Phosphoinositide 3-kinase p110δ promotes lumen formation through the enhancement of apico-basal polarity and basal membrane organization

Signalling triggered by adhesion to the extracellular matrix plays a key role in the spatial orientation of epithelial polarity and formation of lumens in glandular tissues. Phosphoinositide 3-kinase signalling in particular is known to influence the polarization process during epithelial cell morphogenesis. Here, using Madin–Darby canine kidney epithelial cells grown in 3D culture, we show that the p110δ isoform of phosphoinositide 3-kinase co-localizes with focal adhesion proteins at the basal surface of polarized cells. Pharmacological, siRNA- or kinase-dead-mediated inhibition of p110δ impair the early stages of lumen formation, resulting in inverted polarized cysts, with no laminin or type IV collagen assembly at cell/extracellular matrix contacts. p110δ also regulates the organization of focal adhesions and membrane localization of dystroglycan. Thus, we uncover a previously unrecognized role for p110δ in epithelial cells in the orientation of the apico-basal axis and lumen formation. Kidney epithelial cells grown in 3D culture form polarized cysts in which basal membranes surround a lumen. Peng et al.show that the p110δ subunit of phosphoinositide 3-kinase contributes to the formation of these structures by influencing focal adhesion at the basal membrane.

A pical-basal polarity is a fundamental feature of epithelial cells required for the structural and functional integrity of the epithelial tissue 1,2 . Conversely, the disruption of polarity in epithelial cells is often associated with infectious diseases and cancer 3 . Apical-basal polarity is initiated by signalling from cell-cell and cell-to-extracellular matrix (ECM) contacts 4,5 . Events such as intracellular trafficking 6 , cytoskeletal organization 7 and actions of evolutionary conserved complexes mediate further development of the apical and basolateral membrane domains 8 . Several recent reports have demonstrated that mechanical factors are also important regulators of cell polarity 9,10 . Establishment of polarity in epithelial cells results in the segregation of plasma membrane in an apical domain facing the outside surface of the body, or the lumen of internal cavities, and the basolateral domain oriented away from the lumen 11 .
The phosphoinositide 3-kinase (PI3K) family of lipid kinases is divided into three different classes based on primary structure, regulation and lipid substrate specificity. The class I PI3Ks are the best characterized and are frequently deregulated in cancer 12 . Mammals have four Class I PI3K isoforms, all of which are heterodimeric PI3K enzymes consisting of a regulatory subunit in complex with a 110-kDa catalytic subunit, either p110a (PI3Ka), b (PI3Kb), g (PI3Kg) or d (PI3Kd). p110a and p110b are ubiquitously expressed, whereas p110g and p110d are enriched in cells of haematopoietic lineage 13 . All class I PI3Ks produce the phosphatidylinositol (3,4,5)-triphosphate (PtdIns(3,4,5)P 3 ) lipid that controls a complex cellular signalling network, with crucial roles in regulation of apico-basal polarity and epithelial cell morphogenesis [14][15][16][17][18] . PtdIns(3,4,5)P 3 signals can be terminated through the action of the 3-phosphatase PTEN or the 5phosphatases SHIP1/2 to produce PtdIns(4,5)P 2 and PtdIns(3,4)P 2 , respectively 19,20 . In polarized epithelial cells, PtdIns(3,4,5)P 3 and PI3Ks are mainly present on the basolateral membrane 21 , while PTEN plays a central role in the formation of the apical membrane 22 . We have recently reported that SHIP2 is present at the basolateral membrane and regulates cell polarization 23 .
Although the involvement of PtdIns(3,4,5)P 3 has previously been established in the polarization of epithelial cells, the underlying mechanisms and the function of the specific PI3K isoforms in this process have not yet been addressed. Here we demonstrate the presence of p110d at the basolateral membrane of polarized MDCK cysts. Our data also reveal that p110d is required to orient the apical-basal axis and lumen formation through both focal adhesions and basal membrane organization.
Results p110d activity controls apico-basal axis and lumen formation. All class I PI3K isoforms generate PtdIns(3,4,5)P 3 and thus the kinase activity of individual PI3K isoforms cannot be readily distinguished in cells by detection of their lipid product or by using broad-spectrum PI3K inhibitors such as wortmannin and LY294002.
We therefore tested ATP-competitive isoform-selective inhibitors of PI3K (ref. 24) on apico-basal polarity of MDCK grown in three-dimensional (3D) as a model system. These include PI-103, a multi-targeted inhibitor for p110a/b/d/g, which inhibits p110a most efficiently but also targets p110b at 10 to 30-fold higher concentrations, and the isoform-selective inhibitors AS-605240, TGX115 and IC87114 that target p110g, p110b/p110d and p110d, respectively. The specificity profile of these compounds was previously determined by measuring IC 50 values in vitro against purified PI3K family members 24 . We also used CAL-101, now called Idelalisib, a derivative of IC87114 with increased potency, which is in clinical trials for B-cell malignancies 25,26 .
To test these pharmacological inhibitors on apico-basal polarity, single MDCK cells were grown on matrigel to form cysts. Twenty-four hours after plating, cells were treated or not with the different PI3K inhibitors for 72 h, followed by confocal microscopy analysis. In non-treated control cysts, b-catenin staining revealed a monolayer of polarized cells surrounding an open central lumen as visualized by the apical marker podocalyxin (PCX; Fig. 1a, left panel). Following the localization of PCX enabled the investigation of the impact of PI3K inhibition on the establishment of epithelial apico-basal polarity. The four major phenotypes (multi-lumens, no lumen, small and inverted polarized cysts) obtained following treatments with the different inhibitors are represented in Fig. 1a. The percentages of each phenotype depending on the concentrations of inhibitors were gathered in the Supplementary Fig. 1 and presented as pie charts in Fig. 1b. Cysts treated with PI-103 showed a dose-dependent change in lumen formation and cyst growth. At a concentration of 10 mM, 62.33% of the cysts presented no lumen and were very small in size. AS-605240, a selective p110g inhibitor, had effects similar as PI-103 even though the number of small cysts was reduced (36.68% at 10 mM). Treatment of cysts with TGX115 led to the formation of cysts with multi-lumens and no lumen. Moreover, a small proportion (6.52% at 10 mM) of cysts with inverted polarity was seen, detected via the persistence of PCX localized at the cyst periphery. Treatment with the p110d inhibitor IC87114 increased the number of cysts with inverted polarity significantly in a dose-dependent manner, from 3.14 to 34.31% at 0.1 mM and 10 mM of IC87114, respectively. These data were confirmed using CAL-101 (Idelalisib), another specific inhibitor of the p110d. Together these data indicated that all the isoforms of class I PI3K participate in the establishment of epithelial cell polarity. However, only the inhibition of p110d elicited an inversion of apico-basal polarity in a population of cysts, which is suggestive of a specific role of the latter isoform of PI3K in the establishment of the apico-basal polarity axis.
To validate this proposition, we further investigated the effect of IC87114 at 10 mM on the different stages of lumen formation (Fig. 2a) as described in a previous report 27 . As presented in the top panels of Fig. 2a when plating individual MDCK cells in 3D culture at the early stage of cyst formation, PCX initially localized at the ECM-contacting surface in early aggregates with b-catenin accumulating at cell-cell junctions. As cysts develop polarity, PCX is displaced at the centre of the cell-cell contact at a site called the apical membrane initiation site (AMIS). The lumen then starts to develop at a site termed the 'pre-apical patch' (PAP) at 48 h of culture and becomes visible at 72 h and then continues to develop as seen at 96 h of culture (Fig. 2a). In striking contrast to control cysts, on treatment with IC87114, PCX was found to remain at the ECM surface until 96 h, while the cells continued to proliferate and maintain cell-to-cell contacts as shown by bcatenin and nuclei staining (Fig. 2a). We also analysed the localization of the tight junction protein Zonula occludens protein 1 (ZO-1) at the early stages of lumen formation. As observed for PCX, treatment with IC87114 resulted in the inappropriate accumulation of ZO-1 at the cell periphery, while tight junctions formed at the apico-lateral boundary in control cysts (Fig. 2b). Statistical analysis of the impact of p110d inhibition on the observed lumen morphologies indicated a significant decrease in the proportion of cysts with a central lumen (or lumen precursor) regardless of days in culture (Fig. 2c). These data strongly suggest that the enzymatic activity of p110d is required to orient the apico-basal polarity axis and to drive the formation of AMIS, which is necessary for lumen formation.
p110d is expressed at the basolateral membrane. The results described above indicated a crucial function of p110d in establishment of cell polarity. p110d is known to be mainly expressed in cells of the haematopoietic lineage, and although several studies have reported expression in epithelial cells 28,29 , its possible role in epithelial morphogenesis has not been determined. Thus, we decided to further characterize the p110d protein expression and localization in MDCK cells.
First, we examined by immunoblot the expression of p110d in MDCK cells using as a positive control a lysate from HeLa cells, a non-leukocyte cell that is known to express high levels of p110d (refs 13,30). A 110-kDa protein was detected in both cell lines; however, the intensity of the signal was very weak in MDCK cells (Supplementary Fig. 2a; the specific band is marked with an arrow), which is in line with the quantitative reverse transcription-PCR (qRT-PCR) analysis for p110d mRNA abundance in these cells ( Supplementary Fig. 2b). The amplified qRT-PCR product from MDCK cells presented a product of the same size as the product isolated from HeLa cells ( Supplementary  Fig. 2c) and with similarity to the p110d canine sequence, as shown in (Supplementary Fig. 2d). Then we examined the presence of the different class I PI3K isoforms in MDCK cells using qRT-PCR. Data presented in Supplementary Fig. 2e indicated that the p110b isoform was the most highly expressed. Together these results indicated the presence of the p110d in MDCK cells. Next we validated the antibody and the subcellular localization of p110d in polarized MDCK cells through transfection of the cells with different p110d constructs. First, MDCK cells were transfected using an untagged p110d cDNA (UT-p110d) and the efficiency of the transfection was validated by immunoblot analysis (Fig. 3a) using a p110d-specific antibody (sc: anti-p110d).
In transfected cells, we observed an intense signal at 110 kDa corresponding to the size of the signal observed in the lysate of HeLa cells, used here as a positive control. In parallel, we assessed the subcellular localization of the exogenous p110d protein in transfected cells plated to form cysts in matrigel culture (Fig. 3b). Staining of the aforementioned cysts with one of the commercially available p110d-specific antibodies (sc: Anti-p110d) revealed a basolateral signal in both control and the untagged p110d cDNA transfected cells, as expected. As anticipated, based on the results of the immunoblotting analysis using this antibody (Fig. 3a), the signal was weaker in control cells when compared with the transfected cells ( Supplementary Fig. 3). Similar results were obtained using a different commercial antibody raised against p110d (ab: anti-p110 d; Supplementary Fig. 3; Fig. 3b). On the basis of these immunolocalization data, it was determined that the antibody from Abcam was more appropriate for downstream immunofluoresence studies because of the higher signal intensity achieved using this reagent in these cells. Next we used Myc-tagged p110d cDNAs expressed from two different expression vectors (Myc1-p110d, Myc2-p110d, see methods). Immunoblot analysis revealed that the two exogenous proteins were detected with both p110d-and Myc-specific antibodies as shown in Fig. 3c and Supplementary Fig. 4a, respectively, for Myc2-p110d and Myc1-p110d. Furthermore, we used the Myc tag and p110d antibodies to analyze by immunofluorescence the cysts from cells transfected or not (Ctrl) with Myc2-p110d cDNA. The two antibodies specifically labelled the Myc-tagged-p110d protein at the basolateral membrane (Fig. 3d). The signal of p110d expression in the Myc2-p110d expressing cysts was measured by ImageJ analysis and compared to endogenous p110d signal (Fig. 3e). These data confirmed a significantly increased signal intensity of p110d on enforced expression of exogenous protein (Fig. 3e). 3D reconstructions for Ctrl and Myc2-p110d cysts were presented in Fig. 3d. Basolateral localization was also observed for Myc1-tagged p110d ( Supplementary Fig. 4b). Thus, together these data indicated that p110d-specific antibodies detected both endogenous and the exogenous proteins. We concluded that the localization of exogenous p110d is similar to the subcellular distribution observed for the endogenous protein at the basolateral membrane. Co-staining of p110d in Myc2-p110dexpressing cysts with b-catenin indicated that the distribution pattern of p110d is different from that of b-catenin, which is mainly present in the lateral domain ( Supplementary Fig. 4c).
p110d membrane localization depends on its enzymatic activity.
In subsequent experiments, we analysed p110d protein cellular localization during cystogenesis (Fig. 4) and we confirmed its basolateral staining at the different stages of cyst formation (Fig. 4a). Treatment with IC87114 induced the displacement of p110d from the membrane at the ECM contact site to the cytoplasm as demonstrated by analysis of the p110d staining signal using imageJ (see Methods; Fig. 4b,c). We next sough to evaluate the effect of IC87114 on the exogenous Myc-tagged-p110d membrane localization (Fig. 4d). Similar to the endogenous protein, IC87114 treatment significantly decreased the membrane signal of Myc with a concomitant shift into the cytoplasmic area (Fig. 4e). Finally, we used the kinase-dead mutant (R894P) of p110d (Myc-p110d kinase-dead) to validate the effects of the pharmacological inhibitors. MDCK cells were transfected with the corresponding cDNA and grown on matrigel to form cysts. Myc-p110d kinase-dead protein, revealed with Myc antibody, was significantly localized in the cytoplasm (Fig. 4d,e) demonstrating that membrane localization of p110d is dependent on its enzymatic activity. Of interest, these cells expressing the mutant protein formed inverted polarized cyst (about 29%) with actin present at the cell periphery, while in control cysts actin marks the central lumen (Fig. 4d). These data are analogous to those obtained with treatment of cells using CAL-101 or IC87114 (Fig. 1) and confirmed the requirement of active p110d to orient polarity and allow lumen formation.
PtdIns(3,4,5)P 3 , monitored using a specific antibody, was mainly observed at the basolateral membrane, in line with previous studies using the PH domain of Akt as a PtdIns(3,4,5)P 3 probe 21 . In accordance with the observations made for the p110d protein localization in cysts ( Supplementary Fig. 4d,e), inhibition of its activity significantly reduced the observed PtdIns(3,4,5)P 3 membrane localization with a concomitant shift of the immunoreactive population to a cytoplasmic pool. Taken together, these results strongly suggest that localization of the majority of PtdIns(3,4,5)P 3 at the basal membrane in MDCK cysts is dependent on p110d kinase activity.
p110d controls Rac1 activation and BM assembly. Cell contacts with the ECM trigger numerous signalling events that provide the initial cue to define the polarity axis. The laminins are major and important biologically active components of the basement membrane (BM), a thin layer of ECM underlying the epithelial cells. Several publications have now reported the crucial role of laminins in orientation of apico-basal polarity axis through the activation of Rac1 (refs 31-34). PI3Ks are important regulators of Rac1 (refs 21,35) and its implication in Rac1-mediated regulation of lumen formation has been evoked 36 .
Therefore, and according to the observed effects of p110d inhibition on orientation of the apico-basal axis, we evaluated Rac1 activation and, importantly, it is increased with lumen formation. Rac1 activation was significantly reduced by treatment of cysts with IC87114 ( Supplementary Fig. 5), suggesting that p110d activity could affect activation of Rac1. Next we examined the possible role of p110d activity on laminin assembly. As presented in Fig. 5a, (the xz section was presented below each picture and the 3D reconstruction was shown on the right part of the panel), laminin was visible at the cyst periphery and formed a prominent structure covering the control cysts that disappeared on treatment of cells with IC87114, while laminin was still present in the culture gel. Under such conditions, PCX is externalized (red staining) and further confirms the inverted polarization of the IC87114 treated cysts. The treatment of cysts with CAL-101 led to data similar to those obtained using IC87114. Then, we analysed laminin staining in cysts from MDCK cells expressing the Myc-p110d kinase-dead mutant (Fig. 5b) and no laminin was found surrounding these cysts suggesting that the recruitment of laminin at the site of cell contact with the ECM required p110d activity. Along with laminins, type IV Collagen is a key ECM protein, essential for maintaining the structural integrity of BM 37 . Accordingly, we also analysed type IV Collagen assembly in the system. Of interest, the treatment of cysts with IC87114 or CAL-101 removed the layer of type IV Collagen underlying the cysts as observed in controls (Fig. 5c). Taken together, these results strongly suggested that p110d is necessary for the assembly of the BM at the cell surface and that it probably fulfills this role by regulation of a signalling pathway that involved Rac 1 activation.
ECM regulates p110d expression. To further characterize the interplay between PI3K and ECM, first MDCK cells were transfected with Myc-tagged p110d cDNA (Myc2-p110d) and plated on matrigel to form cysts, then laminin and type IV Collagen distributions were assessed at the different stages of cystogenesis. As presented in Fig. 6a, laminin was visible at the cyst periphery and the signal significantly increased from the AMIS stage (24 h of culture) to large lumen at 96 h of culture (Fig. 6a,b). The laminin layer surrounding the cysts became thicker whatever the stage of polarization was due to the enforced expression of Myc 2-p110d (Fig. 6a,b). These data are consistent with those obtained from the analysis type IV Collagen distribution in these cysts (Fig. 6c,d).
Next we examined the possible role of ECM in regulating the expression of p110d. To this end, multiwell plates were either coated or not with various ECM components, including: type I collagen, laminin, fibronectin or poly-D-Lysine. MDCK cells were plated for 24 h on the different ECM-coated supports and analysis for the impact of different ECM components on p110d expression was performed by immunoblotting. P110d expression is increased in response to cell interactions with all ECM substrates analysed in comparison with expression levels of p110d expression detected on an untreated support (Ctrl; Fig. 6e). Similar data were obtained using RT-PCR analysis of p110d gene expression under these treatments (Fig. 6f). Moreover, we demonstrated that p110d mRNA level increased significantly in cells grown on matrigel (Fig. 6g). Although there were no statistically significant differences between the effects of the different ECM proteins tested in our experimental conditions at the p110d protein level, collectively these data are supportive of a role for ECM components in positively regulating p110d expression.

Depletion of p110d inverts polarity and impairs BM assembly.
To validate the specificity of the observed effects with the pharmacological inhibitors of p110d, we used three different short interfering RNAs (siRNAs; si1, si2 and si3) that isoform specifically reduced the abundance of the p110d transcript, since there was no effect of these siRNAs on the mRNA levels of p110a, b and g (Fig. 7a). The mechanism by which PtdIns(3,4,5)P 3 regulates cellular signalling was first revealed for Akt 38 . Thus, we analysed the effects of p110d expression on the phosphorylation status of Akt. No significant change in pAkt was observed after the depletion of p110d using each of the three different siRNAs and on the stimulation of cells with epidermal growth factor ( Supplementary Fig. 6). Consistent results were obtained on the overexpression of Myc-tagged p110d, strongly suggesting that p110d is not a key signalling molecule for Akt phosphorylation (Supplementary Fig. 6). Similar data were previously observed in endothelial cells 39 .
Next MDCK cells treated or not with siRNA (si3) were grown on matrigel to form cysts. The specificity of si3 was further verified by immunoblot (Fig. 7b). In control cysts, p110d was found at the basal membrane (as described in Fig. 3) and the signal was profoundly reduced in cysts from siRNA-treated cells confirming a specific signal for the antibody (Fig. 7c). Consistent with our inhibitors studies, PCX was located at the cyst periphery in siRNA-treated cysts as evidenced by the 3D reconstruction of the confocal data ( Fig. 7c right panels). The tight junctions marker (ZO-1), normally present at the tip of b-catenin positive cell-cell contacts in control cysts was now found at the basal region of cysts formed from siRNA-treated cells ( Supplementary  Fig. 7). Analysis of confocal data indicated that 35% (±2.76%) of the total cysts presented an inversion of polarity, with PCX present at the external surface (Fig. 7d). These data are consistent with those obtained on treatment of cells with IC87114 or NATURE COMMUNICATIONS | DOI: 10.1038/ncomms6937 ARTICLE CAL-101 ( Supplementary Fig. 1). The percentage of cysts with inverted polarity observed in our study is also consistent with the outcomes (30-50%) recently published in different studies 34,40 . Moreover, in the p110d depleted and inverted polarized cysts, no laminin (Fig. 7e), or type IV collagen (Fig. 7f), was detected at the cyst periphery. Taken together, these data confirmed that p110d is required for the orientation of the apico-basal axis, lumen formation, laminin and type IV Collagen assembly through an inside-out signal.
p110d is required for focal adhesion stability. The spatial cues from cell adhesion to the ECM are mediated primarily through integrins, an adhesion molecule superfamily of heterodimeric  33,43 . The b1 integrin is involved in different combinations with many alpha subunits including a 2 and a 5 showing a preference for collagen and fibronectin receptors, respectively. The a2b1 integrin combinations are also able to interact with laminin 41 . Therefore, we analysed the b1, a 2 and a 5 integrins in p110d-disrupted cells. In control cysts, the staining of b1 integrin, a 2 integrin and a 5 integrin were mainly detected at the basolateral membrane of polarized cysts. Knockdown of p110d expression using siRNA as well as treatment of cells using CAL-101 dramatically reduced the staining for the different integrins and particularly of b1 integrin, where the signal was largely abolished ( Supplementary  Fig. 8).
The binding of integrins to ligand initiates the connection between the cytoskeleton and the ECM and contributes to the recruitment of additional cytoskeletal and signalling proteins such as talin, vinculin and paxillin that create focal adhesions and regulate cell shape and polarity 1,44 . Thus, to further characterize the effect of p110d on ECM signalling and focal adhesions, we examined both paxillin and vinculin expression and organization. In control cysts, paxillin was organized at the basolateral membrane adjacent to laminin, while the protein became more cytoplasmic and diffuse on treatment of cysts with IC87114 (Fig. 8a,b). Similar observations were made for vinculin (Fig. 8c,d), indicating that p110d activity regulates paxillin and vinculin subcellular localization and partly co-localized with them ( Supplementary Fig. 9). We next investigated focal adhesions by growing cells on glass to better visualize these structures. Under such conditions, vinculin localized at the ends of actin stress fibres, whereas in cells treated with p110d siRNA vinculin formed smaller and punctuate foci dispersed in the cytoplasm (Fig. 8e). A prominent effect of p110d siRNA was also observed on paxillin distribution (Fig. 8f). The level of tyrosine phosphorylation of paxillin is known to positively correlate to dynamism of adhesion sites 45 . Decreased paxillin phosphorylation reduces the turnover of focal contacts and their stabilization, which is required for cell differentiation and polarization. Immunoblotting revealed a decrease in the phosphorylation of paxillin on tyrosine-118 (pPaxillin) on p110d overexpression (Fig. 8g), suggesting that p110d could promote the stabilization of focal contacts at the basal membrane.
p110d co-localizes with b-DG and regulates its localization. Although the most intensively studied receptors for ECM compounds are integrins, dystroglycan (DG) is another high-affinity laminin receptor, originally identified in studies on muscular dystrophy 46,47 . DG is a heterodimeric glycoprotein that consists of a transmembrane b-subunit and a non-covalently associated, extracellular a-subunit. The cytoplasmic domain is linked to the actin cytoskeleton, whereas the extracellular domain is implicated in laminin assembly and subsequently shown to play important roles in the cell-ECM interactions, BM assembly, cell polarity and morphogenesis [48][49][50][51] .
Consistent with the absence of ECM at the periphery of inverted polarized cysts, we hypothesized that DG could also be regulated by p110d. In control cysts, b-dystroglycan (b-DG) was clearly visible at the basal membrane at the ECM contacts and some co-localization with p110d was observed, which increased on p110d overexpression (Fig. 9a). In contrast, siRNA-mediated p110d knockdown strongly reduced b-DG membrane localization with a drastic reduction of the co-localization signal (Fig. 9a). Treatment of cysts with IC87114 yielded similar results to those observed using the isoform specific siRNA strategy (Fig. 9b). We further reinforced these observations by using CAL-101 to inhibit p110d enzymatic activity. As shown in 2D and 3D reconstructions, the b-DG signal that outlined cell periphery in untreated cysts disappeared and instead accumulated at the bottom of the cysts (Fig. 9c). Collectively, these data emphasize the functional contribution of p110d in b-DG localization at the basolateral membrane. Next we investigated the existence of a feedback control loop of b-DG on p110d expression and localization. In concert with our predictions, siRNA-mediated knockdown of b-DG led to cysts exhibiting inversion of polarity and to the absence of laminin accumulation at the cyst periphery (Fig. 9d,e). Under these experimental conditions, p110d was found mostly in the cytoplasm (Fig. 9f). In conclusion, these data demonstrate that p110d localization at the basal membrane is also regulated by an outside-in signal from the ECM via b-DG that drives the orientation of polarity axis.

Discussion
PI3Ks transmit signals from transmembrane receptors such as receptor tyrosine kinases and G-protein-coupled receptors to control a diverse array of signalling pathways through spatiotemporal recruitment of multiple effectors containing PI-specific binding domains. Thus, they regulate key cellular processes including, cell migration, proliferation, differentiation and survival and as such are targeted for intensive research across the fields of inflammation, immunity, cancer and metabolic diseases 12,38,52 . PtdIns(3,4,5)P 3 , the major product of class I PI3K, plays a crucial role in the generation of epithelial polarity 16 . In line with these previous publications, our results indicate that all of the isoforms of class IA PI3Ks participate in MDCK cyst morphogenesis and the development of epithelial polarity (Fig. 1). The different phenotypes observed using specific isoform inhibitors (Fig. 1) suggested that each of the isoforms has a particular function at a different step of the intricate polarity programme. Unexpectedly, only inhibition of the isoform delta, the least abundantly expressed isoform in MDCK cells, affects the initiation of lumen formation and orientation of the apico-basal polarity axis.
Through the use of mice with inactivated p110d and also selective inhibitors, such as the IC87114 and CAL-101 compounds, the p110d isoform of PI3K has been extensively studied in leukocytes and is designated as a potential therapeutic target in immunity, inflammation and haematological malignancies [53][54][55][56][57] . Nevertheless, mice lacking PI3Kd have thus far only been studied under basal conditions. Yet, mice lacking PI3Kd have thickened intestinal mucosa with hyperplasic glands and mixed inflammatory infiltration of the lamina propria 54 . Even though these effects were attributed to the neutrophilic inflammation, one could not exclude an effect on the epithelial cells, which composed the mucosa. Thus, further in vivo studies were required to determine the polarity defects associated with PI3Kd knockdown in vivo. The preponderance of studies of p110d function in haematological systems has contributed to the dogma of a restricted role of p110d in cells of haematopoietic lineage. Of interest, the development of PI3K isoform-selective inhibitors 12,58-61 and the determination of its crystal structure 62 has advanced the emerging studies of p110d function in nonhaematopoietic cells. Still, most of these studies are related to p110d expression in cancer where p110d has been shown to be the predominant isoform expressed in different human solid tumour cells including; glioblastoma, neuroblastoma, prostate carcinoma and breast cancer 63 . A recent report indicated a somatic mutation in PI3Kd that is associated with recurrent infections and airway damage 64 . Thus, a better understanding of the regulation and function of p110d in non-haematopoietic cells and their connections to oncogenesis is both warranted and timely.
In the present study, we provided important insights into the fundamental role of p110d in the establishment of apico-basal polarity axis and lumen formation in MDCK cells. It is generally accepted that the acquisition of epithelial polarity depends on signalling generated through cell interactions with neighbouring cells and with the surrounding ECM 4,5,31 . Integrins are critical components of this signalling pathway and are known modulators of PI3K and PtdIns(3,4,5)P 3 levels 33,65,66 . Integrinmediated PI3K activation induces Rac1 activation and thereby organizes the actin cytoskeleton and recruitment of adaptor proteins within adhesion complexes (that contain a large set of proteins, including paxillin and vinculin 67 ). Our data revealed a role for p110d in the establishment of the apico-basal polarity axis acting via outside-in signalling cascade that is triggered by cell adhesion to the ECM where it acts to establish apico-basal polarity axis and lumen formation that are crucial to epithelial morphogenesis. Of note, previous studies have reported the contribution of other isoforms of PI3K, including p110a and p110b, in the control of integrin signalling 66,68 and Rho GTPases 69 . Thus, the contributions of the other PI3K isoforms in epithelial organization were not excluded.
We also demonstrated that the inhibition of p110d activity or knockdown of its expression affected laminin and type IV collagen assembly at the cyst periphery. Moreover, these treatments profoundly reduced the integrin expression and impaired both paxillin and vinculin focal distribution, where instead these proteins were found diffuse in the cytoplasm   (Fig. 8). Furthermore, we showed that different ECM proteins positively regulated p110d expression (Fig. 6e,f). Together these data also demonstrated a role for p110d in inside-out signalling to control basal membrane organization. However, further investigations are necessary to establish the precise influence of the type of ECM component, the concentration and the stiffness of the ECM on p110d expression and on the cell polarization process in general.
Using a specific antibody, we demonstrated the presence of PtdIns(3,4,5)P 3 at the basolateral membrane of polarized cysts as previously reported 21 . Interestingly, the treatment of cysts with IC87114 strongly reduced PtdIns(3,4,5)P 3 accumulation at the membrane. It is noteworthy that the intensity of the staining for PtdIns(3,4,5)P 3 in the inhibitor-treated cells was not impacted, which indicates that p110d is not the only enzyme responsible for PtdIns(3,4,5)P 3 production in our system. However, the displacement of PtdIns(3,4,5)P 3 from the basal membrane on treatment with IC87114 suggested that p110d is the PI3K isoform activated at the site of ECM contact during initiation of the epithelial cell polarization process. These findings are in concordance with our previous description of PtdIns (3,4,5)P 3 as a compass for cell polarization and as a regulator of basolateral membrane formation 21 .
PI3K is a known effector of small family of GTPases including Rac1 and the contribution of PI3K downstream of the laminin signal in the regulation of cell polarity has been reported 36 . However, it must be noted that the majority of the previous studies were performed using the broad-spectrum inhibitors of class I PI3K, including LY294002 or wortmannin, and as such they do not enable the identification of the specific PI3K isoform involved. Here we provide the first evidence that the p110d isoform of PI3K regulates Rac1 activation in the context of apicobasal polarization in epithelia ( Supplementary Fig. 5).
DG is another receptor for laminin that plays critical roles in BM organization and epithelial polarity through the assembly of extracellular laminin 47,48,51 . However, the underlying mechanisms remain poorly understood. This study revealed that p110d regulated the maintenance of b-DG at the basal membrane and reciprocally b-DG controlled membrane localization of both p110d and its lipid product PtdIns(3,4,5)P 3 . DG regulates actin organization and participates in Rac1 activation 70 , suggesting that p110d could act at the crossroads between integrin and DG mediated ECM signalling to drive the cytoskeleton organization and Rac1 activation that are required for spatial orientation of polarity. However, Rac1 does not seem to be the only downstream effector of ECM/integrin signalling during cell polarization 40 . Understanding the source of this apparent discordance regarding the role of Rac1 requires further investigation. There are important changes in ECM composition according to the cell type and the differentiation stage 70 . Moreover, there are different receptors for the myriad ECM proteins and almost 100 different proteins are present in focal complexes, including signalling molecules and adaptors. The apparent complexity of cell-matrix adhesions suggests that only a subset of proteins participate in different signalling pathways, clearly indicating that more studies are required to elucidate the intricacies of the signalling pathways that link cell/ECM contacts to the polarization process.
PAR-1, a downstream target of the aPKC-PAR-3-PAR-6 complex, regulates the basolateral localization of the DG complex 49 . At the early stage of cyst formation, PCX is localized at the cell periphery in contact with the ECM. It was shown that membrane trafficking regulated by Rab proteins in cooperation with Par3-aPKC drives vesicles to the apical surface to initiate lumen formation 27 . We have shown here that inhibition of p110d activity induced maintenance of PCX at the cell periphery (Figs 1 and 3). Thus, one could speculate that PI3Kd signalling could play an important role in the targeting of these apical cargoes through determination of the polarity axis, although this remains to be determined.
In epithelial cells, dynamic remodelling of the ECM is essential for the development and normal tissue homeostasis. However, loss of polarity and uncontrolled ECM remodelling often leads to fibrotic diseases and cancer. Thus, the analysis of p110d in carcinoma, in particular those associated with fibrosis, will be of potential interest for cancer research.
Methods the manufacturer's instructions. For 3D culture, 24 h after transfection, cells were detached with trypsin and plated on Matrigel for 4 days as indicated above.
Immunofluorescent staining. MDCK cells were rinsed with ice cold Dulbecco's PBS (DPBS) and fixed with 3.7% paraformaldehyde for 30 min. Samples were permeabilized and saturated with DPBS supplemented with 0.7% fish gelatin and 0.025% saponin for 30 min at 37°C and then incubated with primary antibodies. After the washes, staining was performed with secondary fluorescent antibodies, phalloidin and Hoechst. Samples were observed with a Leica TCS SP5 AOBS tandem 30 confocal microscope. Images were collected as TIF files and analysed with PhotoShop.
Immunoblot. Cells were lysed in 1 Â lysis buffer (20 mM Tris-HCl (pH 7.5), 100 mM NaCl, 1 mM EDTA, and 1% Triton-X-100) containing 1 Â complete protease and phosphatase inhibitor cocktail (Roche diagnostics). An aliquot of quantitated proteins was resolved on SDS-polyacrylamide gel electrophoresis and then electrotransferred onto polyvinylidene difluoride membrane. After transfer, the membrane was saturated in DPBS containing 0.1% Tween 20 and 5% milk. Primary antibodies were added overnight at 4°C. After washes in the presence of DPBS, appropriate secondary antibodies coupled with peroxidase were added. Immunoblotting was revealed with chemiluminescent peroxidase substrate (Chemiluminescent Peroxidase Substrate-3, Sigma-Aldrich) and exposure on Kodak Biomax Light film or with Gbox. Blot scans are presented cropped in figures and at least one marker position is presented. Uncropped scans are supplied in Supplementary Fig. 10.
Rac1 activation assay. Rac1 activation was quantified using G-LIZA Biochem Kit (luminescence based) from TEBU-BIO as described by the manufacturer. In brief, cells lysates as described above for immunoblot (1 mg ml À 1 for protein concentration) were incubated in the affinity wells for 30 min at 40°C. Antigen-presenting buffer was added for 2 min. Primary and secondary antibodies were incubated for 45 min at room temperature. Intensity was measured by luminometer at 0.3 s with horseradish peroxidase detection reagent. Cell lyses buffer was used as blank control (RLU, relative light unit).
Quantitative reverse transcription-PCR. Total RNA was isolated using RNeasy Mini Kit 50 (Qiagen) and applied to reverse transcription using RevertAid First Strand cDNA Synthesis Kit (Fermentas). The cDNA was analysed by qPCR using Quanti Tect SYBR Green PCR Kit (QIAGEN) with 7500 Fast Real-Time PCR System (Applied Biosystems). Reaction parameters were 50°C for 30 min, 95°C for 15 min, followed by 45 cycles of 94°C for 15 s, 55°C for 30 s and 72°C for 30 s. The triplicate mean values were calculated using GAPDH gene transcription as reference for normalization. The primers used are indicated in Supplementary Table 1.
Image analysis. Images obtained by confocal microscopy were analysed by Icy software for 3D reconstruction. For co-localization analysis, images were treated by ImageJ software, the plugin 'Intensity Correlation Analysis' was used. Co-localization index was obtained via Pearson's correlation coefficient (Rr). The values for Pearson's coefficient range from 1 to À 1. A value of 1 represents perfect correlation, À 1 represents perfect exclusion and zero represents random localization. For intensity distribution in cells, the plugin 'Radial profil' was used. For analysis of a cyst, a circle was defined at the periphery of each cell and the plugin produces a profile plot of normalized integrated intensities around concentric circles as a function of distance from a point in the image, considered here as the centre of the cell. The concentric circles were assembled in three circle bands with equal radius. The intensity in each band was calculated from the total integrated intensities around concentric circles present in the band.