GOLPH3 and GOLPH3L maintain Golgi localization of LYSET and a functional mannose 6-phosphate transport pathway

To assess whether loss of LYSET also affects Golgi-localized enzymes apart from GNPT, we re-analyzed our previously reported intracellular proteome and secretome data of Lyset KO mouse embryonic fibroblasts (MEF) (Richards et al, 2022) specifically for type II membrane proteins and known SPPL3 substrates. Surprisingly, we found that only the galactosyltransferase B4GALT5 was markedly increased in the Lyset KO secretome (Fig. 1A), while secretion of other type II membrane proteins, including the previously reported SPPL3 substrates B4GAT1 (Voss et al, 2014) and GALNT2 (Hobohm et al, 2022) was largely unaffected by Lyset deficiency. Immunoblotting of conditioned media from 293FT (Fig. 1B) and HAP1 (Fig. 1C) cells using two distinct B4GALT5-specific monoclonal antibodies confirmed the pronounced B4GALT5 secretion into media of LYSET KO cells, whereas no B4GALT5 was detectable in media of parental (WT) cells. The steady-state concentration of B4GALT5 in the enriched membrane fraction was too low for antibody detection. In both 293FT and HAP1 LYSET KO cells, secretion of the well-established SPPL3 substrates MGAT5, B4GALT1, B4GAT1, and GALNT2 (Voss et al, 2014; Kuhn et al, 2015; Hobohm et al, 2022) was not altered (Fig. 1B,C), demonstrating that loss of LYSET did not globally influence secretion of known SPPL3 substrates and thus does not generally affect SPPL3 activity. To determine whether B4GALT5 secretion from LYSET KO cells was SPPL3-dependent, we generated LYSET- and SPPL3-double-deficient 293FT cells using Cas9 genome editing. In three clones, successful targeting of the SPPL3 locus and loss of SPPL3 protein were confirmed by Sanger sequencing (Fig. 1D) and by immunoblotting (Fig. 1E), respectively. LYSET/SPPL3 KO 293FT clones displayed markedly reduced secretion of SPPL3 substrates, of which B4GALT1 and MGAT5 concomitantly accumulated in the membrane fraction (Fig. 1F). Importantly, no soluble B4GALT5 could be detected in conditioned media of LYSET/SPPL3 KO clones, whereas hypersecretion of the soluble lysosomal lipid synthase CLN5 and the protease cathepsin D due to their impaired M6P-tagging was still apparent (Fig. 1G), excluding a generalized secretion defect. This demonstrated that cell-endogenous SPPL3 is required for cleavage of the membrane anchor and subsequent secretion of B4GALT5 from LYSET-deficient 293FT cells.

LYSET KO cells hypersecrete numerous soluble lysosomal proteins due to impaired M6P-tagging that precludes their lysosomal targeting (Pechincha et al, 2022; Richards et al, 2022; Zhang et al, 2022). Therefore, we assessed whether B4GALT5 secretion from LYSET-deficient cells is caused by the lack of M6P moieties and analyzed isogenic GNPTAB KO 293FT (Fig. 2A) and HAP1 cells (Fig. 2B). These cells express correctly localized LYSET (Richards et al, 2022), but are unable to modify lysosomal enzymes with M6P-tags due to a loss of the catalytic subunits of GNPT encoded by GNPTAB. As expected, LYSET and GNPTAB KO cell lines displayed cathepsin L hypersecretion (Fig. 2A,B). Importantly, also GNPTAB-deficient cells displayed markedly increased secretion of endogenous B4GALT5 resembling the observation made for LYSET-deficient cells. This suggests that B4GALT5 secretion is not specifically caused by the absence of its LYSET-mediated retention but by a lack of M6P formation in these cell lines.

To further substantiate that B4GALT5 is subject to M6P-tagging and MPR-dependent sorting, we treated WT 293FT cells with ammonium chloride. Ammonium chloride prevents the low pH-dependent dissociation of MPR-ligand complexes and leads to occupied MPRs along all transport routes. Thus, all newly synthesized M6P-tagged proteins are secreted and not targeted to lysosomes (Braulke et al, 1987). Following ammonium chloride treatment, both B4GALT5-specific antibodies detected a distinct band in conditioned media of WT 293FT cells (Fig. 2C), which was absent in conditioned media of B4GALT5 KO cells (Fig. 2D). The lower electrophoretic mobility of B4GALT5 secreted from LYSET-deficient cells is due to complex oligosaccharide modifications instead of M6P-modified high mannose-type oligosaccharides, which was abolished upon deglycosylation using PNGase F (Fig. 2C). Next, we knocked out SPPL3 in Flp-In T-REx 293 cells and re-introduced either doxycycline-inducible SPPL3 WT or the catalytically inactive D271N mutant (Voss et al, 2014) (Appendix Fig. S1). Using these models, we found that catalytic activity of SPPL3 is indeed required for ammonium chloride-induced secretion of B4GALT5 (Fig. 2E) and B4GALT5 secretion was similarly dependent on endogenous SPPL3 in hTERT RPE-1 cells (Appendix Fig. S2). Finally, affinity purification of exogenously expressed V5-tagged B4GALT5 using an M6P-recognizing single-chain antibody fragment unambiguously demonstrated M6P-tagging of B4GALT5 (Fig. 2F). In contrast, V5-tagged B4GALT1 could not be precipitated using the M6P-binding antibody fragment. As we did not detect intracellular endogenous B4GALT5 by immunoblot, possibly due to low intracellular protein levels or low sensitivity of the antibodies used, we employed the RUSH system (Boncompain et al, 2012) to visualize intracellular maturation of ER-retained B4GALT5-mCherry upon addition of biotin. Recapitulating our observations for secretion of endogenous B4GALT5, LYSET KO cells over time secreted B4GALT5-mCherry following the biotin-induced synchronized release of the reporter, while only a faint release could be detected in media of WT cells (Fig. 2G). Immunoblotting using an mCherry-specific antibody enabled detection of the B4GALT5 reporter in cell lysates and revealed a differential intracellular maturation of B4GALT5 in WT and LYSET KO cells following release from the ER, again reflecting the complex glycosylation of B4GALT5 lacking M6P-tags in LYSET KO cells. In conclusion, our data reveal that B4GALT5 is subject to M6P-tagging and is targeted to lysosomes for degradation once liberated from its membrane anchor by SPPL3-mediated intramembrane proteolysis. These findings explain the SPPL3-dependent B4GALT5 hypersecretion observed in LYSET and GNPTAB KO cells (Fig. 2H).

The COPI adaptor proteins GOLPH3 and GOLPH3L were recently reported to maintain the cis-trans Golgi polarity and proper Golgi localization of numerous Golgi enzymes, leading to depletion of these enzymes, including B4GALT5, from the Golgi in cells lacking GOLPH3 or both GOLPH3 and GOLPH3L (Rizzo et al, 2021; Welch et al, 2021). To examine whether loss of GOLPH3/GOLPH3L is also accompanied by hypersecretion of B4GALT5, we established GOLPH3/GOLPH3L-deficient HEK293 cells (Appendix Fig. S3) and performed quantitative comparative whole-cell proteome and secretome analyses of GOLPH3/GOLPH3L-deficient and parental HEK293 cells (Appendix Fig. S4; Dataset EV1). When specifically filtering our secretome dataset for type II membrane proteins, we indeed found that secretion of soluble forms of type II membrane proteins was profoundly changed in GOLPH3/GOLPH3L KO HEK293 cells (Fig. 3A). Specifically, we also observed a robust hypersecretion of B4GALT5 from GOLPH3/GOLPH3L KO HEK293 cells. However, direct comparison revealed that LYSET KO cells displayed a substantially stronger increase in B4GALT5 secretion than GOLPH3/GOLPH3L KO cells (Fig. 3B). Unexpectedly, we observed a marked loss of LYSET from all four GOLPH3/GOLPH3L-deficient HEK293 clones analyzed (Fig. 3B,C). Apart from GOLPH3/GOLPH3L clients such as GPP130/GOLIM4, which was destabilized as reported previously (Welch et al, 2021), very few other integral or peripheral Golgi membrane proteins were reduced in GOLPH3/GOLPH3L KO HEK293 cells (Fig. 3C,D), suggesting that the prominent loss of LYSET may be a direct consequence of the loss of GOLPH3/GOLPH3L and not due to global changes within the Golgi membrane proteome. Notably, in GOLPH3 single KO HEK293 cells, the LYSET protein was stable (Fig. 3E), and levels of established Golgi type II membrane protein GOLPH3/GOLPH3L clients (Welch et al, 2021) were also not reduced (Appendix Fig. S3D). Similar results were obtained in GOLPH3 KO HAP1 cells (Fig. 3F; Appendix Fig. S5). In HEK293 cells, targeting of the GOLPH3L locus alone did also not cause a notable reduction of LYSET and Golgi type II membrane proteins (Appendix Fig. S3E–G), suggesting that GOLPH3 and GOLPH3L have redundant functions. Confocal imaging confirmed that LYSET and GNPT were largely lost from GOLPH3/GOLPH3L-deficient HAP1 cells (Fig. 3G,H). Interestingly, we also noted dispersed GM130-immunolabeled cis-Golgi compartments in GOLPH3/GOLPH3L-deficient HAP1 cells (Fig. 3G,H), which was supported by electron microscopy revealing a partial destruction and disappearance of mid-Golgi stacks as well as high numbers of transport vesicles surrounding the Golgi stacks (Appendix Fig. S6).

As expected (Pechincha et al, 2022; Richards et al, 2022), reduced LYSET levels in GOLPH3/GOLPH3L KO HEK293 cells were also accompanied by a loss of the α-subunit of the GNPT complex (Fig. 3C). Therefore, we directly assessed M6P-tagging in GOLPH3/GOLPH3L-deficient HEK293 and HAP1 cells. To this end, media of ammonium chloride-treated cells were incubated with a myc-tagged anti-M6P single-chain antibody fragment bound to magnetic beads. Following elution with M6P and immunoblot analysis, several M6P-tagged proteins between 20 and 130 kDa were detectable in media of WT cells but neither in media of GOLPH3/GOLPH3L, nor of GNPTAB and LYSET KO 293FT and HAP1 cells (Fig. 4A). Accordingly, only in WT anti-M6P-precipitates the precursor forms of cathepsin L and CLN5 were found. In line with impaired M6P-tagging, GOLPH3/GOLPH3L KO HEK293 cells had markedly reduced intracellular levels of mature cathepsin B, L, C, and hexosaminidase B which was accompanied by hypersecretion of the respective precursor forms (Fig. 4B). GOLPH3/GOLPH3L-deficient cells similarly released increased amounts of immature precursor forms of CLN5 and cathepsin D and the maturation of CLN5 and the proteolytic processing of the remaining intermediate form to the mature form of cathepsin D was delayed (Appendix Fig. S7A). The impact on soluble lysosomal proteins was also confirmed by our proteome and secretome data (Fig. 4C,D; Appendix Fig. S7B; Dataset EV1). Peptide data further confirmed the immature nature of cathepsins secreted by GOLPH3/GOLPH3L-deficient HEK293 cells (Appendix Fig. S7C; Dataset EV1) and lysosomal enzymes were similarly affected in GOLPH3/GOLPH3L-deficient HAP1 cells (Fig. 4E; Appendix Fig. S7A). Of note, these observations did not apply to β-glucocerebrosidase (GBA, Fig. 4C,D), which is transported to lysosomes in an M6P-independent manner by the lysosomal membrane protein LIMP2 (Reczek et al, 2007). In the absence of M6P-tagging, M6P-independent transport routes via alternative receptors have been demonstrated to assure lysosomal targeting of a subset of lysosomal hydrolases, among them cathepsin D, which might explain the residual mature cathepsin D form in GOLPH3/GOLPH3L KO cells (Rijnboutt et al, 1991; Markmann et al, 2015; Braulke et al, 2024). As GOLPH3 can be amplified in tumors (Scott et al, 2009), we further generated GOLPH3- and GOLPH3/GOLPH3L-deficient clones of the non-cancerous diploid human cell line hTERT RPE-1 (Appendix Fig. S8). Again, GOLPH3/GOLPH3L KO, but not GOLPH3 KO, led to a reduction of cellular LYSET levels, hypersecretion of immature lysosomal enzymes, and reduced intracellular abundance of the respective mature proteoforms. The decreased intracellular abundance of soluble lysosomal proteins was also apparent when we re-analyzed proteome data published for GOLPH3/GOLPH3L KO U2-OS cells (Welch et al, 2021) (Appendix Fig. S7D). In conclusion, these data demonstrate that GOLPH3/GOLPH3L-deficiency in multiple human cell lines is accompanied by a substantial loss of LYSET and GNPT, leading to impaired M6P tagging of soluble lysosomal proteins and ultimately to their aberrant targeting and maturation.

Transient (Fig. 5A) or stable, moderate-level cumate-inducible (Fig. 5B) re-expression of WT GOLPH3 in GOLPH3/GOLPH3L-deficient HEK293 cells led to restoration of cellular levels of the previously reported GOLPH3/GOLPH3L clients GALNT7 and MAN1B1, but also of LYSET (Fig. 5A,B). In addition, the restoration of mature hexosaminidase B (Fig. 5B) demonstrates a rescue of LYSET function by re-expression of GOLPH3. GOLPH3 was shown to bind Golgi membranes through its phosphatidylinositol-4-phosphate (PI4P)-binding properties located in a positively charged region containing R90 (Dippold et al, 2009; Wood et al, 2009) (Fig. 5C). Moreover, GOLPH3 associates with the coatomer through an N-terminal cluster of arginine residues (R14/R15) (Tu et al, 2012). To examine whether these functional motifs are required for its ability to stabilize LYSET, we similarly re-introduced the GOLPH3 R90L and R14A/R15A mutants into GOLPH3/GOLPH3L KO HEK293 cells. GOLPH3 R90L failed to restore cellular MAN1B1, LYSET and mature hexosaminidase B levels when compared to GOLPH3 WT re-expression (Fig. 5B). In contrast, the ability of GOLPH3 R14A/R15A to rescue MAN1B1, LYSET, and hexosaminidase B levels were only slightly reduced compared to GOLPH3 WT. Collectively, these results indicate that PI4P- and COPI-binding properties of GOLPH3 are required for LYSET stabilization.

In GOLPH3/GOLPH3L-deficient HEK293 cells LYSET accumulated following overnight treatment with Bafilomycin A1, chloroquine or ammonium chloride, which all cause an increase of the lysosomal pH (Fig. 5D). Enrichment of lysosomal and Golgi compartments by differential centrifugation of HEK293 (Fig. 5E) and HAP1 cells (Fig. 5F) revealed that LYSET, but also GNPTα levels were strongly reduced in both the Golgi and the lysosomal fraction of GOLPH3/GOLPH3L KO cells. However, following lysosomal protease inhibitor treatment, LYSET was detectable in the lysosomal fraction (Fig. 5E,F). The GNPT α-subunit also accumulated in the lysosomal fraction of inhibitor-treated HAP1 cells, while in HEK293 cells no significant accumulation was noted, possibly owing to lower expression levels, low antibody sensitivity and/or residual activity of different lysosomal proteases (Pechincha et al, 2022). Upon high-resolution immunofluorescence imaging of parental HAP1 cells, LYSET was almost exclusively found in the Golgi apparatus (Fig. 3G) and consequently did not colocalize with the lysosomal membrane protein LAMP2 (Fig. 5G; Appendix Fig. S9A). In contrast, LYSET-specific staining was largely lost from GOLPH3/GOLPH3L KO HAP1 cells, with only little LYSET remaining in GM130-positive regions (Figs. 3G and 5G; Appendix Fig. S9A). Notably, when GOLPH3/GOLPH3L KO HAP1 cells were treated with protease inhibitors, a brighter, vesicular LYSET staining was observed, and LYSET was partially co-localizing with LAMP2-positive lysosomes (Fig. 5G; Appendix Fig. S9A). Reminiscent of the observations we reported for LYSET-deficient cells (Richards et al, 2022), also the GNPT complex disappeared from GOLPH3/GOLPH3L KO cells and was mostly found in lysosomes upon treatment with protease inhibitors (Fig. 5G,H; Appendix Fig. S9). Taken together, this demonstrates that Golgi localization of LYSET and, as a consequence, the GNPT complex, is abolished in the absence of GOLPH3/GOLPH3L and LYSET is subsequently trafficked to and degraded in lysosomes. Degradation of LYSET in lysosomes is inhibited by increased lysosomal pH or pre-treatment with lysosomal protease inhibitors.

Molecularly, the peripheral Golgi membrane proteins GOLPH3 and GOLPH3L are believed to facilitate retrieval of clients through interactions with the COPI coat on the one hand and their clients’ cytosolic tails on the other hand (Schmitz et al, 2008; Tu et al, 2008; Chang et al, 2013; Isaji et al, 2014; Pereira et al, 2014; Eckert et al, 2014; Welch et al, 2021; Rizzo et al, 2021). We could not directly assess co-localization of cell-endogenous GOLPH3/GOLPH3L and LYSET due to a lack of sensitive antibodies from different host species, but endogenous GNPT and GOLPH3 co-localized in HAP1 cells (Fig. 5H; Appendix Fig. S9B). All known GOLPH3/GOLPH3L clients are type II membrane proteins which harbor positively charged amino acids in their short cytoplasmic, N-terminal regions directly preceding their transmembrane domain (TMD) (Rizzo et al, 2021; Welch et al, 2021). LYSET deviates markedly from known clients, as it is annotated as double-pass membrane protein. However, its cytosolic N-terminus harbors numerous positively charged amino acid residues (Fig. EV1A). We therefore used Alphafold2 to model a potential interaction of GOLPH3 and LYSET (Fig. EV1B,C). With its PI4P-binding pocket (formed by e.g. R90 and R171) facing the cytosolic membrane boundary, GOLPH3 was predicted to primarily be in contact with the arginine-rich cytoplasmic region of LYSET immediately preceding its TMD1 and with the loop region directly following LYSET TMD2 (Fig. EV1C,D). In this model, LYSET R29, R32, R37, and R39 are in close contact with a negative surface region formed by GOLPH3 D247, E250, D258, and D262 (Fig. EV1C,D) which had previously been suspected as client binding site (Welch et al, 2021). Taken together, these structural predictions further suggested that membrane-proximal positively charged amino acid residues in the LYSET N-terminal cytosolic region could facilitate GOLPH3 binding.

To test whether GOLPH3 is able to bind LYSET, we performed co-immunoprecipitation experiments. Indeed, LYSET was co-precipitated from lysates of cells transfected with both GOLPH3 and LYSET following GOLPH3 immunoprecipitation with two distinct commercial antibodies (Figs. 6A and EV1E). We additionally observed co-precipitation of endogenous GOLPH3 with the LYSET–GNPT-complex in HEK293 cells overexpressing both LYSET-FLAG and GNPTAB-myc in a doxycycline-inducible manner (Fig. 6B). Overexpressed GOLPH3 also co-precipitated cell-endogenous COPB (Fig. 6A), a subunit of the COPI coatomer and a known interactor of GOLPH3 (Tu et al, 2012; Welch et al, 2021). Interestingly, whereas LYSET-FLAG was strongly enriched compared to input levels in the FLAG-immunoprecipitation, this was not the case for GOLPH3, which is in line with GOLPH3 having multiple competing Golgi membrane proteins clients (Welch et al, 2021) besides LYSET. In conclusion, our data support that LYSET is a novel, non-type II membrane protein client of GOLPH3/GOLPH3L.

Next, we transiently expressed LYSET constructs, in which individual arginine residues in the N-terminal cytosolic region predicted to be in close contact with GOLPH3/GOLPH3L had been mutated, in HEK293 cells. Replacement of the membrane-distal residues R22, R29 or R32 by alanine and also the R39W (known to cause a mucolipidosis type II/III-like phenotype in humans (Ain et al, 2021)) and R39A mutations had no impact on LYSET protein levels following transient transfection (Figs. EV1F and 6C). However, in case of the R37A mutant, LYSET levels in cell lysates of transfectants were markedly reduced compared with LYSET WT and this was similarly observed for the R22A/R37A, R29A/R37A and R37A/R39A double mutants and to a lesser extent for the R32A/R37A double mutant (Fig. EV1F). A double mutant, in which arginine residues R37 and R39 were replaced by positively charged lysines displayed no reduced expression in transfected cells (Fig. 6C). Using CRISPR/Cas9 genome editing, we then established LYSET R37A/R39A knock-in hTERT RPE-1 and HEK293 cells (Fig. 6D). In agreement with the overexpression data, LYSET levels in R37A/R39A knock-in hTERT RPE-1 (Fig. 6E) and HEK293 (Fig. 6F) clones were markedly reduced compared to the parental cells. Moreover, knock-in cells hypersecreted immature hexosaminidase B and cathepsin L and displayed markedly reduced intracellular levels of the mature proteoforms of these enzymes, although this effect was not as pronounced as in LYSET KO clones. Collectively, our transfection and knock-in data show that R37, one of the arginine residues in the N-terminal cytosolic region predicted to facilitate an interaction with GOLPH3/GOLPH3L, is a key determinant of LYSET stability. Whether this is a direct consequence of reduced interaction with GOLPH3/GOLPH3L remains to be elucidated.

Finally, considering that LYSET interacts with and stabilizes GNPTAB (Richards et al, 2022; Pechincha et al, 2022) we used the more recent Alphafold 3 (Abramson et al, 2024) to predict a putative ternary complex formed by GOLPH3, LYSET and GNPTAB (Figs. 6G–I and EV2). In agreement with the Alphafold2 prediction, the arginine-harboring N-termini of LYSET isoform 1 but also isoform 2 were predicted to be in contact with the negatively charged surface of GOLPH3. Interestingly, however, Alphafold 3 also predicted that the TMD close to the C-terminus of the β-subunit of GNPT is in proximity to TMD1 and TMD2 of LYSET and that positively charged arginine residues in the C-terminal cytosolic tail of the GNPT β-subunit are positioned above a separate negatively charged patch on the GOLPH3 surface (Figs. 6G–I and EV2C–E). Additional modeling suggested that dimerization of GNPTAB would not interfere with the formation of a GOLPH3:LYSET:GNPT complex, as the dimer interface, in agreement with recent cryo-EM data of a soluble GNPT construct (Li et al, 2022), was primarily formed by ectodomain contacts (Fig. EV2F). However, the predicted interaction does not take into account the activation of GNPTAB by site-1-protease-mediated cleavage (Marschner et al, 2011). Interestingly, in this predicted complex, the GNPTα N-termini, which include a stretch of positively charged amino acids previously reported to directly interact with the COPI coat (Liu et al, 2018), are similarly found in close proximity of the GOLPH3 adapter proteins. In sum, these predictions raise the possibility that the COPI adapters GOLPH3 and GOLPH3L may physically interact with LYSET but also with GNPT subunits.

HEK293 and hTERT RPE-1 cells were from the American Type Culture Collection (ATCC), and 293FT cells were from Thermo Fisher Scientific. HAP1 cells as well as LYSET KO and GNPTAB KO HAP1 cells were from Horizon Discovery and were reported earlier (Richards et al, 2022). B4GALT5/B4GALT6 KO HEK293 cells (clone 1F6) were reported earlier (Narimatsu et al, 2019). Flp-In T-REx 293 cells were from Thermo Fisher Scientific. LYSET KO, GNPTAB KO (Richards et al, 2022) and SPPL3 KO (Hobohm et al, 2022) HEK293 cells were reported previously. All cells were cultivated under standard conditions and confirmed negative for mycoplasma contamination by routine PCR. HEK293 and Flp-In T-REx 293 cells were kept in DMEM GlutaMAX™ medium (Thermo Fisher Scientific) supplemented with 10% (v/v) fetal calf serum (PAN biotech or Serana), non-essential amino acids (Sigma-Aldrich or Thermo Fisher Scientific) and penicillin/streptomycin (Sigma-Aldrich). Medium for Flp-In T-REx 293 cells was also supplemented with 15 μg/ml blasticidin and 100 μg/ml zeocin (Invivogen), and tetracycline-free fetal calf serum (Capricorn Scientific) was used for these cells. HAP1 cells were kept in IMDM medium (Sigma-Aldrich or Thermo Fisher Scientific) supplemented with 10% (v/v) fetal calf serum, l-glutamine (Sigma-Aldrich), and penicillin/streptomycin. For hTert RPE-1 cells, DMEM/F12 medium containing HEPES and L-glutamine (Sigma-Aldrich) supplemented with 10% (v/v) fetal calf serum and penicillin/streptomycin was used.

HEK293 cells that stably overexpress both LYSET-FLAG and GNPTAB-myc were generated using lentiviral transduction. Lentivirus was produced in 293FT cells by co-transfection of the transgene-expressing plasmid (GNPTAB-myc-IRES-LYSET-FLAG in pLenti_CMVTRE3G-Puro-DEST, see “Molecular cloning”) and packaging plasmids ∆VPR, VSV-G and pAdvantage using TransIT-LT1 transfection reagent (Mirus, Cat# MIR 2300). The lentivirus-containing supernatant was harvested and filtered (0.45 µm). The lentivirus solution was then used to transduce the HEK293 Tet-On 3 G cell line (TaKaRa, Cat# 631182), which was supplemented with protamine sulfate at a final concentration of 8 µg/mL. Approximately 48 h post transduction, 2 μg/mL puromycin was added to select cells that had been successfully transduced.

Genome editing was performed using Cas9 RNPs delivered by electroporation. Target-specific pre-designed crRNAs (Appendix Table S1) as well as generic tracrRNA and Cas9 protein were purchased from Integrated DNA Technologies. crRNA(s) and tracrRNA were mixed at a molar ratio of 1:1 (total crRNA:total tracrRNA). Cas9 loading was performed according to the manufacturer’s instructions. 0.2 × 10⁶ HEK293, 293FT, hTERT RPE-1 or HAP1 cells were electroporated with Cas9 RNP (30 pmol Cas9 & 36 pmol crRNA:tracrRNA duplex) in 10 μl buffer R using a Neon electroporation device (Thermo Fisher Scientific) with 2 pulses, each 1150 V and 20 ms width, for HEK293 cells, 2 pulses (1300 V and 20 ms width) for hTERT RPE-1 and 3 pulses (1575 V and 10 ms width) for HAP1 cells. SPPL3 KO in Flp-In T-REx 293 cells was achieved using a multi-guide approach with reagents obtained from Synthego as detailed earlier (Hobohm et al, 2022). LYSET knock-in cells were generated as described earlier (Hobohm et al, 2022; Truberg et al, 2022) with the electroporation conditions detailed above and using a single-strand oligonucleotide (Integrated DNA Technologies). Following electroporation, cells were rested in 12-well plates for 72 h. They were then seeded on 96-well plates for single-cell cloning. Clones were expanded, genotyped and cryopreserved. Genomic DNA was isolated using a Quick-DNA microprep kit (Zymo Research) or DirectPCR lysis reagent (Viagen). Genomic regions flanking the editing sites were PCR-amplified with specific oligos (Appendix Table S1) using DreamTaq (Thermo Fisher Scientific) polymerase. Detailed PCR reaction conditions can be made available upon request. PCR products were purified using DNA clean & concentrator kits (Zymo Research) and Sanger-sequenced using oligos also used for PCR amplification at Eurofins Genomics. Sequencing results were analyzed and visualized using CLC Main Workbench 8 (Qiagen). Sanger read deconvolution to assess genome editing results in heterozygous clones was performed using DECODR (Bloh et al, 2021).

Restriction enzymes, PNK and T4 ligase were from Thermo Fisher Scientific or New England Biolabs. Inserts of all newly generated plasmids and expression constructs were confirmed by Sanger sequencing at Eurofins Genomics. Sequences of all oligonucleotides used for molecular cloning in this study are compiled in Appendix Table S2. Cloning of human SPPL3-V5 into the pENTR1A backbone and the SPPL3-V5 expression construct in pcDNA5/FRT/TO/DEST were described before (Hobohm et al, 2022). Site-directed mutagenesis was performed to introduce the D271N active site mutant into the SPPL3 coding sequence within the pENTR1A backbone. The strategy was designed using the NEBaseChanger (v1) online tool (New England Biolabs). Plasmid DNA was amplified using Q5 polymerase (New England Biolabs) and oligonucleotides listed in Appendix Table S2. Following amplification, the PCR reaction mix was treated with PNK and DpnI for 1.5 h at 37 °C in T4 ligase buffer. T4 ligase was added and the reaction mixture was incubated for 1 h at room temperature and subsequently heat-shock transformed into E. coli DH5a. After Sanger sequencing had confirmed correct introduction of the desired mutant, the SPPL3 D271N coding sequence was shuttled into pcDNA5/FRT/TO/DEST using LR clonase II (Thermo Fisher Scientific). Similarly, the GFP expression construct used was generated by shuttling GFP coding sequence from pENTR1A-GFP-N2 (FR1) (item #19364; Addgene) into pcDNA5/FRT/TO/DEST using LR clonase II. For the generation of GOLPH3 overexpression constructs, the coding sequence of human GOLPH3 was obtained from Addgene (pEF-DEST51-GOLPH3, item #21690). Oligonucleotides were designed to amplify the GOLPH3 coding sequence and replace the sequence coding for a C-terminal V5-His tag with a stop codon. In addition, HindIII and XhoI restriction enzyme cleavage sites were introduced upstream and downstream, respectively, of the GOLPH3 coding sequence during PCR amplification with Q5 polymerase. The purified PCR product was cloned into pcDNA3.1 via HindIII and XhoI (Thermo Fisher Scientific). GOLPH3 point mutations were also introduced by side-directed mutagenesis as detailed above using oligonucleotides listed in Appendix Table S2. To enable cumate-inducible ectopic expression of GOLPH3 WT and its mutants, GOLPH3 coding sequence (with C-terminal tag) was again amplified by PCR and cloned into pBQM812A (System Biosciences) via NheI/NotI. Coding sequences for LYSET isoforms 1 and 2 were amplified from previously described constructs (Richards et al, 2022) and inserted into pcDNA6 myc/His via HindIII and BamHI restriction sites. LYSET mutants were generated as outlined above or using a Q5 side-directed mutagenesis kit (New England Biolabs) following the manufacturer’s instructions. To generate a B4GALT1-V5 expression construct, cDNA encoding B4GALT1 (corresponding in sequence to NP_001488.2) was PCR-amplified from Str-KDEL-flGALT-SBP-tagBFP (Addgene item #65274) and cloned via HindIII and BamHI sites into a previously described pcDNA3.1-derived plasmid containing coding sequence for a C-terminal V5 tag (Hobohm et al, 2022). Similarly, a B4GALT5-V5 expression construct was cloned following PCR amplification of B4GALT5 coding sequence (corresponding to RefSeq ID NM_004776) kindly provided by Marco Trinchera (Insubria University, Italy). To implement synchronized B4GALT5 trafficking using the RUSH system, B4GALT5 coding sequences was PCR-amplified and cloned into Str-KDEL_ST-SBP-mCherry (Addgene item #65265) using PteI and EcoRI. The GNPTAB-myc-IRES-LYSET-FLAG expression construct for lentivirus transduction was generated using Gibson Assembly (New England Biolabs). The coding sequence for GNPTAB was PCR-amplified from pcDNA3.1-GNPTAB (item # 78108; Addgene), and the myc-IRES-LYSET-FLAG-containing fragment was synthesized at IDT. Both fragments were assembled into the pLenti_CMVTRE3G-Puro-DEST (w1118-1) vector (item # 27565; Addgene) and digested with EcoRV using Gibson Assembly.

For some cell culture experiments, HEK293 and Flp-In T-REx 293 cells were plated on poly-L-Lys (Sigma-Aldrich)-coated cell culture plastics. All other cell lines were plated on uncoated cell culture plastics. Cells were plated in the corresponding cell culture medium lacking selection antibiotics, were grown to confluency and then harvested (see below). To collect conditioned cell culture supernatants, cells were washed with PBS and subsequently warm OptiMEM GlutaMAX (Thermo Fisher Scientific) was added. Before analysis, cells were cultivated for another 16 to 24 h in OptiMEM GlutaMAX at 37 °C. Transient transfections in HEK293 cells were performed using X-tremegene 360 (Roche Life Sciences), TransIT-LT1 (Mirus Bio) or Lipofectamine 3000 (Thermo Fisher Scientific) transfection reagents following the manufacturers’ instructions. For treatment with small-molecule inhibitors, these compounds (listed in Appendix Table S3) were added to cell culture media and cells there then cultivated for another 16–24 h at 37 °C. Protease inhibitor treatments for subcellular fractionation and immunofluorescence microscopy experiments were performed for 24 h. Stable re-introduction of SPPL3 into SPPL3 KO Flp-In T-REx 293 cells was achieved following transfection with the Flp recombinase-encoding plasmid pOG44 (Thermo Fisher Scientific) and the corresponding SPPL3-encoding pcDNA5/FRT/TO plasmid (see above) at a ratio of 9:1 using the TransIT-LT1 reagent. Two days after transfection, the selection regimen was changed to 10 μg/ml blasticidin and 100 μg/ml hygromycin B gold (Invivogen) and resistant clones were expanded. SPPL3 expression in these cells was induced by supplementing cell culture media with doxycycline (Sigma-Aldrich) for 24–48 h. For the B4GALT5 RUSH experiment, cells kept in medium supplemented with dialyzed FCS (Capricorn Scientific) were transiently transfected with Str-KDEL_B4GALT5-SBP-mCherry using TransIT-LT1. Twenty-four hours after transfection, cells were washed twice with prewarmed PBS. Serum-free DMEM with or without 50 µM D-biotin (Carl Roth) was added and cells were kept at 37 °C as indicated. To establish GOLPH3/GOLPH3L-deficient HEK293 cells with a cumate-inducible re-introduction of GOLPH3 or mutants thereof, clone VF3 was transfected with the pBQM812A plasmid containing the corresponding GOLPH3 coding sequence along with an expression plasmid for a hyperactive PiggyBac transposase (ratio of both plasmids 5:2) using TransIT-LT1. One day after transfection, cells were expanded in fresh medium. At day three after transfection, cells were diluted into fresh culture medium supplemented with 1 µg/ml puromycin (Invivogen) for selection of pools of stable transfectants. These cells were kept under puromycin selection, expanded, cryoconserved and used for experiments. GOLPH3 expression was induced by supplementing media with 1× cumate (diluted from a 10,000× water-soluble cumate solution, System Biosciences).

To harvest cells, culture medium was removed and plates were washed once with ice-cold PBS. Cells were then gently scraped off the cell culture dish in 1 ml of ice-cold PBS and collected by centrifugation (500 × g, 4 °C, 5 min). The supernatant was completely removed, and pellets were either directly used or frozen at −20 °C. Whole-cell lysates were obtained by resuspending cell pellets in ice-cold RIPA buffer (150 mM NaCl, 50 mM Tris HCL (pH 7.4), 1 mM EDTA, 0.5% (w/v) sodium deoxycholate, 0.1% (w/v) SDS, 1% (v/v) Triton X-100 and protease inhibitor mix (Roche)). Following incubation on ice, lysates were cleared by centrifugation (21,000 × g, 30 min, 4 °C). For membrane isolation, cells were resuspended in ice-cold hypotonic buffer (10 mM Tris (pH 7.6), 1 mM EDTA, 1 mM EGTA, pH 7.6) and sheered using a 23 G needle. Following sedimentation of nuclei (2400 × g, 5 min, 4 °C), membrane pellets were obtained by centrifugation at 100,000 × g. Membrane pellets were washed twice with carbonate buffer (0.1 M Na₂CO₃, 1 mM EDTA, pH 11.3), once with STE buffer (150 mM NaCl, 50 mM Tris, 2 mM EDTA, pH 7.6), and then lysed in RIPA. The total protein content of whole-cell lysates and membrane lysates was determined using BCA and Micro-BCA kits, respectively (both from Thermo Fisher Scientific). Samples were mixed with 4× LDS sample buffer, supplemented with DTT (final concentration 50 mM) and subjected to SDS-PAGE.

Conditioned OptiMEM medium was collected from cell culture dishes and immediately centrifuged for 5 min at 500 × g and 4 °C to remove detached cells. The supernatant was transferred to ultracentrifuge tubes (Beckman Coulter) and subjected to 100,000 × g centrifugation to remove exosomes. The supernatant was carefully removed and frozen. Total protein precipitation from conditioned supernatants was performed using trichloroacetic acid (TCA). Thawed conditioned supernatants were mixed with freshly prepared aqueous solution of 100% (w/v) TCA (Carl Roth or Sigma-Aldrich) to reach a final TCA concentration of 20% (w/v). Solutions were vortexed extensively and incubated on ice for one hour. Following centrifugation at 21,000 × g for 45 min, the supernatant was removed and precipitates were washed twice with acetone (−20 °C). Following acetone evaporation, precipitated proteins were resuspended in 2× LDS sample buffer with 100 mM DTT. For enrichment and subsequent deglycosylation of secreted proteins, collected culture supernatants were centrifuged at 500 × g and 4 °C for 5 min. The supernatant was then transferred to Amicon Ultra-4 centrifugal filter units (10 kDa MW cut-off). It was concentrated roughly 40-fold by centrifugation at 2000×g and 4 °C. SDS (final concentration: 1% (w/v)) was added to the concentrated supernatant, and proteins were denatured at 95 °C for 10 min. Subsequently, the solution was mixed with deglycosylation buffer (50 mM Na₂PO₄, pH 7.2, 12 mM EDTA, 0.4% (v/v) NP-40) to adjust the SDS concentration to 0.2% (w/v). In total, 1 µl PNGase F (Roche) was added and the reactions were incubated at 37 °C overnight and stopped by the addition of sample buffer. Control samples were incubated overnight without the addition of glycosidase.

Immediately after harvesting, PBS-washed cells were lysed in an appropriate amount of lysis buffer (50 mM Tris pH 7.6, 150 mM NaCl, 2 mM EDTA, 0.5% (v/v) Igepal CA-630 (Sigma-Aldrich), freshly supplemented with 1× Protease Inhibitor Cocktail (Roche)) for 30 min on ice. Lysates were cleared by centrifugation at 21,000 × g and 4 °C for 20 min. Protein content was quantitated using a BCA assay (Thermo Fischer Scientific). In all, 200 µg total protein was brought to a total volume of 1 ml with lysis buffer and agitated at 4 °C overnight in the presence of 1 µl GOLPH3 antibody (Protein Tech Group #19112-1-AP or Abcam #ab98023) and 30 µl Protein A Dynabeads equilibrated in lysis buffer (Thermo Fisher Scientific). The next day, the samples were washed four times with lysis buffer and resuspended in 2× LDS sample buffer supplemented with 100 mM DTT. Samples were denatured for 10 min at 70 °C and analyzed by immunoblotting.

Unless described otherwise, immunoblotting was performed as follows: Samples containing LDS sample buffer supplemented with DTT were denatured (10 min at 70 °C in most cases; 30 min at 37 °C in case of SPPL3 immunoblots), separated on user-cast Bis-Tris gels using 1× MOPS buffer (50 mM MOPS, 50 mM Tris, 0.1% (w/v) SDS, 1 mM EDTA, pH 7.7) and transferred to methanol-activated PVDF membranes with 0.45 µm pore width (Merck) by tank blotting in 25 mM Tris, 192 mM glycine, pH 8.3 supplemented with 10% (v/v) methanol. Blocking and antibody incubation steps were done in 5% (w/v) non-fat dry milk (Carl Roth) dissolved in TBS supplemented with 0.2% (v/v) Tween-20 (TBS-T), with the exception of anti-SPPL3 immunostainings which were done using 5% (w/v) bovine serum albumin (Carl Roth) dissolved in TBS-T. All primary (Appendix Table S4) and secondary antibodies (purchased from Dianova, Cytiva or R&D Systems) were diluted in blocking solution and incubation steps were performed overnight at 4 °C and for 1 to 2 h at room temperature, respectively. For immunoblot detection using crude hybridoma supernatants, PVDF membranes were blocked with 5% (w/v) non-fat dry milk dissolved in TBS-T and incubated in hybridoma supernatant diluted with blocking buffer (1:1). Antibody binding was visualized using ECL reagents purchased from Merck (Immobilon ECL Ultra), Thermo Fisher Scientific (SuperSignal West Pico Plus, Femto or Atto), Cytiva (ECL prime) or Bio-Rad Laboratories (Clarity Max) and signals were acquired using an Amersham ImageQuant 800 (Cytiva) or Chemidoc (Bio-Rad Laboratories) system. Image cropping and adjustment was conducted using Adobe Photoshop or Image Lab (Bio-Rad Laboratories). Total protein stain on PVDF membranes was done using the No-Stain protein labeling reagent (Thermo Fisher Scientific) and signals (570/20 bandpass filter) were documented using the Typhoon scanner following dye excitation at 488 nm.

Subcellular fractionation was performed as previously described (Richards et al, 2022). Briefly, HAP1 or HEK293 cells were washed three times with ice-cold PBS and once with 250 mM Sucrose 10 mM HEPES pH 7.4, harvested by centrifugation at 500 × g for 5 min, and homogenized by 25 strokes in a tight-fitting 7 ml Douncer (Wheaton) in 2 ml in homogenization buffer (250 mM sucrose, 1 mM EDTA, 10 mM HEPES, pH 7.4). Nuclei were pelleted by 10 min centrifugation at 800 × g and 4 °C. The post-nuclear supernatant (PNS) was recovered, the nuclear pellet re-homogenized using 15 strokes in 1 mL homogenization buffer and centrifuged as above. The PNS were pooled, and the lysosomes were pelleted by centrifugation at 20,000 × g for 30 min and washed once with homogenization buffer (20k fraction). The 20,000×g supernatants were centrifuged at 100,000 × g for 1 h in a Sorvall Discovery M120 ultracentrifuge to obtain the 100k Golgi-enriched fraction. Both 20k and 100k pellets were solubilized in RIPA buffer (Thermo Fisher Scientific) containing complete protease inhibitors (Roche), lysed on ice for 20 min and cleared by centrifugation at 17,000 × g for 20 min. Protein concentrations were determined using the DC Protein Assay (Bio-Rad Laboratories). Equal amounts of protein were solubilized in LDS sample buffer supplemented with 100 mM DTT (final concentration), incubated at 70 °C for 10 min, and run on 4–12% NuPAGE Bis-Tris gels (Thermo Fisher Scientific) using 1× MES running buffer (Thermo Fisher Scientific). Blotting was performed using tank transfer in Bis-Tris transfer buffer (Thermo Fisher Scientific) containing 10% MeOH and nitrocellulose (for detection of the GNPTα subunit) or RTA transfer stacks (Bio-Rad Laboratories). Cells were treated with protease inhibitors (5 μg/mL E64d (Sigma-Aldrich), 100 μM Leupeptin (Sigma-Aldrich), 100 μM Pepstatin A (Roche)) in growth medium for 24 h where indicated.

HEK293 or HAP1 cells were washed three times with PBS and incubated in OptiMEM containing 10 mM NH₄Cl for 24 h. Conditioned media were collected, detached cells were pelleted by centrifugation at 500 × g for 10 min, and the supernatants concentrated 10x using Amicon Ultra-15 centrifugal filter units (10 kDa cut-off, Millipore). The concentrate was adjusted to 10 mM Tris pH 7.4 and 0.2% (w/v) Triton X-100. In all, 3 µg of scFv M6P-1 (Müller-Loennies et al, 2010) and 30 µl myc magnetic beads (Thermo Fisher Scientific) washed with PBS containing 1 mM EDTA were used to precipitate M6P-tagged proteins at 4 °C for 3 h. Beads were washed three times for 5 min with 500 µl (PBS with 0.2% (w/v) Triton X-100 and complete protease inhibitors), three times for 5 min with 500 µl 10 mM glucose 6-phosphate, 10 mM mannose in PBS containing 0.2% (w/v) Triton X-100 containing complete protease inhibitors, and three times for 5 min with 500 µl 10 mM NaPi (pH 7.4). M6P-tagged proteins were eluted in 2×10 min using 100 µl 10 mM Disodium M6P in 10 mM NaPi (pH 7.4) containing complete protease inhibitors. All steps were performed at 4 °C and on ice, respectively. Overall, 30 µl of the eluates and concentrated media corresponding to 5% of the input were used for immunoblotting using 10% Tris-Glycine gels. For M6P detection, blocking and antibody incubations were performed using 3% (w/v) receptor grade BSA in TBS-T as described previously (Richards et al, 2022).

Confluent HEK293 GNPT-myc-IRES-LYSET-FLAG cells grown on 100 mm culture dishes were treated with 1 µg/mL Doxycycline (Clontech) for 24 h in DMEM supplemented with GlutaMAX (Gibco), 1% penicillin/streptomycin (P/S) and 10% tetracycline-free FBS (tet-FBS; Clontech). After removal of the medium, the cells were washed and harvested and lysed in lysis buffer, (25 mM Tris/HCl pH 7.4, 150 mM NaCl containing 1 mM EDTA, 5% (v/v) glycerol, containing 30 mM n-octylglucopyranoside (OGP) detergent (Sigma-Aldrich), protease inhibitors (Sigma-Aldrich) and PhosStop (Roche) for 30 min on ice, gently vortexing every 10 min. Cell lysates were then centrifuged at 100,000 × g for 15 min in a Sorvall Discovery M120 ultracentrifuge. The supernatant was collected, and the remaining membrane pellet re-extracted in the same volume of lysis buffer for 30 min on ice followed by centrifugation at 100,000 × g for 15 min. Both supernatants were combined and processed for a preabsorption using TBS-washed anti-HA agarose beads (Thermo Fisher) for 2 h on a rotating wheel at 4 °C. The beads were separated by centrifugation. An aliquot of the clarified lysates was saved as an input fraction, while the rest was incubated under rotation overnight at 4 °C with anti-FLAG M2 agarose beads (Sigma-Aldrich). The beads were separated from the non-bound fraction (NB) by centrifugation. The beads were then washed five times with TBS containing 15 mM OGP (TBS-OGP). For the elution of LYSET-FLAG and specifically LYSET interacting proteins the beads were incubated with 250 µl using the FLAG® peptide (Thermo Fisher) at a final concentration of 150 ng/µL TBS-OGP for 30 min on a rotating wheel at 4 °C followed by centrifugation at 5000 × g for 1 min (E1). The elution was repeated four more times (E2-E5). Samples were solubilized using 4X LDS sample buffer (Thermo Fisher) supplemented with 400 mM DTT, separated on NuPAGE 4–12% Bis-Tris Protein Gels (Thermo Fisher), and transferred by tank blotting to nitrocellulose membranes using Tris-Glycine transfer buffer supplemented with 20% (v/v) methanol. Blocking was performed with 5% non-fat milk in TBS containing 0.05% Tween-20 (TBS-T) except for anti-FLAG blots, where 5% receptor grade BSA (Serva) in TBS-T was used. Membranes were incubated overnight at 4 °C with the following antibody dilutions in blocking buffer: rabbit anti-TMEM251 (LYSET) (1:1000; Invitrogen, PA5-61769), rat anti-GNPT α-subunit hybridoma supernatant (1:50) (De Pace et al, 2014), mouse anti-FLAG (1:1000; Sigma-Aldrich, F1804), mouse anti-MYC (1:1000; Cell Signaling Technology, 2276), rabbit anti-GOLPH3 (1:1000; Abcam, 230809).

Horseradish peroxidase (HRP)-coupled anti-rabbit IgG (Biozol, JIM-111-035-003), anti-mouse IgG (Jackson, 115-035-003), or anti-rat IgG (Jackson,112-035-003) were used as secondary antibodies and incubated in blocking buffer for 1 h at room temperature at a dilution of 1:5000. Immunoreactive proteins were detected by enhanced chemiluminescence (Clarity Western ECL substrate kit; Bio-Rad, 1705060) and imaging in a Chemidoc Imaging system (Bio-Rad).

HAP1 cells were fixed using 4% (w/v) PFA in PBS (Santa Cruz) for 12 min at room temperature, washed once with 100 mM Glycine in PBS and two times with PBS. Blocking, permeabilization and antibody incubations were done in 3% (w/v) BSA, 0.2% (w/v) saponin in PBS and washed in PBS. Commercial primary antibodies are listed in Appendix Table S4 and rat anti-GNPTα (1:100) as well as mouse anti-LAMP2 (clone 2D5, 1:200) were reported previously (De Pace et al, 2014; Radons et al, 1992). Alexa Fluor Plus 488 or 594-coupled secondary antibodies and Hoechst 33342 (all Thermo Fisher Scientific) were diluted 1:1000. Coverslips were mounted using Aqua/Polymount (Polysciences) and 16-bit images acquired using a Plan-Apochromat 63×/1.4 Oil DIC M27 oil immersion objective on a LSM880 confocal microscope equipped with an Airyscan detector in SR-mode (Zeiss). Images were acquired, and Airyscan processing was performed using Zen Black (Zeiss). Image J2 (v2.14.0/1.54f; Build:c89e8500e4) was used for adjusting maximal and minimal thresholds and merging single-channel images. Thresholds were equally set for all images analyzed in each experiment. Cells were treated with protease inhibitors (5 μg/mL E64d (Sigma-Aldrich), 100 μM Leupeptin (Sigma-Aldrich), 100 μM Pepstatin A (Roche)) in growth medium for 24 h where indicated.

HAP1 cells were fixed with 1% (w/v) glutaraldehyde and 4% (w/v) paraformaldehyde in 0.1 M phosphate buffer overnight. Thereafter, they were harvested and spun down at 1000×g for 10 min in a 1.5 mL microcentrifuge tube. The cells were spun down three times after each washing step with 0.1 M cacodylate buffer and embedded in 3% (w/v) agarose. The solidified agarose-containing pellet was cut on ice into small pieces of about 1–2 mm³. After three more washing steps, the tissue blocks were prepared as described previously (West et al, 2010). Briefly, blocks were post-fixed in 0.1% (w/v) potassium ferrocyanide-reduced 2% (w/v) osmium tetroxide in cacodylate buffer for 1 h. Thereafter, the blocks were rinsed in distilled water and treated with 0.1% aqueous thiocarbohydrazide, which stains carbohydrates, for 20 min. After further rinsing in distilled water, the tissue was liganded with 2% (w/v) osmium tetroxide again for 30 min before staining with uranyl acetate and lead aspartate. The pieces were further dehydrated using ascending ethyl alcohol concentration steps, followed by two rinses in propylene oxide. Infiltration of the embedding medium was performed by immersing the samples in a 1:1 mixture of propylene oxide and Epon and finally in neat Epon and polymerized at 60 °C. Semithin sections (0.5 µm) were prepared for light microscopy, mounted on glass slides and stained with 1% Toluidine blue. Images were acquired with a JEM- 2100Plus Transmission Electron Microscope at 200 kV (Jeol) equipped with a XAROSA CMOS camera (Emsis).

Parental and GOLPH3/GOLPH3L KO HEK293 cells (from four separately frozen passages) were thawed and seeded onto uncoated 15 cm plastic dishes. After three washes with PBS, cells were incubated with OptiMEM (Thermo Fisher Scientific) for 24 h. The medium was collected, centrifuged 10 min at 500 × g to pellet detached cells and the supernatant concentrated 10x using Amicon-15 centrifugal filter units (cut-off 10 kDa, Millipore). Cells were washed 3× with ice-cold PBS, harvested, pelleted at 500 × g and 4 °C for 10 min, and snap-frozen. Cell pellets were resuspended in 700 μL lysis buffer (4% SDS, 100 mM HEPES pH 8.0) and incubated at 95 °C for 10 min at 1000 rpm in a thermomixer. Subsequently, the samples were lysed using a sonicator (Branson) with three cycles of 30 s each at 4 °C. Cell lysates were centrifuged at 20,000 × g for 30 min at room temperature (RT) and the clear supernatants were transferred to new microtubes. The protein concentration of both cell lysates and concentrated media samples was determined using the DC protein assay (Bio-Rad Laboratories) and 150 μg of protein of each sample (both cell lysates and media) was precipitated by ice-cold acetone. The protein pellet was resuspended in 0.5% RapiGest (Waters), 100 mM NH₄HCO₃, pH 7.8 (Mosen et al, 2021). Proteins were reduced with DTT (final concentration of 5 mM) at 56 °C for 45 min, alkylated with 20 mM acrylamide (final concentration) for 30 min at RT, and the reaction quenched by adding 5 mM DTT (Müller and Winter, 2017). Samples were diluted with 100 mM NH₄HCO₃ to a final concentration of 0.1% RapiGest and proteins digested with sequencing-grade modified trypsin (Promega) at an enzyme/protein ratio of 1:100 (w/w) at 37 °C overnight. The next day, RapiGest was hydrolyzed by adding 1% trifluoroacetic acid (TFA) (final concentration) and incubation at 37 °C, 800 rpm for 30 min. Samples were centrifuged at 20,000 × g for 10 min, and the clear supernatants were transferred to new microtubes. Peptides were desalted with Oasis HLB cartridges (Waters) and dried in a vacuum centrifuge. Dried peptides were resuspended in 5% acetonitrile (ACN), and the peptide concentration was determined using the quantitative fluorometric peptide assay (Thermo Fisher Scientific). Finally, peptides were dried again and resuspended in 5% ACN/5% formic acid (FA).

Peptides were analyzed using a Dionex Ultimate 3000 nanoUHPLC system coupled to an Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific). Peptide separation was performed using a 40 cm × 100 µm (inner diameter) in-house produced spray tip generated with a P-2000 laser puller (Sutter Instrument, Novato, CA, USA) packed with 3-μm Reprosil C18 AQ particles (Dr. Maisch) as analytical column. After loading of peptides directly to the analytical column at a flow rate of 850 nl/min with 100% solvent A (0.1% FA in water), they were separated with a 165 min linear gradient from 5% to 35% solvent B (95% ACN/0.1% FA). Eluting peptides were ionized in the positive ion mode and mass spectrometry (MS) analyses were performed in the data-independent acquisition (DIA) mode. All scans were acquired in the Orbitrap part of the instrument, MS1 scans at a resolution of 120,000, a mass range from m/z 350 to 1200, and a maximum injection time of 20 ms; and MS2 scans at a resolution of 30,000, a dynamic mass range, and a maximum injection time of 60 ms.

For library generation, DIA *.raw files were analyzed with the Pulsar search engine available in Spectronaut (version: 18.6.23). Data were searched against UniProt Homo sapiens (entries: 20,375, release date: 01/2022) in combination with the cRAP database containing common contaminants. The following parameters were defined: enzyme: trypsin (except proline was the next amino acid); number of allowed missed cleavage sites: 2; minimum peptide length: 5 amino acids; mass tolerance: dynamic; fixed modification: propionamide at cysteine; variable modifications: oxidation at methionine and acetylation at protein N-terminus. The high-precision indexed retention time (iRT) concept (dynamic) was applied for retention time alignment. Mass tolerances for matching of precursors and fragment ions as well as peak extraction windows were determined automatically by Spectronaut. Data were normalized based on median abundance, and the results were filtered with 1% FDR on the precursor and protein level (q value < 0.01). For the determination of P values, the post-analysis pipeline of Spectronaut was used with default parameters (differential abundance testing: unpaired t test). Volcano plots were generated using the ggplot2 package in R. Datasets were filtered for type II membrane proteins based on annotations in Uniprot as detailed earlier (Hobohm et al, 2022). Filtering for Golgi membrane proteins was done based on GO term annotations with manual corrections. Soluble lysosomal proteins were identified based on a previously reported manual selection of confidently characterized lysosomal enzymes (Markmann et al, 2017).

The structure of binary LYSET:GOLPH3 and LYSET:GOLPH3L protein complexes was predicted using ColabFold v1.5.5 (Mirdita et al, 2022) based on AlphaFold2 using MMseqs2 (Jumper et al, 2021) and AlphaFold2 Multimer (Evans et al, 2022). In case of GOLPH3, the experimentally determined structure (PDB ID3KN1, Wood et al, 2009) was matched to the monomeric structure of the AlphaFold2 prediction to control for proper folding. In case of GOLPH3L and LYSET where such experimental data were missing predictions of AlphaFold2 Multimer on ColabFold were compared to the AlphaFold Structure Database (Varadi et al, 2022) and ColabFold v.1.5.5 predictions based on RoseTTAFold2 (Baek et al, 2023) and OmegaFold (Wu et al, 2022). The best resulting protein complex structure was visualized in UCSF ChimeraX v1.7.1 (Meng et al, 2023). The pLDDT score, electrostatic potential, and transmembrane domains predicted by DeepTMHMM v1.0.24 (Hallgren et al, 2022) were used to apply the different coloring. The structures of ternary GNPTAB:GOLPH3:LYSET complexes were predicted using AlphaFold 3 (Abramson et al, 2024) through the official AlphaFold 3 server. The seed was set to auto and the main UniProt isoform sequences for human GNPTAB and GOLPH3 (or GOLPH3L) as well as the corresponding sequences for human LYSET isoform 1 and isoform 2 were used. Predicted complexes were visualized as detailed above.

Proteome data published for Lyset KO MEFs (Richards et al, 2022) and GOLPH3/GOLPH3L KO U2-OS cells (Welch et al, 2021) were retrieved from the online supplements. In R, these datasets were filtered for proteins annotated as type II membrane proteins in the human reference proteome in Uniprot as described earlier (Hobohm et al, 2022). In addition, datasets were also filtered for proteins confidently considered SPPL3 substrates (as defined in (Hobohm et al, 2022)). Filtered datasets were then plotted using the ggplot2 package in R.