Epithelial cells assemble into three-dimensional aggregates to generate lumen-containing organ substructures. Cells therein contact the extracellular matrix with their basal surface, neighbouring cells with their contact surface and the lumen with their apical surface. We investigated the development of single MDCK cells into aggregates with lumen using quantitative live-cell imaging to identify morphogenetic rules for lumen formation. In two-cell aggregates, membrane insertion into the contact surface established a preapical patch (PAP) characterized by the presence of the apical marker gp135, microvilli and the absence of E-cadherin. This PAP originated from a compartment that had hallmarks of an apical recycling endosome, and matured through Brefeldin-A-sensitive membrane trafficking and the establishment of tight junctions around itself. As a result of the activity of water and ion channels, an optically resolvable lumen formed. Initially, this lumen enlarged without changes in aggregate volume or cell number but with decreasing cell volumes. Additionally, the ROCK1/2-myosin-II pathway counteracted PAP and lumen formation. Thus, lumen formation results from PAP establishment, PAP maturation, lumen initiation and lumen enlargement. These phases correlate with distinct cell surface and volume patterns, which suggests that such morphometric parameters are regulated by trafficking, ROCK-mediated contractility and hydrostatic pressure or vice versa.
- Epithelial cell polarity
- Cell shape
- Surface surveillance
- 3D cell culture
Organs are elaborate cell communities that perform specialized functions. The cell as their smallest unit plays an important role in shaping organs. Therefore, the ability of a cell to control its own shape and contacts with its neighbours is central to organogenesis. Epithelia developed to serve as a specialized interface between the organism and the outside world. Mammalian organs made of epithelial tissue such as kidney and lung consist of two building blocks, cysts and tubules (Metzger et al., 2008; O'Brien et al., 2002). These are characterized by a central and optically resolvable volume: a lumen, which is surrounded by a monolayer of polarized cells. How is this space generated in the first place?
Understanding lumen formation reflects our understanding of how cells polarize during organogenesis and how regulation of cellular mechanics coordinates cell shape changes. Different ways of lumen formation exist (Lubarsky and Krasnow, 2003; Martin-Belmonte and Mostov, 2008). The lumen can be formed either by changing the shape of previously formed epithelial sheets or de novo (e.g. in breast and kidney development). During the generation of a nephron, the functional unit of a kidney, solid cell aggregates form `comma-shaped' structures, which are composed of epithelial cells enclosing a lumen (Bard et al., 2001). Similarly, MDCK (Madin-Darby canine kidney) cells, can form de novo lumina when cultured in a collagen matrix (McAteer et al., 1987; Wang et al., 1990a) and thereby generate cysts. A cyst is a spherical cell aggregate consisting of a single layer of polarized epithelial cells enclosing a liquid-filled lumen (Kroschewski, 2004; Lin et al., 1999; Martin-Belmonte and Mostov, 2008; O'Brien et al., 2002). These cells expose their apical surface towards the lumen. The contact surface connects to neighbouring cells and the basal surface links the cell to the extracellular matrix (ECM) (Nelson, 2003; Wang et al., 1990b). De novo lumen formation is neither well described nor molecularly understood, despite the fact that the function of some molecules was analyzed in fixed specimens (Martin-Belmonte et al., 2007; Yu et al., 2007). Specifically, it is not known which cellular changes precede and generate the new functional apical surface.
A lumen is an optically resolvable volume surrounded by two or more cells that is generated within a solid cell aggregate. It might be the result of changes in cell shape or in the number or position of cells constituting the cell aggregate. Conceptually, several possibilities exist to generate de novo a lumen via cell shape changes. The plasma membranes at the cell-cell contact surface of adjacent cells might be physically separated through actomyosin-driven contraction. This would tear the adhering membranes apart and/or facilitate membrane traffic to enrich apical properties and remove lateral properties in the membrane region that is converted into an apical surface. Alternatively, luminal expansion might be driven by hydrostatic pressure generated by directional ion transport into the intercellular space. Ion transport might be combined with or substituted by apical secretion of highly glycosylated proteins or hydrophilic matrix. Indeed, apical Cl– transport through the cystic fibrosis transmembrane conductance regulator (CFTR) facilitates cyst growth (Li et al., 2004) and apical secretion of the glycoprotein eyes shut is crucial for lumen formation in the sensory organs of Drosophila (Husain et al., 2006). Concentration of ions and proteins in the luminal space might require the presence of a diffusion barrier sealing the space. In fully polarized MDCK cells, tight junctions border the apical cell surface, forming a selective gate and barrier for both paracellular and lateral membrane diffusion (Matter and Balda, 2003).
Apoptosis or programmed cell death, contributes to lumen clearance primarily in large aggregates (Lin et al., 1999; Martin-Belmonte and Mostov, 2008). Chemical composition and physical rigidity of the ECM also strongly influence lumen formation, as they orient the apical surface opposite to the basal surface and modify the number of cells at which a lumen occurs, respectively (Andriani et al., 2003; Guo et al., 2008; O'Brien et al., 2001; Yu et al., 2007; Zeng et al., 2006). However, the cellular changes contributing to de novo lumen formation are not resolved.
The shape and topology of cells in an aggregate determine the multicellular architecture and can be best described by physical terms. As cysts are spherical and reproducibly generated, we hypothesized that physical parameters, such as surfaces or volumes either guide, or are the result of, cell polarization and morphogenesis. Therefore, we analyzed the morphometric changes and polarization stages in MDCK cell aggregates of two to four cells. We found four distinct morphogenetic steps during lumen formation in the absence of cell number changes and we identified three separate mechanisms. An E-cadherin-free and gp135-positive preapical membrane patch (PAP) was generated in the contact surface causing a separation of the adjacent cell membranes, which was, however, not resolvable optically, because of the presence of interdigitating microvilli. The established PAP matured during its enrichment with channels and the establishment of tight junctions along its border. The subsequent generation of an optically resolvable lumen depended on ion, water channels and tight junction activity and was executed within minutes (lumen initiation). Initially, the lumen enlarged in an aggregate of constant volume (isochoric). Further increase in luminal size was later accompanied by cell volume and consequently aggregate volume increase. We show that membrane trafficking is necessary for PAP establishment and maturation but not for lumen initiation. Furthermore, this trafficking occurs through an organelle with properties of the apical recycling endosome (ARE). Finally, specific inhibition of ROCK1/2-dependent myosin II facilitated PAP formation and lumen initiation. Taken together, MDCK cells generate lumina in a sequential multi-step process using targeted membrane traffic and two force-generation mechanisms – hydrostatic pressure in a barrier-sealed volume and regulation of ROCK-mediated contractility – which act in parallel, to open up a lumen. These mechanisms might either regulate or be regulated by the accompanying morphometric changes.
Apicobasal polarity is established before lumen initiation
Single MDCK cells embedded in collagen form cysts within a few days. Quantitative studies on lumen formation in 3D MDCK aggregates showed that lumina appear with maximal probability in aggregates of two to four cells (Zeng et al., 2006). Thus, we analyzed such aggregates to reveal the cellular changes accompanying the initiation of an optically resolvable lumen.
We aimed to correlate lumen formation with the establishment of apicobasal cell polarity. Gp135, gp114 and CFTR localize in fully polarized epithelial cells to the apical domain (Crawford et al., 1991; Le Bivic et al., 1990; Ojakian and Schwimmer, 1988). E-cadherin and gp58, however, localize to the basolateral domain, which is separated from the apical one by tight junctions (Balcarova-Stander et al., 1984). Therefore, we investigated the distribution of these proteins to elucidate the molecular content of surfaces in small aggregates with and without lumen (Fig. 1, supplementary material Fig. S1).
Gp135, gp114 and CFTR occurred in four localization patterns (Fig. 1, supplementary material Fig. S1). In one third (33±3%) of the aggregates, gp135 localized to a part of the contact surface where cortical F-actin was also present (Fig. 1, supplementary material Fig. S1). This region was depleted of the basolateral markers E-cadherin (Fig. 1A) or gp58 (supplementary material Fig. S1B,C). We defined this gp135-positive membrane domain, surrounded by contact surface and facing another cell, preapical patch (PAP) in analogy to the preapical domain described in 2D MDCK cultures (Meder et al., 2005). Gp114 (Fig. 1B,C) and CFTR (Fig. 1C,D) localized to the PAP but not as restricted as gp135; in the presence of a PAP, these proteins occurred exclusively concentrated in the PAP or also at gp135-negative contact and basal surfaces.
In half of the two-cell aggregates (49±2%), gp135 localized at the basal surface and was excluded from the contact surface where E-cadherin or gp58 localized. In other aggregates, gp135 localized in vesicles and in a bigger cluster of membranes (Fig. 1D, arrowheads) concentrated close to the contact surface in a `compact structure' (13±3%) or in aggregates with lumen at the apical cell surface (5±1%) (Fig. 1A). The majority of gp135-positive vesicles contained CFTR (Fig. 1D), were free of E-cadherin (Fig. 1A) and partially colocalized with F-actin (supplementary material Fig. S1A,D). Gp114 partially colocalized with CFTR in vesicles (Fig. 1C). In all aggregates containing a lumen, gp135 (Fig. 1A,D), gp114 (Fig. 1B,C) and CFTR (Fig. 1C,D) localized as expected to the apical cell surface, which was free of E-cadherin (Fig. 1A,B) or gp58 (supplementary material Fig. S1B,C).
These data show that gp135 in the plasma membrane of cells of solid aggregates localized either at the basal surface or in the PAP. In aggregates with lumen it localized at the apical surface. Therefore, we suggest that gp135 relocalizes from the basal surface to the PAP, which is converted into an apical surface in aggregates with a lumen. Not all established apical markers localized identically at the contact surface. As gp135 was the most strictly localized, we postulate that (1) several routes to the PAP exist and (2) gp135 is the earliest tested marker in the PAP. Importantly, the establishment of a gp135 domain (PAP) within the contact surface is not sufficient to produce an optically resolvable lumen.
Gp135 redistributes to establish a preapical patch
To elucidate the mechanism of PAP establishment, we monitored the localization of mRFP-gp135 and E-cadherin-GFP in two-cell aggregates initially devoid of gp135 in the contact surface (Fig. 2A, supplementary material Fig. S2, Movie 1). Using quantitative 4D microscopy, we visualized the insertion of compact gp135-positive, E-cadherin-negative membranes into the plasma membrane of the contact surface. Gp135-positive membranes entered the plane of the plasma membrane at the same location and time as a gap in the E-cadherin-positive surface appeared (supplementary material Fig. S2, 2 μm Z-slice). This localized insertion of preapical membrane resulted in the establishment of a PAP, which persisted over 480 minutes (Fig. 2A, supplementary material Movie 1). Thus, the gp135-positive compact structure converted into a PAP.
We measured the basal (Sb) and contact (Sc) surfaces during PAP establishment to observe the systemic surface effect (Fig. 2B). Sb did not decrease and Sc did not increase during the time of recording (480 minutes) indicating that a surface surveillance mechanism exists where insertion of preapical membranes is counterbalanced by a reduction of the E-cadherin-positive membrane (supplementary material Fig. S3A).
To obtain high-resolution information about the PAP, we visualized the contact surface in two-cell aggregates by transmission electron microscopy (TEM) and identified two different zones (Fig. 2C, inset). Zone I is characterized by the close juxtaposition of two cell membranes; the absence of a resolvable paracellular space indicates the presence of intercellular adhesion complexes. Correspondingly, regions with electron-dense material in zone I could be detected (supplementary material Fig. S1F). Zone II, surrounded by zone I, is characterized by interdigitating microvilli (Fig. 2C, box 4) and devoid of direct cell-cell contacts. These interdigitating plasma membranes could not be resolved by light microscopy but cause a relatively thick appearance of the signal of fluorescent proteins in the PAP (see for example, Fig. 2C and Fig. 4B). Small vesicles close to the plasma membrane of cell 1 (Fig. 2C, box 1) and vesicles of different sizes concentrated in the subcortical region close to zone II (Fig. 2C, box 2) were detected. Multivesicular bodies (Fig. 2C, box 3) and cytoskeletal filaments (Fig. 2C, box 5) were also in close proximity to zone II. Thus, a PAP is a non-adhesive microvilli-rich surface, which is close to many vesicles.
Where does gp135 in the PAP come from? In two-cell aggregates without lumen, gp135 was detectable in the compact structure and in vesicles (see for example, Fig. 1 and Fig. 2A). The trans-Golgi network marker TGN38, localized in elongated structures (arrowheads) and puncta throughout cells. The Golgi reporter giantin localized predominantly in clusters close to the compact structure. Both makers, which to the best of our knowledge have not been found in polarizing cells, poorly colocalized with the gp135-positive compact structure (Pearson's coefficient 0.37 and 0.38, respectively) (Fig. 3) (Banting and Ponnambalam, 1997; Whyte and Munro, 2002). Thus, Golgi membranes do not significantly contribute to the gp135-positive compact structure. The compact structure is not an early endosomal compartment because its reporter molecule, EEA1, was clearly absent in the compact structure resulting in the low Pearson's coefficient of 0.41 (Fig. 3) (Whyte and Munro, 2002). However, EEA1 colocalized extensively with gp135 in peripheral and central vesicles (iso-colocalization signal in Fig. 3). In the compact structure, gp135 colocalized strongest with recycling endosomal markers, such as caveolin 1, rab8 and rab11a (Pearson's coefficients 0.81, 0.89 and 0.92, respectively) (Gagescu et al., 2000; Henry and Sheff, 2008; Prekeris et al., 2000). The highest colocalization was detected with rab11a, the apical recycling endosome reporter (Casanova et al., 1999). The microvilli marker prominin 1 (Fig. 3), was not only present in the PAP (supplementary material Fig. S1E) but also extensively colocalized with gp135 in the compact structure (Pearson's coefficient 0.69) (Corbeil et al., 1999). Consistently, the compact structure and the PAP were embedded in an F-actin network (Fig. 3, supplementary material Fig. S1D). Therefore the gp135-positive compact structure consists of apical recycling endosomal membranes, in which E-cadherin was absent (Figs 1 and 2).
Basolateral recycling endosomes are more acidic (pH 5.8) than apical recycling endosomes (ARE), which have a pH of 6.2-6.5 (Wang et al., 2000). To characterize PAP formation, we performed ratiometric fluorescence pH measurements using pHluorin-mRFP-gp135 (Fig. 4A). The quantitative analysis of the relative and regional pH values (compact structure, PAP and cell surfaces facing the lumen) revealed that the pH in the compact structure is similar to that of the PAP and that in the gp135-positive membranes facing the lumen (P=0.79). It is significantly higher (7.3±2.4%) than the pH in central vesicles (Fig. 4A, right panel) (P=0.0046). Thus, the compact structure environment is less acidic than that of central vesicles but identical to that of the PAP and the apical surface. The gp135-positive central vesicles resemble the vesicles that strongly colocalize with EEA1 (Fig. 3). Thus, if they were indeed these basolateral recycling endosomes, then their reported absolute pH is around 5.8. We found that the pH in the compact structure is 7.3% higher, which results in a pH of 6.2. This value fits well to the values reported for the ARE compartment. Together, the quantified colocalization analyses and the absence of E-cadherin in the gp135-positive compact structure demonstrate that the compact structure has properties of an ARE; this is also supported by the relative pH measurements.
To distinguish whether the compact structure forms the PAP in one or in multiple fusion events, we analyzed fluorescence intensity distributions along the contact surface for gp135, caveolin 1 and gp58 in PAP-forming aggregates, because we had observed that aggregates with a PAP can occur either with a compact structure or without it (Fig. 2A, compare 60-240 minutes with 300 minutes, Fig. 4B). In PAP-forming aggregates with compact structure, gp135 and caveolin 1 intensities followed Gaussian distributions peaking in the centre of the contact surface (R2=0.93 and R2=0.94, respectively). Remarkably, gp58 was inhomogenously distributed along the entire surface (Gaussian distribution: R2=0.11). In PAP-forming aggregates without compact structure, however, the intensities of gp135 and caveolin 1 along the contact surface increased steeply to reach a central plateau; therefore they deviated more from a typical Gaussian distribution than in aggregates with compact structure (R2=0.88). At this stage, the gp58 intensity pattern along the contact surface was complementary to gp135. The different gp58 distribution in aggregates with and without compact structure suggests that during PAP establishment gp58 is removed from the PAP membrane region and is enriched in the lateral membrane. Together, the Gaussian-like distribution of gp135 and caveolin 1 supports the possibility that these apical markers are delivered from the compact structure to the contact surface via stochastically regulated, discrete and directional processes. Indeed, the presence and dynamic behaviour of multiple vesicles near the PAP suggests that the PAP is formed by multiple fusion events (Fig. 2C, supplementary material Fig. S3B). Therefore, the compact structure does not resemble a vacuolar apical compartment (VAC) despite similarities, such as the presence of actin and microvilli (Fig. 2C, Fig. 3, supplementary material Fig. S1D,E) (Cohen et al., 2004; Vega-Salas et al., 1988). Additionally, the presence of endosomal markers in a VAC is controversial and they colocalized with the compact structure only partially (Fig. 3) (Cohen et al., 2004; Low et al., 2000).
Brefeldin A (BFA) is a specific inhibitor of ARF1-dependent vesicular trafficking, which, despite the absence of morphological changes at the TEM level (Hunziker et al., 1991; Prydz et al., 1992), affects Golgi-mediated secretion in MDCK cells (Low et al., 1991) and induces mis-sorting of basolateral proteins in the transcytotic pathway to the apical plasma membrane (Wang et al., 2001). Exposure of a 3D MDCK cultures to BFA significantly reduced the number of two-cell aggregates with lumen and increased proportionally the number of aggregates with PAP (Fig. 4C). Thus, gp135 trafficking leading to the formation of a PAP was unaffected by BFA, but BFA disturbed maturation of the gp135-positive PAP that led to lumen generation.
Altogether, gp135 seems to redistribute from the basal surface to form a PAP via basal early endosomes and an ARE-like compact structure. After insertion of multiple gp135-positive vesicles into the contact surface, an immature PAP is established independently of the biosynthetic pathway. Subsequent BFA-sensitive targeting was necessary to open up a lumen.
Morphometric analysis of de novo lumen formation
To study the mechanism of lumen generation further, we viewed MDCK aggregates by dual channel 4D microscopy. Lumen formation occurred at constant cell number in the absence of cell division and apoptosis (Fig. 5, supplementary material Movie 2). Remarkably, the lumen is formed during rotation of the entire aggregate, with the cells maintaining their positions relative to each other (Zeng et al., 2006).
An example sequence of apical surface and thus lumen formation in a three-cell aggregate is shown in Fig. 5. In this aggregate, all cells express plasma-membrane-targeted double-palmitoylated YFP (pp-YFP), reporting the plasma membrane position, whereas only one cell transiently expresses mRFP-gp135 (Fig. 5, cell 1). Gp135 was initially (–90 minutes) enriched in a PAP spanning almost the entire length of the contact surface (similarly to Fig. 1A and Fig. 2A at 300 minutes), confirming that the PAP is established before lumen appearance. The time between PAP establishment and detection of an optically resolvable lumen was highly variable in our recordings (between 2 and 8 hours) indicative of the maturation phase. A resolvable lumen appeared within 30 minutes and established equivalence between the PAP and the apical surface (Fig. 5, supplementary material Movie 3). Taken together, a PAP can exist over hours and a lumen appears within tens of minutes converting a PAP into an apical surface.
From the images of live-cell recordings (n=7), we measured the morphometric parameters, such as the volume, surface and sphericity of individual cells, aggregates or lumina. The individual measurements were synchronized to lumen appearance (0 minutes) and normalized to the respective values at this time point (Fig. 6, supplementary material Fig. S4C). Data were fit to a linear regression model to test for significant changes between 60 minutes before and after lumen initiation. During this period, aggregate volumes (isochoric) and their sphericities remained constant (P=0.06 and P=0.57, respectively). The increase in luminal volume (P<0.001) was accompanied by a decrease in cell volume (P=0.01). Thus, the cells changed volume and shape within an aggregate of constant shape to accommodate the emerging lumen (Figs 5 and 6, supplementary material Fig. S4). Importantly, the sum of contact and apical cell surfaces of individual cells did not change (P=0.18). Thus, generation of the lumen coincided with a reduction in contact surface, which compensated for the expansion of the apical surface. Similarly to results obtained during the PAP establishment phase, the basal surface did not change during this time (P=0.5) (Fig. 2B, supplementary material Fig. S4C). Lumen enlargement `Phase a' is defined as the period of isochoric aggregate volume after lumen appearance. The subsequent `Phase b' is characterized by a coordinated and strong increase of the aggregate, the lumen and individual cell volumes (Fig. 9). Remarkably, the initial lumen was a prolate spheroid with the major axis aligned with the cell-cell contact plane. Luminal volume increased over time by expanding predominantly in the main equatorial semi-axis, thus increasing the sphericity of the lumen (Fig. 6F).
Inhibition of ROCK1/2-dependent myosin II contraction enhances lumen formation
The shape of individual cells in an aggregate changed upon lumen initiation whereas the aggregate shape remained constant (Fig. 5, Fig. 6B,D,F). Rho-dependent kinases 1/2 (ROCK1/2)-mediated cellular contractility was shown to regulate the morphology of cell aggregates in 3D cultures (Wozniak et al., 2003). To test whether this pathway influences lumen formation, we quantified lumen formation frequencies in two-cell aggregates where the activities of myosin II or its major activator kinases, ROCK1/2 and myosin light chain kinase (MLCK), were inhibited. Surprisingly, inhibition of myosin II, ROCK1/2 or MLCK did not prevent lumen initiation but the inhibitors of myosin II and ROCK1/2 specifically increased the fraction of PAP- and lumen-containing aggregates and reduced the fraction of solid aggregates with basal gp135 localization (supplementary material Fig. S5A,B).
To define precisely the phase(s) in which the inhibitors were interfering, we calculated the probability of developmental transitions using a Markov chain (Markov, 1971). We set the probability of a reversion event (from `PAP' to `no PAP' and from `lumen' to `PAP' or to `no PAP') to 0 as reversions were never observed in any of our recordings (a total of 84 movies covering nearly 100 hours of observation). We found that inhibition of ROCK1/2 and myosin II significantly enhanced both PAP formation and lumen initiation (Fig. 7A,B). The levels of monoSer19-P and diThr18/Ser19-P myosin light chain (pMLC and ppMLC) were not altered upon treatment with ML-7 whereas they were strongly decreased upon treatment with Y-27632 at all applied concentrations (3 μM, 10 μM and 30 μM) (Fig. 7C, supplementary material Fig. S5C). Thus, the inhibitor-mediated reduction of pMLC and ppMLC possibly causes the increase in PAP formation and lumen initiation. Remarkably, aggregates that had already formed a lumen before treatment with Y27632, ML7 or blebbistatin were morphologically unchanged and had unaltered gp135 localization patterns, revealing that lumen maintenance does not require modulation of ROCK-mediated contractility (Fig. 7D).
Together, these results show that ROCK1/2-dependent myosin II activity negatively regulates lumen formation during PAP formation and lumen initiation. Also, ROCK1/2-regulated myosin-II-dependent cell contractility is not necessary for PAP formation or lumen formation or its maintenance. However, ROCK1/2-dependent myosin II activity counteracts PAP and lumen formation but not lumen maintenance.
Channel activity and the paracellular barrier function of tight junctions allow lumen formation
Transepithelial liquid transport driven by ion gradients has been proposed as a mechanism to control lumen size in MDCK cysts (Li et al., 2004). The establishment of ion gradients requires polarized distribution of channels (e.g. apical localization of Cl– channels). As functional tight junction constitutes a paracellular diffusion barrier that facilitates ion gradient formation (Schneeberger and Lynch, 2004), we localized tight junctions in lumen-forming aggregates. Occludin localized in the cytoplasm in 82±2% of aggregates with a PAP (Fig. 8A); otherwise, a tight-junction-like pattern of occludin surrounding the PAP was observed (18±2%). Tight junctions were always present and close to the apical surface in aggregates with lumen (Fig. 8A). Similar results were obtained for the tight junction associated protein ZO-1 (supplementary material Fig. S6A). Thus, tight junctions are positioned after PAP establishment and before lumen appearance around the gp135-positive domain thereby contributing to PAP maturation.
The functionality of paracellular diffusion barriers was tested by staining with anti-gp135 antibodies fixed, but not permeabilized, aggregates transiently transfected with mRFP-gp135 (Fig. 8B). Under these experimental conditions anti-gp135 antibodies did not label the compact structure confirming that this compartment is not in continuity with the PAP surface. However, in aggregates with immature PAP, these antibodies accessed the entire contact surface confirming that the PAP is established before the formation of paracellular diffusion barriers. Anti-gp135 antibodies were also sometimes excluded from the PAP. As such a pattern for PAPs surrounded by tight junctions is expected, we classified these aggregates as aggregates with a more mature PAP. Correspondingly, in aggregates with lumen, the apical surface was not stained by the anti-gp135 antibodies (Fig. 8B).
We analyzed the paracellular barrier function of tight junctions during lumen formation using cells in which ionic permeability of the junctions can be increased by the inducible overexpression of the transcription factor mSnail (Carrozzino et al., 2005). After mSnail expression, E-cadherin and desmoplakin levels were, as expected, altered (supplementary material Fig. S6B) (Cano et al., 2000). However, the 3D aggregate structure was not affected and neither E-cadherin nor desmoplakin are paracellular diffusion barriers, we therefore continued our analysis (supplementary material Fig. S6C). An RNAi approach would require the downregulation of several tight junction components, however, because some act as transcription factors, data would have been difficult to interpret. The fraction of solid aggregates with PAP was increased by mSnail overexpression (–dox), whereas the number of lumen-containing aggregates was proportionally and significantly decreased (P<0.05) (Fig. 8C). Thus, mSnail functioned during PAP maturation to block lumen initiation.
The arrest in the PAP stage was reversible after discontinued mSnail expression (Fig. 8D). During the first 8 hours after mSnail removal, the amount of lumen-containing aggregates gradually increased, whereas the amount of PAP-containing aggregates decreased, maintaining the total number of cells with PAP or lumen constant (Fig. 8D, 0, 4, 8 hours, no mSnail). Eight hours after mSnail removal, the number of aggregates with PAP and lumen was identical to those that never expressed mSnail at the corresponding time point (Fig. 8D, 8 hours, no mSnail +dox) and did not significantly change during the next 40 hours (data not shown). Thus, removal of mSnail triggered rapid lumen formation in PAP-containing cell aggregates.
Lumen initiation after mSnail removal was not affected by general inhibition of membrane trafficking because of incubation at 20°C nor by BFA treatment. However, it was hampered by the inhibition of aquaporins with Ag+ and by the inhibition of Na+ K+ and Cl– channels with bumetanide (Fig. 8D). Thus, additional membrane trafficking is not necessary for the conversion of a mature PAP enriched with all apical components (e.g. gp135, gp114, CFTR) into an apical surface. Since the block of Cl– channels and aquaporins reduced lumen initiation, we suggest that ion-gradient-driven water gathers between the lumen-forming cells, rapidly producing a resolvable separation between the two PAP membranes. Functional tight junctions could retain water, because claudin-1-knockout mice lose tight junctions in the skin and die because of water loss (Furuse et al., 2002). Since PAPs without tight junctions exist, it cannot be excluded that ion and water channels are already positioned and functional before effective diffusion barriers are established. In this situation, water transport might result in a cell volume decrease in the absence of lumen generation (Fig. 6). In fact, statistical comparisons (Student's t-test) of the total cellular volume differences and the lumen between 0 minutes and +60 minutes after lumen appearance revealed no significant difference. However, the loss of total cellular volumes between –60 minutes and +60 minutes is significantly larger than the increase in luminal volume (P=0.026), indicating that the cells already lose volume before lumen initiation.
Remarkably, tight junctions assemble typically in MDCK cells within 4 to 6 hours, which correlates with our time-dependent lumen formation results (Fig. 8C, 0-8 hours no Snail) (Carrozzino et al., 2005; Gonzalez-Mariscal et al., 1985; Wang et al., 1990b). This correlation, the presence of tight junctions around a PAP, and the negative effect of aquaporin and Cl– channel inhibition on lumen formation (Fig. 8A,D, Ag+, bumetanide) suggest that the paracellular barrier function of tight junctions is required for lumen initiation. Downregulation of E-cadherin as a result of mSnail overexpression might also have prevented lumen initiation. However, this possibility is excluded because BFA-treated cells, where E-cadherin is largely removed from the plasma membrane (Sheen et al., 2004), initiated lumen at control frequencies after mSnail removal (Fig. 8D, BFA).
By using time-resolved quantitative analysis of cell polarization and morphometric changes in developing MDCK aggregates, we found that four sequential phases occur during lumen formation (Fig. 9). These stages were defined as PAP establishment, PAP maturation, lumen initiation and lumen enlargement (Figs 1, 2, 3, 4, 5, 6).
Here, we showed that three distinct mechanisms contribute to lumen appearance: regulation of membrane trafficking, ROCK-mediated contractility and generation of hydrostatic pressure. Fixed-cell analysis and live-cell recordings demonstrated that prior to lumen appearance, gp135-positive membranes insert into the contact surface forming a gp135-positive and E-cadherin-negative patch (Figs 1, 2, 3, 4, 5, supplementary material Fig. S1 and Fig. S2). TEM analysis revealed that the adjacent cell membranes in this patch are already separated but highly interdigitating, due to the presence of microvilli protruding from both cell membranes (Fig. 2C).
Prior to PAP establishment, surface-exposed gp135 is found at the basal surface. Thus, to generate a PAP, the cells must either relocalize gp135 to the contact surface or target newly synthesized gp135 directly to the PAP. We exclude the latter because BFA did not inhibit PAP establishment and gp135 did not significantly colocalize with the Golgi markers TGN38 and giantin. In support of transcytotic relocalization, however, is the colocalization of gp135 with vesicular EEA1 close to the basal surface and with caveolin1, rab8 or rab11a in a compact ARE-like structure (Fig. 3). This structure is not continuous with the contact membrane (Fig. 8B). TEM analysis, the Gaussian distribution of fluorescence intensities along the contact surface and the presence of an anti-gp135 antibody-inaccessible compact structure in non-permeabilized cells suggest that multiple vesicles contribute to establishment of the PAP (Fig. 2C, box 2; Fig. 4B; Fig. 8A). This contrasts with a one-step VAC-like membrane insertion (Kamei et al., 2006; Vega-Salas et al., 1988). Taken together, we suggest that gp135 is basally endocytosed and transported in vesicles to the ARE-like compact structure, from where it is targeted via vesicles to the contact surface to generate an immature PAP. Thus, PAP formation seems to require targeted vesicular trafficking.
Remarkably, E-cadherin was always excluded from the PAP (Fig. 1A,B, Fig. 2, supplementary material Fig. S1B). Also, PAP establishment left both the contact and the basal surface unaltered, indicating the presence of a surface surveillance mechanism (Fig. 2B, Fig. 6C). Thus, the maintenance of contact surface, despite local insertion of apical-like membranes, might occur alongside a sub-resolution folding of the lateral E-cadherin-positive membranes (von Bonsdorff et al., 1985). However, this is unlikely because the fluorescence intensity of lateral E-cadherin was constant upon PAP formation in living cells (Fig. 2A, supplementary material Movie 1). E-cadherin removal could alternatively be obtained by local endocytosis, as suggested by the presence and dynamic of E-cadherin-positive structures near the contact surface during PAP formation (supplementary material Fig. S3A). Accordingly, we suggest that establishment of a PAP results from balanced endo- and exocytosis ensuring the maintenance of the contact surface during PAP establishment.
We could resolve two distinct PAP phases, namely the establishment of an immature PAP without tight junctions and Gaussian-like gp135 distribution and its subsequent maturation allowing lumen opening. The latter is also dependent on membrane trafficking, because it was sensitive to BFA (Fig. 4C). Vesicle-mediated establishment of a PAP was necessary but not sufficient to generate a lumen (Fig. 2A, Fig. 4B-C, supplementary material Fig. S1). Thus, membrane trafficking ensured first establishment and then maturation of a PAP, which is a prerequisite for lumen initiation.
The second mechanism required for lumen formation involves the regulation of cellular tension. We demonstrated that inhibition of the ROCK1/2-myosin-II pathway enhanced both PAP formation and lumen initiation (Fig. 7). Therefore, modulation of ROCK-mediated contractility is relevant for both steps. ROCK1/2 activity is required for correct recruitment of adherens junction proteins to the contact surface (Miyake et al., 2006). This pathway has also been shown to negatively control the stability and, thus, the functionality of already established tight junctions (Utech et al., 2005). Remarkably, the establishment of functional tight junctions around the PAP was necessary for lumen initiation and loss of tight junction functionality did not affect establishment of the PAP (Fig. 8). Tight junction functionality was not affected by E-cadherin depletion (Capaldo and Macara, 2007; Theard et al., 2007). Thus, we speculate that the inhibition of actomyosin contraction might have two mainly non-exclusive effects. It might increase cortical actin dynamics to enhance either endocytosis of E-cadherin from the contact surface or insertion of gp135-positive membranes into the contact membrane. Alternatively, it might increase the stability of tight junctions to facilitate lumen appearance. In either way the aggregate would achieve a permissive condition for lumen initiation.
The third mechanism contributing to lumen appearance generates hydrostatic pressure that might cause cell shape changes. Cell shapes changed dynamically upon lumen initiation to accommodate the expanding lumen and to maintain a constant aggregate shape (Fig. 5, Fig. 6D-F). Membrane trafficking was not sufficient to open an optically resolvable lumen (Fig. 2A, Fig. 8D). Surprisingly, the inhibition of actomyosin contraction using specific inhibitors for ROCK1/2, myosin II and MLCK, had either no effect or even enhanced lumen initiation, indicating that ROCK-mediated contractility can inhibit cell shape changes (Fig. 7, supplementary material Fig. S5A,B). However, we reveal (1) recruitment of CFTR channels to the emerging apical membrane (Fig. 1C,D), (2) a requirement of functional paracellular diffusion barriers to open a lumen (Fig. 8), and (3) a reduction of lumen-containing aggregates after inhibition of water or ion channels in the PAP stage (Fig. 8D, Ag+, bumetanide). Together, all these facts suggest that polarized fluid transport into the extracellular space mediated by ion and water channels initiates a lumen that is delineated by a diffusion barrier. Hence, liquid influx into the tight-junction-enclosed compartment can drive the cell shape changes that accompany lumen initiation.
In summary, in the present work, we revealed a sequence of cellular events causing de novo lumen formation (Fig. 9). As each morphogenetic phase builds on the previous one, unidirectional development to a lumen-containing cell aggregate is ensured. Targeted membrane trafficking is followed by the establishment of diffusion barriers, reinforcing cell polarization. Later, lumen initiation requires these functional barriers, channels and the regulation of cellular contractility to fill the luminal space with liquid.
Lumen formation is a fundamental morphogenetic event in nephron development (Kroschewski, 2004; Saxen and Sariola, 1987) and therefore ensures a functional kidney. By combining morphometric measurements with cell polarization analysis, we revealed that lumen formation is based on the surveillance of cell surfaces and volumes. Therefore, the way is paved for the molecular dissection of this morphogenetic system that is based on physical boundary conditions.
Materials and Methods
Antibodies and chemicals
The following antibodies were used: anti-β-actin and anti-E-cadherin DECMA1 (Sigma); anti-mono(Ser19)- and di(Thr18/Ser19)-phospho-MLC (Cell Signaling); anti-occludin-1, anti-ZO-1 antibodies, phalloidin-fluorescein, and anti-IgG secondary goat antibodies labelled with Alexa Fluor dyes (Molecular Probes); mouse anti-EEA1 (BD Biosciences), and rat anti-prominin 1 (Chemicon); anti-gp135 supernatant (Ojakian and Schwimmer, 1988); anti-gp58 (6.23.3) and anti-gp114 (Y-652) (Balcarova-Stander et al., 1984), rabbit anti-giantin (Covance Research Products) and rabbit anti-caveolin 1 (SantaCruz). Y-27632, ML-7 and blebbistatin originated from Calbiochem. Rat anti-TGN38 was a kind gift from Barbara Reaves (Addenbrooke's Hospital, Cambridge, UK). If not otherwise stated, all chemicals were from Sigma.
Rab8-eGFP and rab11a-eGFP were kind gifts from Marino Zerial (MPI-MCB, Dresden Germany). mRFP-gp135: full-length (residues 22-551) rabbit gp135/podocalyxin gene (Meder et al., 2005) (provided by Joachim Fullerkrug, University of Heidelberg, Germany) was subcloned as a PCR-product into HindIII/BamH1 sites of the ptdimer2(12) vector (Campbell et al., 2002). The tdimer gene was exchanged with a PCR-product encoding human CD8α signal sequence (MALPVTALLLPLALLLHAARP) fused to monomeric RFP (mRFP1 gene and ptdimer(12) vector were provided by James Nelson, Stanford University School of Medicine, Stanford, CA) using AgeI and XhoI sites. The resulting mRFP-gp135 construct was verified by dideoxy sequencing.
mRFP-pHluorin-gp135 the superecliptic pHluorin-containing linker was amplified by PCR from TI-VAMP pHluorin (Alberts et al., 2006) using primers (CCCAAGCTTGCTTGTACAGCTCGTCCATGCC; CCGCTCGAGCAAGCGGCGGAAGCGGCGGGACC). mRFP-gp135 plasmid and PCR product were digested and ligated using XhoI and HindIII restriction sites. Fidelity of the construct was confirmed by sequencing.
Cell culture and inhibitor treatment
MDCK II cells were cultured in MEM (Sigma), 10% FCS with 2 mM glutamine (Invitrogen), 100 U/ml penicillin and 100 μg/ml streptomycin. MDCK II cells stably expressing a fusion protein encompassing a GAP43-peptide (MLCCMRRTKQ) that is double-palmitoylated followed by monomeric YFP and a hexa-His tag (pp-YFP; the cells were provided by David A. Zacharias, University of Florida, Gainsville, FL) were cultured as described (Zeng et al., 2006). Stable MDCK II cells expressing E-cadherin-GFP were cultured as described (Adams et al., 1998). Stable MDCK II cells overexpressing mouse mSnail under the control of a tetracycline-repressible transactivator were obtained from Roberto Montesano (Geneva Medical Centre, Switzerland) and cultured as described (Carrozzino et al., 2005). Stable MDCK I cells expressing GFP-CFTR obtained from Bruce A. Stanton (Dartmouth Medical School, NH) were cultured as described (Moyer et al., 1998). Reconstructed soluble collagen type I (Vitrogen 100®, Cohesion, Palo Alto, CA) was mixed with medium containing cells to generate 3D cultures for confocal live cell microscopy and cultured as described (Zeng et al., 2006).
MDCK cells were grown in floating collagen gels supplemented with 0.3% v/v DMSO, 30 μM ML-7, 3 μM to 30 μM Y-27632, 5 μM blebbistatin, 1 μM to 10 μM Brefeldin A (BFA), 0.1% ethanol, 100 μM bumetanide, 10 μM AgNO3 or in gels without inhibitors (control). The drugs were dissolved in DMSO (ML-7, blebbistatin), ethanol (BFA, bumetanide), or in distilled water (Y-27632, AgNO3) and added to the culture medium at the time of plating in collagen (24 hour cultures), or 24 hours before fixation.
MDCK II cells were plated on plastic dishes in medium containing inhibitors, 0.3% v/v DMSO or no additives (control). After 24 hours of growth the cells were lysed; the lysates were loaded on SDS gels and immunoblotted. Band intensities were measured with a GS-800 densitometer (Bio-Rad).
Immunostaining and transfection
Cells were fixed and immunostained against gp135, gp114, gp58 and E-cadherin, as described (Pollack et al., 1998). Briefly, the cells were fixed at 37°C in 4% formaldehyde, permeabilized with 0.1% or 0.5% Triton X-100 in PBS, washed with 75 mM NH4Cl, 20 mM glycine in PBS, blocked in 0.1% BSA, 0.05% Tween-20 in PBS with 10% goat serum (blocking buffer) and incubated with primary antibodies and secondary antibodies labelled with Alexa Fluor 594 (Molecular Probes) overnight at 4°C on a rocking plate; all antibodies were diluted in blocking buffer. Occludin and ZO-1 staining was done as described (Balda et al., 1996) with modifications. Briefly, the cells were pre-extracted for 2 minutes on ice in 0.2% Triton X-100, 100 mM KCl, 3 mM MgC12, 1 mM CaCl2, 200 mM sucrose, and 10 mM HEPES (pH 7.1), fixed for 30 minutes on ice with 96% ethanol, blocked for 1 hour at room temperature in blocking buffer and incubated for 16 to 24 hours at 4°C with primary and secondary antibodies. F-actin was stained with phalloidin-fluorescein; nuclei were stained with Hoechst 33258. Gels were mounted on glass slides in Mowiol, 1.4% w/v DABCO. MDCK II cells were transfected using the Amaxa-T kit (Amaxa Biosystems) and a microporator (Digital Bio Technology).
Specimens for high-resolution confocal microscopy of fixed collagen-grown cells were prepared as described (Pollack et al., 1998) with modifications. Briefly, MDCK cells were trypsinized, resuspended in low-calcium MEM (Invitrogen). Single cells (>90%) were plated in collagen gels (5×104 cells/ml and collagen concentration 2.0 g/l). The gels were detached from plastic immediately after jellification. Z-stacks of fixed cell aggregates were acquired with a Leica SP2 AOBS confocal microscope (Leica Microsystems) as described (Zeng et al., 2006).
Live-cell wide-field microscopy
To obtain high-resolution live-cell images, single MDCK cells were plated on coverslip-bottomed 35-mm dishes, pre-coated with dried collagen, at 5.0×104 cells/ml and with 0.4 g/l collagen or 0.35 g/l collagen supplemented with 3% growth-factor-reduced Matrigel (BD Bioscience). Imaging was started 24 to 48 hours after plating. 4D images were obtained using Deltavision Spectris system (Applied Precision LLC) equipped with an incubation chamber (Solent Scientific), a motorized stage (Nanomotion, Applied Precision), and a cooled CCD camera (CoolSnap HP, Roper Scientific). Image acquisition was done using SoftWorx software (Applied Precision). High-resolution Z-stacks of the cells were obtained with an oil immersion objective (×60, 1.4 NA, PlanApo, Olympus Corporation, Japan).
Transmission electron microscopy
Single MDCK cells were plated on 3 mm diameter sapphire discs pre-coated with dried collagen and grown for 40 hours as described in live-cell wide-field microscopy set up. The cell culture was fixed in situ by high-pressure freezing (HPF) in an HPM 010 high-pressure freezer (Bal-Tec AG Balzers, Liechtenstein). For this, the sapphire discs were covered with a metal spacer ring and a standard HPF Al-specimen carrier (Reipert et al., 2004). The cavity was filled with cell culture medium. The frozen samples were dehydrated by freeze-substitution in water-free acetone with 2% uranyl acetate and 0.5% glutaraldehyde. The samples were kept at –90, –70 and –50°C for 8 hours, respectively, before infiltration with a graded series of HM20 (Polysciences, Eppelheim, Germany) in ethanol (33%, 66% and twice in 100% for 45 minutes each). UV polymerization was done in fresh HM20 for 24 hours at –50°C. The samples were embedded with the cell culture side facing into the resin block. After polymerization the sapphire discs were removed and excess resin was trimmed away. Sections of 50 nm were post-stained with uranyl acetate and alkaline lead citrate and examined in a Philips CM12 transmission electron microscope.
Image processing and volume reconstruction
Fluorescent Z-stacks of cell aggregates obtained from confocal or wide-field microscopy were deconvolved with the Huygens (Scientific Volume Imaging BV Hilversum, The Netherlands) or SoftWorx (Applied Precision).
The 3D geometry of lumina reported in Fig. 6F was manually reconstructed from the deconvolved Z-stack using Amira (Konrad-Zuse-Centrum, Berlin, Germany). The total volume of the reconstructed 3D objects was obtained using the `Measure' function of Amira. X-Y and X-Z projections were obtained using either `Easy 3D' function of Imaris (Bitplane AG) or the `Pitch' and `Yaw' functions of the Zeiss LSM Image Browser (Carl Zeiss GmbH, Germany). The volumes and surfaces of aggregates, individual cells, and lumina were obtained from planar manually segmented individual cell profiles of the deconvolved wide-field 4D movies using Amira. The resulting 3D geometry of a cell was then obtained using the `surface generation' function of Amira. The geometry of the eventual lumen was obtained filling the space between the cells facing the lumen, whereas the geometry of the aggregate was obtained merging reconstructed cells and lumen geometries. Thus the sources of segmentation error were restricted to individual cell segmentation. All other parameters were deduced from this segmentation. A measure for the basal surface (Sb) was obtained reconstructing the entire aggregate as a single 3D object. Similarly a measure for the apical surface (Sa) was retrieved rendering the lumen. From the rendering of individual cells the entire membrane surfaces present in the aggregate (the sum of basal, lateral, and apical surfaces for each individual cell in the aggregate, Sϵ) could be obtained. From the measured surfaces we derived values for the total contact surface (Sc) as: Sc= (Sϵ–Sb–Sa).
The accuracy of a 3D reconstruction from a set of optical images depends on the optical resolution of the imaging system, which set in our case, a theoretical limit to the detection of volumes smaller than 0.05 μm3, and the signal-to-noise ratio of the Z-stack images, which was below 10. Our method applied to a defined fluorescent sphere with r=2.5 μm resulted in an absolute error of 2 μm3, corresponding to 3% of the sphere volume. The exact shape and size of the cells under investigation are unknown, thus only independent segmentations of the same Z-stack can report about a systematic error.
Segmentation errors were calculated by comparing two independent segmentations of the same aggregate. The resulting average relative errors are of 4% on cell volume, 14% on lumen volume, 4.2% on aggregate volume, 3.9% on cell surface, 11.7% on Sa, and 3.8% on Sb.
The channel colocalization in deconvolved 3D Z-stacks was measured using the `colocalization analyzer tool' of Huygens as described (Le Clainche et al., 2007). The iso-colocalization surfaces presented in Fig. 3 were generated using the `Compute colocalization map' function of Huygens and imposing an object threshold of 50%.
The fluorescence intensities were measured in maximum projections of deconvolved confocal Z-stacks for each channel along a ROI overlapping the PAP using ImageJ. Results were smoothed (Savitzky-Golay, first order, ±10 data points), normalized, and fit to a Gaussian distribution (Origin, OriginLab). The gp135 signals were fitted with the Gaussian function: where y0 is the baseline offset, A is the total area under the curve from the baseline offset, xc is the centre of the peak, and w is the full width at half maximum.
MDCK cells transiently expressing mRFP-pHluorin-gp135 were live-imaged using a confocal microscope mounting a water-immersion ×63 objective (1.2 NA). Images were deconvolved and pH maps were generated using the pH PlusGR plug-in in ImageJ (Bizzarri et al., 2006) imposing a pK of 7.18 (Miesenbock et al., 1998) and an R0 (channel ratio in areas of the image where pHluorin is fully quenched corresponding to a pH below 4.75) of 0. The Rf (channel ratio in areas of the image where pHluorin is fully fluorescent corresponding to a pH above 7.5) was calibrated for each image setting the basal pH to neutrality.
Normality of the data was evaluated using the `Shapiro-Wilk Normality Test' in Origin (α=0.05). Unless otherwise stated, data sets were compared using chi-square test separately for each category (e.g. lumen) with P-values pre-calculated for datasets with one degree of freedom. Morphometric data shown in Fig. 6A-E and supplementary material Fig. S4C (time-points: –60 minutes, –30 minutes, 0 minutes, 30 minutes, 60 minutes) were modelled by linear regression analysis in the `R' statistical environment (software obtained from The R Project for Statistical computing, http://www.r-project.org/; data sets used for linear regression shown in supplementary material Fig. S4D-H). Two-sided Mann-Whitney tests (α=0.05) were performed for not normally distributed data sets in supplementary material Fig. S4A,B. Two-sided t-test was used to analyse the normally distributed data in Fig. 7A,D. Significances in differences in relative pH (Fig. 4A) were calculated by a two-way repeated measures ANOVA in `R'. A Markov chain is a stochastic process with Markov property and a standard mathematical method (Markov, 1971).
Quantification of cell aggregates at distinct stages of lumen formation
The frequencies in Fig. 1A, Fig. 4C,E, Fig. 8A,C,D and in supplementary material Fig. S3C, Fig. S5A,B were calculated by dividing the number of aggregates with a certain gp135 localization pattern (e.g. basal gp135) by the total number of examined aggregates. The probabilities of PAP formation and lumen initiation in Fig. 7A were calculated assuming a linear sequence of irreversible events shown in Fig. 7B. The probability of each stage was calculated by dividing the number of aggregates that have reached the investigated stage (e.g. lumen formation), by the total number of aggregates that had reached the previous stage (e.g. PAP formation). The probability values were normalized by the control values (medium and DMSO for Y-27632 and other, respectively). Frequencies in Fig. 7D were calculated by averaging the score of lumen presence in the control and drug-treated cultures. Each aggregate with a lumen scored 1, each aggregate without lumen scored 0, aggregates where the assignment to the above categories could not be unambiguously performed scored 0.5 (Zeng et al., 2006).
The authors thank Mirko Klingauf for expression construct preparation, David A. Zacharias, George Ojakian, Joachim Füllekrug, Bruce A. Stanton, Roberto Montesano, Thierry Galli, Gero Miesenböckand James Nelson for providing antibodies, cells or constructs, and Vartan Kurtcuoglu for image processing expertise. We acknowledge the support of the LMC (light microscopy center), and the WBL (study of applied statistics) at ETH-Zurich, Switzerland. Our work was financially supported by CO-ME, SNF, the KTI, the Roche Research Foundation, the Novartis Foundation and the UBS (AG) on behalf of a client.
Supplementary material available online at http://jcs.biologists.org/cgi/content/full/121/21/3649/DC1
↵* Current address: Scuola Normale Superiore, IIT Research Unit, Piazza dei Cavalieri 7, 56126 Pisa, Italy
↵‡ These authors contributed equally to this work
- Accepted August 31, 2008.
- © The Company of Biologists Limited 2008