The ciliary GTPase Arl3 maintains tissue architecture by directing planar spindle orientation during epidermal morphogenesis

Cell polarity is a fundamental feature of all epithelia, and results from the differential distribution of cellular components whose asymmetry is required for proper tissue function (Vladar et al., 2012; Devenport, 2014). Diverse types of cell polarity are required for the establishment and maintenance of tissue architecture during its development and homeostasis, and drive essential processes such as oriented cell division (OCD) (Lu and Johnston, 2013). During mammalian skin development, OCDs are bimodal (Kulukian and Fuchs, 2013). Well-established polarity pathways drive asymmetric cell division (ACD), positioning the mitotic spindle perpendicular to the tissue plane in a conserved process that is essential for epidermal differentiation (Lechler and Fuchs, 2005; Williams et al., 2011). Basal stem cells (SCs) must also undergo self-amplifying planar cell divisions during skin development and regeneration, but it is unknown whether defects in this process can have deleterious effects on the maintenance of epidermal architecture.

Planar cell polarity (PCP) can modulate mitotic spindle orientation in the context of tissue morphogenesis, but the underlying regulatory mechanisms are largely unknown (Smith et al., 2017; Segalen and Bellaïche, 2009). PCP globally aligns cells in the plane of the tissue (Mitchell et al., 2009), and conserved PCP pathway proteins Celsr1 and Vangl2 are required for the establishment and maintenance of PCP in mammalian epidermis (Devenport and Fuchs, 2008; Devenport et al., 2011). The planar polarized distribution of these atypical cadherins in basal cells of inter follicular epidermis (IFE) dictates the global alignment of hair follicles (HFs) along the A-P tissue plane (Devenport and Fuchs, 2008). In Drosophila and zebrafish, PCP signaling can regulate mitotic spindle orientation along the animal-vegetal axis (Ségalen et al., 2010). The PCP signaling molecules frizzled 6 (Fzd6) and Celsr1 have recently been implicated in the cell contact-dependent specification of planar cell divisions during mammalian skin development, although specific regulatory mechanisms have not been delineated (Oozeer et al., 2017). Whether PCP signaling in basal SCs non-autonomously influences progenitor populations and subsequent tissue morphogenesis outside of the HF, in the stratifying IFE, is currently unknown.

In the course of our work characterizing the function of Arl/ARF family ciliary GTPases during epidermal development and Notch signaling (Ezratty et al., 2011, 2016), we have uncovered an important function for the ciliary small GTPase Arl3 in regulating epidermal integrity, mitotic spindle orientation and PCP signaling during skin development. Arl/ARF family GTPases were initially described having roles in membrane trafficking and microtubule dynamics (Zhou et al., 2006; Jiang et al., 2007), and recent studies implicate these GTPases in the regulation of ciliary signaling and trafficking (Li et al., 2012). Arl3 has been identified as a cargo release factor in primary cilia, and has been associated with various ciliary signaling function(s) (Ismail et al., 2011; Wright et al., 2011, 2016; Kim et al., 2014; Lokaj et al., 2015; Hanke-Gogokhia et al., 2016). Arl3 KO mice display ciliopathy-associated defects in kidney development and retinal photoreceptor function (Schrick et al., 2006).

Using in utero gene targeting in developing mouse embryos, we show that depletion of ciliary GTPase Arl3 from basal SCs causes severe defects during skin development: abnormal expansion of progenitor cell populations, loss of epidermal integrity and skin barrier deficiency. Interestingly, Arl3 knockdown (KD) resulted in defects to planar mitotic spindle orientation. Self-amplifying planar cell divisions were diminished in basal SCs depleted of Arl3, but perpendicular divisions increased and led to an expansion of the suprabasal cell layer. These observations suggest that an Arl3-dependent mechanism maintains cell division polarity along the plane of the tissue, and that disruption of planar mitotic spindle orientation has detrimental consequences to epidermal architecture.

We hypothesized that loss of planar mitotic spindle orientation could be a consequence of defective PCP signaling. In Arl3 KD epidermis, the PCP signaling molecule Celsr1 fails to maintain its polarized distribution at cellular junctions across the A-P tissue plane. This resulted in defective hair follicle angling, a hallmark of disrupted PCP during skin development. Clonal analysis in mosaic embryos revealed that Arl3 is required for the maintenance of Celsr1 polarity and transduction of PCP to neighboring basal cells. In the absence of Celsr1, Fzd6 loses its asymmetrical distribution in the plane of the tissue and becomes abnormally segregated to the apical cortex of basal SCs. Celsr1 and transferrin internalization experiments suggest that Celsr1 is normally endocytosed during mitosis, but that Arl3-dependent endosomal trafficking of PCP components may be required for their polarized membrane segregation. We propose that Arl3 regulates polarized trafficking of PCP components to orient cell divisions in the plane of the epithelium. Cell-cell communication via small GTPase signaling therefore plays a crucial role specifying planar mitotic spindle orientation, a process that is crucial to the maintenance of tissue architecture during development.

Skin epidermis is an archetypal regenerative epithelium maintained by SCs, and an excellent model system for addressing questions regarding lineage determination, cell differentiation and tissue patterning (Blanpain et al., 2004). Arl3 expression and subcellular localization have never been reported in developing epidermis. Immunofluorescence (IF) and confocal microscopy were employed to characterize the subcellular localization of this small ciliary GTPase during epidermal stratification and HF morphogenesis. At E14.5, Arl3 was first detected in epidermal SCs of the basal layer (Fig. 1A). As stratification and HF morphogenesis proceed through E15.5 to E16.5, Arl3 expression was still restricted to basal cells comprising the IFE, but became undetectable in invaginating HF placodes (Fig. 1B,C, arrows). High-magnification imaging of Arl3 revealed a filamentous localization pattern reminiscent of microtubules (Fig. 1D). To determine whether Arl3 localization is dependent on an intact microtubule cytoskeleton, E14.5 epidermis was treated with nocodazole for 1 h to depolymerize the microtubule cytoskeleton. This treatment disrupted the subcellular localization of Arl3 in comparison with DMSO controls (Fig. 1E). To better visualize the subcellular localization of Arl3 in HF placodes, we used confocal microscopy on whole-mount samples of developing skin at E16.5, which showed that Arl3 puncta label primary cilia of invaginating HF placodes (Fig. 1F). Next, we quantitatively analyzed the subcellular distribution of Arl3 in primary cilia of HF placodes versus cilia of the IFE from E15.5 WM epidermal tissue. In both the HF and the IFE, ∼80% of primary cilia detected with an antibody against acetylated tubulin were also immunolabeled with Arl3 (Fig. 1G,H), suggesting that Arl3 GTPase activity could regulate ciliary signaling in both epidermal compartments. Interestingly, Arl3 labeling was significantly brighter in primary cilia of the HF placodes, when compared with cilia from the neighboring IFE (Fig. 1G-H). Quantification of the relative fluorescent intensity revealed a statistically significant twofold increase in Arl3 immunolabeling at HF cilia versus primary cilia of the IFE (Fig. 1I). We conclude that Arl3 expression is developmentally regulated, and that changes in Arl3 subcellular localization (the primary cilium and microtubules) correlates with differential cell fates.

To develop tool(s) to study Arl3 function in vivo, in utero gene targeting methods were employed to efficiently and selectively deliver shRNAs to E9.5 embryos at a stage when skin exists as a single layer of epidermal SCs (Williams et al., 2011; Beronja et al., 2010). The technique has been used previously to ablate cilia function by E12.5 (several days earlier than K14-Cre-mediated epidermal ablation; Ezratty et al., 2011, 2016), and is useful for evaluating protein function early during skin development prior to lineage-determining signaling and establishment of planar cell polarity (Beronja et al., 2010). Arl3 and scrambled control shRNAs were cloned into lentiviral expression vectors for H2B-RFP (to control for infection efficiency in vivo) and KD efficiencies analyzed in cultured primary mouse keratinocytes. IF and western blot analysis demonstrated that Arl3 protein levels were significantly reduced using two different and specific shRNAs in comparison with scrambled control (Fig. S1A-C). Next, high titer (>10⁹ MOI) lentiviruses (LV) were produced and injected into amniotic sacs of E9.5 mouse embryos (Fig. S1D) (Beronja and Fuchs, 2013), and transduction of 85% of the mouse epithelium was routinely observed (Fig. S1E). Confocal microscopy on WM tissue demonstrates transduction throughout the basal and suprabasal layers of the epidermis, as visualized by histone H2B-RFP expression (Fig. S1F), and Arl3 immunolocalization in basal cells was diminished upon transduction with shRNA targeted against Arl3 (Fig. S1G). These results establish that shRNA-mediated KD of Arl3 is specific in vitro and in vivo, and demonstrate that this technology can be used to study the function of Arl3 early during epidermal morphogenesis. KD of Arl3 via in utero delivery of RNAi resulted in gross phenotypic abnormalities that appeared to be distinct from those generated upon loss of ciliary signaling via shRNA-mediated disruption of intraflagellar transport (IFT). P0 pups displayed wrinkled, shiny skin and did not survive postnatally; suggesting a defect in epidermal barrier acquisition, similar to ciliary cKOs (n=4 P0 embryos) (Fig. S1H). However, P0 pups (n=4) also displayed an open eye phenotype and curly tail, defects associated with abnormal PCP signaling (Lu and Johnston, 2013; Lu et al., 2004; Saburi et al., 2008). Fifty percent of embryos isolated at P0 displayed limb defects similar to those seen upon conditional loss of Vangl2 (Gao et al., 2011) but no signs of polydactyly (data not shown). This is in contrast to IFT KD embryos, which display polydactyly, a hallmark of disrupted sonic hedgehog signaling due to the complete ablation of cilia (Ezratty et al., 2011). As Arl3 depletion led to early postnatal death from what appears to be compromised epithelial integrity, embryos were collected and analyzed either at P0 (immediately after birth) or at peri-natal timepoints from E15.5 to E18.5 in order to ascertain the role of Arl3 during early epidermal differentiation and HF morphogenesis.

Epidermal differentiation was ascertained at multiple developmental time points using antibodies against keratin 14 (K14), which labels epithelial SCs of the basal layer, and keratin 10 (K10), a marker of the suprabasal differentiated spinous cells. At E15.5, early during epidermal stratification and differentiation, we did not observe any defects in expression of K10 in suprabasal layers (Fig. S2). At E16.5, K14 was uniformly expressed in Arl3 KD tissue (Fig. 2A), and there was no statistically significant difference in the thickness of the basal layer of Arl3 KD tissue in comparison with non-transduced littermates or scrambled control (Fig. 2A,B). Although K10 was expressed in E16.5 Arl3 KD tissue, the K10⁺ spinous layer appeared disorganized, thickened and non-uniform in comparison with littermate or scrambled controls (Fig. 2C). Quantitative analysis of the thickness of K10⁺ spinous layer revealed a significant expansion upon Arl3 KD generated using two different and specific shRNAs, in comparison with control epidermis (Fig. 2D). In P0 postnatal pups, this phenotype was readily apparent as the K10⁺ spinous layer was extremely expanded, and ectopic K10 expression was visible throughout the upper terminally differentiated layers of the epidermis (Fig. 2E, arrow). The phenotype of Arl3 KD epidermis, which demonstrates an expanded progenitor cell population of K10⁺ suprabasal cells, is distinct from the epidermal phenotypes generated from loss of IFT/cilia, which show basal layer expansion, diminished expression of K10 and failure to undergo Notch-dependent epidermal differentiation (Ezratty et al., 2011; Croyle et al., 2011). In contrast to ciliary cKOs, the hyper-thickened suprabasal layer in Arl3 KD embryos still showed expression of terminal differentiation markers involucrin and filagrin (Fig. 2F-G), suggesting that the program of epidermal differentiation was operative.

Hematoxylin and Eosin staining demonstrated that Arl3 KD tissue displays expanded suprabasal layers and hyper-keratinization of the cornified layer (Fig. 2H, arrow). This phenotype has been observed in other studies in which the terminal differentiation proteins involucrin, periplakin and envoplakin were genetically ablated during skin development (Sevilla et al., 2007). The similarity of our phenotype to these other mouse models with compromised barrier function led us to examine epidermal integrity in Arl3 KD embryos. The barrier permeability assay showed increased uptake of Hematoxylin dye in E16.5 Arl3 KD embryos when compared with control embryos (Fig. 2I). This uptake of dye suggests a defect in barrier formation following Arl3 KD, which likely explains the early postnatal death we observed at P0. Interestingly, nuclear NICD1 was detected in Arl3 KD epithelium at E16.5 (Fig. S3A), and quantitative analysis of the downstream Notch target gene Hes1 via qPCR did not reveal any difference upon Arl3 KD in comparison with scrambled control (Fig. S3B). Western blot analysis of activated Notch1 (NICD1) and Notch3 (NICD3) cleavage products in differentiating cultured keratinocytes revealed a slight decrease upon Arl3 KD when compared with scrambled control (Fig. S3C,D). Taken together, these data suggest that Notch signaling is operative but the disrupted barrier formation we observe upon Arl3 KD may be due to loss of a different signaling mechanism. This is consistent with our observation that epidermal progenitors in suprabasal layers still express, albeit ectopically, downstream Notch target gene K10 (Fig. 2C-E). These data demonstrate that the ciliary GTPase Arl3 is required for the maintenance of proper tissue architecture during epidermal morphogenesis and subsequent integrity of the epidermal barrier.

Given that ciliary mutants display defects in epidermal differentiation and integrity of the epidermal barrier (Ezratty et al., 2011; Croyle et al., 2011), we analyzed ciliogenesis in the absence of Arl3 function. Cultured primary mouse keratinocytes induced to differentiate with 2 mM Ca²⁺ still generated cilia to the same extent as scrambled controls upon shRNA-mediated depletion of Arl3 in vitro (Fig. S4A,B). Ciliogenesis was observed in vivo in both uninjected littermate controls and Arl3 KD IFE immunolabeled for Arl13b to detect primary cilia (Fig. S4C,D). This is particularly striking in confocal projections of E16.5 WM epidermis from Arl3 KD tissue, where large areas H2B-RFP⁺ regions are clearly ciliated (Fig. S4C). From all these approaches, similar results were observed for uninjected littermate control and scrambled control embryos, where the scrambled control has been shown not to interfere with ciliogenesis, cell proliferation or cell fate specification in developing embryos (Ezratty et al., 2011; Beronja et al., 2010). Arl3 KD, H2B-RFP⁺ HF placodes also displayed apically oriented cilia (Fig. S4E). Quantitative analysis of cilia length in E16.5 embryos did not reveal any statistically significant differences between scrambled control and Arl3KD (Fig. S4F). In cultured keratinocytes depleted of Arl3, the overall organization of the microtubule cytoskeleton appeared normal when compared with scrambled control (Fig. S4G). Western blot analysis consistently demonstrated an 85-90% reduction in Arl3 protein levels with two different and specific shRNAs (Fig. S1C), and our previous studies have shown that similar levels of shRNA-mediated IFT KD in vitro or in vivo are sufficient to completely ablate cilia under these assay conditions (Ezratty et al., 2011; Croyle et al., 2011). Taken together, these observations suggest that Arl3 is likely not required for ciliogenesis per se either in vivo during epidermal morphogenesis or in vitro in differentiating cultured keratinocytes.

Defects in epidermal architecture and barrier function can arise when the balance between proliferation and differentiation goes awry in rapidly proliferating basal SCs (Kulukian and Fuchs, 2013). Quantification of Ki67⁺ cells showed a twofold increase in cycling cells in suprabasal layers of E16.5 Arl3KD epidermis, and a statistically significant increase in the total number of K10⁺ suprabasal cells upon Arl3 KD (Fig. 3A-C) (n=4 embryos, 100-200 cells/embryo). Fig. 3A shows an example of Ki67⁺ suprabasal cells (arrow) in Arl3KD epidermis, which are not normally observed in suprabasal layers of control skin at this developmental stage (quantified in Fig. 3B). The basal layer showed a modest increase in total percentage of Ki67⁺ labeling (Fig. 3D), but no statistically significant difference in the total number of K14⁺ basal cells (Fig. 3E). These data suggest that the abnormal cell proliferation in differentiating suprabasal progenitors is a consequence of Arl3 loss in basal SCs, likely contributing to the significant expansion of K10⁺ layers observed. This is in marked contrast to ciliary cKOs/KD, which display basal layer hyperplasia, but no expansion of the K10⁺ progenitor compartment or increase in suprabasal mitoses (Ezratty et al., 2011; Croyle et al., 2011).

Basal SCs must balance planar divisions that amplify their numbers with asymmetric divisions that self-renew a basal progenitor while driving differentiation of a suprabasal cell (Kulukian and Fuchs, 2013). We hypothesized that defects in oriented cell divisions (OCD) in basal SCs could promote the expansion of suprabasal layers observed upon Arl3 KD. To determine whether loss of Arl3 resulted in defect(s) in oriented cell division (OCD), we quantified the number of planar versus perpendicular or ‘asymmetric’ cell divisions (ACD) in Arl3 KD versus wild-type E16.5 epidermis. OCDs can readily be identified via nuclear morphology in WM epidermal tissue, where both planar and ACD are numerous in rapidly proliferating E16.5 epidermis (Fig. 3F, arrows). Loss of Arl3 generated via two different and specific shRNAs consistently resulted in significantly fewer planar divisions, which were balanced by an increased number of perpendicular divisions (Fig. 3G,H). Quantitative analyses show that while 40% of control basal SCs divide asymmetrically at E16.5, upon Arl3 loss up to 60% of cells undergo ACD (n=4 embryos, 100-200 cells/embryo). Diminished planar cell divisions are therefore balanced by a 20% increase in ACD, likely contributing to the expansion of suprabasal layers observed in Arl3 KD tissue. We did not see a significant difference in mitotic spindle positioning in E15.5 Arl3 KD epidermis in comparison with control E15.5 tissue (Fig. S2C). These data suggest that Arl3 plays a role in orienting the mitotic spindle during epidermal differentiation.

Next, we tested whether Arl3 function was required to position the mitotic spindle within the plane of the developing epidermis. Pericentrin staining was used to define the axis of division in mitotic basal cells, and angles were measured with respect to the anterior-posterior (AP) and dorsal-ventral (DV) axes. Planar angles of cell divisions were designated into the following quadrants: anterior-ventral (AV: 90°-180°), anterior-dorsal (AD: 0°-90°), dorsal-ventral (DV: 45°-135°) and anterior-posterior (AP: 0°-45° and 135°-180°) (Fig. 3I) (Oozeer et al., 2017). The orientations of planar divisions in the AP versus DV quadrant or the AV versus AD quadrants in Arl3 KD epidermis were not significantly altered when compared with scrambled control (P≥0.5 by Watson's U² test) (Fig. 3J). There seems to be a slight bias towards DV versus AP planar divisions under control conditions, but as this is not changed upon Arl3 KD, we conclude that Arl3 does not control mitotic spindle orientation within the plane of the developing epidermis.

Defects in apico-basal polarity (ABP) can affect mitotic spindle positioning (Nakajima and Gibson, 2015). In basal cells, the apical positioning of centrosomes is determined by ABP, and measurements of pericentrin localization with respect to the epithelial plane are used as a read out of ABP during skin development (Luxenburg et al., 2015). Our quantifications showed that pericentrin-labeled centrosomes maintained their apical positioning, even in Arl3 KD epidermis that displays a disorganized suprabasal epithelial architecture (Fig. S5A-C). These data indicate that Arl3 function is not required for the apical positioning of centrosomes, and strongly implies that the defects in mitotic spindle orientation observed in Arl3 KD epidermis cannot be attributed to disrupted ABP. Taken together, our observations suggest there is an Arl3-dependent mechanism that maintains planar mitotic spindle orientation across developing mammalian embryonic skin. Disruption of planar polarized divisions may contribute to defects in epithelial tissue architecture, proliferation and differentiation observed in Arl3 KD epidermis.

We hypothesized that loss of planar OCD could be a consequence of defective planar cell polarity (PCP) signaling in the absence of Arl3. PCP signals in Drosophila and zebrafish can direct planar mitotic spindle orientation (Ségalen et al., 2010), and the core PCP proteins Celsr1 and Fzd6 have been implicated in cell contact-mediated orientation of planar cell division in developing mammalian skin (Oozeer et al., 2017). It is well established that genetic ablation of core PCP proteins leads to defects in the coordinated alignment of HFs along the AP axis during skin development (Devenport and Fuchs, 2008). In WT E16.5 epidermal WMs, HFs align along the AP tissue axis (Fig. 4B, arrows). In contrast, E16.5 WM epidermis from Arl3 KD tissue displayed HFs that were not as precisely aligned along the A-P axis, a phenotype indicative of defective PCP signaling (Fig. 4A,B, arrows). Quantification of HF angles showed that Arl3 KD HFs were polarized with angles distributed from +30° to −30° (Fig. 4C,D). Loss of HF polarity and alignment were observed with two distinct shRNA KD LVs, in comparison with either WT or scramble (data not shown) controls (n=4 embryos per condition). These data suggest that disruption of Arl3 function during skin development leads to loss of planar cell polarity, as evidenced by aberrant hair follicle polarity and alignment observed in Arl3 KD tissue. Again, the defects in HF morphogenesis generated upon Arl3 KD were distinct from those caused by loss of ciliary signaling: quantitative analysis of the distribution of HFs in either the peg or germ state showed that they are not statistically different in Arl3 KD versus control epidermis (Fig. 4D). In contrast, in Kif3a cKO or IFT KD epidermis, which effectively ablate cilia, hair follicles are stunted at the hair peg stage (Fig. S6A), and do not undergo further morphogenesis due to defects in Shh signaling (Ezratty et al., 2011). Axin2 and Lef1 levels were not significantly altered upon Arl3 KD, as evidenced by IF and qPCR analysis (Fig. S6B,C). Patched 1 and smoothened expression was also unchanged, although Gli1 showed a two-fold increase following Arl3 KD (Fig. S6C,D). These data suggest that Shh and Wnt signaling remain operative once Arl3 function is ablated. Arl3 KD HFs therefore undergo early steps of Wnt- and Shh-dependent HF morphogenesis, but display defects in coordinated angling and polarity along the A-P axis in the plane of the epidermis – a hallmark of defective PCP signaling.

Observation of misaligned HFs at E16.5 prompted us to look for specific defects in planar cell polarity (PCP) signaling. Components of the Drosophila PCP signaling pathway are conserved in mammalian epidermis, where Celsr1 and Vangl2 are polarized along the A-P axis in basal SCs of developing epidermis and are required for the coordinated alignment of HFs in the plane of the epidermis (Devenport and Fuchs, 2008; Simons and Mlodzik, 2008; Vladar et al., 2009). PCP is first established by E14.5 in developing epidermis (Devenport and Fuchs, 2008), when the membrane-localized atypical cadherin Celsr1 becomes A-P polarized at cell-cell contacts of basal epidermal cells. To determine the mechanism by which Arl3 might regulate PCP signaling, we performed clonal analysis in mosaic KD embryos from early stages of skin development to ascertain whether Arl3 is required for: (1) the establishment or (2) the maintenance of Celsr1 polarity during epidermal morphogenesis. In these experiments, H2B-RFP⁺ Arl3 KD clones were generated by varying the MOI of injected LV, and we analyzed WM epidermal tissue isolated from Arl3-1 shRNA KDs, which significantly depleted Arl3 protein levels in vivo (Fig. S2).

By E15.5 in control WM epidermis, Celsr1 is uniformly polarized at cell-cell contacts along the A-P plane of the epithelium (Fig. 5A). In H2B-RFP⁺ Arl3 KD epidermis Celsr1 immunolabeling is polarized but appears slightly reduced at cellular junctions (Fig. 5A, outlined clone). By E16.5, Celsr1 maintains its polarized subcellular localization along the AP axis in both H2B-RFP⁺ scrambled control and non-transduced control (littermate) E16.5 epidermis (Fig. 5B). In contrast, in Arl3 KD epidermis, where the majority of cells were uniformly H2B-RFP⁺, Celsr1 immunolabeling appears either extremely punctate or essentially undetectable at cellular junctions (Figs 5B,C). Digitalization of Celsr1 orientation across the epidermal tissue plane illustrates that Celsr1 is uniformly polarized at cellular junctions along the A-P axis in scrambled control or non-transduced littermate controls, but appears randomly oriented in mosaic Arl3 KD tissue (Fig. 5D). Loss of Celsr1 immunolabeling and A-P polarity between E15.5 and E16.5 suggests that Arl3 may be required for the maintenance of PCP/Celsr1 polarity during epidermal morphogenesis.

Next, clonal analysis was leveraged to determine whether loss of Arl3 results in non-autonomous disruption of PCP signaling in neighboring clones of wild-type basal cells. To quantify Celsr1 loss upon Arl3 KD at E15.5-16.5, colocalization of Celsr1 and E-cadherin was determined in H2B-RFP⁺ epidermal clones, and compared with adjacent wild-type tissue (Fig. 5C,E-G). Although 60% of E-cadherin IF colocalized with Celsr1 (polarized along A-P axis) in wild-type cells, RFP⁺ Arl3 KD cells showed <5% of colocalization of the Celsr1/E-cadherin signal by E16.5 (Fig. 5C,G). Notably, although Celsr1 was detectable at cellular junctions in non-transduced tissue, it did not appear to be A-P polarized (Fig. 5E, outlined clone). This appearance of unpolarized junctional Celsr1 in non-transduced epidermal cells adjacent to H2B-RFP⁺ Arl3 KD clones suggests that loss of Arl3 disrupts the transduction of a PCP signal to adjacent tissue in a non-autonomous manner. This is further illustrated in WM uninjected epidermis, where quantitative analysis of the polarity of the Celsr1 signal across the plane of the epidermis gives a sharp peak, corresponding to its A-P polarization throughout the WM epidermal back-skin sample (Fig. 5H,I). In contrast, in H2B-RFP tissue with clonal patches of Arl3 KD cells, the distribution of Celsr1 orientation is randomized (Fig. 5H,J). As Celsr1 is barely detectable in RFP⁺ cells at this embryonic timepoint (E16.5), the randomization of signal intensity illustrates lack of Celsr1 polarity in adjacent, non-transduced epidermis. This is readily apparent when comparing diminished Celsr1 signal in large H2B-RFP⁺ regions with unpolarized Celsr1 seen in patches of non-transduced epidermis (Fig. 5C,G-I). We conclude that depletion of the small GTPase Arl3 non-autonomously impacts transduction of PCP to adjacent cells during epidermal morphogenesis. These observations position the ciliary GTPase Arl3 as a new player in PCP signaling somewhere ‘upstream’ of Celsr1, where it may function as part of the molecular mechanism that preserves tissue polarity during epidermal morphogenesis.

Next, we examined whether persistent loss of polarized Celsr1 at cell junctions of basal SCs depleted of Arl3 was due to defects in its mitotic-dependent internalization. Celsr1 undergoes clathrin-mediated endocytosis (CME) during mitosis via a Plk1-dependent mechanism that is required to preserve PCP post-mitotically (Shrestha et al., 2015). To directly determine whether Arl3 may regulate mitotic internalization of Celsr1, we quantified the amount of internalized Celsr1 puncta from mitotic basal cells (n=50) in E16.5 Arl3 KD epidermis, when junctional Celsr1 is reduced and no longer A-P polarized. Analysis of Celsr1 fluorescence intensity in control versus Arl3 KD mitotic basal cells revealed no significant difference in the amount of internalized Celsr1 puncta (Fig. 6A,B). To ascertain whether CME was operative upon Arl3 KD, internalization of fluorescent transferrin was measured in cultured keratinocytes. Quantitative analysis demonstrated that overall levels of internalized transferrin were not reduced upon Arl3 KD (Fig. 6D). These observations suggest that CME and Celsr1 mitotic internalization occur normally when Arl3 function is perturbed.

Alternatively, defective endosomal trafficking of PCP components post-mitotically could disrupt their polarized membrane localization. It has been proposed that PCP components could undergo a post-mitotic, Rab11-dependent recycling to their polarized membrane compartments. To determine whether Arl3 loss led to defects within the endosomal trafficking pathway, transferrin recycling was monitored in cultured keratinocytes transduced with either scrambled control or Arl3 shRNA. Interestingly, Arl3 depletion from cultured keratinocytes quantitatively increased overall transferrin uptake (Fig. 6D). IF analysis revealed the presence of aberrant transferrin⁺ structures in Arl3 KD cells (Fig. 6C, arrows). These structures were rarely observed in the scrambled control, and suggested that transferrin might be abnormally accruing within an endosomal compartment. This observation is consistent with the recently proposed role for Arl3 in regulating release of dynactin-bound cargos from dynein during vesicle trafficking (Jin et al., 2014). Rab11, a small GTPase that functions at the endocytic recycling compartment (ERC) and recycling endosomes, normally colocalizes with 60% of internalized transferrin (10 min pulse, Fig. 6D,E). Upon Arl3 depletion, >80% of pulsed transferrin colocalized with Rab11⁺ endosomal structures (Fig. 6D,E). These data demonstrate that loss of Arl3 function can lead to defects within the ERC/recycling endosomes that result in aberrant trafficking of cargo (transferrin) through the endosomal system. Defects in Celsr1 membrane localization appear to be specific to this cadherin, as localization of E-cadherin to cell-cell contacts is not affected upon Arl3KD at E16.5 (Fig. S7A,B). Importantly, Celsr1 membrane localization is not defective in ciliary mutants, suggesting that Arl3-dependent localization of Celsr1 is likely independent of its function at primary cilia (Fig. S7C,D).

Our results are consistent with the model that Arl3 regulates an endosomal trafficking pathway necessary for the maintenance of polarized membrane domains. Celsr1 is normally polarized at both anterior and posterior membrane domains at cell-cell contacts of epidermal basal SCs, where it may recruit and maintain Vangl2 and Fzd6 to their exclusively anterior and posterior positions in the tissue plane (Devenport, 2014). At E15.5, Vangl2 is still localized at cell junctions in Arl3 KD epidermis, but by E16.5 this subcellular localization is diminished upon Arl3 KD (Fig. 7A,B). Immunolabeling of Vangl2 and Dsh2 at cell-cell contacts was significantly reduced upon Arl3-dependent loss of Celsr1 membrane localization in P0 epidermis (Fig. 7C-E). Western blot of protein levels showed that dishevelled 2 protein levels were unchanged upon Arl3 KD, suggesting that loss of immunolabeling was likely due to defective membrane recruitment/retention, rather than diminished protein expression or changes in protein stability (Fig. 7F). Interestingly, we did not observe the same loss of junctional A-P polarization of Fzd6 in Arl3 KD embryos. Frizzled 6, a non-canonical Wnt receptor required for planar polarized cell divisions of basal epithelial cells, is normally polarized to the posterior membrane of basal SCs (Fig. 7G,H) (Devenport, 2014). In Arl3 KD tissue, this asymmetrical partitioning is lost, and membranous Fzd6 localization is observed throughout the apical domain of basal cells (Fig. 7H, arrows). Quantification of A-P versus apical-basal (A-B) Fzd6 signal showed a twofold increase in A-B membrane localization, an abnormal subcellular localization for a protein that is typically planar polarized in basal SCs (Fig. 7H,I). We conclude that, in the absence of Arl3-dependent, Celsr1 membrane segregation, the asymmetrical partitioning of Fzd6 to the posterior domain is lost and becomes abnormally distributed toward the A-B axis.

We propose that Arl3 regulates polarized endosomal trafficking necessary for the maintenance of PCP protein membrane domains during epidermal morphogenesis (Fig. 7J). In the absence of post-mitotic Celsr1 trafficking, Celsr1 planar polarity is not maintained after multiple rounds of basal SC division. Fzd6 becomes unpolarized and achieves an aberrant apical localization, which may further drive ACDs at the expense of planar divisions. Thus disruption of planar cell polarity via Arl3 loss contributes to defective spindle orientation, hyper-proliferation of suprabasal progenitors and subsequent deleterious consequences for the maintenance of tissue architecture during skin development (Fig. 7J).

Observations from this study suggest that the small GTPase Arl3 regulates two important aspects of PCP during development: the maintenance of PCP throughout the tissue and the decision of whether an epidermal SC undergoes a planar self-amplifying division versus a perpendicular division that alters cellular fate. When these functions go awry in basal SCs during epidermal morphogenesis, the integrity of the developing epithelium becomes compromised. Our study establishes a new framework with which to study GTPase regulation of PCP, and offers novel insights into PCP-mediated regulation of the orientation of the cell division axis and its consequences for cellular fate in a regenerative mammalian epithelium.

shRNA-mediated depletion of Arl3 via in utero gene targeting in developing mouse embryos leads to defects in the morphogenesis and alignment of HFs, a hallmark of disrupted PCP signaling during epidermal development. Abnormal HF alignment and polarity result from defects in the subcellular localization of Celsr1, where the normal asymmetric distribution of Celsr1 established at E13.5 is lost between E15.5 and E16.5 in the absence of Arl3 function. Clonal analysis demonstrated that loss of Arl3 results in non-autonomous defects in Celsr1 polarity in neighboring wild-type clones, suggesting concomitant loss of PCP transduction. These data establish Arl3 as part of the molecular machinery that maintains PCP during tissue morphogenesis.

Arl3 GTPase activity could either directly control the subcellular localization of Celsr1 via a role in trafficking or indirectly regulate Celsr1 polarity via transduction of an upstream PCP cue, the absence of which affects Celsr1 planar polarity once Arl3 function is lost. Previous studies have demonstrated that mitosis-specific endocytosis, and subsequent re-delivery of PCP protein to the plasma membrane, is required for the preservation of tissue polarity in epidermal basal SCs. While internalization of PCP components occurs via clathrin and Plk3-dependent pathway, the molecular mechanisms regulating their post-mitotic re-distribution have not been delineated (Devenport et al., 2011; Shrestha et al., 2015). Cells depleted of Arl3 still undergo CME of Celsr1 during mitosis, but display defects in transferrin recycling through Rab11⁺ endosomal membranes. These data suggest that Arl3-dependent endosomal trafficking is required to direct PCP proteins to their compartmentalized plasma membrane domains.

Our study suggests that PCP signaling specifies the axis of cell division in basal SCs via an Arl3 GTPase-dependent mechanism. Extrinsic polarizing cues transduced via the PCP signaling pathway have previously been shown to orient the mitotic spindle relative to cell-cell contacts or the symmetry axis of developing embryos (Bellaïche, 2016). However, the molecular players that control mitotic spindle orientation downstream of Wnt-Frizzled PCP signaling have not been well defined. Loss of Arl3 in basal cells results in decreased planar cell divisions in developing epidermis, implying that an Arl3-dependent mechanism maintains planar mitotic spindle orientation. In the absence of Arl3-dependent Celsr1 polarity, the asymmetrical partitioning of Fzd6 to the posterior membrane domain is lost and becomes randomized toward the A-B axis. Taken together, these observations suggest that Arl3 is required for the compartmentalization of polarized membrane domains at cell-cell contacts, and that PCP signaling maintains planar spindle orientation during skin development. Alternatively, Arl3 could have two separate functions: one in regulating Celsr1/PCP, and another directly regulating spindle orientation independently from PCP. Further studies will be necessary to delineate between these possibilities. Loss of Arl3-dependent planar cell divisions has potential consequences for basal stem cell self-renewal. Interestingly, cKO of Fzd6 in hematopoietic stem cells leads to loss of self-renewal and increased progenitor cell populations: analogous to what we observe in the expansion of K10⁺ suprabasal skin layers (Abidin et al., 2015). When Arl3 function is lost in basal SCs, the balance between self-renewal and the generation of differential cell fate is therefore perturbed, with deleterious effects on overall epidermal architecture.

Analysis of Arl3 subcellular localization at microtubules near cell-cell contacts and primary cilia suggests that its GTPase activity could regulate cilia-dependent and cilia-independent functions in epidermal basal SCs. Arl3 was originally identified as a MT-binding protein and, consistent with this function, can drive polarized trafficking from the Golgi apparatus to the primary cilium (Kim et al., 2014) and regulate release of dynactin bound cargos from dynein during vesicle trafficking in the cytoplasm (Jin et al., 2014). Arl3 has more recently been studied in complex with UNC119 and RP2, where it functions as a cargo-release factor required for the ciliary targeting of NPHP3 (Wright et al., 2011; Ismail et al., 2012). These and other studies suggest that Arl3 is required for a membrane-targeting GTPase cycle that delivers myristoylated proteins to the ciliary membrane.

How the proposed ciliary functions of Arl3 might relate to our observations that Arl3 regulates aspects of PCP signaling remains unclear. It is intriguing to speculate that ciliary-localized Arl3 may transduce upstream PCP cues, and that loss of this polarity signaling impacts Celsr1 localization and spindle positioning. Further work will be required to determine whether signaling through ciliary-localized Arl3 can transduce PCP cues during epidermal morphogenesis, but the lack of PCP phenotypes in ciliary epidermal mutants argue against this mechanism. We propose that Arl3 could play an analogous, non-ciliary role, in maintaining polarized membrane domains via regulated cargo release (of recycled PCP signaling components) to cell-cell junctions of basal SCs. These new findings underscore the diverse ways Arl3 might regulate the compartmentalization of polarized components to membrane domains, and suggest that regulation of polarized transport in the cilia versus the cytoplasm occurs via common pathways.

Last, our observations that Arl3 regulates planar mitotic spindle orientation through PCP are intriguing in light of the cystic kidney phenotype reported in Arl3 KO mice (Schrick et al., 2006). Defective planar cell polarity signaling and misregulated mitotic spindle orientation have both been proposed to impact the formation of cystic kidneys in polycystic kidney disease (Skalicka et al., 2017; Happé et al., 2011). The study of non-ciliary roles for proteins such as Arl3 will contribute to our understanding of how defects in cell polarity during development can impact the etiology of complex human genetic diseases, such as ciliopathies.

Lentiviral ShRNA was constructed using pLKO vector either with GFP or H2B-mRFP as a reporter. The oligos used to generate the shRNA hairpin cloned into pLKO3.1H2B-RFP vector (Beronja et al., 2010) are as follows: Arl3-2 forward, CCGGATACTGATATTCTCATCTATGCTCGAGCATAGATGAGAATATCAGTATTTTTTG (target sequence ATACTGATATTCTCATCTATG); Arl3-2 reverse, AATTCAAAAAATACTGATATTCTCATCTATGCTCGAGCATAGATGAGAATATCAGTAT; Arl3-1 forward, CCGGGTTGCTGCTTTCTGACCAAATCTCGAGATTTGGTCAGAAAGCAGCAACTTTTTG (target sequence GTTGCTGCTTTCTGACCAAAT); Arl3-1 reverse, AATTCAAAAAGTTGCTGCTTTCTGACCAAATCTCGAGATTTGGTCAGAAAGCAGCAAC.

In utero LV injection and high-titer LV production were based on methods previously described (Beronja et al., 2010). Needles for injection were made by pulling borosilicate glass (Sutter Instrument, inner diameter of 50 mm) using a Sutter Instrument Needle Puller (Model P-97). Needles were then beveled for 10 min at a 45° angle using BV-10 beveller (Sutter Instrument). A Vivo 2100 ultrasound system was used for the visualization of the injection procedure. Each of the E9.5 embryos were injected with 1 μl of high titer virus into the amniotic sac using the Drummond Nanoject II system.

CD-1 mice were used for all injection experiments (Charles River), and were maintained and used according to Columbia University IACUC approved protocols. Embryos were fixed in 4% paraformaldehyde in 0.1 M phosphate buffer (pH 7.4). Embryos were placed in a 30% sucrose solution in 1×PBS at 4°C overnight, and then frozen in OCT at −80°C. Embryos were cut sagittally into 10 μm sections using a cryostat. For WM, embryos were fixed and skins were dissected followed by immunostaining, as previously described (Ezratty et al., 2011). E15.5 tissue was incubated with DMSO or 10 μM nocodazole (Sigma) for 1 h at 37°C to depolymerize the microtubule cytoskeleton prior to fixation and processing for IF.

The isolation, preparation and transduction of keratinocytes were based on methods previously described (Ezratty et al., 2011). Keratinocytes were seeded at 70% confluence. Cells were transduced with LV in media containing 100 μl/ml polybrene and centrifuged for 30 min at 1100 g at 37°C. For immunostaining, transduced cells were passaged to coverslips coated with fibronectin (100 μg/ml) after 48 h. To induce differentiation, cover-slips were grown on media containing either 300 μM or 2 mM Ca²⁺.

Tissues and frozen sections were first blocked with a solution containing 10% normal donkey serum, 0.3% Triton-X, 2% fish gelatin and 10% BSA in PBS for 1 h. Sections and tissues were then incubated with primary antibodies overnight at 4°C. Primary antibodies used were rabbit anti-Arl3 (1:500, Proteintech, 10961-1-AP), rabbit anti-Arl13B (1:500, Proteintech, 17711-1-AP), mouse anti-β-catenin (1:500, BD Biosciences, 610153), rabbit anti-Lef1 (1:500, Cell Signaling, C12A5), rat anti-E-cadherin (1:500, Abcam, DECMA-1), rabbit anti-Ki67 (1:500, Abcam, 15580), rabbit anti-phospho-histone H3 (1:500, Abcam, 47297), rabbit anti-K10 (1:500, Abcam, ab76318), rabbit anti-K14 (1:500, Abcam, ab18159S), rabbit anti-pericentrin (1:500, Biolegend, 923701), rabbit anti-fillagrin (1:500, Abcam, ab53112), rabbit anti-involucrin (1:500, Abcam, ab53112), mouse anti-acetylated tubulin (1:500, Sigma Aldrich, T7451), rabbit anti-Dvl2 (1:500, Cell Signaling, 3216), guinea pig anti-Celsr1 (1:500, a gift from the Fuchs Lab, The Rockefeller University, NY, USA), rabbit anti-Vangl2 (1:500, EMD Millipore, ABN373) and rabbit anti-Frizzled6 (1:200, Abcam, ab150545). Tissues and sections were washed with PBS for 2 h after primary antibody labeling. Primary antibody labeling was detected using donkey anti-mouse Alexa 488 (1:400, Invitrogen, A32766), donkey anti-rabbit Alexa 488 (1:400, Invitrogen, A32790), donkey anti-mouse Alexa 555 (1:400, Invitrogen, A32773), donkey anti-rabbit Alexa 555 (1:400, Invitrogen, A32794). Tissues were mounted on slides using Prolong Gold anti-fade media containing DAPI (Molecular Probes).

Primary keratinocytes on coverslips were fixed with 4% PFA for 10 min followed by permeablization with 0.3% Triton-X for 5 min. The cells were then blocked with 10% NDS and stained with primary antibodies for 1 h, washed and labeled with goat or donkey secondary antibodies. The coverslips were then mounted on slides with Prolong Gold anti-fade media containing DAPI.

Primary keratinocytes were incubated with media containing 10 μg/ml Alexa Fluor-488-conjugated transferrin (Molecular Probes) for 2 min, 10 min and 30 min at 37°C. Membrane-bound transferrin was removed by first incubating with 0.5 M NaCl and 0.2 M NaOAc for 2 min, followed by washing with the same solution. Coverslips were then processed for immunofluorescence as described above.

Total RNA was isolated from primary keratinocytes using a RNA isolation kit (Invitrogen). 1 μg of isolated RNA was used for cDNA synthesis (Verso cDNA synthesis kit, Fisher). qPCR was performed in triplicate for each gene using SYBYR green master mix (Fisher Scientific) in Mx3000P qPCR real-time machine (Agilent Technologies). Primers used were: Axin2F, ACTGACCGACGATTCCATGT; Axin2R, CTGCGATGCATCTCTCTCTG; Hes1F, AGAAGAGGCGAAGGGCAAGAA; Hes1R, CATGGCGTTGATCTGGGTCAT; Gli1F, CAAAATAGGGGTGGGAGAGCC; Gli1R, GGAGACAAAGACCCACGGT; Ptch1F, CCACAGAAGCGCTCCTACA; Ptch1R, CTGTAATTTCGCCCCTTCC; 18SF, CGGCTACCACATCCAAGGAA; 18SR, GCTGGAATTACCGCGGCT.

Western blot was based on the methods described by Ezratty et al. (2011). Cells were lysed and scraped off using RIPA buffer mixed with protease inhibitor. The homogenized mixture was centrifuged at 2000 g for 10 min at 4°C. Primary antibodies used were rabbit anti-Arl3 (1:1000, Proteintech, 10961-1-AP), rabbit anti-Dvl2 (1:1000, Cell Signaling, 3216), mouse anti-GAPDH (1:1000, Abcam, ab8245), rabbit anti-activated Notch1 (1:500, Abcam, ab52301) and rabbit anti-Notch3 (1:500, Abcam, ab23426). IRDye 800 donkey anti-mouse (1:10,000, LI-COR, 926-32212) and IRDye 680 donkey anti-rabbit (1:10,000, LI-COR, 926-68073) were used to detect the primary antibodies. Images were collected and processed using a LI-COR Odyssey system.

Confocal images were acquired with either an Olympus IX83 DSU unit with Hamamatsu Orca-r2 or Nikon A1R laser-scanning microscope (Carl Zeiss MicroImaging) through a 63× oil objective (N.A. 1.4). For WM imaging, z-stacks of 20-40 planes (0.25 μm) were captured and either representative single z-planes or maximum projections (three images) are presented. Images were recorded at either 512×512 or 1024×1024 square pixels. For wide-field epi-fluorescence, images were acquired using an Olympus IX84 DSU 20×/0.8 air or 63×/1.4 oil Plan-Apochromat objectives and equipped with the following Chroma filter sets: 49008 ET TR C94094 (mRFP1), 49004 ET dsR C94093 (Cy3, DyLight549), 41008 Cy5 (Cy5) or 41001 FITC (AlexaFluor 488/GFP). Image acquisition was carried out using either NIS-Elements (Nikon) or Metamorph (Olympus). Maximally projected images (maximum of three images) were created using ImageJ software.

Arl3 localization at cilia was analyzed by counting the number of Arl13b or acetylated tubulin⁺ cilia that displayed colocalization of Arl3, using ImageJ software. For analysis of the intensity of Arl3 signal at cilia of HFs versus IFE, the average gray value of Arl3 signal in ciliated HF regions versus ciliated IFE regions was calculated using Metamorph. Data shown in histograms are normalized to the averaged gray value from ciliated HF regions. The thicknesses of K10 and K14 layers were calculated using Skintools plug-in for ImageJ. The plug-in generates multiples random vertical lines perpendicular to the epidermal border, and from these lines mean thickness was calculated for K14⁺ or K10⁺ regions.

Cell proliferation was quantified by determining the number of Ki67⁺ K14⁺ basal cells, and the number of Ki67⁺ K10⁺ suprabasal cells using Image J software to count objects. For analysis of planar versus perpendicular cell division and Celsr1 intensity in mitotic cells, mitotic cells versus interphase cells were identified based on their nuclear morphology by Hoechst staining (DAPI). The number of planar versus ACDs were manually quantified from 20× or 60× WM confocal stacks obtained from multiple E15.5 or E16.5 backskins from three or four embryos per condition. For measurements of planar angles of cell division, pericentrin staining was used to define the axis of division, and angles were measured with respect to the anterior-posterior (AP) or dorsal-ventral (DV) axis. For measurements of apical pericentrin localization in basal cells from sagittal sections of P0 tissue (Fig. S5C), a perpendicular line was drawn from the apical pericentrin-labeled centrosome toward the basement membrane, and Image J was used to calculate the angle of this line relative to the A-P epidermal plane, visualized with nidogen. These measurements were taken from P0 sagittal sections of skin from control or Arl3 KD pups.

For HF angles, single confocal planes were taken with 5× lens and the angles were measured by drawing a line parallel to the HF. The angles were plotted as rose plots in Georose software. OrientationJ plug-in was used to generate vectors for Celsr1, plotted in Fig. 6D,H,I. In Fig. 6I,J, the data showing the distribution of Celsr1 orientation was generated from single confocal plane images of WM E16.5 tissue stained with Celsr1 and shown in Fig. 6H. The colocalization study between Ecad and Ceslr1 was performed by first outlining RFP-positive cells and RFP-negative cells, and the same region of interest was then used for analysis in Metamorph using the colocalization function in the software, where the percentage of colocalizing pixels/area was determined. A similar approach was used for measurements of transferrin and Rab11 colocalization. The average gray value of Fzd6 was measured by outlining apical-basal and anterior-posterior areas of cells using the object function, followed by quantitative analysis of Fzd6 intensity. The ratio of AB to AP was then analyzed statistically.

For measurements of Celsr1 intensity in mitotic basal cells: z projections (three to five slices) were created from 60× images of E16.5 WM epidermal tissue. E-cadherin staining was used to draw a region of interest around mitotic cells identified as either in metaphase or telophase on the basis of their nuclear morphology, and the average gray value of Celsr1 pixel intensity was measured using Image J software. The average gray value data were normalized to the control condition and plotted as shown in Fig. 6B.

ImageJ and Photoshop CS6 were used to arrange and assemble all images and figures. Statistical analysis was performed using Excel and Graphpad Prism software. For most experiments, data were tested for normalcy and two-tailed Student's t-test was used to determine significance. For analysis of Celsr1 mitotic internalization, data were non-parametric and subjected to a Mann-Whitney test to determine significance. Watson's U² was used to test the distribution of planar angles of division between control and Arl3 KD conditions, using the quadrants/binned data as indicated in Fig. 3K,L.

We thank members of the Ezratty lab for thoughtful discussions and critical reading of the manuscript.