Potential conflict of interest: Nothing to report.
This work was supported by grants (to S.W.L.) from the National Science Council (grant nos.: NSC93-2314-B002-276, NSC94-2320-B002-092, 95-2320-B-002-054, NSC95-3112-B-002-017, 96-3112-B-002-023, and NSC98-3112-B-002-030), the National Health Research Institutes (grant nos.: NHRI-EX90-9142SI, NHRI-EX91-9142SI, NHRI-EX92-9192SI, and NHRI-EX99-9729BI), National Taiwan University Hospital, Taiwan (grant nos.: 90A01 and 91A06), and a grant (to H.P.H.) from the National Science Council (NSC 97-2321-B-002-020-MY3).
The liver architecture plays an important role in maintaining hemodynamic balance, but the mechanisms that underlie this role are not fully understood. Hepsin, a type II transmembrane serine protease, is predominantly expressed in the liver, but has no known physiological functions. Here, we report that hemodynamic balance in the liver is regulated through hepsin. Deletion of hepsin (hepsin−/−) in mice resulted in enlarged hepatocytes and narrowed liver sinusoids. Using fluorescent microbeads and antihepsin treatment, we demonstrated that metastatic cancer cells preferentially colonized the hepsin−/− mouse liver as a result of the retention of tumor cells because of narrower sinusoids. The enlarged hepatocytes expressed increased levels of connexin, which resulted from defective prohepatocyte growth factor (pro-HGF) processing and decreased c-Met phosphorylation in the livers of hepsin−/− mice. Treatment of hepsin−/− mice with recombinant HGF rescued these phenotypes, and treatment of wild-type mice with an HGF antagonist recapitulated the phenotypes observed in hepsin−/− mice. Conclusion: Our findings show that the maintenance of hepatic structural homeostasis occurs through HGF/c-Met/connexin signaling by hepsin, and hepsin-mediated changes in liver architecture significantly enhance tumor metastasis to the liver. (HEPATOLOGY 2012;56:1913–1923)
Type II transmembrane serine proteases (TTSPs) have important physiological functions and pathological implications in iron metabolism,1 blood pressure regulation, and metastasis of cancers.2 More than 20 TTSPs exist and they are divided into four subfamilies. Among these families, the hepsin/enteropeptidase subfamily is recognized structurally by a scavenger receptor cysteine-rich domain linked to a serine protease domain contained within an extracellular stem region. Although hepsin may be involved in the progression of several human cancers,3 its physiological function has not yet been fully characterized. Hepsin is predominantly expressed in the liver.4 Antisense-oligonucleotide blockade of hepsin affects cell growth and enlarges and flattens hepatoma cells.5 Several in vitro studies have identified substrates for hepsin, including coagulation factor VII,6 prohepatocyte growth factor (pro-HGF),7 and prourokinase-type plasminogen activator.8 In addition, hepsin colocalizes with desmoplakin at the sites of desmosomal junctions.9
Previously, Wu et al.10 and our group11 successfully deleted the hepsin gene (hepsin−/−) in mice and generated hepsin−/− mice that are viable, fertile, and grow normally, suggesting that hepsin is not essential for normal development. No significant differences in coagulation function or liver regeneration ability were found in hepsin−/− and wild-type (WT) littermates. Unexpectedly, a subsequent study showed that hepsin−/− mice exhibit profound hearing loss because of a developmental abnormality in the cochlear and auditory nerve.12 The molecular mechanisms underlying such phenotypes, especially those linked to the regulation of hepsin substrates and the physiological functions of hepsin in the liver, where hepsin is highly expressed, are still unclear.
Liver architecture is mostly determined by hepatocytes, which occupy 80% of the liver by volume. The plasma membranes of hepatocytes can be divided into the sinusoidal, bile canalicular, and gap junctional protein-enriched basolateral domains. The sinusoidal domains are closely associated with discontinuous endothelial cells (ECs), and thus hepatocytes are in direct contact with circulating components, and hepatocyte size is influenced by microenvironmental changes, such as hormones and oxidative stress.13 Consequently, the diameter of the sinusoids can be altered by hepatocyte size.14 The diameter of sinusoids is critical for cancer cell invasion and plays an important role in hepatic metastasis, which begins with the retention of circulating cancer cells in the liver sinusoids.15
In this study, we characterized the liver architecture of hepsin−/− mice by transmission electron microscopy (TEM) and intravital multiphoton microscopy (IVM) and employed tumor cell metastasis assays to indicate the pathophysiological significance of changes in liver architecture in hepsin−/− mice. We further elucidated a possible mechanism by which hepsin transmits signals to maintain liver architecture in vivo.
Cx26, connexin 26; Cx32, connexin 32; Cx43, connexin 43; ECs, endothelial cells; EGF, endothelial growth factor; GJIC, gap junctional intercellular communication; HGF, hepatocyte growth factor; IS, intrasplenically; IV intravenous; IVM, intravital multiphoton microscopy; MAPK, mitogen-activated protein kinase; NK4, natural killer transcript 4; PBS, phosphate-buffered saline; TEM, transmission electron microscopy; TTSPs, type II transmembrane serine proteases; WT, wild type.
Materials and Methods
Hepsin−/− mice11 were back-crossed into the C57BL/6Jnarl genetic background for >10 generations. WT mice were C57BL/6Jnarl mice (National Laboratory Animal Center, Taipei, Taiwan). Male mice were used throughout the study, unless otherwise specified. All animal experiments were approved by the Board of Animal Welfare of National Taiwan University College of Medicine (Taipei, Taiwan) and performed according to its guidelines.
Procedures were conducted as previously described16 and are summarized in the Supporting Information. IVM was used for image acquisition of mice 24 hours after hydrodynamic injection with DNA or 3 days after injection with antihepsin (200 μg/0.2 mL of phosphate-buffered saline [PBS]/mouse daily; antibody characterization is shown in Supporting Fig. 1). For treatment of mice with oleamide (Sigma-Aldrich, St. Louis, MO), the reagent was dissolved in dimethyl sulfoxide (15.6 mg/mL), diluted in olive oil, and injected intraperitoneally at 25 mg/200 μL/kg body weight once a day for 2 days before subsequent visualization. For treatment of mice with HGF or natural killer transcript 4 (NK4; a four-Kringle domain antagonist of HGF; gifts from Dr. Toshikazu Nakamura, Osaka University Medical School, Osaka, Japan), the proteins were injected intrasplenically (IS) at 2-4 μg/0.1 mL of PBS. Mice were visualized 1 hour later.
Additional methods are described in the Supporting Information.
Hepatocyte Size Is Altered in Hepsin−/− Mice.
Using TEM, we found that although the morphology of the hepatocytes from hepsin−/− mouse livers was similar to that of hepatocytes from WT mouse livers, the hepsin−/− hepatocytes were larger than those from WT mice, with a 22.6% increase in mean volume density (VV), as compared to WT (Fig. 1A). The size of hepsin−/− hepatocytes was also measured by in vivo live imaging of mice by IVM, which showed an average 27.7% increase in the cross-sectional area of the hepatocytes of hepsin−/− mice, as compared to WT hepatocytes (Fig. 1B); these results ruled out possible interference from fixation and dehydration artifacts that might alter cell size or liver architecture. IVM also revealed that the hepsin−/− mice, but not WT mice, that received hydrodynamic delivery of hepsin DNA for transient hepsin expression (Supporting Fig. 2) had a reduced hepatocyte size (Fig. 1C). Moreover, antibody blockade by intravenous (IV) injection of mice with antihepsin altered hepatocyte size in the WT, but not the hepsin−/−, mice (Fig. 1D). Further analysis by flow cytometry also confirmed that hepsin−/− mouse hepatocytes were larger than those of WT mice, and that the hepsin−/− hepatocyte phenotype, but not the WT-hepatocyte phenotype, was reversed by reexpression of WT, but not mutant, hepsin (Supporting Fig. 3). Together, these results suggest that hepsin expression level is associated with the regulation of hepatocyte size in vivo.
Hepsin Is Involved in Regulation of Liver Sinusoidal Diameter.
Along with the change in hepatocyte size in hepsin−/− mice, TEM also revealed that liver sinusoids from hepsin−/− mice (5.63 ± 0.66 μm) were significantly narrower than those from WT mice (7.66 ± 1.26 μm; Fig. 2A). IVM also showed that the liver sinusoids of 4- and 8-week-old hepsin−/− mice were significantly narrower than those of age-matched WT mice (Fig. 2B). Graphic representation of the diameter distribution from 2,880 sinusoids measured in hepsin−/− livers showed a bell-shaped and left-shifted distribution, as compared to that from WT livers, suggesting that hepsin−/− liver sinusoids were generally narrower, but had the similar vessel densities to those of the WT livers (Supporting Fig. 4).
To verify that expression of hepsin affects the diameter of liver sinusoids, we performed blocking experiments on the WT and hepsin−/− mice with two different hepsin antibodies. A significant decrease in the diameter of the liver sinusoids was observed with antihepsin treatment of the WT, but not the hepsin−/−, mice (Supporting Fig. 5A). Reexpression of WT, but not mutant, hepsin by hydrodynamic delivery of hepsin DNA to mice (Supporting Fig. 5B) resulted in a significant increase in the sinusoidal diameter in hepsin−/−, but not WT, mouse livers (Supporting Fig. 5C). These results indicate that hepsin is causally related to the width of liver sinusoids, and the width of liver sinusoids can be regulated postnatally.
Hemodynamic Retention and Preferential Liver Colonization of Tumor Cells in Hepsin−/− Mice.
We hypothesized that the narrower sinusoids observed in the hepsin−/− mice could result in increased hemodynamic retention of cells that flow through the liver. To examine this possibility, we treated mice with fluorescent microbeads as well as tumor cells. Thirty to sixty minutes after the targeted injection of microbeads of different sizes through the spleen to the liver, nearly twice as many microbeads were retained in the hepsin−/− liver sinusoids as in the WT liver sinusoids (Fig. 3A). These results support the finding that hepsin−/− sinusoids were narrower than WT sinusoids.
Because the physical trapping of circulating cancer cells in the liver sinusoids because of size restriction is an important initial step in liver metastasis,17 we further examined whether hepsin can affect this process by challenging hepsin−/− mice IS with the syngeneic tumor cell line, B16F1. We detected a gradual accumulation of tumor cells in the liver, particularly near the periportal (zone 1) and sinusoidal (zone 2) areas and, eventually, in the pericentral space (zone 3). Overall, a greater number of tumor cells was consistently detected in the hepsin−/− liver sinusoids than in the WT liver sinusoids (Fig. 3B). In correlation with the preferential retention of metastatic tumor cells, there was a 7-fold increase in the number of tumor colonies in hepsin−/− mouse livers, in comparison to WT mouse livers, 12 days after the injection (Fig. 3C). Survival curves associated with tumor-injected hepsin−/− mice were also greatly reduced, as compared to those of tumor-injected WT mice, because of severe tumor burdens (Supporting Fig. 6). A similar phenomenon was also observed when the experiment was repeated with Lewis lung carcinoma cells (Supporting Fig. 7). There was no significant difference in the number of apoptotic cells (Fig. 3B) in the retained tumor cells or the level of hepatic nitric oxide production induced by tumor cells in the WT and hepsin−/− mice (Supporting Fig. 8), nor were there differences in the host immune response or growth advantage in the liver microenvironment between WT and hepsin−/− mice (data not shown). Systemic delivery of tumor cells by tail-vein injection also showed similar preferential colonization of tumor cells in the liver of hepsin−/− mice, compared to that of WT mice, although colonization of the lungs and other organs by tumor cells was similar in WT and hepsin−/− mice (Fig. 3D; Supporting Table 1). We next examined the effect of hepsin reduction on tumor cell colonization in WT mice by systemic challenge with an IV injection of B16F1 tumor cells and then administration of either antihepsin or control antibody. Although a similar tumor burden was detected in the lungs of both models, mice treated with antihepsin were remarkably more susceptible to tumor colonization in their livers than mice treated with control antibody (Fig. 3E). Taken together, these results strongly suggest that loss of hepsin enhances the colonization of livers by tumor cells, probably through increased retention of tumor cells because of narrower sinusoids.
Narrowed Liver Sinusoids in Hepsin−/− Mice Are Primarily Related to Enlarged Hepatocytes.
To investigate the mechanisms responsible for the narrow sinusoids in hepsin−/− mice, we measured the liver weight, liver protein levels (Supporting Fig. 9), and the number and distribution of other nonparenchymal cells surrounding the sinusoids (Supporting Figs. 10 and 11), as well as the amount and distribution of extracellular matrix Proteins (e.g., collagen, laminin, and fibronectin) and adhesion molecules (e.g., intracellular adhesion molecule, vascular cell adhesion molecule, and E-selectin; data not shown). All the results were comparable for both hepsin−/− and WT mice, except that the size of stellate cells was also increased in hepsin−/− mice (Supporting Fig. 11C). Because increased hepatocyte size was the only major factor confirmed to be strongly correlated with decreased sinusoidal width in hepsin−/− mice, we hypothesized that livers of hepsin−/− mice accommodate an increase in hepatocyte size by decreasing the area of sinusoidal spaces.
Expression of Connexins Is Associated With Hepatocyte Size and Sinusoidal Diameter.
To further investigate the mechanism(s) responsible for the changes in hepatocyte size that are the result of the loss of hepsin, we evaluated the subcellular components that may affect cell size, including several ion channels and junction proteins, such as desmoplakin. Although we did not find any differences in the expression of ion channels or desmoplakin in WT and hepsin−/− liver tissues (data not shown), we found that hepatocytes from hepsin−/− mice expressed more than twice as much connexin 32 (Cx32) and connexin 26 (Cx26) as hepatocytes from WT mice (Figs. 4 and 5A). The gap junctions were larger and more numerous in the hepsin−/− liver tissue than in the WT liver tissue. Moreover, consistent with a previous study that showed that connexins can exist as hemichannels in the free border that affect cell permeability and size,18 we found that the livers of hepsin−/− mice had higher numbers of hemichannel-like connexin expression than the livers of WT mice (Fig. 4B). The increase in connexin expression associated with hepsin−/− mice appeared to be mediated post-transcriptionally, because Cx messenger RNA levels were comparable in WT and hepsin−/− mice (data not shown). Correspondingly, hepsin antibody blocking caused an increase in the immunofluorescence of Cx32 and Cx26 in WT mice (Supporting Fig. 12). These results suggest that hepsin regulates the expression of connexins in the liver.
Hepsin−/− mice also had a higher transport efficiency of gap junctional intercellular communication (GJIC)-specific Lucifer yellow staining (Fig. 5B). Abnormal transport efficiency is known to can affect cell size.19 In addition, oleamide, an inhibitor of GJIC,20 increased the sinusoidal diameter and decreased hepatocyte size (Fig. 5C) in hepsin−/− mouse livers. Increased GJIC and/or the presence of more hemichannels also significantly affected hepatocyte cell viability in retrograde ethylene glycol tetraacetic acid/collagenase perfusion isolation of primary hepatocytes from hepsin−/− mice, whereas blocking the GJIC/hemichannels with the junctional blockers, glycyrrhetinic acid and carbenoxolone, increased the number of viable hepsin−/− hepatocytes that were recovered to that of WT hepatocytes (Supporting Fig. 13). Furthermore, overexpression of connexins in human HeLa cells and SK-HEP-1 hepatoma cells increased cell size in both cell lines (Supporting Fig. 14). The effect of hepsin on connexin levels and cell size was further examined in human cells. Treating PLC/PRF/5 cells and Huh7 cells with two independent hepsin antibodies resulted in an increase in both cell size and connexin levels. In contrast, treatment of SK-HEP-1 cells, which lacks hepsin,21 with the same two hepsin antibodies did not affect cell size (Supporting Fig. 15). These results are consistent with our in vivo studies that demonstrated the regulation of cell size and connexin levels by hepsin. Loss of hepsin may thus increase the connexin-mediated gap-junctional communication of hepatocytes, resulting in an expansion of hepatocyte size and a concomitant narrowing of sinusoids.
Hepsin-Mediated Pro-HGF Processing In Vivo.
To elucidate the mechanism that leads to increased connexin expression and increased hepatocyte size, we first determined the cell types in the liver that express hepsin. Among the cell types analyzed, hepatocytes were the major hepsin-expressing cells (Supporting Fig. 16). Because the liver is proposed to be the major organ for subsequent pro-HGF activation22 and pro-HGF is a potential substrate for hepsin, we examined whether hepsin−/− mice were defective in activating pro-HGF. In a pro-HGF processing/activation assay in which recombinant pro-HGF was incubated with WT or hepsin−/− mouse liver lysates, hepsin−/− mouse lysates had significantly lower rates of both pro-HGF processing and HGF-α (the alpha chain of mature HGF) generation, as compared to WT mouse lysates (Fig. 6A). These decreases indicate a reduction in pro-HGF-processing activity in the hepsin−/− liver. Moreover, we found a significantly lower level of HGF in the serum from the hepatic vein of hepsin−/− mice, in comparison to that of WT mice (Fig. 6B). An immunoprecipitation assay confirmed that the mature form of HGF (represented by HGF-α), but not pro-HGF, was the major component of HGF in the serum, and its level was lower in hepsin−/− mice than in WT mice (Fig. 6C).
The liver is a major organ for HGF synthesis, but the decrease in the mature form of HGF in the hepsin−/− mice was not caused by decreased synthesis of pro-HGF, because western blotting analysis of liver lysates revealed that there was no significant difference in the level of pro-HGF in WT and hepsin−/− mice (Fig. 6D). Hepsin−/− mouse livers may therefore be defective in converting pro-HGF produced in the liver into mature HGF that is released into the serum after processing; such a decreased level of mature HGF would be expected to cause diminished HGF signal transduction in the livers of hepsin−/− mice. Correspondingly, we observed that the level of c-Met phosphorylation (HGF activation site, residues Y1234 and Y1235, in the tyrosine kinase domain) was significantly decreased in hepsin−/− livers, as compared to WT livers, whereas the total c-Met level appeared unchanged (Fig. 6E).
Furthermore, when both WT and hepsin−/− mice were treated with an antibody against hepsin, only WT mice exhibited a decrease in HGF and phosphorylated c-Met (Supporting Fig. 17). All of these results indicate that the c-Met-signaling pathway was down-regulated in the hepsin−/− mouse liver because of the defect in pro-HGF activation in the liver.
HGF Levels Are Associated With Connexin Expression and Sinusoidal Width.
It has been shown that HGF down-regulates the level of connexin expression in vitro.23 In addition, we observed increased connexin expression and decreased HGF/c-Met signaling in hepsin−/− mouse livers. Therefore, we hypothesized that the decreased HGF level in hepsin−/− mice caused an increase in both the expression of connexins and hepatocyte size in the liver. To test this, we first analyzed the level of connexin expression in WT and hepsin−/− mouse livers treated with HGF or an antagonist of the HGF receptor, NK4. HGF treatment decreased the expression of connexins in hepsin−/− mice (Fig. 7A), whereas NK4 increased the expression of connexins in WT mice (Fig. 7B). Consistently, hepsin−/− mice had significantly enlarged liver sinusoids after HGF treatment (Fig. 8A), and WT mice had significantly narrowed liver sinusoids after NK4 treatment (Fig. 8B). A dose-dependent increase in the level of phosphorylated c-Met was also detected after HGF treatment (Supporting Fig. 18). Overall, these results suggest that hepsin regulates the liver architecture through the HGF/c-Met/connexin-signaling axis.
The identification of novel phenotypes in our hepsin−/− mice establishes a strong connection in vivo between hepsin and the maintenance of liver architecture. We propose that hepsin deficiency reduces HGF maturation and downstream c-Met phosphorylation that is required for expressing proper levels of connexins, which are, in turn, critical for the maintenance of normal hepatocyte size and, ultimately, normal sinusoidal diameter (Supporting Fig. 19).
Our results are consistent with previous reports showing that overexpression of Cx26 or Cx32 can increase the size of human hepatoma cells24 and mouse hepatocytes.19 Moreover, the increase in Cx26 and Cx32 levels in hepsin−/− mice was correlated with an increase in hepatocyte size in vivo, and this phenomenon was reversed by the GJIC blocker, oleamide. In addition to forming gap junctions, Cx26 and Cx32 also form hemichannels, which, when opened in a reduced-calcium environment, can lead to an increase in cell size because they form a nonselective leak pathway to permit free ions into the cytoplasm, followed by water uptake to maintain isoosmotic conditions.18 Such changes in cell size do not occur in connexin-deficient cells and can also be inhibited by oleamide and other gap-junctional blockers.20 A similar phenomenon might have occurred in our hepsin−/− mice, with increased hepatocyte size induced by excessive GJIC/hemichannels derived from overexpression of connexins on the cell surface.
We propose that HGF/c-Met may regulate connexin expression in hepatocytes in vivo. The detailed mechanism(s), however, remains to be elucidated. Using isolated primary rat hepatocytes in vitro, Ikejima et al.23 showed that down-regulation of Cx32 protein amounts by HGF occurs very quickly, starting at 3 hours after exposure to HGF, most likely by post-transcriptional modifications. In our study, HGF and NK4 affected Cx26 and Cx32 protein levels in vivo as early as 1 hour after exposure (Fig. 7), a result which further supports post-transcriptional mechanisms. Moreover, in rat hepatocytes, the reduction in connexins caused by HGF is prevented by genistein, an inhibitor of c-Met, which also indicates that c-Met signaling is likely to mediate this process.23 There are several modification pathways downstream of c-Met (i.e., the mitogen-activated protein kinase [MAPK], phosphoinositide 3-kinase, and signal transduction and activator of transcription 3 signaling pathways) that are coupled to HGF/c-Met.25 Although there is no direct evidence demonstrating which of these pathways is responsible for the decrease of Cx32 and Cx26, previous findings for connexin 43 (Cx43) may provide some clues. Turnover of Cx43 is regulated by endothelial growth factor (EGF) at multiple levels, including enhancing phosphorylation, ubiquitination, internalization, and degradation of this protein.26 Moreover, these EGF-induced modifications of Cx43 may be caused by the MAPK pathway.26 Because both Cx3227 and Cx2628 proteins turn over with a short half-life (1.5-5.0 hours), similar to that of Cx43,29 and because Cx32 and Cx43 have comparable responses to proteasome inhibitors,30 it is possible that similar signaling pathways or post-transcriptional modification mechanisms may be involved in the down-regulation of the levels of Cx32 and Cx26 by HGF/c-Met. Nevertheless, other regulatory mechanisms may exist, because many of the modifications of Cx43 are the result of the phosphorylation of the carboxyl terminus,26 and Cx32 and Cx26 contain short carboxyl termini with fewer potential phosphorylation sites than Cx43.29
Our results cannot completely rule out the systemic effect of the general loss of hepsin on the new phenotypes. Hepsin is, however, more highly expressed in the liver than in any other organs.4 Furthermore, in the mouse liver, hepatocytes are the only cell type that expresses hepsin (Supporting Fig. 16), whereas only hepatocytes and stellate cells express c-Met. We consistently found that in addition to hepatocytes, stellate cell size was also increased, whereas the sizes of Kupffer cells and ECs that lack c-Met expression were not (Supporting Figs. 10 and 11). This further supports our hypothesis that hepsin affects cell sizes through the HGF/c-Met pathway. The possibility, therefore, exists that the increase in the hepatocyte and stellate cell size could be partially attributed to the effect of hepsin loss in other organs that also coexpress HGF,31 but the significance of such a possibility is unknown. A more-detailed characterization of the liver phenotypes of mice with a liver-specific loss of hepsin, HGF, and c-Met32 will help address this issue and rule out the possibility that the current liver phenotypes are caused by the systemic effect of the general disruption of hepsin expression.
The diameter of sinusoids plays an important role in the pathogenesis of several common human liver diseases.14,33 Whether hepsin−/− mice can offer clues to the pathogenesis of these diseases awaits further study. Here, we used syngeneic tumor cell lines to show that hepsin−/− mice may retain more tumor cells in the liver than WT mice because of the narrower liver sinusoids. Because hepsin overexpression in tumor cells is involved in enhancing metastasis in some tumor models, hepsin has been proposed as a possible target for therapy.34 Our experiments indicate that systemic administration of antihepsin (Fig. 3E) enhances the hemodynamic retention of tumor cells in liver sinusoids similar to that observed in hepsin−/− mice. This is a significant concern that should be seriously considered before using antihepsin as an anticancer strategy.
In summary, we have unveiled a novel phenotype in hepsin−/− mice characterized by hepatocyte enlargement and diminished sinusoidal diameter. Hepsin−/− mice may be a valuable animal model for the study of the pathogenesis of human diseases related to stenotic hepatic sinusoidal spaces and metastatic liver tumors.
The authors thank Dr. Toshikazu Nakamura (Professor Emeritus of Osaka University Medical School) for the HGF and NK4 proteins, Drs. Ya-Chien Yang, Lih-Hwa Hwang, Chien-Kuo Lee, Yung-Li Yang, Jau-Tsuen Kao, Woei-Horng Fang, Pao-Hsien Chu, and Ruey-Bing Yang for their helpful discussions, and Chieh-Lin Wu, Yi-Tzu Chen, Bi-Huei Yang, Ying-Hui Su, Fang-Chun Su, Tzu-Ming Jao, and Chi-Hsung Yu for their excellent assistance. The authors also thank the Transgenic Mouse Models core facility and Microscopy Biophysics Laboratory (which receive grants from the National Research Program for Genomic Medicine of Taiwan and National Taiwan University Center of Genomic Medicine), the Second Core Lab, the Department of Medical Research, National Taiwan University Hospital, and the Instrument Center at National Tsing Hua University for technical support during this study.