To capture the centrosome position and the distribution of Gp135 vesicles, we performed immunostaining at several stages in the transition from a single MDCK cell to a two-cell cyst. As a single non-polarized MDCK cell is embedded into Matrigel, most Gp135 proteins cluster around the interphase centrosome (Figure 1A, i) and strongly co-localize with the Rab11a-labeled recycling endosomes (Figure 1—figure supplement 1A). During mitosis, the Gp135 proteins are well dispersed in the cytosol at metaphase (Figure 1A, ii), weakly cluster near the centrosomes in early telophase (Telo, Figure 1A, iii), and gradually accumulate at the centrosomes from late telophase to the stage of cytokinetic bridge pre-abscission (pre-Abs) (Figure 1A, iii-iv), when the acetyl-tubulin-labeled cytokinesis bridge condenses significantly. Cytokinesis in most animal cells begins in the middle point of anaphase and ends shortly after the completion of mitosis in telophase. Cytokinetic abscission leads to the physical cut of the cytokinetic bridge connecting the daughter cells and concludes cell division. After cytokinetic bridge abscission (post-cytokinesis), the Gp135-positive vesicles are transported to the middle membrane region, which will form the AMIS between the two daughter cells (Figure 1A, v). Subsequently, exocytosis continues to expand the apical membrane and produces a lumen structure (Figure 1A, vi). The relative Gp135 signals near the centrosome regions of different stages show a clear maximum at the cytokinetic pre-abscission stage (Figure 1A–C). Notably, before bridge abscission, we detected a stronger signal of Gp135-positive vesicles surrounding the centrosomes compared to the cytosol and cytokinetic bridge (Figure 1A–B and D).

We quantify the degree of polarization by the centrosome positions using the following quantities. The center point between two nuclei is set as the origin (O), representing the cyst center. The polarity index is defined as the nucleus (N)–centrosome (C) distance, L_N-C, normalized by the average nuclear radius, r_N. In addition, the nucleus–centrosome orientation (θ_N-C, the angle between the N-C and N-N axes) characterizes the movement of the centrosome (Figure 1E, see Materials and methods) (Burute et al., 2017; Rodriguez-Fraticelli et al., 2012). Compared to the non-polarized 1-cell stage, the polarity index gradually increases from the cytokinetic pre-abscission to the post-cytokinesis stage, reaching the highest level during the lumen-open (LO) stage (Figure 1F). A high θ_N-C value indicates that the centrosomes are located at the outside of two nuclei, which also corresponds to a larger L_O-C (origin–centrosome distance) and L_IC (inter-centrosomal distance) (Figure 1—figure supplement 1B–E). From telophase to post-cytokinesis, centrosomes move from outside to the inside of two nuclei, and thus θ_N-C decreases (Figure 1G).

Because recycling endosomes play a crucial role in trafficking apical membrane components such as Gp135, we examined other apical membrane components such as Crb3, Cdc42, and aPKC (Bryant et al., 2010; Klinkert et al., 2016; Schluter et al., 2009). We found that aPKC appears to diffuse in the cytoplasm during the M phase and concentrates at the AMIS only after cytokinesis (Figure 1—figure supplement 1F). This result may be due to a lower aPKC concentration, contributing to a weak immunostaining signal. Nevertheless, the signals of enhanced green fluorescent protein (EGFP)-Crb3 and EGFP-Cdc42 in MDCK cells exhibited similar patterns near the centrosome during cytokinetic pre-abscission and were transported to the AMIS from the centrosome in the post-cytokinesis stage, identical to Gp135 (Figure 1—figure supplement 2A–B).

Taken together, our data suggest that the centrosome acts as a hub of apical recycling endosomes and centrosome migration during cytokinetic pre-abscission before apical membrane components are targeted to the AMIS. Thus, the center migration of centrosomes may facilitate the trafficking of apical recycling endosomes to the AMIS (Figure 1H).

To precisely define the hierarchy of centrosome migration, apical membrane formation, and apical-basal polarity establishment, we performed live cell imaging using MDCK cells that expressed fluorescently labeled centrosome protein, PACT-mKO1 (the PACT domain is a conserved centrosomal targeting motif of PCNT) (Gillingham & Munro, 2000), and apical membrane protein, EGFP-Gp135, cultured in Matrigel. The onset of anaphase is set as 0:00 (h: min). Consistent with our fixation data, live cell imaging showed that EGFP-Gp135-positive vesicles did not surround PACT-labeled centrosomes during metaphase. When cleavage furrow ingression occurred, EGFP-Gp135-positive vesicles were recruited to a region surrounding the centrosome (Figure 2A, 0:10). The movement of these EGFP-Gp135-positive vesicles followed the same path as centrosome migration (Figure 2A, 0:20–0:40), with the vesicles finally fusing to the middle membrane of the cell doublet (Figure 2A, 0:55). Using our live cell movies, we measured the polarity index and found that cell polarization was approximately begun after the onset of anaphase (Figure 2B). θ_N-C and L_O-C decreased significantly within 30 min after the anaphase onset and then plateaued (Figure 2C, D).

These results suggest that centrosomes migrate toward the center of the cell doublet during the initial 30 min following the onset of anaphase, followed by a stable phase with reduced movement. Intriguingly, we observed this type of centrosome movement only in 3D culture cells embedded in Matrigel. In contrast, in adherent two-dimensional (2D) culture cells grown on a cover glass, the centrosomes exhibited random migration patterns (Figure 2—figure supplement 1A–C).

To investigate the temporal process of apical membrane formation, we measured the intensity profile of EGFP-Gp135 along the line connecting the two nuclei; we took the central value as an indicator of the amount of Gp135 on the AMIS (Figure 2E, white arrows). The intensity continually increased after the onset of anaphase (Figure 2F). As shown in Figure 2A, the EGFP-Gp135 signal at the middle membrane of the cell doublet increased significantly after 30 min from the onset of anaphase, coinciding with a reduced movement of the centrosomes (Figure 2C, D). These findings are consistent with our immunostaining results, indicating that centrosome migration precedes the targeting of Gp135 to the AMIS.

To investigate the role of centrosomes in guiding Gp135-labeled vesicle trafficking for polarization, we depleted centrosomes in MDCK cells by applying centrinone, a reversible inhibitor of PLK4 kinase for centriole duplication in vertebrate cells. Centrinone treatment effectively blocked centriole duplication in MDCK cells and progressively depleted centriole numbers during cell cycle progression (Figure 3—figure supplement 1A, B). Three days after centrinone treatment, nearly 80% of cells lost their centrioles (Figure 3—figure supplement 1A), indicating the effective removal of centrosomes by centrinone treatment in MDCK cells.

Previous studies have reported that a loss of centrosomes in cells activates p53, followed by activation of p21, which causes cell cycle arrest in non-transformed cells (Fong et al., 2016; Lambrus et al., 2016; Meitinger et al., 2016). However, cell division is required in our experimental setup to initiate epithelial polarization. To overcome the problem of cell cycle arrest, we generated a p53 knockout (KO) line of MDCK cells using the clustered regularly interspaced short palindromic repeats (CRISPR)/Cas9 editing system (Figure 3—figure supplement 1C, D). We found that the MDCK cells carry normal p53 activity and exhibit increased p21 protein levels after centrinone treatment (Figure 3—figure supplement 1E, F). As expected, the p53-KO MDCK cells do not induce p21 and continue proliferating after centrosome depletion (Figure 3—figure supplement 1F). Compared with wild-type (WT) cells, the p53-KO cells show no difference in polarity index, θ_N-C, or L_O-C during de novo polarization (Figure 3A–C and Figure 3—figure supplement 1G), indicating that the loss of p53 does not interfere with centrosome movement.

Next, we treated the p53-KO MDCK cells with centrinone for three days, seeded them into Matrigel, and observed them by live cell imaging (Figure 3—figure supplement 1H). Our results showed that, in the absence of centrosomes, EGFP-Gp135 did not cluster in the cytoplasm at any time points (Figure 3A), and the accumulation of EGFP-Gp135 at the middle membrane was significantly reduced (Figure 3A, D). Together, our time-lapse imaging and centrosome depletion studies suggest that centrosome depletion causes a reduction in Gp135 accumulation at the apical site due to a lack of Gp135 near the centrosome during cytokinesis. We then performed immunostaining, which supported our live cell imaging observations (Figure 3—figure supplement 2A, B).

We extended the observation time to 48 and 72 hours and found that more than 60% of centrosome-depleted p53-KO cells, which overcome the proliferation arrest, form cysts with a central lumen (Figure 3E, F). For both WT and p53-KO cells, the percentage of single-lumen cysts is higher than centrosome-depleted p53-KO cells at 24 hours, but the differences decrease over time (Figure 3E, F). However, the cell heights of the centrosome-depleted cysts are less uniform, and the Gp135 levels at the apical region remained lower than those of the controls (WT and p53-KO) (Figure 3G). Although p53-KO can overcome cell cycle arrest after centrosome depletion, its cell proliferation rate is still reduced (Figure 3—figure supplement 2C). We, therefore, compared the percentage of lumen formation in two-cell stage cysts to eliminate the effect of proliferation variation. 24 hours after seeding in Matrigel, more than 50% of WT and p53-KO two-cell cysts formed a single central lumen. In contrast, only 30% of centrinone-treated p53-KO two-cell cysts exhibited a single central lumen (Figure 3—figure supplement 2D). Taken together, our data suggest that centrosome depletion primarily affects the early steps of epithelial polarization but not the later steps of lumen formation.

Centrosomes are a large protein complex; thus, we sought to determine which components affect the position of centrosomes and the trafficking of Gp135 vesicles by perturbing centrosome components without completely removing the centrosome. We perturbed molecules associated with centriole/centrosome structures, such as CEP164 (an outer-layer protein of the DAs), ODF2 (the base of the SDAs), and PCNT (the scaffold protein of the PCM) (Figure 4A). Each of these molecules has been reported to affect centrosome migration in different experimental systems (Hannaford et al., 2022; Hung et al., 2016; Krishnan et al., 2022; Pitaval et al., 2017).

We generated KO lines of MDCK cells for CEP164, ODF2, and PCNT (Figure 4—figure supplement 1 and 2) and a KO line of CEP120 (in the p53^−/− background) (Figure 4—figure supplement 2E–H), an important regulator of centriole elongation (Comartin et al., 2013; Lin et al., 2013), to produce centrioles lacking both DAs and SDAs simultaneously (Figure 4A) (Tsai et al., 2019). Our results showed no significant differences in θ_N-C or L_O-C between the WT and KO cells in the post-cytokinesis stage despite the presence of structural defects on the centriole/centrosome (Figure 4B–D). Gp135-labeled vesicles are still targeted to the center of cell doublets in the Matrigel culture (Figure 4E). Depletion of these centriole/centrosome structures does not affect polarized vesicle trafficking or centrosome positioning.

To determine whether defective centrosomes affect the timing of centrosome migration, we examined the centrosome position at the cytokinetic pre-abscission stage, as indicated by θ_N-C and L_O-C (Figure 4F–H); at this stage, centrosomes have initiated migration but have not yet reached their final position (Figure 1A). We found defective centrosomes also initiated migration at the cytokinetic pre-abscission stage in various KO lines (Figure 4F). The θ_N-C and L_O-C values showed no significant differences between the KO and WT cells (Figure 4G, H). These findings suggest that the timing of centrosome migration was not delayed in KO cells in which the SDA, DA, or PCM components of the centrosomes were individually disrupted. This observation may be attributed to the compensatory nature of microtubule anchoring and nucleation on centrosomes, which involve various molecules (Gavilan et al., 2018). The complexity of these interactions may explain the challenge of disrupting a single molecule on the centrosome to affect centrosome migration.

We next examined molecules that have been previously reported involved in MDCK polarization (Bryant et al., 2010; Horikoshi et al., 2009; Martin-Belmonte et al., 2007; Roland et al., 2011; Schluter et al., 2009) to assess whether these molecules affect centrosome positioning. We first applied short hairpin RNA (shRNA) interference to deplete the molecules involved in vesicle trafficking, such as Rab11a, Sec15a, and myosin Vb (MyoVb) in WT cells (Figure 5—figure supplement 1A–C). Although depletion of the regulator of vesicle trafficking leads to a dispersion of Gp135, the centrosomes showed normal localization (Figure 5A–D). These results agree with our findings above, indicating that centrosome positioning governs the efficiency of polarized trafficking (Figure 3). Inhibition of vesicle transport mediated by Rab11a, Sec15a, or MyoVb does not affect centrosome positioning.

Next, we applied shRNA to deplete polarity regulators, such as Par3, Cdc42, and Crb3, in WT cells (Figure 5—figure supplement 1D–F). We found that Par3 depletion, but not Cdc42 or Crb3, significantly affects the centrosome positioning, as shown by a large deviation in θ_N-C, L_O-C, and L_IC compared with the control (Figure 5A–D). The centrosomes do not move toward the AMIS after cytokinesis. Interestingly, we sometimes observed that in Par3-depleted cysts, Gp135- and Rab11a-positive vesicles are concentrated near mislocalized centrosomes in the post-cytokinesis stage (Figure 5A and Figure 5—figure supplement 1G). These data suggest that Par3 acts upstream of both centrosome positioning and polarized vesicle trafficking.

We next examined Par3 localization during de novo epithelial polarization through cytokinesis (Figure 5—figure supplement 1H). For MDCK cells in Matrigel culture, our results show that Par3 first emerges at the cell-cell interface of the mitotic cleavage site at the cytokinetic pre-abscission stage (Figure 5—figure supplement 1H), in contrast to the apical membrane components, which concentrate around the centrosome at this stage (Figure 1A, B and Figure 1—figure supplement 2A, B). Indeed, we found that the cell-cell junction components Par3, actin, ZO-1, and E-cadherin did not surround migrating centrosomes; rather, they localized to the cytokinesis site and cell-cell contact interface at the pre-abscission stage (Figure 5—figure supplement 2A, B). The cell-cell junction components, especially Par3, are recruited to the center of the cell doublets before positioning the centrosomes and targeting apical membrane components to the AMIS. The spatial-temporal localization of proteins involved in epithelial polarization is summarized in Table 1. The sequential order of these events and our knockdown experiments (Figure 5A-D) suggest that Par3 acts upstream of centrosome positioning and polarized vesicle trafficking.

To establish causality between polarity marker Par3 and polarized transport, we generated Rab11a-KO cells (Figure 5—figure supplement 2C–E). In Matrigel culture, we found that the Par3 signal still appears first at the cytokinesis site and that the centrosome begins to migrate toward the center of the cell doublet in Rab11a-KO cells (Figure 5E). Similarly, the complete depletion of centrosomes in centrinone-treated p53-KO cells did not affect Par3 localization (Figure 5F). These data provide evidence in the reverse direction, suggesting that centrosome positioning and polarized vesicle trafficking play a downstream role in Par3 recruitment.

Researchers have demonstrated an association between cytokinesis and the de novo polarization of epithelial cells (Klinkert et al., 2016; Li et al., 2014; Mangan et al., 2016; Rathbun et al., 2020; Schluter et al., 2009; Wang et al., 2014). Based on our results, Par3, which first localizes to the cytokinesis bridge, affects the centrosome positioning in Matrigel culture of MDCK cells. Here, we prevented cell division or changed culture conditions, which may affect the Par3 distribution in cells, to shed light on the relationship among the centrosome position, polarity regulator, and apical membrane components.

We observed cell doublets in Matrigel formed through cell-cell aggregation, not cell division (Figure 6—figure supplement 1A, see Materials and methods). Live-cell imaging showed that Gp135 is enriched on the membrane facing outward and that centrosomes remain stationary for several hours (Figure 6A). In early aggregated cell doublets, Gp135 polarity is inverted, consistent with previous studies (Bisi et al., 2020; Bryant et al., 2010; Bryant et al., 2014; Chou et al., 2016; Martin-Belmonte et al., 2007; Roman-Fernandez et al., 2018). We further performed immunostaining to examine polarity molecules. We found that Par3 was recruited to the cell-cell interface in the aggregated doublets (Figure 6B). Rab11a-positive recycling endosomes were concentrated around the centrosome (Figure 6C). Quantitative measurement θ_N-C revealed that, although centrosomes are located between the two nuclei in the doublet generated by cytokinesis (Ctrl, Figure 6D, E), centrosomes are widely distributed across both nuclei in the aggregated doublets (+Aphi, Figure 6D, E). The results of the aggregated doublets indicate that the relationships among Gp135-associated apical membrane formation, Par3 location, and centrosome positioning are distinct from that of cytokinesis-induced de novo epithelial polarization.

We next create inverted cysts by culturing suspended single cells in low-adhesion microwells (Figure 6—figure supplement 1B) (Huang et al., 2020; Yu et al., 2005); in these conditions, the normal cell cycle progresses, but cells exhibit an inverted Gp135 polarity from cells in Matrigel culture. Immunostaining shows that Par3 proteins form a ring at the edge of the cell-cell interface (Figure 6F), distinct from the Par3 patch observed at the center of the cell-cell interface for cells in Matrigel (Figure 5—figure supplement 1H).

Before and at the onset of anaphase (Figure 6G, −0:15–0:00), the centrosomes form two spindle poles at two sides of the nuclei. At this time point, there is a dispersed Gp135 signal inside the cytoplasm. When cleavage furrow ingression occurs (Figure 6G, 0:05–0:20), Gp135 is concentrated at the furrow membrane between the two daughter cells. As cytokinesis progresses, Gp135 dissociates from the middle membrane and becomes concentrated around the migrating centrosomes and external membranes (Figure 6G, 0:30, white and yellow arrows). While Gp135 spreads toward the outside membrane, the centrosomes migrate toward the center of the cell doublets (Figure 6G, 0:40–1:05, and H-J). Finally, Gp135 is concentrated at the membranes, corresponding to locations where cells are not in contact, and forms a polarization opposite to that of cysts in Matrigel. Taken together, our results identify an interesting condition in which the polarity indicated by the centrosome and nucleus is opposite to the apical-basal polarity. In this case, cytokinesis drives centrosome migration toward the center of the cell doublets, even in the absence of ECM and under conditions of inverted polarity.

Finally, we investigated the movement of centrosomes at the cytokinetic pre-abscission stage in highly polarized cysts (Figure 6—figure supplement 1C) and in an epithelial sheet (Figure 6—figure supplement 1D-E), of which both have apical-localized cytokinetic bridges. Both live cell imaging and immunostaining show that centrosomes are laterally localized at the mid-plane of the apical-basal axis in telophase cells and move apically during the cytokinetic pre-abscission stage (Figure 6— figure supplement 1C–E). Like the de novo epithelial polarization, we found that centrosomes move toward the AMIS direction before bridge abscission (Figure 6—figure supplement 1C–F). Taken together, these findings indicate that the centrosome position is strongly associated with the cytokinesis site but not with the apical membrane.

MDCK (NBL-2) cells (Bioresource Collection and Research Center, Taiwan) were tested for mycoplasma, cultured in minimum essential medium (MEM) with glucose, 2 mM L-glutamine, and 10% fetal bovine serum, and supplemented with 0.1 mM non-essential amino acids and 1 mM sodium pyruvate, penicillin, and streptomycin. We conducted 3D cell culture assays for cyst formation using a previously described method (Lee et al., 2007; Martin-Belmonte et al., 2007). We trypsinized 70% of confluent cells to a single cell suspension of 4 × 10⁴ cells/mL in complete MEM containing 2% growth factor reduced Matrigel (356231, Corning, Glendale, AZ). We plated cells in 2% Matrigel medium (300 μL) in 8-well cover-glass chamber slides (80827, ibidi, Gräfelfing, Germany) pre-coated with 8 μL of 100% Matrigel. The cells were incubated for the indicated period, and the medium was changed every 2 days.

For the centrosome depletion assay, we pre-treated MDCK WT and p53-KO cells with 300 nM centrinone (LCR-263, HY-18682, MedChemExpress, Monmouth Junction, NJ) (Wong et al., 2015) for 3 days on a petri dish to achieve >80% of cells without centrosomes. The centrosome-depleted cells were then trypsinized to a single-cell suspension and plated into a 3D culture in the continuous presence of 300 nM centrinone.

To assess cell aggregates, we treated MDCK cells with aphidicolin (Aphi) for 15 hours, arresting almost all cells in the S phase. Subsequently, we released these synchronized cells from the S phase and simultaneously cultured them in Matrigel. Based on past experience, MDCK cells require approximately 8 hours to progress from the S phase to the M phase. Therefore, we observed cell doublets at 6 hours after release to ensure that these cell doublets were formed through aggregation rather than cell division.

We used rabbit polyclonal antibodies against Gp135 (IF: 1:400, a kind gift from Dr. Jou Tzuu-Shuh, NTU, Taiwan) (Lim et al., 2017), Rab11a (IF: 1:400, WB: 1:2000, 20229-1-AP, Proteintech, Rosemont, IL), aPKC (IF: 1:200, 610175, BD Biosciences, Franklin Lakes, NJ), CEP120 (IF: 1:1000, WB: 1:2000, human residues 639–986, described in our previous paper) (Lin et al., 2013), p53 (IF: 1:100, 9282, Cell Signaling Technology, Danvers, MA), p21 (WB: 1:2000, 10355-1-AP, Proteintech), CEP164 (IF: 1:400, NBP1-81445, NOVUS, Centennial, CO), PCNT (IF: 1:200, ab4448, Abcam, Cambridge, UK), Par3 (IF: 1:200, WB: 1:1000, 07-330, EMD Millipore, Burlington, MA), and beta-actin (IF: 1:200, A2066, Sigma-Aldrich, St. Louis, MO); rabbit monoclonal antibodies against E-cadherin (IF: 1:200, 24E10, #3195, Cell Signaling Technology); and mouse monoclonal antibodies against γ-tubulin (IF: 1:400, GTU-88, T6557, Sigma-Aldrich), acetylated tubulin (IF: 1:400, T6793, Sigma-Aldrich), Gp135 (IF: 1:10, the culture medium of hybridoma clone 3F2D8, a kind gift from Dr. Jou Tzuu-Shuh, NTU, Taiwan) (Lim et al., 2017), polyglutamylated tubulin (IF: 1:100, GT335, AG-20B-0020, Adipogen, San Diego, CA), ODF2 (IF: 1:400, H00004957-M01, NOVUS), α-tubulin (WB: 1:5000, DM1α, T9026, Sigma-Aldrich), CDC42 (WB: 1:1000, 610928, BD Biosciences), and ZO-1 (IF: 1:400, 33-9100, ThermoFisher Scientific, Waltham, MA).

For western blots, we used the following secondary antibodies: HRP-labeled goat anti-rabbit and mouse IgG (1:10000, NEF812 and NEF822, PerkinElmer, Waltham, MA). For immunofluorescence, we used Alexa Fluor 488, 561, and 647 conjugated goat antibodies against rabbit, mouse IgG1, and mouse IgG2b (1:400, all purchased from ThermoFisher Scientific) as secondary antibodies, together with Phalloidin-iFluor 647 reagent (1:1000, ab176759, Abcam) and 4’,6-diamidino-2-phenylindole (DAPI) (1:10000, D8417, Sigma-Aldrich).

We stained 3D cell cultures according to a previously published protocol, with some modifications (Lee et al., 2007). We pre-fixed whole cultures in each well with 4% paraformaldehyde at room temperature for 10 min. If the staining included centrosome markers, we post-fixed the samples with pure cold methanol at −20°C for another 10 min. After fixation, cells were permeabilized with 0.25% Triton X-100 in phosphate-buffered saline (PBS). We performed primary antibody staining in 100 μL PBS plus 10% normal goat serum at 4°C for 48 h. After four 15 min washes in 250 μL PBS with 0.1% Tween 20 at room temperature, we added secondary antibodies with DAPI in 100 μL PBS plus 10% normal goat serum for another 48 hours at 4°C. Finally, we performed another four 15 min washes, followed by adding 100 μL mounting medium (VectaShield, Vector Laboratories, Newark, CA) to each well. We analyzed the samples on the LSM 880 confocal (Zeiss, Jena, Germany) using a Plan-Apochromat 63x oil immersion objective with a numerical aperture (NA) 1.4.

For fluorescence intensity measurements, confocal images were acquired with constant parameters (pinhole size, laser power, master gain, and z-stack interval) and introduced to Fiji. We applied the maximum intensity projection to project the z-stack. The oval selection tool was fixed on size to select the region of interest (ROI) (e.g., pericentrosomal region, cytokinesis bridge, cytosol, AMIS, or apical lumen) and to measure the mean fluorescence intensity of the ROI.

We considered the polarity index, θ_N-C, L_O-C, and L_IC of cell doublets cultured in 3D conditions to assess centrosome positions. We obtained the centroid of each cell nucleus and centrosome, as well as the volume of the cell nucleus, using Imaris software (version 9.8.0, Bitplane, Belfast, UK). The cell nucleus radius, r_N, was determined by dividing its volume by 4/3π and then taking the cube root. We calculated the polarity index from the nucleus–centrosome distance, L_N-C, divided by the r_N.

The nucleus–centrosome orientation (θ_N-C) is represented by the angle between the line from the nucleus to the centrosome and the line from the nucleus to the other nucleus. The central points of each nucleus and centrosome were marked as three points (nuclei: n1 and n2; centrosome: c) in 3D space to give three x, y, z coordinates. We computed the angle between the vector of the nucleus to the centrosome and the vector of the nucleus to the other nucleus using the vector dot product formula. Centrosomes oriented within 90° from the nucleus–nucleus axis were considered to be correctly oriented toward the apical region.

The distance between the centrosome and apical membrane was represented by the distance from the origin point (the center point of the two nuclei) to the centrosome, L_O-C, because the AMIS and apical lumen will form at the center of the cell doublet in Matrigel culture. We also calculated the distance using x, y, z coordinates in 3D space. A smaller distance indicates that the centrosome and apical membrane are closer. The distance between the centrosomes in each cell of a cell doublet is referred to as the inter-centrosomal distance, L_IC.

We obtained stable MDCK cell lines that consistently expressed EGFP-rabbit podocalyxin (Gp135, complementary DNA [cDNA] was a kind gift from K. Simons), EGFP-Crb3 (cDNA was obtained by reverse transcription polymerase chain reaction [RT-PCR] following previously described procedures) (Schluter et al., 2009), EGFP-Cdc42, or PACT-mKO1 using the following protocol. MDCK cells were transduced with lentiviruses that package with the pLAS2w.Ppuro vector (RNAi Core, Taiwan) containing the relevant cDNA. The transduction processes were performed in complete MEM with 8 μg/mL polybrene. The transduced cells were sterile-sorted by a cell sorter (FACSJazz, BD Biosciences) for EGFP or mKO1. Fluorescence-positive cells were selected and expanded as stable lines.

We resuspended stable MDCK lines in a culture medium containing 2% Matrigel and plated the cells in a rose chamber with the cover-glass bottom pre-coated by 100% Matrigel. We allowed the cells to remain in the incubator for 2 hours to ensure that the cells fell and were fully adherent on the Matrigel. We labeled the nucleus in live cells with a far-red fluorescence probe, SiR-DNA (500 nM, SC007 Spirochrome, Cytoskeleton Inc., Denver, CO), in phenol red-free medium containing 2% Matrigel.

After a 1 hour incubation, we performed live cell imaging on a 40x NA1.1 water immersion objective (Leica, Wetzlar, Germany) on an inverted microscope (DMI 6000, Leica) with a confocal scan head (CSU22 Spinning Disk, Yokogawa Electric, Tokyo, Japan), a laser merge module containing 405-, 491-, 561-, and 642-nm laser lines (ILE-400, Andor, Belfast, UK), a DV885 EMCCD (Andor), an MS-2000-XYZ motorized stage with a piezo top plate (ASI, Eugene, OR), and a stage-top incubator set to 37°C under 5% CO₂. The microscope apparatus was controlled by MetaMorph 7.7.2 (Molecular Devices, San Jose, CA). We set the time interval to 5 min and z steps to 0.5 μm.

We collected images of mitotic cells to track centrosome migration during de novo epithelial polarization. These images were imported to Imaris software (version 9.8.0, Bitplane) for 3D reconstruction and analysis. We marked the centroid of each nucleus and centrosome as x, y, z coordinates in 3D space for each time point. We calculated the polarity index, θ_N-C, and L_O-C from the first frame of anaphase onset.

To measure the fluorescence intensity of EGFP-Gp135 on the middle membrane of the 3D cell doublet, we measured the intensity profile of EGFP-Gp135 along the line connecting the two nucleus spots. Then, we recorded the average value of the five center pixels representing the region of the middle membrane from the first frame of anaphase onset.

To generate MDCK lines lacking either p53, CEP164, ODF2, CEP120, PCNT, or Rab11a, we used CRISPR/Cas9 technology. Because CEP120-null cells cannot survive in the presence of WT p53 (Tsai et al., 2019), we performed a knock-out of CEP120 in MDCK p53-null cells. We transfected MDCK cells using a Lipofectamine 2000 transfection reagent (11668019, ThermoFisher Scientific) with the pSpCas9(BB)-2A-GFP (PX458) vector (Ran et al., 2013) (648138, Addgene, Watertown, MA) bearing the appropriate targeting sequence:

p53 #1 (+): 5’-TCCCAGAGAGCGTCGTGAAC-3’; p53 #2 (-): 5’-ACTTGGCTGTCCCGGAATGC-3’; CEP164 #1 (-): 5’-GTCCCACATGGACTGCCCGT-3’;CEP164 #2 (+): 5’-AACTTGGTGATTCAAGAGCG-3’; ODF2 #1 (+): 5’-CCTCATGTCCAAGCTGGTAG-3’; ODF2 #2 (+): 5’-ATCGGGAAGCTGAAGACGGT-3’;CEP120 #1 (+): 5’-ACAGTTGGCTACCGACCCTG-3’; PCNT #1 (+): 5’-TCGCCAGGGAGCAGCATGCG-3’; Rab11a #1 (-): 5’-GCGACGACGAGTACGACTAT-3’;Rab11a #2 (-): 5’-AGCTATAACATCAGCGTAAG-3’

Cells were sterile-sorted by a cell sorter (FACSJazz, BD Biosciences) for GFP. Fluorescence-positive cells were serially diluted to pick single colonies and were expanded to pure clones. Genomic DNAs isolated from each clone were subjected to PCR and sequencing for verification. We examined protein expression levels in each clone via immunofluorescence microscopy or western blotting.

We achieved RNA interference (RNAi) using pLKO.1-EGFP (TRC011, RNAi Core, Taiwan) lentiviral vector. RNAi target sequences were based on previously published reports (Bryant et al., 2010; Horikoshi et al., 2009; Roland et al., 2011; Schluter et al., 2009) for canines, as follows:

Rab11a #1: 5’-AAGGCACAGATATGGGACACG-3’; Rab11a #2: 5’-ATGGTTTGTCATTCATTGAGA-3’; Sec15a #1: 5’-AAGAGGATGAGAATGAAGAGG-3’;Sec15a #2: 5’-AAGCACGGGTCATGATAGTTT-3’; Par3 #1: 5’-GACAGACTGGTAGCAGTGT-3’; Par3 #2: 5’-AGGATAAAGCTGGCAAAGA-3’; MyoVb #1: 5’-AAGGTGGAGTATCTTTCAGAT-3’; MyoVb #2: 5’-AACGGGTAACTGTGGCCTTTA-3’; Cdc42 #1: 5’-GCGATGGTGCCGTTGGTAA-3’; Cdc42 #2: 5’-GATTACGACCGCTGAGTTA-3’; Crb3: 5’-GCTTAAGAGTAGAAGGGAA-3’

We verified knockdown by western blot or quantitative real-time PCR procedures normalized to GAPDH expression. The designed primer sequences were as follows:

Sec15a Fwd: 5’-GTCAGCCTGCCAGCATCTGT-3’; Sec15a Rev: 5’-CTGCTGAACAGCTCCCATGC-3’; MyoVb Fwd: 5’-CAAGTTCCCACTGGTGGCTG-3’;MyoVb Rev: 5’-GAAGACGAGCCCTTCCCAGA-3’; Crb3 Fwd: 5’-GAGAACGGCACCATTACACC-3’; Crb3 Rev: 5’-GAGGGAGAAGACCACAATGAT-3’;GAPDH Fwd: 5’-AGTCAAGGCTGAGAACGGGAAACT-3’; GAPDH Rev: 5’-TGTTTGTGATGGGCGTGAACCATG-3’

Following a previously published protocol (Huang et al., 2020), we generated multiarray spherical microwells for an ECM-free suspension culture. In brief, 100 μL polyacrylamide (PA) solution containing 7% acrylamide monomer (Bio-Rad, Hercules, CA), 0.35% bisacrylamide crosslinker (Bio-Rad), 0.15% N, N, N’, N’-tetramethylethylenediamine catalyst (Sigma-Aldrich), and 0.5% ammonium persulfate (Sigma-Aldrich) was sandwiched between an epoxy mold and 3-(trimethoxysilyl) propyl methacrylate silanized cover glass (22 × 22 mm). Waiting for the complete gelation of PA, we carefully separated the cover-glass-bound microwell substrate from the epoxy template. To avoid cell adhesion, we conjugated poly-L-lysine/polyethylene glycol to the PA substrate using the bi-functional crosslinker sulfo-SANPAH (ThermoFisher Scientific). We sterilized low-attachment microwell substrates with antibiotics and then seeded a single-cell suspension of MDCK stable lines onto the substrate surface. After cells fell into the microwells, we rinsed the PA substrate with the culture medium to remove excess cells outside the microwells. The samples were assembled into a rose chamber and imaged by a CSU22 spinning disk confocal microscope, as described above, for 3D live cell imaging in Matrigel.

To grow polarized epithelial cell monolayers suitable for observation under an inverted microscope, we cultured MDCK cells on the outer surface of Transwell inserts (3470, Corning). Typically, 50 μL of a suspension containing 4.25 × 10⁵ cells/mL was seeded into each 6.5-mm Transwell insert. Cells were cultured for 3 days before immunofluorescence staining or live cell imaging. We mounted Transwell inserts in a custom-made glass-bottom chamber for confocal imaging. For live cell imaging, we stained epithelial cell monolayers with SiR-tubulin (500 nM, SC002, Spirochrome) and monitored the cells using LSM 980 confocal microscopy (Zeiss).

Statistical analyses utilized GraphPad Prism 9 and OriginPro 2021, employing the Mann-Whitney U test and one-way and two-way analysis of variance (ANOVA) with multiple comparisons. Details on statistical tests and sample sizes are provided in the figure legends. n denotes the sample size, and the boxplot boxes indicate the standard deviation (SD).