The phospholipid PI(3,4)P2 is an apical identity determinant

Apical-basal polarization is essential for epithelial tissue formation, segregating cortical domains to perform distinct physiological functions. Cortical lipid asymmetry has emerged as a determinant of cell polarization. We report a network of phosphatidylinositol phosphate (PIP)-modifying enzymes, some of which are transcriptionally induced upon embedding epithelial cells in extracellular matrix, and that are essential for apical-basal polarization. Unexpectedly, we find that PI(3,4)P2 localization and function is distinct from the basolateral determinant PI(3,4,5)P3. PI(3,4)P2 localizes to the apical surface, and Rab11a-positive apical recycling endosomes. PI(3,4)P2 is produced by the 5-phosphatase SHIP1 and Class-II PI3-Kinases to recruit the endocytic regulatory protein SNX9 to basolateral domains that are being remodeled into apical surfaces. Perturbing PI(3,4)P2 levels results in defective polarization through subcortical retention of apically destined vesicles at apical membrane initiation sites. We conclude that PI(3,4)P2 is a determinant of apical membrane identity.

T he most common cell and tissue type is epithelium. The simplest epithelium is a monolayer of cells lining a biological cavity, such as a lumen. To generate such tissue, epithelial cells must form distinct cortical domains 1 . In a prototypical epithelium, the apical surface faces the lumen, the lateral surface interacts with neighboring cells, whereas the basal surface interacts with the extracellular matrix (ECM). The basal and lateral domains are contiguous and termed basolateral. The mechanisms controlling protein delivery to, and maintenance at, cortical domains in polarized cells have been extensively studied 2 . How epithelial cells become polarized and form a lumen de novo remains poorly understood, yet it is an outstanding problem in both development and disease.
MDCK cells grown inside ECM to form three-dimensional (3D) cysts have been widely used as a model system of polarization and lumen formation. In 3D, these undergo stereotyped morphogenesis, transitioning from a single cell to an apical-basal polarized monolayer radially organized around a central lumen 3 . During this process, each cell generates apical-basal polarization de novo. A number of polarization mechanisms first demonstrated in MDCK cysts are conserved in vivo [4][5][6][7][8][9][10] . Thus, MDCK cystogenesis is a powerful reductionist system to study epithelial polarization.
Upon 3D plating, single-MDCK cells divide into doublets with inverted polarity; some apical proteins, such as Podocalyxin/gp135 (Podxl), are found at the ECM-abutting surface but excluded from cell-cell contacts 11,12 . Integrin-dependent ECM sensing triggers Podxl endocytosis and transcytosis to the apical membrane initiation site (AMIS), a zone at doublet cell-cell contacts which remodels into the nascent lumen 13 . Remodeling involves conversion of a basolateral domain into an apical protein delivery zone. This stage is titled the pre-apical patch (PAP) 14 . The luminal space expands as the lumen matures. Delivery to the AMIS is regulated by the Rab11a GTPase. Rab11a influences molecular motors and vesicle docking and fusion machinery recruitment to ensure apical protein delivery to the AMIS 12,13,[15][16][17] . Therefore, Rab11aregulated exocytosis to the AMIS is crucial to generate apical polarity 1 .
Phosphatidylinositol phosphate (PIP) asymmetry is essential for cell polarization 18 . PIPs can be modified by reversible phosphorylation of the 3-, 4-, or 5-position of their inositol ring 19 . Asymmetric PIP production at the cortex, or in organelles, determines membrane identity by scaffolding distinct PIPbinding proteins at these locales. In MDCK cysts apical-basal polarization depends on cortical PIP asymmetry regulated by the 3-phosphatase PTEN 11 : PI(4,5)P 2 is apically enriched, whereas PIP 3 is basolateral. This lead to a model proposing PI(4,5)P 2 as an apical identity determinant; this model is problematic, given that PI(4,5)P 2 is the precursor to PIP 3 and is also basolateral 11,18 . Whether alternate PIP species may fulfill an apical-specific function is unknown.
These advances focus attention on the key question of how existing cell surfaces are remodeled. Specifically, what controls cell-cell contact remodeling into an AMIS? We elucidate a molecular mechanism of de novo polarization through cortical PIP conversion to promote apical identity.
To determine when PIP 3 asymmetry emerges, we examined early lumenogenesis. At all stages, Par3 coincides with PIP 3 depletion. At the cell doublet stage, after initially peripheral Podxl internalized and transcytosed near cell-cell contacts, a single-Par3 punctum formed adjacent to each cell-cell contact (Fig. 1c, blue arrowheads), associated with PIP 3 depletion (white arrowheads). As the AMIS formed, the Par3-positive/PIP 3 -depleted zone expanded. At the PAP, PIP 3 was further depleted and Par3 relocalized to the PIP 3 /apical boundary (Fig. 1c, yellow arrowheads; Fig. 1b). This required the 3-Phosphatase PTEN, which localizes to the AMIS during polarity formation 21 . PTEN depletion attenuated lumen formation via subcortically stalling apically destined Podxl vesicles ( Supplementary Fig. 1b-d). Thus, a PTEN-regulated Par3-positive/PIP 3 -depleted zone marks the AMIS, the site for apical vesicle delivery.
We inhibited SHIP1 by chemical and genetic means. To ensure robust polarity phenotype detection, we built a Pipeline for semiautomated polarity analysis, PAPA, quantifying hundreds to thousands of cysts per condition ( Supplementary Fig. 4b), and performing comparably to manual quantitation ( Supplementary  Fig. 1c). SHIP1 chemical inhibition 27 at the time of plating strongly disrupted lumen formation, causing subcortical apical vesicle retention (Fig. 4d, Supplementary Fig. 4c). In contrast, chemical inhibition of SHIP1 after lumens formed failed to perturb cell polarity (Fig. 4d), in line with requirement to convert basolateral PIP 3 to PI(3,4)P 2 only during AMIS formation. Endogenous SHIP1 depletion resulted in similar phenotypes ( Fig. 4e-g). Control cysts displayed luminal Podxl (Fig. 4h, blue arrowheads), PIP 3 reporter or β-catenin at basolateral membranes, and Par3 at the boundary between these zones (Fig. 4h, yellow arrowheads). SHIP1 depletion resulted in defective lumen formation and aberrant retention of PIP 3 , Par3, and β-catenin adjacent to rudimentary lumens (Fig. 4h, white arrowheads). SHIP1 PI(3,4)P 2 -producing activity was required as only expression of RNAi-resistant WT, but not phosphatase-deficient, SHIP1 reversed SHIP1 depletion phenotypes (Fig. 4h, i). In contrast, filter-grown 2D monolayers displayed no defects in apical-basal polarization upon SHIP1 depletion (Supplementary Fig. 4a). SHIP1 is therefore a regulator of de novo apical membrane biogenesis in 3D.
PIK3C2A/B displayed distinct localizations, though both overlapped with different pools of Rab11a. GFP-PIK3C2A was closely associated with Rab11a vesicles throughout polarization (Fig. 6d). A pool of GFP-PIK3C2B was basolateral and a Schematic representation of a Pipeline for semi-automated phosphoinositide intensity analysis, PAPI. MDCK cysts stably expressing GFP-tagged PIP reporters were cultured in 3D for 48-72 h, fixed and stained with Podxl to mark the apical domain, Phalloidin the cortex, and Hoescht the nucleus. Confocal optical sections of MDCK cysts were imaged and automated processing selected the medial plane based on the maximum lumen area. Separated cyst regions were defined based on differential localization of the above markers. PIP probe intensity was measured in each domain, followed by mathematical and statistical analysis to extract the relative PIP probe intensity ratio within compartments of the same object (cyst). b Cysts at the open lumen stage expressing either EGFP-2xPH-TAPP1 (WT) or a mutant EGFP-2xPH-TAPP1 unable to bind PI(3,4)P 2 (ΔPIP), stained for F-actin (red) and nuclei (blue). Luminal (white arrowheads), basolateral (yellow arrowheads), and nuclear (red arrowheads) localization is highlighted in magnified fields. c Quantitation of relative apical to cytoplasm (left), basolateral to total (center) or nuclear to total (right) PIP reporter intensity compared to GFP-overexpressing control MDCK cells. Box-and-whiskers: 10-90 percentile; +, mean; dots, outliers; midline, median; boundaries, quartiles. n ≥ 208 cysts assessed from three wells/condition/experiment, three independent experiments (2 for EGFP-2xPH-TAPP1 WT clone 2). P-values (One-way ANOVA): *P ≤ 0.05, **P ≤ 0.005, ***P ≤ 0.0001. d, e Forty-eight hours MDCK cysts stained for endogenous PI(3,4)P 2 (greyscale or green), F-actin (red) and nuclei (blue) using four different fixation and staining protocols, i 43 , ii 12 , iii 24 , iv 44 . Note that PI(3,4)P 2 can be observed at the luminal domain in all cases (magenta arrowheads). Nuclear localization was also detected in some conditions (iv, e, yellow arrows). f Sequential optical sections of the medial region of an MDCK cyst expressing EGFP-2xPH-TAPP1 (green) stained for endogenous PI(3,4)P 2 (red). Note co-localization in the luminal membrane (white arrowheads) and nuclei. g Inverted polarized MDCK cyst expressing EGFP-2xPH-TAPP1 (green) stained for PI(3,4)P 2 (red) and Hoescht (blue). overlapped with clusters of Rab11a that were close to the cortex (Fig. 6d, e). GFP-PIK3C2B was initially at doublet cell-cell contacts, but became progressively removed from the forming AMIS (Fig. 6f). In cysts with an open lumen, GFP-PIK3C2B could be seen both at the basolateral membrane and in vesicular compartments near Rab11a. Thus, PIK3C2A is constitutively Rab11a-endosome adjacent, while PIK3C2B may be involved in cortical Rab11a events controlling apical polarization.

Discussion
We describe a transcriptionally regulated PIP-modifying enzyme network essential for lumen formation. Notably, the enzymes required in 2D and 3D are not interchangeable, though both conditions are apical-basal polarized. This revealed an unanticipated role for the PI(3,4)P 2 as a determinant of 3D apical identity. Specific to the 3D context is basolateral cell-cell contact remodeling into an apical domain 12 . Several PIP-modifying enzymes participate in this process.
The doublet cell-cell contact contains a PI→PI(4)P→PI(4,5) P 2 →PIP 3 cascade that generates basolateral identity (for model, see Fig. 10a-c). During polarity rearrangement, PTEN locally reverses the last step in the PIP 3 cascade, inducing a PIP 3 depletion zone: the AMIS 11,21 (Fig. 10b). Due to this early apical enrichment, PI(4,5)P 2 was described as an apical targeting factor 34 . However, PI(4,5)P 2 is also basolateral. Our current data suggest an updated model wherein lack of PIP 3 , rather than the mere presence of PI(4,5)P 2 , encodes part of apical identity.
We report that the long-unidentified apical domain identity determinant is PI(3,4)P 2 , which is enriched apically and on recycling endosomes. That PI(3,4)P 2 is asymmetric in localization and function to PIP 3 is unexpected. PI(3,4)P 2 was considered as part of PIP 3 signaling cascades, given that both lipids bind the Akt PH domain in vitro 35 . Such studies have examined PI(3,4)P 2 in single or poorly polarized cells, where apical domains may not occur. The functions and asymmetry of PI(3,4)P 2 may have been underappreciated in such contexts.
If PI(3,4)P 2 is both in recycling endosomes and the apical domain, how can it be an apical determinant? It may be that discrimination between these two PI(3,4)P 2 -positive compartments is provided by the absence or presence of PI(4,5)P 2 , respectively. Such a combinatorial PIP code for membrane identity greatly expands the possible number of compartment identities than can be generated by the seven PIPs alone. In line with this, although exogenous addition of PI(3,4)P 2 enhanced lumen formation efficiency, it could only do so when SHIP1 was functional. The apical identity code may therefore be the presence of PI(3,4)P 2 and PI(4,5)P 2 , and the absence of PIP 3. In support of this, exogenous PIP 3 addition to an apical membrane causes rapid conversion to basolateral identity 28 . The recycling endosome code may rather be PI(3,4)P 2 without PI(4,5)P 2 .
Our studies underpin that whereas there are in theory multiple ways to make a PIP, these are transcriptionally and spatiotemporally regulated during development. What is important to note is that PIP production and morphogenesis go hand-in-hand: inhibiting SHIP1-mediated PIP 3 to PI(3,4)P 2 conversion results in basolateral domains with more PIP 3 , but not with apical domains with less PI(3,4)P 2 . Rather, an apical domain will not form if sufficient apical PI(3,4)P 2 is not generated, or if cortical PIP 3 is not removed. This represents a conceptual difference from 2D studies where the cortex is a constant structure where PIP levels can change. In multicellular 3D contexts, such changes in cortical PIP levels induces alternate morphogenesis, such as conferring apical or basolateral identity.
In addition to their function in cell polarization, phosphoinositides participate in cell growth and survival pathways 18,40 . For instance, correct cell polarization may lay upstream to decisions of cell death or survival during morphogenesis. Likewise, perturbation of PIP-modifying enzymes, such as the PTEN or INPP4B loss observed in cancer 40,41 , may be facilitative of the disrupted polarity and overgrowth of tumors. The extent to which these processes are distinct pathways or are intimately linked remains unclear.
We describe an unappreciated function of PI(3,4)P 2 in apical domain morphogenesis in MDCK cysts. Although an in vitro reductionist system, the molecular mechanisms elucidated in MDCK cysts have been demonstrated as conserved in a variety of other systems [4][5][6][7][8][9] . Yet, there are several developmental mechanisms for lumen formation; the polarized exocytosis and membrane remodeling we describe here can be bypassed by apoptotic cavitation of multicellular clusters or folding and joining of tissue sheets 1,42 . In these instances, we predict that apical PI(3,4)P 2 is likely supplied by the endosomal pool, rather than conversion from basolateral PIP 3 by SHIP1. In spite of differences in the way it can form, the apical domain is essential for exchange of proteins, nutrients, solutes and lipids. Given, for example, that the kidney lumen is an extremely active site of endocytosis in vivo, that PI(3,4)P 2 has emerged as a regulator of rapid endocytosis aligns with such a physiological requirement [23][24][25] . It will thus now be important to determine the involvement and regulation of apical and recycling endosome PI(3,4)P 2 in development and disease.
Antibodies and immunocytochemistry. Cysts immunolabelling was adapted from previously described protocols 12 : Cultures were fixed in 4% paraformaldehyde (PFA, Affimetrix) for 5-15 min at room temperature (RT), washed twice in PBS, Fig. 4 SHIP1 converts PIP 3 -rich basolateral membrane into apical domains. a Conversion between the phosphoinositide species, all derived from phosphatidylinositol (PI), occur via the action of kinases (in green) and phosphatases (in red). Dashed arrows, pathway whose occurrence or regulatory enzyme is still unknown. b Heat map of Manhattan-clustered, differentially expressed PIP kinases and phosphatases in cells grown as a monolayer for 48 h or as cysts in Matrigel for 23 or 48 h. Relative mRNA expression levels (log2 values), and clustering categories, are shaded as indicated. Four independent experiments. c SHIP1 localization during lumen formation in MDCK cells expressing SHIP1-EGFP (green) and stained for Podxl (white) and β-catenin or Factin (red). d Quantitation of cyst phenotypes treated with ethanol control or SHIP1 inhibitor either at the time of plating (d0) and fixed at day 3 (3 day, left), or treated at day 3 (d3) and fixed at day 5 (5 day, right). Values are mean ± s.d. For 3 day, n ≥ 300 cysts assessed from three wells/condition/ experiment, four independent experiments, P-values (two-way ANOVA): **P ≤ 0.001, ***P ≤ 0.0001). For 5 day, n ≥ 1900 cysts assessed from three wells/ condition/experiment, three independent experiments. e RNA extracts from MDCK cells stably expressing scramble or SHIP1 shRNA were analyzed by RT-qPCR to detect SHIP1 mRNA levels (n = 1 well per condition, from four independent experiments). P-value (Student's t-test): *P ≤ 0.05. f, g Western blot of WT and phosphatase-dead SHIP1 and GAPDH in total cell lysates of parental (MDCK) or SHIP1-EGFP-expressing cells expressing scramble or SHIP1 shRNA f, and quantitation of cyst phenotypes g. Mean ± s.d., n ≥ 400 cysts from four wells/condition/experiment. P-values: two-way ANOVA): *P ≤ 0.05, **P ≤ 0.001, ***P ≤ 0.0001. h PIP 3 [EGFP-PH-Grp1], Podxl and Par3 (left panels) or β-catenin (right panels) localization in cysts expressing scramble or SHIP1 shRNA. Yellow arrowheads, Par3 (left panels) or β-catenin (right panels); blue arrowheads, Podxl. White arrowheads, overlap of apical and basolateral domains. i Immunolabelling of above conditions, with Podxl and β-catenin. Two independent experiments for parental MDCK versus SHIP1-EGFP WT cells and one experiment for parental MDCK cells versus SHIP1-EGFP WT cells or SHIP1-EGFP phosphatase-dead cells. Scale bars, 10 µm blocked for 1 h in PFS buffer (PBS, 0.7% w/v fish skin gelatin (Sigma-Aldrich), 0.5% saponin (Sigma-Aldrich)), and incubated with primary antibodies diluted in PFS at 4°C overnight with gentle rocking. Then, cyst cultures were washed thrice with PFS and incubated with secondary antibodies diluted in PFS for 1 h at RT, followed by washing twice in PFS and twice in PBS. Primary antibodies are as described below. Alexa fluorophore-conjugated secondary antibodies (1:250) or Phalloidin (1:200) (both Invitrogen) and Hoescht to label nuclei (10 μg ml −1 ), were utilized. For validation of endogenous PI(3,4)P 2 staining in 3D, we optimized staining protocols. We tested four staining protocols in addition to the commercially recommended protocol (PI(3,4)P 2 Ab, Echelon, Z-P034), (i) 43 , (ii) 12 , (iii) 24 , (iv) 44 . Buffer four produced endogenous labeling patterns mirroring EGFP-2xPH-TAPP1 and was used for all further PI(3,4)P 2 staining. Briefly, staining in buffer four is as follows: cysts were fixed in 4% PFA followed by three washes in glycine buffer (100 mM glycine in PBS), and two washes in PBS. Cysts were permeabilized  in glycine buffer containing 0.1% of saponin for 20 min, and blocked in PBS containing 10% fetal calf serum and 0.1% saponin. Primary and secondary antibody incubation in blocking buffer were performed in the conditions specified above, followed by three washes in blocking buffer. List of used antibodies, in Supplementary Table 1.
Image acquisition and analysis, PAPA and PAPI. Confocal images were acquired either on Zeiss LSM 880 Airyscan confocal microscope, a Zeiss LSM 510 confocal, or an Opera Phenix Z9501 high-content imaging system (PerkinElmer). 3D culture has the imaging challenge of sparsely positioned objects (cysts) that can be positioned in different Z-planes. This has previously required manual imaging of each object, precluding large sample number analyses. We have overcome this by building two analysis pipelines to allow quantitation of cyst phenotypes or phosphoinositide distribution from hundreds to thousands of 3D cysts per condition: (1) a Pipeline for semi-Automated Polarity Analysis, PAPA, and (2) a Pipeline for semi-Automated Phosphoinositide Intensity analysis, PAPI. PAPA and PAPI make use of an Opera Phenix Z9501 high-content imaging system (PerkinElmer). To direct imaging only to cysts (and not the surrounding non-cyst areas) we used the PreciScan module (PerkinElmer) to scan entire wells to find cysts by performing z-stacks of each well at ×5 magnification, on-the-fly processing of images using user-defined rules to identify cysts, and then directing the microscope to perform high-resolution z-stacks of objects (cysts). In the case of PAPI, this requires cysts to express GFP-tagged PIP probes above a baseline level defined as optimum for imaging. At this step, we perform at least 8 optical sections every 2 µM, imaging at least 25 fields or objects (×20 and ×63, respectively). For PAPA, cysts are stained for apical marker (Podxl), the cortex (Phalloidin), whole-cell stain (cytoplasm), and nuclei (Hoescht). For PAPI, GFP-tagged PIP reporters were imaged in place of whole cell stain. For PAPA, as lumens can occur in different Zplanes of a given cyst, a maximum projection of the medial Z-planes was applied. Using Harmony imaging analysis software (PerkinElmer), user-defined phenotype classification rules to detect polarity orientation based on the localization and intensity of apical markers and the detection and quantification of number of lumens was applied to each cyst. For benchmarking of accuracy, PAPA was compared to manual imaging and counting of polarity phenotypes for control vs SHIP1-inhibited cysts. PAPA produced concordant quantitation of phenotypes to expert manual quantitation. For PAPI, as maximum projection of different Z-planes could introduce quantification artefacts, analyses were performed only on the most medial Z-plane, which was automatically calculated based on the maximal luminal area. To detect PIP probe intensity in different subcellular locales, cysts were fixed and stained with Podxl to mark the apical domain, Phalloidin to mark the cortex, and Hoescht to mark the nucleus. Combinations of these stainings were used to create masked regions to calculate mean total, cortical, apical, basolateral, cytoplasmic (excludes nuclear region), and nuclear intensity per area in each region. For depiction of these regions, see Fig. 2a. Note that for all images taken at super-resolution for subcellular analysis, such as co-localization between Rab11 and PIPs or PIPmodifying enzymes, a Zeiss LSM 880 Airyscan confocal microscope was used from a single plane only. Statistical analysis was performed to calculate percentages of polarity phenotypes, normalized to control, and the relative PIP intensity in different cellular regions. Data were processed using KNIME analytics platform, and GraphPad Prism to generate graphs.
Statistics. Cyst phenotypes were binned into four categories: (a) single lumen, (b) multiple lumens/vesicular accumulation of Podxl, (c) inverted polarity, (d) other/ no lumen. Relative percentages were normalized to control. For RNAi rescue experiments, only cysts expressing exogenous transgene were scored. Relative percentages from each category were normalized to control. Values are mean ± S.D. from 3 to 9 replicate experiments, with n ≥ 300 cysts per replicate, unless otherwise indicated. For qPCR, expression was normalized as fold change (log2) from the mean expression for all conditions. Significance was calculated using a paired, twotailed Student's t-test, Mann-Whitney test, one-way ANOVA or two-way ANOVA test. No statistical methods were used to predetermine the sample size. No randomizations were used. The investigators were not blinded to allocation during experiments and outcome assessment. Statistical tests used are stated on every figure legend with P-values as appropriate. Data distribution should meet the normal distribution requirements. No estimate of variation. Data were analyzed using KNIME analytics platform, Excel (Microsoft) or Prism (Graphpad).
Live cell dual-color imaging. Live imaging was performed on cysts in eight-well chamber slides (Labtek II, Nunc). Cysts co-expressing EGFP-2xPH-TAPP1 and either TagRFP-T-Rab11a or Membrane-tdTomato were imaged on an inverted spinning-disk confocal microscope system (Yokogawa/Zeiss) with a 37°C and 5% CO 2 controlled environment (Zeiss) and heated stage (PECON) through a 20 × 1.49NA lens (Zeiss), using 488/561 nm laser lines. Images were captured via an AxioCam Mrm (Zeiss). A stack of three images at 5 µm intervals was taken using an automated stage controlled via the ZEN software package (Zeiss). As cysts develop and move during image acquisition, the optimal focal plane was manually chosen for each time point post-acquisition from the stacked images, before compilation of image series of each movie. Movies and images were processed using ImageJ (NIH), as adapted from Stehbens et al. 45 . Briefly, the functions 'Remove Outliers' (radius, 2; threshold, 5), and 'Unsharp Mask' (radius, 7; mask, 0.5) were applied to improve image contrast.
RNAi. Lentiviral shRNAs were used to stably deplete proteins, described in Virus production and transduction methods section. ShRNA sequences were generated using iRNA software, as per Addgene (https://www.addgene.org/tools/protocols/ plko/). RNAi target sequences are listed in Supplementary Table 2. Knockdown was verified by western blot or RT-qPCR procedures, normalized to GAPDH expression. Due to the canine origin of MDCK, shRNAs were custom-designed, and RNAi sequences chosen not to match human were utilized when rescuing with transgenes.
RT-qPCR. For detection of phosphoinositide kinases and phosphatases expression or for shRNA-mediated knockdown verification, RT-qPCR was used. Primers were designed based on the canine genome using Primer3Plus software and are listed in Supplementary Table 3. RT-qPCR was performed using EXPRESS One-Step SYBR® GreenER™ Universal (Thermo Fisher Scientific) per manufacturer's instructions.
Delivery of exogenous PI(3,4)P 2 . Exogenous PIP was delivered to cysts as previously reported 11 with appropriate adaptations: PI(3,4)P 2 and Histone carriers (Shuttle PIP TM , Echelon, Salt Lake City, UT) were freshly prepared in PIP solution (150 mM NaCl, 4 mM KCl, 20 mM HEPES at pH 7.2) to a final concentration of 300 or 100 µM, respectively. Complexes were combined for 10 min, and diluted 1:10 in Hank's buffered salt solution, solution added for 30 min to MDCK cysts that had previously been cultured in 3D for 48 h. Cells were fixed, stained, and imaged.
Repeatability of experiments. Immunofluorescence images of 3D MDCK cysts: representative image from one field of~1 × 10 4 cells in shown. Each experiment was repeated at least twice unless otherwise indicated. Immunofluorescence images of MDCK cells grown in Transwell: representative image from two image sets.
Movies: Representative movie from five movies of single lumen MDCK cysts. Western blot: Figure 4f: representative image from two blots.   Figure 8c, e, f: ≥25 independent objects/condition, one independent experiment.

Data availability
All relevant data are available from the authors upon request.