BAIAP3, a C2 domain–containing Munc13 protein, controls the fate of dense-core vesicles in neuroendocrine cells

Zhang et al. conducted a siRNA screen of C2 domain proteins involved in regulated peptide secretion. One of the hits, a Munc13 family member BAIAP3, was characterized as endosome localized involved in post-exocytic dense-core vesicle protein recycling to the TGN. BAIAP3 knockdown inhibited dense-core vesicle maturation/stability in neuroendocrine/endocrine cells.


Introduction
Dense-core vesicle (DCV) exocytosis is essential for peptidergic and aminergic signaling in the nervous, endocrine, and immune systems. DCV biogenesis in the TGN, DCV transport across the cytoplasm, and DCV docking/priming and fusion at the plasma membrane use similar mechanisms in different secretory cell types, but many aspects remain incompletely understood. After budding from the TGN, immature DCVs undergo maturation by endosomal retrieval to become fully functional (Kögel and Gerdes, 2010). Mature DCVs are recruited to the plasma membrane where they are docked and primed by several protein factors that assemble vesicle and plasma membrane SNA RE proteins into trans-complexes (Rizo and Xu, 2015). The Ca 2+ -triggered fusion of DCVs with the plasma membrane is mediated by Ca 2+ sensors acting on SNA RE complexes and the plasma membrane (Südhof and Rothman, 2009;Jahn and Fasshauer, 2012). Unlike synaptic vesicles that recycle locally, DCVs undergo compensatory endocytosis, which recycles DCV membrane proteins through retrograde trafficking to the TGN (Farquhar, 1983;Bauer et al., 2004).
C2 domain-containing proteins play essential roles at multiple steps in anterograde DCV-mediated trafficking. C2 domains are comprised of eight antiparallel β-strands connected by surface loops (Sutton and Sprang, 1998). Some C2 domains bind phospholipids in a Ca 2+ -dependent manner through negatively charged surface loop residues that form Ca 2+ binding sites with the membrane (Corbalan-Garcia and Gómez-Fernández, 2014). This enables C2 domain-containing proteins to mediate Ca 2+ -dependent membrane processes such as for Munc13-1/2 (C2B domain) function in vesicle priming or for synaptotagmin function in fusion triggering (Südhof, 2014). Some C2 domains do not bind Ca 2+ but mediate protein-protein interactions, such as for Munc13-1/2 (C2A domain) and CAPS (calciumdependent activator protein for secretion 1) function in vesicle docking and priming (Lu et al., 2006;Petrie et al., 2016).
Munc13 proteins are an important class of C2 domaincontaining proteins, with Munc13-1, -2, and -3 regulating Ca 2+ -dependent vesicle priming in neurons and endocrine cells (Koch et al., 2000;Kwan et al., 2006;Man et al., 2015). Similarly, Munc13-4 is essential for the Ca 2+ -dependent priming of secretory granules in immune cells (Feldmann et al., 2003;Boswell et al., 2012;Johnson et al., 2016) but also functions in the Ca 2+ -dependent regulation of late endosomal fusion events Dense-core vesicle (DCV) exocytosis is a SNA RE (soluble N-ethylmaleimide-sensitive fusion attachment protein receptor)dependent anterograde trafficking pathway that requires multiple proteins for regulation. Several C2 domaincontaining proteins are known to regulate Ca 2+ -dependent DCV exocytosis in neuroendocrine cells. In this study, we identified others by screening all (∼139) human C2 domain-containing proteins by RNA interference in neuroendocrine cells. 40 genes were identified, including several encoding proteins with known roles (CAPS [calcium-dependent activator protein for secretion 1], Munc13-2, RIM1, and SYT10) and many with unknown roles. One of the latter, BAI AP3, is a secretory cell-specific Munc13-4 paralog of unknown function. BAI AP3 knockdown caused accumulation of fusionincompetent DCVs in BON neuroendocrine cells and lysosomal degradation (crinophagy) of insulin-containing DCVs in INS-1 β cells. BAI AP3 localized to endosomes was required for Golgi trans-Golgi network 46 (TGN46) recycling, exhibited Ca 2+ -stimulated interactions with TGN SNA REs, and underwent Ca 2+ -stimulated TGN recruitment. Thus, unlike other Munc13 proteins, BAI AP3 functions indirectly in DCV exocytosis by affecting DCV maturation through its role in DCV protein recycling. Ca 2+ rises that stimulate DCV exocytosis may stimulate BAI AP3-dependent retrograde trafficking to maintain DCV protein homeostasis and DCV function. (Ménager et al., 2007;He et al., 2016;Woo et al., 2017). The fifth and most recently discovered Munc13 protein BAI AP3 is expressed in the central nervous system and neuroendocrine cells (Wojcik et al., 2013;Man et al., 2015). It is similar to Munc13-4 (46% amino acid similarity), with N-and C-terminal C2 domains predicted to bind Ca 2+ . However, the cellular role of BAI AP3 and the Ca 2+ -regulated membrane fusion process in which it may function have not been determined.
There are ∼139 genes encoding C2 domain-containing proteins in the human genome. The majority of these have not been studied for a role in regulated DCV exocytosis. We conducted a high throughput siRNA screen for C2 domaincontaining proteins involved in regulated DCV exocytosis in neuroendocrine cells. Of the 40 siRNA pools that inhibited Ca 2+ -triggered secretion, the screen identified several wellcharacterized proteins known to function in DCV exocytosis as well as many proteins of unknown function. The role of the novel Munc13 protein BAI AP3 was fully characterized as a C2-domain protein that controls the activity of DCVs in neuroendocrine and endocrine cells. BAI AP3 was found to operate by localizing to endosomes and regulating DCV protein recycling at the TGN. We propose that, in stimulated cells, Ca 2+bound BAI AP3 accelerates the retrograde trafficking of DCV proteins to the TGN to maintain DCV protein balance during the Ca 2+ -stimulated anterograde trafficking of DCVs.

C2-domain proteins required for DCV exocytosis
Pancreatic neuroendocrine BON cells contain numerous DCVs and exhibit Ca 2+ -triggered release of peptides (neurotensin, pancreastatin, and chromogranins), biogenic amines, or an expressed neuropeptide Y (NPY)-GFP (Parekh et al., 1994;Karatekin et al., 2008). We established an assay with a BON cell line that stably expresses NPY-Venus as a DCV cargo protein ( Fig. 1 A). The percent secretion (34.6 ± 0.4%) of cellular NPY-Venus in response to Ca 2+ influx (ionomycin stimulation) over a basal level (5.7 ± 0.1%) was used to measure acutely stimulated DCV exocytosis ( Fig. 1 B). We performed a triplicated siRNA screen of all human C2 domain-containing proteins (Table S1). Z score, which measures the number of SDs from a sample siRNA to the nontargeting siRNA (Birmingham et al., 2009), was used to identify inhibitory siRNAs. siRNA pools targeting three genes required for regulated exocytosis (CAD PS [CAPS], SNAP25, and STX1A [syntaxin 1A]) were used as controls. siRNAs targeting CAD PS strongly inhibited the percent secretion of NPY-Venus (z score = −8.7), whereas targeting SNAP25 and STX1A resulted in milder inhibition (z score = −2.4 and −2.1, respectively). Therefore, we set a z score = −2 as the threshold for hit identification. 40 inhibitory siRNA pools were identified with this threshold (Fig. 1 C and  Table S2). Of these, multiple genes with characterized roles in regulated exocytosis, such as CAD PS (Walent et al., 1992;Ann et al., 1997), UNC13B (Varoqueaux et al., 2002), RIM1/3 (Deng et al., 2011), PLA2 (Brown et al., 2003), DOC2A (Yao et al., 2011), and Synaptotagmin 10 (SYT10; Cao et al., 2011), were identified ( Fig. S1 B), which validated the screen. Additional assays were conducted to assess the quality of the screen. For example, UNC13A siRNAs did not inhibit NPY-Venus secretion in pilot experiments, and Munc13-4 (UNC13D) is not ex-pressed in BON cells (Fig. S1 A). In accord with this, neither gene was identified as a hit, whereas UNC13B (Munc13-2) was identified as a significant hit (Fig. S1 B). All vesicle-associated membrane proteins (VAMPs) were included in the screen as controls. VAMP2 is not expressed in BON cells (Fig. S1 A), and the screen did not identify VAMP2 as a hit but did identify VAMP1, 3, and 8 as hits (Fig. S1 B). The numerous controls, combined with the identification of known C2 domain-containing proteins for regulated DCV exocytosis, indicated the robustness of the screen.

BAI AP3 is required for regulated protein secretion
The screen identified BAI AP3 as a novel gene required for regulated protein secretion. BAI AP3 encodes a protein originally identified in a yeast two-hybrid assay as a brain-specific angiogenesis inhibitor 1 (BAI1)-interacting protein (Shiratsuchi et al., 1998). BAI AP3 is a Munc13 protein comprised of two Munc13 homology domains (MHDs; Koch et al., 2000) bracketed by two C2 domains predicted to bind Ca 2+ (Fig. 1 D and Fig. S6) and exhibits >70% sequence similarity to Munc13-4 in MHD and the C2 domain. BAI AP3 knockout mice display increased seizure propensity and anxiety behavior (Wojcik et al., 2013), which could result from abnormal neuropeptide or serotonin secretion (Kovac and Walker, 2013). A BAI AP3 missense mutation was detected in a search for hypothalamic signaling genes related to extreme obesity (Mariman et al., 2015). However, the precise function of BAI AP3 was unknown, so we characterized its cellular phenotype and its role in membrane trafficking.
Protein expression of BAI AP3 in BON cells was confirmed by Western blotting (Fig. 1 E). BAI AP3 siRNAs reduced the expression of BAI AP3 without reducing the expression of CAPS ( Fig. 1 E). We tested four different RNAi duplexes that targeted BAI AP3 and found that three significantly inhibited NPY-Venus secretion ( Fig. 1 F). In addition, endoribonucleaseprepared siRNAs (esiRNAs) targeting two regions of BAI AP3 mRNA were found to generate a similar knockdown phenotype (Fig. S1 C). To eliminate off-target effects of siRNA knockdown, we conducted rescue studies. Overexpression of a siRNAresistant BAI AP3 partially rescued stimulated NPY-Venus secretion inhibited by siRNA ( Fig. 1 G). Lack of full rescue was likely caused by incomplete transduction efficiency in expressing the siRNA-resistant BAI AP3. We used an orthogonal assay consisting of [ 3 H]serotonin secretion stimulated by ionomycin and found that BAI AP3 siRNA also inhibited Ca 2+ -induced [ 3 H]serotonin secretion ( Fig. 1 H). We noted that the reduced percent secretion of NPY-Venus or [ 3 H]serotonin in BAI AP3 knockdown cells was mainly because of an increase in DCV cargo pool size (Fig. 1, I and J), which suggested that the number of DCVs may be increased by BAI AP3 knockdown (see the BAI AP3 knockdown caused accumulation of defective DCVs in BON cells section).

BAI AP3 affects spontaneous DCV exocytosis
An explanation for an increase of DCV cargo induced by BAI AP3 knockdown ( Fig. 1 J) could be that spontaneous DCV exocytosis ( Fig. 2 A) during the 2-d siRNA incubation was affected. Indeed, spontaneous DCV exocytosis was found to occur in BON cells based on the analysis of the 2 d-conditioned medium from control cells, which contained NPY-Venus as well as mature prohormone convertase 1 (PC1; PC1/PCSK1), which is a specific DCV cargo (Fig. 2, B and C; Carraway et al., 1994). Proteins secreted into the culture medium from control cells in 1 d were analyzed by liquid chromatography tandem mass spectrometry and compared with proteins secreted during an acute 10-min stimulation with ionomycin, which represents soluble DCV cargo proteins. Similar protein components (PC1/PCSK1, chromogranin A [CgA/CMGA] and derived peptides, secretogranin 1/2 [SCG1/2], neurotensin [NTS], and VGF nerve growth factor inducible [VGF]) were detected in both samples ( Fig. 2 D and Fig. S2). Collectively, the results indicate that DCVs in resting BON cells undergo spontaneous exocytosis.
We assessed the effects of BAI AP3 siRNA treatment over a 2-d interval on spontaneous DCV exocytosis by quantifying NPY-Venus in cell lysates and culture medium. BAI AP3 siRNAs increased NPY-Venus in cell lysates and reduced amounts in the culture medium with minimal effects on total NPY-Venus (Fig. 2, E and F). The results indicate that the knockdown of BAI AP3 protein leads to the accumulation of DCV cargo in the cells.

BAI AP3 knockdown caused accumulation of defective DCVs in BON cells
The intracellular accumulation of DCV cargo during the BAI AP3 siRNA incubation might be because of an increased number of DCVs. Electron microscopy of BON cells confirmed that there were 39% more DCVs [mean value] after BAI AP3 knockdown (Fig. 3, A and B). However, despite the increased pool of DCVs, acute ionomycin stimulation of BAI AP3 knockdown cells resulted in amounts of secretion comparable with that of control cells ( Fig. 1 I), which indicated that DCVs accumulated during the 2-d BAI AP3 knockdown were defective for exocytosis. Because immature DCVs respond poorly to stimuli (Tooze et al., 1991;Eaton et al., 2000;Kögel and Gerdes, 2010;Bonnemaison et al., 2013), we determined whether BAI AP3 knockdown cells had an increased number of immature DCVs. Golgi VAMP4 is present on immature but not mature DCVs (Eaton et al., 2000). Only a small fraction of CgA + DCVs colocalized with VAMP4 in control cells, but this significantly increased after BAI AP3 knockdown (Fig. 3, D and E), indicating an increased number of immature DCVs. In addition, a small reduction in the mean size of DCVs was detected by electron microscopy (Fig. S3 A), consistent with a DCV maturation defect (Tooze et al., 1991). A defect in DCV maturation likely explains the accumulation of DCVs during the 2-d BAI AP3 siRNA treatment.
In addition to a maturation defect, DCV trafficking was disrupted in BAI AP3 knockdown cells. In electron micrographs of control cells, most DCVs were close to the plasma membrane and enriched in cell protrusions ( Fig. 3 A). In contrast, DCVs in BAI AP3 knockdown cells were more randomly distributed throughout the cytoplasm (Fig. 3, A and C). To specifically view plasma membrane-proximal DCVs, we imaged cells by total internal reflection fluorescence (TIRF) microscopy and found plasma membrane-proximal DCVs to be significantly reduced in BAI AP3 knockdown cells compared with control cells (Fig. 3, F and G), even though the total number of DCVs in the cells was increased (Fig. 3 B). The results indicate that BAI AP3 knockdown in resting cells caused DCV trafficking defects, which could indicate that the accumulated defective DCVs lack key proteins required for DCV transport on the cytoskeleton.

BAI AP3 knockdown caused a loss of insulin granules
To determine whether BAI AP3 knockdown affected DCV maturation in other cell types, we conducted studies in pancreatic β cell lines. BAI AP3 is expressed at relatively high levels in the β cell line INS-1 832/13 (called INS-1) and in primary pancreatic islets but not in pancreatic exocrine cells ( Fig. 4 A). Efficient BAI AP3 knockdown in INS-1 cells dramatically reduced the cellular content of PC2 (PC2/PCSK2) and insulin (Fig. 4, B-D). However, proinsulin and pro-PC2, which reside in the Golgi, were reduced to a much lesser extent or not at all. The preferential reduction of mature PC2 and insulin suggested a loss of insulin granules, which was confirmed by finding a strong reduction of Syt-9, an insulin granule membrane marker (Fig. 4 E, top;Iezzi et al., 2004), and also by electron microscope imaging ( Fig. S3 B). Residual insulin granules in BAI AP3 knockdown cells were also significantly smaller (by 18%) in electron microscopy (Fig. S3 C). Consistent with the loss of insulin granules, glucosestimulated insulin secretion was markedly reduced in BAI AP3 knockdown cells (Fig. 4 G). The loss of insulin granules could be caused by enhanced spontaneous DCV exocytosis, by reduced DCV biogenesis, or by increased DCV turnover. We did not detect a significant increase of insulin or proinsulin in cell culture medium from BAI AP3 knockdown cells (not depicted), which excludes increased spontaneous release as responsible. However, blocking degradation with lysosomal protease inhibitors resulted in a restoration of PC2/PCSK2 levels ( Fig. 4 F) as well as of Syt-9-and insulin-immunoreactive structures (Fig. 4 E,bottom). Electron microscopy ( Fig. 4, H and I) revealed that these structures were multigranular bodies that are characteristic of crinophagy or microautophagy (Orci et al., 1984;Marsh et al., 2007). The results indicate that INS-1 cells with BAI AP3 knockdown generate defective insulin granules that are preferentially targeted for lysosomal degradation. (E) Spontaneous NPY-Venus secretion into culture medium from wild-type, BAI AP3 knockdown, or CAPS knockdown cells. n = 3. (B, C, and E) Molecular mass is shown in kilodaltons. (F) Quantification of NPY-Venus in cell lysates, culture medium, or total (culture medium + cell lysate) for conditions similar to E after normalizing to cell number. n = 3 independent experiments. See also Fig. S2. Data are expressed as mean ± SD. P-values were obtained by a two-tailed Student's t test. *, P < 0.05; **, P < 0.01. KD, knockdown; NT, nontargeting.

BAI AP3 localizes to late and recycling endosomes
To determine how BAI AP3 knockdown caused the generation of defective DCVs, we determined the cellular localization of BAI AP3. C-terminal HA-tagged BAI AP3 mainly localized to cytoplasmic punctate structures in BON cells that did not colocalize with the DCV marker CgA (Fig. 5, A and B). However, we found substantial colocalization of BAI AP3-HA with GFP-Rab11 + recycling endosomes and with GFP-Rab9 + late endosomes. In contrast, BAI AP3-HA did not colocalize with GFP-Rab5 + early endosomes or GFP-Rab7 + late endosomes (Fig. 5,C and D). Rab9 and Rab7 occupy different subdomains of the late endosome (Barbero et al., 2002), which could account for differences in the colocalization of BAI AP3 with late endosomal Rab proteins.
To confirm an endosomal location for BAI AP3, we fractionated INS-1 cells and separated cellular organelles on equilibrium sucrose gradients. BAI AP3 migrated to light fractions, which were positive for the endosomal markers Rab7, Rab11, and Rab9, but not to heavy fractions, which were positive for the DCV marker α-granuphilin (Fig. 5, E and F). Endogenous BAI AP3 also localized to intracellular punctate structures in INS-1 cells (Fig. S4), similar to BAI AP3-HA overexpressed in BON cells. We conclude that BAI AP3 localizes to recycling and late endosomes but not to DCVs.

BAI AP3 promotes endosome recycling to the TGN
The preceding results show that defective DCVs are generated in cells lacking BAI AP3. This could indicate that endosomelocalized BAI AP3 functions in a recycling process that retrieves DCV proteins (e.g., VAMP4) from immature DCVs or that recycles DCV proteins that are lost to the plasma membrane during exocytosis . To test whether BAI AP3 functions in retrograde trafficking, we assessed the localization of TGN46, which recycles to the TGN via early and recycling endosomes (Johannes and Popoff, 2008). TGN46 mainly localized to the perinuclear TGN in control cells but was widely distributed in punctate structures in BAI AP3 knockdown cells (Fig. 6, A and B), consistent with a lack of recycling back to the TGN. Perinuclear TGN46 in control cells and cytoplasmic TGN46 in BAI AP3 knockdown cells colocalized with the TGN marker Golgin-97 ( Fig. 6 C), indicating that TGN integrity was disrupted after BAI AP3 knockdown. Disruption of steady-state TGN46 localization by BAI AP3 knockdown was quite similar to that observed for the depletion of other proteins required for retrograde trafficking to the TGN, such as VAMP4, STX6, STX16, Vti1A, conserved oligomeric Golgi 6 (COG6) of the COG complex, and VPS52 of the Golgi-associated retrograde protein (GARP) complex (Ganley et al., 2008;Pérez-Victoria et al., 2008;Laufman et al., 2011;Shitara et al., 2013).
We addressed the possible mechanism by which BAI AP3 could promote endosome recycling to the TGN. Other Munc13 proteins exhibit phospholipid-and SNA RE-binding properties that mediate membrane fusion reactions (Guan et al., 2008;Boswell et al., 2012;James and Martin, 2013;Woo et al., 2017). As BAI AP3 localizes to Rab11 + and Rab9 + endosomes, which can fuse with the TGN, we anticipated that BAI AP3 could interact with SNA REs and lipids on the TGN. The fusion of recycling endosomes with the TGN utilizes SNA RE complexes formed by STX6, STX16, Vti1A, and VAMP3 or VAMP4 (Mallard et al., 2002). GST-tagged cytoplasmic domains of STX6, STX16, VAMP3, and VAMP4 but not Vti1A were found to retain BAI AP3-HA from cell lysates (Fig. 6, D and E). BAI AP3 contains two C2 domains predicted to bind Ca 2+ (Fig. S6), and we found that Ca 2+ increased STX6, STX16, and VAMP4 interactions with BAI AP3 (Fig. 6, D and E). In a protein-lipid overlay assay (Fig. S5), recombinant BAI AP3 bound to PI(4)P and PI(3)P, characteristic of Golgi and endosomal membranes, respectively. Collectively, the results suggest that BAI AP3 could promote SNA RE complex assembly on PI(4)P-containing membranes to mediate retrograde trafficking to the TGN.
Lastly, we assessed the Ca 2+ -dependent properties of BAI AP3 in live cells. mCherry-BAI AP3 was mainly cytoplasmic with some punctate structures in resting BON cells. When intracellular Ca 2+ levels were increased by ionomycin stimulation, BAI AP3 was rapidly recruited to the plasma membrane and to GFP-VAMP4 + Golgi structures but not to DCVs (Fig. 6, F and G and Video 1). After ∼60 s, mCherry-  Fig. S4. Data are expressed as mean ± SD. P-values were obtained by a two-tailed Student's t test. *, P < 0.05; **, P < 0.01. BAI AP3 at the Golgi returned to the resting level before stimulation. The recruitment of BAI AP3 to the Golgi is consistent with a proposed role in promoting endosome fusion with the TGN in retrograde trafficking. Recruitment to the plasma membrane was consistent with previous studies (Lecat et al., 2015). Whether BAI AP3 also functions in endocytic retrieval of exocytosed DCVs at the plasma membrane needs further investigation.

C2-domain proteins in the regulated secretory pathway
Several C2 domain-containing proteins function in Ca 2+ -dependent vesicle exocytosis (Südhof, 2012;Pinheiro et al., 2016). We report the first systematic screen of C2 domain-containing proteins in the regulated secretory pathway of neuroendocrine cells. Ionomycin directly mediates Ca 2+ entry so that the screen only identified proteins that exert their actions downstream of Ca 2+ entry. The quality of the screen was verified by multiple controls and by the successful identification of genes for proteins known to function in regulated exocytosis as the top hits CAD PS and UNC13B (Table S2). Other strong hits involved in regulated exocytosis were RIMS1 encoding a Rab3-and Munc13-1-interacting protein (Deng et al., 2011), SYTL1/2 encoding Rab effectors involved in DCV docking (Fukuda, 2013), DOC2A encoding a protein required for DCV exocytosis (Li et al., 2014), and SYT10 encoding a Ca 2+ sensor for peptidergic granule fusion (Cao et al., 2011). These proteins operate directly in the docking, priming, and fusion steps of DCV exocytosis. In contrast, the knockdown of other C2-domain proteins appeared to affect DCV exocytosis indirectly through other trafficking pathways. PLA2 siRNAs strongly reduced the percent secretion in the screen, possibly by affecting cargo protein trafficking to the plasma membrane by altering Golgi tubulation (Ha et al., 2012). A large number of hits likely impacted NPY-Venus secretion by affecting endocytosis (ITSN1, TOL LIP, and MCTP1/2) or endosomal trafficking (RAB11FIP2/5, MYOF, FER1L5, FER1L6, and DYSF; Shin et al., 2005;Tarbutton et al., 2005;Capelluto, 2012;Gubar et al., 2013;Qiu et al., 2015;Redpath et al., 2016). The latter group may function at intersections between the regulated secretory pathway and endosomal trafficking by affecting DCV biogenesis, maturation, or recycling. Lastly, siRNAs directed against several targets (CPNE4/7, PLCB1, PLCL1, PRK CA, CAPN6, and ESYT2/3) likely involve Ca 2+ -dependent signaling that affects posttranslational modifications of proteins and cytoskeletal elements in the regulated secretory pathway (Tonami et al., 2011;Sugiyama et al., 2013;Reinhard et al., 2016;Yu et al., 2016). Overall, the screen provided a foundation for characterizing novel C2 domain-containing proteins for their direct or indirect roles in the regulated secretory pathway.

BAI AP3 controls the fate of DCVs
The screen identified BAI AP3 as a novel gene required for optimal function in the regulated secretory pathway. BAI AP3 is a paralog of Munc13-4, but the tissue expression of these proteins differs, with BAI AP3 expressed predominantly in the hypothalamus and other brain regions, the pituitary, pancreatic islets (this study), and in natural killer and CD8 + T cells, whereas Munc13-4 is predominantly expressed in secretory myeloid cells (Genevestigator, GTExPortal: BAI AP3 or UNC13D). Mouse and human genetic studies suggested a role for BAI AP3 in the nervous system in regulating behavior and food intake (Lauridsen et al., 2011;Wojcik et al., 2013;Mariman et al., 2015). BAI AP3 was also proposed to participate in growth factor secretion by tumor cells (Palmer et al., 2002). Our results provide evidence that BAI AP3 indirectly affects DCVmediated peptide secretion.
The detection of endogenous DCV cargo, especially mature peptide hormones that are only processed within DCVs, in conditioned medium indicated that basal secretion in BON cells was mediated by spontaneous DCV exocytosis (Fig. 2 D  and Fig. S2), as reported for other peptides in endocrine cells (Mains and Eipper, 1984;Sambanis et al., 1990;Matsuuchi and Kelly, 1991;Sirkis et al., 2013). BAI AP3 siRNA appeared to block vesicle maturation, which was reflected by increased VAMP4 retention on DCVs, decreased DCV size, and disrupted DCV trafficking. Immature DCVs accumulated in resting cells over 2-d incubations with BAI AP3 siRNAs as a result of spontaneous DCV exocytosis with suboptimal recycling (Fig. 7, top). When challenged with Ca 2+ /ionomycin, the accumulated DCVs failed to release cargo, resulting in decreased percent secretion. A parsimonious explanation of the results in BON cells is that DCVs accumulated during the 2 d of BAI AP3 siRNA treatment were defective for exocytosis, whereas other DCVs participated in exocytosis, thereby decreasing percent secretion. Thus, BAI AP3 knockdown did not affect exocytosis per se but conferred a loss of function to a pool of DCVs produced during the siRNA incubation. Our results are compatible with a study in chromaffin cells from BAI AP3 knockout mice, indicating that the Ca 2+ -triggered exocytosis of a small rapid release pool of DCVs was not affected (Man et al., 2015), but the status of other DCVs in the chromaffin cells was not assessed.
The fate of DCVs was different in INS-1 cells compared with BON cells after BAI AP3 knockdown (Fig. 7). The greater In resting cells, DCVs undergo maturation and slow spontaneous exocytosis. DCV membrane proteins are recycled to the TGN by retrograde trafficking with the basal activity of BAI AP3. Knockdown of BAI AP3 disrupts protein recycling and may deplete DCV membrane proteins, causing DCV maturation and trafficking defects that lead to immature DCV (iDCVs) accumulation in BON cells or DCV turnover in INS-1 cells. (Bottom) When cells are stimulated with Ca 2+ , increased BAI AP3 activity may stimulate retrograde trafficking to meet the increased demand for protein recycling during Ca 2+ -triggered DCV exocytosis. efficiency of BAI AP3 knockdown in INS-1 cells may be part of the reason. Nonetheless, BAI AP3 knockdown in INS-1 cells caused a dramatic decrease in insulin granules by promoting their lysosomal degradation (Fig. 4). Insulin granules undergo spontaneous exocytosis both in primary β cells (Del Prato et al., 2002) and in INS-1 cells (Fig. 4 G, 1.5 mM glucose). Therefore, BAI AP3 knockdown for 2 d may similarly disrupt the recycling of insulin granule membrane proteins after spontaneous exocytosis. Insulin granules did not accumulate during the 2-d BAI AP3 knockdown, likely because of an efficient stress response in β cells that targets defective insulin granules for crinophagy or microautophagy (Orci et al., 1984;Marsh et al., 2007;Goginashvili et al., 2015). For example, insulin granules lacking Rab3, a protein that recycles from DCVs through an endosomal intermediate to the TGN (Jena et al., 1994;Slembrouck et al., 1999), undergo lysosomal degradation (Marsh et al., 2007). Although the fate of DCVs is different in BON and INS-1 cells, the mechanisms underlying the generation of defective DCVs in cells lacking BAI AP3 are likely similar, involving the Golgi biogenesis of DCVs with an altered balance of membrane proteins such that mature functional DCVs are not generated.

BAI AP3 plays a role in retrograde trafficking
Lack of localization to DCVs implied that BAI AP3 might not play a direct role in DCV exocytosis. Our evidence indicated that endosome-localized BAI AP3 played a role in endosomemediated retrograde trafficking. BAI AP3 knockdown strongly disrupted the steady-state localization of TGN46 as well as TGN integrity, which is similar to the knockdown phenotype of proteins required for retrograde trafficking such as COG and GARP complexes (Pérez-Victoria et al., 2008;Laufman et al., 2011) or the SNA RE proteins VAMP4, STX6, STX16, and Vti1A (Ganley et al., 2008;Shitara et al., 2013). Retrograde trafficking plays an important role in the DCV cycle. When DCVs undergo exocytosis, transmembrane DCV proteins such as phogrin (Vo et al., 2004), peptidylglycine α-amidating monooxygenase (Bäck et al., 2010), carboxypeptidase E (Arnaoutova et al., 2003), VAMP2, and synaptotagmin (Cárdenas and Marengo, 2016) are lost to the plasma membrane and need to be recycled to support new DCV biogenesis (Bauer et al., 2004). Proteins such as STX6, VAMP4, and SYT4 are transient residents on immature DCVs that are recycled to the TGN during DCV maturation (Bonnemaison et al., 2013). The recycling of VAMP4 from DCVs in BON cells was deficient in BAI AP3 knockdown cells. Collectively, the results indicate that BAI AP3 plays an important role in endosome-mediated protein recycling, which indirectly affects the regulated secretory pathway.

Coupling of retro-and anterograde trafficking
Because of the coupling of retrograde and anterograde trafficking, the disruption of retrograde trafficking by BAI AP3 knockdown could deplete essential proteins in the TGN, leading to the generation of defective DCVs that fail to mature or are subjected to degradation by crinophagy or microautophagy. There are several examples where the perturbation of endosomal proteins exert effects on DCV function, such as for the knockdown of endosomal AP3 (Asensio et al., 2010) or the knockout of Vti1A (Walter et al., 2014), which impair DCV biogenesis and maturation. Knockdown of the endosomal proteins VAMP3, Rab11FIP2/5, and ITSN2 also disrupted NPY-Venus secre-tion in our screen. These studies reveal the significance of endosome-mediated retrograde trafficking for DCV-mediated anterograde trafficking.
BAI AP3 contains a CAT CHR (complex associated with tethering containing helical rods) domain found in GARP and COG tethering complexes, which are required for endosome fusion with the TGN (Pei et al., 2009;James and Martin, 2013;Hong and Lev, 2014). Similarly to COG and GARP (Hong and Lev, 2014), BAI AP3 interacts with SNA RE proteins (STX6, STX16, VAMP3, and VAMP4) required for endosome-mediated retrograde trafficking. Thus, BAI AP3 may function in resting cells in parallel with COG and GARP complexes to enable endosome-TGN trafficking especially for DCVs retrieved after spontaneous exocytosis. As a paralog of Munc13-4, BAI AP3 likely also exhibits Ca 2+ -stimulated activity at endosome-TGN interfaces. Indeed, we found that BAI AP3 interactions with TGN SNA REs was enhanced by Ca 2+ and that BAI AP3 was recruited to the TGN in response to Ca 2+ elevations. Because BAI AP3 is expressed in cells with a regulated secretory pathway, Ca 2+ increases could activate the C2 domains of BAI AP3 to enhance retrograde trafficking in concert with the Ca 2+dependent activation of the C2 domains of Munc13-2 and SYT10, which accelerate DCV exocytic trafficking. Thus, BAI AP3 may serve as a key Ca 2+ -dependent effector in retrograde trafficking that functions to meet the increased demand for protein recycling during Ca 2+ -stimulated DCV exocytosis in secretory cells (Fig. 7, bottom).
An alternative possibility is that BAI AP3 controls protein exchange between endosomes and immature DCVs without the need to recycle proteins to the TGN. In this model, transient proteins retrieved from immature DCVs are delivered to endosomes, and DCV membrane proteins recycled from the plasma membrane are delivered to immature DCVs by endosomes. This model was partly supported by studies of Rab2 and its effectors in Caenorhabditis elegans (Edwards et al., 2009;Hannemann et al., 2012;Ailion et al., 2014) and could explain some of our results. The knockdown of BAI AP3 may cause loss of proteins from immature DCVs or a failure to deliver essential proteins for exocytosis. A careful examination of possible two-way traffic between endosomes and DCVs is needed to test this model.
In conclusion, we identified BAI AP3 as essential for optimal function in the regulated secretory pathway (Fig. 7). However, BAI AP3 acts on endosomal pathways to control the fate and activity of DCVs in endocrine and neuroendocrine cells. We propose that BAI AP3 functions in retrograde trafficking to maintain a homeostatic balance of a critical subset of DCV transmembrane proteins during their recycling. BAI AP3 may function similarly to other Munc13 proteins in regulating SNA RE-dependent fusion but, unlike other Munc13 proteins, appears to uniquely regulate membrane fusion at an endosome-TGN interface.
Plasmids. NPY-Venus construct encoding human prepro-NPY (without C peptide) tagged with Venus was inserted into the BamHI/ EcoRI sites of lentiviral vector pWPXL (cat. no. 12257; Addgene). TfR-mCherry lentiviral vector was purchased from Clontech (cat. no. 632580). BAI AP3 was cloned from its cDNA (cat. no. MHS4426-99240349; Dharmacon) and inserted into a lentiviral vector. An HA epitope sequence was added to the C terminus of BAI AP3 before the stop codon. mCherry-BAI AP3 construct was generated by inserting the mCherry in an N-terminal structurally flexible region after E108, and the construct was made with a pEGFP vector. GFP-Rab5 and GFP-Rab11 were generated by inserting canine Rab5 and Rab11 coding sequences into KpnI and BamHI sites of pEGFP-C1. GFP-Rab7 and GFP-Rab9 plasmids were purchased from Addgene (cat. nos. 12605 and 12663) and modified to remove the HA-coding sequence within two KpnI sites. The shRNAs were generated by inserting the seed sequence into pLKO.1 (cat. no. 10878; Addgene) following the vendor's protocol. The seed sequence for rat BAI AP3 is 5′-GAC CGT CCG GTG TCA CTAC-3′; the nontargeting sequence is 5′-CAA CAA GAT GAA GAG CAC CAA-3′.
Lentiviral packaging and stable cell generation 20 µg of lentiviral construct, 6 µg of envelop plasmid (pMD2.G; cat. no.12259; Addgene), and 15 µg of packaging plasmid (psPAX2; cat. no.12260; Addgene) were cotransfected into a 10-cm dish of preseeded HEK293FT cells by calcium phosphate transfection. The medium was replaced twice with viral production medium (30% FBS) 6-8 h and 24 h later. Cell debris and intact cells were pelleted at 1,000 g for 5 min and filtered out (0.45 µm) from conditioned culture medium after 60 h. Viral particles were pelleted by ultracentrifugation at 33,000 rpm for 2 h with a Ti-70 rotor (Beckman Coulter) and resuspended in 100 µl of cold PBS. Concentrated virus was used to transduce BON or INS-1 cells in the presence of 10 µg/ml protamine sulfate (cat. no. 4020; Sigma).
To generate stable cells for the NPY-Venus secretion assay, BON cells were successively transduced with NPY-Venus and TfR-mCherry lentivirus and diluted into 96-well plates. Single cell-derived colonies were expanded and tested for Ca 2+ -stimulated exocytosis. Clone H8 showed robust secretion and was used for the high throughput screen. TfR-mCherry did not undergo stimulated secretion and was used as a quality control parameter. BAI AP3-HA-expressing cells were generated using a similar procedure and used as a pool without clonal selection.
BAI AP3 knockdown INS-1 cells were generated by transducing an shRNA-encoding lentivirus. A nontargeting lentivirus was used as the control. Successfully transduced cells were selected with 0.4 µg/ml puromycin treatment.

NPY-Venus secretion assay
BON Clone H8 cells were seeded in 96-well plates at 1.8 × 10 4 cells/ well and incubated for 48 h. For acute stimulation, cells were washed once with 200 µl PSS-Na (145 mM NaCl, 5.6 mM KCl, 2.2 mM CaCl 2 , 0.5 mM MgCl 2 , 15 mM Hepes, and 5.6 mM glucose) after removing culture medium and incubated in 100 µl PSS-Na/ionomycin (1.25 µM) or PSS-Na/DMSO (same concentration as ionomycin) for 10 min at 37°C. The stimulation buffer was transferred to a black-bottom plate. Cells were lysed with 100 µl PSS-Na/1% Triton X-100, and the cell lysate was transferred to a separate black-bottom plate. The fluorescence of NPY-Venus and TfR-mCherry of both stimulation buffer and cell lysates was determined by a Safire II plate reader (Tecan). Background fluorescence of the buffer was subtracted from the samples. The ratio of NPY-Venus in stimulation buffer and total (stimulation buffer + cell lysate) was used to calculate percent secretion: % secretion =

NPY − Venus in buffer + NPY − Venus in cell lysate × 100 % .
For spontaneous secretion, BON clone H8 cells were seeded in 96-well plates and incubated for 2 d. Culture medium was collected after siRNA treatment, and cells were lysed in the same volume of 1% Triton X-100 in PBS. Then, samples were mixed with loading buffer without boiling and loaded onto SDS-PAGE gels for separation. NPY-Venus was quantified by the fluorescence of Venus under AutoQuant software (GE). PC1 was quantified by quantitative Western blotting. Cell number was determined by counting nuclei after Hoechst 33342 staining before lysing cells. An extended version of the protocol for the NPY-Venus and serotonin secretion assays can be found in Bio-protocol (Zhang and Martin, 2018).
siRNA and plasmid transfection siRNAs were delivered into BON cells by reverse transfection. In brief, siRNA (50 nM final) was mixed with Metafectamine SI (Biontex Laboratories GmbH) or RNAiMAX (Life Technologies) reagent and transfection buffer in 96-well plate wells, and the cell suspension was added (1.8 × 10 4 cells per well). Plasmid was transfected into BON cells by Lipofectamine 2000 reagent (Life Technologies) following the vendor's protocol. 2.5 µg of total plasmid was used for each transfection in 6-well plates or 0.5 µg of total plasmid for 24-well plates. Cells were imaged 24 h after transfection.
High throughput siRNA screen siRNAs for C2 domain-containing proteins were selected from the Dharmacon human siGEN OME SMA RTpool library (GE) and were delivered into cells by Metafectamine SI reagent (Biontex Laboratories GmbH) in 96-well plates. An ionomycin-stimulated NPY-Venus secretion assay was performed 48 h after transfection. The BON NPY-Venus cells (Clone H8) also express a nonsecretable protein, TfR-mCherry. Wells that showed abnormally high TfR-mCherry were excluded from analysis. Nontargeting siRNAs were added to the plate at random positions. Z score was calculated based on the percent secretion of NPY-Venus for each well compared with nontargeting siRNAs. The screen was conducted on three separated occasions, and the mean z score for each gene was used to select hits. The threshold was set as z score = ±2. Cell number was determined by counting nuclei after Hoechst 33342 staining in a separate experiment. The equation to calculate z score was Y siRNA stands for secretion of sample siRNA-treated cells. Y_ and SD_ stand for mean and SD of secretion from control wells.
Generation of esiRNA esiRNA is composed of hundreds of different RNA duplexes with each one comprising only a minor fraction; therefore esiRNAs trigger efficient gene silencing with minimal off-target effect (Kittler et al., 2007). esiRNAs were generated following a previously described protocol (Kittler et al., 2005). In brief, two target regions in the BAI AP3 coding sequence were chosen based on RiDDLE prediction and amplified by PCR. Double-strand RNA of target regions (nucleotides 578-1,047 and 2,516-3,035) were generated by an in vitro transcription assay and treated with Dicer to generate esiRNA. esiRNA was purified by Q-Sepharose, and the quality of final product was verified on a 4% agarose gel.
siRNA rescue studies BON cells stably expressing NPY-Venus were transduced with lentivirus encoding BAI AP3-myc or BAI AP3-myc-rescue. The rescue construct contains six silent mutations for siRNA no. 3 ( Fig. 1 F): 5′-GAG/tcCGTC/gCGT/gTGC/tCAT/cTAC-3′ (underlined nucleotides before the slash are WT, and the nucleotides in lowercase letters after the slash are rescue construct). The resistance of the rescue construct to siRNA knockdown was confirmed by Western blotting. Two cell populations expressing BAI AP3-myc (in ∼49.6% of cells) and BAI AP3myc-rescue (in ∼43.3% of cells) at similar levels were used for the NPY-Venus secretion assay following the same protocols described in the siRNA and plasmid transfection section, except that the final siRNA concentration was 12.5 nM. To count DCVs that reached the TIRF footprint of BON cells, live-cell images of randomly selected cells were captured with a STO RM/TIRF microscope. DCVs were determined by the Find maxima module of ImageJ using the same threshold for all images. Source code is deposited at Github (https ://github .com /kingmanzhang /ImageJ _macro _microscopy; vesicle counting under TIRF.txt).
For the BAI AP3-mCherry recruitment assay, doubly transfected (BAI AP3-mCherry and GFP-VAMP4) BON cells were stimulated with 2.5 µM ionomycin in the presence of 2.2 mM external Ca 2+ . Every 10 s, a dual-channel image of the equatorial section of the cells was captured by a confocal microscope (A1; Nikon). To determine the fluorescence of BAI AP3-mCherry on the Golgi, on the plasma membrane, and in the cytoplasm, CellProfilers was used to create masks for those regions. The GFP channel was used to create a mask for the TGN region. Cell boundary and the nucleus were automatically detected by CellProfiler. Regions within five pixels to cell boundary were regarded as the plasma membrane. Cytoplasmic regions were calculated by subtracting the nucleus, Golgi, and plasma membrane regions from the entire cell. Fluorescent percent changes (ΔF divided by F 0 ) were reported.

Electron microscopy
Cells on coverslips were fixed by 2.5% glutaraldehyde and 2% paraformaldehyde and postfixed by 1% osmium tetroxide and 1% potassium ferricyanide. After fixation, cells were stained with saturated aqueous uranyl acetate and dehydrated in ethanol and propylene oxide. Cells were infiltrated and embedded in Durcupan ACM resin. Cell sections were imaged by a transmission electron microscope (CM120; Philips). The number and size of DCVs in EM micrographs was determined with the Particle Analysis plugin under ImageJ.

Mass spectrometry
BON cells were plated on nine 10-cm dishes. The next day, three plates were replaced with serum-free culture medium (5 ml per plate) after five extensive washes with PBS. After 22 h, serum-free culture medium that contained spontaneously secreted protein was collected from the three plates. For the other six plates, cells were washed extensively and incubated in PSS-Na (5 ml per plate) in the presence of 1.25 µM ionomycin or DMSO (same concentration) for 10 min at 37°C. All samples were centrifuged to remove cell debris, protease inhibitor (Roche) was added, and samples were snap frozen. Samples were separated into protein and peptide fractions by a 10,000-Dalton molecular weight cutoff centrifugal filter (EMD Millipore), and the protein fraction was digested with trypsin. Samples were separated by liquid chromatography and analyzed by tandem mass spectrometry using a Q-Exactive Orbitrap mass spectrometer (Thermo Fisher Scientific). Relative protein abundance was determined by label-free quantification using the median of area under the curve of MS1 peaks (Cox and Mann, 2008;Cox et al., 2011).

INS-1 fractionation
INS-1 cells harvested from five dishes (15 cm) were resuspended in 2 ml KGlu-EGTA (20 mM Hepes, pH 7.1, 120 mM potassium glutamate, 20 mM potassium acetate, and 2 mM EGTA). Cells were passed through a ball-bearing cell homogenizer (4-µm clearance) 18 times. Nuclei and unbroken cells were pelleted by low speed centrifugation at 1,000 g for 10 min. The postnuclear supernatant was subjected to sucrose gradient (0.27-2 M, total volume 11 ml) separation at 114,000 g for 18 h. The gradient was separated into 12 fractions (1 ml each). Proteins in each fraction were concentrated by TCA precipitation and analyzed by Western blotting.

Lipid blot assay
Recombinant His-MBP-BAI AP3 was purified from HEK293FT cell lysates after transfection with pcDNA3-His-MBP-BAI AP3 using Ni-NTA beads. After blocking the lipid blot for 1 h with 5% BSA, PIP strips (Echelon) were incubated with 0.5 µg/ml His-MBP-BAI AP3 in the presence of 10 µM Ca 2+ or 50 µM EGTA for 1 h. After washing three times, bound BAI AP3 was determined with an MBP antibody (cat. no. E8032; New England Biolabs) and HRP-conjugated secondary antibody. His-MBP was used as a control. Three independent experiments were conducted.

Statistical analysis
All statistical analysis was conducted in R software. Significance was determined using Student's t test, unless indicated otherwise. Bar graphs are presented as mean ± SD unless otherwise stated. *, P < 0.05; **, P < 0.01. Figures were generated with Prism (GraphPad) or ggplot2 in R.

Online supplemental material
Fig. S1 shows the lack of Munc13-4 and VAMP2 in BON cells, selected result from the siRNA screen, and BAI AP3 knockdown phenotype by esiRNAs. Fig. S2 shows the identification of CgA-derived peptides from acutely stimulated and resting BON cells. Fig. S3 shows DCV size distribution in BON cells and insulin granule number and size changes in INS-1 cells. Fig. S4 shows the immunofluorescent staining of endogenous BAI AP3 in INS-1 cells. Fig. S5 shows the lipid binding specificity of BAI AP3 in a protein-lipid overlay assay. Fig. S6 shows the conservation of Ca 2+ -binding residues in the C2 domains of BAI AP3. Table S1 lists all human C2 domain-containing proteins in the siRNA screen. Table S2 lists all hits identified from the siRNA screen. Video, related to Fig. 6 F, shows Ca 2+ -stimulated recruitment of BAI AP3 to the Golgi and plasma membrane in BON cells.