Lipid Cooperativity as a General Membrane-Recruitment ... - Cell Press

Resource

Lipid Cooperativity as a General MembraneRecruitment Principle for PH Domains Graphical Abstract

Authors Ivana Vonkova, Antoine-Emmanuel Saliba, Samy Deghou, ..., Jan Ellenberg, Peer Bork, Anne-Claude Gavin

Correspondence [email protected] (P.B.), [email protected] (A.-C.G.)

In Brief Vonkova et al. systematically quantify the lipid-binding properties of 91 pleckstrin homology (PH) domains using a physiological, quantitative, liposome microarray-based assay. The data set reveals that cooperativity between lipids is a key mechanism for membrane recruitment of PH domains.

Highlights d

The lipid-binding properties of 91 pleckstrin homology domains have been quantified

d

Over 10,000 individual protein-lipid interaction experiments have been performed

d

The binding specificity and affinity imply cooperativity between signaling lipids

d

Mutations found in cancer biopsies fine-tune lipid-binding specificities

Vonkova et al., 2015, Cell Reports 12, 1–12 September 1, 2015 ª2015 The Authors http://dx.doi.org/10.1016/j.celrep.2015.07.054

Please cite this article in press as: Vonkova et al., Lipid Cooperativity as a General Membrane-Recruitment Principle for PH Domains, Cell Reports (2015), http://dx.doi.org/10.1016/j.celrep.2015.07.054

Cell Reports

Resource Lipid Cooperativity as a General Membrane-Recruitment Principle for PH Domains Ivana Vonkova,1,7 Antoine-Emmanuel Saliba,1,2,5,7 Samy Deghou,1,7 Kanchan Anand,1 Stefano Ceschia,1 Tobias Doerks,1 Augustinus Galih,1 Karl G. Kugler,1 Kenji Maeda,1 Vladimir Rybin,3 Vera van Noort,1,6 Jan Ellenberg,2 Peer Bork,1,4,* and Anne-Claude Gavin1,4,* 1Structural and Computational Biology Unit, European Molecular Biology Laboratory (EMBL), Meyerhofstrasse 1, 69117 Heidelberg, Germany 2Cell

Biology and Biophysics Unit, EMBL, Meyerhofstrasse 1, 69117 Heidelberg, Germany Expression and Purification Core Facility, EMBL, Meyerhofstrasse 1, 69117 Heidelberg, Germany 4Molecular Medicine Partnership Unit, EMBL, Meyerhofstrasse 1, 69117 Heidelberg, Germany 5Present address: Institute for Molecular Infection Biology, University of Wu ¨ rzburg, Josef-Schneider-Straße 2, 97080 Wu¨rzburg, Germany 6Present address: KU Leuven, Centre of Microbial and Plant Genetics, Kasteelpark Arenberg 22, 3001 Leuven, Belgium 7Co-first author *Correspondence: [email protected] (P.B.), [email protected] (A.-C.G.) http://dx.doi.org/10.1016/j.celrep.2015.07.054 This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). 3Protein

SUMMARY

Many cellular processes involve the recruitment of proteins to specific membranes, which are decorated with distinctive lipids that act as docking sites. The phosphoinositides form signaling hubs, and we examine mechanisms underlying recruitment. We applied a physiological, quantitative, liposome microarray-based assay to measure the membranebinding properties of 91 pleckstrin homology (PH) domains, the most common phosphoinositidebinding target. 10,514 experiments quantified the role of phosphoinositides in membrane recruitment. For most domains examined, the observed binding specificity implied cooperativity with additional signaling lipids. Analyses of PH domains with similar lipid-binding profiles identified a conserved motif, mutations in which—including some found in human cancers—induced discrete changes in binding affinities in vitro and protein mislocalization in vivo. The data set reveals cooperativity as a key mechanism for membrane recruitment and, by enabling the interpretation of disease-associated mutations, suggests avenues for the design of small molecules targeting PH domains. INTRODUCTION A eukaryotic cell usually produces more than 1,000 different lipid species that possess a wide range of structural, physical, and biochemical properties. These lipids have a role in virtually all biological processes (van Meer, 2005) through their extensive, regulated association with other lipids and proteins. The significance of these regulatory circuits is evident from the variety of human disorders arising from altered protein-lipid interactions

(Bayascas et al., 2008; Lindhurst et al., 2011; Zu¨chner et al., 2005; Ko¨berlin et al., 2015), which constitute attractive targets for pharmaceutical drug development (Hussein et al., 2013). Protein-lipid interactions can drive the recruitment of peripheral membrane proteins to specific subcellular membranes and thereby contribute to the organization of many cellular functions. This process involves a group of specialized lipid-binding domains (LBDs) that recognize distinctive membrane features, such as specific lipid species or head groups and membrane curvature or charge (Lemmon, 2008). For example, the seven phosphoinositide species—produced by the reversible phosphorylation of the inositol head group of phosphatidylinositol (PI)—are enriched in distinct organelles (Kutateladze, 2010). They form a molecular signature that defines the different subcellular membranes and which is read by a subgroup of LBDs, namely the phosphoinositide-binding domains (Lemmon, 2003). However, many phosphoinositide-binding domains have remarkably low affinity and specificity—if at all—for individual phosphoinositide species (Dowler et al., 2000; Lemmon, 2008; Yu and Lemmon, 2001; Yu et al., 2004). Their efficient and specific recruitment to subcellular membranes sometimes even occurs by a coincidence-sensing mechanism, implying that cooperative associations with membrane proteins and/or additional signaling lipids such as phosphatidylserine, phosphatidic acid, and sphingolipids may play a role (Anand et al., 2012; Di Paolo and De Camilli, 2006; Gallego et al., 2010; Moravcevic et al., 2012; Burger et al., 2000; Huang et al., 2011; Knight and Falke, 2009; Kutateladze et al., 2004; Lee and Bell, 1991; Lucas and Cho, 2011; Macia et al., 2000; Stahelin et al., 2003, 2004; Ziemba and Falke, 2013). However, only a few studies have so far explored this phenomenon, and a global and unbiased understanding of membrane-recruitment principles—i.e., the elusive ‘‘phosphoinositide code’’—is yet to be achieved. We therefore systematically studied the mechanisms underlying membrane recruitment in a prototypic family of phosphoinositide-binding domains—the pleckstrin homology (PH) domain, which is the most common membrane-targeting motif in eukaryotes. We have developed an approach termed a liposome Cell Reports 12, 1–12, September 1, 2015 ª2015 The Authors 1


microarray-based assay (LiMA) (Saliba et al., 2014) to quantitatively profile the recruitment of 91 PH domains to a large variety of surrogate cellular membranes composed of the main classes of signaling lipids and their systematic combinations. The resulting data represent one of the largest sets of physically quantified protein-lipid interactions to date (data available at http://vm-lux. embl.de/deghou/data/ph-domain/). They reveal some of the basic features that enable PH domains to specifically recognize membranes: that is, the selective recognition of individual phosphoinositide species and the frequent, context-specific tuning mechanisms driven by the presence of additional lipids such as phosphorylated sphingoid long-chain bases (LCBs) and phosphatidylserine, for which rheostasis might represent an important mode of action, contributing to the effective spatiotemporal fine-tuning of cell signaling. RESULTS Analysis of Membrane-Recruitment Mechanisms for 91 PH Yeast Domains Using LiMA We recently developed LiMA, a method that integrates biochemical principles—that is, the assembly of surrogate of biological membranes on a thin agarose layer—with quantitative fluorescence microscopy-based imaging and microfluidics (Saliba et al., 2014). LiMA measures protein recruitment to membranes in a quantitative and multiplexed manner and is thus well suited for charting cooperative binding mechanisms on a large scale. This miniaturized array was further developed to accommodate 122 different types of liposomes, each comprising 26 different signaling lipids present in different combinations and concentrations (Figure 1A; Table S1) in assay buffer (10 mM HEPES [pH 7.5] and 150 mM NaCl) containing no phosphate that could act as competitor of the interactions (see the Supplemental Experimental Procedures). For comparison, we also included seven non-physiological analogs that are synthesized in higher eukaryotes, but not in yeast (Table S1A). These non-physiological lipids in yeast represent interesting controls to assess LBDs binding specificity and can be used as chemical analogs, which means they are included in the final data set—albeit flagged as such—but to avoid confusion, they are not included in the final analysis. As controls, liposomes carrying biotinylated phosphatidylethanolamine (PE) were placed at ten specific positions on the array, and binding to streptavidin-AF488—spiked in each cell extract—served as a general indicator for the quality of the assay (Figure S1A). The different lipid mixtures of phosphoinositide phosphates (PIPs) and other signaling lipids were chosen based on the few examples from the literature demonstrating cooperative-binding mechanisms for a limited number of LBDs (Gallego et al., 2010; Moravcevic et al., 2012; implying both PIPs and glycerophospholipid and PIPs and sphingolipids). Given the paucity of data on the exact local organization and concentration of signaling lipids in membranes in vivo (van den Bogaart et al., 2011), we opted for standard concentrations (5–10 mol %) that have been used for in vitro studies and that represent an approximation of the in vivo situation (Tables S1B and S1C). For example, PS (phosphatidylserine), PA (phosphatidic acid), PE (phosphatidylethanolamine), and PI (phosphatidylinositol) are abundant lipids, and each account for 10 mol % of 2 Cell Reports 12, 1–12, September 1, 2015 ª2015 The Authors

the whole-cell lipidome. This is in contrast to PIPs and sphingolipids, in which overall averaged cellular abundance is generally very low ( 1, that is when the binding affinity to liposomes containing two signaling lipids was stronger than the sum of the binding affinities to liposomes containing the individual lipids. Simple additive interactions (NBIL1+L2 = NBIL1 + NBIL2) were thus not considered. To account for technical variability, NBIL1+L2 also has to be higher than the highest value obtained for NBIL1+NBIL2 among the replicates (Figures 3A, S3A, and S3B). Importantly, as the lipids were routinely analyzed at one concentration—which might already be saturating for some protein-liposome pairs—the data likely provide


Figure 2. Clustering Analysis of Liposomes Based on Similarities in the Recruitment of PH Domains (A) Hierarchical clustering of 90 liposome types composed of different combinations of phosphoinositides and an additional signaling lipid. Each liposome type is colored based on the phosphatidylinositol phosphate (PIP) species present in the mixture. The pie charts show the representation of charges of the auxiliary lipids within indicated subgroups, which correspond to depth of four from the root. # indicates an inactive (in our assay) dipalmitoyl variant of PI(3,5)P2. The statistical significance was calculated on clades where the liposome types containing different PIP were randomly distributed (data not shown). (B) Effect of the auxiliary lipid’s charges on the PH domains-liposomes interactome within the clades highlighted in (A). Only interactions of PH domains that specifically interacted with liposomes containing auxiliary lipids of negative, positive, or neutral charge are shown. (C) Principal coordinates analysis of PH-domain-binding profiles of all liposome types composed of combinations of signaling lipids. Red, liposomes containing organelle (Org) PIPs; blue, liposomes containing plasma membrane (PM) PIPs. MDS, multidimensional scaling; PI(3)P, early endosome; PI(4)P, Golgi, PM, late endosomes/lysosomes; PI(3,5)P2, endosome/lysosome. See also Figure S2.

a lower estimate for the fraction of cooperative binding events (see the high fraction of false negatives in Figure S3B, for which cooperativity was observed at lower lipid concentrations and that we marked with a star). We also observed instances of negative cooperativity (NBIL1+L2 < max{NBIL1;NBIL2}; CI = NBIL1+L2/ max{NBIL1;NBIL2}). However, they were very rare (33/247 = 13.4% of all cooperative events) and might give an estimate of our false-positive rate. Cooperative associations, implying phosphoinositide species and other auxiliary signaling lipids, were observed for the vast majority of the PH domains that bound liposomes (56/60; 93.3%). As a complementary approach, we performed detailed binding studies for 17 randomly selected cooperative interactions and could confirm 14 (82%); for the remaining three, the results were ambiguous (Figures 3B and S3B; protein concentrations are in Table S5B). The binding curves fit a model in which the binding intensity is proportional to the product of the two lipid concentrations, which is consistent with the view that the two lipids cooperate to efficiently recruit PH domains. Importantly, the dose-response experiments show that cooperativity also takes place when the concentration of PI(4,5)P2 is lower than the one used for the initial screen (for example, YRB2, SWH1, and OPY1N; Figures 3B and S3B). Phosphorylated LCBs were the preferred phosphoinositide partners (Figure 3C). They contributed up to two times more frequently to cooperative interactions than other structurally related lipids (Figure 3D). Notably, signaling lipids that predomi-

nantly localize to the plasma membrane (PM), such as phosphatidylserine (PS) (Leventis and Grinstein, 2010) or ceramides (Schneiter et al., 1999), mostly partnered with phosphoinositide species also present at the PM—i.e., PI(4,5)P2, PI(3,4)P2, and PI(3,4,5)P3 (Di Paolo and De Camilli, 2006) (Figure 3C). Overall, this indicates that these cooperative events are specific and apparently restricted to discrete pairs of signaling lipids. Cooperative Mechanisms Fine-Tune the Recruitment of PH Domains to Biological Membranes In Vivo The impact of coincidence sensing on the recruitment to artificial membrane is diverse (Figure 4A; Table S5C). For the 16 PH domains (out of the 54 implied in cooperative associations involving PIPs) that are selective for specific phosphoinositide species, the presence of auxiliary lipids caused significant changes in the phosphoinositide-binding specificity (Figure 4A, class 1). For instance, the PH domain of the kinase AKT1 strongly and specifically binds to liposomes containing either PI(3,4,5)P3 or PI(3,4)P2, yet the presence of LCBs induces a selective, 5fold increase in the affinity for PI(4,5)P2 (Figures 4A and 4B), whereas the affinity to LCBs alone remains unchanged (data not shown). This is reminiscent of the behavior of an oncogenic form of AKT1 that carries an E17K mutation in its PH domain (Landgraf et al., 2008). When probed in our assay, this mutation triggers an increased affinity for PI(4,5)P2 (Figure 4B) in vitro and constitutive targeting to the plasma membrane in vivo (Landgraf Cell Reports 12, 1–12, September 1, 2015 ª2015 The Authors 5


Figure 3. Landscape of Cooperating Lipids in the Targeting of Yeast PH Domains to Membranes (A) Heatmap of cooperative indexes (CIs) calculated for PH domains (columns) and membranes containing combinations of physiological signaling lipids (rows; the wedge indicates low and high lipid concentrations, respectively). Only 50 PH domains with high-confidence CI > 1 for at least one liposome type are shown. RI: reproducibility index. (B) Dose responses measured for the PH-domain interaction with liposomes containing the indicated concentration of signaling lipids. Values are means (n R 2). (C) The summary of the propensity of different driver lipids (PIPs, right) and auxiliary lipids (left) to cooperate based on data shown in (A). Organelle PIPs and PM PIPs: as in Figure 2; ceramides: Cer, Cer1P, phytocer, and dihydrocer; LCB1Ps: S1P, DHS1P, and PHS1P; LCBs: sphingosine, DHS, and PHS. (D) Cooperative interactions with regard to auxiliary lipids encountered. The bar plot gives the proportion of cooperative interactions of all experiments performed for each group of auxiliary lipids. LCB1Ps, LCBs, and ceramides are as in (C). See also Figure S3.

6 Cell Reports 12, 1–12, September 1, 2015 ª2015 The Authors


Figure 4. Cooperativity Fingerprints Reflect Protein Function and Localization (A) Classification of different impacts of lipid cooperation on the membrane affinity of PH domains. The pie chart indicates the proportion of each class in our data set. The bar plots show the NBIs measured for liposomes containing PIPs alone or mixtures of PIPs with auxiliary lipids. The order of auxiliary lipids in the mixtures is (left to right) PS, DHS, PHS, sphingosine, DHS1P, PHS1P, S1P, Cer, Cer1P, dihydrocer, and phytocer. (B) Influence of lipid cooperativity on membrane recruitment of AKT1-PH. Comparison of AKT1-PH wild-type (wt) and E17K NBIs to membranes of various lipid compositions. The NBI values for each AKT1 variant are normalized to the value of PI(3,4)P2 only (relative NBI = 1). Stars indicate high-confidence cooperative interactions. (C) Principal-component analysis of the S. cerevisiae PH domains with at least one CI > 1 (n = 37). Only CI values for liposomes containing PI(4,5)P2 with PS/ DHS1P/PHS1P were considered. The box plots represent the difference between nucleus and non-nucleus groups (bottom, PC1; right, PC2), and CDC42interacting and non-interacting groups (right, PC2). (D) Box plots of the CI values of PI(4,5)P2:PS (top) or PI(4,5)P2:DHS1P/PHS1P (bottom) liposomes calculated for the groups of PH domains defined in (C). (E) Proteins targeted by the same cooperating lipid pairs are functionally related. Histograms show NBIs for PI(4,5)P2 alone or in the presence of cooperating auxiliary lipids (CI > 1). Bars are normalized to the highest value for each individual PH domain. Stars indicate high-confidence cooperative interactions; crosses indicate incomplete data where the cooperativity could not be assessed. (F) Impact of phosphatidylserine and PI(4,5)P2 metabolism on the localization of selected GFP fusions in S. cerevisiae. D CHO1, phosphatidylserine synthase deletion; MSS4ts, thermosensitive mutant of the phosphatidylinositol-4-phosphate 5-kinase. All scale bars represent 3 mm. See also Figure S4.

et al., 2008). Notably, LCBs further increased the affinity of E17K AKT1 for PI(4,5)P2 to levels that are similar to those observed for one of its physiological ligands PI(3,4)P2, suggesting that cooperativity might also in part contribute to E17K AKT1-induced oncogenicity. For the remaining PH domains (38/54; 70.4%) that did not (20/54; 37.1%) or very poorly and non-specifically (18/54; 33.3%) bind to liposomes containing phosphoinositides alone (Yu et al., 2004), the presence of auxiliary lipids increased their affinity for specific phosphoinositide-containing liposomal membranes (Figure 4A, classes 2a and b; Table S5C). This might explain why previous studies, based on the probing of single

phosphoinositide species in vitro, failed to detect efficient PHdomain recruitment to membranes (Yu et al., 2004). For example, the PH domain of BEM3—a membrane-associated GTPaseactivating protein (GAP) for CDC42, a key regulator of polarized cell growth (Aguilar et al., 2006)—is known to poorly (if at all) bind artificial membranes containing only PIPs (Yu et al., 2004). However, it efficiently targets membranes that contain PI(4,5)P2 and auxiliary signaling lipids; e.g., PS (see below). To test the functional relevance of the identified cooperating lipid pairs (i.e., whether they contribute to a physiological membrane code), we related the in vitro binding profiles to Cell Reports 12, 1–12, September 1, 2015 ª2015 The Authors 7


physiologically derived in vivo data. We first made use of annotations on protein-protein interactions and localization provided by STRING (Szklarczyk et al., 2011) and SGD (Nash et al., 2007), respectively. We observed that the proteins targeted by the same cooperating lipid pairs are functionally related and/or colocalize with their recruiting lipids (Figures 4C–4E). For instance, PI(4,5)P2 and PS—two lipids that accumulate at the sites of polarized growth through mechanisms that imply vectorial delivery via secretory vesicles, active transport, and compartmentalized metabolism (Fairn et al., 2011)—most frequently cooperate to recruit PH domains from components of the CDC42 network that also generally localize at the budding sites (BEM2, BEM3, BOI1, BOI2, BUD4, CDC24, CLA4, OSH2, OSH3, SEC3, SKM1, and SWH1; Figures 4C–4E). The cooperativity between PI(4,5)P2 and PS appear specific (Figure S4A). We also used live-cell imaging to determine the effect of perturbation of PS or PI(4,5)P2 metabolism on the cellular localization of five proteins from the CDC42 network, BEM3, BOI1, BOI2, CDC24, and BUD4 fused to GFP (Figure 4F). We observed that both PS and PI(4,5)P2 are required for their association with the sites of bud growth. This is consistent with the view that the two lipids might also cooperate in vivo. As controls, we also tested two proteins that localize at the bud neck but are not part of the CDC42 network, SKG3, and CAF120. The recruitment of SKG3 and CAF120 to both artificial membranes and yeast bud neck was unaffected by the absence of PS (Figure S4B). Similarly, the PH domains present in nuclear proteins (ASK10, a component of RNA polymerase II; NUP2 and YRB2, involved in nucleocytoplasmic transport; PSY2, a DNA-damage checkpoint protein; RTT106, a histone chaperone; and SIP3, a transcription co-factor) are predominantly targeted by combinations of phosphoinositides and phosphorylated LCBs (e.g., DHS1P and/or PHS1P; Figures 4C–4E; Table S5C). This is consistent with recent evidence suggesting that a nuclear pool of these lipids plays a role in various nuclear functions (Lucki and Sewer, 2012; Viiri et al., 2012). Our data thus support the notion that cooperativity is a general and functionally relevant attribute of PH domains that frequently integrates affinity and specificity to expand the lipid code beyond the set of available phosphoinositides. Targeting of PH Domains to Organelle and PM Phosphoinositides Our data set indicates that the specificities of the PH domains encompass all seven phosphoinositide species (Figures 2A, S2A, and S2B) (Dowler et al., 2000). When grouping liposomes according to their PH-domain recruitment profiles, we observed that those containing phosphoinositides known to predominantly localize to the PM (PI(4,5)P2, PI(3,4)P2, and PI(3,4,5)P3) and those containing other phosphoinositides also present in the organelles (PI(3)P, PI(4)P, PI(5)P, and PI(3,5)P2; Hammond et al., 2014; van Meer et al., 2008) form two clusters that determine the propensity of the two types of membranes to recruit PH domains (Figures 2C, S2C, and S2D). By comparing the sequence of all PH domains that bind organelle phosphoinositides (OPs) with those that do not (Figure S5A), we derived a OPbinding consensus motif that comprises four residues located near the known lipid-binding site (b1-b2 loop and b7 strand) and five in other regions (e.g., a carboxy-terminal a helix; Fig8 Cell Reports 12, 1–12, September 1, 2015 ª2015 The Authors

ure 5A: OBM labeled in pink). The OP-binding consensus motif differs from the previously characterized BSM (Figure 5A; KXn(K/R)XR labeled in blue) that is present in all PH domains that bind phosphoinositides (Isakoff et al., 1998; Moravcevic et al., 2012; Park et al., 2008; Yu et al., 2004). A Conserved, Organelle-Phosphoinositide-Binding Motif Is Perturbed in Some Cancer Biopsies Next, we inspected the sequences of 1,205 PH domains annotated in the Uniprot database (UniProt Consortium, 2014) for the presence of this OP-binding motif (Figures 5B and S5B; Table S5A) and found that it is also conserved in higher eukaryotes, including humans. It is frequently—but not exclusively (see below)—present in the PH domains of lipid-transfer proteins that often associate with membrane contact sites specialized in specific lipid metabolism and signaling (Lev, 2010). This includes the PH domains of CERT, FAPP1, OSBP1, OSH2, and SWH1 that were known to bind OPs (Dowler et al., 2000; Levine and Munro, 2002; Roy and Levine, 2004). Importantly, the OP-binding motif is not restricted to lipid-transfer proteins, and 56% of the PH domains with the best OBM scores belong to other protein families (see Figures 5B and S5A; Table S5A). To validate the significance of the motif, we selected 13 PH domains (including ten from humans) for experimental confirmation (Figures 5B–5G; Table S5B). The selection also included four PH domains known to bind OPs (CERT, FAPP1, OSBP1, and SWH1) as controls (Dowler et al., 2000; Levine and Munro, 2002; Roy and Levine, 2004). The PH domains with a high OP-binding score (CERT, FAPP1, OPY1C, OSBP1, OSBP2, OSBPL3, OSPBL7, and SWH1) also mostly retain the basic sequence motif KXn(K/R)XR. These eight domains bound strongly to both organelle and PM phosphoinositides in vitro (Figure 5B), and the five domains we expressed in S. cerevisiae (CERT, FAPP1, OPY1C, OSBP2, and SWH1) colocalized with intracellular membrane structures, including some representing the Golgi (Figures 5C and 5G). By contrast, PH domains with the KXn(K/R)XR signature but with a largely incomplete OP-binding motif (OSBPL10, OSBPL11, and PLCD1) bound only weakly to OPs in vitro, whereas their affinities for PM phosphoinositides was high (Figures 5B and 5C). Finally, PH domains largely lacking both the OP-binding motif and the KXn(K/R)XR signature (OSBPL5 and OSBPL8) did not bind to any artificial membranes tested (Figure 5B). We also introduced mutations to disrupt the OP-binding motif in four PH domains that bind OPs in vitro and in vivo (Figures 5D– 5G and S5C; Table S5B). They were either artificially engineered (CERT-PH R98Q, FAPP1-PH K74Q, and SWH1-PH K360Q) or naturally observed in some human cancer biopsies (FAPP1-PH T9A and OSBP2-PH R262L; Imielinski et al., 2012; Network, 2012). For two domains, SWH1-PH K360Q and OSBP2-PH R262L, we also performed additional biophysical measurements using CD spectroscopy and confirmed that the two mutants were folded and stable under our experimental conditions (Figures S1B and S1C). In all cases, the mutations induced a very selective decrease in the affinity of the PH domain for OPs both in vitro and in vivo, whereas binding to PM phosphoinositides remained largely unaffected. This demonstrates that the OPbinding motif plays an important role in the proper subcellular localization of PH-domain-containing proteins, including families


Figure 5. Characterization of a New Motif for Binding Organelle PIPs (A) PH domain secondary structure representation with the positions of the organelle PIP-binding motif (OBM; pink) and basic sequence motif (BSM; blue). (B) 1,205 PH domains are scored according to the OBM (color range) and BSM (circle size) conservation (left). The pie chart shows gene ontology annotations for the 47 highest-scoring PH domains. ***p < 3.2 3 10 12; **p < 1.1 3 10 6; *p < 0.005. The NBIs of 100 PH domains for either organelle PIPs (pink) or PM PIPs (blue) are shown (right). (C) Intracellular localization of selected PH domains (mCherry-fusion) and trans-Golgi marker, KEX2 (GFP-fusion), in S. cerevisiae. The wedge indicates high and low OBM/BSM motif score. (D and E) The NBIs measured for recruitment of WT and K/R to Q mutants of SWH1 (K360Q), FAPP1 (K74Q), and CERT PH (R98Q) domains (D) or WT and R262L mutant of OSBP2 PH domain (E) to liposomes containing organelle PIPs (pink) or PM PIPs (blue; n = 2). (F) Dose-response recruitment of WT (filled squares) and T9A mutant (open squares) of FAPP1 PH domain to liposomes containing increasing concentrations of PI(4)P (pink) or PI(4,5)P2 (blue; mean ± SD; n = 3). (G) GFP fusions of selected WT and mutated PH domains expressed in a thermo-sensitive PIK1ts S. cerevisiae strain. All scale bars represent 3 mm. See also Figure S5.

Cell Reports 12, 1–12, September 1, 2015 ª2015 The Authors 9


of lipid-transfer proteins. Our data are also able to reveal the consequences of disease-associated mutations on the binding specificity of PH domains, which may contribute to current efforts in the design of small-molecule inhibitors of PH domains (Hussein et al., 2013). DISCUSSION Here, we report a large, systematic, and quantitative data set on the membrane-binding properties of PH domains, one of the most common domains in the human genome and mutations in which underlie several human diseases and syndromes. Our data reveal that the membrane recruitment of PH domains requires the presence of phosphoinositides but also that binding affinity and specificity are frequently determined by the presence of additional, auxiliary, signaling lipids that function as molecular rheostats. The most prominent among these are phosphorylated LCBs and PS, which are all conserved bioactive lipids with elusive mechanisms of actions. They are known to be targeted by only a small number of domains, suggesting that rheostasis might represent their primary, fundamental signaling mode. LCBs and PS, like many other lipids, are heterogeneously distributed within cellular and organelle membranes, and this compartmentalization—implying spatially regulated metabolism, lateral lipid segregation, and active lipid transport—may contribute to the generation of the coincidence signals. Coincidence sensing might work via the presence of a lipid-binding pocket containing two binding sites. Structural and biophysical studies of some prototypic PH domains suggest the presence of additional positively charged binding sites for anionic, auxiliary lipids (Gallego et al., 2010). Otherwise, we can speculate that the auxiliary lipids induce local reorganization of the membrane and the formation of nanoscale PIP-containing lipid domains (van den Bogaart et al., 2011). The emergence of a class of synthetic lipids with a caged head group (Ho¨glinger et al., 2014) will allow to further discriminate between these mechanisms. The frequency of cooperativity and the apparent diversity of the auxiliary lipids involved all point to the importance of such events. These basic membrane recruitment principles, which might also hold for other LBDs (Gallego et al., 2010), may contribute to the spatiotemporal tuning of signaling and thus expand the lipid code beyond the set of available phosphoinositides. Finally, our study can be considered as a proof of principle for the feasibility of comprehensive and systematic analyses of protein recruitment to membranes and indicates that our quantitative screening approach (Saliba et al., 2014) is scalable to entire proteomes and lipidomes. EXPERIMENTAL PROCEDURES LBDs Expression and Preparation of Cell Extracts The lipid-binding domains (LBDs) were expressed as N-terminal His6-SUMO3 and C-terminal sfGFP (Pe´delacq et al., 2006) fusions in E. coli (BL21 STAR; Invitrogen; see Table S5D for sequences of primers). Cell lysis was performed as described previously (Saliba et al., 2014). LiMA Experimental Procedure The fabrication of liposome microarrays, the experimental procedure of protein-liposome interaction assay, the image analysis, and the calculation of


normalized binding intensity (NBI) values were described previously (Saliba et al., 2014). Briefly, liposomes were formed in a buffer with a physiological salt concentration (150 mM NaCl) from lipid mixtures containing various combinations of 26 signaling lipids (Table S1). The liposomes were incubated 20 min in the cell extracts containing sfGFP-tagged proteins. Subsequently, the unbound material was washed and the interactions were monitored by automated microscopy. Fluorescence intensities from pixels matching liposomal membranes were extracted and used for calculation of NBIs. The details of the selection of LBDs, the protocol for the production of recombinant proteins, the detailed composition of all liposome microarrays used, the preparation and imaging of yeast strains, and the computational data analysis are described in the Supplemental Experimental Procedures. SUPPLEMENTAL INFORMATION Supplemental Information includes Supplemental Experimental Procedures, five figures, and five tables and can be found with this article online at http:// dx.doi.org/10.1016/j.celrep.2015.07.054. AUTHOR CONTRIBUTIONS I.V., A.-E.S., P.B., and A.-C.G. designed the research with the expert help of J.E.; I.V., A.-E.S., S.D., S.C., A.G., K.G.K., and V.v.N. conducted the experiments and performed the analysis; I.V. and A.-E.S. did almost all of the experimental work and S.D. did almost all of the bioinformatic analyses presented here; K.M. and V.R. contributed to the biochemical and biophysical protocols; K.A. and T.D. performed the PH-domain prediction; and I.V., A.-E.S., S.D., P.B., and A.-C.G. discussed results and wrote the manuscript with support from all the authors. ACKNOWLEDGMENTS We are grateful to E. Hurt and M. Kaksonen for inspiring comments on the manuscript and to the EMBL Advanced Light Microscopy and the Protein Expression and Purification Core Facilities, G. Zeller, C. Tischer, C. Besir, M.M. Hemberger, and C. Merten for expert help and the sharing of reagents. We thank N. Silva Martin (EMBL) for cDNA from C. thermophilum, J. Holthuis (University of Osnabru¨ck) for providing HA-CERT-pcDNA3.1 plasmid, E.C. Hurt (BZH, Heidelberg University) for pRS315-mCherry vector, and S. Emr (Weill Institute for Cell and Molecular Biology, Cornell University) for pik1ts strain. We thank Y.P. Yuan for providing Information Technology infrastructure. We also thank other members of P.B.’s, J.E.’s, and A.-C.G.’s groups for continuous discussions and support. This work is partially funded by the DFG in the framework of the Cluster of Excellence, CellNetworks Initiative of the University of Heidelberg (ExIni, EcTop) to K.M. A.-E.S. is supported by the European Molecular Biology Laboratory and the EU Marie Curie Actions Interdisciplinary Postdoctoral Cofunded Programme. K.A. thankfully acknowledges support from Marie Curie reintegration grant (ERG). Funding for open access charge was supported by the European Molecular Biology Laboratory. The authors declare competing financial interests in the form of a patent application based on the methods, LiMA, used for this work. Received: July 28, 2014 Revised: June 30, 2015 Accepted: July 27, 2015 Published: August 20, 2015 REFERENCES Aguilar, R.C., Longhi, S.A., Shaw, J.D., Yeh, L.Y., Kim, S., Scho¨n, A., Freire, E., Hsu, A., McCormick, W.K., Watson, H.A., and Wendland, B. (2006). Epsin N-terminal homology domains perform an essential function regulating Cdc42 through binding Cdc42 GTPase-activating proteins. Proc. Natl. Acad. Sci. USA 103, 4116–4121. Amlacher, S., Sarges, P., Flemming, D., van Noort, V., Kunze, R., Devos, D.P., Arumugam, M., Bork, P., and Hurt, E. (2011). Insight into structure and


assembly of the nuclear pore complex by utilizing the genome of a eukaryotic thermophile. Cell 146, 277–289.

Lemmon, M.A. (2003). Phosphoinositide recognition domains. Traffic 4, 201–213.

Anand, K., Maeda, K., and Gavin, A.C. (2012). Structural analyses of the Slm1PH domain demonstrate ligand binding in the non-canonical site. PLoS ONE 7, e36526.

Lemmon, M.A. (2008). Membrane recognition by phospholipid-binding domains. Nat. Rev. Mol. Cell Biol. 9, 99–111.

Bayascas, J.R., Wullschleger, S., Sakamoto, K., Garcıá-Martıńez, J.M., Clacher, C., Komander, D., van Aalten, D.M., Boini, K.M., Lang, F., Lipina, C., et al. (2008). Mutation of the PDK1 PH domain inhibits protein kinase B/ Akt, leading to small size and insulin resistance. Mol. Cell. Biol. 28, 3258–3272. Burger, K.N., Demel, R.A., Schmid, S.L., and de Kruijff, B. (2000). Dynamin is membrane-active: lipid insertion is induced by phosphoinositides and phosphatidic acid. Biochemistry 39, 12485–12493. Di Paolo, G., and De Camilli, P. (2006). Phosphoinositides in cell regulation and membrane dynamics. Nature 443, 651–657. Dowler, S., Currie, R.A., Campbell, D.G., Deak, M., Kular, G., Downes, C.P., and Alessi, D.R. (2000). Identification of pleckstrin-homology-domain-containing proteins with novel phosphoinositide-binding specificities. Biochem. J. 351, 19–31. Fairn, G.D., Hermansson, M., Somerharju, P., and Grinstein, S. (2011). Phosphatidylserine is polarized and required for proper Cdc42 localization and for development of cell polarity. Nat. Cell Biol. 13, 1424–1430. Gallego, O., Betts, M.J., Gvozdenovic-Jeremic, J., Maeda, K., Matetzki, C., Aguilar-Gurrieri, C., Beltran-Alvarez, P., Bonn, S., Fernańdez-Tornero, C., Jensen, L.J., et al. (2010). A systematic screen for protein-lipid interactions in Saccharomyces cerevisiae. Mol. Syst. Biol. 6, 430. Hammond, G.R., Machner, M.P., and Balla, T. (2014). A novel probe for phosphatidylinositol 4-phosphate reveals multiple pools beyond the Golgi. J. Cell Biol. 205, 113–126. Ho¨glinger, D., Nadler, A., and Schultz, C. (2014). Caged lipids as tools for investigating cellular signaling. Biochim. Biophys. Acta 1841, 1085–1096. Huang, B.X., Akbar, M., Kevala, K., and Kim, H.Y. (2011). Phosphatidylserine is a critical modulator for Akt activation. J. Cell Biol. 192, 979–992. Huh, W.K., Falvo, J.V., Gerke, L.C., Carroll, A.S., Howson, R.W., Weissman, J.S., and O’Shea, E.K. (2003). Global analysis of protein localization in budding yeast. Nature 425, 686–691. Hussein, M., Bettio, M., Schmitz, A., Hannam, J.S., Theis, J., Mayer, G., Dosa, S., Gu¨tschow, M., and Famulok, M. (2013). Cyplecksins are covalent inhibitors of the pleckstrin homology domain of cytohesin. Angew. Chem. Int. Ed. Engl. 52, 9529–9533. Imielinski, M., Berger, A.H., Hammerman, P.S., Hernandez, B., Pugh, T.J., Hodis, E., Cho, J., Suh, J., Capelletti, M., Sivachenko, A., et al. (2012). Mapping the hallmarks of lung adenocarcinoma with massively parallel sequencing. Cell 150, 1107–1120. Isakoff, S.J., Cardozo, T., Andreev, J., Li, Z., Ferguson, K.M., Abagyan, R., Lemmon, M.A., Aronheim, A., and Skolnik, E.Y. (1998). Identification and analysis of PH domain-containing targets of phosphatidylinositol 3-kinase using a novel in vivo assay in yeast. EMBO J. 17, 5374–5387. Knight, J.D., and Falke, J.J. (2009). Single-molecule fluorescence studies of a PH domain: new insights into the membrane docking reaction. Biophys. J. 96, 566–582. Ko¨berlin, M.S., Snijder, B., Heinz, L.X., Baumann, C.L., Fauster, A., Vladimer, G.I., Gavin, A.-C., and Superti-Furga, G. (2015). A conserved circular network of coregulated lipids modulates innate immune responses. Cell 162, 170–183. Kutateladze, T.G. (2010). Translation of the phosphoinositide code by PI effectors. Nat. Chem. Biol. 6, 507–513. Kutateladze, T.G., Capelluto, D.G., Ferguson, C.G., Cheever, M.L., Kutateladze, A.G., Prestwich, G.D., and Overduin, M. (2004). Multivalent mechanism of membrane insertion by the FYVE domain. J. Biol. Chem. 279, 3050–3057. Landgraf, K.E., Pilling, C., and Falke, J.J. (2008). Molecular mechanism of an oncogenic mutation that alters membrane targeting: Glu17Lys modifies the PIP lipid specificity of the AKT1 PH domain. Biochemistry 47, 12260–12269. Lee, M.H., and Bell, R.M. (1991). Mechanism of protein kinase C activation by phosphatidylinositol 4,5-bisphosphate. Biochemistry 30, 1041–1049.

Lev, S. (2010). Non-vesicular lipid transport by lipid-transfer proteins and beyond. Nat. Rev. Mol. Cell Biol. 11, 739–750. Leventis, P.A., and Grinstein, S. (2010). The distribution and function of phosphatidylserine in cellular membranes. Annu. Rev. Biophys. 39, 407–427. Levine, T.P., and Munro, S. (2002). Targeting of Golgi-specific pleckstrin homology domains involves both PtdIns 4-kinase-dependent and -independent components. Curr. Biol. 12, 695–704. Lindhurst, M.J., Sapp, J.C., Teer, J.K., Johnston, J.J., Finn, E.M., Peters, K., Turner, J., Cannons, J.L., Bick, D., Blakemore, L., et al. (2011). A mosaic activating mutation in AKT1 associated with the Proteus syndrome. N. Engl. J. Med. 365, 611–619. Lucas, N., and Cho, W. (2011). Phosphatidylserine binding is essential for plasma membrane recruitment and signaling function of 3-phosphoinositidedependent kinase-1. J. Biol. Chem. 286, 41265–41272. Lucki, N.C., and Sewer, M.B. (2012). Nuclear sphingolipid metabolism. Annu. Rev. Physiol. 74, 131–151. Macia, E., Paris, S., and Chabre, M. (2000). Binding of the PH and polybasic C-terminal domains of ARNO to phosphoinositides and to acidic lipids. Biochemistry 39, 5893–5901. Mitra, P., Zhang, Y., Rameh, L.E., Ivshina, M.P., McCollum, D., Nunnari, J.J., Hendricks, G.M., Kerr, M.L., Field, S.J., Cantley, L.C., and Ross, A.H. (2004). A novel phosphatidylinositol(3,4,5)P3 pathway in fission yeast. J. Cell Biol. 166, 205–211. Moravcevic, K., Oxley, C.L., and Lemmon, M.A. (2012). Conditional peripheral membrane proteins: facing up to limited specificity. Structure 20, 15–27. Nash, R., Weng, S., Hitz, B., Balakrishnan, R., Christie, K.R., Costanzo, M.C., Dwight, S.S., Engel, S.R., Fisk, D.G., Hirschman, J.E., et al. (2007). Expanded protein information at SGD: new pages and proteome browser. Nucleic Acids Res. 35, D468–D471. Network, C.G.A.; Cancer Genome Atlas Network (2012). Comprehensive molecular characterization of human colon and rectal cancer. Nature 487, 330–337. Park, W.S., Heo, W.D., Whalen, J.H., O’Rourke, N.A., Bryan, H.M., Meyer, T., and Teruel, M.N. (2008). Comprehensive identification of PIP3-regulated PH domains from C. elegans to H. sapiens by model prediction and live imaging. Mol. Cell 30, 381–392. Pe´delacq, J.D., Cabantous, S., Tran, T., Terwilliger, T.C., and Waldo, G.S. (2006). Engineering and characterization of a superfolder green fluorescent protein. Nat. Biotechnol. 24, 79–88. Roy, A., and Levine, T.P. (2004). Multiple pools of phosphatidylinositol 4-phosphate detected using the pleckstrin homology domain of Osh2p. J. Biol. Chem. 279, 44683–44689. Saliba, A.E., Vonkova, I., Ceschia, S., Findlay, G.M., Maeda, K., Tischer, C., Deghou, S., van Noort, V., Bork, P., Pawson, T., et al. (2014). A quantitative liposome microarray to systematically characterize protein-lipid interactions. Nat. Methods 11, 47–50. Schneiter, R., Bru¨gger, B., Sandhoff, R., Zellnig, G., Leber, A., Lampl, M., Athenstaedt, K., Hrastnik, C., Eder, S., Daum, G., et al. (1999). Electrospray ionization tandem mass spectrometry (ESI-MS/MS) analysis of the lipid molecular species composition of yeast subcellular membranes reveals acyl chainbased sorting/remodeling of distinct molecular species en route to the plasma membrane. J. Cell Biol. 146, 741–754. Stahelin, R.V., Burian, A., Bruzik, K.S., Murray, D., and Cho, W. (2003). Membrane binding mechanisms of the PX domains of NADPH oxidase p40phox and p47phox. J. Biol. Chem. 278, 14469–14479. Stahelin, R.V., Ananthanarayanan, B., Blatner, N.R., Singh, S., Bruzik, K.S., Murray, D., and Cho, W. (2004). Mechanism of membrane binding of the phospholipase D1 PX domain. J. Biol. Chem. 279, 54918–54926.

Cell Reports 12, 1–12, September 1, 2015 ª2015 The Authors 11


Szklarczyk, D., Franceschini, A., Kuhn, M., Simonovic, M., Roth, A., Minguez, P., Doerks, T., Stark, M., Muller, J., Bork, P., et al. (2011). The STRING database in 2011: functional interaction networks of proteins, globally integrated and scored. Nucleic Acids Res. 39, D561–D568. Trajkovic, K., Hsu, C., Chiantia, S., Rajendran, L., Wenzel, D., Wieland, F., Schwille, P., Bru¨gger, B., and Simons, M. (2008). Ceramide triggers budding of exosome vesicles into multivesicular endosomes. Science 319, 1244–1247. UniProt Consortium (2014). Activities at the Universal Protein Resource (UniProt). Nucleic Acids Res. 42, D191–D198. van den Bogaart, G., Meyenberg, K., Risselada, H.J., Amin, H., Willig, K.I., Hubrich, B.E., Dier, M., Hell, S.W., Grubmu¨ller, H., Diederichsen, U., and Jahn, R. (2011). Membrane protein sequestering by ionic protein-lipid interactions. Nature 479, 552–555. van Meer, G. (2005). Cellular lipidomics. EMBO J. 24, 3159–3165. van Meer, G., Voelker, D.R., and Feigenson, G.W. (2008). Membrane lipids: where they are and how they behave. Nat. Rev. Mol. Cell Biol. 9, 112–124.


Viiri, K., Ma¨ki, M., and Lohi, O. (2012). Phosphoinositides as regulators of protein-chromatin interactions. Sci. Signal. 5, pe19. Yu, J.W., and Lemmon, M.A. (2001). All phox homology (PX) domains from Saccharomyces cerevisiae specifically recognize phosphatidylinositol 3-phosphate. J. Biol. Chem. 276, 44179–44184. Yu, J.W., Mendrola, J.M., Audhya, A., Singh, S., Keleti, D., DeWald, D.B., Murray, D., Emr, S.D., and Lemmon, M.A. (2004). Genome-wide analysis of membrane targeting by S. cerevisiae pleckstrin homology domains. Mol. Cell 13, 677–688. Ziemba, B.P., and Falke, J.J. (2013). Lateral diffusion of peripheral membrane proteins on supported lipid bilayers is controlled by the additive frictional drags of (1) bound lipids and (2) protein domains penetrating into the bilayer hydrocarbon core. Chem. Phys. Lipids 172-173, 67–77. Zu¨chner, S., Noureddine, M., Kennerson, M., Verhoeven, K., Claeys, K., De Jonghe, P., Merory, J., Oliveira, S.A., Speer, M.C., Stenger, J.E., et al. (2005). Mutations in the pleckstrin homology domain of dynamin 2 cause dominant intermediate Charcot-Marie-Tooth disease. Nat. Genet. 37, 289–294.

Cell Reports Supplemental Information

Lipid Cooperativity as a General Membrane-Recruitment Principle for PH Domains Ivana Vonkova, Antoine-Emmanuel Saliba, Samy Deghou, Kanchan Anand, Stefano Ceschia, Tobias Doerks, Augustinus Galih, Karl G. Kugler, Kenji Maeda, Vladimir Rybin, Vera van Noort, Jan Ellenberg, Peer Bork, and Anne-Claude Gavin

SUPPLEMENTAL DATA Supplemental figures Figure S1. Assessment of the quality of the dataset and overview of the detected PH domain-liposome interactions (related to Figure 1). A. Control for potential position-dependent artifacts on the liposome array. Each cell of the heatmap corresponds to a position on the liposome array and indicates the logarithmic value of the ratio between the number of experiments of NBI > 0.037 (i.e., interactions) and the total number of experiments measured at the particular coordinate. Squared cells represent the positions of the positive control (liposomes containing a PE-biotin lipid), which remained fixed across all replicates. B. Control for PH domain-sfGFP fusions folding using circular dichroism (CD). C. For each PH domain studied with CD a secondary structure prediction from the CD data are given (column two and three; structured [%] represents a sum of helix, beta and turn elements predictions). The calculated melting temperature is given in column four. D. Determination of NBI cutoff maximizing sensitivity and specificity of LBDsliposome interaction detection. Top, ROC curve and, bottom, precision-recall curve analyses of the NBIs of the screen dataset. The NBI cutoff of 0.037 yields a true positive rate (recall) of 75.2%, a false positive rate of 3.4%, precision of 86% (dashed lines). AUC (area under curve) of ROC curve was 0.95. E. Qualitative reproducibility of the screen. All LBDsliposome experiments were assigned as interaction or no binding according to the mean NBI value calculated from all available replicates (interaction if NBI ≥ 0.037, otherwise no binding). The big pie chart shows reproducibility of these annotations based on NBI annotations of the corresponding replicates. In cases of inconsistent annotation, reproducibility based on visual inspection of the images is shown (small pie charts). F. Determination of the reproducibility index (RI) for each domain-liposome type experiment 1

studied. The upper plot represents the standard error (SENBI) as a function of the NBI (log10 transformed). The lower plot represents the RI as a function of the NBI (log10 transformed). Interactions yield a positive RI while no binding events yield a negative RI. The closer the RI is to 0, the more confident is the datapoint (interaction or no binding). The plot on the right side of the panel represents randomly picked examples of domain-liposome type experiments which yielded different ranges of RI. G. Assessment of the ability of LiMA to recover the specific lipid-binding profile of four positive control LBDs (EEA1-FYVE, HSV2, Lactadherin-C2, p40phox-PX). The numbers behind the lipid names indicate lipid concentration (mol %). H. For the four positive controls shown in panel (G), the boxplots show the NBI and 1/RI for the known specific lipid(s) partners and other lipids and lipid mixtures. The P values show the statistical significance between indicated pairs of boxplots. I. Comparison of dissociation constants (Kd) calculated from titration experiments using LiMA on a set of PH domains with a reference Kds (Ref. Kd). J. Correlation between NBI measurements and literature derived Kds for selected PH domain-lipid combinations. The Pearson correlation coefficient of the presented interactions was 0.74 and associated P = 2.6.10-3. K. Visualization of a relationship between the time delay in the imaging (x axis) and NBI measurements between replicates (y axis). L. Correlation analysis of NBIs measured on the same liposome array imaged twice, once at time 0 (imaging 1) and then two hours later (imaging 2), for PLCD1 (upper plot) and EEA1 (lower plot). M. Correlation analysis of LBDs concentration versus NBIs of all interactions detected in the screen. N. Pairwise correlation analysis of NBIs for LBDs with more than 1.5 fold difference in protein concentration between replicates. The NBIs from the corresponding replicates were compared pairwise (Wilcoxon test followed by Bonferroni correction) and corrected P values, reflecting statistical significance of difference in NBIs, were plotted for each pair. P value 0.05 (solid line) was used as a threshold for statistically significant difference. The domain name together with the concentration (μM) of the two corresponding replicates (Repl.) is given. O. Number 2

of interactions detected per PH domain. The proteins for which data from only one replicate are available are labeled in red, the proteins for which the mean RI of the interactions was ≥ 2 are marked in blue.

3

4

5

Figure S2: Hierarchical clustering of liposome types according to the similarities of their PH domain-binding profiles (related to Figure 2). A. Top, hierarchical clustering of single signalling lipid-containing liposomes according to the similarities in their PH domain-binding profiles. The colours indicate lipid families to which the signalling lipids belong. The lipids that are not physiological in S. cerevisiae are marked with §. Middle, barplot represents the number of PH domains that were recruited to membranes of particular liposome type. Bottom, boxplot giving the NBI values of interacting PH domains. The numbers behind the lipid names indicate lipid concentration (mol %) used for each signalling lipid. B. Detailed view on the hierarchical clustering shown in the Figure 2A. The liposome types composed of combination of signalling lipids are clustered according to their PH domain binding profiles. Right, the number of PH domains interacting with each liposome type and left, the distribution of NBI values are shown. The lipids that are not physiological in S. cerevisiae are marked with §. C. The optimum number of clusters in the hierarchical clustering shown in (B) and Figure2A was decided based on a partitioning around medoids. Two clusters composed of liposome types containing PM PIPs (triangles) or organelle PIPs (circles), respectively, were found. D. The silhouette plot assesses the robustness of the two clusters identified in (C). The average silhouette width of 0.54 indicates that a robust structure has been found. # indicates the dipalmitoyl variant of PI(3,5)P2.

6

7

Figure S3: Cooperation of lipids in the targeting of PH domains to membranes (related to Figure 3). A. Landscape of cooperating lipids in the targeting of mammalian PH domains to liposomal membranes. B. Validation of interactions of PH domains with cooperating lipids using dose response experiments. The figure gives the results of dose response experiments (heatmaps) for 31 selected PH domain-lipids combination pairs and compares them with the results obtained from the PH domain screen (bar plots and RI values). The results are divided into four groups: true positive, true negative, false positive and false negative. Each cell in the heatmaps gives the NBI value (violet) measured for the PH domain interaction with liposomes containing the indicated concentration of signalling lipids. The grey colour indicates missing data. Values are mean (n ≥ 2). The bar plots show mean NBI values (± s.d.) measured in the PH domain screen for each liposome type containing indicated signalling lipids or their combinations (black, NBI ≥ 0.037; white, NBI < 0.037). The concentration of signalling lipids in liposomes used in the PH domain screen was 10 mol %, except the combination of PI(4,5)P2 and DHS1P, for which data from liposomes containing 7 mol % of both lipids are shown. The RI values for each experiment is given under the bar plots. In the group of false negative, the PH domains recognizing cooperating lipids at lower lipid concentrations are marked with a star.

8

9

10

Figure S4: Correlation of cooperativity indices and in vivo validation (related to Figure 4). A, Correlations between CIs calculated for combinations of PI(4,5)P2 with PS and other negatively charged auxiliary lipids (DHS1P, PHS1P, S1P and Cer1P). B, Summary of in vivo experiments of GFP fusions of SKG3 and CAF120 proteins in S. cerevisiae. The metabolism of phosphatidylserine and PI(4,5)P2 was perturbed with CHO1 (phospahtidylserine synthase deletion) and MSS4ts (thermosentsitive mutant of the phosphatidylinositol-4-phosphate 5kinase). Scale bars, 3 μm. The following columns summarize information on the classification as member of CDC42 network (based on STRING database) and on the subcellular localization in bud (based on Yeast GFP fusion localization database). The last two columns show in vitro data for N terminal PH domains of SKG3 (SKG3N) and CAF120 (CAF120N). They indicate if any high confidence detection of PI(4,5)P2 and phosphatidylserine cooperation (as shown in Figure 3A) were detected by giving CI and RI calculated for the interactions with PI(4,5)P2:phosphatidylserine combination.

11

Figure S5: PIP-binding profile of PH domains and details of validation of the organelle PIP binding motif (OBM) (related to Figure 5). A. Heatmap representing the PIP-binding profile of PH domains. The solid line separates groups of PH domains interacting and not interacting with organelle PIPs, respectively, which were used as training sets during identification of the OBM. The barplot on the left shows the mean protein concentration (μM) across the replicates used for each PH domain. The PH domains for which data from only one replicate are available are marked in red. B. OBM (pink) and basic sequence motif (BSM, blue) scoring scheme. Each position in OBM and BSM is associated with a score and is pointed out on the secondary structure scheme of a PH domain. C. Selected segments of PH domains [corresponding to the parts schematically represented above in (B)] from CERT, FAPP1, OSBP2 and SWH1 proteins with the mutated residues highlighted. (*) indicates naturally occurring mutations.

12

13

Supplemental table legends Table S1: List of lipids (related to Figure 1) A. List of lipids (related to Figure 1) 30 different lipids (including 1 fluorescent lipid, 1 PEGylated lipid and 1 biotinylated lipid) have been used in this study. The first column provides the category of the lipids and the second one the abbreviation used in this study [(§) indicates that the lipid is not physiological in yeast]. Common name, systematic name and synonyms are listed in the third, fourth and fifth column, respectively. Cas number is given in the column six. Pubchem substance ID and LIPID MAPS ID (http://www.lipidmaps.org) are indicated in the columns seven and eight. The lipid supplier, the catalog number and the lipid origin (yeast, Saccharomyces cerevisiae) are given in the columns nine, ten and eleven. Lipid solubility is given in the column twelve. The column thirteen lists the overall charge calculated for each lipid using the data from Gallego et al (PMID: 21119626). * and # indicate different variants of PI(4)P and PI(3,5)P2 lipid, respectively. (n/a: not available) B. Current knowledge of in vivo lipid concentration and commonly used lipid concentration in physiological in vitro protein-lipid interaction assay (related to Figure 1) For every lipid species used in this study (column 1), the in vivo probes available in literature are reminded in column 2. In column 3, the in vivo concentration available from lipidomics studies for every lipid used in this study are reported with the associated organism and the Pubmed ID number (PMID). Remark: since sterols are not present, the total of mol% of glycerolipids, glycerophospholipids, glycerophosphoinositol phosphates and sphingolipids 14

does not reach 100%. For every lipid the subcellular enrichment is reported in column 4. Finally, the commonly used in physiological in vitro protein-lipid interaction assay are compiled. n/a: not available, (§) lipids not physiological in yeast. C. List of lipid mixtures (related to Figure 1) In total, 125 lipid mixtures were used for the liposome arrays. In all lipid mixtures, PC is used as a carrier lipid (second column) complemented with up to four signaling lipids (columns three to six). Each lipid mixture contained a PEGylated lipid (column seven) and a fluorescent lipid (column eight). The corresponding molar ratio for each lipid mixture is indicated in the columns labeled mol % (columns nine to fifteen). The lipid abbreviations are as defined in Table S1A. (§) indicates that the lipid is not physiological in S. cerevisiae. # indicates different variant of PI(3,5)P2 lipid.

15

Table S2: Proteins used in the screen (related to Figure 1) In total, 95 LBDs were tested, out of which 91 were PH domains. The first column gives protein names as used in this study. The suffixes _1 or _2 indicate two variants of the same PH domain. The suffixes _A or _B indicate two different proteins from C. thermophilum that are orthologous to the same protein from S. cerevisiae. The second column gives systematic (ORF) names, where applicable. The third and the fourth column contain Uniprot ID and species, respectively. The type of LBD is given in the fifth column. The prediction score for each PH domain derived from SMART database is shown in the sixth column. In case of high SMART e-value, the e-values from BLAST or NCBI CDS are given. The column seven indicates the studies (PMID) that studied or predicted the particular LBD. The eight and ninth columns contain information about amino acid (AA) borders and exact AA sequence of probed LBDs, respectively. Source of the DNA template for all LBDs is indicated in the tenth column. The expression level in E. coli and the solubility are given in the column eleven and twelve, respectively. The column thirteen shows median protein concentration in μM across all replicates. Proteins with more than 1.5 fold difference in protein concentration between at least two replicates are marked (‡). Proteins for which the different protein concentrations had impact on the affinity to liposomes are marked (*). Columns fourteen to twentytwo give information about protein concentration (μM) used in each replicate. The column twentythree displays if the average RI of detected interactions indicates high confidence (0 < RI < 2) and the column twentyfoure gives the cooperativity class. The column twentyfive gives the annotations based on in vivo localization from Yeast GFP fusion localization database (Huh et al., 2003; PMID :14562095). The column twentysix gives the annotations used in Fig. 4C, D and E. CDC42 network annotation was derived from STRING database, nucleus annotation is based on information available in Yeast GFP fusion localization database or GO biological process annotation (marked with #). (n/a: not available; ND: not determined) 16

Table S3: Comprehensive view on the results of all domain-liposomes experiments probed in the systematic analysis of PH domains (related to Figure 2 and 3) Summary of mean NBI values from all domain-liposome experiments tested. The domainliposome experiments with the mean NBI value above the threshold (0.037) are in green. The experiments with the mean NBI value below threshold are in white. Domains are indicated in the first column and liposome types in the first row. The number behind the name of a liposome type indicates concentration (mol %) of the lipid(s) used. (n/a: not available) The worksheet named "RIs" gives the RI values for all domain-liposome experiments tested. The worksheets named "NBIs replicate 1-4" gives all the NBI values measured for each replicate. Table S4: Current knowledge of lipid-binding properties of the LBDs used in this study (related to Figure 1) A. The table shows 45 PH domains that were previously tested for their lipid-binding properties. The first column gives the name of the PH domain as used in this study. The second and third columns indicate if the interaction with lipid was detected in previous studies (refered to with PMID) and in this study with high confidence (0 < RI < 2), respectively. The last column summarizes the match of our data with the literature-based benchmark dataset. B. The table summarizes the current knowledge about the lipid ligand preferences and the protein recruitment to biological membranes of the LBDs tested in this study that were used as a benchmark. The data of lipid ligands were obtained from liposome based/SPR studies or supported by a structure information (Pleckstrin). The first column gives protein names as used in this study. The second column indicates the LBD type that was used in each particular study. The third column refers to the studies (PMID) showing in vivo protein/domain recruitment to the biological membranes detected either by imaging or Ras rescue assay 17

(marked with *). The preferred lipid ligands are listed in the column four. The information about the affinity of the interaction (if available) is given in the fifth column (Kd). The column six indicates the source of the information (PMID). The interactions recapitulated in this study are marked green (interactions with only one replicate available are in pale green), the interactions not recapitulated are marked in red. (n/a: not available)C, The table summarizes the PH domains for which new high confident (0 < RI < 2) interactions were detected. Novel interaction indicates the 34 PH domains that were not previously reported to interact with any membranes. New specificity/mechanism indicates the 26 PH domains for which the interaction with lipids was previously reported and for which we propose additional specificity and/or binding mechanism. The 30 PH domains that did not interact with any liposomes with high confidence in this study are in italics. (n/a: not available)

Table S5: Summary of cooperative interactions detected, BSM/OBM scores, and proteins and primers used for additional experiments (related to Figures 3, 4 and 5, and Experimental Procedures) A. Basic sequence motif (BSM) and Organelle PIP-binding motif (OBM) scores (related to Figure 5) The first column gives the names of protein either using the Uniprot ID or, for the PH domains used in this study, the PH domain name as defined in Table S2. The second and the third column give the OBM and BSM score. The fourth column gives the total score that is the sum of OBM and BSM scores. B. Proteins used in additional experiments (related to Figures 3, 4 and 5) The table summarizes information about proteins used in the additional, validation experiments. The first column gives domain names as used in this study. The second and the 18

third columns give standard and systematic (ORF) protein names where applicable. The fourth and the fifth column indicate Uniprot ID and species, respectively. The amino acid sequence borders of the PH domains and the exact amino acid sequences are in the column six and seven, respectively. The source of the DNA template for the constructs is indicated in the column eight. The columns nine to fifteen summarize information about protein concentration (μM) used in the individual experiments shown in Figure 3B, Figure4B, Figure5B,D,E,F; and FigureS1B,C,I and FigureS3b. C. Summary of the cooperative interactions detected (related to Figure 3 and 4) The PH domains are divided into three groups according to their cooperativity class (class 1, 2a and 2b). The barplots show NBI values measured with liposome types composed of either PIP species alone (yellow bars), or combination of PIPs and auxiliary lipids (colour/white bars, order indicated in the legend in the figure, from left to right). The colour bars indicate cooperative interactions - the combinations composed of lipids physiological in S. cerevisiae in blue, the combinations composed of at least one lipid non-physiological in S. cerevisiae in green, the incomplete data for which the cooperativity could not be assessed are in gray, and the white bars indicate non-cooperative interactions or no binding. The horizontal line indicates the threshold (NBI = 0.037). Only data of PH domains with at least one cooperativity event of physiological lipids are shown. (§) indicates that the lipid is not physiological in S. cerevisiae, # indicates different variants of PI(3,5)P2 lipid. D. Primers used for cloning (related to Experimental Procedures) The summary of primers used for cloning of all domains and full length proteins. The first column indicates the domain name. The second and the third column give the sequences of forward and reverse primers, respectively. The lower case letters represent the specific sequence matching the template DNA. The upper case letters indicate extra nucleotides 19

needed for cloning (red - restriction site; blue - additional nucleotides added to secure proper function of restriction enzymes). The fourth column indicates the final vector where the cloned genes were inserted to.

20

SUPPLEMENTAL EXPERIMENTAL PROCEDURES Selection of Lipid Binding Domains (LBDs) The protein candidates were identified using Hidden Markov Models (HMMs) of PH domains from the Pfam database (http://pfam.sanger.ac.uk/) to search against S. cerevisiae and C. thermophilum proteomes. The candidate list was supplemented with additional S. cerevisiae and mammalian PH domains found in the literature. The prediction of amino acid sequences of PH domains was refined by secondary (Psipred; bioinf.cs.ucl.ac.uk/psipred/) and 3D structure (Phyre; http://www.sbg.bio.ic.ac.uk/~phyre2/html/page.cgi?id=index) prediction modeling, which enabled us to set borders of each PH domain more precisely. In cases of ambiguous predictions of boundaries, multiple variants of particular domain were selected. The amino acid sequences of all PH domains were submitted to SMART database (smart.embl-heidelberg.de/) and NCBI CDS (Conserved Domain Search; www.ncbi.nlm.nih.gov/Structure/cdd/wrpsb.cgi) to obtain the e-value score ( Table S2). Recombinant protein expression DNA frgments encoding 74 LBDs were codon optimized for expression in E. coli and synthesized (Entelechon). Sequences coding remaining LBDs were cloned either from S. cerevisiae genomic DNA (14) or cDNA isolated from C. thermophilum (5) (kind gift of N. Silva Martin, EMBL). Full length sequence of HSV2 was cloned from S. cerevisiae genomic DNA. C2 domain of lactadherin was cloned from p416-GFP-Lact-C2 vector (Haematologic technologies). PH domain of CERT was cloned from HA-CERT-pcDNA3.1 plasmid (kind gift of J. Holthuis, University of Osnabrück), FAPP1 PH domain was cloned from PHFAPP1pBGP plasmid (Levine and Munro, 2002). PH domains of OSBP proteins (OSBP1, OSBP2, OSBPL3, OSBPL7, OSBPL8, OSBPL10 and OSBPL11) were cloned from human cDNA (Openbiosystems, Table S5B), except PH domain of OSBPL5 that was cloned from 21

synthetic gene (Entelechon). For sequence details of all primers see Table S5D. Single point mutations were introduced by the QuikChange Lightning Site-Directed Mutagenesis Kit (Agilent Technologies). All sequences were cloned into pETM11 vector and proteins were expressed as N-terminal His6-SUMO3 and C-terminal sfGFP fusions in E. coli (BL21 STAR, Invitrogen) (Saliba et al., 2014). Proteins were produced in 5 mL cultures in 24 deep-well plates in auto-inducing ZY media. Cells were grown at 37 °C up to OD600 ~2, subsequently the temperature was shifted to 15 °C (synthetic genes) or 18 °C (cloned genes) and proteins were produced o/n for 14-15 h. Cells were pelleted at 3,000 g for 20 min and washed in cold PBS. Final pellets (volume ~100 μL) were snap frozen in liquid nitrogen and stored at -80 °C for further use. Preparation of cell extracts Cell lysis was performed as described previously (Saliba et al., 2014). Expression level and protein solubility were evaluated on a Coomassie stained gel and a western blot with anti-GFP antibody (Miltenyi Biotec GmbH, MACS Molecular). Fluorescence intensity of sfGFP-tagged protein in the cell extracts was measured on a microplate reader (BioTek) at excitation and emission wavelengths of 485 nm and 528 nm, respectively. Concentrations of sfGFP-fusions in the cell extracts were estimated from the fluorescence intensity as described previously (Saliba et al., 2014). Subsequently, a small amount (final concentration 50 μg/ml) of streptavidin-AF488 (Life Technologies) was spiked into each cell extract. Cell extracts were centrifuged for 10 min at 16,000 g at 4°C prior use. Protein purification Protein purification was performed as described previously (Saliba et al., 2014). For the titration experiments and the circular dichroism (CD) spectroscopy analysis, the gel filtration

22

elution buffer (10 mM HEPES pH 7.5, 250 mM NaCl) was changed for the assay buffer (10 mM HEPES pH 7.5, 150 mM NaCl) via an overnight dialysis. Circular dichroism (CD) spectroscopy CD spectra were collected on a Jasco-815 spectropolarimeter in a quartz cuvette (2 mm path length) at 20°C with protein concentrations in the range of 4.5-7.0 μM ( Table S5B) in the assay buffer (10 mM HEPES pH 7.5, 150 mM NaCl). Secondary structure analysis was performed using Jasco Spectra Manager software package. Thermal denaturation experiments were performed by monitoring the CD signal at 222 nm while heating protein solutions from 20°C to 90°C with a constant rate 1°C per minute. Liposome microarrays preparation Fabrication and principle of the liposome microarray-based assay (LiMA) have been previously described (Saliba et al., 2014). All lipid mixtures were prepared in chloroform based solvent ( Table S1A) and stored in 1.5 ml glass vials (Sigma) under argon at −20 °C. The final concentration of each lipid mixture was 0.26 mM. Each lipid mixture was composed of a carrier lipid (phosphatidylcholine PC, up to 95 mol %), a signalling lipid (up to 10 mol %) or a combination of signaling lipids (up to 10 mol %), and completed with phosphatidylethanolamine (PE)-Atto 647 (0.1 mol %) to ensure autofocusing during automated microscopy, and PE-PEG350 (0.5 mol %) that facilitates the generation of liposomes. The signalling lipids consisted of phosphoinositides, sphingolipids, glycerophospholipids and DAG. Altogether 122 different lipid mixtures were used ( Table S1C). For phosphatidylinositol PI(3,5)P2 two variants differing in fatty acyl chains were probed - dioleyl [DOPI(3,5)P2] and dipalmitoyl [DPPI(3,5)P2] ( Tables S1). Unlike liposomes containing DOPI(3,5)P2, the liposomes containing DPPI(3,5)P2 were not recognized by the specific PI(3,5)P2 sensor Hsv2 protein. The interactions with DOPI(3,5)P2 represented the 23

majority of binding events reported for PI(3,5)P2-containing liposomes, therefore only those were considered for further analysis and only DOPI(3,5)P2 was used in validation experiments. A thin agarose layer (TAL) of 250 nm height was prepared by coating coverslips (30x45 mm, #1; Menzel) with 1% low melting agarose (Sigma) in water. Lipid mixtures were spotted on the TAL by a syringe-driven spotter (Automatic TLC-spotter4, Camag). Each spot on the array had 800 μm x 800 μm with a distance between two spots of 200 μm (the overall spot density was 100 spots per 1 cm2). The array consisted of 120 spots in 4 groups of 30 spots on area of 10 mm x 18 mm. A polydimethylsiloxane (PDMS) device containing 4 separate chambers was then bonded onto the array as described before (Saliba et al., 2014). Assembled microfluidic devices were stored at 4 °C under inert atmosphere before use. The impact of mutations in the organelle PIP binding motif (OBM) was tested on arrays containing plasma membrane (PM) PIPs [PI(3,4)P2 and PI(4,5)P2] and organelle PIPs [PI(3)P, PI(4)P, PI(5)P and PI(3,5)P2] of two different concentrations, 5 and 10 mol % (Figures 5D and E). For the dose response experiments, arrays containing liposomes with increasing concentrations (from 0 to 7 mol %) of PI(4)P* (porcine brain extract; Table S1A) and PI(4,5)P2 were used (Figure 5F). For dose response experiments for validation of cooperative interactions, the liposome arrays containing liposomes of increasing concentrations (0, 3, 6 and 10 mol %) of two lipids [PI(4,5)P2/ PI(3,5)P2 and DHS1P/ PHS1P/ PS/DHS/PHS/Phytocer] were used (Figure 3B and Figure S3B). For combination of PI(4,5)P2 and DHS1P only concentration up to 6 mol % were used for PI(4,5)P2.

24

For titration experiments for Kd calculations, the liposome arrays containing various PIPs [PI(3)P, PI(4)P*, PI(3,5)P2, PI(3,4)P2 and PI(4,5)P2] with a concentration of 3 mol % were used ( Figure S1I). Measurement of membrane recruitment using LiMA Giant liposomes (> 5 μm) were spontaneously formed within 5 minutes after injection of the assay buffer (10 mM Hepes, 150 mM NaCl, pH 7.4) into the microfluidic chambers. Successful formation of liposomes was confirmed by microscopy. Each liposome spot was automatically imaged in two fluorescence channels, Atto 647 (3 ms exposure time) for monitoring the presence of liposomes, and GFP (100 ms exposure time) to check for possible autofluorescence. After acquisition of the blank images, 20 μL (= volume of one chamber) of cell extracts were injected in each chamber using a syringe pump (KD scientific). After 20 minutes incubation, unbound proteins were washed by 4 chamber volumes (80 μL) of assay buffer. Subsequently, the microfluidic device was coupled to the microscope. Images of every single liposome spot were acquired automatically. Each protein domain was tested on the array in multiple replicates. The position of the liposome spots on the array was shuffled for each replicate to control for potential position bias. On each array ten liposome spots containing lipid mixture with PE-biotin were placed on different positions providing a specific ligand for streptavidin-AF488 (Figure S1A). Positive signal from these control liposomes served as assurance of a successful experimental protocol. Image acquisition and image processing Images were acquired using the same equipment and settings as described previously (Saliba et al., 2014). One constant exposure time of 3 ms was used for Atto 647 images, representing the position of liposome membranes, and exposure times of 5 ms, 10 ms, 30 ms, 50 ms, 75 ms 25

and 100 ms were used for GFP images, representing protein-liposome binding events. Multiple exposure times for GFP were selected in order to capture a broad range of binding intensities of tested domains. Three liposome arrays (3x 120 spots) were always prepared in parallel and imaged in a sequence one after another. The whole procedure of imaging of the three arrays took about 3 hours altogether. Images were processed with the same software and settings as described previously (Saliba et al., 2014). Normalized binding intensity (NBI) values were calculated as described previously (Saliba et al., 2014). Titration experiments for Kd calculations The purified sfGFP fusions of PH domains of Boi2, Swh1 and AKT1 were diluted in the assay buffer (10 mM HEPES pH 7.5, 150 mM NaCl) to reach the following protein concentrations (BOI2: 0.2, 0.5, 1, 3, and 7 μM; AKT1: 0.05, 0.1, 0.2, 0.5, 1, 2, 3, 4, and 6 μM; SWH1: 0.1, 0.2, 0.5, 1, 1.5, 2, 3, 5, and 7 μM). Subsequently, all these dilutions were separately probed on arrays made of various PIPs-containing liposomes [PI(3)P, PI(4)P*, PI(3,5)P2, PI(3,4)P2 and PI(4,5)P2] with a constant PIPs concentration of 3 mol % and NBI values were measured for all interactions (see below). The Kd calculation was performed using a non-linear regression analysis with Origin 7.5 software. Assessment of the relationship between the time delay in the imaging and NBI measurements The liposome arrays were automatically imaged in the same way, starting from the initial position (top left, A1) and the end position (bottom right, J12) position ( Figure S1A). Since the positions of the individual lipid mixtures were reshuffled between replicates, each interaction between a particular domain with a liposome type can be assigned to a certain ‘time delay in imaging’ (ie. the time needed for the automatic microscope to reach the 26

position the specific liposome spot starting from the initial position of the array). We have then calculated the time delay (ie. the difference in ‘time delay in imaging’) between the respective replicates and correlated these delays with the differences in corresponding NBIs measured (Figure S1K). To further assess possible bias introduced by time delay in imaging, we also performed two experiments (using PLCD1 and EEA1 proteins) in which we imaged the same liposome arrays twice, once at time 0 and once two hours later, and then we correlated the NBIs obtained during the both time points (Figure S1L). No decrease in NBIs was detected for the time ranges tested using for the data collection. ROC curve-based evaluation of interactions In order to use only the high quality data for analysis, all acquired images were visually inspected and bad quality images (no liposomes formed, unfocused images and/or presence of protein precipitates) were removed. The images were manually annotated as “interaction”, “no binding” or “dubious”. We performed a ROC analysis (package ROCR (Sing et al., 2005)) using the manual annotation in order to extract an NBI threshold that was subsequently used as interaction predictor. The threshold was set to NBI value 0.037 (interactions with NBI ≥ 0.037, otherwise no binding), leading to 75.2% true positive rate and 3.4% false positive rate and a precision of 86%, AUC of ROC curve = 0.95 (Figure S1B). The final NBI value for each domain-liposome experiment was calculated as a mean NBI from all available replicates. The mean NBI values were used for further analysis. Assessment of effect of protein concentration and positional bias NBI values for all images manually annotated as interaction were collected for all replicates of domains with more than 1.5 fold difference in protein concentration between replicates. Wilcoxon test followed by Bonferroni correction (R package stats) were performed to detect 27

significant differences between measured NBIs (threshold was set to P < 0.05 after correction). In order to look for position bias, we computed the logarithmic value of the ratio between number of data points of NBI > 0.037 (i.e., interactions) and the total number of data points measured at the particular coordinate of the array. Assessment of data reproducibility The quantitative reproducibility of the data was computed by calculating the Pearson correlation between the NBIs of the corresponding replicates (Figure 1B). The qualitative reproducibility (Figure S1E) was assessed by comparing the annotations of protein-liposome experiments based on both, NBI threshold and manual annotation. Each domain-liposome experiment was assigned as interaction or no binding based on the mean NBI value. Subsequently, the annotations of all individual replicates were compared. For the cases of inconsistent annotation based on NBI, the manual annotation was used. Comparison of S. cerevisiae and C. thermophilum binding profiles To assess the similarity of the binding profiles of 27 pairs of S. cerevisiae and C. thermophilum PH domains orthologs, we computed a Pearson correlation coefficient of the NBIs measured on the entire dataset (Figure 1C). Clustering analysis The uncentered Pearson correlation was used to cluster the liposome types according to the similarities between their PH domain binding profiles (i.e., what PH domains interacted with them and with what affinity) using the log(NBI) values of base 1. The liposomes were clustered using the hclust function from the R package stats. The dissimilarity matrix was computed using the Dist function from the amap package. The same approach was used for 28

clustering liposomes containing single signalling lipids and liposomes containing combination of signalling lipids. Fisher test (R package stats) was used for assessment of statistical significance of differences in auxiliary lipid charge distribution between individual subgroups (Figure 2A). To further validate the separation of clusters of plasma membrane PIPs and organelle PIPs observed in the hierarchical clustering analysis, we performed a Principal Coordinates Analysis (PCoA) and represented the two first dimensions, MDS1 and MDS2, of the two different clusters in boxplots (Figure 2C). The PCoA was performed using the implementation of the R package Vegan (Dixon, 2003). Subsequently, we performed a Wilcoxon test between those two dimensions (MDS1 and MDS2). We demonstrated further the robustness of these clusters using the partitioning around medoids (Hennig, 2010) (Figure S2C) and the silhouette coefficient (Rousseeuw, 1987) (Figure S2D). Calculation of the Reproducibility Index (RI) The reproducibility index (RI) is computed for each interaction that was measured at least twice. For each interaction between a protein domain and a liposome type, we calculated the mean NBI of the replicates (NBI) as well as the associated standard error (SENBI). We then calculated the RI of the interaction as such, where Th is the binding threshold (0.037):

If NBI > Th then

If NBI < Th then

Based on the RI calculation, the experiments have been assigned to four categories: high confidence interactions (RI ]0;2[), high confidence no binding (RI ]-2;0[), low confidence interactions (RI ≥ 2) and low confidence no binding (RI ≤ -2). 29

Cooperativity assessment To evaluate the role of cooperative associations or rheostasis (i.e. the changes in binding affinity for one lipid owing to the presence of another lipid) in the recruitment of PH domains to phosphoinositide-containing membranes, we derived a cooperativity index (CI = NBIL1+L2 / [NBIL1+NBIL2]). An interaction was considered cooperative when CI > 1, that is when the binding affinity to liposomes containing two signaling lipids was stronger than the sum of the binding affinities to liposomes containing the individual lipids (NBIL1+L2 > NBIL1+NBIL2 and, to account for technical variability, NBIL1+L2 also has to be higher than the highest value obtained for NBIL1+NBIL2 amongst the replicates). We also observed instances of negative cooperativity (NBIL1+L2 < max{NBIL1;NBIL2}; CI = NBIL1+L2 / max{NBIL1;NBIL2} and, to account for technical variability, NBIL1+L2 also has to be lower than the lowest value obtained for NBIL1 or NBIL2 amongst the replicates). The dose response experiments of 31 randomly selected PH domain-lipids combination pairs were performed on liposome arrays containing liposomes of increasing concentrations of two lipids (Figure S3B). The PH domains were tested at the same protein concentration as used in the PH domain screen where the cooperative/non-cooperative interactions were observed. The screen results (“predictions”) along with the validation results (the “truth”) were used to estimate a true positive rate and false negative rate of the cooperativity heatmap. The PH domains were separated into three cooperativity classes (Figure 4A) based on manual inspection of data represented in plots in Table S5C.

Principal component analysis of proteins based on the cooperativity index values Principal component analysis (PCA) was performed using the R package ade4 (Dray and Dufour, 2007). The PCA was performed on a set of S. cerevisiae PH domains with CI >1 for 30

at least one liposome type (total 37 PH domains). Only CI values obtained for liposome types containing PI(4,5)P2:PS (10:10 mol %), PI(4,5)P2:DHS1P (10:10 mol %), or PI(4,5)P2:PHS1P were considered. The PH domains were annotated using the information of the corresponding full length proteins available in STRING database (Franceschini et al., 2013) and Yeast GFP fusion localization database (Huh et al., 2003) (Table S2). From the STRING database, proteinprotein interactions with a high experimental score (>0.8) were considered. With this filter, the protein CDC42 was identified as the common and specific interactor of the majority of proteins containing PH domains of high PC2 values (> quantile 75%) as defined by the PCA, which correspond to high CI for PI(4,5)P2:PS liposomes. In order to make sense of one outliner (BUD4), we subsequently refined our filter and considered also interactions whose STRING text mining score was superior to 0.5. The refined filter identified only the BUD4 (score 0.649) as the additional CDC42 interactor. The nuclear localization annotation was derived from Yeast GFP fusion database (nucleus and nuclear periphery annotations considered) and supplemented with proteins manually curated as being involved in positive regulation of transcription from RNA polymerase II promoter in Gene Ontology biological process annotations (http://www.yeastgenome.org). Search of the organelle PIP binding motifs (OBM) We first defined two groups of PH domains - the PH domains with ability to interact with organelle PIPs (16) represented the foreground set, and all remaining PH domains (interacting with PM PIPs only, or not interacting with any individual PIPs at all) represented the background set (74) (Figure S5A). All PH domain sequences from both sets were aligned (hmmer version 3.0) against the Pfam PH domain seed HMM. Two PH domains were removed from the foreground set - AVO1 PH domain (high SMART e-value prediction 31

score), and PDK1 PH domain (not aligned properly with the other PH domains). The multiharmony algorithm (Brandt et al., 2010) was then run on the multiple alignments of the remaining sequences (14 PH domains of the foreground set and 74 PH domains of the background set). The algorithm computes how the amino acids content of one group at one position differs from the amino acids content of the other group and thus identify residues/positions in the alignment that are specific for one group. The algorithm proposed 17 best scoring positions. Out of these 17 positions we have selected nine that: (i) showed no overlap in overrepresented amino acids in both, the foreground and the background, and (ii) were located inside the predicted PH domain secondary structure. The sequences of 90 PH domains tested in this study and 1,115 PH domains extracted from Uniprot database (http://www.uniprot.org/; Uniprot release 2013_11) were ranked according to a scoring scheme which was decided manually and works as follows: 1) conservation of OBM residues - 3 points for the two positions of lowest variability (T in the β1-β2 loop and K in the β7-strand) and 1 point for the other seven. The maximal score for OBM was 13. 2) conservation of BSM (basic sequence motif) residues - 1 point for each. The maximal score for BSM was 3. When summed, OBM and BSM could give a maximum score of 16 (Figure S5B, Table S5A). The statistical significance of the difference between NBI values measured for interactions of PH domains of different scores with liposomes containing either PM, or organelle PIPs was calculated using the Wilcoxon test. The glutamine residue used for lysine/arginine substitution in PH domains with engineered mutation was selected based on the PH

32

consensus sequence coming from Pfam database (based on 163 PH domains) (Levine and Munro, 2002). Naturally occurring variants in OBM were searched in COSMIC database (Forbes et al., 2010) (http://cancer.sanger.ac.uk/cancergenome/projects/cosmic/). The statistical significance of the differences in NBI values measured for wild types and mutated forms of PH domains were calculated using the Wilcoxon test. Yeast strains Yeast strains with deleted or depleted lipid metabolizing genes (CHO1, MSS4ts) were created in the SGA Y7039 strain (a gift from C. Boone, University of Toronto) derived from the BY4741 background (Tong and Boone, 2007) (MAT can1:: STE2pr-LEU2 lyp1 ura30 leu20 his31 met150) by standard yeast molecular biology procedures (Janke et al., 2004). These mutant strains were mated with selected yeast strains from Yeast GFP Clone Collection (Invitrogen) (BEM3, BOI1, BOI2, CDC24, CLA4) using robot facilitated mating, sporulation and strain selection of the standard SGA protocols (Huh et al., 2003; Tong and Boone, 2007). The genotypes of the final strains were confirmed by PCR. CERT-PH, FAPP1-PH, OSBP2-PH, OPY1C-PH and PLCD1-PH domains and their mutants were cloned into pRS315-mCherry vector (kind gift of E.C. Hurt, BZH, Heidelberg). SWH1PH domain was cloned from yeast genomic DNA (amino acid sequence as for E. coli expression) (Table S5D) and then inserted into the pRS315-mCherry vector. Single point mutation was introduced by the QuikChange Lightning Site-Directed Mutagenesis Kit (Agilent Technologies). All domains were expressed with C-terminal mCherry tag from ADH1 promotor. For colocalization experiments a yeast strain with GFP-tagged trans-Golgi marker KEX2 (Yeast GFP Clone Collection) was used. Effect of lowered PI(4)P level in Golgi membranes was tested with thermosensitive PIK1ts strain (Audhya et al., 2000) (kind gift of S. Emr, Weill Institute for Cell and Molecular Biology, Cornell University, Ithaca)

33

[MATa leu2-3,112 ura3-52 his3-D200 trp1-D901 lys2-801 suc2-D9, pik1::HIS3, harboring pRS314pik1-83 (Amp, TRP1 CEN6 pik1-83)]. Imaging of yeast strains Yeast strains harbouring mCherry-tagged PH domains (KEX2-GFP and PIK1ts strains) were grown in SD medium without tryptophan overnight at 30 °C (KEX2-GFP) or 25 °C (PIK1ts). Cells were diluted to OD600 < 0.05 and further grown for ~ 3 h in respective temperature. The cells were adhered on glass-slides (BioTek) coated with Concavalin A (Sigma) and imaged. The PIK1ts strains tested at nonpermissive temperature were incubated at 37 °C for 50 minutes prior imaging. Images were acquired with an Olympus IX-81 microscope with ×100 oil objective /NA 1.45 objective lens and Orca-ER camera (Hamamatsu). For live-cell imaging of yeast strains harbouring both GFP-fused genes and deletions/depletions of genes of lipid metabolic enzymes cells were inoculated in SD medium without tryptophan and histidine, supplemented in addition with 1.0 mM ethanolamine in case of cho1, and grown overnight at 30 °C. Cells were diluted to OD600 = 0.1, adhered on Concanavalin A-coated 96 well glass bottom plates, and imaged on fully motorized Olympus fluorescence microscope system (Olympus IX-81) at 30 °C (temperature controlled incubator, EMBL manufacture). The mss4ts strains tested at nonpermissive temperature were incubated at 37 °C for 50 minutes prior imaging. Images were acquired using a ×100 /NA 1.45, low noise highly sensitive ORCA-R camera (Hamamastu), MT20 illumination system, and Uniblitz Electro-Programmable Shutter system. All acquired images were processed with ImageJ software.

34

SUPPLEMENTAL REFERENCES Audhya, A., Foti, M., and Emr, S.D. (2000). Distinct roles for the yeast phosphatidylinositol 4-kinases, Stt4p and Pik1p, in secretion, cell growth, and organelle membrane dynamics. Molecular biology of the cell 11, 2673-2689. Brandt, B.W., Feenstra, K.A., and Heringa, J. (2010). Multi-Harmony: detecting functional specificity from sequence alignment. Nucleic Acids Res 38, W35-40. Dixon, P. (2003). VEGAN, a package of R functions for community ecology. Journal of Vegetation Science 14, 927-930. Dray, S., and Dufour, A.-B. (2007). The ade4 package: implementing the duality diagram for ecologists. Journal of statistical software 22, 1-20. Forbes, S.A., Tang, G., Bindal, N., Bamford, S., Dawson, E., Cole, C., Kok, C.Y., Jia, M., Ewing, R., Menzies, A., et al. (2010). COSMIC (the Catalogue of Somatic Mutations in Cancer): a resource to investigate acquired mutations in human cancer. Nucleic Acids Res 38, D652-657. Franceschini, A., Szklarczyk, D., Frankild, S., Kuhn, M., Simonovic, M., Roth, A., Lin, J., Minguez, P., Bork, P., von Mering, C., et al. (2013). STRING v9.1: protein-protein interaction networks, with increased coverage and integration. Nucleic Acids Res 41, D808815. Hennig, C. (2010). fpc: Flexible procedures for clustering. R package version Available: http://cranr-projectorg/web/packages/fpc/. Huh, W.K., Falvo, J.V., Gerke, L.C., Carroll, A.S., Howson, R.W., Weissman, J.S., and O'Shea, E.K. (2003). Global analysis of protein localization in budding yeast. Nature 425, 686-691. Janke, C., Magiera, M.M., Rathfelder, N., Taxis, C., Reber, S., Maekawa, H., MorenoBorchart, A., Doenges, G., Schwob, E., Schiebel, E., et al. (2004). A versatile toolbox for PCR-based tagging of yeast genes: new fluorescent proteins, more markers and promoter substitution cassettes. Yeast 21, 947-962. Levine, T.P., and Munro, S. (2002). Targeting of Golgi-specific pleckstrin homology domains involves both PtdIns 4-kinase-dependent and -independent components. Current biology : CB 12, 695-704. Rousseeuw, P.J. (1987). Silhouettes: a graphical aid to the interpretation and validation of cluster analysis. Journal of computational and applied mathematics 20, 53-65. Saliba, A.E., Vonkova, I., Ceschia, S., Findlay, G.M., Maeda, K., Tischer, C., Deghou, S., van Noort, V., Bork, P., Pawson, T., et al. (2014). A quantitative liposome microarray to systematically characterize protein-lipid interactions. Nat Methods 11, 47-50. Sing, T., Sander, O., Beerenwinkel, N., and Lengauer, T. (2005). ROCR: visualizing classifier performance in R. Bioinformatics 21, 3940-3941. Tong, A.H.Y., and Boone, C. (2007). High-Throughput Strain Construction and Systematic Synthetic Lethal Screening in Saccharomyces cerevisiae. Methods in Microbiology 36, 369– 386, 706–707.

35