Apoc3, Ces3, Cyp3a11, ApoH, Apoc2 and Klf15) showed reduced expression in Prox1^ΔLIV livers as early as E12.5.

Although in general hepatocyte transcripts were largely decreased in Prox1^ΔLIV livers, we identified a handful of hepatocyte metabolic transcripts (e.g. Apoa4, Cyp7a1 and Ldhb) that were upregulated when Prox1 function is absent. Two of those transcripts (Cyp7a1 and Ldhb) are known targets of hepatic nuclear receptors (NRs): Cyp7a1 was a target of Nr5a2 and Hnf4α, whereas Ldhb is a potential target of ERRα. Interestingly, protein-protein interactions between Prox1 and each of those NRs have been described in HepG2 cells or adult liver extracts (Charest-Marcotte et al., 2010; Qin et al., 2004; Song et al., 2006), with Prox1 acting as a transcriptional co-repressor in the resulting complex. This negative effect of Prox1 on NR function could explain why the expression of Cyp7a1 and Ldhb increased following Prox1 ablation. Of note, defective regulation of Nr5a2 and Hnf4α activities could have a broader impact in hepatocyte transcription, because these NRs are important core components of the hepatic transcription factor network (Kyrmizi et al., 2006).

In summary, we demonstrated that Prox1 is a novel, crucial regulator of cell differentiation and morphogenesis during hepatogenesis. In addition, we found that Prox1 activity is necessary to establish metabolic transcription correctly in hepatocyte precursors. Therefore, we propose that Prox1 is a novel component of the hepatocyte transcriptional regulatory network. Our findings bear significance to improving in vitro programming of stem cell differentiation to hepatocytes for cellular therapy of liver disease.

Prox1^floxP/+ mice and Prox1^GFP-Cre/+ mice (G. Oliver, St Jude Children’s Research Hospital, Memphis, TN, USA), Foxa3cre mice (K. H. Kaestner, University of Pennsylvania, Philadelphia, PA, USA) and AlbCre mice (B6.Cg-Tg[Alb-cre]21Mgn/J, Jackson Laboratories, Bar Harbor, ME, USA) were maintained and genotyped as described previously (Harvey et al., 2005; Lee et al., 2005; Postic et al., 1999; Srinivasan et al., 2010). Rosa26^EYFP mice (Srinivas et al., 2001) were also obtained from the Jackson Laboratory. HNF6/OC2 double-knockout mice were obtained as previously described (Clotman et al., 2005). Mice were treated according to the criteria outlined in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All experiments were reviewed and approved by the St Jude Animal Care and Use Committee and by the Ethical Committee of the Université Catholique de Louvain.

Tissues of dissected embryos or livers of newborn mice were prepared for immunohistochemical or in situ hybridization analyses as previously described (Wang et al., 2005). Paraffin-embedded tissues were processed for immunohistochemical or histological analyses as described by Westmoreland et al. (Westmoreland et al., 2009).

For immunohistochemical analysis, tissue sections were incubated in primary antibody overnight at room temperature. Supplementary material Table S1 provides a full list of the antibodies used and the experimental conditions. Images were obtained with a Zeiss Axioskop 2 microscope. Images were processed using Adobe Photoshop version 7.0 (Adobe Systems).

In situ hybridization was performed on sections, as described by Wang et al. (Wang et al., 2004). The Ahsg probe was obtained by RT-PCR using total RNA isolated from the E14.5 mouse liver and the following primers: 5′-GCTGCCTTCAACACACAGAA-3′ (forward) and 5′-ATGTCCTGTCTGCCAAAACC-3′ (reverse) to amplify a 500-bp fragment. The plasmid used to prepare the Tgfb1 RNA probe was kindly provided by H. L. Moses (Vanderbilt University, Nashville, TN, USA).

RNAs were isolated using TRIzol (Invitrogen) or an RNeasy Micro Kit for E12.5 livers (Qiagen), and cDNA was prepared using the SuperScript First-Strand Synthesis System for RT-PCR (Invitrogen). Quantitative real-time PCR (qRT-PCR) was performed on a Mastercycler realplex machine (Eppendorf). Expression levels were determined with gene-specific primers (supplementary material Table S5) and SYBR Green reagent. Gapdh expression was used to normalize gene expression levels.

Gene expression analyses were performed at the Hartwell Center for Bioinformatics and Biotechnology at St. Jude Children’s Research Hospital. The Affymetrix Mouse Genome 430 2.0 GeneChip array (Affymetrix) was used. Total RNA was prepared using TRIzol. An Agilent 2100 Bioanalyzer was used to confirm RNA quality (Agilent Technologies). Total RNA samples (5-10 μg) were processed according to the Affymetrix Gene Expression Technical Manual (2006; http://media.affymetrix.com/support/downloads/manuals/expression_analysis_technical_manual.pdf). Expression signals were calculated by the MAS5 statistical algorithm and the Affymetrix GCOS software (version 1.4), and detection calls were determined using the GeneChip Operating software (GCOS; Affymetrix). Expression profiles in mutant and wild-type samples were compared using the pair-wise method in GCOS. Microarray data have been deposited in Gene Expression Omnibus under accession number GSE36583.

Liver samples from mouse embryos were homogenized in RIPA buffer containing protease inhibitors and phosphatase inhibitors. Antibodies recognizing phospho-Smad3 or phospho-Smad2 proteins used for western blotting are listed in supplementary material Table S1.

E12.5 livers were dissected into four parts that were maintained separately in culture on Millicell-CM culture plate inserts (Millipore) as previously described (Clotman et al., 2005), with or without recombinant TGFβ or activin A (200 ng/ml of either agent) (R&D Systems). Explants were maintained in culture for 24 hours.

HepG2 cells were treated for 15 minutes with 3% paraformaldehyde in PBS, washed in PBS, and collected. Whole-cell extracts were sonicated five times for 10 seconds each at 15-μm amplitude (Sanyo Soniprep 150), pre-cleared using normal rabbit serum and salmon sperm DNA, and incubated with specific antibodies (supplementary material Table S1) overnight at 4°C. After washing the immunoprecipitates, we eluted the DNA-antibody complexes; the crosslinking was reversed, and DNA was purified by QIAquick kit (Qiagen). Quantitative real-time PCR analysis was performed using 2 μl ChIP sample per 25 μl reaction, and normalized to input DNA. The qPCR primers were verified using tenfold serially diluted DNA. Primers used to amplify the Prox1-binding element in the Fst promoter were 5′-TGTCACTGAACAGGTGTGGT-3′ and 3′ TCACCATGACTCTTGCCATC-5′.

Dlk1⁺ hepatoblasts were isolated from E12.5 wild-type mouse livers via fluorescence-activated cell sorting (FACS) per published methods (Tanimizu et al., 2003). The primary antibody was anti-Dlk1/Pref-1 (MBL #D187-3), and the secondary antibody was phycoerythrin-conjugated anti-rat IgG. RNA from FACS-sorted Dlk1⁺ cells and Dlk1^- cells was extracted using the Qiagen RNeasy Micro Kit.

All experiments were performed at least thrice. Values are expressed as mean ± s.e.m. Results were analyzed using unpaired Student’s t-test. Differences with P<0.05 were considered statistically significant.

We thank G. Oliver for providing the Prox1^loxP/+ and Prox1^GFP-Cre/+ mouse strains; L. Paul for help isolating fetal liver RNA; the Hartwell Center, the FACS Core Facility, and the Cell and Tissue Imaging Core of St. Jude; N. Shiojiri and S. Hupert for technical advice; and A. McArthur for editing the manuscript.