Control of apico–basal epithelial polarity by the microtubule minus-end-binding protein CAMSAP3 and spectraplakin ACF7

Microtubules regulate cell polarity by serving as rails for intracellular transport and by controlling organelle positioning and signaling. Although in many types of cultured cells microtubules form a radial array, in polarized epithelial cells, microtubules are arranged along the apico–basal axis, with the minus ends located at the apical side (Akhmanova and Hoogenraad, 2015; Bartolini and Gundersen, 2006; Dammermann et al., 2003). This non-centrosomal microtubule organization can be controlled by the relocalization of γ-tubulin-binding centrosomal microtubule-anchoring proteins, such as ninein, to non-centrosomal apical sites (Moss et al., 2007) and by proteins that bind to microtubule minus ends independently of γ-tubulin. The most notable example of the latter are the members of calmodulin-regulated spectrin-associated protein (CAMSAP)/Patronin family, which can recognize and decorate free microtubule minus ends and protect non-centrosomal microtubules from depolymerization (Goodwin and Vale, 2010; Hendershott and Vale, 2014; Jiang et al., 2014; Tanaka et al., 2012). Recent work in worms has demonstrated that a ninein homolog and Patronin can act both redundantly and non-redundantly in organizing non-centrosomal arrays in different cell types (Wang et al., 2015).

In mammalian cells, the minus-end-binding protein CAMSAP3 (also known as Nezha or Marshalin) was initially shown to interact with components of adherens junctions and to anchor microtubule minus ends at these sites (Meng et al., 2008). CAMSAP3 is also very abundant in supporting cells of the organ of Corti, a specialized type of epithelial cell with extremely dense microtubule bundles that are anchored at centrosomal and non-centrosomal microtubule-organizing centers at the cell cortex (Takahashi et al., 2016; Zheng et al., 2013). Similar to other CAMSAP/Patronin-family members, CAMSAP3 has a C-terminal conserved domain, the CKK, as well as several coiled-coil regions and an N-terminal calponin homology domain, the function of which is currently unknown (Akhmanova and Hoogenraad, 2015; Baines et al., 2009). Through the CKK domain and the adjacent positively charged microtubule-binding regions, CAMSAP3, as well as its homolog CAMSAP2 recognize and decorate polymerizing microtubule minus ends, forming stretches of stabilized microtubule lattice that can serve as ‘seeds’ for non-centrosomal microtubule outgrowth (Jiang et al., 2014). Recent work shows that CAMSAP3 is an important factor in organizing apico–basal microtubule arrays in enterocytes, where it strongly localizes to the apical cell cortex and tethers microtubule minus ends (Toya et al., 2016). Mice homozygous for the C-terminal truncation of CAMSAP3, in which the CKK domain is lost, are viable but show increased lethality at early postnatal stages (Toya et al., 2016). Intestinal epithelial cells of those mice could still polarize and form a normal brush border, in spite of the fact that the microtubules were disorganized, and the nuclei and Golgi membranes mispositioned (Toya et al., 2016). Experiments in Caco-2 cells, a human intestinal cancer cell line, have demonstrated that the first coiled-coil region of CAMSAP3 is involved in recruitment of CAMSAP3-bound microtubule minus ends to the apical side (Toya et al., 2016). In CAMSAP1, this coiled-coil region was shown to bind to spectrin (King et al., 2014). Whether the apical localization of CAMSAP3, indeed, depends on spectrin or some other proteins has not been determined.

Here, we set out to investigate the mechanism of CAMSAP3 recruitment to the apical side of epithelial cells. We found that CAMSAP3, but not CAMSAP2, interacts with the spectraplakin ACF7 (also known as MACF1; Karakesisoglou et al., 2000; Kodama et al., 2003; Wu et al., 2008) and that ACF7 depletion leads to the release of CAMSAP3-decorated microtubule stretches from the apical cell surface. ACF7, CAMSAP2 and CAMSAP3 were not essential for the distribution of early cell polarity markers in two-dimensional (2D) cultures; however, loss of ACF7 but not of the two CAMSAP proteins affected the formation of epithelial cysts in 3D cultures, indicating that ACF7 plays a more profound role in epithelial polarity than the CAMSAP proteins. Loss of CAMSAP3 led to more centrosome-centered microtubule organization, and this affected the organization of the actin cytoskeleton at the apical side of the cell, probably owing to mislocalization of recycling endosomes that are positive for Rab11A, which are an important component in apical polarity and brush border formation in intestinal cells (Bryant et al., 2010; Dhekne et al., 2014; Knowles et al., 2015; Overeem et al., 2015; Sobajima et al., 2015). Taken together, our study identifies ACF7 as an important regulator of apico–basal polarity in mammalian intestinal cells and demonstrates the importance of non-centrosomal microtubule organization for efficient epithelial cell polarization.

To facilitate visualization of CAMSAP3 in Caco-2 cells, we have generated a homozygous knock-in line in which the endogenous CAMSAP3 protein was tagged with GFP (Fig. 1A). In agreement with previous publications (Meng et al., 2008; Tanaka et al., 2012; Toya et al., 2016), in this cell line, CAMSAP3–GFP localized to microtubule ends, colocalized with endogenous CAMSAP2 and was concentrated at the apical side of polarized Caco-2 monolayers (Fig. 1B–D). Upon microtubule depolymerization through nocodazole treatment, no apical CAMSAP3 stretches were detected, indicating that the apical recruitment of CAMSAP3 strongly depends on the presence of microtubules (Fig. 1E). We next used fluorescence recovery after photobleaching (FRAP) assays to investigate the turnover of CAMSAP3–GFP at the apical side of polarized Caco-2 monolayers. We found that in 1-day-old Caco-2 monolayers, a ∼75% recovery of CAMSAP3–GFP signals at the apical surface occurred within ∼15 min (Fig. 1F,G). This recovery was associated with movements of unbleached CAMSAP3–GFP-labeled microtubule minus ends into the photobleached area (Fig. 1F,H). Interestingly, ∼15 min after photobleaching in 5-day-old monolayers, the CAMSAP3–GFP signal at the apical surface recovered only to ∼40% (Fig. 1F,G). The mobility of CAMSAP3–GFP stretches in older cultures was much lower (Fig. 1H), indicating that the apical cytoskeleton undergoes maturation on a scale of several days, resulting in immobilization of CAMSAP3 stretches.

When Caco-2 cells were cultured in Matrigel, they formed cysts, as described previously (Jaffe et al., 2008). Already at the two-cell stage, GFP–CAMSAP3 colocalized with apical markers, such as atypical protein kinase C (aPKC), and the area of enhanced actin accumulation, which was located at the membrane interface between the two daughter cells, as shown previously (Jaffe et al., 2008) (Fig. 1I). After 4 days in 3D culture, GFP–CAMSAP3 strongly accumulated at the apical side (Fig. 1J). In such polarized cysts, microtubules were aligned along the apico–basal axis (Fig. 1K).

We never observed any accumulation of GFP–CAMSAP3 at the cell–cell junctions in either 2D or 3D cultures, which is in line with the observations in mouse intestine but contradicts the data previously obtained by immunostaining of CAMSAP3 in Caco-2 cells (Meng et al., 2008; Toya et al., 2016). Our data are thus in full agreement with the role of CAMSAP3 in apical microtubule organization but provide no support for CAMSAP3 function at the adherens junctions.

We reasoned that the capture and stable association of CAMSAP3 stretches at the apical surface is caused by the presence of a binding partner. To identify CAMSAP3-interacting proteins, we performed streptavidin pull-down assays from HEK293T cells that co-expressed biotinylation (Bio) and GFP-tagged CAMSAP3 and the biotinylated ligase BirA, and analyzed the isolated proteins by performing mass spectrometry (Jiang et al., 2014). The highest scoring hit in this screen was the spectraplakin ACF7 (Fig. 2A), whereas no peptides that were derived from spectrin, which has been previously reported to bind to CAMSAP1 (King et al., 2014), were found. Whether this was owing to the absence of appropriate spectrin isoforms in HEK293T cells, the biochemical procedure used or the lack of interaction between CAMSAP3 and spectrin is unclear.

Immunostaining showed colocalization of GFP–CAMSAP3 with the puncta of endogenous ACF7 at the apical side of Caco-2 monolayers (Fig. 2B). However, in contrast to CAMSAP3, which was only located apically, ACF7 was spread throughout the cytoplasm (Fig. 2D). As an alternative model of enterocyte polarity, we used the LS174T-W4 cell line, in which the activation of LKB1 kinase through doxycycline-induced expression of the adapter protein STRAD (also known as STRADA) is sufficient to induce polarization and formation of microvilli on one side of the cell in the absence of cell–cell contacts (Baas et al., 2004). In doxycycline-treated LS174T-W4 cells, both endogenous and GFP-tagged CAMSAP3 colocalized with endogenous ACF7 beneath the polarized regions of actin enrichment (Fig. 2C), supporting complex formation between CAMSAP3 and ACF7.

We next investigated the interaction between CAMSAP3 and ACF7 biochemically. ACF7 is a very large protein (∼5000 amino acids) containing an N-terminal actin-binding calponin homology domain, a plakin domain, 30 spectrin repeats, EF-hand motifs that could mediate Ca²⁺ binding, a microtubule-binding GAS2-related domain and the EB1-binding SxIP motif (Karakesisoglou et al., 2000; Kodama et al., 2003; Wu et al., 2008) (Fig. 2E). Because the full-length ACF7 is difficult to efficiently express in cells, we generated GFP-tagged ACF7 fragments covering the whole ACF7 protein. We performed streptavidin pull-down assays from HEK293T cells co-expressing these fragments together with biotin ligase BirA and CAMSAP2 or CAMSAP3 tagged with an mCherry and biotinylation tag (Bio–mCherry), or just the Bio–mCherry protein as a negative control. We found that CAMSAP3, but not CAMSAP2, specifically bound to the ACF7 fragment encompassing spectrin repeats 20–26 (fragment ACF7_5, Fig. 2F). Deletion mapping further showed that the coiled-coil region 1 (CC1) of CAMSAP3, which has been previously implicated in the apical localization of the protein (Toya et al., 2016), was essential for this interaction (Fig. 2G).

The binding between CAMSAP3 and ACF7 was confirmed by performing colocalization experiments in HeLa cells. Overexpression of CAMSAP3 and CAMSAP2 caused strong bundling of microtubules, and CAMSAP3, but not CAMSAP2, could specifically recruit fragment ACF7­_5 to these microtubule bundles (Fig. 2H). Furthermore, we found that endogenous ACF7 was recruited to microtubule bundles that had been induced through overexpression of full-length CAMSAP3 and of the CAMSAP3 deletion mutant that lacked the calponin homology domain, but not of the CAMSAP3 deletion mutants that lacked the CC1 region (Fig. S1). Taken together, these data indicate that CAMSAP3 and ACF7 interact with each other and can cooperate in binding to microtubules. Deletion mapping showed that the CC1 region of CAMSAP3 and a specific spectrin-repeat region of ACF7 are required for the interaction.

We next investigated whether ACF7 participates in the apical recruitment of CAMSAP3. ACF7 was depleted with two different small interfering (si)RNAs, and the efficiency of depletion was confirmed by immunofluorescence staining and real-time PCR (Fig. 3A–C). Although actin enrichment was still observed at the apical side of ACF7-depleted Caco-2 monolayers, both the apical localization and the total number of CAMSAP3–GFP stretches were very strongly reduced, indicating that ACF7 regulates the localization as well as the formation or stability of CAMSAP3-decorated microtubule stretches (Fig. 3D,E). In contrast, other early polarity markers, such as ezrin, aPKC and zonula occludens 1 (ZO-1; also known as TJP1) were still apically located in 2D cultures of ACF7-depleted cells (Fig. 3D; Fig. S2A). Also, the height of polarized monolayers was not significantly affected in the absence of ACF7 (Fig. S2B). Further, in line with the fact that ACF7 depletion reduced the abundance of CAMSAP3-decorated microtubule stretches, we found an increase, albeit a mild one, in the number of centrosomal microtubules, indicating that ACF7 contributes to some extent to the maintenance of a non-centrosomal microtubule network (Fig. S2C,D).

To further address the effect of ACF7 knockdown, we analyzed cyst formation in 3D cultures. To accelerate lumen formation in Caco-2 cysts in order to make 3D analysis compatible with a transient protein depletion protocol, we used dense Caco-2 3D cultures and, after the initial cyst formation, treated them with cholera toxin, which strongly stimulates lumen formation by inducing cAMP signaling (Jaffe et al., 2008). Under these conditions, we found a clear well-formed or mildly disorganized lumen in ∼75% of cysts that had been treated with control siRNAs, whereas after ACF7 depletion, only 10% of the cysts developed normally, and cysts completely lacking the lumen were prevalent (Fig. 3F,G). Analysis of early (3–4 cell) 3D cultures showed that although the apical polarity markers localized to intercellular junctions in control cells, as described previously (Jaffe et al., 2008), they were mislocalized in ACF7-knockdown cells (Fig. 3H).

We conclude that although ACF7 depletion does not cause mislocalization of all polarity markers in 2D cultures, it does lead to the loss of apical localization of CAMSAP3. When embedded in a 3D matrix, the distribution of all investigated polarity markers and cyst formation were strongly affected by ACF7 depletion, indicating that distinct mechanisms with differential sensitivities to ACF7 levels control the ability of cells to polarize in 2D and 3D cultures.

We next set out to address the effects of the loss of CAMSAP2 and CAMSAP3 on epithelial polarity. Using CRISPR-Cas9 technology, we knocked out CAMSAP2 or CAMSAP3 genes in the CAMSAP3-GFP knock-in cell line (Fig. 4A,B). Apical CAMSAP3–GFP signals were unperturbed in CAMSAP2-knockout cells, whereas in CAMSAP3-knockout cells, CAMSAP2 stretches became longer, as described previously (Tanaka et al., 2012) (Fig. 4A). CAMSAP-decorated microtubule stretches are generated through microtubule polymerization from the minus end, and the stretches formed in the presence of CAMSAP3 and CAMSAP2 are likely to be shorter than those formed in the presence of CAMSAP2 alone, because CAMSAP3 is a more efficient inhibitor of microtubule minus-end growth than CAMSAP2 (Jiang et al., 2014). Microtubule organization was not visibly changed in CAMSAP2-knockout cells, but a more centrosome-centered array was observed in CAMSAP3-knockout cells, in line with the findings of a previous study (Tanaka et al., 2012) (Fig. 4C,D).

Next, we analyzed polarity in CAMSAP2- and CAMSAP3-knockout clones. In 2D cultures, junctional ZO-1 and E-cadherin, as well as the early markers of apical polarity aPKC and ezrin, were unperturbed (Fig. S3A,B). When cultured in 3D, two out of three CAMSAP2-knockout clones and three out of three CAMSAP3-knockout clones formed morphologically normal cysts (Fig. S3C,D). Because two CAMSAP2-knockout lines could form morphologically normal cysts, we think that the morphological abnormalities in the third clone were due to secondary mutations. This conclusion was confirmed by siRNA-mediated depletion of CAMSAP2, CAMSAP3 or both CAMSAPs simultaneously (Fig. S3E). In addition, CAMSAP2- and CAMSAP3-knockout clones showed no defects in the recruitment of the apical markers aPKC and ZO-1 in early 3D cysts (Fig. S3F). We conclude that CAMSAP2 and CAMSAP3 are not required for the formation of lumen-containing 3D cysts, indicating that the strongly affected cyst formation in ACF7-depleted cells is unlikely to be due to CAMSAP3 mislocalization.

We then focused on late markers of epithelial polarity that are associated with the ability of the cells to form a brush border. Apical accumulation of actin and the phosphorylated form of ezrin, an actin-binding protein important for the formation of brush border (Dhekne et al., 2014; Overeem et al., 2015; Viswanatha et al., 2012) was unaffected in CAMSAP2-knockout cells (Fig. 4E,F; Fig. S3A), even though the same cell line showed abnormalities in cyst morphology in 3D culture (Fig. S3C,D; CAMSAP2 KO #1). This result indicates that the ability of cells to form 3D cysts and to localize actin regulators does not necessarily correlate.

Strikingly, in CAMSAP3-knockout cells, the apical accumulation of actin and phosphorylated ezrin was concentrated around the cell–cell junctions and reduced in the central part of the cell (Fig. 4E,F; Fig. S3A). Previous work has shown that during brush border formation, the apical actin cytoskeleton is strongly regulated by signaling and/or trafficking processes that are dependent on the apical accumulation of Rab11A-positive recycling endosomes (Dhekne et al., 2014; Knowles et al., 2015; Overeem et al., 2015; Sobajima et al., 2015). We analyzed the distribution of Rab11A endosomes by using cells that had been stably transfected with a construct for doxycycline-inducible expression of dTomato–Rab11A. Strikingly, in control and CAMSAP2-knockout cells, Rab11A endosomes were distributed under the apical surface, whereas in CAMSAP3-knockout cells they displayed a single focus of accumulation, which coincided with the centrosome (Fig. 4G,H; Fig. S3G). In fact, the fluorescence intensity of the Rab11A clusters in CAMSAP3-knockout cells appeared to be ∼5-fold higher than the fluorescence intensity of apically located Rab11A vesicles in control cells, revealing a very strong clustering of endosomes (Fig. S3G). This result was confirmed in cells in which CAMSAP3 had been depleted using siRNA: in these cells, the formation of a single centrally located Rab11A cluster strongly correlated with the efficiency of CAMSAP3 knockdown (Fig. 4I). We also analyzed Rab11A in ACF7-knockdown cells and found that the endosomes were equally distributed over the cell, indicating defects in the apical targeting (Fig. S3G). However, actin still accumulated at the apical surface of ACF7-depleted cells (Fig. 3D), suggesting that clustering of Rab11A endosomes in the middle part of the cell is more deleterious for actin organization than their random distribution.

Our data support the notion that the localization of Rab11A endosomes strongly correlates with the position of microtubule minus ends, in line with the idea that it is controlled by cytoplasmic dynein (Horgan et al., 2010; Khanal et al., 2016; Riggs et al., 2007). When the microtubule array becomes more radial, Rab11A endosomes accumulate in the central part of the cell, and the formation of apical actin-rich structures is inhibited.

In this study, we showed that the spectraplakin ACF7 is a binding partner of CAMSAP3 and that it is required for regulating the abundance and apical localization of CAMSAP3-decorated microtubule stretches. While this paper was in preparation, the Drosophila homologues of CAMSAP3 and ACF7, Patronin and Shortstop (Shot), respectively, were demonstrated to be present in the same apical complex and cooperate with each other, as well as an apical form of spectrin, in organizing apico–basal microtubule arrays in fly epithelia (Khanal et al., 2016) and in early fly embryos (Nashchekin et al., 2016). These findings are also consistent with earlier work in flies, which has demonstrated the importance of Shot for microtubule minus-end organization during tubulogenesis of the embryonic salivary glands (Booth et al., 2014). ACF7 appears to act upstream of CAMSAP3 as a cortical recruitment factor but, given that it has a microtubule-interacting domain, it is also likely to synergize with CAMSAP3 to some extent for microtubule binding and stabilization. We note that the accumulation of CAMSAP3 at the apical cortex occurred in the form of microtubule stretches and thus depended on the presence of intact microtubules. This is similar to the recruitment of CAMSAP2 to the Golgi membranes, where binding of CAMSAP2-decorated microtubule minus ends but not of the cytosolic CAMSAP2 was observed (Wu et al., 2016). It is possible that the presence of multiple CAMSAP molecules on the same microtubule is required for efficient binding to intracellular structures such as the Golgi and the actin-rich cortex by increasing avidity.

Because ACF7 does not show an exclusively apical localization in Caco-2 cells, an additional pathway ensuring tethering of CAMSAP3-decorated stretches specifically to the apical surface must exist. It is possible that, similar to flies, a specific apical isoform of spectrin might be involved; this would mean that the same region of CAMSAP3, the CC1, mediates apical localization by binding to two different proteins. Alternatively, some signaling pathway restricting CAMSAP3 binding to microtubules at the apical cell side might be involved. Interesting in this respect is our observation that CAMSAP3-decorated microtubule stretches become less mobile during monolayer maturation, suggesting that as epithelial cells differentiate, more numerous connections between these stretches and the cortical cytoskeleton are formed, or the cortical cytoskeleton itself becomes less dynamic. It is also important to note that analysis of 3D cysts showed that ACF7 depletion has a stronger impact on cell polarity than loss of CAMSAP proteins, indicating that ACF7 performs additional CAMSAP-independent functions during epithelial polarization.

Furthermore, our data showed that the knockout of CAMSAP3 affected apical actin organization, probably because the more centrosomal microtubule system in CAMSAP3-knockout cells prevents apical accumulation of Rab11A endosomes. How can these results be reconciled with the presence of a normal brush border in mice expressing nonfunctional CAMSAP3 (Toya et al., 2016)? The most likely explanation is the gradual centrosome inactivation observed during long-term differentiation of epithelial cells. In line with this idea, in CAMSAP3-knockout mice, microtubules in epithelial cells become disorganized but do not form a radial pattern (Toya et al., 2016). A disorganized non-centrosomal array is likely to delay but not to block apical localization of Rab11A endosomes or other yet unknown polarity factors that are transported towards microtubule minus ends. In contrast, the presence of a centrosome seems to act in a dominant-negative manner, by sequestering these factors in the central part of the cell. In line with this idea, actin accumulation at the apical side was inhibited more strongly in CAMSAP3-knockout cells, in which Rab11A endosomes were clustered, than in ACF7-knockdown cells, in which the endosomes were distributed throughout the cell. These data suggest that although the centrosome can promote cell polarization under certain conditions (Bornens, 2012), in epithelial cells it acts as a polarity inhibitor and its microtubule-anchoring activity must be suppressed to allow the cells to efficiently complete their polarization program.

Caco-2 cells were obtained from Alpha Yap (University of Queensland, Australia) (Ratheesh et al., 2012) and cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum (FCS). LS174-W4 cells (Baas et al., 2004) were obtained from Johannes L. Bos (UMC Utrecht, The Netherlands) and cultured in RPMI supplemented with 10% FCS. HEK293T cells were obtained from American Type Culture Collection (ATCC) and cultured in DMEM supplemented with 10% FCS. The cell lines were routinely checked for mycoplasma contamination using LT07-518 Mycoalert assay (Lonza). The identity of the cell lines was monitored by immunofluorescence-staining-based analysis with multiple markers. To grow a polarized monolayer of epithelial cells, Caco-2 cells were cultured for 5 days after reaching a fully confluent monolayer. 3D cultures of Caco-2 cells were generated by seeding cells on top (lumen formation assay) or within (3D immunofluorescence staining) 9.7 mg/ml Matrigel (Corning). One day after seeding, cells were treated with 100 ng/ml cholera toxin (Sigma-Aldrich).

FuGENE 6 (Promega) was used to transfect different plasmids into Caco-2 and LS174T-W4 cells. Polyethylenimine (PEI, Polysciences) was used to transfect HEK293T cells for streptavidin pull-down assays. HiPerFect (Qiagen) was used to transfect Caco-2 cells with siRNAs at 20 nM. Corresponding experiments were performed 72 h after siRNA transfection.

Streptavidin pull down and western blotting were performed as described previously (Jiang et al., 2014). Samples were prepared from pull-down assays of biotinylated proteins from extracts of transfected HEK293T cells using streptavidin beads, as described previously (Jiang et al., 2014), and were resuspended in 10% formic acid with 5% DMSO and were analyzed with an Agilent 1290 Infinity (Agilent Technologies) liquid chromatography instrument, operating in reverse-phase (C18) mode, coupled to a TripleTOF 5600 (AB Sciex) spectrometer. Mass spectrometry spectra (350–1250 m/z) were acquired in high-resolution mode (R>30,000), whereas tandem mass spectrometry (MS2) spectra were acquired in high-sensitivity mode (R>15,000). Raw files were processed using Proteome Discoverer 1.4 (version 1.4.0.288, Thermo Scientific, Bremen, Germany). The database search was performed using Mascot (version 2.4.1, Matrix Science, UK) against a SwissProt database (taxonomy human). Carbamidomethylation of cysteines was set as a fixed modification and oxidation of methionine was set as a variable modification. Trypsin was specified as the enzyme and up to two missed cleavages were allowed. Data filtering was performed using percolator, resulting in 1% false discovery rate (FDR). Additional filters were: search engine rank 1, peptide confidence high and ion score >20.

Caco-2 CAMSAP3–GFP knock-in in CAMSAP-knockout cells were generated by using CRISPR–Cas9 technology (Ran et al., 2013). The Caco-2 CAMSAP3–GFP knock-in line was generated by transfecting the cells with the pSpCas9(BB)-2A-Puro (PX459) vector (Addgene, #62988) bearing the appropriate targeting sequence (5′-ACCGCCCGGGTGGGGCTATT-3′) and the donor plasmid. Caco-2 CAMSAP-knockout cell lines were generated by transfecting the cells with PX459 vector bearing appropriate targeting sequences (CAMSAP2: 5′-CATGATCGATACCCTCATGA-3′; CAMSAP3: 5′-GTACGATTTCTCGCGGGCCA-3′).

To establish clonal stable knock-in and knockout lines, cells were subjected to selection with 20 µg/ml puromycin for 3 days. After selection, cells were allowed to recover in normal medium for 5 days, and the efficiency of knock-in or knockout was checked by immunofluorescence staining. Depending on the efficiency, 20–100 individual clones were isolated and characterized by immunofluorescence staining.

Protein domains were predicted using the web-based simple modular architecture research tool (SMART) (Letunic et al., 2015; Schultz et al., 1998). Full-length ACF7 was purchased from Promega Kazusa Genome Technologies (pFN21AE0600). ACF7 constructs were amplified by performing PCR and cloned into pEGFP-C2 (Clontech). Bio–mCherry–CAMSAP2, bio–mCherry–CAMSAP3 and GFP–CAMSAP3 have been described previously (Jiang et al., 2014). Deletion mutants were cloned by using PCR-based strategies.

Gibson assembly was used to assemble a pUC19-based CAMSAP3–GFP knock-in donor plasmid. Primers used to amplify the 5′ homology arm were: 5′-CCTTGGCCCCTCTGCACATA-3′ and 5′-TTTGGGGGTGCCGCCGCC-3′. Primers used to amplify the 3′ homology arm were: 5′-CCCCACCCGGGCGGTCCA-3′ and 5′-TTAGTCCTAAGCCTGGGAAGC-3′.

pSIN-TRE-rtTA-IRES-Hygro-iLID-dTomato-Rab11A was constructed from pSIN-TRE-rtTA-IRES-Puro (kindly provided by Benjamin Bouchet, Utrecht University, The Netherlands) by replacing the puromycin-resistance cassette with the hygromycin-resistance cassette from pCDNA5-FRT-TO (Invitrogen) and encodes amino acids 2–216 human Rab11A (derived from GFP–Rab11A WT, a gift from Richard Pagano, Mayo Clinic College of Medicine, Rochester, MN; Addgene plasmid #12674), N-terminally fused to the light-sensitive dimerization module iLID (derived from pLL7.0:Venus-iLID-Mito, a gift from Brian Kuhlman, University of North Carolina, Chapel Hill, NC; Addgene plasmid #60413) and dTomato.

siRNAs targeting the following sequences were purchased from Sigma-Aldrich: ACF7 #1, 5′-TTGCAGCAGGTGAATGGAC-3′; ACF7 #2, 5′-CCAAAGTGACTTGAAGGAT-3′ (Drabek et al., 2006); CAMSAP2, 5′-GTACTGGATAAATAAGGTA-3′ (Jiang et al., 2014); CAMSAP3, 5′-GCATTCTGGAGGAAATTGA-3′; and control targeting luciferase, 5′-CGTACGCGGAATACTTCGA-3′ (Dambournet et al., 2011).

Rat monoclonal antibody against tyrosinated α-tubulin (clone YL1/2) was purchased from Abcam (#ab6160). Mouse polyclonal antibody against ACF7 was purchased from Abnova (#H00023499-A01). Mouse monoclonal antibodies against the following proteins were used: CAMSAP3 (#SAB4200415) and γ-tubulin (#T5326) (Sigma-Aldrich); Ku80 (#611360), ZO-1 (#610966), E-cadherin (#610181), mCherry (#632543) (Clontech) and ezrin (#610602) (BD Biosciences). Rabbit polyclonal antibodies against the following proteins were used: GFP (#ab290, Abcam); CAMSAP2 (#NBP1-21402 Novus, #17880-1-AP Proteintech); aPKC (#H00023499-A01, Santa Cruz); Actin (#20-33) and CEP135 (#SAB4503685) (Sigma-Aldrich); phosphorylated ERM (#3141, Cell Signaling); E-cadherin (Alpha Yap, University of Queensland, Australia). Alexa-Fluor-405- (#A-31553 and #A-31556), Alexa-Fluor-488- (#A27023 and #A-11034), Alexa-Fluor-594- (#A-11032, #R37117 and #A-11007) and Alexa-Fluor-647- (#A-21235) conjugated goat antibodies against mouse, rabbit and rat were purchased from Life Technologies. For Western blotting, IRDye-800CW-conjugated goat antibody against mouse and rabbit IgG (#P/N 925-32210 and #P/N 925-32211) and IRDye-680LT-conjugated goat antibody against mouse and rabbit IgG (#P/N 925-68020 and #P/N 925-68021) were purchased from Li-Cor Biosciences. See Table S1 for the antibody dilutions used.

Alexa-Fluor-594-conjugated phalloidin was purchased from Life Technologies (#A12381). Doxycycline was purchased from Sigma-Aldrich and used at a concentration of 1 µg/ml. Nocodazole was purchased from Sigma-Aldrich and used at a concentration of 10 µM. Cell were treated with nocodazole for 2 h prior to fixation.

Lentivirus packaging was performed by using MaxPEI-based co-transfection of HEK293T cells with psPAX2, pMD2.G and the lentiviral vector pSIN-TRE-rtTA-IRES-Hygro-iLID-dTomato-RAB11A. Supernatant of packaging cells was harvested up to 72 h of transfection, filtered through a 0.45-µm filter and incubated with a polyethylene glycol (PEG)-6000-based precipitation solution overnight at 4°C. After precipitation, virus was concentrated up to 100× by centrifugation and dissolution in 1× phosphate buffered saline (PBS). Target cells were incubated for 4 h in complete medium supplemented with 8 µg/ml polybrene before infection. To establish clonal Caco-2 stable lines carrying doxycycline-inducible dTomato–Rab11A, medium was replaced 24–48 h after infection and 100 µg/ml hygromycin (Invitrogen) was added.

For immunofluorescent staining of 2D cultures, cells were fixed with either 4% PFA for 20 min at room temperature (ezrin, CEP135, phosphorylated ERM, phalloidin) or −20°C MeOH for 10 min (α-tubulin, CAMSAP2, aPKC, ZO-1, ACF7, CAMSAP3, actin, E-cadherin, γ-tubulin) followed by permeabilization with 0.2% Triton X-100 for 2 min. Next, samples were blocked with 1% BSA diluted in 1× PBS supplemented with 0.05% Tween-20 for 45 min and sequentially incubated with primary antibodies for 1 h and fluorescently labeled secondary antibodies for 45 min. Finally, samples were washed, dried and mounted in DAPI-containing Vectashield (Vector laboratories).

For immunofluorescence staining of 3D cultures, cells were fixed with 4% PFA for 20 min at room temperature, rinsed with 1× PBS with glycine (10× stock: 38.00 g NaCl, 9.38 g Na₂HPO₄, 2.07 g NaH₂PO₄, 37.50 g glycine in 500 ml milliQ water, pH 7.4) followed by permeabilization with 0.5% Triton X-100 in 1× PBS with glycine for 20 min. Next, samples were blocked with 1% BSA diluted in 1× immunofluorescence wash buffer (10× stock: 38.00 g NaCl, 9.38 g Na₂HPO₄, 2.07 g NaH₂PO₄, 2.5 g NaN₂, 10 ml Triton X-100, 2.5 ml Tween-20 in 500 ml milliQ water, pH 7.4) for 45 min and sequentially incubated with primary antibodies for 2 h at 37°C and fluorescently labeled secondary antibodies for 1.5 h at room temperature. Finally, samples were washed and mounted in DAPI-containing Vectashield.

2D confluent monolayers and 3D cultures were imaged by using confocal fluorescence illumination on a Nikon Eclipse Ti microscope equipped with a perfect focus system (PFS, Nikon), a spinning-disc-based confocal scanner unit (CSU-X1-A1, Yokogawa), an Evolve 512 EMCCD camera (Photometrics) attached to a 2.0× intermediate lens (Edmund Optics), a super-high-pressure mercury lamp (C-SHG1, Nikon), a Roper Scientific custom-made set of Stradus 405-nm (100 mW, Vortran), Calypso 491-nm (100 mW, Cobolt) and Jive 561-nm (100 mW, Cobolt) lasers, a set of ET-BFP2, ET-EGFP, ET-mCherry and ET-EGFP-mCherry filters (Chroma) for wide-field fluorescence observation, a set of ET460/50m, ET525/50m or ET535/30m (green), ET630/75m (red) and ET-EGFP/mCherry filters (Chroma) for spinning-disc-based confocal imaging and a motorized stage MS-2000-XYZ with Piezo Top Plate (ASI). The microscope setup was controlled by MetaMorph 7.7.11.0 software. Images were acquired using Plan Fluor 10× NA 0.3 air, Plan Fluor 20× MI NA 0.75 oil, Plan Apochromat λ 60× NA 1.4 oil and Plan Apo VC 60× NA 1.4 oil objectives. This system was also used for FRAP experiments using the iLas2 system (Roper Scientific).

z-series images of 2D confluent layers were acquired using a 0.1-µm-step confocal-based scan. Side views were reconstructed by projecting maximum fluorescence intensities of 50×0.11-µm side view slices. The apical–basal polarity index was determined by using side views as described above in the following equation: ⁠, in which F_a is the fluorescent intensity in the upper one-third of the side view corrected for the area and F_b is the fluorescence intensity in the lower two-thirds of the side view corrected for the area. For presentation, images were adjusted for brightness using ImageJ 1.50b. FRAP measurements were performed by bleaching a 8.8-µm-diameter circle in the apical side of a cell monolayer followed by 15 min of imaging with a frame interval of 10 s.

Phase contrast images were acquired on a Nikon Ti instrument equipped with a perfect focus system (PFS, Nikon), a super-high-pressure mercury lamp (C-SHG1, Nikon), Lambda SC Smart Shutter controllers (Sutter), a Plan Apochromat DM 20× NA 0.75 (Ph2) or a Plan Fluor DLL 10× NA 0.3 (Ph1), an ET-mCherry filter (Chroma), a CoolSNAP HQ² CCD camera (Photometrics), a motorized stage MS-2000-XYZ with Piezo Top Plate (ASI) and a stage top incubator INUG2E-ZILCSD-DV (Tokai Hit) for 37°C under 5% CO₂. The microscope setup was controlled by the open source microscopy software Micro-manager.

To determine protein knockdown efficiency by immunofluorescence staining, cells were imaged on a Nikon Eclipse 80i upright microscope equipped with a CoolSNAP HQ² CCD camera (Photometrics), an Intensilight C-HGFI precentered fiber illuminator (Nikon), ET-DAPI, ET-EGFP and ET-mCherry filters (Chroma), controlled by Nikon NIS Br software and using a Plan Apo VC 100× NA 1.4 oil, Plan Apo VC 60× NA 1.4 oil or a Plan Fluor 20× MI NA 0.75 oil objective (Nikon). For knockdown efficiency tests, fluorescence intensity was measured per cell and corrected for the cell surface area using ImageJ 1.50b.

Microtubule organization in CAMSAP-knockout lines was imaged by confocal fluorescence illumination on a Leica TCS SP8 STED3X SMD FLIM Super Resolution instrument and confocal microscope equipped with adaptive focus control, a filter-free Spectral Leica SP detector and HyD detector, a 405-nm DMOD and a fully tunable supercontinuum white laser (470 to 670 nm). Images were acquired using a HC PL APO Cs2 100×1.40 oil objective. z-series were acquired using 0.16-µm- (Control) and 0.22-µm- (CAMSAP2 and CAMSAP3 knockout) step confocal-based scan. Images are presented as maximum projections and were adjusted for brightness using ImageJ 1.50b.

Total RNA was extracted from Caco-2 cells using Trizol reagent (Thermo Fisher Scientific) and used as a template for reverse transcription (RT-PCR) with random hexamer primers (Thermo Fisher Scientific). After DNase treatment (Thermo Fisher Scientific), cDNA was submitted to quantitative real-time (q)PCR using Sybrgreen technology (Applied Biosystems) on a ViiA7 apparatus (Applied Biosystems). Two ACF7 and two reference gene (HPRT1 and GAPDH) primer pairs were used to quantify ACF7 relative expression following siRNA transfection using the ΔΔCt method. Primer sequences were as followed: ACF7 #1, 5′- GGTCCTCTCAGGCATCAAAC-3′, 5′-AGTTTCACCTGTCGCTGCTT-3′; ACF7 #2, 5′-TGCACTCATTCACCGATACC-3′, 5′-CCCAGTCTTTCTGCCACTTC-3′; HPRT1, 5′-TGCAGACTTTGCTTTCCTTGGTCAGG-3′, 5′-CCAACACTTCGTGGGGTCCTTTTCA-3′; GAPDH, 5′-CTTCGCTCTCTGCTCCTCCTGTTCG-3′, 5′-ACCAGGCGCCCAATACGACCAAAT-3′.

All experiments were conducted at least twice. Statistical analysis was performed using GraphPad Prism 5.

We thank J. L. Bos (UMC Utrecht, The Netherlands), A. Yap (University of Queensland, Australia), B. Bouchet (Utrecht University, The Netherlands), R. Pagano (Mayo Clinic College of Medicine, Rochester, MN), B. Kuhlman (University of North Carolina, Chapel Hill, NC) for the gifts of materials, and J. Demmers (Erasmus MC, Rotterdam, The Netherlands) for performing mass spectrometry analysis.