Multiscale dynamics of tight junction remodeling

Epithelia aid in organizing the body by creating tissue barriers – sheets of connected cells with the ability to regulate the quantity and type of materials that cross the tissue. In addition to acting as barriers, epithelial tissues are involved in specialized functions, such as absorption, secretion, transportation and signaling. Almost every organ in our body is covered with or lined by an epithelial tissue that is specifically suited for the task of the organ.

Molecules can cross tissues by going either through cells (transcellular transport) or between cells (paracellular transport), and different tissues have different requirements for the strength and selectivity of these barriers. Tight junctions (TJs) are responsible for sealing the paracellular space and discriminating among small solutes, such as ions, water and small uncharged molecules, while generally restricting the passage of larger molecules and microorganisms (Marchiando et al., 2010). These barriers are crucial for establishing concentration gradients across tissues, thereby creating specialized environments (e.g. a low pH in the stomach), and preventing infection.

TJs regulate the flux of molecules through the paracellular spaces with a family of transmembrane proteins called claudins, which are capable of forming size- and charge-selective pores, as well as restricting flux. Vertebrate genomes encode a large number of claudins (more than 20 in humans) with diverse pore- and barrier-forming properties (Günzel and Yu, 2013 ; see also Box 1). TJs are connected to the actomyosin cytoskeleton through scaffolding proteins, including ZO family proteins (Box 1), and these cytoskeletal connections are crucial for the structure and function of TJs. The organization and tension of perijunctional actomyosin is regulated by the small GTPase RhoA; thus, RhoA activity must be tightly regulated in space and time for correct TJ formation, maintenance and remodeling (Arnold et al., 2017).

TJ selectivity is regulated through claudin pores – channels formed by the extracellular domains of claudins – that conduct ions and small molecules on the basis of size and charge. Paracellular permeability through claudin pores is commonly referred to as the ‘pore pathway’. Selectivity is based on the composition of pore-forming claudins (both cation and anion pores) and barrier-forming claudins, which restrict the flux of small solutes (Günzel, 2017). Barrier properties of TJs can change in response to various stimuli by ‘claudin switching’, e.g. by replacing a pore-forming claudin with a barrier-forming claudin (Günzel, 2017). Owing to the large number of highly selective pores, the pore pathway is characterized by its high capacity and high selectivity (Anderson and Van Itallie, 2009; Shen et al., 2011). Transepithelial electrical resistance (TER) measures the magnitude of ion flux across epithelia (increased TER indicates decreased ion permeability), whereas more-refined techniques can measure size and charge selectivity of junctions (Box 2; see also Shen et al., 2011).

A second pathway, known as the ‘leak pathway’, allows larger molecules to cross TJs, albeit with less selectivity and a much lower capacity than the pore pathway (Shen et al., 2011). The predominate hypothesis is that leak pathway flux occurs due to the breaking and annealing of claudin strands, which may be influenced by intermolecular associations between claudins, occludin, ZO-1 and the actin cytoskeleton (Fanning et al., 2012; Turner et al., 2014; Van Itallie et al., 2015). Another leak pathway regulator, JAM-A, is sufficient to limit the flux of macromolecules in the absence of claudin strands, suggesting that JAMs serve as a stop-gap measure to protect barrier integrity when claudin strands are disrupted or newly forming (Otani et al., 2019). The presence of redundant mechanisms to regulate the leak pathway underlines the importance of restricting flux of macromolecules via the leak pathway.

To effectively limit flux, it is necessary for TJs to be continuous around the perimeter of epithelial cells. However, multiple physiological processes in epithelial tissues involve cell-shape changes that require cell–cell boundaries to expand and contract, thus posing challenges to the continuity of TJs. For example, actomyosin-driven changes of cell shape and cell rearrangements induce the stretching, bending and folding of tissues during developmental morphogenesis (Harris and Tepass, 2010). Other mechanical challenges arise from organ-specific function, including bladder expansion and gut peristalsis. However, it is poorly understood how such dynamic tissues can also act as stable barriers. Because the correct regulation of barrier function is essential for optimal organ function (Marchiando et al., 2010), epithelial cells must possess robust mechanisms for maintaining barrier function while remaining plastic enough to adapt to a changing environment. This Review focuses on how TJ remodeling on several scales contributes to sustained barrier function in the face of cell and tissue dynamics.

TJ dynamics encompass multiple scales – from the molecular composition and dynamics of individual proteins and protein complexes to tissue-scale events, such as mechanical stress – which affect TJ remodeling (Fig. 1). Molecular- and tissue-scale dynamics of TJs have been studied in some detail, although our knowledge of these processes is still incomplete. An even greater gap in knowledge lies in the intermediate scales – at the strand network and cellular scales (Fig. 1). In this section, we will discuss TJ dynamics across scales, the technical limitations that prevent us from fully understanding these remarkable structures, and the interplay between these levels of TJ organization and remodeling.

The basic unit of the TJ is the claudin-based strand (Fig. 1, strands). Multiple claudin-based strands join together to form elaborate strand networks that vary considerably in complexity and organization from tissue to tissue (Claude and Goodenough, 1973). Because claudins possess diverse pore-forming and barrier-forming properties (Günzel and Yu, 2013; see also Box 1), the composition of claudins within the strand network has a profound effect on ion flux across the tissue (Fig. 1, molecules and strands; Box 2). Less well-understood is how the morphology of the strand network itself impacts barrier function.

Owing to the tight packing of TJ strands and the relatively poor z-resolution of light microscopy, the morphology of the TJ strand network has traditionally been viewed by freeze fracture electron microscopy (FFEM). One FFEM study comparing the epithelial strand networks that exhibit different barrier properties found that an increased strand number is generally associated with decreased ion flux (Claude and Goodenough, 1973). However, there are also examples where TJ permeability is independent of strand number (Colegio et al., 2003), suggesting the claudin composition – not just the number of strands – determines permeability.

Some treatments that affect barrier properties correspond to rapid changes in strand network configuration. For example, exposure of Necturus (salamander) gallbladder to plant cytokinins causes a rapid but reversible increase in TER (Bentzel et al., 1976; see also Box 2). This treatment also results in the reorganization of TJ strand networks from being mostly parallel relative to the apical membrane to more-disorganized, vertically oriented networks with a higher number of free strand ends (Bentzel et al., 1980). Removal of cytokinins corresponds to a drop in TER and a reorientation of strands to a parallel orientation with very few vertical crosslinks (Bentzel et al., 1980). Similarly, treatment with the actin-depolymerizing drug cytochalasin D, which decreases TER in guinea pig ileum, results in the transformation of the strand network from a regularly crosslinked meshwork to one that is sparse and disorganized (Madara et al., 1986). Thus, in addition to claudin composition, strand density, orientation and crosslinking may impact barrier function (Fig. 1).

Mechanical changes in junctional or tissue tension can also alter the strand network morphology by reorienting the strand network or inducing strand breaks (Fig. 2; see also Hull and Staehelin, 1976; Pitelka and Taggart, 1983). Interestingly, radial tension generated by contraction of the apical actomyosin array resulted in networks with more vertically oriented strands, whereas tension generated by tissue stretch generated more parallel TJ strands (Pitelka and Taggart, 1983). This indicates that tissue mechanics may influence strand network morphology, in turn impacting barrier function.

Although FFEM is considered the ‘gold standard’ tool for visualizing TJ strands because it provides high-resolution information about the morphology of the strand network, this technique presents limitations, including limited molecular specificity and dependence on tissue fixation, meaning that strand dynamics cannot be observed over time (Bartle et al., 2018). Improvements to super-resolution imaging techniques might make high-resolution live imaging of the dynamics of epithelial cell TJ strands and proteins that associate with them possible in the future.

In the meantime, expression of claudins in fibroblasts, which do not normally make TJs, has been an effective approach to reconstitute dynamic strand networks similar to those seen by FFEM in epithelial cells (Furuse et al., 1998). Because the plane of cell–cell contact in fibroblasts is parallel to the imaging plane and the strands are less densely packed than in native TJs, it is possible to resolve individual strands by using light microscopy or super-resolution microscopy (Sasaki et al., 2003; Kaufmann et al., 2012; Van Itallie et al., 2017). However, it should be noted that reconstituted claudin strands lack the full complement of TJ proteins, meaning that strand dynamics in native TJs could vary substantially. These reconstitution experiments revealed that claudin strands can be remarkably dynamic within the membrane, capable of breaking and annealing, as well as forming transient end-to-side and side-to-side interactions (Sasaki et al., 2003; Van Itallie et al., 2017). These observations led to the hypothesis that the breaking and annealing of TJ strands is the mechanistic basis that underlies the leak pathway (Sasaki et al., 2003; Van Itallie et al., 2017; Zihni et al., 2016), the non-selective flux of macromolecules across tissues (Box 2).

To better understand how interactions between claudins and other TJ proteins influence strand dynamics, the TJ proteins ZO-1 and occludin were also introduced into fibroblasts expressing claudin-2. The reconstituted strands that were linked to actin through ZO-1 aligned with actin filaments (Van Itallie et al., 2017). Although the overall size of strand patches was stabilized when they were linked to actin through ZO-1, the dynamic breaking and joining of strands was not changed, indicating that this interaction alone is not sufficient to stabilize strand breaks (Van Itallie et al., 2017). Interestingly, occludin tends to localize to free strand ends and branch points, which are more susceptible to breaking and appear to be sites of claudin incorporation (Van Itallie et al., 2017; see also Fig. 2). Because reconstituted strands are likely to behave differently than native strands, future studies that can observe TJ strands and single-molecule dynamics in vivo would be a major leap forward in our understanding of how proteins, such as ZO-1 and occludin, modify TJ dynamics, as well as how strand dynamics contribute to barrier function.

In summary, claudin-based strands form the selectively permeable seal that generates the barrier function of the TJ. Strand organization and function is regulated by multiple factors including mechanical forces that change tissue tension, connection of the transmembrane TJ proteins to the actin cytoskeleton and molecular composition of the TJ. We next discuss the dynamics of TJ proteins at the molecular scale.

Fluorescence recovery after photobleaching (FRAP) is a powerful tool for understanding TJ remodeling at the molecular level. FRAP of claudins in individual reconstituted strands in fibroblasts revealed that, once incorporated into strands, claudins are locked in place – there is no fluorescence recovery (Sasaki et al., 2003; Van Itallie et al., 2017). Rather, newly synthesized claudin molecules are likely to be added at strand ends (Van Itallie et al., 2017).

In epithelial TJs, FRAP of claudins reveals that they, too, are remarkably stable, with typical mobile fractions between 20 and 35% (Shen et al., 2008; Yu et al., 2010; Raleigh et al., 2011; Capaldo et al., 2014; Higashi et al., 2016). Several studies suggest that claudin stability is associated with claudin function and incorporation into claudin strands. For example, claudin-2 is a pore-forming claudin that is permeable to cations. When occludin serine residue 408 is not phosphorylated, it forms a complex with ZO-1 and claudin-2, and claudin-2 mobility increases. Under these circumstances, TER increases (i.e. ion permeability decreases), indicating loss of claudin-2 pore functionality (Raleigh et al., 2011). A similar pattern is observed when Caco-2 cells are treated with the pro-inflammatory cytokines interferon gamma (IFN-γ) and tumor necrosis factor alpha (TNF-α) (Capaldo et al., 2014). Cytokine treatment decreases TER, which is associated with increased mobility of claudin-4, a barrier-forming claudin and decreased mobility of claudin-2 (Capaldo et al., 2014). Additionally, claudins that are capable of forming reconstituted strands in the epithelial-like cell line SF7 have lower mobile fractions than other claudins that are unable to do so (Yamazaki et al., 2011). The simplest interpretation of these results is that incorporation of claudins into strands reduces their diffusion and allows them to act as either pores or diffusion barriers, according to the type of claudin.

FRAP and complementary techniques, such as fluorescence loss in photobleaching (FLIP) and photoactivation, have also provided information about the molecular dynamics of TJ proteins over short periods of time (seconds to minutes), demonstrating that FRAP recovery of occludin and claudins occurs through lateral diffusion within the membrane; although diffusion of claudins is more spatially restricted than that of occludin, and the mobile fraction of occludin is higher, between 50 and 80% (Shen et al., 2008; Yu et al., 2010; Raleigh et al., 2011; Van Itallie et al., 2015; Ronaghan et al., 2016; Janosevic et al., 2016). Over longer periods of time (hours), claudins and occludin also behave differently. A recent study using pulse labeling of SNAP- and CLIP-tagged TJ proteins has shown that occludin and claudins move within the TJ in unique ways (Van Itallie et al., 2019). Newly synthesized claudins are trafficked to the basolateral membrane and are incorporated into free strand ends on the basal side of the TJ. As strands mature, claudins move apically and old claudins are removed by endocytosis. By contrast, occludin trafficking is not polarized, and occludin can exchange between TJ and basolateral pools. This further implies that claudins are ‘locked in’ to strands, whereas occludin is more transiently associated with strands or other TJ structures (Van Itallie et al., 2019).

ZO-1 exchange occurs not through diffusion along the membrane but through exchange with a cytoplasmic pool (Shen et al., 2008; Yu et al., 2010). According to FRAP studies, ∼10–50% of ZO-1 is immobile (Shen et al., 2008; Yu et al., 2010; Raleigh et al., 2011; Van Itallie et al., 2015; Higashi et al., 2016). Stabilization at the junction requires the actin-binding region of ZO-1 and might also reflect stable association with structures, such as claudin strands. Recent evidence suggests that ZO family proteins form phase-separated compartments (Beutel et al., 2019, Schwayer et al., 2019), and many TJ proteins, including scaffolding proteins, signaling proteins and the cytoplasmic tails of TJ membrane proteins, are concentrated within these compartments – in some cases as much as 40-fold higher than the cytoplasmic concentration (Beutel et al., 2019). The dense concentration of TJ proteins within these phase-separated compartments is likely to allow for emergent TJ properties, such as the polymerization of claudins into strands, which requires ZO-1 or ZO-2 and their ability to multimerize, as well as the ability of ZO-1 to bind actin (Beutel et al., 2019; Otani et al., 2019; Schwayer et al., 2019; Umeda et al., 2006). Understanding the molecular dynamics and organization of TJ proteins will give us a stronger foundation to explain how TJ strands and molecules respond to cell- and tissue-scale dynamics.

Adherens junctions (AJs) are typically viewed as the main load-bearing actin-associated cell–cell junctions, and their role in force sensing during cell shape change and development has been extensively studied (Charras and Yap, 2018). However, it is less clear how TJs respond to cell shape changes, particularly when existing cell–cell interfaces are elongated or new cell–cell interfaces are established, such as during cell division and cell extrusion (Fig. 1, cells).

During cell division, a relatively small portion of the cell–cell junction experiences direct pulling forces from the contractile ring. In Xenopus laevis embryos, FRAP of AJ proteins showed that they are stabilized at the cleavage furrow; however, FRAP does not detectably change for TJ proteins, indicating that AJs are coupled to the forces that are generated by the actomyosin contractile ring (Higashi et al., 2016). Interestingly, ZO-1 intensity at the furrow does increase relative to that at the poles, indicating that some TJ remodeling occurs during cytokinesis (Higashi et al., 2016). Indeed, FFEM showed that TJs in mitotic cells are more disorganized, comprising a higher number of free strand ends than non-mitotic cells (Tice et al., 1979).

Cell extrusion presents another potential challenge to epithelial barrier integrity. During extrusion, which occurs due to cell death or overcrowding of healthy cells in the epithelial monolayer, cells are squeezed out of epithelial tissues by their neighbors (Gudipaty and Rosenblatt, 2017). During this process, TJs extend basally to maintain contact between neighboring cells and the extruded cell (Madara, 1990; Guan et al., 2011; Marchiando et al., 2011). Following cell extrusion, new cell–cell interfaces must be formed.

During events involving cell-scale dynamics, such as cell division and cell extrusion, major shape changes occur not only in cell interfaces that are directly involved in the process but also in neighboring cells. These cells must then change their shape to accommodate the movements of the dividing or extruding cells. During both cytokinesis and extrusion, the epithelial barrier is largely maintained (Higashi et al., 2016; Rosenblatt et al., 2001; Madara, 1990; Guan et al., 2011; Marchiando et al., 2011). Therefore, gaining a better understanding of the mechanisms that allow TJs to expand and contract, while simultaneously restricting ion and macromolecule flux, is an important goal for this field of research.

Tissue-scale dynamics include TJ remodeling over extended periods of time and greater distances during development, morphogenesis and in diseased states. For example, maturation of TJs as tissues differentiate during development involves the expression of different sets of tissue-specific TJ proteins and signaling proteins (Günzel and Yu, 2013). Furthermore, tissue-scale mechanical stretching during developmental morphogenesis or organ-specific functions (e.g. bladder filling and voiding) requires significant TJ remodeling. Additionally, disease states, such as pathological inflammation, can influence the morphology, protein composition and barrier function of TJs at the tissue scale by modulating TJ protein expression and trafficking, as well as actomyosin structure (please see the excellent reviews by Luissint et al., 2016; Garcia-Hernandez et al., 2017; Marchiando et al., 2010).

During developmental morphogenesis, dynamic remodeling of TJs is not only required to maintain barrier function as epithelial cells undergo shape changes and rearrangements, but TJs also actively help to drive development. Several studies have shown that TJ proteins are crucial for apical-basal polarity, blastocyst and lumen formation, and convergent extension (Moriwaki et al., 2007; Navis and Bagnat, 2015; Baumholtz et al., 2017; Otani et al., 2019). For example, loss of ZO-1 or occludin results in multi-lumen formation in 3D epithelial culture (Odenwald et al., 2018; Otani et al., 2019). Additionally, disruption of claudins impairs expansion of the blastocoel, a fluid-filled lumen, due to a leaky barrier and loss of hydrostatic pressure (Moriwaki et al., 2007). Furthermore, TJ-dependent fluid accumulation generates the hydrostatic pressure essential for lumen formation and expansion during organogenesis, as clearly demonstrated in the zebrafish gut (Navis and Bagnat, 2015). Together, these studies suggest that TJs play a crucial role during development that extends beyond their role in regulating paracellular permeability.

Having defined a multiscale framework within which we can consider TJ remodeling, we next highlight a recent study from our lab that unifies TJ and barrier dynamics across scales.

We recently demonstrated that, when cell–cell junctions elongate in the Xenopus laevis embryonic epithelium, local discontinuities in ZO-1 and occludin, but not claudin-6, often occur along the expanding junction (Stephenson et al., 2019). By using a highly sensitive barrier assay, the zinc-based ultrasensitive microscopic barrier assay (ZnUMBA), we identified that the local breaks correspond to sites of leak pathway flux (Stephenson et al., 2019). Importantly, these leaks were quickly repaired by Rho flares, local activations of RhoA that promote local actin accumulation and myosin II-mediated contraction, which concentrates TJ proteins to reinforce the barrier at sites of local leaks (Fig. 3).

Breaking and annealing of TJ strands is thought to be the means that allows macromolecules to cross the TJ (Zihni et al., 2016). Owing to its indiscriminate nature and increased prevalence in diseased tissues, leak pathway flux could be assumed to be an ‘accidental’ feature of TJs rather than a purposeful one. We propose that the ability of TJ strands to easily break gives cells the flexibility to change their shape without being constrained by a rigid paracellular barrier. Actomyosin contraction then drives reinforcement of the TJ barrier by locally increasing protein concentration. Because reconstituted strand ends readily re-anneal with one another (Van Itallie et al., 2017), locally increasing protein concentration may facilitate this process (Fig. 1, strands and cells). Additionally, there might be yet-unknown functions of transient increases in permeability; perhaps increased flux contributes to junction remodeling or these strand breaks are necessary for transepithelial migration.

Not every instance of junction elongation causes a leak or is followed by contraction. Therefore, Rho flares might be an emergency repair mechanism that kicks in when a TJ is more severely damaged. However, this raises the question: what is the routine maintenance mechanism that allows TJs to elongate gradually while maintaining the barrier? In tissues that have been experimentally stretched, TJ strands align with the axis of tension; they appear highly taut and are reduced in number (Pitelka and Taggart, 1983; Hull and Staehelin, 1976). These features indicate that a reorientation of the strand network in response to stretch could help to cover an expanding area without de novo synthesis of TJ strands (Fig. 2). Breaks in strand networks and free strand ends are also a common feature of remodeling TJ networks (Van Itallie et al., 2019). In reconstituted TJ strands, free ends can rapidly reanneal to existing strands and are also the sites of new claudin incorporation (Van Itallie et al., 2017). Thus, strand reannealing and claudin addition may help TJ strands to cover more linear distance, ensuring that a stretched cell maintains barrier function.

In the following section, we speculate on how our knowledge about global perturbations of TJ proteins can be applied to understand TJ remodeling on local, short-term scales.

The function and remodeling of TJs have been studied by using genetic manipulations of specific TJ proteins, as well as by modulating cytoskeletal tension using small-molecule inhibitors or mechanical stretch. In this section, we focus on experiments that test the functions of ZO proteins, occludin and cytoskeletal tension in TJ establishment and barrier function.

Double knockout of ZO-1 and ZO-2 in MDCK II cells results in failure to form TJ strands and, thereby, increased flux of ions through the pore pathway and of macromolecules through the leak pathway (Otani et al., 2019). Previous knockdown studies that investigated partial loss of ZO-1 or ZO-2, or of both ZO proteins resulted in a number of less-severe phenotypes, also pointing to the importance of ZO proteins in the regulation of TJ structure and epithelial barrier function (Van Itallie et al., 2009; Fanning et al., 2012; Rodgers et al., 2013; Odenwald et al., 2018; Umeda et al., 2006). Possible reasons for variability among these studies include redundancy between ZO-1 and ZO-2, the use of different epithelial cell lines comprising different baseline levels of TJ proteins, and variations regarding the level of ZO protein knockdown. Nevertheless, common themes that emerged from these studies include that complete or partial loss of ZO proteins correlates with the robust accumulation of F-actin and myosin II, as well as an increased flux of macromolecules through the leak pathway (Van Itallie et al., 2009; Fanning et al., 2012; Rodgers et al., 2013; Odenwald et al., 2018; Umeda et al., 2006; see also Fig. 4A). Notably, this aligns with our finding that the transient TJ breaks prior to Rho flares are, indeed, associated with local reduction of ZO-1, as well as with increased flux via the leak pathway, which is soon followed by local F-actin accumulation (Stephenson et al., 2019; see also Fig. 3).

Like ZO-1 depletion, genetic perturbation of occludin is also associated with increased flux via the leak pathway (Turner et al., 2014; Buschmann et al., 2013). Interestingly, global knockdown or knockout of occludin has no observable effect on strand network morphology when using FFEM (Yu et al., 2005); thus, how exactly occludin perturbation increases flux via the leak pathway is unknown. However, the local loss of ZO-1 and occludin we observed prior to Rho flares (Stephenson et al., 2019) suggests that Rho flares have a direct role in regulating the barrier to macromolecules. For instance, it is possible that ZO-1 and occludin change the dynamics of TJ strands, thereby, modulating barrier function. However, such a change in TJ strand dynamics would not be visible by FFEM.

The function and remodeling of TJs can be altered over shorter periods of time by modulating actomyosin cytoskeletal tension. Acute treatments that inhibit actin polymerization lead to increased paracellular permeability due to mislocalization of TJ proteins, including ZO-1 and occludin (Van Itallie et al., 2009, 2015; Fanning et al., 2012; Shen and Turner, 2005). Further, inhibition of myosin II activity mediated by Rho kinase (ROCK) or myosin light chain kinase (MLCK) alters the contractility of the apical actomyosin bundle, thus, affecting the permeability of TJs (Nusrat et al., 1995; Clayburgh et al., 2004; Walsh et al., 2001). For example, inhibition of ROCK results in increased flux of macromolecules and ions (Walsh et al., 2001). By contrast, inhibition of MLCK stabilizes ZO-1 at TJs by coupling it to the apical actomyosin array, thereby, restricting the flux of ions (Yu et al., 2010). Taken together, these results suggest that, although ROCK and MLCK both regulate myosin II activity through phosphorylation of myosin II regulatory light chain, these kinases have different effects on TJ barrier function. Interestingly, the local reductions in ZO-1 and occludin we detected prior to Rho flares are followed by accumulation of F-actin and myosin II (Fig. 3), and ROCK-mediated myosin II activity is required for efficient reinstatement of ZO-1 at the sites of barrier leaks (Stephenson et al., 2019).

Mechanical forces change tissue tension and promote associated changes in the actin cytoskeleton and TJ remodeling at a global scale. For instance, when alveolar epithelial cells are stretched biaxially in a cyclic manner, they exhibit an increase in paracellular permeability and in perijunctional actin levels (Cavanaugh et al., 2001; DiPaolo and Margulies, 2012). Additionally, upon biaxial stretch, the overall levels of ZO-1 and occludin at TJs are reduced, and regions of alternating high and low TJ protein intensities are visible along the junction (Cavanaugh et al., 2001). A similar phenotype, i.e. loss of ZO-1 and occludin at TJs, was observed in biaxially stretched Caco-2 cells (Samak et al., 2014; see also Fig. 4B). These studies suggest that stretch-induced cytoskeletal rearrangements perturb the uniform, continuous localization of TJ proteins along cell–cell junctions, thus disrupting barrier function. However, it remains unclear whether (i) the observed discontinuities in ZO-1 and occludin, indeed, correspond to local barrier breaches and (ii) the discontinuities in ZO-1 and occludin are cause or consequence of stretch-induced cytoskeletal rearrangement. To this end, Samak et al. have shown that inhibition of MLCK prior to experimental tissue stretch attenuated the loss of ZO-1 and occludin, suggesting that MLCK-mediated cytoskeletal rearrangement promotes discontinuities of TJs (Samak et al., 2014). Together, these results highlight the importance of feedback regulation between the actomyosin cytoskeleton and TJ proteins, as TJ proteins respond to actomyosin-mediated mechanical forces, and partial or complete loss of ZO proteins leads to robust accumulation of F-actin and myosin II. However, further studies are required to determine the mechanistic nature of this feedback.

So far, we have discussed multiple temporal and spatial scales at which TJs are dynamic, and our current understanding of some of the mechanisms that drive TJ remodeling at global and local scales. We conclude by highlighting two emerging areas in TJ research, mechanotransduction and super-resolution imaging, and outline how future work may enhance our understanding of mechanisms that drive TJ remodeling.

Mechanotransduction is a process by which cells sense mechanical force and respond by transmitting biochemical signals through cellular signaling pathways. Mechanotransduction at AJs is evident in response to both tissue-scale and locally applied forces (Charras and Yap, 2018; Pinheiro and Bellaïche, 2018). Tensile forces, for example, can change the conformation of mechanosensitive AJ proteins, resulting in the recruitment of additional proteins to reinforce and stabilize junctions (Yonemura et al., 2010; Le Duc et al., 2010; Liu et al., 2010). However, much less is known about mechanotransduction at TJs (Sluysmans et al., 2017).

A recent study identified a mechanosensitive pathway that maintains epithelial integrity and barrier function through activation of Rho (Acharya et al., 2018). In that study, the mechanical stretching of cultured epithelial cells activates Rho through the recruitment of a specific Rho guanine nucleotide exchange factor (RhoGEF), p114 RhoGEF. Active Rho then activates the formin mDia1 (also known as Diaph1) to promote actin assembly, which, in turn, strengthens the junctions (Fig. 4B; see also Acharya et al., 2018). Of note, p114 RhoGEF also promotes Rho activation during TJ assembly (Terry et al., 2011; Xu et al., 2013), suggesting that this mechanosensitive pathway may also be at play at TJs.

Seminal EM studies carried out in epithelial chick hair cells identified two distinct, yet interconnected, actin populations at junctions: (i) bundles of actin filaments that run parallel to the membrane at the AJ and, (ii) a meshwork of filaments that lies just beneath the TJ (Hirokawa, 1982; Hirokawa et al., 1983). Indeed, these EM data underlie most of the currently held hypotheses about connections of actin to TJs and AJs, leading us to assume that transmission of tension primarily takes place at the AJ. However, more evidence is needed to determine whether AJs are, indeed, the primary load-bearing actin-associated cell–cell junction or whether TJs can also be load-bearing structures.

Several pieces of data suggest that reassessment of our current understanding of actin connections to the TJ and AJ is warranted. First, we recently quantified the apico-basal localization of actin with respect to the TJ and AJ in the Xenopus embryonic epithelium, and found that TJ-associated F-actin is more abundant than AJ-associated F-actin in these cells (Higashi et al., 2019). Second, tricellular junctions, the points where three cells come together, may utilize specialized actin connections (see the recent review by Higashi and Miller, 2017). Recent studies have suggested that F-actin makes end-on (versus side-on) connections at tricellular junctions (Yonemura, 2011; Choi et al., 2016). We propose that these end-on actin connections are important for tension transmission and barrier function at tricellular junctions, and recruit different protein partners than side-on actin connections to offset the increased force applied on tricellular junctions (Higashi and Miller, 2017). Third, by using platinum replica EM, a recent examination of connections between actin and junctions in vascular endothelial cells revealed that AJs in these cells connect to Arp2/3-containing branched actin networks but to not bundled actomyosin (Efimova and Svitkina, 2018), raising the question of whether connections to branched actin networks are also important in polarized epithelial cells. Fourth, together, these studies suggest that actin connections to apical cell–cell junctions can differ from the commonly held model of a perijunctional actomyosin ring that runs parallel to the AJ. Future studies should directly investigate whether TJs are load-bearing structures. To this end, the use of tension biosensors – similar to those developed to measure mechanical stresses at focal adhesions and AJs at the molecular level (Grashoff et al., 2010; Hoffman and Yap, 2015) – but, instead, designed to measure tensile force applied on specific TJ proteins, would be immensely useful.

By applying our knowledge of mechanotransduction at AJs to that of TJs, it seems possible that TJ proteins also undergo tension-dependent conformational changes analogous to the conformational changes exhibited by AJ proteins, such as α-catenin (Yonemura et al., 2010), and/or that specific proteins are recruited to the TJ under increased tension. Indeed, ZO-1 has recently been shown to undergo a tension-dependent conformational change, opening to a stretched conformation in response to actomyosin-generated force (Spadaro et al., 2017). Here, unfolding of ZO-1 occurs at a level of force similar to that needed to unfold α-catenin in AJs (>5 pN for ZO-1 compared with 5–15 pN for α-catenin) (Spadaro et al., 2017; Yao et al., 2014). This conformational change allows ZO-1 to recruit occludin and a transcription factor to TJs (Spadaro et al., 2017). Supporting the conclusion that ZO-1 is a mechanosensitive protein, FRAP experiments have shown that ZO-1 is stabilized in Xenopus embryos under increased tension (Higashi et al., 2016), suggesting that TJs can, indeed, sense increased tension and respond by stabilizing ZO-1.

What are the potential functional effects of mechanotransduction at TJs? Several pieces of evidence suggest that mechanosensitive pathways are important for maintenance of barrier function under tensile stress. Proteins that are known to be recruited to cell–cell junctions in response to tension, such as afadin and vinculin, are necessary to maintain tissue integrity and barrier function under high tension (Choi et al., 2016; Le Duc et al., 2010). Furthermore, a recently identified mechanosensitive pathway helps to maintain epithelial integrity and barrier function by activating Rho at junctions (Acharya et al., 2018). In the future, it will be interesting to further explore mechanotransduction pathways in epithelial tissues that exhibit cyclic stretch and release, such as filling and emptying of the bladder or the lung (Carattino et al., 2013), or gut peristalsis (Trepat et al., 2007), as in these tissues, the ability to sense and respond to force is likely to be essential for barrier maintenance. Additionally, mechanotransduction at TJs was recently shown to play a role in epithelial tissue spreading during zebrafish gastrulation (Schwayer et al., 2019). In that study, accumulation of TJ components, not AJ components, scaled with the tension of the adjacent actomyosin network, indicating that that these TJs are mechanosensitive. Transportation of phase-separated non-junctional ZO-1 clusters towards the TJ by actomyosin-driven flow was required for the mechanosensitivity of these TJs. This study opens the door for future studies of functional roles for TJs in mediating mechanotransduction during developmental or homeostatic processes. Finally, because mechanosensitive junction proteins can also be activated to undergo conformational changes and recruit other proteins in response to locally applied forces, such as the cytokinetic contractile ring (Higashi et al., 2016), it will be interesting to explore whether local TJ remodeling events involve mechanotransduction and whether they are important for barrier function.

Traditionally, EM techniques, including transmission electron microscopy (TEM) and FFEM, have been used to obtain an ultrastructural view of TJs. Indeed, these approaches have defined our understanding of the structural organization of TJs (e.g. Farquhar and Palade, 1963; Chalcroft and Bullivant, 1970). However, EM cannot capture dynamic changes in TJ strands or the associated cytoskeleton, which underlie the dynamic regulation of barrier function. Light-microscopy techniques, by contrast, are able to reveal the dynamic nature of TJs but, due to the limit the diffraction of light imposes on spatial resolution, these approaches are limited to a resolution of ∼250 nm (under optimal conditions) (Bartle et al., 2018; Schermelleh et al., 2019). Because the size of TJs and other epithelial cell–cell junctions are near this resolution limit (Bartle et al., 2018), the information one can obtain about TJ composition, structure and dynamics is restricted when using light microscopy.

Super-resolution microscopy can bridge the gap between light microscopy and EM. By using different strategies to measure just a subset of fluorophores in the sample, super-resolution microscopy techniques – including stimulated emission depletion (STED) and structural illumination microscopy (SIM), as well as single-molecule localization microscopy techniques, such as fluorescence photoactivated localization microscopy (F-PALM) or stochastic optical reconstruction microscopy (STORM) – overcome the resolution limit of light microscopy, leading to an improved resolution in the range of 20–100 nm (Bartle et al., 2018; Schermelleh et al., 2019; Lambert and Waters, 2017). These techniques combine molecular specificity (i.e. labeling proteins of interest) with high-resolution spatial information. Additionally, they provide an advantage over EM for imaging the actin cytoskeleton associated with TJs, because cells do not have to be treated with harsh fixatives, which often disrupt the cytoskeleton. Recently, several studies have applied super-resolution microscopy techniques to study TJ proteins, and some of these studies reveal novel information about TJ remodeling (Table S1).

The emergence of more widely accessible super-resolution microscopy techniques opens new avenues in TJ research and is expected to reveal new insights into the structure and dynamic remodeling of TJs. There are several aspects of barrier biology, in particular, that could benefit from super-resolution imaging: (i) determining the nanoscale organization of claudins and occludin in strand networks in reconstituted systems, epithelial cells, organoids, and/or developing model organisms; (ii) investigating the mechanisms of delivery and incorporation of TJ proteins into sealing strands, as well as TJ protein turnover; (iii) investigating remodeling of TJs during cytokinesis to understand how barrier function is maintained during cell division and; (iv) determining how the actin cytoskeleton is connected to TJs and tricellular junctions, and how these connections are remodeled under different tension conditions.

In this Review, we have highlighted the dynamic remodeling of TJs to maintain an effective barrier in response to cell- and tissue-scale forces. We argue that, in order to understand TJ remodeling, we must investigate TJ remodeling at multiple scales. Although the functional roles of many TJ proteins in maintaining an intact barrier have been characterized on an individual basis, it is largely unknown how TJ strands and strand networks are dynamically remodeled in response to physiological or pathological stimuli. To address these questions, we suggest several areas of interest for further investigation. First, understanding the kinetics of claudin strand assembly, will help to answer the question of whether claudins self-assemble or whether other proteins facilitate strand formation. Second, additional structural information about claudins and occludin, including their binding interfaces, will shed light on how strands branch, break and anneal, and might help reveal the exact function of occludin. Finally, observing the dynamics of TJ strand remodeling in epithelial cells will determine how strand networks are dynamically remodeled in vivo.

We thank members of the Miller lab for useful discussion and feedback, Tomohito Higashi for providing critical feedback on a draft of this Review, and the three anonymous referees for their insightful comments and suggestions.