Segregation in the Golgi complex precedes export of endolysosomal proteins in distinct transport carriers

The endolysosomal system of eukaryotic cells comprises an array of membrane-enclosed organelles, including early, late, and recycling endosomes, as well as lysosomes. Transmembrane proteins that reside in these compartments (hereafter referred to as endolysosomal proteins) are synthesized in the ER and subsequently transported through the cis, medial, and trans cisternae of the Golgi stack and the TGN (collectively referred to as the “Golgi complex”; Braulke and Bonifacino, 2009). The proteins are eventually delivered to the endolysosomal system either directly from the Golgi complex (Harter and Mellman, 1992; Waguri et al., 2003; Ang et al., 2004) or indirectly after transport to the plasma membrane and endocytosis (Lippincott-Schwartz and Fambrough, 1986; Braun et al., 1989; Janvier and Bonifacino, 2005). Key determinants of sorting to the endolysosomal system are signals present in the cytosolic domains of the proteins (Bonifacino and Traub, 2003), which are recognized by adaptor proteins (APs) that are components of protein coats (Robinson, 2004). Signal–adaptor interactions promote incorporation of the proteins into coated transport carriers that participate in the delivery of proteins to the endolysosomal system. Despite progress in the characterization of these molecular mechanisms, however, many aspects of endolysosomal protein sorting in the context of the whole cell remain poorly understood. These aspects include the step in the biosynthetic pathway when endolysosomal proteins diverge from plasma membrane proteins, the extent to which specific endolysosomal proteins follow the direct or indirect pathways, the nature of the transport carriers involved in either pathway, the particular signal–adaptor interactions that mediate protein incorporation into these carriers, and the possible existence of other sorting determinants. Addressing these issues in intact, living cells has proved difficult because of limitations in the ability to visualize the transport of newly synthesized endolysosomal proteins with sufficient temporal and spatial resolution and without temperature or drug manipulations that perturb the structure and function of the Golgi complex.

In this study, we have taken advantage of recent methodological developments that allow synchronization of protein transport through the biosynthetic pathway, as well as live-cell and superresolution imaging, to examine how endolysosomal proteins are sorted in the Golgi complex. Specifically, we have used the “retention using selective hooks” (RUSH) system (Boncompain et al., 2012) to track the biosynthetic transport of three transmembrane proteins with different steady-state distributions in the endolysosomal system: the cation-dependent mannose-6-phosphate receptor (CD-MPR; localized to the TGN and early/late endosomes), the transferrin receptor (TfR; plasma membrane, early endosomes, and recycling endosomes), and lysosomal-associated membrane protein (LAMP) 1 (late endosomes and lysosomes). Our analyses reveal an unexpected level of complexity in the mechanisms of endolysosomal protein sorting at the Golgi complex. We find that CD-MPR undergoes an early segregation from TfR and LAMP1 in the Golgi complex, well before their export in transport carriers. This segregation is independent of signals in the cytosolic tails but dependent on the transmembrane and luminal domains of the proteins. The CD-MPR subsequently leaves the Golgi in a population of predominantly vesicular transport carriers in a manner dependent on a cytosolic dileucine-based signal that interacts with clathrin-associated GGA adaptors. These carriers do not translocate toward the plasma membrane but directly deliver the CD-MPR to endosomes. The TfR and LAMP1, on the other hand, are exported in a population of predominantly tubular carriers destined for the plasma membrane, independently of cytosolic sorting signals and their cognate adaptors. The sorting signals in TfR and LAMP1 and the clathrin-associated AP-2 complex are, however, required for endocytosis of TfR and LAMP1 as a requisite for their eventual delivery to endosomes and lysosomes, respectively. These findings demonstrate that early segregation of different sets of endolysosomal proteins in the Golgi complex precedes their export in two distinct populations of transport carriers involved in the direct and indirect pathways. Our study also highlights distinct requirements for signal–adaptor interactions in the exit of different endolysosomal proteins from the Golgi complex.

Transport of newly synthesized TfR, LAMP1, and CD-MPR through the biosynthetic pathway was analyzed using the RUSH system (Boncompain et al., 2012). The endolysosomal proteins (i.e., “reporter” proteins) were genetically fused to a streptavidin-binding peptide (SBP) and a fluorescent protein (GFP or mCherry) and coexpressed with streptavidin fused to the ER-retrieval signal KDEL (Munro and Pelham, 1987; i.e., “hook” protein; Fig. 1 a). For simplicity, the reporter proteins will hereafter be referred to as TfR, LAMP1, and CD-MPR, with the implicit understanding that they are modified for use in the RUSH system. As expected, coexpression of the reporter proteins with the hook proteins resulted in their accumulation in the ER (Fig. 1 b, 0 min). Addition of the vitamin biotin competed off the SBP–streptavidin interaction, resulting in synchronous release of the proteins from the ER (Fig. 1 b and Video 1), and their eventual transport to their corresponding locations in the endolysosomal system (Fig. S1 a). Coexpression of combinations of the reporter proteins showed that they all exited the ER in the same transport carriers (Fig. S1 b) and arrived simultaneously in the Golgi complex at 15–25 min after the addition of biotin (Fig. 1, b and c). At 20–35 min, the proteins began to exit the Golgi complex in pleiomorphic transport carriers similar to those previously shown to mediate various post-Golgi transport events (Hirschberg et al., 1998; Polishchuk et al., 2000, 2003, 2006; Puertollano et al., 2003). Interestingly, we noticed the existence of two distinct populations of carriers: predominantly tubular carriers containing both TfR and LAMP1 but not CD-MPR (Fig. 2, a, b, d, and e; and Video 2), and predominantly vesicular carriers containing CD-MPR but not TfR (Fig. 2 c and Video 3; best observed with higher time resolution in Video 4). For simplicity, we refer to these carriers as “tubular” and “vesicular,” respectively, notwithstanding that they display substantial variation in shape and size. Further analyses of the tubular carriers showed that they were enriched in the vesicular stomatitis virus glycoprotein (VSV-G; Fig. 2 f), a marker of transport carriers destined for the plasma membrane (Hirschberg et al., 1998; Polishchuk et al., 2006; Micaroni et al., 2013). In contrast, they were devoid of internalized transferrin (Tf; Fig. S2, a and b), indicating that they were not derived from recycling endosomes. In addition, total internal reflection fluorescence (TIRF) microscopy showed direct fusion of carriers containing TfR (Fig. S3 a and Video 5) but not internalized Tf (Fig. S3 b and Video 6), with the plasma membrane. Together, these observations indicated that TfR/LAMP1 tubular carriers bud from the Golgi and directly fuse to the plasma membrane without passing through recycling endosomes.

Airyscan superresolution imaging of cells fixed 30 min after the addition of biotin confirmed the presence of TfR and LAMP1 in tubules budding from the Golgi complex (Fig. 3, a and b) and CD-MPR in a distinct population of vesicles (Fig. 3, d and e). Surprisingly, whereas TfR and LAMP1 colocalized throughout the entire Golgi structure (Fig. 3, a and c; Pearson’s coefficient = 0.95, similar to that of the same reporter protein tagged with different fluorescent proteins; Fig. S4, a and b), TfR and CD-MPR were largely segregated to different Golgi domains (Pearson’s coefficient = 0.37; Fig. 3, d and f). Live-cell Airyscan imaging after the addition of biotin showed that TfR and LAMP1 continuously colocalized from their entry into the Golgi complex to their exit in tubular carriers (Fig. 3, g and i; and Video 7). In contrast, TfR and CD-MPR started to segregate shortly after their entry into the Golgi complex, and their segregation increased over time (Fig. 3, h and i; and Video 8). Similar Golgi segregation and budding patterns were observed in a different cell line, U2OS (Fig. S4, c and d). These observations thus revealed that segregation of the CD-MPR from other endolysosomal proteins begins in the Golgi complex, before their incorporation into distinct transport carriers. The sorting receptor sortilin behaved similarly to the CD-MPR (Fig. S4, e and f), demonstrating that it belongs to the same subset of endolysosomal cargos as the CD-MPR.

Sorting of TfR and LAMP1 to endosomes and lysosomes, respectively, is dependent on sorting signals fitting the YXXØ motif (where X is any amino acid and Ø a bulky hydrophobic amino acid) in the cytosolic tail of the proteins (Traub and Bonifacino, 2013; Fig. 4 a). In the case of TfR, an additional signal comprising the sequence GDNS (Fig. 4 a) contributes to sorting to the basolateral plasma membrane in polarized epithelial cells (Odorizzi and Trowbridge, 1997). Strikingly, mutation of the YXXØ and GDNS signals in TfR (residues Y20, G31, D32, N33, and S34 to alanines) and the YXXØ signal in LAMP1 (residue Y404 to alanine; Fig. 4 a) did not prevent incorporation of these proteins into the Golgi-derived tubular carriers, although it resulted in subsequent accumulation of the proteins at the plasma membrane (Fig. 4, b and c; and Video 9). YXXØ motifs are recognized by the AP complexes AP-1, AP-2, AP-3, and AP-4 (Fig. 5 a; Traub and Bonifacino, 2013) and the GDNS motif by AP-1 (Gravotta et al., 2012). CRISPR/Cas9 knockout (KO) of subunits of these complexes (Fig. 5 b) had no effect on exit of TfR and LAMP1 from the Golgi complex in tubular carriers (Fig. 5 c). KO of the AP-2 μ2 subunit, however, caused accumulation of TfR and LAMP1 at the plasma membrane (Fig. 5 d and Fig. S5, a–c), as previously shown by siRNA knockdown (Motley et al., 2003; Janvier and Bonifacino, 2005). KO of subunits of the other complexes did not prevent transport of LAMP1 to lysosomes (Fig. 5 d). These experiments indicated that interactions of cytosolic sorting signals with AP complexes are dispensable for export of TfR and LAMP1 from the Golgi complex in tubular carriers, but an interaction with AP-2 is subsequently required for endocytic delivery of the proteins to endosomes and lysosomes.

In contrast to TfR and LAMP1, CD-MPR has been shown to exit the Golgi by virtue of the interaction of a DXXLL motif in the cytosolic tail of the receptor (Fig. 4 a) with the monomeric clathrin adaptors GGA1, GGA2, and GGA3 (Puertollano et al., 2001, 2003; Zhu et al., 2001). Indeed, we observed that CD-MPR-containing vesicles budding from the Golgi were decorated with GGA1 (Fig. 4 d). Moreover, mutation of the DXXLL signal (residues L274 and L275 to alanine; Fig. 4 a) prevented exit of CD-MPR in the GGA1-coated carriers (Fig. 4 e). Hence, unlike Golgi export of TfR and LAMP1 in tubular carriers, export of CD-MPR in vesicular carriers depends on a specific signal–adaptor interaction. It is worth noting that the DXXLL mutant of the CD-MPR was not diverted to tubules by default but was retained in the Golgi complex. We also observed that mutation of the cytosolic sorting signals in both the TfR and CD-MPR did not prevent their segregation into different Golgi subdomains (Fig. 4 f), indicating that this phenomenon is independent of interactions with APs.

What then are the determinants of segregation in the Golgi complex and exit into tubules? To address this question, we constructed chimeric proteins having different combinations of the luminal, transmembrane, and cytosolic domains of LAMP1 and CD-MPR (both type I transmembrane proteins; Fig. 6 a) and compared their transport with that of the TfR (a type II transmembrane protein) using the RUSH system (Fig. 6, b–h). We observed that LAMP1 chimeras having the transmembrane and/or luminal domains of the CD-MPR (termed MML, LML, and MLL) were not incorporated into Golgi-derived tubules (Fig. 6 b) and were segregated from the TfR in the Golgi complex (Fig. 6, e, g, and h). In contrast, a chimera having the transmembrane and luminal domains of LAMP1 and the cytosolic tail of CD-MPR (LLM) was exported into tubules (Fig. 6 b) and colocalized with TfR in the Golgi complex (Fig. 6 f). These results indicated that transmembrane and/or luminal domains determine protein segregation within the Golgi complex that precedes exit into distinct transport carriers.

The TGN has been classically regarded as the Golgi subcompartment where newly synthesized proteins destined for secretory vesicles, the plasma membrane, endosomes, and lysosomes are sorted into distinct populations of transport carriers (Griffiths and Simons, 1986). In this classical view, all cargo proteins remain mixed throughout the Golgi complex until they are packaged into their corresponding transport carriers. Several findings, however, challenge the notion that all cargo sorting in the Golgi complex occurs at the TGN. First, transmembrane cargos were shown to partition from Golgi enzymes during their transport through the Golgi cisternae (Patterson et al., 2008). In addition, the contents of different types of secretory granule were found to segregate from one another as early as in the cis-Golgi cisternae (Clermont et al., 1992). Other studies showed that proteoglycans aimed for the apical or basolateral surfaces of polarized epithelial cells acquire different carbohydrate modifications in the Golgi complex (Tveit et al., 2005; Vuong et al., 2006), and high-mannose forms of basolateral and apical proteins exhibit different detergent solubility, all suggestive of segregation in the early Golgi complex (Alfalah et al., 2005). Using superresolution and live-cell imaging, we now provide direct evidence that two sets of transmembrane proteins destined for the endolysosomal system undergo progressive segregation into distinct Golgi domains before their export into distinct transport carriers. These findings indicate that cargo sorting can occur early in the Golgi complex, even for proteins that are targeted to the same organellar system. The Golgi domains to which proteins are segregated could correspond to the center or the rims of the same cisternae or to different cisternae. They could also be discrete ministacks laterally connected as part of a larger Golgi ribbon (Yano et al., 2005; Puthenveedu et al., 2006). Ultrastructural methods in combination with the RUSH system will be required to determine the exact identity of these domains.

Early segregation likely determines the sites of cargo export from the Golgi complex. For proteins that require interaction with clathrin adaptors (e.g., CD-MPR, sortilin; Nielsen et al., 2001; Puertollano et al., 2001; Zhu et al., 2001), the early segregation domain must be connected to the TGN, which is where clathrin coats are located in the Golgi complex (Klumperman et al., 1993). Proteins that exit independently of clathrin adaptors (e.g., TfR, LAMP1, and plasma membrane proteins; Pols et al., 2013), on the other hand, could be exported from a nonclathrin TGN domain or an earlier cisterna. In support of this latter possibility, electron tomography studies showed the presence of a still unidentified nonclathrin, “lace-like” coat at the rims of Golgi cisternae proximal to, but distinct from, the TGN (Ladinsky et al., 1994). These sites were proposed to mediate protein export to the plasma membrane (Ladinsky et al., 1994), although evidence for this function remains to be obtained. Furthermore, fluorescence microscopy of nocodazole-fragmented Golgi ministacks suggested that plasma membrane–directed cargos such as VSV-G exit from the Golgi stack before reaching the TGN (Tie et al., 2016). Thus, early cargo segregation in the Golgi stack likely determines export from different Golgi subcompartments.

Our studies also show that endolysosomal proteins leave the Golgi complex in two types of transport carrier. Both types have variable sizes and shapes, although they tend to be predominantly tubular or vesicular in appearance. The characteristics of these carriers likely reflect the Golgi compartments from which they arise. The carriers containing TfR and LAMP1 probably correspond to the VSV-G Golgi–to–plasma membrane carriers previously characterized by correlative light-electron microscopy (Polishchuk et al., 2000). These carriers were found to be large (up to 1.7 µm in diameter), tubular-saccular, and devoid of protein coats (Polishchuk et al., 2000), suggestive of their origin in noncoated areas of the Golgi complex. As in our study, they were shown to fuse with the plasma membrane en bloc, without intersecting any other compartment along the way (Hirschberg et al., 1998; Polishchuk et al., 2000). Other immunoelectron studies also demonstrated export of LAMP1 from the TGN in noncoated carriers devoid of cation-independent mannose-6-phosphate receptor and AP-1 but containing VPS41 and VAMP7 (Pols et al., 2013). The latter carriers may correspond to a population that follows the direct pathway and that are less than the level of detection in our assays. Correlative light-electron microscopy of CD-MPR- and GGA1-containing carriers also showed them to be large, convoluted tubular-vesicular structures. However, they have associated clathrin-coated profiles, indicative of their origin at the TGN (Polishchuk et al., 2006). Previous studies showed that, unlike the VSV-G/TfR/LAMP-1 carriers, the CD-MPR/GGA1 carriers did not translocate toward the plasma membrane but merged with endosomes (Puertollano et al., 2003; Waguri et al., 2003; Polishchuk et al., 2006). These properties of the TfR/LAMP1 and CD-MPR carriers are consistent with their being the mediators of transport in the indirect and direct pathways, respectively, to the endolysosomal system. The use of the direct pathway by the CD-MPR fits in with its role as an intracellular sorting receptor for lysosomal hydrolase precursors.

Molecular dissection of the endolysosomal proteins used in our study revealed that they have different types of sorting determinant. The initial segregation in the Golgi stack is independent of sorting signals in the cytosolic tails but dependent on the transmembrane and/or luminal domains of the proteins. Transmembrane domains could mediate partitioning into specific lipid domains or interactions with other transmembrane proteins, as shown for other sorting events (Nishikawa and Nakano, 1993; Emery et al., 2003; Alfalah et al., 2005; Patterson et al., 2008; Kaiser et al., 2011). Luminal domains could segregate proteins by promoting oligomerization or aggregation in the special environment of the Golgi complex, as also shown in other settings (Compton et al., 1989; Dintzis et al., 1994; Colomer et al., 1996; Wolins et al., 1997; Paladino et al., 2004). In this regard, it is worth noting that some constitutive secretory cargos, such as the cartilage oligomeric matrix protein and lysozyme C, bind in a Ca²⁺-dependent manner to the Golgi protein Cab45, which facilitates their export into a specific population of secretory carriers (Crevenna et al., 2016). It is conceivable that a similar mechanism could operate for segregation of transmembrane proteins through their luminal domains. The subsequent packaging of proteins into Golgi export carriers has long been thought to depend on interactions of cytosolic sorting signals with TGN-associated adaptors such as the GGAs, AP-1, and AP-4. However, for the proteins examined in our study, this appears to be true only for the CD-MPR and sortilin, which require a GGA-binding signal for exit into vesicular carriers. In contrast, exit of TfR and LAMP1 in tubular carriers is independent of sorting signals and of the AP-1, AP-2, AP-3, and AP-4 adaptors. This is in line with the carriers’ being the same that transport plasma membrane proteins. AP-2 is subsequently required for endocytosis of TfR and LAMP1 from the plasma membrane, as previously shown by RNAi studies (Motley et al., 2003; Janvier and Bonifacino, 2005). The fact that AP-2 is the only YXXØ-interacting adaptor required for sorting of TfR and LAMP1 to endosomes and lysosomes demonstrates the critical role of endocytosis in this process. Collectively, these observations lend further support to the notion that CD-MPR follows mainly the direct pathway, and TfR and LAMP1 the indirect pathway, for transport to endosomes and lysosomes.

Although the use of the RUSH system has allowed us to track the biosynthetic transport of endolysosomal transmembrane proteins in unprecedented detail, there are several caveats in the interpretation of our experiments. The most important one is that the expression levels of the reporter proteins are likely higher than those of their endogenous counterparts. The mechanics of RUSH could generate a wave of newly synthesized reporter proteins moving through the secretory pathway, potentially creating abnormal structures or altering the properties of the organelles along the way. Overexpression of the reporter proteins could also saturate sorting dependent on signals and adaptors (Marks et al., 1996). To avoid these problems, in our study we imaged cells expressing moderate levels of the reporter proteins: high enough for detection of transport intermediates but not so high that they changed the appearance of the organelles. In this regard, electron microscopy of cells expressing reporter proteins 25 min after their release from the ER showed normal appearance of the Golgi complex (unpublished data). In addition, saturation of sorting mechanisms would have been expected to homogenize the distribution of different reporters among Golgi domains, transport carriers, and destination organelles, but this was clearly not the case in our studies. These considerations notwithstanding, we cannot rule out the existence of alternative processes, such as populations of CD-MPR following the indirect pathway and TfR and LAMP1 following the direct pathway to some extent.

Our results suggest a two-step process for the sorting of endolysosomal proteins in the Golgi complex (Fig. 7). In the first step, two sets of proteins become segregated to different domains of the Golgi stack by virtue of transmembrane and/or luminal domains. In the second step, proteins segregated to one domain (e.g., CD-MPR) exit the Golgi complex in vesicular carriers bound for the endolysosomal system, in a process that is dependent on recognition of cytosolic sorting signals by clathrin adaptors (i.e., the GGAs). Proteins in the other domain, in contrast, leave the Golgi complex in tubular carriers directed to the plasma membrane, independently of sorting signals and clathrin adaptors. This model differs from the classical model in that different sets of endolysosomal proteins are presorted in the early Golgi and that some of those proteins leave the Golgi complex independently of signal–adaptor interactions.

Plasmid constructs to synchronize the traffic of the TfR, LAMP1, CD-MPR, VSV-G, sortilin, and LAMP1–CD-MPR chimeric reporter proteins (Figs. 1 a, 4 a, and 6 a) through the secretory pathway were generated by replacing sequences encoding the reporter proteins in the original bicistronic RUSH constructs (gift of F. Perez and G. Boncompain, Curie Institute, Paris, France; Boncompain et al., 2012) in which streptavidin KDEL was used as the hook. The type I transmembrane proteins LAMP1 (UniProt accession no. P14562), CD-MPR (UniProt accession no. P20645), sortilin (UniProt accession no. Q99523), and VSV-G (UniProt accession no. P04882) were modified by insertion of the SBP and a fluorescent protein (EGFP or mCherry) at their luminal N termini, immediately after the signal peptide. The type II transmembrane protein TfR (UniProt accession no. P02786) was tagged with SBP and a fluorescent protein at its luminal C terminus. Tagging in the luminal domain avoided potential interference with cytosolic sorting signals. TfR-Y20A/GDNS31-34AAAA, LAMP1-Y404A, and CD-MPR-L274A/L275A mutants were generated using the QuickChange Site-Directed Mutagenesis kit (Agilent Technologies). Chimeras combining LAMP1 and CD-MPR domains (Fig. 6 a) were generated in the bicistronic RUSH construct using Gibson assembly. LAMP1: luminal, amino acids 22–371; transmembrane, amino acids 372–395; cytosolic, amino acids 396–407. CD-MPR: luminal, amino acids 27–185; transmembrane, amino acids 186–210; cytosolic, amino acids 211–277. EGFP-tagged Rab4a and GGA1 were described previously (Puertollano et al., 2001; Chen et al., 2012).

Mouse monoclonal antibodies to AP-1 γ (1:5,000 for immunoblotting, anti–Adaptin γ; catalog no. 610385), AP-2 μ2 (1:5,000 for immunoblotting, anti-AP50; catalog no. 611351), AP-3 δ (1:5,000 for immunoblotting, anti–Adaptin δ; catalog no. 611329), AP-4 ε (1:5,000 for immunoblotting, anti–Adaptin ε; catalog no. 612018), and actin (1:5,000 for Western blotting, anti-actin; catalog no. 612657) were purchased from BD Biosciences. HRP-conjugated goat anti–mouse antibody (1:5,000 for immunoblotting; catalog no. sc-2004) was purchased from Santa Cruz Biotechnology. Mouse monoclonal antibody to cation-independent mannose-6-phosphate receptor (1:100 for immunofluorescence microscopy, MEM-238; ab8093) was purchased from Abcam. Rabbit monoclonal antibody to LAMTOR4 (1:1,000 for immunofluorescence microscopy; catalog no. 13140) was purchased from Cell Signaling Technology. Alexa-conjugated secondary antibodies including Alexa Fluor 488–conjugated donkey anti–rabbit antibody (1:1,000 for immunofluorescence microscopy; catalog no. A21206) and Alexa Fluor 555–conjugated donkey anti–mouse antibody (1:1,000 for immunofluorescence microscopy; catalog no. A31570) antibodies were purchased from Invitrogen.

For fixed-cell imaging experiments, 40,000 HeLa cells per well were seeded on a 12-well plate containing 18-mm cover glasses (Marienfeld) coated with fibronectin 1 d before transfection. For live-cell imaging experiments, 40,000 HeLa cells per well were seeded on 2-well Nunc Lab-Tek chambers coated with fibronectin 1 d before transfection. Cells were transfected using FuGENE 6 (E2691; Promega). 6 μl FuGENE transfection reagent was diluted into 80 µl Opti-MEM (31985-070; GIBCO BRL), and, separately, 2 µg DNA was diluted into 20 µl Opti-MEM. For cotransfections, DNA plasmids were combined in an equimolar ratio. After 5 min, the DNA and transfection solutions were mixed and incubated for 20 min before being added to the cells. 20 h after transfection with the indicated plasmids, cells were imaged in 37°C prewarmed phenol red–free medium (20163-029; GIBCO BRL), supplemented with 25 mM Hepes. D-biotin (Sigma-Aldrich) at a final concentration of 40 µM was added to the chamber at time 0.

Immunofluorescence microscopy was performed as previously described (Schindler et al., 2015). In brief, HeLa cells were fixed for 30 min at RT in 4% PFA, 4% sucrose, 0.1 mM CaCl₂, and 1 mM MgCl₂ in PBS. Cells were then permeabilized for 10 min at RT with 0.2% saponin in PBS, followed by incubation with the indicated antibodies. Live-cell imaging was conducted with an Eclipse Ti Microscope System (Nikon) equipped with an environmental chamber (temperature controlled at 37°C and CO₂ at 5%) and NIS-Elements AR microscope imaging software. Spinning-disk confocal images were taken with a Plan Apo VC 60× objective (NA 1.40) and a high-speed electron-multiplying charge-coupled device camera (Evolve 512; Photometrics) mounted on the left portal. TIRF microscopic images were taken with an Apo TIRF 100× Oil DIC N2 objective (NA 1.49) and an electron-multiplying charge-coupled device camera (DU-897; Andor) mounted on the right portal. TIRF position was calibrated for each imaging experiment, and the focus was maintained using a Perfect Focus system. Dual-color imaging was done by fast switching of the excitation lasers, and images from green and red channels were aligned automatically. For TIRF imaging of Alexa Fluor 488–conjugated Tf, transfected cells were incubated in DMEM without FBS, with 1% BSA, for 30 min before being incubated for 1 h in Alexa 488–Tf (working concentration 25 µg/ml; T13342; Thermo Fisher Scientific) at 37°C. RUSH was performed under the constant presence of Alexa Fluor 488–Tf after addition of biotin (which doubled the volume of media, effectively halving the concentration of Tf for the duration of the imaging). Superresolution microscopic images were taken using an LSM 880 microscope with Airyscan (Zeiss) and a Plan Apochromat 63× objective (NA 1.40) with the settings recommended by the manufacturer.

HeLa-KO cell lines were generated using the CRISPR/Cas9 system (Ran et al., 2013). Target gRNA sequences (AP-1 γ1: 5′-TACATACCGATGTCGGAATG-3′; AP-2 μ2: 5′-CGATGTCATCTCGGTAGACT-3′; AP-3 δ: 5′-CCTTGTGGTTACGGATGCCG-3′; AP-4 ε: 5′-GCAATCAAGTTAGCCCAACA-3′) were cloned into the px330 CRISPR/Cas9 vector using the restriction enzyme BbsI. CRISPR/Cas9 constructs were transfected into HeLa cells, and cell lines derived from single colonies were validated by immunoblotting to confirm the loss of the target proteins. For AP-2 μ2, small amounts of the target protein were found in the validation screening, perhaps because of the lethality of complete AP-2 μ2 KO. Rapid accumulation of TfR–SBP–mCherry and SBP–mCherry–LAMP1 on the surface of AP-2 μ2 KO cells was considered confirmation of effective abrogation of AP-2 function (Fig. 5 d and Fig. S5 a).

HeLa cells were plated onto six-well plates and transfected with plasmids encoding SBP–GFP–CD-MPR or SBP–GFP–LAMP1 using FuGENE 6. 20 h after transfection, cells were incubated with biotin for 0 and 60 min. All further manipulations were on ice or at 4°C. Cells were detached from the plate by incubating with 10 mM EDTA in PBS for 20 min, pipetting up and down every 5 min. Cells were transferred to 1.5-ml microcentrifuge tubes followed by fixation in PBS containing 4% PFA for 20 min. Cells were washed four times by repeated centrifugation (4°C, 500 g, 5 min) in PBS to remove residual PFA. Fixed cells were stained with anti-GFP conjugated with Alexa Fluor 647 (565197; BD Biosciences) at a concentration of 2 µg/ml in PBS containing 3% BSA. Cells were filtered using Cell-Strainer-capped 5-ml round-bottom tubes (352235; Corning). A minimum of 50,000 cells per sample was analyzed using an LSRFortessa cell analyzer (BD Biosciences), gating for GFP-positive cells indicative of expression of the transgene. Data were analyzed using FlowJo software. Surface expression of protein was deduced by relative intensity of Alexa Fluor 647.

All numerical results are reported as the mean ± SEM and represent data from a minimum of three independent experiments. Line plots were performed in ImageJ. For Fig. 3 (g–i), the first observable robust tubulation from the Golgi was considered time 0 to normalize time points. For live-cell imaging, Imaris was used to calculate the Pearson’s correlation of the voxels from the whole image in a z-stack. Images were thresholded at 0.05% of total intensity to reduce background. For each sample, at least seven cells were analyzed per sample. Data were normalized to time point minus 15 min in each data set. A two-tailed Student t test for unpaired data were used to evaluate single comparisons between different experimental groups using Microsoft Excel.

For the kinetic analysis of RUSH, SBP–mCherry–LAMP1, SBP–GFP–CD-MPR, or TfR–SBP–mCherry was transfected into HeLa cells 1 d before the experiment. The images were taken by spinning-disk microscopy for 60 min with 3-min time intervals. Only cells with a total intensity at time 0 between 0.5 × 10⁷ and 10⁷ arbitrary units were selected to be neither overexpressing nor too affected by bleaching across the 60 min. Because of random lateral movement of the microscopy stage, some time course data sets were stabilized with the image stabilizer plugin for ImageJ (http://www.cs.cmu.edu/~kangli/code/Image_Stabilizer.html). The Golgi was masked and intensity measured across the whole time course in that region. The data set from each cell was normalized so the highest value was equal to 1.

Fig. S1 and Video 1 show synchronized transport and eventual destinations of RUSH cargos. Videos 2, 3, 4, 7, and 8 show distinct carriers derived from the Golgi complex for RUSH cargos, and Video 9 shows those for TfR mutant. Figs. S2 and S3 and Videos 5 and 6 show direct fusion of TfR-containing carriers with the plasma membrane. Fig. S4 shows control experiments for cargo segregation in the Golgi complex. Fig. S5 shows different routes for cargos to reach endosomes and lysosomes.

We thank Frank Perez and Gaelle Boncompain for kindly providing the original RUSH constructs and Michal Jarnik for help with electron microscopy.

This work was funded by the Intramural Program of the National Institute of Child Health and Human Development, National Institutes of Health (grant ZIA HD001607).

The authors declare no competing financial interests.

Author contributions: Y. Chen and J.S. Bonifacino conceived the project. Y. Chen, D.C. Gershlick, and S.Y. Park performed experiments and analyzed the data. All authors participated in the preparation of the figures and the writing of the manuscript.