An early endosome–derived retrograde trafficking pathway promotes secretory granule maturation

In eukaryotic cells, the secretory pathway is responsible for targeting proteins and lipids to the plasma membrane and the extracellular space. Hence, secretion is essential for cell membrane integrity, cell polarity, and cell-to-cell communication. In conventional secretion, the process is classified into constitutive and regulated secretory pathways. For both pathways, cargo proteins are translated in the ER and trafficked through the Golgi to reach the TGN. At the TGN, cargoes are sorted into secretory vesicles for constitutive secretion or into secretory granules (SGs) for regulated secretion. Unlike constitutive secretion, in which secretory vesicles are rapidly transported to the plasma membrane, cargoes undergoing regulated secretion are stored and processed in SGs that are secreted only on demand.

Professional secretory cells from the endocrine and exocrine systems secrete biologically active molecules through the regulated secretory pathway. In general, cargoes are concentrated at the TGN and bud off as immature SGs, either through aggregation or through the help of various coat and adapter proteins (Asensio et al., 2013; Burgess et al., 2011; Lui-Roberts et al., 2005). Immature SGs then undergo a maturation process through homotypic fusion and remodeling to ensure proper packaging of cargoes and removal of unneeded material (Bonnemaison et al., 2013). Although SGs were initially believed to be solely post-Golgi derived, recent studies suggest that proteins from the endocytic pathway might be required for proper SG biogenesis or function. Endocytic contribution to SG maturation and function has been described in both cultured cell lines and model organisms. In mouse bone marrow–derived mast cells, the early endosome (EE) regulator Rab5 controls SG size and content, whereas Rab27B localizes to SGs and recruits the tetraspanin CD63 from endosomes (Azouz et al., 2014; Klein et al., 2017; Mizuno et al., 2007). In rat pancreatic acinar cells, unwanted content of immature zymogen granules enters the endolysosomal system during remodeling and maturation in a Rab5- and tumor protein D52–dependent manner (Messenger et al., 2013). Dense core vesicles in Caenorhabditis elegans neurons require Rab2 and retrograde transport from endosome-associated recycling protein (EARP) complex to mature, and loss of Rab5 or Rab10 activity leads to defects in regulated secretion of dense core vesicles (Ailion et al., 2014; Edwards et al., 2009; Sasidharan et al., 2012; Topalidou et al., 2016). In addition, mucocysts from Tetrahymena thermophila need endosomal class C core vacuole/endosome tethering (CORVET) complex subunits for proper maturation (Briguglio et al., 2013; Kontur et al., 2016; Sparvoli et al., 2018).

The larval salivary glands of Drosophila melanogaster are a highly accessible genetic model system for studying SG biogenesis (Biyasheva et al., 2001; Burgess et al., 2011, 2012; Torres et al., 2014). During early larval development, the salivary gland produces saliva to help moisten food for feeding (Goldstein and Fyrberg, 1994). 24 h after entering the third instar larval stage (L3), the larvae stop feeding and begin production of glue proteins that are packaged into SGs known as glue granules. These SGs mature over the next 18 h and are released in response to a pulse of the hormone ecdysone to stick the pupa onto a solid surface during metamorphosis (Biyasheva et al., 2001; Burgess et al., 2011). Mature SGs are two- to fourfold larger in cross-sectional area than immature SGs, enabling us to employ the fluorescently tagged cargo protein Sgs3-DsRed to visually identify genes required for proper SG maturation. Using this system, we previously demonstrated the importance of type II phosphatidylinositol 4-kinase (PI4KII) in this process. Loss of PI4KII results in failure of SG maturation and defects in late endosome (LE) trafficking, suggesting a potential endocytic contribution to SG maturation via PI4KII activity (Burgess et al., 2012).

PI4KII is one of three PI4Ks in Drosophila. PI4Ks phosphorylate PI to form PI 4-phosphate (PI4P), which serves as the precursor to PIP2 and PIP3 and is a crucial signaling molecule on cellular membranes (Tan and Brill, 2014). Unlike the type III PI4Ks, PI4KIIIα and PI4KIIIβ, PI4KII is palmitoylated and inserted in the cytoplasmic leaflet of organelle membranes (Jung et al., 2008). The three PI4Ks produce PI4P on different cellular membranes (Tan and Brill, 2014). PI4KIIIα produces plasma membrane PI4P, whereas PI4KIIIβ (called Four wheel drive [Fwd] in Drosophila) synthesizes Golgi PI4P (Polevoy et al., 2009; Tan et al., 2014). PI4KII localizes to the TGN and endosomes, where its catalytic activity is needed for formation of endosomal tubules (Burgess et al., 2012).

In this study, we aimed to discover potential PI4KII mediators involved in SG maturation through a targeted RNAi screen. This screen revealed that post-Golgi–derived SGs require an endocytic contribution to mature. We showed that CD63, an endolysosomal protein that decorates lysosome-related organelles (LROs; Raposo et al., 2007), localizes to the limiting membrane of SGs and regulates SG size. Taking advantage of CD63’s ability to promote SG maturation, we performed a secondary screen with CD63 and discovered that PI4KII, Syntaxin 16 (Syx16), and Past1, a PI4P effector (Jović et al., 2009; Koles et al., 2015), act in the same pathway. Moreover, Past1 and Syx16 are essential for PI4KII-mediated endosomal tubulation that is associated with accumulation of PI4P on mature SGs. In summary, we show that Past1, Syx16, and PI4KII support an EE–to–TGN retrograde trafficking pathway that is needed for SG maturation. Our results indicate that SG maturation in the larval salivary gland is surprisingly complex and may share similarity with a subset of LROs.

We previously demonstrated that PI4KII, but not Fwd, is required for SG maturation (Burgess et al., 2012). Because PI4KII and Fwd localize near each other at the TGN and Golgi, we hypothesized that an endosomal defect due to loss of PI4KII might be the underlying cause for failed SG maturation. To identify potential regulators of PI4KII-mediated SG maturation, we screened 254 genes involved in intracellular trafficking, cell polarity, or cytoskeletal dynamics by RNAi (Table S1 A). To induce RNAi, genetically encoded haripin RNAs specific to each gene were expressed in developing salivary gland cells via the galactose metabolism 4/upstream activating sequence (GAL4–UAS) system using the AB1-GAL4 driver (Burgess et al., 2011; Costantino et al., 2008; see Materials and methods). The severity of the effect of gene knockdown on SG biogenesis was scored qualitatively on a scale of 0 to 4 based on the size of the resulting SGs revealed by the fluorescently tagged cargo protein Sgs3-DsRed (Fig. 1 A), where 0 resembled WT controls, 1 showed a slight reduction in SG size, and 4 exhibited severe defects in the production of nascent SGs. Knockdowns resulting in a score of 2 or 3 exhibited SGs that were small enough to be considered as positive hits. More than half (48 of 80) of the positive hits encode proteins involved in endolysosomal trafficking or function (Fig. 1 G). It is possible that this list includes false positives, as we did not verify the specificity of the RNAi lines. However, many of the genes, including those discussed below, were tested with multiple RNAi lines targeting the same gene, showed phenotypes similar to other subunits of the same protein complex, or were verified with null mutants. These include genes encoding the SNAREs Syx16 and Snap29, which localize to both Golgi and endosomal compartments (Csizmadia et al., 2018; Morelli et al., 2014; Xu et al., 2002); the GTPase Rab5, which regulates trafficking at EE; and the ADP-ribosylation factor-like protein Arl1 and the Eps15 homology domain-containing protein 1 (EHD1) orthologue Past1, which are involved in EE–to–TGN retrograde trafficking (Fig. 1, B–F; Christis and Munro, 2012; D’Souza et al., 2014; McKenzie et al., 2012; Sharma et al., 2009). Thus, SG maturation requires endosomal trafficking proteins.

Because of the large contribution of endocytic genes, we hypothesized that SG maturation might employ pathways similar to those involved in LRO maturation. Although most LROs originate from the endolysosomal system, a subset of LROs, namely Weibel-Palade bodies (WPBs), originate from the TGN and acquire endocytic material during maturation (Marks et al., 2013). To investigate whether SGs share similar properties to these LROs, we expressed a well-characterized LRO marker, fluorescently tagged mouse tetraspanin CD63 (CD63-GFP). In contrast to controls lacking CD63-GFP expression, cells expressing one copy of the CD63-GFP transgene had larger SGs, and CD63-GFP localized to the membranes of SGs containing the fluorescent cargo protein Sgs3-DsRed (Fig. 2, A–B′′). In addition, SGs became abnormally large when CD63-GFP was expressed at higher levels using two copies of the transgene (Fig. 2, C–C′′). Quantification of the cross-sectional area of these SGs revealed a positive correlation between CD63-GFP expression level and SG size (Fig. 2 D; see Materials and methods). As CD63-GFP expression levels increased, the distribution of SG size shifted toward SGs >15 µm² in cross-sectional area (Fig. 2 E). The ability of CD63-GFP to promote SG maturation was not an artifact of expression of the mammalian protein, as overexpression of the Drosophila CD63 homologue Tsp29Fa also increased SG size (Fig. 2, F–I). To examine whether CD63 and Tsp29Fa act similarly in this process, colocalization of the two proteins was examined by coexpressing and immunostaining CD63-GFP and Tsp29Fa-Flag-HA in early and late L3 salivary glands (Fig. S1, A–B′′). We performed Pearson’s correlation coefficient and Mander’s overlap coefficient analyses. Pearson’s measures the correlation between the fluorescence intensity of two proteins based on their overlapping pixels, and Mander’s measures the degree of overlap of one protein with another protein. In early L3 salivary gland cells, Tsp29Fa-Flag-HA distribution correlated with CD63-GFP (Fig. S1 C, Pearson’s = 0.43 ± 0.13). In late L3 salivary gland cells, CD63-GFP and Tsp29Fa-Flag-HA showed weaker correlation (Fig. S1 D, Pearson’s = 0.22 ± 0.041), but the two proteins exhibited strong cooccurrence on SG membranes (Mander’s CD63 = 0.51 ± 0.084, Mander’s Tsp29Fa = 0.47 ± 0.14). Moreover, cooverexpression of one copy of Tsp29Fa-HA and one copy of CD63-GFP resulted in abnormally large SGs, similar to overexpressing two copies of CD63-GFP (Fig. S1 E) and further indicating that Tsp29Fa and CD63 share a conserved function.

To determine how CD63-GFP and Tsp29Fa overexpression affects intracellular transport, we examined markers for secretory and endosomal trafficking. To examine secretion, we expressed fluorescently tagged mouse transmembrane glycoprotein CD8-RFP, which is delivered to the apical membrane in control salivary gland cells. When coexpressed, both CD63-GFP and CD8-RFP localized to the apical membrane (Fig. S1 F), indicating that the increase in SG size is not caused by a failure in apical secretion. To investigate endosomal transport, we compared the distribution of the EE protein Hrs in control cells with no Tsp29Fa-Flag-HA expression to cells overexpressing Tsp29Fa-Flag-HA (Fig. S1 G). Tsp29Fa-Flag-HA colocalized with Hrs and shifted Hrs toward the cup-like structure formed by the cis-Golgin Lava lamp (Lva; Fig. S1 G). The localization of this compartment is reminiscent of the TGN (Burgess et al., 2011), suggesting that Tsp29Fa overexpression might influence EE–to–TGN retrograde trafficking.

To evaluate whether endogenous tetraspanins play a role in SG maturation, we knocked down Tsp29Fa expression using RNAi. SGs in Tsp29Fa knockdowns were significantly smaller than in controls (Fig. 3, A, B, and G), and expression of CD63-GFP suppressed the Tsp29Fa RNAi phenotype (Fig. 3, C and G). To further confirm the role of Tsp29Fa in SG maturation, we generated two deletion mutants using CRISPR-Cas9 gene editing (Fig. S2 A). The first mutant, Tsp29Fa^J1a, is a predicted null allele of Tsp29Fa, and the second mutant, Df(2L)L7a, deletes nearly all of the coding sequences of Tsp29Fa and the neighboring tetraspanin Tsp29Fb (Fig. S2, B and C). Tsp29Fb has weaker alignment score with CD63 compared with Tsp29Fa (CD63 vs. Tsp29Fa E-value = 5.0 × 10⁻⁴¹; CD63 vs. Tsp29Fb E-value = 5.0 × 10⁻³²). However, both proteins are highly conserved in their transmembrane domains (Fig. S2, B and C). SGs from Tsp29Fa^J1a homozygotes had a mild but significant defect in SG maturation (Fig. 3, D, E, and H), whereas SGs from Df(2L)L7a homozygotes were even smaller (Fig. 3, E, F, and H). To address whether Tsp29Fb expression is up-regulated in response to loss of Tsp29Fa, quantitative PCR was performed on WT and Tsp29Fa^J1a L3 larvae, and no significant change in Tsp29Fb RNA expression level was observed (Fig. S2 D). Thus, CD63 family members share the ability to promote SG maturation.

To identify genes required for CD63 function in SG maturation, we expressed CD63-GFP in the background of different RNAi knockdowns from our initial screen (Table S1 B). Expression of CD63-GFP suppressed the knockdown phenotypes of some but not all of the genes (Fig. 4, A–F′). Altogether, 21 of 37 knockdowns tested were suppressed by CD63-GFP (Table S1 B). For example, the knockdown phenotype of clathrin adapter subunit AP-2σ was suppressed (Fig. 4, A and A′), and EE regulator Rab5 appeared to be partially suppressed (Fig. 4, B and B′) by CD63-GFP expression. Notably, expression of CD63-GFP had no effect on SG maturation in the remaining knockdowns tested (16 of 37). These included genes involved in retrograde trafficking such as Syx16, Past1, and Arl1. Localization of CD63-GFP in knockdowns of Syx16 and Past1 (Fig. 4, E–F′) resembled CD63-GFP localization in PI4KII null mutants (PI4KIIΔ; Fig. 4, D and D′). CD63-GFP remained on small SGs and decorated enlarged endosomes that often contained a faint Sgs3-DsRed signal. This similarity suggested that Syx16, Past1, and PI4KII might act in the same pathway, and that endosome–to–TGN retrograde trafficking might be required for CD63 to promote SG maturation. In contrast to Syx16, Past1, and PI4KII, the defects in knockdowns of Arl1 were enhanced by CD63-GFP expression, revealing accumulation of CD63 on extremely large (>5-µm–diameter) endosomes (Fig. 4, C and C′). Because many of the nonsuppressible genes act at EEs or in retrograde trafficking, it is likely that retrograde trafficking from EEs, rather than LEs, promotes SG maturation.

Based on our genetic analysis, we suspected that PI4KII, Past1, and Syx16 mediate EE–to–TGN retrograde trafficking to promote SG maturation. To investigate whether loss of PI4KII affects EE function, we expressed the PI3P binding EE probe GFP-2xFYVE (Gillooly et al., 2000; Jean et al., 2012; Juhász et al., 2008) in WT and PI4KIIΔ salivary glands. In controls, GFP-2xFYVE localized to small and large puncta near SGs marked by Sgs3-DsRed (Fig. 5 A). In PI4KIIΔ, GFP-2xFYVE marked additional enlarged vesicular structures that were occasionally filled with Sgs3-DsRed (Fig. 5 B). Counting the number of hollow endosomes in WT and PI4KIIΔ showed that their occurrence was significantly higher in PI4KIIΔ (Fig. 5 C). To test whether PI3P normally accumulates on immature SGs, early L3 salivary glands were examined, and no PI3P was observed on immature SGs of either WT or PI4KIIΔ cells (Fig. S3 B). To further evaluate the role of PI4KII at EE, we coexpressed GFP-2xFYVE with a kinase-dead PI4KII tagged with mCherry (mCh-PI4KII^K311M; Burgess et al., 2012). In controls, mCh-PI4KII^K311M partially colocalized with GFP-2xFYVE on endosomes and decorated tubular structures extending from these endosomes that lacked the GFP-2xFYVE signal (Fig. 5 D). In PI4KIIΔ, mCh-PI4KII^K311M partially colocalized with GFP-2xFYVE on enlarged endosomes that exhibited significantly fewer and less extensive mCh-PI4KII^K311M–positive tubules (Fig. 5, E and F). These observations indicate that PI4KII activity is required for proper EE organization.

To investigate whether trafficking of CD63-GFP occurs via PI4KII tubules, we coexpressed CD63-GFP and mCh-PI4KII (Fig. 6, A–A′′; and Fig. S4 A). No correlation was observed between CD63-GFP and mCh-PI4KII (Fig. 6 B, Pearson’s = −0.053 ± 0.025). However, some overlap was observed, as mCh-PI4KII tubules occasionally formed contacts with CD63-GFP–positive SG membranes or endosomes (Fig. 6 B, Mander’s PI4KII = 0.36 ± 0.11). When we examined the localization of mCh-PI4KII with CD63-GFP in the PI4KIIΔ mutant background, it appeared that mCh-PI4KII also localized to SGs and partially decorated SG membranes (Fig. S3 B), raising the possibility that mCh-PI4KII localizes less efficiently to SGs than the endogenous, untagged PI4KII. In contrast, coexpressing kinase-dead mCh-PI4KII^D465A with CD63-GFP in the PI4KIIΔ mutant background resulted in concentration of mCh-PI4KII^D465A signal inside CD63-GFP–positive organelles (Fig. S3 B), suggesting that kinase-dead PI4KII is sorted into multivesicular bodies when CD63-GFP is expressed in the absence of WT PI4KII. CD63-GFP localized to the limiting membranes of small SGs in PI4KIIΔ mutants and knockdowns of Syx16 and Past1 (Fig. 4, D′, E′, and F′), raising the question of how SGs acquired CD63-GFP if the retrograde trafficking pathway was compromised. We hypothesized that immature SGs would acquire some CD63-GFP as they bud from the TGN, because CD63 is normally secreted to the plasma membrane before entering the endosomal system (Duffield et al., 2003; Pols and Klumperman, 2009; Schulze et al., 2017; Yu et al., 2016). This is likely the case, as we observed CD63-GFP on immature SG membranes in early L3 salivary gland cells from both WT controls and PI4KIIΔ mutants (Fig. S3 C).

Because CD63-GFP and mCh-PI4KII showed poor colocalization, we decided to investigate whether the lipid product of PI4KII colocalizes with CD63-GFP. To test this, we expressed CD63-GFP with the pan-PI4P probe mCh-2xP4M (see Materials and methods; Hammond et al., 2014) in WT salivary gland cells. Both markers localized to SG membranes and endosomes with good correlation (Fig. 6, C and D; and Fig. S4 A, Pearson’s = 0.41 ± 0.067). In contrast, distribution of GFP-PI4KII and mCh-2xP4M showed a weaker correlation, but PI4KII overlapped significantly with mCh-2xP4M (Fig. 6, E and F; and Fig. S4 A, Pearson’s = 0.32 ± 0.085, Mander’s PI4KII = 0.90 ± 0.036, Mander’s 2xP4M = 0.36 ± 0.050). The lower overlap of 2xP4M with PI4KII was expected, because P4M labels cellular PI4P ubiquitously and is not restricted to PI4P synthesized by PI4KII. To determine whether PI4P distribution on SG membranes depends on PI4KII activity, we coexpressed Sgs3-GFP with mCh-2xP4M in WT and PI4KIIΔ salivary gland cells. In control cells, mCh-2xP4M decorated most of the SGs marked by Sgs3-GFP (Fig. 6, G–G′′), whereas in PI4KIIΔ cells, the number of mCh-2xP4M–positive SGs was significantly reduced (Fig. 6, H and I; and Fig. S4 B). Hence, mCh-2xP4M colocalizes strongly with CD63-GFP on SG membranes, and this association requires PI4KII.

Our CD63-GFP secondary screen results suggested that Past1 and Syx16 might act in the same pathway as PI4KII to promote SG maturation. Past1 is important for the initiation and dynamics of EE membrane tubules and is a potential PI4P-binding protein (Jović et al., 2009; Koles et al., 2015). Thus, we evaluated the effect of a Past1 null mutant (Past1^110-1) on PI4KII tubule formation. Tubules marked by GFP-PI4KII were less extensive in Past1^110-1 salivary gland cells, and GFP-PI4KII decorated abundant enlarged organelles that were absent in controls (Fig. 7, A and B; and Fig. S4 C). Knockdown of Syx16 caused a similar phenotype (Fig. 7 C and Fig. S4 C). To examine whether loss of Past1 also affected the dynamics of PI4KII-positive tubules, we performed live imaging of GFP-PI4KII in control and Past1^110-1 salivary glands (Video 1). These experiments revealed that GFP-PI4KII tubules in Past1^110-1 extended and retracted significantly faster than in controls (Fig. 7, D–F; *, P = 0.002). To assess whether loss of Past1 compromised PI4KII-mediated retrograde trafficking that is important for SG maturation, we examined changes in PI4P distribution in Past1^110-1 salivary gland cells. Localization of mCh-2xP4M to SGs was reduced in Past1^110-1 mutants (Fig. 7, G–H′′; and Fig. S4 D), similar to PI4KIIΔ mutants. In addition, mCh-2xP4M accumulated on endosomes in Past1^110-1 mutants, but only rarely in PI4KIIΔ mutants (compare Fig. 7, G–H′′ with Fig. 6, G–H′′). Thus, Past1 (and likely Syx16) is required for the abundance and stability of PI4KII tubules that contribute to accumulation of PI4P on mature SGs, which, in turn, is important for CD63 to promote SG maturation.

Regulated SGs are thought to mature, in part, through homotypic fusion (Farkaš and Šutáková, 1998; Burgess et al., 2011), yet how this maturation is regulated has remained poorly understood. In this study, we provide evidence that post-Golgi–derived immature SGs require input from the endocytic pathway to fully mature. In particular, we establish that retrograde transport from EEs via PI4KII, Past1, and Syx16 is important for CD63-mediated SG maturation (Fig. 8 A). PI4KII is required for normal EE–TGN retrograde trafficking, and loss of PI4KII results in small SGs and enlarged EEs that contain mistrafficked Sgs3 cargo proteins. Also, loss of PI4KII leads to reduced levels of PI4P on both SG and EE membranes (Fig. 8 B). Past1 and Syx16 are needed for PI4KII activity in retrograde trafficking, which leads to similar phenotypes for SGs and EEs. However, PI4KII remains active in the EEs and increases PI4P levels at EEs (Fig. 8 C).

Because we observed defects in both SG maturation and LE dynamics in PI4KIIΔ mutants (Burgess et al., 2012), we included a large number of endosomal genes in our initial RNAi screen to identify genes that might act in the same pathway as PI4KII. Many of the positive hits are involved in EE sorting, EE–LE transition, or retrograde trafficking, whereas most of the LE genes tested were negative. The lack of positive LE hits suggests that EE sorting is more important to SG maturation than LE-dependent transport. Moreover, not all retrograde trafficking pathways are critical for SG maturation. For example, none of the retromer or AP-3 subunits tested was required for SG maturation, consistent with our previous observation that these were dispensable for the formation of PI4KII tubules (Burgess et al., 2012).

CD63 was originally named granulophysin and is found on various types of SGs in human tissues (Hatskelzon et al., 1993; Nishibori et al., 1993). However, it was unclear whether CD63 plays a functional role in the biogenesis of these SGs or simply decorates SG membranes. Our results identify a positive correlation between CD63 expression level and SG size. This correlation is not a result of compromised exocytosis, as transport of CD8 to the apical membrane is unaffected by CD63 expression. CD63 and Drosophila Tsp29Fa are functionally conserved; they localize to SG membranes and act synergistically to promote SG maturation. Although Tsp29Fa and CD63 overlap in their distribution on SG membranes, their correlation is relatively weak, suggesting they might localize to different subdomains. Interestingly, overexpression of Tsp29Fa results in increased localization of EEs in proximity to the TGN, suggesting that overexpression of Tsp29Fa and CD63 may alter EE sorting and trafficking in a manner that facilitates retrograde transport to the Golgi. Indeed, it is possible that overexpression of CD63 promotes SG maturation by up-regulating the retrograde trafficking pathway. Although it remains uncertain how CD63 acts in this process, our results demonstrate that CD63 not only decorates SG membranes but also plays a functional role in SG maturation.

Loss of Tsp29Fa does not affect SG maturation to the same extent as in PI4KIIΔ or Past1^110-1 mutants. The defect in SG maturation is stronger when both Tsp29Fa and Tsp29Fb are deleted in the Df(2L)L7a mutant. This suggests that Tsp29Fa and Tsp29Fb may be partially redundant. Because tetraspanins can associate to form membrane microdomains (Perez-Hernandez et al., 2013; Verweij et al., 2011), it is likely that that loss of a single tetraspanin could be compensated for by repurposing other tetraspanins. Thus, most animal models of tetraspanin deficiency exhibit relatively mild phenotypes (Geisert et al., 2002; Knobeloch et al., 2000; Maecker and Levy, 1997; Miyazaki et al., 1997; Tarrant et al., 2002). For example, even though CD63 is well characterized for it is importance in melanosome biogenesis, CD63-deficient mice show pronounced pigmentation defects only in retinal pigment epithelia and not other tissues (van Niel et al., 2011). Moreover, Drosophila has other putative CD63 homologues, namely Tsp39D and Tsp47F. Unlike Tsp29Fa and Tsp29Fb, which are highly expressed in the gut and salivary gland, Tsp39D is highly expressed in the central nervous system, and Tsp47F is expressed at very low levels in the adult head. Tsp39D knockdown resulted in small SGs, yet this phenotype is not fully suppressed by expression of CD63-GFP, indicating that the role of Tsp39D in SG maturation is distinct from Tsp29Fa. On the other hand, Tsp39D expression is crucial for maintenance of synaptic vesicles in Drosophila photoreceptor cells (Tsai et al., 2019), suggesting that CD63 homologues might carry out analogous functions in regulated secretion in different tissues.

Expression of CD63-GFP in RNAi knockdowns that lead to a small granule phenotype provides a sensitive system to identify pathways required for CD63/Tsp29Fa-mediated SG maturation. Suppression of the small granule phenotype could be due to the ability of CD63 to stimulate trafficking through a pathway that is only partially compromised by the RNAi knockdown or to entirely bypass the function of the target gene. For example, CD63 has been reported to increase endocytic flux of H⁺/K⁺-ATPase in Cos-7 cells and promotes hemocyte-mediated phagocytosis in clams (Duffield et al., 2003; Yu et al., 2016). Thus, CD63-GFP expression may suppress the AP-2σ RNAi phenotype by restoring endocytic flux that is important for proper SG maturation. On the other hand, failure of CD63-GFP expression to suppress the RNAi phenotype indicates that the pathway being altered is necessary for CD63-mediated SG maturation. RNAi knockdowns of many EE genes exhibited either weak or no suppression by CD63-GFP, revealing that proper EE sorting is critical for CD63-mediated SG maturation. Moreover, CD63-GFP expression failed to suppress knockdowns of many retrograde trafficking genes, suggesting it is likely that normal EE–to–TGN retrograde trafficking is important for proper SG maturation. Expression of CD63-GFP in the background of Syx16 and Past1 knockdowns resembled CD63-GFP expression in PI4KIIΔ mutants. CD63-GFP decorated enlarged endosomes that occasionally contained faint Sgs3-DsRed signal. The faint signal could be a result of impairment in endosomal degradation or expansion of organelle volume. Interestingly, expression of CD63-GFP in the knockdowns of Arl1 and subunits of Golgi-associated retrograde protein complex (GARP) resulted in abnormally large endosomes. It is possible that knockdowns of Arl1 and GARP complex subunits result in blockade of multiple retrograde trafficking pathways and that expression of CD63 leads to fusion of the resulting enlarged endosomes.

Prior biochemical studies have suggested that CD63 and PI4KII may act in concert. PI4KII activity is present in immunoprecipitates of CD63 from human cultured cells (Berditchevski et al., 1997; Yauch and Hemler, 2000). Nonetheless, the physiological consequences of this interaction have remained unclear. Although we were unsuccessful at coimmunoprecipitating CD63 with PI4KII (not depicted) and we did not directly observe CD63 on PI4KII-positive tubules, our data provide evidence that CD63 requires PI4KII activity to promote SG maturation. For example, PI4KII tubules formed contacts with CD63-positive SG membranes in WT salivary glands. Moreover, mCh-PI4KII decorated SG membranes together with CD63-GFP when coexpressed in PI4KIIΔ mutants. It is possible that mCh-PI4KII weakly localizes to SG membranes and that this is more apparent in the absence of endogenous PI4KII. Furthermore, strong colocalization was observed between CD63 and PI4P, the enzymatic product of PI4KII, on SG membranes. Generation of PI4P on SG membranes was strongly dependent on PI4KII activity, as loss of PI4KII significantly reduced the number of PI4P-positive SGs. In mammalian systems, PI4KII has been reported to associate with SGs, but it is unknown whether PI4KII synthesizes PI4P on SG membranes (Guo et al., 2003; Kristiansen et al., 1998; Panaretou and Tooze, 2002; Del Vecchio and Pilch, 1991; Xu et al., 2006). Moreover, the role of PI4P on SG membranes is unclear. One possibility is that proper levels of PI4P might be required for recruitment of PI4P-binding adapter proteins to facilitate remodeling of SGs during maturation (Mills et al., 2003; Wang et al., 2007, 2003). Another possibility is that CD63 requires PI4P to promote SG maturation.

We demonstrate that Past1 and Syx16 are responsible for the formation and stability of PI4KII-positive tubules. Loss of PI4KII-positive tubules results in an accumulation of PI4P on endosomes, reduction of PI4P on SGs, and failure of SG maturation. Mammalian EHD proteins are associated with endosomal tubulation and PI4P binding (Jović et al., 2009). However, it was initially unclear whether these proteins tubulate membranes, stabilize membrane tubules, or promote tubule vesiculation (Sharma et al., 2009). Indeed, EHD1 is needed for tubule vesiculation, whereas the related family member EHD3 promotes membrane tubulation and tubule stabilization (Bahl et al., 2016; Cai et al., 2013). In terms of sequence homology, Drosophila Past1 is more similar to human EHD1 (Gramates et al., 2017). However, the requirement for Past1 in tubulation and stabilization of PI4KII-positive endosomal tubules reveals functional similarity to EHD3, at least in the context of the larval salivary gland.

PI4KII orthologues were recently shown to be required for the maturation of WPBs, which, like large dense core vesicles, are post-Golgi–derived granules that are often considered as part of the LRO family because they require endocytic trafficking during their biogenesis. However, although AP-3 is required for targeting CD63 to mature WPBs and large dense core vesicles (Grabner et al., 2006; Harrison-Lavoie et al., 2006), AP-3 is dispensable for glue granule maturation (Burgess et al., 2012; this study). Nevertheless, the requirement for CD63- and PI4KII-dependent retrograde trafficking indicates that glue granules are among a subset of SGs that require endosomal trafficking (Ailion et al., 2014; Azouz et al., 2014; Cattin-Ortolá et al., 2017; Klein et al., 2017; Sparvoli et al., 2018; Topalidou et al., 2016).

In conclusion, our work provides additional evidence that SG maturation is not isolated in the post-Golgi compartment and requires endocytic contribution. We demonstrate that Past1 and Syx16 regulate PI4KII-mediated retrograde transport that is critical for CD63 to promote SG maturation. It will be of interest to determine if endocytic contribution is a general feature of SG maturation. Importantly, defects in SG maturation could result in improper packaging of cargoes such as hormones, neurotransmitters, and mucins that may lead to physiological disorders. Therefore, unraveling mechanisms required for SG maturation could provide novel therapeutic insight into diseases associated with SG dysfunction.

Flies were cultured on standard cornmeal molasses agar at 25°C (Ashburner, 1989). RNAi stocks expressing hairpin RNAs directed against individual genes were acquired from Bloomington Drosophila Stock Center (BDSC) or Vienna Drosophila RNAi Center (VDRC) as noted (Table S1). Additional stocks included P{GawB}AB1-GAL4 (BDSC); P{w⁺, CG14671}, P{neoFRT}82B, Df(3R)730 (PI4KIIΔ; Burgess et al., 2012); w¹¹¹⁸;;Past1^110-1 (Olswang-Kutz et al., 2009; a gift from M. Horowitz, Tel Aviv University, Tel Aviv, Israel); P{w⁺, UAS-CD63-EGFP} (Corrigan et al., 2014; Panáková et al., 2005; Redhai et al., 2016; a gift from C. Wilson, University of Oxford, Oxford, UK); P{w+, UAS-GFP-myc-2xFYVE} (BDSC); P{w+, UAS-mCherry-myc-2xFYVE} (Jean et al., 2012; a gift from A. Kiger, University of California San Diego, San Diego, CA); P{w⁺, Sgs3-DsRed} (Costantino et al., 2008); P{w⁺, Sgs3-GFP} (Biyasheva et al., 2001); P{w⁺, UAS-CD8-EGFP} (BDSC); P{w⁺, UAS-CD8-RFP} (BDSC); and all of the PI4KII transgenic stocks were previously described. P{w⁺, αTub84B-GFP-PI4KII}; P{w⁺, αTub84B-mCh-PI4KII}; P{w⁺, αTub84B-mCh-PI4KII^ATP}; and P{w⁺, αTub84B-mCh-PI4KII^CAT} transgenic lines were previously described (Burgess et al., 2012). Note that mCh-PI4KII^K311M (ATP-binding mutant) and mCh-PI4KII^D465A (catalytic site mutant) were previously designated mChPI4KII^ATP and mCh-PI4KII^CAT. Both mutations result in a catalytically inactive (kinase-dead) PI4KII enzyme in vitro and behave similarly in rescue experiments (Burgess et al., 2012). P{w⁺, Sgs3-DsRed} was recombined with P{w⁺, UAS-CD63-EGFP} for CD63 coexpression experiments. UAS lines were expressed in salivary gland cells under control of the AB1-GAL4 driver (P{GawB}AB1; BDSC).

For the Tsp29Fa target sequence (5′-GGA​GTT​GGC​GCC​GTC​TAT​AC[GGG]-3′), the forward gRNA was 5′-CTT​CGG​AGT​TGG​CGC​CGT​CTA​TAC-3′, and the reverse was 5′-AAA​CGT​ATA​GAC​GGC​GCC​AAC​TCC-3′. For the Tsp29Fb target sequence (5′-GCA​GTT​GGA​TAA​CGG​AGA​GC[AGG]-3′), the forward gRNA was 5′-CTT​CGC​AGT​TGG​ATA​ACG​GAG​AGC-3′, and the reverse was 5′-AAA​CGC​TCT​CCG​TTA​TCC​AAC​TGC-3′. gRNA oligonucleotides from Thermo Fisher Scientific were annealed and subcloned into pU6-BbsI-chiRNA (Addgene plasmid #45946).

Individual CRISPR-Cas9 lines were genotyped using primers flanking Tsp29Fa and Tsp29Fb. The Tsp29Fa forward primer sequence was 5′-GTG​AAG​TAT​ACG​CTT​TTC​GG-3′, and the reverse was 5′-CGG​ATT​TTA​ATC​TCA​CGT​GC-3′. The Tsp29Fb forward primer sequence was 5′-CGC​CAA​GTA​CAT​GCT​CAT​C-3′, and the reverse was 5′-GTG​TAG​GTG​TAG​GGT​TCG​TG-3′. Primers were synthesized by Thermo Fisher Scientific.

P4M is the PI4P-binding domain of the Legionella protein SidM (Hammond et al., 2014). 2xP4M sequences were PCR-amplified from a mammalian vector containing GFP-2xP4M (Addgene plasmid #51472) and subcloned into a modified pCaSpeR4 vector containing the αTub84B promoter (Marois et al., 2006) and mCherry. The forward primer sequence was 5′-GCT​CTA​GAA​ATT​CTG​GTG​GTA​GTG​CTT​C-3′, and the reverse was 5′-GCT​CTA​GAG​CAT​TCA​TTT​TAT​GTT​TCA​GG-3′. Primers were synthesized by Thermo Fisher Scientific.

RNA was extracted from 10 w¹¹¹⁸ and Tsp29Fa^J1a late L3 larvae using PureLink RNA Mini Kit (Thermo Fisher Scientific). cDNA library was generated using SensiFAST cDNA Synthesis Kit (Bioline). qRT-PCR was performed using PowerUp SYBR Green Master Mix (Thermo Fisher Scientific) on a ViiA 7 Real-Time PCR System (Thermo Fisher Scientific). Forward primer for α-tubulin at 84B was 5′-TGT​CGC​GTG​TGA​AAC​ACT​TC-3′. Reverse primer for α-tubulin at 84B was 5′-AGC​AGG​CGT​TTC​CAA​TCT​G-3′ (Ponton et al., 2011). Tsp29Fa and Tsp29Fb primer sequences were obtained from FlyPrimerBank (Harvard Medical School, Cambridge, MA). Primers for Tsp29Fa were forward, 5′-AGC​GCC​AAT​GCT​GTG​AAG​TAT-3′, and reverse, 5′-GCT​TAT​AGC​CCG​TAT​AGA​CGG​C-3′. Primers for Tsp29Fb were forward, 5′-CCA​TCG​GAT​TCA​TTC​TGA​TCG​C-3′, and reverse, 5′-AAG​GCA​CAC​TCC​GTA​CAG​ATT-3′. Primers were synthesized by Thermo Fisher Scientific.

pU6 vectors containing gRNAs targeting Tsp29Fa and Tsp29Fb (Fig. S2) were injected into Cas9-expressing embryos (PBac{vas-Cas9}VK00027, BDSC #51324). Injections were done by BestGene. Individual fly lines were screened by PCR for deletions in Tsp29Fa and Tsp29Fb. UAS-Tsp29Fa-FLAG-HA lines were generated by injecting the UFO04029 plasmid (Drosophila Genomics Resource Center) into embryos containing attP sites (PBac{y⁺-attP-3B}VK00002, BDSC #9723; and PBac{y⁺-attP-9A}VK00027, BDSC #9744) for phiC31 integrase-mediated transgenesis. Injections were done by BestGene. mCh-2xP4M lines were generated by P-element transformation, where the plasmid containing αTub84B-mCh-2xP4M was injected into w¹¹¹⁸ embryos. Injections were done by BestGene.

For live imaging, salivary glands were dissected from early or late L3 larva in 50 µl Drosophila Ringer’s solution (10 mM Tris, 182 mM KCl, 46 mM NaCl, and 3 mM CaCl₂·2H₂O, pH 7.2) using fine forceps. Dissected salivary glands were transferred to 30 µl Drosophila Ringer’s on a microscope slide and sealed using a coverslip edged with high-vacuum M grease (Apiezon). Samples were imaged using a Quorum spinning-disc confocal coupled with an Olympus IX81 microscope (Quorum Technologies), a Nikon A1R point scanning confocal microscope, or a Leica SP8 Lightning confocal microscope. Images were acquired from the spinning-disc confocal using a 60× oil objective (NA 1.4), a Hamamatsu C9100-13 electron multiplying charge-coupled device camera, and Volocity 6.3 software (PerkinElmer). Serial optical sections were acquired at an interval of 0.3 µm for a total of 20–30 µm. Images acquired from Nikon A1R used a 60× oil objective (NA 1.4) and NIS-Elements software (Nikon). Images acquired from Leica SP8 Lightning confocal used a 40× oil objective (NA 1.3) and LAS-X software (Leica). Leica SP8 images were processed with Lightning using adaptive strategy from the LAS-X software. All images were adjusted for brightness and contrast using Adobe Photoshop CS6.

Salivary glands were dissected from early or late L3 larva in 50 µl PBS, pH 7.4. Dissected glands were fixed with 4% PFA PBS for 20 min at room temperature and permeabilized in PBST (PBS and 0.1% Triton X-100). Primary antibody incubation was performed overnight in PBST with 5% normal goat serum (Thermo Fisher Scientific). Secondary antibodies were diluted in PBST and incubated for 1 h at room temperature. Samples were mounted in 8 µl ProLong Diamond (Thermo Fisher Scientific) using self-adhesive reinforcement labels (Avery #32203) as spacers. Mounted samples were cured overnight at room temperature before sealing with nail polish. Primary antibodies were 1:1,000 rabbit anti-Lva (a gift from O. Papoulas and J. Sisson, The University of Texas at Austin, Austin, TX; Sisson et al., 2000); 1:30 mouse anti-Hrs (final concentration 3 µg/ml; Developmental Studies Hybridoma Bank; Riedel et al., 2016); 1:1,000 chicken anti-GFP (#ab13970, Abcam); and 1:100 rat anti-HA clone 3F10 (final concentration 1 µg/ml; Roche). Secondary antibodies conjugated to Alexa Fluor 405, 488, 568, and 647 were used at 1:500 (Thermo Fisher Scientific).

Cross-sectional areas of SGs were quantified using Volocity 6.3. Images were first deconvolved using the iterative restoration function with a confidence level of 99% and a maximum of 30 iterations. The find object function was used to identify and measure the surface area of SGs. Objects with a cross-sectional area <4 µm² were filtered out. Analyses were performed on three z-slices from a single acquisition (roughly one to three cells in the field of view) from three individual salivary glands.

Colocalization analyses were performed on original images using Volocity 6.3. Intensity threshold was chosen using the find object function to remove unwanted background signal. For each graph, five images from three independent experiments were used to calculate the average and standard deviation.

The percentage of SGs with P4M was quantified using Imaris 9.3.1 (Oxford Instruments). Original images were cropped into a smaller area encompassing mostly a single cell with three z-planes. Surfaces were rendered for SG and P4M from Sgs3-GFP and mCh-2xP4M signals with thresholding. Distance transformation was performed to measure the shortest distance for each of the SG surfaces to reach a P4M surface. Distances >0.5 µm represent SGs without nearby P4M surfaces surrounding them. The dynamics of GFP-PI4KII tubules was quantified by measuring the change in distance between frames using the manual tracking function from ImageJ (National Institutes of Health).

Raw data from ImageJ and Volocity 6.3 were imported into Microsoft Excel 2019. Student’s t tests (two-tailed, unequal variance) were performed to test for statistical significance: **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

Box-and-whisker plots were generated using Microsoft Excel 2019. Each line of the box from bottom to top represents the 25th, 50th (median), and 75th percentile of the population. Bottom and top whiskers represent minimum and maximum values determined by 1.5× the interquartile range. Dots outside of the whiskers are samples considered statistical outliers.

Table S1 shows the RNAi screen and CD63-GFP coexpression screen. Fig. S1 is a characterization of CD63 and Tsp29Fa in salivary gland cells. Fig. S2 shows CRISPR mutagenesis of Tsp29Fa and Tsp29Fb. Fig. S3 is a PI4KIIΔ characterization. Fig. S4 shows PI4KII and P4M analyses. Video 1 shows that GFP-PI4KII-positive tubules are more dynamic in Past1^110-1 mutant.

We thank J. Brumell, U. Tepass, and members of the Brill laboratory for helpful discussions; M. Marks, H. Krämer, T. Balla, and U. Tepass for comments on the manuscript; A. Kiger, C. Wilson, M. Horowitz, Bloomington Drosophila Stock Center, and Vienna Drosophila RNAi Center for fly stocks; the Drosophila Genomics Resource Center for plasmids; O. Papoulas and the late J. Sisson for antibodies; and P. Paroutis and K. Lau (SickKids Imaging Facility) for assistance with microscopy and image analysis.

We acknowledge funding from the following sources: SickKids Restracomp Studentship, Canadian Institutes of Health Research Strategic Training Fellowship #TGF-53877, University of Toronto Open Fellowship, and Natural Sciences and Engineering Research Council of Canada Postgraduate Scholarships—Doctoral Program Studentship (to C-I.J. Ma); Lunenfeld Summer Studentship (to Y. Yang); Canadian Institutes of Health Research Operating Grants #PJT-162165 and #MOP-119483 and Canadian Institutes of Health Research Institute of Genetics #IG1-115714 (to J.A. Brill).

The authors declare no competing financial interests.

Author contributions: Conceptualization: C-I.J. Ma, J. Burgess, J.A. Brill; Data curation: C-I.J. Ma, J.A. Brill; Formal Analysis: C-I.J. Ma, Y. Yang; Funding acquisition: J.A. Brill; Investigation: C-I.J. Ma, Y. Yang, T. Kim, C.H. Chen, M. Vissa, J. Burgess; Methodology: C-I.J. Ma., G. Polevoy, J. Burgess, J.A. Brill; Project administration: J.A. Brill; Resources: C-I.J. Ma, Y. Yang, T. Kim, M. Vissa, G. Polevoy, J. Burgess, J.A. Brill; Supervision: J.A. Brill; Validation: C-I.J. Ma, Y. Yang, T. Kim, C.H. Chen, M. Vissa, J. Burgess, J.A. Brill; Visualization: C-I.J. Ma, J.A. Brill; Writing – original draft: C-I.J. Ma, J.A. Brill; Writing – reviews & editing: C-I.J. Ma, Y. Yang, T. Kim, C.H. Chen, M. Vissa, J. Burgess, J.A. Brill.