Ursodesoxycholic acid, which is rapidly conjugated with taurine in vivo,1 is widely used for the treatment of cholestatic liver disease.2-4 Its beneficial effect is thought to involve a stimulation of hepatocellular bile secretion5, 6 as well as cytoprotective and antiapoptotic effects.7-10 The choleretic action of tauroursodeoxycholic acid (TUDC) is largely due to a rapid insertion of intracellularly stored transport ATPases into the canalicular membrane, such as the bile salt export pump (Bsep) and multidrug resistance protein-2 (Mrp2).11 However, the molecular basis of TUDC-sensing is still unknown. Evidence has been presented that the TUDC-induced insertion of Bsep into the canalicular membrane involves an activation of focal adhesion kinase (FAK), phosphatidylinositol 3-kinase (PI3 kinase), and c-Src, which trigger downstream a dual activation of extracellular signal-regulated kinases (Erks) and p38 mitogen-activated protein kinase (p38MAPK).6, 11, 12 Interestingly, TUDC-induced choleresis and kinase activation were inhibited in the presence of an RGD-motif containing hexapeptide,12 suggesting an involvement of integrins in TUDC-induced signaling toward canalicular secretion. Although TUDC per se does not affect hepatocyte volume,13 the signaling events triggered by TUDC strongly resemble those initiated in response to hypoosmotic or insulin-induced hepatocyte swelling.12, 14, 15 Here, mechano/swelling-sensitive α5β1 integrins become activated and trigger an FAK/Src/MAP kinase-dependent signaling toward choleresis with Bsep and Mrp2 insertion into the canalicular membrane.16, 17 In view of the recent finding that urea can activate α5β1 integrins in liver directly in a swelling-independent way,18 we studied the interaction between TUDC and α5β1 integrins, which are the predominant integrin isoform in liver.19 The data show that α5β1 integrins act as a long-searched TUDC receptor, which triggers TUDC-dependent choleresis.6, 11, 12 Molecular dynamics (MD) simulations revealed that TUDC, when interacting with the head region of α5β1 integrin, introduces an allosteric conformational change that has been linked to integrin activation before.20-22
Ursodeoxycholic acid, which in vivo is converted to its taurine conjugate tauroursodeoxycholic acid (TUDC), is a mainstay for the treatment of cholestatic liver disease. Earlier work showed that TUDC exerts its choleretic properties in the perfused rat liver in an α5β1 integrin-mediated way. However, the molecular basis of TUDC-sensing in the liver is unknown. We herein show that TUDC (20 μmol/L) induces in perfused rat liver and human HepG2 cells the rapid appearance of the active conformation of the β1 subunit of α5β1 integrins, followed by an activating phosphorylation of extracellular signal-regulated kinases. TUDC-induced kinase activation was no longer observed after β1 integrin knockdown in isolated rat hepatocytes or in the presence of an integrin-antagonistic hexapeptide in perfused rat liver. TUDC-induced β1 integrin activation occurred predominantly inside the hepatocyte and required TUDC uptake by way of the Na+/taurocholate cotransporting peptide. Molecular dynamics simulations of a 3D model of α5β1 integrin with TUDC bound revealed significant conformational changes within the head region that have been linked to integrin activation before. Conclusions: TUDC can directly activate intrahepatocytic β1 integrins, which trigger signal transduction pathways toward choleresis. (HEPATOLOGY 2013)
Materials and Methods
The experiments were approved by the responsible local authorities. Livers from male Wistar rats (120-150 g body mass), fed a standard chow, were perfused as described23 in a nonrecirculating manner. The perfusion medium was the bicarbonate-buffered Krebs-Henseleit saline plus L-lactate (2.1 mM) and pyruvate (0.3 mM) gassed with O2/CO2 (95/5 v/v). The temperature was 37°C. The osmolarity was 305 mosmol/L. Hypoosmotic exposure (225 mosmol/L) was performed by lowering the NaCl concentration in the perfusion medium. The addition of inhibitors to inflowing perfusate was made either by use of precision micropumps or by dissolution into the Krebs-Henseleit buffer. Viability of the perfused livers was assessed by measuring lactate dehydrogenase leakage into the perfusate, which did not exceed 20 milliunits min−1 g liver−1. The portal pressure was routinely monitored with a pressure transducer (Hugo Sachs Electronics, Hugstetten, Germany).24 If not stated otherwise, the compounds used in this study did not affect portal perfusion pressure. The effluent K+ concentration was continuously monitored with a K+-sensitive electrode (Radiometer, Munich, Germany).
Immunofluorescence Staining of the α5β1 Integrin Dimer and the Active Conformation of the β1 Integrin Subunit.
Cryosections of perfused livers were analyzed for total α5β1 integrin expression, induction of the active conformation of the β1 integrin subunit, and integrity of the actin cytoskeleton by using the LSM510 laser scanning microscope from Carl Zeiss AG (Oberkochen, Germany) For immunohistochemistry of liver slices as primary antibodies, the polyclonal rabbit anti-α5β1 integrin dimer antibody (1:100) and the monoclonal mouse anti-β1 integrin subunit active conformation antibody (1:100) were used. To visualize filamentous actin in order to document cell shape, FITC-coupled phalloidin (1 μg/mL) was applied.
Western Blot Analysis.
Lysed samples from perfused liver or HepG2 cells were transferred to sodium dodecyl sulfate / polyacrylamide gel electrophoresis (SDS-PAGE) and proteins were then blotted to nitrocellulose membranes using a semidry transfer apparatus (GE Healthcare, Freiburg, Germany). Blots were blocked for 1 hour in 5% (w/v) bovine serum albumin (BSA) or milk powder containing 20 mmol/L Tris, pH 7.5, 150 mmol/L NaCl and 0.1% Tween 20 (TBS-T), then incubated at 4°C overnight with the respective first antibody (antibodies used: anti-phospho-EGFR-Y845, -Y1045, Y1173, anti-β1 integrin [1:2,500], anti-phospho-Erk-1/-2, anti-Erk-1/-2, anti-rat Ntcp [K4], anti-EGFR [1:5,000], and anti-γ-tubulin 91:10,000]). Following washing with TBS-T and incubation with horseradish peroxidase-coupled antimouse, or antirabbit IgG antibody (all diluted 1:10,000) at room temperature for 2 hours, the blots were washed extensively and developed using enhanced chemiluminescent detection (GE Healthcare). Blots were then analyzed densitometrically and normalized to total protein amount.
Cloning of FLAG-Tagged Ntcp-EGFP.
Two different FLAG tags were cloned to the NH2 terminus of rat Ntcp using partially overlapping oligonucleotides. The sequence of the first primer pair was 5′-ATG GAC TAC AAG GAC GAT GAC GAT AAG AGG-3′ and 5′-CTT ATC GTC ATC GTC CTT GTA GTC CAT CCT-3′ and coded for a start codon, followed by the conventional FLAG tag (DYKDDDDK). The second primer pair was 5′-CC GCC ATG GAC TAC AAG GAC GAT GAC GAT AAG AGG-3′ and 5′-CTT ATC GTC ATC GTC CTT GTA GTC CAT GGC GGC CT-3′, which codes for a FLAG tag with an improved start codon (Kozak sequence). The free ends of the annealed products were complementary to the restriction sites of the endonuclease Van91I. The products were cloned into the Ntcp-enhanced green fluorescent protein (EGFP) plasmid after digestion with Van91I in-frame to the 5′-site of Ntcp.25 The accuracy of the resulting plasmids was confirmed by sequencing.
Generation of Stably Expressing FLAG-Ntcp-EGFP Cells, Transfection, and Cultivation.
HepG2 cells were cultured in Dulbecco's modified Eagle's medium Nutrimix F12 (Biochrom, Berlin, Germany) containing 5% fetal calf serum (PAA, Coelbe, Germany), as described.26 FLAG-Ntcp-EGFP was transfected into HepG2 cells using Lipofectamine 2000 (Life Technologies, Darmstadt, Germany) according to the manufacturer's guidelines. Stable cell lines were established with 0.5% of geneticin as selection agent.27 HepG2 cells stably expressing FLAG-Ntcp-EGFP and wildtype HepG2 cells were cultured in 6- or 12-well culture plates until subconfluence and stimulated with different concentration of TUDC for various timepoints.
Small Interfering RNA (siRNA) Knockdown of β1 Integrin.
siRNA was prepared by Qiagen (HP GenomeWide siRNA, Qiagen, Hilden, Germany) targeting the β1 integrin sequence 5′-AAA AGT CTT GGA ACA GAT CTG-3′. Cells were washed three times with serum-free Dulbecco's modified Eagle's medium Nutrimix F12 + 0,5% geneticin followed by siRNA transfection using HiPerFect transfection reagent (Qiagen) according to the manufacturer's instruction.
Results from at least three independent experiments are expressed as means ± SEM (standard error of the mean). Results were analyzed using Student's t test: P < 0.05 was considered statistically significant.
Generation of α5β1 Integrin-Ligand Complex Structures and MD Simulations.
A detailed description of how starting structures for MD simulations of α5β1 integrin bound to either TUDC, TC, or GRGDSP were generated and how MD simulations of in total 1.050 μs length of these systems were performed and analyzed is provided in the Supporting Text. Integrin sequence numbers are according to Uniprot.
Activation of Integrins by TUDC.
In isolated perfused rat liver, addition of TUDC at a concentration of 20 μmol/L induced within 1 minute the appearance of the active conformation of β1 integrin, whereas in the absence of TUDC the active β1-isoform was only scarcely detectable (Fig. 1A; see Supporting Fig. 1 for total α5β1 integrin staining). TUDC-induced β1 integrin activation was predominantly observed inside the hepatocytes (Fig. 1B). Equimolar concentrations of other bile acids, such as taurocholic acid (TC), glycochenodeoxycholic acid (GCDC), taurochenodeoxycholic acid (TCDC), or tauro-lithocholic acid 3-sulfate (TLCS) were ineffective with regard to β1 integrin activation (Supporting Fig. 2). None of the bile acids had any effect on the immunostaining for total α5β1 integrins (Supporting Fig. 3). TUDC-induced integrin activation was sensitive to inhibition by the RGD-motif containing hexapeptide GRGDSP, which also prevented swelling-induced integrin activation,14, 15 whereas the control hexapeptide GRADSP was ineffective (Fig. 1A). In line with previous data,12 TUDC induced within 1 minute phosphorylation of extracellular signal regulated kinases Erk-1/-2, which was abolished in the presence of the RGD-motif containing hexapeptide but not in presence of the inactive control hexapeptide (Fig. 2). TUDC also induced activation of the epidermal growth factor receptor (EGFR) in an RGD-hexapeptide-sensitive way (Fig. 2). TUDC-induced EGFR tyrosine phosphorylation involved tyrosine residues 845 and 1173, but not Tyr1045 (Fig. 2). Tyr845 is a known Src kinase target, which triggers an activating autophosphorylation at Tyr1173.28, 29 A similar EGFR phosphorylation pattern is induced by hypoosmotic hepatocyte swelling,30 a condition in which α5β1-integrins act as osmosensors.
Swelling-induced β1-integrin activation largely occurred in the plasma membrane12 (Fig. 1B), in line with the concept that β1 integrins in the plasma membrane serve as osmo-/mechanosensors through a swelling-induced attachment to extracellular matrix proteins.15 In contrast, TUDC-induced β1 integrin activation predominantly occurred in the cytosol of hepatocytes (Fig. 1; Supporting Fig. 2), which suggests that cumulative TUDC transport by way of the Na+/taurocholate cotransporting polypeptide (Ntcp) into the hepatocytes is required. This possibility was tested in a human HepG2 cell line, which expresses α5β1 integrins, but not Ntcp (Fig. 3A). As shown in Fig. 3A, TUDC, even at a concentration of 100 μmol/L, did not induce the active conformation of β1 integrin in parental HepG2 cells. However, in Ntcp-FLAG-expressing HepG2 cells (Fig. 3A), TUDC induced the appearance of the active β1 integrin conformation inside the cells (Fig. 3A). In line with a requirement of TUDC uptake into the cells for TUDC-induced activation of β1 integrins, TUDC did activate Erks in Ntcp-expressing HepG2 cells, but not in the Ntcp-deficient parental cell line (Fig. 3B).
The requirement of β1 integrins for TUDC-induced Erk activation was also investigated in Ntcp-transfected HepG2 cells after β1 integrin knockdown using an siRNA approach (Supporting Fig. 4). β1 integrin knockdown fully abolished the bile acid induced Erk activation in these cells (Fig. 3B).
TUDC/TC Interactions with Regard to α5β1 Integrin Activation.
Whereas TC, even at a concentration of 100 μmol/L, had no β1 integrin-activating activity (Fig. 4), TUDC concentrations as low as 5 μmol/L induced β1 integrin activation (Fig. 4). When TUDC was added on top of TC (100 μmol/L), considerably higher concentrations of TUDC were required in order to induce a comparable β1 integrin activation. This indicates that TC may interfere with TUDC-induced α5β1 integrin activation.
MD Simulations of α5β1 Integrin Bound to TUDC, TC, and GRGDSP.
The sensitivity of TUDC-induced integrin activation to GRGDSP (Figs. 1A, 2) led us to hypothesize that the hexapeptide might compete with TUDC binding in the head region of the integrin between propeller domain and βA domain. This region has been identified as the binding site of RGD-peptides.20, 31 We thus performed docking of TUDC and, as a control, TC to a model of the α5β1 ectodomain18 (Fig. 5A; Supporting Fig. 6). We assumed that the sulfonate moieties of the two bile acids mimic the interaction of the RGD-peptides' Asp sidechain with the Mg2+ ion located at the MIDAS site (metal-ion dependent adhesion site) of the βA domain. The two representative docking solutions showed a similar binding mode of the cholan scaffold, which extends to the propeller domain.
The two complex structures were investigated by MD simulations of 200 ns each, as was a complex structure of the antagonistic GRGDSP peptide binding to the ectodomain of α5β1 integrin. Furthermore, three simulations of 150 ns each of these complex structures with a truncated version of the ectodomain were performed. Unless stated otherwise, all results refer to the simulations with the nontruncated ectodomains. The simulations with the full ectodomain reveal minor structural changes in the propeller domain (root mean-square deviations of the coordinates of Cα atoms [RMSD] ≈ 2.0-2.5 Å) in all cases, and in the βA domain in the case of TC and GRDGSP (Supporting Fig. 7A,B). Larger conformational changes (RMSD ≈ 3.5 Å) were observed for βA in the case of the TUDC complex (Supporting Fig. 7B). The bile acids show stable binding modes that deviate by ∼4 Å RMSD from the docking solutions (Supporting Fig. 7C): the cholan scaffold binds almost all the time to a groove between the α5 and β1 subunits, and the interaction between the sulfonate moiety and the MIDAS ion was never broken. The hexapeptide shows larger conformational changes (RMSD ≈ 7 Å) compared to the starting geometry, which arise mostly from a higher mobility of the N-terminus (Supporting Fig. 7C). This can be explained by Asp180 of αV being mutated to Ala200 in α5, leading to a loss of salt bridge interactions involving Arg of the peptide compared to the αVβ3 complex structure.31 Again, the interaction between Asp of the hexapeptide and the MIDAS ion was never broken. Similar results were obtained for the simulations of the truncated ectodomains (data not shown).
Significant Conformational Changes Involving Helices α1 and α7 Between TUDC- Versus TC- and GRGDSP-Bound α5β1 Integrin.
Considerable variation between the βA domains of the complex structures is found in the region of the center of the helix α1 and the N-terminus (“top”) of helix α7, with the structures of TC- and GRGDSP-bound βA being similar to each other but significantly differing from that of the TUDC complex. First, the distance between Cβ atoms of Leu165 of α1 and Ile371 of α7 is smaller by ∼2 Å in the TUDC complex (Fig. 5E), indicating a tighter packing between the top of α7 and the center of α1. Second, the kink angle of α1 is larger by more than 10° in the case of the TUDC complex (Fig. 5F). A similar albeit less pronounced difference in the kink angles was also observed in the simulation of the truncated ectodomains (data not shown). Thus, in the TUDC case, α1 straightens and starts to become a continuous helix structure (Fig. 6A). This is also corroborated by residues Lys163-Ser164-Leu165 being in a helical conformation during 98% of the simulation time of the TUDC complex. A similar degree of helicality of α1 is observed for TUDC bound to the truncated ectodomain (data not shown). In contrast, a break existing in the unliganded structure of αvβ3 (32), which has served as template for the α5β1 model, at Gly166 is largely maintained in the TC and GRGDSP cases (Fig. 6B).
The straightening of α1 leads to an inward movement of the central region of the helix (see arrow in Fig. 5C) and the formation of a region of novel hydrophobic packing (“T-junction”20, 22) between residues of this central region and those located at the top of α7 and the end of the β6-α7 loop for the TUDC complex (Figs. 5C, 6). As a consequence, the C-terminus (“bottom”) of helix α7 is moved outwards in the direction of the C-terminus of helix α1 (Fig. 5D). The motion becomes amplified in the TUDC complex when the position of the βA domain relative to the propeller domain is considered (Fig. 5D) in the wake of a shift of the center of mass of the βA domain by 1.2-1.8 Å compared to that in the TC and GRGDSP complexes (Fig. 5G). Furthermore, the outward movement of α7 is overlaid with a downward movement of the helix (see arrows in Fig. 5D). In contrast, no T-junction formation is observed for TC- and GRGDSP-bound integrins (Fig. 6) as is no outward and downward movement of helix α7 (Fig. 5D).
TUDC is known for its choleretic and hepatoprotective effects. As shown previously, TUDC-induced choleresis is triggered by a p38MAPK and Erk-dependent insertion of intracellularly stored Bsep and Mrp2 into the canalicular membrane of the hepatocyte.6, 12 TUDC-induced choleresis and signal transduction towards MAP kinases was recently shown to involve integrins12 and to resemble strongly osmosignaling events, which are initiated by hypoosmotic hepatocyte swelling.12 In line with this, TUDC also induced EGFR activation (Fig. 2), as does hypoosmotic hepatocyte swelling.30 As shown here, TUDC directly, i.e., nonosmotically, interacts with α5β1 integrins, resulting in an integrin activation and initiation of integrin signaling involving c-Src, FAK, EGFR, PI3 kinase, and MAP-kinases.6, 12 In line with this, β1 integrin knockdown abolished TUDC signaling towards Erks. These data suggest that β1 integrins are a long-searched sensor for TUDC in the liver. Integrin activation by TUDC was not only found in rat liver, but also in human HepG2 cells and was not mimicked by other bile acids (TC, GCDC, TCDC, TLCS). This may explain at least in part the unique hepatoprotective and choleretic properties of TUDC compared to other bile acids. Nevertheless, as the experiments reported herein have been performed in noncholestatic livers and hepatocytes, it remains unclear to what extent other mechanisms come into play in the cholestatic liver, such as Ca2+/type II InsP3 receptor-33, 34 or cPKCα/PKA-dependent pathways.35 In order to effectively trigger integrin activation, TUDC has to be taken up by and/or to be concentrated inside the hepatocyte. In line with this, TUDC-induced integrin activation was most pronounced in the cytosol and only found in HepG2 cells that express Ntcp. This requirement for concentrative TUDC uptake and the liver-specificity of Ntcp-expression may explain why TUDC acts primarily in the liver. Higher TUDC concentrations were required for β1 integrin activation when TC was simultaneously present. This is probably not explained by a competition of TUDC with TC for entry into the hepatocyte by way of Ntcp. This view is supported by the previous finding5 that TUDC at concentrations of 10-50 μmol/L stimulates TC excretion into bile by up to 30% in perfused rat liver when TC is present at a concentration of 100 μmol/L in the perfusate. This would not be expected if bile acid entry into the hepatocyte would become rate-controlling.
An alternative explanation for the TC-mediated inhibition of TUDC-induced β1 integrin activation is offered by the results obtained from MD simulations of TUDC, TC, and GRGDSP bound to a 3D model of the ectodomain of α5β1. The starting structure of the α5β1 ectodomain is based on an unliganded structure of αVβ3,32 which has been proposed to represent a low-affinity state of the integrin headpiece.21, 36 The βA domain of this structure shows a break in helix α1 and lacks a tight hydrophobic packing between residues of the central region of helix α1 and the top of helix α7 and the end of the β6-α7 loop. These characteristics persist during the simulation of the TC and GRGDSP complexes. In stark contrast, the simulation of the TUDC-bound ectodomain reveals a T-junction formation between these two regions that was first observed in the structure of liganded αIIbβ3,20 and afterwards in computational studies of agonist-bound integrin ectodomains.21, 22 As a consequence an outward and downward movement of helix α7 results that resembles that of a rod connecting a piston (the region of the T-junction) to a crankshaft (the hybrid domain). This mechanical model has been proposed as a mechanism to change the interdomain βA/hybrid domain hinge angle, leading to an unbending of the integrin structure.37 According to current models, such an unbending is required for activation.32
The activating structural changes in the case of the TUDC-bound ectodomain are only observed in the vicinity of the binding pocket and remain located within the βA domain (Supporting Fig. 8). This finding is in line with the timescale of integrin activation in the absence of force (microseconds38 to seconds39), which is orders of magnitude larger than the simulation time of 200 ns investigated here. The limited simulation time may also explain why no separation of the β6-α7 loop from the β1-α1 loop is observed in either simulation (Supporting Fig. 8). Further support for the observed T-junction formation in the β1-subunit is provided by the fact that residues involved in it (Fig. 6) are highly conserved across the eight β-subunits (Leu165: 6x, Val 367: 4x) as are residues in the immediate neighborhood (Leu169: 8x, Val370: 8x, Ile371: 6x). In total, the MD simulations suggest that TUDC, when interacting with the head region of α5β1 integrin, introduces an allosteric conformational change in the β1 subunit that propagates to the interdomain βA/hybrid domain hinge, which should lead to integrin activation, in agreement with experiment. In contrast, TC and GRGDSP do not introduce such a conformational change and, hence, should not activate integrin, again in agreement with experiment. As these ligands do remain bound in the head region, they effectively block this region for an interaction with TUDC, which can explain the TC-mediated inhibition of TUDC-induced β1 integrin activation.
As to a possible mechanism of TUDC-induced β1 integrin activation, TUDC forms a monodentate MIDAS coordination with its sulfonic acid moiety (Fig. 6C), and with a second oxygen of this moiety a charge-assisted hydrogen bond with the amide nitrogen of Asn244 (Supporting Table 1). The accompanying conformational change of the loop between helices α2 and α3 leads to formation of a hydrogen bond (hydrophobic interaction) between Ser242 (Ile241) of this loop and Tyr153 (Lys156) of the β1-α1 loop. These interactions presumably stabilize the β1-α1 loop region, with a further stabilization arising from a water-mediated hydrogen bond between TUDC's sulfonic acid moiety and the amide nitrogen of Tyr153. Apparently, the stabilization leads to helix α1 straightening and becoming continuous, which results in an inward movement of the central region of α1 and the T-junction formation. In contrast, the sulfonic acid moiety of TC interacts simultaneously with MIDAS and LIMBS, reminiscent of the coordination of carboxylate groups of an RGD peptide40 or eptifibatide41 bound to αvβ3 or a mutant of αIIbβ3, respectively. No further interaction is observed between the sulfonic acid moiety and the surrounding protein and, hence, no stabilization of the β1-α1 loop region can be expected. Accordingly, the break in helix α1 persists, and no inward movement of the helix is observed.
The difference in β1 integrin activation between TUDC and TC must be rooted in the differences in the substitution pattern of the cholan scaffolds (Supporting Scheme 1): although the simulations started from very similar binding modes of the bile acids (Supporting Fig. 6), different, yet stable, orientations of the cholan scaffold develop in the course of the simulation (Fig. 6D). The cholan scaffold of TC is oriented almost perpendicular to the one of TUDC, which is favored by hydrogen bond formation of the 7α-OH group of TC with Ser265 and Asp268 (Supporting Table 1). In TUDC, the configuration at C7 is inverted, which drastically reduces hydrogen bond interactions of the 7β-OH group with the α5 subunit (Supporting Table 1). In contrast, the presence of the 12α-OH group in TC does not seem to be responsible for the nonactivating behavior of TC because the group does not make any interactions with the α5 subunit in the binding mode found. Support for the hypothesis that it is the configuration of C7 that determines whether β1-integrin becomes activated or not is provided by the fact that TCDC does not activate β1-integrin either: whereas TCDC lacks a 12α-OH group, in contrast to TC, it does have a 7α-OH group, as does TC (Supporting Scheme 1). Overall, the differences in the orientation of the cholan scaffold lead to differences in the orientation of the sulfonic acid moieties, with the above-described consequences for β1-integrin activation.
In summary, TUDC has the unique property to directly interact with α5β1 integrins inside the hepatocyte. The resulting conformational change triggers β1 integrin activation and initiates integrin-dependent signaling, which explains not only the choleretic and cytoprotective effects of this therapeutically used bile acid but also its hepatocyte-specificity.
The authors thank the “Zentrum für Informations und Medientechnologie” (ZIM) at the Heinrich Heine University for computational support.