GRAF2, WDR44, and MICAL1 mediate Rab8/10/11–dependent export of E-cadherin, MMP14, and CFTR ΔF508

Integral proteins can reach the plasma membrane via different routes. Here, Lucken-Ardjomande Häsler et al. identify three proteins that are associated with dynamic intracellular tubules, closely aligned with the ER, and involved in the transport of specific cargos whose export is particularly sensitive to ER stress.


Introduction
In eukaryotic cells, the ER is the birthplace of the majority of membrane proteins, secreted proteins, and lipids. Despite the canonical belief that they follow the same route from the ER through the ER-Golgi intermediate compartment (ERGIC), Golgi, and TGN to reach the plasma membrane, differences between individual cargos exist. Lipids can be transferred between membranes at contact sites (Nishimura and Stefan, 2020). Some proteins transit via tubular endosomes (Desclozeaux et al., 2008;Henry and Sheff, 2008;Ang et al., 2004;Monis et al., 2017), others such as MMP14 (also called MT1-MMP) and GLUT4 are stored in vesicles for timed or targeted release (Bravo-Cordero et al., 2007;Watson et al., 2004), and a few, such as CFTR and interleukin-1β, can enter a route opened by ER stress (Dupont et al., 2011;Gee et al., 2011). These differences might arise from binding to different adapters or partitioning in membrane domains, which could lead to proteins exiting the classical pathway of secretion at any step (Marie et al., 2009;Chen et al., 2017;Hoffmeister et al., 2011;Stephens and Pepperkok, 2004). Under certain conditions, some integral proteins and lipids are still exported when cells are incubated with Brefeldin A (BFA), which among other things leads to dissolution of the Golgi into the ER (Fujiwara et al., 1988). These cargos have been proposed to bypass the Golgi and are said to follow an unconventional pathway of secretion. Among the cargos that have been reported to reach the plasma membrane in the presence of BFA are E-cadherin (Low et al., 1992), MMP14 (Deryugina et al., 2004), CFTR (Rennolds et al., 2008;Gee et al., 2011), and the ciliary protein Polycystin-1 (Gilder et al., 2018). Whether these cargos actually bypass the Golgi and follow the same route out of the ER is unclear, but what they also have in common is that their export depends on a small group of Rabs.

Results
GRAF1b and GRAF2 colocalize with Rab8a/b and Rab10 In HeLa cells, a prototypical epithelial cell line with wellcharacterized membrane trafficking pathways, GRAF1b and GRAF2 are associated with dynamic tubules and vesicles (Lucken-Ardjomande Häsler et al., 2014). To identify the Rabs associated with GRAF-mediated trafficking, we screened live cells for colocalization with GRAF1b and GRAF2. GRAF1b/2 colocalized with Rab8a/b and displayed a striking contiguous distribution with Rab10 on the same tubules ( Fig. 1 A and Fig. S1 A). GRAF1b/2 colocalized with Rab8a/b throughout the lifetime of the tubules, but GRAF1b/2 were recruited to preexisting Rab10 tubules ( Fig. 1 B and Fig. S1 B). In fixed cells, GRAF1b/2 also colocalized with endogenous Rab8 (Fig. 1 C and Fig. S1 C). Down-regulation of GRAF2 expression (the main GRAF of HeLa cells; Fig. S1, D-F) with transfection of an shRNA-encoding plasmid specific for GRAF2 (shGRAF2a; Fig. S1 G) did not change the percentage of cells with RFP-Rab8 or Rab10 tubules (Fig. 1 D). It led however to a decrease in the total skeletal length of the tubules ( Fig. 1 E and Fig. S1 H). The BAR domain of GRAF2 was essential, as GRAF2 ΔBAR was cytosolic and failed to colocalize with Rab8a/10 tubules (Fig. 1 F).
Rab8 and Rab10 regulate the same intracellular trafficking pathways (Sato et al., 2014;Sano et al., 2008;Sun et al., 2010;Homma and Fukuda, 2016). In agreement, Rab8a and Rab10 labeled the same network of intracellular membranes (Fig. 1 G). Our data suggest that Rab8 may promote the recruitment of GRAF1b/2 to Rab10 tubules. Indeed, upon coexpression of the dominant negative mutant Rab8a T22N (Peränen et al., 1996), GRAF1b/2 were cytosolic ( These observations show that GRAF1b/2 are associated with the same intracellular transport intermediates as Rab8 and Rab10 and demonstrate a requirement of Rab8 activation for GRAF-positive compartments to form.

GRAF1b and GRAF2 bind to MICAL1 and WDR44
We sought to determine how GRAF1b/2 might be connected to Rab8 and Rab10, since there was no direct interaction. We performed pull-downs from rat brain and HeLa cell lysates using GRAF SH3 domains as baits, as they are their only proteinprotein interaction domain and SH3 domains are enriched in proteins associated with small GTPase signaling (Xin et al., 2013;Fig. S2 A). 47 proteins were identified, of which 8 were pulled down from both lysates (Fig. S2 B). Among these eight, two-MICAL1 and WDR44-are Rab-binding proteins. We confirmed pull-down of endogenous MICAL1 and WDR44 by GRAF1/2 SH3 (Fig. 2 A) and coimmunoprecipitation of the overexpressed proteins in 293T cells (Fig. S2 C). 293T cells were chosen for biochemical assays because of their high efficiency of transfection and protein expression. In both cases, GRAF1b/2 bound a proline-rich motif, as binding was abolished by deletion of the proline-rich region of WDR44 (Fig. 2,B and C;and Fig. S2 D) and by substitution of a single aa (K832A) within the PPKPP motif of MICAL1 (Fig. 2,B and D;and Fig. S2 E). Binding was direct, since GRAF2 SH3 bound to WDR44 ΔC and MICAL1 ΔMOΔCH in vitro (Fig. S2 F). Western blot analysis showed that WDR44 and MICAL1 were expressed in a variety of cell lines (Fig. S2 G). Endogenous GRAF2 coimmunoprecipitated WDR44 from lysates of HeLa cells (Fig. 2 E) and MICAL1 from lysates of SH-SY5Y cells (Fig. 2 F), which have a higher MICAL1/WDR44 expression ratio (Fig. S2 G). WDR44 and MICAL1 are thus two novel direct GRAF1b/2-binding proteins.
In live HeLa cells, GFP-tagged MICAL1 was mostly cytosolic and diffuse but was associated with dynamic tubules in 15%-20% of the cells (Fig. 2 G, Fig. S3 A, and Video 1), where it colocalized throughout time with GRAF1b/2 ( Fig. 2 H, Fig. S3 B, and Video 2). Although MICAL1 PPAPP could sometimes be found on GRAF1b/2 tubules (Fig. S3 C), colocalization was significantly reduced (Fig. 2 I). MICAL1 antibodies did not label any clear structures in HeLa cells (Fig. 2 J) and were only faintly associated with GRAF1b/2 tubules (Fig. 2 K and Fig. S3 D). GFP-tagged WDR44, on the other hand, labeled a heterogeneous ensemble of dynamic tubules and vesicles but also, in ∼30% of the cells, peripheral amorphous patches (Fig. 2 G, Fig. S3 A, and Video 1). These patches were an artifact of protein overexpression, as endogenous WDR44 was only associated with puncta and tubules (Fig. 2 J and Fig. S3 E). In addition, unlike WDR44 tubules, they did not colocalize with GRAF1b/2 ( Fig. 2 H, Fig. S3 F, and Video 3). In agreement with its failure to coimmunoprecipitate GRAF1b/2, WDR44 ΔPro did not colocalize with them (Fig. 2 I and Fig. S3 G). Endogenous GRAF2 was found on GFP-WDR44 tubules and vice versa (Fig. 2,K and L). The number of HeLa cells with endogenous WDR44 tubules increased upon overexpression of GRAF1b/2, but not MICAL1 (Fig. 2 M). Reciprocally, it was significantly decreased by down-regulation of GRAF2 expression, but not MICAL1 (Fig. 2 N and Fig. S3 H). The BAR domain of GRAF2 was necessary, as GRAF2 ΔBAR was diffuse and, when transfected in shGRAF2a-expressing cells, led to the association of RFP-WDR44 with puncta instead of long tubules (Fig. S3 I). In HeLa cells, the extent of endogenous WDR44 tubulation was thus directly related to the expression level of GRAF2.
We have shown that WDR44 and MICAL1 bind the same domain of GRAF1b/2, suggesting that they may compete for GRAF1b/2 binding. In agreement, although WDR44 and MICAL1 were found on the same tubules, they were enriched on complementary segments (Fig. 2 O). Live imaging showed the initial formation of MICAL1-positive tubules to which WDR44 was later recruited (Video 4). These experiments show that WDR44 and MICAL1 are dynamically associated with the same intracellular tubules as GRAF1b/2 and suggest a temporal transition from GRAF1b/2-MICAL1 to GRAF1b/2-WDR44 complexes.

MICAL1 connects GRAF1b/2 to Rab8 and Rab10
Based on the literature, MICAL1 may connect GRAF1b/2 to Rabs. To investigate this possibility, we first confirmed earlier reports that MICAL1 interacts with activated forms of Rab8a, Rab8b, and Rab10, but not Rab1 or Rab11a Rai et al., 2016), by testing coimmunoprecipitation of the overexpressed proteins in 293T cells (Fig. 3 A and Fig. S4 A) and colocalization in HeLa cells (Fig. 3 B). Time-lapse imaging showed that MICAL1 was recruited simultaneously to Rab8a and Rab10 on newly formed tubules (Video 5 and Video 6). Endogenous Rab8 was found on overexpressed MICAL1 tubules and reciprocally (Fig. 3,C and D). MICAL1 knockdown significantly decreased the extent of tubulation associated with RFP-Rab8a and Rab10 (Fig. 3,E and F;and Fig. S4 B). This suggests a role for endogenous MICAL1 in the growth of Rab8/10 tubules. In agreement with recent data (Rai et al., 2016), a protein made of the last 167 aa of MICAL1, MICAL1 tail , was sufficient to coimmunoprecipitate Rab8a (Fig. 3 G). The GRAF-binding domain of MICAL1 lies outside this region, as MICAL tail did not coimmunoprecipitate GRAF1b/2, unlike MICAL1 Pro (aa 800-850), which contains the PPKPP motif ( Fig. 3 H and Fig. S4 C). This suggests that MICAL1 may bind simultaneously to GRAFs and Rabs. In agreement, Rab8a coimmunoprecipitated GRAF1b/2 upon coexpression of MICAL1, but not MICAL1 PPAPP (Fig. 3 I and Fig. S4 D). MICAL1 can thus bridge GRAF1b/2 and Rab8a and, by extension, is expected to bridge GRAF1b/2 and Rab8b/10. WDR44 tubules form in a Rab11-, Rab8-, and Rab10-dependent manner and are induced by Cytochalasin D WDR44 was originally identified as a Rab11-binding protein (Zeng et al., 1999;Mammoto et al., 1999). In agreement, Rab11a, but not Rab8a/b or Rab10, coimmunoprecipitated WDR44 when overexpressed in 293T cells (Fig. 4 A). When transfected in HeLa cells, Rab11a was principally associated with puncta that colocalized with Transferrin-positive endosomes (Fig. S4 E). Rab11a was also found on endogenous WDR44 tubules, but WDR44 was not recruited to Rab11a-positive endosomes (Fig. 4 B and Fig. S4 F). In addition, neither GRAF1b/2 nor MICAL1 colocalized with Rab11a (Fig. 4 C and Fig. 3 B). Whereas overexpression of Rab11a or of the constitutively active mutant Rab11a Q70L had no significant effect on WDR44 tubulation, overexpression of the dominant negative mutant Rab11a S25N, which did not bind WDR44 (Fig. S4 G), decreased it ( Fig. 4 D). Down-regulation of Rab11a expression with transfection of the shRNA-expressing plasmid shRab11a (Fig. S4 H) also led to a significant decrease in WDR44 tubulation (Fig. 4 E). Therefore, whereas Rab11a is not a stable component of the MICAL1/GRAF/WDR44 tubules, its activation is important for WDR44-positive tubules to form.
Even though WDR44 did not bind to Rab8 or Rab10, it displayed a contiguous colocalization with them on intracellular tubules (Fig. 4,F and G;and Fig. S4 I). Imaging of live cells showed that WDR44 was recruited to preexisting Rab8a-and Rab10-positive tubules (Fig. 4 H, Video 7, and Video 8). Rab10 then dissociated from WDR44-positive tubules, but Rab8a remained. By similarity to Rab8 and Rab10 (Etoh and Fukuda, 2019;Radhakrishna and Donaldson, 1997;Hattula et al., 2006), WDR44 tubules were destabilized by incubation of cells with Nocodazole, which disrupts microtubules, and induced by incubation of cells with Cytochalasin D, which fragments the actin cytoskeleton (Fig. 4,I and J;and Fig. S4,J and K). Cytochalasin D also improved detection of endogenous MICAL1 along WDR44positive tubules (Fig. 4 K) and on GRAF1b/2 tubules (Fig. 4,L and M;and Fig. S4 L). Overexpression of Rab8a and of the constitutively active mutants Rab8a Q67L and Rab10 Q68L increased WDR44 tubulation (Fig. 4 D). Conversely, overexpression of the dominant negative mutant Rab10 T23N decreased it. These experiments suggest that Rab8 and Rab10 actively promote the formation of intracellular tubules, to which WDR44 is recruited. In agreement, knocking down Rab8a or Rab10 inhibited endogenous WDR44 tubulation (Fig. 4 E;and Fig. S4 M). Therefore, although WDR44 only binds to Rab11a, WDR44 tubule formation is differentially controlled by the three GTPases Rab8, Rab10, and Rab11. Our experiments also show that the formation, stability, and turnover of MICAL1-GRAF-WDR44 complexes and associated membrane tubules are intricately connected to the status and dynamics of the actin and microtubule cytoskeleton. MICAL1 G3W, WDR44 ΔC, and GRAF1/2 BAR-PH are dominant negative mutants interfering with Rab8/10-mediated trafficking and WDR44 tubulation The MICAL family of proteins is thought to coordinate Rab binding or receptor activation with rearrangements of the cytoskeleton (Giridharan and Caplan, 2014;Frémont et al., 2017). MICAL1-3 have a MonoOxygenase (MO) domain that can oxidize proteins (Vitali et al., 2016;Schmidt et al., 2008). So far, two MICAL substrates have been identified, Actin and CRMP2, whose oxidation results in destabilization of F-actin and microtubules, respectively (Lee et al., 2013;Morinaka et al., 2011). Deletion (MICAL1 ΔMO) or inactivation (MICAL1 G3W; Terman et al., 2002) of the MO domain of MICAL1 increased its association with intracellular puncta and tubules (Fig. 5 A). These structures colocalized with GRAF1b/2, Rab8a, and Rab10 (Fig. 5,B and C;and Fig. S5,A and B), which were still coimmunoprecipitated by MICAL1 G3W (Fig. S5 C). Since the subcellular distribution of Rab8/10 was dramatically altered by MICAL1 G3W (compare Rab8a/10 in Fig. 5 B with Fig. 4 F), we tested whether overexpression of MICAL1 G3W interfered with their function. Indeed, MICAL1 G3W blocked insulin-stimulated export of GLUT4 (Fig. 5 D), which is dependent on Rab10 in adipocytes (Sano et al., 2008). In addition, whereas MICAL1 G3W did not colocalize with WDR44 (Fig. 5,C and E;and Fig. S5 D), overexpression of MICAL1 G3W decreased endogenous WDR44 tubules and prevented their induction by Cytochalasin D (Fig. 5 F). This suggests that the MO activity of MICAL1 promotes its dissociation from intracellular membranes and the turnover of its complex with GRAF2. This would limit its effects to a local destabilization of the cytoskeleton that could facilitate the recruitment of WDR44 by GRAF2 but still allow the elongation of WDR44 tubules along intact microtubules. MICAL1 G3W is thus a dominant negative mutant that interferes with Rab10-(and presumably Rab8-) mediated trafficking and with WDR44 tubule formation.
By analogy, we then looked for a WDR44 mutant that might interfere with Rab-mediated trafficking. WDR44(1-504) was reported to inhibit Rab11-mediated recycling of the Transferrin receptor (Zeng et al., 1999). In agreement, WDR44 ΔC was mostly found on peripheral patches (Fig. 5 G), where it recruited Rab11a (Fig. 5,H and I) and colocalized with internalized Transferrin and the recycling endosome marker VAMP3 (Fig. S5 E). Unlike Rab10, Rab8a was not found on patches of WDR44 ΔC; but as WDR44 ΔC colocalized with both of them on cytoplasmic tubules and puncta (Fig. 5,H and I), it may still interfere with their functioning. In agreement, WDR44 ΔC inhibited insulinstimulated export of GLUT4 (Fig. 5 D). WDR44 ΔC is thus a dominant negative mutant that interferes not only with Rab11mediated recycling but also with Rab10-(and presumably Rab8-) dependent trafficking.
We have shown that WDR44 and MICAL1 bind to the SH3 domain of GRAF1/2. But even though the membrane-binding regions of GRAF1/2 (GRAF1/2 BAR-PH) did not colocalize with endogenous MICAL1, endogenous WDR44 was found on the same structures (Fig. 5 J and Fig. S5 F). Live imaging showed that these tubules were relatively immobile (Video 9). By similarity to the full-length proteins, GRAF1/2 BAR-PH induced endogenous WDR44 tubules (Fig. 5 F). Cytochalasin D, however, did not further stimulate it. This suggests that overexpression of GRAF1/2 BAR-PH interfered with the normal dynamics of WDR44 tubules. In addition, GRAF1/2 BAR-PH colocalized with Rab8a and Rab10 on puncta and tubules (Fig. 5,K and L;and Fig. S5 G) and interfered with insulin-stimulated export of GLUT4 (Fig. 5 D). Pull-down of HeLa cell proteins by GST, GST-GRAF1 SH3, or GST-GRAF2 SH3, analyzed on two replicate membranes. One was probed with α-WDR44, the other with α-MICAL1. They were then stained with Coomassie, revealing GST-tagged proteins. (B) Schematic representation of human GRAF2, MICAL1, and WDR44 and of mutants used. GRAF2 has two membrane-binding domains, BAR and PH, a GTPase Activating Protein (GAP) domain, and an SH3 domain. MICAL1 has an NADPH-binding MO domain, CH and LIM domains, a proline-rich region with a PPKPP motif, and a C-terminal coiled coil (CC). * indicates K832 mutated in MICAL1 PPAPP. WDR44 has an N-terminal FFAT motif, a proline-rich region, and seven WD repeats. (C and D) Immunoprecipitation (IP) of transfected 293T cells with α-GFP (C) or using α-myc-coated beads (D). The efficiency of GRAF2 binding (untagged GRAF2 [uGRAF2]; C) or GRAF2-GFP (D) was quantified, using coimmunoprecipitation with GFP-WDR44 (C) or myc-MICAL1 (D) as reference. n = 3 or 4. Binding was decreased by removal (C) or mutation (D) of proline-rich regions. (C) Note that GFP-WDR44 underwent proteolysis. (E) Endogenous WDR44 and GRAF2 were coimmunoprecipitated from a HeLa cell lysate using α-GRAF2 and α-WDR44, but not without antibody (−). The membrane was probed with α-GRAF2, followed by α-WDR44. The left and right parts of the membrane were treated identically. (F) Endogenous MICAL1 was coimmunoprecipitated from an SH-SY5Y cell lysate using α-GRAF2, but not without antibody (−). The membrane was probed with α-GRAF2, followed by α-MICAL1.  This also indicates that the intrinsic characteristics of Rab8a/10 tubules are sufficient to recruit the membrane-binding region of GRAF1/2. MICAL1 G3W, WDR44 ΔC, GRAF2 BAR-PH, and GRAF1 BAR-PH are thus four dominant negative proteins affecting the normal functioning of different components of Rab8/10/11-and MICAL1/ GRAF/WDR44-mediated trafficking.
WDR44 tubules are in close contact with the ER via binding to VAPA/B To gain more insight into the identity of MICAL1/GRAF/WDR44 tubules, we first examined colocalization of endogenous WDR44 with dextran as a marker of clathrin-independent endocytosis. There was none (Fig. 6 A). Endogenous WDR44 tubules also did not colocalize with markers of the ERGIC (ERGIC53), Golgi (GM130), TGN (TGN46), lysosomes (LAMP2), or recycling endosomes (internalized Transferrin; Fig. 6 B). They were, however, often aligned with markers of the ER (Calnexin-GFP), and the patches associated with WDR44 ΔC overlapped with endogenous Calreticulin, an ER protein (Fig. 6 C). We searched for binding partners of WDR44 ΔC by coimmunoprecipitation of endogenous proteins after overexpression in 293T cells and identified VAPA (Fig. S6 A). VAPA and its closely related homologue VAPB are transmembrane proteins of the ER (Fig. S6 B). They act as receptors for many cytoplasmic proteins, which often bind the MSP domain of VAPs via an FFAT motif (EFF-DAxE; Murphy and Levine, 2016). Indeed, WDR44 has an FFAT sequence ( Fig. 2 B), and WDR44 ΔFFAT was not coimmunoprecipitated by VAPA (Fig. 6 D) or VAPB (Fig. S6 C). Reciprocally, VAPA ΔMSP and VAPA DD, with two point mutations in its FFAT-binding site (K94D M96D; Kaiser et al., 2005), were not coimmunoprecipitated by WDR44 (Fig. 6, E and F).
In transfected HeLa cells, WDR44 tubules and patches colocalized with VAPA and VAPB (Fig. 6 G and Fig. S6 D). WDR44 tubules also colocalized with endogenous VAPB (Fig. 6 H and Fig. S6 E) and other endogenous markers of the ER (Fig. S6 F). The dynamics of WDR44 tubules closely followed the dynamics of the ER (Fig. 6 I). Super-resolution microscopy however showed that even though WDR44 and endogenous ER markers were in close apposition, their fluorescence intensity maxima were sometimes slightly shifted from one another (Fig. 6 J). In addition, WDR44 ΔFFAT was still found on tubules ( GRAF/WDR44 label a subset of tubular endosomes but are not involved in recycling To identify the compartments with which WDR44 tubules communicate, we used BFA, a drug that leads to the tubulation of certain organelles and the mixing of groups of intracellular compartments (Lippincott-Schwartz et al., 1991). Incubation of cells with BFA increased WDR44 tubulation (Fig. 7 A and Fig. S7 A). WDR44 was not responsible for the tubulation of the TGN (Video 10), but WDR44 tubules colocalized with TGN46 after 5 min and internalized Transferrin after 15 min (Fig. 7, B and C). Colocalization was also seen with other markers of the TGN (STX16) and recycling endosomes (VAMP3; Fig. S7, B and C). There was no increase in colocalization with ERGIC53, GM130,or LAMP2 (Fig. 7 C and Fig. S7 D). These observations suggest that WDR44 tubules are part of the TGN-endosome-plasma membrane homotypic membrane system (Lippincott-Schwartz et al., 1991). In agreement, when all proteins were overexpressed, STX16 and VAMP3 were found in WDR44 tubules, even without BFA (Fig. 7 D).
GRAF2 and WDR44 are involved in the specific export of neosynthesized E-cadherin, MMP14, CFTR, and CFTR ΔF508 In addition to recycling, Rab8, Rab10, and Rab11 regulate the export of a subset of neosynthesized proteins. Since WDR44positive tubules were induced by BFA ( Fig. 7 A), we focused on candidates that were also reported to use an unconventional pathway of export. We first examined E-cadherin, an adhesion protein essential for epithelial morphogenesis and behaving as a tumor suppressor (Pal et al., 2018). In agreement with Lock and Stow (2005), we found that export of neosynthesized E-cadherin in HeLa cells was inhibited by Rab11a S25N (Fig. 8 A). It was also inhibited by Rab8a T22N, Rab10 T23N, MICAL1 G3W, GRAF1/2 BAR-PH, and WDR44 ΔC ( Fig. 8 A and Fig. S9 A). This was specific, as export of the tight junction protein Occludin was unaffected by these mutants (Fig. S9 B). Shortly after transfection, E-cadherin was found in intracellular puncta that colocalized with Rab8 and GRAF2 and were often associated with GRAF2-positive tubules ( Fig. 8 B). When GRAF2 or WDR44 was knocked down, E-cadherin was exported less efficiently (Fig. 8 C and Fig. S9 C) and was trapped in a perinuclear compartment (Fig. 8 D). Knockdown of MICAL1 led to a modest decrease (12%), albeit not statistically significant, while down-regulation of VAPA/B did not have an effect (Fig. 8 C). Export of Occludin was not affected by down-regulation of GRAF2 or WDR44 (Fig. S9 C).
Neither was it inhibited by down-regulation of GRAF2 or WDR44 (Fig. 8 C).
MMP14 is a matrix metalloproteinase that targets components of the extracellular matrix and is tightly associated with tumor progression and cell migration (Egeblad and Werb, 2002). In agreement with earlier studies, we found that in HT 1080 cells grown on collagen-coated dishes, export of MMP14 was inhibited by overexpression of Rab8a T22N (Fig. 8 A;Bravo-Cordero et al., 2016Wiesner et al., 2013). It was also inhibited by Rab10 T23N, MICAL1 G3W, GRAF1/2 BAR-PH, and WDR44 ΔC. Imaging of live cells showed that MMP14 was transported in Rab8a-and GRAF2-positive tubules ( Fig. 8 E). In addition, export of MMP14 was decreased in cells in which GRAF2 or WDR44 was knocked down (Fig. 8 C), where it accumulated in perinuclear structures (Fig. 8 F). In shWDR44aexpressing cells, export of MMP14 was restored upon cotransfection of the shRNA-resistant protein RFP-WDR44 res , but not RFP-WDR44 res ΔPro ( Fig. 8 G), a mutant that cannot bind GRAF1b/2 ( CFTR is a channel responsible for the transport of anions across the plasma membrane of a variety of epithelial cells. It can follow several routes of export as it progresses through the secretory pathway (Yoo et al., 2002;Egan et al., 2002;Rennolds et al., 2008;Gee et al., 2011). Knockdown of Rab8 or Rab11 was shown to inhibit its apical delivery (Vogel et al., 2015). Of particular relevance for patients with cystic fibrosis (Elborn, 2016), deletion of F508 leads to intracellular retention of the channel and degradation. Surface exposure of CFTR ΔF508 can be helped by small molecules, such as VX-809 and C4. VX-809, which aids protein folding (Farinha et al., 2015;Loo and Clarke, 2017), limits the degradation of misfolded CFTR ΔF508 and, as such, leads to an increase in the expression level not only of the fully glycosylated Golgi-modified protein but also of the immature unglycosylated protein (Farinha et al., 2013(Farinha et al., , 2015. C4, on the other hand, limits the degradation of CFTR ΔF508 by lysosomes (Farinha et al., 2013;Holleran et al., 2013). The combination of VX-809 and C4 is potent in stabilizing plasma membrane levels of CFTR ΔF508, irrespective of its pathway of export. When expressed in HeLa cells, CFTR and CFTR ΔF508 colocalized with Rab8a and WDR44 (Fig. 8 H). In 293T cells, overexpression of Rab10 T23N and MICAL1 G3W significantly decreased export of CFTR and CFTR ΔF508 (Fig. 8 I). Down-regulation of GRAF2, WDR44, MICAL1, and VAPA/B expression inhibited export of CFTR ΔF508, which, in the case of GRAF2 and WDR44, was n = 4-12. (E) Confocal images of transfected HeLa cells stained with α-WDR44 and α-Rab8. Unlike WDR44, endogenous Rab8 colocalized with GFP-MICAL1 G3W. (F) Percentage of transfected HeLa cells left untreated or after incubation with Cytochalasin D (0.5 µg/ml, 30 min), with endogenous (endo.) WDR44 tubules. n = 3-8. (G) Confocal stack of HeLa cells transfected with GFP-WDR44 ΔC showing its association with large peripheral patches but also, in a few cells, with a thin reticular network. (H) Confocal images of transfected HeLa cells. GFP-WDR44 ΔC recruited RFP-tagged Rab11a and Rab10 to amorphous peripheral patches and colocalized with Rab8a and Rab10 on intracellular puncta and tubules. (I) Manders colocalization coefficients for the indicated proteins on GFP-WDR44 ΔC structures. n = 20-31 cells. (J) Confocal images of transfected HeLa cells stained with α-MICAL1 and α-WDR44. Unlike MICAL1, WDR44 was recruited to sections of GRAF2 BAR-PH-positive membranes. (K) Confocal images of transfected HeLa cells. GRAF2 BAR-PH-GFP colocalized with RFP-tagged Rab8a and Rab10. (L) Manders colocalization coefficients for the indicated proteins on GRAF1/2 BAR-PH-GFP structures. n = 21-26 cells. (C, D, F, I, and L) Data are means ± SEM; *, P < 0.05; **, P < 0.01; ***, P < 0.001; and ****, P < 0.0001. (A, B, E, G, H, J, and K) Insets show magnifications of the boxed areas. Scale bars: 10 µm; scale bars of insets: 2 µm. . Export of CFTR was also sensitive to knockdown of GRAF2, WDR44, MICAL1, and VAPA/B, but to a lesser degree, suggesting that for the wild-type protein, other pathways may compensate more efficiently. Down-regulation of GRAF1, which is also expressed in 293T cells (Fig. S1 E), did not significantly perturb export of CFTR ΔF508 (Fig. 8 J).
Export of neosynthesized CFTR, CFTR ΔF508, E-cadherin, and MMP14 is sensitive to ER stress In previous studies, ER stress was shown to induce export of CFTR ΔF508 through a Golgi bypass route (Gee et al., 2011). In agreement, overexpression of Sar1 H79G, a dominant negative mutant interfering with COPII coat dynamics, stimulated export of CFTR ΔF508 ( Fig. 9 A). This was also seen after incubation of the cells with BFA or with thapsigargin, an ER Ca 2+ -ATPase inhibitor ( Fig. 9 B). Strikingly, Sar1 H79G-induced export of CFTR ΔF508 was inhibited by MICAL1 G3W (Fig. 9 C). By similarity, BFA and thapsigargin-induced export were inhibited by Rab10 T23N (Fig. 9 B). These results suggest that basal and stressinduced exports of CFTR ΔF508 converge in Rab10-associated tubular endosomes.
Contrasting with CFTR ΔF508, we found that export of E-cadherin and MMP14 was inhibited by Sar1 H79G overexpression or BFA addition, even more so than export of GFP-GPI and α 2B -AR (Fig. 9, A and B). Moreover, unlike GFP-GPI and α 2B -AR, E-cadherin and MMP14 were also trapped in cells in the presence of thapsigargin (Fig. 9 B). This suggests that their export is particularly sensitive to ER stress in general, which prompted us to verify that knocking down WDR44 and GRAF2 did not induce it (Fig. S9 D;Lee, 2005). As BFA increased endogenous WDR44 tubules (Fig. 7 A), this might not be the ER stress-sensitive step of the export pathway of E-cadherin and MMP14. Alternatively, BFA-induced WDR44 tubules might not be functional for protein export, as they were still induced in the presence of Rab10 T23N or MICAL1 G3W (Fig. S9 E). To examine at which point E-cadherin diverges from the classical pathway of secretion, we expressed a fusion protein developed for the Retention Using Selective Hooks (RUSH) system (Boncompain et al., 2012). Streptavidin Binding Protein (SBP)-GFP-E-cadherin consists of the N-terminal signal sequence of interleukin-2 followed by SBP, GFP, and finally mouse E-cadherin without its prodomain (Fig. 9 D). In agreement with Boncompain et al. (2012), SBP-GFP-E-cadherin was trapped in the ER and released with biotin addition (Fig. 9 E and Fig. S9 F). This was also true for SBP-GFP-GPI ( Fig. 9 E). However, when the full sequence of human E-cadherin was used to make E-cadherin-SBP-GFP ( Fig. 9 D), the protein already reached the plasma membrane in untreated cells (Fig. 9 F). Biotin addition failed to induce a significant increase in plasma membrane E-cadherin-SBP-GFP, contrary to VSVG-SBP-GFP ( Fig. 9 E). This suggests that E-cadherin-SBP-GFP already diverges from the classical pathway of secretion in the ER, as it escapes retention by an ER-resident hook.

Discussion
In this study, we report the identification of three novel proteins-GRAF2, WDR44, and MICAL1-participating in the export of a subset of neosynthesized proteins. E-cadherin, MMP14, CFTR, and CFTR ΔF508 were previously found to reach the plasma membrane in a Rab8-and/or Rab11-dependent manner (Bravo-Cordero et al., 2007;Wiesner et al., 2013;Lock and Stow, 2005;Desclozeaux et al., 2008;Vogel et al., 2015). We now show that their export is also sensitive to overexpression of dominant negative Rab10, MICAL1, GRAF1b/2, and WDR44 mutants and inhibited by knockdown of GRAF2 or WDR44. This is specific, as exports of a GPI-anchored protein, of α 2B -AR, and of Occludin are not affected. Apart from CFTR ΔF508, their export is only moderately inhibited by MICAL1 knockdown, maybe because a catalytic amount of MICAL1 is sufficient to target GRAF2 to Rab10/8-positive membranes. This would be coherent with the lower MICAL1 expression in 293T cells (in which CFTR ΔF508 export was assessed) and with the fact that we did not see endogenous MICAL1 tubules under resting conditions. Alternatively, other proteins might substitute for MICAL1, although we α-ERGIC53, α-GM130, α-TGN46, or α-LAMP2. For visualization of recycling endosomes, cells were incubated with Alexa Fluor 546-Transferrin (10 µg/ml, 1 h); for the ER, cells were transfected with Calnexin-GFP. WDR44 tubules only colocalized with Calnexin-GFP. (C) Confocal images of transfected HeLa cells stained with α-Calreticulin, which colocalized with GFP-WDR44 ΔC patches. (D and F) Immunoprecipitation (IP) of transfected 293T cells with α-GFP. The efficiency of binding of myc-VAPA (D) and myc-WDR44 (F) was quantified using binding to GFP-WDR44 (D)  have shown that none of the other MICAL family members interact with GRAF1b/2. Contrasting with our original expectation, but in agreement with others (Geng et al., 2012;Proulx-Bonneau et al., 2011;Chu et al., 2004;Zucker et al., 2002), we found that BFA inhibits the export of E-cadherin and MMP14. We further show that this is true for other inducers of ER stress and is specific, as export of α 2B -AR and of a GPI-anchored protein are unaffected by incubation of the cells with thapsigargin. As inhibition of ER to Golgi transport invariably leads to ER stress (Hikiji et al., 2015;Gee et al., 2011), we now show that it can also specifically block protein export indirectly and therefore cannot be used as proof that a protein enters the classical pathway of secretion. Our results do not allow us to determine whether E-cadherin and MMP14 bypass the Golgi or not, but the fact that E-cadherin-SBP-GFP was not trapped in the ER by an ER-resident hook suggests that E-cadherin already segregates from other cargo in the ER. In agreement, the exit of E-cadherin from the ER was proposed to require binding of the phosphatidylinositol-(4,5)-bisphosphate (PI(4,5)P 2 )-generating enzyme PIPKIγ, something that was reported to be abolished by ER stress (Geng et al., 2012). Nevertheless, as E-cadherin and MMP14 accumulate in a perinuclear compartment in cells in which GRAF2 or WDR44 has been knocked down, GRAF2/WDR44 tubules are probably not directly involved in transport out of the ER, but at a later step. This step might be shared with stressinduced unconventional secretion that bypasses the Golgi, as MICAL1 G3W and Rab10 T23N also inhibit stress-induced export of CFTR ΔF508. Our results therefore support the existence of at least three modes of protein export: (1) a classical route using COPII-coated vesicles followed by GPI-anchored proteins and α 2B -AR, insensitive to ER stress but blocked by Sar1 H79G and BFA; (2) a route inhibited by ER stress followed by E-cadherin, MMP14, CFTR, and CFTR ΔF508 that may or may not use COPIIcoated vesicles and at some stage depends on Rab8, Rab10, GRAF2, and WDR44; and (3) a stress-induced pathway followed by CFTR ΔF508 that bypasses the Golgi and is also dependent on Rab10.
In recent years, proteins of the GRAF family have attracted attention because they are mutated in cancer (Li et al., 2017;Aissani et al., 2015;Bozóky et al., 2013;Qian et al., 2010) and have been associated with neuropsychiatric diseases (Barresi et al., 2010;Jarius and Wildemann, 2015;Dahm et al., 2014), but little mechanistic information has been published. We show here that direct binding of proline-rich regions of MICAL1 and WDR44 to the SH3 domain of GRAF1/2 allows GRAF-mediated trafficking to be controlled by Rab8, Rab10, and Rab11 (see model in Fig. 10). Rab8, Rab10, MICAL1, GRAF1b/2, and WDR44 are dynamically associated with the same intracellular tubules. Activation of Rab8a is essential for GRAF1b/2 to be recruited. Our results suggest that GRAF1b/2 can be brought to preformed Rab8/10 tubules via MICAL1 binding and that both MICAL1 and GRAF1b/2 contribute to Rab8/10 tubule extension. The MO activity of MICAL1 regulates the turnover of its complex with GRAF1b/2, as overexpression of an enzymatically inactive MI-CAL1 G3W mutant prevents recruitment of WDR44. Knockdown of GRAF2, Rab8, or Rab10 inhibits formation of endogenous WDR44 tubules, while overexpression of GRAF1b/2 or of constitutively active mutants of Rab8 and Rab10 induces them. With seven WD repeats that fold into β-propellers (Li and Roberts, 2001), WDR44 is likely to constitute an interaction platform for the recruitment of yet-to-be-identified proteins that could contribute to tubule elongation, membrane scission, transport, or fusion. Our results also indicate that although Rab11a is not a stable component of GRAF1b/2 tubules, it indirectly regulates the pathway via WDR44. MICAL1/GRAF/WDR44 constitutes a new molecular path that can connect Rab11 to Rab8/10, as suggested before for Rabin-8 (Knödler et al., 2010) and Myosin Vb (Roland et al., 2011).
Overexpression of Rab8 is known to induce a decrease in F-actin and focal adhesions in a RhoA-dependent manner (Hattula et al., 2006;Bravo-Cordero et al., 2016). We now show that GRAF1b/2 and MICAL1 could be involved, GRAF1b/2 via their GTPase Activating Protein (GAP) domain (Ren et al., 2001;Taylor et al., 1999;Doherty et al., 2011a) and MICAL1 via its MO domain (Giridharan and Caplan, 2014). Under resting conditions, both MICAL1 and GRAF1b/2 are in an auto-inhibited state (Eberth et al., 2009;Giridharan et al., 2012;Schmidt et al., 2008). This could be released in response to Rab8 and MICAL1 binding, respectively, and allow the three proteins to cooperate in triggering rearrangements of the cytoskeleton (Nüchel et al., 2018;Frémont et al., 2017). Reciprocally, we also report that similar to Rab8 tubules, endogenous WDR44 tubules are stabilized on microtubules and are induced by destabilization of F-actin. In the case of Rab8, Cytochalasin D was proposed to accelerate macropinocytosis, leading to an increase in trafficking through recycling endosomes (Hattula et al., 2006). This may also be responsible for the increase in WDR44 tubules, as similar to Rab8 and Rab10 (Henry and Sheff, 2008;Ang et al., 2003;Gerges et al., 2004;Etoh and Fukuda, 2019;Shibata et al., 2016), we have shown that WDR44 tubules are part of the TGNendosome-plasma membrane homotypic membrane system (Lippincott-Schwartz et al., 1991). GRAF2-dependent WDR44  . Export of neosynthesized CFTR, CFTR ΔF508, E-cadherin, and MMP14 is sensitive to ER stress. (A) Export efficiency of GFP-Extope CFTR or GFP-Extope CFTR ΔF508 in 293T cells, of E-cadherin-RFP, HA-α 2B -AR-RFP, or GFP-GPI in HeLa cells and of MMP14-GFP in HT 1080 cells. n = 3-6. Sar1 H79G increased export of GFP-Extope CFTR ΔF508 but inhibited export of E-cadherin-RFP and MMP14-GFP. (B) Export efficiency of GFP-Extope CFTR ΔF508 in 293T cells and of E-cadherin-GFP, MMP14-GFP, HA-α 2B -AR-RFP, or GFP-GPI in HeLa cells. For GFP-Extope CFTR ΔF508, cells were left untreated, incubated with BFA (1.25 µg/ml, 12 h), or thapsigargin (1 µM, 2 h), the latter followed by a 2-h chase (Gee et al., 2011). Cells were collected 36 h after transfection. HeLa cells were incubated with BFA (1.25 µg/ml, 6 h) or thapsigargin (1 µM, 6 h) and collected 12 h after transfection. Basal and ER stress-induced export of GFP-Extope CFTR ΔF508 was inhibited by RFP-Rab10 T23N. ER stress inhibited export of E-cadherin-GFP and MMP14-GFP. n = 4-6. (A and B) The dashed line represents the export efficiency under control conditions. (C) Biotinylation of surface proteins in transfected 293T cells. Export efficiency of GFP-CFTR tubules could thus also be described as tubular endosomes, which correspond to a heterogeneous ensemble of compartments, at the crossroad between endocytic and exocytic pathways (Cresawn et al., 2007;Baetz and Goldenring, 2013;Powelka et al., 2004;Xie et al., 2014). We have shown partial colocalization of endogenous WDR44 with another protein associated with tubular endosomes, MICAL-L1. Recently, overexpression of GRAF1c was reported to decrease the number of endogenous MICAL-L1 tubules (Cai et al., 2014). We made the reciprocal observation: overexpression of MICAL-L1 inhibits WDR44 and GRAF1b/2 tubulation. This suggests competition rather than cooperation between GRAF/WDR44-and MICAL-L1-mediated trafficking. In agreement, unlike GRAF1b/2 and WDR44, the subcellular distribution of MICAL-L1 is insensitive to knockdown of Rab8 and Rab10 (Kobayashi et al., 2014;Rahajeng et al., 2012); and unlike silencing of MICAL-L1 (Sharma et al., 2009), silencing of GRAF2 and WDR44 does not interfere with Integrin-β1 recycling. As knockdown of GRAF1 in 293T cells does not significantly interfere with the export of CFTR ΔF508, it is possible that GRAF1c participates in MICAL-L1-mediated protein recycling, but that together with WDR44, GRAF2 is specifically involved in exocytosis of neosynthesized proteins.
Our results also reveal a new feature of GRAF2/WDR44 tubules: they are closely aligned with ER membranes via binding of WDR44 to VAPA/B. Even though VAPA/B were proposed to play a role in protein export and in membrane trafficking (Peretti et al., 2008;Dong et al., 2016;Wakana et al., 2015;Kuijpers et al., 2013;Saita et al., 2009), it is the first time that contact sites between a membrane trafficking intermediate and the ER have been observed. We have shown that downregulation of VAPA/B expression inhibits endogenous WDR44 tubules. However, among the cargos tested, this only affected export of CFTR ΔF508, indicating that VAPA/B are not mechanistically essential for GRAF2/WDR44-derived trafficking and that the effect on CFTR ΔF508 could also be related to other changes resulting from VAPA/B knockdown (Dong et al., 2016;Wakana et al., 2015). VAP-mediated ER contacts might therefore simply contribute to the formation and/or stabilization of membrane tubules on a preexisting reticular scaffold, control their dynamics (Dong et al., 2016;Raiborg et al., 2015;Rocha et al., 2009), or help target the tubules, for example, to PI(4,5) P 2 -enriched microdomains of the plasma membrane (Pan et al., 2018). But they might also allow lipid exchange with the ER (Peretti et al., 2008;Perry and Ridgway, 2006;Mesmin et al., 2013;Kim et al., 2015), indirectly promoting cargo selection. Of interest, Rab10 was proposed to play a role in the growth of new ER tubules by regulating the formation of a domain enriched in lipid-synthesizing enzymes (English and Voeltz, 2013). Therefore, if GRAF2/WDR44 were also associated with these domains, they could play a role in transporting newly synthesized lipids directly out of the ER and indirectly regulate the morphology and dynamics of the ER.
Our results have deepened our understanding of the mechanism of GRAF2 and Rab8/Rab10-mediated trafficking. Of interest, although little was known about the function of mammalian WDR44 and MICAL1, they were both proposed to play a role in protein export (Mammoto et al., 2000;Van Battum et al., 2014;Grauzam et al., 2018). Given the importance of the proper regulation of plasma membrane levels of E-cadherin and MMP14 in cancer and of CFTR in cystic fibrosis, it is essential to identify the molecular pathways that control their trafficking.
Our results now show that in addition to Rab8/10/11, GRAF2, WDR44, and MICAL1 are also important.
All constructs were verified by sequencing.

Antibodies and reagents
Purified rabbit polyclonal α-GRAF1 and α-GRAF2 antibodies were described previously ( The secondary reagents for Western blotting were goat α−mouse and α−rabbit IgG-HRP conjugates (Bio-Rad). HRP-Protein A (Thermo Fisher Scientific) was used for analysis of beads when antibodies used for the immunoprecipitation and for Western blotting were raised in the same species.
For knockdown of protein expression in HeLa cells, cells were transfected with an shRNA-encoding plasmid, selected 24 h later with puromycin (10 µg/ml, 16 h), and then split. Experiments were performed after 88-96 h of knockdown. If shRNA cells were to be transfected, transfection was performed 72 h after the transfection of the shRNA. For knockdown of protein expression in 293T cells, cells were cotransfected with an shRNA-encoding plasmid and pCI N-RFP DEST, pCI GRAF2 res -RFP, pCI RFP-WDR44 res , or pCI RFP-WDR44 res ΔPro. Cells were split, transfected after 52-56 h, and collected 36 h later. For knockdown of protein expression in HT 1080 cells, cells were infected with lentivirus, selected 24 h later with puromycin (1 µg/ml, 16 h), and split. Cells were transfected 72 h after the original infection with pCI MMP14-GFP, alone or together with pCI RFP-WDR44 res or pCI RFP-WDR44 res ΔPro, and analyzed 16 h later. To make lentivirus, 293T cells were grown on Poly-L-Lysine-coated plates and cotransfected with shRNA, pMD2.G, and PSPAX2. Medium was changed to DMEM supplemented with 4% FBS and 25 mM Hepes after 24 h, and lentivirus was harvested as tissue culture supernatant after a further 48-60 h.

Immunofluorescence
Cells grown on glass coverslips were fixed in 3.2% PFA diluted in culture medium (20 min, 37°C). For immunostaining of cytoplasmic proteins, coverslips were then washed in PBS and blocked in 5% normal goat serum (NGS) and 0.1% saponin (Sigma) in PBS. All further incubations were done in 1% NGS and 0.1% saponin in PBS. Coverslips were mounted on slides in a buffered PVA glycerol mountant containing 2.5% DABCO (Sigma). For staining of F-actin, Alexa Fluor 546-labeled phalloidin was added to the primary antibody dilution (final concentration, 0.4 U/ml). A similar protocol was followed for immunostaining of surface proteins under nonpermeabilizing conditions, but saponin was omitted from the buffers.

Confocal microscopy
For live imaging, cells were grown on glass-bottom culture dishes (MatTek). Immediately before imaging, medium was replaced with phenol red-free DMEM containing 5% FBS and placed in a temperature-controlled chamber on the microscope stage with air/CO 2 95:5 and 100% humidity.
Most imaging data were acquired using a fully motorized inverted microscope (Eclipse TE-2000; Nikon) equipped with a CSU-X1 spinning disk confocal head (UltraVIEW VoX; Perki-nElmer) using a 60× oil immersion lens (Plan Apochromat V, 1.4 NA; Nikon) under control of Volocity 6.0 (PerkinElmer). 14-bit digital images were acquired with a cooled EMCCD camera (9100-02; Hamamatsu Photonics). Three 50-mW solid-state lasers (488, 561, and 647 nm; CVI Melles Griot) coupled to individual acoustic-optical tunable filters (AOTF) were used as a light source to excite fluorescent proteins and dyes, as appropriate. Imaging of RUSH proteins was performed with a Zeiss LSM 780 using standard photomultiplier tubes, a 63× oil immersion lens of NA 1.4, and ZEN acquisition software. ImageJ 1.48s (National Institutes of Health) was used to process the images, adjusting only the brightness and contrast of the different channels. Due to the high dynamic range of the signals and the fact that we were particularly interested in the dim structures in the cell's periphery, we adjusted the scaling of the images to bring this information into the dynamic range of a computer display. No filters were applied. We followed the principles of Color Universal Design. On the merged channel, images acquired in the red channel are colored in red; images acquired in the green channel are colored in cyan; and images acquired in the far red channel are colored in magenta. Representative images of live cells and fixed cells collected in a minimum of three independent experiments are shown. Unless otherwise mentioned, images are from a single focal plane. For confocal stacks, maximum intensity projections of images acquired at 0.3-µm intervals, in single channels and as a merged image, are shown.
For quantification of the proportion of HeLa cells with RFP-Rab8a/10 tubules, cells were fixed 20 h after transfection. The number of cells with at least one tubule was counted and expressed as percentage of the total number of cells examined (minimum 25 cells per sample per experiment). For quantification of the total skeletal length of RFP-Rab8a/10 structures and for quantification of the total volume of GRAF1b/2-GFP structures per cell, cells were fixed 20 h after transfection. Confocal stacks were acquired. Each cell was identified as a region of interest. Volocity was used to identify fluorescent objects and measure their combined skeletal length or volume. Manders colocalization coefficients were quantified using Volocity, setting thresholds for each cell using a region of cytosol. For quantification of the proportion of HeLa cells with endogenous WDR44 tubules, cells were fixed 24 h (overexpression) or 96 h (shRNA) after transfection. The number of cells with at least one WDR44 tubule was counted and expressed as a percentage of the total number of cells examined (minimum 100 cells per sample per experiment). For quantification of the total WDR44 length per cell, confocal stacks were acquired. Each cell was identified as a region of interest. WDR44 objects with a nonspherical shape (shape factor <0.5) were identified using the Volocity software, and their skeletal length was measured. The total skeletal length per cell was quantified. For displaying fluorescence intensity profiles, fluorescence intensities were measured on a 1-px line using ImageJ software. Values were then normalized using the minimum and maximum values measured along that line. For detailed numbers of independent experiments, see the figure legends.
The microscope was a Leica TCS SP8 STED 3X equipped with a pulsed white light laser and HyD hybrid detectors enabling gating. Gating was adjusted to detect signal emitting between 1.7 and 6.8 ns after the excitation pulse. Excitation of Alexa Fluor 488 was performed at 488 nm and of Alexa Fluor 514 at 535 nm. Suitable clean-up filters were in place. The depletion laser was at 592 nm.
The primary lens was a 100×/1.46 NA (orange, HC PL APO 100×/1.40 oil), chromatically corrected across the used spectrum. Sampling frequency was set to 20 nm, and the depletion laser was operated at settings that allowed a lateral resolution below 60 nm. As the photon budget was limited, line averaging (16×) and frame accumulation (8×) were used to achieve a suitable signal-to-noise ratio. ImageJ was used to process the images and to measure fluorescence intensity profiles across tubules.
For pull-downs of cell lysates, beads were washed three times in cell lysis buffer (20 mM Hepes, pH 7.4, 150 mM NaCl, 2 mM EDTA, 1% IGEPAL CA-630 [Sigma], 1 mM DTT, and cOmplete Protease Inhibitor Cocktail). In parallel, 15-cm dishes of HeLa cells or rat brains were lysed in the same buffer (30 min, 4°C). Lysates were cleared by centrifugation (30 min, 100,000 g), and protein concentration of the supernatants was determined with a Bradford assay (Bio-Rad). Equivalent protein amounts of GST, GST-GRAF1 SH3, or GST-GRAF2 SH3 bound to glutathione sepharose beads were incubated with a volume of lysate corresponding to ∼7 mg protein or with the same volume of cell lysis buffer for 30 min at 4°C under rotation. Beads were pelleted (5 min, 500 g) and washed five times in cell lysis buffer without protease inhibitors. For identification of bound proteins by Western blot, beads were directly incubated with sample buffer, heated at 95°C, and loaded on gels. For identification of bound proteins by mass spectrometry, beads were incubated with Thrombin (90 min, 4°C). Eluted proteins were collected in the supernatant after centrifugation (5 min, 500 g), sample buffer was added, and proteins were separated using 4%-12% NuPAGE Novex Bis-Tris gels. Protein bands were revealed by Coomassie staining. Bands were excised, and proteins were identified by liquid chromatography-tandem mass spectrometry (LC-MS/MS; Thermo Orbitrap). Results were analyzed with Scaffold 3 (Proteome Software).
For in vitro pull-downs, GRAF2 SH3 was released from GST-GRAF2 SH3 beads by incubation with Thrombin (90 min, 4°C) and collected in the supernatant after centrifugation. WDR44 and MICAL1 domains bound to glutathione sepharose beads were then incubated in the presence or absence of an equal amount of GRAF2 SH3 for 30 min at 4°C and washed three times in bacterial lysis buffer. Proteins bound to the beads were separated using 4%-12% NuPAGE Novex Bis-Tris gels and revealed by Coomassie staining.
Protein lysates, immunoprecipitation of endogenous proteins, and Western blotting Untreated cells or shRNA-expressing cells (96 h after transfection) were scraped in cold PBS and pelleted. They were resuspended in lysis buffer (20 mM Hepes, pH 7.4, 150 mM NaCl, 2 mM EDTA, 1% IGEPAL CA-630, and cOmplete Protease Inhibitor Cocktail) and incubated 30 min at 4°C. Lysates were cleared (5 min, 14,000 g), and protein concentration was determined using a Bradford assay. Rat brains were homogenized in the same buffer using a Dounce homogenizer, but lysates were cleared at 100,000 g, 30 min. When lysates were to be directly analyzed, sample buffer was added to equal protein amounts (75 µg protein per sample); in general, for immunoprecipitation of endogenous proteins, 150 µg protein was incubated with purified antibodies or serum in a total volume of 100 µl and incubated 3 h at 4°C under rotation. 12.5 µl of a 50% suspension of Protein G Sepharose 4 Fast Flow beads (GE Healthcare Life Sciences) was added for another 30 min before beads were spun down (5 min, 500 g), washed three times in lysis buffer, and finally resuspended in sample buffer. Samples were heated 3 min at 95°C, loaded on 4%-12% NuPAGE Novex Bis-Tris gels, and analyzed by Western blot. Proteins were transferred to polyvinylidene fluoride membranes; powder milk was used as blocking agent. Secondary antibodies were detected using Amersham ECL detection reagent (GE Healthcare Life Sciences), and results were developed on x-ray film. The same protocol was followed for the coimmunoprecipitation of endogenous proteins from SH-SY5Y cells but using a volume of lysate corresponding to 100 cm 2 of a confluent layer of cells for each immunoprecipitation.
For the identification of the GRAF proteins expressed in HeLa cells, 500 µg protein were used for each immunoprecipitation. For the identification of WDR44-binding proteins, lysates corresponding to a 40 cm 2 confluent layer of 293T cells were pooled. In these two cases, a similar protocol was followed, but proteins were revealed by Coomassie staining of the gels. Bands were excised, and proteins were identified by LC-MS/MS.
For coimmunoprecipitation of endogenous proteins from HeLa cells, cells were resuspended in Triton X-100-containing lysis buffer (20 mM Hepes, pH 7.4, 150 mM NaCl, 2 mM EDTA, 1% Triton X-100, and cOmplete Protease Inhibitor Cocktail) and incubated 30 min at 4°C. Lysates were centrifuged (5 min, 14,000 g). The pellets were resuspended in the same volume of lysis buffer and cleared at low speed (5 min, 1,000 g), and a volume equivalent to 125 cm 2 of a confluent layer of cells was used for each immunoprecipitation.

Coimmunoprecipitation after overexpression in 293T cells
On the first day, 293T cells were plated in six-well dishes and a few hours later were transfected with the plasmid coding for the prey. The next day, cells were transfected with plasmids encoding the baits. 20-24 h later, the medium was removed, and cells were detached by pipetting in 1 ml cold PBS. Cells were spun down, and pellets were resuspended in 100 µl lysis buffer (20 mM Hepes, pH 7.4, 150 mM NaCl, 2 mM EDTA, 1% IGEPAL CA-630, and cOmplete Protease Inhibitor Cocktail) and incubated on ice for 30 min. Lysates were cleared (5 min, 14,000 g). 10 µl of each supernatant was collected, and 5 µl sample buffer was added (lysates). The rest of the supernatants were incubated with 12.5 µl of a 50% suspension of Protein G Sepharose 4 Fast Flow beads for 30 min at 4°C under rotation. Beads were then spun down (5 min, 500 g) and washed three times with 500 µl lysis buffer. Beads were resuspended in 10 µl sample buffer. Beads and lysates were heated 3 min at 95°C, loaded on 4%-12% NuPAGE Novex Bis-Tris gels, and analyzed in parallel by Western blot. Membranes from immunoprecipitations done with rabbit α-GFP antibody were first probed with mouse α-myc antibody. Corresponding beads were then probed with mouse α-GFP antibody while lysates were probed with rabbit α-GFP antibody. Membranes from immunoprecipitations done with mouse α-myc antibody were first probed with rabbit α-GFP antibody. Corresponding beads were then probed with rabbit α-myc antibody while lysates were probed with mouse α-myc antibody. For coimmunoprecipitation of uGRAF2 with rabbit α-GFP, uGRAF2 was detected in the beads with rabbit α-GRAF2 and HRP-Protein A. In general, the membranes corresponding to the beads were chosen to show the level of expression of the baits, but when the background was too strong, the lysates were shown. For immunoprecipitations using α-myc-coated beads, a similar procedure was followed, but 100-mm dishes were used. Cells were resuspended in 200 µl lysis buffer, 15 µl of each lysate was loaded per well, and 5 µl of α-myc-coated magnetic beads (Cat# 5698, RRID:AB_10707161; Cell Signaling Technology) were used per sample. In all cases, results displayed are representative of a minimum of three independent experiments.
For quantification of coimmunoprecipitations, band intensities in the beads and in the lysates were measured in triplicate using ImageJ software; background was removed. Within each experiment, the intensity of the prey found in the beads was normalized by the value obtained for the corresponding lysate and expressed as a ratio to the chosen reference. The reference sample and the numbers of independent experiments used for the quantifications are mentioned in each figure legend.
For transfection, cells were detached with accutase at 37°C, spun down at low speed, washed in PBS, and microporated (1500 V, 3 pulses, 10 ms) using the Neon Transfection system (Thermo Fisher Scientific). Cells were plated on 35-mm dishes coated with Poly-L-Lysine and Collagen from calf skin. 16 h later, cells were serum starved in DMEM-GlutaMAX containing 0.1% BSA for 2 h and, where indicated, stimulated with insulin (1 µg/ml, 30 min). Cells were delicately scraped in cold PBS and fixed in 3.2% PFA (15 min, on ice). Cells were spun down and blocked in 5% NGS in PBS at RT under gentle shaking. After 30 min, cells were pelleted (5 min, 500 g) and resuspended in 1% NGS in PBS containing DyLight 650 α-HA antibody (5 µg/ml) or unconjugated α-HA (1 µg/ml) and incubated for 90 min at RT. Cells were spun down, washed three times in 1% NGS in PBS, and resuspended in Alexa Fluor 647-conjugated α-mouse antibody in 1% NGS in PBS. Cells were incubated for another hour, washed three times, fixed once more in 3.2% PFA, and resuspended in PBS. Cells were analyzed by FACS using an LSRFortessa analyzer (BD Biosciences), and results were analyzed with FlowLogic 7.2.1 (Inivai Technologies Pty. Ltd.). After comparison with undifferentiated cells, differentiated cells were selected based on their light-scattering properties. GFP or GFP/RFP-positive cells were selected. Background was subtracted from the GFP and Alexa Fluor 647 fluorescence intensities. The ratio of the geometric mean of the Alexa Fluor 647 fluorescence intensity to the geometric mean of the GFP fluorescence intensity was calculated and expressed as a ratio to the value obtained for the corresponding insulin-treated control. This was used as a measurement of HA-GLUT4-GFP export efficiency. For all cotransfections of nonfluorescent proteins, coexpression of myc-tagged proteins or of MICAL1 was verified in parallel following a similar protocol but using buffers containing 0.1% saponin and α-myc (2 µg/ml) or α-MICAL1 (1.9 µg/ml). 50-1,000 differentiated cells were analyzed in each independent experiment.
Integrin-β1 internalization shRNA-expressing HeLa cells (96 h after transfection) were incubated for 1 h with Alexa Fluor 647-conjugated α-Integrin-β1 antibody (1 µg/ml) at 37°C. Cells were then washed in PBS, acid stripped (0.5% acetic acid, 0.5 M NaCl) for 45 s (Reineke et al., 2015), and either directly scraped and fixed in 3.2% PFA (uptake) or chased for 4 h at 37°C in prewarmed medium before being washed and stripped once more, scraped, and fixed (chase). Cells were pelleted (5 min, 500 g) and incubated in blocking solution (5% NGS in PBS) for 30 min at RT. All further incubations and washes were done in PBS with 1% NGS. Cells were pelleted and incubated with unconjugated α-Integrin-β1 antibody (5 µg/ml) for 90 min and washed three times. They were then incubated with Alexa Fluor 546-conjugated α-mouse antibody (2 µg/ml) for 1 h, washed three times, and fixed once more in 3.2% PFA before being resuspended in PBS. Cells were analyzed by FACS using an LSRFortessa analyzer, and results were analyzed with FlowLogic 7.2.1. Within each experiment, the geometric means of the Alexa Fluor 647 and the Alexa Fluor 546 intensities were normalized by the values obtained for the shControl uptake sample. For each shRNA, the ratio of the normalized intensity of internalized Integrin-β1 (Alexa Fluor 647) to the one of surface Integrin-β1 (Alexa Fluor 546) was calculated as a measurement of the internalized Integrin-β1 ratio. Any defect in degradation of the internalized Integrin-β1 antibody or in the recycling of the receptor should lead to a higher internalized Integrin-β1 ratio. Within each experiment, between 1,000 and 30,000 cells were analyzed in each sample.

Protein export by FACS analysis
This assay was used to quantify the export of E-cadherin-GFP, E-cadherin-RFP, HA-α 2B -AR-GFP, HA-α 2B -AR-RFP, GFP-GPI, SBP-GFP-E-cadherin, SBP-GFP-GPI, E-cadherin-SBP-GFP, and VSVG-SBP-GFP in HeLa cells, of MMP14-GFP in HT 1080 cells, and of GFP-Extope CFTR and GFP-Extope CFTR ΔF508 in 293T cells. HeLa and HT 1080 cells were grown in 35-mm dishes. Unless otherwise specified in the figure legend, they were delicately scraped in 1 ml culture medium 16 h after transfection. 293T cells were grown in 12-well dishes and collected 36 h after transfection by pipetting. In all cases, cells were fixed in PFA (3.2%, 20 min, 37°C). Following fixation, all further incubations and centrifugations were done at RT. Cells were spun down (5 min, 500 g), resuspended in blocking solution (5% NGS in PBS), and incubated 30 min. Cells were then pelleted and resuspended in 50 µl of buffer (1% NGS in PBS) containing the appropriate primary antibody (α-E-cadherin: 8 µg/ml; α-MMP14: 4 µg/ml; Alexa Fluor 647 α-GFP: 2 µg/ml; rabbit α-GFP: 2.5 µg/ml; α-HA: 1 µg/ml; and α-VSVG: 5 µg/ml) and incubated for 90 min under gentle rocking. Cells were washed three times in 250 µl wash buffer (0.1% BSA in PBS). When unlabeled primary antibodies were used, cells were then incubated in a dilution of Alexa Fluor 647-labeled secondary antibody (2 µg/ml) in buffer (1% NGS in PBS) and incubated for 1 h under gentle rocking; cells were then washed three times in 250 µl wash buffer. After the final washes, cells were fixed once more in 3.2% PFA in PBS (20 min), washed, and resuspended in PBS. Cells were analyzed by FACS using an LSRFortessa analyser, and results were analyzed with FlowLogic 7.2.1. For single transfections, GFP-positive or RFPpositive cells were selected; for cotransfections, GFP-and RFPpositive cells were selected. Background was subtracted from the GFP, RFP, and Alexa Fluor 647 fluorescence intensities. The ratio of the geometric mean of the Alexa Fluor 647 fluorescence intensity to the geometric mean of the GFP or RFP (depending on the tag of the export candidate) fluorescence intensity was calculated and expressed as a ratio to the value obtained for the control. This was used as a measurement of protein export efficiency. Within each experiment, between 1,000 and 20,000 cells were analyzed in each sample.

Biotinylation of surface proteins
Cells were delicately washed three times in cold PBS. Fresh EZ-Link Sulfo-NHS-LC-Biotin (Thermo Fisher Scientific; 0.170 mg/ ml in biotinylation buffer [10 mM triethanolamine, pH 9, 140 mM NaCl]) was added to each well, and cells were gently rocked for 30 min at 4°C. Reaction was stopped by adding 1 ml of 100 mM glycine + 0.1% BSA in PBS. Cells were scraped, pelleted, washed delicately in Tris buffer (50 mM Tris, pH 7.4, 140 mM NaCl), and lysed (100 µl of 50 mM Tris, pH 7.4, 140 mM NaCl, 2 mM EDTA, 1% IGEPAL CA-630, 0.25% deoxycholate, and cOmplete Protease Iinhibitor Cocktail) for 30 min on ice. Lysates were cleared (5 min, 14,000 g), and 5 µl of each supernatant was removed (lysate). 10 µl of a 50% suspension of Neutravidin agarose beads (Thermo Fisher Scientific) was added to each sample, and they were then incubated 30 min at 4°C under rotation. Beads were pelleted (5 min, 500 g), washed three times in lysis buffer, and resuspended in sample buffer. Lysates and beads were heated 3 min at 95°C and analyzed by Western blot. For quantification of protein export efficiency, band intensities in the beads and in the lysates were measured in triplicate using ImageJ software; background was removed. Within each experiment, the intensity of the band in the beads was normalized by the corresponding value in the lysates and expressed as a ratio to the chosen reference.
Statistical data analysis Statistical data analysis was performed using Prism version 8.0.0 for Mac, GraphPad Software. For all quantifications provided, the means and SEM are shown. Unless otherwise stated, statistical data analysis was performed with one-way ANOVA of unmatched data. Gaussian distribution was assumed. Homoscedasticity was tested using a Brown-Forsythe test. The number of independent experiments (n) is detailed in the figure legends. In all cases, significant adjusted P values are represented as *, P < 0.05; **, P < 0.01; ***, P < 0.001; and ****, P < 0.0001.
In Fig. 1 D and Fig. 3 F, results were analyzed together; for each Rab, Dunnett's multiple comparisons test with corresponding shControl-transfected cells was used. In Fig. 1 Fig. 3 E, results were analyzed together; for each Rab, Dunnett's T3 multiple comparisons test with corresponding shControltransfected cells was used. In Fig. 1, I and J, in each case and for each GRAF, Dunnett's T3 multiple comparisons test with Rab8a-transfected cells was used. In Fig. 2, C and D;Fig. S2,D and E;Fig. 3 A;Fig. 4 A;and Fig. 6,D and F), in each case, Dunnett's multiple comparisons test with the reference was used. In Fig. 2 I, Dunnett's multiple comparisons test between pairs of GRAF1b/2-expressing cells transfected with mutant and corresponding wild-type protein was used. In Fig. 2 M, Fig. 4 D,Fig. 5 F,and Fig. 7 G,results were analyzed together with Dunnett's multiple comparisons test using GFP-transfected cells as control; n = 24 for GFP-transfected cells. In Fig. 2 N, Fig. 4 E, and Fig. 6 M, results were analyzed together with Dunnett's multiple comparisons test using shControl-transfected cells as control; n = 24 for shControl-transfected cells. In Fig. 4 I, Dunnett's multiple comparisons test using DMSO-treated cells as control was used. In Fig. 4 J, Dunnett's T3 multiple comparisons test using DMSO-treated cells as control was used. In Fig. 4 M, Dunnett's T3 multiple comparisons test between corresponding Cytochalasin D and DMSO-treated samples was used. In Fig. 5, C, I and L, for each mutant, Tukey's multicomparisons tests between all combinations of transfected proteins were used. In Fig. 5 D, for testing the effects of mutants on insulin-induced GLUT4 translocation, analysis of paired insulin-treated samples with Dunnett's multiple comparisons tests using pCI or RFP-transfected cells as controls was used. For analyzing insulininduced GLUT4 translocation within each transfection, Sidak's multiple comparisons tests between paired untreated and insulin-treated samples were used. In Fig. 5 F, Sidak's multiple comparisons tests between corresponding control and Cytochalasin D-treated cells were used. In Fig. 7 A, Dunnett's multiple comparisons tests using methanol-treated cells as control were used. In Fig. 7 C and Fig. S7 C, for each protein, Dunnett's T3 multiple comparisons test with corresponding methanol-treated cells was used. In Fig. 7 I, Sidak's multiple comparisons tests between corresponding paired uptake and chase samples were used. In Fig. 8 A and Fig. 9 A, results were analyzed together. For each cargo, Dunnett's multiple comparisons test using GFP/RFPtransfected cells as control was used. In Fig. 8 C, for each cargo, Dunnett's multiple comparisons test using shControl-transfected cells as control was used. In Fig. 8 G, Tukey's multiple comparisons test between all combinations of transfections was used. In Fig. 8 I, for each cargo, Dunnett's multiple comparisons test using RFP-transfected cells as control was used. In Fig. 8 J, for each cargo, Dunnett's multiple comparisons test using shControl + RFP-transfected cells as control was used. To assess rescue, paired t tests between shGRAF2a cells cotransfected with RFP and RFP-GRAF2 res and between shWDR44a cells cotransfected with RFP and RFP-WDR44 res or RFP-WDR44 res ΔPro were used. In Fig. 9 B, for each cargo, Dunnett's multiple comparisons test with untreated cells as control was used. For GFP-Extope CFTR ΔF508, Sidak's multiple comparisons test between corresponding RFP and RFP-Rab10 T23N-transfected cells was used. In Fig. 9 C, in each case, Dunnett's multiple comparisons test of matched data with the corresponding control was used. In Fig. 9 E, Sidak's multiple comparisons tests between samples incubated with or without biotin was used. In Fig. S9 E, unpaired t test between untreated and BFA-treated GFP-transfected cells and Dunnett's multiple comparisons tests between BFA-treated transfected cells using BFA-treated GFP-transfected cells as control were used.

E and
Online supplemental material Fig. S1 shows that similar to GRAF2, GRAF1b colocalizes with Rab8a/b and Rab10 and that GRAF2 is the predominant GRAF of HeLa cells. Fig. S2 contains the complete results for the protein pull-downs by GRAF1 SH3 and GRAF2 SH3 and additional biochemical data on the interaction of GRAF1b/2 with MICAL1 and WDR44. Fig. S3 contains additional imaging of GRAF1b/2 together with MICAL1 and WDR44 proteins. In Fig. S4, the interactions of MICAL1 and WDR44 with Rabs and the response of WDR44, GRAF1b/2, and MICAL1 to cytoskeleton-perturbing drugs are further detailed. Fig. S5 contains supplementary data on the dominant negative mutants MICAL1 G3W, WDR44 ΔC, and GRAF1/2 BAR-PH. Fig. S6 contains additional data on the WDR44-mediated contacts of GRAF1b/2 with the ER proteins VAPA/B. Fig. S7 contains supplementary results detailing the response of endogenous WDR44 to BFA. Fig. S8 contains additional data on the interplay between GRAF/WDR44, other members of the MICAL family, and recycling endosomes. Results presented in Fig. S9 further support a role for GRAF2 and WDR44 in the export of E-cadherin but not Occludin; Western blots show that knocking down WDR44 and GRAF2 does not cause ER stress; and additional imaging of SBP-GFP-E-cadherin is presented. In Video 1, time-lapse movies of HeLa cells transfected with GRAF1b, GRAF2, WDR44, and MICAL1 are run in parallel. The following videos correspond to time-lapse imaging of binary protein combinations: GRAF2 and MICAL1 in Video 2; GRAF2 and WDR44 in Video 3; MICAL1 and WDR44 in Video 4; Rab10 and MICAL1 in Video 5; Rab8a and MICAL1 in Video 6; Rab10 and WDR44 in Video 7; Rab8a and WDR44 in Video 8; and GRAF2 BAR-PH and WDR44 in Video 9. Video 10 shows the BFAinduced response of TGN46 and WDR44. Proteins only identified in one pull-down are in gray (26/47 hits); those found both in GST-GRAF1 SH3 and in GST-GRAF2 SH3 pull-downs but only in one lysate are in black (13/47 hits); and those found both in rat brain and in HeLa cell lysates are in red (8/47 hits). Since GRAF1 and GRAF2 have a 67.9% identity and a 91.1% similarity in their SH3 domains, we focused our attention on these eight robust interacting partners. (C-E) Immunoprecipitation (IP) of transfected 293T cells with α-GFP (C and D) or α-myc-coated beads (E). (C) uGRAF2 was coimmunoprecipitated by GFP-tagged WDR44 and MICAL1. myc-tagged WDR44 and MICAL1 were coimmunoprecipitated by GFP-tagged GRAF1b and GRAF2. In the beads, GRAF2-GFP and GRAF1b-GFP overlapped with the remaining myc-MICAL1 band (*). (D and E) The efficiency of GRAF1b binding was quantified using coimmunoprecipitation with GFP-WDR44 (D) or myc-MICAL1 (E) as reference. n = 3-5. Data are means ± SEM; **, P < 0.01; ***, P < 0.001; and ****, P < 0.0001. Binding was abolished by removal (D) or mutation (E) of a proline-rich region. (D) Overexpressed WDR44 underwent proteolysis. (F) Pull-down test of GRAF2 SH3 by GST-WDR44 ΔC, GST-WDR44 ΔN2, GST-MICAL1, GST-MICAL1 ΔN, and GST-MICAL1 ΔMOΔCH. Beads were analyzed by electrophoresis and Coomassie staining showing direct binding of GRAF2 SH3 to WDR44 ΔC and to MICAL1 ΔMOΔCH. GRAF2 SH3 did not bind MICAL1 or MICAL1 ΔN. This suggests that the PPKPP motif of MICAL1 is not accessible in the isolated fulllength protein and that in order to be exposed, MICAL1 has to undergo a conformational change, such as the one induced by Rab-binding (Schmidt et al., 2008). (G) Western blot analysis of an equal amount of BSC1, hTERT-RPE1, 293T, HeLa, U-87 MG, and SH-SY5Y cell lysates loaded on two replicate gels. One membrane was probed with α-WDR44, the other with α-MICAL1. β-tubulin was used as loading control. Figure S7. WDR44 labels a subset of tubular endosomes. (A) Confocal stacks of hTERT-RPE1, 293T, HeLa, U-87 MG, HT 1080, and COS-7 cells incubated with BFA (5 µg/ml, 15 min) and stained with α-WDR44. All cell types showed an increase in endogenous WDR44 tubules (compare with Fig. S3 E). (B) Confocal images of transfected HeLa cells left untreated or after incubation with BFA (5 µg/ml, 15 min) and stained with α-WDR44. While there was no colocalization of endogenous WDR44 tubules with GFP-STX16 or GFP-VAMP3 under resting conditions, colocalization was seen after incubation with BFA. (C) Manders colocalization coefficients for GFP-STX16 and GFP-VAMP3 with endogenous WDR44 structures. n = 10-20 cells. Data are means ± SEM; **, P < 0.01; and ****, P < 0.0001. (D) Confocal images of untransfected HeLa cells incubated with BFA (5 µg/ml) for 5 or 15 min and stained with α-WDR44 and α-ERGIC53, α-GM130, or α-LAMP2. Endogenous WDR44 tubules did not colocalize with any of these proteins. (A, B, and D) Insets show magnifications of the boxed areas. Scale bars: 10 µm; scale bars of insets: 2 µm.