The Importance of Membrane Microdomains for Bile Salt

CORRECTION Correction: The importance of membrane microdomains for bile salt-dependent biliary lipid secretion (doi:10.1242/jcs.211524) Johannes Eckstein, Hermann-Georg Holzhütter and Nikolaus Berndt

There was an error published in J. Cell Sci. 131, jcs211524 (doi:10.1242/jcs.211524).

The formatting of Eqn 6 in the above paper was incorrect; the correct equation should read as below.

The reversible exchange of BSs between the aqueous phase and the patch we describe by the master equation: dPi;N ðBSÞ Li;N ¼ BS kþ½Li;N hðBS 1Þ=N Pi;N ðBS 1ÞkPi;N ðBSÞ 1 BS 1 þ ; ð6Þ dt c h where k+ and k− are the on and off rate constants for the uptake and release of a single BS molecule. Li,N is the number of lipid molecules in the patch and BSC is the number of BS molecules in the canalicular space per membrane unit. The parameter η is a solubility index that represents the number of lipid molecules that are required to accommodate one BS molecule. The larger η, the smaller is the membrane solubility of a BS.

We apologise to readers for any confusion that this error might have caused. Journal of Cell Science

RESEARCH ARTICLE The importance of membrane microdomains for bile salt- dependent biliary lipid secretion Johannes Eckstein, Hermann-Georg Holzhütter*,‡ and Nikolaus Berndt*

ABSTRACT Later, PC, CH and BS mixed micelles are formed (Small, 2003). In Alternative models explaining the biliary lipid secretion at the this model, the secretion rate of individual lipids into the bile is canalicular membrane of hepatocytes exist: successive lipid independent from each other and mainly determined by the activity extraction by preformed bile salt micelles, or budding of membrane of ABC flippases transporting PC and CH from the inner to the outer fragments with formation of mixed micelles. To test the feasibility of the leaflet. The E-model is consistent with several experimental findings latter mechanism, we developed a mathematical model that describes on changes in the lipid composition of the bile of transgenic animals the formation of lipid microdomains in the canalicular membrane. Bile lacking one or two of the ABC transporters. Disruption of the gene salt monomers intercalate into the external hemileaflet of the for the lipid transporter Mdr2 (also known as Abcb4) in mice, which ‘ ’ canalicular membrane, to form a rim to liquid disordered domain is responsible for the flopping of PC from the inner to the outer patches that then pinch off to form nanometer-scale mixed micelles. membrane leaflet, was accompanied by a complete absence of Model simulations perfectly recapitulate the measured dependence of phospholipid from the bile (Smit et al., 1993). Similarly, knockdown – bile salt-dependent biliary lipid extraction rates upon modulation of the of the CH transporter Abcg5 Abcg8 in mice reduced the secretion – membrane cholesterol (lack or overexpression of the cholesterol rate of CH by 70 90% (Kosters et al., 2006; Yu et al., 2002a). transporter Abcg5–Abcg8) and phosphatidylcholine (lack of Mdr2, also Importantly, however, there was no significant change of the −/− known as Abcb4) content. The model reveals a strong dependence of CH secretion rate in Abcg8 mice that have an altered lipid the biliary secretion rate on the protein density of the membrane. Taken composition of the canalicular membrane due to the simultaneous together, the proposed model is consistent with crucial experimental knockdown of the serine flippase Atp8b1 (Groen et al., 2008). This – findings in the field and provides a consistent explanation of the central finding suggested that mechanisms other than the Abcg5 Abcg8- molecular processes involved in bile formation. mediated loading of CH on BS micelles are involved in the secretion of CH. One possible alternative mechanism is the BS-dependent KEY WORDS: Bile formation, Bile salts, Mathematical modeling, budding of mixed micelles from the canalicular membrane. This Membrane lipid, Microdomain budding model (B-model) assumes that budding of membrane patches is a central mechanism of bile micelle formation instead of INTRODUCTION extraction of single lipids from the membrane by preformed BS Production of the bile is an important liver function required for micelles as assumed in the E-model. This model relies on electron the efficient digestion of dietary lipids and the elimination of microscopy studies (Crawford et al., 1995) and mathematical xenobiotics and endogenous waste products. In the process of bile modeling (Eckstein et al., 2015; Marrink and Mark, 2002). formation, bile salts (BSs) are actively pumped from the cytosol of According to the B-model, the lipid composition of the bile hepatocytes into the extracellular space where they act as detergents should reflect the lipid composition of membrane region from solubilizing phospholipids, cholesterol and proteins from the outer which bile micelles are being released. Experimental studies suggest leaflet of the canalicular membrane. that BS-dependent extraction of membrane constituents takes place The molecular mechanism by which the central lipids of the bile, preferentially from liquid disordered (Ld) microdomains that are phosphatidylcholine (PC) and cholesterol (CH), are secreted from enriched in the membrane lipid PC (Tietz et al., 2005). Hence, the canalicular membrane into the extracellular space and packed interpreting experimental data on biliary secretion rate on the basis together with BSs in mixed micelles or vesicles is still a matter of of the B-model requires knowledge of the size distribution and lipid debate. The extraction model (E-model) assumes that BS micelles composition of Ld microdomains. While the chemical composition formed in the canalicular space successively extract PC and CH of the outer leaflet as a whole can be assessed by various molecules from the outer leaflet. Initially, PC is the preferentially experimental techniques [chemical labeling with non-penetrating extracted lipid thereby increasing the CH-solubilizing capacity. agents, immunological methods, phospholipase digestion of membrane phospholipids, use of phospholipid-exchange proteins, and physicochemical methods such as X-ray diffraction and NMR Charité– Universitätsmedizin Berlin, Institute of Biochemistry, Charitéplatz 1, 10117 (Tietz et al., 2005)] the distribution of lipids and proteins across Berlin, Germany. liquid ordered (Lo) and Ld microdomains can only be indirectly *These authors contributed equally to this work inferred from the chemical composition of detergent-resistant ‡Author for correspondence ([email protected]) membranes (DRMs) (Mazzone et al., 2006). Whether this method truly informs about the composition of microdomains in the intact H.-G.H., 0000-0002-5054-6023 membrane under in vivo conditions is highly questionable This is an Open Access article distributed under the terms of the Creative Commons Attribution (Lichtenberg et al., 2005). License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution and reproduction in any medium provided that the original work is properly attributed. We thus strived in this work to construct a reliable mathematical model of biliary lipid secretion that picks up the concept of the

Received 29 September 2017; Accepted 23 January 2018 B-model. To this end, we first extended our previous mathematical Journal of Cell Science

1 RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs211524. doi:10.1242/jcs.211524 model of microdomain formation (Eckstein et al., 2015) by factor γ weights the relative contribution of the phase-ordering including lipid–protein interactions. This was necessary because energy J and the interaction energy W to the total energy E. The our previous model had some limitations in correctly describing the index σ designates the ordering state of the lipid: σ=− 1 denotes the size distribution and lifetime of microdomains because only lipid– low-ordered state where the bulky conformation of the fatty acid lipid interactions were considered. In this case, membrane lipids chains allows high flexibility and thus rapid movement of the lipid; progressively de-mixed and eventually segregated into single large σ=+ 1 denotes the high-ordered state where the fatty acid chains are Lo and Ld ‘macrodomains’. Formation of such stable macro-scale stretched and tightly packed thus restricting lipid mobility. domains is indeed observed in artificial protein-free membrane The set of six cells adjacent to cell i is denoted by n(i). wσ(Xi, Xk) systems (AMSs) (Kahya et al., 2003), but not in biological is the matrix of pairwise lipid interactions. There are two such s s membranes. Hence, to circumvent the problem of complete lipid matrices for each ordering state σ=± 1. The elements jðXi ; Xk Þ of de-mixing, we had to introduce in our previous model an average the matrix J quantify the strength with which a lipid with ordering domain life time at which the simulation of domain growth was state σ ‘enslaves’ the neighboring lipid to adopt the same ordering interrupted owing to the virtual extraction of a membrane patch. To state. Numerical values for the elements of the matrices wσ(Xi, Xk) s s get rid of this obvious limitation, we extended our model by and jðXi ; Xk Þ were taken from our previous work (Eckstein et al., including protein–lipid interactions as a possible mechanism that 2015). may constrain the continuous growth of domains and enable the Minimization of the total lattice energy E was carried out by using formation of time-stable microdomains. the Gillespie algorithm with periodic boundary conditions. For We further combined our extended model of microdomain more technical details of the minimization procedure see Eckstein formation with a model of micelle budding from Ld domains that et al. (2015). is in concordance with biophysical principles governing the solubilization of biological membranes by detergents (Helenius and Particle movement Simons, 1975; Lichtenberg et al., 2013). According to this model, BS The movement of mobile lipids is restricted to the pairwise monomers preferentially intercalate into Ld domains and facilitate the exchange between neighboring cells i, j with the exchange rate: budding of nanometer-scale mixed micelles. We demonstrate that the w þ w X model can perfectly recapitulate experimentally observed variations r ði; jÞ¼r i j ; w ¼ w ðX ; X Þ; ð Þ L L exp k T i s i k 2 in the lipid composition of the bile and BS-dependent secretion rates B k[nðiÞ under normal conditions as well as at altered activity of the ABC transporters for PC and CH. We discuss several other aspects wi is the total energy of the lipid at lattice position i and rL is a substantiating the feasibility the membrane budding concept as the scaling factor used to relate the time scale of the lattice simulation to dominating mechanism of biliary lipid secretion. the time scale of experiments. Movement of a protein is modeled as a series of consecutive shifts RESULTS of protein units to adjacent lattice cells. We implemented two Model description variants for the re-distribution of the lipids displaced by the moving Lattice model front of the protein: (1) shift to lattice cells that prior to the move are The modeling approach developed in our preceding work (Eckstein occupied by the protein units constituting the back of the protein, (2) et al., 2015) was extended by including membrane proteins and their successive sequential displacement of lipids along the surface of the interaction with membrane lipids. In brief, the model represents the protein like a bow wave. Both variants yielded identical simulation outer membrane leaflet as a triangular lattice of 122×122 cells with results. The rate of protein movement is calculated from Eqn 3: 1nm2 of cell size corresponding to an upper boundary of cross- w þ w X sectional area occupied by a typical membrane lipid. Each lattice cell is P L rPði; jÞ¼rP exp ; wP ¼ wk ; occupied either by a lipid or a protein unit. We distinguish between kB T k[AðiÞ ð Þ three basic types of membrane lipids: CH, phosphatidylcholine (PC) X 3 w ¼ w ; and sphingomyelin (SM). They represent the prevailing membrane L j lipids in the outer leaflet of the canalicular membrane, and serve as l[Mð jÞ prototypes for other membrane lipids with similar chemical properties. where wP denotes the energy of the protein at position i, with A(i) the Proteins are represented as hexagonal-shaped non-deformable set of cells defining its rim. w is the total energy of all M( j) lipids ‘ ’ L clusters of units occupying 19, 37 or 61 lattice cells of the lattice. moved by the movement of the protein in the direction of lattice cell According to the ‘lipid annulus’ concept (Contreras et al., 2011) the j. The scaling factor rP relates the time scales of protein to those of surface of integral membrane proteins is covered with tightly bound the lipid movement. ‘ ’ rim lipids acting like a lubricating layer. The interaction energies In our model, the plethora of different membrane proteins is of rim lipids with adjacent mobile lipids determine the energetic represented by two generic proteins, which we will refer to in the state of the protein and thus the preference of the protein for the Lo following as raft protein (RP) and non-raft protein (NRP) owing to or Ld phase, respectively. their preferences for the Lo and Ld phase, respectively. This lipid preference is controlled by the interaction energies of the rim lipids Interaction energies and phase-ordering energies XP of the protein with the adjacent mobile lipids Xk, Movement of lipids and proteins is driven by minimization of the E total energy of the system according to: wsðXP; Xk Þ¼0:66 wsðCH; Xk Þþ0:34 wsðSM; Xk ÞðRPÞ; w ðX ; X Þ¼ : w ð ; X Þþ : w ð ; X ÞðÞ: XN X s P k 0 14 s CH k 0 86 s PC k NRP s s ð Þ E ¼ W þ gJ ¼ wsðXi; Xk ÞþgjðXi ; Xi Þsisk ; ð1Þ 4 i¼1 k[nðiÞ The coefficients in Eqn 4 equal the fraction of CH and SM in the Lo where X=CH, PC or SM and specifies the lipid type, the scaling phase and of CH and PC in the Ld phase, respectively. Journal of Cell Science

2 RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs211524. doi:10.1242/jcs.211524

Diffusion and mobility over BS numbers equal or larger than UN+1 (i.e. large enough to form The ‘diffusivity’ of a particle (lipid, protein) is expressed through a ring around the patch of size N). the diffusion coefficient D that up to a scaling factor relates to the The reversible exchange of BSs between the aqueous phase and mean squared displacement Δx(t)2=4Dt where Δx(t) is the distance the patch we describe by the master equation: that the particle travels during the time span t. The diffusivity of a P ð Þ particle has to be distinguished from its ‘mobility’ which in our d i;N BS ¼ k ½L ð Þ=NP ð Þ t BSc þ i;N h BS 1 i;N BS 1 model refers to the number of elementary moves accomplished d within time t. A particle may have a large mobility but a small Li;N kPi;N ðBSÞ 1 BS 1 þ ; ð6Þ diffusivity if boundaries exist restricting the possible directionality h of the movement, an effect called tortuosity. where k+ and k− are the on and off rate constants for the uptake and release of a single BS molecule. Li N is the number of lipid Definition of the biliary lipid extraction rate , molecules in the patch and BSC is the number of BS molecules in Bile salts are specific organic detergents. Therefore, their interactions η with membranes obey the biophysical principles of detergent- the canalicular space per membrane unit. The parameter is a dependent membrane solubilization. The general consensus is a solubility index that represents the number of lipid molecules that are required to accommodate one BS molecule. The larger η, the three-stage model originally proposed by Helenius and Simons smaller is the membrane solubility of a BS. (Helenius and Simons, 1975) comprising: (1) partitioning of Assuming a fast equilibrium {i.e. [dPi N(BS)/dt]≈0}, Eqn 6 detergent between lipid bilayers and the aqueous media, described , K reduces to a recursive algebraic equation the solution of which by a partition coefficient , (2) formation of unstable membrane P patches rimmed by detergent BS, and (3) release of these unstable yields an explicit expression for i,N(BS): membrane patches in the aqueous phase under formation of mixed P ð Þ¼ 1 ; micelles. According to this scheme, our model assumes that BSs i 0 V intercalate into the Ld domains, with the hydrophilic part remaining at k BS YBS L P ð Þ¼ 1 BSc þ ½L ðk Þ þ i;N ; the surface and facilitating membrane budding and release of nascent i;N BS Nk i;N h 1 1 BS 1 ð7Þ V k¼ h N i 1 micelles. We defined a membrane patch of size at position as þ = BS 1 PLi;N h k QBS being extractable if it is fully encircled by lipids belonging to the Ld ¼ þ BSc þ ½L ðk Þ: V 1 Nk i;N h 1 phase: BS¼1 k¼1

X The ratio KBS=k+/k− of the forward and backward rate constants in patch of size N extractable $ sj ¼UN : Eqn 7 defines the partition coefficient of the BS. j[CN ðiÞ Using Eqn 7 in Eqn 5 yields an explicit expression for the extraction rate of a single patch. To examine the size distribution of The sum counts the ordering states of the lipids at the rim CN(i)ofthe extracted patches, we sum up Eqn 5 over all extractable patches of patch. UN denotes the length of the rim. In our lattice model it holds the same size N to obtain the mean extraction rate of patches with that: size N: rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ! N GN 4 1 r X X BS UN ¼ 3 1 : r ðNÞ¼ 0 P ð Þ: ð Þ 3 ex i BS 8 G0 N i¼1 BSUNþ1 Numerous experimental studies have demonstrated that the The secretion rate of any membrane species X (lipid or protein) solubilized membrane fraction increases with increasing associated with the extraction of patches with size N is: concentration of detergents. We thus made the plausible assumption that the patch extraction rate is proportional to the BS r XGN X rX ðNÞ¼ 0 BS P ð ÞX ; ð Þ concentration within the patch. Furthermore, it has been shown that ex N i BS i;N 9 G0 i¼ U detergents like BSs or polymers can arrange into pore-like membrane 1 BS Nþ1 structures facilitating the excision of the enclosed membrane where Xi,N is the number of molecules of species X contained in the fragment (Mazer et al., 1984; Schubert and Schmidt, 1988). We patch. Summing up Eqn 9 over all patch sizes N the mean secretion thus further postulated that the extraction of a membrane patch may rate of an arbitrary membrane species X finally reads: occur only if the number of intercalated BSs is large enough to form a r X XGN X rim of BS molecules around the patch (see also Fig. 7). Based on rX ðNÞ¼ 0 BS P ð ÞX : ð Þ i ex i BS i;N 10 these assumptions we define the rate of the -th hexagonal membrane G0 N N i¼1 BSUNþ1 patch from the set of all i ¼ 1; 2; ...; GN extractable patches with size N being extracted from the model membrane as follows: Comparing simulation results with published data r X r ði; NÞ¼ 0 BS P ð Þ; ð Þ Diffusion of membrane proteins ex N i;N BS 5 G0 First, we fixed the scaling factor rP in Eqn 3 such that the simulated BSUNþ1 diffusion coefficients of membrane proteins were in the expected where r0 is a scaling factor that is used to relate model-based lipid range. Putting rP/rL to 0.38 and 0.41 for the NRP and RP, respectively, extraction rates to in vivo measurements, 1=G0 is the probability for the model simulations correctly recapitulated the measured randomlyP choosing a specific patch from the whole set of relationship between the diffusivity of three different non-raft G0 ¼ N GN of possible patches and Pi,N(BS) is the probability proteins and their membrane densities (Fig. 1A). The overall (fraction) of BSs in the patch. The factor (BS/N) relates the extraction diffusion coefficients obtained as time averages across the diffusion 2 rate linearly to the BS density. The summation term in Eqn 5 runs coefficients for both types of domains were 0.8 μm /s for the NRPs Journal of Cell Science

3 RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs211524. doi:10.1242/jcs.211524

Fig. 1. Simulated diffusion of lipids and proteins. (A) Impact of protein density on the diffusion coefficient of a NRP imbedded into the Ld phase. Circles designate experimental data taken from Ramadurai et al. (2009): empty circles, dimeric lactose transporter (R=3 nm); filled circles, lipids [75% dioleoylglycerophosphocholine (DOPC), 25% dioleoylphosphatidylglycerol (DOPG)]. Squares designate simulated data: empty squares, NRPs comprising N=37 units (equal to the lattice cells); filled squares, lipids (14% CH and 86% PC ensuring adoption of a monophasic Ld phase). In order to make both systems comparable the diffusion coefficients are shown normalized to the diffusion coefficient of the protein-free lipid system. Very low protein densities (<0.1%) correspond to less than one protein on the lattice and thus could not be simulated. Conversely, high protein densities (>10%) were not accessible to the experiment. (B) Trajectories of selected NRPs and RPs monitored in time steps of 10 µs over a time interval of 1 ms in the steady state of a simulation of the canalicular membrane. Circles, NRPs moving in the Lo phase; dots, NRPs moving in the Lo phase; squares, RPs moving in the Ld phase; triangles, RPs moving in the Lo phase. and 0.07 μm2/s for the RPs. These values are in the range reported in In contrast, the presence of membrane proteins resulted in the several experimental studies (Dietrich et al., 2002; Kenworthy et al., formation of numerous small domains, irrespective of whether the 2004; Owen et al., 2009). The simulations uncovered large differences proteins were mobile or immobile (Fig. 2A,B). Simulated domain in the protein mobilities in the Ld and Lo domain (see typical patterns in the presence of mobile or immobile proteins yielded trajectories of the RP and NRP in Fig. 1B). Both the RPs and NRPs identical mean domain sizes. However, the temporal and spatial have much lower mobilities in Lo domains compared to what was stability of the domains were substantially different. Domains found for Ld domains. This is a consequence of the low lipid mobility forming in the presence of immobile proteins were stable, retained in Lo domains restricting the displacement of the lipid shell in front of the spatial location determined by the initial distribution of proteins the moving protein (equal to the motion in a viscous/crystalline and displayed only fluctuations of their phase borders. In contrast, environment). During random transits from one domain type to the domains formed in the presence of mobile proteins continuously other, the proteins switch between low- and high-mobility regimes. changed their spatial position and shape, and were occasionally completely disintegrated and replaced by newly formed domains. It Influence of membrane proteins on domain formation is worthwhile to note, that the domain patterns do not fit a We used our model to study how the presence of membrane proteins conventional ‘islands in the sea’ view [e.g. as concluded in Meder influences the formation of domains in the canalicular membrane of et al. (2006) from long-range diffusion of model proteins] but rather hepatocytes. In these simulations, we probed the effect of variations resemble a maze with ruffled border lines with both phases being in the protein density (the percentage of the membrane fraction continuous (percolating). occupied by proteins) and the protein mobility. The latter aspect is We used the total length of the phase borders separating Lo and important as membrane proteins can be mobile (free diffusion or Ld domains from each other to quantify the average domain size energy-dependent transport) or virtually immobile due to anchoring (see Fig. 2C). Constancy of its mean over a time interval of 50 ms to the cytoskeleton (Goiko et al., 2016; Kikuchi et al., 2002). was used as criterion for the accomplishment of a stationary domain The simulations were started with a random distribution of lipids size distribution. and proteins. The RP:NRP ratio was 47:53 corresponding to the ratio of membrane areas covered by Lo and Ld domains in the BS-dependent lipid secretion rates of wild-type mice protein-free case. This ratio is only slightly higher than the estimated The simulations outlined in this section were carried out for mobile share of ∼33% of total membrane protein resident in lipid rafts proteins for which (in contrast to immobile proteins) the domain purified from hepatocyte plasma membranes in a non-detergent patterning is an emergent property that is not predefined by the affinity chromatography method (Atshaves et al., 2007). choice of the initial protein distribution. As above, the reference case In the protein-free membrane, domains progressively grew and was defined by a protein density of 40%, a ratio NRP:RP=53:47 and collapsed finally into two completely segregated ‘macrodomains’. a lipid composition of SM:CH:PC=15.7:37.8:46.5. Based on Journal of Cell Science

4 RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs211524. doi:10.1242/jcs.211524

Fig. 2. Domain formation in the apical membrane of hepatocytes. The lipid composition of the model membrane was SM:CH:PC=15.7:37.8:46.5; protein composition was NRP:RP=53:47; and total protein:total lipid=0.4:0.6. The size of the model protein was N=37 nm2 corresponding to a protein radius of R≈3 nm. (A) Domain structures at different time points (upper row: 0.1 ms, lower row: 125 ms) for a protein-free leaflet (left column), randomly distributed immobile proteins (middle) and diffusing proteins (right column). (B) Phase borders between the Lo and Ld domains shown in A. The total length of the phase borders correlates inversely with the average domain size. (C) Time evolution of the total length of the phase borders used as measures of domain sizes. Proteins are diffusive. Protein densities are changed in steps of 10% from 0% (bottom) to 40% (top). Gray-shaded areas indicate the standard deviation of the stochastic simulations.

Eqns 8–10 we calculated the extraction rate of membrane patches BS-dependent lipid secretion rates of mice defective in lipid and membrane components (lipids, proteins) for various transporters concentrations of BS. The simulations were compared with data MDR3 (Mdr2 in mice) is an ATP-dependent transporter of from two independent experiments in which the taurocholate (TC)- amphipathic cationic and neutral compounds implicated in the dependent biliary lipid secretion rates of PC and CH were measured transmembrane transport of PC from the inner to the outer leaflet in wild-type mice and in mice either affected by a Mdr2 gene (Smit et al., 1993). Hence, genetic ablation of this transporter should disruption (Oude Elferink et al., 1995) (Fig. 3A,B) or carrying a lead to a decrease of the PC content in the outer leaflet. We used the genetic defect in the CH transporter ABCG5–ABCG8 (Groen et al., model to estimate the reduction of the PC content in the outer leaflet 2008) (Fig. 3C,D). As the absolute lipid secretion rates, half- that might account for the reduced PC and CH secretion rates saturation concentrations of TC and the biliary CH:PC ratio differed observed in mice heterozygous or homozygous for an Mdr2 gene in these two experiments, we used different values for the model disruption. To this end, we varied the PC content of the membrane parameter KBS=k+/k− (membrane/water partition coefficient) and from 0% to 50% while fixing the CH:SM ratio to 71:29, the protein slightly differing lipid compositions of the membrane to fit the density to 40% and putting the RP:NRP ratio equal to the size ratio wild-type experiments depicted in Fig. 3. of the membrane fraction occupied by Lo and Ld domains in the With increasing BS concentrations, the lipid secretion rates of PC protein-free membrane. The latter setting was made to take into and CH tend towards upper limit values. This saturation account that PC depletion of the membrane reduces the size and characteristic is well reproduced by the model. The half-maximal relative fraction of Ld domains thereby reducing the membrane area secretion rates for both PC and CH are obtained at BS secretion rates available for the accommodation of NRP. The best concordance of ∼150 nmol/min/100 g and ∼600 nmol/min/100 g. Differences in with the measured PC and CH output rates for the Mdr2+/− mice was these values are possibly due to differences in the design of the two obtained by lowering the PC content of the outer leaflet from the experiments (see Discussion). According to our secretion model, initial 47% to 37% for the heterozygous mice (Fig. 3B). This result the existence of a maximum for biliary lipid secretion reflects the suggests that an even modest reduction of the membrane PC content saturation of the membrane with BS because the solvation of each by about 20% was sufficient to account for the observed 30% drop BS molecule in the membrane ‘consumes’ η lipid molecules so that of the maximal PC secretion rate of the Mdr2+/− mice. Notably, with the secretion rate of an individual membrane patch cannot be further the estimated reduction of PC to 37%, the model also correctly elevated if all lipids are consumed. recapitulated the drop of the maximal CH secretion rate to 70% of As shown in Table 1, over the wide range of BS secretion rates Mdr2+/− mice. considered in the model simulations, the lipid composition of the We also studied how the CH content of the outer leaflet may normal bile shows a remarkable constancy. PC, CH and SM are influence BS-dependent lipid secretion rates. Membrane depletion extracted relative to each other in the ratio of about 88:11:1, in good of CH may result from an impaired delivery of CH to the membrane agreement with reported experimental values (≈85%, 15%, <1% or a genetic defect of the CH transporter Abcg5–Abcg8 (Groen

(Nibbering et al., 2001). et al., 2008). In these computations, we kept the PC content of the Journal of Cell Science

5 RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs211524. doi:10.1242/jcs.211524

Fig. 3. Taurocholate-dependent biliary secretion of PC and CH. (A,B) Secretion rates of PC and of CH for wild-type mice (filled circles) and the heterozygous Mdr2+/− mice (open circles). Experimental data were taken from Oude Elferink et al. (1995). The solid line represents the calculated mean secretion rates for the wild-type mice (PC content of the outer leaflet=47%). The dotted line represents the calculated mean secretion rates for the Mdr2+/− mice (PC content of the outer leaflet lowered to 37%). Shaded areas mark the range of computed secretion rates obtained from six recurrent independent stochastic model simulations. −4 +/− The model parameters were: KBS=5 · 10 and η=1, lipid compositions are 38% CH, 47% PC, 15% SM (wild type) and 45% CH, 37% PC, 18% SM (Mdr2 ). (C,D) Secretion rates of PC and of CH for wild-type mice (squares) and the homozygous Abcg8−/− mice (triangles). Experimental data were taken from Groen et al. (2008). The solid line represents the calculated mean secretion rates for the wild-type mice (CH content of the outer leaflet=25%). The dashed line represents the calculated mean secretion rates for the Abcg8−/− mice (CH content of the outer leaflet lowered to 13%). Shaded areas mark the range of −4 computed secretion rates obtained from 12 recurrent independent stochastic model simulations. The model parameters were: KBS=1 · 10 and η=1, lipid compositions are 24% CH, 47% PC, 29% SM (wild type) and 14% CH, 47% PC, 40% SM (Abcg8−/−). outer leaflet constant. This is equivalent to compensating for the content from an initial 38% to 25%. Such a reduction of the drop of CH with a corresponding increase of SM. Remarkably, this membrane content of the Abcg5–Abcg8-deficient mice was indeed setting did not change the relative abundance and size distribution of reported in Kosters et al. (2006). microdomains and hence preserved the integrity of the membrane. We also investigated the effect of a pathologically high CH This model finding is in concordance with the observation that no content in the outer leaflet on biliary CH secretion. Varying the obvious physical differences were apparent between the Abcg5– membrane content of CH in a broad range between 11% and 51% at Abcg8-deficient mice and wild-type mice (Yu et al., 2002a). The a typical BS output of 600 nmol/min/100 g yielded a stable domain-structure-preserving effect achieved by replacing CH with coexistence of Lo and Ld microdomains. The relative CH content SM is due to the fact that SM alone is capable of creating a stable Lo of the bile increased in a hyper-linear fashion up to 25% at a phase. This model-based finding has also been demonstrated membrane CH content of 51% (see Fig. 4A). Such a fivefold experimentally (Frisz et al., 2013; Wang and Silvius, 2003). As increased CH content of the bile is in good agreement with reported shown in Fig. 4A, a decrease of the CH content of the membrane values for the Abcg5–Abcg8-overexpressing mice (Yu et al., was paralleled by a pronounced decrease of the CH secretion rate, 2002b). At high CH loading of the membrane, the model predicts a whereas the PC secretion rate remained virtually unaffected. A good critical range of the biliary BS content (at ∼70%) at which the lipid concordance between the simulated and measured BS dependence composition of mixed micelles may even favor the formation of bile of the CH secretion rate was obtained by reducing the membrane CH crystals (see red curve in Fig. 4B). This model prediction is

Table 1. Extraction rates of membrane components at different concentrations of BS 50 nmol/min BS 150 nmol/min BS 450 nmol/min BS μg/min nmol/min μg/min nmol/min μg/min nmol/min CH 0.41 (5.4%) 1.07 (11.5%) 1.43 (5.6%) 3.69 (11.3%) 2.19 (5.7%) 5.66 (11.2%) PC 6.39 (84.4%) 8.13 (87.5%) 22.46 (88.4%) 28.57 (87.7%) 34.66 (89.6%) 44.09 (87.8%) SM 0.06 (0.7%) 0.09 (0.9%) 0.19 (0.8%) 0.30 (0.9%) 0.29 (0.8%) 0.46 (0.9%) RP 0.02 (0.2%) 0.0003 (≈0%) 0.03 (0.1%) 0.0005 (≈0%) 0.03 (0.1%) 0.0005 (≈0%) NRP 0.70 (9.2%) 0.01 (0.1%) 1.30 (5.1%) 0.02 (0.1%) 1.50 (3.9%) 0.03 (0.1%) The BS (taurocholate) concentration was set at 50, 150 or 450 nmol/min. Extraction rates are present as μg/min and nmol/min. The percentages of each membrane component is given in parentheses. The average molecular mass of membrane proteins was set to 55 kDa. Journal of Cell Science

6 RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs211524. doi:10.1242/jcs.211524

Fig. 4. Impact of membrane CH on biliary CH content and formation of bile crystals. (A) Computed lipid composition of the bile (red, SM; green, PC; blue, CH) as function of the CH content of the membrane. The sum of CH and SM in the membrane was kept at 53.5%. (B) Equilibrium CH–BS–phospholipid ternary phase diagram for the BS taurocholate adopted from Portincasa et al. (2002). The blue and red line represent the computed BS-dependent biliary content of PC, CH at a membrane CH content of 38% (normal case) and 51% (cholesterol hyper saturation). consistent with the observation in Yu et al. (2002b) showing that the content of the external leaflet and the physico-chemical properties bile from transgenic animals was opaque and had large amounts of of the BS), we have made the assumption that the predicted increase amorphous material as determined by microscopic inspection. of the maximal secretion rate can be fully counter-balanced by the rate of membrane replenishment. Size distribution and lipid composition of extracted membrane In our model, increasing membrane solubility of the BSs is patches represented by decreasing values of the solubility index η As shown in Fig. 5A, our model predicts a rather narrow size determining the number of PC molecules required to adopt one BS distribution of secreted membrane patches (which are nascent mixed molecule. Changing this parameter from 1.5 to 0.75 results in an micelles) around the peak value at N=20 lipids. The absence of larger about fourfold increase in the maximal lipid secretion rate (Fig. 6B). patches (N>100) reflects the size distribution of Ld domains, which We conclude that elevating the production of BS with high membrane on average have a size of about 35 lipids. It has to be noted that the solubility in PC-rich Ld microdomains could be a convenient means predicted size range of extracted patches (≈20–60 lipids) is below the to maintain an almost normal lipid secretion rate in cases where the size range of 100–400 lipids reported for primary PC micelles export of otherwise physiologically abundant BS is hampered, for (Hofmann and Small, 1967; Small et al., 1969) and for micelles in the example, due to a genetic defect in the bile salt export pump. For native bile from the dog (Mazer et al., 1984). This discrepancy example, secretion of an unexpectedly large amount of tetra- suggests that the extracted nascent micelles are unstable and rapidly hydroxylated bile acids acid was observed in mice carrying a aggregate to larger stable micelles (Nichols and Ozarowski, 1990). mutated bile salt pump. The total bile salt output in these mutant mice Notably, despite the large impact of variations in the protein density was still ∼30% of wild type although the secretion of cholic acid was and PC content on absolute extraction rates of patches, there is little greatly reduced to 6% (Wang et al., 2001). From this observation, one impact on the relative composition of the bile (Fig. 5B). might conclude that the solubility index η correlates inversely with the hydrophilicity of the BS. On the other hand, the effect of five Biliary lipid secretion at various protein density and for different BS different BSs on bile flow revealed a higher membrane-solubilizing species potency for hydrophobic BS (Loria et al., 1989). Finally, other studies Finally, we studied the possible impact of changing protein have failed to establish a significant relationship between bile flow concentrations and membrane solubility of BS on the lipid and BS hydrophobicity (Gurantz and Hofmann, 1984). Hence, there extraction rate. As expected from the strong influence of the is apparently no simple relationship between the hydrophobicity of protein density on the size of Ld domains (see Fig. 2A), even BS and the solubility index η. moderate changes of the protein density from 40% to 35% or 45%, respectively, gave rise to a change of lipid secretion rates for PC and DISCUSSION CH by a factor of about two (Fig. 6A). Intriguingly, an increase of In this work, we developed a mathematical model of microdomain the protein content to 45% lowers the lipid secretion rate to the same formation in the canalicular membrane of hepatocytes and the extent as a decrease of the PC content to 32%, which is below the solubilization of bile micelles from Ld microdomains. Our aim was value associated with the lowered capacity of the heterozygous to check how the lipid and protein composition of the canalicular

MDR2 (37%). Note that in these simulations (changing the protein membrane might control the biliary secretion rate of PC and CH and Journal of Cell Science

7 RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs211524. doi:10.1242/jcs.211524

Fig. 5. Impact of membrane protein and PC on patch extraction and secretion rates. (A) Size distribution of extracted membrane patches for membrane protein densities of 40%, 30% and 20% (membrane PC content: 47%) and different membrane PC content (inset) of 47%, 32% and 22% (membrane protein density: 40%). −4 Values are calculated at the half-maximal BS secretion of 150 nmol/min/100 g with parameters KBS=5 · 10 and η=1. (B) The secretion rate of PC (green), CH (blue), SM (red) and proteins (black) are shown for different membrane protein densities (membrane PC content: 47%) and different membrane PC content −4 (membrane protein density: 40%). Values are calculated at the half-maximal BS secretion of 150 nmol/min/100 g with parameters KBS=5 · 10 and η=1. whether variations in the lipid and protein content of the membrane membrane proteins in the structural organization and secretory might account for a diverse set of seemingly contradictory function of the canalicular membrane, the mere presence of experimental findings. membrane proteins appears to have a significant impact on the structure and size of microdomains and thereby on the lipid Membrane proteins restrain the size of lipid domains secretion rate (Eckstein et al., 2015). Depletion of the protein Segregation of membrane lipids into domains differing in lipid density of the plasma membrane, for example, owing to enhanced composition, ordering state of lipids and preference for specific protein internalization at reduced cellular ATP level, as reported for membrane proteins has been implicated in the organization and several epithelial cell membranes (Doctor et al., 2000; Mandel et al., sorting of functional protein complexes, formation of connections 1994), should render the membrane more vulnerable to BS- between the membrane and the cytoskeleton and budding of dependent destruction and accelerate the cytotoxic effect of BS. membrane fragments in endocytosis and exocytosis (Laude and Another example for the protective role of membrane proteins Prior, 2004). With respect to the canalicular membrane of against potentially toxic bile acids may be the decreased resistance hepatocytes, a specific function of coexisting Lo and Ld of the hepato-canalicular membrane to hydrophobic bile salts microdomains is enabling the extraction of membrane fragments observed in progressive familial intrahepatic cholestasis type 1 without endangering membrane integrity. With the aim to better (PFIC1), which is caused by a deficiency of the phospholipid understand the molecular mechanisms governing the size, shape flippase ATP8B1 (Paulusma et al., 2006). The liver of PFIC1 and distribution of microdomains, we have extended our previous mice lack immunohistochemically demonstrable canalicular mathematical model of domain formation (Eckstein et al., 2015) by ectoenzymes. In the light of our model findings, this protein including the interaction of lipids with membrane proteins. The depletion of the outer leaflet may substantially contribute to main finding of model simulations is that the presence of membrane the reduced membrane resistance against BS, as well as the proteins with differing preferences for specific lipid species phospholipid membrane asymmetry accompanying the gene defect. prevents the complete segregation of lipids into the large ‘macrodomains’ that are typically observed in artificial lipid The mechanism of BS-dependent biliary lipid extraction membranes. The protein:lipid ratio of the membrane determines In concordance with the general biophysical principles of detergent- the average domain size and hence the size of membrane patches dependent membrane solubilization (Helenius and Simons, 1975; extracted into the bile. Whereas a domain-inducing effect of Lichtenberg et al., 2013), we postulated a lipid secretion mechanism proteins has been already addressed in several experimental studies (B-model) according to which BSs may accumulate in Ld domains (reviewed in Jacobson et al., 2007) and also by model simulations in a saturating manner. If the BS concentration in a membrane patch (Hinderliter et al., 2001), this is the first report demonstrating, by is sufficiently high enough to form circular structures that shield the means of a carefully validated membrane model, that proteins – rim of the patch, extraction of the patch may occur (see Fig. 7). The irrespective of their mode of motion in the membrane – are capable assumption of such an ‘encircle and cut out’ mechanism of BS- of restraining the formation of lipid domains to the nanoscale size triggered patch release is further backed up by observations showing range. Hence, besides the specific functions of the numerous that polymeric detergents as well as bile acids may form pore-like Journal of Cell Science

8 RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs211524. doi:10.1242/jcs.211524

Fig. 6. Impact of membrane protein content and the membrane solubility of BS on lipid secretion into the bile. (A) Secretion rates of PC at normal lipid composition of the membrane and three different protein densities. The middle line refers to a normal density of 40%, the upper and lower line refer to a reduced (35%) and increased (45%) density of membrane proteins. Shaded areas mark the range of computed secretion rates obtained from 12 recurrent −4 independent stochastic model simulations. The parameters used were KBS=5 · 10 and η=1. (B) Secretion rates of PC at normal lipid composition of the membrane but altered model parameters for the interaction of BS with the membrane: η=1.5 (low membrane solubility of the BS), the middle to η=1.0 (normal −4 solubility), η=0.75 (high solubility). The solid lines were computed by setting the value of the partition coefficient as KBS=5 · 10 . The shaded areas show the −4 −3 influence of varying KBS; the lower margin of the shaded areas refer to KBS=2.5 · 10 , the upper margins refer to KBS=1 · 10 .

−4 structures in the membrane (Mazer et al., 1984; Schubert and obtained with setting KBS=k+/k−=5 · 10 and η=1, that is, one lipid Schmidt, 1988). For the BS taurocholate, the best agreement molecule is ‘consumed’ to dissolve one BS molecule in the Ld between the measured and simulated lipid extraction rates was phase of the membrane. Of note, these two model parameters are BS specific. According to our model simulations, the maximal lipid secretion rates shown in Fig. 3 are accounted for by the saturation of Ld domains with the BS taurocholate. From this it follows that the capacity of lipid transporters and of the intracellular transport of lipids and proteins to the apical membrane is not a limiting factor for TC-dependent lipid extraction as long as the size of Ld microdomains can be kept at an average size of 20–40 lipids and the total leaflet area occupied by Ld domains is not larger than 50%. However, a shift of the canalicular BS composition towards BSs with a higher membrane solubility (i.e. lower values of the solubility index η and/or larger values of the partition coefficient KBS)may give rise to a significant left- or right-shift of the lipid secretion curves (see Fig. 6B) and thus may lead to an imbalance between the membrane-replenishing capacity of the hepatocyte and the elevated extraction of membrane patches. Along this line of argumentation, the differences in the absolute values of the lipid secretion rates obtained in the two independent experiments shown in Fig. 3 may at Fig. 7. Schematic of the proposed membrane budding model (B-model) least be partially due to variations in the BS composition of the bile of biliary lipid secretion. (1) Trans-bilayer movement of PC from the inner to actually present in the canalicular lumen. The infused TC is taken up the outer leaflet mediated by MDR3. (2) Trans-bilayer movement of CH from by the hepatocyte and can be chemically modified into different BS the inner to the outer leaflet mediated by Abcg5–Abcg8. (3) Trans-bilayer flip- species having a higher or lower hydrophobic index than TC. flop of BS between the inner and outer leaflet. (4) Transport of BS by BSEP Of note, the half-saturating TC secretion rate of 150 nmol/min/ probably operating as a ‘liftase’ presenting CH to luminal BS. (5) Reversible 100 g and 600 nmol/min/100 g obtained for the wild-type mice partition of BS monomers between the aqueous lumen and Ld microdomains (see Fig. 3) is above the rate that is required to solubilize 8% of of the membrane. (6) Formation of BS micelles if the critical micellar concentration of the BS is reached. (7) Arrangement of BS monomers in ring- the canalicular membrane per minute as estimated by Crawford like structures facilitating membrane budding. (8) Membrane release of disk- (Crawford, 1996). Given a liver mass of 3–5% of body weight in 6 like nascent micelles (Hjelm et al., 1992) wrapped by a layer of BS molecules. mice (Nesteruk et al., 2014), 135×10 hepatocytes per g mice liver Journal of Cell Science

9 RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs211524. doi:10.1242/jcs.211524

(Sohlenius-Sternbeck, 2006) and 218 μm² canalicular membrane redistribution of CH, the Ld phase becomes depleted in CH and surface per hepatocyte (Crawford, 1996), the total area of the apical accordingly the released micelles have a lower relative CH content. membrane available for bile formation amounts to 117.72×109 μm²/ 100 g. Assuming 106 lipids per μm² area of the outer leaflet of the Arguments in favor of the budding model membrane, the lipid secretion rate can be roughly estimated as There are several aspects conferring a high degree of credibility to 16 nmol/min/100 g. With TC as the dominant BS in the experiments our proposed membrane budding model (B-model) of biliary lipid shown in Fig. 3, such a secretion rate is achieved with TC secretion secretion (Fig. 7). First, the model simulations recapitulate a large rates of 110 nmol/min/100 g in Mdr2+/− mice and 85 nmol/min/ array of reported data on the lipid composition of the bile and BS- 100 g in the corresponding wild-type mice or 250 nmol/min/100 g dependent lipid secretion rates under normal conditions and in Abcg8−/− mice and 200 nmol/min/100 g in the corresponding perturbation of trans-bilayer lipid transport. Second, the model is wild-type mice. All these TC secretion rates are below the estimated built on numerous experimental findings on the formation of half-saturation secretion rates. microdomains obtained in artificial membrane systems and on top takes into account the interaction of lipids and proteins as a crucial The importance of the membrane lipid composition on biliary domain-stabilizing factor in real biological membranes. Third, the extraction rates disproportional appearance of PC and CH in the bile observed under Experiments in mice with knockdown or overexpression of specific extreme situations (CH overload, knock-down or overexpression of ABC transporters involved in trans-bilayer lipid translocation have lipid translocase) is explained by corresponding variations in the provoked a renewed thinking about the mechanism of biliary lipid lipid composition of Ld microdomains. In our simulations, the CH: secretion. PC ratio in the Ld microdomain varied between 0.01 and 0.39. A Mice with a disruption of the Mdr2 gene (the mouse homolog of maximal solubility limit of up to 66% has been determined for PC MDR3), were found to be unable to transport PC into bile (Smit bilayers (Huang et al., 1999). Fourth, the secretion mechanism et al., 1993). Human MDR3 (ABCB4) mutations result in a wide obeys the general biophysical principles governing the detergent- spectrum of phenotypes, ranging from progressive familial mediated solubilization of biological membranes in general intrahepatic cholestasis type 3 (PFIC3) to adult cholestatic liver (Helenius and Simons, 1975; Lichtenberg et al., 2005). In disorders (Jacquemin et al., 2001). Intriguingly, our model particular, the controlled loss of membrane fragments in the simulations suggest that a 30% decrease in the lipid secretion rate, process of bile secretion and the disproportional membrane- as observed in Mdr2 heterozygous mice can be accounted for by a damaging loss of membrane fragments at critically high BS modest PC depletion (from 47% to 37%) of the membrane. Usually concentrations can be understood as two facets of one and the the heterozygous genotype with a functional allele and a non- same basic mechanism. The high sensitivity of the patch extraction functional allele produces 50% of the standard activity. In this case, rate against changes in the protein content of the membrane (see the PC content of the membrane should be ∼23% and not 37% as Fig. 6A) has to be seen in the context of BS-dependent membrane predicted by the model. Hence, there appears to be a compensation damage. If the BS concentration reaches a critical threshold where for the loss of the non-functional allele, either by enhanced gene the extraction of membrane proteins exceeds their replenishment by expression and anterograde transport of the protein to the membrane intra-cellular vesicular transport, the protein:lipid ratio of the and/or reduced retrograde transport from the membrane to the membrane is decreasing. This may invoke a self-amplifying circle lysosome. as the lowering of the protein:lipid ratio increases the patch Mutations in human genes encoding for ABCG5 or ABCG8 have extraction rate. Several studies (Billington et al., 1980; Holdsworth been demonstrated to cause sitosterolemia, characterized by an and Coleman, 1976) have reported that a protein loss of ∼20% accumulation of sterols in blood and tissues, consequent to the precedes BS-dependent membrane lysis. According to our model, a enhanced intestinal absorption and decreased biliary removal of CH protein depletion of that size would result in an about 30-fold and plant sterols (Berge et al., 2000). Yu et al. (2002a) have found increase of the membrane extraction rate which may define the that the CH molar ratio in the bile was significantly lower in the upper limit for the increase of intracellular processes involved in Abcg5–Abcg8−/− mice (0.37%) than in the littermate controls membrane re-filling. In addition, the model may also explain the (2.2%) whereas the PC content of the bile was not significantly paradoxical finding that inhibition of the bile salt export pump altered. Conversely, overexpression of Abcg5–Abcg8 in mice was (BSEP; also known as ABCB11) results in an even higher secretion accompanied by a more than fivefold increase of biliary CH (Yu rate of CH and PC in BSEP−/− mice (Wang et al., 2001). Bile acids et al., 2002b). The outcome of these experiments can be well may intercalate into the outer leaflet not only from the luminal side reproduced by our model if we make the following assumptions: (1) but also through translocation from the inner leaflet (Donovan and that similar to other trans-bilayer lipid transporters, Abcg5–Accg8 Jackson, 1997) (process number 3 in Fig. 7). This represents a works primarily as a flippase translocating CH from the inner to the special secretion mode that works without luminal presence of BS in outer leaflet. Indeed, in the Abcg5–Abcg8-deficient mice, the CH micellar concentrations as required for CH secretion according to content of the membrane was found to be reduced by ∼50% the E-model (Vrins et al., 2007). In response to the blocked BS (Kosters et al., 2006) suggesting that changes of the Abcg5–Accg8 transporter, BSEP−/− mice upregulate the production of tetra- activity are accompanied by analogous changes of the CH content in hydroxy BSs, which owing to their very high hydrophilicity should the membrane. (2) That changes of the membrane CH are be able to maintain a secretion rate comparable with that of the wild- counterbalanced by inverse changes of the SM content. In our type mice. Changing in our model the parameters KBS and η three-lipid model, this condition appears to be indispensable in determining the solubility of BS in the Ld phase has a high impact −4 order to preserve the microdomain architecture of the outer leaflet on the secretion rate. Increasing KBS from KBS=2.5 · 10 to −3 and hence the functional and structural integrity of the membrane. KBS=10 shifts the half-saturation constant from 220 down to Replacement of CH by SM preserves the size distribution of Lo and 60 nmol/min/100 g, and decreasing the parameter η from η=1 to Ld domains because increasing SM stabilizes Lo microdomains by η=3/4 raises the saturation level of the secretion by 30% (see dragging more CH from the Ld into the Lo phase. Owing to this Fig. 6B). These changes together may elicit a 6-fold higher Journal of Cell Science

10 RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs211524. doi:10.1242/jcs.211524 solubilization efficiency. Finally, budding from the external hemi- Holdsworth, G. and Coleman, R. (1976). Plasma-membrane components can be leaflet provides an explanation of how luminal BSs can extract large removed from isolated lymphocytes by the bile salts glycocholate and taurocholate without cell lysis. Biochem. J. 158, 493-495. quantities of phospholipid on the basis of their detergent action, Huang, J., Buboltz, J. T. and Feigenson, G. W. (1999). Maximum solubility of without disrupting the integrity of the detergent-resistant canalicular cholesterol in phosphatidylcholine and phosphatidylethanolamine bilayers. plasma membrane. Biochim. Biophys. Acta. 1417, 89-100. Jacobson, K., Mouritsen, O. G. and Anderson, R. G. W. (2007). Lipid rafts: at a crossroad between cell biology and physics. Nat. Cell. Biol. 9, 7-14. Jacquemin, E., De Vree, J. M., Cresteil, D., Sokal, E. M., Sturm, E., Dumont, M., Competing interests Scheffer, G. L., Paul, M., Burdelski, M., Bosma, P. J. et al. (2001). The wide The authors declare no competing or financial interests. spectrum of multidrug resistance 3 deficiency: from neonatal cholestasis to cirrhosis of adulthood. Gastroenterology 120, 1448-1458. Author contributions Kahya, N., Scherfeld, D., Bacia, K., Poolman, B. and Schwille, P. (2003). Probing Conceptualization: H.H., N.B.; Methodology: J.E., H.H., N.B.; Software: J.E.; Formal lipid mobility of raft-exhibiting model membranes by fluorescence correlation analysis: J.E.; Investigation: J.E.; Data curation: J.E.; Writing - original draft: J.E., spectroscopy. J. Biol. Chem. 278, 28109-28115. H.H., N.B.; Writing - review & editing: J.E., H.H., N.B.; Visualization: J.E., H.H., N.B.; Kenworthy, A. K., Nichols, B. J., Remmert, C. L., Hendrix, G. M., Kumar, M., Supervision: H.H., N.B.; Project administration: H.H., N.B.; Funding acquisition: H.H. Zimmerberg, J. and Lippincott-Schwartz, J. (2004). Dynamics of putative raft- associated proteins at the cell surface. J. Cell Biol. 165, 735-746. Funding Kikuchi, S., Hata, M., Fukumoto, K., Yamane, Y., Matsui, T., Tamura, A., This work was supported by Deutsche Forschungsgemeinschaft (Graduate School Yonemura, S., Yamagishi, H., Keppler, D., Tsukita, S. et al. (2002). Radixin deficiency causes conjugated hyperbilirubinemia with loss of Mrp2 from bile 1772 ‘Computational Systems Biology’) and by Bundesministerium für Bildung und canalicular membranes. Nat. Genet. 31, 320-325. Forschung (LiSyM, grant no. 31 L0057). Deposited in PMC for immediate release. Kosters, A., Kunne, C., Looije, N., Patel, S. B., Oude Elferink, R. P. and Groen, A. K. (2006). The mechanism of ABCG5/ABCG8 in biliary cholesterol secretion in References mice. J. Lipid Res. 47, 1959-1966. Atshaves, B. P., Mcintosh, A. L., Payne, H. R., Gallegos, A. M., Landrock, K., Laude, A. J. and Prior, I. A. (2004). Plasma membrane microdomains: Maeda, N., Kier, A. B. and Schroeder, F. (2007). SCP-2/SCP-x gene ablation organization, function and trafficking. Mol. Membr. Biol. 21, 193-205. alters lipid raft domains in primary cultured mouse hepatocytes. J. Lipid Res. 48, Lichtenberg, D., Goni, F. M. and Heerklotz, H. (2005). Detergent-resistant 2193-2211. membranes should not be identified with membrane rafts. Trends Biochem. Sci. Berge, K. E., Tian, H., Graf, G. A., Yu, L., Grishin, N. V., Schultz, J., Kwiterovich, 30, 430-436. P., Shan, B., Barnes, R. and Hobbs, H. H. (2000). Accumulation of dietary Lichtenberg, D., Ahyayauch, H. and Goñi, F. M. (2013). The mechanism of cholesterol in sitosterolemia caused by mutations in adjacent ABC transporters. detergent solubilization of lipid bilayers. Biophys. J. 105, 289-299. Science 290, 1771-1775. Loria, P., Carulli, N., Medici, G., Tripodi, A., Iori, R., Rovesti, S., Bergomi, M., Billington, D., Evans, C. E., Godfrey, P. P. and Coleman, R. (1980). Effects of bile Rosi, A. and Romani, M. (1989). Determinants of bile secretion: effect of bile salt salts on the plasma membranes of isolated rat hepatocytes. Biochem. J. 188, structure on bile flow and biliary cation secretion. Gastroenterology 96, 321-327. 1142-1150. ̈ Contreras, F. X., Ernst, A. M., Wieland, F. and Brugger, B. (2011). Specificity of Mandel, L. J., Doctor, R. B. and Bacallao, R. (1994). ATP depletion: a novel intramembrane protein-lipid interactions. Cold Spring Harb Perspect Biol, 3, method to study junctional properties in epithelial tissues. II. Internalization of Na a004705. +,K(+)-ATPase and E-cadherin. J. Cell Sci. 107, 3315-3324. Crawford, J. M. (1996). Role of vesicle-mediated transport pathways in Marrink, S. J. and Mark, A. E. (2002). Molecular dynamics simulations of mixed hepatocellular bile secretion. Semin. Liver Dis. 16, 169-189. micelles modeling human bile. Biochemistry 41, 5375-5382. Crawford, J. M., Mockel, G. M., Crawford, A. R., Hagen, S. J., Hatch, V. C., Mazer, N. A., Schurtenberg, P., Carey, M. C., Preisig, R., Weigand, K. and Barnes, S., Godleski, J. J. and Carey, M. C. (1995). Imaging biliary lipid Kanzig, W. (1984). Quasi-elastic light scattering studies of native hepatic bile from secretion in the rat: ultrastructural evidence for vesiculation of the hepatocyte the dog: comparison with aggregative behavior of model biliary lipid systems. canalicular membrane. J. Lipid Res. 36, 2147-2163. Biochemistry 23, 1994-2005. Dietrich, C., Yang, B., Fujiwara, T., Kusumi, A. and Jacobson, K. (2002). Mazzone, A., Tietz, P., Jefferson, J., Pagano, R. and Larusso, N. F. (2006). Relationship of lipid rafts to transient confinement zones detected by single Isolation and characterization of lipid microdomains from apical and basolateral particle tracking. Biophys. J. 82, 274-284. plasma membranes of rat hepatocytes. Hepatology 43, 287-296. Doctor, R. B., Dahl, R. H., Salter, K. D., Fouassier, L., Chen, J. and Fitz, J. G. Meder, D., Moreno, M. J., Verkade, P., Vaz, W. L. C. and Simons, K. (2006). (2000). ATP depletion in rat cholangiocytes leads to marked internalization of Phase coexistence and connectivity in the apical membrane of polarized epithelial membrane proteins. Hepatology 31, 1045-1054. Donovan, J. M. and Jackson, A. A. (1997). Transbilayer movement of fully ionized cells. Proc. Natl. Acad. Sci. USA 103, 329-334. taurine-conjugated bile salts depends upon bile salt concentration, Nesteruk, M., Hennig, E. E., Mikula, M., Karczmarski, J., Dzwonek, A., Goryca, hydrophobicity, and membrane cholesterol content. Biochemistry 36, K., Rubel, T., Paziewska, A., Woszczynski, M., Ledwon, J. et al. (2014). 11444-11451. Mitochondrial-related proteomic changes during obesity and fasting in mice are Eckstein, J., Berndt, N. and Holzhutter, H.-G. (2015). Computer simulations greater in the liver than skeletal muscles. Funct. Integr. Genomics 14, 245-259. suggest a key role of membranous nanodomains in biliary lipid secretion. PLoS Nibbering, C. P., Groen, A. K., Ottenhoff, R., Brouwers, J. F. H. M., Vanberge- Comput. Biol. 11, e1004033. Henegouwen, G. P. and Van Erpecum, K. J. (2001). Regulation of biliary Frisz, J. F., Lou, K., Klitzing, H. A., Hanafin, W. P., Lizunov, V., Wilson, R. L., cholesterol secretion is independent of hepatocyte canalicular membrane lipid Carpenter, K. J., Kim, R., Hutcheon, I. D., Zimmerberg, J. et al. (2013). Direct composition: a study in the diosgenin-fed rat model. J. Hepatol. 35, 164-169. chemical evidence for sphingolipid domains in the plasma membranes of Nichols, J. W. and Ozarowski, J. (1990). Sizing of lecithin-bile salt mixed micelles fibroblasts. Proc. Natl. Acad. Sci. USA 110, E613-E622. by size-exclusion high-performance liquid chromatography. Biochemistry 29, Goiko, M., De Bruyn, J. R. and Heit, B. (2016). Short-lived cages restrict protein 4600-4606. diffusion in the plasma membrane. Sci. Rep. 6, 34987. Oude Elferink, R. P., Ottenhoff, R., Van Wijland,, M., Smit, J. J., Schinkel, A. H. Groen, A., Kunne, C., Jongsma, G., Van Den Oever, K., Mok, K. S., Petruzzelli, and Groen, A. K. (1995). Regulation of biliary lipid secretion by mdr2 P- M., Vrins, C. L., Bull, L., Paulusma, C. C. and Oude Elferink, R. P. (2008). glycoprotein in the mouse. J. Clin. Invest. 95, 31-38. Abcg5/8 independent biliary cholesterol excretion in Atp8b1-deficient mice. Owen, D. M., Williamson, D., Rentero, C. and Gaus, K. (2009). Quantitative Gastroenterology 134, 2091-2100. microscopy: protein dynamics and membrane organisation. Traffic 10, 962-971. Gurantz, D. and Hofmann, A. F. (1984). Influence of bile acid structure on bile flow Paulusma, C. C., Groen, A., Kunne, C., Ho-Mok, K. S., Spijkerboer, A. L., Rudi and biliary lipid secretion in the hamster. Am. J. Physiol. 247, G736-G748. De Waart, D., Hoek, F. J., Vreeling, H., Hoeben, K. A., Van MARLE, J. et al. Helenius, A. and Simons, K. (1975). Solubilization of membranes by detergents. (2006). Atp8b1 deficiency in mice reduces resistance of the canalicular Biochim. Biophys. Acta. 415, 29-79. membrane to hydrophobic bile salts and impairs bile salt transport. Hepatology Hinderliter, A., Almeida, P. F. F., Creutz, C. E. and Biltonen, R. L. (2001). Domain 44, 195-204. formation in a fluid mixed lipid bilayer modulated through binding of the C2 protein Portincasa, P., Venneman, N. G., Moschetta, A., Van den Berg, A., Palasciano, motif. Biochemistry 40, 4181-4191. G., Vanberge-Henegouwen, G. P. and Van Erpecum, K. J. (2002). Quantitation Hjelm, R. P., Thiyagarajan, P. and Alkan-Onyuksel, H. (1992). Organization of of cholesterol crystallization from supersaturated model bile. J. Lipid. Res. 43, phosphatidylcholine and bile salt in rodlike mixed micelles. J. Phys. Chem. 96, 604-610. 8653-8661. Ramadurai, S., Holt, A., Krasnikov, V., Van Den BOGAART, G., Killian, J. A. and Hofmann, A. F. and Small, D. M. (1967). Detergent properties of bile salts: Poolman, B. (2009). Lateral diffusion of membrane proteins. J. Am. Chem. Soc.

correlation with physiological function. Annu. Rev. Med. 18, 333-376. 131, 12650-12656. Journal of Cell Science

11 RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs211524. doi:10.1242/jcs.211524

Schubert, R. and Schmidt, K. H. (1988). Structural changes in vesicle membranes Vrins, C., Vink, E., Vandenberghe, K. E., Frijters, R., Seppen, J. and Groen, A. K. and mixed micelles of various lipid compositions after binding of different bile (2007). The sterol transporting heterodimer ABCG5/ABCG8 requires bile salts to salts. Biochemistry 27, 8787-8794. mediate cholesterol efflux. FEBS Lett. 581, 4616-4620. Small, D. M. (2003). Role of ABC transporters in secretion of cholesterol from liver Wang, T.-Y. and Silvius, J. R. (2003). Sphingolipid partitioning into ordered into bile. Proc. Natl. Acad. Sci. USA 100, 4-6. domains in cholesterol-free and cholesterol-containing lipid bilayers. Biophys. J. Small, D. M., Penkett, S. A. and Chapman, D. (1969). Studies on simple and mixed 84, 367-378. bile salt micelles by nuclear magnetic resonance spectroscopy. Biochim. Biophys. Wang, R., Salem, M., Yousef, I. M., Tuchweber, B., Lam, P., Childs, S. J., Acta. 176, 178-189. Helgason, C. D., Ackerley, C., Phillips, M. J. and Ling, V. (2001). Targeted Smit, J. J., Schinkel, A. H., Oude Elferink, R. P., Groen, A. K., Wagenaar, E., van inactivation of sister of P-glycoprotein gene (spgp) in mice results in Deemter, L., Mol, C. A., Ottenhoff, R., van Der lugt, N. M., van Roon, M. A. & nonprogressive but persistent intrahepatic cholestasis. Proc. Natl. Acad. Sci. et al. (1993). Homozygous disruption of the murine mdr2 P-glycoprotein gene USA 98, 2011-2016. leads to a complete absence of phospholipid from bile and to liver disease. Cell 75, Yu, L., Hammer, R. E., Li-Hawkins, J., Von BERGMANN, K., Lutjohann, D., 451-462. Cohen, J. C. and Hobbs, H. H. (2002a). Disruption of Abcg5 and Abcg8 in mice Sohlenius-Sternbeck, A.-K. (2006). Determination of the hepatocellularity number reveals their crucial role in biliary cholesterol secretion. Proc. Natl. Acad. Sci. USA for human, dog, rabbit, rat and mouse livers from protein concentration 99, 16237-16242. measurements. Toxicol. In Vitro 20, 1582-1586. Yu, L., Li-Hawkins, J., Hammer, R. E., Berge, K. E., Horton, J. D., Cohen, J. C. Tietz, P., Jefferson, J., Pagano, R. and Larusso, N. F. (2005). Membrane and Hobbs, H. H. (2002b). Overexpression of ABCG5 and ABCG8 promotes microdomains in hepatocytes: potential target areas for proteins involved in biliary cholesterol secretion and reduces fractional absorption of dietary canalicular bile secretion. J. Lipid Res. 46, 1426-1432. cholesterol. J. Clin. Invest. 110, 671-680. Journal of Cell Science