Introduction

Many types of lipids are heterogeneously distributed among various organelles of the cell¹. Due to the hydrophobic nature of lipids, their trafficking needs to be facilitated. While vesicular trafficking contributes only to a small portion of lipid trafficking, the majority of lipid trafficking is achieved by a group of lipid transfer proteins (LTPs) that carry lipids using their hydrophobic cavities to stabilize lipids in the aqueous intracellular environment². LTPs often contain targeting domains, motifs, transmembrane regions, amphipathic helices or surface charge to define their donor and acceptor organelles³. Many LTPs localize at two organelles simultaneously at membrane contact sites, where these organelles come closer to exchange information and material, including lipids^{4, 5}.

The human genome encodes over 150 LTPs; new LTPs and LTP families are being characterized every so often^{2, 6–13}. About 50 of these proteins are secreted and are involved in carrying lipids, metals, lipopolysaccharides, and other small molecules in blood plasma². The remaining intracellular LTPs are mainly responsible for intracellular lipid trafficking among organelles. Since many lipid-modifying enzymes are localized to different organelles, LTPs also feed metabolic lipid fluxes, eventually defining cellular lipid levels¹⁴. For example, LTP-mediated trafficking of ceramide from the endoplasmic reticulum (ER) to the trans Golgi is needed for sphingomyelin synthesis¹⁵. While achieving a considerable understanding of some LTPs in the last years, the function of many LTPs in defining cellular levels and distributions remains unclear.

To study LTPs systematically, we designed and created an arrayed gene knockout library targeting 90 intracellular LTPs based on the CRISPR/Cas9 technology. Lipidomics analysis of the library-generated LTP knockout cells confirmed known and identified novel lipid disturbances emerging from loss of LTPs. These included CERT, GLTP, NPC1 and NPC2 knockout cells with altered sphingolipid levels. Furthermore, we identified major sphingolipid imbalances in ORP9 and ORP11 knockout cells, two proteins of unknown function in sphingolipid metabolism. ORP9 and ORP11 form a heterodimer to localize at the ER-trans Golgi membrane contact sites. At this contact site, the ORP9-ORP11 dimer transfers phosphatidylserine (PS) from the ER to the Golgi and phosphatidylinositol-4-phosphate (PI(4)P) in the opposite direction. Consequently, loss of either protein causes phospholipid imbalances in the Golgi apparatus that results in lowered sphingomyelin synthesis capacity at this organelle. Collectively, our LTP knockout library toolbox identified various proteins controlling cellular lipid levels. Among these, we found the ORP9-ORP11 heterodimer defining phospholipid composition of the trans Golgi as a critical step in sphingomyelin synthesis. These findings highlighted how phospholipid and sphingolipid gradients along the secretory pathway are linked at the ER-Golgi membrane contact sites.

Results

1) A CRISPR knockout library targeting lipid transfer proteins

To understand the function of LTPs in defining cellular levels and distributions, we created an arrayed CRISPR/Cas9 knockout library targeting LTPs. We targeted only the intracellular LTPs in the library and excluded extracellular proteins, such as LIPOCALIN and LBP-BPI-CETP families². In this library, 90 wells of a 96-well plate were used for targeting LTPs and 6 for non-targeting (NT) controls (Fig. 1A). For optimal gene disruption, we used a lentiviral gene delivery system and 3 guide RNAs per gene – a method also used by others to increase the frequency of gene disruption¹⁶. We applied the LTP knockout library to MelJuSo, a human melanoma cell line. The efficiency of the knockout strategy was confirmed by western blotting of selected wells (Fig. 1B).

Figure 1:

Next, we performed whole-cell lipidomics analysis of the LTP knockout cells. To this end, library-generated MelJuSo cells were grown in lipid-depleted serum and analysed using a lipidomics platform detecting 17 different lipid classes thrice on average (Fig. 1A). One of the detected lipid classes, hexosylceramide, can correspond to glucosylceramide and galactosylceramide – two isomeric lipids in mammalian cells. By metabolic chasing of a fluorescent ceramide analogue in cells silenced for glucosylceramide synthase or galactosylceramide synthase; or treated with the glucosylceramide synthase inhibitor PDMP, we validated that glucosylceramide is the major hexosylceramide in MelJuSo cells (Fig. S1). This further substantiated the previous reports of galactosylceramide being present mainly in oligodendrocytes¹⁷.

We next calculated the z-scores within each lipid class based on their relative abundance per measurement and plotted their average for LTP knockout cell lines. The analysis revealed many LTP knockout cells with lipid imbalances (Fig. 1C, Fig. S2A). Meanwhile, no z-score above the absolute value of 2 was observed for NT control cells (Fig. S2B). Overall, the lipidomics analysis of the LTP knockout library uncovered numerous candidate regulators of cellular lipid levels.

2) NPC1 and NPC2 knockout cells accumulate sphingomyelin in lysosomes and the plasma membrane

Lipidomics analysis of the LTP knockout library revealed various anticipated imbalances. Among these candidate regulators of cellular lipid levels, those related to sphingolipid metabolism are notable. This is likely due to many enzymatic steps of sphingolipid metabolism being more linear and lacking the elasticity of phospholipids^{1, 14}. In the analysis, we identified GLTP knockout cells with lowered glucosylceramide levels (Fig. 2A, Fig. S3A-C). Loss of GLTP is recently reported to block ER-to-Golgi anterograde vesicular trafficking that results in lowered glucosylceramide synthesis¹⁸.

Figure 2:

In addition, we observed NPC1 and NPC2 knockout cells with elevated sphingolipid levels (Fig. 1B). Mutations in NPC1 and NPC2 genes are the cause of Niemann-Pick disease, type C that leads to lysosomal accumulation of cholesterol¹⁹. Cells lacking NPC1 or NPC2 function also accumulate sphingolipids²⁰. Our observation of elevated sphingolipid levels in NPC1 and NPC2 knockout cells was present in all sphingolipid classes (Fig. 2B, C). Meanwhile phospholipid levels were not altered dramatically in these cells (Fig. S3D). To confirm whether this increase is associated with lysosomal accumulation of sphingolipids, we used a sphingomyelin-binding biosensor based on the equinatoxin secreted by the beadlet anemone Actinia equina (Fig. S3E)^{21, 22}. Immunofluorescence staining of NPC1 and NPC2 knockout cells using the biosensor showed lysosomal accumulation of sphingomyelin (Fig. 2D). Flow cytometry analysis using the sphingomyelin biosensor and the GM1 glycosphingolipid biosensor, the B subunit of cholera toxin, demonstrated that sphingolipid accumulation is also reflected on the cell surface of NPC1 and NPC2 knockouts (Fig. 2E, F). Meanwhile, the surface protein levels were not altered as detected by staining for the tetraspanin protein CD63. Collectively, our results corroborate the sphingolipid accumulation in NPC1 and NPC2 deficient cells and illustrate that the LTP knockout library is a viable tool to study LTPs and their role in regulating cellular lipid levels.

3) CERT, ORP9 and ORP11 knockout cells demonstrate reduced sphingomyelin levels

Sphingomyelin is the major sphingolipid in mammalian cells, and it is mainly synthesized in the trans Golgi from ceramide delivered from the ER. The non-vesicular trafficking of ceramide from the ER to the trans Golgi is mediated by the ceramide transfer protein CERT^{15, 23} (Fig. 3B). Our analysis identified CERT knockouts with the lowest sphingomyelin levels (Fig. 1C, 3A). These cells also demonstrate increased glucosylceramide levels, likely as a compensation mechanism (Fig. 3A). Furthermore, we found decreased sphingomyelin and increased glucosylceramide levels in ORP9 and ORP11 knockout cells – two proteins of unknown role in sphingolipid metabolism (Fig. 1C, 3A). The sphingolipid imbalances in all three knockout cells –CERT, ORP9, and ORP11– were also represented in many sphingolipid subspecies (Fig. 3C, Fig. S4B). Lowered sphingomyelin levels of the three knockout cells were further confirmed by their sensitivity towards methyl-β-cyclodextrin treatment (Fig. 3D). Meanwhile, none of the three knockout cells demonstrated phospholipid imbalances to a compelling degree (Fig. S4A).

Figure 3:

The observation that ORP9 and ORP11 knockout cells mimic CERT knockouts implied a role for ORP9 and ORP11 in de novo sphingomyelin synthesis, whereas loss of neither ORP9 nor ORP11 reduced CERT protein levels (Fig. 3E). ORP9 and ORP11 each contain a PH domain mediating localization at the trans Golgi, the site of de novo sphingomyelin synthesis (Fig.3F, G). As loss of either ORP9 or ORP11 did not alter the Golgi ultrastructure as detected by light or electron microscopy (Fig. S5), it further suggested that ORP9 and ORP11 affect sphingomyelin levels by a previously unknown mechanism. Next, we investigated the possible mechanism by which ORP9 and ORP11 regulate cellular sphingolipid levels.

4) ORP9 and ORP11 dimerization is critical for their localization to the ER-Golgi contact sites

ORP9 and ORP11 interact with each other via their regions between their PH and ORD domains²⁴. As no other domain is found in this region, we speculated some secondary structures must facilitate the dimerization. AlphaFold and PCOILS coiled coils prediction tool suggested that ORP9 and ORP11 contain two alpha helices each in this region – hereafter referred as “coils” (Fig. 4A, S6A, B). AlphaFold-Multimer suggested that the coils of ORP9 and ORP11 interact with each other (Fig. 4B, S6C-D). Identical analysis of the full-length proteins predicted that the coils drive dimerization of the two proteins (Fig. 4C, D, S6E). This was tested by targeting the coils of ORP9 to the mitochondrial outer membrane using the N-terminal sequence of TOM70. In cells expressing mitochondria-targeted ORP9 coils, the coils of ORP11 located to this organelle (Fig. 4C, D). The absence of either coil prevented the colocalization, suggesting that the coils of ORP9 and ORP11 are sufficient for their dimerization. This finding was validated by co-immunoprecipitations (Fig. 4E).

Figure 4:

ORP9, but not ORP11, contains a FFAT, two phenylalanines in an acidic tract, motif that drives ER localization by interacting with the ER-resident VAPA and VAPB²⁵. Despite not containing a FFAT motif, ORP11 could interact with VAPA and VAPB by its dimerization with ORP9. We tested this by co-immunoprecipitations, where ORP11 could be co-isolated with VAPA from cells expressing VAPA, ORP9 and ORP11 (Fig. 4J). This interaction was diminished significantly when ORP9 was not co-expressed. Furthermore, a VAPA mutant unable to interact with FFAT motifs failed to interact with either ORP9 or ORP11, showing that ORP11 interacts with VAPA indirectly via the ORP9-FFAT motif.

We and others previously showed that in addition to VAPA and VAPB, the human proteome contains three other proteins interacting with FFAT and related short linear motifs: MOSPD1, MOSPD2 and MOSPD3^{26, 27}. In our earlier efforts to identify interaction partners of VAPA, VAPB and other motif-binding proteins, we performed BioID, proximity biotinylation followed by proteomics, identification of membrane contact sites proteins²⁷. In this analysis, ORP9 and ORP11 were found in proximity with VAPA and VAPB but not with other motif-binding proteins (Fig. S6F). This further confirmed that despite lacking a FFAT motif, ORP11 together with ORP9 is part of VAPA- and VAPB-mediated membrane contact sites.

The Golgi localization of ORP9 and ORP11 is mediated by their PH domain interacting with phosphatidylinositol phosphates²⁴. We observed that the PH domains of ORP9 and ORP11 localized only partially to the Golgi, unlike the PH domains of OSBP and CERT that show exclusive localization to this organelle^{28, 29} (Fig. 4H, S7D). We hypothesized that the dimerization of ORP9 and ORP11 is a mechanism to increase their avidity towards trans Golgi membranes. To test this, we created a chimera containing ORP9- and ORP11-PH domains that localized to the Golgi more efficiently than the individual PH domains (Fig. 4H, I). This also implied that loss of either protein should reduce the other’s localization to the trans Golgi. Immunofluorescence staining of ORP9 and ORP11 knockout cells for the same proteins confirmed that both ORP9 and ORP11 are required for the Golgi localization of the other protein (Fig. 5). Meanwhile, loss of either protein did not influence the levels of the other (Fig. 3E). Our results collectively show that the dimerization of ORP9 and ORP11 via their coils is required for the localization of ORP9-ORP11 dimer to the ER and the trans Golgi simultaneously.

Figure 5:

5) ORP9 and ORP11 are essential for PS and PI(4)P levels in the Golgi apparatus

ORP9 and ORP11 belong to the OSBP-related protein (ORP) family that transfers sterols or PS while transferring phosphatidylinositol phosphates in the opposite direction^30–33 (Fig. 6A, S7B). ORP9 and ORP11 carry the conserved PS binding site and are recently shown to relocate to the site of lysosomal damage to supply PS for membrane repair³⁴. Using an in vitro lipid transfer assay, we confirmed that the lipid transfer domains (ORD) of ORP9 and ORP11 are sufficient to traffic PS in vitro (Fig. 6B, S7C). Moreover, similar to ORP5 and ORP8, the PS-trafficking activity of ORP9- and ORP11-ORDs was improved when acceptor liposomes were decorated with PI(4)P^{35, 36} (Fig. 6C). This indicated that ORDs of ORP9 and ORP11 transport PI(4)P in the opposite direction of PS trafficking.

Figure 6:

PS is synthesized in the ER and its concentration increases along the secretory pathway as it is enriched on the cytosolic leaflet of the plasma membrane¹⁴. PI(4)P, on the other hand, is abundant in the trans Golgi, where it recruits many proteins to this membrane³⁷. Another LTP localized at the ER-trans Golgi membrane contact sites, OSBP transfers cholesterol from the ER to the trans Golgi, while counter transporting PI(4)P to the ER for its hydrolysis by the ER-resident PI(4)P phosphatase SAC1 (Fig. 8). Homologous to OSBP, the ORP9-ORP11 dimer resides in the ideal intracellular interface to traffic PS in anterograde and PI(4)P in retrograde direction between the ER and trans Golgi (Fig. 8). This also suggested that the loss of either protein would cause PS and PI(4)P imbalances between the ER and trans Golgi. We tested this idea by using PI(4)P- and PS-binding biosensors^{38, 39} (Fig. S7D). Golgi localization of these sensors indicated PI(4)P accumulation in the Golgi of ORP9 and ORP11 knockout cells (Fig. 6D, E). Also, PS levels were decreased in the Golgi of these cells (Fig. 6F, G). Both phenotypes were pronounced more strongly in ORP9 knockout cells compared to ORP11 knockouts. Furthermore, both phenotypes, lowered PS and increased PI(4)P levels, could be rescued by reconstitution of the missing protein. Overall, these findings suggested that the ORP9-ORP11 dimer is required for maintaining PS and reducing PI(4)P levels in the Golgi apparatus.

6) ORP9 and ORP11 are needed for de novo sphingomyelin synthesis in the Golgi apparatus

Next, we set out to investigate the possible role of ORP9 and ORP11 in regulating cellular sphingomyelin levels. By performing a sphingomyelin synthase activity assay that uses the fluorescent ceramide analogue NBD-ceramide as a substrate⁴⁰, we found that ORP9 and ORP11 knockout cells do not have reduced sphingomyelin synthesis capacity (Fig. 7A, B). The same was also found in CERT knockout cells. Whereas all three knockouts demonstrated increased glucosylceramide synthesis capacity, supporting the lipidomics analysis (Fig. 3A). Unaltered sphingomyelin synthesis capacity was further validated by the unreduced protein and mRNA levels of both human sphingomyelin synthases, SMS1 and SMS2 (Fig. S8A, B).

Figure 7:

SMS1 is localized at the trans Golgi, while SMS2 shows dual localization at the plasma membrane but is also found at the trans Golgi⁴¹. As the enzymatic activity assay in lysates reports on the global sphingomyelin synthesis capacity, we next investigated de novo sphingomyelin synthesis that occurs at the trans Golgi, where CERT as well as the ORP9-ORP11 dimer localize. For this purpose, we chased the metabolic fate of palmitic acid alkyne (Fig. 7C). Palmitic acid is the precursor of sphinganine and all sphingolipids (Fig. 3B, 7C). To enter the sphingolipid pathway, palmitic acid first needs to travel to the ER and later to the Golgi for conversion to sphingomyelin, unlike ceramide analogues, such as NBD-ceramide, that can be converted to sphingomyelin by the plasma membrane-resident SMS2. Furthermore, the palmitic acid alkyne does not contain a bulky fluorescent group and the terminal alkyne allows visualization by click chemistry (Fig. 7C). As this method enabled us to monitor de novo sphingomyelin production in intact cells, we confirmed the reduced de novo sphingomyelin synthesis in CERT knockout cells using this method (Fig. 7D, E, S8C). In addition, ORP9 and ORP11 knockouts also showed decreased conversion of palmitic acid to sphingomyelin, suggesting a similar defect as CERT knockouts.

The observations that loss of ORP9 or ORP11 does not affect CERT protein (Fig. 3E) or reduce CERT localization to the Golgi (Fig. S9) implied that CERT-mediated transfer routes are not affected in ORP9 or ORP11 knockouts. Lipidomics analysis showing accumulation of ceramide and glucosylceramide in ORP9/ORP11 knockouts further suggested the conversion of ER-bound ceramides to sphingomyelin in the Golgi is hampered.

To distinguish between two possibilities that may result in lowered sphingomyelin levels – a ceramide delivery defect to the trans Golgi or an inability to convert ceramide to sphingomyelin in the Golgi, we established an immunoisolation protocol for trans Golgi membranes using a monoclonal antibody against the trans Golgi marker Golgin-97 (Fig. 7F). These isolates were enriched for the trans Golgi marker Golgin-97 and were devoid of other organelle markers, such as the ER, mitochondria, and lysosomes (Fig. 7G). We confirmed the membrane integrity of Golgin-97-enriched fractions by their sphingomyelin synthase activity, as these multi-pass transmembrane proteins require intact membranes for activity⁴¹ (Fig. S10A). Lipidomics analysis of these fractions validated that the loss of ORP11 and especially that of ORP9 results in lowered PS levels in the trans Golgi (Fig. 7H, S10B-D). The trans Golgi fractions also showed reduced sphingomyelin levels, confirming the limited de novo sphingomyelin synthesis (Fig. 7D, E, I S10E). Moreover, the same fractions showed elevated ceramide levels in these fractions (Fig. 7J, S10F), revealing that the knockouts do not have a ceramide delivery defect, but instead a lowered capacity to convert ceramide to sphingomyelin in their trans Golgi membranes. We tested this notion by performing a sphingomyelin synthase activity assay in the isolated Golgi membranes, where supplying excess substrate to these membranes bypasses the ceramide delivery routes (Fig. 7K, L). This assay further substantiated that the Golgi of ORP11 and especially of ORP9 knockout cells have a lowered sphingomyelin synthesis capacity. Meanwhile, localization of both sphingomyelin synthases was not reduced in the Golgi, as detected by three different antibodies (Fig. S11). Collectively, our results indicate that PS-PI(4)P exchange between the ER and trans Golgi mediated by the ORP9-ORP11 LTP complex is critical for de novo sphingomyelin synthesis in trans Golgi.

Discussion

In this study, we describe a gene knockout library for systematic characterization of intracellular LTPs. The arrayed nature of the library enables high and low throughput analysis. As new LTPs are identified in regular basis, the arrayed feature of the library allows expansion to include newly identified LTPs, e.g. ATG2A, SHIP164, KIAA0100 (Hobbit in Drosophila), KIAA1109 (Csf1 in yeast) and RMDN3 are new human LTPs that are identified through the course of this study^{8–10, 12, 13}.

Lipidomics analysis of the library demonstrated many lipid imbalances. We further validated loss of NPC1, NPC2, CERT, and GLTP with sphingolipid imbalances^{15, 18, 20} (Fig. 1C). In addition to sphingolipids imbalances, we found STARD7 knockout cells with increased levels of phosphatidylglycerol (PG), a lipid class exclusive to mitochondria in mammals (Fig. 1C). STARD7 is a phosphatidylcholine (PC) transfer protein that localizes to the mitochondrial intermembrane space to supply PC to the mitochondrial inner membrane^42–44. Consequently, loss of STARD7 reduces PC levels in this membrane as well as decreasing cardiolipin levels and respiratory capacity^{43, 44}. While PC in the inner membrane is not a precursor of cardiolipin, it is possible the reduced PC levels in this membrane impair the activity of the cardiolipin synthase that uses PG as its substrate, suggesting an explanation for the PG accumulation in STARD7 knockout cells.

LTPs often localize at membrane contact sites to facilitate lipid exchanges between organelles. We found ORP9 and ORP11 localizing at the ER-trans Golgi contact sites to exchange PS and PI(4)P. We show that ORP9-ORP11 dimerization is critical for their contact site localization, and consequently, loss of either protein is sufficient to disturb PS and PI(4)P levels in the Golgi. In return, this phospholipid imbalance reduces sphingomyelin synthesis capacity of the Golgi. The contact site localization of ORP9-ORP11 dimer is homologous to that of ORP9-ORP10 dimer localizing at the ER-endosome contact sites. ORP10 also lacks a FFAT motif and interacts with ORP9⁴⁵. In both LTP complexes, ORP9 provides the FFAT motif required for ER contact. Compared to ORP11 knockout cells, ORP9 knockout cells provide more pronounced phenotypes of cellular sphingolipid imbalances (sphingomyelin and glucosylceramide), phospholipid imbalances in the Golgi apparatus (PS and PI(4)P), as well as reduced capacity of sphingomyelin synthesis in trans Golgi membranes. Furthermore, loss of ORP9 led to a more dramatic effect on the Golgi localization of ORP11 than vice versa. This is most likely due to ORP9 providing a critical factor, i.e. the FFAT motif, for contact site localization, whereas it is possible the loss of ORP11 can be partially compensated by ORP10. ORP10 can also deliver PS to Golgi membranes⁴⁶, however, our lipidomics analysis of ORP10 knockout cells did not demonstrate any sphingolipid imbalances. In response to lysosomal membrane damage, ORP9, ORP10 and ORP11, together with OSBP and ORP1L localize at the lysosomal membrane to provide PS and other lipids for the repair of the damaged membrane. This localization is driven by a burst of PI(4)P generation as a result of membrane damage, highlighting the role of PI(4)P in spatiotemporal control of lipid trafficking and membrane contact site formation^{34, 47}.

An intriguing observation is that the loss of ORP9 causes more accumulation of ceramide than CERT knockouts, despite loss of either protein leading to a comparable reduction of sphingomyelin levels. As the loss of CERT would lead to a ceramide accumulation in the ER, it is possible that such an accumulation in the ER is sensed to reduce ceramide production in this organelle. An ER-resident candidate ceramide sensor for this purpose was suggested previously⁴⁸. Meanwhile, our findings show a ceramide accumulation in the trans Golgi of ORP9 knockouts. Accordingly, ceramide accumulation in the Golgi caused by the loss of ORP9 would fail to “turn on” an ER-localized ceramide sensing machinery, thus leading to further ceramide accumulation compared to CERT knockout cells. This notion also supports our finding that the loss of the ORP9-ORP11 dimer causes ceramide accumulation primarily in the Golgi.

Asymmetric distribution of lipids between two bilayer leaflets is a characteristic feature of the plasma membrane. This is owed to the build-up of sphingomyelin and cholesterol on the outer leaflet and PS on the inner leaflet¹⁴. The transition of the thin, symmetrical ER membrane to a thicker, rigid and asymmetrical one takes place in the Golgi (Fig. 8)¹⁴. Sphingolipid, cholesterol and PS concentrations also increase along the secretory pathway. Various mechanisms are described to drive the sphingolipid and cholesterol trafficking against the concentration gradient, including thermodynamic trapping of cholesterol due to complex formation with sphingomyelin or energy release from the retrograde trafficking of PI(4)P followed by its hydrolysis in the ER^{14, 31, 49}. The same PI(4)P gradient between the ER and Golgi could also power the anterograde trafficking of PS, which, unlike sphingomyelin or cholesterol, is still exposed on the cytosolic leaflet of the Golgi where it has a higher concentration than the ER – thus, maintaining lipid trafficking against the gradient requires energy. In brief, our finding that the ORP9-ORP11 dimer mediated phospholipid exchange promoting the sphingomyelin synthesis reveal further intertwining of lipid gradients along the secretory pathway at the ER-trans Golgi membrane contact sites.

Figure 8:

Conflict of Interest

Authors declare no conflict of interest.

Acknowledgements

Authors thank Ruud Wijdeven, Robbert Kim, Carolina Jost and the members of the Neefjes lab for their technical and critical input.

Materials and methods

Cell culture

MelJuSo and HEK293T (ATCC CRL-3216) cells were cultured in IMDM (Gibco #21980) and DMEM (Gibco #41966) supplemented with 8% fetal calf serum (Biowest #S1810), respectively. MelJuSo cell line authentication was performed by Eurofins Genomics (19-ZE-000487).

Library design, generation and transduction

Targeting and non-targeting guide sequences (Supplementary Table 1) were obtained from the Brunello Human CRISPR Knockout Pooled Library⁵⁰ and cloned into lentiCRISPR v2 plasmid⁵¹ as described previously⁵². Cloning of guide RNAs were individually confirmed by Sanger sequencing.

For library transduction, HEK cells were seeded on a 96-well plate on the day before transfection and transfected with library plasmids and lentiviral packaging plasmids pRSVrev, pHCMV-G VSV-G, and pMDLg/pRRE using polyethyleneimine (Polysciences #23966). A day after replacing the medium, virus was harvested and MelJuSo cells were transduced in the presence of 4µg/ml polybrene (EMD Milipore #TR-1003-G). Transduced cells were subjected to selection using 2µg/ml puromycin.

Lipid extractions and lipidomics analysis

For lipidomics analysis, library-generated cells were expanded to 15-cm dishes and cultured in IMDM supplemented with lipid-depleted serum (Pel-Freez #37217-5) for 3-4 days. Next, cells were scraped in 2% NaCl solution and lipids were extracted following the Bligh-Dyer protocol. In brief, harvested cells were resuspended in 200µL 2% (w/v) NaCl followed by addition of 500µL methanol and 250µL chloroform. Samples were vortexed for 5 min. Following phase separation by adding 250µL chloroform and 250µL 0.45% (w/v) NaCl, samples were centrifuged for 5min at 15,000xg. Next, bottom fractions were collected and dried under continuous nitrogen stream.

Comprehensive, quantitative shotgun lipidomics was carried out as described in detail elsewhere^{53, 54}. Briefly, dried lipid extracts were spiked with 54 deuterated internal standards and dissolved in methanol:chloroform 1:1 containing 10 mM ammonium acetate. Lipids are then analysed with a flow injection method at a flow rate of 8 µL/min applying differential ion mobility for lipid class separation and subsequent multiple reaction monitoring in positive and negative electrospray ionization mode. Using the Shotgun Lipidomics Assistant (SLA) software individual lipid concentrations are calculated after correction for their respective internal standards.

Data analysis & statistics

Data analyses, including lipidomics data analysis, were performed using R (4.1.0 “Camp Pontanezen”) and RStudio (2022.12.0+353) with following packages: ggplot2, dplyr, readr, ggrepel, ggcorrplot2, reshape2, ggbeeswarm, ggsignif, purr, tidyr, and tibble. For Z-score calculations, first the percentages of each lipid class within individual measurements were calculated, followed by calculating Z-scores for individual data points using the following formula: “Z = (x - μ)/σ”, where µ is the mean percentage of a lipid class and σ is the standard deviation. Next, average Z-scores for LTP knockout cell lines were calculated. For POPC normalized data analysis for the analysis of acyl chain distributions and Golgi lipidomics, each lipid subspecies normalized to palmitoyl-oleoyl-phosphatidylcholine (POPC) levels. For Golgi lipidomics, POPC-normalized values were normalized again to the mean of POPC-normalized values for MJS and gNT samples. Raw lipidomics data available at Supplementary Table 2.

Mean values are denoted in all graphs except for boxplots. When present, error bars indicate standard deviations. In boxplots, middle line denotes median, box boundaries denote the first (Q1) and third (Q3) quartiles, and the lower and upper whiskers denote “Q1-1.5*IQR” and “Q3+1.5*IQR”, respectively. IQR: inter-quartile range. For statistical analysis, student’s t-test is used, unless stated otherwise. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001.

Co-immunoprecipitation, SDS-PAGE and Western blotting

Co-immunoprecipitation from HEK293T cells, SDS-PAGE and Western blotting were performed as described previously²⁷.

Immunofluorescence staining and quantifications

Immunofluorescence staining was performed as described previously²⁷. For equinatoxin staining, cells were fixed with 4%PFA:PBS (v/v) and permeabilized with 10µg/ml digitonin in PBS for 15min at RT. Further stainings were performed in PBS. All microscopy images were acquired using a Zeiss LSM 900 with Airyscan. Images were analysed and quantified using ImageJ/FIJI software. To quantify the Golgi localization of lipid biosensors (OSBP-PH and LactC2), first a mask was created using Golgin-97 counter staining and this mask was used to quantify signal from trans Golgi as well as the cytoplasm. Golgi localization of immunofluorescence staining of endogenous proteins (ORP9, ORP11, CERT) was performed first by creating a mask using Golgin-97 and later using this mask to quantify signal of interest relative to signal of Golgin-97. Manders’ coefficient for colocalization was calculated using the Jacop plugin for ImageJ/FIJI.

Flow cytometry analysis

MelJuSo cells were brought to suspension by trypsinization and stained with the indicated probes on ice prior to analysis using a BD LSR-II equipped with 488 and 561nm lasers. Data was analysed using FlowJo v.10 software.

Electron microscopy

Cells cultured in 6-cm dishes were fixed for an hour at room temperature by adding double concentrated fixative to the medium (final concentration: 1.5% glutaraldehyde in 0.1M cacodylate buffer). After three times rinsing with 0.1 M cacodylate buffer, the cells were postfixed with 1% osmium tetroxide and 1.5% uranyl acetate. Cells are dehydrated with a series of ethanol, followed by a series of mixtures of ethanol and EPON (LX112, Leadd) and at the end pure EPON. BEEM capsules filed with EPON were placed on the dishes with the open face down. After EPON polymerization at 40°C the first night and 70°C the second night, the BEEM capsules were snapped off. Ultrathin sections 80 nm were made parallel to the surface of the BEEM capsules containing the cultured cells. The sections were contrasted with uranyl acetate and lead hydroxide and examined with a Tecnai Twin transmission electron microscope (Thermo Fisher, Eindhoven, Netherlands). Overlapping images were automatically collected and stitched together into a composite image as previously described⁵⁵.

Recombinant protein purification

A codon optimized version of cDNA encoding the non-toxic version of equinatoxin²² was ordered from IDT technologies, GFP sequence was obtained from meGFP-C1 vector and a cDNA encoding the fusion protein was synthesized using Gibson Assembly. cDNAs encoding ORD domains of ORP9 and ORP11, and EQT-GFP fusion protein were cloned into a pNKI1.8/GST-expression vector.

E. coli BL21/Rosetta were grown in 2xYT medium and expression was induced when OD₆₀₀ was at 1 for overnight at 18°C. Bacterial pellets were resuspended in GST purification buffer (50mM Tris pH 8.0, 250mM NaCl, 1mM EDTA, 1mM DTT) and lysed by tip sonication. Lysates were cleared by centrifugation at 12,000xg for 30 min. Proteins were purified using Glutathione Sepharose™ 4 Fast Flow (GE Healthcare #17-5132-03) and cleaved using 3C protease protein prior to reverse purification using glutathione beads.

Methyl-β-cyclodextrin and MTT viability assay

Cells were seeded in a 96-well plate and were cultured in OptiMEM (ThermoFisher #31985047) for 3 days. Cells were treated with 10mM methyl-β-cyclodextrin (Sigma #C4555) and cell viability was tested using MTT viability assay (Cayman #21795). Absorbance at 560nm was measured using a BMG ClarioStar plate reader.

Liposome preparation and lipid transfer assay

Donor liposomes composed of 2% NBD-PS (Avanti #810194), 2% Rhodamine-PE (Avanti #810150), 10% DOPE (Avanti #850725), 86% DOPC (Avanti #850375) and acceptor liposomes composed of DOPC with or without 5% Brain PI(4)P (Avanti #840045) were prepared as follows: Lipids were mixed in a glass container and dried under a constant nitrogen flow to create a film. Lipids were freeze-dried at least for two hours under vacuum. Next, lipid films were rehydrated in HKM (50mM HEPES pH 7.2; 120mM potassium acetate; 1mM MgCl₂) buffer for 30 minutes followed by 5 cycles of freeze thaw. Lipids were extruded using Avanti MiniExtruder using a 100nm filter. Uniformity and the concentration of lipids were confirmed using a Wyatt Nanostar Dynamic Light Scattering. For lipid transfer assays, final concentrations of 80µM of each donor and acceptor vesicles were used in 100µL volume and the indicated amounts of protein of interest was added in 5µL. NBD fluorescence in time was measured using a BMG ClarioStar plate reader.

AlphaFold & AlphaFold-multimer predictions

Structure predictions were performed using AlphaFold, AlphaFold-Multimer and ColabFold prediction tools, as described^56–58.

Golgi isolation and quantification of Golgi lipid levels

Cells were washed twice with 0.25M Sucrose and scraped in IM buffer (0.25M Mannitol, 0.5mM EGTA, 5mM HEPES pH 7.4). Samples were passed through a 26xG needle for lysis. Lysates were cleared from nuclei by centrifugation twice at 600xg for 5min. Post-nuclear supernatant was subjected to protein determination using BCA protein determination kit. Equal amounts of protein were supplemented with 150mM NaCl and incubated with Protein A/G Magnetic Beads (ThermoFisher #88802) preloaded either or not with Golgin-97 antibody. Next, samples were washed 3 times with IM buffer supplemented with 150mM NaCl before lipid extraction for lipidomics analysis (described above) or enzymatic activity assay (described below). Amounts of Golgi fractions isolated from different knockouts were confirmed by western blotting.

siRNA transfections and metabolic chasing of BODIPY-C5-ceramide in intact cells

siRNA transfections were carried out as described previously²⁷. siRNA for UGCG (M-006441-02-0005) and UGT8 (M-010270-02-0005) were obtained from Horizon Discovery. MelJuSo cells treated with siRNAs or 10µM PDMP (Cayman #62595) were washed twice with PBS and incubated with serum-free medium supplemented with 0.25µM BODIPY-C5-Ceramide complexed to BSA (Thermo Fisher #B22650). After 3 hours, cells were subjected to lipid extraction and thin layer chromatography as described below. The following lipids were used as standards: TopFlour-C11-galactosylceramide (Avanti Polar Lipids #810266), TopFlour-C11-glucosylceramide (Avanti Polar Lipid #810267) and BODIPY-C5-Lactosylceramide (Thermo Fisher #B34402).

Metabolic chasing of de novo sphingomyelin synthesis

Cells were incubated with 20µM ethanolic palmitic acid alkyne (Cayman Chemical #13266) in serum-free medium for 6 hours. Next, cells were washed twice in PBS and scraped in 2%NaCl solution. Lysates were subjected to protein determination using Pierce BCA Protein Kit (Thermo #23225) and equal amounts of proteins were used for lipid extractions as above. For alkaline hydrolysis, dried lipids were resuspended in 200µL methanolic sodium methoxide and incubated at room temperature for 1h. Samples were added 30µL acetic acid/water (1:4, v/v), 120µL 2% NaCl, and 400µL chloroform to re-extract lipids. Dried lipids were resuspended in 15µL chloroform and 65µL of click mix containing 400µM 3-Azido-7-hydroxycoumarin (Baseclick #BCFA-047-1) and 900µM tetrakis(acetonitrile)copper(I) tetrafluoroborate (Sigma #677892) in acetonitrile/ethanol (7:3, v/v) was added and incubated 3h at 45°C prior to TLC analysis.

Quantitative RT-PCR

RT-PCR was performed as described previously²⁷ using the following primer sets: CERT_1: ATGTCGGATAATCAGAGCTGGA / ATCCTGCCACCCATGAATGTA, CERT_2: TCCATCTGTCTTAGCAAGGCT / GCTGTTCAATGGCATCTATCCA, SMS1_1: TGTGCCGAGTCTCCTCTGA / CCGTTCTTGTGTGCTTCCAAA, SMS1_2: CAGCATCAAGATTAAACCCAACG / TGGTGAGAACGAAACAGGAAAG, SMS2_1: TCCTACGAACACTTATGCAAGAC / CCGGGTACTTTTTGGTGCCT, SMS2_2: CAAATTGCTATGCCCACTGAATC / GTTGTCAAGACGAGGTTGAAAAC

Enzymatic activity assay in lysates and isolated Golgi fractions

For enzymatic activity measurements in lysates, cells were washed twice in 0.25M Sucrose and scraped in IM buffer (0.25M Mannitol, 0.5mM EGTA, 5mM HEPES pH 7.4). Samples were passed through a 26xG needle for lysis. Lysates were cleared from nuclei by centrifugation twice at 600xg for 5min. Post-nuclear supernatant was subjected to protein determination using BCA protein determination kit. Equal amounts of protein in a 50µL volume were mixed with 50µL IM buffer supplemented with 5µM NBD-C6-ceramide (Avanti Polar Lipids #8102109) from an ethanolic solution. Reactions were incubated 1h at 37°C in dark with constant shaking. Next, samples were subjected to lipid extractions and thin-layer chromatography as described. For the enzymatic activity assays in Golgi, isolated Golgi fractions were resuspended in 50µL IM buffer and incubated with NBD-C6-ceramide as above.

Thin-layer chromatography

Dried lipids were spotted on a thin-layer chromatography (TLC) Silica gel 60 (Merck #1.05554.0001) plate and developed in chloroform:methanol:acetone:acetic acid:water (50:10:20:10:5, v/v/v/v/v). Fluorescent images were acquired using a Typhoon FLA9500 equipped with a 488nm laser and a BPB1 filter (530DF20).