The polarity protein Dlg5 regulates collective cell migration during Drosophila oogenesis

Collective migration plays critical roles in animal development, physiological events, and cancer metastasis. However, the molecular mechanisms of collective cell migration are not well understood. Drosophila border cells represent an excellent in vivo genetic model to study collective cell migration and identify novel regulatory genes for cell migration. Using the Mosaic Analysis with a Repressible Cell Marker (MARCM) system, we screened 240 P-element insertion lines to identify essential genes for border cell migration. Two genes were uncovered, including dlg5 (discs large 5) and CG31689. Further analysis showed that Dlg5 regulates the apical-basal polarity and cluster integrity in border cell clusters. Dlg5 is enriched in lateral surfaces between border cells and central polar cells but also shows punctate localization between border cells. We found that the distribution of Dlg5 in border cell clusters is regulated by Armadillo. Structure-function analysis revealed that the N-terminal Coiled-coil domain and the C-terminal PDZ3-PDZ4-SH3-GUK domains but not the PDZ1-PDZ2 domains of Dlg5 are required for BC migration. The Coiled-coil domain and the PDZ4-SH3-GUK domains are critical for Dlg5’s cell surface localization in border cell clusters.


Introduction
Cell migration plays a critical role in embryonic development, wound healing and physiological processes. Clarifying the molecular mechanism of cell migration helps better understand the abnormal cell migration involved pathological processes such as tumor metastasis and inflammation. Cells can move singly, they also can migrate as groups in the form of collective migration [1]. Collective migration has been observed in embryo gastrulation, wound healing and tumor metastasis [1][2][3]. During collective migration, how individual cells interact with each other and maintain coherence as a collective group during migration is poorly understood. The BC migration in Drosophila ovary is an excellent and genetically tractable system to study collective cell migration [2,4,5]. At early stage 9 of Drosophila oogenesis, border cells (BCs) are first specified and selected out from the anterior follicle epithelium that contain following: dlg5 EP2087 /CyO (Szeged stock Center), Ecad-GFP [38], UAS-Par6.mCherry [39], UAS-aPKC [40]. UAS-Dcr2 was kindly provided by Dr. Barry Dickson. The GFP trap lines, Nrg-GFP, Jupiter-GFP, GalT-GFP, ATPalpha-GFP, Lac-GFP and CG31689 CB03239 , were gifts from Dr. Lynn Cooley. USG-Gal4 was kindly provided by Dr. Stephane Noselli [41]. All the RNAi lines were obtained from Vienna Drosophila RNAi Center. The UAS-dlg5.RNAi strains employed were: GD22496, GD46234 and KK101596 (only the first one was shown). The UAS-aPKC.RNAi, UAS-Par6.RNAi and UAS-Arm.RNAi lines were KK100874, KK101015 and KK102545 respectively. Ubi-Dlg5.EGFP, Dlg5-TagRFP-T and all the dlg5 truncated transgenes (Δ1, Δ2, Δ3, Δ4, Δ5, Δ6, Δ7, Δ8, Δ9, C1, C2, C3, C4, N1, N2, N3, N4, M1, M2, M3, M4) were descripted previously [37]. The FRT P-element insertion lethal lines used in this study were obtained from the Howard Hughes Medical Institute. The dlg5 KG748 mutant was backcrossed to isogenized w 1118 for more than 15 generations to outcross the background mutations. Clonal analyses in the adult ovaries were performed using the hs-Flp/FRT systems with GFP or RFP as marker for wild-type cells. Three days after eclosion, adult flies were heat-shocked at 37˚C for one hour every day in three days and then fed with fresh yeast for one day before dissection. Experiments with the temperature sensitive Gal80 ts system were carried out at 18˚C to repress GAL4-mediated transcriptional activation. Three days after eclosion, adult flies were transferred to 29˚C and induced for three days before dissection. Genotype list was provided in S1 Table.

P-element Screen with MARCM system
The MARCM tool yw, hs-Flp, tub-Gal4, UAS-mCD8::GFP; tub-Gal80, FRT40A/CyO was used to generate GFP positive MARCM clones [24]. To perform the screen, the virgins of yw, hs-Flp, tub-Gal4, UAS-mCD8::GFP; tub-Gal80, FRT40A/CyO were collected and mated to FRT-Pelement lethal lines. Then the 2 nd and 3 rd instar larvae from these crosses were heat-shocked at 37˚C for two hours every day in three days. After eclosion, non-curly females were fed with fresh yeast for 2-3 days. Ovaries were dissected in PBS and mounted in medium. For BC migration examination, fluorescent images of stage 10 egg chambers was taken using Olympus BX51 fluorescent microscopy. According to the position of BC clusters at stage 10, each egg chamber was classified into two classes, Normal or Delay. The Normal class was identified when BC cluster reached the border between nurse cells and the oocyte at stage 10. The Delay class was identified when BC cluster did not reach the border and stayed among nurse cells at stage 10.

The P-element based loss of function screen
To better uncover the molecular mechanisms of BC migration, we performed a P-element based loss-of-function screen using MARCM method. In this screen, we utilized 240 P-element lines with insertions on the chromosomal arm 2L from the UCLA "Bruinfly" FRT-lethal P collection [43]. All the 240 lines are homozygous lethal and a FRT site have been recombined with these P-element insertions [43]. Thus, our screen covers about 240 lethal genes. The number of essential genes present on chromosomal arm 2L is estimated to be about 1000. Therefore, the screen we performed was not saturated. The screen data was collected in S2 Table. Of the 240 lines we have tested, 25 (~10%) displayed no MARCM clones, suggesting that the corresponding mutations cause cell lethality and may affect genes required for cell viability. Alternatively the FRT site could be absent for various reasons, making recombination no longer possible. Two mutations, taiman k15101 and Rack1 EY128 , already known to be required for BC migration were identified (S1C and S1E Fig). Taiman is a steroid hormone receptor coactivator and has been reported to be required for BC migration [10]. All of the taiman k15101 mutant clones show BC migration delay (S1I Fig). Rack1 EY128 is a null allele of Rack1, which encodes a cytoplasmic scaffolding protein with seven WD repeats [44]. We have previously shown Rack1 is required for BC migration and morphology maintenance [45]. Two mutations not previously described were also uncovered. They are dlg5 KG748 and Scim13 1 . The two mutations show weak phenotype of BC migration defects (S1D, S1F and S1I Fig). About 30% mutant clones of the two mutant lines show BC migration delay (S1I Fig). The P-element in Scim13 1 disrupts the gene CG31689 that encodes an ATP-binding cassette subfamily G protein. We found that CG31689 is specifically expressed in BC clusters from the analysis of its GFP trap mutant CB03239 [46] (S1G-S1H" Fig). The mutation dlg5 KG748 contains a KG P-element insertion 1158bp upstream of the translation start site of CG6509. CG6509 encodes the Drosophila homolog of dlg5, a member of MAGUK family. MAGUK family proteins mainly function as scaffolding proteins involved in cell junction assembly, apical-basal polarity, signaling transduction and cell proliferation [47][48][49][50]. Dlg5 contains a coiled-coil domain, four PDZ domains, an SH3 domain and a GUK domain. All these modular motifs in Dlg5 are protein-protein interaction domains that are consistent with features of a scaffold protein.

Dlg5 is required for BC cluster migration and integrity maintenance
To validate the dlg5 KG748 phenotype observed in the screen, we utilized the other dlg5 mutation dlg5 EP208 to analyze the BC migration phenotype. The lethal mutation dlg5 EP2087 , which was not covered in this screen, contains an EP P-element insertion 1129bp upstream of the translation start site of CG6509. The two P-element mutations, dlg5 KG748 and dlg5 EP208 , failed to complement each other. Our previous work showed both of the two mutants were hypermorphic alleles of dlg5, and dlg5 EP208 was stronger than dlg5 KG748 [37]. We examined the mutant clones of the two alleles in BC clusters using genetic mosaic methods (Flp/FRT). The results showed that mosaic BC clusters containing clone mutant for either dlg5 KG748 or dlg5 EP208 displayed significant migration delay (Fig 1A-1C'). In wild-type stage 10 egg chambers, all of the BC clusters reached the oocyte border, whereas 33% (dlg5 KG748 ) or 40% (dlg5 EP208 ) of the mutant mosaic BC clusters failed to reach the border by stage 10 (Fig 1A-1C' and 1N). In some cases, the dlg5 EP208 mosaic BC clusters showed disruption of the cluster integrity and two or more parts of the dissociated clusters were found in dlg5 mutant clones (Fig 1D-1E'). All the phenotypes could be fully rescued by dlg5 full-length transgene Ubi-Dlg5.EGFP, which confirmed that defects in BC migration were caused by loss of function of dlg5 [37]. These data demonstrate that Dlg5 is required for BC migration and cell-cell adhesion in the BC clusters. We also performed RNAi analyses in the BC clusters using c306-Gal4. UAS-Dicer2 was used to enhance the RNAi efficiency [51]. c306-Gal4 drives expression not only in entire BC clusters but also in anterior and posterior follicle cells from early stage egg chambers [52,53]. Expressing dlg5. RNAi in both BCs and PCs using c306-Gal4 inhibited BC migration in about 40% of stage 10 egg chambers (Fig 1F, 1G and 1N), which was a little stronger than previously reported [35]. Expressing dlg5.RNAi also caused dissociated clusters (Fig 1H and 1I), which was similar to the phenotype of dlg5 mutants. Expressing dlg5.RNAi in the BC clusters by slbo-Gal4 or USG--Gal4 (a combination of upd-Gal4 and slbo-Gal4) produced similar migration defects (S2 Fig). All the results indicated that Dlg5 is required not only for BC migration but also for adhesion of BCs to each other and to the PCs.

Apical-basal polarity of BC clusters
BC clusters are originally from the anterior follicle epithelium that maintains apical-basal polarity. Turning from epithelial cells to the part-epithelial and part-mesenchymal BCs, their apical-basal polarity is significantly remodeled. Cell-cell adhesions of follicle cells are partially maintained in BC clusters. We have previously shown that Dlg5 regulates the apical polarity of follicle epithelia [37]. In this study, we first examined the apical-basal polarity of wild-type BC clusters in detail. Armadillo (Arm), aPKC and Neuroglian (Nrg) are used as markers for adherens junctions (AJs), apical junctions and septate junctions respectively. At early stage 9, BC clusters detach from the follicle epithelia and invaded the nurse cells [5]. At middle stage 9, BC clusters also undergo rotation during migration. Therefore, both side views and top views (including apical views and basolateral views) of BC clusters were analyzed. In single confocal sections or z-series projections, the asymmetric distribution of aPKC was observed in the BC clusters from early stage 9 to stage 10 (Fig 2A-2J and 2L). At stage 9, aPKC accumulated at the junctions between adjacent BCs and was found at high levels near the apical junction of the BC cluster (Fig 2A-2J). The localization of aPKC was like a claw shape in the projection of confocal z sections of BC clusters (Fig 2F). At stage 10, aPKC were found at high levels at the surface of the BC cluster facing the oocyte (Fig 2L). aPKC were not detected at the contacts between BCs and PCs (Fig 2A-2J and 2L). The adherens junctional protein Arm was detected throughout the lateral junctional surface of the BCs and enriched in both the apical and basal domains of the region between BCs and PCs (Fig 2A-2F and 2K). At some of the apical junctional area, Arm was overlapped with aPKC (Fig 2A-2D). These findings are consistent with the previous studies on the localization of apical proteins Crumbs, Par6 and Baz and the adherens junction component Ecad in the BC cluster [16,17,20]. It has been reported that N-cadherin (Ncad) is expressed in follicle cells in early to mid-oogenesis but disappears at stage 10 oogenesis [55]. In our previous study, Ncad is the most significantly reduced protein among polarity proteins in dlg5 mutant follicle cell clones [37]. However, we found that no Ncad was expressed in BC clusters from early stage 9 to stage 10 since no significant signal was detected by the Ncad antibody in BC clusters (S3 Fig). The septate junctional protein Nrg was detected in the lateral domain of the surfaces between the BCs and the PCs (Fig 2A-2K). From the side view, we found that Nrg-GFP signal complemented with that of Arm (Fig 2A-2D). In the Nrg-GFP localization region, Arm was not enriched (Fig 2A-2D). The schematic diagram in Fig

Dlg5's expression pattern and subcellular localization in BC clusters
To determine the expression pattern and subcellular localization of Dlg5, we utilized three transgenic flies expressing GFP-or TagRFP-T-tagged Dlg5. The genomic construct transgene Dlg5-TagRFP-T and the Ubi-Dlg5.EGFP transgene were previously generated [37]. We have reported that both of the two transgenes are functional and can rescue the lethality of the dlg5 mutants [37]. The Ubi-Dlg5.TagRFP-T transgene generated in this study was inserted at the 51C1 attP site on the second chromosome. To perform lethality rescue assay, we generated a chromosome containing both Ubi-Dlg5.TagRFP-T and dlg5 KG748 . Flies with this chromosome were able to be kept as a homozygous stock for more than 20 generations and the chromosome also complements with the deficiency chromosome Df(2L)BSC242 that deletes the dlg5 locus. This result indicate that the chimera proteins expressed by the construct Ubi-Dlg5.TagRFP-T can function normally even in the almost complete absence of endogenous Dlg5.
First, we utilized the genomic construct Dlg5-TagRFP-T for analysis of the expression and localization of Dlg5 in BC clusters. PCs in the BC clusters could be distinguished by Arm-GFP, Ecad-GFP, Nrg-GFP or Jupiter-GFP markers (Fig 3). We found at different stages of BC migration, Dlg5-TagRFP-T mainly localized at the basolateral surfaces between BCs and PCs (Fig 3A-3J). At stage 8 or early stage 9 before invading, Dlg5-TagRFP-T was not colocalized with Ecad-GFP but colocalized with Nrg-GFP around PCs (Fig 3A and 3B). At early stage 9 or mid stage 9 after invading, Dlg5-TagRFP-T was partially colocalized with Arm (Fig 3C and  3D). This distribution pattern was similar to the Nrg localization (Fig 3E and 3F). At mid stage 9, Dlg5-TagRFP-T was also colocalized with Nrg in the contacts between BCs and PCs (Fig 3E  and 3F). However, there were also some punctate Dlg5-TagRFP-T localization in the junctional region between BCs (Fig 3D-3F, 3I and 3J), suggesting that Dlg5 might be also involved in cell adhesions between BCs. Dlg5-TagRFP-T was strongly localized around PCs but almost not localized in the contacts between BCs and nurse cells (Fig 3G-3J), suggesting that Dlg5 contributes to the cell-cell adhesion inside BC clusters but not the outside. Due to a significantly higher level of expression from the Ubi transgenes than that from the genomic transgene, we also utilized the Ubi transgenes Ubi-Dlg5.EGFP and Ubi-Dlg5.TagRFP-T for analysis of Dlg5's subcellular localization (Fig 4A-4H'). Dlg5-GFP expressed by Ubi-Dlg5.EGFP is colocalized with Dlg5-TagRFP-T expressed by the genomic construct Dlg5-TagRFP-T in BC clusters (Fig 4A), indicating that the subcellular localization of Dlg5-GFP is normal and not altered by Ubi promoter. To mimic the endogenous distribution of Dlg5, we expressed Dlg5-GFP in the homozygous dlg5 EP2087 animals by the Ubi-Dlg5.EGFP transgene (Fig 4C-4F'), and expressed Dlg5-RFP in the homozygous dlg5 KG748 animals by the Ubi-Dlg5. TagRFP-T transgene (Fig 4G). The subcellular localization pattern of the Ubi expressed Dlg5-GFP was very similar to that of the genomic expressed Dlg5-TagRFP-T, displaying a lateral localization in surfaces between BCs and PCs (Fig 4A-4F'). Both simple confocal images ( Fig 4C) and high resolution confocal images (Fig 4E-4F') showed that Dlg5-GFP was partially colocalized with Arm, and almost completely colocalized with Dlg in BC clusters (Fig 4D). At small region of the apical domain, Dlg5-GFP also overlapped with aPKC ( Fig 4C' and 4D'). There were also punctate Dlg5-GFP localization in junctional region between BCs (Fig 4B). From analysis of the colocalization between Ubi-Dlg5.EGFP and Sqh-mCherry and the threedimensional reconstructions of Ubi-Dlg5.TagRFP-T in BC clusters, we found that the strong punctate localization of Dlg5 in surfaces between BCs and nurse cells was due to the strong expression of the Ubi transgenes in nurse cells (Fig 4B and 4H'). Dlg5-TagRFP-T expressed by the genomic construct Dlg5-TagRFP-T also shows no distribution in contacts between BCs and nurse cells (Fig 4A). We previously reported that Dlg5 colocalized with Arm and apical proteins in follicle cells of early stage egg chambers but colocalized with Dlg in follicle cells of stage 10 egg chambers [37]. The distribution of Dlg5 in BC clusters are similar to its localization pattern in follicle cells of stage 10 egg chambers. Besides Nrg and Dlg, we also found Dlg5 was colocalized with the basolateral proteins ATPα (Na + ,K + -ATPase α subunit) and Lac (Lachesin) (S4 Fig). Taken together, these results indicate that Dlg5 is expressed in BC clusters and mainly localized in surfaces between BCs and PCs with similar localization pattern of septate junctional protein Nrg or Dlg. Both the dissociated cluster phenotype of Dlg5 deficiency and the cell-cell contact localization pattern of Dlg5 in BC clusters suggest that Dlg5 functions in cell adhesions and helps keep BCs together in a collective cluster.

Dlg5 deficiency affects apical-basal polarity in BC clusters
Next, we attempted to determine whether apical-basal polarity was affected by Dlg5 deficiency. A previous study using dlg5 RNAi did not find gross defects in the localization of adherens junction and polarity proteins [35]. We examined the localization of the apical protein aPKC, the adherens junctional marker Arm and the basolateral markers Dlg and Nrg in dlg5 mutant clones or dlg5 RNAi BCs (Fig 5, S5 Fig and S6 Fig). We found that the distribution of Arm was disrupted in dlg5 RNAi BC clusters as compared with the controls (single sections in Fig 5A, 5B, 5E and 5F, corresponding z-series projections in S5A, S5B, S5E and S5F Fig). The apically localized aPKC in BC clusters was strongly reduced in dlg5 RNAi BC clusters as compared with the controls (single sections in Fig 5C, 5D, 5G and 5H, corresponding z-series projections in S5C, S5D, S5G and S5H Fig). The misdistribution of Arm and reduction of aPKC were also found in dlg5 EP2087 mutant clones (Fig 5I and 5J). The aPKC reduction phenotype was more severe in dlg5 RNAi BC clusters when combined with dlg5 KG748 heterozygous background (S5I and S5J Fig). The basolateral markers Dlg and Nrg were not affected in dlg5 RNAi BC clusters (S6A-S6D Fig). The basolateral protein Fasciclin 3 (Fas3), which localized at the membrane interface between the two PCs, was also not affected in dlg5 RNAi BC clusters (S6E and S6F  Fig). These data indicate that Dlg5 specifically functions in apical polarity and adherens junctions, and regulates the distribution of aPKC and Arm during BC migration.

The function and localization of Dlg5 in BC clusters are dependent on Arm
The immunostaining results demonstrated that distribution of aPKC and Arm were significantly affected in dlg5 deficient BC clusters (Fig 5). To determine whether they were required for Dlg5's function in BC migration, we overexpressed aPKC or Arm to perform rescue of the BC migration delay that was resulted from RNAi knockdown of dlg5. As mentioned above, expressing dlg5.RNAi in BC clusters caused~40% migration delay (Fig 1F, 1G and 1I). We found overexpression of Arm was able to partially rescue the migration delay caused by dlg5. RNAi, showing a 22% migration delay (Fig 1L and 1I), though the aPKC reduction was not rescued by the overexpression of Arm (S7A, S7B and S7D Fig). Whereas, overexpression of aPKC could not rescue the migration delay caused by dlg5.RNAi, which showed 36% migration delay (Fig 1J and 1I), and the misdistribution of Arm was also not rescued (S7A, S7C and S7E Fig). This result suggests that Dlg5's function in BC migration is partially dependent on Arm. However, overexpression of Arm or aPKC could not rescue the dissociated phenotype caused by dlg5.RNAi (Fig 1K and 1M), suggesting that Arm or aPKC was not sufficient for the cell-cell adhesions in BC clusters. Conversely, to determine whether apical proteins or Arm were required for the distribution of Dlg5 in BC clusters, we overexpressed apical proteins (aPKC and Par6) or performed RNAi knockdown of apical proteins (aPKC and Par6) and Arm in Ubi-Dlg5.EGFP background. We found that overexpression or RNAi knockdown of aPKC or Par6 were not able to affect Dlg5's localization as compared with the UAS-LacZ control ( Fig  6A-6E). Whereas, RNAi knockdown of Arm could apparently affect the membrane localization of Dlg5 in the BC cluster as compared with the UAS-LacZ control (Fig 6A and 6F). In addition, the RNAi knockdown efficiency of Arm was validated by antibody staining (S8 Fig). This result suggested that Dlg5 localization in BC clusters is regulated by Arm. Interestingly, we found neither RNAi knockdown of aPKC and Par6 nor knockdown of Arm could affect the apical localization of Dlg5 in follicle cells (S9 Fig), suggesting that the interaction between Dlg5 and Arm is strictly critical in the migratory BC cells but not in epithelial cells.

Structure-function analysis
Drosophila Dlg5 belongs to the MAGUK superfamily and contains a coiled-coil motif, four PDZ domains, an SH3 domain, and a GUK domain (Fig 7A). To clarify the subcellular localization ability and function of these domains in BC clusters, we utilized a series of dlg5 transgenes that were previously generated to analyze the subcellular localization and the BC migration delay rescue ability of these transgenes [37]. These dlg5 transgenes were inserted in the same attP site and expressed a full set of different Dlg5 fragments or truncations (Fig 7A). We first analyzed the subcellular localization ability of these transgenes in BC clusters (   (Fig 7D and 7F and S10D Fig). On the contrary, expression of the C-terminal MAGUK core domains (C3, PDZ4-SH3-GUK) showed strong uniform membrane targeting including the entire BC-PC surfaces (S10G Fig). Moreover, deletions of the N-terminus (Δ1, deleting the N-terminal Coiled-coil-PDZ1-PDZ2 region), PDZ domains (Δ2 and Δ5, deleting PDZ1-PDZ2 or PDZ3-PDZ4 region respectively), or the linker region (Δ3) could not completely disrupted the BC-PC surfaces localization (Fig 7B, 7C and 7E and S10B Fig). These results suggested that the MAGUK core tandem domains PDZ4-SH3-GUK are required for the BC-PC surface localization. The coiled-coil domain of Dlg5 is a unique feature that distinguishes it from all other Dlg family members. Expression of N-terminal N3 (Coiled-coil) fragment displayed uniform BC-PC surfaces localization (S10J Fig), suggesting the N-terminal coiled-coil domain is important for Dlg5's cell surface localization. The middle M2 (the linker region) and M4 (Linker-PDZ3-PDZ4) fragments were specifically localized to the BC-PC or BC-BC apical surfaces and colocalized with Arm (S10N and S10P Fig), suggesting that the linker region contributes to the Dlg5 regulates collective cell migration during Drosophila oogenesis specific localization of Dlg5 in BC-PC surfaces where Dlg5 partially colocalized with Arm. It has been reported that the emerging function of PDZ domain of MAGUK proteins is to bind the extreme carboxy-terminal cytoplasmic tail of transmembrane proteins [47]. However, we found that specific expression of PDZ1-PDZ2 domains (M1) or PDZ3-PDZ4 domains (M3) exhibited a diffused and cytoplasmic distribution pattern lacking any cell membrane localization (S10M and S10O Fig), indicating that the PDZ domains contain no membrane targeting function on their own. Compared to the M1 localization, M3 showed PC enrichment (S10M and S10O Fig), suggesting that PDZ3-PDZ4 but not PDZ1-PDZ2 contributes to BC-PC surfaces enrichment of Dlg5. Compared to the localization of FL, the deletion of PDZ1-PDZ2 (Δ2) or PDZ3-PDZ4 (Δ5) showed strong punctate aggregation in cytosol (Fig 7C and 7E), which might be due to the disruption of the association between Dlg5 and membrane proteins when PDZ domains were absent. Among the deleting fragments of Dlg5, only the deletion of N-terminal Coiled-coil-PDZ1-PDZ2 (Δ1) and the deletion of C-terminal SH3-GUK (Δ6) displayed no punctate aggregation in cytosol or cell surfaces (Fig 7B, S10C Fig), suggesting that both the N-terminal region and the C-terminal region contribute to the aggregation of Dlg5. This result also indicated a possibility that Dlg5 molecules interact with each other in a headto-tail manner and form large protein complexes that can scaffold and stabilize polarity proteins at the cell surfaces.
We next determined which domains or regions were required for the Dlg5's function in BC migration. We utilized the truncated transgenes to rescue the migration delay phenotype of BC clusters in dlg5 mutant clones (Fig 8). The full-length transgene (FL) could fully rescue the migration delay of dlg5 mutant clones (Fig 8A and 8B). Besides the FL, we found that only the deletion of the PDZ1-PDZ2 region (Δ2) was able to fully rescue the migration delay of dlg5 mutant mosaic clones (Fig 8E and 8L), suggesting the PDZ1-PDZ2 domains are not required for Dlg5's function in BC migration. Deletions of the other regions, such as the Coiled-coil deletion (Δ1), the PDZ3-PDZ4 deletion (Δ5), the SH3-GUK deletion (Δ6), the SH3 or GUK specific deletions (Δ7 and Δ8), all failed to rescue the BC migration delay of dlg5 EP2087 mosaic clones (Fig 8D and 8F-8I), indicating that both N-terminal coiled-coil domain and the C-terminal PDZ3-PDZ4-SH3-GUK domains are required for BC migration. Expression of the Cterminal fragment, C4 (including PDZ3-PDZ4-SH3-GUK), also failed to rescue the BC migration delay (Fig 8C). We previously showed that the expression of C4 was not able to rescue the lethality of dlg5 EP2087 , but could partially rescue the lethality of dlg5 KG748 [37]. From the escapers of the dlg5 KG748 /dlg5 KG748 ; C4/+ flies, we found the egg chambers developed to stage 10 with normal morphology, but all the BC clusters stayed at the anterior of stage 10 egg chambers with no significant migration (Fig 8J-8K'). Immunostaining of the egg chambers showed misdistribution of Ecad and apparent reduction of aPKC. This result indicates that the C-terminal PDZ3-PDZ4-SH3-GUK region cannot replace Dlg5's function in the migration and the polarity maintenance of BC clusters.

Discussion
Our study demonstrated that Dlg5 is required for BC migration and the maintenance of apical-basal polarity of BC clusters, since dlg5 deficiency caused BC migration delay (Fig 1A-1I), reduction of apical polarity protein aPKC, and misdistribution of Arm (Fig 5). We previously shown aPKC is a key polarity determinant in coordinating the function of three distinct cell polarities (apical-basal polarity, front-back polarity and inner-outer polarity) in BC migration [37]. Dlg5 functions as a regulator in BC migration probably through regulating the key polarity determinant protein aPKC in BC clusters [20]. However, the distribution of Dlg5 is affected by RNAi knockdown by Arm but not aPKC (Fig 7), indicating that the localization of Dlg5 is We examined the apical-basal polarity of BC clusters in detail (Fig 2), and described the expression and subcellular localization of Dlg5 in BC clusters for the first time (Fig 3 and Fig 4). We found Dlg5 colocalized with septate junctional proteins and mainly localized at the basolateral surfaces between BCs and PCs. Dlg5 also localized at cell-cell contacts between BCs in a punctate pattern, partially overlapping with the adherens junctional protein Arm and the apical protein aPKC. This result suggested that Dlg5 may act as a scaffold and function in the stabilization of apical polarity components including aPKC in the apical junction, as consistent with its roles in the follicle epithelium [37]. The distribution of Dlg5 in BC clusters was different from that in follicle cells of early stage egg chambers but similar to that in follicle cells of stage 10 egg chambers [37]. Both the dissociated phenotype of dlg5 deficiency and the BC-PC surfaces enrichment of Dlg5 protein suggested that Dlg5 functions in cell-cell adhesions in BC clusters. Furthermore, the distribution of Dlg5 in BC clusters was affected by the RNAi knockdown of Arm but not aPKC or Par6, indicating that the distribution of Dlg5 in BC clusters is regulated by Arm but not apical proteins.
As a MAGUK family member, Dlg5 has many protein-protein interaction domains, such as the Coiled-coil domain, the PDZ domain, the SH3 domain, and the GUK domain that has no catalytic activity [50]. Previous works have determined which domains of mammalian Dlg5 physically interact with junctional and membrane-bound proteins like β-catenin, vinexin, smoothened, and trafficking regulators syntaxin 4 in vitro or in cultured cells [27,29,33]. We previously reported that the PDZ3-PDZ4 domains of Dlg5 were critical for its function in the follicle cells [37]. In this study, we found that both the N-terminal Coiled-coil domain and the C-terminal PDZ3-PDZ4-SH3-GUK domains were required for Dlg5's function in BC migration, and the PDZ1-PDZ2 domains were not necessary for BC migration. The PDZ1-PDZ2 domains might have redundant function with the PDZ3-PDZ4 domains. Lastly, similar to the previous study in follicle cells, we found the middle linker region and the MAGUK core domains (PDZ-SH3-GUK) could be individually targeted to the AJs and the membranes respectively in BC clusters.
The role of human DLG5 in cancer proliferation, migration, and cancer development has been reported [28,56,57]. Human DLG5 may suppress cell proliferation via interacting with p55 [28]. DLG5 is involved in pancreatic caricinogenesis by suppressing the growth of pancreatic ductal adenocatcinoma (PDAC) [56]. A genome wide RNAi screen showed that DLG5 contributes to invasion and metastasis in breast cancer cell [57]. These studies revealed that DLG5 is a potential molecular target for tumor therapy. In this work, we showed that Drosophila Dlg5 promotes collective cell migration by regulating the cluster polarity, providing clues to DLG5's roles in cancer metastasis.
In conclusion, we found the MAGUK family gene dlg5 (discs large 5) regulates the apicalbasal polarity and cluster integrity during border cell migration. Dlg5 is enriched in lateral surfaces between border cells and central polar cells but also shows punctate localization between border cells. The localization of Dlg5 is regulated by Arm. We found that the N-terminal Coiled-coil domain and the C-terminal PDZ3-PDZ4-SH3-GUK domains but not the PDZ1-PDZ2 domains of Dlg5 are required for border cell migration. The Coiled-coil domain and the PDZ4-SH3-GUK domains are critical for Dlg5's cell surface localization in border cell clusters.
were marked by the absence of the GFP or RFP (green). Transgenes and aPKC were marked by red and white respectively. J-K', Escapers from dlg5 KG748 homozygous rescued by C4 showed completely no migration of BC clusters and misdistribution of Ecad-GFP (green, K) and reduction of aPKC (white, K') compared to the wild-type Ecad-GFP (J, J'). L, Quantification of BC migration delay in different genotypes. EP2087, dlg5 EP2087 . DAPI was marked by blue. Scale bars: 20μm. https://doi.org/10.1371/journal.pone.0226061.g008 Dlg5 regulates collective cell migration during Drosophila oogenesis Supporting information S1 Fig. The P-element screen result. A, Schematic diagram of an egg chamber during BC migration. Left, the BC cluster initiates and invades into the nurse cells at early stage 9. Middle, the BC cluster migrates between nurse cells at middle stage 9. Right, the BC cluster reaches the border between nurse cells and oocyte at stage 10. B, A wild-type BC cluster clones reaches the border between nurse cells and oocyte in a stage 10 egg chamber. C-F, BC migration is delayed in P-element mutants, including tainman k15101 (C), Scim13 1 (D), Rack1 EY128 (E) and dlg5 KG748 (F). Green, GFP. Mutant clones are marked by GFP. G-H", The GFP expression pattern of CG31689 CB03239 in ovaries. CG31689 specifically expresses in BCs at both stage 9 (G-G") and