The lipid transfer protein Saposin B does not directly bind CD1d for lipid antigen loading

Background: Lipid antigens are presented on the surface of cells by the CD1 family of glycoproteins, which have structural and functional similarity to MHC class I molecules. The hydrophobic lipid antigens are embedded in membranes and inaccessible to the lumenal lipid-binding domain of CD1 molecules. Therefore, CD1 molecules require lipid transfer proteins for lipid loading and editing. CD1d is loaded with lipids in late endocytic compartments, and lipid transfer proteins of the saposin family have been shown to play a crucial role in this process. However, the mechanism by which saposins facilitate lipid binding to CD1 molecules is not known and is thought to involve transient interactions between protein components to ensure CD1-lipid complexes can be efficiently trafficked to the plasma membrane for antigen presentation. Of the four saposin proteins, the importance of Saposin B (SapB) for loading of CD1d is the most well-characterised. However, a direct interaction between CD1d and SapB has yet to be described. Methods: In order to determine how SapB might load lipids onto CD1d, we used purified, recombinant CD1d and SapB and carried out a series of highly sensitive binding assays to monitor direct interactions. We performed equilibrium binding analysis, chemical cross-linking and co-crystallisation experiments, under a range of different conditions. Results: We could not demonstrate a direct interaction between SapB and CD1d using any of these binding assays. Conclusions: This work strongly indicates that the role of SapB in lipid loading does not involve direct binding to CD1d. We discuss the implication of this for our understanding of lipid loading of CD1d and propose several factors that may influence this process.


Introduction
The presentation of antigenic molecules is fundamental to host defence and immune regulation. The MHC class I and II molecules present peptide antigens to T cells and the molecular mechanisms of peptide loading and editing are relatively well understood [1][2][3] . However, the mechanism by which lipid antigens are loaded onto the CD1 family of antigen-presenting molecules remains poorly understood. CD1 molecules have structural similarity to MHC class I, including association with β-2microglobulin (β 2 m), but unlike MHC molecules, CD1 molecules are non-polymorphic and the antigen-binding grooves of CD1 molecules are extremely hydrophobic, allowing the binding of most classes of lipids including phospholipids, glycerolipids, lysolipids and glycolipids possessing a huge variety of glycosylated head groups 4 .
Of the five CD1 family members, antigen presentation by CD1d is best characterised. CD1d undergoes complex cellular trafficking in a manner similar to that of MHC class II. Following synthesis in the endoplasmic reticulum, CD1d traffics via the Golgi apparatus to the plasma membrane, and is then internalised and trafficked through early and late endocytic compartments before being returned to the plasma membrane 5 , where CD1d-lipid complexes are recognized by the T cell receptor (TCR) of natural killer T (NKT) cells. Invariant NKT (iNKT) cells, which bear an invariant TCR and are characteristically potently activated by recognition of CD1d loaded with the synthetic glycolipid α-galactosylceramide (α-GalCer), are innatelike T cells involved in the orchestration of immune responses 6 .
Lipid antigens are embedded in cellular membranes 7 and thus are not directly accessible to the lumenal, lipid-binding groove of CD1 molecules. The loading of lipid antigen onto CD1 molecules therefore requires the action of lipid transfer proteins (LTPs) for efficient antigen presentation. The loading of lipids onto CD1d can occur in the ER, at the cell surface and in endocytic compartments. Relatively little is known about how lipid specificity is determined, and studies using secreted or recycling, surface-cleavable human CD1d found differences in the lipid repertoire bound to CD1d, reflecting the lipids present in the compartments through which it had trafficked [8][9][10] . Different LTPs are present in the different cellular locations where lipid antigen loading occurs and are likely to determine the lipid repertoire bound to CD1 molecules 11 . Interestingly, the late endocytic/lysosomal compartment is the site of both glycolipid catabolism and lipid loading of CD1d and the resident LTPs function in both these processes. Importantly, the antigenicity of glycolipids can be regulated by processing of their glycan headgoups prior to presentation to T cells, highlighting the complex interplay between these two pathways 12,13 .
The LTPs present in the lysosomal compartments include ganglioside monosialic 2 activator (GM2A), Niemann-Pick type C 2 (NPC2) and four saposin (Sap) proteins. SapA, SapB, SapC and SapD are the products of the proteolytic cleavage of a prosaposin (PSAP) precursor and are each small, non-enzymatic proteins possessing three disulphide bonds that stabilise an extremely heat-resistant helical structure. The saposins can exist in a "closed" monomeric, globular conformation or, at low pH, adopt a more "open" conformation, forming higher-order oligomers enclosing a hydrophobic cavity into which lipid acyl chains can be buried [14][15][16][17][18][19] . Each saposin functions in conjunction with specific hydrolases to facilitate the degradation of different glycosphingolipids and the loss of saposin function phenotypically resembles the loss of the associated hydrolases 20-25 . There are two proposed mechanisms for how saposins assist in lipid presentation to hydrolases: the "solubiliser" model, whereby saposins encapsulate lipids within a hydrophobic oligomeric complex for presentation to soluble enzymes, and the "liftase" model, where saposins bind directly to membranes destabilising them, allowing membrane-associated hydrolases access to lipid substrates. Structural and cell-based evidence exists to support both these models 19,26-29 . These two models are also relevant for the proposed mechanisms by which saposins load lipids onto CD1 molecules 30 . SapB is thought to behave as a "solubiliser", forming protein-lipid complexes that enhance lipid loading of CD1 molecules, while SapC associates with membranes and may facilitate lipid loading at the membrane surface.
The evidence for saposin-mediated lipid loading of CD1 molecules comes from animal knockout studies, cell-based assays and in vitro assays. Knockout of PSAP in mice results in the dramatic loss of presentation of endogenous iNKT cell ligands by CD1d 31 . The presentation of exogenous lipid antigens, such as α-GalCer, is impaired in PSAP knockout or knockdown cells [31][32][33][34] . In vitro assays have identified that all the saposins, as well as the other lysosomal LTPs, GM2A and NPC2, are capable of transferring lipids onto CD1 molecules 31,35 . However, some LTPs appear to be more efficient than others at loading specific lipid antigens onto different CD1 molecules 30,33,36 . Specificity within the saposin family in this process has been explored via in vitro assays identifying SapB as the dominant saposin for the loading of lipid antigens onto CD1d. Specifically, incubation of CD1d with SapB and α-GalCer has been shown to enhance the stimulation of iNKT cells 33,34 and SapB

Amendments from Version 1
We thank the reviewers for their insightful comments. We have revised the manuscript following their helpful suggestions as follows: We acknowledge that the lack of a detectable interaction does not necessarily mean that there is no interaction. We have edited the abstract and discussed why an interaction too weak to be detected by the assays used in this study is unlikely to be physiologically relevant. This is supported by the fact that strong direct protein-protein interactions are crucial for several different lipid transfer pathways as detailed in the Discussion.
We have performed an additional equilibrium binding experiment where SapB was pre-loaded with the lipid α-GalCer while CD1d remained loaded with the endogenous human lipids with which it co-purifies. In agreement with our earlier results, no interaction was detected in this new experiment. This additional data is now included as part of Figure 2.
We have clarified the nature and relevance of the lipids expected to be associated with CD1d and SapB in the experiments reported in this manuscript and clarified that glycosylation of SapB is not required for lipid loading of CD1d.
Any further responses from the reviewers can be found at the end of the article REVISED can mediate lipid binding to CD1d in T cell independent assays 33 suggesting a direct role for this specific saposin in lipid loading of CD1d. The ability of SapB to facilitate and enhance lipid exchange on CD1d has also been monitored in vitro via iNKT TCR binding to CD1d following incubation with lipids alone or lipids plus SapB 34 . These in vitro assays implicate SapB as a "lipid editor" that facilitates the loading and unloading of lipid antigens onto CD1d.
Due to the hydrophobic property of lipids, their exposure to aqueous solutions is thermodynamically unfavourable. Therefore, lipid transfer between hydrophobic lipid-binding cavities of proteins requires direct interaction of the protein components to provide continuous shielding of the lipid from the aqueous phase 37 . This hypothesis is supported by the fact that other known lipid transfer proteins directly interact with the protein receiving the lipid 38-40 . However, several publications report a failure to detect a direct interaction between saposins and human CD1d 11,33,34 . Unfortunately, these reports do not show any data or detail the approaches used to monitor binding and the implication is that these interactions exist but are weak and very transitory. Therefore we have used a number of robust and sensitive biochemical and structural techniques to monitor direct interactions between SapB and CD1d. Using multiple approaches in a range of conditions we do not observe any evidence of a direct interaction between CD1d and SapB.

Methods
Cloning and cell line production Codon-optimised cDNA was synthesized (GeneArt) encoding human β 2 m (residues 21-119, Uniprot P61769) and the extracellular domain of human CD1d (residues 20-301, Uniprot P15813) with a C-terminal hexahistiding tag (CD1d-H 6 ). A modified version of the piggyBac target protein plasmid PB-T-PAF 41 was constructed retaining the N-terminal secretion signal but with the Protein A fusion removed. CD1d-H 6 and β 2 m were individually cloned into this modified vector, using SpeI and AscI restriction-endonuclease sites, to produce PB-T-CD1d-H 6 and untagged PB-T-β 2 m. For production of an inducible, stable co-expression cell line, HEK293F cells were quadrupletransfected with PB-T-CD1d-H 6 , PB-T-β 2 m, PB-RN and PBase using a DNA mass ratio of 5:5:1:1 and transfected cells were selected with geneticin using the established protocol for the piggyBac expression system 41 .
Codon-optimised cDNA was synthesized (GeneArt) encoding the segment of human PSAP corresponding to SapB (residues 195-274, Uniprot P07602) and subcloned into the bacterial expression vector pET-15b using NcoI and XhoI restriction endonuclease sites to produce untagged protein. SapB was also subcloned into the mammalian expression vector pHLsec 42 using the restriction enzymes AgeI and KpnI to encode a C-terminally H 6 -tagged protein.

Protein expression and purification
Protein expression of CD1d-β 2 m complex in the HEK293F co-expression cell line was induced with 2 µg/mL doxycycline and expression media was collected over a 2 month period. CD1d-H 6 was purified from conditioned media by nickel affinity chromatography in phosphate-buffered saline (PBS) pH 7.4 and eluted using PBS containing 300 mM imidazole. CD1d-β 2 m was further purified by size exclusion chromatography on a Superdex 200 column (GE Healthcare) equilibrated in cross-linking buffer (20 mM HEPES pH 7.0, 150 mM NaCl) for interaction assays or distinct buffers (described below) for crystallisation experiments. CD1d-β 2 m was concentrated to 2 mg/mL and stored at 4 °C for up to one month.
To produce deglycosylated CD1d-β 2 m, cells were treated with 5 µM kifunensine at the same time as protein expression was induced. CD1d-β 2 m was purified by nickel affinity purification and eluted with 100 mM citrate pH 4. CD1d-β 2 m protein was buffer exchanged into 100 mM citrate pH 5.5, using repeated rounds of dilution and concentration using a centrifugal concentrator, concentrated to 2 mg/mL and incubated with 3125 units of Endoglycosidase H (Endo H; New England Biolabs) per milligram of protein for 3 hours at room temperature. Deglycosylated CD1d-β 2 m was purified by size exclusion chromatography (as described above).
Untagged SapB was expressed in Escherichia coli Origami (DE3) cells and purified as described previously 43 . Briefly, cleared lysate was heat-treated, precipitated proteins were cleared by centrifugation and supernatant containing SapB was dialysed overnight in the presence of 20 µg/mL DNAse against anion exchange buffer (50 mM Tris pH 7.4, 25 mM NaCl). SapB was further purified by anion exchange chromatography (HiTrap QSepharose column) followed by size-exclusion chromatography (HiLoad 16/600 Superdex 75 column) in 50 mM Tris pH 7.4, 150 mM NaCl. Purified SapB was concentrated to 16 mg/mL and stored at 4 °C.
His 6 -tagged SapB was expressed in HEK293F cells by polyethylenimine-based transient transfection. After 4 days, SapB-H 6 was purified from conditioned media by nickel affinity and size exclusion chromatography with a Superdex 75 column (GE Healthcare) in cross-linking buffer. Prior to crystallisation trials (but not cross-linking assays), the tag was removed by incubation with 1.2 U/mL Carboxypeptidase A-agarose at 25°C overnight followed by incubation with Ni-NTA resin to remove tagged protein. The supernatant containing untagged SapB was separated from Carboxypeptidase A-agarose and Ni-NTA agarose by passing through a gravity flow column.
α-GalCer loading Endogenously purified lipids were exchanged for the lipid α-GalCer (Avanti Lipids) based on a protocol used to obtain crystal structures of α-GalCer-loaded CD1d 44,45 . Lipid was resuspended in DMSO at 1 mg/mL and dissolved by heating the solution at 80°C for 10 min, with brief sonication and vortexing every 3 minutes. CD1d-β 2 m and SapB were incubated with a 3-fold molar excess of α-GalCer in PBS overnight at room temperature. α-GalCer-loaded proteins were then separated from free α-GalCer by size exclusion chromatography with a Superdex 200 10/300 column equilibrated in 20 mM HEPES pH 7.0, 150 mM NaCl. α-GalCer-loaded proteins were used within 24 hours after size exclusion chromatography to avoid significant hydrolysis of the lipid.

Equilibrium binding assay
Ni-NTA agarose beads were loaded with a saturating amount of CD1d-β 2 m, and complete capture of CD1d-β 2 m was verified by measuring the absorbance of the supernatant at 280 nm. Saturated beads were washed three times in equilibrium assay buffer (50 mM HEPES pH 7.0, 150 mM NaCl, 0.05% digitonin, 2.5 mM CaCl 2 , 2.5 mM MgCl 2 ). Serial dilutions (1:1) of CD1d-β 2 m-saturated beads were made in equilibrium assay buffer supplemented with unloaded Ni-NTA beads to equalize the bead volume across all samples. SapB was added at a final concentration of 680 nM in equilibrium assay buffer. The reaction was incubated with shaking at room temperature for 1 hour. The supernatant containing unbound SapB was separated from the beads by centrifugation (800g, 2 min) followed by separation with a Micro-Spin column (Pierce) (800g, 2 min) to remove all beads. The supernatant was analysed by SDS-PAGE followed by staining with SYPRO Ruby. For analysis of bead samples, these were boiled in SDS-PAGE loading dye, separated by SDS-PAGE and detected by Coomassie staining.
Cross-linking assay Amine-reactive cross-linking agents disuccinimidyl sulfoxide (DSSO) and PEGylated bis(sulfosuccinimidyl)suberate (BS(PEG) 5 ) were dissolved in DMSO and added to 10 µM CD1d-β 2 m alone, 20 µM SapB alone or 10 µM CD1d + 20 µM SapB with final cross-linker concentrations of 0, 200 µM or 1 mM in cross-linking buffer. The reaction was incubated for 30 min at room temperature and terminated by incubation with 20 mM Tris pH 8 for 15 min. The reaction products were analysed by SDS-PAGE followed by Coomassie staining.
X-ray data collection and molecular replacement Crystals were cryoprotected in reservoir solution supplemented with 20% v/v glycerol and flash-cooled by plunging into liquid nitrogen. Diffraction data were recorded at Diamond Light Source (DLS) on beamlines I03, I04 and I04-1. Diffraction data were indexed and integrated using the automated data processing pipelines at DLS 46 implementing XIA2 DIALS 47 , XIA2 3dii or autoPROC and STARANISO 48 depending on the extent of anisotropic diffraction, then scaled and merged using AIMLESS 49 . Version numbers for crystallography data processing software are detailed in Table 1. The model for molecular replacement was generated using the published structure of CD1d-β 2 m bound to lysophosphatidylcholine 50 (PDB ID 3U0P, chains A and B), using phenix.sculptor (Phenix v1.14-3260) 51,52 to remove the lipid ligand from the model and revert the glycosylation site mutations present in this structure. Molecular replacement was performed using phenix.phaser 53 and was followed by rigid body refinement using phenix.refine 54 . Molecular replacement into the highly anisotropic data (Crystal form (iii)) was carried out using both the isotropic and anisotropically processed data, resulting in the same solution. Maps were inspected using COOT (v0.8.9) 55 . Structure figures were rendered using PyMOL (v2.1.0, Schrödinger LLC).

SDS-PAGE analysis of crystals
For each crystallisation condition tested, several crystals were harvested from the crystallisation drop and washed by transferring to a drop of reservoir solution to remove uncrystallised protein components. Crystals were then transferred to a second drop of reservoir solution and dissolved in SDS-PAGE loading buffer. Alternatively, large crystals from which data had been collected at the synchrotron were recovered into SDS-PAGE loading buffer from the crystal-mounting loop. Individual crystal components were analysed by SDS-PAGE followed by Coomassie staining.

Results
Purification of CD1d-β 2 m complex from a human cell line Several strategies have been developed previously for the purification of CD1d including refolding from E. coli inclusion bodies 56 , baculovirus-insect cell expression 50 and mammalian expression systems 57 . An important consideration for CD1d protein expression and purification is the inclusion of the essential partner protein β 2 m. In a manner similar to the heavy chain of MHC class I, CD1d non-covalently associates with β 2 m and this association is required for maturation of its four glycan chains 58 . In order to obtain high quality, near-native CD1d protein for in vitro studies, we chose to co-express human CD1d with β 2 m in the mammalian suspension cell line HEK293F as this would confer correct folding, native post-translational modification and full glycosylation of CD1d. Initial trials using polyethylenimine-based transient co-transfection of CD1d and β 2 m resulted in very low yields. Thus we used a piggyBac-expression system 41 to establish stable and inducible co-expression of full-length, untagged β 2 m with the ectodomain of CD1d possessing a C-terminal hexahistidine tag (H 6 ) ( Figure 1A). Both proteins were expressed with an N-terminal secretion signal sequence targeting synthesis to the endoplasmic reticulum, resulting in protein secretion into the media. This expression strategy results in CD1d molecules that are bound to endogenous human lipids acquired during folding in the endoplasmic reticulum and trafficking through the Golgi apparatus. CD1d-H 6 was purified from conditioned media by nickel affinity chromatography followed by size exclusion chromatography ( Figure 1B and C). Untagged β 2 m co-purified with CD1d, indicating that both proteins are correctly folded and form a stable complex. Although some CD1d alone was purified, we were able to separate this from CD1d-β 2 m following size exclusion ( Figure 1C). Using this expression system, we obtained a high yield (1 mg/100 mL media) of very pure CD1d-β 2 m protein. The yield was over 10 times higher than that obtained with transient co-transfection in the same cell line.

Purification of SapB
The four saposin proteins (SapA-D) are produced from the proteolytic cleavage of the precursor protein PSAP which normally occurs upon delivery to the lysosome. To produce recombinant SapB alone, the segment of human PSAP corresponding to SapB ( Figure 1D) was expressed in E. coli. Untagged SapB was purified following a well-established protocol 43 that exploits the heat stability of saposins followed by anion exchange and size exclusion chromatography ( Figure 1E and F). Previous structural studies with E. coli-expressed SapB identified that it exists as a tight dimer that co-purifies with E. coli phosphatidylethanolamine 14 .
CD1d binding to SapB is not detectable using in vitro interaction assays Classical pull-down experiments involve the capture and immobilisation onto resin of a tagged bait protein followed by incubation with prey protein. Subsequent wash steps remove unbound protein leaving behind protein complexes for detection. However, as the concentration of free prey protein is reduced at each wash, the equilibrium is pushed towards complex dissociation, meaning this method is not well suited for the capture of weak or transient interactions. Using this approach, no interaction was detectable between immobilised CD1d-β 2 m and SapB. Therefore, an alternative approach was adopted to overcome the limitations of prey-depletion by using an in vitro equilibrium binding assay (Figure 2A). In this assay, the bait is bound to beads at a range of concentrations and the prey is added at a low, limiting concentration. Following incubation, and no wash steps, the supernatant is separated from the beads and unbound prey is detected using the sensitive protein dye SYPRO Ruby. Monitoring the depletion of prey from the supernatant by the large excess of bait protein, rather than monitoring complex formation directly, represents a more sensitive measure of binding and thus is optimised for weak or transient interactions 59 .
Equilibrium binding assays were carried out using CD1d-β 2 m as bait at a concentration range of 0-40 µM and with SapB as the prey at 680 nM. Previous in vitro studies have identified that SapB can load α-GalCer onto CD1d at a pH range between 5 and 8 33 . Although this exchange appears to be most efficient at pH 6, this pH causes partial dissociation of the CD1d His-tag from the Ni-NTA resin. Therefore, the equilibrium binding assay was performed at pH 7. Previous work, and our preliminary experiments, showed that β 2 m dissociates from CD1d in the presence of detergent 60 . However, the gentle detergent digitonin was found to cause minimal disruption of the CD1d-β 2 m complex 60 and was used in this assay. Despite the use of this optimised approach, no binding of SapB to CD1d could be detected ( Figure 2B). In these conditions, at high CD1d-β 2 m concentrations, a small amount of β 2 m is detectable in the supernatant due to dissociation from CD1d ( Figure 2B). By comparing how little appears to be lost based on the Coomassie-stained bead samples this highlights the sensitivity of this technique and of the SYPRO Ruby stain for protein detection.  Previous studies have shown that the ectodomain of CD1d co-purifies with a broad range of cellular lipids following synthesis and trafficking through the secretory pathway, but is likely to be primarily loaded with sphingomyelin 8-10 . SapB copurifies with E. coli phosphatidylethanolamine 14 , a lipid species that has also been shown to bind CD1d 61 . Although it seems likely that this lipid composition for SapB and CD1d should be compatible with protein-protein interactions required for lipid exchange, lipid loading onto CD1d has only been formally demonstrated in vitro for a few lipids, and in particular α-GalCer 33,34 . Therefore, both proteins were pre-loaded with α-GalCer prior to performing the equilibrium binding assay. However, this variation also did not result in a detectable interaction between SapB and CD1d ( Figure 2C). To exclude the possibility that a strong interaction between CD1d and SapB only occurs when SapB exchanges a lower affinity antigen for α-GalCer, we repeated the equilibrium binding assay where SapB was pre-loaded with α-GalCer while CD1d retained the co-purified endogenous lipids ( Figure 2D). Again, there was no substantial depletion of SapB from the supernatant in this assay. These results strongly suggest that CD1d and SapB do not directly interact in vitro, or that their interaction is extremely weak with a K d > 40 µM.
In vitro cross-linking does not capture an interaction between CD1d and SapB Cross-linking of proteins using short chemical tethers that react with surface-exposed lysine side-chains can be used to capture transient or weak interactions. Both CD1d and SapB have several surface-exposed lysines, and of particular importance, lysine residues are present near the lipid-binding groove of CD1d where we would predict SapB to bind ( Figure 3A and B). Therefore, it is possible that CD1d and SapB interact in a way that is amenable to intermolecular cross-linking of surface lysines by amine-reactive cross-linkers. We used cross-linkers of different lengths to maximise the chances of capturing an interaction: DSSO, with a 10.3 Å spacer arm, and BS(PEG) 5 , with a 21.7 Å spacer arm. CD1d-β 2 m and SapB were incubated with these cross-linkers either alone or following mixing ( Figure 3C). For the CD1d-β 2 m samples alone, cross-linking between CD1d and β 2 m is clearly evident, as expected. There is also some evidence of additional higher order oligomers, probably due to the large number of surface lysines on both CD1d and β 2 m. However, no additional cross-linking products were observed when CD1d-β 2 m and SapB were incubated together with the crosslinkers. This indicates that CD1d-β 2 m and SapB do not interact together in a conformation that is amenable to cross-linking. SapB possesses one N-linked glycosylation site on residue Asn21 ( Figure 3B), which remains unglycosylated when expressed in E. coli. Although unglycosylated SapB loads lipids onto CD1d 33,34 , it is possible that this glycan on SapB may increase its affinity for CD1d. Therefore, we repeated the cross-linking experiment following expression of His-tagged SapB in mammalian HEK293F cells, enabling the purification of glycosylated protein. Following purification, two SapB-H 6 species were present, with molecular weights of approximately 10 and 13 kDa, likely corresponding to unglycosylated and glycosylated SapB-H 6 ( Figure 3D). It was not possible to separate these two SapB species, likely due to SapB dimerization, and so cross-linking experiments were carried out using this mixed sample ( Figure 3D). However, we again did not observe formation of any additional cross-linked products when CD1d-β 2 m was incubated with glycosylated SapB-H 6 . We note that although the same amount of unglycosylated SapB ( Figure 3C) and glycosylated SapB ( Figure 3D) were used in cross-linking assays, the Coomassie-stained bands corresponding to glycosylated SapB appear much stronger. This increased staining may be due to the additional histidine residues in the affinity tag increasing the amount of Coomassie reagent binding to this small protein.
CD1d-β 2 m and SapB do not co-crystallise An alternative method to capture weak, transient protein-protein interactions is to attempt co-crystallisation. In this experiment both protein components are mixed together at a range of concentrations and different molar ratios, then subjected to crystallisation trials. The benefit of using the sitting-drop vapour diffusion approach is that during the experiment the drop containing protein and precipitant gradually equilibrates with the reservoir, generally resulting in reduced drop volume and the progressive increase in concentration of the drop components. In this way, the protein components are steadily concentrated together with the aim of capturing an interaction that is otherwise extremely hard to measure using biochemical and biophysical assays 62 .
Extensive crystallisation trials containing both CD1d-β 2 m and SapB were performed, varying the following parameters: (1) protein concentration (6-13 mg/mL CD1d-β 2 m, 2-6 mg/mL SapB), (2) molar ratio of SapB to CD1d, taking into account the predicted dimeric state of SapB (2:1-3:1 SapB to CD1d); (3) full glycosylation or deglycosylation of protein components; and (4) lipid exchange for α-GalCer. Figure 4A shows an example of the high purity of the protein components that went into crystallisation trials. These experiments yielded several different crystals in a range of conditions, a subset of which are shown in Figure 4B. To determine the composition of these crystals, two approaches were used depending on the quality of the crystals grown. The first approach, which can be used for any protein crystal, involves harvesting and washing the crystals in reservoir solution to remove residual uncrystallised components, then separating the proteins by SDS-PAGE followed by detection using Coomassie staining ( Figure 4C, each gel lane corresponds to the above crystal in panel 4B). This revealed that both SapB and CD1d-β 2 m were capable of forming crystals but in none of the crystals grown did both components crystallise together. The second approach used was to collect diffraction data for any crystals that diffracted to 5 Å resolution or better, then use molecular replacement to confirm the contents of the asymmetric unit (Table 2). Three crystal forms satisfied this criteria, including two that had also Image of the crystal that diffracted to highest resolution as described in Table 2. Scale bar represents 100 μm. Abbreviations: gCD1d-glycosylated CD1d, dgCD1ddeglycosylated CD1d, gSapB-glycosylated SapB, and ugSapB-unglycosylated SapB. Redundancy 3.4 (3.5) 3.4 (3.4) 6.9 (6.6) 6.7 (6.6)
been tested using SDS-PAGE ( Figure 4B (iii) and (iv) and 4D (v)). These three samples all contained CD1d-β 2 m only as confirmed by molecular replacement trials using multiple copies of all potential components as well as manual inspection of maps. In all datasets additional density was observed for unmodelled glycans on CD1d (deglycosylation of CD1d using Endo H leaves a residual acetylated glycosamine), confirming the validity of the molecular replacement solution. For several CD1d molecules, additional unmodelled density near the lipid-binding groove was evident that likely corresponds to partially-ordered lipid components. Further inspection of electron density maps revealed no evidence for the presence of SapB. For most molecules of CD1d in all three different crystal forms the predicted binding site for SapB (over the lipid-binding groove) was occupied by another copy of CD1d. The use of both SDS-PAGE analysis of all grown crystals and molecular replacement solution of higher quality crystals confirmed that, despite extensive efforts with a broad range of CD1d-β 2 m/SapB combinations, these do not form a stable, crystallisable complex.

Discussion
Using highly purified components we have been unable to identify a direct interaction between CD1d-β 2 m and SapB in vitro. Several biochemical and structural approaches were implemented that have the capacity to capture weak and transient interactions, including binding equilibrium assays, protein crosslinking, and co-crystallisation trials. Various factors that may influence the strength or specificity of these interactions were tested, including different pH and buffer systems, exchange of the lipids loaded onto both protein components, and changing the glycosylation state of each protein. None of these variants resulted in the detection of a direct interaction. Our previous work studying saposin interactions with lysosomal hydrolases showed that although these interactions are sensitive to pH and salt concentrations, they were always detectable, and while detergent/lipid was required, neither the nature of the lipid/detergent nor the glycosylation state mattered 19 .
Direct protein-protein interactions are crucial for several different lipid transfer pathways. The lipopolysaccharide (LPS) transfer cascade resulting in Toll-like receptor 4 (TLR4) activation involves several lipid transfer steps. LPS binding protein (LBP) extracts LPS and transfers it onto the receptor CD14. CD14 then transfers LPS to the TLR4 co-receptor MD-2. A direct, high affinity (K d = 5.52 nM) interaction between LBP and CD14 is necessary for lipid transfer, and lipid transfer promotes rapid dissociation of CD14:LPS from LBP. Similarly, a direct interaction between CD14 and the TLR4:MD-2 complex is required for lipid transfer 38 . The transfer of lipids between proteins is also seen in the formation of Apolipoprotein B (apoB)-containing lipoproteins. Microsomal Triglyceride Transfer protein assists this process by binding to apoB with high affinity (K d 10-30 nM) 63 and transferring lipids onto apoB through an unknown mechanism. Both functions are required for lipoprotein assembly 39 . As another example, Niemann-Pick C1 (NPC1) and Niemann-Pick C2 (NPC2) proteins collaborate to export cholesterol from the lysosome. Soluble NPC2 binds cholesterol in the lysosomal lumen, interacts with the middle lumenal domain of the multipass transmembrane protein NPC1 and transfers cholesterol to NPC1 via a transfer tunnel linking the cholesterol-binding pockets of both proteins. The interaction between NPC2 and NPC1 (K d 0.4-2 µM) is required for cholesterol export 40 . In all cases, a high-to-medium affinity direct interaction between proteins engaged in lipid transfer is detectable. In this study, we were unable to detect a direct interaction between CD1d and SapB despite the assays being capable of detecting very weak interactions. Specifically, the equilibrium assay used in this study can detect protein interactions where the dissociation constant is as weak as 40 µM. In the absence of avidity effects, interactions weaker than 40 µM are unlikely to be physiologically relevant. Furthermore, the cross-linking assay can detect short-lived, transient interactions and the co-crystallography technique uses very high concentrations of protein components allowing for the detection of very weak interactions. It therefore seems unlikely that an interaction that is undetectable by these methods could permit direct lipid exchange between SapB and CD1d.
The study of lipid loading of CD1 molecules is technically challenging as biochemical methods often require the inclusion of detergents, which may interfere with lipid solubilisation and binding. Also, the detection of changes to CD1d-lipid complexes is limited by available reagents including specific antibodies and iNKT TCRs. However, the inability to identify a direct interaction between CD1d and SapB raises a number of important questions regarding the role of saposins in lipid antigen presentation.
It is well established that knockout of the saposin precursor protein PSAP results in impaired presentation of lipid antigens by CD1d. Due to the dual role of saposins in lipid loading and lipid processing, the absence of PSAP will not only alter CD1d lipid loading, but will also significantly alter the composition of the lysosomal lipidome. Indeed, the loss of a single saposin results in the cellular accumulation of multiple, different GSLs. For example, loss of SapB alone, in mice or in humans, results in the accumulation of sulfatides, globotriaosylceramide, and lactosylceramide 64,65 . Therefore, in saposin knockout and rescue models, the effects on antigen processing and the changes to local lipid concentrations may have indirect and unpredictable effects on CD1d loading in the lysosome. Similarly, cell-based assays that involve recognition of CD1d-lipid complexes by NKT cells are complicated by the requirement of some glycolipid headgroups to be processed by lysosomal hydrolases 12 , a process that is itself dependent upon saposins.  34 . This comparison is driven by the crucial roles that Tapasin and SapB play in the binding and exchange of peptides or lipids, respectively, to the antigen-binding grooves of MHC class I or CD1d molecules. In order for Tapasin to carry out this role, it binds directly to MHC class I but as part of a larger complex including additional proteins 2,68 . Although the in vitro assays demonstrating the importance of SapB for CD1d lipid presentation did not require other proteins, it is possible that there may be additional factors involved in enhancing the affinity of SapB for CD1d in cells.
Additional factors may influence the efficiency of lipid loading onto CD1d in cells. For example, the relative abundance of different LTPs in the lysosome may influence the lysosomal lipid composition, altering exchange kinetics. Therefore, it may be important that SapA-D exist in equimolar concentrations in the endolysosome, as ensured by their production from a common precursor. If this is the case, studies involving individual saposins or partial PSAP rescues will be difficult to interpret, potentially explaining some experimental inconsistencies. The lysosomal lipidome is not only influenced by the activity of saposins and sphingolipid processing hydrolases but also by the careful balance of other membrane components such as cholesterol and bis(monoacylglycero)phosphate that are crucial for endosomal vesicle formation. Lipid loading of CD1 in endolysosomal compartments is highly complex, requiring the delicate interplay of lipid binding, lipid processing and lipid exchange. For now, the mechanism of transfer of lipids onto CD1 molecules, as well as the molecular determinants of LTP binding specificity, remains unclear. absence of lipids, influence of NaCl on electrostatic interactions. To this regard, however, it is not clear if the authors have tried the equilibrium binding assay of unloaded CD1d-b2m with aGalCer loaded Saposin B as additional control in Figure 2. This could be an important discriminator in the ability of the two proteins to interact.

Data availability
Although a negative result is always difficult to interpret (and publish), the authors discuss the limitations of their in vitro results, as well as that of previous assumptions stemming from knock out models where the lysosomal lipidome is severely altered. Overall, the results are in agreement with the fundamental distinct mode of action of Saposin C (which has been shown to interact with CD1b and CD1c) and Saposin B. Future studies looking at the ligandome of CD1d associated proteins in the lysosome may shed light on perhaps a larger loading complex which includes Saposin B.

If applicable, is the statistical analysis and its interpretation appropriate? Not applicable
Are all the source data underlying the results available to ensure full reproducibility? Yes

Are the conclusions drawn adequately supported by the results? Partly
No competing interests were disclosed.

Competing Interests:
Reviewer Expertise: Cellular Immunology, innate/adaptive immunity I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.  Figure 2. This could be an important discriminator in the ability of the two proteins to interact. This important concern is similar to that raised by reviewer 1. We have addressed this concern by performing an additional equilibrium binding assay to test the binding of α-GalCer loaded SapB to CD1d loaded with endogenous lipids (Fig. 2D). Although this does not represent an "unloaded CD1d-β2m", as mentioned in this comment, this experiment includes the endogenous lipid that State that SapB used might lack PTM required for binding in Fig 2, although this was tested in Fig 3. State whether conditions in Fig 2C were shown to permit transfer of lipid. Ideally, this should be shown for a parallel experiment under the same conditions.

Is the work clearly and accurately presented and does it cite the current literature? Yes
Is the study design appropriate and is the work technically sound? Partly

If applicable, is the statistical analysis and its interpretation appropriate? Not applicable
Are all the source data underlying the results available to ensure full reproducibility? Yes

Are the conclusions drawn adequately supported by the results? Partly
No competing interests were disclosed.
I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.
Author Response 14 Oct 2019 , University of Cambridge, Cambridge, UK

Maria Shamin
Thank you for your thoughtful comments and constructive feedback. In-line responses to your queries are included below.

My major point about this work is that I could not tell which, if any, experiments also included relevant lipid, particularly pre-loaded into saposin. Either key sentences in the MS should be rewritten to avoid that impression (if it is wrong), or the MS should clearly include the caveat that some aspects of the physiological situation were not reproduced in the binding experiments.
Most experiments were carried out with SapB and CD1d-β2m loaded with the lipids they co-purified with following protein expression. Based on previous studies, these are likely to be phosphatidylethanolamine for SapB and endogenous human lipids that bind during folding in E.coli the endoplasmic reticulum and trafficking through the Golgi apparatus for CD1d-β2m (primarily sphingomyelin). As phosphatidylethanolamine is a lipid species that has been shown previously to bind CD1d then the proteins are not necessarily in a non-physiologically relevant as purified lipid-bound state. However, we accept that in light of no detectable interaction this is a crucial factor to be clear about in the text. To ensure that the lipids used in each experiment is clear, we lipid-bound state. However, we accept that in light of no detectable interaction this is a crucial factor to be clear about in the text. To ensure that the lipids used in each experiment is clear, we have modified the relevant sections of the main text and figures (also see next comment).

On that basis, it has not been addressed if prolonged interaction only occurs when one of the proteins is occupied by lipid. This might even be asymmetric so that prolonged interaction requires lipid in Saposin, but CD1 must be empty.
We agree with the reviewer that the lipid-bound state of the protein components may be a crucial factor in the affinity and on/off-rates of their potential interaction. An experiment to try and address this, suggested by reviewer 1, is now included where the SapB has been exchanged to bind α-GalCer and the CD1d used with endogenous lipids, similar to those that would be bound upon delivery to the endocytic compartments where exchange should occur. This did not result in significant depletion of SapB in equilibrium binding assays (Fig. 2D) suggesting that if the lipid-bound state of these components is a driving factor in the interaction it is sufficiently complex to be outside the scope of this study. Regarding the reviewer's statement that CD1d "must be empty", it is unclear that CD1d actually ever exists in a truly empty form as it appears to always co-purify, from insect or mammalian cells, with lipids bound.
The rationale for this work is explained in 2 sentences of the final paragraph of the Introduction. This rationale is not explained as clearly as it might be. In the 1st sentence, do the authors mean that the 3 refs were written to indicate that the lipid binding proteins must be able to bind each other even when lipid is absent to have a productive relationship for lipid handover? Statements in the referenced papers are not specific about what conditions the individual proteins might be in regarding lipid content for an interaction to occur. Despite this, all references agree that Saposins are crucial for lipid loading of CD1. In the review by Teyton 2018, he states "We have not been able to reliably measure an interaction between any mouse or human CD1 molecules and the four Saps…". In both Yuan 2007 and Salio 2013 they reference previous work identifying interactions between CD1b and/or CD1c with Saposin C but both state that they have been unable to identify direct interactions of saposins with CD1d despite these papers being specifically focused on the role of CD1d and SapB in lipid loading. As the referenced papers are not specific enough in their statements to address the reviewers points we do not feel we can substantially re-write this statement. The lack of detail in the published statements were part of the motivation for our own studies and the range of techniques and conditions we tested.
The different models should be described more, and the limits to the data mentioned in the 2nd sentence should be outlined. As above, we do not feel we can provide more detail as there were no models proposed and no data provided in the references. We have however, extended the first part of the last paragraph of the Introduction based on this reviewer's next comment to try and clarify that a direct interaction was expected for this process.

In addition, the Introduction and Discussion should both describe (in brief and extensive terms, repsectively) what is known for other examples of lipid handover: NPC1/NPV2 MPT/ApoB LBP/CD14 CD14/MD2.
We thank the reviewer for highlighting these additional protein complexes involved in lipid handover and agree that this is an important addition as it helps put the current work in perspective regarding the requirement, in these other systems, for direct protein-protein interactions for lipid transfer. This is especially compelling as the affinities described for these interactions would have been clearly detectable in the assays used here. We have rewritten the final paragraph of the introduction to cite these examples of lipid handover that involve a direct interaction between proteins involved in lipid transfer. We have also added a substantial paragraph to the discussion proteins involved in lipid transfer. We have also added a substantial paragraph to the discussion where we discuss the mechanisms of lipid transfer between these proteins and the requirement for medium-to-strong, direct interactions between proteins engaged in lipid transfer.
Also in discussion of tapasin, describe if the relationship parallels the one envisaged here. Finally, is there handover of peptide from another binding protein to MHC? The original comparison of saposin function to tapasin function was made by Salio in their et al. 2013 paper. The parallels envisaged between these systems are the requirement for a direct protein-protein interaction (MHC I to Tapasin versus CD1d to SapB) and the functional outcome of this interaction being a change to the antigen content of the binding groove (MHC I-peptide versus CD1d-lipid). Tapasin has been shown to bind directly to MHC I and functions to load and optimize the peptide binding to MHC I. There are clearly potential differences between these systems particularly as the relevant antigens have different solubility and thus peptides do not require "transfer proteins". We have added a sentence to the discussion describing more clearly the potential parallels between these systems. However, we feel that any further speculation at this stage is unwarranted.
The estimated maximum level of Kd that these techniques would find is 40 μM (Fig 2B). The authors should discuss the physiological relevance of an interaction weaker than that. We thank the reviewer for this point and how it relates also to the additional protein-protein interactions mentioned above involved in lipid transfer. The published affinities for these interactions range from 5 nM to 2 µM thus readily detectable using the techniques adopted in our study. An interaction weaker than 40 µM falls outside what would be considered a physiologically relevant interaction affinity in the absence of a highly-multivalent interaction, which we do not expect for SapB and CD1d. We have now added additional text to the discussion explicitly stating this point.
They should also discuss how proximity labelling (BioID, APEX) might find that. Although BioID and APEX are excellent techniques for detecting transient and weak protein-protein interactions in cells, these techniques are not yet available for the analysis of interactions within the endolysosome due to the low pH of these compartments that significantly reduce the efficacy of the biotinylation reaction. We would be very excited to implement these sorts of techniques when they become available.
State that SapB used might lack PTM required for binding in Fig 2, although this was tested in Fig  3. The reviewer is correct that the SapB used in Fig 2 is Fig. 3 was performed in the hope that this may increase the affinity to detectable levels, rather than be required for it. We have ensured this point is now clearer in the text. Fig 2C were  Saposin B has been reported to mediate the lipid loading to CD1d in lysosomal compartments. This report employed several sensitive analytical methods to dissect whether saposin B directly interacts with CD1d/beta2m to deliver the lipid antigens to CD1d. The authors have extensively explored different experimental conditions to analyze the potential interactions with equilibrium binding assay, chemical cross-linking and co-crystallization methods. None of these assays detected direct saposin B-CD1d/beta2m interaction. The authors concluded that saposin B does not directly bind CD1d lipid loading.

State whether conditions in
While the authors' studies are particularly thorough and extensive, the reviewer still has reservation to make a statement that this work could establish "comprehensively" that the role of saposin B in lipid loading does not involve direct interaction with CD1d. As the authors mentioned that the " " in vivo condition may be understandably different from the experimental conditions the authors have in vitro extensively tried. For example, as the authors pointed out, in lysosomes, there are multiple saposins present at possibly higher local concentrations, the potential "loading complex" may requires additional proteineous factors, etc. Therefore, while all results strongly suggested there might not be direct interaction, it is still hard to conclusively state there is none.
The additional factor is the pH in experimental condition may still affect the potential interaction even though it may not be critical for saposin and other lipid hydrolases.
Other comments: The amount of cross-linked saposin/CD1d may be lower than detection limit in Commassie blue staing. The authors may consider using western blotting to increase the detection sensitivity.
In equilibrium binding assay with a-GalCer, the lipid exchange/editing can be imagined as unloading of "endogenous" lipids in CD1d (co-purified from bacteria or mammalian cells) and replacing with a-GalCer. It is unknown if both saposin and CD1d loaded with the same a-GalCer lipid, whether this editing and saposin/CD1d interaction would occur. As antigen editing would be stronger affinity antigen replacing the lower affinity antigen. It is strongly suggested that the same experiment in Fig. 2C to be repeated with purified CD1d (no a-GalCer) with a-GalCer-loaded saposin B.

Is the work clearly and accurately presented and does it cite the current literature?
Yes Is the work clearly and accurately presented and does it cite the current literature?

Is the study design appropriate and is the work technically sound? Yes
Are sufficient details of methods and analysis provided to allow replication by others? Yes

If applicable, is the statistical analysis and its interpretation appropriate? Yes
Are all the source data underlying the results available to ensure full reproducibility? Yes

Are the conclusions drawn adequately supported by the results? Partly
No competing interests were disclosed.

Competing Interests:
Reviewer Expertise: Innate immunology I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.
Author Response 14 Oct 2019 , University of Cambridge, Cambridge, UK

Maria Shamin
Thank you for your thoughtful comments and constructive feedback. In-line responses to your queries are included below.
While the authors' studies are particularly thorough and extensive, the reviewer still has reservation to make a statement that this work could establish "comprehensively" that the role of saposin B in lipid loading does not involve direct interaction with CD1d. As the authors mentioned that the "in vivo" condition may be understandably different from the in vitro experimental conditions the authors have extensively tried. For example, as the authors pointed out, in lysosomes, there are multiple saposins present at possibly higher local concentrations, the potential "loading complex" may requires additional proteineous factors, etc. Therefore, while all results strongly suggested there might not be direct interaction, it is still hard to conclusively state there is none. We acknowledge that the lack of a detectable interaction does not necessarily mean that there is no interaction. We have therefore removed the word "comprehensively" from the abstract so that the conclusion now reads: "This work strongly indicates that the role of SapB in lipid loading does not involve direct binding to CD1d." The additional factor is the pH in experimental condition may still affect the potential interaction even though it may not be critical for saposin and other lipid hydrolases. The equilibrium binding assays and cross-linking experiments were performed at pH 7. As described in the text, this is a pH previously demonstrated to be able to facilitate SapB loading of lipids onto CD1d. Although this may not represent the most optimal pH for the interaction it was a necessary compromise that supported both the technical requirements of our assays and a pH at lipids onto CD1d. Although this may not represent the most optimal pH for the interaction it was a necessary compromise that supported both the technical requirements of our assays and a pH at which an interaction should have been observable.
The amount of cross-linked saposin/CD1d may be lower than detection limit in Commassie blue staining. The authors may consider using western blotting to increase the detection sensitivity. Although we agree that Western blot would have the capacity to visualize weaker bands following cross-linking, there is one primary reason we have not used this approach here. If the interaction between SapB and CD1d is so weak that, even in the presence of high concentrations of purified proteins and crosslinker, the proportion forming a complex must be detected by Western blot then to us it seems likely this interaction is too weak for further structural analysis and unlikely to be of physiological relevance (in agreement with our equilibrium binding analysis).
In equilibrium binding assay with a-GalCer, the lipid exchange/editing can be imagined as unloading of "endogenous" lipids in CD1d (co-purified from bacteria or mammalian cells) and replacing with a-GalCer. It is unknown if both saposin and CD1d loaded with the same a-GalCer lipid, whether this editing and saposin/CD1d interaction would occur. As antigen editing would be stronger affinity antigen replacing the lower affinity antigen. It is strongly suggested that the same experiment in Fig. 2C to be repeated with purified CD1d (no a-GalCer) with a-GalCer-loaded saposin B. We thank the reviewer for this suggestion and have now performed this additional experiment. We have repeated the equilibrium binding assay with SapB pre-loaded with α-GalCer, and CD1d loaded with the endogenous lipids with which it co-purifies following expression in a human cell line. In agreement with our other results, this alternate lipid loading of the protein components did not result in depletion of SapB from the supernatant even in the presence of 40 µM CD1d-β2m. As this additional experiment addresses an important point regarding relevant lipid loading (mentioned also by the other reviewers), we have updated the manuscript to include this new equilibrium binding assay in Fig.2, panel D.
No competing interests were disclosed. Competing Interests: