Regulation by a TGFβ-ROCK-actomyosin axis secures a non-linear lumen expansion that is essential for tubulogenesis

Tubular structures carry out important functions in multicellular organisms (Lubarsky and Krasnow, 2003). Their morphogenesis involves an orchestrated sequence of changes in cell state and behaviour, as well as the communal activities of cells to construct a functional lumen. The regulation of kinetics of lumen morphogenesis is crucial to ensure a correct morphology, size and function of the tubular structure. Loss of tube size regulation leads to tubule dysfunction conditions, such as polycystic kidney disease (Luschnig and Uv, 2014; Lubarsky and Krasnow, 2003).

A general principle for lumen formation is to inflate the forming lumen by accumulating osmolarity-increasing components, including ions and proteoglycans (Datta et al., 2011). In parallel, as the plasma membrane has limited extensibility, the luminal membrane is expanded through membrane biogenesis. In cases where the lumen is contained in a uniform membrane without cell junctions, as in an intracellular vacuole, the lumen can grow isotropically over time due to the uniform distribution of osmotic pressure, while maintaining a spherical shape continuously. However, in most tubulogenesis models, particularly in vasculogenesis, lumen formation or/and growth occur in a multicellular context, and cellular junctions interrupt the lining of the lumen (Baer et al., 2009). This discontinuity introduces new parameters to lumen morphogenesis, and suggests additional regulatory mechanisms working at the level of the junctions.

During Ciona intestinalis notochord tubulogenesis, extracellular lumen pockets emerge and grow between pairs of cells in tandem, and subsequently connect through extensive cell shape changes and tissue remodelling (Denker and Jiang, 2012; Dong et al., 2009) (Fig. 1A). The formation of an individual lumen represents the simplest model system for multicellular tube formation, as the lumen is produced and enclosed by only two cells. The surface of the lumen is thus interrupted by a ring of cell-cell junctions (apical/lateral junction) connecting the apical membranes of the two adjacent cells (Denker and Jiang, 2012; Denker et al., 2013) (Fig. 1B). The secretory pathway of the notochord cells is responsible for apical membrane growth, and partially for the lumen formation, which is primarily driven by an osmolarity-based lumen inflation process (Deng et al., 2013). A previous fixed time point study (Dong et al., 2009) suggests an isotropic growth of the lumen over time, resembling the growth of an intracellular vacuole. However, the individual lumens in the Ciona notochord are fundamentally different in that they are bound not entirely by plasma membrane but also by a ring of cell junctions. In the static state of an established epithelium, the integrity of cell-cell junctions is actively and dynamically maintained so that the epithelium serves as a barrier with selective permeability (Guillot and Lecuit, 2013). During lumen formation, the cell-cell contact is constantly and dramatically modified, but at the same time it must nevertheless contain the lumen. In notochord tubulogenesis, as the apical membranes and lumen enlarge, the apical/lateral junction enlarges as well, but it is not known how these processes are coordinated.

Here, we show through detailed time-lapse analyses that the lumen growth is not linear but exhibits a lag phase, during which the lumen continues to grow, whereas the tight junction ring around it does not expand. This intricate kinetics is regulated by both the contractility and the dynamics of actin and myosin filaments that reside at the lateral cell-cell junctions, and are themselves subjected to dynamic regulation. Specifically, a TGFβ-ROCK-actomyosin axis negatively regulates the junctional opening that is powered by the osmotic pressure within the lumen. These results reveal an elaborated program concocted by the cells to cope with the complication of the cell-cell junction in constructing a simple lumen.

During Ciona early development, notochord cells undergo convergent extension and form a one-cell-thick rod. Each cell is cylindrical and possesses a single and circumferential basal domain that contacts the basal lamina (Dong et al., 2009; Fig. 1A,B). The cohesion of the notochord tissue is maintained by broad cadherin-based adherens junction disks between consecutive cells, which define the lateral domains of the cells (two per cell). The first step of lumen formation is the formation of an apical domain in the centre of the lateral domains; this is followed by the formation of a lumen pocket between apical domains of two neighbouring cells (Denker et al., 2013; Fig. 1B). The growing lumen is sealed by a de novo-formed ring of tight junctions (ZO-1 positive) at the place where two apical domains juxtapose. As the lumen expands, the lateral domains reduce and the tight junction ring concurrently enlarges (Denker et al., 2013). Once the lumen pockets have reached a crucial size, they tilt in alternate directions as a result of extensive cell shape changes, and eventually contact each other to form a continuous lumen (Fig. 1Ad-f).

To analyse the dynamics of lumen growth, we focused on the period from lumen initiation up to the beginning of the tilting process (∼120 min, Fig 1Aa-c). We monitored the change of three parameters over this period: the lumen volume, the diameter of the apical/lateral junction ring (or transverse diameter, magenta in Fig. 1Bb) and the longitudinal radius (green in Fig. 1Bb), which is along the anterior-posterior axis of the embryo and perpendicular to the junctional plane. As we observed that the junctional ring is circular at all stages of lumen growth (supplementary material Fig. S1A), we concluded that the last two parameters were sufficient to geometrically describe the lumen, thus allowing us to quantify its growth. We observed that the lumen volume increased steadily over time, suggesting a continuous supply of luminal material and apical membrane (Fig. 1C; supplementary material Movie 1). However, behind the linear growth of lumen volume, the longitudinal radius and transverse diameter evolved with different and unexpected kinetics (Fig. 1Da). The longitudinal radius grew in a relatively linear fashion for ∼55 min, then reached a plateau of 6.3 µm on average, which persisted until the end of lumen growth and the onset of lumen tilting. By contrast, the transverse diameter increased linearly for a much shorter time, then stopped at a plateau of around 6.5 µm. This pause lasted, on average, for 30 min, before the transverse diameter increased again, beginning at the same time as the growth of longitudinal radius reached plateau.

Thus, three phases for lumen growth can be distinguished (Fig. 1D). In phase 1 (0 to 25 min), both longitudinal radius and transverse diameter grow linearly. Phase 2 (25 to 55 min) experiences the continuous increase of the longitudinal radius at the same rate as before (for statistical tests, see supplementary material Fig. S1Ba), whereas the change of transverse diameter is stalled. We therefore designated this period as the lag phase. The third phase (55 to 120 min) sees the resumption of the growth of transverse diameter, which proceeds at a significantly higher rate than prior to the lag phase (for statistical tests, see supplementary material Fig. S1Bb), whereas the growth of longitudinal radius stops permanently. The consequence of these successive phases is a stepwise change in the globe shape of the lumen pocket (Fig. 1Db). Initially spherical, it becomes a prolate spheroid during phase 1. During the lag phase, the lumen pocket undergoes dramatic transformation to become an oblate spheroid, with the longitudinal radius doubling and becoming the largest radius of the spheroid. In phase 3, the lumen returns to a slight prolate spheroid, due to the twofold increase of the transverse diameter, which now becomes the largest radius of the spheroid. These dynamic changes of lumen shape accompanying the linear growth of the lumen volume suggest a coordination of two relatively independent processes: lumen formation and apical/lateral junctional opening.

The oblate spherical shape of the lumen in lag phase suggests, according to the Laplace's Law, the presence of a higher surface tension (T) at the junctional region, T₁=P₁×R₁ (Fig. 1Db), than at the areas away from the junctions, T₂=P₂×R₂, considering hydrostatic pressure P₁=P₂ based on the Pascal's Principle and radius R₁>R₂ (Schwarz and Safran, 2013; Clark and Paluch, 2011). Thus, this tension must be actively neutralized, and therefore the pause maintained, by the activity of a mechanical force that directly opposes lumen expansion. Our best candidate was non-muscle myosin II, which when bound to actin filaments generates contractile forces that regulate cell shape changes during morphogenesis (Vicente-Manzanares et al., 2009).

We found that active, S19-phosphorylated, myosin regulatory light chain (MRLC) was abundant in the lateral membrane (Fig. 2Aa, yellow arrow). pS19-MRLC overlapped significantly with a lateral actin population (Fig. 2Aa′-a‴, white arrow). Anti pT18/pS19 MRLC antibody staining showed a similar, though relatively faint, localization pattern (supplementary material Fig. S2). MRLC-mCherry and Lifeact-mEGFP expressed under the notochord-specific brachyury promoter reproduced this localization pattern in live embryos (Fig. 2Ab-b‴). The lateral domain could be subdivided into three regions. The highest concentration of myosin was found in the proximal region close to the apical domain (region 1 in Fig. 2Aa‴,b,b‴). The density of myosin became lower in the region away from the apical domain (region 2), and was undetectable in the distal region next to the basal domain, where only actin is present (region 3). As the lateral domain is the site of E-cadherin-positive adherens junctions and, more proximally, of ZO1-positive tight junctions at the apical/lateral transition (Denker et al., 2013; Dong et al., 2009), we examined in which specific junctional domain the MRLC is present. Co-expression analysis with the basolateral marker Lgl showed a tripartite expression pattern, with a myosin-only proximal domain, an Lgl/MRLC overlapping domain centrally and a distal Lgl-only domain (Fig. 2Ba,b). This organization is identical to that of the E-cadherin/Lgl localization (Denker et al., 2013), suggesting myosin is present in the same area as the E-cadherin. Furthermore, we found little overlap between myosin and ZO1 (Fig. 2Bc), consistent with the idea that the myosin is primarily present at the level of the adherens junctions and not the tight junctions, which define a narrow region most proximal to the apical domain (region t in Fig. 2Bc,Ac″,c‴).

To test the role of myosin in the control of lumen opening, and more specifically in the establishment of the lag phase, we used blebbistatin, which specifically inhibits myosin activity (Straight et al., 2003). We first incubated embryos at the end of the cell elongation but before any visible lumen could be observed; this resulted in an absence of lumen formation (data not shown), suggesting myosin II is necessary for the initiation of lumen formation. We thus treated the embryos at the time when the first sign of lumen could be observed, and followed lumen growth for 2 h. Blebbistatin treatment led to the formation of considerably narrower lumens, following a significantly different kinetics (Fig. 3A-C; supplementary material Movie 2), beginning during phase 2. The longitudinal radius plateaued significantly earlier than the controls (Fig. 3B), while, more strikingly, the transverse diameter increased continuously without a pause, following a relatively linear kinetics (Fig. 3Ca). As a consequence of this altered growth pattern, the lumen underwent relatively homothetic growth, maintaining its initial prolate spheroid shape with the larger radius lying in the plane of junctions (Fig. 3Cb). During the third phase, because the longitudinal radius remained unchanged and only the transverse diameter increased, the lumen was modified into a tall and narrow lenticular shape in the median view. Interestingly, the growth of lumen volume was similar to that in control animals (Fig. 3D), except in the last 30 min, the growth of lumen volume became retarded in the blebbistatin-treated animals. Three hours after lumen tilting began, most lumens fused to each other in the notochord of control embryos, whereas fusion of the lumens in blebbistatin-treated embryos was delayed and partial (supplementary material Fig. S3A). After an extended time, the lumens in blebbistatin-treated embryos are able to connect to form a continuous but narrower tube (Dong et al., 2011). These observations indicate that impairing myosin activity has strong consequences in the lumen growth pattern specifically in the lag phase, and the pause in the opening of the junctions is functionally important for notochord tubulogenesis.

Whereas the inhibition of myosin activity led to an accelerated opening of the junctional ring and a rapid reduction of the actomyosin-rich lateral domain (Fig. 3Ea-f′), overexpression of a constitutively active, phosphomimetic form of MRLC (T18E-S19E), produced the opposite effect. The opening of the junctional ring was arrested and, as consequence, the lumen was constricted at the site of the junction (Fig. 3Eg-i′; supplementary material Movie 3). This phenotype was associated with a dramatic concentration of the T18E-S19E-mCherry fusion protein in the region close to the apical-lateral junction (Fig. 3Eg,h,h′). Interestingly, the altered localization of MRLC did not affect the localization of the cortical actin (Fig. 3Ei,i′). These results indicate that the elevation of myosin activity above normal can counteract lumen expansion, and suggest that an intermediate level of myosin activity is required for normal lumen growth.

The concentration of T18E-S19E MRLC at the apical/lateral junction prompted us to examine whether the lag phase observed in the normal process was associated with a transient accumulation of myosin or a local rise in phosphorylated myosin II, close to the apical/lateral junction. We performed immunostaining before lumen formation and at three phases of lumen expansion in normal embryos (Fig. 4). Interestingly, pS19-MRLC localization remained unchanged throughout lumen development. However, we noted a very local and transient accumulation of actin filaments at the apical/lateral junction during the lag phase (Fig. 4Cc,d).

We next sought to manipulate the actin cytoskeleton to see whether it could effect a change in lumen growth kinetics and lumen shape. Disassembly of actin filaments by latrunculin abolishes lumen formation completely (Dong et al., 2011). We thus used jasplakinolide, which disrupts actin dynamics by promoting polymerization and preventing depolymerization of actin filaments (Holzinger, 2010). The drug was added either before lumen formation or during lumen expansion. Surprisingly, the jasplakinolide treatment resulted in the formation of intracellular vacuoles instead of extracellular lumen pockets in both cases (supplementary material Movie 4 and Fig. 5A, treatment began before extracellular lumens were visible; supplementary material Movie 5, treatment began after significant lumens had formed). Notochord cells in suspension following the protease treatment of jasplakinolide-treated embryos confirmed the intracellular location of the vacuoles (supplementary material Fig. S3B).

These vacuoles were positive for Slc26aα (Fig. 5Ab,d,f,i; supplementary material Fig. S3Bb′), an anion transporter normally localized in the apical/luminal domain (Deng et al., 2013) (supplementary material Fig. S3Bb), but negative for the polarity marker Lgl (Fig. 5Ad), which is normally localized in the basal/lateral domain (Fig. 2Ba,b). They also contained ZO1, and the polarity protein Par6, which is normally co-localized with ZO1 at the apical/lateral junction in normal notochord cells (Denker et al., 2013). Both proteins were concentrated into a single discrete patch (arrowheads in Fig. 5Af,i), in contrast to Slc26aα (arrows). Intriguingly, the E-cadherin, which is normally present in the lateral domain (Denker et al., 2013), was contracted to the vacuole that bears a Par6-positive patch (Fig. 5Ak,l). In a few cases, the jasplakinolide treatment caused the evagination of a small amount of cytoplasm that protruded into neighbouring cell; these evaginations were also enclosed by Slc26aα-positive membrane and decorated with single Par6 patch (asterisks on Fig. 5Ag,i).

To understand how the vacuoles form, we performed time-lapse imaging. Strikingly, vacuole formation was not the consequence of a switch from the production of extracellular lumen to the growth of the intracellular vacuole at the onset of lumen formation, but a dramatic reversal of the lumen formation process and conversion of extracellular lumen pockets into vacuoles. When treatment was initiated before lumen formation, lumen formed initially, indicated by the opening of a tight junction ring (supplementary material Fig. S3Ca, Movie 6). However, 6 min later, the expansion of lumen was halted, and the tight junction ring began to close at a speed of 0.2 μm/min. It eventually converged into a single patch at the lateral cell surface, after which the tight junction-positive membrane, along with a fraction of adherens junctions-positive membrane, detached from the lateral surface and moved into the cytoplasm to form a single small vacuole (also see Fig. 5Al). When treatment was initiated when extracellular lumens were already large, the tight junction ring began to contract within 5 min, at a speed of 0.4 μm/min, resulting in a constriction of the pre-existing lumen in the plane of the ring (supplementary material Fig. S3Cb1, Movie 7). Five minutes later, the contraction led to the pinching of the lumen pockets into two individual intracellular vacuoles. Lifeact labelling revealed a redistribution of actin filaments from the lateral surface to exclusively at the apical/lateral junctions as lumen closure proceeded (supplementary material Fig. S3b2,b3, Movies 8,9). After closure, the intracellular vacuoles stayed juxtaposed to the lateral membrane for various amounts of time as actin filaments continued to accumulate, before finally detaching from the lateral membrane and plunging into the cytoplasm with the actin patch attached to the surface of the vacuole. The vacuoles were able to continue to grow in the presence of jasplakinolide, indicating the treatment does not affect lumen secretion and luminal membrane formation. These results suggest that inducing actin accumulation at the apical/lateral junctions can dramatically affect extracellular lumen expansion, ultimately leading to a reversal of the opening process. Moreover, the facts that when the treatment begins at the onset of phase 1, the effect is observed only in phase 2, whereas when the drug is applied during phase 2 the effect is immediate suggest that it is during phase 2 that this actin-based regulation is at work (supplementary material Fig. S3Cc).

The speed of junctional ring closure is significantly greater than that of its opening during normal lumen growth: 0.2 µm/min before lumen opening and 0.4 µm/min after lumen opening versus 0.1 µm/min. This suggests the potential for a mechanical force at the cell/cell junction that counters the expansive pressure of the lumen. Once this mechanical force becomes unopposed, it can rapidly close the lumen. Supporting this hypothesis, we found that the formation of vacuoles was also associated with an accumulation of myosin and actin at the receding junctions (supplementary material Fig. S3Cb2,b3).

To further test this hypothesis, and considering the apparent opposite effects of blebbistatin and jasplakinolide, we asked whether the inhibition of myosin activity could rescue the jasplakinolide phenotype. Indeed, addition of blebbistatin prevented the conversion of lumen pockets into vacuoles (Fig. 5Be-f′; supplementary material Movie 10). While actin clearly clustered at apical/lateral junctions because of the jasplakinolide, the lumen pockets remained extracellular and grew in dimensions (supplementary material Fig. S4). Similarly, overexpression of the non-phosphorylatable form of MRLC (T18A-S19A) also rescued the effect of jasplakinolide (Fig. 5Bg-h′). These results together suggest that the effect of jasplakinolide is myosin dependent.

Addition of blebbistatin dramatically abolished the jasplakinolide phenotype, but the kinetics of lumen expansion was different from the wild type and, importantly, was also significantly different from the kinetics of blebbistatin alone. Jasplakinolide attenuated the lumen volume growth that blebbistatin alone could attain throughout the entire lumen formation stage (Fig. 5Ca). Further analysis reveals that the kinetics of lumen formation after blebbistatin/jasplakinolide treatment exhibited an ‘intermediate’ phenotype, and displayed the features of both individual treatments. The slope of longitudinal radius growth was less steep than in the wild type, but after 60 min, the radius increased more rapidly to reach a final value close to the control, showing a rescue of the blebbistatin effect (Fig. 5Cb). The growth of transverse diameter did not experience a lag phase, similar to the blebbistatin treatment, although it quickly reached a plateau after 60 min due to the antagonistic effect of jasplakinolide (Fig. 5Cc, the associated lumen shape changes are illustrated in Fig. 5Cd). This intermediate phenotype is also apparent when we plot the longitudinal radius against the transverse diameter (Fig. 5Ce). In control, both parameters increased in phase 1. During phase 2, only the longitudinal radius increased precipitously, whereas the transverse diameter stays relatively constant (highlighted in green in Fig. 5Ce). The opposite occurred in phase 3, and the lumen growth was essentially the widening of the apical/lateral junction ring (highlighted in purple). During blebbistatin treatment, the phase 2 is abolished, whereas the phase 3 is kept. By contrast, jasplakinolide is able to fully restore a phase 2 (highlighted in green), albeit after a slight delay, and also to prevent the phase 3.

These results show that modifying actin dynamics can have a clear effect on the contractile activity at the level of the junctions. As we see that changes in actin concentration in the control animals correlated with phase 2, we can hypothesize that changes in actin dynamics are responsible for the contractile events taking place during phase 2.

A number of distinct pathways regulate the actomyosin contractility and actin filament dynamics. Among them, Rho-associated protein kinase (ROCK) promotes contractility partly through direct phosphorylation of myosin light chain and inhibitory myosin light chain phosphatase, and partly through inhibition of actin filament depolymerization indirectly (Riento and Ridley, 2003). ROCK activity has been shown to influence lumen formation in an in vitro MDCK model, where its inhibitor, Y27632, promotes lumen initiation and opening (Ferrari et al., 2008). In Ciona notochord cells, Y27632 caused an accelerated opening of the junction ring (rapid increase of transverse diameter), leading to a prolate spheroid morphology of the lumen, very similar to that in blebbistatin-treated embryos (Fig. 6Aa-b′). This supports the idea that ROCK activity is upstream of actomyosin contractility and is crucial for normal kinetics of lumen expansion.

In Ciona notochord, the tight junction apparatus at the apical/lateral junction ring, which circumscribes the lumen, undergoes highly dynamic remodelling, disassembling and reassembling constantly as the ring moves centrifugally, with a net increase in total tight junction molecules, such as ZO-1. This ring is the site of apical polarity proteins, including Par6, Par3 and aPKC (Denker et al., 2013). Work on epithelial plasticity has revealed a TGFβ-dependent tight junction disassembly pathway in epithelial-to-mesenchymal transition (Ozdamar et al., 2005). Activated TGFβ receptor cytoplasmic kinase domains phosphorylate Par6 and phosphorylated Par6, in turn, recruits a ubiquitin-dependent mechanism that targets RhoA (the small GTPase upstream of ROCK) for degradation at the tight junction site. The consequent inhibition of actomyosin activity is important for the dissolution of the tight junctions. Interestingly, it has been shown that notochord cells strongly express the only member of Ciona TGFβ family protein (Reeves et al., 2014), as well as a TGFβ type I receptor (TGFβRIc, also called Ci-ALK4/7; (Imai et al., 2004). Based on these observations, we hypothesized that TGFβ pathway may regulate the lumen expansion in notochord cells through the ROCK-actomyosin contractility pathway. We first examined the localization of TGFβRIc. TGFβRIc-tGFP localized diffusely in the cytoplasm but was not found at the cell membrane before lumen formation (Fig. 6Ba-c). During lumen formation, a substantial amount of TGFβRIc-tGFP was found at the plasma membrane, including the apical and lateral domains (Fig. 6Bd-f). We next treated the embryos with SB431245, an inhibitor for TGFβ type I receptor (Inman et al., 2002; Hudson and Yasuo, 2005). SB431245 treatment (20 µM) reproducibly caused a constriction of the lumen, indicating that, whereas the apical membrane expansion and lumen production were not significantly affected, the apical/lateral junction opening was considerably inhibited (Fig. 6C; supplementary material Movie 11), similar to the phenotype of constitutively active MRLC (MRLC T18E-S19E; Fig. 3Eg-i′). This phenotype could be reproduced by overexpression of a dominant-negative version of TGFβRIc (Fig. 6D) (Hudson and Yasuo, 2005), suggesting lumen constriction is likely the result of the specific targeting of TGFβRIc. To test whether the impaired opening of the apical/lateral junctional ring resulted from an excessive actomyosin contractility, we treated the embryos with blebbistatin in addition to SB431245 (Fig. 6Ea-d′) or overexpressed a non-phosphorylatable MRLC (T18-S19A) then treated the embryos with SB431245 (Fig. 6Ee-f′). Both treatments effectively abolished the constriction phenotype of SB431245, and the apical/lateral junction opened significantly. Occasionally, in SB431245/blebbistatin double-treated embryos, an additional central tight junction spot split the lumen pocket in two, suggesting a defect in proper clearance of central tight junctions to form a single lumen. To test whether the RhoA-ROCK pathway could be an intermediate effector between TGFβRIc and actomyosin, we co-treated embryos with SB431245 and the ROCK inhibitor Y27362. This double treatment effectively rescued the phenotype of SB431245 (Fig. 6Eg-h′), suggesting that, under normal conditions, TGFβRIc activation causes an inhibition of the RhoA-ROCK pathway, which leads to a reduced actomyosin contractility. These results suggest the presence of a TGFβ-ROCK-actomyosin axis in the notochord cells that functionally regulates the opening of apical/lateral junction ring and the kinetics of lumen formation.

We have shown that a balanced actomyosin activity (regulated by the TGFβ-ROCK pathway) is crucial for proper lumen opening, specifically for the presence of phase 2 during the lumen expansion, when the longitudinal radius increases but the transverse diameter does not. The rationale for the existence of phase 2 is unknown, but we hypothesized that cells may during this time build a necessary osmolarity gradient that is required for lumen inflation, and the osmotic pressure is responsible for forcing the apical/lateral junctional ring to widen. Slc26aα is a Ciona Slc26 family anion transporter localized exclusively in the apical domain during lumen formation. It has been implicated in the regulation of luminal osmolarity: disruption of its expression abolishes lumen formation without affecting other aspects of tubulogenesis, including apical domain expansion (Deng et al., 2013). We tested whether an increase of Slc26aα could have an effect on lumen formation. Notochord cells overexpressing Slc26aα-mCherry correctly targeted the fluorescent fusion proteins exclusively to the apical domain, and developed extracellular lumens without overt morphological differences from the control. However, detailed quantitative examination revealed that the kinetics of lumen expansion was altered. The kinetics of the transverse diameter increase deviated from the control significantly and followed that of blebbistatin treatment (Fig. 7A; supplementary material Movie 12), whereas the kinetics of longitudinal radius increase remained remarkably unchanged (Fig. 7B). As a consequence of the absence of phase 2, when the junctional opening would have been halted, the shape of lumen at 55 min was close to spherical instead of being an oblate spheroid (Fig. 7C), and the lumen volume was significantly larger than in the control (Fig. 7D). Equally intriguingly, treating embryos overexpressing Slc26aα with blebbistatin did not accelerate the junctional opening further than either blebbistatin or Slc26aα overexpression alone. These results suggest that luminal osmotic pressure is the driving force for the opening of the apical/lateral junction during lumen expansion. The halt of junctional opening can be abrogated when the osmotic pressure is artificially elevated to a level that can overcome opposing force effected by the contraction of actomyosin at the junctions.

Much effort has been made to understand the specification of the apical domain and initiation of the lumen in tubulogenesis; less is known about the process of lumen expansion after lumen formation is initiated. This is confounded by the complexity of cellular context in most tubulogenic scenarios. This study takes advantage of the Ciona notochord – the simplest multicellular lumen formation model, and reveals unexpected and complex lumen expansion kinetics. The most striking feature of this kinetics is the presence of a lag phase during which the apical/lateral junctional opening is halted. The molecular mechanism behind this lag phase relies on actin filament stability, and very importantly on actomyosin contractility at the cell-cell junctions. As supplementary material Fig. S5 summarises, there appears to be a TGFβ-ROCK-actomyosin axis in operation to secure a gradual and proper progression of lumen expansion. The connection with actin dynamics has been shown in other models, where actin depolymerization is necessary for junction disassembly, as treatment with jasplakinolide blocks the process (Gliem et al., 2010; Ivanov et al., 2004). In addition, inhibiting depolymerization using jasplakinolide or cofilin knockdown leads to an over-contractility that can be rescued by blebbistatin (Flannagan et al., 2010; Wiggan et al., 2012).

The tandem arrangement of cylindrical notochord cells determines the circular geometry of apical/lateral junction. As a consequence, when the cortical actomyosin elements exert their force they reduce the junctional ring, much like the purse-string type epithelial apical constriction mechanism involved in gastrulation and dorsal closure in early development (Takeichi, 2014; Lecuit et al., 2011; Dawes-Hoang et al., 2005). The evidence for the existence of this centripetal contractile force comes also from the observation that the closing of junctional ring after jasplakinolide treatment is much more rapid than that of opening. Importantly, in notochord and other reported systems, Rho-dependent kinase ROCK plays an essential role in actomyosin contraction. The regulation of lumen expansion by actomyosin contractility reported here is unlikely to be an isolated case, as the work by Gutzman and Sive has demonstrated that actomyosin might negatively regulate the size of lumen in zebrafish neural tube (Gutzman and Sive, 2010). Of note, regulation of myosin contractility has been reported to play important roles at other stages of tubulogenesis. For example, in MDCK cells, hepatic cells and mouse blood vessels, peripheral actomyosin inhibits cell polarization, preapical patch (PAP) formation and lumen initiation, whereas inhibition of ROCK or myosin II facilitates lumen initiation (Ferrari et al., 2008; Chen and Macara, 2005; Herrema et al., 2006; Xu et al., 2011). Inhibition of ROCK or myosin II in these contexts enhances cortical actin dynamics, increases E-Cadherin endocytosis and increases tight junction stability. Intriguingly, inhibition of ROCK or myosin II at a later stage does not enhance lumen expansion (Ferrari et al., 2008). In Ciona notochord, myosin contractility is required for lumen initiation, as blebbistatin treatment before any sign of lumen opening blocks tubulogenesis.

An outcome of the cortical actomyosin contraction during tissue remodelling is strengthening of the cell-cell junctions (Jamora and Fuchs, 2002; Lecuit et al., 2011; Delanoë-Ayari et al., 2004; Rossier et al., 2010; Shewan et al., 2005; Cavey et al., 2008). This is particularly relevant in notochord cells as we observed that wrenching up actomyosin contractility is accompanied by maintenance of or an increase in cell-cell junctions, whereas weakening of the actomyosin activity accelerates tight junction opening and removal of lateral interfaces. Thus, actomyosin activity has to be finely calibrated so that the junctions of luminal tissue remain impermeable and simultaneously plastic enough for tubulogenesis to complete. Interestingly, the TGFβ pathway has been reported to exert opposite effects on junction homeostasis and remodelling (supplementary material Fig. S5). First, it promotes actomyosin contractility through ROCK (Moustakas and Heldin, 2008), which would halt lumen opening. But it has also been shown to cause a local disassembly of actin filaments (Ozdamar et al., 2005; Zhang et al., 2013) and, thus, a disassembly of the tight junctions, which would favour lumen opening and expansion. Similarly, it has been shown that the pathway can promote adherens junction disassembly by facilitating E-Cadherin endocytosis through the FLRT3/Rnd1 pathway (Ogata et al., 2007), which would also favour lumen opening and expansion. Thus, TGFβ could, by itself, balance actomyosin activity and lumen opening through differentially and quantitatively activating several pathways. Interestingly, the fact that after blebbistatin-induced rescue of the SB431245 treatment there are sometimes additional tight junction spots (Fig. 6Ec-d′) could suggest that only the contractility phenotype is rescued, but the proper tight junction disassembly is not in notochord cells. In addition, the TGFβRI receptor is expressed more strongly in the posterior part of the notochord (Reeves et al., 2014), and concomitantly we observed a stronger pS19 MRLC and actin staining in the most posterior notochord cells (supplementary material Fig. S6). These differential expressions are intriguing and their significance is worthy of further study.

The primary force of lumen expansion is intra-lumen pressure that is established by an osmotic gradient across apical membrane of the adjacent cells. This is supported by the observation that lumen formation fails after the disruption of apical-localized anion transporter Slc26aα (Deng et al., 2013), and overexpression of Slc26aα produces a lumen larger than control (Fig. 7D). The fact that the volume increase is linear suggests there is a linear addition of material content. During lumen formation the total volume of notochord increases by twofold, whereas the total volume of notochord cells remains constant (Dong et al., 2009). The mechanism by which the bulk of the lumen is added is little known. It is conceivable that some liquid comes through the tight junctions. The TGFβ pathway has been shown to regulate permeability in several epithelial models (Ronaldson et al., 2009; Lebrin et al., 2005; Birukova et al., 2005). The lumen inflation mechanism might be altered by the blebbistatin treatment through reducing the transport of transporters to the apical membrane, or altering the permeability of the junctions, as the concomitant overexpression of Slc26aα and blebbistatin treatment does not further enhance the respective phenotypes. We can also consider that actomyosin contractility, by opposing junctional opening, also helps build tension to power luminal growth together with osmotic pressure.

Unlike an ideal and uniformed biological entity such as a vacuole growing in a uniformed medium, a growing lumen in Ciona notochord is enclosed by two apical domains interspaced by an apical/lateral junction ring. Hence, the lumen/cell interface is discontinuous, and in phase 2, when the lumen takes up the shape of an oblate spheroid, the wall tension at the interface becomes heterogeneous. Based on Laplace's law, the force of tension T exerted on the interface is directly proportional to the pressure P_l within the lumen and the radius of interface curvature R, i.e. T=P×R. Hence, at a given point (a) on the apical membrane the wall tension T_a is approximated to be R_a×P_l (supplementary material Fig. S7) and the wall tension at the apical/lateral junction (j) is the highest (R_j×P_l). To withstand such a high wall tension and to prevent the cells from being torn apart at the junction, notochord cells have recruited an elaborated system centred at actomyosin contractility. Further study will explore the two additional questions: how are the transitions, from phases 1 to 2 and 2 to 3, triggered, and how do the cells sense and respond to the hydrostatic pressure? We speculate the presence of a sensing system relaying pressure/mechanical information in the lumen to the TGFβ receptors, which in turn steer the actomyosin balance in one way or another. Alternatively, as it has been shown in the kidney or lung, mechanical stretch can locally activate or induce the expression of the TGFβ ligand (Quinlan et al., 2008; Warburton et al., 2003). The pathway could then be activated by autocrine signalling, which is well known in development or cancer (Dumont and Arteaga, 2003; Janda et al., 2002; Nakajima et al., 1998; Badalucco et al., 2013). Many proteins have been proposed, either directly or indirectly, to be mechanosensors (notably in blood vessels). They include cadherins, cytoskeleton proteins, tyrosine kinase receptors and luminal membrane proteins (ion channels and glycocalyx proteins). In the kidney, polycystin has been suggested to sense flow and, when mutated, to be responsible for polycystic kidney (Hahn and Schwartz, 2009). Ultimately, this system would optimize the dynamics of lumen opening so that the notochord tube reaches the correct size at the correct moment, which is important in the life of a pelagic larva that uses its tubular notochord as a hydroskeleton to swim away and settle before it can feed itself.

Adult C. intestinalis were purchased from the Service Expédition de Modèles Biologiques, Station Biologique de Roscoff (CNRS/University Pierre and Marie Curie, France). The animals were kept in the Sars Centre ascidian culture facility in running filtered seawater for 7 days under constant light to accumulate gametes. Eggs were then extracted through dissection and mixed in seawater with sperm from other individuals. Ten minutes after fertilization, the eggs were washed with seawater through a nylon filter to remove sperm and debris. Embryos were then dechorionated as described previously (Jiang et al., 2005).

These plasmids have been described previously: DE-Cad-mCherry (Dong et al., 2009), MRLC-mCherry, ensconsin-3GFP and Lifeact-mEGFP (Dong et al., 2011); Slc26aα-mCherry (Deng et al., 2013); and Lgl-tGFP, ZO1-tGFP, tGFP-Par6 and mCherry-Par3 (Denker et al., 2013). MRLC T18E-S19E-mCherry and MRLC T18A-S19A-mCherry were generated from MRLC-mCherry by directed mutagenesis using the Stratagene Quick Change II Mutagenesis Kit. TGFβRI was amplified from cien220421 clone in the Ciona intestinalis gateway-compatible full-length cDNA library (by Mike Gilchrist, kindly provided by Janet Chenevert) using primers described previously (Hudson and Yasuo, 2005), and cloned into the pCR 8/GW/TOPO vector (Invitrogen) to obtain entry clone. A TGFβRI entry clone, and a Nup50 entry clone from Patrick Lemaire (The Centre de Recherches de Biochimie Macromoléculaire, Montpellier, France) (Roure et al., 2007), were recombined into the Minos-B3-eBra-bpFOG-B5::R1-ccdB/CmR-R2-tGFP or the Minos-B3-eBra-bpFOG-B5::R1-ccdB/CmR-R2-mCherry destination vector (Dong et al., 2009, 2011) using the Gateway LR reaction (Invitrogen) to generate TGFβRI-tGFP and Nup50-mCherry expression constructs, respectively.

Electroporation was performed as described previously with some modifications (Corbo et al., 1997). Dechorionated fertilized eggs (200 μl) were mixed with 80 μg plasmid DNA (adjusted to a volume of 100 μl using double distilled H₂O) and the osmolarity was adjusted by adding 400 μl of 0.95 M mannitol in double distilled H₂O. Electroporation was performed in 4 mm cuvettes with a Gene Pulser Xcell System (BIO-RAD), using a time-constant protocol (15 ms, 50 V). After electroporation, embryos were washed once and kept at 13 or 16°C to the desired stages. Live embryos were transferred into eight-well Lab-Tek chamber slides (ThermoScientific/Nunc) for confocal observation.

We used anti-phospho-myosin light chain 2 (Ser19) (Cell Signaling Technology, 3671) and anti-diphospho-myosin light chain 2 (Thr18/Ser19) (Cell Signaling Technology, 3674) at 1/50, and anti-rabbit Alexa-568 (Invitrogen A11011, dilution 1/600) as secondary to stain the embryos following the fixation and immunostaining protocols described previously (Dong et al., 2009). Animals were counterstained with phallacidin (3 units/ml). Samples were mounted using VectaShield medium and Secure Seal Spacers (13 mm×0.12 mm, Electron Microscopy Science) for confocal observation.

Blebbistatin (Sigma, B0560), jasplakinolide (Sigma, J7473) and SB431245 (Sigma, S4317) were dissolved in DMSO and Y27362 (Merck Millipore, 688002) in milliQ water. Final concentrations were 100 µM for blebbistatin, 1 µM for jasplakinolide, 20 µM for SB431245 and 200 µM for Y27632.

Confocal images were taken with a Leica TCS SP5 confocal laser scanning microscope (CLSM) equipped with a 40× oil immersion objective (numerical aperture 1.25). Z-series were taken at intervals of 0.5 to 1 μm, resulting in stacks of 20-50 images. Image analysis and processing were performed with Leica TCS SP5 systems LAS AF software packages, Adobe Photoshop and ImageJ. Unless otherwise specified, the projections presented are maximum projections. For temporal analyses of lumen development, longitudinal radius and transverse diameter were measured on ImageJ at regular time intervals from the start of lumen formation until the lumen fully formed. Lumen volume was calculated with the values thus obtained.

Statistical analysis of lumen expansion was performed using SigmaPlot 12.5.

We thank Agnès Roure and Patrick Lemaire for providing Gateway destination vectors, Dm-E-cadherin, Nup50 and ensconsin entry clones; Janet Chenevert for providing the TGFβRI full-length clone from the Ciona intestinalis gateway-compatible full-length cDNA library; François Robin for providing ZO1 and Lgl entry clones; and the ‘Liaozhai 101’ group for discussion on biophysics.