Cell–cell junctions in focus – imaging junctional architectures and dynamics at high resolution

Cell–cell junctions are crucial for direct intercellular communication, collective cell rearrangements and barrier function – key processes underpinning tissue development, homeostasis and disease. Early electron microscopy (EM) studies revealed the distinctive spatial organization of junction complexes and the membrane morphology along the apical-basal axis of epithelial cells (Farquhar and Palade, 1963; Tsukita et al., 2001; Yonemura et al., 1995). Meanwhile, fluorescence microscopy of junction-associated proteins has provided detailed insights into diverse junctional molecular landscapes (Herrenknecht et al., 1991; Knudsen et al., 1995; Knudsen and Wheelock, 1992; Oda and Takeichi, 2011; Vasioukhin et al., 2000). Notably, live imaging has enhanced our knowledge of the dynamics and regulatory mechanisms of junction assembly and trafficking (Cadwell et al., 2016; Grimsley-Myers et al., 2020; Malinova and Huveneers, 2018; Peglion et al., 2014). It has also delivered important insights into the molecular events underlying junctional mechanotransduction (Angulo-Urarte et al., 2020; Charras and Yap, 2018; Ehrlich et al., 2002; Khalil and de Rooij, 2019; Vaezi et al., 2002). However, a major obstacle to a deeper understanding of junctional dynamics is that the molecular organization of the core junctional complexes and associated cytoskeletal components is in fact below the resolution of conventional fluorescence microscopy, which is limited by the optical diffraction barrier.

Moreover, there is a strong association between membrane morphology and junction organization. In addition, the presence of membrane curvature-sensing proteins at junctions highlights the role of nanoscale membrane properties in junctional dynamics (Dorland et al., 2016; Rolland et al., 2014; Saarikangas et al., 2011). Transmission EM (TEM) visualizes the membrane and achieves nanometer resolution but suffers from low molecular specificity and is limited by 2-dimensionality. A coherent overview of the junctional landscape requires visualization with nanoscale resolution of the architecture of cell–cell adhesions and the membranes in which adhesions are embedded.

Advancements in EM and super-resolution (SR) imaging approaches have resulted in exciting new opportunities to unravel the junctional landscape in such detail. SR (Betzig et al., 2006; Gustafsson, 2000; Hell and Wichmann, 1994; Neil et al., 1997; Rust et al., 2006) and expansion microscopy (Chen et al., 2015) approaches overcome the diffraction barrier, achieving nanoscale resolution (20–100 nm) of fluorescently labeled proteins. Volume electron microscopy (vEM) techniques provide three-dimensional (3D) visualization of entire cells and tissues (Collinson et al., 2023; Peddie et al., 2022). When combined with correlative light and electron microscopy (CLEM), these techniques can incorporate molecular localization information (Hoffman et al., 2020).

This Review aims to showcase the impact of SR and vEM in elucidating concepts of junctional organization in epithelial and endothelial cells (Fig. 1A), and across the intercalated discs of cardiomyocytes (Box 1). We also outline advances in junction-related fluorescence resonance energy transfer (FRET) tension sensors (Table 1), as well as Rho GTPase biosensors and optogenetic tools (Table 2). These techniques expand the microscopy toolkit by enabling direct sensing and subcellular modulation of junctional dynamics.

First, we will overview classical models of the organization of different cell–cell junctions. Junction complexes consist of transmembrane adhesion receptors that link to the cytoskeleton via intracellular adaptor proteins. In textbook examples of confluent epithelial cells, junctional complexes segregate along the apical-basal axis (Fig. 1A,B). Tight junctions (TJs) are located closest to the apical surface and form the zona occludens (ZO), which selectively regulates paracellular passage of solutes and ions (Zihni et al., 2016). TJs consist of transmembrane proteins, including claudins, the TJ-associated MARVEL domain-containing proteins (TAMPs) [occludin (Furuse et al., 1993), tricellulin (also known as MarvelD2) (Ikenouchi et al., 2005) and MarvelD3 (Steed et al., 2009)], the immunoglobulin-like junctional adhesion molecule (JAM) family proteins (Garrido-Urbani et al., 2014; Martin-Padura et al., 1998), coxsackievirus and adenovirus receptor (CXADR, also known as CAR) (Cohen et al., 2001) and endothelial selective adhesion molecule (ESAM) (Hirata et al., 2001), as reviewed by (Martin-Padura et al., 1998; Otani and Furuse, 2020; Piontek et al., 2020). Intracellularly, TJs connect to the actin and microtubule (MT) cytoskeletons through the zonula occludens proteins ZO-1–ZO-3 (also known as TJP1–TJP3) (Itoh et al., 1999) and cingulin adaptor proteins (Fig. 1B) (Niessen, 2007; Zihni et al., 2016).

At the lateral membrane, the zonula adherens (ZA) is connected to circumferential actomyosin bundles via adherens junctions (AJs), which include cadherin–catenin and nectin–afadin complexes as adhesion subunits (Fig. 1A,B). Nectins are a family of IgG-like single transmembrane adhesion proteins that form homophilic and heterophilic adhesions and intracellularly bind to the actin-binding protein afadin (Mandai et al., 1997; Reymond et al., 2001; Satoh-Horikawa et al., 2000; Takahashi et al., 1999). During formation of cell–cell contacts, nectin–afadin-based adhesions assemble first, followed by cadherin-based adhesions (reviewed in Takai et al., 2008; Takai and Nakanishi, 2003). Classical cadherins (e.g. E-cadherin, VE-cadherin and N-cadherin) form extracellular homotypic adhesions and intracellularly link to the cytoskeleton via the armadillo proteins β-catenin and α-catenin (Fig. 1B) (Gumbiner, 2005).

Desmosomes are localized basally from the ZA and connect to the intermediate filament (IF) cytoskeleton (Farquhar and Palade, 1963; Takeichi, 2014) (Fig. 1A,B). They are formed by the desmosome cadherins – desmogleins (Dsg1–Dsg4) and desmocollins (Dsc1–Dsc3) (Garrod et al., 2002; Getsios et al., 2004; Kljuic et al., 2003; Whittock and Bower, 2003), which in turn bind to the armadillo proteins plakophilin (Pkp1–Pkp3) and plakoglobin (Pg, also known as JUP) (Green and Simpson, 2007) (Fig. 1B). Desmoplakin (DP, also known as Dsp) links the desmosome core to IFs (Stappenbeck et al., 1993). Dsg, Dsc and Pkp proteins exhibit tissue-specific expression (Johnson et al., 2014) and their association with the IF cytoskeleton generates strong adhesions in cardiac and epithelial tissues, which are generally exposed to significant mechanical stress (Najor, 2018).

Junctions directly sense and respond to mechanical forces through mechanosensitive proteins that are recruited and undergo conformational changes upon tension, such as ZO-1 at TJs (Haas et al., 2020; Spadaro et al., 2017), synaptopodin, α-actinin (Kannan and Tang, 2015), vinculin and α-catenin at AJs (Angulo-Urarte et al., 2020; Buckley et al., 2014; Huveneers et al., 2012; le Duc et al., 2010; Yao et al., 2014; Yonemura et al., 2010) and DP at desmosomes (Price et al., 2018).

Several advanced microscopy approaches have proven particularly useful for imaging junctional architecture and dynamics. SR microscopy achieves nanoscale resolution by overcoming the optical diffraction limit of light. Structured illumination microscopy (SIM) (Gustafsson, 2000) and stimulated emission depletion (STED) (Hell and Wichmann, 1994) are based on spatial control of fluorescence emission and detection, whereas single-molecule localization microscopy (SMLM) controls temporal excitation of single fluorophores (Betzig et al., 2006; Rust et al., 2006). SIM uses a grid to generate a line-patterned illumination of the sample; multiple images using different illumination orientations and phases are then mathematically reconstructed into an image with ∼100–130 nm lateral and ∼100–250 nm axial resolution (Gustafsson, 2000). This method is compatible with standardly available fluorophores and fluorescent proteins and can be used for live imaging (Li et al., 2015). STED achieves 20–50 nm lateral and 100–300 nm axial resolution by using an outer ring-shaped laser that depletes the region surrounding a small area of labeled fluorophores excited by an inner beam (Hell and Wichmann, 1994). STED requires little post-processing (Hell and Wichmann, 1994) and is suitable for multi-color labeling (Pellett et al., 2011) and live-imaging (Hein et al., 2008; Westphal et al., 2008). SMLM methods, including stochastic optical reconstruction microscopy (STORM) and photoactivated localization microscopy (PALM), use sequential rounds of excitation to localize individual fluorophores, thereby achieving 20–50 nm lateral and 40–100 nm axial resolution (Betzig et al., 2006; Rust et al., 2006).

Expansion microscopy (ExM) achieves nanoscale resolution with conventional labeling strategies and microscopy. Samples are embedded into a swellable polymer gel that expands isotropically, thereby separating labeled proteins and revealing their relative distribution (Chen et al., 2015; Chozinski et al., 2016; Humpfer et al., 2024; Tillberg and Chen, 2019; Yao et al., 2021). The application of ExM in junctional biology is promising. For instance, a study combining ExM with STED demonstrated that CAMSAP3, a minus-end capping protein, tethers MTs to TJs through paracingulin. Disrupting this interaction affects cytoplasmic MT orientation (Flinois et al., 2024). Additionally, ExM has been used to investigate dynamin localization at AJs throughout an entire C. elegans specimen (Yu et al., 2020).

Various types of EM are used to assess 3D junctional ultrastructure in cells and tissues. One of the earliest EM techniques, freeze-fracturing EM (FFEM), involves rapid freezing and fracturing through the lipid bilayer of a sample to expose the membrane interior. Application of FFEM enabled the visualization of the meshwork-like structures formed by TJs (Staehelin, 1973; Staehelin et al., 1969). Current developments focus on enhancing the applicability and accessibility of vEM imaging, which involves physically sectioning and imaging a sample at continuous depths, followed by digital reconstruction into 3D EM datasets (Collinson et al., 2023; Peddie et al., 2022). These techniques encompass single-section electron tomography (ssET), serial-block face scanning electron microscopy (SBEM) and focused ion beam-scanning electron microscopy (FIB-SEM).

ssET obtains axial (z) information by obtaining multiple images of 200–300-nm-thick sections tilted at different angles (±60–70°), which are then reconstructed into high-resolution volumes (2×2×2 nm) (Gan and Jensen, 2012; Peddie et al., 2022). Although it has notoriously low throughput, this technique has been improved by photo-micropatterning, in which UV light creates customizable grids for extracellular matrix substrates. This allows for the controlled growth of cells in specific orientations, which makes finding the optimal imaging locations within samples much easier (Engel et al., 2019; Toro-Nahuelpan et al., 2020). For example, ssET with lattice micropatterning was used to identify F-actin-rich membrane protrusions at cell–cell contacts in endothelial cells (Engel et al., 2021). SBEM uses an automated microtome to slice samples into sections of equal thickness, achieving 10 nm lateral and ∼50 nm axial resolution (Denk and Horstmann, 2004), and is used for 3D reconstructions of membranes, organelles and junctional complexes within cells and tissues (Motta et al., 2019). FIB-SEM uses a focused ion beam to slice a sample with even higher precision, achieving <5 nm isotropic resolution (Knott et al., 2008; Xu et al., 2017); making it more suitable for ultrastructural 3D reconstruction of nanoscale membrane curvatures and organelles. CLEM approaches that combine SR with vEM can add molecular specificity to 3D junctional ultrastructure. For example, a combination of FIB-SEM with SIM of TJ proteins in neurons has demonstrated that JAM-C (also known as JAM3)-positive membrane regions contain smooth membranes, whereas the surrounding plasma membrane areas are highly curved. These observations suggest that TJs either require a flat membrane or induce membrane flattening (Hoffman et al., 2020).

Finally, platinum replica electron microscopy (PREM) is a technique that uses transmission EM to resolve cytoskeletal topography at single-filament resolution in 3D (Svitkina, 2022, 2017). PREM exposes internal cell structures by removing the plasma membrane through detergent extraction or by mechanically rupturing the exterior of the cell to expose the cytoplasmic face of the plasma membrane (Svitkina, 2022). PREM offers fast sample preparation and compatibility with immunolabeling but is limited to thin samples. Of note, the removal of the membrane might introduce uncertainty in distinguishing cell boundaries (Efimova and Svitkina, 2018). For a selection of representative SR and vEM images of junctions, see Fig. 2.

Many studies have investigated the assembly, integration and segregation of junctional complexes along the apical-basal axis in epithelial cells. However, the adhesion complexes in established junctions appear as diffraction-limited spots in light microscopy, making it a challenge to visualize their nanoscale organization (Gonschior et al., 2020). Notably, imaging the desmosome plaque presents optical challenges due to its size (0.5–1.0 μM) and molecular complexity. By EM, the plaque is visible as two electron-dense regions – the outer dense plaque, found adjacent to the plasma membrane, and the inner dense plaque, located further intracellularly (Al-Amoudi et al., 2011). Recently, SR and vEM have helped introduce new concepts regarding the nanoscale organization and maturation of tight, adherens and desmosome junctions in confluent epithelial monolayers.

Early FFEM microscopy studies indicated that TJs organize into individual strands interwoven into meshworks (Staehelin et al., 1969). These strands are formed by the claudin protein family, which, depending on the claudin type, can generate tight barriers or selective ion and water channels (Piontek et al., 2020). SR studies greatly contributed to integrating claudin organizations with their function. For instance, SR established that claudin-3- and claudin-5-based strands are organized into differently sized meshworks (Kaufmann et al., 2012). Also, live SIM of claudin-2 demonstrated that assembly of such meshworks occurs through the remodeling of individual strands in a ZO-1- and actin-dependent manner (Van Itallie et al., 2017). Given the existence of diversely organized claudin meshworks, it is essential to investigate the various claudins simultaneously to fully understand their role in TJ function. STED imaging of 26 different fluorescently labeled mammalian claudins in COS-7 cells was used to characterize the nanoscale organization of all claudin protein types into meshworks with different densities (Fig. 2A) (Gonschior et al., 2022). Furthermore, multi-color STED revealed that channel-forming and barrier-forming claudins have distinct organizations. Interestingly, differences in the observed organizations of channel-forming claudins confer specificity of ion passage. The functional segregation of these different structures is dictated by the extracellular domain of claudins and is independent of membrane cholesterol or anchoring to actin through ZO-1 (Gonschior et al., 2022). Thus, the incorporation and segregation of claudins into finely regulated meshworks with specific sizes, elicits the tunability and selective properties of TJs.

SR has also contributed to understanding how TJs sense and respond to mechanical stimuli. SIM imaging has reached a resolution that has enabled assessment of the distance between the N- and C-termini of ZO-1 (Spadaro et al., 2017). Using SIM, the researchers showed that ZO-1 unfolds under 2–4 pN of applied force (Spadaro et al., 2017), establishing ZO-1 as a mechanosensitive protein, which was later confirmed with a ZO-1 FRET sensor (Haas et al., 2020). STED microscopy was used to investigate the role of other TJ transmembrane proteins, showing that JAM-A (also known as F11R) and claudin-1, -2, -3, -4 and -7 regulate the conformation of ZO-1. In turn, their depletion caused F-actin and non-muscle myosin II (NMII) disruption, increased tension on the AJ and disrupted barrier integrity. Furthermore, JAM-A and claudin-1, 2, -3, -4 and -7 together with CXADR were necessary to anchor the ZO-1 N-terminus at the membrane, and their combinatory depletion disrupted nanometer-scale organization of ZO-1 (Nguyen et al., 2024). Thus, SIM and STED have been used to show that the mechanosensitive properties of ZO-1 are controlled by the TJ transmembrane proteins that anchor it in the membrane. This anchoring is crucial for enhancing mechanical resistance and junctional integrity.

The first studies applying SR microscopy to the AJ addressed its nanoscale molecular organization. 3D stochastic optical reconstruction microscopy (3D-STORM) (Wu et al., 2015) and interferometric photoactivated localization microscopy (iPALM) (Truong Quang et al., 2013) have achieved the high resolution needed to quantify the number of E-cadherin molecules within junctional assemblies or clusters. The size of the cadherin clusters depended on the number of adhesion molecules, the cortical actin network (Wu et al., 2015) and the rate of cadherin endocytosis (Truong Quang et al., 2013). iPALM and surface-generated structured illumination were used to investigate the organization of junctional cadherin–catenin complexes along the axis orthogonal to the plasma membrane. Whereas regular SIM uses a grid to create a structured illumination pattern across the sample, surface-generated structured illumination uses variable scanning angles to obtain interference patterns, thus enhancing observational detail close to thin surfaces such as the cell membrane (Ajo-Franklin et al., 2005; Bertocchi et al., 2017; Paszek et al., 2012). AJ components were found to be organized into a multi-layered stratified architecture localized ∼30 nm from the actin cytoskeleton, a distance that can be bridged by adaptor proteins, such as the unfolded vinculin protein (Bertocchi et al., 2017). SR methods have thus provided the resolution necessary to assess the nanoscale distribution of specific molecules in AJs.

SIM of junctions in immature epithelial monolayers has demonstrated that the AJ consists of separate cadherin and nectin clusters that are both attached to the circumferential actin belt (Indra et al., 2013). Recently, STED microscopy with physical cryosectioning was used to overcome the optical challenge posed by thick apicobasal polarized epithelial cells, enabling the study of AJ organization during 14 days of epithelial monolayer maturation (Mangeol et al., 2024). Surprisingly, this revealed that during epithelial junction maturation, the nectin–afadin complexes segregate from more basal cadherin–catenin complexes and associate with the majority of the F-actin belt (Fig. 3A). Together with previous observations showing that afadin is necessary to assemble the circumferential actin belt and that its depletion disrupts actomyosin organization at the ZA (Sakakibara et al., 2018), these findings emphasize that nectin-based adhesions, rather than cadherin-based adhesions, are major-load bearing structures in mature epithelial junctions (Mangeol et al., 2024) (Fig. 3A). In conclusion, these studies highlight the application of SR to address the intramolecular organization and dynamic regulation of the AJ and reveal that nectin–afadin complexes play a major role in force regulation in mature epithelial tissue, thus redefining our understanding of force-dependent adhesion mechanisms.

Live imaging studies established the sequential recruitment of various desmosome cadherins and integration of cytoplasmic proteins into the desmosome core complex. Nascent desmosome formation depends on prior formation of cadherin-based adhesions (Nekrasova and Green, 2013). Dsg and Dsc cadherins are then recruited and form trans heterodimers at cholesterol-rich plasma membrane regions (Harrison et al., 2016; Nekrasova and Green, 2013; Nekrasova et al., 2011). In turn, the adaptor proteins DP, PKP2 and Pg associate with the desmosome core and the IF cytoskeleton, thereby controlling adhesion clustering (Godsel et al., 2005; Green et al., 2019; Najor, 2018; Nekrasova and Green, 2013). During maturation of cell–cell contacts, desmosome cadherins transition from Ca²⁺-dependent to Ca²⁺-independent adhesions, leading to a hyper-adhesive state (Fig. 3A) (Garrod, 2013; Kimura et al., 2007). However, how desmosome assembly and maturation dynamics occur at the nanoscale, as well as their relationship with the plasma membrane and the cellular environment, remained unclear. Recent studies using SR and vEM have begun to address these questions.

SIM has revealed that desmosome assembly requires E-cadherin trans-homodimerization and heterophilic binding with Dsg2. At later stages of junction maturation, E-cadherin is replaced by Dsc2, which creates a more stable Dsg2–Dsc2 dimer (Shafraz et al., 2018). A combination of scanning EM and SIM has shown that DP is necessary to integrate and stabilize small clusters of Dsg–Dsc cadherins into larger clusters (Wanuske et al., 2021). DP depletion leads to a lack of desmosomes and disrupted intercellular adhesion whereas Pg depletion affects anchoring of IFs and desmosome morphology (Wanuske et al., 2021) (Fig. 3A). Furthermore, live STED showed that Dsg3 resides in both desmosome and extradesmosome pools (Fuchs et al., 2023; Wanuske et al., 2021), which associate with the IF network and actin cytoskeleton respectively (Fuchs et al., 2023). Upon incorporation of Dsg3 into the desmosome core, its interaction with the actin cytoskeleton is replaced with interactions with the IF network (Fuchs et al., 2023) (Fig. 3A).

SIM also demonstrated that a mutation in the transmembrane domain region of Dsg1 reduced its incorporation into desmosomes in an epidermal carcinoma cell line (Lewis et al., 2019), resulting in smaller and weaker desmosomes (Zimmer et al., 2022). This effect might relate to the notion that desmosome assembly requires proper embedding of junctional transmembrane proteins into cholesterol- and sphingolipid-enriched membrane lipid rafts (Brennan et al., 2012; Resnik et al., 2011; Zimmer and Kowalczyk, 2020). Further, cryo-electron tomography (cryoET) has shown that desmosome-associated lipid rafts are ∼10% thicker than the surrounding plasma membrane (Lewis et al., 2019). Therefore, membrane composition could play an important role in regulating desmosome function.

In summary, SR studies have demonstrated that E-cadherin guides initial desmosome assembly, after which DP-dependent clustering and stable Dsg–Dsc dimer formation takes over. Additionally, proper embedding of Dsg within lipid rafts is likely important for desmosome structure and function.

After the initial desmosome plaque is formed, it switches to a hyper-adhesive state. SMLM (specifically, direct stochastic optical reconstruction microscopy; dSTORM) has been used to characterize the architecture of maturing desmosome plaques by resolving the nanoscale organization of Dsg3, Pg and the C-terminal, N-terminal and rod domains of DP (Fig. 2B) (Beggs et al., 2022; Stahley et al., 2016). Generally, the plaque resembles a trapezoid, with the short side proximal to the plasma membrane and the long side at a distance of ∼120 nm from the membrane. The C-terminal and rod domains of DP are oriented at an angle to the plasma membrane (Stahley et al., 2016); during desmosome maturation, they move closer to the membrane. This coincides with a loss of E-cadherin from the desmosomes and an increase in adhesive strength (Beggs et al., 2022) (Fig. 3A). Experiments with a DP-based tension-sensing FRET module have indicated that the DP rod domain unfolds and extends in presence of mechanical stress (Price et al., 2018). This suggests that a conformational change occurs within DP during desmosome maturation that could act to absorb junctional force.

Although desmosomes are known to strongly associate with the IF network, ultrastructural information regarding their interaction within the cellular environment is less clear. FIB-SEM imaging of epidermal carcinoma cells has revealed that an elaborate network of keratin filaments connects desmosomes to the nucleus (Jorgens et al., 2017). Cryo-SIM with FIB-SEM have further revealed the 3D organization of a desmosome–keratin–endoplasmic reticulum (ER)-linked complex (Bharathan et al., 2023). The keratin bundles and peripheral ER tubules are positioned orthogonally to the plasma membrane and physically associate with the desmosome junction (Fig. 3A). Live imaging has confirmed that anchoring of keratin filaments by desmosomes is crucial for ER stability (Bharathan et al., 2023) (Fig. 2C). Combined with observations of ER–plasma membrane contact sites (Chung et al., 2022) and ER extensions near cadherin-based junctions (Joy-Immediato et al., 2021), these findings support a new role for desmosomes in ER organization. In conclusion, SR and vEM methods have shed light on the molecular events that occur during desmosome assembly and how desmosome architectures associate with the ER and nucleus.

Pulling and pushing forces generated by the cytoskeleton are sensed by junction-associated mechanosensitive molecules to promote intercellular interaction and maintenance and reinforcement of junctions. The molecular effects of these forces at junctions are increasingly well characterized. Cytoskeletal remodeling is driven by signaling pathways initiated by Rho family GTPases that control pulling and pushing forces (Yamada and Nelson, 2007). At the junction, actomyosin-induced pulling forces cause the unfolding and stabilization of actin-bound α-catenin and recruitment of vinculin (Angulo-Urarte et al., 2020; Khalil and de Rooij, 2019; Ladoux and Mege, 2017). This is mediated by RhoA-ROCK-myosin light chain kinase (MLCK) signaling (Wozniak and Chen, 2009), which regulates the phosphorylation of NMII on actin bundles, generating actomyosin contractions. Conversely, Rac1 GTPase activity induces actin-based protrusions that produce pushing forces, which are locally generated by junctional association of actin elongation factors, such as Mena (also known as ENAH) (Leerberg et al., 2014; Vasioukhin et al., 2000), vasodilator-stimulated phosphoprotein (VASP) (Leerberg et al., 2014), formins (Grikscheit et al., 2015), Ena/VASP-like protein (EVL), collapsin response mediator protein 1 (CRMP-1) (Yu-Kemp et al., 2017) and actin-branching proteins, such as ARP2/3 proteins (Helwani et al., 2004; Kovacs et al., 2002; Verma et al., 2004). These protrusive actin regulators support both the formation of junctions by enhancing cell-cell contact and maintenance of established junctions.

To understand the nature and direction of cytoskeletal-derived forces and their impact on junctional dynamics, visualizing the spatial organization of the cytoskeleton and its components in relation to the junction and the plasma membrane is key. For instance, EM has demonstrated the presence of circumferential actin filaments at AJs (Yonemura et al., 1995). Moreover, PREM offers the ability to visualize the organization of the actin, IFs and MT cytoskeleton in various cellular compartments at single-filament resolution (Svitkina et al., 1995). PREM-based visualization of the cytoskeletal geometry at endothelial junctions has revealed how actin-driven pulling and protrusive forces correlate with specific AJ subtypes (Efimova and Svitkina, 2018). Mature linear VE-cadherin junctions associate with ARP2/3-positive branched actin networks, whereas NMIIA-positive circumferential contractile actin bundles localize at a distance from the AJs and connect to the junction-linked branched actin network through small oblique actin bundles (Efimova and Svitkina, 2018) (Figs 2C, 3B). Based on previous studies using conventional fluorescence microscopy, contractile actin bundles were thought to directly contact AJs (Huveneers et al., 2012; Millan et al., 2010; Vasioukhin et al., 2000). By visualizing junctions subjected to inward pulling force, PREM revealed that they in fact terminate at the junction-associated branched actin network (Efimova and Svitkina, 2018). Thus, PREM has helped to show that contractile actin bundles are physically separated from junctions, whereas branched actin networks are directly associated (Efimova and Svitkina, 2018).

SIM has been used to study the spatial arrangements of the non-muscle myosin isoforms NMIIA and NMIIB at nascent (Heuze et al., 2019) and mature epithelial junctions (Gomez et al., 2015). Consistent with the findings from PREM studies (Efimova and Svitkina, 2018), NMIIA localizes on actin bundles parallel to the junction (Gomez et al., 2015; Heuze et al., 2019) and exerts contractile forces downstream of RhoA signaling, enabling junctional elongation and growth (Heuze et al., 2019; Smutny et al., 2010). NMIIB, however, localizes and exerts anisotropic pulling forces at the junctional membrane (Gomez et al., 2015), co-occurring with ARP2/3-and cortactin-positive branched actin networks (Heuze et al., 2019). The depletion of NMIIB increases the extent of the branched actin network, which leads to weakening of junctional adhesions due to excessive protrusive activity (Heuze et al., 2019). At the junction, NMIIB contractility rigidifies and counterbalances the extent of the branched network while ensuring junction-cytoskeletal coupling through α-catenin (Heuze et al., 2019). The branched actin network links to NMIIA-associated peri-junctional actin bundles, which control reinforcement and elongation of the junction (Fig. 3B). Different NMII isoforms thus exert forces at distinct areas of the junctional actin cytoskeleton.

How is the branched actin network assembled at the cell–cell contact interface? Thin section EM of lateral AJs in epithelial cells has indicated that the local plasma membrane is highly curved (Li et al., 2021, 2020). Interestingly, SBEM and 3D electron tomography have shown that the membrane morphology at lateral AJs includes folds with negative curvatures (Fig. 2D) (Senju et al., 2023). The negative membrane curvature-sensing protein metastasis suppressor-1 (MTSS1) localizes at these folds, where it interacts with WAVE-2 (Senju et al., 2023). WAVE-2 is an ARP2/3-activating protein downstream of junctional Rac1 signaling (Dawson et al., 2012) required for junctional actin assembly (Senju et al., 2023). Concomitantly, depletion of MTSS1 reduces actin-based protrusions, weakens intercellular interaction and compromises epithelial junctional integrity (Saarikangas et al., 2011; Senju et al., 2023) (Fig. 3B). Thus, the morphology of the plasma membrane can control actin-driven pushing forces, which in turn promote junctional adhesion.

From SR and vEM microscopy, a model emerges in which junction-proximal pushing forces exerted through branched actin networks are concentrated at the membrane to maintain intercellular interactions, while junction-distal pulling forces promote junctional reinforcement. What happens when the junctional pushing and pulling continuum is challenged? Inhibition or depletion of ARP2/3 dissolves the branched actin network and RhoA-induced pulling forces subsequently destabilize the AJs (Efimova and Svitkina, 2018; McEvoy et al., 2022). Inhibition of NMII-mediated contractility also destabilizes AJs by causing excessive Rac1-driven branched actin formation, pushing forces that are associated with weakened junctional adhesions, and the formation of intercellular gaps as a result of reduced AJ-cytoskeletal coupling (Efimova and Svitkina, 2018; McEvoy et al., 2022). Thus, in mature AJs, a balance of pushing and pulling forces maintains adhesion and promotes junctional repair (Fig. 3B).

A particular example of cytoskeletal imbalance between connected cells is the formation of asymmetric AJs, or cadherin fingers, at the interface of leader and follower cells during endothelial collective cell migration. VE-cadherin-based AJs are found at protrusions that extend from the leader cell into the follower cell; these are formed through actomyosin-mediated pulling by the leader cell and Rac1-dependent cytoskeletal pushing by the follower cell (Fig. 3C). In the follower cell, the cadherin complex proteins are internalized (Brevier et al., 2008; Dorland et al., 2016; Hayer et al., 2016). PREM has shown that the cytoskeletal structure of the asymmetric AJ consists of thin actin filaments within the protrusion. Furthermore, a funnel-shaped branched actin network that is oriented towards the cell–cell contact has been visualized in the follower cell (Efimova and Svitkina, 2018). 2D EM revealed that the plasma membrane at the asymmetric junction resembles an extended membrane tube (Hayer et al., 2016). STED microscopy demonstrated that the F-BAR protein PACSIN2 is specifically recruited to the positively curved junctional membrane to control internalization of VE-cadherin in follower cells (Fig. 2E) (Dorland et al., 2016). In general, BAR domain-containing proteins sense and generate nanoscale membrane curvatures (Simunovic et al., 2019; Simunovic et al., 2015), control protein trafficking and regulate Rho GTPase-mediated cytoskeletal remodeling (Carman and Dominguez, 2018). In addition, SIM and STED microscopy in alveolar epithelial cells reveals that claudin-18- and ZO-1-based adhesions have similar asymmetric junctional structures, which associate with the trafficking regulator dynamin-2 (Lynn et al., 2021). Thus, BAR proteins are active at the junction–membrane interface and innovative imaging approaches are likely to obtain new insights into the relationship between plasma membrane organization and trafficking of junctional proteins.

Collectively, these studies show that a pushing and pulling continuum ensures junctional interaction and integrity. Curved plasma membranes support ARP2/3-driven protrusive actin networks that enhance intercellular adhesion. In turn, anisotropic NMIIB contractility reinforces junction–cytoskeletal coupling and restricts the extent of the protrusive network. Meanwhile, parallel NMIIA pulling forces promote junction elongation. Slight deviations from this force continuum facilitate junctional remodeling and tissue dynamics. However, excessive imbalance of forces compromises the integrity of junctions and cell–cell contacts.

SR and EM are powerful approaches to explore cell–cell junction ultrastructures in unprecedented detail. However, fully grasping the subcellular processes controlling junction dynamics requires monitoring changes in the repertoire of junctional proteins, associated cytoskeletal networks and their regulatory signaling pathways with high temporal resolution. Biosensors measuring, for instance, Rho GTPase activity (Kreider-Letterman et al., 2023; Mahlandt et al., 2023a; Stephenson et al., 2019) and junctional forces [such as FRET tension sensors for cadherins (Borghi et al., 2012; Conway et al., 2013; Ringer et al., 2017), α-catenin (Acharya et al., 2017), ZO-1 (Haas et al., 2020) and desmoplakin (Price et al., 2018)] offer unique insights into spatiotemporal activation and inactivation of junctional (mechano)signals. Optogenetic advancements have led to the development of novel imaging tools that enable spatiotemporal control of junctional remodeling (Mahlandt et al., 2023b; Nzigou Mombo et al., 2023; Varadarajan et al., 2022). These tools are reversible, non-invasive and inducible, ideal for use in living cells and model systems. This section highlights developments in microscopy-based biosensors and optogenetic tools and their contributions to understanding junction dynamics.

Unimolecular Förster resonance energy transfer (FRET) biosensors rely on energy transfer between two fluorescent proteins, where emitted light from one is absorbed and emitted by another at a different wavelength. When the proteins are in close proximity, energy transfer occurs, and an increase in FRET is observed. Insertion of FRET fluorophores separated by an elastic linker into a protein of interest allows one to monitor tension and compressive forces, which can reveal mechanical properties of individual proteins and their responses to external mechanical stimuli (Fischer et al., 2021). Multiple FRET tools are available to assess the impact of forces on junction components (Table 1). The first FRET tension sensors (TSs) for vinculin (Grashoff et al., 2010), the cytoplasmic domain of E-cadherin (EcadTSmod) (Borghi et al., 2012) and VE-cadherin (Conway et al., 2013) were used to measure actomyosin-generated tension at junctions. These tension measurements were later confirmed by an α-catenin-based TS (Acharya et al., 2017). Interestingly, combining FRET imaging of α-catenin TSs with force activation by magnetic twisting cytometry has demonstrated that mechanical perturbations directly induce stretch-mediated conformational changes in α-catenin (Kim et al., 2015). Although early studies used ratiometric FRET, more recently fluorescence lifetime imaging microscopy (FLIM) has been used to measure the fluorescent lifetime of the donor fluorophore, which decreases upon efficient energy transfer. FLIM is independent of fluorophore concentration, photobleaching and optical path length, and thus provides more quantitative, reliable and robust measurements of FRET efficiency. Additionally, improved FRET modules containing ferredoxin-like linkers are more accurate for measuring forces in the low pN region (Ringer et al., 2017) and have been implemented in a new VE-cadherin TS (Arif et al., 2021). This sensor was used to reveal that leukocytes rapidly increase tension on, and promote tension-dependent dephosphorylation of, VE-cadherin, leading to endothelial junction destabilization and leukocyte diapedesis (Arif et al., 2021).

FRET sensors have revealed key differences between junctional structures. To investigate mechanotransduction at desmosomes, a DP FRET sensor has been used to demonstrate that desmosomes, unlike AJs, are not sensitive to actomyosin-generated stresses but absorb stress from externally applied pulling forces (Price et al., 2018). For TJs, a ZO-1-based FRET sensor was used to show that these structures also experience actomyosin-derived tension (Haas et al., 2020). Further studies have indicated that depleting the TJ scaffolding protein ZO-2 increases tension across ZO-1 at TJs but does not affect AJ tension (Pinto-Duenas et al., 2024). These findings suggest that distinct force-bearing components reside at the junctional interface. To assess changes in tension across distinct junctional components, multiplexing FRET biosensors (Windgasse and Grashoff, 2023) might enable simultaneous monitoring of different junction structures during adhesion, maturation and remodeling.

Another junction-related FRET sensor has been developed by inserting a spectrin-repeat tension-sensing module (Meng and Sachs, 2012) within a flexible linker region in the actin-binding protein α-actinin-4 (Morris et al., 2022). Experiments with this α-actinin-4–sstFRET522 sensor in confluent epithelial cells showed higher FRET levels at junctions compared to those in junction-free protrusive areas, suggesting that α-actinin-4 is under compressive forces at junctions. FRET levels further increased within the junction-associated contractomeres, myosin motor-driven structures that control sliding of junctional vertices during apical constriction (Morris et al., 2022).

The development of biosensors and optogenetic tools now permits inducible control of intra- and extra-cellular interactions of the junctional complexes and their associations with the cytoskeleton. The tools discussed here are based on molecular systems that unfold in blue light, causing inducible recruitment of proteins to subcellular compartments or reversible separation of intra and interprotein domains (Table 2).

An optogenetic E-cadherin (opto-E-cad) tool, with a LOV2 domain inserted at an extracellular Ca²⁺-binding site, was recently developed (Nzigou Mombo et al., 2023). Blue light causes unfolding of the LOV2 Jα-helix and disrupts Ca²⁺-controlled adhesion, impairing cohesion, collective cell migration and invasion in 2D, 3D and in vivo tissues. The reversibility of LOV2 folding allows for temporal modulation of adhesion by controlling light cycles (Nzigou Mombo et al., 2023). Opto-E-cad is compatible with live-imaging and with other fluorescence channels, enabling simultaneous study of extracellular adhesion and molecular dynamics, and facilitating tunable control of junctional remodeling.

Optogenetic tools have also been developed to study desmosomes (Sadhanasatish et al., 2023). Here, the light-inducible dissociating module AsLOV2–Zdk2 (Wang et al., 2016) was inserted into two DP fragments, with AsLOV2 in the N-terminal-rod domain and Zdk2 in the tail domain. Blue light causes these DP domains to separate, resulting in loss of desmosome association with IFs (Sadhanasatish et al., 2023). This confirmed previous research (Wanuske et al., 2021) showing that interactions with IFs are dispensable for desmosome assembly under homeostatic conditions, but necessary for maintaining desmosome adhesion and cell–cell cohesion under external mechanical loads (Sadhanasatish et al., 2023).

Junction adhesion depends on cytoskeletal force generation and Rho GTPases to remodel the actin cytoskeleton. Traditional methods track Rho GTPase activation using FRET sensors; however, Rho FRET sensors are difficult to combine with other fluorophores and require advanced imaging set-ups. Location-based Rho GTPase biosensors, which rely on fluorescently tagging subdomains of Rho effector proteins, overcome these limitations. These biosensors permit the close monitoring of localization dynamics of active Rho GTPases (Mahlandt et al., 2021, 2023a).

Finally, active modulation of cytoskeletal forces could provide better insights into junctional dynamics. Traditionally, constitutively active or inactive proteins or inhibitors have been used to affect cytoskeletal dynamics at the cellular scale. Recently developed optogenetic Rho GTPases provide local tuneability of cytoskeletal dynamics. Illumination-dependent heterodimerization of plasma membrane-bound anchors like LOVpep (Strickland et al., 2012; Wagner and Glotzer, 2016) or iLID (Guntas et al., 2015; Mahlandt et al., 2023b) have been used to recruit specific RhoGEFs to the plasma membrane on demand and control local cytoskeletal remodeling in a reversible manner. Optogenetically activating Cdc42 and Rac enhances barrier function in endothelial cells (Mahlandt et al., 2023b) whereas optogenetic RhoA activation induces local contractions and junction remodeling (Mahlandt et al., 2023b). Interestingly, a combination of Rho and Ca²⁺ biosensors with RhoA optogenetic tools has been used to follow the temporal sequence of TJ reinforcement in Xenopus embryos (Varadarajan et al., 2022). That study demonstrates that the epithelial barrier integrity depends on mechanical activation of Ca²⁺ channels, which causes local Rho-dependent TJ remodeling (Varadarajan et al., 2022).

vEM and SR have enabled the creation of increasingly accurate models of the interplay between junction complexes, the cytoskeleton, the plasma membrane and the cellular environment. SIM and STED are relatively easy to use, but this ease comes at the cost of resolution. Higher resolution can be obtained from SMLM, but this approach involves significant technical and analytical challenges. Biosensors and optogenetics allow for the assessment of mechanisms of junctional and cytoskeletal regulation in diverse systems, enhancing our understanding of adhesion across development, homeostasis and disease. Future studies combining these approaches are necessary to further evaluate the role of the junctional cell membranes. For instance, do folded membranes actively create microdomains and how do they contribute to junction trafficking? vEM methods are uniquely capable of characterizing the membrane ultrastructure, and the combination of these methods with correlative light microscopy techniques that can reveal molecular localizations of junctional components makes them promising for the future of junction biology. However, their application remains limited due to their low-throughput capacity and requirements for high technical expertise and computing power. Going forward, international and trans-continental initiatives by advanced imaging centers, open-source analysis platforms and public data repositories can help increase global access to complex techniques and datasets needed to expand our understanding of junctional architecture and dynamics (Collinson et al., 2023; Czymmek et al., 2024).

The authors thank Dr J. Goedhart and Dr M. Grönloh for reading and commenting on the manuscript and N. Nojszewska for scientific discussions.