PIP3-binding proteins promote age-dependent protein aggregation and limit survival in C. elegans

Class-I phosphatidylinositol 3-kinase (PI3KI) converts phosphatidylinositol 4,5-bisphosphate (PIP2) to phosphatidylinositol 3,4,5-triphosphate (PIP3). PIP3 comprises two fatty-acid chains that embed in lipid-bilayer membranes, joined by glycerol to inositol triphosphate. Proteins with domains that specifically bind that head-group (e.g. pleckstrin-homology [PH] domains) are thus tethered to the inner plasma-membrane surface where they have an enhanced likelihood of interaction with other PIP3-bound proteins, in particular other components of their signaling pathways. Null alleles of the C. elegans age-1 gene, encoding the catalytic subunit of PI3KI, lack any detectable class-I PI3K activity and so cannot form PIP3. These mutant worms survive almost 10-fold longer than the longest-lived normal control, and are highly resistant to a variety of stresses including oxidative and electrophilic challenges. Traits associated with age-1 mutation are widely believed to be mediated through AKT-1, which requires PIP3 for both tethering and activation. Active AKT complex phosphorylates and thereby inactivates the DAF-16/FOXO transcription factor. However, extensive evidence indicates that pleiotropic effects of age-1-null mutations, including extreme longevity, cannot be explained by insulin like-receptor/AKT/FOXO signaling alone, suggesting involvement of other PIP3-binding proteins. We used ligand-affinity capture to identify membrane-bound proteins downstream of PI3KI that preferentially bind PIP3. Computer modeling supports a subset of candidate proteins predicted to directly bind PIP3 in preference to PIP2, and functional testing by RNAi knockdown confirmed candidates that partially mediate the stress-survival, aggregation-reducing and longevity benefits of PI3KI disruption. PIP3-specific candidate sets are highly enriched for proteins previously reported to affect translation, stress responses, lifespan, proteostasis, and lipid transport.

AKT/FOXO pathway, pleiotropic effects of age-1-null mutations greatly exceed those of other disruptions to that signaling cascade, suggesting the possible involvement of other PIP 3 -binding proteins [2,3]. There are >200 human proteins with Pleckstrin Homology (PH) domains, augmented by additional phosphoinositide-bindingdomain families (see below). However, very few proteins in these families have been shown to preferentially bind PIP 3 , and protein homology has proven insufficient to identify PIP 3 -specific binding [4]. In contrast, structural analysis has shown promise for predicting PH-domain binding preferences [5]. We used PIP 3 -and PIP 2 -affinity enrichment to compare proteins from isogenic strains that differ genetically in their ability to make PIP 3 , coupled to proteomic identification of the binding proteins. Molecular modeling of proteins that show preferential binding to PIP 3 over PIP 2 allows us to extend structural prediction beyond PH-domain proteins, with the potential for confirmation by functional testing.
While disruption of insulin and insulin-like signaling (IIS) pathways has been shown to extend lifespan in diverse species, the elimination of class-I PI3K confers at least 4-fold greater life extension than any other IIS mutation [3]. The basis for this heightened dependence on PI3K I remains unresolved, but may be related to the ability of PI3K I mutation to reduce protein aggregation [27], which accompanies normal aging but is elevated and neurotoxic in most or all neurodegenerative diseases. IIS, and in particular PI3K I , modulate neuronal processes including learning and neuron survival [28,29].
In the current study we identified the PIP 3 -binding proteins from C. elegans and showed that some are involved in age-related traits including oxidative stress resistance and longevity in normal worms, as well as protein aggregation and associated functional impairment in C. elegans models of neurodegeneration-associated proteinopathy.

PI3K I contributes to protein aggregation in diverse nematode models
Components of the IIS pathway, including PI3K I , have been implicated in diverse neuropathologies including Alzheimer's disease [30][31][32]. PI3K I is a key mediator of protein aggregation [27] and the unfolded protein response [33], leading us to test whether its knockdown (by RNAi targeting the age-1 gene that encodes the PI3K I catalytic subunit) would rescue nematode models of protein aggregation. In adult C. elegans with muscle expression of a Q40::YFP transgene, age-1 knockdown reduced the number of fluorescent aggregates by >35% ( Figure 1A; P < 10 -4 ). Moreover, in worms expressing human Aβ 1-42 in muscle, amyloidinduced paralysis declined 46% after age-1 knockdown ( Figure 1B; P = 0.02).

PIP 3 deficiency reduces the yield of membrane proteins and especially of membrane-associated PIP 3 -binding proteins
The two fatty-acid chains of PIP 3 are embedded in the inner plasma membrane, whereas PIP 3 -binding domains such as PH bind the phosphorylated inositol ring that projects into the cytoplasm. In this way, key signaling proteins such as AKT, PDK-1, PLCs, and aPKC are tethered to the inner membrane surface, where they are in proximity to other signaling kinases via clustering [34]. We used PIP 3 -coated agarose beads to isolate PIP 3binding proteins from C. elegans membranes, in which they are expected to be enriched. Based on staining of electrophoresed proteins, most but not all membrane proteins isolated from wild-type (N2) adults were less abundant in age-1-null mutant adults ( Figure 2, solid arrows). Many protein bands, not necessarily the same ones, increased in abundance in worms fed a diet supplemented with exogenous PIP 3 (open arrows, Figure  2). Gels like those of Figure 2 were sliced and analyzed by high-resolution proteomics to identify the proteins in each lane. Complete spectral counts are listed in Supplementary Data, Table S1, and summarized as Venn diagrams in Figure 3A. Of the 708 membrane proteins identified from N2, 632 (89%) were also seen in age-1(mg44) F2 adults lacking active PI3K I and having no

Figure 2:
A strong nonsense mutation in the age-1 gene (allele mg44) reduces the recovery of many membrane proteins relative to wild-type controls, whereas feeding PIP 3 to worms restores some bands. Polyacrylamide/SDS gels, stained with SYPRO Ruby after electrophoresis, show A. isolated membrane proteins, and B. proteins recovered from PIP 3 -coated beads, bound after isolation of membrane proteins as in A. detectable PIP 3 , so only 11% of these proteins may be membrane-associated via PIP 3 tethering. Feeding PIP 3 to PI3K-null worms restored 40 proteins that were identified in N2 (5.6%). Considering just those membrane proteins that bind PIP 3 far more than PIP 2 based on relative spectral counts ( Figure 3B), 560 proteins from N2 adults met these criteria but just 286 of those (51%) were also identified in very long-lived age-1(mg44) F2 adults. Feeding PIP 3 restored 81 proteins found in N2 (15%), or 30% of the 274 N2-specific proteins that might have been rescued.

Proteins that preferentially bind PIP 3
In Table 1, results are compiled from 3 typical experiments (of 5 that were run). The table lists many proteins that consistently preferred PIP 3 over PIP 2 binding (Experiments 1 and 2) and/or were greatly depleted in age-1(mg44) worms lacking active PI3K I and hence PIP 3 (Experiment 3). Noteworthy differences are highlighted in yellow. Examples include AKT-1, a Ser/Thr kinase known to contain a PIP 3 -specific PH domain [4], which strongly preferred binding to PIP3 over PIP2 but was not effectively recovered from membranes of well-fed worms; muscle M-line assembly protein (UNC-89) and both nematode 14-3-3 proteins (PAR-5 and FTT-2), all reported to contain PH domains of uncertain specificity [4]; disorganized muscle protein 1; a variety of vitellogenins, protein precursors of LDL proteins also related to ApoB-100; chaperonins HSP60, HSP70, HSP90, and 4 subunits of the T complex, important in protein folding and refolding; protein disulfide isomerase, involved in refolding oxidized proteins; V-type proton ATPase; 3 fatty-acid binding proteins; 14 ribosomal proteins; and 5 α (noncatalytic) subunits of the 26S proteasome, most of Related proteins that behaved similarly have been grouped together as indicated. Expt. 1: Worm (N2) proteins, recovered after affinity binding to PIP 2 -or PIP 3 -coated beads, were electrophoresed on polyacrylamide/SDS gels, and identified from trypsindigested gel slices by LC-MS/MS proteomics. Expt. 2: Membrane proteins were isolated from wild-type worms (N2), and associated proteins were recovered using a detergent that dissociates protein complexes (unlike Expts. 1 and 3). Expt. 3A: Membrane proteins were isolated from wild-type worms (N2), a PI3K-null mutant (mg44) or mg44 adults fed PIP 3 . Expt. 3B: Proteins from 3A were bound to PIP 3 -beads, eluted, resolved by electrophoresis, and identified from trypsin-digested gel slices by LC-MS/MS proteomics. Spectral counts are shown, indicating the number of significant peptide identifications per protein, a crude measure of relative protein abundance. Deep yellow highlighting indicates ratios of ≥5; lighter yellow indicates suggestive differences. *RNAi knockdown extends lifespan; **RNAi reduces lifespan; †RNAi alters protein aggregation.   which are in the 20S core assembly.
In silico modeling predicts preferential interaction with PIP 3 over PIP 2 , for most proteins with higher observed affinity for PIP 3 Molecular docking programs were used to estimate affinities for PIP 3 vs. PIP 2 of proteins empirically observed to prefer PIP 3 . The docking predictions are not expected to support all of the observed PIP 3 -specific candidate proteins, since retention on PIP 3 -coated beads could reflect either direct or indirect binding (e.g., via a complex). We began by retrieving crystallographic or NMRbased structures from the PDB database, for structuredefined orthologs of candidate proteins that had shown preferential affinity for PIP 3 in multiple experiments. The corresponding C. elegans protein structures were then derived using molecular modeling with I-TASSER or MODELLER 9.13 (see Methods). Similarly, the NMR-based structures of PIP 3 and PIP 2 were refined in MODELLER, chiefly by truncating the fatty-acid chains to limit their contributions to interactions.
Docking of each protein structure was simulated with PIP 2 and PIP 3 separately by energy minimization, using AutoDock-Vina to calculate ΔG binding (the change in Gibbs free energy on binding) for each docking interaction. Table 3 shows ΔG binding for protein binding to PIP 3 or PIP 2 and the difference between them. That energy difference, ΔΔG = ΔG(PIP 3 )-ΔG(PIP 2 ), indicates the binding preference for either phosphoinositide. Docking models that emphasize contact points ( Figure 4A & B) illustrate the precise fit of PIP 3 in PH domains of both human and nematode AKT proteins. Predicted ΔΔG values for 15 of 31 candidate proteins (48%) surpassed all 40 randomly chosen control proteins ( Figure 4C; rank-order P < 3 x 10 -4 ), and 16 of those (bold lines, Table 3) would be considered significant at an empirical threshold of P < 0.05. For all 16, ΔG(PIP 3 ) was < -7.5, indicating relatively stable interactions.

PIP 3 -binding proteins influence protein aggregation in C. elegans model systems
The same 18 PIP 3 -binding candidates were tested to determine whether they contribute to protein aggregation. Strain CL4176 can be induced to express human Aβ 1-42 in body-wall muscle [37]; it forms β-amyloid aggregates leading to paralysis soon after induction, or progressively with age if not induced [36]. RNAi knockdown of 5 genes (28%) encoding RAD-50, AKT-1, CAND-1, FAT-2 and DHC-1, reduced age-dependent paralysis in adult worms For each protein structure listed (obtained from PDB or derived as described), docking was simulated with PIP 2 or PIP 3 by energy minimization using AutoDock-Vina. ΔG binding was calculated for each docking, and proteins were ranked by the difference between PIP 3 and PIP 2 binding energies: ΔΔG = ΔG(PIP 3 ) -ΔG(PIP 2 ). Control protein structures were taken at random from PDB for 40 proteins, of which the top-ranked 10 (based on ΔΔG) are shown.
with "leaky" Aβ 1-42 expression [36], shown at day 12 in Figure 6. Three of these knockdowns (rad-50, cand-1 and fat-2) also significantly blocked paralysis 48 h after induction of Aβ 1-42 (data not shown), but we consider uninduced paralysis to be a more appropriate model of age-dependent protein aggregation.
Muscle-specific expression of human α-synuclein (in strain NL5901), a model of Parkinson's disease, was also assessed during adult aging. Significant reductions in the number of aggregates, at least as deep as that elicited by RNAi to age-1, were observed at 9 and 10 days post-hatch after knockdown of genes encoding 6 (33%) of 18 PIP 3 -binding proteins tested: RAD-50, FAT-2, TCT-1, PRDX-3, KAT-1, and PAS-6 ( Figure 7). KD of ifb-1 appeared to increase aggregates, although without statistical significance.

RNAi knockdowns of several PIP 3 -binding proteins confer multiple fitness benefits
Previous studies in C. elegans had reported lifespan extension upon knockdown for two of those 5 genes, akt-1 [38] and fat-2 [39]. We confirmed a significant (P < 0.05) life extension upon akt-1 knockdown (data not shown), and somewhat stronger effects of RNAi targeting cand-1 (**P < 10 -4 ) or rad-50 (*P < 3x10 -4 ) ( Table 4, Figure  8), suggesting that these PIP 3 -binding proteins may also contribute to the unique longevity of age-1-null mutants that cannot form PIP 3 [3]. Significant but less pronounced life extension was observed when RNAi was begun only at the L4 (late-larval) stage to avoid effects on development ( Table 4, Expt. 4). However, no life extension was seen in a daf-16 mutant ( Figure 8B), confirming that cand-1 and rad-50 act via the insulin/IGF-1 signaling pathway. Of 18 PIP 3 -binding candidates tested by RNA interference, 5 improved fitness by at least two measures, survival of peroxide stress and protection in protein-aggregation models (Table 5), and at least 4 extend lifespan.   PIP 3 , the product of class-I PI3K, is normally embedded in eukaryotic cell plasma membranes where it is thought to contribute to multiple kinase-cascade signaling pathways. C. elegans mutants lacking the PI3K I catalytic subunit exhibit extreme longevity, improved stress resistance, delayed development, and reproductive defects [3]. We infer that PIP 3 plays critical roles in these pleiotropic physiological traits by binding and recruiting signaling proteins. Those proteins, and the pathways they participate in, must confer developmental and reproductive benefits early in life, presumably through roles in anabolic metabolism and cell proliferation. Nonetheless, they appear to have deleterious long-term consequences so that their continued activity in post-gravid adults promotes aging.

Proteins that bind PIP 3 are tethered to the membrane, dependent on PIP 3 availability
We used a combined proteomic/genetic strategy to isolate and identify PIP 3 binding proteins in C. elegans. Although any affinity-capture procedure can produce false positives, we set several criteria by which to evaluate candidate proteins identified in at least 3 independent experiments. Preferential binding to PIP 3 over PIP 2 PIP 3 -specific binding was assessed by comparing the capture of any given protein based on affinity to each ligand. Many proteins showed differential binding, of which 10 (highlighted in bright yellow, Table 1/Expt. 1) met our arbitrary criteria of (i.) at least 5-fold higher binding to PIP 3 than to PIP 2 and (ii.) at least 5 peptide "hits". Several caveats should be considered. The PIP 3 / PIP 2 ratio in normal cells ranges from 0.001 to 0.02, suggesting that moderately high affinity ratios might not be sufficient to prevent binding to PIP 2 due to its higher abundance in vivo. Moreover, affinity purification may not precisely mirror physiological dependence on ΔG binding , since binding also depends on protein abundance. Protein levels are equal for the two PI ligands in lysates, but certainly differ among cell types in vivo, which might also vary in their PIP 3 levels. Binding propensities in vivo can also be altered by additional factors such as competing ligands, other interacting proteins in a complex, and other hydrophobic or electrostatic features in the immediate vicinity of membrane-embedded PIP 3 .
Some proteins may show preferential affinity for PIP 3 through indirect binding, in which the protein is part of a stable complex containing a PIP 3 -specific binding protein. Our data suggest that this may be quite common. We note that Experiments 1 and 2 in Table 1 differed in only one respect: protein complexes were disrupted in the second experiment, leading to loss of many "PIP 3 -specific" proteins that had ceased to be significantly differential. The value of candidate PIP 3 -binding proteins as potential  pharmacologic targets, however, does not depend on whether membrane-tethering is direct or indirect.
Many proteins are integral to membranes, and if these are in vast excess they might mask proteins that are recruited to membranes via tethering to PIP 3 . Thus it is reassuring that most PIP 3 -binding proteins that were affinity-isolated from the membrane fraction, although depleted in overall abundance by the two-step isolation, were further reduced in age-1(mg44) adult worms relative to N2 controls (Table 1, Experiment 3B). Apparent exceptions include pyruvate carboxylase 1, CAND-1, TCT-1, peroxiredoxin and proteasome α subunits.

Restoration of membrane tethering by exogenous PIP 3
In principle, it should be possible to reverse all "age-1" traits in worms genetically deficient in PI3K I and hence in PIP 3 , by supplying exogenous PIP 3 , provided that it is taken up, is not degraded, and reaches all tissues. Of the proteins listed under (2.) above, fatty-acid desaturases and elongases showed essentially no reversal (≤3%) whereas all other proteins were partially reversed (17-80%) by PIP 3 feeding.
This wide variation in efficacy of PIP 3 supplementation appears paradoxical: If PIP 3 is able to enter cells to reverse any of the age-1 phenotypes, should it not reverse all of them? Variable extents of rescue could result from our use of very short-acyl-chain PIP 3 , which may be insufficient to tether large proteins or complexes to the plasma membrane. A plausible alternative explanation is tissue heterogeneity of PIP 3 uptake, since PIP 3 added to the medium or the bacterial lawn would reach intestinal cells first, and unless it saturated the membranes of those cells it might not distribute any further. Our data appear inconsistent with this scenario, however, since the proteins least effectively "reverted" (restored to membrane fractions in age-1(mg44) F2 worms) by PIP 3 feeding are fatty-acid desaturases and elongases, expected to be concentrated in the intestine since it serves as the lipidstorage organ of nematodes [40,41].

Molecular modeling can predict proteins that preferentially bind to PIP 3
Park et al. [5] used "machine learning" to define the amino-acid residues and positional constraints within PH domains that distinguished between proteins demonstrated previously (or, for the test group, demonstrated in that paper) to have binding specificity for particular PI moieties including PIP 3 . Apart from that study, no systematic attempt has been made to identify all proteins with PIP 3specific binding. Over the last few years, molecular modeling has proven itself capable of screening proteindrug and protein-protein interactions with ever-increasing reliability. Nevertheless, we were uncertain whether it would prove equal to the challenge posed by discriminant PIP 3 -binding.
Just over half of the 31 tested proteins that had been identified as strong PIP 3 -specific candidates substantially exceeded the minimal energetic requirements for avid and selective binding of PIP 3 , with ΔΔG values outside the range observed in randomly-chosen control proteins (Table  4; Figure 4C). We note that PIP 3 would be considered a reasonable candidate ligand for all 31 proteins, given that their ΔG levels ranged from -5.5 to -13.6; protein ligands predicted previously by AutoDock agreed well with confirmed binders provided that ΔG binding was < -4 [42]. Our results strongly support the premise that molecular modeling can predict substrate specificity for PtdInsPbinding proteins. The 15 proteins with unexceptional ΔΔG values may not selectively bind PIP 3 themselves, but instead participate in larger protein complexes that include a PIP 3 -specific component. Such indirect tethering of complexes may be quite common, but the candidate proteins nevertheless remain drug targets of interest if they are components of complexes that require PIP 3 or membrane localization to function.
The frequent recovery of ribosomal proteins, specific to PIP 3 affinity and to worms with PI3K I activity, suggests that PIP 3 may be involved in tethering ribosomes to rough endo plasmic reticulum. This conjecture is currently supported only by indirect evidence [55], but would account for the 7.4-fold GO enrichment for "translation" ( Table 2).

PIP 3 -binding proteins contribute to diverse pathways mediating extreme age-1 traits
Although it was not a criterion for PIP 3 -specific binding, a goal of the current study was to assess whether any of the proteins downstream of PI3K I , if depleted by RNA interference, could confer some part of the beneficial survival traits displayed so strikingly by very-long-lived age-1-null mutant worms [3]. In fact, as shown in Figure  5, RNAi directed against genes rad-50, cand-1, cct-1, fat-2, and akt-1, encoding candidate PIP 3 -binding proteins, extended the length of time that adult worms could survive a lethal oxidative stress (5-mM H 2 O 2 ), emulating the greatly enhanced peroxide-resistance of age-1(mg44) F2 adults [3]. Moreover, RNAi targeting rad-50, cand-1, fat-2, and dhc-1 rescued 92-100% of the paralysis that otherwise progressively afflicted worms expressing Aβ 1-42 in body-wall muscle ( Figure 6). RNAi directed against akt-1, although also significant, was less effective (69% rescue) -perhaps due to functional redundancy between AKT-1 and AKT-2 proteins. Two of these RNAi treatments were previously reported to increase longevity, and three (targeting akt-1, cand-1 and rad-50) produced moderate increases in lifespan (9-20%) in our hands.
In view of the importance of PI3K I as a driver of cell proliferation, inhibitors have been actively sought as potential chemotherapeutic agents for cancer. In pursuit of novel anti-cancer drugs, and equally in seeking drugs to prevent or ameliorate Alzheimer's and other agedependent diseases, potential therapeutic benefits of PI3K I inhibitors have been over shadowed by the virtual certainty that they would also be detrimental to stem cell niches. This concern has motivated our search for PIP 3binding proteins as alternative targets downstream of PI3K I . Among these, AKT is believed to drive most of the proliferative effects of PI3K [45], whereas RAD50, CAND1, FAT2 and TCT1 constitute novel targets that may preserve survival benefits of PI3K I disruption, uncoupled from blockage of cell proliferation.

RNA interference
Targeted genes were subjected to RNAi knock-down by feeding worms (either from the time of hatching, or from the L4 (last larval) stage to avoid developmental effects of RNAi) on HT115 bacterial sublines from the Ahringer RNAi library [56]. Briefly, synchronized eggs were recovered after alkaline hypochlorite lysis and transferred to plates seeded with HT115 (DE3) bacteria, deficient in RNAse III and containing (a.) IPTG-inducible T7 RNA polymerase, and either (b.) the L4440 plasmid with a multiple cloning site (MCS) between two inwarddirected T7 RNA polymerase promotors for "feeding vector (FV) controls", or (c.) L4440 containing an exonic segment of the targeted gene, cloned into its MCS [56].

Isolation of membrane proteins
Synchronized day-3 C. elegans adults were collected after washing in S buffer, drained of excess liquid, and flash frozen in liquid nitrogen. The worm pellets were pulverized with a dry-ice-cooled mortar and pestle, and suspended in buffer with nonionic detergent (20-mM Hepes pH 7.4, 300-mM NaCl, 2-mM MgCl 2 , 1% NP40), and protease/phosphatase inhibitors (MilliporeSigma, Darmstadt, Germany) at 0°C. Worm or cell debris was removed by brief centrifugation of lysate (5 min. at 3000 rpm). Native membrane-associated proteins were isolated from lysates with ProteoExtract membrane purification kit (MilliporeSigma) following the manufacturer's protocol, and either (a.) used for PIP 3 binding (see next section), or (b.) suspended in Laemmli buffer containing 2% SDS (w/v) and 0.3-M β-mercaptoethanol, heated 5 min at 95°C to dissolve proteins, and electro phoresed on 4-20% polyacrylamide gels (SDS-PAGE). Gels were stained with SYPRO Ruby (ThermoFisher) to visualize total protein.

Isolation of PIP 3 -binding membrane proteins
Isolated membrane proteins were pre-adsorbed to uncoated control beads; the unbound fraction was collected and incubated 6 h at 4°C with PIP 2 -or PIP 3coated agarose beads (echelon, Salt Lake City, UT). After extensive washing, bound proteins were eluted from PIP 2and PIP 3 -coated beads, suspended in Laemmli buffer containing 2% SDS (w/v) and 0.3-M β-mercaptoethanol, and heated 5 min at 95°C to dissolve proteins prior to separation on 4-20% polyacrylamide/SDS gels as above.

Identification of membrane and/or PIP 3 -binding proteins
Proteins isolated from membranes, or membranes followed by PIP 3 -coated beads, were dissolved in Laemmli buffer as described above, and separated in one dimension on 1% SDS, 4-12% acrylamide gradient gels. They were then stained with SYPRO Ruby (ThermoFisher) or Coomassie Blue to visualize total protein, and 1-mm slices were excised. Proteins were digested in situ with trypsin, and peptides analyzed by high-resolution LC-MS/MS with a Thermo-Velos Orbitrap mass spectrometer (ThermoFisher) coupled to a nanoACQUITY liquid chromatography system (Waters, Milford MA) as previously reported [20]. Proteins were identified by MASCOT (www.matrixscience.com) matching of peptide fragmentation patterns to a database of previously observed fragment patterns [20].

PIP 3 feeding to "rescue" age-1-null worms
Very long-lived age-1(mg44) mutant worms are maintained as genetically mixed cultures, with a recessive visible-trait marker (dumpy) on a balancer chromosome carrying wild-type age-1. Synchronized worms that are not dpy/dpy (and thus not age-1 +/+ ) are placed singly on individual nutrient-agar plates and classified based on the developmental rate of their progeny: heterozygotes (mg44/+) produce a majority of offspring that develop normally, reaching adulthood in 2.5 days, while all progeny of first-generation mg44/mg44 homozygotes are second-generation "F2" age-1(mg44) homozygotes that uniformly develop quite slowly (>8 days at 20 o C, from hatch to the L4/adult moult [3]). These F2 mutants were fed 30-μM phosphatidylinositol 3,4,5-triphosphate di-C4 (echelon), beginning at 3 days post-hatch (as soon as they could be distinguished with certainty from their less longlived siblings). PIP 3 di-C4 differs from normal PIP 3 by having very short (4-carbon) fatty-acid chains to improve water solubility.

Paralysis assays
CL4176 worms were synchronized as described above, and eggs were transferred to 60-mm NGM-agar plates seeded with either FV-control bacteria or RNAiexpressing bacteria to target each gene that encodes a protein of interest. Paralysis of worms with muscle expression of Aβ 1-42 was assayed as described previously [36]. Briefly, worms were upshifted from 20 to 25°C at the L3-L4 transition, and triplicate groups of 50-100 worms were scored 29 h later. Alternatively, age-dependent paralysis was monitored over 10-13 days post-hatch, maintaining worms at 20°C without upshift. Paralysis was defined by movement of the head, but not the body, in response to a touch stimulus.
Aggregation assays for Q40::YFP and α-synuclein::YFP AM141 worms with muscle expression of an unc-54p/Q40::YFP transgene [27] were synchronized and fed from hatch with RNAi targeting each individual gene that encodes a PIP 3 -binding protein, or empty feeding vector (FV controls). They were assessed by imaging yellow fluorescence, 4 days post-hatch as described [36,56]. NL5901 worms, with muscle expression of an unc-54p/α-synuclein::YFP fusion protein [57], were grown on RNAi (or FV) bacteria as described above and imaged at days 9 and 10 post-hatch. To record images, worms were immobilized on glass slides in S buffer containing 0.3% (w/v) sodium azide to block muscle contraction, and fluorescence images were captured on a DP71 camera mounted on a BX51 fluorescence microscope (Olympus, Tokyo) with a 10x objective. YFP-containing aggregates (fluorescent foci) were counted for >15 worms (5-10 fields) per group with dotcount (reuter.mit.edu/software/ dotcount). Statistical significance of differences in counts/ worm were based on 5-10 fields per group, by 2-tailed heteroscedastic t tests.

Longevity survivals
Synchronized eggs (or L4 larvae, to avoid developmental effects) were plated on control bacteria containing an empty feeding-vector plasmid, or on RNAi bacteria selected from the Ahringer library [56] transcribing double-stranded RNA of an exonic segment from a PIP 3 -binding protein. Worms were transferred to fresh plates daily for 7 days, and on alternate days thereafter, scoring worms as alive if they moved spontaneously or in response to gentle prodding [3]. Worms lost for reasons other than natural death were censored (removed from mortality calculations) from the date of first annotation onward.

Hydrogen peroxide stress survivals
Worms were hatched and maintained on RNAi or control plates as in the preceding section. Day-1 adult worms (50 worms, 24 h after the L4/adult molt) were washed free of bacteria and incubated at 20 o C in 300 μl of 5-mM H 2 O 2 in a 24-well plate. Worms were scored at hourly intervals for survival based on movement in response to touch [36] to assess each RNAi for effect on H 2 O 2 -stress survival.

Computer modeling of protein structures
Sequences of all proteins were compiled from WormBase and UniProt databases, and used to perform BLASTP searches to retrieve known structures from PDB (Protein Data Base). The 3-dimensional structures of candidate proteins were modeled in two ways. In the case of nematode proteins for which a reasonable template exists (i.e. ≥70% identity to a protein of known structure), template modeling was performed in MODELLER 9.13 (salilab.org/modeller) with default parameters [36]. Proteins lacking good templates were modeled in I-TASSER (zhanglab.ccmb.med.umich.edu/I-TASSER) to predict structures by ab initio "multi-threading" methods [58]. Ramachandran plots were generated for all structures generated in Modeller or I-TASSER using the RAMPAGE server (mordred.bioc.cam.ac.uk/~rapper/rampage.php), and amino acids in dis-allowed regions were looprefined using MODELLER. This process was continued iteratively until the entire plots fell within permitted limits. The final minimum-energy conformers were then used for further docking analyses as described.

Protein docking to PIP 3 and PIP 2
The structure of PIP 3 (Pubchem database) was processed by truncation of fatty-acyl chains using ChemSketch, and hydrogen atoms added as required with MGL Tools and Python Molecular Viewer while monitoring torsions. The PIP 2 structure was then generated from that of PIP 3 by removing the 3-phosphate. AutoDock Vina 4.2 (http://vina.scripps.edu) was used with default parameters [59] to predict interactions of candidate proteins with PIP 3 and PIP 2 , and to score the complexes for interaction energies. For control interactions, protein structures were retrieved at random from the PDB database (http://www.rcsb.org). Heteroatoms (ligand), if present in downloaded structures, were manually removed before docking, as above, in AutoDock Vina 4.2 (run with Raccoon interface on a 32-core Linux cluster). To avoid bias, entire proteins were made available for docking. Grid dimensions were defined manually for each interaction. The Gibbs free energy of binding was calculated as ΔG binding = ΔG vdW + ΔG elec + ΔG H-bond + ΔG desolv + Δg tors , where ΔG vdW = the Lennard-Jones van der Waals potential with 0.5Å smoothing; ΔG elec = the Solmajer-Mehler distance-dependent dielectric potential; ΔG H-bond = hydrogen-bonding potential with Goodford directionality; ΔG solv = charge-dependent version of Stouten pairwise atomic solvation energy; and Δg tors is a function of the number of rotatable bonds in the ligand only (see http:// autodock.scripps.edu for details).

Statistical analyses
Differences between groups were assessed for significance by the Fisher-Behrens heteroscedastic t test (appropriate to samples of unequal or unknown variance). Differences in relative peptide abundance, based on spectral counts relative to the total per sample, were assessed for significance by chi-squared or Fisher exact tests. Significance of longevity-survival differences was ascertained by Gehan-Wilcoxon log-rank tests. Significance of differences in peroxide-stress survival was assessed by Fisher exact tests at the earliest assay time for which control survival was < 15%.