PS Integrins and Laminins: Key Regulators of Cell Migration during Drosophila Embryogenesis

During embryonic development, there are numerous cases where organ or tissue formation depends upon the migration of primordial cells. In the Drosophila embryo, the visceral mesoderm (vm) acts as a substrate for the migration of several cell populations of epithelial origin, including the endoderm, the trachea and the salivary glands. These migratory processes require both integrins and laminins. The current model is that αPS1βPS (PS1) and/or αPS3βPS (PS3) integrins are required in migrating cells, whereas αPS2βPS (PS2) integrin is required in the vm, where it performs an as yet unidentified function. Here, we show that PS1 integrins are also required for the migration over the vm of cells of mesodermal origin, the caudal visceral mesoderm (CVM). These results support a model in which PS1 might have evolved to acquire the migratory function of integrins, irrespective of the origin of the tissue. This integrin function is highly specific and its specificity resides mainly in the extracellular domain. In addition, we have identified the Laminin α1,2 trimer, as the key extracellular matrix (ECM) component regulating CVM migration. Furthermore, we show that, as it is the case in vertebrates, integrins, and specifically PS2, contributes to CVM movement by participating in the correct assembly of the ECM that serves as tracks for migration.


Introduction
Cell migration plays a key role in a wide variety of biological phenomena. During embryogenesis, many cells travel substantial distances to reach their final destinations, where they aggregate to form tissues. In the adult organism migration remains prominent in both normal physiological conditions, as well as pathological situations. During this process, a migratory cell first breaks the adhesive bonds with their neighboring cells and surrounding matrix. Concomitantly, the cell establishes new dynamic contacts with the substratum over which it will migrate, to serve as traction points that will propel its movement. This behaviour, an intricately-coordinated and controlled processes in normal cells, becomes destructive and damaging when acquired by cancerous cells. Hence a better understanding of the molecular mechanisms that transform stationary cells into migratory cells would not only be useful to gain a deeper insight of organogenesis, but also help to understand, treat or even prevent cancer metastasis.
Among the adhesion receptors found to be involved in the migration of different cell types, integrins constitute a major family of receptors promoting cell migration. Integrins are heterodimeric receptors consisting of a and b chains that are present and conserved in all metazoan animals. However, while in mammals, eight b and eighteen a subunits have been characterized, the Drosophila genome encodes two bs (bPS and bn) and five as (a1-a5) subunits. The role of integrins in cell migration includes both a structural and a signalling aspect. On one hand, they act as links between the ECM and the actin cytoskeleton, allowing cells to grasp to the substratum and move. On the other hand, they modulate signalling components that control cell migration, such as members of the Rho family of GTPases, focal adhesion kinase, Src kinase, and the Erk and JNK pathways. Finally, during development, integrins have diverse ways of contributing to cell migration. They can be required in migrating cells for their movement and/or in the surrounding cells to assemble an ECM substratum for migration [1].
In the Drosophila embryo, three tissues of epithelial characteristics are known to require integrin function for their proper migration: the midgut endoderm, the visceral branches of the developing trachea and the salivary glands [2,3,4,5,6]. These three cell types use the same substratum for their migration, the visceral mesoderm (vm) [4,6,7,8]. For both the trachea and the endoderm, a requirement for PS2 was demonstrated in the vm substrate [3,4]. In contrast, different integrins are involved on the side of the migrating cells. Thus, PS1 is required in the visceral branches of the trachea [4], while both aPS1bPS and aPS3bPS and, with a limited contribution bn, are required in the endodermal cells [3,5]. PS1 and PS2 functions during cell migration seem to be distinct and specific, as they cannot substitute for each other [3,4]. Because PS1 is expressed in epithelial cells whereas PS2 is found in mesodermal cells, the specificity of the functions may arise from the presence of unique downstream effectors in the different cell types. Alternatively, PS1 integrin function in cell migration could be due to its ability to mediate ligand-affinity interactions necessary to promote the migratory function of integrins. If this were the case, one would expect PS1 integrin to mediate migration in different cell types and not only in epithelial cells. We decided to test this by analysing the migration of the caudal visceral mesodermal (CVM) cells, a group of mesodermal cells that also uses the vm as a substratum.
CVM cells are the progenitor of the longitudinal muscles, the outer sheet of muscles surrounding the midgut endoderm [9,10]. Initially, they are located at the posterior tip of the mesodermal germ layer. At the onset of germ band retraction, they split into two bilaterally symmetrical clusters on each side of the embryo and begin to move anteriorly over the visceral mesoderm. During their last phase of migration CVM cells spread regularly over the visceral mesoderm, acquire a spindle shape and form regularly spaced longitudinal fibers ( Fig. 1; [9,10]. In mutant embryos where the vm is abnormal or absent, such as twist or binou (bin) mutants, the CVM cells internalize and arrange in two clusters ventrolaterally but cannot migrate anteriorly [10,11]. These results have led to the proposal that the vm serves as a substratum for CVM migration. Furthermore, although it has not been formally proven, the current model is that cells do not migrate over the vm directly but rather they might use an unidentified ECM substratum present on the vm [12].
In this work, we show that PS integrins are required for CVM migration. CVM cells lacking the PS integrins constantly form and retract protrusions without efficient movement, suggesting a PS function for the adhesion of protrusions to the substrate. We demonstrate that while PS2 is required in the vm, PS1 is expressed and specifically required in CVM cells. Since these motile cells are of mesodermal origin, our results strongly suggest that PS1 is capable of mediating cell migration irrespective of the origin of the tissue. Furthermore, we show that PS1 function in both mesodermal and epithelial migrating cells is highly specific, as it cannot be replaced by PS2. This specificity resides both in the cytoplasmic and the extracellular domains of the aPS1 subunit. In addition, we show that removal of laminins, and specifically lamininW (a1,2;b1;c1), inhibits CVM migration, demonstrating that CVM cells use an ECM substratum present on the vm for their migration. Furthermore, we show that PS2 integrins contribute to CVM movement by the assembly of an optimal ECM substrate.

Role of integrins during CVM migration
The Drosophila genome contains two integrin b subunits bPS and bn. The bPS integrin, encoded by the gene myospheroid (mys), is widely expressed while bn is mostly expressed in the midgut endodermal cells [5]. The bPS subunit forms heterodimer with all five aPS subunits encoded by the Drosophila genome. Since only ab heterodimers are transported to the cell surface [13], elimination of the bPS subunit results in a complete absence of all PS integrin function. Thus, in order to study the possible role of PS integrins during CVM migration, we first examined CVM migration in embryos lacking the bPS subunit. Around the onset of germ band retraction, CVM cells split into two bilaterally symmetrical clusters on each side of the embryo and by late stage 11, they begin to move anteriorly over the vm ( [9,10], Movie S1 and Fig. 1). During subsequent stages, the clusters migrate along the dorsal and ventral edge of the midgut in close contact with the vm until, by stage 13, they reach the midgut-foregut transition, where the proventriculus inserts ( [9,10] and Fig. 1 and Fig. 2A, B). Finally, as the midgut encloses the yolk, CVM cells acquire a spindle shape, stretch and form regularly spaced longitudinal oriented fibers, which spread regularly over the underlying circular muscles ( [9,10] and Fig. 1 and Fig. 2C, D). The bPS subunit is maternally deposited in the embryo, thus to completely eliminate PS integrin function, we generated embryos lacking both maternal and zygotic bPS expression (referred to hereafter as bPS 2 mutant embryos, see material and methods). To visualize the morphology of CVM cells, we have expressed cell membrane markers, such as Src-GFP and CD2, in CVM cells using the GAL4 lines G447.2 or 5053A ( [9,14] and Fig. 2). Quantification of CVM migration over the vm was done by scoring all four CVM branches at stage 13, which is when wild type CVM cells have reached the midgut-foregut transition and have stopped their migration. We used the segmental grooves of the embryo as reference points since we noticed that the midgut-foregut transition is at the level of the boundary between the first (T1) and second (T2) thoracic segments. It is worth mentioning here that the initial steps of CVM migration seem to be independent of the vm. CVM cells can internalize, arrange in two clusters ventrolaterally and disperse over the posterior midgut in embryos lacking the vm, such as twist or biniou (bin) [10,11]. By analyzing bin mutant embryos at stage 13, when germ band retraction is complete, we found that CVM cells become positioned at the level of the fourth abdominal segment, A4 ( Figure S1B). In this context, we can use A4 as a landmark for the position that CVM cells reach independently of their movement over the vm. Taking this into account, we used the following criteria to quantify CVM migration over the vm: no migration indicates CVM branches which have stopped at the level of A4, strong delay indicates CVM branches that stay in A3 and A2, mild delay corresponds to CVM branches that stop their migration at the level of A1 and T3 segments and no delay refers to CVM branches that have reached the T1-T2 boundary (Fig. 2L). Using this criterion, we found that 87.5% of wild type CVM branches show no delay and 12.5% show a mild delay (Fig. 2K). In contrast, even though CVM cells can invaginate and split in two clusters in bPS 2 mutant embryos, 1,8% of CVM branches do not migrate, 25.8% show a strong delay, 60.4% show a mild delay and only 12% migrate as wild type (Fig. 2E, F, K). These results show that while integrins are not required for the initial vm-independent migration of CVM cells, they play an important role during their migration over the vm. Later in development, at stages 15 and 16, integrins are also required for the close apposition of the longitudinal muscles to the vm and for proper spreading of the longitudinal fibers (Fig. 2G, H).
The migration defects observed in bPS 2 mutant embryos could be attributed to a failure in CVM differentiation. The normal expression in these mutants of reporter genes driven by two independent CVM GAL4 lines, G447.2 or 5053A (see above), argues against this. Nevertheless, we further tested this possibility by analyzing the expression of other CVM markers, such as bHLH54F-LacZ, dorsocross-GFP (doc-GFP), crocodile-LacZ (croc-LacZ) and the anti-couch-potato (cpo) antibody [15,16], in bPS 2 mutant embryos and found it was not affected ( Figure S2). These results show that loss of integrin function does not affect CVM cell fate.
Although bn is mainly expressed in the midgut, low levels might be expressed in other tissues where it could provide compensatory integrin function. To test this possibility we analyzed CVM migration in the absence of bn and found it was not affected (data not shown). Next, we tested whether the function of bn was masked by the presence of bPS by examining CVM migration in embryos lacking zygotic and maternal bPS and bn integrins. We found that removal of bn function did not enhance the strength of the bPS phenotype, demonstrating that bn is not required for CVM migration (data not shown).
Talin, the first integrin-binding protein identified [17], is essential for most bPS integrin functions [18]. However, the role of Talin in cell migration has only been analyzed in the endoderm, where Talin is required for the early bPS-dependent phase of migration [5]. Thus, we have analyzed CVM migration in embryos lacking both maternal and zygotic contributions of Talin, and found that the phenotype was identical to the loss of bPS ( Figure S1C, D). This suggests that Talin is required for CVM migration.
Integrins are required at different steps of a migratory process: protrusion formation, adhesion, cell traction, de-adhesion and tail retraction [19]. In two Drosophila tissues, the migrating endoderm and the amnioserosa, integrins are required for the formation of protrusions [3,20]. However, we could detect projections in the CVM cells of fixed bPS 2 mutant embryos (Fig. 2E). To better analyze the role of integrins in protrusion formation during CVM migration, we analyzed this process in vivo using a membrane tagged version of GFP, Src-GFP. Wild type CVM cells sent out protrusions as they migrate anteriorly (Movie S2 and Fig. 2I). We found that bPS 2 CVM cells could indeed send out protrusions (Movie S3 and Fig. 2J). However, in contrast to wild type, mutant CVM cells went through cycles of extending and retracting protrusions without an efficient movement (Movies S2, S3). These results led us to propose that bPS function is most likely required for the proper adhesion of the protrusions to the substrate and not for the formation of these structures per se. aPS1 and aPS2 cooperate to mediate CVM migration Three out of the five a subunits, aPS1 (encoded by the gene mew, [21]), aPS2 (encoded by the gene if, [22]) and aPS3 (encoded by the gene scab, [23]), have been implicated in different migratory processes during Drosophila embryogenesis, such as the migration of the midgut endoderm primordium, the salivary glands and the tracheal visceral branches [2,3,4,5,6]. In order to test the specific requirements of these a subunits in mediating CVM cell migration, we analysed the phenotype of embryos lacking the different individual aPS subunits. These a subunit genes do not have any detectable maternal contribution [21,23,24]. We found that in embryos lacking aPS1 1,1% of CVM branches do not migrate, 17,5% show a strong delay, 53,1% show a mild delay and only 28,3% migrate as wild type (Fig. 3A, B, D, E, and M). Similarly, in aPS2 2 mutant embryos, 17.2% and 53.8% of the CVM branches show a strong and a mild delay, respectively, and 29% migrate normally (Fig. 3A, B, G, H, M).
In addition, we found that aPS1 2 aPS2 2 double mutant embryos showed a more severe delay in cell migration than the single mutants alone, phenocopying the migration defects seen in bPS 2 mutant embryos (Fig. 3K, M). Later, attachment and spreading of fibers is also affected in both single and double mutant embryos (Fig. 3F, I, L). We could not detect any defect in CVM migration in embryos mutant for aPS3 (data not shown). Taken altogether, these results suggest that aPS1 and aPS2 are the a subunits that mediate bPS integrin function during CVM migration.
aPS1 is expressed in CVM cells during their migration CVM cells originate from the posterior tip of the mesoderm anlage. Here, we show that both aPS1 and aPS2 subunits are required for CVM cell migration over the vm. However, previous analysis have shown that while aPS2 is expressed in all mesodermal derivatives, including the vm, aPS1 is only expressed in the embryonic ectoderm and endoderm derivatives [13,22,25]. How can we then explain the role of PS1 in CVM migration? To answer this question, we decided to re-examine aPS1 and aPS2 expression patterns during CVM migration. To visualize CVM cells, we used the CVM specific marker croc-LacZ [15]. We could detect expression of aPS1 mRNA in CVM cells before germ band retraction, when cells migrate into the space between the germ band and the posterior midgut anlage (Fig. 4A). CVM cells continue to express aPS1 as they separate into two lateral populations and start their migration over the vm (Fig. 4B, C). This expression is maintained through stage 13, when CVM cells become distributed along the trunk region of the embryo (Fig. 4D). The expression persists as the cells begin to surround the midgut and it disappears by stage 14. We could not detect expression of aPS1 in the vm. In contrast, aPS2 was found in the vm, as previously reported, but not in CVM cells (data not shown).
These results show that aPS1 integrin is specifically expressed in cells that migrate over the vm independently of their origin, suggesting that PS1 integrins could have specialized to accomplish the migratory function of integrins. We next decided to test this by analysing the ability of aPS2 to substitute for aPS1 during cell migration.

Specific requirements for aPS1 function in migrating cells
The above results suggest a model in which PS1 integrin is required in motile cells for their migration. If this were the case, we should be able to rescue the migration phenotype of mew mutant embryos by expressing the aPS1 subunit specifically in the migrating cells. In order to test this, we tried to rescue the CVM migration phenotype of aPS1 2 mutant embryos by expressing the aPS1 subunit in these cells using the GAL4 line G447.2. We found that the migration defect is substantially rescued, as now 71.5% migrate as wild type and we do not find any branch with a strong delay (Fig. 5C, G). We next tested the specific requirement of PS1 in CVM cells by assessing whether the PS2 integrin could fulfill the role of PS1. Expression of aPS2 in CVM cells is less efficient than aPS1 at rescuing the defects in CVM migration of aPS1 mutant embryos, as in these embryos there is still a proportion of CVM branches, 7.5%, that show a strong delay and only 42.8% migrate normally (Fig. 5D, G). In all cases, we have obtained similar data using at least two independent UAS lines. These results indicate that it is PS1 function in particular, rather than PS integrin function in general, which is required in CVM cells for their migration over the vm.
Previous reports analyzing the specificity of PS integrin function have shown that exchanging the cytoplasmic domains had no effect in the ability of each a subunit to rescue the embryonic lethality due to its own mutation [26]. From these results, it was concluded that the functional differences between the two a subunits lies primarily in the extracellular domains [26]. Here, we have examined the requirement of the aPS1 cytoplasmic tail during CVM migration. We found that although a aPS1 subunit without its cytoplasmic domain, aPS1Dcyt, or with that of aPS2, aPS1-PS2, can substantially rescue the migration defects, it is less efficient than the normal aPS1 subunit (Fig. 5E, G). These results suggest that even though the functional differences between aPS1 and aPS2 lie primarily in their extracellular domains, the cytoplasmic domains may also play a role. Finally, a aPS2 subunit with the aPS1 cytoplasmic tail, aPS2PS1, behaves similar to aPS2 in its inability to rescue the aPS1 2 mutant phenotype (Fig. 5F, G).
The specificity and requirement of the different domains of aPS1 that we have observed might be particular to CVM migration, due to their mesodermal origin, or they may reflect a general characteristic of aPS1 function during cell migration. In order to test this we repeated the rescue experiments in a population of migratory cells of epithelial origin that also uses the vm as their migratory substrate, the visceral branch (VB) of the embryonic tracheal tree. We had previously shown that aPS1 is also specifically expressed and required in the VB cells for their migration over the vm, as targeted expression of aPS1 in tracheal cells using the breathless-GAL4 (btl-GAL4) line could rescue the visceral branch phenotype of aPS1 2 mutant embryos [4]. Using the same strategy, we have tested the abilities of the different aPS1 chimeras to fulfill the role of aPS1 in the migrating VB cells, and we have found that they are the same as those here described for the CVM cells ( Figure S3). Thus, we have found similar specific requirements for the different domains of the aPS1 subunit in regulating the migration of cells of two different origins. These results support a model in which aPS1 might have evolved to promote the migratory function of integrins, which relies on a combination of interactions with downstream effectors out and inside the cell.
PS2 is required in the vm to assemble a proper ECM substrate PS2 has been shown to be required for the migration of different cell populations over the vm [2,3,4,5,6]. The current model is that PS2 is expressed and required in the vm for these migratory processes. However, the function of PS2 in the vm remains still unknown. PS2 mutant embryos show irregularities in the visceral mesoderm [2,3]. Therefore, the delay in migration observed in PS2 mutant embryos could be explained as a consequence of irregularities in the substratum. Alternatively, as integrins are required for assembly and/or accumulation of ECM components [27,28], PS2 could also be required in the vm to assemble an optimal ECM substrate for migration. To test these possibilities, we examined the organization of ECM components over the vm during CVM migration. LamininW (a1,2; b1; c1) seemed a good candidate as its mRNA was detected in the visceral mesoderm at the time of CVM migration [29], However, we found, in agreement with previous results, that the LamininW protein could not be detected around this tissue until stage 13, when CVM migration has been completed [29]; our own results, data not shown). In contrast, Nidogen (Ndg), another ECM component, was abundantly found closely apposed to the vm from stage 12 onwards (Fig. 6A). Thus, we decided to use Ndg to assess matrix assembly over the vm. We found that in bPS 2 mutant embryos the accumulation of Ndg was impaired and a spot-like distribution was observed (Fig. 6B). Furthermore, aPS2 2 , but not aPS1 2 , mutant embryos also show an aberrant distribution of Ndg around the vm (Fig. 6C and 6D). These results suggest that an important role for the PS2 integrin in the vm is to assemble a proper ECM substrate required for CVM migration.
Finally, we tested the ability of aPS1 to substitute for aPS2 in the visceral mesoderm. We found that expression of aPS1 in the vm is as efficient as aPS2 at rescuing the CVM migration and Ndg distribution defects of aPS2 mutant embryos (data not shown). These results suggest that both PS1 and PS2 integrins have equivalent abilities in providing a proper ECM substrate for CVM migration.
The laminin a1,2; b1; c1 trimer is specifically required for CVM migration As mentioned above, failure of CVM migration in mutant embryos where the vm is abnormal or absent have led to the proposal that the vm serves as a substratum for CVM migration [10,11]. However, a model has been proposed in which CVM cells do not migrate over the vm directly, but rather they use an ECM substratum present on the vm [12]. This model has not yet been proven and the composition of this putative ECM substratum remains unknown. Here, we have tested this model by analyzing the consequences of eliminating several ECM components, such as laminins, collagen IV and perlecan, on CVM migration. To examine the requirements of laminins during CVM migration, we analyzed embryos mutant for the unique Laminin b chain (Lamb), which results in the complete elimination of the two laminins present in Drosophila [30]. We found that in Lamb 2 mutant embryos most CVM cells (81.7%) invaginate but do not move anteriorly (Fig. 7A, B, I). As we have recently shown that the vm of Lamb 2 mutant embryos appears highly disorganized at stage 16 [30], the observed disruption of CVM migration in Lamb 2 mutant embryos could reflect an earlier requirement for laminins in the vm. However, this does not seem to be the case as no morphological defects were detected in the vm of Lamb 2 mutant embryos during CVM migration, stages 10-13 (Fig. 7D). In addition, the normal expression in Lamb 2 mutant embryos of reporter genes driven by CVM GAL4 lines strongly suggests that loss of laminin function does not affect CVM specification (Fig. 7B,  D). Contrary to vertebrates, in Drosophila there are just two laminin trimers: lamininA (a3,5; b1; c1) and lamininW (a1,2; b1; c1). To determine whether CVM migration requires either both laminin trimers or a specific trimer, we analyzed CVM migration in embryos mutant for the individual a subunits. We found that while CVM migration was normal in embryos lacking the a3,5 subunit (encoded by Laminin A, LanA) (Fig. 7E, I), embryos mutant for a1,2 (encoded by wing blister, wb) showed a CVM migration phenotype that phenocopied the one observed in Lamb 2 mutant embryos (Fig. 7F, I). We have recently shown that in absence of all laminins the deposition and assembly of other ECM components, such as Perlecan and Collagen IV, were severely impaired [30]. In this scenario, the role of lamininW during CVM migration could be indirect, whereby lamininW would be required to assemble a proper ECM substrate, or direct, where lamininW is itself required for CVM migration. To test this, we analyzed the distribution of Collagen IV and Perlecan in embryos mutant for Lam a1,2 and found they appeared normal ( Figure S4B and data not shown, respectively). In addition, we found that CVM migration was not affected in embryos mutant for Collagen IV, the collagen IV processing factor SPARC or Perlecan ( Figure S4D and data not shown). Altogether, these results strongly suggest that lamininW (a1,2; b1; c1) is one of the main substrates used by CVM cells for their migration. In addition, similarly to integrin mutant embryos, stg15 lamininW mutant embryos also show defects in attachment and spreading of fibers (Fig. 7G, H).
The CVM migration phenotype of embryos lacking laminins is more severe than that found in integrin mutant embryos (compare Fig. 2F with 7B). An explanation for this result could be that CVM cells use other laminin receptors besides integrins to migrate over the vm. Dystroglycan (Dg) is another laminin receptor that plays a critical role in linking the ECM and the cytoskeleton. We have analyzed CVM migration in Dg mutant embryos and found it was normal (data not shown). In addition, embryos double mutant for integrins and Dg, such as aPS1 2 ; Dg 2 ( Figure S4F) or bPS 2 ; Dg 2 (data not shown), did not show any enhancement of the integrin phenotype. These results show that Laminin function during CVM migration is not mediated by Dg.

Hemocytes are not required for CVM migration
The major sources for ECM molecules in the Drosophila embryo are hemocytes and the fat body [31]. Hemocytes derive exclusively from the head mesoderm and migrate along several invariant paths throughout the embryo [32]. During late stage 11, hemocytes move anteriorly and ventrally, populating the clypeolabrum and gnathal buds, and posteriorly, beneath the amnioserosa, toward the tail of the embryo [32] and Fig. 8A). As it is around late stage 11 when CVM cells begin to move anteriorly over the vm, they could use the ECM proteins produced by the hemocytes that have populated the end of the germ band. To test this hypothesis, we analyzed how CVM cells move relatively to hemocytes (Fig. 8A, B). We found that CVM migration initiates well before hemocytes have entered the tail end of the embryo (Fig. 8A). Thus, as CVM migration takes place in a hemocyte free region, this suggests that the laminin matrix required for CVM migration might not be laid down by migrating hemocytes. To further test this hypothesis, we blocked completely hemocyte movement by overexpressing a dominant negative form of Rac1, Rac1 N17 , and analyzed the consequences on CVM migration [33]. We found that while hemocytes have not left the head region, CVM cells migrate normally (Fig. 8C). One possible explanation for these results is that hemocytes could produce ECM proteins that are diffused throughout the embryo and maintained over the vm by binding to ECM receptors present in this tissue, such as the PS2 integrins. To test this possibility, we analyzed CVM migration in serpent (srp) mutant embryos [34]. In srp 3 embryos, the primordium of the hemocytes invaginates with the ventral furrow, but cells fail to proliferate or migrate and subsequently die. As a consequence, srp 3 embryos are devoid of mature hemocytes. In addition, although fat body precursors are present, they do not proliferate, do not arrange in a continuous sheet and do not express early differentiation markers such as seven-up [34]. Even though srp 3 embryos display strong germ band retraction phenotype [35], we found that CVM cells moved anteriorly (Fig. 8D), in contrast to lamininW mutant embryos where CVM cells do not initiate migration (Fig. 7F). These experiments demonstrate that the main sources of ECM molecules, hemocytes and fat body, are not required for CVM migration.
We next decided to test whether the CVM cells and/or the vm could be the source of the ECM substrate by performing RNAi knockdown experiments. RNAi knockdowns of wb and LanB1 via CVM or/and vm Gal4 drivers of the respective UAS-RNAi constructs (see materials and methods) did not affect CVM migration (data not shown). Unfortunately, we believe these experiments are not conclusive, as RNAi does not always work in embryos. In fact, the embryonic CVM cells seem to be particularly refractory to this treatment. Thus, RNAi Knockdown of other genes known to be expressed and required in the CVM, such as HLH54F and daughterless, has been shown to work in the adult CVM cells but not in the embryo [16].

Discussion
Integrin cell surface adhesion receptors play an essential role in mediating cell migration during development. During Drosophila embryogenesis, integrins are implicated in the movement of three groups of epithelial cells over the vm [3,4,6]. In all cases, migration requires the function of PS1, and in some cases PS3, in migrating cells and an unknown function for PS2 in the visceral mesoderm. Here, we show that PS1 and PS2 integrins are also required for the migration over the vm of cells of mesodermal origin, the longitudinal visceral muscle progenitors or CVM. We show that PS1 is expressed and required in migrating CVM cells and that this function is highly specific, as it cannot be replaced by PS2. Furthermore, we demonstrate that this specificity lies mainly in the extracellular domain. In addition, our results revealed that a function of PS2 in the vm is to assemble an ECM substrate necessary for migration. This function of PS2 is not specific, as it can be substituted by PS1. Finally, we identify the trimer lamininW (a1,2; b1; c1) as the main ECM component supporting CVM migration.
Previous analysis of the capacities of PS1 and PS2 integrins to substitute for each other in several cellular processes have revealed specific requirements for the different functional abilities of the two integrins [2,3,4,26,36,37]. In a classical view, the diversification of the function of a progenitor gene may occur through the partitioning of regulatory sequences in the original duplication event. The differential tissue expression would then lead to functional refinement and diversification [38]. Phylogenetic analyses of a integrins support this model for the PS1 and PS2 families [39]. An ancestral single a subunit would have duplicated to give rise to aPS2, which retained mesodermal regulatory sequences and evolved to perform stable adhesions, and aPS1, which retained ectodermal and endodermal regulatory elements, and became specialized in mediating dynamic adhesions [26]. Our results showing that aPS1 is expressed in a mesodermal derivative suggest that the scenario might not be that simple. In an alternative model, protein functional divergence may have preceded changes in expression patterns [40]. After the duplication event, rapid changes in functionally important sequences of the protein may have gradually led to functional divergence. Then, the two duplicate genes might have undergone degeneration of some of their cis-regulatory motifs. In this context, aPS1 would have diverged to perform a migratory function. Finally, coding-sequence and expression divergence between aPS2 and aPS1 could have been coupled.
We show here that during cell migration, as it is the case for most integrin-dependent developmental events, PS1 and PS2 cannot substitute for each other [3,4,26,41]. Our results demonstrate that the specificity resides mainly in the extracellular domain. PS1 and PS2 integrins show distinct binding specificity for all ligands identified to date. Thus, one way to explain why the aPS2 extracellular domain cannot substitute the aPS1 could be that the interaction between PS2 and its own ligand might promote an adhesion that it is not appropriate for cell migration. The cytoplasmic domain of the a subunits plays important roles in specifying responses upon ligand-binding. For example, while a5 and a2 cytoplasmic domains support collagen gel contraction, a4 cytoplasmic domain promotes cell migration on collagen [42]. Our demonstration that the aPS2 cytoplasmic tail cannot fully replace the aPS1 tail suggests that during cell migration the cytoplasmic domains may also participate by transmitting distinct intracellular responses. Alternatively, the downstream effectors of PS2 may not be present in migrating cells. Finally, as PS2 has been shown to promote cell spreading on tiggrin [43], a third explanation could be that the ECM components that serve as PS2 ligands for migration are not present on the vm.
In contrast, we show here that aPS2 can be replaced by aPS1 in the vm, as it is the case during the migration of the tracheal cells [4]. This result suggests that PS integrin function during the assembly of an ECM substrate necessary for cell migration does not require specificity of the a subunit. This is also the case for integrin function on the regulation of gene expression in the embryo [37].
During amphibian gastrulation, integrins contribute to cell movement in several ways. They are required in migrating cells to provide traction and to transmit guidance signals and in the surrounding cells to assemble an ECM substrate that serves as tracks for migration [44,45]. In Drosophila, integrins have been shown to be required for the assembly of ECM components in several developmental processes, including dorsal closure, wing imaginal disc morphogenesis and maintenance of the stem-cell niche in the gonads. Here, we show that in flies, as it is the case in vertebrates, integrins, and in particular PS2, contribute to cell migration by assisting in the assembly of an ECM substrate. PS2, and not PS1, mutant embryos show irregularities in the visceral mesoderm. Thus, there are at least two possible ways in which PS2 could function in the assembly of an ECM over the vm. PS2 could be required for proper morphogenesis of the vm, which in turn could be necessary for proper expression and/or localization of ECM components. Alternatively, PS2 could directly assemble an ECM essential for both cell migration and proper vm morphogenesis. Experiments in different model systems have demonstrated the importance of the ECM for the migration of different cell populations [46]. We recently showed that removal of all laminins in the Drosophila embryo affects the migration of several cell populations over the vm [30]. Here, we show that laminins are also required for the migration of CVM cells. Furthermore, our results demonstrate that lamininW is the main ECM molecule supporting CVM migration, and that lamininA, perlecan and collagen IV appear of lesser importance. The migration defects observed in embryos mutant for lamininW are more severe than those found in embryos lacking both the PS integrins and Dg. One explanation for this result is that other laminin receptors might be required for CVM migration. Alternatively, laminin function during CVM migration may not only be to provide an ECM substratum, but also to present binding sites for guidance cues necessary for CVM migration. Recent studies in Drosophila showing genetic interactions between laminins and the secreted guidance cue Slit during embryonic cardiac cell migration and axons across the midline support this option [47,48]. We have also found that Ndg, a glycoprotein that forms a non-covalent complex with laminin and collagen IV, accumulates over the vm at the time of CVM migration. Besides a structural role in the generation and maintenance of basement membranes, Ndg, free of laminins, has its own biological functions that include cell migration. Ndg is required for the migration of trophoblasts cells, neutrophils and Schwann cells [49,50]. Interestingly, a3b1 integrins, which is most similar to aPS1bPS, seem to mediate this non-structural function of Ndg [51]. A role for Drosophila Ndg in cell migration awaits the isolation of mutants in this gene. Finally, we show that the main source of ECM molecules in Drosophila, hemocytes and fat body [52,53], do not provide the ECM molecules required for CVM migration, suggesting that there might be an alternative source. The alternative source could be either the vm itself, as lamininW has been detected in this tissue in stage 11 embryos [29,54], or the CVM themselves. This is the case for human keratinocytes, which deposit laminin 332 to promote their linear migration [55].
During embryonic development, many cells use as their migratory routes ECM components found in the interstices and/ or basement membrane surrounding different tissues. The interactions between moving cells and ECM components has to be highly coordinated to guide cells towards their final destinations during the process of organogenesis. The understanding of how this is regulated is still rather fragmentary due to the complexity of the cellular and molecular interactions involved. We have recently shown that in the Drosophila embryo, ECM components, such as laminins, are required for the migration of many cell populations, including endoderm, macrophages, salivary glands, trachea and mesodermal cells [30] and this work). Thus, we can use the power of Drosophila as a model system to gain insights into the developmental roles of the cell-ECM interactions during cell migration. In addition, migrating tumour cells destroy and consequently rearrange the ECM that surrounds them in order to promote proliferation and tumour invasion. Thus, a detailed and comprehensive analysis of the role of the different ECM molecules and their regulation during cell migration is important to further understand not only embryonic development but also tumor metastasis.

Materials and Methods
Drosophila strains and techniques  [65], Rb-Ndg 1/100 [66]. Alexa-conjugated secondary antibodies used were Alexa 488 (green), Alexa 568 (red) (Molecular Probes TM ). For nonfluorescent staining, embryos were incubated in biotinylated secondary antibodies followed by incubation with Elite ABC complex (Vector Laboratories) and revealed with DAB (Gibco-BRL). For fluorescent in situ hybridization, the Streptoavidinfluorescein was used in combination with the Tyramide Signal Amplification kit (NEL700A, Perkin-Elmer). Images were collected with a Zeiss Axioplan 2 microscope or a Leica TCS-SP2 confocal microscope.
To compare rescue abilities of the different transgenes, a mean out of an average of 35 embryos per genotype was calculated. These mean values were ranked transformed and analysed with the analysis of variance (ANOVA). Differences among genotypes were tested using post-hoc Fisher (LSD) [68].

In vivo time-lapse recording
Embryos were collected, dechorionated and mounted on a glass coverslip. They were immersed in Voltalef oil 10S and the coverslip was attached to a steel frameslide (no. 11505151; Leica, Wetzlar, Germany). GFP-fluorescent images were captured using a Leica TCS-SP2 confocal microscope. Movies were assembled from single focal plane images using ImageJ.