Matrix Metalloproteinases are required for membrane motility and lumenogenesis during Drosophila heart development

Matrix Metalloproteinases (Mmps) degrade glycoproteins and proteoglycans of the extracellular matrix (ECM) or cell surface and are crucial for morphogenesis. Mmps and their inhibitors are expressed during early stages of cardiac development in vertebrates and expression is altered in multiple congenital cardiomyopathies such as cardia bifida. Drosophila genome encodes two copies of Mmps, Mmp1 and Mmp2 whereas in humans up to 25 Mmps have been identified with overlapping functions. We investigated the role of Mmps during embryonic heart development in Drosophila, a process which is morphogenetically similar to early heart tube formation in vertebrates. We demonstrate that the two Mmps in Drosophila have distinct and overlapping roles in cell motility, cell adhesion and cardiac lumenogenesis. We determined that Mmp1 and Mmp2 promote Leading Edge membrane dynamics of cardioblasts during collective migration. Mmp2 is essential for cardiac lumen formation, and mutants generate a cardia bifida phenotype. Mmp1 is required for luminal expansion. Mmp1 and Mmp2 both localise to the basal domains of cardiac cells, however, occupy non-overlapping domains apically. Mmp1 and Mmp2 regulate the proteoglycan composition and size of the apical and basal ECM, yet only Mmp2 is required to restrict ECM assembly to the lumen. Mmp1 negatively regulates the size of the adhesive Cadherin cell surface domain, whereas in a complementary fashion, Mmp2 negatively regulates the size of the Integrin-ECM domain and thereby prescribes the domain to establish and restrict Slit morphogen signalling. Inhibition of Mmp activity through ectopic expression of Tissue Inhibitor of Metalloproteinase in the ectoderm blocks lumen formation. Therefore, Mmp expression and function identifies ECM differentiation and remodelling as a key element for cell polarisation and organogenesis.


Introduction
Matrix Metalloproteinases (Mmps) are zinc dependent proteases which digest components of the extracellular matrix (ECM) and embedded signalling molecules. Mmps enable morphogenesis by modifying cell migration, cell polarization, ECM remodelling and lumenogenesis PLOS ONE | DOI: 10.1371/journal.pone.0171905 February 13, 2017 1 / 22 a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 [1]. Mmps are regulated by signalling pathways such as Wnt/β-catenin and can modulate signalling of growth factors such as Vascular Endothelial Growth Factors (VEGF) [2][3][4]. Due to genetic redundancy in the mammalian genome, it is challenging to examine Mmp function when genetic compensation can affect mutant analysis [1,[5][6][7]. In contrast, the Drosophila genome encodes two Mmps, a secreted protease Mmp1 and a glycophosphatidylinositol (GPI) anchored Mmp2 [8,9]. Drosophila has a single Tissue Inhibitor of Metalloproteinase (Timp), which has been shown experimentally to be a potent inhibitor of vertebrate Mmps, Drosophila Mmp1 and Mmp2, and other extracellular proteases [10]. An in vitro model of human vasculogenesis reveals that endothelial cells require a membrane associated Mmp (MT1-Mmp) for luminal expansion and formation of vascular guidance tunnels [11]. Branching morphogenesis in organs such as lungs, mammary and submandibular glands requires activity of Mmps for cell motility and lumen formation, mediated by ECM degradation [12]. During tumour invasion, MT1-Mmp, targeted to the invadopodia, promotes metastasis by degrading ECM barriers [13,14]. Although vertebrate Mmps have received considerable attention, their contributions to embryonic morphogenesis are less characterised because of the barriers to genetic approaches [5,7,15,16]. Using Drosophila embryogenesis as a genetic model, we can study the effect of complete elimination or inhibition of Mmp activity in vivo. Conservation of activity in Drosophila homologues of Mmps has been well established [8,9]. Intriguingly however, Mmp activity is not required for embryonic viability in Drosophila since single or double mmp mutant embryos hatch and survive until mid or late larval stages [17]. Nevertheless, developmental processes such as motor axon fasciculation during embryogenesis require Mmp activity [18,19]. Expression data demonstrated that Mmps are upregulated during late stages of embryogenesis in multiple tissues [17]. Vertebrate Mmps and Timps are expressed in the cardiomyocytes during early heart tube assembly [20,21] and modulate cardiac morphogenetic events such as heart tube formation, directional looping [22] and differentiation of ostial cells [23].
In this study, we tested the genetic requirement of Mmps during heart development in Drosophila, a process with striking similarities with early stages of vertebrate cardiogenesis [24,25]. To form the embryonic heart, cardioblasts (CBs) in the lateral mesoderm align into bilateral rows and collectively migrate towards the dorsal midline. There, the CBs form specific medial adhesions with their contralateral partners and form an apical lumen, which expands as the embryo transitions through larval stages [26]. We demonstrate that Mmp1 and Mmp2 are expressed in the CBs and both influence Collective Cell Migration (CCM) of CBs by regulating lateral adhesions between CBs and the activity of the migratory Leading Edge (LE). After migration, Mmp1 and Mmp2 play complementary roles during ECM remodelling, cell polarization and lumen formation. Mmp2 is essential for forming adhesions with contralateral CBs. Mmp1 is required to limit the domain of E-Cadherin based contralateral adhesions and its activity contributes to luminal expansion. We show that Mmp2 regulates apical localization of guidance signalling molecules Slit and Robo, and ECM receptors Integrin and Dystroglycan (Dg). Composition of the CB apical luminal ECM is modulated by both Mmps, whereas positioning is only dependent on Mmp2. Overall, we demonstrate that Mmps play essential roles promoting CCM, ECM remodelling, cell polarization and lumen formation during Drosophila cardiogenesis.

mmp1 and mmp2 mutants have distinct heart phenotypes
In Drosophila, polarised positioning of molecularly distinct ECMs to the basal and luminal domains is essential for the formation of the heart [27][28][29]. Since Mmps contribute to ECM degradation and remodeling in other systems, we reasoned that Mmps may contribute to Drosophila heart formation. To address this hypothesis, we examined the phenotypes of loss of function single and double mmp1 and mmp2 mutant hearts with luminal, junctional and nuclear markers and assessed embryonic heart structure (Fig 1). The cardioblasts of stage 16 wildtype embryos were aligned in bilateral rows and migrated to the midline collectively ( Fig  1A). At stage 17, contralateral CBs contact at the midline and then reshape to enclose a medial lumen (Fig 1A'). Dystroglycan (Dg), an ECM receptor, localises to the luminal domain, whereas Discs-large (Dlg), an apical polarity scaffold protein, labels the junctions at the apical attachment sites (Fig 1E and 1E'). Although loss of either or both mmp1 and mmp2 resulted in disorganised CB arrangement along the bilateral rows, CBs eventually reached the midline (Fig 1B, 1B',1C, 1C', 1D and 1D'). Following migration, mmp1 and mmp2 mutant phenotypes In mmp1 Q112 / mmp1 Q112 null mutants (from here on referred to as mmp1), organization of CB rows is uneven and delayed migration of CBs is observed (asterisk), however they reach the midline and form a reduced size lumen (arrow). (C-D') In mmp2 w307 / mmp2 w307 and mmp1 q112 , mmp2 w307 / mmp1 q112 , mmp2 w307 null mutants (from here on referred to as mmp2 and mmp1,mmp2, respectively), organization of CBs along the bilateral rows is disrupted and CB migration is delayed (asterisk). The CBs in the heart region do not reach the midline by stage 17 (arrow). (E-H) Cross sectional views of stage 17 hearts labelled with α-Mef-2, α-Dg and α-Dlg are visualised (E-H). (E) In wildtype, Dg localises to the CB luminal domain (arrowhead) and pericardial cells (asterisk) while Dlg localises to the junctional domains (arrow). (F) In mmp1 mutants, a reduced Dg labeled lumen forms (arrowhead) and Dlg labels the enlarged junctional domain (arrow). (G) In mmp2 and mmp1,mmp2 mutants, CBs fail to adhere to contralateral partners (arrowhead). CBs exhibit rounded morphology and Dg localises along an extended apical domain (arrowhead). (E'-H') Representation of CBs at stage 17 in wildtype and mmp1, mmp2 and mmp1,mmp2 mutants summarising distribution of Dg (green), Dlg (red) and Mef2 (blue). Posterior is to the right in dorsal view images (A-D). Scales-25 μm in A-D' and 10 μm in E-H'. diverge. In mmp1 mutants, a Dg rich lumen forms, but it is reduced in size and is enclosed by extended Dlg marked cell junctions (Fig 1F and 1F'). In mmp2 and mmp1,mmp2 double mutants, a lumen fails to form (Fig 1G, 1G', 1H and 1H'). Prior to lumen closure, wildtype CBs curve to form a medial pocket. CB somas in mmp2 mutants are spherical and Dg localisation is circumferentially distributed, whereas Dlg marked junctions are absent ( Fig 1G, 1G', 1H and 1H'). We scored 20-25 βPS Integrin labelled posterior heart cross-sections and determined that 75% of the embryos mutant for either mmp2 or both mmp1,mmp2 were rounded and lacked contralateral contacts, whereas the most prevalent phenotype in mmp1 mutants was the presence of a small apical lumen (Table 1). These results suggest that Mmp2 is required to develop a heart tube, whereas Mmp1 is required for luminal expansion.

Remodelling of apical and basal ECM requires Mmp1 and Mmp2 activity in the CBs
In Drosophila, Mmp2 is required to cleave components of the basement membrane and regulate its polarised deposition [19,[30][31][32][33]. To explore Collagen-IV deposition in the developing heart, we analysed time-lapse movies of embryos expressing a CollagenIV-GFP gene trap construct (VkgGFP). In wildtype embryos, Vkg is present at the basal and medially oriented apical surface of the CBs (Fig 2A, S1 Movie). At the apical side of CBs, Vkg is absent from the apical extensions and is restricted to the pre-luminal domain during CCM, and to the lumen thereafter (Fig 2A"). In mmp1 mutants, VkgGFP localises normally and is absent from the adhesive domains ( Fig 2B-2B", S2 Movie). In mmp2 and mmp1,mmp2 double mutants, VkgGFP localisation extended to the entire apical region of the CBs (Fig 2C-2C" and 2D-2D", S3 and S4 Movies). The junctional (the dorsal-apical domain of CBs which adheres to contralateral CBs), luminal (the apical domain encompassing the lumen), basal and lateral (domain connecting ipsilateral CBs) domains in mmp2 and mmp1,mmp2 mutants significantly accumulated GFP signal relative to wildtype (Fig 2C',2D' and 2G). We reasoned that if spatial restriction of ECM around the CBs requires Mmp2, then Mmp2 would not be active in the pre-luminal domain. By employing a Mmp2-GFP construct [31] we determined that Mmp2 was enriched in the apical outgrowths and basal domains and was absent from the pre-luminal domain (Fig 2E and  2E'). After heart assembly, enriched localisation of Mmp2 was not detected at any particular CB cell domain ( Fig 2F and 2F'). Mmp1 was detected in the pre-luminal and luminal domain of CBs during migratory and lumen formation stages, respectively, supporting a model in which Mmp1 excludes Cadherin from this domain (S1 Fig).
If Mmp2 limits the extent of the luminal ECM, does it also play a role in defining ECM identity? To address this question we examined the distribution of Pericardin (Prc), an ECM component expressed and deposited by the pericardial cells exclusively at the basal side of CBs [29]. In contrast to wildtype, where Prc is excluded from the lumen (Fig 3E and 3E'), we found Prc localisation at the apical ECM as well as the basal ECM in mmp1, mmp2, and mmp1,mmp2 mutants ( Fig 3F, 3F', 3G, 3G', 3H and 3H'). These results suggest that Mmp1 and Mmp2 collectively regulate the identity of the apical ECM, whereas Mmp2 is required to limit ECM to the pre-luminal and luminal domain.

CB polarization requires Mmp2 activity
Apicalisation of Integrin is an early event in CB polarization [34]. Localised Integrin enables CBs to stabilise polarising morphogens such as Slit and its receptor, Robo, which are required for lumen formation [34][35][36][37]. However, it remains unclear what signals coordinate early Integrin apicalisation. Since Vkg accumulates ectopically in mmp2 mutants (  Signalling by the Slit morphogen is required for heart lumen formation, perhaps by destabilising Cadherin based CB adhesion [35][36][37]. If ECM assembly is altered in Mmp mutants, then the resulting expansion of Integrin distribution would extend the domain of Slit and Robo stabilisation. In wildtype stage 17 embryos, Slit and Robo accumulate along the apical domain and the lumen develops as a single, continuous midline tube (Fig 4A, 4E, 4I and 4I'). In mmp1 mutants, Slit and Robo localise predominantly to the luminal domain, however lateral accumulation is occasionally observed (Fig 4B, 4F, 4J and 4J'). More dramatically, in mmp2 and mmp1,mmp2 mutants, Slit and Robo fail to apicalise, but accumulate at the lateral domains ( Fig 4C, 4D, 4G, 4H and 4K-4L'). These lateral domains resemble ectopic pockets with luminal composition. Integrin and Vkg are present in the ectopic lumens in mmp2 mutants, confirming the luminal characteristics of these lateral structures (data not shown). Even though Integrin localisation includes the apical domain in mmp mutants, the ability to direct Slit and Robo to the apical domain is lost in mmp2 and mmp1,mmp2 mutants and retained in mmp1 mutants. Collectively these results suggest that targeting of guidance molecules Slit/Robo to the luminal domain, downstream of Integrin function, requires Mmp2 activity. In the Drosophila nervous system, Mmp2 dependent cleavage of Faulty Attraction, an ECM protein containing multiple epidermal growth factor (EGF) domains, is required for motor axon targeting [19]. Similarly, Slit, an EGF domain containing protein, can be cleaved in to a C-terminus fragment of unknown function and an active N-terminus fragment [38,39]. Since Slit normally localises to the luminal domain we sought to establish if Mmp2 processed Slit outside the lumen. However, subsequent to ubiquitous transgene expression in embryos, no differences in full-Slit or C-Slit levels in Mmp2 overexpressing embryos were seen relative to control (data not shown). If Slit processing by Mmp2 is unlikely, other mechanisms account for Slit mislocalisation in mmp2 mutants. One possibility is that lateral accumulation of Slit and Robo in mmp2 and mmp1,mmp2 mutants is due to perturbed Integrin dependent targeting and ectopic stabilisation.

Mmp1 and Mmp2 are required for the motility of CB Leading Edge during collective cell migration
CBs collectively migrate towards their final destination-the dorsal midline. Since Mmps regulate cell migration in multiple vertebrate and invertebrate models [40][41][42], we assessed whether either Mmp contributes to CCM of CBs. To observe CCM, we analysed time-lapse movies of heart development in embryos expressing a nuclear marker, tail-up-GFP (tupGFP), and a fluorescent actin binding protein, Moesin-mCherry (MoeRFP) (Fig 5). This method allowed us to quantify aspects of CCM such as migration velocity, filopodial and lamellopodial activity [43]. We noticed a significant reduction in migration velocity in both mmp2 and mmp1,mmp2 mutant CBs, and a non-significant reduction was observed in mmp1 mutants (Fig 6A, S2  Table). These results strongly suggest that ECM modification is required for the normal  To better understand why migration speed declined, we quantified the number of filopodia and lamellopodia at the Leading Edge (LE). As CBs approach the midline, the number of Moesin-labelled filopodia and lamellopodia increases at the LE (Fig 5A and 5A',  S2A Fig, S5 Movie). These protrusions mediate the formation of apical adhesions with the contralateral CBs upon medial contact (Fig 5A") [36]. In mmp1, mmp2, and mmp1,mmp2 mutants, filopodial and lamellopodial activities were significantly reduced (Fig 5B, 5B',5C, 5C', 5D and 5D', Figs 6B and 5C, S6-S8 Movies, S2 Table). Although both Mmp1 and Mmp2 activity were required to promote LE activity and outgrowth formation during CCM stages of heart development, only Mmp2 was required for timely migration of the CBs. Additionally, the negative correlation between distance to the midline and number of filopodial extensions was reduced in mmp1, mmp2 and mmp1,mmp2 mutants compared to wildtype (S2B- S2D Fig, S1 Table). During early stages of migration, CBs maintain direct contact with Amnioserosa cells and follow behind the ectoderm LE as it migrates towards the dorsal midline [44]. However, following ectoderm dorsal closure, CBs initiate autonomous movement to complete their migration to the midline. We used complimentary approaches to test whether Mmp2 is autonomously required in the CBs for CCM and lumen formation. First, we attempted to rescue mmp2 mutants by expressing a UAS-Mmp2 transgene using a mef2-GAL4 construct that drives the expression of GAL4 in heart and body wall muscle cells, but not in the Amnioserosa or the Ectoderm which are immediately ventral and dorsal to the heart tissue. Migration velocity, filopodial, lamellopodial activity of CBs and lumen formation (data not shown) were rescued (Fig 5E-5E", Fig 6A-6C). The negative correlation between distance to the midline and number of filopodia per segment however was not restored (S2F Fig, S1 Table). Second, we reduced Mmp2 levels in the CBs by driving a UAS-mmp2-RNAi construct under the control of mef2-GAL4 driver. Recapitulating the phenotype of mmp2 mutants, CCM of CBs was delayed and frequent gaps (data not shown) and clumps of CBs were also observed in MMP2 reduced embryos (S3B Fig, S9 Movie). Reduction of filopodial and lamellopodial activities were also apparent and migration velocity was reduced (Fig 6A-6C). Formation of intermittent lumens was observed in Mmp2 reduced embryos (S3F Fig). Altogether these results suggest that CBs autonomously regulate CCM and lumenogenesis primarily via Mmp2 activity.

Overexpression of Mmp2 in the CBs disrupts ECM, CCM and lumen formation
The observation that migratory and lumen formation phenotypes in mmp1 mutant were relatively less pronounced compared to mmp2 mutants suggests that Mmp2 plays the predominant role in regulating cardiac development. Since removing Mmp2 activity in the CBs results in disruption of CCM, lumen formation and accumulation of ECM around the CBs, we hypothesised that elevating Mmp2 levels in the CBs would increase ECM degradation and decrease the size of the CB lumen. We employed a UAS-Mmp2 construct under the control of heart specific mef2-GAL4. During CCM, the regular CB alignment was disrupted, but the disjointed CB rows were still able to extend filopodia and lamellopodia (Fig 7A and 7A', S10 mmp2-RNAi knockdown embryos. Knockdown and control embryos were raised at 29˚C to maximise the effect of the GAL4/UAS system. Bars represent mean±s.e.m. Filopodial (B) and lamellopodial (C) activities are significantly reduced in mmp1, mmp2, mmp1,mmp2 mutants and mmp2-RNAi knockdown embryos. Migration velocity (A), filopodial (B) and lamellopodial (C) activities are restored in embryos where MMP2 rescue was performed. For Box plots, center lines show the medians and box limits indicate the 25th and 75th percentiles as determined by R software [60]. Whiskers extend to the 5 th and 95 th percentile. Width of the boxes is proportional to the square root of the sample size. Notches represent 95% confidence intervals. Onetailed T-test was performed to determine significance of difference. *-P<0.05, **-P<0.01, ***-P<0.001. Movie). Quantification further demonstrated that migration velocity and lamellopodial activity in the CBs were not affected, however filopodial activity was reduced (Fig 6A-6C). The negative correlation between distance to the midline and number of filopodia per segment was also reduced (S2E Fig, S1 Table). Once migration was complete, the dramatic misalignment of the CB rows persisted in to late embryonic stages (Fig 7B and 7B'). We further assessed whether Integrin localisation was affected in the CBs. In control and Mmp2 overexpressing embryos, Integrin localised to the pre-luminal and luminal domains of the CBs (Fig 7E-7E" and 7F-7F"). At stage 17, in Mmp2 overexpressing embryos however, Integrin was observed at multiple small lateral and midline luminal pockets (Fig 7B and 7B'). Consistent with a role for Mmps in ECM degradation, overexpression of Mmp2 resulted in severe reduction of VkgGFP (C-D") Embryos expressing vkg-GFP and mef2>moesin-mCherry demonstrate the reduction of Vkg-GFP levels at the CB ECM in mef2>mmp2 embryos relative to control (arrows). In control embryos, ECM localises at the pre-luminal (C,C' arrowhead), luminal (C", arrowhead) and basal domains (C-C", diamonds). After Mmp2 overexpression, ECM is reduced at the pre-luminal (D,D' arrowhead), luminal (D", arrowhead) and basal domains (D-D", diamonds) (C-D', arrow indicates the dorsal midline). (E-F') Embryos labelled with α-βPS and α-Mef2 reveal the disorganization of CBs at the heart proper in Mmp2 overexpressing hearts relative to control (arrows). In control embryos, Integrin localises to the pre-luminal (E' arrowheads) and luminal domains (E,E", arrowheads). In mef2>mmp2 embryos, Integrin localises to the pre-luminal domain (F', arrowheads) and ectopic lumens (F,F", arrowhead). (G) Mean Vkg-GFP fluorescence intensity in control and Mmp2 overexpressing embryos for junctional, luminal and basal domains. Junctional, luminal and basal fluorescence intensity, respectively, differ significantly in Mmp2 overexpressing embryo compared to both controls as determined by one-tailed T test. *** -p<0.001. Scale-10 μm.
doi:10.1371/journal.pone.0171905.g007 levels surrounding the CBs relative to control embryos (Fig 7C-7C" 7D-7D", 7G). This suggests that presence of Vkg in the CB pre-luminal ECM is not essential for Integrin targeting but might be required for expansion of luminal ECM. However, disruption of the ECM during lumen formation fragments the continuous Integrin-ECM complex and generates ectopic luminal pockets.

Ectopic expression of TIMP in the ectoderm inhibits lumen formation in CBs
Mmp activity is regulated by Timp which inhibits both Mmp1 and Mmp2 in Drosophila [10]. Timps are expressed in both mammalian and Drosophila hearts during cardiac development [45,46], suggesting that CBs tightly regulate Mmp activity to control ECM assembly. Surprisingly, overexpressing Timp in the CBs using a mef2-GAL4 driver did not affect CCM or lumen formation (data not shown). Therefore, we expressed Timp ectopically in the tissue closest to the apical extensions of the CBs where Mmp2 accumulates. We used the paired-GAL4 driver to express Timp to higher levels in the ectodermal stripes immediately dorsal to the heart ( Fig  8A). We reasoned that since Mmp2 is detected at the apical domain of CBs, expression of TIMP in the dorsally located ectodermal stripes would be sufficient to inhibit local Mmp2 activity. All CBs, located ventral to the paired expressing and the more distant non-paired expressing ectodermal stripes, failed to extend apical protrusions and form a lumen when Timp was ectopically expressed (Fig 8B, 8B', 8C and 8C'). Next, we assessed whether CBs retained their ability to localise ECM receptor Dg to the appropriate CB domain. In the control embryo, Dg localised to the luminal and basal domains of the CBs (Fig 8D and 8D'). Subsequent to ectopic Timp expression, CBs failed to form a lumen and Dg label was dispersed around the CBs (Fig 8E and 8E'). In addition, misaligned and rounded CBs were also observed, phenocopying mmp2 and mmp1,mmp2 mutants. This phenotype, coupled with the in situ hybridisation data [45], suggests that Timp modulates Mmp function during heart development, and that expression by the CBs is spatially regulated.

Discussion
Multiple Mmps and Timps are expressed in embryonic vertebrate hearts and are required for key events such as tube formation, cardiac looping and heart septation [20,21]. Although vertebrate studies highlight the importance of Mmps and Timps, the large number of Mmp and Timp genes complicate analysis of their function. Employing Drosophila as a genetic model, we demonstrate that Mmps are required for CCM, ECM degradation, cell polarization and lumen formation in Drosophila hearts.
Loss of Mmp2 activity results in severe embryonic heart development phenotypes that persist to larval stages. This activity is autonomously required in the CBs. Surprisingly, despite a malformed heart, mmp2 mutant larvae appear healthy and develop through to pupation, whereas mmp1 and mmp1,mmp2 mutant larvae die during late larval stages [17]. This suggests that an effective heart is not required for embryonic and larval development. Peristaltic movements within the larva may be sufficient to circulate the hemolymph around the larval body in the absence of a functional heart.

Mmp1 and Mmp2 have distinct roles during embryonic heart development
Previous studies suggest that Mmp1 and Mmp2 play distinct roles in Drosophila. During fat body disintegration in pupae, Mmp2 is required to cleave components of the basement membrane and thereby disable cell-ECM interaction while Mmp1 disrupts cell-cell adhesions by cleaving cell adhesion molecule, E-Cadherin [30]. Consistent with this, our data indicates that Mmp1 and Mmp2 perform divergent functions during heart development, specifically during CCM and lumen formation. Mmp1 is required to expand lumen size, which is consistent with E-cadherin as a key substrate [30]. Since Mmp1 localises at the pre-luminal and luminal domains, its activity may restrict E-cadherin to less exposed dorsal and ventral attachment sites and remove E-cadherin from the more exposed medial luminal domain. In a complementary manner, Mmp2 localises to the apical outgrowths in CBs and assists in clearing the Collagen-IV containing apical ECM from the LE. The significant reduction of filopodia and lamellopodia at the LE of CBs and accumulation of Vkg around the CBs in mmp2 mutants suggests that Mmp2 activity promotes the formation of invasive protrusions by degrading the ECM barrier at the dorsal apical domain. Enlargement of the Cadherin domain in mmp1 mutants may reduce the concentration of LE motility factors below threshold for filopoda formation.
The contrasting roles for Mmp1 and Mmp2 reflect different substrate specificities and localization during heart development. Mmp1 can degrade Collagen-IV and Fibronectin in vitro, however cannot degrade laminins, Fibrinogen or type I and type II fibrillar collagens [8]. Mmp2 on the other hand is proposed to degrade Collagen-IV as well as laminins [9]. Laminins are required for proper localisation of ECM components around the CBs and are required for heart development [27,28,47,48]. Stability of Laminin is key to lumen formation. In both lanb1 and mmp2 mutants, abnormal accumulation of Collagen-IV is observed around the CBs [47].
In mammals, GPI anchored MT1-Mmps accumulate at podosomes of normal cells and invadopodia of metastasising cancer cells and are responsible for removing the basement membrane barrier [13,14,49]. CBs extend apical outgrowths, morphologically similar to mammalian podosomes, which localise Mmp2. Therefore, Mmp2 may weaken or remove the apical ECM barrier to create the necessary environment for the formation of actin-based membrane extensions (Fig 9). Degradation of the ECM might further release embedded signalling molecules such as Slit and Netrin which are required for formation of filopodia and lamellopodia in the CBs [43].

How might Mmps promote cell polarization?
An early feature of CCM of CBs is the targeting of Integrin and ECM components to the apical pre-luminal domain [34]. Subsequently, Integrin at the pre-luminal domain stabilises lumen size determinants, Slit and Robo. In wildtype heart tubes, the apical domain is comprised of the luminal sub-domain bracketed by dorsal and ventral junctional domains that seal the heart tube medially. In mmp1 mutants this fundamental polarity is maintained, but the luminal subdomain is reduced and the junctional domains are extended. In contrast, in mmp2 mutants, apical junctional domains are not established and a modified ECM domain extends over the entire CB apical surface. The mmp2 mutant apical ECM contains ectopic Pericardin, and Slit and Robo do not accumulate; neither is characteristic of a pre-luminal domain. Mmp2 is not required to target Integrin apically, since βPS1 accumulates apically in mutants. Rather, Mmp2 is required to limit the extent of the ECM within this apical domain. Mmp2 is also not required for proper targeting of Integrin in the Drosophila pupal fat body [30]. Both Mmp function and Integrin accumulation at the incipient CB apical surface are required to stabilise (and possibly target) the luminal determinants Robo and Slit [34]. In both the fat body and heart, Integrin localisation is not affected when Mmp2 is overexpressed suggesting that Mmp2 activity regulates Slit/Robo apicalisation downstream of Integrin activity and that Integrin is not a target substrate for Mmp2 in Drosophila. The spatial restriction of Fibroblast Growth Factor signalling activity by Mmps described during branching morphogenesis in Drosophila [50] may analogously apply to heart lumen formation signals Netrin and Slit.
Mmp2 sculpts the location and molecular composition of the apical CB lumen. In mmp2 mutant embryos, ectopic lumens (described by [51,52]) form between lateral CBs which localise luminal markers, Integrin, Dg, Slit, Robo and Vkg. This suggests that the complex of luminal determinants can assemble in the absence of Mmp2, and thereby displace Cadherin from lateral pockets. In converse, increased Mmp2 expression results in degradation of Collagen-IV on all cell surface domains. Collagen-IV is not required for formation of luminal space between CBs since lateral luminal pockets containing Integrin and not Collagen-IV are also observed between Mmp2 overexpressing CBs.
Ectopic lumens, containing the canonical luminal determinants Slit and Robo, form when Mmp2 is absent or if Mmp2 is overexpressed. Both perturbations alter the molecular identity of the apical surface, but this does not reveal how pre-luminal Integrin leads to Slit and Robo accumulation. Two possible mechanisms emerge. The Syndecan family of proteoglycans work cooperatively with Integrins in ECM adhesion [53]. Syndecans are substrates of MT1-Mmps [41,54] and stabilise Slit and Robo in the heart lumen [55]. Therefore, it is possible that Drosophila Sdc is an Mmp2 substrate, and its cleavage would prevent Slit accumulation. Alternatively, Timp may be localised by a component of the pre-luminal ECM and acts locally to inhibit Mmp2 activity.

Conclusion
Heart morphogenesis in vertebrates and Drosophila begins with polarised cell migration, medial cardiac cell fusion and lumen formation. Mmp action plays a key role in all three activities. Mmps regulate the sub-cellular pattern of ECM stabilisation and the local molecular composition of the ECM. This is required to localise membrane responses to guidance signals, including those for LE motility and for lumen specification.

Time-lapse live imaging, quantification and statistics
Embryos were prepared according to the hanging drop protocol [59]. Mutants were identified by the absence of GFP expressing balancer. Live imaging was performed using a Leica SP5 confocal microscope.
Quantification of Viking levels was performed by measuring the GFP intensity of a 2 μm 2 area at the junctional, luminal or basal domains of CBs by employing ImageJ. All values were normalised to wildtype embryos (vkgGFP/+) (n = 19 embryos).
Quantification of filopodial and lamellipodia activity is as previously described by Raza and Jacobs [43]. One-tailed T-tests were performed to compare migration velocity, filopodial and lamellopodial activity between a specific pair of genotypes, with a p threshold of 0.05. Single cross-sectional images at the heart proper region per embryos were scored blindly and the averaged values are shown in Table 1.  Table. Comparison of correlation coefficients of 'distance of CBs to the midline' and 'number of filopodia per heart segment' in wildtype, mutant and rescued embryos. Z values were obtained through Fischer r to z transformation. Z values were used to determine the significance of difference (p-value). 'N' represents the number of heart segments scored. (DOCX) S2 Table. Sample size for migration velocity, filopodial activity and lamellopodial activity quantification. Genotypes which display a significant difference in correlation coefficient compared to wildtype are shaded (p<0.05, -1.645<z<1.645). (DOCX)