Results

BioID can successfully tag proteins colocalized with secreted proteins

We first investigated if intracellular PPIs between each SecP and their supporting SecMs can be measured using the BioID method. To do this, each bait-BirA was expressed in HEK293 cells using the Flp-In™ system (see materials and methods) for targeted integration of the transgenes into the same genomic locus to ensure comparable transcription rates of each transgene. Variations in mRNA level caused by random integration can trigger adaptive response such as the unfolded protein response in some cell lines which reciprocally alters the active PPIs network involved in the secretion. We observed successful secretion of bait-BirA proteins into culture supernatant, evaluated by Western blot (Fig. 2a). Thus, the BirA fusion did not change the intracellular localization of the model proteins, and it is expected that they enter the secretory pathway where they are processed and packaged for secretion. We also verified the biotinylation profile by western blot for each cell line in the presence and absence of biotin. The biotinylation profile of the bait-BirA cells is different when biotin is added to the culture with substantial increased biotinylation of specific proteins, while no obvious change is observed for WT (Fig. 2b), suggesting that BioID2 successfully tagged specific proteins within the cells. Colocalization of the bait-BirA proteins and the biotinylated proteins was then studied by multicolor co-immunofluorescence microscopy to test whether biotinylated proteins are actual partners of the model proteins. The results demonstrated successful labeling of the interactors by BirA through colocalization of the biotin-labeled proteins and bait-BirA, while WT did not show increased biotinylation under the same experimental conditions (Fig. 3). To quantify the colocalizations, we calculated different colocalization metrics (see methods) from the images and compared to the WT (Table 1 and Fig. S1), and the results confirmed the specificity of the BioID labeling system to tag the proximal proteins (closer Li’s ICQ value to 0.5 or Pearson’s R value to one demonstrates a dependent protein staining pattern between the red and green channels).

WT cells revealed endogenous biotinylation landscape

After successful tagging of the proximal proteins we aimed to identify the interactions with each bait protein. For this, cells were lysed, and biotinylated proteins were purified using streptavidin. Purified proteins were digested by trypsin, and peptides were subjected to LC-MS/MS, and the biotinylated proteins in samples were mapped using MaxQuant (Tyanova, Temu, & Cox, 2016). Differentially biotinylated proteins were then identified in each sample compared to the WT using Perseus (Tyanova & Cox, 2018). When implemented in a WT control cell line, we identified proteins that are biotinylated endogenously along with bona fide interactors. These include proteins that bind biotin as a cofactor such as carboxylases which is problematic with streptavidin-based protein detection (Tytgat et al., 2015). Although the extent to which the general endogenous biotinylation has not been systematically quantified, the biotinylated proteins isolated from the WT sample showed considerable overlap with interacting proteins detected in other model protein samples, suggesting endogenous biotinylation may be more pervasive than previously believed. Thus, using the interaction partners detected from the WT sample as background, we filtered out interactions detected in each model bait-BirA sample that were likely a result of endogenous biotinylation. Among the top differentially biotinylated proteins in bait-BirA samples, the bait proteins showed the highest log-fold change (LFC) (Fig. 4 and TableS1). This observation is expected because the bait protein is a potential substrate for BirA located in the closest vicinity of the enzyme and is considered as evidence to show the biotinylation system is working properly within the cell.

Interactors are enriched for secretory pathway components and co-secreted proteins

For each secreted protein we identified probable PPIs as interactors having 3-fold or greater enrichment and an adjusted p-value < 0.1, in model proteins compared to WT control (Fig. S2 for all thresholds and Fig. 5b for all significant interactions) based on secretory pathway-related gold standards compiled from 3 independent gene sets (see methods). We saw a significant enrichment for the secretory pathway machinery, secretory-resident and co-secreted proteins among probable PPIs across all model protein samples (Fig. 5a and Fig. S3). The secretory machinery components are more enriched among the top 300 hits for all model proteins than other co-secreted proteins, suggesting more frequent interactions between the secretory pathway machinery and their products than the crosstalk between co-secreted proteins. Probable PPIs detected in all model proteins (n=19) and hits shared among all SERPIN gene products are significantly enriched for proteins involved in protein folding (Fig. 5b). Indeed, molecular chaperones are highly promiscuous when assisting protein folding due to their inherent flexibility (Mayer, 2010). Apart from the shared interactions, PPIs for each model protein differ substantially. Thus, the question remained if these private interactions correspond to unique properties of each model protein.

Private interactors reflect post-translational and structural features of model proteins

PPIs in secretory pathway mediate the folding, modification and transportation of secreted proteins (Bonifacino & Glick, 2004; Calakos, Bennett, Peterson, & Scheller, 1994; Ikawa et al., 1997; Pearl & Prodromou, 2006; Watanabe et al., 2019). Incidentally, co-expression analysis has linked certain PTMs across the secretome to the expression of their responsible enzymes. For example, PDIs are consistently upregulated in tissues secreting disulfide-rich proteins (Feizi et al., 2017). As the bait-BirA proteins differ in structural composition and PTMs (Fig. 6), we wondered if bait-BirA proteins with shared features have higher affinity for specific interactors. More specifically, we hypothesize that proteins requiring a specific PTM would preferentially interact with the secretory machinery components responsible for the PTM synthesis. To establish a more comprehensive connection around product-specific interactions with the secretory pathway, we aggregated various PTM and structural properties across all model proteins and analyzed their associations with the corresponding secretory machinery using a Bayesian modeling framework (see methods). Among the studied PTMs, bait-BirA proteins with disulfide bonds and N-linked glycans demonstrated higher affinity towards specific interactions (fig. 7a) that are known to help secretion of proteins with the corresponding PTMs. Thus, we analyzed the detected interactions associated with glycosylation, disulfide bond addition, and protein folding.

Proteins with increased glycosylation are associated with quality control pathways

We detected significant interactions in the Calnexin/Calreticulin cycle and related processes for more heavily glycosylated proteins (Fig. 7a). For example, the glycosylated baits interacted with calreticulin (CALR), a calcium-binding chaperone that promotes folding, oligomeric assembly, and quality control of glycoproteins in the ER (Nauseef, McCormick, & Clark, 1995). They also interacted with UGGT1, which recognizes glycoproteins with minor folding defects and reglucosylates single N-glycans near the misfolded part of the protein. Reglucosylated proteins are then recognized by CALR for recycling to the ER and refolding or degradation (Ferris, Jaber, Molinari, Arvan, & Kaufman, 2013). Two members of the PDI family, PDIA3 and ERp29, which form a complex with calreticulin/calnexin, also showed association with N-glycosylated baites (Fig. 7a) suggesting their role in glycoprotein folding and quality control. Calnexin/Calreticulin-PDIA3 complexes promote the oxidative folding of nascent polypeptides (Sakono, Seko, Takeda, & Ito, 2014) and ERp29 promotes isomerization of peptidyl-prolyl bonds to attain the native polypeptide structure (Sakono et al., 2014; Tannous, Pisoni, Hebert, & Molinari, 2015). Proteins with chaperone activity, such as HSPA5 (Fig. 7a), were also found to interact with N-linked glycan-containing bait-BirA. HSPA5 is a component of the glycoprotein quality-control (GQC) system which recognizes glycoproteins with amino acid substitutions, and targets them for ER‐associated degradation (ERAD) (Ferris, Kodali, & Kaufman, 2014). EDEM3, another interactor associated with the N-glycan containing proteins (Fig. 7a), is a glycosyltransferase involved in ERAD mediated degradation of glycoproteins by catalyzing mannose trimming from Man8GlcNAc2 to Man7GlcNAc2 in N-glycans (Ninagawa et al., 2014). Given that most of these molecular chaperones and enzymes are involved in ERAD mediated degradation of the misfolded glycoproteins these findings suggest the quality control pathways are critical for synthesizing and secreting proteins with N-linked glycans.

Disulfide bond formation is rate-limiting in protein secretion

Several members of the PDI family including P4HB, PDIA3, PDIA4 and PDIA6 significantly interacted with model-BirAs containing more disulfide bonds (Fig. 7a). These enzymes catalyze the formation, breakage and rearrangement of disulfide bonds through the thioredoxin-like domains (Kozlov, Määttänen, Thomas, & Gehring, 2010). The identification of various PDIs highlights the importance of the oxidative folding enzymes in protein folding and maintaining stability that can limit the efficiency of protein secretion. The proteins with more disulfide bonds also interact with major ER chaperones HSAP5 and DNAJB11, a co-chaperone of HSAP5, that play a key role in protein folding and quality control in the ER lumen (Ng, Watowich, & Lamb, 1992; Yu, Haslam, & Haslam, 2000), highlighting their important role in secretion of the disulfide bond enriched proteins. The PDI, ERp44 showed the strongest association (LFC > 8, Fig. S4) with disulfide bond enriched proteins i.e. SERPINC1 and SERPING1. ERp44 mediates the ER localization of the oxidoreductase Ero1α (an oxidoreductin that reoxidizes P4HB to enable additional rounds of disulfide formation) through the formation of reversible mixed disulfides (Anelli et al., 2003). Hence, the strong association of ERp44 suggests the importance of the thiol-mediated retention in disulfide bond formation, particularly when secretory is loaded with the proteins with dominant disulfide bond. In addition, ERO1LB, PRDX4 and SIL1 were ER-localized enzymes that were associated with disulfide bond formation. ERO1LB efficiently reoxidizes P4HB (Mezghrani et al., 2001), PRDX4 couples hydrogen peroxide catabolism with oxidative protein folding by reducing hydrogen peroxide (Zito, 2013), and SIL1 can reverse HSAP5 cysteine oxidation which alters its chaperone activity to cope with suboptimal folding conditions (Siegenthaler, Pareja, Wang, & Sevier, 2017). The identification of these oxidoreductase enzymes highlights the importance of ER redox homeostasis in disulfide bond formation and protecting cells from the consequences of misfolded proteins.
To validate the importance of the ERp44 interaction, which showed the highest fold change, on productivity of proteins with more disulfide bonds we knocked down ERP44 in the cells expressing SERPINC1-BirA, using an orthogonal RNAi approach, esiRNAs, to target gene knockdown (Kittler, Heninger, Franke, Habermann, & Buchholz, 2005). HEK293 cells expressing SERPINC1-BirA were transfected with esiRNA against ERP44, with EGFP and KIF11 as negative/positive controls, respectively. The supernatants of esiRNA experiments collected 48 and 72 h post-transfection for measuring the secretion of SERPINC1-BirA by ELISA. All quantifications were repeated in duplicate. The results showed ERP44 knockdown in SERPINC1-BirA cells led to the 44% and 41% reduction in the secretion of SERPINC1 at day two and three post-transfection, respectively, supporting the importance of the Thiol-mediated retention on secretion of the disulfide bond enriched protein.

Identified PPIs are associated with structural motifs on bait proteins

In addition to PTMs, the Bayesian modeling framework found associations between SecP structural features and the SecMs (Fig. 7b). For example, model proteins depleted in the asx motif (Golovin & Henrick, 2008) showed higher tendency to interact with DNAJB1, a molecular chaperone of the HSP70 family. The asx motif impacts N-glycan occupancy of Asn-X-Thr/Ser sites, depending on the ability of the peptide to adopt an Asx-turn motif (Imperiali, Shannon, & Rickert, 1992; Imperiali, Shannon, Unno, & Rickert, 1992). As another example, NUCB1, a chaperone-like amyloid binding protein that prevents protein aggregation (Bonito-Oliva, Barbash, Sakmar, & Graham, 2017), interacted more strongly with our proteins with more ST-turns (Fig. 7b). ST turns occur frequently at the N-termini of α-helices (Doig, Stapley, Macarthur, & Thornton, 2008) and are regarded as helix capping features which stabilize α-helices in proteins (Aurora & Rosee, 1998). Thus, the enriched interaction of the NUCB1 with St-turn suggests that it can help stabilize folding of protein with a predominant α-helical secondary structure.