In a first step, we examined the protein interactions of the Arabidopsis BDCP SPI with LIP5 and SKD1. As large size of the SPI cDNA renders the molecular analysis extremely difficult, we used the evolutionarily conserved C-terminal part containing the PH-BEACH-WD40 domains (called SPI-PBW hereafter, Supplementary Figure S1A) for protein-interaction assays. Yeast two-hybrid assays revealed interactions between SPI-PBW and both, LIP5 and SKD1 (Figure 1A). The interaction between SPI-PBW and SKD1 was confirmed by co-precipitations (co-Ps) of bacterially expressed Glutathione-S-Transferase (GST)/His₆-fusions of SPI-PBW and His₆-fusions of SKD1 (Figure 1B). His₆-SKD1 did not bind to GST labeled resins, confirming the specificity of the SPI-PBW/SKD1 interaction. As the bacterial expression of His₆-LIP5 failed, we confirmed the SPI/LIP5 interaction in co-immunoprecipitation (co-IP) experiments with proteins expressed in Nicotiana benthamiana (N. benthamiana) leaf cells. Here, the expression of SPI-PBW was very weak and just above the detection level. We therefore used a slightly smaller SPI fragment lacking the WD40-repeats (termed SPI-PB; Supplementary Figure S1A). SPI-PB was C-terminally fused to a HA-tag (SPI-PB-HA). LIP5 was used as N-terminal fusion with three FLAG tags (3xFLAG-LIP5). These experiments revealed a clear co-IP suggesting an interaction between SPI-PB and LIP5. 3xFLAG-LIP5 did not bind to unlabeled beads, indicating that the SPI-PB/LIP5 interaction is specific (Figure 1C). The specificity of the immunoprecipitation was confirmed, by showing that SPI-PB-HA and 3xFLAG-LIP5 did not bind to α-ProteinA beads (Supplementary Figure S1B).

To monitor the intracellular localization of SPI, LIP5 and SKD1, we transiently expressed fluorescently marked proteins. As shown before, we observed YFP-SDK1 either evenly distributed in the cytoplasm or at one or several cytoplasmic dots in transiently transformed Arabidopsis leaf epidermal cells (Supplementary Figures S2A,B) (Shahriari et al., 2010b). Quantification of YFP-SKD1 localization revealed its localization to cytoplasmic dots in about one thirds of the cells (Supplementary Figure S2C). The YFP-SKD1 dots partially co-localized with two endosomal marker proteins, the Rab5 related small GTPases ARA6 (RabF1) and ARA7 (RabF2B) (Ueda et al., 2004; Ebine et al., 2011). Sixty-two percentage of YFP-SKD1 positive dots were labeled with ARA6-mCHERRY and 47.4% were marked with mCHERRY-ARA7 (Figures 2A–D). As reported before, LIP5-YFP localized to endosomes (Brandt et al., 1975; Shim et al., 2008; van Balkom et al., 2009). 59.3% of LIP5-YFP labeled dots showed also a ARA6-mCHERRY signal and 58.2% co-localized with mCHERRY-ARA7 (Figures 2E–H). In contrast to SKD1 and LIP5, the N-terminal YFP-fusions of SPI-PBW and full length genomic SPI (genSPI) were evenly distributed in the cytoplasm (Supplementary Figures S2D,E).

The intracellular localization of the interaction between SPI-PBW, LIP5, and SKD1 was assessed by Bimolecular fluorescence complementation (BiFC) assays in transfected N. benthamiana leaf epidermis cells. We found BiFC interactions of SPI-PBW with LIP5 and SKD1 in cytoplasmic dots. The Manders co-efficient of the SPI-PBW/SKD1 BiFC signal was 54.96% with ARA6-mCHERRY and 45.76% with mCHERRY-ARA7 (Figures 3A,B). 49.3% of the SPI-PBW/LIP5 BiFC signal co-localized with ARA6-mCHERRY and 44.6% with mCHERRY-ARA7 (Figures 3C,D). As SPI protein fusions are ubiquitously distributed in the cytoplasm we used also a cytoplasmically localized negative control. We found no interactions of SPI-PBW and LIP5 with AtMYC1 (MYC related protein1) (Pesch et al., 2013) (Supplementary Figure S2F). The functional integrity of AtMYC1 N-terminally fused to YFP-fragments was confirmed by demonstrating its BiFC interaction with TTG1 (TRANSPARENT TESTA GLABRA1) (Zimmermann et al., 2004; Zhao et al., 2012) (Supplementary Figure S2G). TTG1-YFP_C was included as negative control for YFP_N-SKD1 (Supplementary Figure S2F). In summary, our data suggest that the interaction between SPI-PBW and LIP5 and SKD1 can occur at endosomes.

Because SPI contains a PH domain known to mediate binding to phospholipids (Ferguson et al., 1995; Lemmon, 2007) we assessed the binding-ability of bacterially expressed GST-SPI-PBW-His₆ to a selection of membrane-anchored lipids in protein-lipid overlay assays (Supplementary Figures S3A–C). GST was included as a negative control and the PH domain from PLC-d1 as a positive control (Supplementary Figure S3B). We found no binding of GST-SPI-PBW-His₆ to any of the phospholipids tested, including the endosomal enriched Phosphatidylinositol-3-phosphate (PI3P) (Voigt et al., 2005; Vermeer et al., 2006) (Supplementary Figure S3C). These data suggest that the association of SPI to endosomes is not mediated by phosphoinositides.

To test the functional relevance of the SPI/LIP5/SKD1 interactions, we performed a genetic analysis. We focused on the seed coat mucilage phenotype because plants expressing dominant negative SKD1 produce seeds lacking seed mucilage completely (Shahriari et al., 2010a). Seed coat mucilage is enriched in polymeric sugars that are secreted from the outer seed coat cells. Its presence can be monitored by Ruthenium Red staining of whole seeds (Western et al., 2000). We found no difference between spi and wild-type seeds (Figures 4A,B). The mucilage layer of lip5 mutant seeds was still formed, but its area was significantly smaller than in wild-type seeds. This observation is consistent with the finding that LIP5 acts as an enhancer of SKD1 ATPase activity (Haas et al., 2007). The spi lip5 double mutant seeds exhibited a strong reduction of the mucilage layer that was significantly different to wild type as well as lip5 single mutant seeds (Figures 4A,B). Apart from this mucilage phenotype we did not note any differences to spi mutants in spi lip5 double mutant plants with respect to growth and morphological features. These data indicate that the secretion of seed coat mucilage requires the SKD1/SPI/LIP5 pathway and that SPI and LIP5 act synergistically.

To test whether SPI is functionally involved in the endosomal-vacuolar transport route, we examined the intracellular localization of two soluble vacuolar enzymes CPY and AALP in spi mutants (Rojo et al., 2003; Shahriari et al., 2010b). Both, CPY-CFP and AALP-mCHERRY were found in the vacuole in more than 90% of the cells. Their distribution was unaffected in both spi and lip5 single mutants. However, in spi lip5 double mutants we found a severe reduction of cells displaying vacuolar CPY and AALP (Figures 5A–D). The vacuolar sorting defect was even more obvious after controlled induction of CPY-mCHERRY expression using an EtOH-inducible vector system (Roslan et al., 2001) (Figures 5E,F and Supplementary Figure S4).

We reasoned that the vacuolar sorting defect might result in an increased secretion similar as observed in plants expressing dominant negative SKD1 (Shahriari et al., 2010b). We therefore performed a secretion assay using Col-0 and spi lip5 leaf mesophyll protoplasts. We compared the amounts of CPY present in the fraction of protoplasts with those secreted to the surrounding medium after 1 h of incubation (Figure 6). As compared to wild type, we found a stronger signal of CPY in the medium of spi lip5 protoplasts in two independent experiments (one shown in Figure 6). We excluded that the CPY signals from the medium samples were caused by impurities from damaged protoplast by using α-cytoplasmic Fructose-1.6-Bisphosphatase (cF1.6BPase) as a control (Figure 6).

The finding that the spi lip5 double mutants share many phenotypes with the dominant negative SKD1 mutants (mucilage phenotype, vacuolar transport defects, aberrant protein secretion) suggests that they may act together in promoting SKD1 activity. In this case, one would expect that the overexpression of SKD1 can compensate the reduced activation of SKD1 in spi lip5 double mutants. We therefore tested whether overexpression of SKD1 can rescue the spi lip5 vacuolar transport phenotype. Toward this end, we assessed the localization of AALP-mCHERRY and CPY-CFP in spi lip5 double mutants co-expressing SKD1. The vacuolar transport of AALP was restored in 100% (Figures 7A,B) and of CPY-CFP in 80% of cells (Figures 7C,D). Co-expression of the late endosomal marker ARA7 did not restore the vacuolar transport of CPY-CFP (Figures 7E,F), highlighting the SKD1-specificity of the rescue. These data suggest that LIP5 and SPI regulate endosomal-vacuolar transport by acting as positive regulators of SKD1.

The common phenotypes of BDCP mutants points to their function in the control of membrane fusion and fission events at different steps along the endomembrane system (Cullinane et al., 2013). Recently, a working model was developed that proposes BDCPs to act as scaffold proteins that regulate membrane dynamics in a binding partner-dependent manner (Cullinane et al., 2013). However, only few binding partners of BDCPs have been identified (Adam-Klages et al., 1996; Segui et al., 2001; Tchernev et al., 2002; Filimonenko et al., 2010; Hocking et al., 2010) and their contribution to the pathophysiology observed in BDCP mutants was not characterized in detail so far. A possible link between BDCPs and the LIP5/SKD1 complex had been suggested by the findings that LIP5 interacts with LYST in yeast two-hybrid assays (Tchernev et al., 2002) and that over-expression of a dominant negative SKD1 protein leads to an association of LYST to membranes (Fujita et al., 2004). Our finding that SPI directly interacts not only with LIP5 but also with SKD1 strongly suggest that a functional link between BDCPs and LIP5/SKD1 is evolutionarily conserved.

As shown for human LYST, the SPI protein is evenly distributed in the cytoplasm (Perou et al., 1997; Fujita et al., 2004). The previous finding that LYST relocates to membranes only upon co-expression of dominant negative SKD1 (E235Q) suggests that it is either recruited under these conditions, or that it is present at endosomal membranes at low levels and becomes detectable because of the formation of large endosomal class E compartments. Consistent with both scenarios, we observe SPI interactions with LIP5 and SKD1 at endosomes in BiFC experiments. As BiFC tends to stabilize protein–protein interactions (Kerppola, 2008) a recruitment of SPI to endosomes by its binding partners is conceivable.

Membrane association of other BDCPs has been shown to depend on their binding to phospholipids. The human BDCP factor associated with neutral sphingomyelinase activation (FAN) binds directly to plasma membrane associated phosphoinositides (Haubert et al., 2007). Other BDCPs such as the mammalian homologs NBEA and lipopolysaccharide-responsive and beige-like anchor protein (LRBA) do not bind to phospholipids. Crystal structure analysis of LRBA revealed an uncommon PH domain, lacking a cluster of basic residues essential to form the classical phospholipid binding fold (Gebauer et al., 2004; Haubert et al., 2007; Lemmon, 2007). SPI protein does not appear to bind to phosphoinositides suggesting that the recruitment of SPI to membranes may be mediated by binding to other proteins.

The ATPase activity of SKD1 is essential for the disassembly of the ESCRT machinery from membranes, and thereby the formation of ILVs in late endosomal compartments (Babst et al., 1998; Azmi et al., 2006; Cai et al., 2014). The depletion of SKD1 and expression of its dominant negative SKD1 versions cause the formation of aberrantly enlarged pre-vacuolar structures (class E compartments) in various species (Fujita et al., 2003; Shahriari et al., 2010b). In plants, the expression of dominant negative SKD1 causes vacuole fragmentation suggesting that the membrane transport between MVBs and the vacuole is impaired (Shahriari et al., 2010b). Although spi mutants exhibit fragmented vacuoles (Saedler et al., 2009), we found no obvious defects in endosomal trafficking. Neither the internalization of the styryl dye FM4-64 (Saedler et al., 2009) nor the vacuolar transport of soluble cargo proteins (this study) were affected in spi mutants. Thus, the cellular spi mutant phenotype differs to other class E (Fujita et al., 2003; Haas et al., 2007) or BDCP mutants (Faigle et al., 1998; Cornillon et al., 2002). Similar results were reported for lip5 mutant plants (Cai et al., 2014). Although LIP5 promotes the enzymatic activity of SKD1, lip5 mutants exhibit no global mutant phenotypes (Haas et al., 2007). However, a role of LIP5 in MVB biogenesis was reported in the context of stress responses (Wang et al., 2014, 2015). In addition, trafficking defects of PIN proteins in roots were correlated with impaired gravitropism (Buono et al., 2016). Together these data suggest that LIP5 is a regulator of SKD1 but not an absolute requirement for SKD1 activity. Our genetic studies, showing a redundant function of LIP5 and SPI suggest that the two proteins act together in the regulation of SKD1. As the additional expression of SKD1 can overcome the requirement of LIP5 and SPI it is likely that SKD1 acts genetically downstream of LIP5 and SPI. It is conceivable that SPI acts together with LIP5 as a positive regulator of SKD1. This conclusion appears to be in conflict with data obtained in human cell culture systems. Here, overexpression of SKD1 was not sufficient to restore the giant lysosome phenotype in LYST deficient fibroblasts suggesting that LYST acts downstream of SKD1-mediated membrane transport (Fujita et al., 2004). This suggests that the molecular function we describe for the BDCP SPI is mechanistically different to that of BDCPs in humans. An alternative explanation would be that a rescue of protein/membrane trafficking by additional expression of SKD1 is independent of a rescue of endosomal sizes.

Arabidopsis thaliana spi-2 (GK_205G08), spi-3 (SALK_065311), spi-4 (GK_420D09), lip5-1 (SAIL_854F08) and lip5-2 (GK_351B10) have been previously described (Haas et al., 2007; Wang et al., 2014); and were obtained from the National Arabidopsis Stock Center. Positions of T-DNA insertions were confirmed by sequencing the flanking genomic regions. Seeds were grown on soil under long day conditions (22°C; 110 μmol m^-2 s^-1). Ruthenium Red Staining of seed coat mucilage was performed in H₂O (+0.1% Tween20) as described (Western et al., 2000). Isolation of protoplasts from Arabidopsis leaves was performed according to the Tape-Arabidopsis Sandwich method (Wu et al., 2009). Fifty microliters protoplasts were harvested and washed with 150 μl of W5 solution (154 mM NaCl, 125 mM CaCl₂, 5 mM KCl, 5 mM glucose, and 2 mM MES, pH 5.7) in a 1.5 ml Eppendorf tube at 100g for 5 min. For determination of protein secretion 50 μl of washed protoplasts were incubated in 50 μl of W5 solution for 60 min at room temperature in the dark. After centrifugation (5 min, 100g), supernatant (secretion fraction) and intact protoplasts were divided and boiled in protein loading buffer for further analysis.

Coding sequences (CDS) of SPI-PBW (AT1G03060) was described before (Steffens et al., 2015). The CDS of SPI-PB and AALP (AT5G60360) were amplified from Col-0 cDNA. Primer sequences are provided in Supplementary Table S1. Entry clones of SKD1 (AT2G27600), LIP5 (AT4G26750), CPY (AT3G10410), ARA6 (AT3G54840) and ARA7 (AT4G19640) were published before (Shahriari et al., 2010b). All constructs were confirmed by sequencing. A list of GATEWAY^®-vectors used in this study is provided in Supplementary Table S2.

Yeast two-hybrid assays were performed as described (Gietz et al., 1995). Interactions were analyzed on dropout media lacking leucine (L), tryptophan (W) and histidine (H) [+20 mM 3-Amino-1,2,3-Triazole (3AT)]. For biochemical assays GST-SPI-PBW-His₆, GST and His₆-tagged SKD1 were expressed and purified as described before (Steffens et al., 2014). Lysis of bacteria expressing GST-SPI-PBW-His6 was performed in STE-Buffer (10 mM Tris pH8.0, 150 mM NaCl, 1 mM EDTA). After centrifugation of the lysates, PBS (pH 7.5) was used for washing and elution according to (Steffens et al., 2014). GST-SPI was eluted in PBS, supplemented with 40 mM reduced L-glutathione. His₆-fusions were eluted in Tris-Elution buffer (100 mM Tris pH 8.0, 150 mM NaCl) supplemented with 500 mM imidazole. For co-precipitation and protein-lipid binding assays (Echelon bioscience; performed with 1.5 μg purified protein according to the manufacturers instructions), purified proteins were concentrated using Amicons Ultra-4 Centrifugal Filters (Merck Millipore) and dialyzed (Xpress Micro Dialyzer Cartridge MD300; Scienova) against Tris-Elution Buffer. Purifications and co-precipitations were analyzed by immunoblotting (Steffens et al., 2014). HA- and 3xFLAG fusions were (co)-immunoprecipitated using antibody conjugated μ-columns according to the manufacturer’s instructions (Miltenyi; μMACS^TM) from total protein lysates derived from infiltrated N. benthamiana leaves. Bound proteins were eluted and analyzed by immunoblotting, using either a monoclonal horseradish peroxidase (HRP) coupled with anti-HA antibody from rat (Roche; 1:1000 dilution) or a primary monoclonal anti-FLAG antibody from mouse (Sigma–Aldrich; 1:2000 dilution) and a secondary anti-mouse antibody from goat coupled to HRP (Sigma–Aldrich; 1:10,000).

Nicotiana benthamiana leaves were transformed as described (Yang et al., 2000). Biolistic transformation of Arabidopsis leaves was performed as described (Mathur et al., 2003) and analyzed after 12 to 16 h by Confocal Laser Scanning Microscopy (CLSM). CLSM was performed as described (Steffens et al., 2014). Expression of alcR_pro-fusion proteins (pCAMPARI-vector) was induced by fumigation of transfected leaves with 2% ethanol (v/v) for 4 h (Roslan et al., 2001). Before induction of protein expression, transfected cells were identified by the presence of constantly co-expressed free YFP. To ensure comparisons between equally strong transfected cells, laser intensities and exposure times were fixed.