New factors for protein transport identified by a genome-wide CRISPRi screen in mammalian cells

The molecular machinery underlying the processes of protein transport and secretion has been conserved from yeast to mammalian cells, being essential to maintain specificity and communication between organelles, exocytosis, and endocytosis. Fully one third of proteins navigate the secretory pathway, entering this system via the ER. They are then transported via cellular compartments, including the Golgi apparatus and transport carriers, to be targeted to their final destination (Lee et al., 2004). Recent demonstrations indicate that cell compartments establish cross-regulatory mechanisms through numerous membrane contact sites (Wu et al., 2018), are endowed with tightly regulated dynamics (Valm et al., 2017), and stand at the crossroad of signaling pathways where inputs and outputs are integrated and coordinated (Luini and Parashuraman, 2016). Thus, protein transport and secretion processes are clearly more complex than previously thought.

Yeast genetic studies and in vitro biochemical approaches were initially used to discover the basal protein transport machinery (Novick et al., 1980; Braell et al., 1984). More recently, arrayed RNA interference screens at the genome scale or targeting kinases/phosphatases expanded the list of functional and regulatory components of secretory pathways in metazoan cells (Bard et al., 2006; Asensio et al., 2010; Wendler et al., 2010; Chia et al., 2012; Simpson et al., 2012). In these studies, artificial secreted reporters preceded by a signal sequence such as HRP (ss-HRP) or firefly luciferase were used for detection (Bard et al., 2006; Wendler et al., 2010). Alternatively, secretory pathway organization or transport to the cell surface of fluorescently labeled exogenous transmembrane proteins such as vesicular stomatitis virus G (VSV-G) were analyzed using high throughput imaging or flow cytometry systems (Asensio et al., 2010; Chia et al., 2012; Simpson et al., 2012). Although these approaches were useful, their implementations in different cell types, with various cargo proteins and environmental conditions, remain difficult.

Indeed, a current challenge is to understand how secretory pathways are adapted and regulated in response to intrinsic demands or environmental cues, and eventually altered in diseases. Toward this objective, versatile platforms are needed to uncover factors involved in protein transport in various physiological and pathophysiological contexts. Recent developments of the bacterial CRISPR-associated nuclease Cas9 technology paired with libraries of single-guide RNAs (sgRNAs) have been successfully used to perform pooled genome-wide screening (Shalem et al., 2015; Kampmann, 2018), where targeted genes can be disrupted (Shalem et al., 2014) and gene expression can be inhibited or activated (Gilbert et al., 2014).

Here, we developed an efficient strategy using a pooled genome-wide CRISPR interference (CRISPRi) screen to identify genes involved in protein trafficking and secretion in mammalian cells, and we highlight the contribution of newly identified factors in these processes.

We first established a HeLa cell line stably expressing the GFP-tagged TAC protein (interleukin-2 receptor) and the catalytically inactive Cas9 (dCas9) fused to the KRAB repressor protein (HeLa TAC-GFP dCas9-KRAB cells). The TAC protein, which contains a single transmembrane domain (TMD) and localizes at the cell surface, represents a straightforward reporter to investigate protein transport along the ER-Golgi secretory pathway (Stanley and Lacy, 2010). A GFP signal allows monitoring of the total expression of the TAC protein, whereas its cell surface expression can be assessed by immunofluorescence microscopy and flow cytometry analysis using a phycoerythrin (PE)–conjugated antibody, which recognizes the extracellular domain of the TAC protein (Fig. 1, A–C; and Fig. S1). On the other hand, the dCas9-KRAB fusion protein in association to sgRNA targeting the transcription start site (TSS) of specific genes (CRISPRi system) is a powerful tool to inhibit gene expression (Gilbert et al., 2013). We first tested our reporter system by incubating cells with brefeldin A (BFA), an inhibitor of protein transport through the ER and Golgi membranes (Misumi et al., 1986). Flow cytometry analysis demonstrated a decrease of TAC surface expression upon BFA treatment (Fig. 1 D). Incubation of cells with trypsin, a protease that cleaves most proteins expressed at the cell surface, strongly reduced TAC surface expression that recovered after 18 h in trypsin-free medium. In these conditions, BFA treatment prevented TAC surface expression recovery (Fig. 1 E).

We then validated our cell line by analyzing TAC expression after knockdown of genes essential for secretory pathway function. Cells were infected with lentivirus to express control sgRNA (Gal4) or sgRNAs targeting the TSS of SCFD1, which regulates SNARE complex assembly (Carr and Rizo, 2010), Sec24A, a COPII vesicular coat subunit (Wendeler et al., 2007), or Sec61A, which promotes translocation of secreted and membrane proteins into the ER (Voorhees and Hegde, 2016). A tight quality control is associated with the cotranslational insertion of nascent polypeptide chains upon emergence from the ribosomal exit tunnel, and a translocation defect of nascent polypeptide chains in the ER leads to their degradation (Brandman and Hegde, 2016). For this reason, down-regulation of Sec61A is an appropriate control to monitor TAC expression when TAC translation is altered. 7 d after infection, total and surface TAC expressions were monitored by flow cytometry. Sec61A knockdown, as expected, reduced both surface and total TAC expression. However, SCFD1 and Sec24A knockdown strongly reduced TAC surface transport without affecting the total TAC expression (Fig. 1, F and G). Cells were also analyzed for the surface/total TAC expression ratio. Upon SCFD1 and Sec24A knockdown, a strong enrichment in the lower quartile was observed related to a decrease of TAC surface expression. Interestingly, no enrichment in one or the other quartile was detected in Sec61A-depleted cells, where both surface and total TAC expression were reduced (Fig. 1, G and H). Altogether, these results established our cell line as an efficient and specific cellular platform to monitor secretory pathway function, and prompted us to use it to perform a pooled genome-wide CRISPRi screen.

HeLa TAC-GFP dCas9-KRAB cells were transduced with the CRISPRi-v2 library (Horlbeck et al., 2016a) at a low MOI (0.3) and selected for stable integration with puromycin. 7 d after transduction, cells were separated by FACS into a high and low surface/total TAC expression ratio, selecting upper and lower quartiles, respectively. Genomic DNA was then isolated, and read counts of sgRNAs were obtained by deep sequencing to determine the abundance of each sgRNA in the two cell fractions (Fig. 2 A). Subsequent analysis using MAGeCK (Li et al., 2014) allowed the identification of many genes whose down-regulation modified TAC transport at the cell surface (Table S1 and Fig. 2 B).

On one hand, in the lower quartile, gene ontology term enrichment analysis of the 100 top-ranked genes revealed that knockdown of genes encoding components of “Golgi vesicle transport” was the main functional category inhibiting TAC transport (Fig. 2, B and C). Genes involved in “toxin transport,” “antigen processing and presentation,” and “Golgi organization” were also highly enriched. More specifically, we identified as top hits most of the COPI subunits (COPA, COPB1, COPB2, COPG1, COPZ1, and ARCN1) as well as the COPII subunits Sec24A, Sec24B, and Sec13. We also identified the SNARE Sec22B and the SNARE-associated factors SCFD1 and NBAS, several subunits of the TRAPP complex and the conserved oligomeric Golgi complex, the small GTPase SAR1 and the Rab GTPase RAB1A, the Sec23-interacting protein Sec23IP, and the exocyst complex component EXOC2, among others (Table S1 and Fig. 2 B). Interestingly, factors belonging to small ADP-ribosylation factor (ARF) and ADP-ribosylation factor-like (ARL) GTPases, Golgins, or additional SNAREs were not identified among the top hits. This is consistent with previous reports (Wendler et al., 2010; Simpson et al., 2012), highlighting the high degree of redundancy between paralog genes that could explain why the knockdown of these components did not alter robustly TAC transport. Of note, genes involved in mRNA homeostasis and protein translation were also identified among hit genes. Although it is likely that their knockdown indirectly alters TAC transport, a subset of them may also be part of transcriptional or translational programs required for the expression of components specifically involved in the secretory pathway. We can also not exclude the possibility that some of these genes may have additional nonbiosynthetic functions.

On the other hand, in the upper quartile, the knockdown of genes encoding known components involved in “protein exit from the ER,” “response to topologically incorrect protein,” “ER to cytosol transport,” and “response to ER stress,” as well as components involved in posttranslational modification, were the main functional categories promoting TAC transport (Fig. 2, B and D). This suggests that inhibition of the machinery required for the maintenance of ER homeostasis and quality control along the secretory pathway may favor transport of cargo proteins that are most likely misfolded or incompletely processed.

Altogether, these results clearly demonstrated the value of our approach to identify genes involved in secretory pathway function and organization.

62 genes with unknown or poorly characterized function that were identified either in the lower or upper quartiles as candidate genes (Table S2) were selected for a secondary screen using a secretion assay based on the detection of ss-HRP (von Blume et al., 2009; see Materials and methods for selection criteria of candidate genes). Genes were knocked down in HeLa–ss-HRP after transfection of specific smart pool siRNAs prearrayed in a 96-well plate. After 3 d, cells were washed and incubated with complete medium, and cell lysates and medium were harvested after 8 h for HRP quantification. As a positive control, in SCFD1-depleted cells, a strong signal in the cell lysate associated with a low signal detected in the medium demonstrated a clear defect of HRP secretion resulting in its accumulation inside cells (Fig. 3 A and Table S2). This approach identified several genes for which the knockdown inhibited HRP secretion. They included, for example, WDR7, YPEL5, TMEM161, FAM162B, and GPR162 that encode an orphan G protein coupled receptor (GPCR). Interestingly, we also identified few genes where the knockdown promoted HRP secretion such as TMEM167A, FAM46A, as well as the ubiquitin specific peptidase USP32 (Fig. 3 A and Table S2). Future analyses will help to dissect their roles.

We selected four genes for further analysis that encode the coiled-coil domain-containing protein 151 (CCDC151), CCDC157, C10orf88, and the tetratricopeptide repeat protein 17 (TTC17). They were highly enriched in the lower quartile of the CRISPRi screen, their knockdown strongly inhibited HRP secretion, and they encode proteins with a poorly defined or unknown function. First, we aimed to confirm the effect of their knockdown on TAC protein transport. HeLa TAC-GFP cells were transfected with specific individual siRNAs and, after 3 d, knockdown efficiency was monitored by RT-PCR (Fig. 3 B). Flow cytometry analysis revealed that knockdown of the selected genes inhibited TAC expression at the cell surface, confirming that they were true hits (Fig. 3, C and D). We also found that their siRNA-mediated knockdown reduced the transport of the major histocompatibility complex I (MHC-I), an endogenous cargo protein trafficked along the ER-Golgi secretory pathway to the cell surface (Fig. 3, E–G). Altogether, these results demonstrated that the selected candidate genes were required for both exogenous and endogenous cargo protein transport.

Next, we assessed the role of newly identified factors on the morphology of ER, ER exit sites, cis Golgi membranes and TGN by immunofluorescence microscopy, using antibodies targeting calnexin (CNX), Sec31A, GM130, and TGN46. As a control, in SCFD1-depleted cells, Golgi membranes were completely fragmented, characterized by GM130- and TGN46-positive punctae dispersed throughout the cytoplasm. Interestingly, this phenotype was also observed after CCDC151 and C10orf88 knockdown (Fig. 4, A–C). Conversely, after TTC17 knockdown, Golgi membranes remained localized in the perinuclear area, but presented as an expanded structure (Fig. 4, A, B, and D). In CCDC157-depleted cells, Golgi membranes exhibited a characteristic ring-shaped organization as observed for both GM130 and TGN46 staining (Fig. 4, A, B, and E). Electron microscopy studies confirmed that Golgi stacks had altered ultrastructure in TTC17-depleted cells, characterized by enlarged and swollen Golgi cisternae, and in CCDC157-depleted cells, most likely caused by accumulated and coalesced transport carriers in close proximity of Golgi membranes (Fig. 4 F). Along with the perturbed Golgi complex architecture, the distributions of ER exit sites were also altered upon candidate gene knockdown (Fig. S2 A), with no obvious changes in the distribution and morphology of the ER (Fig. S2 B).

How could the selected factors play a role in protein transport and secretion, and secretory pathway organization? To address this question, we looked for interacting partners using the BioPlex network, which encompasses results aiming at identifying binding partners of proteins encoded by the human genome (Huttlin et al., 2015, 2017). Interrogation of this resource identified TTC17-interacting proteins, including exclusively ER- and Golgi-resident proteins and cargo proteins transported along the secretory pathway (Fig. 5 A). Interacting partners of CCDC151 include the SNARE-associated protein SNAPIN, GTPase Rab-Like protein 5, and Golgin A5 (Fig. 5 B). Interacting partners of C10orf88 include Rab protein GDI2, clathrin heavy chain, clathrin adaptors accessory protein CCDC91, and plasma membrane proteins NCKAP1 and LILRB4 (Fig. 5 C). The BioPlex resource does not contain a network for CCDC157. Taken together, these data supported a direct role of the selected factors on the secretory pathway function, prompting us to probe their site of action.

A Pfam domain analysis (Finn et al., 2016) or TMHMM software (Krogh et al., 2001) did not predict TMD for the selected factors, whereas the HMMTOP algorithm (Tusnády and Simon, 2001) yielded one TMD for TTC17 (TMHMM and HMMTOP are two transmembrane protein topology prediction methods, based on a hidden Markov model). To assess the subcellular localization of the selected factors, we performed immunofluorescence microscopy on HeLa cells using specific antibodies directed against the endogenous proteins, after permeabilization with digitonin and extensive washing. This procedure allows the removal of soluble cytoplasmic pool of proteins, highlighting their potential association with intracellular compartments. This approach revealed a close proximity between CCDC151 and the centrosomal marker centrin-3 (Fig. 5, D and E). This location is consistent with the interacting partners of CCDC151 revealed by BioPlex and the Golgi membrane phenotype observed in CCDC151-depleted cells (Fig. 4, A and B). Indeed, interacting partners of CCDC151 included CEP170 (Fig. 5 B), a key component of the centrosome that functions as a microtubule-organizing center, essential for the perinuclear localization of Golgi membranes (Rios and Bornens, 2003). Hence, we hypothesize that CCDC151 is a bridging factor between the centrosome and Golgi membranes via interactions with Golgin A5 and CEP170, thereby contributing to the maintenance of Golgi architecture. Disruption of this functional link after depletion of CCDC151 may result in Golgi membrane dispersion throughout the cytoplasm. Of interest, our results also revealed the presence of TTC17 and CCDC157 on Golgi membranes (Fig. 5, F, G, and I). A residual signal of C10orf88 was mainly diffuse in the cytoplasm (Fig. 5, H and I). The subcellular location of TTC17 and CCDC157, and the consequence of their knockdown on Golgi membranes, prompted us to further characterize these two factors.

A permanent flux of membrane traffics to and from the Golgi complex. BFA has been regularly used to study Golgi membrane dynamics, inducing a complete fusion of Golgi membranes with the ER and endosomes via tubule formation (Lippincott-Schwartz et al., 1989) and the relocation of peripheral matrix Golgi proteins such as GM130 and GRASP65 to the ER exit site (Mardones et al., 2006). First, we tested whether Golgi membrane reorganization was altered upon BFA treatment in TTC17- and CCDC157-depleted cells. HeLa cells were transfected with specific individual siRNAs and, after 3 d, treated with BFA. Immunofluorescence microscopy revealed that GM130 completely dispersed upon 30 min of BFA treatment in control cells and in most of the CCDC157- and TTC17-depleted cells (Fig. 6, A and B). Next, we tested the effects of CCDC157 and TTC17 knockdown on Golgi membrane reassembly after BFA washout, which is mediated by fusion events. In control and TTC17-depleted cells, Golgi membrane reassembly occurred in most cells after 60 min, while in CCDC157-depleted cells, GM130 remained completely dispersed (Fig. 6, A and B). Given these results demonstrating that Golgi membrane reassembly after BFA washout was strongly altered upon CCDC157 knockdown, we assessed the distribution of the ER-Golgi intermediate compartment (ERGIC) by immunofluorescence microscopy using an anti-ERGIC53 antibody. ERGIC is a collection of tubulovesicular membrane clusters that allow delivery of secretory cargo from ER exit sites to the Golgi complex in a COPII vesicle-dependent manner via homotypic and heterotypic fusion events (Lorente-Rodríguez and Barlowe, 2011). In control and TTC17-depleted cells, ERGIC53 staining mainly localized in the proximity of Golgi membranes, whereas in CCDC157-depleted cells, ERGIC53 staining was more dispersed throughout the cytoplasm (Fig. 6, C and D). These results suggested that CCDC157 could act as an important factor for the fusion of transport carriers with the Golgi complex.

To further test a role in membrane fusion, we assessed the impact of CCDC157 knockdown on the distribution of transport carriers derived from the endocytic pathway that fuse with Golgi membranes (Johannes and Popoff, 2008). We performed immunofluorescence microscopy experiments using an antibody targeting the transferrin (Tf) receptor (TfR), which, although it is mainly recycled to the plasma membrane after endocytosis (Huebers and Finch, 1987), is also retrieved to Golgi membranes (Snider and Rogers, 1985; Woods et al., 1986; Jin and Snider, 1993), and an antibody targeting EEA1, a marker of early endosomes, critical for endosomal trafficking (Barysch et al., 2009). We hypothesized that while TfR-containing transport carriers do not mainly fuse with Golgi membranes (Snider and Rogers, 1985; Woods et al., 1986), alteration of these fusion events could lead over time to the accumulation of transport carriers in the vicinity of the Golgi apparatus. In TTC17-depleted cells, the distribution of TfR and EEA1 was not markedly affected compared with control cells (Fig. 6, E and F; and Fig. S3). However, in CCDC157-depleted cells, the distribution of both TfR and EEA1 was strongly altered, most of the signal being coalesced to the perinuclear area in the form of enlarged TfR- and EEA1-positive vesicles, surrounded by Golgi membranes (Fig. 6, E–H; and Fig. S3). In line with these observations, the uptake as well as the recycling of Tf were reduced upon CCDC157 knockdown (Fig. S4). Altogether, these results suggested that in the absence of CCDC157, incoming carriers deriving from the endocytic pathway accumulated and coalesced in close proximity to Golgi membranes, strengthening CCDC157 as being an important factor for the fusion of transport carriers with the Golgi complex.

In addition to being a gate entrance and a major sorting station, the Golgi apparatus allows posttranslational modifications such as glycosylation. This process is spatially regulated, being dependent on accurate localization of specific enzymes acting sequentially from the cis- to trans-Golgi cisternae. Does altered Golgi membrane organization in CCDC157- and TTC17-depleted cells impact Golgi polarization and create glycosylation defects? To address these questions, we first assessed the localization of GM130 and TGN46, cis- and trans-Golgi markers, respectively, in cells incubated in the presence or absence of nocodazole. Nocodazole is a microtubule-depolymerizing agent that causes Golgi ribbon dissociation and hence Golgi stack dispersion (Cole et al., 1996), facilitating the visualization of Golgi-resident protein segregation in individual Golgi stacks (Fig. 7, A and D). In control cells, GM130 and TGN46 staining only partially colocalized (Pearson’s correlation coefficient r = 0.507). Similar results were obtained in CCDC157-depleted cells (r = 0.495). Conversely, in TTC17-depleted cells, GM130 and TGN46 staining showed a high degree of colocalization (r = 0.805; Fig. 7, A–D), suggesting that protein localization in their respective cisternae was perturbed with no further segregation of cis- and trans-Golgi markers. In addition, we observed that upon TTC17 knockdown, nocodazole failed to induce complete Golgi membrane fragmentation, although the microtubule cytoskeleton was entirely depolymerized (Fig. 7 C). We hypothesized that the dramatic alteration of Golgi membrane architecture upon TTC17 depletion prevents efficient Golgi stack dissociation.

We next tested whether the loss of Golgi polarization in TTC17-depleted cells impairs cargo protein glycosylation. The lysosomal protein LAMP2 is highly glycosylated during its transport across the secretory pathway, increasing its molecular weight (Stewart et al., 2017). In control and CCDC157-depleted cells, immunoblotting analysis revealed LAMP2 mainly as a fully processed 100-kD band. However, in TTC17-depleted cells, a smaller form of ∼80 kD became apparent, indicating incomplete LAMP2 processing, likely related to a glycosylation defect (Fig. 7 E). These results established TTC17 as an important factor for Golgi membrane structure and polarized arrangement of Golgi cisternae, enabling cargo protein processing and transport.

Although a large number of gene products involved in protein transport was identified in recent decades (Novick et al., 1980; Braell et al., 1984; Bard et al., 2006; Wendler et al., 2010), the list is far from complete, and several crucial issues are emerging. Indeed, in eukaryotic cells, evolution has given rise to an incredible diversity and complexity of cargo molecules, whose efficient and accurate transport along the secretory pathway is of paramount importance for organismal development, cell and tissue homeostasis, immunity, metabolic regulation, nerve transmission, and healthy aging (Rothman and Orci, 1992; Söllner and Rothman, 1994; Lee et al., 2004; Zanetti et al., 2012; Guo et al., 2014; Malhotra and Erlmann, 2015). An increasing number of studies has demonstrated that cells are endowed with specific structural and functional machinery that can be tightly regulated to cope with the whole spectrum of cargo proteins (Cancino et al., 2014; Ma et al., 2018; Raote et al., 2018; Lopes-da-Silva et al., 2019). In this context, it is essential to develop versatile and high-throughput screening strategies to uncover how secretory pathways are adapted to and are regulated in response to intrinsic demands and environmental cues, and altered in diseases.

In this study, we implemented an unbiased pooled genome-wide CRISPRi screen allowing us to efficiently identify new components of the secretory pathway. Major benefits have already been associated with pooled CRISPR screens, establishing such approaches as powerful tools for systematically defining gene function in mammalian cells (Gilbert et al., 2014; Shalem et al., 2015; Canver et al., 2018). The efficiency of our strategy relies on the combination of key parameters. First, we used a complementary reporter system associated with cell selection based on the ratio of two fluorescent signals. This reporter system allowed monitoring surface/total TAC expression ratio variations and therefore differential FACS-based cell sorting, directly relative to the transport of TAC protein to the cell surface. The strengths of this strategy were exemplified with Sec61A knockdown, where we could avoid following up undesirable phenotypes such as defective TAC protein synthesis. Second, as the secretory pathway is of critical importance for cell homeostasis and viability, gene expression was systematically down-regulated using the dCas9 fused to a KRAB effector domain for CRISPRi, instead of the active Cas9 for CRISPR knockout (Shalem et al., 2015; Kampmann, 2018). The resulting transcriptional repression of target genes preserved minimal gene expression, preventing massive cell death after down-regulation of key factors for protein and membrane trafficking. Third, the combination of the dual fluorescent reporter with the CRISPRi system allowed us to monitor TAC transport and perform FACS-based cell sorting at a relatively early time point (7 d after lentivirus transduction), which also would reduce excessive cell death after gene knockdown. Altogether, combining these technical parameters makes our approach a noteworthy advance compared with previous RNA interference–based genome-wide screens and resulted in a powerful platform allowing us to focus on the identification of genes involved in secretory pathway function and organization.

Our work highlighted several factors whose knockdown inhibited or promoted the trafficking of both exogenous and endogenous cargo proteins. Various genes encoding proteins known to be involved in calcium homeostasis, signaling pathways, or lysosomal activity, among others (Table S2), strengthen the idea that there are functional links between these cellular processes and the secretory pathway. In addition, several proteins with unknown or poorly characterized function were identified. In particular, TMEM167A, FAM46A, and USP32, which promote secretion upon knockdown, as well as C10orf88, WDR7, YPEL5, TMEM161, FAM162B, and GPR162, which inhibit secretion upon knockdown, deserve greater attention. For example, USP32 is a membrane-bound ubiquitin protease localized in the Golgi apparatus and overexpressed in breast cancer (Akhavantabasi et al., 2010); and GPR162 is an orphan GPCR. Compelling evidence suggests that the activation of GPCRs on Golgi membranes is critical for protein and membrane trafficking, but their identities remain elusive (Díaz Añel and Malhotra, 2005; Cancino et al., 2014; Eichel and von Zastrow, 2018). Altogether, our results provide a valuable resource to gain future important insights into fundamental mechanisms governing protein and membrane transport.

Our study further characterized two of the newly identified factors, TTC17 and CCDC157, as new actors of critical importance for the structure and function of Golgi membranes. Although additional studies will be required to fully decipher their roles, our results suggest that like other coiled-coil proteins (Wong and Munro, 2014; Cheung and Pfeffer, 2016) CCDC157 could be part of a tethering complex required for fusion events at the Golgi membranes. Furthermore, phenotypes observed in TTC17-depleted cells associated with TTC17 interacting partners involved in sphingomyelin metabolism (ENPP7 and NAAA; Fig. 5 A; Duan et al., 2003; Tsuboi et al., 2007) suggest that TTC17 could play a role in lipid production and distribution, critical parameters for Golgi membrane organization (van Galen et al., 2014; Campelo et al., 2017). Deciphering how CCDC157 and TTC17 are recruited to membranes and how their functions are coordinated with other structural factors and components of the tethering and fusion machinery will be of major interest.

In conclusion, while initial screens performed in yeast revealed the basic principles conserved across species (Novick et al., 1980), and, more recently, arrayed RNA interference screens revealed key players that function in metazoans (Bard et al., 2006; Saito et al., 2009; von Blume et al., 2009), our pooled CRISPRi screen unveiled new components that further refine the steps along the secretory pathway. In addition, we anticipate that the adaptation of our screening platform to specific cargo proteins, professional secretory cell types, and intrinsic or environmental challenges will open new and stimulating perspectives for a better understanding of the secretory pathway architecture in health and disease, for both conventional and unconventional secretion (Zhang and Schekman, 2013; Cruz-Garcia et al., 2018; Villeneuve et al., 2018; Chiritoiu et al., 2019).

Antibodies used in this study included mouse monoclonal antibodies anti-TAC conjugated to PE (BioLegend; 356103), anti-GM130 (BD Biosciences; 610822), anti-Sec31A (BD Biosciences; 612351), anti–MHC-I (Thermo Fisher Scientific; MA1-70111), anti-EEA1 (BD Biosciences; 610457), anti–α-tubulin (Thermo Fisher Scientific; 62204), anti-LAMP2 (BioLegend; 354302) and anti-centrin3 (Novus Biologicals; H00001070-M01), a mouse polyclonal antibody anti-C10orf88 (Abcam; ab169106), a sheep polyclonal antibody anti-TGN46 (BioRad; AHP500), a rabbit monoclonal antibody anti-ERGIC53 (Abcam; ab125006), rabbit polyclonal antibodies anti-CNX (Abcam; ab22595), anti-CCDC151 (Abcam; ab151469), anti-TTC17 (Atlas Antibodies; HPA038508), anti-CCDC157 (GeneTex; GTX45090), anti-TfR (Abcam; ab84036); secondary antibodies donkey anti-sheep IgG–Alexa Fluor 568 (A21099), goat anti-mouse IgG–Alexa Fluor 488 (A11001), goat anti-mouse IgG–Alexa Fluor 568 (A11004), goat anti-rabbit IgG–Alexa Fluor 568 (A11011), and goat anti-rabbit IgG–Alexa Fluor 488 (A11034; Thermo Fisher Scientific); and a secondary antibody mouse IgGκ conjugated to HRP (Santa Cruz Biotechnology; sc-516102).

Reagents used in this study were obtained from the following sources: Dulbecco’s PBS (D8662), BFA (B6542), nocodazole (M1404), puromycin (P9620), G-418 (4727878001), protease inhibitor cocktail (05892953001), trypsin (T3924), penicillin/streptomycin (P0781), and L-glutamine (G7513) were from Sigma-Aldrich. Q5 HF DNA polymerase (M0491L), Phusion HF DNA polymerase (M0530L), Phusion HF reaction Buffer (B0518S), and dNTPs (N0447S) were from New England Biolabs. T4 DNA ligase (EL0011), FBS (10270-106), Lipofectamine 2000 transfection reagent (12566014), DH5α bacteria (18258012), Restore Plus reagent (46430), ProLong Gold antifade reagent with DAPI (P36935), OptiMEM (319885062), and Tf conjugated with Alexa Fluor 647 (T23366) were from Thermo Fisher Scientific. TransIT-2020 transfection reagent (5405) and TransIT-293 transfection reagent (MIR2700) were from Mirus. Polybrene (TR-1003-G) and digitonin (11024-24-1) were from Merck. 0.45 µm low protein binding membrane (28145-479) was from VWR International. NucleoSpin Blood L kit (740569.10) was from Macherey-Nagel. ECL Western Blotting Detection Reagent (RPN2106) was from GE Healthcare. DMEM (D6546) was from Molecular Probes. SPRI AMPure XL beads were from Beckman Coulter. White CulturPlate 96-well plates (6005680) were from PerkinElmer. RNAeasy Plus Mini kit (74134) was from QIAGEN. Cloned AMV First-Strand cDNA Synthesis kit (12328032) was from Invitrogen. Mouse IgGκ Isotype (400111) was from BioLegend. Universal mycoplasma detection kit (30-1012K) was from American Type Culture Collection (ATCC).

Plasmids included pCMV6-AC-IL2R-GFP (Origene plasmid RG215768). pHR-SFFV-dCas9-BFP-KRAB was a gift from S. Qi (Stanford University, Stanford, CA) and J. Weissman (University of California, San Francisco, San Francisco, CA; Addgene plasmid 46911; Gilbert et al., 2013). pU6-sgRNA EF1Alpha-puro-T2A-BFP was a gift from J. Weissman (Addgene plasmid 60955; Gilbert et al., 2014). pCMV-VSV-G was a gift from B. Weinberg (Whitehead Institute for Biomedical Research, Cambridge, MA; Addgene plasmid 8454; Stewart et al., 2003). psPAX2 was a gift from D. Trono (École polytechnique fédérale de Lausanne, Lausanne, Switzerland; Addgene plasmid 12260). hCRISPRi-v2 genome-wide library was a gift from J. Weissman (Addgene pooled library 83969; Horlbeck et al., 2016a).

All cell lines, HeLa cells (ATCC), HeLa–ss-HPR cells (a gift from V. Malhotra, Centre for Genomic Regulation, Barcelona, Spain), HeLa TAC-GFP cells (this paper), HeLa TAC-GFP dCas9-KRAB cells (this paper), and HEK293T cells (ATCC) were grown in complete medium consisting of DMEM (Molecular Probes; D6546) containing 10% FBS (Thermo Fisher Scientific; 10270-106), 100 U/ml penicillin, 100 µg/ml streptomycin (Sigma-Aldrich; P0781), and 2 mM L-glutamine (Sigma-Aldrich; G7513) at 37°C with 5% CO₂. All cell lines were tested every month using ATCC universal mycoplasma detection kit (ATCC; 30-1012K) to confirm they were clear of contamination by mycoplasma.

To generate a HeLa cell line stably expressing GFP-tagged TAC protein, we transfected HeLa cells with pCMV6-AC-IL2R-GFP (Origene; plasmid RG215768) using TransIT-2020 transfection reagent (Mirus; 5405), following the manufacturer’s recommendations, and after 2 d, transfected cells were selected with 600 µg/ml G-418 (Sigma-Aldrich; 4727878001) during 1 wk. Then, GFP-positive cells were isolated by FACS, and single cells were collected in 96-well plates. After expansion into the 6-well format, TAC-GFP expression was assessed from clonal lines by flow cytometry and immunofluorescence microscopy analysis. The GFP signal allows monitoring the total TAC expression, whereas the pool expressed at the cell surface can be assessed using a PE-conjugated antibody, which recognizes the extracellular domain of the TAC protein. To generate HeLa TAC-GFP dCas9-KRAB cells, we transduced HeLa TAC-GFP cells with lentivirus to express dCas9-KRAB-BFP fusion protein (see section Lentivirus production and transduction). After 2 d, transduced cells were sorted for GFP- and BFP-positive cells.

Specific individual siRNA and smart pools siRNA were transfected in HeLa cells, HeLa–ss-HPR cells, and HeLa TAC-GFP cells using Lipofectamine 2000 transfection reagent (Thermo Fisher Scientific; 12566014) following the manufacturer’s recommendations. Individual siRNA and smart pools siRNA were purchased from Dharmacon. The siRNA sequences used were the following: control 5′-CGU​ACG​CGG​AAU​ACU​UCG​A-3′, SCFD1 5′-AGA​CUU​AUU​GAU​CUC​CAU​A-3′, TTC17 5′-GAA​UAC​GGG​UGC​UGA​AGA​A-3′, 5′-GGA​GAG​AGU​UAA​UCU​UUC​U-3′, 5′-GGA​CCU​GGA​UCU​AUA​UGA​U-3′, 5′-GGA​CCA​GCC​UGU​ACG​CUA​U-3′, CCDC157 5′-GGA​AAG​ACC​UGA​CGC​GCC​U-3′, 5′-ACU​ACG​GAC​CUG​CGG​CUA​A-3′, 5′-CUG​AGU​GGG​AGC​ACG​ACA​A-3′, 5′-CAG​CGU​GGA​AUC​CCA​GAU​A-3′, CCDC151 5′-GCU​CAA​GGC​CUA​UCU​AAU​G-3′, 5′-GAA​GGC​UGC​CCG​AAU​ACA​A-3′, 5′-CGG​GAA​CGC​UAC​AUA​AGU​G-3′, 5′-GCA​CAU​UAC​CAG​CGU​GUA​C-3′, and C10orf88 5′-GCA​CAU​UGA​UGA​UAA​GAU​U-3′, 5′-GAA​AUA​AGA​CCG​AGU​GUC​A-3′, 5′-GCU​CAG​CAG​UUG​AUG​GAU​A-3′, 5′-GGG​AUA​CCU​CUA​AGA​CAU​U-3′. After incubation for 72 h, total RNA was extracted and knockdown efficiency was assessed by RT-PCR.

Gene-specific CRISPRi sgRNA oligonucleotide sequences used were the following: Gal4 5′-GAA​CGA​CTA​GTT​AGG​CGT​GTA-3′, Sec61A 5′-GGC​TAG​CAC​TGA​CGT​GTC​TCT-3′, SCFD1 5′-GAG​CAG​CCA​GTA​TTC​GGG​AA-3′, and Sec24A 5′-GAC​ATG​ATG​ACT​GGG​TTG​GA-3′. Forward sgRNA oligonucleotides were designed with 5′-TTG and 3′-GTTTAAGAGC overhangs, and reverse sgRNA oligonucleotides were designed with 5′-TTAGCTCTTAAAC and 3′-CAACAAG overhangs. Oligonucleotides were diluted to 100 µM in water, and 1 µl of forward and reverse oligo was combined in a 50-µl reaction containing 200 mM potassium acetate, 60 mM Hepes, pH 7.4, and 4 mM magnesium acetate. Oligonucleotides were incubated for 5 min at 95°C and slow-cooled (0.1°/s) for annealing. Annealed oligos were diluted 1/40, and 1 µl of insert was ligated into 10 ng of digested vector (pU6-sgRNA EF1Alpha-puro-T2A-BFP; Addgene; plasmid 60955 digested with BstXI and Blpl) using T4 DNA ligase (Thermo Fisher Scientific; EL0011) in a final volume of 10 µl. Ligation was allowed to proceed for 1 h. Ligated products were transformed into DH5α bacteria (Thermo Fisher Scientific; 18258012). After lentivirus production (see section Lentivirus production and transduction), TAC-GFP-dCas9-KRAB HeLa cells were infected, and knockdown efficiency was monitored 7 d after transduction by RT-PCR. TAC-GFP expression was assessed by flow cytometry analysis.

Lentivirus was generated in HEK293T cells using TransIT-293 transfection reagent (Mirus; MIR2700) following the manufacturer’s recommendations. At 50% confluence, HEK293T cells incubated in complete medium lacking penicillin/streptomycin were transiently transfected with a lentiviral vector, either pHR-SFFV-dCas9-BFP-KRAB (Addgene plasmid 46911) for dCas9-KRAB fusion protein expression, pU6-sgRNA EF1Alpha-puro-T2A-BFP (Addgene plasmid #0955) containing gene-specific CRISPRi sgRNA oligonucleotides, or hCRISPRi-v2 library (Addgene pooled library 83969) for a pooled genome-wide CRISPRi screen, together with packaging plasmids pCMV-VSV-G (Addgene plasmid 8454) and psPAX2 (Addgene plasmid 12260) at a 4:1:3 ratio. Transfection was performed using one 10-cm dish for dCas9-KRAB fusion protein expression and for each specific sgRNA, and using two 15-cm dishes for hCRISPRi-v2 library. After transfection (24 h), the medium was refreshed, and after an additional 24 h, virus was collected and filtered through a 0.45-µm low protein binding membrane (VWR International; 28145-479). For dCas9-KRAB fusion protein expression and CRISPRi-mediated knockdown, 1 ml of medium was used immediately for infection of ∼500,000 target cells plated in 6-well plate in a final volume of 2 ml. For a pooled genome-wide CRISPRi screen, ∼200 million TAC-GFP-dCas9-KRAB HeLa cells plated in thirty 15-cm dishes were infected at an MOI of 0.3. Medium was supplemented with 8 µg/ml polybrene (Merck; TR-1003-G) lacking penicillin/streptomycin. After 24 h, medium was replaced with fresh medium, and after an additional 24 h, high BFP-positive cells expressing dCas9-KRAB fusion protein were sorted by FACS. For CRISPRi-mediated knockdown and a pooled genome-wide CRISPRi screen, 48 h after infection, 2 µg/ml puromycin (Sigma-Aldrich; P9620) was added to select transduced cells.

Approximately 200 million HeLa cells stably expressing TAC-GFP and dCas9-KRAB fusion protein were transduced with the hCRISPRi-v2 library (Addgene; 83969) at an MOI of 0.3 to ensure no more than one viral integration event per cell. The CRISPi-v2 library comprised 104,535 sgRNAs targeting 18,905 human genes. Each sgRNA of the library targeted the TSS of the respective genes in regions of low nucleosome occupancy avoiding genomic DNA compaction to impact Cas9 binding (Horlbeck et al., 2016a,b). 2 d after infection, an aliquot of cells was analyzed by FACS to confirm transduction efficiency, and then cells were cultured with 2 µg/ml puromycin (Sigma-Aldrich) to remove uninfected cells. After five additional days, cells were trypsinized, washed extensively, and plated with fresh medium for 18 h, allowing recovery of transmembrane protein expression at the cell surface. Cells were then detached using 0.5 mM EDTA, resuspended in blocking buffer (OptiMEM containing 5% FBS), and stained for TAC surface expression using a PE-conjugated anti-TAC antibody during 1 h at 4°C. Finally, live cells were sorted by FACS following the PE/GFP fluorescence ratio, selecting the upper and lower quartile, respectively. 10 million cells were sorted for each fraction (Parnas et al., 2015; Park et al., 2017; Menzies et al., 2018). Cell sorting was performed using a FACSAria sorter. Two independent screens were performed. After FACS sorting, cells were centrifuged, and pellets were frozen until processing.

Genomic DNA was extracted from four distinct cell populations, corresponding to the upper and lower quartile of the PE/GFP fluorescence ratio, from two independent screens, using the NucleoSpin Blood L kit (Macherey-Nagel; 740569.10) according to the manufacturer’s instructions. Then, for each sample, all genomic DNA was amplified by PCR to enrich sgRNA sequences. Each 100 µl PCR reaction contained 1× Phusion HF reaction Buffer (NEB; B0518S), 3% vol/vol DMSO, 0.4 µM of each forward and reverse primer mix (Integrated DNA Technologies), 0.2 mM dNTPs (NEB; N0447S), 40 U/ml Phusion HF DNA polymerase (NEB; M0530L), and 2 µg genomic DNA. PCR primer sequences for the forward primers with specific barcode for de-multiplexing were the following: 5′-AAT​GAT​ACG​GCG​ACC​ACC​GAG​ATC​TAC​ACG​ATC​GGA​AGA​GCA​CAC​GTC​TGA​ACT​CCA​GTC​ACC​TTG​TAG​CAC​AAA​AGG​AAA​CTC​ACC​CT-3′, 5′-AAT​GAT​ACG​GCG​ACC​ACC​GAG​ATC​TAC​ACG​ATC​GGA​AGA​GCA​CAC​GTC​TGA​ACT​CCA​GTC​ACG​CCA​ATG​CAC​AAA​AGG​AAA​CTC​ACC​CT-3′, 5′-AAT​GAT​ACG​GCG​ACC​ACC​GAG​ATC​TAC​ACG​ATC​GGA​AGA​GCA​CAC​GTC​TGA​ACT​CCA​GTC​ACA​GTT​CCG​CAC​AAA​AGG​AAA​CTC​ACC​CT-3′, and 5′-AAT​GAT​ACG​GCG​ACC​ACC​GAG​ATC​TAC​ACG​ATC​GGA​AGA​GCA​CAC​GTC​TGA​ACT​CCA​GTC​ACT​AGC​TTG​CAC​AAA​AGG​AAA​CTC​ACC​CT-3′, The PCR primer sequence for the common reverse primer was 5′-CAA​GCA​GAA​GAC​GGC​ATA​CGA​GAT​CGA​CTC​GGT​GCC​ACT​TTT​TC-3′. PCR cycling conditions used included 1× 98°C for 30 s, 23× 98°C for 30 s, 56°C for 15 s, 72°C for 15 s, and 1× 72°C for 10 min. For each sample, PCR products were pooled, size-selected using SPRI AMPure XL beads (Beckman Coulter) following the manufacturer’s recommendations, and quantified on a Qubit and run on a Bioanalyzer (Agilent). The four libraries were pooled to a total concentration of 10 nM and subjected to DNA sequencing on a HiSeq2500 (Illumina) sequencer. Samples were sequenced with the rapid run mode of the single-ended 50 bp, following the manufacturer’s recommendations, using a custom sequencing primer: 5′-GTG​TGT​TTT​GAG​ACT​ATA​AGT​ATC​CCT​TGG​AGA​ACC​ACC​TTG​TTG-3′.

For data analysis, raw sequencing reads were aligned to the v2CRISPRi library sequences using Bowtie 2, and data were analyzed using the MAGeCK algorithm (Li et al., 2014). For genes with multiple independent TSS targeted by the sgRNA library, we calculated phenotype and P values independently for each TSS.

As we intended to validate hit genes found in our primary screen with a secondary screen, hit genes were selected as interesting candidates among those with a P value <0.01 and a log2 fold change (lfc) less than −0.5 for genes enriched in the lower quartile and a lfc >0.3 for genes enriched in the upper quartile (see Tables S1 and S2). The difference of lfc cutoff applied for the selection of candidate genes relied on the weaker phenotypes observed on TAC transport for genes enriched in the upper quartile compared with those enriched in the lower quartile. Thus, 62 genes identified either in the lower (39 genes) or upper quartiles (23 genes) as candidate genes with an unknown or poorly characterized function were selected for a secondary screen using a secretion assay based on the detection of ss-HRP. HeLa cells stably expressing ss-HRP were transfected in a 96-well plate prearrayed with specific smart pools siRNA (Dharmacon) targeting selected genes, using Lipofectamine 2000 transfection reagent (Thermo Fisher Scientific; 12566014). Briefly, for each well, 3,000 HeLa–ss-HRP cells in 80 µl of complete medium lacking penicillin/streptomycin were plated into a 96-well plate containing 20 µl of 250 nM siRNA and 0.5% (vol/vol) Lipofectamine 2000 in OptiMEM (Thermo Fisher Scientific; 319885062). Medium was replaced after 24 h. After two additional days, cells were washed with medium and incubated with 150 µl of complete medium for 8 h. Then 50 µl of the medium was collected and mixed with ECL reagent (GE Healthcare; RPN2106) in white CulturPlate 96-well plates (PerkinElmer; 6005680), and the chemiluminescence was measured in a Tecan Microplate reader M1000 PRO. For normalization, cells were lysed with lysis buffer supplemented with protease inhibitors to quantify intracellular HRP activity.

After CRISPRi- or siRNA-mediated target gene knockdown, cells were washed with PBS and detached by incubation with 0.5 mM EDTA in PBS for 10 min. Cells were then resuspended in blocking buffer (OptiMEM containing 5% FBS), and live cells were incubated for 1 h at 4°C with an anti-TAC antibody conjugated to PE. GFP and PE fluorescence were collected on a LSR Fortessa (BD Biosciences) and analyzed using FlowJo software (FlowJo). For surface MHC-I expression, cells were washed with PBS and detached by incubation with 0.5 mM EDTA in PBS for 10 min. Cells were then fixed with 4% (wt/vol) PFA in PBS for 15 min at room temperature, washed, and incubated for 1 h at 4°C with an anti–MHC-I antibody. The epitope recognized by the anti–MHC-I antibody localizes in the extracellular domain of MHC-I. Cells were washed again, incubated with a goat anti-mouse IgG–Alexa Fluor 488 and fluorescent signals collected on a LSR Fortessa (BD Biosciences) and analyzed using FlowJo software (FlowJo). For Tf internalization and recycling, see section FACS-based endocytosis and recycling assays.

Control HeLa cells and cells knocked down for TTC17 and CCDC157 were detached with 0.5 mM EDTA for 10 min, washed in serum-free medium, and incubated at 4°C for 30 min in the presence of 50 µg/ml Tf conjugated with Alexa Fluor 647 (Thermo Fisher Scientific; T23366). Cells were then incubated at 37°C and at different type points (0, 2, 5, 10, and 15 min), Tf internalization was stopped by placing cells on ice for 10 min. Cells were then fixed with 4% (wt/vol) PFA in PBS for 15 min at room temperature and analyzed by flow cytometry on a LSR Fortessa (BD Biosciences). For recycling assay, cells were incubated with 50 µg/ml Tf conjugated with Alexa Fluor 647 for 30 min at 37°C, washed, and incubated at 37°C with 100 µg/ml unlabeled Tf for different time points (0, 5, 10, 15, and 30 min). Cells were then fixed with 4% PFA and analyzed by flow cytometry.

Cells were lysed for 30 min on ice in lysis buffer (50 mM Tris, pH 7.4, 150 mM NaCl, 0.1% SDS, 1% (vol/vol) Triton X-100, and 0.5% sodium deoxycholate) supplemented with protease inhibitor cocktail (Sigma-Aldrich; 05892953001), 1 mM Na₃VO₄, and 25 mM sodium fluoride and centrifuged at 16,000 × g for 15 min. Samples were incubated with 1× SDS sample buffer at 95°C for 10 min, resolved by 12% SDS-PAGE, and transferred to methanol activated Immobilon-P 0.45 µm polyvinylidene difluoride membranes for blotting. Membranes were blocked with 5% (wt/vol) BSA in PBS containing 0.1% Tween 20 (PBS-T) for 30 min at room temperature and probed with appropriate primary antibodies overnight at 4°C in PBS-T containing 5% (wt/vol) BSA. Membranes were then washed 3 × 15 min in PBS-T and incubated with the corresponding HRP-conjugated secondary antibody for 1 h at room temperature in PBS-T containing 5% (wt/vol) BSA. Membranes were then washed again for 3 × 15 min in PBS-T, and signals were detected with ECL Western Blotting Detection Reagent (GE Healthcare; RPN2106) and acquired with the Chemidoc MP Imaging System (Bio-Rad). Membranes were stripped with Restore Plus (Thermo Fisher Scientific; 46430) according to the manufacturer’s instructions.

Cells grown on coverslips were fixed with cold methanol for 10 min at −20°C or 4% (wt/vol) PFA in PBS for 15 min at room temperature. Cells fixed with PFA were permeabilized with 0.1% Triton X-100 in PBS at room temperature and then incubated with blocking buffer (2.5% [wt/vol] FCS and 0.1% Triton X-100 in PBS) for 30 min at room temperature. Cells were then incubated with primary antibody diluted in blocking buffer for 1 h at room temperature followed by PBS wash and incubated with secondary antibody. Secondary antibodies conjugated with Alexa Fluor 488, 568, or 647 were diluted in blocking buffer and incubated 1 h at room temperature. Samples were mounted using ProLong Gold antifade reagent with DAPI (Thermo Fisher Scientific; P36935). Images were acquired with a Leica SP8 laser confocal scanning microscope with a 40× objective. Images displayed in figures are representative single Z-slices. After acquisition, images were processed using an Airyscan processing tool on the ZEN software provided by Zeiss.

For cytosol washout prior to immunofluorescence microscopy, cells were washed twice with room temperature KHM buffer (125 mM potassium acetate, 25 mM Hepes-KOH, pH 7.2, and 2.5 mM magnesium acetate). Cells were then permeabilized by incubation in KHM buffer containing 30 µg/ml digitonin for 5 min on ice followed by a wash for 5 min at room temperature with KHM buffer. Cells were subsequently fixed with 4% PFA and processed for immunofluorescence microscopy. This procedure allows the removal of soluble cytoplasmic pool of proteins, highlighting their potential association with intracellular compartments.

The level of colocalization was determined by acquiring confocal images from ∼30 cells per condition. Colocalization quantifications were performed using Coste’s method of thresholding with object Pearson’s analysis using Imaris 8.2.0 by Bitplane AG. Intensity profile graphs of the confocal images were calculated by a pixel-based method by ImageJ (National Institutes of Health).

For unbiased analysis of Golgi membrane surface area, images were analyzed with the CellProfiler software (Broad Institute; Dao et al., 2016). For each image analyzed, the area of Golgi membranes, identified as object, was extracted using the module “MeasureObjectSizeShape.” For unbiased analysis of Golgi membrane morphology, the classifier of the CellProfiler Analyst was trained using cells randomly chosen from the whole experiment. These were classified as having a ring shape structure, fragmented, or intact Golgi. The classifier of the CellProfiler Analyst was then used to define, for all cells, Golgi membrane morphology on object level.

In all experiments, images shown in individual panels were acquired using identical exposure times and scan setting and adjusted identically for brightness and contrast using Photoshop CS5 (Adobe).

3 d after transfection with individual specific siRNA, control HeLa cells and cells knocked down for TTC17 and CCDC157 were fixed with 2% PFA–2% glutaraldehyde solution in sodium cacodylate buffer for 1 h at room temperature. Cells were carefully detached using a plastic cell scraper, collected into Eppendorf tubes and centrifuged to obtain the pellet. Cells were then post-fixed for 30 min in 1% OsO4 at room temperature, washed three times in distilled water and post-fixed for 1 h in 1% uranyl acetate. The pellets were dehydrated in graded steps of ethanol (50, 70, 90, 96, and 100%), two times in 100% of propylene oxide, and embedded into epon. Sections (60 nm thick) were cut on a Leica UC7 ultramicrotome and examined with a Fei Tecnai 12 BioTwin Spirit transmission electron microscope.

Total RNA was extracted from HeLa cells using the RNAeasy Plus Mini kit (QIAGEN; 74134), and cDNAs were synthesized with the Cloned AMV First-Strand cDNA Synthesis kit (Invitrogen; 12328032) and oligo d(T)₂₀ according to the manufacturer’s instruction. cDNA of indicated genes was amplified by PCR using Q5 HF DNA polymerase (NEB; M0491L) and the following specific primers: GAPDH forward primer 5′-ACC​ACC​ATG​GAG​AAG​GCT​GG-3′, reverse primer 5′-CTC​AGT​GTA​GCC​CAG​GAT​GC-3′, PCR product size: 527 bp; SCFD1 forward primer 5′-GGG​GAA​GAT​GAA​GGA​GCC​ATA-3′, reverse primer 5′-TCA​GCC​TCA​GAA​GGT​GCT​TG-3′, PCR product size: 319 bp; Sec24A forward primer 5′-CCA​TTG​TCC​TCG​TGC​ATC​AT-3′, reverse primer 5′-TGC​CAA​TAA​GCC​ACC​TCC​TT-3′, PCR product size: 369 bp; Sec61A forward primer 5′-AGT​GGA​CCT​GCC​AAT​CAA​GT-3′, reverse primer 5′-CTT​TGG​CAG​AGG​AAC​CTG​AG-3′, PCR product size: 379 bp; TTC17 forward primer 5′-CAG​ATG​ACC​ATG​CAC​GAA​AAA-3′, reverse primer 5′-GTT​GGC​CAA​GTT​GAC​AAG​AGG-3′, PCR product size: 248 bp; CCDC157 forward primer 5′-GCA​GCC​CAC​CCA​CTG​TAA​TAA-3′, reverse primer 5′-TGC​CCT​GGG​TGT​ACT​CTC​CTA-3′, PCR product size: 236 bp; C10orf88 forward primer 5′-TAC​ACA​ACT​GCC​TGG​TGG​AGA-3′, reverse primer 5′-CGG​TCT​TAT​TTC​CCA​CAT​GGA-3′, PCR product size: 218 bp; and CCDC151 forward primer 5′-TGA​ACC​AAG​AGG​CCC​TCA​AT-3′, reverse primer 5′-GCG​CTC​GTT​CTC​CAG​TTT​CT-3′, PCR product size: 148 bp. The PCR products were resolved on 1.5% agarose gel.

The statistical details of all experiments are indicated in the figure legends, including statistical analysis performed, error bars, statistical significance, and experiment numbers (n). Statistics were performed using GraphPad Prism 6 software.

Fig. S1 shows the establishment of the cell line expressing a dual fluorescent reporter for protein transport based on the TAC-GFP expression, and the optimization of experimental conditions to assess TAC surface expression by flow cytometry. Fig. S2 shows the ER exit site and ER organization in control cells and in TTC17- and CCD157-depleted cells visualized by immunofluorescence microscopy. Fig. S3 shows the effect of TTC17 and CCDC157 knockdown on EEA1 distribution visualized by immunofluorescence microscopy. Fig. S4 shows the effect of TTC17 and CCDC157 knockdown on the uptake and recycling of Tf analyzed by flow cytometry. Table S1 shows MAGeCK analysis of the pooled genome-wide CRISPRi screen replicate. Table S2 shows the 62 candidate genes selected from the CRISPRi screen and results of the secondary screen using a secretion assay based on the detection of ss-HRP by chemiluminescence.

We thank members of the Moreau laboratory for helpful discussion.

S.J. Popa acknowledges support from a Wellcome Trust PhD studentship (109152/z/15/z). S.E. Stewart acknowledges support from a Biotechnology and Biological Sciences Research Council Future Leader Fellowship (BB/P010911/1). F. Campelo acknowledges financial support from the Spanish Ministry of Economy and Competitiveness (“Severo Ochoa” program for Centres of Excellence in R&D [SEV-2015-0522], FIS2015-63550-R, FIS2017-89560-R, and BFU2015-73288-JIN, AEI/FEDER/UE), Fundació Privada Cellex, and the Generalitat de Catalunya through the CERCA program. A. Ashok acknowledges support from a La Caixa Foundation fellowship. J. Villeneuve acknowledges support from a Marie Curie fellowship within the the European Union’s Horizon 2020 research and innovation program (842919). D.C. Rubinsztein is grateful for support from the UK Dementia Research Institute (funded by the Medical Research Council, Alzheimer’s Research UK, and the Alzheimer’s Society) and the Roger de Spoelberch Foundation. This work was supported by a Wellcome Trust Strategic Award (100574/Z/12/Z), the Medical Research Council Metabolic Diseases Unit (MRC_MC_UU_12012/5), and the Isaac Newton Trust/Wellcome Trust Institutional Strategic Support Fund/University of Cambridge research grant.

The authors declare no competing financial interests.

Author contributions: J. Villeneuve, K. Moreau, V. Malhotra, R. Schekman, D.C. Rubinsztein, J. Weissman, and J. Ripoche conceptualized the experiments. J. Villeneuve and L. Bassaganyas performed and analyzed most of the experiments, with contributions from S.J. Popa, M. Horlbeck, C. Puri, S.E. Stewart, F. Campelo, A. Ashok, C.M. Butnaru, N. Brouwers, and K. Heydari. J. Villeneuve wrote the original and revised manuscripts with contribution from the other authors.