The FAM114A proteins are adaptors for the recycling of Golgi enzymes

The Golgi is the major sorting hub in the secretory pathway. It receives newly made lipids and proteins from the endoplasmic reticulum (ER) and then sorts them to the cell surface or the compartments of the endocytic system. Following arrival from the ER, proteins and lipids move through the stack of Golgi cisternae before leaving in carriers forming at the trans side of the Golgi. The Golgi stack contains many resident enzymes that modify glycoproteins and glycolipids as they move through the stack (Moremen et al., 2012; Schjoldager et al., 2020). These enzymes, along with the transporters that deliver nucleoside-sugars and ions, must all maintain their residence within the stack while their substrates arrive and then depart. Furthermore, the enzymes are typically arranged within the stack in the order in which they act and so are localised to only a subset of cisternae. There has been much debate about the mechanism by which cargo proteins move past the enzymes that modify them, but the current widespread consensus is that the cisternae form on the cis side and then mature as they progress through the stack (Glick and Nakano, 2009; Pantazopoulou and Glick, 2019). As the cisternae progress, the resident enzymes are continuously recycled in vesicles and delivered to earlier cisternae in the stack and so maintain a constant distribution in a manner analogous to hopping down an upward moving escalator (Lujan and Campelo, 2021; Welch and Munro, 2019).

Vesicle budding from the Golgi stack is dependent on the COPI coat that is formed from the small GTPase Arf and coatomer, a heptameric complex distantly related to the clathrin adaptors proteins (Arakel and Schwappach, 2018; Beck et al., 2009; Taylor et al., 2022). COPI is responsible for the recycling of ER residents back to the ER from the cis Golgi, and it can bind directly to recycling signals in the cytoplasmic tails of ER membrane proteins. However, COPI is also present on later cisternae in the stack, and in vitro budding assays have shown that it can concentrate Golgi resident proteins into vesicles (Adolf et al., 2019; Eckert et al., 2014). This raises two questions: how COPI can collect different cargo in different parts of the stack, and how are these vesicles then delivered to distinct destinations depending on where they originated from. In vertebrates, two of the coatomer subunits are present as pairs of paralogues, but in vitro vesicle budding assays did not detect significant differences between the contents of vesicles formed using coatomer containing different combinations of these paralogues (Adolf et al., 2019). Beyond the coat itself, another factor that could influence cargo recruitment is adaptors that bind specific subsets of cargo as is the case for clathrin-coated vesicles (Traub and Bonifacino, 2013). For COPI, the clearest example of such an adaptor is the cytosolic protein GOLPH3, which binds directly to the short basic tails found on many Golgi glycosylation enzymes (Ali et al., 2012; Schmitz et al., 2008; Tu et al., 2008; Welch and Munro, 2019). Removal of GOLPH3 results in loss of Golgi retention for a subset of Golgi residents, and addition of GOLPH3 to an in vitro COPI budding assay stimulates the incorporation of specific cargo into vesicles (Eckert et al., 2014; Rizzo et al., 2021; Welch et al., 2021).

Given that the Golgi contains a diverse population of resident proteins that varies between cell types, it seems likely that there are additional adaptors that enable the COPI coat to select specific cargo in specific circumstances. To seek such adaptors, we have made use of one of the mechanisms by which the vesicles that bud from the Golgi are captured by their destination compartments. Capture of vesicles within the stack has been shown to depend, at least in part, on long coiled-coil proteins called golgins, with different golgins found in different parts of the stack (Gillingham and Munro, 2016; Muschalik and Munro, 2018; Witkos and Lowe, 2017). When these golgins are relocated to an ectopic location they are sufficient to cause ectopic vesicle capture, and different golgins capture different classes of vesicle (Park et al., 2022; Wong and Munro, 2014). Three golgins have been shown to capture Golgi-derived vesicles at an ectopic location: GMAP-210, golgin-84 and TMF, encoded in humans by the genes TRIP11, GOLGA5 and TMF1. GMAP-210 and golgin-84 are located on the cis- and medial Golgi, whereas TMF is later in the stack, and consistent with this, the vesicles captured by TMF contain proteins from the later part of Golgi (Bascom et al., 1999; Mori and Kato, 2002; Sato et al., 2015; Wong and Munro, 2014). The mechanism by which these golgins capture vesicles remains unclear, although it is known that motifs at the N-terminus are sufficient, and in the case of GMAP-210 it has been suggested that this region recognises the lipid composition of the vesicle (Magdeleine et al., 2016; Wong et al., 2017). Nonetheless, the ectopic capture of vesicles by golgins provides a means to isolate them from the rest of the stack and examine their contents. This has been successfully applied for the golgins at the trans-Golgi, which capture carriers coming from endosomes and has allowed the identification of vesicle-resident proteins and a linker protein that connects the golgins to the vesicles (Shin et al., 2017, 2020). In this paper, we use the golgin GMAP-210 to identify the FAM114A proteins as being associated with intra-Golgi transport vesicles and demonstrate that they can bind Golgi residents and are required for normal Golgi function in vivo.

To identify novel components of intra-Golgi vesicles, a MitoID proximity-dependent labelling assay was applied to the intra-Golgi golgin tether GMAP-210 as undertaken previously for trans-Golgi golgins and Rab GTPases (Gillingham et al., 2019; Shin et al., 2020). In short, the basis of the screen was to ectopically relocate intra-Golgi vesicles to the mitochondria and biotinylate the proximal vesicle-resident proteins to allow purification by streptavidin pulldown and identification by mass spectrometry (Fig. 1A). To do this, the N-terminal vesicle-binding region of GMAP-210 was fused to the promiscuous biotin ligase BirA*, the coiled-coil protein GCC185 as a spacer that lacks tethering activity, an HA epitope tag and the mitochondrial targeting sequence of monoamine oxidase (MAO). Controls included a version where a conserved tryptophan residue is mutated, as this has been shown to cause GMAP-210 to lose the ability to capture giantin (GOLGB1)- and GALNT2-containing vesicles (Wong et al., 2017). These golgin chimeras were stably expressed in HEK293 cells, and to capture vesicles, expression was induced, and cells treated with nocodazole to convert the Golgi into ministacks. Previous work has shown that this treatment increases the efficiency of capture of Golgi-derived vesicles as it places many more mitochondria in the proximity of a Golgi stack (Wong and Munro, 2014). Following a biotin pulse, labelled proteins were isolated with streptavidin and identified by mass spectrometry.

Of the top 15 proteins enriched with the wild-type GMAP-210 bait, several known Golgi-resident proteins were identified including ERGIC-53 (LMAN1), golgin-84 (GOLGA5), TMEM87A and FAM114A2, all of which were substantially reduced by the W4A mutation in the GMAP-210 vesicle capture motif (Fig. 1B; Table S1). ERGIC-53 and golgin-84 are integral membrane proteins that are well characterised as cargo of Golgi-derived vesicles (Adolf et al., 2019). In contrast, FAM114A2 is a cytosolic protein of unknown function. The Drosophila orthologue, CG9590, was identified as a Rab2 effector in an affinity chromatography proteomic screen using S2 cell lysate, and this was subsequently supported by findings in a Rab MitoID screen in mammalian cells (Gillingham et al., 2014, 2019). The Drosophila protein and human FAM114A1 were both shown to colocalise with cis-Golgi markers, but their function is unknown. All are predicted by AlphaFold2 to have an unstructured N-terminal half and a C-terminal helical bundle (Fig. 1C,D; Fig. S1A). The unstructured N-terminus contains a highly conserved motif containing three tryptophan-glycine pairs. This motif is part of a region previously assigned as a ‘domain of unknown function’ (DUF719) by the Pfam database, but its role is unclear. The C-terminal helical bundle is predicted by AlphaFold to be the part that binds Rab2 (Fig. S1B).

To elucidate the function of the FAM114A proteins, GST-tagged forms of both were produced in bacteria and used as baits in pulldowns to isolate binding partners from HEK293T cell lysate. The interacting proteins were identified by mass spectrometry and compared to a control of GST alone using volcano plots. Both FAM114A proteins enriched Rab2A and Rab2B, and a selection of Golgi-resident integral membrane proteins (Fig. 1E; Table S1). They also efficiently enriched METTL26 (C16orf13), a cytosolic protein of unknown function. However, this was not found as an interactor with the Drosophila orthologue (see below), is not predicted by AlphaFold2 to interact with FAM114A1 or FAM114A2, when knocked out in mice does not affect viability or fertility (Muñoz-Fuentes et al., 2018), and so it was not investigated further. FAM114A1 pulled down relatively few proteins whereas FAM114A2 specifically enriched a vast array of Golgi-resident membrane proteins, many of which are presumptive intra-Golgi COPI cargo proteins, such as glycosyltransferases.

FAM114A proteins are predicted to be peripheral membrane proteins and so it is likely that they are binding the cytoplasmic tails of Golgi-resident proteins. This is reminiscent of what is seen for the COPI adaptors GOLPH3 and GOLPH3L, which have been shown to interact with the cytoplasmic tails of Golgi residents through membrane-proximal polybasic stretches (Ali et al., 2012; Rizzo et al., 2021; Tu et al., 2008; Welch et al., 2021). The predicted structure of the FAM114A helical bundle also has large electronegative surfaces like the surface observed on the crystal structure of GOLPH3 (Fig. S1C). To examine the interaction with Golgi enzymes, we applied to FAM114A proteins a binding analysis like that applied to GOLPH3 (Welch et al., 2021). In this assay, the signal anchor region of a plasma membrane protein, sucrase-isomaltase (SI), is expressed as a fusion to GFP, and the cytoplasmic tail replaced with those of various type II Golgi-resident proteins (Fig. 2A). The fusion proteins are expressed in mammalian cells and after solubilisation in detergent their binding to FAM114A-coated beads can be assayed. We initially tested direct binding of FAM114A proteins to the tail of GALNT2, a Golgi-resident O-linked mucin type glycosyltransferase, which was pulled down by FAM114A2. The GALNT2–SI–GFP–FLAG chimera was recombinantly expressed and purified from HEK293T cell lysate using FLAG affinity chromatography. The purified chimera was then assayed for binding to beads coated with GST-tagged FAM114A proteins. The GALNT2 tail chimera exhibited strong, direct and specific binding to GST-tagged FAM114A2 but not to GST-tagged FAM114A1 or GST alone (Fig. 2B). Next, we generated a series of chimeras with tails from a range of different Golgi enzymes and tested the ability of GST-tagged FAM114A2 to pull them out from HEK293T cell lysates, as we had undertaken previously for GOLPH3 (Welch et al., 2021). As with GOLPH3, FAM114A2 bound to diverse tails with a preference for tails with membrane-proximal polybasic clusters. In contrast, tails with a paucity of positive residues or those containing negative residues, such as the plasma membrane protein SI, bound poorly to FAM114A2 (Fig. 2C; Fig. S2A). Of the 14 Golgi enzyme tails tested in Fig. 2, 12 were tested in our previous study on GOLPH3, and the overall pattern of binding was broadly similar, and where there were differences, it was that some (but not all) proteins tails bound better to GOLPH3 (Welch et al., 2021). Membrane proximal insertion of three arginine or lysine residues, but not histidine residues (which are not protonated at a cytosolic pH of 7.4), into the tail of SI was sufficient to bestow FAM114A2 binding, as was also the case for GOLPH3. In summary, FAM114A2 is comparable to the COPI adaptors GOLPH3 and GOLPH3L as it binds directly to the tails of cargo with membrane-proximal polybasic stretches.

Deletion of GOLPH3 and GOLPH3L from U2OS cells leads to reduced levels of a subset of Golgi enzymes and downstream defects in glycosylation (Welch et al., 2021). Thus, CRISPR-Cas9 gene-editing was used to delete the FAM114A genes in wild-type U2OS cells and also in ΔΔGOLPH3, GOLPH3L double deletion U2OS cells in case of functional redundancy between the GOLPH3 and FAM114A families (Fig. 3A). Multiplexed quantitative mass spectrometry was used to compare relative protein abundances between wild-type and ΔΔFAM114A1, FAM114A2 double deletion U2OS cell lines. In contrast to what was seen with ΔΔGOLPH3, GOLPH3L cells, there was no clear difference in the levels of Golgi-resident proteins in wild-type U2OS cell versus ΔΔFAM114A1, FAM114A2 cells (Fig. S2B). We have previously shown that the deletion of GOLPH3 genes perturbs the Golgi retention of a GALNT2 cytoplasmic chimera in an in vivo Golgi retention assay (Welch et al., 2021). We found that deletion of both FAM114A genes in a wild-type U2OS background had a far smaller effect on Golgi retention of the GALNT2 reporter when analysed by flow cytometry or immunofluorescence (Fig. 3B; Fig. S2C). Furthermore, there was no detectable additive effect when the FAM114A genes were deleted in a GOLPH3 double-knockout background. As reported previously, a panel of cell surface lectins revealed strong defects in glycosylation in the ΔΔGOLPH3, GOLPH3L U2OS cells, but in contrast knockout of the FAM114A genes did not result in consistent changes in lectin labelling apart from a small increase in binding by Wisteria floribunda agglutinin (WFA) which recognises mucin-type O-linked glycans (Fig. S2D).

It is possible that the FAM114A proteins are of more importance in specific cell types, and we also examined primary fibroblasts which express a different ratio of the FAM114A proteins than do U2OS cells (Fig. 3C). However, siRNA knockdown of the FAM114A proteins did not result in detectable changes in lectin staining, although we did detect a small but reproducible reduction in the levels of GALNT7, a Golgi enzyme whose level is also particularly sensitive to loss of GOLPH3 (Welch et al., 2021) (Fig. 3D–F).

For many proteins involved in Golgi function, their removal only causes detectable phenotypes in particular tissues, perhaps reflecting plasticity and robustness in intracellular trafficking pathways (Bem et al., 2011; Lowe, 2019; Marin-Valencia et al., 2017; Schmidt et al., 2007). Therefore, we used the Drosophila system to examine FAM114A activity in a multicellular organism. Drosophila have a single FAM114A orthologue, CG9590, which has previously been shown to bind to Rab2 but is otherwise uncharacterised (Gillingham et al., 2014). Expression atlas data shows that it is expressed in most tissues, with an elevation in cells with high secretory activity, such as the salivary and accessory glands (Krause et al., 2022). The genes that match this tissue profile most closely are other proteins involved in Golgi function, such as COPI subunits and SNAREs.

CG9590 is predicted to have a structure very similar to that of its mammalian orthologues, with an N-terminal unstructured domain including a region with WG motifs, and a C-terminal helical bundle with an electronegative surface (Fig. 4A). However, unlike the mammalian relatives, CG9590 also has near its N-terminus a motif containing two tryptophan residues embedded in acidic residues (²⁴WDDW), and this feature is conserved amongst insects. A similar Wx_n(1–6)[W/F] motif has been shown to bind to the μ-homology domain of the δ-COP subunit of coatomer and typically contains two tryptophan or phenylalanine residues separated by two or three residues and positioned within a highly acidic stretch (Suckling et al., 2015). As with the mammalian FAM114A proteins, we initially used affinity chromatography to identify potential binding partners of CG9590. CG9590 and, for comparison, the Drosophila orthologue of GOLPH3, Sauron, were expressed in bacteria as GST fusions and used to enrich interacting partners from S2 cell lysates. When compared to what was seen with GST alone, GST-tagged CG9590 specifically enriched a plethora of membrane proteins that are resident in the Golgi (including several O-linked mucin-type glycosylation enzymes) or are likely to cycle between the ER and Golgi (Fig. 4B and Table S1). As expected, Sauron also enriched Golgi residents, albeit with differing efficiencies compared to CG9590 (Fig. 4C). Neither protein showed an enrichment of the Drosophila ortholog of METTL26/C16orf13 (CG18661), the protein found enriched with the mammalian proteins (Fig. 4C,D). In addition to the lack of METTL26, another striking difference with the results with the human proteins was that CG9590 showed a strong enrichment of the subunits of the COPI coat. In order to determine whether CG9590 was interacting with the COPI coat via the ²⁴WDDW sequence that resembles a Wx_n(1-6)[W/F] motif, the two tryptophan residues were mutated to alanine residues, and the protein interactome of the mutant was compared to that of the wild-type protein. Relative to wild-type CG9590, the enrichment of the COPI coat subunits, Vap33 and subunits of the OST complex was markedly reduced in the mutant CG9590 sample (Fig. 4D). In contrast, there was little or no difference in binding to intra-Golgi proteins, suggesting that the ability of CG9590 to bind Golgi residents is independent of COPI binding, whereas some ER residents are possibly binding directly to the COPI coat itself. Direct comparison of the proteins binding to Drosophila GOLPH3 and CG9590 showed that some Golgi enzymes preferred GOLPH3, but most showed comparable levels of binding. It is of course possible that these in vitro binding assays that are done in the absence of membrane might not capture all interactions made in vivo. Nonetheless, the proteins that were highly enriched with both GOLPH3 and CG9590 are all Golgi-resident membrane proteins, indicating that the assay does at least have a considerable degree of specificity (Fig. 4E). Given the structural and functional similarities between the mammalian FAM114A proteins and Drosophila CG9590 we will refer to the latter as FAM114A.

To investigate the role of Drosophila FAM114A we used CRISPR-Cas9 to delete the entire gene from the genome (Fig. 5A). Flies lacking both alleles were viable and fertile, and an antibody raised against the Drosophila protein revealed that the protein was absent as expected. The antiserum was not suitable for immunofluorescence and, hence, we generated Drosophila lines expressing a GFP-tagged form of FAM114A under UAS control (Fig. 5B). The tissue reported to have the highest level of expression of FAM114A is the larval salivary gland, which secretes large amounts of glue proteins and has an abundance of secretory organelles and proteins (Loganathan et al., 2021). FAM114A–GFP was expressed in the salivary gland using a fkh-Gal4 driver and was found to accumulate on the Golgi (Fig. 5C). Comparison with other Golgi markers showed that the protein localised on the cis side of the Golgi stack being distributed between ER exit sites and the earliest Golgi markers, such as the golgins GMAP and GM130. This is consistent with a role in recycling of Golgi residents or escaped ER residents from the earliest compartments of the stack.

The glue proteins produced in the salivary gland include secretory mucins that are heavily modified with O-linked glycans and have thus proven useful for detecting defects in Golgi-dependent glycosylation (Biyasheva et al., 2001; Reynolds et al., 2019; Tran and ten Hagen, 2013). Salivary gland proteins were prepared from flies lacking FAM114A, and when separated by gel electrophoresis, it could be seen that the major glue protein Sgs3 migrated faster than in wild type, indicating reduced glycosylation; this was rescued by expression of FAM114A–GFP (Fig. 6A). The O-linked glycans that are attached to the glue proteins are initiated by the addition of N-acetylgalactosamine (GalNAc) in the Golgi, which is typically modified with galactose to form Galβ1,3GalNAc, and then extended further with glucuronic acid (Ji et al., 2018; Tian and ten Hagen, 2007). The lectin Vicia villosa agglutinin (VVA) recognises O-linked GalNAc but not the extended structure, and so it labels the Golgi, and loss of this staining has been previously observed in mutants of the golgin coiled-coil proteins where Sgs3 mobility is also increased (Park et al., 2022). By contrast, peanut agglutinin labels the extended structure and so labels the secretory granules. Loss of FAM114A results in smaller granules (Fig. 6B), and loss of clear Golgi labelling by VVA is consistent with reduced glycan modification of the glue proteins as they transit the Golgi (Fig. 6C). Taking these results together we conclude that FAM114A is required for normal Golgi glycosylation in Drosophila.

In this study, we report that the FAM114A proteins are associated with the Golgi and intra-Golgi transport vesicles and that they can bind directly to the tails of Golgi resident enzymes via membrane-proximal basic residues. Removal of the single orthologue from Drosophila results in defects in glycosylation. From this, we propose that the FAM114A proteins act as adaptors to help recruit Golgi enzymes into COPI-coated vesicles that recycle membrane proteins within the Golgi stack and hence maintain the levels and organisation of glycosylation enzymes and other Golgi residents.

Such a role for the FAM114A proteins would be analogous to that of GOLPH3, which also binds to the tails of Golgi-resident proteins and has been shown to promote their inclusion into COPI vesicles and to maintain their Golgi localisation (Rizzo et al., 2021; Tu et al., 2008; Welch et al., 2021). Unlike the FAM114A proteins, GOLPH3 is localised toward the trans side of the Golgi and thus is likely to function in a distinct set of COPI-coated vesicles. By contrast, FAM114A appears to be in the cis side of the stack and between the ER exit sites and the Golgi. Thus, FAM114A might be in COPI vesicles recycling within the early Golgi and also in those returning to the ER. The FAM114A proteins appear to be more minor players in Golgi recycling than GOLPH3 given that more proteins are affected in cells lacking the latter, and that the GOLPH3 orthologue in Drosophila is essential for viability whereas FAM114A is not. Individual knockouts for the two FAM114A paralogues in mice show no detectable effect, although a double mutant has not been reported (Khan et al., 2021; Muñoz-Fuentes et al., 2018). In contrast, GOLPH3 knockout in mice causes a range of severe phenotypes even though its less highly expressed paralogue GOLPH3L is still present (Muñoz-Fuentes et al., 2018). This seemingly more minor role for the FAM114A proteins is also suggested by it being absent from some invertebrates, such as C. elegans, and from the fungal phylum, whereas GOLPH3 is conserved in both. However, it is interesting to note that there appears to be a distant FAM114A orthologue in plants given that one enigma about GOLPH3 is that it is absent from plants despite them typically having well organised Golgi stacks populated by many different enzymes. In Arabidopsis, this putative orthologue is encoded by the gene AT2G15860, with orthologues present in all plants examined, including green algae. For reasons that are obscure, AT2G15860 is annotated in UniProt as having a BAT2 domain, a term applied to the unstructured N-terminus of PRRC2 proteins, but such a domain is not detected by either the Pfam or InterPro domain databases. AlphaFold2 predicts a structure like that of the FAM114A proteins with an unstructured N-terminal region containing a tryptophan-rich motif and a C-terminal helical bundle. No functional characterisation has been reported in any plant species, but it seems a plausible candidate for role in the organisation of the Golgi in plants.

The FAM114A proteins are known to bind Rab2 in both humans and Drosophila. Our affinity chromatography with FAM114A1 and FAM114A2 found that the former binds most efficiently to Rab2, but this might reflect this in vitro assay system, as an interaction between Rab2A and FAM114A2 was readily detected in our previous in vivo proximity biotinylation screen. AlphaFold2 gives a high confidence prediction for a complex between Rab2 and the FAM114A proteins, indicating that the interaction is with the C-terminal helical bundle (Fig. S1B). The predicted Rab2-binding side is away from the acidic part of the FAM114A surface and so would not interfere with binding to a membrane-proximal and basic cytosolic tail, and the unstructured hypervariable region of the Rab is long enough that the lipidated C-terminus of the Rab could remain in the membrane. Rab2 is localised to the Golgi and was initially reported to act in trafficking between the ER and Golgi (Cheung et al., 2002; Tisdale and Balch, 1996). However, in C. elegans, which lack FAM114A, Rab2 appears to be primarily involved in the formation of dense-core vesicles (Ailion et al., 2014; Sumakovic et al., 2009), and in Drosophila there is genetic evidence for roles in constitutive secretion, lysosome function and dense core vesicle production (Fujita et al., 2017; Götz et al., 2021; Ke et al., 2018; Lorincz et al., 2017). A role in regulated secretion might explain the lack of a widespread phenotype, but the two proteins appear to be expressed in a wide range of tissues even if the ratio between them varies somewhat. Moreover, Rab2 binds to quite a wide range of effectors and so it might simply serve to recruit a diverse set of proteins to the correct part of the Golgi, and this set of proteins can vary between species. Finally, it should be noted that the tryptophan-glycine repeat motif conserved in all FAM114A proteins has properties similar to those of neutral amphipathic helixes, which have been proposed to direct binding of proteins to lipid bilayers, and so it could potentially augment the action of Rab2 in targeting of the FAM114A proteins to membranes (Drin et al., 2007; Van Hilten et al., 2024).

There is clearly much that remains to be learnt about the in vivo role of FAM114A proteins, but our work clearly indicates that they have a role in Golgi function and appear to be new additions to the growing list of proteins that serve to allow COPI-coated vesicles to transport different cargo in different parts of the Golgi stack.

For a list of the plasmids used in this study, see Table S2. GFP-tagged cytoplasmic tail chimeras were generated as described previously (Welch et al., 2021). Plasmids designed to delete FAM114A1 and FAM114A2 were generated by annealing complementary oligonucleotide pairs encoding single gRNAs with BbsI-compatible overhangs and cloning them into the BbsI-digested bicistronic CRISPR-Cas9 mammalian expression vector pX458 (pSpCas9[BB]-2A-GFP). The coding sequence of FAM114A1 and FAM114A2 was fused to a C-terminal GAGA linker and a GST tag and cloned into the baculoviral expression vector pAcebac1 (Geneva Biotech) and the bacterial expression vector pOPC. Drosophila CG9590 was fused to a C-terminal GSGSGS linker and a GST tag and cloned into pOPC.

For a list of the antibodies used in this study see Table S2. To raise an antibody against CG9590, GST-tagged CG9590 was produced in bacteria and affinity purified with glutathione beads (see below). Purified GST-tagged CG9590 was freeze dried before being used for five rounds of immunisations of rabbits over 2 months (Davids Biotechnologie, Regensburg, Germany). Rabbit serum was depleted of GST-specific antibodies and subsequently affinity purified using the GST-tagged CG9590 antigen immobilised on beads, being eluted from beads using a low pH buffer and immediately neutralised on elution.

U2OS (ATCC), HEK293T cells (ATCC) and FibroGRO Xeno-Free human foreskin fibroblasts (Merck, gift from Martin Lowe, School of Biology, University of Manchester, UK) were maintained in Dulbecco's modified Eagle's medium (DMEM, Thermo Fisher Scientific) with 10% fetal bovine serum (FBS, Thermo Fisher Scientific) and 1% penicillin-streptomycin (PS; Gibco) in a humidified incubator at 37°C with 5% CO₂. Flp-In T-Rex HEK293 cell lines (Thermo Fisher Scientific) expressing mitochondrial-relocated golgin-BirA* fusions (from John Shin, MRC Laboratory of Molecular Biology, UK) were maintained in DMEM with 10% FBS, PS 5 μg/ml blasticidin (Generon) and 100 μg/ml hygromycin (Thermo Fisher Scientific). Cells were passaged 1:10 every 3–4 days by trypsinisation and were regularly screened for mycoplasma contamination (Mycoalert, Lonza).

D.Mel-2 cells were maintained in Schneider's Drosophila medium (Thermo Fisher Scientific) with 10% FBS and PS at 24°C. Cells were subcultured at a ratio of 1:10 every 3–4 days by detaching cells through tapping of the flask and the cell suspension diluted in fresh medium in a fresh flask. Large scale D.Mel-2 cultures were prepared by diluting cells to a density of 10⁶ cells/ml in Insect Xpress culture medium (Lonza) in Erlenmeyer flasks. Cells were incubated at 25°C with shaking at 140 rpm and subcultured by dilution at a ratio of 1:10 every 3–4 days.

GST-tagged FAM114A proteins were produced in Sf9 cells using the MultiBac baculoviral expression system (Geneva Biotech). Sf9 cells were seeded at a density of 10⁶ cells/cm² in 6-well plates in Insect Xpress culture medium at 27°C and allowed to adhere for at least 10 min. Cells were transfected with 2 μg of bacmid DNA using Fugene HD transfection reagent (Promega) according to the manufacturer's protocol. Cells were incubated for 3–5 days, and the medium containing virus was used to inoculate a 50 ml culture of cells at a density of 2×10⁶ cells/ml in 250 ml Erlenmeyer flasks. Cells were cultured at 27°C with shaking at 140 rpm for 3–5 days. Cells were pelleted at 2500 g for 10 min and the pellet stored on ice or snap frozen in liquid nitrogen prior to protein purification. Alternatively, the supernatant containing the virus was used to inoculate larger cultures or to enhance the viral titre. The supernatant was preserved in the presence of 2% FBS at 4°C in darkness for medium term storage.

CRISPR-Cas9 gene editing was used to simultaneously knockout FAM114A1 and FAM114A2 in U2OS cells by targeting early constitutive exons with small out-of-frame deletions. Exon 3 of FAM114A1 was simultaneously targeted at 5′-GTGCAGGGGCTGCCGCCATT-3′ and 5′ CCAACACCAGCTGACCCCAG-3′ and exon 2 of FAM114A2 was targeted at 5′-ACTCTCTGGTTTGGCACCT-3′ and 5′-GGGGCTGCTTCAGTTAGCAG-3′. U2OS cells were seeded in T-75 flasks in culture medium and maintained in a humidified incubator with 5% CO₂ at 37°C. Once cells reached 50–80% confluency they were transfected with CRISPR-Cas9 plasmids using polyethylenimine. At 24 h after transfection, single GFP-positive clones (i.e. cells expressing Cas9-2A-GFP) were sorted into 96-well plates (MoFlo Cell Sorter, Beckman Coulter). Candidate knockout clones were validated by immunoblotting and the lead clone was further validated by sequencing of PCR-amplified genomic regions and also mass spectrometry of the cell lines. GOLPH3 and GOLPH3L were simultaneously deleted in the ΔΔFAM114A1, FAM114A2 U2OS background as described previously (Welch et al., 2021).

Stable cell lines expressing GFP-tagged cytoplasmic tail chimeras under a cumate-inducible promoter were generated by PiggyBac transposition (System Biosciences). Wild-type, ΔΔFAM114A1, FAM114A2 and ΔΔΔΔFAM114A1, FAM114A2, GOLPH3, GOLPH3L CRISPR knockout U2OS cells were cultured to 50% confluency in 6-well plates and subsequently transfected with 0.2 μg PiggyBac transposase (PB210PA-1) and 0.5 μg of the PiggyBac-compatible expression plasmid. Cells were expanded to T-75 flasks 2 days after transfection and cells were subject to selection in culture medium with 0.5–1 μg/ml puromycin (Sigma) (selection medium) at 3 days after transfection. Cells were cultured in selection medium for several weeks until the polyclonal pool of integrants had reached confluency. Cell lines were immediately cryopreserved and maintained in selection medium containing 60 μg/ml cumate (System Biosciences) for at least one passage prior to assay.

Foreskin fibroblasts at 60–80% confluency 24 h after seeding were treated with ON Targetplus SMARTpool siRNA oligonucleotides targeting FAM114A1 and FAM114A2 separately or simultaneously or were treated with a non-targeting negative control siRNA (Horizon Discovery) using Lipofectamine RNAiMAX transfection reagent according to the manufacturer's instructions (Thermo Fisher Scientific). Cells were treated with siRNA on day 1 and 3 after seeding, and on day 6 were washed once gently with PBS and lysed in plate with 1× LDS sample buffer (Novex) with 10% TCEP. The lysate was sonicated and clarified by centrifugation before being subject to immunoblot analysis.

Doxycycline-inducible stable Flp-In T-REx HEK293 cell lines expressing mitochondrial-relocated golgin-BirA* fusion proteins were induced with 1 μg/ml doxycycline (Sigma) in culture medium once they reached ∼80% confluency. At 24 h after induction, cells were treated with 0.5 μM nocodazole (Sigma), 50 μM biotin (Sigma) and 1 μg/ml doxycycline in culture medium for a further 9 h. Cells were harvested and lysed for a streptavidin pulldown. Dynabeads One Streptavidin T1 beads (Thermo Fisher Scientific) were washed once in lysis buffer [50 mM Tris-HCl pH 7.4, 150 mM NaCl, 1 mM EDTA, 0.5% Triton X-100, 1 mM PMSF, 1× cOmplete EDTA-free protease inhibitor (COMP; Roche)] using a DynaMag-2 magnetic stand (Thermo Fisher Scientific). Cell lysates were added to the washed beads and incubated overnight with agitation at 4°C. The beads were washed twice in wash buffer 1 (2× SDS with 1× COMP) for 8 min, three times in wash buffer 2 (50 mM Tris-HCl pH 7.4, 500 mM NaCl, 1 mM EDTA, 1% Triton X-100, 0.1% deoxycholate, 1× COMP) for 8 min and three times in wash buffer 3 (50 mM Tris HCl pH 7.4, 50 mM NaCl, 1× COMP) for 8 min. Proteins were eluted by boiling at 98°C in 1× LDS, 10% β-mercaptoethanol and 6 mM biotin for 5 min. Samples were resolved by SDS-PAGE, and gel slices were sent for mass spectrometry analysis (see below).

Pelleted mammalian, bacterial and insect cells were resuspended in lysis buffer (as described for the streptavidin pulldown). Large scale cultures were sonicated on ice for 1 min with 10 s on-off cycles (bacteria cells) or for only 10 s (mammalian and insect cells) using a lance sonicator (Sonic Vibra-Cell, 45% amplitude). Smaller scale cultures were sonicated using a water sonicator for 1 min (mammalian and insect cells) or 3 min (bacteria cells) with 10 s on-off cycles (Misonix 300, amplitude 5.0). Lysates were immediately placed on fresh ice for at least 5 min to mitigate heat generation from sonication. Lysates were subject to agitation at 4°C for a further 10 min prior to clarification by centrifugation at 16,000–32,000 g for 10 min at 4°C. Where required, protein content was quantified using the Pierce BCA Protein Assay Kit and lysates normalised. Protein samples were kept on ice prior to downstream purification or were resolved by SDS-PAGE.

Glutathione–Sepharose 4B beads (GE Life Sciences) were equilibrated in lysis buffer (as above) prior to pelleting at 100 g for 1 min and removal of the supernatant. Lysates containing GST-tagged fusion proteins were incubated with beads for 30 min with agitation at 4°C. Beads were then washed once with lysis buffer with 150 mM NaCl, once with lysis buffer with 500 mM NaCl and then another four times with lysis buffer with 150 mM NaCl. For the purification of GST-tagged CG9590 for rabbit immunisations, the fusion protein was eluted in buffer consisting of 50 mM Tris-HCl pH 7.4 with 25 mM reduced glutathione. For pulldowns upstream of mass spectrometry analysis, lysates containing prey proteins were preincubated on glutathione–Sepharose beads at 4°C for 30 min to preclear non-specific interactors. Prey lysates were mixed with beads loaded with GST fusion baits, and the mixtures were incubated at 4°C for 1 h with agitation. Beads washed five times in lysis buffer prior to elution in lysis buffer with 1.5 M NaCl. The prey proteins in the eluate precipitated with TCA in acetone and resolubilised in 1× LDS with 10% β-mercaptoethanol or TCEP. Bait proteins were eluted by boiling in 2× LDS with 10% β-mercaptoethanol or TCEP.

Wild-type and knockout U2OS cell lines were seeded at 2×10⁴ cells/cm² in T-75 flasks in culture medium at 37°C with 5% CO₂. At 80–90% confluency, cells were washed in EDTA solution (0.5 mM EDTA in PBS) and detached using Accutase (Sigma) for 2 min at 37°C. Cells were resuspended in ice-cold FACS buffer (2% FBS in PBS) and ∼10⁶ cells were transferred to round-bottom 96-well plates. Cells were pelleted by centrifugation at 300 g for 5 min, the supernatant removed, and cells washed by resuspension in FACS buffer. Cells were stained with fluorescein-labelled lectins at 20 μg/ml (Vector Biolabs) and a fixable eFluor 780 viability dye diluted 1:1000 (Thermo Fisher Scientific) in FACS buffer on ice in darkness for 30 min. Non-specific binding was controlled by preincubation of the lectin with saturating concentrations of competing sugars at least 30 min prior to addition to cells. Finally, cells were washed three times in FACS buffer, fixed in 4% paraformaldehyde (PFA) diluted in PBS for 20 min at room temperature and washed a further two times in FACS buffer. Suspensions were kept at 4°C in darkness until required and were filtered using a 100 μm plate filter prior to loading on an LSRII flow cytometer (BD Biosciences). Gates were applied and density curves generated using FlowJo V10. Briefly, singlets were gated based on forward and side scatter, dead cells were excluded from analysis using the viability dye.

Inducible stable cell lines expressing GFP-tagged cytoplasmic tail chimeric reporters were cultured in 6-well plate format in selection medium (see above) containing 60 μg/ml cumate for at least a week prior to analysis. Once cells reached 80–90% confluency, they were washed once with EDTA solution and were detached from the plate in Accutase for 2 min at 37°C. Cells were resuspended in selection medium, and cell suspensions were transferred into a deep 96-well plate. Cells were pelleted at 300 g for 5 min and were resuspended in ice-cold FACS buffer. Suspensions were transferred to a round-bottomed 96-well plate and were resuspended in a cocktail consisting of an Alexa Fluor (AF) 647-conjugated anti-GFP antibody (BioLegend) and an eFluor 780 fixable viability dye diluted in FACS buffer. Cells were incubated on ice in darkness for 30 min before being washed, fixed and analysed as described for lectin stains.

Protein samples generated from the MitoID assay and GST affinity chromatography were resolved by SDS-PAGE and gels stained with InstantBlue Coomassie stain (Expedeon). Gel slices were excised for trypsin digestion and analysis by Nanoflow reverse-phase liquid chromatography-mass spectrometry using the Velos Orbitrap mass spectrometer (Thermo Fisher Scientific) as described previously (Gillingham et al., 2019). For spectral count analysis of the results of the MitoID assay, Mascot (Matrix Science) was used to search for peptides against the UniProt human proteome and further filtered using Scaffold (Proteome Software Inc). MitoID spectral counts were compared using D-score analysis from the open-source ComPASS platform (Sowa et al., 2009).

For whole-cell proteomic analysis, cells were lysed in 8 M urea with 20 mM Tris-HCl before being sonicated using a Misonix 300 water sonicator; 10 s on, 10 s off for 1 minute at amplitude 5.0. Lysates were subsequently cleared by centrifugation at 16,100 g for 10 min at 4°C. Total protein concentration was measured using a BCA assay (Pierce) and adjusted to 200 μg/ml. Protein samples were reduced with 5 mM DTT, alkylated with 10 mM iodoacetamide and subject to sequential protein digestion with Lys-C and trypsin (Promega). Digestion was halted with formic acid (final concentration 0.5%), precipitates were cleared by centrifugation at 16,100 g for 8 mins and supernatants were desalted using C18 StageTips (3M Empore) containing 4 mg of Poros R3 resin (Applied Biosystems). Peptides were labelled using TMT 10plex reagent and separated on an offline HPLC. Finally, peptides were resolved on a 3000 RSLC Nano System (Thermo Fisher Scientific) and peptides were analysed via a nanospray ion source into a Q Exactive Plus hybrid quadrupole Orbitrap mass spectrometer (Thermo Fisher Scientific).

Mass spectrometry data generated from GST affinity chromatography and whole-cell proteomic analysis was analysed using MaxQuant and peptides were searched against the UniProt human or Drosophila proteome using Andromeda (Cox and Mann, 2008; Cox et al., 2011). The Perseus platform was used to filter samples and to convert protein LFQ intensities to logarithmic values (Tyanova et al., 2016). Missing values were imputed using the default settings, statistical tests made using Welch's or Student's two-sample two-sided t-tests, and volcano plots generated. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifiers PXD054030, PXD054187 and PXD054197.

U2OS cells were trypsinised, added to culture medium, seeded onto microscope slides (Hendley-Essex), and incubated at 37°C with 5% CO₂. The next day, cells were washed in PBS, fixed in 4% PFA in PBS for 20 min at room temperature and washed again in PBS. Cells were permeabilised in 10% Triton X-100 in PBS for 10 min and detergent was removed with five PBS washes. Cells were blocked in 20% FBS with 1% Tween-20 in PBS for 1 h, blocking buffer was removed and cells were incubated with antibody diluted in blocking buffer for 1 h. Cells were washed first in PBS, then in blocking buffer and then they were incubated with an anti-rabbit-IgG conjugated to AF555 secondary antibody (Thermo Fisher Scientific) and an AF488–GFP booster (Chromotek) diluted in blocking buffer for 1 h. They were washed again in PBS, then in blocking buffer and finally in PBS before most liquid was aspirated, and cells were mounted in Vectashield (Vector Biolabs). Slides were imaged using a 63× oil-immersion objective on a Leica TCS SP8 confocal microscope.

Drosophila melanogaster stocks and crosses were kept on Iberian food [5.5% (w/v) glucose, 3.5% (w/v) organic wheat flour, 5% (w/v) yeast, 0.75% (w/v) agar, 16.4 mM nipagin (methyl-4-ydroxybenzoate) and 0.004% (v/v) propionic acid] at 25°C and 50% relative humidity with a repeating 12-h-light–12-h-dark cycle. The following stocks were used: Oregon R as a control, CFD2_nos-Cas9 (Port et al., 2014), Df(3R)BSC569 (BDSC #25670) – a genomic deficiency that includes the CG9590 locus, fkh-Gal4 on the second and third chromosome (BDSC #78061 and #78060), FAM114A^Δ43 (CG9590 null mutant, this study) and UAS-FAM114A-GFP (this study).

CRISPR/Cas9 was used to generate a Drosophila CG9590 null mutant. To remove the entire coding region of CG9590 a pair of gRNAs was chosen targeting either end of the genomic locus. Both were cloned separately into pCFD3 as previously described (Port et al., 2014; https://crisprflydesign.org/). pCFD3-gRNA-CG9590_2 and pCFD3-gRNA-CG9590_4 were then co-injected into CFD2_nos-Cas9 embryos at a concentration of 100 ng/µl each. G0 flies were crossed to balancer stocks and F1 males were used to set up single crosses to generate stable lines. Once crosses were going, males were removed from vials and used for diagnostic PCRs and sequencing. The genomic DNA was isolated using microLYSIS Plus (Clent Life Sciences). We recovered FAM114A^Δ43, which removes the entire CG9590 genomic locus. The stock is viable and fertile, but we noticed that homozygous nulls do not persist in the presence of a balancer, indicating reduced fitness.

The CG9590 cDNA and a C-terminal eGFP with a GHGTGSTGSGSSR linker in between were cloned into pUAS-K10attB using NEBuilder HiFi DNA Assembly (NEB). Briefly, UAS-K10attB was cut with NotI and XbaI and the CG9590 cDNA and eGFP amplified by PCR with homology arms and the linker sequence added to the oligos for HiFi DNA assembly. The UAS-FAM114A-GFP construct was then injected into embryos carrying an attP40 landing site and expressing the phiC31 Integrase under the vasa promoter. Injections were performed by John Overton (Gurdon Institute, Cambridge, UK). Successful transformants were identified by the presence of red eyes and used to make stable lines.

Wandering third-instar larvae were collected and salivary glands were dissected in PBS. Tissues were fixed in fresh 4% PFA for 30 min and then permeabilised four times for 30 min each in PBS containing 0.3% Triton X-100 (Sigma). Tissues were then blocked four times for 30 min each in PBS containing 0.1% Triton X-100, 5% BSA (Cell Signaling Technology) and primary antibodies were incubated in PBS containing 0.1% Triton X-100, 5% BSA overnight at 4°C. Tissues were washed four times for 30 min each in PBS containing 0.1% Triton X-100 and secondary antibodies were incubated in PBS containing 0.1% Triton X-100, 5% BSA overnight at 4°C. The Chromotek GFP-Booster Atto 488 (Proteintech) was used to boost the GFP signal. Tissues were washed four times for 30 min each in PBS containing 0.1% Triton X-100 and equilibrated in Vectashield with DAPI (2BScientific) overnight at −20C. Tissues were then mounted in Vectashield with DAPI. Images were taken with a Zeiss LSM 900 with Airyscan 2 and processed in Fiji. Nine larvae were analysed per genotype in three technical repeats, and 45 line profiles were obtained with five profiles per imaged larva. Primary and secondary antibodies are in Table S2.

For each genotype five salivary gland pairs from wandering third-instar larvae were dissected in PBS and immediately transferred into RIPA buffer (Sigma) plus protease inhibitors (cOmplete, Roche, PMSF, Sigma) on ice. Samples were homogenised using a Kimble pellet pestle (DWK Life Science) and left on ice for 25 min. NuPAGE 4×LDS sample buffer (Invitrogen) and 5% β-mercaptoethanol (Sigma) were added and the samples heated at 90°C for 10 min. Samples were run on a NuPAGE 4-12% Bis-Tris mini gel (Invitrogen) using MES buffer (Formedium). After transfer to nitrocellulose (Amersham) the membrane was blocked in PBS, 0.1% Tween 20, 3% skimmed milk powder and 1% BSA for 1 h at room temperature. The membrane was cut and primary antibodies were added overnight at 4°C in PBS, 0.1% Tween 20, 1% BSA and 3% skimmed milk powder (Marvel). Membranes were washed and secondary antibodies were added for 1 h at room temperature. After washing the blot was developed using SuperSignal West Femto Maximum Sensitivity Substrate (Thermo Fisher Scientific). A BioRad Chemidoc MP imaging system (Bio-Rad) was used to acquire images. Primary and secondary antibodies in Table S2.

Lectin staining was done as described previously (Tian et al., 2013). Briefly, tissues were fixed for 30 min in fresh 4% PFA at RT. VVA conjugated with TRITC (Rhodamine, Stratech Scientific) or PNA with Alexa Fluor 568 (Thermo Fisher Scientific) were applied at 1 μg/ml (PNA) or 5 µg/ml (VVA), with or without competing sugar: N-acetyl-D-galactosamine (GalNAc, 0.3 M, Sigma) for VVA and D(+)-galactose (0.3 M, Formedium) for PNA. Images were taken on a Zeiss LSM 900 with Airyscan 2 and processed in Fiji software. To quantify granule size, the granule area was measured manually using Fiji software. Six to nine salivary glands from different larvae were analysed for each genotype in two or three technical repeats. All fully visible granules in each image were measured and the data graphed using GraphPad Prism (v10).

We thank Catherine Rabouille, Gunter Merdes, Jennifer Richens, John Kilmartin, John Shin, Kelly Ten Hagen, and Martin Lowe for reagents and advice. Mass spectrometry analysis was performed at the Biological Mass Spectrometry and Proteomics Facility of the MRC LMB.

The peer review history is available online at https://journals.biologists.com/jcs/lookup/doi/10.1242/jcs.262160.reviewer-comments.pdf