Introduction

Intercellular adhesion molecule (ICAM)-1 is the counterreceptor of leukocyte β2-integrins and mediates firm adhesion of leukocytes to epithelial and endothelial cells (Rothlein et al, 1986; Sumagin et al, 2014; Reglero-Real et al, 2012; Vestweber, 2015). ICAM-1 is not only a passive endothelial surface anchor for circulating immune cells, but also signals and remodels the plasma membrane and the underlying actin cytoskeleton to promote transendothelial migration (TEM) and extravasation of leukocytes in many inflammatory diseases (Reglero-Real et al, 2012), including those involving cholestasis, in which inhibition of ICAM-1 function reduces liver injury (Gujral et al, 2004). ICAM-1 is apically confined in polarized intestinal and hepatic epithelia (Sumagin et al, 2014; Reglero-Real et al, 2014, 2016), which confers on these cellular barriers the capacity to establish a haptotactic gradient between apical and basolateral membrane domains in order to guide infiltrated immune cells (Sumagin et al, 2014). The apical membrane domains of hepatocytes and cholangiocytes form bile canaliculi and bile ducts, respectively, which drain bile acids and other hepatic molecules into the gastrointestinal tract. The composition of the apical plasma membrane domains in these small channels differs from that of basolateral membranes and from plasma membranes in non-polarized cells, and is determined by an actin cytoskeletal network, which concentrates in these apical domains, and by mechanisms of polarized intracellular sorting of proteins and lipids towards these membrane regions (Meyer et al, 2020; Tsukada et al, 1995; Fu et al, 2010; Cacho-Navas et al, 2022; Musch, 2014). The filamentous actin cytoskeleton in the bile canaliculus forms small membrane protrusions called microvilli, which contain actin but not myosin. However, early electron microscopy analyses and studies of bile canaliculus dynamics in vivo revealed the additional existence of an actomyosin mesh surrounding the basal regions of these microvilli. This cytoskeletal scaffold regulates the diameter and dynamics of the biliary network and is involved in bile trafficking and cholestasis (Meyer et al, 2020; Tsukada et al, 1995; Gupta et al, 2017).

The population of hepatic ICAM-1 molecules that are segregated into bile canaliculi and ducts is not accessible to circulating immune cells, which are in contact with the basolateral membranes of hepatic epithelial cells through the space of Disse. Indeed, the adhesion of leukocytes increases when these epithelial cells lose their polarity and expose this receptor pool. Studies of ICAM-1 dynamics with photoactivatable ICAM-1 chimeras have demonstrated that basolateral ICAM-1 is highly dynamic, but the apical pool of receptors is highly stable and confined within BCs (Reglero-Real et al, 2014). This confinement depends on the interaction of ICAM-1 with the underlying canalicular F-actin (Reglero-Real et al, 2014). However, it is currently unknown whether segregation and immobilization of ICAM-1 into these hepatic apical plasma membrane domains play any role in canalicular dynamics and function.

Here, we have followed experimental strategies that facilitate the analysis of ICAM-1 function in absence of interactions with circulating immune cells. We have generated and characterized CRISPR-CAS9-edited human hepatic epithelial cells lacking ICAM-1 expression, and liver organoids derived in 3D from hepatic bipotent ductal stem cells of wildtype (WT) and ICAM-1_KO mice. We show that ICAM-1 regulates the size and dynamics of bile canalicular-like structures (BCs) in absence of leukocyte adhesion by controlling the actomyosin meshwork that surrounds these apical lumens. Hepatic epithelial cells lacking ICAM-1 expression have their BCs massively enlarged, although these are properly sealed and retain their ability to confine other canalicular proteins. In contrast, upregulation of ICAM-1 expression by inflammatory cytokines greatly reduced BC frequency in a myosin-II-dependent manner. By combining super-resolution microscopy and proximal interactomics analyses we also reveal that microvillar ICAM-1 is organized into membrane nanoclusters with EBP50, a membrane-cytoskeleton protein connector that has been linked to canalicular dysruption during intrahepatic cholestasis (Bryant et al, 2014; Reczek & Bretscher, 2001; Hsu et al, 2010; Li et al, 2015). Indeed, we show that EBP50 controls ICAM-1-mediated regulation of polarity and actomyosin. Our findings indicate that ICAM-1 acts not only as a mediator of leukocyte infiltration in inflammatory and cholestatic diseases, but also as a master regulator of cellular scaffolds that maintain the apical membranes that form the biliary network.

Results

ICAM-1 regulates the size and morphology of apical canalicular-like structures in human hepatic epithelial cells

We had previously shown that ICAM-1 is expressed in sinusoids but also concentrates in the bile canaliculi and bile ducts of polarized hepatic epithelial cells of the human liver parenchyma (Figure 1A) (Reglero-Real et al, 2014). HepG2 cells are spontaneously polarized human hepatic epithelial cells that form actin-rich, bile canalicular-like structures (BCs) and are thus a prototypical in vitro model for studying hepatic apicobasal polarity (Reglero-Real et al, 2014; Lazaro-Dieguez & Musch, 2017; Madrid et al, 2010; van Ijzendoorn et al, 1997; Cacho-Navas et al, 2022). To investigate the role of ICAM-1 on hepatic epithelial cell polarity, we generated HepG2 cells in which the ICAM-1 gene was edited using CRISPR/CAS9 (ICAM-1_KO cells) (Figure 1B and C). ICAM-1_KO cells exhibited larger, less spherical and more elongated BCs than did parental WT cells, suggesting a role for this receptor in regulating BC morphology (Figure 1C and D). We called these structures enlarged BCs (eBCs) because their area was more than 2.5 times that of BCs in WT cells, although the frequency of eBCs per cell was similar to that of BCs (Figure 1E). Similar to BCs, eBCs were enriched in microvilli, sealed by tight junctions (TJs), and contained canalicular markers that reach BCs by a direct route of intracellular transport, such as MDR1, or by an indirect route, such as CD59 (Figure 1D) (17,18). eBCs were also surrounded by a subapical compartment (SAC), as observed by expressing the SAC marker GFP-Rab11 (Supplementary Figure 1A). The enlargement of these canalicular structures was rescued by expressing ICAM-1-GFP in ICAM-1_KO cells, which reduced the size of BCs to those in control HepG2 cells (Figure 1F). ICAM-1 knockdown (KD) also increased the BC area, confirming the effect of ICAM-1 reduction on BC size (Supplementary Figures 1B and C). To analyze ICAM-1-mediated morphological changes further, we performed correlative cryo-soft X-ray tomography (cryo-SXT) in control and ICAM-1_KO HepG2 cells that stably expressed GFP-Rab11 to enable the localization of BCs in live cells. Cryo-SXT makes it possible to image three-dimensional structures of hydrated biological samples close to their native state without chemical pre-treatment (Oton et al, 2012, 2017). Correlative Cryo-SXT revealed that eBCs from ICAM-1_KO cells contained all the morphological features of BCs, although in an enlarged state, including the abundance and length of the microvilli (Figure 1G).

Figure 1.

Potential fusion events between eBCs were also observed by Cryo-SXT (Figure 2A) and by time-lapse fluorescence microscopy (Figure 2B, left panel, central graph, movies 1 and 2) in ICAM-1_KO cells. However, no differences in the relative changes of canalicular areas between frames were observed between KO and WT cells in these time-lapse experiments (Figure 2B, right graph). Indeed, long-term culture of ICAM-1_KO (Figure 2C and Supplementary Figure 1D) and ICAM-1_KD (Supplementary Figure 1E) cells induced the appearance of microvillus-rich structures that were much larger than eBCs observed between 48 and 72 h, covering areas of up to 100 μm². These structures were surrounded by TJs that had a cobblestone morphology, although the TJ polygons did not enclose an underlying nucleus and, hence, did not correspond to single cells establishing cell–cell contacts (Figure 2C and D). Spherical, canalicular-like domains enriched in F-actin (Figure 2D and Supplementary Figure 1F) and similar to the BCs observed in control cells were found in contact with these large canalicular structures, suggesting that the fusion of several BCs might contribute to the enlargement of canalicular domains in ICAM-1-depleted cells. Finally, to test whether these massive eBCs were exposed to the extracellular milieu, WT and ICAM-1_KO cells were incubated in the cold with sulfo-NHS-biotin, which labels lysine residues from proteins that are exposed on the cell surface and in contact with the medium. Cells were fixed and biotinylated proteins stained with TRITC-conjugated neutravidin. This staining revealed that streptavidin did not penetrate the interior of the eBCs, indicating that these large structures are not exposed to the extracellular milieu (Figure 2E). Taken together, our analyses indicate that eBCs fuse more frequently than BCs and become considerably expanded after four days of polarization.

Figure 2.

ICAM-1 upregulation by inflammatory cytokines reduces hepatic apicobasal polarity

ICAM-1 editing or silencing increases the size of apical surfaces maintaining the frequency of BCs, indicating that reducing this receptor promotes apicobasal polarity in hepatic epithelial cells. ICAM-1 is basally expressed in hepatic epithelial cells in vitro and in vivo (Smith & Thomas, 1990; Adams et al, 1989; Park et al, 2010) but its expression is strongly upregulated by cytokines during the inflammatory response (Volpes et al, 1990; Kvale & Brandtzaeg, 1993). We next addressed the effect of exposing polarized WT and ICAM-1_KO cells to a set of cytokines involved in liver inflammation that potentially increase hepatic ICAM-1 expression (Park et al, 2010; Lacour et al, 2005; He et al, 2021). IL-1β, IFN-γ and the combination of TNF-α and IFN-γ induced the expression of ICAM-1 protein by a factor of 4-6, whereas TNF-α had a moderate effect (Figure 3A), as previously reported (Reglero-Real et al, 2014). Quantification of the apical-to-basolateral ratio revealed that the expression increase was greater in the basolateral domains than in the apical ones (Figure 3B and C). These changes of expression and distribution were correlated with a reduction in the number of BCs in cytokine-stimulated cells (Figure 3D). However, BC loss was prevented when ICAM-1_KO cells were stimulated with these cytokines (Figure 3D, left graphs). In addition, BC size was not reduced upon cytokine stimulation (Figure 3D, right graphs), indicating that either the morphological remodeling of the remaining BCs is controlled by additional expression changes and signaling events triggered by these cytokines, or that these inflammatory mediators provoke the collapse of the smallest BCs, so only the enlarged ones are preserved after the treatment, giving rise to a moderate increase in the average size. However, collectively, these findings demonstrate that ICAM-1 expression levels modulate BC size and/or frequency and thus have a role in hepatic epithelial apicobasal polarity.

Figure 3.

ICAM-1 controls canalicular actomyosin, thereby regulating BC size

Upon leukocyte adhesion, ICAM-1 signals to actomyosin in endothelial cells (Reglero-Real et al, 2012; Millan & Ridley, 2005; van Buul & Hordijk, 2004; van Buul et al, 2007). We hypothesized that hepatic ICAM-1, which is abundant and highly confined within BCs (Reglero-Real et al, 2014), could also signal locally to actomyosin and thereby regulate the size and dynamics of these apical membrane domains in absence of any interaction with immune cells. To analyze the distribution of actomyosin, polarized hepatic epithelial cells expressing GFP-tagged myosin light chain (GFP-MLC) were incubated with SirActin, a fluorogenic, cell permeable probe specific for actin, which concentrated into BCs similar to the phalloidin staining (Figure 4A and 1C). GFP-MLC formed a spherical network that surrounded SirActin positive BCs (Figure 4A, movie 3). Single planes of the confocal time lapse projections showed that MLC formed a ring at the base of microvilli-like actin domains, but did not localized in the most central parts of microvillar structures. High-resolution confocal microscopy of endogenous F-actin and myosin light chain phosphorylated in the T18/S19 residues (p-MLC) confirmed that active non-muscle myosin-II complexes were distributed in the most distal parts BC where it overlapped with F-actin, which also formed microvilli-like structures (Figure 4B). In addition, the regions of F-actin that overlapped with ICAM-1 and the actomyosin network clearly differed, and the colocalization of p-MLC and ICAM-1 staining was much less extensive that that of these two proteins with F-actin, as shown by the Manders’ analyses of the confocal images (Figure 4B, central and right panels). Importantly, the canalicular staining of p-MLC (Figure 4C) and of the myosin heavy chain-IIb (MHCIIb) (Figure 4D) were significantly reduced in ICAM-1_KO cells, whereas the F-actin levels were maintained (Figure 4E). In contrast, F-actin and p-MLC distribution in the basolateral membrane domains were higher in ICAM-1_KO cells than in WT cells (Figure 4E and Supplementary Figure 2A). The exogenous expression of ICAM-GFP in ICAM-1_KO cells restored the canalicular actomyosin network, indicating that ICAM-1 regulates the actomyosin cytoskeleton surrounding BCs (Figure 4C-E and Supplementary Figure 2A). Moreover, cells treated with blebbistatin, the ATPase inhibitor for non-muscle myosin-II, induced eBCs in a manner similar to that observed in ICAM-1_KO cells (Figure 4F). Blebbistatin also prevented the ICAM-1-dependent loss of BCs in response to IL-1β (Figure 4G).

Figure 4.

MLC phosphorylation and actomyosin contraction are regulated by the Rho kinases (ROCKs), which are effectors of the RhoA subfamily of GTPases. ROCK inhibition with Y27632 phenocopied the effect of ICAM-1 editing and blebbistatin treatment on BC size (Supplementary Figure 2B). The RhoA subfamily comprises the small GTPases RhoA, RhoB and RhoC (Hodge & Ridley, 2016; Riento & Ridley, 2003), which share 88% amino acid identity (Wheeler & Ridley, 2004) and are activated and translocated to the plasma membrane in response to different stimuli (Marcos-Ramiro et al, 2016, 2014). The expression of GFP-RhoB and GFP-RhoC in polarized hepatic cells suggested that the RhoA subfamily significantly concentrates in the plasma membrane of BCs (Supplementary Figure 2C). GFP-RhoA expression completely depolarized these hepatic epithelial cells, so its apicobasal distribution could not be addressed (not shown), whereas another Rho GTPase, GFP-Rac1, was almost evenly distributed between canalicular and basolateral plasma membrane domains (Supplementary Figure 2C). Together, these results indicate that BC concentrate the machinery regulating the Rho-ROCK-actomyosin signaling pathway.

ICAM-1 signaling to actin and non muscle myosin-II prevents BC morphogenesis

To address whether hepatic ICAM-1 signaling to actomyosin affects hepatic cell polarity, while taking into account that canalicular ICAM-1 cannot be accessed from the canalicular lumen (Figure 2E), we adopted an alternative strategy and investigated the effect of ICAM-1 surface immobilization on BC morphogenesis of epithelial cells that initially were non-polarized. A variety of antibodies and fibronectin were immobilized on coverslips, cells were trypsinized and then seeded on the antibody-coated coverslips to analyze the effect of ICAM-1-mediated signaling. Cells seeded on cultured plates previously precoated with anti-ICAM-1 antibody were unable to form BCs after 24 h of culture and had more stress fibers than did cells seeded on surfaces coated with control immunoglobulin(Ig)-G, anti-transferrin receptor (TfR) or the integrin ligand, fibronectin, all of which formed BCs with a frequency similar to that of cells plated on uncoated surfaces (Figure 5A-C). This result indicated that receptor engagement signaled to F-actin, thereby preventing the acquisition of polarity. Moreover, cells on anti-ICAM-1 antibodies frequently displayed stellate stress fibers that were similar to those arising in response to the expression of the constitutively active mutant of ROCK (Figure 5A-C) (Garg et al, 2008; Nakano et al, 1999). We found that in addition to the fraction of receptor localized on the basal surface after 24 h of adhesion, stellate stress fibers concentrated surface ICAM-1 in their proximities (Figure 5A and Supplementary Figure 3A), whereas p-MLC localized at the F-actin-rich regions where stellate stress fibers arised (Figure 5B and Supplementary Figure 3A). These cellular structures did not constitute aberrant BCs, since they neither had a canalicular morphology nor concentrated other canalicular markers such as MDR1 (Figure 5D). When cells seeded on anti-ICAM-1-coated coverslips were treated with blebbistatin, canalicular morphogenesis was restored to a level comparable with that observed in ICAM-1_KO cells on anti-ICAM-1 antibody-coated coverslips or in cells cultured on control coverslips (Figure 5B, C and Supplementary Figure 3B). This implies that the impairment of BC formation in response to surface ICAM-1 immobilization depends on non-muscle myosin-II. As previously described, the myosin inhibitor blebbistatin affected only the distribution of MLC (Figure 5B and Supplementary Figure 3B), and not the phosphorylation of the residues recognized by the antibody (Ponsaerts et al, 2008; Calaminus et al, 2007). In addition, the overall signaling to p-MLC was comparable in cells plated on anti-ICAM-1 antibodies and on FN (Supplementary Figure 3C), indicating that the impaired morphogenesis was specific to ICAM-1 immobilization, and independent of the general actin remodeling and myosin-II activation occurring in response to the adhesion and spreading of cells previously in suspension. Indeed, ICAM-1 surface engagement by clustering with a secondary antibody, independently of epithelial adhesion and spreading, was sufficient to induce the phosphorylation of MLC (Figure 5E), which demonstrates that the receptor signals to actomyosin in these cells.

Figure 5.

ICAM-1 is the main adhesion receptor for T-cells in the hepatic epithelium (Reglero-Real et al, 2014). We had previously shown that T-cell adhesion can be prevented by blocking antibodies against this receptor and LFA-1 or by ICAM-1 gene silencing (Reglero-Real et al, 2014). To investigate the relevance of our findings in the contest of leukocyte interaction, we analyzed the effect of T-cell adhesion on the apicobasal polarity of hepatic epithelial cells. First, ICAM-1_KO cells had also impaired their ability to interact with T-cells, comfirming the essential role of epithelial ICAM-1 in T-cell adhesion (Supplementary Figure 3D). In addition, co-culture of T-lymphocytes with polarized HepG2 cells for at least 6 h, at an approximate ratio 1:2, reduced the number of BCs found in cells prior to adhesion by approximately 40%. Hence, these results indicate that the signaling events that control epithelial apicobasal polarity by ICAM-1 in absence of interaction with immune cells, also mediate the reduction of epithelial polarity and BC frequency upon leukocyte adhesion (Figure 5F).

The proximal interactome of ICAM-1 regulates BC size and ICAM-1-mediated signaling to actomyosin

To investigate further the mechanisms of ICAM-1-mediated signaling to canalicular actomyosin, we analyzed the proteome in the proximity of ICAM-1 by BioID (Roux et al, 2012). We expressed a chimeric ICAM-1-BirA* protein, performed a biotinylation assay and analyzed the neutravidin pulldown fractions by mass spectrometry. Two independent experiments revealed a restricted set of proteins that were biotinylated in both assays and only in cells expressing the ICAM-1-BirA* chimera (Supplementary Figure 4A-C). Of these, proteins such as SNAP23 and VAMP5 have been shown to co-cluster or interact with ICAM-1 in other cell types (Carman et al, 2007; Li et al, 2015; Schoppmeyer et al, 2022). The ezrin-radixin-moesin phosphoprotein (EBP)-50/NHERF1/SLC9A3R1 was also detected and further validated (Figure 6A). EBP50 is an scaffolding protein that interacts with Rho and Rab GTPase activity regulators (Hsu et al, 2010; Reczek & Bretscher, 2001) and can directly bind to transmembrane proteins or indirectly through the interaction with ERM proteins, which in turn also bind to Ig-superfamily receptors (Barreiro et al, 2002; Reglero-Real et al, 2014). However, no proximal biotinylation was detected for ERM proteins, despite SNAP23, another known ICAM-1 interactor, was also validated (Supplementary Figure 4D). EBP50 has been found to interact with ICAM-1 in whole rat hepatic tissue and regulate the liver inflammatory response after bile duct ligation, although the precise molecular mechanisms underlying such regulation, and the cellular identity and localization where such interaction takes place have not been addressed (Li et al, 2015). We found that EBP50 was highly polarized and concentrated and colocalized with ICAM-1 in BCs (Figure 6B). Confocal analysis of hepatic tissue also revealed showed a sinusoidal (S) and canalicular (red arrows) distribution of EBP50 in vivo (Figure 6C). However, triple co-staining revealed that EBP50 was exclusively localized in the actin-positive canalicular microvilli but not in the peripheral actomyosin mesh (Figure 6D). EBP50 is a modular protein that is maintained in a closed conformation by a head-to-tail intramolecular interaction that masks its association with ligands through its two N-terminal PDZ domains (Morales et al, 2007). Immunoprecipitation experiments showed a weak interaction between ICAM-1 and ectopically expressed full-length EBP50, which was strongly increased when the two N-terminal PDZ domains were expressed alone, without the masking C-terminal EB domain (Figure 6E). The binding of EBP50 fragments containing only one PDZ domain was clearly reduced and suggested a preferential interaction with the second PDZ domain. The expression of these PDZ domains was sufficient to target EBP50 to BCs (Supplementary Figure 4E). Thus, ICAM-1 interacts throgh this PDZ tandem in addition to a potential indirect interaction through ERM proteins. This latter interaction, which occurs through the the EB domain of EBP50, could not be properly analyzed because of the very low levels of ectopic expression obtained for this fragment alone, which suggests the EB domain is unstable and prone to degradation in absence of the rest of the protein (not shown). To better understand the ICAM-1-EBP50 molecular complexes, we analyzed the canalicular distribution of these two proteins by super-resolution stimulated emission depletion (STED) confocal microscopy (Figure 6F). STED revealed that both ICAM-1 and EBP50 distribution was not even, but displayed patterns of overlapping nano-scale domains along F-actin-rich microvilli. These nanodomains had a size between 80 and 200 nm. In contrast, super-resolution images showed that microvillar F-actin did not form nanoscopic clusters in these filaments (Figure 6G). Furthermore, in plasma membrane regions containing few microvilli, such as those of non-polarized cells, ICAM-1 was more evenly distributed and did not overlap with EBP50 (Figure 6H). These EBP50 nanodomains, however, responded to ICAM-1 engagement by aggregating themselves, often forming circular structures of several nanoclusters, which only partially overlapped with the clustered receptor (Figure 6H and I). Furthermore, such redisitribution caused EBP50 to become more insoluble to mild-non inonic detergents (Supplementary Figure 4F), indicating that ICAM-1 induces changes in the condensation properties of EBP50, either by increasing the association of EBP50 to cytoskeletal structures or by changing its oligomerization properties (Fouassier et al, 2000). Importantly, ICAM-1 and EBP50 nanoclusters were in closer proximity in apical BCs than in non-polarized plasma membrane regions, even if non-polarized cells were exposed to anti-ICAM-1 antibody, which increased nanocluster overlapping, as shown by Manders’ analyses (Figure 6I). This indicates that the proximal interaction and the potential ICAM-1-EBP50-mediated signaling mainly occur in BCs in absence of receptor engagement. In addition, it also suggested that, in addition to associating to this scaffold protein, ICAM-1 also signals and reorganizes EBP50 in macromolecular complexes that do not contain the receptor.

Figure 6.

The absence of ICAM-1 in KO cells caused no overall dispersion of EBP50 from BCs, indicating that EBP50 is localized in these apical plasma membrane domains through additional interactions (Figure 6J). Nevertheless, quantification of EBP50 by STED microscopy revealed an increase in the intensity of the nanodomains in ICAM-1_KO cells with respect to WT, but not in their distribution and density, measured by analyzing the nearest neighbor distance (Figure 6K, nnd). These observations suggest that EBP50 nanoclustering does not require ICAM-1 expression, but also that this receptor can modulate EBP50 clustering at BCs. Next, we addressed whether EBP50 regulates canalicular size and dynamics. siRNA-mediated SLC9A3R1 gene silencing reduced EBP50 expression by around 50% and increased BC area by almost a half (Figure 7A and B). Although moderately, this EBP50 depletion decreased pMLC staining at the canalicular actomyosin network. Unlike ICAM-1_KO cells, F-actin levels at the canalicular microvilli of EBP50_KD cells were lower than in control cells, which suggests that EBP50 has functions that are independent of ICAM-1 in these apical domains (Figure 7B). It is particularly notheworthy that EBP50 depletion attenuated the negative effect of ICAM-1 on BC morphogenesis and ICAM-mediated signaling to F-actin (Figure 7C-E). EBP50_KD cells formed more BCs upon ICAM-1 immobilization and reduced the number of stress fibers with respect to control cells (Figure 7E). In addition, EBP50 accumulated with ICAM-1 close to stellate stress fibers, consistent with the interaction between ICAM-1 and EBP50 in membrane domains enriched in underlying F-actin (Figure 7D, top images). These results indicate that EBP50 regulates ICAM-1-mediated signaling towards actomyosin. Such signaling prevents BC morphogenesis and can promote BC contraction and loss of apicobasal polarity. Our findings, therefore, provide a mechanistic rationale to explain the role that the association between ICAM-1 and EBP50 plays in mediating the progression of inflammatory diseases that disrupts the biliary network, such as experimental cholestatic liver injury (Li et al, 2015).

Figure 7.

ICAM-1 regulates the size and morphology of canalicular-like structures in an actomyosin-dependent manner in hepatic organoids

A confocal analysis of ICAM-1 distribution in murine livers revealed that this receptor was highly expressed in hepatic sinusoids and, with less intensity, in bile canaliculi and ducts (Figure 8A) (Colás-Algora et al, 2020). Hence, we performed a morphological study of bile canaliculi in livers from ICAM-1_KO mice and their WT littermates (Bullard et al., 2007), which revealed a significant increase of canalicular width in hepatic tissue lacking ICAM-1 expression (Figure 8B and Supplementary Figure 5A). This suggested that ICAM-1 regulates the biliary network of murine livers. However, the specific cell-autonomous contribution of epithelial ICAM-1 with respect to the sinusoidal expression of the receptor and, above all, with respect to the adhesion of immune cells to hepatic epithelial and endothelial ICAM-1-expressing cells, could not be discriminated by analyzing these tissues. To investigate the potential cell-autonomous role of epithelial hepatic ICAM-1 on regulating canalicular membrane domains and cell polarity, independent of other functions in vessels and of its interactions with leukocytes, we took advantage of recent advances in the generation of hepatic organoids obtained from 3D culture and expansion of ductal bipotent stem cells isolated from human and murine livers (Supplementary Figure 5B) (Broutier et al, 2016; Huch et al, 2015). Murine hepatic organoids expressed hepatocyte markers such as albumin. They also expressed ICAM-1 without previous exposure to inflammatory stimuli (Supplementary Figure 5C) (Broutier et al, 2016; Huch et al, 2013). Hepatic organoids had an heterogenous morphology but mostly formed spheroids of cell agregrates that formed cell-cell junctions (Figure 8C). Interestingly, the bile canalicular marker HA-4 was localized between cells in the peripheral regions of these spheroids. HA-4 exhibited a staining pattern of wide tubular and dotted domains different to that of the cell-cell junction marker ZO-1 (Figure 8C, enlarged areas). Moreover, a time-lapse analysis of carboxyfluorescein diacetate (CFDA), a fluorescent probe of hepatobiliary transport and secretion into the bile canaliculi, revealed that fluorescent CF was also laterally secreted between cells, as shown by the relative distribution of CF with respect to the nuclei, which was similar to that of HA-4 marker (Figures 8C and D). Transmission electron microscopy of these organoids revealed the existence of lateral cavities between cells, which were sealed by electrodense cell–cell junctions and contained long, thin microvilli, different from those observed in the plasma membrane domain facing bigger lumens in the central part of the spheroids (Figure 8E). The cell–cell junctions observed in these lateral lumens resembled the recently described membrane bulkheads that contribute to the elongation of canalicular lumens (Figure 8E) (Belicova et al, 2021). Collectively, these results suggested that differentiated hepatic organoids form BC-like lateral cavities that morphologically and functionally resemble BCs. Interestingly, ICAM-1 accumulated in these domains in mouse and human hepatic organoids (Figure 8F). pMLC and EBP50 were also localized in lateral plasma membrane domains, although EBP50 also concentrated in the plasma membrane facing central lumens (Supplementary Figure 5D). To address whether ICAM-1 regulates the morphology of these BC-like lateral cavities, we generated hepatic organoids from WT and ICAM-1_KO mice. Measuring the area of BC-like lateral lumens revealed larger areas in ICAM-1_KO than in WT differentiated organoids (Figure 8G). Time-lapse analyses of organoids incubated with CFDA and albumin secretion assays indicated that BC-like lateral domains were functional in WT and ICAM-1_KO organoids (Figure 8H). In addition, confocal (Supplementary Figures 5E and F) and electron (Supplementary Figure 5G) microscopy analyses also revealed that ICAM-1_KO organoids contained enlarged cells (Supplementary Figure 5H), suggesting that ICAM-1 may exert a overall control of the submembranal cytoskeletal scaffolds that control cellular size in the hepatic organoids. Finally, incubation with blebbistatin also increased the area of these lateral lumens in a manner almost identical to that observed in ICAM-1_KO organoids (Figure 8I)

Figure 8.

Hence, collectively, our observations from the various polarized hepatic cellular models reveal that ICAM-1 is organized in nano-scale domains in surface microvilli at BCs, from where the receptor signals to actomyosin independent of its role as an adhesion receptor for immune cells, in order to regulate hepatic cell architecture. In a more general perspective and taking into account previous reports (Reglero-Real et al, 2014; Cacho-Navas et al, 2022), these findings also provide direct evidence of a close and reciprocal connection between epithelial apicobasal polarity and the inflammatory response through ICAM-1 (Supplementary Figure 6).

Discussion

We had previously shown that epithelial apicobasal polarity modulates leukocyte adhesion to hepatic epithelial cells by regulating ICAM-1 exposure to the extracellular milieu (Reglero-Real et al, 2014). Our findings here indicate ICAM-1-mediated signaling regulates epithelial cell polarization. This signaling can emanate from receptor engagement upon leukocyte adhesion, in accordance with previous reports in different cellular contexts (Millan & Ridley, 2005; Etienne et al, 1998), but also by confining and enriching ICAM-1 in BCs in a leukocyte-independent manner. The participation of ICAM-1 in immune cell adhesion and trafficking through the vascular system, which are essential immune and inflammatory processes closely involved in liver pathology, makes it difficult to discriminate and quantify in vivo the real contribution of epithelial ICAM-1 to the loss of epithelial architecture during the progression of these pro-inflammatory diseases, independently of its function in leukocyte adhesion. However, our experiments with different cellular models demonstrate that ICAM-1 upregulation upon proinflammatory stimulation is sufficient to depolarize hepatic epithelial cells in an actomyosin-dependent manner. The concentration of this receptor in nano-scale domains in membrane regions enriched in microvilli, such as apical BCs, is sufficient to induce signaling to actomyosin, which regulates the contraction of BCs, leading to BC disruption. Thus, given that many proinflammatory liver diseases, which involve massive leukocyte infiltration into the liver, cause a reduction of epithelial apicobasal polarity in the hepatic parenchyma (Gissen & Arias, 2015), leukocyte-dependent and independent-ICAM-1 signaling may directly contribute to the pathological dysfunction of hepatocytes and cholangiocytes.

Although ICAM-1 localizes to and signals from the basolateral membrane, polarized hepatic cells express basolateral-to-apical transcytotic machinery that removes the receptor from these plasma membrane domains and confines most of ICAM-1 within the BC (Reglero-Real et al, 2014; Cacho-Navas et al, 2022). This machinery is developmentally expressed in other epithelia and is essential for the acquisition of cell polarity and differentiation of gut epithelial cells, highlighting again the molecular connections between epithelial patterning and immune response. By editing and silencing the gene coding for ICAM-1, or by increasing its expression in response to inflammatory cytokines, we have shown that this apical confinement controls the canalicular dynamics by regulating the pericanalicular myosin-II levels. Bile canaliculi are mechanosensory hubs that polarize actomyosin upon changes in canalicular flow and during the regeneration of the hepatic parenchyma (Meyer et al, 2020). In accordance with these findings, we have found that BCs concentrate not only F-actin and actomyosin but also their master regulators, the GTPases of the RhoA subfamily. In vivo, myosin-II exhibits a canalicular distribution, although discriminating between subapical and fully apical distributions by high-resolution light microscopy is more difficult to achieve in vivo under some experimental conditions (Meyer et al, 2020). However, a detailed analysis of liver samples by electron microscopy has identified a subapical myosin-II network surrounding the microvillus-rich canalicular plasma membrane (Tsukada et al, 1995). This myosin-II activity controls the function of the bile canaliculus in vivo (Meyer et al, 2017). Pharmacological inhibition of ROCK is sufficient to increase the bile canalicular diameter in murine livers, which reduces the bile canalicular flow and delays CFDA clearance from this network of channels (Meyer et al, 2017). Conversely, cholestasis-inducing drugs affect hepatic myosin-II activation, suggesting that canalicular actomyosin may be a target when treating diseases related to the bile flow (Sharanek et al, 2016). On the other hand, blebbistatin and Y-27632 treatments are also sufficient to induce lateral, canalicular-like lumen formation in columnar epithelial cells when cell–cell junction formation is prevented. This suggests a central and specific role for the RhoA-ROCK-myosin-II pathway in forming lateral canalicular-like structures (Cohen et al, 2007). The enlargement of hepatic apical membranes has also been observed in the liver parenchyma in various pathological and physiological contexts, and resembles alterations observed in cells lacking proteins regulating apical membrane integrity, such as ABCB4 and Cdc42 (Pradhan-Sundd et al, 2019; Shitara et al, 2019; Sonal et al, 2014).

Myosin-II has been located in the terminal web, the cytoskeletal network at the base of microvilli, in a variety of epithelial cells (Delacour et al, 2016). Myosin-II activity in the terminal web negatively regulates the incorporation of G-actin into microvilli, which affects their length (Chinowsky et al, 2020). The myosin II grid detected in the base of BCs under our experimental conditions display a discrete pattern that colocalizes with the most basal part of microvilli, suggesting that this cytoskeletal structure is also part of the hepatic epithelial terminal web. Therefore, crosstalk appears to exist between myosin-II at the terminal web and microvilli. If ICAM-1 expression and localization in canalicular microvilli determine the pericanalicular myosin-II activation that controls canalicular dynamics, it is plausible that the receptor may regulate the canalicular network in proinflammatory diseases in which hepatic ICAM-1 density is massively increased. However, the effect of inflammatory cytokines on the bile flow has not been adequately addressed so far. We do know that TNF-α and IL-6 treatments clearly reduce bile flow in rat livers and alter canalicular dynamics in vitro (Ikeda et al, 2003). Liver pathologies with an inflammatory component have in common the loss of apicobasal polarity of epithelial cells in the hepatic parenchyma (Shousha et al, 2004). Our results stimulating polarized hepatic cells suggest that cellular depolarization is intrinsic to expression changes occurring in these cells upon exposure to inflammatory cytokines and that ICAM-1 induction contribute, at least in part, to polarity loss without requiring engagement by adhering leukocytes.

ICAM-1 is a type-I transmembrane protein that mediates plasma membrane remodeling upon engagement because it is connected to the underlying actin cytoskeleton by membrane-actin crosslinkers such as ERMs. In endothelial cells, engaged ICAM-1 receptor reorganizes microvilli into docking structures that mediate leukocyte adhesion and promote extravasation (Reglero-Real et al, 2012; Millan et al, 2006; Barreiro et al, 2002). Therefore, in other cellular contexts, ICAM-1 may help form plasma membrane domains in which the receptor is also confined. Photoactivation experiments have demonstrated that ICAM-1 concentrates and is highly stable in BCs, whereas in the basolateral plasma membrane the receptor rapidly diffuses and translocates towards the BC (Reglero-Real et al, 2014). This suggests that molecular mechanisms must exist that selectively stabilize or engage ICAM-1 in these apical membrane domains. Our interactome analyses revealed that ICAM-1 interacts with the scaffold protein EBP50/NHERF1/SLC9A3R1, which is also confined in canalicular microvilli. Moreover, super-resolution microscopy experiments clearly demonstrate that ICAM-1 and EBP50 are organized in overlapping nanoclusters in BC. The nature of these domains remains to be solved, but F-actin does not form clusters and is evenly distributed in microvilli, suggesting that other nano-scale structures should be involved. Liquid ordered membrane condensates or lipid rafts help form membrane microdomains based on the differential affinity of certain membrane lipids for membrane-associated proteins (Levental & Lyman, 2023), thereby regulating clustering and function of transmembrane proteins.

Lipid-ICAM-1 membrane nanodomains may be involved in the control of ICAM-1 nanodomains and their subsequent signaling since these condensates are abundant in BCs (Ismair et al, 2009) and a transient confinement in raft-like domains has been previously reported for ICAM-1 (Millan et al, 2006; Tilghman & Hoover, 2002) and EBP50 (Sultan et al, 2013) in different cellular contexts.

ICAM-1-actin connection through EBP50 or ERM proteins into overlapping nanodomains may therefore have a role in its apical confinement, but additional interactions mediated by ICAM-1 transmembrane and extracellular domains must contribute to its retention and function in the BCs. Ligand-mediated stabilization may be one of the mechanisms contributing to ICAM-1 retention in BCs. It is unlikely that polarized hepatic cells express and accumulate β2 integrin molecules on the canalicular surface that are able to interact in trans with ICAM-1. In addition, hepatic epithelial cells express very low levels of Muc1, which is another ICAM-1 ligand involved in epithelial cancer cell adhesion (Regimbald et al, 1996). However, hepatocytes are the main source of fibrinogen, a large protein involved in coagulation, which binds and clusters the ICAM-1 receptor by interacting with its first Ig-like domain (Tsakadze et al, 2002). As shown by The Human Protein Atlas (Uhlen et al, 2015; Yu et al, 2015), hepatic cells express very high levels of various fibrinogen chains and at least the fibrinogen β-chain is highly concentrated in bile canaliculi in immunohistochemical analyses of human hepatic tissue. This suggests that fibrinogen is not only secreted, but also retained close to hepatic apical membrane domains, where it could potentially interact with, cluster and stabilize canalicular ICAM-1, thereby inducing its intracellular signaling.

In conclusion, our findings indicate that ICAM-1 plays a dual role in hepatic epithelial cells: as an adhesion receptor that interacts with leukocytes and guides immune cell migration, and as a signaling molecule that regulates non-muscle myosin-II through EBP50, independent of its role as a leukocyte adhesion counter receptor (Supplementary Figure 6). ICAM-1 membrane domains can signal to this motor protein in different cellular contexts. From BCs, segregated ICAM-1 controls canalicular actomyosin and BC homeostasis (Supplementary Figure 6, top). However, an excess of ICAM-1 expression in response to inflammatory stimuli can exacerbate this signaling and induce the collapse of canalicular structures. Furthermore, ICAM-1 engagement in non-canalicular membrane domains by immune cells also dramatically reduces epithelial apicobasal polarity (Supplementary Figure 6, bottom). Interestingly, our results bring together molecular machinery that is central to the progression of cholestasis in preclinical models, providing potential mechanistic insights into the canalicular collapse provoked by this pathological condition (Gupta et al, 2017; Li et al, 2015), which is linked to inflammatory responses that induce ICAM-1 expression (Gujral et al, 2004; Li et al, 2017). Further investigation into the leukocyte-independent signaling pathways orchestrated by ICAM-1 may reveal new strategies to preserve the hepatic epithelial architecture in different pathological inflammatory contexts.

Star methods

Key Resources Table

Chemicals

Plasmids

sgRNAs CRISPR/Cas9

PCR oligonucleotides

siRNA oligonucleotides

Lead Contact and Materials Availability

Further information and requests for resources and reagents should be directed to, and will be fulfilled by, the Lead Contact, Jaime Millán (jmillan@cbm.csic.es).

All unique/stable reagents generated in this study will be made available on request, but we may require a Material Transfer Agreement.

Cells and culture

Human polarized hepatic HepG2 cells were grown in high-glucose Dulbecco’s modified Eagle’s medium supplemented with 5% fetal bovine serum. T lymphoblasts were prepared from isolated human peripheral blood mononuclear cells (PBMCs). Nonadherent PBMCs were stimulated with 0.5% phytohemagglutinin for 48 h and maintained in RPMI medium supplemented with 2 U/ml IL-2, as previously described (Millan et al, 2002). Experiments were performed with memory T-lymphocytes cultured for 7-12 days.

Biliary ductal stem cell isolation and organoid culture

Wildtype and ICAM-1_KO mice on a C57BL/6 background were used. ICAM-1_KO mice were originally generated and kindly shared by Prof. Nancy Hogg (London, UK) (Bullard et al., 2007). Mouse organoids derived from hepatic bipotent stem cells were cultured as previously described (Broutier et al., 2016). Briefly, livers were isolated from mice and digested using collagenase/dispase II (0.125 mg/ml in DMEM/F12 medium). Isolated ducts were mixed with Matrigel (BD Bioscience), seeded and cultured in 3D. One week after seeding, organoids were mechanically dissociated into small fragments and transferred to fresh Matrigel. Passage was performed weekly at 1:3-1:6 split ratios. For human organoid culture, frozen organoids were previously obtained from healthy liver resections (∼1 cm³) from liver transplantations performed at the Erasmus Medical Center, Rotterdam, MEC-2014-060. These organoids were cultured and expanded at the CBM Severo Ochoa under a Material Transfer Agreement between the University of Cambridge and the CBM Severo Ochoa (CSIC-UAM). All in vivo experiments were conducted in compliance with UK legislation set out in the Animal Scientific Procedures Act 1986. All procedures were conducted in accordance with UK Home Office regulations.

Liver organoid differentiation

The liver organoid differentiation procedure was performed as described elsewhere (Huch et al., 2015). Briefly, liver organoids were seeded for 3 days in liver expansion medium (Broutier et al., 2016). That medium was then exchanged for differentiation medium (Broutier et al., 2016), which was replenished every 2-3 days for a period of 11-13 days. Culture medium was collected 48 h after the final medium change for albumin secretion assays. Albumin concentration in culture supernatant was determined using the BCG Albumin assay kit (Sigma-Aldrich). For the CFDA secretion assay, organoids were incubated with 0.1 μM of CFDA diluted in phenol red-free DMEM supplemented with 10 mM HEPES, pH 7.4. At 10 min, CFDA fluorescence at the BC-like, lateral lumens of live organoids were detected by time-lapse confocal microscopy under a Nikon AR1 confocal microscope. Time-lapse acquisitions were processed using Fiji image processing software (NIH, Bethesda, Maryland).

Method Details

RNA preparation and PCR analysis

RNA was extracted from organoid cultures using an RNeasy Mini RNA Extraction Kit (Qiagen), and reverse-transcribed using Moloney murine leukemia virus reverse transcriptase (Promega). cDNA was amplified in a thermal cycler (GeneAmp PCR System 9700, Applied Biosystems). The primers used are listed in the corresponding key resource table.

Cell transfection and stable expression of exogenous proteins

5 μg DNA/10⁶ cells or 100 nM siRNA /10⁶ cells were transfected by electroporation (200 mV, 950 μF and 480 Ω; Bio-Rad). Expression was measured 24-48 h post-transfection, and the siRNA effect was analyzed 72 h post-transfection. For stable expression of exogenous proteins, transfected cells were selected by treatment with 0.75 μg/ml G-418 sulfate for at least 4 weeks after transfection. Positive cell clones were selected and maintained in drug-free medium. After several passages in this medium, >80% of cells retained expression of the exogenous protein. For CRISPR/Cas9 gene editing, the cDNA sequence was analyzed using the Breaking-Cas tool (http://bioinfogp.cnb.csic.es/tools/breakingcas), and the selected target sequences were inserted in the pSpCas9(BB)-2A-GFP plasmid, which was a gift from Feng Zhang (Massachusetts Institute of Technology, Cambridge, MA, USA) (Addgene plasmid # 48138; http://n2t.net/addgene:48138; RRID:Addgene_48138). GFP-positive cells were sorted after 24 h of transfection and plated. Individual clones were tested by immunofluorescence and immunoblot analyses.

Laser scanning and spinning disk confocal microscopy

HepG2 cells were grown on coverslips (5x10⁵ cells/well), fixed in 4% paraformaldehyde (PFA) for 15 min, rinsed and treated with 10 mM glycine for 2 min to quench the aldehyde groups. Cells were then permeabilized with 0.2% Triton X-100, rinsed and blocked with 3% bovine serum albumin (BSA) in PBS for 15 min at room temperature (RT). Cells were incubated for 30 min with the primary antibodies, rinsed in PBS and incubated for 30 min with the appropriate fluorescent secondary antibodies. Actin filaments were detected with fluorophore-conjugated phalloidin (Ruiz-Saenz et al, 2011) (see table). Incubation with antibodies and other fluorescence reagents was always carried out at 37°C. Phosphorylated proteins were stained using Tris-buffered saline (20 mM TRIS, pH 7.4; 150 mM NaCl) instead of PBS. Confocal laser scanning microscopy was carried out with a confocal Zeiss LSM710 system coupled to an AxioImager M2 microscope, a confocal Zeiss LSM 800 system coupled to an AxioObserver microscope, and a confocal Nikon AR1+ system coupled to an Eclipse Ti-E microscope. Confocal spinning microscopy was performed with a Confocal Spinning Disk SpinSR10 system from Olympus coupled to an inVerted microscope IX83. Morphological analyses and BC quantification were performed by staining for F-actin, EBP50, ICAM-1, phospho-MLC or ZO-1, depending on the experiment. The basolateral and apical intensity of ICAM-1 in WT and ICAM_KO HepG2 cells was calculated from confocal images of polarized cell colonies by measuring the fluorescence intensity in Fiji of basolateral and apical areas, respectively. Images were processed with Fiji and Imaris software to enable 3D volume reconstruction. When specified, XZ projections, produced by summing the confocal images containing the structure of interest, were shown.

Stimulated emission depletion (STED) super-resolution microscopy

STED super-resolution images were acquired using a confocal microscope Leica TCS SP8 (Leica Microsystems) with an immersion objective HCX PL APO CS 100x NA 1.4 (Leica). For the green channel, samples excitation was achieved using a WLL2 laser settled at 488 nm with 10% of power, fluorescence was collected through a hybrid detector in the 498-542 nm, and STED laser beam intensity (592 nm) was used at 30% of its power. For the red channel, samples excitation was achieved using a WLL2 laser settled at 553 nm with 10% of power, fluorescence was collected through a hybrid detector in the 563-620 nm, and STED laser beam intensity (660 nm) was used at 60% of its power. Images were acquired at 1024 x 1024 pixels, with a 6x zoom factor (to obtain a pixel size corresponding to 18.36 nm), frame average 1 or 2, in the case of the red channel, a frame accumulation of 4, and a scanning speed of 400 Hz. Under these conditions, no significant photo bleaching was detected.

STED images were analyzed using a custom software written in MATLAB (The MathWorks, Inc., Natick, Massachusetts, United States) that detects fluorescent spots and provides their intensity and centroid position through an automatic detection algorithm based on fitting to a Gaussian point-spread function (PSF) profile (Martínez-Muñoz et al, 2018). In our experiments, the PSF profile and its full-width-at-half-maximum (FWHM = 60 nm) describing STED resolution were estimated using fluorescent spots present in the cell cytoplasm, considered as a non-aggregate state of the receptor. This PSF was applied for the identification and analysis of the spot clusters within the biliary channels. The nnd was calculated as the minimum Euclidean distance between each spot and its neighbors. Plot of the mean fluorescence intensities values and statistical analysis, calculated through the unpaired t-test with Welch’s correction, were obtained using the GraphPad Prism 9 software.

T-lymphocyte adhesion assays

To measure the ability of T cells to induce their depolarization, HepG2 cells were plated onto 24-well plates (5x10⁴ cells/well) for 48 h. The cells were co-cultured with T lymphocytes at a 2:1 ratio for the indicated times. After washing, cells were fixed, immunofluorescence was performed and the percentage of CD3-stained T lymphocytes or CD3-stained T cells adhering to HepG2 cells were measured under a fluorescence microscope.

Surface protein labeling

HepG2 cells were incubated for 30 min with 250 μg/ml of sulfo-NHS-biotin at 4°C. Cells were then washed and traces of unbound biotin were blocked by incubation for 10 min with DMEM containing 10% FBS. Cells were then incubated with TRITC-conjugated streptavidin at 4°C for 30 min to label surface-biotinylated proteins. Cells were then fixed and immunofluorescence was performed.

Organoid whole-mount immunostaining

Organoid staining was performed as previously described (Broutier et al., 2016). Briefly, for whole-mount immunofluorescent staining, organoids were washed twice in cold PBS before fixing in 4% PFA for 30 min at 4°C. Then, organoids were rinsed in PBS and incubated with gentle agitation in blocking buffer containing 1% DMSO, 2% BSA and 0.3% Triton X-100. Organoids were rinsed again in PBS and labeled with the primary antibodies at 4°C overnight, rinsed a further time in PBS, and incubated with the appropriate fluorescent secondary antibodies for 2 h. Finally, nuclei were stained with DAPI.

Tissue immunofluorescence and immunohistochemistry

Murine livers were removed and fixed overnight in 10% neutral-buffered formalin (Sigma-Aldrich) at RT. After fixation, tissues were incubated in 30% sucrose overnight and then frozen in Tissue-Tek O.C.T. compound. The sections were allowed to cool at RT and then incubated in blocking buffer (1% DMSO, 2% BSA and 0.3% Triton X-100 in PBS) for 2 h at RT. Primary antibodies were diluted in blocking buffer and incubated overnight at 4°C. Secondary antibodies were diluted in PBS containing 0.05% BSA and incubated for 2 h at RT. To facilitate analyses of hepatic ICAM-1 distribution mice were sometimes previously exposed to intraperitoneal LPS for 17 h. The analyses of bile canalicular width was performed in absence of any inflammatory challenge to reduce the potential interactions of the hepatic receptor with immune cells. More than 80 bile canaliculi from four WT and four ICAM-1_KO mice were quantified.

The immunohistochemical analysis of human hepatic tissue was approved by the Hospital Ethics Committee of the Hospital Universitario de Salamanca. Biopsies from liver allograft rejections and control livers form healthy donors were analyzed. Formalin-fixed, paraffin-embedded sections of 4-µm thickness were deparaffinized in xylene and rehydrated through a decreasing graded ethanol solution series. After suppression of endogenous peroxidase activity (3% hydrogen peroxide, 10 min) and antigen retrieval (boiling in 10 mM citrate buffer, pH 6.0), sections were immunostained with the appropriate primary antibody, using the DakoCytomation Envision Plus peroxidase mouse system (Dako). The stained protein was visualized using DAB solution (Dako), and lightly counterstained with Mayer’s hematoxylin. To ascertain the specificity of the antibody immunoreactivity, a negative control was carried out in the absence of the primary antibody; this did not produce any detectable immunolabeling.

Correlative cryo-soft X -ray tomography

HepG2 cells stably expressing GFP-Rab11 were grown on gold finder Quantifoil R2/2 holey carbon grids. To improve the adhesion of cells, the carbon-coated grids were glow-discharged and coated with FBS before cell seeding. The grids were then vitrified by immersion in liquid ethane on a Leica EM CPC and cryopreserved in liquid nitrogen. Samples selected by cryo-epifluorescence microscopy were transferred to a Mistral Transmission soft X-ray Microscope (TXM, Alba Synchrotron Light Source). Samples were again mapped using an epifluorescence microscope online with the Mistral TXM. The epifluorescence microscope was equipped with an LED light source (Mightex), a Mitutoyo 20x 0.42 NA objective (Edmund Optics) and a Retiga CCD detector (QImaging). Coordinates determined in this way were used to select the X-ray acquisition areas. An X-ray energy of 520 eV was used to illuminate the samples. The photon flux on the sample was on the order of 4.6 x 10¹⁰ photons/s. Data sets were acquired using a Ni zone plate objective lens (ZP) with an outermost zone width of 40 nm (Zeiss) following a tomographic acquisition scheme. The tilt angle of sampling ranged from -65° to 65° at 1° intervals. Exposure time varied with sample thickness from 1 to 5 s per image. The final pixel size was 11.8 nm. Tilt series were normalized relative to the flat-field images, beam current and exposure time using software developed in house. Data were also corrected by the apparent point spread function (APSF) of the microscope (Oton et al, 2017) and deconvolved using a Wiener filter with a k=1/SNR value of 0.1 based on the signal-to-noise ratio (Oton et al, 2016). The pre-processed tilt series were aligned to a single rotation axis using IMOD software (Kremer et al, 1996) and reconstructed using TOMO3D software with 30 iterations of the SIRT algorithm (Agulleiro & Fernandez, 2011). 3D rendering volumes were represented with Chimera and ImageJ (Schneider et al, 2012).

Transmission electron microscopy

Matrigel-embedded liver organoids were in situ fixed with 4% paraformaldehyde and 2% glutaraldehyde in 0.1 M phosphate buffer, pH 7.4 for 2 h at RT. Postfixation was carried out with a mixture of 1% osmium tetroxide and 0.08% potassium ferricyanide for 1 h at 4°C and then with 2% uranyl acetate for 1 h at RT. Samples were dehydrated with ethanol and processed for standard Epon (TAAB-812) embedding. After polymerization, orthogonal ultrathin sections (80 nm) were collected on formvar-coated slot grids and stained with uranyl acetate and lead citrate. Finally, sections were examined in a Jeol JEM-1400Flash transmission electron microscope operating at 100 kV. Images were recorded with a Gatan OneView (4K x 4K) CMOS camera.

Biotinylation of ICAM-1-proximal proteins: BioID assay

To generate an expression plasmid containing the ICAM-1-BirA* construct, the sequence coding for GFP in the ICAM-1-GFP was substituted by the BirA* sequence (Roux et al, 2012) obtained from the Cav1-BirA* plasmid, kindly provided by Prof. I. Correas (Centro de Biología Molecular Severo Ochoa), with BsrGI and AgeI enzymes (New England Biolabs). The expression vector coding for ICAM-1-BirA* was transfected into HepG2 cells by electroporation and cell clones stably expressing ICAM-1-BirA* were selected with G-418 as previously described (Reglero-Real et al, 2014). ICAM-1-BirA* HepG2 cells were cultured on 10-cm-diameter plates and, after 48 h, incubated with 50 μM biotin for 16 h, lysed and subjected to a pull-down assay of biotinylated proteins with neutravidin-agarose (Thermo Scientific) as previously described (Roux et al, 2012; Marcos-Ramiro et al, 2016). In parallel, non-transfected HepG2 cells were used as a control of the precipitation of biotinylated proteins. Lysates and pull-down pellets were analyzed by western blot and by LC-ESI QTOF tandem mass spectrometry. The identified proteins were listed and the proteins of interest were validated by western blot.

Immunoprecipitation assays

HepG2 were transfected with expression vectors coding for FLAG-tagged full length EBP50 and the indicated FLAG-tagged N-terminal fragments. 48 h post-transfection, cells were lysed in 400 μl of TNE buffer (50 mM Tris pH 7.4, 150 mM NaCl, 5 mM EDTA) containing 1% Triton-X100 and a protease and phosphatase inhibitor cocktail. Lysates were incubated with 20 μl protein G-coated Sepharose (Sigma-Aldrich) previously conjugated with 5 μl of anti-FLAG antibody by overnight incubation. Lysates were precleared for 1 h with anti-IG mouse protein-G-sepharose and then incubated with anti-FLAG conjugated protein-G-sepharose for 3 h. Antibody-conjugated beads were rinsed in TNE + TX100 buffer five times and dried by aspiration. Immunoprecipitated proteins were eluted in 20 μl of Laemmlís buffer and analyzed by western blot.

Protein extraction and western blot

Cell lysates and organoids cultures were prepared using Laemmli buffer supplemented with a cocktail of protease inhibitors. The lysates were heated at 95°C for 5 min and cleared by centrifugation at 14x10³ rpm for 5 min. The samples were loaded on acrylamide gels and transferred onto an Immobilon-PVDF membrane (Millipore), which was blocked in PBS containing 5% nonfat dry milk and incubated overnight with the indicated primary antibody. Anti-rabbit or anti-mouse horseradish-peroxidase-(HRP)-conjugated secondary antibodies were then used and the antibody–protein complexes were visualized using ECL (GE-Healthcare). Band intensities were quantified using Fiji software.

To analize the solubility of EBP50 and other proteins in non-ionic detergents upon ICAM-1 engagement, cells were incubated with 2 μg/ml anti-ICAM-1 mAb at 40°C for 30 min, washed and incubated at 37°C for the indicated times. Next, cells were lysed at 4°C for 20 min in 400 μl of TNE buffer (50 mM Tris pH 7.4, 150 mM NaCl, 5 mM EDTA) containing 1% Triton-X100 and a protease and phosphatase inhibitor cocktail, passed five times through a 22G needle, incubated for 5 min at 37°C and then centrifuged at 14x10³ rpm for 5 min at 4°C to separate the supernatant (soluble fraction) from the pellet. Pellets were washed with TNE + 1% Triton lysing buffer before adding laemmli buffer to and sonicate the samples.

Quantification and statistical analysis

Data are summarized as the mean and standard deviation (SD) or the mean and standard error of the mean (SEM). Student’s two-tailed unpaired samples t-tests or two-way ANOVAs were used to establish the statistical significance (p < 0.05) of group differences, depending on the experiment. In all cases, data from at least three independent experiments were used. All calculations were performed using Prism 7 software.

Acknowledgements

We gratefully acknowledge the expert technical advice of the Confocal Microscopy, Electron Microscopy and Genomic facilities of the CBM Severo Ochoa, especially the contribution of Milagros Guerra, from the electron microscopy facility. We thank the staff of the Advanced Light Microscopy and cryo-EM facilities of the CNB-CSIC for their expert technical assistance with the correlative Cryo-SXT. We also thank Dr. Eva Pereiro at ALBA Synchrotron Light Source (Cerdanyola del Vallès, Spain) for her expert technical advice, Prof. Nancy Hogg at the Francis Crick Institute (London, UK) for generating and sharing the ICAM-1_KO mouse, Dr. Lucía Cordero Espinoza at the Gurdon Institute (Cambridge, UK) for her technical support in generating liver organoids, and Ana López Sancha for her technical support with isolating the PBMCs. The work was supported by grants PID2020-119881RB-I00 from AEI (to J.M.) and P2022/BMD-7232 TomoXliver2 (to J.M., J.M.C and A.C.) and IND2019/BMD-17139 (to J.M.) from Comunidad de Madrid. S.B.F is supported by Endocornea, Convenio Colaboración CSIC, funded by Instituto de Investigación Fundación Jiménez Díaz. C.M. acknowledges support through the grant PID2021-125386NB-I00 funded by MCIN/AEI/10.13039/501100011033/ and FEDER “ERDF A way of making Europe”. C.C-N. is a recipient of FPI fellowships from MINECO. N.C-A. is a recipient of an FPU fellowship from MECD. N.R-R is additionally supported by funding from the People Programme (Marie Curie Actions) of the European Union’s Seventh Framework Programme (FP7/2007–2013) under REA grant agreement n° 608765.

Conflict of interest

The authors declare that they have no conflicts of interest.