Cell‐extracellular matrix interactions in the fluidic phase direct the topology and polarity of self‐organized epithelial structures

Abstract Introduction In vivo, cells are surrounded by extracellular matrix (ECM). To build organs from single cells, it is generally believed that ECM serves as scaffolds to coordinate cell positioning and differentiation. Nevertheless, how cells utilize cell‐ECM interactions for the spatiotemporal coordination to different ECM at the tissue scale is not fully understood. Methods Here, using in vitro assay with engineered MDCK cells expressing H2B‐mCherry (nucleus) and gp135/Podocalyxin‐GFP (apical marker), we show in multi‐dimensions that such coordination for epithelial morphogenesis can be determined by cell‐soluble ECM interaction in the fluidic phase. Results The coordination depends on the native topology of ECM components such as sheet‐like basement membrane (BM) and type I collagen (COL) fibres: scaffold formed by BM (COL) facilitates a close‐ended (open‐ended) coordination that leads to the formation of lobular (tubular) epithelium. Further, cells form apicobasal polarity throughout the entire lobule/tubule without a complete coverage of ECM at the basal side, and time‐lapse two‐photon scanning imaging reveals the polarization occurring early and maintained through the lobular expansion. During polarization, gp135‐GFP was converged to the apical surface collectively in the lobular/tubular structures, suggesting possible intercellular communications. Under suspension culture, the polarization was impaired with multi‐lumen formation in the tubules, implying the importance of ECM biomechanical microenvironment. Conclusion Our results suggest a biophysical mechanism for cells to form polarity and coordinate positioning at tissue scale, and in engineering epithelium through cell‐soluble ECM interaction and self‐assembly.

medicine, but also important for cancer biology. The transmembrane glycoprotein gp135 (also called Podocalyxin) is often used as an apical marker located at the luminal surfaces, 5,6 and Rab small GTPases mediate the direct transportation of gp135 and apicobasal polarization. 7,8 Over the past few decades, studies on epithelial morphogenesis have indicated the importance of cell-cell adhesions and cell-extracellular matrix (ECM) interactions. 5,[9][10][11][12][13][14][15][16] The differentiation of epithelial cells in which non-polarized cells form polarized epithelium depends on ECM components. 9 It was shown that breast epithelial cells differentiate into tubules in type I collagen (COL), 17,18 while they form lobular acini in basement membrane (BM, mimicked by Matrigel in experiments). 19,20 Our previous study further showed that cell-collagen interaction permits a long-range morphogenetic coordination at sub-millimetre scale. 21 Nevertheless, how epithelial cells utilize cell-ECM interaction to coordinate their positioning and polarization in response to different ECM components at the whole-tissue scale is not fully understood.
To form long-range coordination, it is generally believed that ECM can serve as scaffolds to guide cell positioning and polarization. 15 Cells constantly secrete soluble ECM molecules and degrade existing ECM scaffolds into soluble fragments in vivo. Theoretically, these soluble forms of ECM can be assembled into scaffolds through two processes. The first is that they self-assemble into new scaffolds or re-incorporate into pre-existing scaffolds. Alternatively, soluble ECM can interact with cells which serve as nucleation cores to assemble ECM scaffolds. An example is the development of renal tubules where BM components are found to be dynamically assembled around the pre-tubular structure. 22 Apicobasal polarization is a typical process during epithelial tubulogenesis, and consistently with the role from ECM, the cellular mechanism involving integrin and RhoA signalling pathways has been identified in triggering the polarity formation. 23,24 Here, we study whether ECM scaffolds created by ECM self-assembled hydrogel or by cell-mediated assembly play the primary roles in the formation and coordination of epithelial morphogenesis. These two processes can hardly be decoupled in vivo or through the conventional ECM reconstitution approaches. Our recent work demonstrated that cell motion promotes fibrillary assembly of soluble COL, in supporting the role of cells in ECM scaffold generation. 25 We therefore use the in vitro open-system assay, and found that spatiotemporal coordination in epithelial morphogenesis and polarization can occur on cell-assembled ECM in the fluidic phase rather than pre-assembled ECM in the solid phase. The coordination depends on native topology of the ECM components such as basement membrane (BM) and type I collagen (COL). Further discovered during tubulogenesis, the apicobasal polarization proceeds in a collective way along the axis of the tubule, implying intercellular communications within the cell groups. Our results suggest a potential mechanism which cells can use to form polarity and coordinate morphogenesis in vivo, and a strategy to engineer epithelial structures through self-assembly in vitro.

| Cell Culture, reagents, DNA constructs and lentivirus
Cell culture medium and reagents were purchased from Invitrogen Gibco. Madin Darby canine kidney (MDCK II) cells (from ATCC) were maintained in Advanced Dulbecco's modified Eagle's medium (serum reduced medium) supplemented with 3% fetal bovine serum, 2 mM L-glutamine, 20 unit/mL penicillin, 20 μg/mL streptomycin and 1 mM sodium pyruvate in a humidified 95% air, 5% CO 2 incubator at 37°C. BD Matrigel (basement membrane matrix, growth factor reduced and phenol red-free) and 3-D Culture Matrix™ Rat Collagen I (5 mg/mL) were purchased from BD Biosciences and R&D Systems, respectively. Rabbit anti-laminin and mouse anti-collagen I primary antibodies were purchased from Sigma, and mouse anti-integrin ɑ6 antibody from Abcam. Pacific Blue-conjugated goat anti-rabbit and anti-mouse IgGs were purchased from Invitrogen, and Rhodamineconjugated goat anti-mouse IgG antibody from Sigma.
The plasmid pcDNA3-gp135-GFP construct was a gift from Dr Joachim Füllekrug (Max Planck Institute of Molecular Cell Biology and Genetics). 26 The plasmid expressing GFP-tagged human β-actin (GFPβ-actin) under endogenous promoter was provided by Dr Beat A. Imhof (Switzerland). 27 Lentivirus encoding mCherry-conjugated histone H2B (H2B-mCherry) was generously provided by Dr Rusty Lansford and David Huss (Biology, California Institute of Technology). 28

| Development of stable fluorescent MDCK cell lines
We first developed the stable MDCK cell line expressing H2B-mCherry. Cells at 20%-30% confluency were infected with lentivirus encoding H2B-mCherry and then diluted in 96-well plates to enable the selection and the expansion of single fluorescent colony. Next, we transfected MDCK_H2B-mCherry cells with pcDNA3-gp135-EGFP to develop the cell line expressing both H2B-mCherry and gp135-EGFP (MDCK_H2B-mCherry/gp135-EGFP). In addition, we have transfected MDCK cells with EGFPβ-actin plasmid (MDCK_EGFP-actin). This cell line was used to compare the results obtained from MDCK_H2B-mCherry/gp135-EGFP cells. The transfection was performed by using Lipofectamine2000 (Invitrogen), followed by the antibiotic selection with G418 (300 µg/mL) for 2 weeks to obtain a cell pool displaying various levels of EGFP. A single colony with an intermediate fluorescence intensity of gp135-GFP or GFPβ-actin was selected through dilutions of the cell pool in 96-well plates and further amplified.

| Chambers for cell culture and microscopy
Cell culture experiments and time-lapse microscopy were performed in custom, stainless steel chambers. These chambers were manufactured to have a rectangular shape with a height of 0.6 cm and a 2 × 5.5 cm 2 internal opening. Nail polish was used to seal 24 × 60 mm No. 1 coverslips on the bottom of the chambers. To perform multiple-well experiments in one chamber, polydimethylsiloxane (PDMS) blocks were cut to fit the chamber containing multiple wells (~0.5 × 0.5 cm 2 ). The surface of PDMS block was then cleaned and airdried to allow for a firm attachment on the coverslip of the chamber.

| Cell culture on BM gels or lessadhesive substrates
To make BM gels, we prepared sterile chambers sealed with coverslips on the bottom, and the stock solution (100%) of BD Matrigel was then spread on the top of the coverslips (40-80 µL/cm 2 ) followed by incubation at 37°C for 20-30 minutes. This allowed forming a layer of gel with a variable height (200-400 µm).
To place cells, individual MDCK cells (~1-2 × 10 4 cells/cm 2 ) were seeded on BM gels in the culture medium containing 2% BM or 20 µg/ mL COL. Then, the chamber was placed into petri dish in the cell culture incubator with medium change every 4 days, or every day (or every other day) during time-lapse microscopy. Here, the concentration of BM in the medium followed the 'on-top' assay developed by Bissell and her coworkers. 29 20 µg/mL COL in the medium was chosen to match the mass concentration of 2% BM (10 mg/mL in stock). Nevertheless, we have applied the different concentrations of COL from 5 to 50 µg/ mL, and cells could form tubular structures within this range.
For cell culture on less-adhesive substrates, we first prepared a layer of agarose gel (1%) on the coverslips in the chambers. A mixture of MDCK cells (~10 4 cells/mL) and culture medium containing 2% BM or 20 µg/mL COL was added into the chambers and moved in the cell culture incubator. To minimize water evaporation, the chambers were covered by coverslips with a small opening (~5 mm) to allow for air exchange. Medium was changed carefully after 7 days to avoid destroying or losing the cell aggregates.

| Immuno-staining experiments
Immuno-staining experiments were conducted at room temperature, except for incubation with primary antibodies at 4°C. In brief, cell samples were fixed with 4% paraformaldehyde for 15 minutes and permeabilized with 0.1% Triton X-100 for 20 minutes. The cells were then incubated with rabbit anti-laminin, or mouse anti-collagen I at 4°C overnight, followed by incubation with goat secondary antibody conjugated with Pacific Blue (410/455 nm) or Rhodamine (550/570 nm) for 2 hours at room temperature. The images were collected by using epifluorescence or scanning microscopy.
For phase-contrast and/or epi-fluorescence microscopy, the imaging system based on Olympus IX71 microscope was equipped with motorized excitation and emission filters with a shutter control (lambda 10-3, Sutter), an Electron-Multiplying CCD camera (ImagEM,

| Cell-ECM interactions in the fluidic phase for polarized epithelium formation
We first examined if cells can develop coordinated polarity on ECM scaffolds formed by ECM self-assembly without any soluble ECM.
Madin-Darby Canine Kidney (MDCK) cells were used in this study, which are a popular model cell line for epithelial morphogenesis and apicobasal polarity formation regulated by Rab GTPases. 3,8 To track the development of apicobasal polarity, we engineered MDCK cells stably expressing mCherry-conjugated histone H2B (H2B-mCherry) and GFP-conjugated gp135 (gp135-GFP). Here, H2B-mCherry is used to indicate cell nucleus, 28 whereas the apical marker gp135 is used to indicate cell polarization. 5,26 To define polarity, we noted that in polarized epithelium, cells are organized into a surrounding and continuous monolayer structure with gp135 primarily confined at their apical surfaces, 5 which are located at the inner space of the organization (illustrated in Figure 1A with the experimental data shown in Figure 1B). Thus, we used the spatial distribution of H2B-mCherry with respect to gp135-GFP to define and track the formation of polarity. The polarization process here refers to the convergence of diffusive gp135-GFP to the apical surface of the lumen or to the intercellular region between cell nuclei. To initiate cell density-dependent long-range coordination, we adopted the cell density characterized and optimized in our previous study. 21 F I G U R E 1 Soluble ECM components are required to form epithelial polarization on Matrigel. (A) Experimental set-up. MDCK cells were cultured on the top of basement membrane (BM) gels with or without soluble ECM components in the medium. Cells express H2B-mCherry and gp135-GFP to indicate their nucleus and polarity, respectively. Note the distribution of nucleus with respect to gp135 in polarized lumens (on the right). (B) Left in each row: Represented phase-contrast (bright field) and 3-D projected fluorescent images (gp135: green, H2B: red, and their overlay) for cells seeded on BM gels and cultured for 12 days (seen in Methods for 3-D projected imaging and processing). The medium contained (top row) no soluble ECM, (middle row) 2% BM or (bottom row) 20 µg/mL type I collagen (COL). Nuc: cell nucleus. Right in each row: Fluorescence intensities of H2B-mCherry and gp135-GFP along the indicated white dotted lines (from left to right). The overlay of the red and green curves indicates the positions of cell nucleus (red) with respect to the apical marker (green). a. u.: arbitrary unit. (C and D) 3-D view of the polarized lobular and tubular structures. The images from confocal scanning microscopy (with 20x objective) were reconstructed into 3-D structures. The 3-D views showed the relative position of cell nucleus (red) and apical marker gp135 (green) in the closed lobular (C) and elongated tubular (D) structures After long-term culture, epithelial cells including MDCK cells can secrete ECM molecules. 16,30,31 We therefore used open systems to dilute and/or remove secreted ECM by changing medium every day or every other day. Cells were cultured on preassembled BD Matrigel gels (to mimicking basement membrane (BM), the primary matrix components underlying polarized epithelium in vivo) open to a large medium space that contained no soluble ECM ( Figure 1A). Under this condition, cells grew and merged into big clusters (hundreds of micrometres in diameter) without forming general apicobasal polarity ( Figure 1B

| Dynamics of cell positioning and polarization dependent on cell-ECM interactions
Having shown that cell-BM (COL) interactions in the fluidic phase lead to the formation of polarized lobules (tubules), we examined how cells coordinate their positioning and polarization in response to soluble BM (COL). Same as the set-ups in Figure 1A, cells were seeded on 3-D BM with or without soluble ECM in the medium, and the dynamic processes of cell aggregation and polarization were continuously recorded by time-lapse imaging in the following days.
We first examined cellular dynamics on BM gels in the presence or absence of soluble BM. Without soluble BM, time-lapse microscopy revealed that individual cells proliferated into small clusters, which continuously grew and merged with each other without general polarity formation (through the 5-day period of observation, Movie S1) (Figure 2A, Movie S1). By contrast, with soluble BM, most clusters stopped merging and became polarized on the 2nd-3rd day ( Figure 2B). Here, the timing of polarization was defined by the conversion of gp135 from the outer layers to the inner areas of clusters/ cysts ( Figure 2B, and Movie S2). The lumen cultured by MDCK cells (H2B-mCherry/GFPβ-actin) displays actin ring at the apical side ( Figure 2C), which was consistent with the previous report. 32 We next examined the cellular dynamics on BM gels with soluble COL. Similar to the observation with soluble BM (Figure 2B), cells were found to form clusters which proliferated and coalesced.
However, the coalescence did not stop on the 3rd day but instead clusters continued to fuse into a long-range (>200 µm), tubule-like structure ( Figure 2D). By then, cells started to polarize through a collective conversion of gp135 from the outer layer to the inner area of tubule ( Figure 2D

| Cell polarization in the fluidic phase with ECM assembly
If the coordination of cell positioning and polarization requires the formation of ECM scaffolds, the results above suggest that it is the scaffold mediated by cell-ECM interaction in the fluidic phase that determines epithelial morphology and coordinates polarity. To see how such scaffolds are formed and affect cell positioning and polarization, we performed immuno-staining of laminin (a major component in BM 33 ) and COL on polarized lobules and tubules.
We first examined how laminin is distributed on cells seeded on BM gels with or without soluble BM. In the presence of soluble BM, condensed laminin was found at the outer layers of clusters after 3 days of culture ( Figure 3A), and the density of laminin increased during the culture time ( Figure 3B), indicating the laminin assembly along with the cluster growth. After 7 days, most clusters were found to form polarized lobules surrounded by dense laminin (Figure 3A).  Similar mechanism may be extended to the observed tubular structure with partial COL coverage. In considering that this is a descriptive data with limited work, we did not have explanation why ECM components were not assembled on the top of the lobule, or dig out more insights at current stage.

| Culture of lobular and tubular epithelial structures under suspension conditions
The results above suggest two distinct, ECM-dependent processes to coordinate epithelial morphogenesis in the fluidic phase. The first is that cells recruit soluble BM components which are known to form branched networks, 35 to create a closed-end (ie, restricted) scaffold surrounding individual cluster. Such scaffold provides a physical barrier to block the coalescence of clusters and allow them to proliferate and polarize within the restricted space ( Figure 3A). The second process is that cells recruit soluble COL, which is known to form linear, bundled fibres, 36,37 to create an open-ended (ie, unrestricted) F I G U R E 2 Distinct dynamics of cell coordination in response to soluble BM/COL. MDCK cells express H2B-mCherry and GFP-gp135 or β-actin-GFP to indicate cell nucleus and apicobasal polarity, respectively. Cells were cultured on BM gel with or without soluble ECM in the medium and positioned on the microscope by change with fresh medium every day. Time-lapse fluorescence images were taken with 22 min interval time per cycle in the following days. (A and B) Represented time series of 3-D projected epifluorescence images (gp135: green, H2B: red) for cells in medium (A) without ECM (n > 10), (B) with 2% basement membrane (BM) components (n > 10) (also seen in Movies 1&2, interval = 22 min). (C) Represented optical sectioning from scanning microscopy β-actin: green, H2B: red) to demonstrate lumen formation with polarized actin distribution after culturing cells in medium containing 2% BM for 14 days. (D) Represented time series of 3-D projected epifluorescence images for cells in medium with 20 µg/mL COL (also seen in Movie S3) (n = 14). The purple arrows indicate the polarization initiating from the local regions of the tubule. (E) The collective apicobasal polarization along the tubular axis (n = 6). The time-sequence images (interval: 22 min*5 = 110 min) showed the spatial distribution of gp135 relative to cell nuclei (H2B) during the polarization. (F) Size quantification of the diameters (along the long axis) of unpolarized clusters without ECM (n = 21) and polarized lobular lumens with BM (n = 50), and the length of polarized tubes with COL (n = 14) around Day 5. (G) The timing quantification when lumens (n = 38) or tubules (n = 14) got polarized under culture with 2% BM or 20 µg/mL COL in the medium. N/A refers to 'not applicable for polarization' without ECM in the medium. The data quantification (mean ± SD) was performed by using ImageJ and Origin. *** and **** represent significant difference with P < 10 -2 and 10 -6 in comparison with the group 'with COL' from Student's t test analysis scaffold ( Figure 3D). In contrast to the first one, scaffold formed by soluble COL allows clusters to continuously merge with one another through long-range interactions in the fluidic/semi-fluidic phase.
These two processes were observed in the assays where cells were supported by pre-assembled ECM, that is, a solid phase. To ascertain whether the solid phase is absolutely required in soluble ECM-mediated epithelial cell polarization and morphogenesis, we  Figure 5D (right)), which seemed similar to the results found in the lobules cultured on top of BM gels ( Figure 3C). It was noted that soluble BM and COL in the medium could form scaffolds alone ( Figure 5B and D), which assisted the formation of polarized epithelium. Thus, cell-ECM interactions are required to initiate epithelial morphogenesis and create polarization at the tissue scale.

| D ISCUSS I ON
Emerging evidence suggests that cell-ECM interactions are crucial in regulating tissue development, homeostasis and repair. 7,39,40 Such interactions can occur on the solid phase (ie, the existing ECM scaffold) and in the fluidic phase where cells interact with soluble ECM components. Here, we showed the following findings based on the engineered MDCK cells stably co-expressing H2B-mCherry (nucleus)

F I G U R E 4
The expansion of dividing cells into polarized lobule. The MCDK cells expressing gp135-GFP and H2B-mCherry were seeded on BM gel with 2% BM in medium, and two-photon confocal imaging started one day later. The wavelengths of excitation light were 890 nm for GFP and ~ 1100 nm for mCherry; the step size in z-direction was 1.0 µm. Time-lapse imaging was taken with the interval time of 3 or 4 h. Medium was changed every day along with focus/position corrections. Acquired confocal images were processed by two channels (GFP and mCherry) overlay and 3-D reconstructions. Each image here shows the 3-D view of 4 or 5 time-points at the labelled time (in hours, and the starting time was set as zero), and two images for each time point were displayed from different angles (generally from the top and the bottom views). (A) The polarity distribution at early stage in the cluster with a few cells. (B) The polarity distribution during growth of the cluster into fine 3-D lobular structure. More detailed 3-D views are shown in Movie S5. A note: the images of (A, B) were acquired at the same position in one experiment, but may not be from the exact same sample during the re-focusing process as cells were motile at the early stage It is likely that ECM scaffolds assembled by cell-soluble ECM interactions possess a more physiological structure. For example, BM gels assembled in the absence of cells were found more resistant to laminin staining than those assembled by cells, as indicated by the difference of laminin staining ( Figure 3A), which might result from difference of porosity. Indeed, previous reports showed that immune-staining of cells in BM or COL gels requires a pre-cleavage of gels by ECM proteinases. 14,45 The selection of topology in the polarized epithelial clusters appears to correlate with the native topology of BM and COL (ie, lobular or linear/tubular).

| Coordination of cell dynamics dependent on soluble ECM components
It has been documented that cell-ECM interactions through integrin signalling are crucial in regulating apicobasal polarity in epitheliums. 46 By application of functional-inhibitory antibodies, it was reported that integrin β1 is involved in the polarity regulation including the apical localization of gp135. 47 Further studies of the downstream mechanism showed that integrin-linking kinase (ILK) modulates laminin assembly and regulates integrin-microtubule network for directional delivery of apical factors. 48,49 Multiple groups also revealed that small GTPases Rac and Rho signalling act downstream of integrins to mediate and maintain appropriate epithelial polarity. 16,23,50,51 Our work here added one piece of information to the scenario that biophysical factors like cells-mediated assembly of ECM structure may play a role in directing the epithelial topology and polarity during morphogenesis. At this stage, nevertheless, we did not investigate how molecular signals from cell-soluble BM/COL interactions lead to distinct dynamics in polarity formation.

| Epithelial polarization with or without direct cell-ECM contact
Our third finding is that cells can maintain polarity without a complete coverage of ECM scaffold around the lobular/tubular structures (Figures 3 and 5D), suggesting that some cells can be polarized without direct cell-ECM interactions. Nevertheless, without ECM in the medium, cells were unable to form polarity (Figure 1 and Movie S1). Time-sequence images from confocal microscopy indicate that cells are polarized in small cluster at early stage, and con-  Figure 1B). This suggests that epithelial tubulogenesis is mediated by ECM microenvironment. Consistently from the previous report, perturbing cell mechanics by inhibition of ROCK-Myosin II pathway also resulted in multiple lumens during tubulogenesis, due to impaired cell motility. 24 The observations of multi-lumen tubes support that the apicobasal polarization is also mediated in mechanical way beside chemical signals.

| CON CLUS IONS
In summary, our results highlight the criticalness of cell-ECM interactions in the fluidic phase for epithelial polarization and morphogenesis. Importantly, we show how a physiological environment can spontaneously emerge through cell-soluble ECM interactions, and the native structure of self-assembled ECM helps define the topology of self-organized epithelium. The collective polarization during lobular/tubular morphogenesis may imply the existence of intercellular communications. In the fluidic phase, the polarization occurs in the small cluster at early stage and is maintained through the ex-

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest in this work.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.