Desmosomes, strong cell–cell junctions of epithelia and cardiac muscle, link intermediate filaments to cell membranes and mechanically integrate cells across tissues, dissipating mechanical stress. They comprise five major protein classes – desmocollins and desmogleins (the desmosomal cadherins), plakoglobin, plakophilins and desmoplakin – whose individual contribution to the structure and turnover of desmosomes is poorly understood. Using live-cell imaging together with fluorescence recovery after photobleaching (FRAP) and fluorescence loss and localisation after photobleaching (FLAP), we show that desmosomes consist of two contrasting protein moieties or modules: a very stable moiety of desmosomal cadherins, desmoplakin and plakoglobin, and a highly mobile plakophilin (Pkp2a). As desmosomes mature from Ca2+ dependence to Ca2+-independent hyper-adhesion, their stability increases, but Pkp2a remains highly mobile. We show that desmosome downregulation during growth-factor-induced cell scattering proceeds by internalisation of whole desmosomes, which still retain a stable moiety and highly mobile Pkp2a. This molecular mobility of Pkp2a suggests a transient and probably regulatory role for Pkp2a in desmosomes.
Vertebrates have evolved several multimolecular systems to provide the cell cohesion and resilience necessary to withstand mechanical stress and deformation without a protective exoskeleton. Epithelia – the functional interfaces around, within and between body organs – have evolved complex multiprotein desmosome cell–cell junctions. Desmosomes connect to cytoplasmic intermediate filament networks to form a mechanically resilient transcellular tissue system. Together with adherens junctions and tight junctions, desmosomes enable integration and stratification of epithelia, whilst ensuring that barrier function is maintained at all times and also that damaged tissue cells can be replaced. Loss of desmosome function is linked to severe diseases, particularly in mechanically challenged tissues such as skin and cardiac muscle (reviewed by Chidgey and Dawson, 2007; Delmar and McKenna, 2010; Dusek and Attardi, 2011; Ishida-Yamamoto and Igawa, 2014; Maruthappu et al., 2019; Spindler et al., 2018; Thomason et al., 2012).
Within the desmosomal macromolecular complex, the extracellular domains of the desmosomal cadherins (DCs), desmocollin (Dsc) and desmoglein (Dsg), bind tightly to their counterparts on the adjacent cells, while their intracellular domains connect via plakoglobin (PG, also known as JUP) and plakophilins (Pkps) to intermediate filament-binding dimers of desmoplakin (DP) (Garrod and Chidgey, 2008; Green et al., 2019). Desmosomes of intact homeostatic tissues exist in a state of hyper-adhesion, rendering them particularly stress resistant (Wallis et al., 2000; Garrod et al., 2005). Whilst newly formed desmosomes need extracellular Ca2+ for adhesion through their DCs, mature hyper-adhesive desmosomes remain robustly attached to each other even after experimental removal of extracellular Ca2+ by chelation.
Hyper-adhesion is developmentally regulated in the mouse embryo and is acquired with time in confluent culture of several cell lines (Kimura et al., 2007, 2012; Wallis et al., 2000). It has been suggested that hyper-adhesion is associated with an ordered arrangement of the extracellular domains of the DCs, indicated by the presence of an electron-dense midline present halfway between the desmosomal cell membranes, and that this may result in Ca2+ being locked into the structure (Garrod et al., 2005). It has been reported that this ordered arrangement of extracellular domains of genetically modified Dsg3 is lost upon removal of extracellular Ca2+, both from pharmacologically-induced hyper-adhesive and Ca2+-dependent desmosomes, suggesting that the ordering of DCs is not the only factor involved in desmosomal adhesion (Bartle et al., 2020, 2017).
Whilst tissue integrity depends upon the maintenance of stable cell–cell junctions, cells must retain the ability to free themselves from their neighbours in order to move or to get rid of damaged cells. During embryonic development, in wound healing and in tissue regeneration, cell adhesion must be temporarily relaxed or lost for cells to undertake tissue remodelling. This is seen in epithelial-to-mesenchymal transition (EMT), the conversion of compact epithelial sheets to solitary or small groups of motile cells, which occurs during development and in metastatic progression of some cancers, and is often triggered by growth factor signalling (Nieto et al., 2016).
Despite the importance of cell adhesion in health and disease, little is known about how desmosomal adhesion is lost in vivo. A limited amount of evidence suggests that loss of desmosome-dependent cell adhesion may occur through internalisation of whole desmosomes by the cells involved (Allen and Potten, 1975; Garrod et al., 2005). This has not been studied in detail because no tissue culture model of the process exists. Studies on the internalisation of half desmosomes, induced by (non-physiological) Ca2+ chelation, have shown that this process is actin dependent, that the internalised half desmosomes remain intact until degraded by a combination of lysosomal and proteasomal activity, and that desmosomal proteins are not recycled (McHarg et al., 2014). Whether similar mechanisms apply to the internalisation of whole desmosomes under physiological conditions remains to be determined.
Desmosomes are undoubtedly highly stable structures, stability being key to their function, and it is becoming clear that the protein exchange rates within desmosomes mirror their stability, as has been shown for Dsg2, Dsg3, Dsc2a, DP and PG (Bartle et al., 2020; Foote et al., 2013; Gloushankova et al., 2003; Lowndes et al., 2014; Moch et al., 2020; Vielmuth et al., 2018; Windoffer et al., 2002). Nevertheless, they are also dynamic, with a life cycle of four phases: de novo assembly from their component molecules, a weakly-adhesive Ca2+-dependent phase compatible with cell migration, a strongly adhesive Ca2+-independent phase in tissue homeostasis, and desmosome removal or breakdown as epithelia become activated by relevant signalling.
We investigated the complex relationships between desmosomal proteins during the last three stages of this life cycle using time-lapse fluorescence microscopy, including fluorescence recovery after photobleaching (FRAP) and fluorescence loss and localisation after photoactivation (FLAP). A culture model for desmosome internalisation was established to mimic the in vivo situation more closely, and we observed that one of the desmosomal proteins, the plakophilin 2 isoform Pkp2a, behaves very differently from the others both in Madin–Darby canine kidney (MDCK) and human colorectal adenocarcinoma (CaCo-2) cells. In Ca2+-dependent desmosomes, the molecular mobility of the Dsc2 isoform Dsc2a, Dsg2, PG and DP is very low and is further reduced as the cells become hyper-adhesive, forming a stable desmosomal moiety or module. DP in MDCK cells is mobile initially but stabilises with the stable moiety with the onset of hyper-adhesion. In contrast, Pkp2a shows a unique and persistently dynamic exchange between its desmosomal and cytoplasmic pools. Similar relative protein dynamics persist when whole desmosomes are internalised during HGF-induced scattering of MDCK cells. These results suggest a uniquely dynamic role for Pkp in desmosomes.
Ca2+-dependent desmosomes exhibit differential protein dynamics
To shed light on the behaviour of the individual desmosomal components, we first analysed their dynamic properties using FRAP. The relative mobility of the proteins was assessed by determining their half-time of fluorescence recovery (t1/2) and mobile fraction (Fm) (Carisey et al., 2011). Dsc2a, Dsg2, PG, DP and Pkp2a tagged with either mNeonGreen (referred to hereafter as neonGreen) or eGFP fluorophores were stably expressed in the simple epithelial cell line MDCK type II (Fig. 1A), a well-established model for the study of cell–cell junctions (Dukes et al., 2011). The cells were plated at confluent density and cultured for 24 h prior to FRAP, at which time the majority of the desmosomes showed Ca2+-dependent adhesion (Fig. S1A,B) (Wallis et al., 2000).
A wide distribution in the Fm of the desmosome proteins in MDCK cells was observed (Fig. 1B; Table S1). The transmembrane proteins Dsg2 (Fm=29.8%) and Dsc2a (Fm=35.7%) showed the lowest mobile fractions, closely followed by the cytoplasmic plaque protein PG (Fm=37.7%). DP was slightly more mobile (Fm=50.2%), whereas Pkp2a showed the highest mobile fraction (Fm=74.6%) (Fig. 1C). Analysis of the recovery times showed similar results, with slower recovery times of DCs (Dsg2, t1/2=75.7 s; Dsc2a, t1/2=62.9 s) and PG (t1/2=60.1 s) in comparison to the significantly faster recovery time of Pkp2a (t1/2=30.1 s) (Table S1). To put the high stability of the DCs into perspective against actin-associated cell–cell junction proteins, we measured the mobility of E-cadherin, the adhesion receptor of adherens junctions. The mobile fraction of E-cadherin tagged with mEmerald (E-Cad) was much higher (Fm=52.8%) than the mobile fractions of the DCs, demonstrating that desmosomal receptors are more stable at the cell–cell interface than E-cadherin in adherens junctions (Fig. 1C; Table S1).
The very low mobile fraction of Dsg2, Dsc2a and PG prompted us to evaluate the contribution of the reversible photobleaching properties of the fluorophores, which can be up to 20% of the fluorescence intensity prior to the bleaching event (Sinnecker et al., 2005). Performing FRAP on paraformaldehyde-fixed cells, we observed Fm=6.8% and t1/2=11.3 s for neonGreen and Fm=10.2% and t1/2=13.1 s for eGFP after 5 min of recordings, which reflect the inherent recovery properties of the fluorescent tags (reversible photobleaching; red dotted line in Fig. 1B). These recovery rates suggest an even lower mobile fraction of the desmosomal proteins and impair the determination of accurate t1/2, in particular for the stable desmosomal proteins (indicated in Table S1). However, these values cannot directly be subtracted from the Fm and adjusted to t1/2 measured in the live-cell recordings, as the fixation process may lead to conformational changes in the fluorophores that may result in varied lifetimes and properties (Becker, 2012).
To bypass the potential issue of reversible photobleaching and for further analysis, we analysed FLAP (Fig. 2A). This allowed us to calculate the fluorescence decay within the activated region of interest (ROI) as well as to track the activated fluorescent proteins within their environment. Using photoactivatable-GFP (PaGFP)-tagged Dsc2a, with mScarlet–Pkp2a used as a desmosome marker, and PG–PaGFP with Dsc2a–mScarlet, FLAP analysis showed similar protein turnover to that observed using FRAP. Both techniques revealed a low mobile fraction for Dsc2a and PG, whereas PaGFP–Pkp2a, with Dsc2a–mScarlet as a desmosome marker, had a significantly higher turnover, as was also seen by FRAP (Figs 1C, 2B).
These results were recapitulated in transiently transfected human colorectal adenocarcinoma cells (CaCo-2) (Fig. S1E–H). In both FRAP and FLAP experiments, Pkp2a showed a significantly higher mobile fraction (72.8% and 79.3%, respectively) in comparison to those of Dsc2a, PG and DP. Interestingly, DP was less mobile in CaCo-2 cells and had a FRAP mobile fraction of 24.1%, which was similar to those of Dsc2a (24.1%) and PG (21.3%) and to the FRAP mobile fraction value of 28% recently reported for DP in transiently transfected HaCaT cells (Bartle et al., 2020), indicating that the mobile fraction of DP may vary between cell types and different adhesion states of desmosome in these cell lines.
We conclude that in Ca2+-dependent desmosomes, DCs and PG form a stable moiety of desmosomes, whereas Pkp2a has a dramatically higher mobile fraction, which suggests a more transient association of this protein with desmosomes. In Ca2+-dependent desmosomes of MDCK cells the mobile fraction of DP is intermediate between that of the stable moiety and the highly mobile Pkp2a.
Plakophilin 2a shows rapid exchange between desmosomes and a cytoplasmic pool
The mobility of desmosomal proteins could be due to a local exchange of proteins at the plasma membrane, an exchange with intracellular pools or both. In order to determine the mechanism of Pkp2a exchange at desmosomes, we performed further FLAP experiments with MDCK cells. FLAP recordings of Dsc2a–PaGFP activated at the plasma membrane were taken over 30 mins, confirming the extraordinarily stable localisation and apparent absence of lateral diffusion of Dsc2a (Fig. 2C). In contrast, photoactivated PaGFP–Pkp2a at the same settings distributed rapidly along the entire membrane (Fig. 2C). To assay the flux between the cytoplasmic pool of Pkp2a and the desmosome-associated Pkp2a, we performed FLAP analysis by targeting the 405 nm laser light to a 10 µm-diameter circular region of the cytoplasm for localised photoactivation and tracking of PaGFP–Pkp2a diffusion out from this region (Fig. 2D). PaGFP–Pkp2a diffused within seconds throughout the entire cytoplasm and was found to become rapidly enriched at the membrane (Fig. 2D,E). Line profile analysis revealed overlapping fluorescence intensity peaks of PaGFP–Pkp2a and Dsc2a–mScarlet, indicating the recruitment of Pkp2a from the cytoplasm to desmosomes (Fig. 2F). Similar attempts to perform FLAP analysis with the other desmosomal components Dsc2a and PG failed because of their extremely low presence in the cytoplasm (Fig. S1C). Taken together, the results reveal a previously unknown property of Pkp2a: that it circulates continuously between the desmosomal plaque and the cytoplasm, in stark contrast to the more stable association of the other studied components with the desmosomes.
Hyper-adhesive desmosomes are more stable yet retain high Pkp2a turnover
We next asked whether the dynamics of the major desmosomal proteins changed as desmosomes matured from Ca2+-dependent to Ca2+-independent hyper-adhesion, with two major questions in mind: (1) do the major proteins acquire greater stability, and (2) does Pkp2a remain more dynamic than the other proteins? We first established in all the MDCK cell lines we tested that most desmosomes had become hyper-adhesive after 3 days of confluent culture (Fig. S1A,B). To investigate whether the adhesive state of desmosomes affects their protein dynamics, we performed FRAP analysis using MDCK cells after 3 days of confluent culture (Fig. 3A,B) and compared the results with those obtained for cells after 1 day of confluent culture (Fig. 1). All the tested desmosomal proteins became significantly less mobile as desmosomes matured to hyper-adhesion (Fig. 3B; Table S1). In contrast, the mobile fraction of E-cadherin was not significantly reduced over the same time period. Dsg2 showed the lowest mobile fraction of 13.6%, followed by Dsc2a and PG (both less than 25%). Taking into account reversible photobleaching of up to 10%, the data suggest that these three proteins remain almost static in cell–cell junctions. The low mobile fraction rendered it impossible to determine a reliable t1/2 for the proteins due to potential interference from rapidly reversible photobleaching seen in fixed cells (neonGreen, t1/2=11.3 s; EGFP, t1/2=13.1 s; Table S1).
A significant decrease of the mobile fractions was also observed for DP and Pkp2a (Fig. 3C; Table S1). DP showed the greatest decrease in its mobile fraction, from 50.2% to 30.0%, which was concomitant with a reduction in the t1/2. Although Pkp2a also showed decreased mobile fraction under hyper-adhesive conditions, it retained a high Fm of 58.8%. This was at least twice the level of mobile fractions of the other desmosomal proteins and was particularly striking when line profiles of the bleached regions were compared (Fig. 3A). These results were confirmed using FLAP (Fig. 3D; Fig. S1D). The mobile fractions of both Dsc2a and PG tagged with PaGFP were significantly reduced after 3 days of confluent culture, but Pkp2a retained a considerably higher turnover.
These results demonstrate a substantial decrease in mobility of the desmosomal components accompanying maturation to hyper-adhesion. Furthermore, they support the idea of a stable desmosomal moiety in which DP joins the module formed by DCs and PG at earlier stages in MDCK cells. Pkp2a, however, has a novel and surprisingly dynamic role.
Intact desmosomes are internalised during cell scattering
The existence of a stable moiety of core desmosomal components, even where the binding of the DCs was Ca2+ dependent, led us to question whether this would be maintained during desmosome downregulation, which is required for morphogenesis, wound healing, tissue remodelling and EMT. Alternatively, would desmosomal components become highly mobile leading to desmosome dissolution? In order to create a culture model of desmosome downregulation that did not require Ca2+ chelation, we used HGF-induced scattering of MDCK cells, which leads to phenotypic changes resembling partial EMT, including increased cell motility and loss of cell–cell adhesion (Balkovetz, 1998; Ridley et al., 1995).
Desmosomes were tracked using real-time imaging of mixed populations of MDCK cells expressing either Dsc2a–YFP or Dsg2-mCherry. Thus, where two differentially labelled cells were in contact, desmosomes were labelled with Dsc2a–YFP (green) on one side and Dsg–mCherry (magenta) on the other (Fig. 4A,B). To test the validity of this system, we first treated such mixed populations with Ca2+ chelation, which has been well documented to induce splitting of desmosomal adhesion and internalisation of half desmosomes. Imaging showed the extracellular separation of the desmosomes into two half desmosomes that were subsequently internalised by their own cell, as indicated by the entry of solely magenta or solely green particles into the cells (Fig. 4A,C). These observations are consistent with the results of previous Ca2+ chelation experiments in which half desmosomes were seen to be internalised (Mattey and Garrod, 1986; McHarg et al., 2014).
In order to monitor the downregulation of desmosomes in the presence of a physiological Ca2+ concentration, such mixed cell populations were grown on fibronectin-coated glass and induced to scatter by treatment with HGF following serum starvation. Under these conditions, cells growing in small islets started spreading and single cells detached from their neighbours (Fig. S2A). Treatment with HGF for 6 h resulted in a significantly greater number of intracytoplasmic desmosome particles (52.2±4.1 cytoplasmic desmosomes per cell; mean±s.e.m.) in comparison to untreated cells (1.9±0.3 cytoplasmic desmosomes per cell) (Fig. S2B). Clustering of internalised desmosomes was observed (Fig. S2A), so the number of internalised desmosomes is likely to be an underestimate. High magnification fluorescence imaging at 5-second intervals revealed that desmosomes remained stable at the sites where cells maintained contact. Surprisingly, during the final stages of cell separation, whole desmosomes were rapidly internalised by one or other of the pair of adjacent cells (Fig. 4B,C; Movie 1), with the entire desmosomal structure becoming engulfed, including cytoplasmic fragments from both cells (i.e. fluorescence signals of Dsc2a–YFP and Dsg2–mCherry). There was no preference for whether the Dsc2a–YFP- or the Dsg2–mCherry-expressing cell internalised the whole desmosomes (Fig. 4D). Quantification of the colocalisation of fluorescent particles after internalisation revealed that concomitant internalisation of the Dsg and Dsc complex from opposing cells was the dominant mode of internalisation following HGF treatment (Fig. 4C). Immunostaining of unpermeabilised HGF-treated cells with an antibody specific for the extracellular domain of Dsc2 and Dsc3 failed to detect internalised desmosomes, indicating that they were fully internalised and no longer had continuity with the cell surface (Fig. S2C,D). These results implied that HGF-induced scattering led to the internalisation of whole desmosomes, rather than separation of the desmosomal adhesion receptors of neighbouring cells or the complete dissolution of desmosomes.
To confirm this hypothesis, we performed electron microscopy (EM). We used correlative light and electron microscopy (CLEM) to identify cells with internalised desmosomes (Fig. S2E). Subsequent ultrastructural analysis of the EM micrographs showed that the internalised, double-labelled structures were indeed morphologically intact desmosomes with attached keratin filaments and remains of the plasma membrane (Fig. 4G). Comparing the width of desmosome plaques as a readout for their intactness showed only a slight reduction from those of intact desmosomes at the membrane (Fig. 4F). The great majority (>90%) of desmosomes internalised following HGF treatment were internalised whole, with a small number of structures identified as half desmosomes (Fig. 4E; Fig. S2F). We conclude that desmosomes are predominately mechanically removed without undergoing any form of dissolution. Rather, they appear to be torn away from one cell and internalised by its neighbour.
Internalised desmosomes include all major desmosome proteins as well as keratin filament fragments
To determine which of the multiple components remained associated with internalised desmosomes after cell separation, we examined their composition using fluorescence super-resolution microscopy following immunostaining for major desmosomal proteins. Structured illumination microscopy (SIM) and colocalisation analysis revealed that the majority of internalised desmosomes maintained their protein composition following internalisation (Fig. 5A,B). The high resolution of SIM clearly revealed the two distinct patches of DP in the intact desmosomes (Fig. 5A, lower left panel). This morphology, referred to as ‘railroad tracks’ (Stahley et al., 2016), was also observed post internalisation (Fig. 5A). Analysis based on images taken using wide-field microscopy showed internalised Dsc2a–YFP colocalised with DP (81% of all internalisation events), PG (91%), Pkp1 (72%) and keratin 8 (KRT8; 88%) (Fig. 5B). These results support the interpretation that the internalised desmosomes were intact, both at the morphological level and in terms of protein composition.
Electron micrographs showed intermediate filaments attached on both sides (plaques) of the internalised desmosomes (Fig. 4G). SIM analysis also suggested that internalised desmosomes were still attached to keratin filaments after internalisation (Fig. 5A, lower right panel). Some of the high-resolution images suggested that keratin filaments were severed at one side of the desmosome originating from the neighbouring cell (Fig. 5C). To examine this possibility further, we carried out live-cell imaging on cells stably expressing Dsc2a–YFP that were transiently co-transfected with mCherry–KRT18. This resulted in all cells possessing YFP-labelled desmosomes but only a subset showing mCherry-labelled keratin filaments. Where pairs of mCherry–KRT18-expressing cells and untransfected cells were tracked through HGF-induced scattering, co-internalisation of mCherry–KRT18 with Dsc2a–YFP into the non-transfected cell was observed (Fig. 5D; Movie 2). These results are consistent with the view that the force involved in cell separation results in rupture or severing of the keratin filaments in one of the opposing cells, rather than pulling the desmosome plaques apart from each other. This reinforces the picture of desmosomes as extremely stable cell–cell attachments.
Whole desmosome internalisation is dependent on actin
The internalisation of half desmosomes following chelation of extracellular Ca2+ has been shown to be dependent on actin (McHarg et al., 2014). To investigate the role of the actomyosin network in the internalisation of whole desmosomes in our model, MDCK cells were pre-treated for 1 h with the p160ROCK (also known as ROCK1) inhibitor Y-27632 (50 µM), followed by co-treatment with HGF and the inhibitor. Y-27632 treatment attenuated the spreading of the cells and the internalisation of desmosomes (Fig. 6A,B; Fig. S2B). Furthermore, staining for actin and myosin regulatory light chain 2 (MYL12B) as an indicator for active non-muscle myosin-2 revealed a high colocalisation of stress fibres with DP at the last points of contact before desmosomes were internalised when treated with HGF alone (Fig. 6A,C) (Heissler and Manstein, 2013). Treatments with the myosin II inhibitor blebbistatin (50 µM) also debilitated the ability of cells to scatter and the internalisation of desmosomes (Fig. 6B; Fig. S2B). These results suggest that actomyosin, and thus force, is required for internalisation of whole desmosomes during cell scattering.
Desmosome protein mobilities remain consistent after internalisation
Since desmosomes remained structurally intact following internalisation, we next asked whether internalisation changed the stability or the dynamic behaviour of the individual desmosomal components. We therefore performed FLAP analysis of Dsc2a as part of the stable desmosomal moiety and the putative signalling component Pkp2a in internalised desmosomes after 4 h of HGF treatment. Both Dsc2a and Pkp2a showed a comparable FLAP profile following HGF treatment to the phosphate-buffered saline (PBS)-treated control (Fig. 7A–C). The results indicate that the functional structure of desmosomes, encompassing both the stable moiety and the transiently interacting Pkp2a, remains consistent after internalisation. It remains to be determined whether these internalised structures therefore perform any further function within the cells.
Our major finding is that desmosomes consist of two contrasting protein moieties or modules. These are: (1) a strikingly stable module composed of the DCs and PG, at the Ca2+-dependent stage, and (2) a high-mobility module, represented by Pkp2a, which exhibits a transient interaction with the desmosome core and a fast turnover between its desmosome-associated pool and a cytoplasmic pool. Additionally, under these conditions DP in CaCo-2 cells can already be counted as part of the stable moiety. Both modules become less dynamic as desmosomes mature from Ca2+ dependence to hyper-adhesion. While DP transits to behave more like a stable component at this stage in MDCK cells, Pkp2a in both MDCK and CaCo-2 cells retains high molecular mobility. This modular composition of desmosomes is maintained even during growth factor-induced epithelial cell separation when whole double-plaque-bearing desmosomes are internalised by the scattering MDCK cells (Fig. 8). We further demonstrate that these internalised desmosomes are fully intact and even retain attached, apparently torn keratin filaments from the formerly adjacent cell. This suggests that force is involved during the loss of cell–cell adhesion and cell separation. Inhibition of actomyosin attenuated both cell scattering and the internalisation of desmosomes. It appears that the cells are literally torn apart by the force generated by their scattering movement.
FRAP and FLAP allow detailed analysis of the integration of proteins in subcellular compartments (Ishikawa-Ankerhold et al., 2012). Our study has revealed a diverse profile of protein dynamics of components within desmosomes that are engaged in cell–cell adhesion, the most striking observation being that Pkp2a is highly dynamic throughout desmosome maturation and downregulation, whereas the DC adhesion receptors are exceptionally stable. This finding is even more surprising because ultrastructural studies have shown that Pkp is located in the outermost region of the outer dense desmosomal plaque (Al-Amoudi et al., 2007; North et al., 1999). Our results suggest that molecular exchange can take place readily within this apparently dense and well-organised structure. For integrin-mediated cell–matrix adhesions (focal adhesions), it has previously been shown that structural proteins that directly link adhesion receptors to the contractile actin cytoskeleton are less mobile than proteins controlling fast signalling processes (e.g. GTPase activity; Stutchbury et al., 2017). By analogy, the interpretation of our data in this study may lead to a model whereby the DCs, PG and DP form the main stable axis to the intermediate filament network and that Pkps can act as the major signalling components distributing information derived from desmosomes to other compartments of the cell. Such model would be in line with the multiple ways that Pkps are involved in a variety of signalling events (Hatzfeld, 2007; Hatzfeld et al., 2014). Two examples that might particularly benefit from fast translocation of Pkp2 to different compartments are the reported roles of Pkp2 in regulating RhoA localisation, thus mediating signals that also affect the actin cytoskeleton, and in transcriptional regulation in the nucleus (Cerrone et al., 2017; Godsel et al., 2010; Mertens et al., 1996; Sobolik-Delmaire et al., 2010). There is increasing evidence that both roles occur in collaboration with components of adherens junctions and the fast-cycling Pkp2, potentially through competitive binding with β-catenin for E-cadherin (Chen et al., 2002).
In order to compare protein dynamics, we have focused on one representative candidate of each desmosomal protein family. However, tissue-specific isoforms of DCs (Dsg1–Dsg4, Dsc1–Dsc3) and Pkp (Pkp1–Pkp3) have been identified and their dynamic properties within desmosomes may differ. In line with our findings, another DC, Dsg3, has been shown to be exceptionally stable in mouse keratinocytes (Vielmuth et al., 2018). The high mobile fraction of Pkp2a in our study also seems to apply to Pkp3, which is present in most simple and stratified epithelia and has been shown to have a mobile fraction of ∼60% in mouse keratinocytes (Keil et al., 2016; Schmidt and Jager, 2005). Pkp1 is more closely related to Pkp2 and similarly exhibits a dual localisation within the karyoplasm and desmosomes although with an expression pattern that is mostly restricted to complex and suprabasal layers of stratified epithelia and thus is reciprocal to that of Pkp2 (Schmidt and Jager, 2005). The mobility of Pkp1 has been shown to be dependent on phosphorylation through AKT2 and appears to differ in a cell-type-dependent manner, having a high mobile fraction of 60% in MCF-7 cells and being considerably more stable (Fm of ∼30%) in mouse keratinocytes (Keil et al., 2016; Wolf et al., 2013). It still remains to be clarified what signalling pathways regulate Pkp dynamics and vice versa. Nevertheless, these previous findings reinforce the dualistic model of desmosome dynamics where Pkps are part of a desmosomal signalling module.
The mobility of all tested proteins significantly decreased during maturation from Ca2+-dependent desmosomes to Ca2+-independent desmosomes that had adopted the hyper-adhesion state in MDCK cells. The already low mobile fractions almost halved for the DCs, PG and DP, and decreased by 15% for Pkp2a. A similar result was found recently when hyper-adhesion was pharmacologically induced using Gö6976, an inhibitor of conventional PKC isoforms, although Pkp was not included in that study (Bartle et al., 2020). The disadvantage of using such an inhibitor is that it might block multiple cellular events, any one of which may directly or indirectly affect the mobility of desmosomal components. Now that we have studied simple time-dependent maturation of desmosomes, we can be fully confident that stabilisation of desmosomal components is indeed a consequence of hyper-adhesion.
Of particular interest is the decrease in mobility of DP that accompanies desmosome maturation in MDCK cells. During the Ca2+-dependent phase, we found DP to be substantially more mobile than the stable moiety represented by the DCs and PG, but upon the onset of hyper-adhesion the mobility of DP was reduced to the extent that it could now also be considered part of the stable moiety. We suggest that this stabilisation of DP makes a significant contribution to the gaining of hyper-adhesion. This view is consistent with data showing that expression of a phospho-null DP mutant with enhanced keratin-binding properties leads to an increased association of the desmosomal plaque with keratin filaments, resulting in decreased protein exchange and increased desmosome stability (Albrecht et al., 2015; Bartle et al., 2020). It has been proposed that inhibition of PKC inhibitors or downregulation of PKCα may promote DP–keratin integration, leading to the onset of hyper-adhesion (Bartle et al., 2020). Furthermore, during formation of the desmosome–keratin scaffold the development of radial and inter-desmosomal keratin filaments coincides with a decrease in DP mobility (Moch et al., 2020). Considering that the mobility of DP was lower in CaCo-2 cells than in MDCK cells, resembling the mobility reported by Bartle et al. in HaCaT cells (Bartle et al., 2020), it is interesting to speculate that the dynamics of DP are cell-type specific and just one factor driving the onset of hyper-adhesion.
There is also evidence for alternative ways of influencing the mobility of desmosomal proteins, one deriving from crosstalk with the actin cytoskeleton. For example, the knockout of α-adducin, a protein involved in the assembly of a plasma membrane-stabilising cortical spectrin–actin network, increased Dsg3 mobility in the membrane (Hiermaier et al., 2021). Adducin is a PKC substrate (Larsson, 2006), as are many of the actin regulatory proteins. Furthermore, actin-mediated endocytosis is regulated by PKCα and it can be blocked by the inhibitor Gö6976 (Hryciw et al., 2005), as can the actin-mediated internalisation of half desmosomes (Holm et al., 1993). It has also been shown that complete absence of PKCα in mice appears to stabilise hyper-adhesion of desmosomes in epidermal wounds (Thomason et al., 2012). It is thus highly probable that PKCα regulates the mobility and the life cycle of the desmosomal complex at multiple levels, even though it is not required for desmosome assembly (Thomason et al., 2012). Further experiments are needed to determine the precise role of PKCα in the life cycle of desmosomes.
Another possible reason for such decrease in mobility of desmosomal components could be the progressive packing of cadherins into a higher-order array that locks DCs in a Ca2+-independent adhesion state; this increasing order undoubtedly occurs during the onset of hyper-adhesion (Garrod et al., 2005; Kimura et al., 2012). This suggestion was disputed by Bartels and colleagues, who found that the ordered packing of Dsg3 with a genetically modified extracellular domain was lost upon chelation of Ca2+ from pharmacologically induced hyper-adhesive desmosomes (Bartle et al., 2020). However, the modified Dsg3 may not reflect the behaviour of wild-type DCs, since it is difficult to understand how adhesion between opposed desmosomal halves can be maintained in the presence of chelating agents, the defining criterion of hyper-adhesion, if the ordered array of their extracellular domains is lost (Garrod, 2013).
Irrespective of their maturation state, desmosomes form extremely strong intercellular bonds. Our data demonstrate that cells appear to move apart by tearing part of the desmosome–intermediate filament complex out from the neighbouring cell, resulting in the internalisation of the desmosomes as a whole unit with associated keratin filaments. This observation was surprising, as others have claimed that cell scattering involves desmosome dissociation or junctional splitting together with partial desmosome disassembly (Boyer et al., 1989; Savagner et al., 1997). We found no evidence for such mechanisms of desmosome dismantling in cells undergoing scattering. The forceful internalisation of the whole complex by one of the attached cells recalls the observation of whole desmosome internalisation in vivo in wound edge keratinocytes and certain carcinomas (Allen and Potten, 1975; Garrod et al., 2005; Schenk, 1975). It is striking that once internalised, desmosomes remained intact as judged by both their ultrastructure and protein composition. The few observed internalised half desmosomes resembled those described previously by Demlehner et al. and Duden and Franke in cells in low Ca2+ medium or uncoupled cells and might represent the internalisation of newly assembled half desmosomes from the membrane (Demlehner et al., 1995; Duden and Franke, 1988). Alternatively, they could represent a small fraction of desmosomes that split in half prior to internalisation (Fig. S2G, model 2). However, our colocalisation experiments and EM analysis revealed that the internalisation of entire desmosomes is by far the dominant mode (>90%) of desmosome internalisation (Fig. 8; Fig. S2G, model 1). Over longer periods after internalisation we have previously shown that components of engulfed desmosomes remain associated and are not recycled, but rather undergo lysosomal and proteasomal degradation (McHarg et al., 2014).
We note that a similar mode of internalisation to that described here – the internalisation of junctions by one member of a pair of cells – has also been reported for tight junctions and gap junctions (Jordan et al., 2001; Matsuda et al., 2004). However, unlike the internalisation of the whole junctional complex, Matsuda et al. have previously revealed that for tight junctions only the transmembrane proteins, claudin-3 and claudin-1, of apposed membranes are co-endocytosed (Matsuda et al., 2004), whereas other tight junction components seem to dissociate from those claudins prior to their internalisation. Jordan et al. have described a similar co-endocytosis of connexin-based gap junctions, but noted that those internalised gap junctions, referred to as annular junctions, are rare and are unlikely to be the only mechanism to clear gap junctions from the plasma membrane (Jordan et al., 2001).
In their study of scattering by NBT II cells, Boyer et al. noted that loss of cell–cell contact is accompanied by the appearance within the cytoplasm of dots containing desmosomal components and associated keratin filaments (Boyer et al., 1989). We suggest that these dots are equivalent to the whole internalised desmosomes that we report here. We have further shown that some of these filaments are contributed by the opposing cell of an adjacent pair, indicating that the filaments have been fragmented during cell separation. If we take the view that cell separation occurs by tearing under the force generated by cell movement, this would suggest that the keratin filaments have simply broken under the applied force. Our data show that stress fibres accumulate at sites of desmosomes after HGF treatment, and the presence of actomyosin indicates that these sites are under tension. This enrichment of actomyosin at desmosomes disappeared in the presence of actomyosin inhibitors, and the cells failed to separate. These experiments suggest that forces are critical for cell separation at the sites of desmosomes. Alternatively, we speculate that a signalling event such as phosphorylation is involved in severing the keratin filaments and that such a signalling process would therefore have a facilitating role in cell separation. This would be analogous to the rapid and specific phosphorylation events that sever cytoskeleton components for daughter cell separation at the end of cytokinesis and during meiotic maturation (Goto et al., 2000; Klymkowsky et al., 1991). Our data do not specifically address the severing mechanism, but, given the particularly high tensile strength of keratins (Qin et al., 2010), it seems likely that the breakage involves additional processes.
In conclusion, our data presented here lead to a new concept of dualistic desmosome organisation. There appear to be two protein moieties or modules within the junctions: a stable moiety, which is presumably responsible for the strong adhesive function of the desmosome, and a much more mobile protein moiety, exemplified by Pkp2a, which has the potential for signalling and regulatory function. This behavioural dichotomy is both present and detectable when desmosomes are relatively newly formed and Ca2+ dependent, and also when they mature to Ca2+ independence. This dualism persists even when desmosomes are being downregulated and removed.
MATERIALS AND METHODS
Cell lines and transfection
Madin–Darby canine kidney II cells (MDCK; ECACC) (Madin and Darby, 1958) and the human colon adenocarcinoma cell line CaCo-2 (CaCo-2; ATCC, kindly provided by I. Roberts, University of Manchester, Manchester, UK) (Fogh et al., 1977) were cultured at 37°C in 5% humidified CO2 in high glucose Dulbecco's Modified Eagle Medium (DMEM; Sigma or GE Healthcare), supplemented with 10% (v/v) foetal calf serum (FCS; Gibco or GE Healthcare) and 100 U/ml penicillin and 100 µg/ml streptomycin (P/S; Gibco), herein referred to as standard medium (SM) with a Ca2+ concentration of ∼1.8 mM. Cells were transfected using Lipofectamine LTX transfection reagent, according to the manufacturer's instructions (Invitrogen). To generate stable cell lines, transfected MDCK cells were selected using 2 µg/ml puromycin (Thermofisher) in SM with medium changes every 2 days for 10 days. MDCK cells stably expressing Dsc2a–YFP were kindly provided by R. E. Leube (University Hospital RWTH, Aachen, Germany; Windoffer et al., 2002). CaCo-2 cells were transiently transfected using Lipofectamine LTX, plated on glass-bottom dishes (MaTek) and imaged 24 h post transfection. For live-cell imaging experiments, MDCK cells were plated on glass-bottom dishes (MatTek or Ibidi). For cell scattering experiments using HGF, the glass-bottom dishes were coated with 10 µg/ml bovine plasma fibronectin (FN; Sigma) diluted in PBS.
Cloning and constructs
DP–eGFP (Addgene #32227), Dsg2–mCherry (Addgene #36991) and mCherry–KRT18 (Addgene #55065) were purchased from Addgene. mEmerald–E-cadherin was obtained from the Michael Davidson collection (#54072 from P. Kanchanawong, Mechanobiology Institute, Singapore). Dsc2a–YFP was a kind gift from R. E. Leube (University Hospital RWTH, Aachen, Germany). To generate constructs containing PaGFP, neonGreen and mScarlet, we used plasmids where the desmosome genes were already cloned into a custom-made vector by Oxford Genetics where the vector pSF (#OG394R1) was modified to have an EF1a promoter and puromycin selection marker. The vector was linearised by restriction digestion followed by gel purification (see Table S2 for restriction enzymes used), and the purified vector was used to clone fragments containing the ORFs of PaGFP, neonGreen and mScarlet, these fragments were amplified by polymerase chain reaction using Phusion High-Fidelity DNA Polymerase (M0530L, New England Biolabs) using 35 cycles with an annealing temperature of 60°C (see Table S2 for list of primers) and cloned using a Gibson Assembly Cloning kit (#EE5510S, New England Biolabs). All primers were designed using SnapGene (GSL Biotech LLC, Chicago, IL) and were synthesised by Eurofins Genomics (Germany). All ORF sequences were confirmed by Sanger sequencing performed by Eurofins.
Antibodies and reagents
The following antibodies were used at the indicated dilutions [in 1% BSA in PBS (Sigma)]: anti-desmoplakin I and II (clone 11-5F, custom-made by D.R.G.; Parrish et al., 1987), 1:400; anti-plakoglobin (clone 15F11, Sigma), 1:100; anti-keratin 8 (clone LE41, custom-made by E.B.L.; Lane, 1982), used as hybridoma culture supernatant; anti-desmocollin 2 and 3 (clone 7G6, Zymed Laboratories), 1:100; anti-plakophilin 1 (clone PP1-5C2, Abnova), 1:50; anti-MYL12B (also known as MRLC2; clone EPR9331, Abcam), 1:100. Secondary antibodies conjugated to Alexa Fluor 488, 594 or 647 were all from Thermofisher (used at 1:500). Alexa Fluor 488- (1:500), Texas-Red-X- (1:500) and Alexa Fluor 647-conjugated (1:200) phalloidin were from Life Technologies. DAPI readymade solution (Sigma) was used at a concentration of 1 µg/ml. Y-27632 dihydrochloride (Tocris Bioscience) was dissolved in water and used at a final concentration of 50 μM. Blebbistatin (Tocris Bioscience) was diluted in DMSO (Sigma) and used at a final concentration of 50 µM.
Ca2+ switch assay
MDCK cells were cultured at confluent density (1.35×105 cells/cm2) in SM for 1 day or 3 days. They were washed thrice with calcium- and magnesium-free PBS (Gibco) and incubated with calcium-free DMEM (Gibco) supplemented with 10% chelated FBS and 3 mM EGTA (Sigma-Aldrich), herein referred to as ‘+EGTA’ for 90 min (unless specified otherwise) at 37°C in 5% humidified CO2. Ca2+ sensitivity of desmosomes was quantified as previously described (Wallis et al., 2000). In brief, following Ca2+ chelation, the cells were fixed in ice-cold methanol for 10 min and immunostained for detection of desmoplakin by immunofluorescence. Cells that remained attached by desmoplakin-positive projections were scored as having Ca2+-independent desmosomes, and the number of these cells was expressed proportional to the total cell number.
Cell scattering assay
Transfected MDCK cells were plated at 1.5×103 cells/cm2 on FN-coated glass-bottom dishes and cultured in SM for 2 days at 37°C. The cells were serum starved for 12 h in DMEM supplemented with 1% P/S (serum-free medium, SFM). Cells were treated with 40 ng/ml human recombinant hepatocyte growth factor (HGF; STEMCELL Technologies) in SFM for 4 h at 37°C prior to live-cell imaging or 6 h prior to fixation and staining. Scattering was assessed by immunostaining for desmoplakin and quantified by scoring cells that remained attached with desmosomes to only two neighbouring cells. The number of these scattering cells was expressed proportional to the total number of cells in the cell islets.
Wild-type MDCK cells were cultured in the same manner as for the cell scattering assay, but prior to treatment with HGF the cells were pre-treated for 1 h with SFM supplemented with either Y-27632 (50 µM) or blebbistatin (50 µM). HGF at a final concentration of 40 ng/ml was added to the drug treatment and cells were cultured for 6 h prior to fixation and immunostaining for DP, actin and MYL12B.
Colocalisation experiments were carried out using a Delta Vision microscope (Applied Precision) with a 60×/1.42 Plan Apo N (Oil) objective and a Sedat Quad filter set, with images collected using a Retiga R6 (Q-Imaging) camera.
3D-SIM was performed using a DeltaVision OMX Version 4 Blaze microscope (GE Healthcare), equipped with 405-, 488-, and 568-nm lasers and a BGR filter drawer. A 100×/1.40 PSF Plan Apo Oil objective and liquid-cooled Photometrics Evolve EM charge-coupled device camera for each channel were used. For 3D-SIM, 15 images per section per channel were acquired (made up of three rotations and five phase movements of the diffraction grating) at a z-spacing of 0.125 µm. Structured illumination reconstruction and alignment were carried out using the SoftWoRx (GE Healthcare) program.
FLAP, FRAP and HGF-induced scattering experimental data of MDCK cells was collected with an inverted Nikon Eclipse Ti microscope equipped with a 100×/NA 1.40 Plan Apo oil immersion objective lens, a focus drift correction system, a piezo-motorised stage, a 37°C on-stage incubation system (LCI), 100 mW diode lasers (405 nm, 491 nm and 561 nm), an EMCCD camera (Evolve 512; Photometrics), CSU-22 spinning disc scan head (Yokogawa) and a 3D FRAP system (iLAS2 Roper Scientific). The microscope system was controlled using MetaMorph (Molecular Devices) and iLAS2 (Roper Scientific) software.
FLAP and FRAP experimental data of CaCo-2 cells was acquired using a CSU-X1 spinning disc confocal (Yokagowa) on a Zeiss Axio-Observer Z1 microscope with a 60×/1.40 Plan-Apochromat objective, Evolve EMCCD camera (Photometrics) and motorised XYZ stage (ASI). The 488 nm, 561 nm and 405 nm lasers were controlled using an acousto-optic tunable filter (AOTF) through the laser stack [Intelligent Imaging Innovations (3I)] allowing both rapid ‘shuttering’ of the laser and attenuation of the laser power. The microscope system was controlled using Slidebook software (3I).
Cells were fixed with 4% (w/v) paraformaldehyde in PBS or with 100% ice-cold methanol for anti-desmoplakin (clone 11-5F) immunostaining. Antibodies were diluted in 1% BSA and added to the cells for 1 h. Images were acquired on the Delta Vision systems (above) and processed using the FIJI ImageJ software (version 1.53 g; https://fiji.sc/). To analyse colocalisation, the background was subtracted using a rolling ball algorithm, thresholding was performed using the triangle algorithm in ImageJ based on the Dsc2a–YFP intensity, and the particle analysis function of ImageJ was used to identify Dsc2a–YFP particles in a ROI. These particles were manually assessed for colocalisation with the protein of interest using the RGB profiler function of ImageJ. Results were calculated as percentages of colocalising events. To quantify cytoplasmic desmosomes, cells were first immunostained for DP. Images were thresholded using the triangle algorithm in ImageJ, and particles of a minimum size of 20 nm2 were measured within the cytoplasm ≥2 µm away from the membrane.
MDCK cells stably expressing neonGreen, mEmerald or eGFP constructs were seeded at confluent density (1.35×105 cells/cm2) on uncoated glass-bottom dishes in SM and cultured for 24 h for Ca2+-dependent desmosomes or for 3 days for desmosomes to acquire hyper-adhesion before imaging. CaCo-2 cells were imaged 24 h following transfection. Images of MDCK cells were acquired using the Nikon Eclipse Ti system (see above) and those of CaCo-2 cells were acquired using the Zeiss Axio-Observer Z1 system (see above). For FRAP quantification, five circular ROIs of 2 µm diameter at the cell–cell junctions were selected and photobleached with a 10 ms burst of the 488 nm laser at 100% power. iLAS2 or Slidebook software was used to capture three images prior to photobleaching and one image every 10 s for 5 min post bleaching. Movies were analysed using Fiji (version 1.53 g) by manually tracking the ROI and measuring the fluorescence signal. The subsequent analysis was performed as described previously (Carisey et al., 2011). In brief, intensities of bleached ROIs and three control unbleached desmosomal ROIs were manually measured using ImageJ. Values were background subtracted, and measurements were corrected with the control values of unbleached ROIs to compensate for any overall loss of fluorescence. The values were then normalised to the intensity of the first postbleach value. Graphs were prepared using Prism 8 (GraphPad).
MDCK cells co-expressing the required PaGFP-tagged constructs and a mScarlet-tagged desmosome marker protein (Dsc2a–mScarlet unless otherwise specified) were plated at confluent density (1.35×105 cells/cm2) on uncoated glass-bottom dishes and cultured for 24 h or 3 days before imaging. To measure protein dynamics following cell scattering, imaging started 4 h post HGF treatment, local sites of cell separation were observed and ROIs of internalised desmosomes were selected. CaCo-2 cells were imaged 24 h following transfection. Images of MDCK cells were acquired using the Nikon Eclipse Ti system (see above) and those of CaCo-2 cells were acquired using the Zeiss Axio-Observer Z1 system (see above). Similarly to FRAP, five circular ROIs of 2 µm diameter were selected and photoactivated using a 405 nm laser at 100% power for 10 ms. iLAS2 or Slidebook software was used to capture three images prior and one image every 10 s for 5 min–1 h post photoactivation. Movies were analysed using Fiji (version 1.53 g). The intensities of the post-activated PaGFP ROIs were measured manually using ImageJ. Values were normalised to the intensity of the first post-activation image. Graphs were prepared using Prism 8 (GraphPad).
MDCK cells stably expressing Dsc2a–YFP were co-cultured with MDCK cells transiently expressing Dsg2–mCherry on a gridded glass-bottom dish (MatTek). Cells were treated with HGF (see above). Live-cell imaging of ROIs was carried out using the Nikon Ti Eclipse microscope (see above). Cells were grown on glass-bottom dishes with a gridded coverslip (MatTek Corporation, P35G-1.5-14-CGRD). After fluorescence images of cells co-expressing Dsc2a–YFP and Dsg2–mCherry were acquired, brightfield (phase-contrast) images of cells and the finder grid index were captured at lower magnification for ROI reference. Cells in dishes were then fixed with 2% glutaraldehyde and 4% formaldehyde in 0.1 M cacodylate buffer, post-fixed in 1% osmium tetroxide, dehydrated in ethanol and embedded in epoxy resin (EPON 812, SERVA). Upon resin polymerisation the coverslip was removed using liquid nitrogen (leaving the finder grid embossed on the block surface) and the ROIs were re-localised using a stereomicroscope. Target cells within each ROI were cut horizontally with an ultramicrotome (LEICA Ultracut-UC7), and the resulting ultrathin sections were collected on formvar/carbon-coated slot grids (Ted Pella Inc., 01805-F) and post-contrasted with uranyl acetate and lead citrate. Sections were analysed under a JEOL JEM-1010 transmission electron microscope operated at 80 kV, and images were acquired with an SIA 12C CCD camera. Desmosome width was measured by taking the average of three line measurements per desmosome using ImageJ. A total of 10 desmosomes from each of 10 cells (N=3) at the membrane and following internalisation were measured.
Graphing and statistical analysis were performed using GraphPad Prism 8 and 9 software. When comparing means, the D'Agostino–Pearson test was used to assess the normality of the data to determine the appropriate statistical tests to use.
We thank the staff of the Bioimaging facility at the University of Manchester and the A*STAR Microscopy Platform for their help with imaging and analysis. The Ballestrem laboratory is part of the Wellcome Trust Centre for Cell-Matrix Research, University of Manchester, which is supported by core funding from the Wellcome Trust (grant number 203128/Z/16/Z). The work in Singapore was supported by institutional core funding to the Skin Research Institute of Singapore from the Biomedical Research Council of Singapore.
Conceptualization: J.B.F., H.H., E.B.L., D.R.G., C.B.; Methodology: J.B.F., H.H.; Software: J.B.F.; Validation: J.B.F., H.H., D.L., J.L., R.A.d.A., B.Y.; Formal analysis: J.B.F., H.H., D.L.; Investigation: J.B.F., H.H., D.L., J.L.; Resources: R.A.d.A., B.Y.; Data curation: J.B.F., H.H., D.L., J.L.; Writing - original draft: J.B.F., E.B.L., D.R.G., C.B.; Writing - review & editing: J.B.F., D.L., E.B.L., D.R.G., C.B.; Visualization: J.B.F., H.H., D.L.; Supervision: G.D.W., E.B.L., D.R.G., C.B.; Project administration: E.B.L., D.R.G., C.B.; Funding acquisition: E.B.L., D.R.G., C.B.
This work was supported by the Faculty of Biology, Medicine and Health at the University of Manchester, UK and by the Agency for Science, Technology and Research (A*STAR), Singapore. C.B. acknowledges the Biotechnology and Biological Sciences Research Council (BBSRC; grant numbers BB/R001707/1, BB/R014361/1) and the Wellcome Trust (grant number 202923/Z/16/Z) for funding of this project. J.B.F. is supported by the Faculty of Biology, Medicine and Health at the University of Manchester and by the Agency of Science Technology and Research (A*STAR). Deposited in PMC for release after 6 months.
The authors declare no competing or financial interests.