PDZD8-mediated lipid transfer at contacts between the ER and late endosomes/lysosomes is required for neurite outgrowth

Membrane contact sites (MCSs) between the endoplasmic reticulum (ER) and late endosomes/lysosomes (LE/lys) are emerging as critical hubs for diverse cellular events, and changes in their extents are linked to severe neurological diseases. While recent studies show that the synaptotagmin-like mitochondrial-lipid-binding (SMP) domain-containing protein PDZD8 may mediate the formation of ER–LE/lys MCSs, the cellular functions of PDZD8 remain largely elusive. Here, we attempt to investigate the lipid transfer activities of PDZD8 and the extent to which its cellular functions depend on its lipid transfer activities. In accordance with recent studies, we demonstrate that PDZD8 is a protrudin (ZFYVE27)-interacting protein and that PDZD8 acts as a tether at ER–LE/lys MCSs. Furthermore, we discover that the SMP domain of PDZD8 binds glycerophospholipids and ceramides both in vivo and in vitro, and that the SMP domain can transport lipids between membranes in vitro. Functionally, PDZD8 is required for LE/lys positioning and neurite outgrowth, which is dependent on the lipid transfer activity of the SMP


INTRODUCTION
Endoplasmic reticulum (ER) and endosomes are two main organelles in the secretory and endocytic pathways. Besides vesicular transport, ER and endosomes can be linked via membrane contact sites (MCSs), microdomains at which the membranes of the ER and endosomes are closely apposed (with a gap of <30 nm) and tethered, and the extent of such contacts increases as endosomes mature (Friedman et al., 2013;Hönscher and Ungermann, 2014;King et al., 2020;Raiborg et al., 2015b;Cohen et al., 2018;Valm et al., 2017). The majority of early endosomes and all late endosomes and lysosomes (hereafter referred collectively to as LE/lys) are stably associated with the tubular ER network, and these contact sites represent long-lasting, stable tethering along the vast tubular ER membrane (Zajac et al., 2013). Thus, it is not surprising that ER-endosome MCSs play essential roles in cell physiology (Phillips and Voeltz, 2016;Wu et al., 2018).
To date, several cellular functions of ER-endosome MCSs have been unveiled. One such function is to enable dephosphorylation of endosomal epidermal growth factor receptor (EGFR) through ER-resident phosphatase PTP1B (Eden et al., 2010), while other ER-endosome MCSs can regulate endosome dynamics, including endosome trafficking along microtubules (MTs) and endosome division (Hoyer et al., 2018;Rowland et al., 2014;Zajac et al., 2013). Another fundamental function of these contact sites is to transfer lipids and ions between the opposing membrane of ER and endosomes (Phillips and Voeltz, 2016;van der Kant and Neefjes, 2014;Wu et al., 2018).
Although vesicular transport is considered as a major pathway for the transfer of lipids between bilayer membranes, this route can be bypassed by lipid transfer proteins (LTPs) that are often localized to MCSs (Scorrano et al., 2019;Wong et al., 2017). LTPs are able to extract lipids from a donor membrane, shield them from the aqueous cytosolic environment within their hydrophobic pockets, and subsequently deliver them to a target membrane. Besides transferring lipids, these LTPs always serve as molecular tethers that physically bridge two opposing organelles and attach them to each other to ensure high efficiency of lipid transfer (Gatta and Levine, 2017;Saheki and De Camilli, 2017). Multiple proteins are proposed as potential tethers at ER-endosome MCSs in metazoans, including ER-resident vesicle-associated membrane proteins (VAPs)  and the ER transmembrane (TM) protein protrudin (ZFYVE27) (Raiborg et al., 2015a;Shirane and Nakayama, 2006), Star-related lipid transfer domain containing 3 (STARD3) (Alpy et al., 2013), and oxysterol-binding proteinrelated protein 1L (ORP1L; also known as OSBPL1A) (van der Kant et al., 2013) as well as vacuolar protein sorting-associated protein 13C (VPS13C) (Kumar et al., 2018). One important characteristic of ER-LE/lys MSCs is that such contact sites directly regulate LE/lys positioning. Protrudin plays critical roles in both formation of ER-LE/lys MCSs and LE/lys positioning (Raiborg et al., 2015a). The ER-resident protrudin directly binds Rab7a and phosphatidylinositol 3-phosphate (PI3P) to facilitate formation of transient ER-LE/lys MCSs. At such MCSs, protrudin delivers kinesin-1 to FYVE and coiled-coil domain containing 1 (FYCO1) on the LE/lys membrane, thus promoting anterograde transport of LE/lys. Protrudin-mediated ER-LE/lys MCSs are essential for neurite outgrowth through promoting LE/ lys anterograde transport and subsequent fusion with plasma membrane (Raiborg et al., 2015a).
While recent studies have suggested that PDZD8 mediates the formation of the ER-LE/lys MCSs (Elbaz-Alon et al., 2020;Guillén-Samander et al., 2019;Shirane et al., 2020), the cellular functions of PDZD8 and the extent to which its cellular functions depend on lipid transfer activities are largely elusive. Consistent with recent studies, we demonstrated that PDZD8 interacted with protrudin, and that PDZD8 acted as a tether at ER-LE/lys MCSs (Elbaz-Alon et al., 2020; Guillén-Samander et al., 2019; Hirabayashi et al., 2017;Shirane et al., 2020). Furthermore, we discovered that the SMP domain of PDZD8 binds glycerophospholipids and ceramides both in vivo and in vitro, and can transport lipids between membranes in vitro. Functionally, PDZD8 is required for LE/lys positioning and neurite growth in a lipid transfer activity-dependent manner.

PDZD8 interacts with the TM domain of protrudin
To identify unknown proteins that may interact with protrudin, we performed GFP-Trap assays in COS7 cells transiently expressing GFP-protrudin followed by mass spectrometry (MS) analysis. After removal of those proteins co-immunoprecipitated (co-IPed) by GFP alone, we found multiple known protrudin-interacting proteins including VAPA, VAPB and Rab7a in the coIP-MS results (Fig. S1A). A SMP domain-containing protein named PDZ domain-containing protein 8 (PDZD8) strongly interested us owing to its potential lipid transfer activities (Fig. S1A). Therefore, we sought to explore the cellular localizations and functions of PDZD8.
To confirm the protrudin-PDZD8 interaction, we performed GFP-Trap assays using top) or PDZD8-GFP as bait (Fig. 1A,bottom). Consistent with recent studies (Elbaz-Alon et al., 2020;Shirane et al., 2020), our GFP-trap assays confirmed the protrudin-PDZD8 interactions in both conditions. We further dissected the protrudin protein, and found that the TM domain (residues 1-208) of protrudin was responsible for interacting with PDZD8 (Fig. 1B). Note that PDZD8 did not interact with FYCO1 (Fig. S1B), a protrudin-interacting protein (Raiborg et al., 2015a). We could not produce purified PDZD8 from either human or other organisms in sufficient quantities for biochemical analysis. Thus, the direct interaction between PDZD8 and TM domain of protrudin could not be examined in this study. Since protrudin is enriched at ER-LE/lys MSCs (Raiborg et al., 2015a), we further examined whether PDZD8 colocalized with protrudin at these MCSs by live-cell confocal microscopy. We showed that PDZD8-Halo indeed colocalized with GFP-protrudin (Pearson's correlation coefficient=0.83) (Fig. 1C). The colocalization could be observed not only along ER tubules (Fig. 1C, box 1) but also on LE/lys, likely at the ER-LE/lys MSCs (Fig. 1C, box 2). However, we noted that small interfering RNA (siRNA)-mediated suppression of PDZD8 did not inhibit the enrichment of protrudin on LE/lys, indicating that the localization of protrudin to such MCSs is independent of PDZD8 (Fig. S1C-E).

PDZD8 localizes to ER-LE/lys MCSs
To explore the role of PDZD8 at MCSs, we started with examination of its cellular localization in animal cells. To avoid the artifacts of cytomegalovirus promoter (CMV)-driven high expression in imaging, we generated a stable single clone COS7 cell line (hereafter referred to as sPDZD8-GFP) that was infected with a lentiviral vector for PDZD8-GFP expression driven by the weak promoter EF1α, which showed a lower and more homogeneous expression level of PDZD8-GFP compared to CMV promoterdriven transient expression, as revealed by flow cytometry analysis (Fig. S1F). We then investigated the cellular localization of PDZD8 in the sPDZD8-GFP line transiently expressing Lamp1-Halo as a LE/lys marker, ER-tagRFP as an ER marker, and the mitochondrial marker mitoBFP by four-color live-cell confocal microscopy. PDZD8 localized to ER membrane (Fig. 1D), which agrees with the results of a previous study (Hirabayashi et al., 2017). Remarkably, a fraction of PDZD8-GFP showed a ring pattern decorating the entire LE/lys membrane (Fig. 1D,E), and targeting of PDZD8 to LE/lys was accompanied by ER recruitment, as revealed by colocalization among ER-tagRFP, PDZD8-GFP and Lamp1-Halo and corresponding linescan analysis (Fig. 1E, row 1), indicating that PDZD8 was enriched at ER-LE/lys MCSs. PDZD8 was not specifically accumulated at ER-mitochondria MCSs, as shown by linescan analysis along ER tubules (Fig. 1E, row 2). Consistent with results from a recent study (Elbaz-Alon et al., 2020), we noted that PDZD8-GFP-decorated LE/lys appear to be in close proximity to mitochondria (Fig. 1E, rows 1 and 2).
To further validate the localization of PDZD8 at ER-LE/lys MCSs, we tracked the dynamics of PDZD8 relative to ER-LE/ lys MCSs in the sPDZD8-GFP stable line transiently labeled with LE/lys and ER markers by live-cell confocal microscopy (Fig. 1F). PDZD8 was gradually recruited to ER-LE/lys MCSs as such contacts expanded and the level of PDZD8 was proportional to the extent of ER-LE/lys tethering ( Fig. 1F; Movie 1), suggesting that PDZD8 is a tether between ER and LE/lys at the MCSs.
We further examined the localization of PDZD8 relative to LE/lys, ER and mitochondria by Lattice-SIM super-resolution microscopy. Compared to non-PDZD8-decorated LE/lys (Fig. 1G, white arrow), PDZD8-decorated LE/lys were completely wrapped around by ER tubules (Fig. 1G, yellow arrow).
We also tested whether PDZD8 was accumulated at ER-plasma membrane (PM) MCSs by live-cell confocal microscopy in sPDZD8-GFP cells (Fig. S2A,B) or HeLa cells transiently transfected with PDZD8-GFP (Fig. S2C,D). PDZD8 was not specifically colocalized with GFP-Mapper puncta, a known marker of ER-PM MCSs (Chang et al., 2013) in both cell lines examined, indicating that PDZD8 does not have a role in ER-PM MCSs.
The TM domain of PDZD8 anchors the ER while the C1-CC region targets LE/lys The function of PDZD8 in tethering the ER to LE/lys requires the presence of binding sites responsible for recognizing these two membranes. Based on bioinformatics analysis, the domain organization of PDZD8 consists of an N-terminal TM domain, a SMP domain, a PDZ domain and a C1 domain, followed by a coiled-coil (CC) domain at its C terminus ( Fig. 2A; Fig. S3A) (Fidler et al., 2016). In addition, the PDZ domain in human PDZD8 is inserted into the C2 domain, and therefore the C2 domain sequence in PDZD8 is split with ∼65 amino acids lying between the SMP and PDZ domains (residues 320-372), then the rest (residues 695-800) being before the C1 domain (Guillén-Samander et al., 2019;Wong et al., 2017). PDZD8 has been suggested to be an ERresident protein with its N-terminal TM domain anchoring into ER membrane (Hirabayashi et al., 2017). Consistent with this finding, PDZD8 with a TM deletion (PDZD8-ΔTM) was cytosolic and failed to target the ER membrane, whereas a fraction of PDZD8-ΔTM was able to recognize LE/lys, as revealed by decoration of PDZD8-ΔTM on the entire LE/lys surface (Fig. 2B). Truncated PDZD8 containing the TM and SMP domains only (PDZD8-TM-SMP) displayed the same distribution as the full-length protein on ER membrane but without LE/lys localization (Fig. 2C). In addition, the SMP domain alone was cytosolic and failed to recognize either the ER or LE/lys membranes (Fig. S3B). Taken together, our results indicate that the N-terminal TM domain of PDZD8 is responsible for ER targeting, which is in agreement with recent studies (Elbaz-Alon et al., 2020;Guillén-Samander et al., 2019;Shirane et al., 2020).
Next, we dissected PDZD8 to identify the domains that were required for recognizing LE/lys by live-cell confocal microscopy in COS7 cells transiently transfected truncated PDZD8 constructs (  domain (PDZD8-ΔTM-ΔC1-ΔCC) (Fig. 2G) nor PDZD8 without the C1 and CC domain (PDZD8-ΔC1-ΔCC) (Fig. 2H) was able to target LE/lys. These findings indicate that the C1-CC domain is necessary and sufficient for PDZD8 to target LE/lys membranes (Fig. 2I). Besides COS7 cells, PDZD8-C1-CC was able to recognize LE/lys membranes in multiple animal cell lines, including HeLa, U2OS and Neuro-2a cell lines (Fig. S4), suggesting LE/lys recognition by PDZD8-C1-CC is conserved among cell types.
The C1-CC region of PDZD8 targets LE/lys through directly binding Rab7 We next sought to understand the molecular mechanisms underlying the recognition of LE/lys membrane by the C1-CC domain. Neither the C1 domain nor the CC domain is likely to directly bind to lipid bilayer of LE/lys membranes (Hurley and Misra, 2000;Truebestein and Leonard, 2016). Thus, we hypothesized that C1-CC domains may bind to proteins on LE/lys membrane to mediate LE/lys recognition. Using GFP-Trap followed by MS analysis, we identified PDZD8-interacting proteins in sPDZD8-GFP COS7 cells. We focused on proteins belonging to the following two categories. First, it was exclusively present in the list after removal of proteins coIPed by GFP alone; second, it was a LE/ lys membrane protein. From the list of protein candidates selected based on the rules described above (Fig. S5A), we performed a brief screening by co-expressing identified protein candidates with PDZD8-C1-CC and examining whether the protein candidates could enhance the LE/lys recruitment of PDZD8-C1-CC by livecell confocal microscopy in COS7 cells. Among seven candidates examined in the screening, only Rab7a expression substantially boosts the C1-CC recruitment to LE/lys membrane (Fig. 3A,E). We further tested the roles of two Rab7a mutants on C1-CC recruitment. Co-expression of Rab7a(Q67L), a constitutively active mutant, greatly enhances C1-CC recruitment to LE/lys (Fig. 3B,E), whereas Rab7a(T22N), a dominant-negative mutant, blocked C1-CC recruitment to LE/lys (Fig. 3C,E). In addition, PDZD8-C1-CC was not substantially recruited to Rab5-positive early endosomes even when Rab5a was co-expressed ( Fig. 3D,E), indicating that the recruitment of C1-CC to LE/lys was Rab7 specific, which is consistent with results from recent studies (Elbaz-Alon et al., 2020;Guillén-Samander et al., 2019;Shirane et al., 2020).
Since PDZD8 interacted and colocalized with protrudin, there was a possibility that PDZD8-C1-CC recognized LE/lys membrane through protrudin. Suppression of protrudin did not substantially hamper PDZD8-C1-CC recruitment to LE/lys compared with the scrambled siRNA-treated cells (Fig. S5B,D,F,H), whereas suppression of Rab7 via Rab7a siRNA-mediated knockdown almost completely blocked PDZD8-C1-CC recruitment ( Fig. S5B,C,F,I), ruling out the possibility that PDZD8-C1-CC recognizes LE/lys through protrudin. Suppression of VPS37B, an ESCRT-I component identified from PDZD8-GFP coIP-MS analysis did not substantially inhibit PDZD8-C1-CC recruitment (Fig. S5E,F). In addition, because protrudin is known to recognize LE/lys membrane in a bipartite manner through binding to Rab7 and PI3P (Raiborg et al., 2015a), we investigated whether PI3P was required for the recruitment of the PDZD8-C1-CC region to LE/lys membranes. Inhibition PI3P synthesis by wortmannin or VPS34IN1 treatment, failed to halt PDZD8-C1-CC recruitment to LE/lys in COS7 cells transiently transfected with PDZD8-C1-CC and Flag-Rab7, although LE/lys were enlarged and clustered near nucleus after these treatments (Fig. S5G).
We further investigated whether PDZD8-C1-CC directly binds to Rab7 in vitro. Glutathione S-transferase (GST) pulldown assays with purified proteins showed that GST-tagged C1-CC but not GST alone, interacted directly with GTP-(but not GDP-) loaded Rab7, and that there was no interactions with Rab5-GTP ( Fig. 3F), consistent with results from a recent study (Guillén-Samander et al., 2019). Thus, PDZD8 is a new Rab7 effector that interacts with this protein through its C1-CC region. Taken together, these findings demonstrate that PDZD8-C1-CC recognizes LE/lys membrane through Rab7, in line with the recent studies (Elbaz-Alon et al., 2020;Guillén-Samander et al., 2019;Shirane et al., 2020).

Rab7 activities control PDZD8-mediated ER-LE/lys MCSs
The presence of Rab7 in PDZD8 mediated ER-LE/lys MCSs raised a question of whether the activity of Rab7 regulated PDZD8-mediated ER-LE/lys tethering. Coexpressing Rab7 or Rab7(Q67L) in the sPDZD8-GFP line resulted in specific and dramatic recruitment of full-length PDZD8 to entire surface membrane of LE/lys, indicating that LE/lys were entirely wrapped by ER tubules, whereas PDZD8 only decorated a few LE/lys membrane in control sPDZD8-GFP cells transiently transfected with the LE/lys marker Lamp1 ( Fig. 4A-C). Conversely, coexpressing Rab7(T22N) completely blocked the recruitment of full-length PDZD8 to LE/lys membrane (Fig. 4D), indicating that Rab7 activity controlled PDZD8-mediated ER-LE/lys MCSs. Notably, upon overexpression of either Rab7 wild-type (WT) or Rab7(Q67L), PDZD8 was strongly recruited to ER-LE/lys MCSs so that ER-localized, non-MCS PDZD8 was hardly detectable, while the amount of PDZD8-decorated LE/lys was much less; ER-resident, non-MCS PDZD8 signals were readily detectable in neighboring non-Rab7-transfected cells (Fig. 4B,C). In contrast, overexpression of Rab7(T22N) completely blocked recruitment of PDZD8 to LE/lys and ER-resident, non-MCS PDZD8 was easily detectable (Fig. 4D), suggesting that Rab7 recruits PDZD8 from ER pools to LE/lys membranes to mediate ER-LE/lys tethering.
To further validate the tethering functions of PDZD8, we examined the level of the ER-LE/lys interactions by live-cell confocal microscopy in COS7 cells. Compared to cells only Immunoblotting with antibodies against Flag and GFP was performed. Bottom, a GFP-Trap assay was performed in COS7 cells expressing PDZD8-GFP. Immunoblotting with antibodies against protrudin and GFP was performed. GFP empty vector was used as control. (B) GFP-Trap assays were performed in COS7 cells expressing PDZD8-GFP and Flag-tagged protrudin fragments. Immunoblotting with antibodies against Flag and GFP was performed. GFP empty vector was used as control.  transfected with the LE/lys marker Lamp1-mCherry and the ER marker GFP-Sec61β, co-overexpression of Rab7 and PDZD8 substantially enhanced the extent of ER-LE/lys interactions, as revealed by the colocalized pixels between these two organelles, whereas overexpression of Rab7 alone could only slightly increase ER-LE/lys interactions ( Fig. 4E-H), indicating that both PDZD8 and Rab7 are required and sufficient for ER-LE/lys tethering, which confirms the key findings in recent results (Elbaz-Alon et al., 2020;Guillén-Samander et al., 2019;Shirane et al., 2020).
PDZD8-Rab7-mediated ER-LE/lys tethering was further examined by four-color live-cell super-resolution lattice-SIM microscopy in the sPDZD8-GFP stable line transiently expressing Flag-Rab7 along with ER, mitochondria and LE/lys markers. Co-expression of Rab7 and full-length PDZD8 caused an extensive ER-LE/lys tethering, as revealed by ER tubules completely ( Fig. 5A-C, box 2 images) or partially ( Fig. 5A-C, boxes 3-7) wrapping the LE/lys with specific enrichments of PDZD8 at such contacts ( Fig. 5A-C).
The SMP domain of PDZD8 binds glycerophospholipids and ceramide both in vitro and in vivo, and mediates lipids transport in vitro The localization of PDZD8 at ER-LE/lys MCSs is consistent with a potential role of SMP domain-containing proteins in lipid transport Yellow arrows indicate PDZD8-decorated LE/lys; white arrows indicate LE/lys without PDZD8 decoration. Quantification of frequency of PDZD8 decorated LE/lys was shown on the right of magnified images. N represents cell number in the analysis. (E-G) Representative confocal images of live COS7 cells expressing ER-Halo (green) and Lamp1-mCherry (E, magenta), mCherry-Rab7 (F, magenta) or co-expression of mCherry-Rab7 and PDZD8-GFP (G, magenta). Left top panel, merged image of whole cell; right top panel, colocalized pixels between ER and LE/lys (white); bottom panels, magnified images from boxed region in whole-cell image. (H) Quantification of number of LE/lys wrapped around by ER at cell periphery. Lamp1 (n=16); Rab7 (n=23); Rab7 and PDZD8 (n=15). Pvalues are as shown (two-tailed unpaired Student's t-test). Mean±s.d. Scale bars: 10 μm (whole-cell images in A-G); 2 μm (enlarged images in A-G).
between adjacent organelles. Therefore, we next tested the possibility that PDZD8 might directly bind and transport lipids. To investigate whether PDZD8 might solubilize lipids, a prerequisite for a lipid transport function, we examined lipid species bound by full-length PDZD8 in sPDZD8-GFP COS7 cells infected with a doxycycline-inducible lentiviral vector encoding Flag-Rab7 (Fig. S6A). Our rationale of using inducible expression of Rab7 in sPDZD8-GFP cells (iRab7-sPDZD8-GFP) for this assay was that PDZD8 presumably binds lipids at ER-LE/lys MCSs and inducible expression of Rab7 would enable a transient increase in PDZD8-mediated contacts. Lipid species associating with PDZD8-GFP were assessed by non-targeted lipidomics using liquid chromatography-tandem mass spectrometry (LC-MS/MS) with rigorous washes of the protein before lipid analysis. PDZD8-GFP mainly associated with glycerophospholipids [lysophosphatidylcholines and phosphatidylcholines (LPC/PC), phosphatidylserine (PS) and phosphatidic acid (PA)], and a few ceramides and sphingomyelin (SM) but no sterols were detected using GFP tag-only as a control (Fig. 6A). Since PDZD8 harbors two potential lipid-binding domains (SMP and C1), we next assessed lipids bound by the SMP domain or the C1 domain alone by LC-MS/MS. We found that both the species and abundance of lipids co-purified with SMP-GFP highly resembled those of lipids co-purified with full-length PDZD8-GFP (Fig. 6A,B). We could not obtain significant results from the C1 domain using LC-MS/MS comparing with GFP tag control (data not shown). Therefore, this line of evidence indicates that PDZD8 solubilized glycerophospholipids and ceramides through the SMP domain. In addition, we calculated the ratio of the specific lipids bound by PDZD8-GFP (Fig. S6B, left panel) or SMP-GFP (Fig. S6B, right panel) to the total amounts of the specific lipids extracted from COS7 cells. We found that both the PDZD8 and the SMP domain bound a very small amount of lipids [for example, only ∼5.8% of total ceramide (d18:1/22:0) content of cellular membranes was bound by PDZD8-GFP; and ∼4.5% of this lipid species bound by SMP-GFP] in vivo.
We further explored lipid-binding activities of PDZD8 by in vitro lipids-binding assays. So far, we cannot produce full-length proteins in sufficient quantities for biochemical analysis. Thus, we used purified PDZD8 fragments for this purpose. Consistent with our LC-MS/MS results, the SMP domain of PDZD8 co-migrated with nitrobenzoxadiazole (NBD)-labeled glycerophospholipids [PS, PC, phosphatidylethanolamine (PE) and PA] and ceramides but not cholesterols, as assessed by native gel electrophoreses (Fig. 6C). Among these lipids, the SMP domain preferentially bound to PC and PS (Fig. 6C,F). We made a space-filling SMP mutant of PDZD8 (SMP-mut), in which two residues with small hydrophobic sidechains were replaced with tryptophan residues, which have with large side-chains (V166W, L285W), according to an approach previously reported in an elegant study (Saheki et al., 2016) (Fig.  S6C). We found that the purified SMP-mut did not bind to these lipids except for a low binding affinity to PC (Fig. 6D,F). We also assessed the lipid-binding activities of the C1-CC region via in vitro lipid-binding assays. C1-CC barely bound NBD-labeled PC, PS, PE and ceramides, but bound NBD-PA at a low level (Fig. 6E,F), consistent with our LC-MS/MS results and indicating that the SMP domain is responsible for the lipid-binding activities of PDZD8.
Based on these observations, we next employed a fluorescence resonance energy transfer (FRET)-based assay to examine whether the SMP domain transported lipids between membranes according to the strategy developed by previous studies (Kumar et al., 2018;Saheki et al., 2016;Valverde et al., 2019;Yu et al., 2016). Briefly, donor liposomes containing DGS-NTA, NBD-PC and Rhodamine-PE (Rh-PE) were mixed with non-fluorescent acceptor liposomes containing phosphatidylinositol 4,5-bisphosphate [PI(4,5)P 2 ], and then purified 'tethered SMP domain' proteins were added. This tethered SMP domain is tethered to donor liposomes via a N-terminal hexahistidine (His) tag and is also tethered to acceptor liposomes through a C-terminal pleckstrin homology (PH) domain, which interacts with PI(4,5)P 2 in the acceptor liposomes. Thus, the linker tethered together donor and acceptor liposomes to mimic the architecture of MCSs, where the PDZD8-SMP likely functions (Fig. 6G).
To confirm that the tethered SMP domain was able to bind donor and acceptor liposomes, we conducted liposome pulldown assays. We found that the tethered SMP domain, but not GST, could efficiently bind both the donor and acceptor liposomes (Fig. 6H). In addition, we further examined the liposome tethering efficiency by measuring the size distribution of liposomes, in the presence or absence of the tethered SMP domain, via dynamic light scattering (DLS). The DLS assays indicated that the addition of tethered SMP (but not the no-protein control or the addition of GST) resulted in a substantial increase in size of liposomes (Fig. S6D), indicating that the tethered SMP efficiently bridges donor and acceptor liposomes. We also observed that the addition of tethered SMP resulted in two peaks, with a smaller one at ∼400 nm (this might suggest about four liposomes cross-linked) and a larger one at ∼5000-6000 nm (tens of liposomes cross-linked) (Fig. S6D). In addition, the DLS assays indicate that the liposomes in absence of tethered SMP are well dispersed, uniform in size (∼100 nm) and stable in solution, suggesting no spontaneous fusion or hemifusion of liposomes without tethered SMP proteins.
Initially, FRET between NBD and Rh in the donor liposomes quenched NBD fluorescence. Transfer of NBD-PC, Rh-PE or both to acceptor liposomes would result in their dilution, decreased FRET and thus an increase in NBD fluorescence. We observed an increase in NBD fluorescence upon addition of the wild-type (WT) tethered SMP (but not the no-protein control; Fig. 6I). We investigated the lipid transfer activities of the space-filling mutant SMP-mut using in vitro lipid transfer assays. Consistent with our in vitro lipid-binding assays, in which the SMP-mut still binds to PC in a low level (Fig. 6D), SMP-mut caused a substantial decrease in the transfer of NBD-PC or Rh-PE comparted to the non-mutant form, but did not completely lose the transfer activities, as revealed by a very small increase in NBD fluorescence in presence of donor liposomes containing NBD-PC, Rh-PE and non-fluorescent acceptor liposomes (Fig. 6I, purple curve). In contrast to a robust increase in NBD fluorescence upon the addition of the WT tethered SMP, the addition of SMP-mut failed to increase NBD fluorescence in presence of donor liposomes containing NBD-ceramide, Rh-PE and acceptor liposomes (Fig. 6J, purple curve), which agrees with our in vitro lipid-binding assay, in which the SMP-mut did not bind either ceramide or PE. This result suggests that PDZD8-SMP is able to transfer PC, PE and ceramides in vitro. However, it should be noted that the FRET-based assays do not discriminate whether the PDZD8-SMP transports PE only, PC only, ceramide only, or the three lipids. We therefore examined whether PDZD8-SMP transported PE using in vitro lipid transfer assays with donor liposomes containing NBD-cholesterol and Rh-PE, and acceptor liposomes. Owing to the lack of suitable commercial Rhodaminelabeled lipids (such as Rh-cholesterol, Rh-PC or Rh-ceramide), in this study we only investigated the PE transfer activity of the SMP domain in vitro. We observed an increase in NBD fluorescence upon addition of the tethered SMP-WT (but not the no-protein control; Fig. 6K). Since our results indicate that the PDZD8-SMP does not bind NBD-cholesterol (Fig. 6C), the increase in NBD fluorescence indicated the transfer of Rh-PE from donor liposomes to acceptor liposomes mediated by PDZD8-SMP.
dithionite (Meers et al., 2000), which quenches NBD fluorescence only in the outer leaflet of the liposome bilayer but does not affect the inner leaflet, ruling out the possibility that the tethered SMPmediated fluorescence increase is due to fusion or hemifusion between donor and acceptor liposomes (Fig. S6E).
Notably, we observed that there was no substantial increase in NBD fluorescence in reactions with the tethered SMP and donor liposomes lacking acceptor liposomes (Fig. 6I-K), corresponding to a scenario in which PDZD8-SMP extracts fluorescent lipids from the donor liposomes, but cannot deliver them to acceptor liposomes. This result indicates that the SMP domain of PDZD8 extracted only a small amount of lipids from acceptor liposomes, and therefore the dilution of NBD-PC and Rh-PE was not strong enough to activate NBD fluorescence. Our results were in accordance with the SMP module of the previously characterized glycerophospholipid transporter extended synaptotagmin 1 (E-Syt1), in which the lipid-extracting activities of E-Syt1 was also very low in similar reactions with purified E-Syt1 and donor liposomes but without acceptor liposomes (Kumar et al., 2018;Saheki et al., 2016;Valverde et al., 2019;Yu et al., 2016). This line of evidence supports that PDZD8-SMP might act as a shuttle transporter that extracts and solubilizes lipids (one or two lipid molecules at once) from the membrane of the donating organelle, ferries them through the cytosol, and then deposits them in the membrane of the acceptor organelle Wong et al., 2017).
Moreover, since SMP domain-containing proteins such as E-Syts form dimers through SMP domain dimerization , we tested whether PDZD8 also formed dimers or oligomers. We first examined the oligomeric status of PDZD8 by GFP-Trap assays in sPDZD8-GFP cells transiently transfected with PDZD8-Myc-Flag. The GFP-trap assay showed that PDZD8-Myc-Flag was co-pelleted with PDZD8-GFP (Fig. S6F). In addition, PDZD8-TM-SMP sufficiently co-pelleted with fulllength PDZD8-Myc-Flag, whereas PDZD8 lacking the SMP domain could not bind PDZD8-Myc-Flag, indicating that PDZD8 formed dimers or oligomers through the SMP domain (Fig. S6F). Moreover, the level of PDZD8-Myc-Flag co-pelleted by TM-SMP was less than that of full-length PDZD8, suggesting that other regions may also contribute or stabilize the oligomerization of PDZD8 (Fig. S6F). Notably, TM-SMP tended to be dimerized even in denatured conditions, as revealed by two bands in western blots, one band with a molecular mass of ∼75 kDa, the other of ∼150 kDa, were present in both input and GFP-trapped samples (Fig. S6F, red asterisks). In addition, we found that PDZD8 lacking C1-CC (PDZD8-ΔC1-ΔCC) could interact with protrudin, and the C1-CC regions may contribute to PDZD8-protrudin interactions because the level of protrudin co-pelleted by PDZD8-ΔC1-ΔCC was less compared to level of protrudin co-pelleted by full-length PDZD8 (Fig. S6F).
Lipid transfer activities of PDZD8 are required for LE/lys positioning and neurite outgrowth We next investigated the role of PDZD8 in the LE/lys positioning since PDZD8 interacted with Rab7 and protrudin, both of which have been shown to play essential roles in facilitating LE/lys anterograde trafficking (Raiborg et al., 2015a). We used siRNAmediated suppression to explore the cellular functions of PDZD8. So far there is no effective commercial PDZD8 antibody for either western blots or immunofluorescence staining, and thus we used quantitative PCR (qPCR) to confirm the efficiency of PDZD8 suppression in COS7 or U2OS cells (Fig. S7A). We found that suppression of PDZD8 resulted in LE/lys clustering at perinuclear regions in comparison with scrambled siRNA-treated cells (Fig. 7A). Fig. S7B. Besides COS7 cells, PDZD8 suppression led to more LE/ lys clustering at perinuclear regions in multiple cell lines including U2OS (Fig. S7C,E) and HeLa (Fig. S7D,E) cells.

The definition of perinuclear regions of cells is demonstrated in
We further investigated whether the role of PDZD8 in LE/lys positioning was dependent on its lipid transfer activities. We made two lipid transfer-deficient mutants on the basis of full-length PDZD8a space-filling SMP mutant of PDZD8 (PDZD8-SMPmut), in which the SMP domain of PDZD8 was replaced by the SMP V166W L285W mutant examined in Fig. 6D, and PDZD8-ΔSMPfor this purpose (Fig. 7B, top). The LE/lys positioning defects was near-completely rescued by siRNA-resistant full-length PDZD8-GFP (Fig. 7B,C). Although these two lipid transferdeficient PDZD8 mutants were still able to mediate the formation of the ER-LE/lys MCSs, neither PDZD8-SMP-mut nor PDZD8-ΔSMP could rescue the LE/lys positioning defects (Fig. 7B,C), suggesting that lipid transfer activity of the SMP domain was essential for the positioning of LE/lys. In addition, we observed abnormally enlarged LE/lys upon expression of these two mutants in PDZD8-suppressed cells, suggesting that lipid transfer activity of PDZD8 is also required for other aspects of LE/lys including maturation, which agrees with a recent study showing that PDZD8 contributes to neuronal integrity (Shirane et al., 2020).
In addition, overexpression of protrudin promoted the anterograde transport of LE/lys, as revealed by the enrichment of Lamp1-Halo labeled LE/lys at cell periphery regions (Fig. S7F, top, yellow arrows). However, the defect of LE/lys positioning resulted from PDZD8 suppression was not substantially rescued by protrudin overexpression (Fig. S7F; Fig. 7C). This line of evidence indicates that PDZD8 is not functionally redundant to protrudin. Since these two proteins interacted with each other at MCSs, it is possible that PDZD8 is required in protrudin-mediated LE/lys anterograde transport because of the lipid transfer activities of PDZD8.
Since LE/lys anterograde transport promoted by protrudinmediated MCSs is essential in neurite outgrowth (Raiborg et al., 2015a;Shirane and Nakayama, 2006), we further investigated whether PDZD8 played a role in the process of neurite outgrowth using PC12 cells (Black and Greene, 1982). Neuronal growth factor (NGF) stimulated PC12 differentiation, as revealed by the formation of extrusions with the length over >30 μm visualized by immunofluorescence staining of tubulin (Fig. 7D,E). Suppression of PDZD8 using lentivirus-encoded small hairpin RNAs (shRNAs) targeting PDZD8 significantly inhibited NGF-induced extrusion formation during PC12 differentiation (Fig. 7F). Both the percentage of cells with extensions longer than 30 μm and the length of extrusions substantially decreased upon PDZD8 shRNA treatment compared with that of scrambled shRNA-infected cells (Fig. 7F,K,L). The inhibition effects induced by PDZD8 suppression were comparable to that of protrudin suppression (Fig. 7G,K,L), whereas double suppression of protrudin and PDZD8 did not result in cumulative effects (Fig. 7H,K,L), supporting a hypothesis that PDZD8 and protrudin might function in a same pathway in neurite extension. The shRNA-mediated suppression of PDZD8 and protrudin were confirmed by qPCR (Fig. S7G). The expression of shRNA-resistant PDZD8-GFP substantially rescued the defects of neurite length in PDZD8suppressed cells (Fig. 7I,K,L), but expression of the lipid transferdeficient mutant PDZD8-SMP-mut did not substantially rescue this defect (Fig. 7J,K,L). This result indicates that lipid transfer activity of PDZD8 has an essential role in this process. PDZD8 is required for LE/lys positioning and neurite growth in a lipid transfer activity-dependent manner.
It has been a mystery whether there are homologs of yeast ERMES, a complex that tethers the ER to mitochondria, in metazoans (Kornmann et al., 2009). A previous work proposed that PDZD8 was a functional ortholog of maintenance of mitochondrial morphology protein 1 (Mmm1) of yeast ERMES and that it acts to mediate the formation of the ER-mitochondria MCSs in mammalian cells based on domain-swapping assays along with complementation assays (Hirabayashi et al., 2017). However, domain-swapping and complementation experiments were insufficient to demonstrate that these two proteins (or even the domains) were functional orthologs, and PDZD8 and Mmm1 were more likely to be paralogs based on phylogenetic analyses (Wideman et al., 2018).
Our results, in accordance with recent studies from other groups (Elbaz-Alon et al., 2020;Guillén-Samander et al., 2019;Shirane et al., 2020), pointed to PDZD8 being a key player at ER-LE/lys MCSs. However, ER, LE/lys and mitochondria are physically associated, rendering it difficult to rule out the possible localizations and/or roles of PDZD8 in mitochondrially associated MCSs (Elbaz-Alon et al., 2020). In addition, a recent work by Kumar and Leonzino et al. revealed that VPS13A acts as a tether and LTP at ER-mitochondria MCSs (Kumar et al., 2018). VPS13A showed a clear and robust localization at ER-mitochondria MCSs in animal cells. VPS13A targets to ER through binding to VAP via its FFAT motif at the N terminus, while it recognizes mitochondria through C-terminal regions. Interestingly, a recent study demonstrated that VPS13A localizes at the interface between two types of mitochondrially associated MCSs -ER-mitochondria and mitochondria-endosome MCSs. VPS13A interacts with Rab7 on LE/lys and is required for efficient lysosomal degradation, suggesting an important role of VPS13A in crosstalk of these two different membrane contacts. Given that a complex homologous to ERMES was unlikely to be present in animal cells, VPS13A, but not PDZD8, might carry out the same roles as yeast ERMES and perform lipid transfer functions at such MCSs.
A growing number of studies have shown that many intracellular LTPs mediate lipid transport between opposing membranes at MCSs through two modes of actionshuttling or bridging. Shuttle transporters typically extract one or two lipid molecules from the membrane of the donor organelle, solubilize it during transport through the cytosol and deposit it in the acceptor organelle membrane, while bridge transporters feature an extended channel, most likely lined with hydrophobic residues that bind tens of lipids at once (Berrak et al., 2020;Lees and Reinisch, 2020;Louise et al., 2019).
It has been reported that a complex of multiple SMP domains shuttle glycerophospholipids across MCSs. For the SMP domaincontaining protein E-Syt2, which transfers a range of glycerophospholipids between the ER and the plasma membrane, its shuttle mode of action was strongly supported because the dimer formed by its SMP domains was too short to bridge the gap between the membranes (Schauder et al., 2014). Another example is the SMP-containing transmembrane protein 24 (TMEM24; also known as C2CD2L), which is a shuttle transporter selective for phosphatidylinositols over other phospholipids. The crystal structure of TMEM24 shows that the lipid cavities of the two monomers do not join together at the head-to-head dimer interface, and therefore, lipids cannot flow along the length of the dimer (Lees et al., 2017). Similar to E-Syt2 and TMEM24, our results showed that PDZD8 forms dimers or oligomers through the SMP domain (Fig. S6F). More importantly, we showed that the SMP domain efficiently mediated glycerophospholipid (NBD-PC) and ceramide transport in vitro but its lipid-extracting activities in reactions with donor liposomes and without acceptor liposomes were very low (Fig. 6I,J). This result suggests that the SMP domain of PDZD8 extracts a very small amount of lipids from the membrane of donor liposomes, similar to what occurs with E-Syt1 (Kumar et al., 2018;Saheki et al., 2016;Valverde et al., 2019;Yu et al., 2016). This line of evidence supports that the SMP domain of PDZD8 adopts the shuttle model. In contrast, the chorein-N family of lipid transporters, including the Vps13 proteins and the autophagy-related protein Atg2, which feature an extended channel, bind tens of lipids at once and show much higher lipid extracting activities than E-Syt1 were reported to be bridge lipid transporters (Kumar et al., 2018;Li et al., 2020;Valverde et al., 2019). Therefore, our results suggest that PDZD8 might be a glycerophospholipid and ceramide shuttle transporter at ER-LE/lys MCSs. The crystal structure of the PDZD8-SMP domain is needed to confirm this hypothesis.
Our results show that the SMP domain of PDZD8 is capable of transporting PE in vitro lipid via transfer assays (Fig. 6K) in which a NBD-Rhodamine fluorescence donor-acceptor FRET pair was used. The unavailability of commercial Rhodamine-labeled lipids (such as Rh-cholesterol, Rh-PC or Rh-ceramide) has hampered our investigations of whether the PDZD8-SMP transports PC or ceramide. Although other red dyes (such as BODIPY TR)-labeled lipids are available, these red dye-labeled lipids were not suitable for the measurement (Redfern and Gericke, 2004). For instance, the excitation and emission of BODIPY-TR (589 nm/617 nm) are different from Rhodamine (558 nm/575 nm), suggesting that the BODIPY-TR may not be excited by the NBD (ex/em: 460 nm/ 534 nm) in FRET-based assays. Therefore, the lipid transfer activities of the PDZD8-SMP on PC or ceramide remain unclear and warrant further investigations.
Our results indicate that PDZD8 interacts and colocalizes with protrudin, which was previously shown to promote LE/lys anterograde transport (Raiborg et al., 2015a). We showed that these two proteins marked the same type of ER-LE/lys MCSs (Fig. 1C), and suppression of PDZD8 resulted in LE/lys clustering at perinuclear regions, which is similar to the phenotype found in protrudin-suppressed cells (Fig. 7A-C; Fig. S7B-D). Interestingly, protrudin is one of the ER-LE/lys MCSs tethers without lipid transport abilities, which raises a question of whether PDZD8 and protrudin might cooperate to regulate the LE/lys transport under certain physiological events, for example, in neurite outgrowth, in which a large number of LE/lys are required to transport to cell periphery. In this scenario, PDZD8 and protrudin have independent roles at the ER-LE/lys MCSs, with protrudin mediating efficient loading of LE/lys with kinesin-1 while PDZD8 transports lipids between ER and LE/lys at such MCSs, both of which are required for proper LE/lys positioning.

Cell culture and transient transfection
The African green monkey kidney fibroblast-like COS7 cell line (ATCC), human osteosarcoma U2OS cells (ATCC), human cervical cancer HeLa cells (ATCC), human embryonic kidney 293T (ATCC) and mouse neuroblastoma Neuro-2a cells (ATCC) were grown in DMEM (Invitrogen) supplemented with 10% fetal bovine serum (Gibco). The rat pheochromocytoma PC12 cell line were grown in RPMI1640 (Invitrogen) supplemented with 10% horse serum (Gibco) and 5% fetal bovine serum (Gibco). All of the cell lines used in this study were confirmed free of mycoplasma contamination.
For transfection, cells were seeded at 4×10 5 cells per well in a six-well dish ∼16 h before transfection. Plasmid transfections were performed in OPTI-MEM (Invitrogen) with 2 µl Lipofectamine 2000 per well for 6 h, followed by trypsinization and replating onto glass-bottom confocal dishes at ∼3.5×10 5 cells per well. Cells were imaged in live-cell medium (DMEM with 10% FBS and 20 mM Hepes with no penicillin or streptomycin) ∼16-24 h after transfection. For all transfection experiments in this study, the following amounts of DNA were used per 3.5 cm well (individually or combined for cotransfection): 500 ng for PDZD8-GFP, mito-BFP, GFP-Mapper, Lamp1-Halo and Lamp1-OFP; 1000 ng for ER-tagRFP and ER-mGFP; 300 ng for pTY-CR-hRab7A, pTY-CR-hRab7A(Q67L) and pTY-CR-hRab7A(T22N). For siRNA transfections, cells were plated on 3.5 cm dishes at 30-40% density, and 2 µl Lipofectamine RNAimax (Invitrogen) and 50 ng siRNA were used per well. At 48 h after transfection, a second round of transfection was performed with 50 ng siRNAs. Cells were analyzed 24 h after the second transfection for suppression.

Live cell super-resolution Lattice-SIM microscopy
Cells were grown on 35 mm glass-bottom confocal MatTek dishes, and medium was changed to OptiMEM prior to loading to Elyra Lattice SIM (Zeiss, Germany). Super-resolution lattice-SIM images were acquired on Elyra 7 with Lattice SIM equipped with a 63×/1.4 NA Plan-Apochromat oil immersion objective, 405, 488, 561 and 642 nm lasers and sCMOS cameras (PCO.Edge CLHS). Cells were imaged with the 405 nm laser for BFP, 488 nm for GFP, 561 nm for tagRFP and 642 nm for Janilia Fluo ® 646 HaloTag ® Ligand. Images were subsequently processed using the Zen software. The end images had the resolution of 120 nm in X-Y and 300 nm in Z.

Immunofluorescence staining
Cells were fixed with 4% paraformaldehyde (PFA, Sigma) in PBS for 10 min at room temperature. After washing with PBS three times, cells were permeabilized with 0.1% Triton X-100 in PBS for 15 min on ice. Cells were then washed three times with PBS, blocked with 1% FBS in PBS for 1 h, incubated with primary antibodies in diluted blocking buffer overnight, and washed with PBS three times. Secondary antibodies were applied for 1 h at room temperature. After washing with PBS three times, samples were mounted on Vectashield (H-1000; Vector Laboratories).

GFP-trap assay
GFP trap (GTA-100, ChromoTek) was used for detection of proten-protein interactions and the GFP-Trap assays were performed according to the manufacturer's protocol. 5% input was used in GFP traps unless otherwise indicated.

Mass spectrometry for identifications of GFP-protrudin-or PDZD8-GFP-interacting proteins
The bound proteins were extracted from IP beads using SDT lysis buffer (4% SDS, 100 mM DTT, 100 mM Tris-HCl pH 8.0). IP beads sample were boiled for 3 min and further ultrasonicated. Undissolved beads were removed by centrifugation at 16,000 g for 15 min. The supernatant, containing proteins, were collected. Protein digestion was performed with FASP method. Briefly, the detergent, DTT and IAA in UA buffer was added to block-reduced cysteine. Finally, the protein suspension was digested with 2 μg trypsin (Promega) overnight at 37°C. The peptide was collected by centrifugation at 16,000 g for 15 min. The peptide was desalted with C18 StageTip for further LC-MS analysis. LC-MS/MS experiments were performed on a Q Exactive Plus mass spectrometer that was coupled to an Easy nLC (Thermo Fisher Scientific). Peptide was first loaded to a trap column (100 μm×20 mm, 5 μm, C18, Dr Maisch GmbH, Ammerbuch, Germany) in buffer A (0.1% formic acid in water). Reverse-phase highperformance liquid chromatography (RP-HPLC) separation was performed using a self-packed column (75 μm×150 mm; 3 μm ReproSil-Pur C18 beads, 120 Å, Dr Maisch GmbH, Ammerbuch, Germany) at a flow rate of 300 nl/min. The RP-HPLC mobile phase A was 0.1% formic acid in water, and B was 0.1% formic acid in 95% acetonitrile. The gradient was set as following: 2%-4% buffer B from 0 min to 2 min, 4% to 30% buffer B from 2 min to 47 min, 30% to 45% buffer B from 47 min to 52 min, 45% to 90% buffer B from 52 min and to 54 min, and 90% buffer B kept until to 60 min. MS data was acquired using a data-dependent top20 method dynamically choosing the most abundant precursor ions from the survey scan (350-1800 m/z) for HCD fragmentation. A lock mass of 445.120025 Da was used as internal standard for mass calibration. The full MS scans were acquired at a resolution of 70,000 at m/z 200, and 17,500 at m/z 200 for MS/MS scan. The maximum injection time was set to 50 ms for MS and 50 ms for MS/ MS. Normalized collision energy was 27 and the isolation window was set to 1.6 Th. Dynamic exclusion duration was 60 s. The MS data were analyzed using MaxQuant software version 1.6.1.0. MS data were searched against the UniProtKB Rattus norvegicus database (36,080 total entries, downloaded 08/14/2018). Trypsin was seleted as the digestion enzyme. A maximum of two missed cleavage sites and the mass tolerance of 4.5 ppm for precursor ions and 20 ppm for fragment ions were defined for database search. Carbamidomethylation of cysteines was defined as a fixed modification, while acetylation of protein N-terminal, oxidation of Methionine were set as variable modifications for database searching. The database search results were filtered and exported with a <1% false discovery rate (FDR) at peptide-spectrum-matched level, and protein level, respectively.

Mass spectrometry for identification of PDZD8-GFP-and PDZD8-SMP-associated lipids
Mass spectrometry for identification of PDZD8-GFP-and PDZD8-SMPassociated lipids was performed as follows. To extract lipids, 1 ml methyl tert-butyl ether (MTBE) was added to GFP-Trap agarose beads (Chromoteck) and the samples were shaken for 1 h at room temperature. Next, phase separation was induced by adding 250 μl water, letting it sit for 10 min at room temperature and centrifuging for 15 min at 14,000 g, 4°C. Because of the low density and high hydrophobicity of MTBE, lipids and lipophilic metabolites are mainly extracted to the upper MTBE-rich phase. The lipid was transferred to fresh tubes and dried with nitrogen.
Additionally, to ensure data quality for metabolic profiling, quality control (QC) samples were prepared by pooling aliquots from representative samples for all of the analysis samples, and were used for data normalization. QC samples were prepared and analyzed with the same procedure as that for the experiment samples in each batch. Dried extracts were then dissolved in 50% acetonitrile. Each sample was filtered with a disposable 0.22 µm cellulose acetate and transferred into 2 ml HPLC vials and stored at −80°C until analysis. For UHPLC-MS/MS analysis, lipid analysis was performed on Q Exactive orbitrap mass spectrometer (Thermo Fisher Scientific) coupled to a UHPLC system Ultimate 3000 (Thermo Fisher Scientific). Samples were separated using a Hypersil GOLD C18 column (100×2.1 mm, 1.9 μm) (Thermo Fisher Scientific). Mobile phase A was prepared by dissolving 0.77 g of ammonium acetate to 400 ml of HPLC-grade water, followed by adding 600 ml of HPLC-grade acetonitrile. Mobile phase B was prepared by mixing 100 ml of acetonitrile with 900 ml isopropanol. The flow rate was set as 0.3 ml/min. The gradient was 30% B for 0.5 min and was linearly increased to 100% in 10.5 min, and then maintained at 100% in 2 min, and then reduced to 30% in 0.1 min, with 4.5 min re-equilibration period employed. Both electrospray ionization (ESI) positive-mode and negative-mode were applied for MS data acquisition. The positive mode of spray voltage was 3.0 kV and the negative mode 2.5 kV. The ESI source conditions were set as follows: heater temperature of 300°C, Sheath Gas Flow rate, 45arb, Aux Gas Flow Rate, 15 arb, Sweep Gas Flow Rate, 1arb, Capillary Temp, 350°C, S-Lens RF Level, 50%. The full MS scans were acquired at a resolution of 70,000 at m/z 200, and 17,500 at m/z 200 for MS/MS scans. The maximum injection time was set to for 50 ms for MS and 50 ms for MS/MS. MS data was acquired using a data-dependent Top10 method dynamically choosing the most abundant precursor ions from the survey scan (200-1500 m/z) for HCD fragmentation. Stepped normalized collision energy was set as 15, 25, 35 and the isolation window was set to 1.6 Th. QC samples were prepared by pooling aliquots that were representative of all samples under analysis, and used for data normalization. Blank samples (75% acetonitrile in water) and QC samples were injected every six samples during acquisition.
For data preprocessing and filtering, lipids were identified and quantified using LipidSearch 4.1.30 (Thermo, CA). Mass tolerance of 5 ppm and 10 ppm were applied for precursor and product ions. Retention time shift of 0.25 min was performed in 'alignment'. M-score and chromatographic areas were used to reduce false positives. The lipids with less than 30% relative standard deviation (RSD) of MS peak area in the QC samples were kept for further data analysis.
SIMCAP software (Version 14.0, Umetrics, Sweden) was used for all multivariate data analyses and modeling. Data were mean-centered using Pareto scaling. Models were built on principal component analysis (PCA), orthogonal partial least-square discriminant analysis (PLS-DA) and partial least-square discriminant analysis (OPLS-DA). All the models evaluated were tested for over fitting with methods of permutation tests. The descriptive performance of the models was determined by R2X (cumulative) [ perfect model: R2X (cum)=1] and R2Y (cumulative) [ perfect model: R2Y (cum)=1] values while their prediction performance was measured by Q2 (cumulative) [ perfect model: Q2 (cum)=1] and a permutation test (n=200). The permuted model should not be able to predict classes -R2 and Q2 values at the Y-axis intercept must be lower than those of Q2 and the R2 of the non-permuted model. OPLS-DA allowed the determination of discriminating metabolites using the variable importance on projection (VIP). The VIP score value indicates the contribution of a variable to the discrimination between all the classes of samples. Mathematically, these scores are calculated for each variable as a weighted sum of squares of PLS weights. The mean VIP value is 1, and usually VIP values over 1 are considered as significant. A high score is in agreement with a strong discriminatory ability and thus constitutes a criterion for the selection of biomarkers.
The discriminating metabolites were obtained using a statistically significant threshold of variable influence on projection (VIP) values obtained from the OPLS-DA model and two-tailed Student's t-test (Pvalue) on the normalized raw data at univariate analysis level. The P-value was calculated by one-way analysis of variance (ANOVA) for multiple groups analysis. Metabolites with VIP values greater than 1.0 and P-value less than 0.05 were considered to be statistically significant metabolites. Fold change was calculated as the logarithm of the average mass response (area) ratio between two arbitrary classes. On the other side, the identified differential metabolites were used to perform cluster analyses with R package.

GST-tag protein purification
GST constructs were transformed into Escherichia coli BL21 (DE3) cells, and cells were incubated at 37°C until the optical density (OD) at 600 nm reached 0.6-0.8. Subsequently, cells were incubated at 16°C for another hour, followed by induction with 1 mM IPTG overnight at 16°C. Cells were lysed via sonication. GST fusion proteins were purified via the GST-tag Protein Purification kit (P2262, Beyotime, China), followed by size-exclusion chromatography (GE Healthcare) with protein standard mix (69385, Sigma) according to the manufacturer's instructions.

Pulldown assays
Purified GST-C1-CC and GST was incubated with 20 μl Glutathione resin (P2262, Beyotime, China) at 4°C overnight, followed by washing beads with freshly prepared HNM buffer (20 mM Hepes, pH 7.4, 0.1 M NaCl, 5 mM MgCl 2 , 1 mM DTT and 0.2% NP-40). Purified His-Rab7 (80629-R07E, Sinobiological Inc.) or Rab5 (PRO-590, Prospec) was incubated with 100 μM GTP or GDP at 25°C for 30 min, followed by addition of 30 mM MgCl 2 and incubation at 25°C for another 1 h. Finally, GTP or GDP-loaded His-Rab7 or Rab5 was subjected to GST pulldown using GST-C1-CC or GST-only loaded glutathione resin overnight at 4°C, respectively. Glutathione resin was washed with freshly prepared HNM buffer. Glutathione resin was resuspended in 100 μl 2× SDS-sampling buffer. Re-suspended resin was boiled for 10 min at 95°C to dissociate protein complexes from beads. Western blotting was performed using anti-His antibodies and anti-Rab5 antibodies. The Coomassie staining was performed for purified GST-C1-CC and GST, with His-Rab7 and Rab5 as loading controls.

Protein purification of tethered PDZD8-SMP-His
The tethered-SMP-His constructs were transfected into 10-cm dishes of HEK293T cells, and the medium was changed 5 h later. After another 24 h, cells were incubated in OptiMEM. Medium was collected 48 h later, and centrifuged at 500 g for 10 min.The supernatant was mixed with 500 µl of balanced HisPur™ Ni-NTA Resin (88221, Thermo Scientific) and incubated in a 4°C shaking bed for 2 h. After centrifugation at 1000 g for 10 min, the supernatant was discarded. Proteins on beads were washed with 20 mM Tris-HCl pH 8.0, 300 mM NaCl buffer containing 20 mM, 50 mM, 100 mM, and 200 mM imidazole in turn. The eluted proteins were examined by Coomassie staining and western blotting experiments. Then, the eluted proteins were concentrated (Millipore UFC900396), and washed with imidazole-free buffer to obtain the imidazole-free and concentrated tethered-SMP-His proteins.

Dithionite assay
After performing a lipid-transfer reaction (as described above), 2.5 μl freshly prepared dithionite buffer (100 mM dithionite in 50 mM Tris-HCl pH 10) was added to reactions, and NBD fluorescence was monitored for an additional 15 min.

Liposome pull-down assay
For the liposome pull-down assay, proteins were diluted to 7 µg/ml in osmotically matched protein dilution buffer (20 mM Hepes, 120 mM NaCl, 1 mM EGTA, 0.2 mM CaCl 2 , 1.5 mM MgCl 2 , 1 mM DTT, 5 mM KCl, pH 7.4, 1% BSA was added to enhance solubility) and were pre-cleared by ultracentrifugation at 120,000 g for 45 min using a TLA100.3 rotor in a Beckmann Optima-MAX benchtop ultracentrifuge. 1 ml of protein mixture was then combined with 10 µl of liposomes and incubated with shaking at 30°C. Liposomes were recovered by ultracentrifugation (120,000 g for 30 min); supernatant and pellet fractions were resuspended in equal volumes of 1× Laemmli buffer and analyzed by western blots.

Dynamic light scattering assay
For the DLS measurements, a Litesizer 500 (Anton Paar) with a semiconductor laser (40 mW, 658 nm) was used. For each sample, three independent runs of 10 s were performed according to the manufacturer's instruction. Measurements were made at 25°C.
Image analysis for quantification of frequency or number of PDZD8-decorated LE/lys and quantification of the positioning of LE/lys All image analysis and processing was performed using ImageJ (National Institutes of Health). The number of PDZD8-decorated LE/lys was counted manually with assistance of ImageJ plugin 'Cell counter'. PDZD8decorated LE/lys represents the LE/lys that are completely coated by PDZD8. For quantification of LE/lys positioning, perinuclear regions of cells were defined as shown in Fig. S7B. To calculate the percentage of perinuclearly LE/lys, we manually counted the number of total LE/lys and perinuclearly LE/lys with the assistance of an ImageJ plugin Cell counter.

Analysis of neurite outgrowth in PC12 cells
Rat pheochromocytoma PC12 cells were grown in RPMI1640 (Invitrogen) supplemented with 10% horse serum (Gibco) and 5% fetal bovine serum (Gibco). PC12 cells were infected using Polybrene overnight before NGF (100 ng/ml) stimulation for 48 h. Cells were subjected to immunofluorescence stainings using anti-tubulin antibodies. The length and number of neurites were measured manually with ImageJ.

Statistical analysis
All statistical analyses and P-value determinations were performed in GraphPad Prism6. To determine P-values, an ordinary one-way ANOVA with Tukey's multiple comparison tests was performed among multiple groups. A two-tailed unpaired Student's t-tests was performed between two groups of data. The error bars in plots represent mean±s.d.