The secreted AdamTS-A metalloprotease is required for collective cell migration

Cell migration is crucial to shaping and positioning tissues during development, with some cells migrating extensive distances to populate various tissues. Cell migration also occurs throughout life, such as during tissue infiltration of immune cells, bone remodeling and wound healing. Tumor metastasis, the process by which tumor cells leave their site of origin to invade a secondary tissue, also requires cell migration. Cells move as individuals, as groups or as intact epithelia. Parts of cells also migrate, as observed with the axonal projections of neurons or fine branches of Drosophila tracheoles.

Several complex and delicately orchestrated events underlie directed cell movement (Alberts et al., 2002). Migrating cells extend actin-rich cytoplasmic protrusions (filopodia, lamellipodia and pseudopodia) in the direction of migration. Such protrusions form by actin polymerization at the leading edge, which pushes the cell membrane forwards. Polymerization of the actin filament plus ends enriched near the leading edge is counteracted by depolymerization of the actin filament minus ends deeper in the cell. For cells to move, they must also attach to a substratum. Attachment is mediated by integrins, which are transmembrane heterodimeric signaling molecules that recognize and bind components of the extracellular matrix (ECM), such as collagen and fibronectin, and that also bind proteins within the cell that are linked to the actin cytoskeleton (Ginsberg et al., 1992; Schwartz, 1992; Sastry and Horowitz, 1993). With force provided by myosins, a cell contracts to release the tension created by the cellular protrusions at the leading edge, bringing the bulk of the cell forward. The trailing edge must simultaneously release from the substratum to allow forward movement.

Cells typically travel through and upon the ECM, a complex mixture of proteins and polysaccharides. The ECM, which is produced and secreted by cells, fills the intercellular space to help determine the shape and mechanical properties of many tissues. The complex fibrillar meshwork of the ECM, once thought to primarily provide structural support and tissue integrity, plays an active role in regulating cell behavior (Rozario and DeSimone, 2010; Brown, 2011; Wolf and Friedl, 2011). ECM proteoglycans sequester and modulate chemical signals, including growth factors and guidance molecules. Importantly, adhesions between cells and the ECM are crucial determinants of the rates and directions of cell movement, with tight adhesions correlating with slower movement and weaker adhesions correlating with more rapid movement. Consequently, too little or too much adhesion can prevent movement entirely (Gullberg and Ekblom, 1995; Streuli, 1999).

Much is known about single cell migration and interactions between the cell and ECM. Much less is known about collective cell migration. In single cell migration, the entire cell contacts the ECM, attaching and detaching from it as the cell moves forward. By contrast, during collective cell migration, cells contact both the ECM and other cells within the collective. Maintaining cell-cell adhesions while adjusting cell-ECM adhesions adds significant complexity to the process. Nonetheless, during both development and tumor metastasis, many cells migrate as collectives, moving as highly polarized epithelial sheets or branches, or as less polarized cell clusters or streams (Rørth, 2009).

Modulation of the ECM, which is crucial to both single cell and collective migration, is mediated by matrix metalloproteases (MMPs), a group of zinc-dependent proteases that regulates ECM composition, organization and function through cleavage of ECM components (Vu and Werb, 2000). MMPs are either secreted or membrane bound, either through a single transmembrane domain or covalently attached membrane anchor. ADAMTS metalloproteases (a disintegrin and metalloprotease with thrombospondin motifs), a subgroup of secreted zinc metalloproteases, have several domains that are distinct from those of classical MMPs (Blelloch and Kimble, 1999; Nishiwaki et al., 2000; Apte, 2004). Based on studies in C. elegans, ADAMTS metalloproteases are thought to clear a path in front of migrating cells to allow forward movement (Blelloch et al., 1999; Blelloch and Kimble, 1999; Nishiwaki et al., 2000).

The human genome encodes 19 ADAMTSs, many with redundant functions (Apte, 2004; McCulloch et al., 2009; Enomoto et al., 2010). Therefore, studies in Drosophila, which has only three ADAMTS proteins, can provide key insights into their mechanism of action. Here, we characterize AdamTS-A (currently known as CG14869), which is expressed in migratory populations, including cells that migrate as individuals and cells that migrate as highly polarized collectives. We show AdamTS-A is essential for migration of multiple tissues. Our studies of AdamTS-A function in the SG reveal that not only do ADAMTS proteins clear a path for migration, as suggested for C. elegans GON-1, but may also function in cell detachment.

The Bloomington Stock Center (Indiana University, Bloomington, IN, USA) provided if^k27e (Wilcox et al., 1989), mew^m6 (Brower et al., 1995), tub-GAL4, Df(3R)Exel⁶¹⁷⁴, UAS-srcEGFP, UAS-TmemGFP and tud¹ (Boswell and Mahowald, 1985). M. Frasch (Universität Erlangen-Nürnberg, Erlangen, Germany) provided bHLH54F⁵⁹⁸ (Ismat et al., 2010), 5053A-GAL4 and croc-lacZ (Häcker et al., 1995). R. Lehmann (Skirball Institute, New York, USA) provided nos-GAL4 (Van Doren et al., 1998). S. Hayashi (RIKEN CDB, Kobe, Japan) provided btl-GAL4 (HVII) (Shiga et al., 1996). N. Perrimon (Harvard University, Cambridge, MA, USA) provided wb^4Y18 (Martin et al., 1999). C. Dahmann (Max Planck Institute, Dresden, Germany) provided Cad99C^57A and Cad99C^120B. Our lab generated trh⁸ (Isaac and Andrew, 1996), fkh-GAL4 (Henderson and Andrew, 2000) and sage-GAL4 (Chung et al., 2009).

The translation start site through the cysteine-rich domain of the AdamTS-A ORF was deleted by homologous recombination (Gong and Golic, 2003) and confirmed by PCR. AdamTS-A knockout lines failed to complement Df(3R)88F5^{e03525-d01653} and Df(3R)Exel⁶¹⁷⁴, and failed to express detectable transcript. The AdamTS-A^KO and deficiencies are maintained over the TM6B, Ubx-lacZ balancer, as both genotypes are lethal over TM3, which carries an AdamTS-A allele (Nelson and Szauter, 1992). Primers used to generate the knock out and other tools are described in supplementary material Table S1.

Two versions of UAS-AdamTS-A and UAS-AdamTS-A-GFP, corresponding to the two ORFs predicted on Flybase, were generated by subcloning a cDNA into the Gateway vectors (http://emb.carnegiescience.edu/labs/murphy/Gateway%20vectors.html). Although transcripts were detected with both versions of UAS-AdamTS-A, protein was detected with only the shorter form, suggesting that only the smaller ORF is made. To generate an enzyme-dead version of ADAMTS-A, the glutamate at position 439 was mutated to an alanine using the QuickChange Site-Directed Mutagenesis Kit (Stratagene).

All tissue-specific rescue experiments were performed with the AdamTS-A^KO in trans to Df(3R)88F5^{e03525-d01653}. All fly lines were balanced over lacZ balancers and stained with βGalactosidase (βGal) to distinguish homozygotes.

TEM analysis was performed on stage 11-12 AdamTS-A^KO and wild-type (Canton S) embryos as previously described (Myat and Andrew, 2000). AdamTS-A^KO homozygotes were selected prior to fixation by the absence of GFP using UAS-srcEGFP/UAS-srcEGFP; AdamTS-A^KO/TM6B, twi-GAL4.

Immunohistochemistry and in situ hybridization were performed as described previously (Reuter and Scott, 1990; Azpiazu and Frasch, 1993; Knirr et al., 1999). Rabbit polyclonal antibodies used: anti-Vasa (1:200, Santa Cruz Biotechnology), anti-βgal (1:3000, Capell), anti-GFP (1:10,000, Molecular Probes), anti-GM130 (1:100, Abcam), anti-Rab11 [1:500 (Satoh et al., 2005); R. Cohen, University of Kansas, Lawrence, KS, USA], anti-Vps16 (1:100, H. Kramer, University of Texas Southwestern Medical Center, Dallas, TX, USA), anti-Cad99C (1:10,000; C. Dahmann, Max Planck Institute, Dresden, Germany), anti-CC3 [1:100, Cell Signaling, anti-SAS (1:500, unpublished)]. Rat polyclonal anti-SG1 (1:10,000) (Abrams et al., 2006) and rat anti-ADAMTS-A (1:500, our lab) were used. Mouse monoclonal antibodies used were: anti-2A12 [1:10, Developmental Studies Hybridoma Bank (DSHB)], anti-Crb (Cq4) (1:100-DAB, 1:10-fluorescence, DSHB), anti-α-spectrin (1:2, DSHB), anti-βgal (1:5000-DAB, 1:500-fluorescence, Promega), anti-GFP (1:200, Molecular Probes), anti-actin (1:2000, MP Biochemicals), anti-CSP2 (1:40, DSHB) and anti-FasIII (1:100, DSHB). Secondary biotinylated anti-mouse, biotinylated anti-rabbit, biotinylated anti-rat, goat anti-mouse-555, goat anti-rat488, goat anti-rabbit488 and goat anti-rat568 were used.

Rat antiserum to ADAMTS-A was generated against inclusion body preparations of E. coli expressing residues 298-469 of the AdamTS-A ORF. The antiserum does not detect endogenous levels of ADAMTS-A protein, but does detect overexpressed protein.

Stage 12 wild-type, AdamTS-A^KO and AdamTS-A^KO Cad99C^57A/120B embryos were stained using antibodies against SAS (apical membrane) and α-spectrin (basolateral membrane), and 0.5 μm sections imaged on a Zeiss LSM 510-Meta confocal microscope with a 100× objective. The lengths of the three distalmost cells were measured from the center of the apical surface to the center of the basal surface using ImageJ software and converted from pixels to μm.

Wild-type (Canton S), AdamTS-A^KO, tub-GAL4::UAS-AdamTS-A and tub-GAL4::UAS-AdamTS-A^E439A embryos were dechorionated and sorted for absence of GFP on an embryo sorter (Union Biometrica). Approximately 50 μl of embryos were resuspended in 200 μl sample buffer [Tris-HCl (pH 6.8), SDS, bromophenol blue, glycerol, β-mercaptoethanol]. Samples (20 μl each) were run on 6% SDS-PAGE gels, transferred to methanol-treated PVDF membrane and exposed to Kodak film using enhanced chemiluminescence (ECL) reagents (GE-Healthcare). Primary antibodies used were: anti-Cad99C (1:10,000) (D’Alterio et al., 2005) (D. Godt, University of Toronto, Canada) and anti-βtubulin (E7) (1:1000, DSHB). HRP-conjugated secondary antibodies were used at 1:10,000.

Mammalian ADAMTS family members are expressed in a variety of tissues, and are important for a wide range of developmental and disease processes (Vázquez et al., 1999; Sandy et al., 2001; Jönsson-Rylander et al., 2005; Dunn et al., 2006; Held-Feindt et al., 2006; Murphy, 2008; McCulloch et al., 2009; Enomoto et al., 2010; Kessenbrock et al., 2010). Human ADAMTS genes have been grouped into four major subfamilies based on sequence and functional similarities (Fig. 1A) (Brocker et al., 2009). Family A includes ADAMTS 9 and ADAMTS 20, which cleave aggrecan, versican and brevican (Sandy et al., 2001; Apte, 2004; Nakada et al., 2005; Held-Feindt et al., 2006; Silver et al., 2008; McCulloch et al., 2009). Family B, the largest subgroup, includes ADAMTS 7 and ADAMTS 12, which cleave cartilage oligomeric matrix protein (COMP) (El-Hour et al., 2010). Family C - ADAMTS 2, ADAMTS 3 and ADAMTS 14 - processes procollagen types I-III (Abbaszade et al., 1999; Apte, 2004; Jones and Riley, 2005). ADAMTS 13, the single Family D member, cleaves von Willebrand factor (Dong et al., 2002). The Drosophila genome encodes three ADAMTS proteins, two of which are completely uncharacterized (Fig. 1A). stall (stl), which is most similar to ADAMTS 13 of the D group, functions in ovarian follicle cell formation (Ozdowski et al., 2009), whereas the uncharacterized CG4096 gene is most similar to the ADAMTS B group. Here, we focus on the previously uncharacterized ADAMTS-A, the single Drosophila A group ADAMTS (supplementary material Fig. S1).

ADAMTS secreted proteases contain several domains, many with undetermined functions. All contain an N-terminal signal peptide (SP), a prodomain, a metalloprotease domain, a disintegrin domain, thrombospondin-type I (TS-1) repeats and a cysteine-rich domain (Fig. 1C). Drosophila ADAMTS-A, human ADAMTS 9 and ADAMTS 20, and C. elegans GON-1 contain an additional C-terminal cysteine-rich Gon1 domain (Fig. 1C; supplementary material Fig. S1) (Llamazares et al., 2003; Somerville et al., 2003).

AdamTS-A is expressed in hemocytes (H) and caudal visceral mesoderm (CVM), cells that migrate as individuals (Fig. 1Ba,b,d-g). The CVM is a population of mesoderm cells that migrates anteriorly, spreads dorsally and later forms the outer layer of longitudinal visceral muscle fibers (LVMF) (Fig. 1Bb,d-h). The visceral branch (VB) of the trachea migrates as part of a polarized collective of cells and expresses AdamTS-A during mid-embryogenesis (Fig. 1Bc,e). The salivary gland (SG), which migrates as a fully intact and highly polarized collective, expresses AdamTS-A in the secretory cells from invagination at stage 11 to the end of migration (Fig. 1Bb-h). Expression of AdamTS-A in the early CVM and tracheal VB was confirmed using mutations in the transcription factor genes HLH54F and trh, which are required for gene expression in each respective cell type (Fig. 1Bi-k) (Ismat et al., 2010; Chung et al., 2011). Thus, AdamTS-A is expressed in four distinct migratory tissues: hemocytes, CVM/LVMF, tracheal VB and SG.

To analyze AdamTS-A function, we created a small deficiency, Df(3R)88F, that removes AdamTS-A and two neighboring genes, and we created a knockout allele, AdamTS-A^KO (Fig. 1D; see Materials and methods). The AdamTS-A^KO allele is RNA null (Fig. 1Bl), and lethal over deficiencies that remove AdamTS-A (supplementary material Table S2). We first examined the effects of AdamTS-A loss in a well-characterized migratory population, the germ cells (GC). Prior to migration, GCs are encapsulated into the pocket of the posterior midgut primordium (supplementary material Fig. S2A, Fig. S3A). GCs then cross the posterior midgut epithelium and migrate to the somatic gonad (supplementary material Fig. S2B-E, Fig. S3B,C). By late embryogenesis, most GCs are found in two tight clusters, one on either side of the embryo (Fig. 2Aa; supplementary material Fig. S2F, Fig. S3D) (Starz-Gaiano and Montell, 2004; Kunwar et al., 2006). We found ∼6% of late stage wild-type embryos had 25% or more GCs outside the clusters (Fig. 2Aa,d). By contrast, ∼45% of AdamTS-A^KO/Df(3R)88F5 embryos (hereafter referred to as AdamTS-A^KO) had 25% or more of their GCs outside the clusters (Fig. 2Ab,d). GCs do not detectably express AdamTS-A, but migrate in close apposition to the CVM, which does (Fig. 1Bb,d-g; supplementary material Fig. S2), suggesting that ADAMTS-A secreted from the CVM could affect GC migration. Interestingly, expressing wild-type AdamTS-A in only the GCs completely rescued the GC migration defect relative to wild type, and to the GC-GAL4 and UAS-AdamTS-A controls (Fig. 2Ac,d; compare bar 8 and bars 1, 3 and 7, supplementary material Fig. S3A-E′). Expressing wild-type AdamTS-A just in the CVM reduced the GC migration defect compared with control animals (bar 6), but only significantly when compared with the UAS-AdamTS-A (bar 3), and not CVM-GAL4 (bar 5) controls (Fig. 2Ad; supplementary material Fig. S3F-J″′). These results suggest ADAMTS-A works better for GC migration when expressed tissue autonomously, although it is presumably normally provided non-autonomously by the CVM (based on AdamTS-A mRNA expression). To determine whether protease activity is required, we tested an enzyme-dead version, which did not rescue GC migration as well as the wild-type protein did (Fig. 2Ad, bar 9).

CVM cells migrate in two rows in a posterior to anterior direction along the length of the embryo, as shown with croc-lacZ, a CVM marker (Fig. 2Ba; supplementary material Fig. S2, Fig. S3F-J″′) (Häcker et al., 1995). In AdamTS-A mutants, many CVM cells migrate away from these two rows, with a greater than 40% increase in the number of embryos displaying mis-migrating CVM cells (Fig. 2Bb,d). Confocal imaging of the CVM revealed an even more severe phenotype than suggested by the HRP-stained samples (supplementary material Fig. S4, arrowheads) (Fig. 2Ba-d). As anticipated, expressing AdamTS-A just in the CVM significantly rescued the migration defect (Fig. 2Bc,d, bar 6; supplementary material Fig. S3J-J″′), signifying a tissue-autonomous function for ADAMTS-A in the CVM. Here, also, the enzyme-dead version of ADAMTS-A did not rescue the migration defect (Fig. 2Bd, bar 7), indicating that metalloprotease activity is required for function.

In AdamTS-A mutants, the tracheal VBs (3-6) mis-migrated with a 40% increase compared with wild type (Fig. 2Ca,b). We expected tracheal expression of AdamTS-A might rescue the VB migration defects; however, as AdamTS-A is also expressed in the CVM cells upon which the VBs migrate (supplementary material Fig. S5) (Boube et al., 2001), expressing AdamTS-A just in the CVM might also rescue VB migration. Expressing AdamTS-A in either the trachea or the CVM alone reduced the VB mis-migration defect compared with controls, although only significantly when compared with the UAS-AdamTS-A control (Fig. 2Cd, compare bar 5 and 3, 4, compare bar 7 and 3, 6; supplementary material Fig. S3F-I,K-N). Simultaneous expression of AdamTS-A in the trachea and CVM significantly rescued VB migration relative to all controls (Fig. 2Cd, compare bar 9 and 3, 8; supplementary material Fig. S3F-I,K-N). Although expression in both tissues provides the best rescue of tracheal migration, we do not know whether expression in the CVM contributes directly by providing ADAMTS-A activity to the VBs, or indirectly by rescuing migration of the CVM cells upon which the VBs migrate. Altogether, the results from the GCs, CVM and tracheal VBs reveal that ADAMTS-A is required for migration of several cell types, including cells that migrate as individuals (GCs and CVM) and cells that migrate as part of an intact epithelium (VBs).

To determine how ADAMTS-A functions in collective cell migration, we turned to the embryonic SGs, which express AdamTS-A prior to and during posterior migration. SGs form from two plates of ∼100 polarized epithelial cells each on the ventral surface of parasegment two. Through coordinated cell shape changes, cell rearrangement, growth and migration, the SG primordia form fully internalized and elongated tubes. Neither cell death nor cell division normally occurs during SG development, and, during the entire process, the SG remains fully polarized (Fig. S3V-Y) (Kerman et al., 2006; Maruyama and Andrew, 2012). The basal surfaces of SG cells form the outside of the tube, exhibiting extensive lamellipodial protrusions during migration (Cheshire et al., 2008). The apical surfaces form the inner lining of the tube, directly contacting the developing matrix-filled lumen. Based on previous work suggesting that matrix metalloproteases function to clear a path for migration, we expected defects similar to those observed with integrin mutants, in which the SGs completely fail to migrate and eventually buckle, forming U-shaped tubular structures (Fig. 4F,H,J) (Bradley et al., 2003). We did not observe integrin-like phenotypes with AdamTS-A loss; instead, we observed apical surface defects. Wild-type SGs displayed apical surfaces that were smooth and uniform in diameter at all stages (Fig. 3A-C,G), whereas AdamTS-A^KO SGs had multiple deformations, including rough irregular surfaces at early stages (Fig. 3D,E,G), and constrictions and bulges at later stages (Fig. 3F,G). Rescue of AdamTS-A SGs was complicated by our finding that overexpression of AdamTS-A in otherwise wild-type SGs also resulted in apical irregularities, suggesting some dose sensitivity (Fig. 3G; supplementary material Fig. S3Q-T). Two GAL4 drivers (fkh-GAL4 and sage-GAL4) were used to express AdamTS-A in the SG (Fig. 3G; supplementary material Fig. S3Q-T,V-Y). The best rescue was achieved with either the sage-GAL4 driver at 25°C (Fig. 3G; supplementary material Fig. S3Q-T,V-Y) or the fkh-GAL4 driver at 18°C, in which GAL4 activity and, presumably, expression levels are lower (Fig. 3G) (Brand et al., 1994). As AdamTS-A is predicted to encode a secreted protease, and as the basal surface of the SG contacts the CVM during mid- to late migration of both tissues (supplementary material Fig. S6C-H), we also asked whether CVM-driven expression of AdamTS-A could rescue the AdamTS-A^KO SG defects. CVM-driven expression of AdamTS-A (supplementary material Fig. S3F-I) did not rescue the SG mutant phenotypes relative to any control (Fig. 3G). We conclude that AdamTS-A functions tissue autonomously in the SG.

The defects associated with AdamTS-A^KO are not due to apoptosis, based on anti-CC3 antibody staining where wild type and AdamTS-A mutants show only the apoptosis that normally occurs during embryogenesis (Fan and Bergmann, 2010) (supplementary material Fig. S7; data not shown). We have also ruled out changes in cell fate, based on staining with a range of tissue-specific markers: CrebA, Sage and SG1 for the SG; Trh, Tango and Knirps for the trachea; Vasa for the GCs; and croc-lacZ (βgal) for the CVM (supplementary material Figs S5-7; data not shown). Whereas the defects in the GCs, CVM and tracheal VBs are likely due to mis-migration (Fig. 2), the SG defects do not appear to be. To determine whether the AdamTS-A SG phenotypes depend on migration, we assayed SGs in which AdamTS-A, integrin function, or both, were disrupted. The SG expresses αPS1 (mew - FlyBase) and invaginates dorsally until it reaches the visceral mesoderm (VM), which expresses αPS2 (if - FlyBase). Upon VM contact, the SG normally turns and migrates posteriorly in close contact with the VM, which also expresses LamininA1,2 (wb - FlyBase) (Bradley et al., 2003; Vining et al., 2005). At stage 12, when the distal region of the SG first contacts the VM, slight irregularities in the apical surface were observed in the distal SG in integrin pathway mutants αPS1, αPS2 and LamininA1,2, similar to those seen in AdamTS-A^KO SGs (Fig. 4C,E,G,I; Fig. 3D). As the SG tube continued to elongate, the failure of the SG to migrate in integrin pathway mutants resulted in buckling of the SG tube (Fig. 4F,H,J). At this stage, the SG apical surfaces in the integrin pathway mutants were relatively smooth and uniform in diameter, distinctly different from the SGs of AdamTS-A mutants (Fig. 4D; Fig. 3E). Removing both AdamTS-A and integrin pathway function resulted in the same buckling defects observed in integrin pathway mutants and the same apical irregularities observed in AdamTS-A mutants (Fig. 4L,N,P). Thus, the AdamTS-A phenotypes do not require cell migration. Instead, we propose that the phenotypes associated with AdamTS-A loss are linked to the cell rearrangements required for tube elongation, a process largely unaffected by integrin loss (Bradley et al., 2003). Indeed, staining with DAPI (white) in combination with the apical marker SAS (green) and the basolateral marker α-spectrin (magenta) revealed irregularities in the arrangement of nuclei around the lumen in the AdamTS-A^KO and mew (αPS1); AdamTS-A double mutants (Fig. 4R,S,U) compared with the regular array of nuclei in both wild type and mew mutant SGs (Fig. 4Q,T). Overall, this analysis suggests that AdamTS-A may normally facilitate the cell rearrangements that accompany SG tube elongation during posterior migration.

Our finding that loss of AdamTS-A may affect cell rearrangement during tube elongation raises the issue of where the protease is secreted. To localize ADAMTS-A, we generated and expressed a GFP-tagged version in the SG (Brand and Perrimon, 1993). ADAMTS-A-GFP was detected throughout SG cells, significantly overlapping the endoplasmic reticulum protein Ph4αSG1 (Fig. 5A-A″′), and the Golgi marker GM130 (Fig. 5B-B″′, arrowheads in B″′), indicating that ADAMTS-A is traveling through the secretory pathway. ADAMTS-A-GFP also overlaps with the secretory vesicle marker CSP2 only at the apical surface (Fig. 5C-C″′) and the apical recycling endosome marker Rab11 (Fig. 5D-D″′), suggesting ADAMTS-A is present in vesicles destined for apical secretion. ADAMTS-A-GFP was also enriched at the apical region where the apical membrane protein Crb is found (Fig. 5E-E″′). Finally, very low levels of ADAMTS-A-GFP were detected in the SG lumen (Fig. 5A-F). ADAMTS-A-GFP also overlapped with Vps16, a marker for endosome to lysosome trafficking (Fig. 5F-F″′), consistent with data showing that the level of ADAMTS-A is important for its proper function (Fig. 3G). Too much ADAMTS-A in the SG may result in targeting much of it for degradation. A similar staining pattern was observed with overexpression of an untagged version of ADAMTS-A, using an ADAMTS-A antibody that detects only overexpressed protein (supplementary material Fig. S3U-U″′). Overall, the staining pattern is consistent with ADAMTS-A travelling through the secretory pathway and being secreted apically.

The apical surface of the SG contacts the lumen, which is filled with a fibrillar matrix first detectable during embryonic stage 11 (Myat and Andrew, 2002; Abrams et al., 2006). Thus, both the basal and apical surfaces of the SG contact an ECM. Staining SGs using antibodies against SAS, an apical membrane marker, and α-Spectrin, a basolateral marker, revealed that SG cells are fully polarized in AdamTS-A mutants and, just like wild type, migrate posteriorly (Fig. 6A-D). The staining also revealed that the apical irregularities in the AdamTS-A^KO SGs localize between neighboring cells (Fig. 6D), supporting the idea that, although AdamTS-A mutant SGs extend in the right direction, the apical cell surfaces could be held back as the gland moves forward. To explore this idea further, we measured the lengths of the three distal-most SG cells (A, B and C) in wild-type and AdamTS-A^KO embryos (yellow, pink and blue arrows in Fig. 6C,D, respectively). On average, two of the distal-most cells (cell A, cell B) of AdamTS-A mutant SGs were significantly longer than wild type (Fig. 6D,E), suggesting that the apical surfaces remain attached to the apical ECM as the basal surfaces extend posteriorly during migration. High magnification transmission electron microscopy (TEM) revealed two additional differences in AdamTS-A versus wild-type SGs: wild-type stage 12 SGs had highly convoluted apical surfaces, as previously observed (Myat and Andrew, 2002; Kerman et al., 2008) (Fig. 6J,J′), whereas AdamTS-A mutant SGs had smoother and flatter apical surfaces, perhaps suggesting a decrease in apical membrane turnover (Fig. 6K,K′). Wild-type SGs also had smooth, intact lateral cell membranes basal to the adherens junctions (Fig. 6L,L′), whereas some AdamTS-A mutant SGs showed separations along the lateral cell membranes, with the most extreme example showing significant separations at multiple sites (Fig. 6M,M′), suggesting increased tension between neighboring cells. Consistent with this idea, much higher levels of F-actin were observed at the apical surface and along the lateral surfaces of SG cells of AdamTS-A^KO mutants compared with wild type (Fig. 6U,W). Very high levels of F-actin were also observed at the apical irregularities in later stage AdamTS-A^KO glands (Fig. 6Y). Thus, the actin cytoskeleton appears to be under increased tension in AdamTS-A mutant SGs. Altogether, the apical separations between cells, the flatter apical surfaces and the increased actin levels in the AdamTS-A^KO support a model in which ADAMTS-A functions to release the apical surfaces of SG cells from the apical matrix, allowing cells to more easily rearrange as the tube elongates and moves to its final position.

We propose that ADAMTS-A functions to release SG cells from the apical ECM through cleavage of either apical ECM substrates or molecules that attach the apical membrane to the ECM. Either way, removal of a link between the apical membrane and apical ECM should reduce SG attachment and allow for fluid forward movement. The atypical cadherin Cad99C localizes to the apical surface of the developing SG, making it an ideal candidate for mediating this attachment (supplementary material Fig. S8B-B″) (S.-Y. Chung and D.J.A., unpublished). Indeed, removing Cad99C resulted in a significant rescue of the SG defects in the AdamTS-A mutants (compare Fig. 7C with 7B,D; Fig. 6E; compare Fig. 4V with 4R,S). To determine whether Cad99C is itself an ADAMTS-A substrate, we examined Cad99C levels in AdamTS-A mutant SGs, but found no differences compared with wild type (supplementary material Fig. S8B-C″). We then asked whether overexpression of ADAMTS-A in embryos would result in decreased Cad99C levels. Neither overexpression nor loss of AdamTS-A affected Cad99C levels (Fig. 7E), indicating that Cad99C is unlikely to be a direct substrate of ADAMTS-A. These results suggest a model in which Cad99C links the apical membrane of SG cells to the apical ECM, and that Cad99C binds an ECM component normally cleaved by ADAMTS-A during the cell rearrangements of tube elongation (Fig. 7F).

Here, we report the characterization of the Drosophila ADAMTS secreted metalloprotease ADAMTS-A. AdamTS-A is expressed in several migratory cell types during embryogenesis and its loss results in migration defects. ADAMTS-A appears to function both in the cells in which it is expressed, and in neighboring cells. In the highly polarized epithelial cells of the migrating SG, loss of AdamTS-A results in apical membrane irregularities that are linked to defects in the arrangement of SG cells around the central lumen. The distal-most cells of the migrating SG are over-elongated, and there is increased tension between SG cells, and between SG cells and the apical surface. We show that ADAMTS-A travels through the secretory pathway and co-localizes with the apical recycling endosomal marker protein Rab1l, and with only the apical pools of the general secretory vesicle marker CSP. Based on these findings, we propose that AdamTS-A functions to release the apical cell surface of individual SG cells from the underlying apical ECM, facilitating the cell rearrangements required for tube elongation during collective migration.

The SG migrates as a polarized collective, which, like a single migrating cell, has a leading edge that forms protrusions in the direction of migration, and a trailing edge that must detach from the substrate to allow forward movement. Contraction and migration of the proximal half of the gland allows the SG to detach from the ventral surface of the embryo and continue migrating posteriorly (Xu et al., 2008). Based on these observations, it has been suggested that the entire SG is comparable with a single migrating cell, with the distal region of the organ acting as the leading edge and the proximal region as the trailing edge. Collective cell migration involves not only the complex interplay between the entire collective and local environment, but also interactions among individual cells within the collective (Rørth, 2009). Interactions between the entire collective and environment include responses to guidance signals provided by surrounding cells, as well as formation and dissociation of adhesions between the migrating tissue and its extracellular environment. Within the collective, individual cells must coordinate their responses to guidance information by adjusting their shapes and position so that polarity is maintained throughout migration. Cells within the collective must also regulate both cell-cell and cell-matrix adhesions to allow coordinated movement relative to one another. Indeed, tube elongation requires SG cells to undergo ‘convergence-extension’, during which the cells must not only detach from one another but also from the apical ECM. Detachment of cells from one another depends on the turnover of E-cadherin (Pirraglia et al., 2010). We propose that detachment of SG cells from the apical matrix is facilitated by ADAMTS-A.

Our hypothesis that AdamTS-A functions to detach cells from the apical ECM during collective migration would be a novel function for this family of proteases. Studies of worm GON-1 in collective gonad migration support its role in clearing a path at the front. The failure of the gonads to migrate in the absence of gon-1 is completely rescued by expressing GON-1 in only the distal tip cell (Blelloch and Kimble, 1999; Blelloch et al., 1999), a finding that is difficult to reconcile with worm GON-1 functioning in cell detachment. These two different activities can be explained by differences in which side of the cell secretes the protease and potential differences in substrates. Discovering two different activities raises the issue of how ADAMTS-A functions in cells that migrate as individuals. Although many GCs and CVM cells do mis-migrate in Drosophila AdamTS-A mutants, many cells in both populations migrate correctly. This suggests no general problems in interpreting and following navigation cues but, instead, a problem in the mechanics of migration, consistent with a role for Drosophila ADAMTS-A at either the leading or lagging edge of individual migrating cells. If detachment were delayed, cells might miss short-lived migration cues and follow an aberrant path. Conversely, if failure to clear the path is the problem, cells might instead follow the path of least resistance and end up at the wrong place. Thus, our data do not distinguish between the two proposed mechanisms for ADAMTS-A action in individually migrating cells.

Removing Cad99C suppressed the SG defects in AdamTS-A mutants; our biochemical data indicate, however, that Cad99C is not a substrate for ADAMTS-A cleavage. We propose that ADAMTS-A may instead cleave the secreted ECM protein(s) to which Cad99C is thought to bind to link SG cells to the apical matrix (Fig. 7). This idea is consistent with studies of mammalian ADAMTS proteins and worm GON-1, where both known and suspected substrates are secreted proteins: aggrecan, brevican, versican and procollagen for mammalian ADAMTS family members, and Collagen IV and/or Fibulin for GON-1 (Rodriguez-Manzaneque et al., 2002; Sandy et al., 2001; Zheng et al., 2001; Kubota et al., 2004; Lee et al., 2005; McCulloch et al., 2009; Enomoto et al., 2010). Indeed, it appears that it is the closely related ADAM proteases that localize and function at the plasma membrane, and specialize in cleaving transmembrane proteins, including members of the cadherin and protocadherin family. ADAM10 cleaves E-Cadherin (Maretzky et al., 2005; Solanas et al., 2011) and Protocadherin 12 (PCDH12) (Bouillot et al., 2011). Similarly, ADAM13 cleaves Cadherin 11 (McCusker et al., 2009). In addition to cadherins, ADAM10 also cleaves Notch in both vertebrates and flies (Bozkulak and Weinmaster, 2009; Wang et al., 2007), and Robo in flies during axon outgrowth (Coleman et al., 2010). Thus, our data are consistent with the idea that secreted ADAMTS proteases cleave secreted ECM proteins, whereas the closely related transmembrane ADAM proteases cleave transmembrane substrates.

Substrates for Drosophila ADAMTS-A may depend on cell type. Indeed, a single mammalian ADAMTS can cleave a variety of targets. For example, ADAMTS 1 cleaves aggrecan, versican or procollagen 1 in a tissue-specific manner (Rodriguez-Manzaneque et al., 2002; Sandy et al., 2001; Lee et al., 2005). Different mammalian ADAMTSs can also cleave the same target protein; ADAMTS 1, ADAMTS 4, ADAMTS 5, ADAMTS 9 and ADAMTS 20 have all been shown to cleave versican in different developmental contexts (Sandy et al., 2001; McCulloch et al., 2009; Enomoto et al., 2010). The fly genome does not encode identifiable aggrecan, brevican or versican homologs, although a clear fibulin homolog exists based on blast searches with the mouse or C. elegans proteins (supplementary material Fig. S9A). Interestingly, the fibulin gene (CG31999) is expressed in embryonic tissues, including the VM, the tissue upon which many of the cell types affected by loss of AdamTS-A migrate (supplementary material Fig. S9B-E). Similarly, there are two canonical Collagen I proteins, both expressed and secreted by hemocytes as they migrate throughout the embryo. Identifying the substrates for ADAMTS-A in the SG and other cells that require its activity should provide excellent tools for clarifying the roles of each of the domains of the ADAMTS proteins and provide insight into the role of cell-ECM interactions in development and disease.

We thank G. Berry, R. Fox, C. Hanlon and two anonymous reviewers for critiquing the manuscript. We thank the Bloomington Stock Center, M. Frasch, S. Hayashi and N. Perrimon for fly stocks, and D. Cavener, R. Cohen, D. Godt and H. Kramer for antisera. We thank FlyBase for information on AdamTS-A gene structure and that of the neighboring genes.