Phosphoinositide-binding proteins mark, shape and functionally modulate highly-diverged endocytic compartments in the parasitic protist Giardia lamblia

Phosphorylated derivatives of phosphatidylinositol (PIPs) are key membrane lipid residues involved in clathrin-mediated endocytosis (CME). CME relies on PIP species PI(4,5)P2 to mark endocytic sites at the plasma membrane (PM) associated to clathrin-coated vesicle (CCV) formation. The highly diverged parasitic protist Giardia lamblia presents disordered and static clathrin assemblies at PM invaginations, contacting specialized endocytic organelles called peripheral vacuoles (PVs). The role for clathrin assemblies in fluid phase uptake and their link to internal membranes via PIP-binding adaptors is unknown. Here we provide evidence for a robust link between clathrin assemblies and fluid-phase uptake in G. lamblia mediated by proteins carrying predicted PX, FYVE and NECAP1 PIP-binding modules. We show that chemical and genetic perturbation of PIP-residue binding and turnover elicits novel uptake and organelle-morphology phenotypes. A combination of co-immunoprecipitation and in silico analysis techniques expands the initial PIP-binding network with addition of new members. Our data indicate that, despite the partial conservation of lipid markers and protein cohorts known to play important roles in dynamic endocytic events in well-characterized model systems, the Giardia lineage presents a strikingly divergent clathrin-centered network. This includes several PIP-binding modules, often associated to domains of currently unknown function that shape and modulate fluid-phase uptake at PVs.

In addition to their structural functions in membranes, in model eukaryotes PIPs are involved in spatiotemporal organization of membrane remodeling processes such as clathrincoated vesicle (CCV) formation during clathrin-mediated endocytosis (CME). In particular, Included in the giardial CHC interactome were three proteins with predicted PIP-binding domains: FYVE domain protein Gl16653 and two PX-domain proteins (Gl7723 and Gl16595), the latter part of a six-member protein family (Table 1; [19,20]). In a previous study, we hypothesized that Gl16653 (GlFYVE), Gl7723 (GlPXD1) and Gl16595 (GlPXD2) act as PIPbinding adaptors to link and maintain static clathrin assemblies at the PM and PV membrane interface in G. lamblia [19]. We further postulated that a perturbation of PIP-binding protein levels and/or function would lead to impaired fluid-phase uptake by affecting PV functionality. To test these hypotheses, we performed an in-depth functional characterization of all previously-identified PIP-binding proteins associated to clathrin at PVs. We assessed their lipidbinding preferences and visualized their subcellular localizations using electron microscopy and both conventional and super resolution light microscopy. By manipulating protein levels and/or function we could elicit novel fluid-phase uptake and PV morphology-related phenotypes, thereby establishing PIPs as a link between the role of clathrin as a membrane remodeling protein and PV-based endocytosis in G. lamblia. Furthermore, we used a combination of co-immunoprecipitation and in silico annotation techniques to expand protein interactomes established previously, thereby discovering a new set of PIP-binding proteins with roles likely reaching beyond the PV compartment. Lastly, we propose an updated working model summarizing the complex networks between PIP-binding proteins and clathrin assemblies at PVs.

Results
The G. lamblia genome encodes at least seven distinct PIP-binding modules Table 1. G. lamblia PIP-binding proteins. A compilation of all PIP-binding domains identified in the Giardia Genome Database (www.giardiadb.org; GDB) using previously characterized domains [24] as baits for HMM-based homology searches (column 1). Predicted giardial orthologs are present for PIP-binding domains ENTH, PH, FYVE, PX, BAR, FERM and PROPPINs (column 2) and mostly retrieve the correct domains when used as baits for reverse HHpred searches (column 4). Except for Glepsin, GlPXD2 and GlPROP1 and 2, all others are currently annotated on GDB as generically "hypothetical", i.e. of unknown function (column 6). Each orthologue was assigned a name used throughout this report (column 7). Functional domain predictions using SMART (http://smart.embl-heidelberg.de/; column 8) and subcellular localization data (column 9) either previously reported or acquired in this study (column 10), are also included.
Since GlPXD1-2, GlFYVE and GlNECAP1 were experimentally shown to be associated to giardial clathrin assemblies [19], we selected these proteins and GlPXD3-6 for more detailed subcellular localization experiments. Stimulated emission-depletion (STED) microscopy in co-labelling experiments with Dextran-OG as a marker for fluid-phase endocytosis unequivocally confirmed accumulation for GlPXD1-4 and 6, GlFYVE and GlNECAP1 epitope-tagged reporters at PVs (Fig 1C-1I). The signal generated by GlPXD5 reporters was insufficient for a conclusive localization using STED microscopy but was shown to localize in close proximity to PVs using conventional confocal microscopy ( Fig 1B).
To extend the initial annotation of giardial PIP-binding proteins we performed multiple sequence alignment (MSA) analyses for each giardial PIP-binding module with selected orthologs to delineate lipid-binding motifs and residues critical for PIP recognition (S1 Fig). In silico structural analyses of the lipid-binding domains of giardial proteins and their closest homologs were performed ab initio using the online tool I-TASSER [32][33][34]. Comparative analysis of structure models generated with I-TASSER clearly demonstrated positional conservation of residues critical for PIP binding (S1 Fig). Taken together, in silico analysis identifies seven distinct PIP-binding module types encoded in the G. lamblia genome, conserved on both sequence and structural levels. Subcellular localization of epitope-tagged variants by fluorescence microscopy indicates clear association to PVs with the exception of Glepsin [21].
Quantification of the chemiluminescence signals shows a marked preference of MBP-GlPXD1 for PI(4,5)P 2 in PIP gradients ( Fig 2B) which was corroborated by experiments using PIP strips (S2C Fig). Under these conditions, GlPXD2, 3, and 6 show unexpectedly promiscuous binding preferences, with GlPXD2 presenting a marked affinity for PI(3)P and PI(4,5)P 2 , GlPXD3 for PI(3)P and to a lesser extent PI(5)P, and GlPXD6 for PI(3)P, PI(4)P, and PI(5)P ( Fig 2B). These data were in line with results from independent PIP strip experiments (S2C and S2D Fig). MBP-GlPXD4 and MBP-GlPXD5 binding preferences could only be probed using PIP strips (S2C Fig), showing in both cases a marked affinity for PI(3,5)P 2 and PI(4,5)P 2 (S2C and S2D Fig). Binding preferences for MBP-GlFYVE could not be determined, given that no signal was ever obtained on both PIP arrays and strips (Fig 2A, S2C

Depletion of free PI(3)P, PI(4,5)P 2 and PI(3,4,5)P 3, but not PI(4)P binding sites in vivo inhibits PV-mediated uptake of a fluid-phase marker
The marked preference of GlPXD1-6 for PIP residues PI(3)P and PI(4,5)P 2 raised the question whether their perturbation would elicit loss-of-function phenotypes in fluid phase uptake by Giardia trophozoites. Using a combination of commercially available antibodies, heterologous reporter constructs and chemical treatment, we reduced the bioavailability of PI3P, PI(4,5)P 2 , and in addition PI(3,4,5)P 3 and PI(4)P. for all identified PIP-binding proteins including positions of repetitive motifs and putative lipid and Zn -binding residues using HHPRED, HMMER and InterProScan. Ptd-Phosphatidylinositol. (B) Conventional confocal light-microscopy analysis of representative non-transgenic trophozoites labelled with Dextran-OG (first panel) to mark PV lumina and of antibody-labelled trophozoites expressing HA-tagged PIP-binding protein reporters. Except for Glepsin-HA and HA-GlFERM, all tested reporter proteins localize in close proximity to peripheral vacuoles (PVs) at the cell cortex. Epitope-tagged HA-GlPXD5 and GlPROP1-HA additionally show signal distribution throughout the cell. Cells were imaged at maximum width, where nuclei and the bare-zone are at maximum diameter. Epitope-tagged Glepsin-expressing cells were imaged at maximum width of the ventral disk. Scale bar: 1 μm. (C-I) Confocal STED microscopy analysis of trophozoites expressing epitope-tagged PIP-binding reporter proteins for GlPXD1-6, GlFYVE-HA and GlNECAP1-HA (red channel) co-labelled with Dextran-OG as a marker for PV lumina (green channel). As shown in the merged insets, although all reporters are clearly PV-associated, reporters for proteins GlFYVE-HA and HA-GlPXD1 and 2 are proximal to the PM with respect to Dextran-OG, indicating they reside at the PV-PM interface. In contrast, reporters for HA-GlPXD3 and GlNECAP1-HA appear to intercalate PVs. Scale bars: 1 μm for full cell and inset images.  , from 100pmol (A) to 1.56pmol/spot (G), were probed with fixed amounts (2.5 μg) of clathrin assemblies-associated epitopetagged PIP-binding domains from GlPXD1-6, GlNECAP1 and GlFYVE, followed by immunodetection of the epitope tag. The protein fusion partner MBP (MBP alone) and for antibodies raised against PI(4,5)P 2 (anti-PI(4,5)P 2 ) were included as negative and positive controls for binding, respectively. No signal using arrays was obtained for MBP-GlPXD4 and MBP-GlPXD5 however, binding preferences for these fusions were determined using lipid strips (S2A and S2B Fig). (B) Plots of densitometric Detection of PI(3)P, PI(4,5)P 2 , and PI(3,4,5)P 3 in chemically fixed trophozoites by immunofluorescence microscopy with primary PIP-targeted antibodies highlights enrichment for all PIP moieties in the cortical region containing PVs (S3 Fig). Ectopic expression of fluorescent high-affinity reporters for PI(3)P and PI(4)P, namely 2xFYVE-GFP and GFP-P4C [42], respectively, in transgenic G. lamblia trophozoites was used to identify membranes enriched for PI(3)P and PI(4)P deposition (Fig 3A-3D). Live microscopy of cells expressing 2xFYVE-GFP shows distinct reporter accumulation in cortical areas consistent with binding to PV membranes ( Fig 3B, green panels), whereas representative cells from line GFP-P4C show a more diffused cytosolic staining pattern, with some accumulation at PVs (Fig 3D, green panels). Fluid-phase uptake of Dextran-R was assessed in cells from both transgenic lines, and compared to wild-type cells using quantification of signal intensity. Wild-type control cells and transgenic cells weakly expressing 2xFYVE-GFP ( Fig 3A) incorporated large amounts of Dextran-R ( Fig 3E). Conversely, a strong 2xFYVE-GFP signal correlated with low amounts of endocytosed Dextran-R detected at the cell periphery and with noticeably enlarged cells ( Fig 3B). In contrast, there was no detectable difference in either Dextran-R uptake efficiency (based on fluorescent signal intensity) or cell width between weak ( Fig 3C) and strong expressors (Fig 3D) of the GFP-P4C line. Cell width ( Fig 3F) and fluidphase uptake (Fig 3G) aberrant phenotypes in 2xFYVE-GFP cells were recorded with respect to wild-type control and GFP-P4C cells and tested for significance (p>0.05) on 100 cells/line selected in an unbiased fashion. These data translate into a significant negative correlation between expression of the PI(3)P-binding 2xFYVE-GFP reporter and fluid-phase uptake ( Fig  3H) whereas only a slight albeit insignificant correlation was found between Dextran uptake and GFP-P4C expression ( Fig 3I). Furthermore, performance of the Mann-Whitney test on our data confirmed the significance of the observed changes in cell width only in 2xFY-VE-GFP-expressing cells, compared with non-transgenic cells. Specifically, the null hypothesis for no change in width was rejected for the WB vs 2xFYVE-GFP comparison (p(0.05)value = 2.22045e-16) and accepted for the WB vs GFP-P4C comparison (p(0.05)value = 0.182377).
The cationic antibiotic neomycin binds tightly to the headgroup of phosphoinositides with a marked preference for PI(4,5)P 2 and, to a lesser extent, PI(3,4,5)P 3 [43,44]. As a means to perturb PI(4,5)P 2 and PI(3,4,5)P 3 availability in Giardia trophozoites, we tested its effect on fluid-phase uptake by treating wild-type trophozoites with 7.2 mM neomycin followed by uptake of Dextran-R. Quantitative light microscopy image analysis revealed a significantly lower level of Dextran-R in treated trophozoites (p<0.05) (Fig 3J and 3K) which remained vital and motile in the presence of neomycin up to 15mM for 50 minutes (S1-S4 Videos). Taken together, the data indicate that depletion of free binding sites for PI(3)P, PI(4,5)P 2 , and PI(3,4,5)P 3 , but not PI(4)P significantly impacts fluid-phase endocytosis through G. lamblia PVs.

Functional characterization of GlPXD1-4 and 6, GlFYVE and GlNECAP1
Manipulation of PIP residue homeostasis elicited PV-dependent fluid-phase uptake phenotypes. We hypothesized that changing expression levels of giardial PIP-binding proteins previously identified in clathrin interactomes would elicit aberrant uptake phenotypes in Giardia trophozoites. In addition, we explored the functional boundaries of each PIP-binding module by defining their protein interactomes. To test this, we used the previously-generated epitope-analyses using FIJI for each MBP-fused PIP-binding domain and each spotted PI/PIP residue based on array data presented in (A). (C) Testing of the binding affinity of the MBP-fused PIP-binding domain from GlNECAP1 on a wider range of lipid residues detects cardiolipin as the preferred substrate. (E) Dextran-R uptake in non-transgenic wild-type cells as negative controls for construct-induced uptake phenotypes. Scale bars: 1 μm (F) Box-plot representing the distribution of cell width (in μm) across at least 100 wild-type, 2xFYVE-GFP-and GFP-P4C-expressing cells selected in an unbiased fashion. A statistically significant (two-sided t-test assuming unequal variances, p<0.05) increase in median cell width with respect to non-transgenic cells is detected for 2xFYVE-GFP-but not GFP-P4C-expressing cells. Asterisks indicate statistical significance. n.s.: not significant. (G) Box-plot representing the distribution of measured Dextran-R signal intensity across at least 100 wild-type, 2xFYVE-GFP-and GFP-P4C-expressing cells selected in an unbiased fashion. A statistically significant (two-sided t-test assuming unequal variances, p<0.05) decrease in Dextran-R signal intensity, normalized to wild-type cells (100%), is detected for 2xFYVE-GFP-but not for GFP-P4C-expressing cells. Asterisks indicate statistical significance. n.s.: not significant. (H) A statistically significant (p<0.5) linear correlation exists between Dextran-R signal (x-axis, intensity_red channel [%]) and 2xFYVE-GFP expression (y-axis, intensity_green channel [%]) measured across 100 cells. (I) The apparent linear correlation between GFP-P4C expression (y-axis, intensity_green channel [%]) and Dextran-R signal (x-axis, intensity_red channel [%]) is not statistically significant (p<0.5). (J) Wide-field microscopy-based immunofluorescence analysis of the impact of neomycin treatment on Dextran-R uptake to deplete PI(4,5)P 2 binding sites in non-transgenic wild-type cells. With respect to non-treated cells (WT; left panel), Dextran-R signal at PVs is visibly impacted in non-transgenic neomycin-treated cells (WT_Neo; right panel). Scale bars: tagged reporter lines for full-length GlPXD1-4 and 6, GlFYVE and GlNECAP1 (Fig 1C-1I) for assessing the effects of ectopic expression on fluid-phase uptake phenotypes. Furthermore, we used the same lines as a source of tagged "baits" in antibody-based affinity co-immunoprecipitation (co-IP) and identification of reporter-associated protein complexes (Table 2). Further investigation of GlPXD5 was abandoned at this stage due to its intractably low levels of expression. Table 2. Overview of interactomes derived from epitope-tagged reporter lines for full-length GlPXD1-4 and 6, GlFYVE and GlNECAP1. Main putative interaction partners are highlighted for each antibody-based affinity co-IP experiment. The type of interaction detected (reciprocal/one-sided) and its relative strength as indicated by the number of exclusive spectral counts associated to each candidate interactor, are also included.

Bait-ORF number Partners Interaction Data Annotation and reference(Zumthor et al., 2016 if not indicated)
One sided-strong S4E Fig
Transmission electron microscopy (tEM) analysis confirmed the presence of randomly distributed peripheral PV clusters in cells expressing HA-GlPXD2 (Fig 5E; left panel) which were not present in representative wild-type control cells (Fig 5E; right panel).
The GlPXD3 interactome is connected to clathrin assemblies and includes a novel dynamin-like protein. GlDRP, GlCHC, and GlAP2 (α/β subunits) were detected in the GlPXD3 interactome, thereby establishing the association of this PX domain protein with clathrin assembly structures at the PV/PM interface (Figs 4 and S4C and S3 Table). A pseudokinase (Gl15411 [48]) previously identified in GlCHC assemblies was also found in the GlPXD3 interactome (Figs 4 and S4C [19]). Furthermore, the GlPXD3 and Gl15411 interactomes share proteins GL50803_16811 (Gl16811) tentatively annotated as a ZipA protein in GDB, and proteins GL50803_87677 (Gl87677) and GL50803_17060 (Gl17060), annotated as a NEK kinase and an ankyrin-domain carrying protein, respectively (Figs 4 and S4C). Unique interaction partners for GlPXD3 include the SNARE protein Gl7309 [47] and GlNSF (GL50803_114776) [49]. In addition, protein GL50803_103709 carrying a predicted N-terminal BRO domain and protein GL50803_9605 were identified as unique GlPXD3 interaction partners (Figs 4 and S4C). Furthermore, the StAR-related lipid-transfer protein Gl16717, already found in the GlPXD6 interactome was also found to be a low-stringency interaction partner for GlPXD3 and Gl15411, thereby connecting the GlPXD3 and GlPXD6-GlFYVE circuits.
Protein Gl9605, the sixth most abundant hit in the GlPXD3 interactome (S3 Table), and currently annotated as having an unknown function, was localized in close proximity to PVs ( Fig 6A) and identified as a highly-diverged dynamin-like protein (Fig 6B).
In support of this, the predicted GTPase domain in Gl9605 contains signature motifs in the P-loop (G1), switch 1 (G2) and switch 2 (G3) regions [50][51][52]. Conserved motifs in the G4 region are only partially maintained (Fig 6B). To test residue conservation on a structural level, Gl9605 was subjected to ab initio modelling using I-TASSER and the resulting tertiary structure was superimposed on that of a dynamin-like protein 2 (DLP2 Cj:5ovW) [53], Gl9605's closest structural homologue (Fig 6C). A structural overlap TM-score of 0.913 suggests an almost perfect structural match, with clear chemical and positional conservation of key residues involved in GTPase activity ( Fig 6C). We sought to elicit a dominant-negative phenotype by engineering Gl9605 K73E and S74N mutants [54]. In contrast to either wildtype cells or cells expressing a wild-type epitope-tagged Gl9605 control, expression of Gl9605 K73E and S74N mutant reporters inhibited fluid-phase uptake of Dextran-R in a statistically significant manner (p<0.05; Fig 6D).
Regulated ectopic expression of GlFYVE variants inhibits fluid-phase uptake. GlFYVE is a confirmed interactor of clathrin assemblies [55] through specific association to GlCHC and GlDRP (Figs 4 and S4D and S6 Table). GlFYVE's extended interactome includes GlPXD6 and GlNECAP1.
To characterize the function of GlFYVE and to test whether a dominant-negative effect on uptake could be elicited, we performed a deletion analysis by generating epitope-tagged C-terminal (pCWP1::NT-GlFYVE-HA) and N-terminal (pCWP1::CT-GlFYVE-HA) truncation constructs. These consist of either the disordered region followed by the FYVE domain ( Fig  7A), residues 1-300) or the armadillo repeat-rich (ARM repeats) domains (Fig 7A), residues 301-990), respectively.
Expression of both constructs is regulated by an inducible promoter which is de-repressed during transient induction of encystation, the process during which a flagellated trophozoite differentiates to a cyst [56]. After a short (6h) induction pulse, transfected cells were subjected to Dextran-R uptake. In cells expressing full-length pCWP1::GlFYVE-HA and truncated variants, the amount of Dextran-R accumulated in PVs was significantly (p<0,05) lower ( Fig 7B,  box plot). Furthermore, IFA analysis of pCWP1::NT-GlFYVE-HA cells revealed the presence of structures which overlapped neither with Dextran-R-labelled PVs ( Fig 7B) nor with encystation specific vesicles (ESVs) labeled with the anti-CWP1 antibody (Fig 7C). In contrast, CT-GlFYVE-HA and full length GlFYVE-HA localized predominantly to PVs (Fig 7B-7D). The subcellular localization of GlCHC in these lines and in a wild-type control overlapped with the truncated CT-GlFYVE-HA variant, but only partially with NT-GlFYVE-HA and GlFYVE-HA ( Fig 7D).
Ectopic expression of GlNECAP1 significantly impairs fluorescent Dextran uptake. Co-IP using epitope-tagged GlNECAP1 confirmed interaction with clathrin assembly compo-  Table).
Three putative conserved AP2-interacting motifs were identified using multi-sequence alignment; the high affinity WxxF motif at the N-terminus, two residues being invariant Overview of core protein interactomes determined from co-IP analyses. Interactomes for GlFYVE, GlNECAP1 and GlPXD3, 4, and 6 defined by co-IP analysis were integrated with previously published data [19] for core clathrin assembly, GlPXD1, GlPXD2 and GlFYVE interactomes. Solid lines: interactions detected at high stringency. Dashed lines: interactions detected at low stringency. Yellow partners are currently annotated on GDB as "hypothetical protein" i.e. proteins of unknown function.
De novo 3D modelling confirms overall structural conservation of all key residues in GlNE-CAP1 compared to mammalian NECAP1 ( Fig 8B). Furthermore, the interacting interface of NECAP1 with the β-linker region of AP2 was also identified in the structural model for GlNE-CAP1 (Fig 8B).
To test whether expression of a GlNECAP1 variant lacking the putative high-affinity motif WVIF could elicit a dominant-negative uptake effect, a deletion construct GlNECAP1ΔW-VIF-HA was synthesised ( Fig 8A) for ectopic expression. Accumulation of Dextran-R into PVs detected by microscopy was significantly lower (p<0.05) in transgenic cells ectopically expressing GlNECAP1-HA or an APEX-and epitope-tagged variant GlNECAP1-APEX2-2HA compared with wild type controls (Fig 8C, box plot). Conversely, ectopic expression of a deletion construct GlNECAP1ΔWVIF-HA (Fig 8C, GlNECAP1ΔWVIF-HA) had no discernible effect on accumulation of Dextran-R in PVs (Fig 8C, box plot). Ectopic expression of the genetically encoded enzymatic reporter [58,59] GlNECAP1-APEX2-2HA was associated to significantly enlarged PVs in tEM compared to wild type controls (Fig 8D; S5 Fig).

GlPXD3 associates specifically to PVs as membrane coat
Co-localization studies with Dextran-OG and ectopically expressed HA-GlPXD3 show apparent coating of the entire PV membrane on the cytoplasmic side by the reporter construct ( Fig  9A).
This provided us with an opportunity to generate measurements of PV organelles in optical sections using 3D STED microscopy followed by reconstruction and rendering with IMARIS. Rendered images show hive-like GlPXD3-labelled structures predominantly in the cortical area of the cell underneath the PM that clearly surround the entire PV membrane (Fig 9B). The major and minor principal axes of these structures measured 437 +/-93 nm and 271 +/-60 nm. Consistent with the subcellular localization of this marker on the cytoplasmic side of PV membranes, these values were significantly higher (p� 0.05) than those obtained from PVs labeled with Dextran-OG (371 +-79 nm and 221 +/-49 nm) (Fig 9C and 9D). Signal overlap of epitope-tagged GlPXD3 with endogenous GlCHC as a marker for the PM-PV interface [19] in fluorescence microscopy is low. The image data indicate that both labels have distinct distributions but may spatially overlap at focal clathrin assemblies in small areas at the PV-PM interface (Fig 9E). Similarly, labelling for both PI(3)P and a reporter GlPXD3 variant showed minimal signal overlap (Fig 9F), despite the strong affinity of the latter for this lipid in in vitro lipid-array binding experiments (Fig 2A and 2B).

PIPs and PIP binders in G. lamblia
PIPs are recognized spatiotemporal organizers and decorate the surface of the eukaryotic cell's plasma and endo-membrane system [1][2][3]. G. lamblia is no exception; despite its significant  reduction in endomembrane complexity, this species maintains a variety of PIP residues, mostly located at the cell periphery. We identified 11 novel proteins, in most cases of unknown function that carry predicted PIP-binding modules and primarily localize in close proximity to PVs.
All hitherto identified PIP-binding proteins in G. lamblia can be loosely grouped in two categories; they are either relatively small proteins (up to 400 amino acid residues) consisting almost entirely of the PIP-binding module (e.g. GlPXD6 and GlNECAP1), or they are large proteins consisting of a single predicted domain for PIP-binding associated to domains of unknown function (e.g. GlPXD2 and GlFYVE). A full functional characterization of the latter is a challenge given the level of genomic sequence divergence in G. lamblia. This makes it currently difficult to determine whether sequences are lineage-specific or so diverged as to be unrecognizable orthologues of previously characterized proteins. Hence, structural annotation of large G. lamblia proteins carrying PIP-binding modules such as GlPXD2 or GlFYVE is limited to the lipid binding domain.
Eight out of 14 identified PIP-binding modules are either directly or indirectly associated to clathrin assemblies. Their PIP binding preferences, as measured using in vitro lipid-binding assays, are clearly distinct despite showing a varying degree of promiscuity, consistent with previously published data [20]. In contrast to previous reports, we could not measure PIP residue binding activity for GlFYVE using in vitro lipid-binding assays [22]. Furthermore, GlNE-CAP1 showed a distinctive and highly specific binding preference for cardiolipin. This is a surprising finding since cardiolipin is an abundant phospholipid of the inner mitochondrial membrane [60] whose presence in Giardia is controversial [61,62]. Although GlNECAP1 lacks canonical motifs for cardiolipin binding [63], previous reports on the identification of cardiolipin-binding PH domains [64,65] lend support to the observation that the PH-like domain in GlNECAP1 could bind cardiolipin, at least in vitro. The evolutionary implications for the presence of cardiolipin in an organism with "bare-bones" mitochondrial remnants i.e. mitosomes, with no maintenance of membrane potential nor ATP synthesis activity [66], provide for an exciting research direction worth pursuing.

An interactome-based model for PIP-binding proteins and clathrin assemblies at PVs
Data derived from APEX-mediated tEM experiments on transgenic trophozoites expressing APEX-tagged clathrin assembly components (GlCHC and GlCLC; [19]) show how larger PVs are associated to more than one PM-derived clathrin-marked invagination (Fig 10A). This is supported by data from IFA and STED microscopy analysis of trophozoites loaded with Dextran-OG and labelled with anti-GlCHC antibodies (Fig 10B). By combining APEXderived tEM data with STED microscopy data for both Dextran-OG and GlPXD3 labelling, a quantified sub organellar model for PV organization can be built which takes into account organelle size and relative distribution of clathrin assemblies (Fig 10C). In this model, GlPXD3 clearly emerges as a membrane coat that surrounds individual PV organelles (Fig 10C, upper  panel) on the cytoplasmic side of clathrin assemblies at the PV-PM interface (Fig 10C, lower  panel).
The PV-associated PIP-binding protein interactome appears as a tightly knit molecular network with GlCHC at its center (Figs 4 and 10D). Despite the high level of interconnectivity of distinct PIP-binder interactomes (Fig 4), specific molecular circuits such as the ones defined by the SNARE quartet (S4F Fig), pseudokinase Gl15411 and novel DLP Gl9605 (S4G Fig), as well as StAR-related lipid-transfer protein Gl16717 (S4H Fig), can be recognized. Notably, involved in the interaction between NECAP1 proteins and AP2 complexes (shaded in blue and green). (C) Wild-type non-transgenic control cells (WT) and cells expressing either epitope-tagged GlNECAP1 reporters GlNECAP1-HA, GlNECAP1-APEX2-2HA or the ΔWVIF deletion construct GlNECAP1ΔWVIF-HA (green) were tested for Dextran-R (red) uptake. Dextran-R signal intensity after uptake was significantly (p<0.05) decreased in GlNECAP1-HA-and GlNECAP1-APEX2-2HAexpressing cells when compared to wild-type controls and GlNECAP1ΔWVIF-HA-expressing cells (box-plot). (D) Quantitative tEM analysis of GlNECAP1-APEX2-2HA-expressing cells (upper panels) and wild-type non-transgenic cells (WT; lower panels) shows visibly enlarged PVs in GlNECAP1-APEX2-2HA-expressing cells, with a statistically significant (p<0,05) increase in the median of measured area of peripheral vacuoles (in μm 2 ; box-plot).
https://doi.org/10.1371/journal.ppat.1008317.g008  system associated to clathrin assemblies (blue) and GlPXD3 coats (red), based on data presented in this report and in [13]. PV membranes and lumina are represented in dark and light green, respectively. Cross-sections at (1) and (2) yield views in the right panel, highlighting foci of clathrin assemblies beneath the PM, above GlPXD3's coat-like deposition pattern surrounding PVs. (D) An overview of the G. lamblia PIP-binding interactome associated to PVs. All represented PIP-binding proteins were found to contact clathrin assemblies (GlCHC) in either reciprocal (double-headed arrows) or one-way (single-headed arrows) modes of interaction, following filtering of co-IP data either at high (black solid lines) or low (grey dashed lines) stringency.
GlPXD1 and 2 are the only PIP-binders who's extended interactomes include the G. lamblia putative clathrin light chain (Fig 4 and S4F Fig), arguably GlCHC's closest binding partner. The GlPXD1 interactome further stands out for enrichment of proteasome-associated components (S1 Table), invoking scenarios concerning clathrin assembly turnover in G. lamblia. Although previous data showed that clathrin assemblies are long-lived stable complexes [19], they would still require remodeling, degradation, and substitution with new components. In the absence of classical components as well as C-terminal motifs on GlCHC for ordered disassembly of clathrin coats, GlPXD1's proteasome-enriched interactome points to proteasomemediated degradation of GlCHC assemblies as an alternative process to achieve turnover albeit without recycling of coat components.
In the context of clathrin assembly dynamics, GlNECAP1 once again comes to the forefront. NECAP1 is characterized as an AP2 interacting partner and an important component of CCVs in the assembly phase [57]. Given that CCVs have not been detected in Giardia, this begs the question of the functional role of a NECAP1 cardiolipin-binding orthologue in G. lamblia, which was found to interact with G. lamblia AP2 subunits and GlFYVE. Recent developments in gene knock-out [67] and CRISPR-Cas9-based knock-down [68,69] methodologies tailored to G. lamblia will be instrumental towards a full functional characterization of GlNE-CAP1's function(s).

PIP binding homeostasis and fluid-phase uptake
We initially hypothesized that perturbation of free PIP binding sites would elicit fluid-phase uptake phenotypes by impacting PV functionality. The hypothesis tested positive for PI3P, PI (3,4,5)P 3 , and PI(4,5)P 2 . A significant effect on cell width was detected when free PI(3)P binding sites were reduced by ectopic expression of 2xFYVE-GFP (Fig 3B and 3F), linking PIP residues to both endocytic homeostasis and overall maintenance of cell size, possibly in connection to membrane turnover. Complementing these data, ectopic expression of both GlFYVE and GlNECAP1 significantly impacted fluid-phase uptake. Furthermore, ectopic expression of GlNECAP1 induced an enlarged PV phenotype similar to that induced by expression of a predicted GTP-locked GlDRP mutant [70]. As with all uptake phenotypes we elicited and measured using fluorescent dextran as a fluid-phase uptake reporter, it is still unclear whether the defect lies in PV-PM fusion or in the sealing-off of PV lumina.
Ectopic expression of a truncated GlFYVE deprived of its ARM repeats, namely NT-GlFYVE, induced formations of vesicle-like structures of undefined origins. ARM folds are superhelical structures mostly involved in protein-protein interactions [71], suggesting that a loss of these domains may impact GlFYVE function and protein complex formation and may lead to protein aggregation. Importantly, these structures are associated to GlCHC which, in cells expressing the NT-GlFYVE recombinant protein, has lost its almost exclusive PV localization. In line with this hypothesis, the NT-GlFYVE epitope-tagged reporter loses association to PVs. In contrast to the GlFYVE-induced uptake phenotype and despite a severe PV clustering phenotype, HA-GlPXD2-expressing cells still appear to perform fluid-phase uptake comparably to wild-type cells. This suggests that PV morphology can be decoupled from effective PVmediated uptake. Taken together, these data link PIPs to clathrin assemblies and fluid-phase PV-mediated uptake, providing new insights on clathrin's hitherto unclear role in Giardia endocytosis.

Beyond clathrin assemblies
Investigation of the molecular milieu within which clathrin-associated PIP-binding proteins operate in G. lamblia revealed two protein sets of special interest. Four predicted SNARE proteins were detected in both the GlPXD2 and GlPXD3 interactomes. Further investigations will be necessary to determine whether the function of this SNARE quartet is indeed fusing PM and PV membranes at contact sites, thereby allowing entry of fluid-phase material into PV organelles.
Another finding of special interest concerns Gl9605, a hitherto unrecognized DLP found in the interactome of GlPXD3 with similarity to bacterial DLPs (BDLPs; S6 Fig). Similar to their eukaryotic counterparts, BDLPs are capable of helical self-assembly and tubulation of lipid bilayers, and were shown to be most closely related to the mitofusins FZO and OPA (S6 Fig) [24,25], but only distantly related to classical dynamins [26]. BDLPs were also found in the Archaea class Methanomicrobia [72], making the family ubiquitously distributed across all kingdoms. These data show how the DLP/DRP family in G. lamblia has now expanded to include the previously unidentified endocytosis-associated Gl9605 BDLP homologue. GlDRP plays a role in the regulation of PV and encystation-specific vesicle (ESV) size [70]. Although its role in fluid-phase uptake has not been determined, expression of a GTP-locked GlDRP mutant inhibited endocytosis of biotinylated surface proteins [70]. On the other hand, a similar mutational analysis of Gl9605 shows that this DLP variant can elicit a dominant-negative fluid-phase uptake phenotype. Although we did not test Gl9605 involvement in surface protein uptake, the data so far suggest that two distinct DLPs play independent albeit complementary roles in the regulation of PV-mediated fluid-phase uptake and organelle homeostasis.
In this work, we report on the detailed functional characterization of PIP-binding proteins in G. lamblia that associate to clathrin assemblies. Our data reveals a previously unappreciated level of complex interplay between lipid residues and their protein binders in marking and shaping endocytic compartments in this parasite. However, several identified PIP-binding modules appear to associate to PVs independently of clathrin. Their extended interactomes and their involvement in fluid-phase uptake have yet to be investigated but current data point towards a complex network of PIP binders of varying binding preference and affinity, all working in the same subcellular environment, yet, in some cases (GlFERM, GlBAR1 and 2, GlPROP1 and 2, and Gl16801), not directly linked to clathrin assemblies. The only known exception is Glepsin whose localization remains controversial due to conflicting reports [21,73]. We systematically did not detect Glepsin in any of the interactomes for clathrin-associated PIP binders, in line with its localization at the ventral disk [21]. Altogether, the variety of PIP residues and PIP-binding modules in the G. lamblia cortical area containing endocytic PVs underscores their necessity for correct functioning of membrane traffic even in a protist so clearly marked by reduction in endomembrane complexity.

Giardia lamblia cell culture, induction of encystation and transfection
G. lamblia WBC6 (ATCC catalog number 50803) trophozoites were cultured and harvested applying standardized protocols [56]. Encystation was induced by the two-step method as previously described [74,75]. Transgenic cell lines were generated using established protocols by electroporation of linearized or circular pPacV-Integ-based plasmid vectors prepared from E. coli as described in [76]. Transgenic lines were then selected for puromycin resistance (final conc. 50 μg ml -1 ). After selection, transgenic trophozoites carrying episomal or integrated reporter constructs were further cultured with or without puromycin, respectively.

Construction of expression vectors
Oligonucleotide sequences used for cloning in this work are listed in S8 Table. pPacV-Integbased [34] expression of epitope tagged reporter constructs was driven using either putative endogenous (pE) or encystation-dependent (pCWP1) promoters. Constructs 2xFYVE-GFP and GFP-P4C [42] were kindly provided by Prof. Dr. H. Hilbi (University of Zurich).

PV labelling using fluid-phase markers
Fluid-phase uptake assay in G. lamblia was performed as described previously [26] using dextran coupled to either Oregon Green 488 (Dextran-OG) (Cat. Nr. D-7171, Thermo Fisher Scientific) or Texas-Red (Dextran-R) (Cat. Nr. D-1863, Thermo Fisher Scientific) fluorophores, both at 1mg/ml final concentration. The same protocol was used following treatment with 7.2mM neomycin (G418, Sigma) in supplemented PBS for 45-50 minutes at 37˚C.

Protein analysis and sample preparation for mass spectrometry (MS)-based protein identification
Protein analysis was performed on 4%/10% polyacrylamide gels under reducing conditions (molecular weight marker Cat. Nr. 26616, Thermo Scientific, Lithuania). Immunoblotting was done as described in [77]. Gels for mass spectrometry (MS) analysis were stained using Instant Blue (Expedeon, Prod. # iSB1L) and destained with ultra-pure water.

Mass Spectrometry, protein identification and data storage
MS-based protein identification of de-stained and diced gel lanes was performed as described in [19]. Free access to raw MS data is provided through the ProteomeXchange Consortium on the PRIDE platform [78]. Accession numbers for datasets derived from bait-specific and corresponding control co-IP MS analyses are the following: PXD013890 for GlPXD1, 3 and 6, PXD013897 for GlFYVE, PXD013896 for GlNECAP1 and PXD013899 for GlPXD2 and 4.

In silico co-immunoprecipitation dataset analysis
Analysis of primary structure and domain architecture of putative components of giardial PIP -binding proteins was performed using the following online tools and databases: SMART for prediction of patterns and functional domains (http://smart.embl-heidelberg.de/), pBLAST for protein homology detection (https://blast.ncbi.nlm.nih.gov/Blast.cgi?PAGE=Proteins), HHPRED for protein homology detection based on Hidden Markov Model (HMM-HMM) comparison (https://toolkit.tuebingen.mpg.de/#/tools/hhpred), PSORTII for sub-cellular localization prediction (https://psort.hgc.jp/form2.html), TMHMM for transmembrane helix prediction (http://www.cbs.dtu.dk/services/TMHMM/), RCSB for 3D structure of homologues (https://www.rcsb.org/), and the Giardia Genome Database to extract organism-specific information such as protein expression levels, predicted molecular sizes and nucleotide/protein sequences (www.giardiaDB.org). The generated co-IP datasets were filtered using a dedicated control-co-IP dataset generated using non-transgenic wild-type parasites. Filtration of the bait-specific co-IP and control-co-IP datasets was done using Scaffold4 (http://www. proteomesoftware.com/products/) with high stringency parameters (95_2_95, FDR 0%) and low stringency parameters (95_2_50, FDR 0%). Furthermore, exclusive hits for bait-specific datasets were manually curated using the following criteria for inclusion into the interactome model: i) exclusive detection with > 3 spectral counts in bait-specific datasets or ii) an enrichment of peptide counts >3 with respect to the ctrl. co-IP dataset. Data presented in S1-S7 Tables show exclusive and non-exclusive protein hits filtered using both stringency levels.

Immunofluorescence analysis (IFA) and light-microscopy
Samples for immunofluorescence analysis of subcellular distribution of reporter proteins by wide-field and laser scanning confocal microscopy (LSCM) were prepared as described previously [33,35]. . Cells were imaged at maximum width, with nuclei and the bare-zone at maximum diameter. Deconvolution was performed with Huygens Professional (Scientific Volume Imaging). Three-dimensional reconstructions and signal overlap quantification (Mander's coefficient) in volume images of reconstructed stacks were performed using IMARIS x64 version 7.7.2 software suite (Bitplane AG) or FIJI [79], respectively.

Live cell microscopy of GFP fusion proteins in transgenic Giardia lamblia
Transgenic trophozoites expressing GFP-fusion proteins and non-transgenic lines were harvested and resuspended in approximately 0.5-1 ml of the medium and transferred to a precooled 24 well culture plate placed directly on ice in an ice bucket. After overnight oxygenation in the dark at 4˚C, cells were washed in ice-cold PBS supplemented (S-PBS) with 5mM glucose (Cat. No. 49139, Fluka) and 0.1mM ascorbic acid (Cat. No. 95209, Fluka) at pH 7.1. An aliquot of cells in S-PBS was placed on a microscopy slide and left to recover at 37˚C for 3 minutes directly before imaging. For the Dextran uptake assay, Dextran-R was added to the cell suspension to a final concentration of 1mg/ml. Cells were incubated in the dark at 37˚C for 20 min and were imaged directly or chemically fixed for further processing by IFA.

Super resolution (gSTED) microscopy
Sample preparation was done as described for wide field microscopy and LSCM. For imaging, samples were mounted in ProLong Gold antifade reagent (Cat. Nr. P36934, Thermo Fisher Scientific). Super resolution microscopy was performed on a LSCM SP8 gSTED 3x Leica (Leica Microsystems) at the Center for Microscopy and Image Analysis, University of Zurich, Switzerland. Nuclear labelling was omitted due to possible interference with the STED laser. Further data processing and three-dimensional reconstructions of image stacks were done as described for LSCM.

Sample preparation for transmission electron microscopy
Transgenic trophozoites expressing GlPXD2 (GL50803_16595) and non-transgenic cells were harvested and analyzed by transmission electron microscopy (tEM) as described previously [70].

Chemical fixation of DAB-stained cells
DAB stained cell suspicions were post-fixed with 1% aqueous OsO4 for 1 hour on ice, subsequently rinsed three rimes with pure water and dehydrated in a sequence of ethanol solutions (70% up to 100%), followed by incubation in 100% propylene oxide and embedding in Epon/ Araldite (Sigma-Aldrich, Buchs, Switzerland). Samples were polymerized at 60˚C for 24h. Thin sections were imaged pre-and post-staining with aqueous uranyl acetate (2%) and Reynolds lead citrate.

Expression and purification of bacterial fusion proteins
For each candidate PIP-binding protein, corresponding nucleotide stretches coding for selected amino acid residues (S9 Table) were modified by including an HA-coding sequence either at the 5' end or the 3' end and then subcloned into the pMal-2Cx E. coli expression vector (New England Biolabs). The resulting recombinant variants were expressed as maltosebinding protein (MBP) fusions in E. coli (strain Bl21) and grown in LB medium either at 37˚C (MBP-GlPXD1, MBP-GlPXD2, MBP-GlPXD3, MBP-GlPXD6, MBP-GlNECAP1 and MBP-GlFYVE) or 30˚C (MBP-GlPXD4 and MBP-GlPXD5) to an OD 600 = 0.4. Induction of expression was performed by adding 0.2 mM IPTG (Isopropyl β-D-1-thiogalactopyranoside, Cat. Nr. 15529019, Thermo Fischer Scientific) to the cultures and incubating for a further 4 hours. Cells were harvested at 4˚C (4,000 x g) and bacterial pellets were resuspended in 5 ml of cold column buffer with 1x PIC (Protease inhibitor cocktail set I; Cat. Nr. 539131-10VL, Merck) and 200 mM PMSF (Cat. Nr. 329-98-6, Sigma Aldrich). Cells were lysed by sonication and centrifuged (20 min, 9,000 x g, 4˚C). Cleared supernatant was incubated with amylose resin slurry (Amylose Resin High Flow, Cat. Nr. E8022L, BioLabs) for 4 hours at 4˚C on a turning wheel, washed with column buffer and then transferred to an empty column (BioRad). Unbound protein was washed using until background OD 280 reached~0.06. Protein fractions were eluted using 10mM maltose solution and pooled for overnight dialysis in a dialysis cassette (Slide-a-Lyzer, Cat. Nr. 66380, Thermo Fischer Scientific) against 25mM NH 4 Ac at 4˚C and later lyophilized or snap-frozen. Protein fractions were stored at -80˚C. floated on ultrapure water for 5 min before incubation in blocking buffer (1xPBS containing 0.1%v/v Tween-20 and 3% fatty-acid free BSA (Sigma A7030)) at RT for 1h. For PLO using lipid strips, E. coli-derived lyophilized were reconstituted in 1x PBS and protein concentration was measured using the Bradford assay. For PLO using PIP arrays, snap-frozen protein samples were used. For both types of PLO assays, the equivalent of 0.5 μg/ml of protein in PBS containing 3% fatty acid free BSA were incubated for 1h at RT with gentle agitation. After washing with 1xPBS containing 0.1% v/v Tween-20, PIP-strips were incubated (1h, RT, agitated) with a monoclonal anti-HA antibody (clone 3F10, monoclonal antibody from Roche) at a dilution of 1:500 in blocking buffer. Subsequently strips were washed and incubated (1h, RT, agitated) with a goat-derived polyclonal anti-rat antibody conjugated to HRP at a dilution of 1:2000 in blocking buffer (Cat. Nr. 3050-05, Southern Biotech). After further washing, strips were developed using a chemiluminescent substrate (WesternBright ECL HRP Substrate, Cat. Nr. K-12045-D50).

Densitometric analysis of lipid strips and arrays
Relative quantification of immunoblotting signal intensity on PIP strips and arrays overlaid with PIP-binding proteins was performed using FIJI [79]. For each strip or array, the spot with the highest pixel number was set as a reference for 100% binding; signals coming from all other spots were normalized against it. The data were visualized as bar charts of relative signal intensity as a measure of lipid-binding preference for each PIP-module.

Identification of Giardia orthologues of known PIP-binding domains
PIP-binding domain representatives were used as bait for in silico searches within the Giardia genome database (GDB) (http://giardiadb.org/) using the online tool HHpred (https://toolkit. tuebingen.mpg.de/) to detect remote giardial homologues using hidden Markov models (HMMs; Table 1) [25]. Outputs were firstly evaluated based on the calculated probability and the corresponding E-value for the prediction, with cut-offs for probability and e-value set to 90 and 1e-10, respectively. Sequence identity and similarity were also considered. To validate the prediction, candidate giardial PIP-binding proteins were then utilized as baits to search PDB databases using HHpred to retrieve orthologous PIP-binding proteins/modules. For additional validation, I-TASSER [32][33][34] was also used to predict hypothetical structures of putative giardial PIP-binding domains next step validation.

Multiple sequence alignment analysis
Multiple sequence alignment using two or more sequences was performed with the Clustal Omega sequence alignment algorithm [80,81]. The sequences used to compile the alignments shown in S1 Fig were chosen based on representative members for each PIP-binding domain type [1,10,82]. Alignments for Figs 6 and 8 were based on previously characterized G1-G4 GTP binding motifs [53] and NECAP1 proteins [57], respectively.

De novo structural modeling and analysis
Ab-initio prediction of hypothetical 3D models presented in S1 Fig was done using I-TASSER [32][33][34]. The best model was chosen based on the C-score predicted by the algorithm. A Cscore is a measure of confidence for a model based on the significance of threading template alignments and the convergence parameters of the structure assembly simulations. It ranges from -5-to 2, with higher C-scores indicating higher confidence. The final 3D structures were displayed using PyMOL (The PyMOL Molecular Graphics System, Version 2.0 Schrödinger, LLC.). The superimposition of Giardia PIP-binding proteins with their closest structural orthologue are based on I-TASSER predictions, with structural similarities expressed by TMscore and RMSD a values. The TM-score is computed based on the C-score and protein length. It ranges from 0 to 1, where 1 indicates a perfect match between two structures. RMSD a is the root mean square deviation between residues that are structurally aligned by TM-align [83]. Specifically for GlBAR1 and 2, the structural overlap analysis was performed by selecting positively-charged residues from previously characterized BAR domains shown to play a role in Table 3. GTPase domain sequences selected for the phylogenetic analysis of Gl9605, a novel Giardia BDLP. For each entry, the origin (species and abbreviation), assigned name (protein), unique identifier (UniProtKB) and amino acid stretch used to reconstruct the tree shown in S6 Fig, are  PIP-binding proteins in Giardia lamblia lipid binding [84]. These were manually superimposed on corresponding residues in the predicted GlBAR1 and 2 structures.

Phylogenetic analysis
Selected sequences of GTPase domains (Table 3) were aligned using Clustal Omega tool. The tree construction was submitted to a PHYLogeny Inference Package (PHYLIP) program [85, 86] using random number generator seed set to 111 and number of bootstrap trials set to 10000. The tree was visualized using the on-line tool iTOL and includes branch lengths as a measure of evolutionary distance [87]. variants used as affinity handles in co-immunoprecipitation experiments identify GlCHC as a strong interaction partner for GlPXD1, 4 and 6. GlPXD1 and 4 further interact with other known clathrin assembly components such as GlCLC, GlAP2 subunits α, β and μ, and GlDRP. GlPXD2, albeit at low stringency, is the only other PXD protein found in all three interactomes. The GlPXD4 interactome includes a putative SNARE protein (5785; [45] while GlPXD6 as an affinity handle pulled down another PIP residue binder, GlFYVE, known to be associated to clathrin assemblies in G. lamblia [19]. (B) The curated extended interactome for GlPXD2 includes all core clathrin assembly components (GlCHC, GlCLC, all GlAP2 subunits, GlDRP [19]) and includes PIP-binders GlPXD4 and GlNECAP1. Three putative SNAREs Gl5785, Gl10013 and Gl14469 were also detected as bona fide GlPXD2 interaction partners, the latter previously detected in the GlPXD4 interactome. (C) Analysis of the extended GlPXD3 interactome using an epitope-tagged variant as affinity handle reveals robust interactions with clathrin assembly components GlCHC, α and β GlAP2 subunits, and GlDRP. Predicted inactive NEK kinase 15411 [48] is similarly associated to clathrin assemblies [19] and further shares proteins Gl16811, Gl87677 and Gl17060 as interaction partners with GlPXD3. Predicted SNARE protein Gl7309, GlNSF (GL50803_1154776) and proteins Gl103709 and Gl9605 are unique GlPXD3 interaction partners. The GlPXD3 interactome is connected to the GlPXD6 circuit both directly and through Gl16717.