Physiological Consequences of Targeting 14-3-3 and Its Interacting Partners in Neurodegenerative Diseases

The mammalian 14-3-3 family comprises seven intrinsically unstructured, evolutionarily conserved proteins that bind >200 protein targets, thereby modulating cell-signaling pathways. The presence of 14-3-3 proteins in cerebrospinal fluid provides a sensitive and specific biomarker of neuronal damage associated with Alzheimer’s disease (AD), Creutzfeldt–Jakob disease (CJD), spongiform encephalitis, brain cancers, and stroke. We observed significant enrichment of 14-3-3 paralogs G, S, and Z in human brain aggregates diagnostic of AD. We used intra-aggregate crosslinking to identify 14-3-3 interaction partners, all of which were significantly enriched in AD brain aggregates relative to controls. We screened FDA-approved drugs in silico for structures that could target the 14-3-3G/hexokinase interface, an interaction specific to aggregates and AD. C. elegans possesses only two 14-3-3 orthologs, which bind diverse proteins including DAF-16 (a FOXO transcription factor) and SIR-2.1 (a sensor of nutrients and stress), influencing lifespan. Top drug candidates were tested in C. elegans models of neurodegeneration-associated aggregation and in a human neuroblastoma cell-culture model of AD-like amyloidosis. Several drugs opposed aggregation in all models assessed and rescued behavioral deficits in C. elegans AD-like neuropathy models, suggesting that 14-3-3 proteins are instrumental in aggregate accrual and supporting the advancement of drugs targeting 14-3-3 protein complexes with their partners.


Introduction
14-3-3 proteins comprise small families of paralogous proteins (seven in mammals, two in nematodes) that serve as modulators of critical signaling and regulatory pathways; e.g., via their interactions with transcription factors, including FOXO and TFEB [1]. The 14-3-3 protein family in mammals comprises seven 30-kDa proteins designated as β, γ, ε, τ, η, σ, and ζ paralogs (beta, gamma, epsilon, tau, eta, sigma, and zeta). These mammalian 14-3-3 genes, situated at seven distinct chromosomal loci, are expressed ubiquitously but are especially abundant in cerebral neurons [2]. They have especially high binding affinities for proteins containing phospho-serine or phospho-threonine and serve as both molecular chaperones and regulatory modulators of their numerous protein ligands [3]. They occur predominantly as homo-and heterodimers, forming complexes with over 200 other ligand proteins. Their highly disordered C-terminal and N-terminal domains [4] are generally stabilized by protein-protein interactions (PPIs) with their binding partners, through which they modulate signal transduction, apoptosis, cell-cycle progression, RNA transcription, and DNA replication [3,5]. Table 1. Spectral hits for three paralogs of 14-3-3 in AD and AMC aggregates, isolated by immunoprecipitation (IP) with antibodies to seed proteins Aβ  or tau, or "total" aggregates without IP.

Molecular Dynamic Simulations and Interactome Analyses of 14-3-3 Paralogs
Previous studies demonstrated that 14-3-3 proteins have disordered N-and C-termini as monomers but attain stable conformations when bound to a target peptide or phosphopeptide [17]. We obtained experimental structures for the conserved core regions of three of the seven mammalian 14-3-3 paralogs (γ/G/gamma, σ/S/sigma, and ζ/Z/zeta) in both the peptide-bound monomeric and dimeric states from the Protein Data Bank (PDB IDs 5N10, 4O46, and 6QHL, respectively). In order to explore the structural stability of these proteins as monomers, we conducted 200-ns atomistic molecular dynamic simulations in triplicate. Based on plots of root-mean-square deviation (RMSD) vs. time (Figure 1a), none of these 14-3-3 monomeric core regions (α/alpha, γ/gamma, and ζ/zeta) attained a stable conformation within 200 ns.

Molecular Dynamic Simulations and Interactome Analyses of 14-3-3 Paralogs
Previous studies demonstrated that 14-3-3 proteins have disordered N-and Ctermini as monomers but attain stable conformations when bound to a target peptide or phosphopeptide [17]. We obtained experimental structures for the conserved core regions of three of the seven mammalian 14-3-3 paralogs (γ/G/gamma, σ/S/sigma, and ζ/Z/zeta) in both the peptide-bound monomeric and dimeric states from the Protein Data Bank (PDB IDs 5N10, 4O46, and 6QHL, respectively). In order to explore the structural stability of these proteins as monomers, we conducted 200-ns atomistic molecular dynamic simulations in triplicate. Based on plots of root-mean-square deviation (RMSD) vs. time (Figure 1a), none of these 14-3-3 monomeric core regions (α/alpha, γ/gamma, and ζ/zeta) attained a stable conformation within 200 ns. In view of their disordered end domains, we expected the 14-3-3 proteins to bind multiple protein partners. We previously used aggregate-permeable "click" reagents to cross-link neighboring proteins within amyloid aggregates from human SY5Y-APPSw neuroblastoma cells [18] and subsequently employed the same protocol for aggregates immunopurified from AMC and AD hippocampal tissue (manuscript in preparation). In the latter study, we observed over twice as many 14-3-3 interacting partners in AD as in AMC aggregates (Figure 1b,c). Consistent with our previous work on defining aggregate interactomes [18], we found that 85-87% of AD β-amyloid proteins in close proximity to 14-3-3G or 14-3-3S were absent from AMC aggregates (Figure 1d). Several proteins interacting with 14-3-3G predominantly in AD are shown in Figure 1b,d along with their relative abundances (spectral counts) in AD vs. AMC aggregates. Our analysis of the AD interactome indicates that 14-3-3G interacts directly with tau (TAU) and other aggregate proteins, including ankyrin-3 (ANK3), plectin (PLEC), kinesin heavy chain 5C (KIF5C), and hexokinase (HXK1) (Figure 1b,c)-all of which play key roles in neurodegenerativedisease aggregation. Amyloid beta interactome sub-networks of 14-3-3G (b) and 14-3-3S (c) in aggregates from AMC and AD, respectively; node color indicates relative protein abundances (see key). (d) Venn diagram showing the 14-3-3G-interacting protein counts unique to AMC or AD, or shared in common.
In view of their disordered end domains, we expected the 14-3-3 proteins to bind multiple protein partners. We previously used aggregate-permeable "click" reagents to cross-link neighboring proteins within amyloid aggregates from human SY5Y-APP Sw neuroblastoma cells [18] and subsequently employed the same protocol for aggregates immunopurified from AMC and AD hippocampal tissue (manuscript in preparation). In the latter study, we observed over twice as many 14-3-3 interacting partners in AD as in AMC aggregates (Figure 1b,c). Consistent with our previous work on defining aggregate interactomes [18], we found that 85-87% of AD β-amyloid proteins in close proximity to 14-3-3G or 14-3-3S were absent from AMC aggregates (Figure 1d). Several proteins interacting with 14-3-3G predominantly in AD are shown in Figure 1b,d along with their relative abundances (spectral counts) in AD vs. AMC aggregates. Our analysis of the AD interactome indicates that 14-3-3G interacts directly with tau (TAU) and other aggregate proteins, including ankyrin-3 (ANK3), plectin (PLEC), kinesin heavy chain 5C (KIF5C), and hexokinase (HXK1) (Figure 1b,c)-all of which play key roles in neurodegenerativedisease aggregation.

Identification of Top 14-3-3 Interactions as Potential Targets to Disrupt Aggregation
We assessed siRNA knockdowns targeting 14-3-3G, 14-3-3Z, and four of their most abundant interacting proteins (ankyrin, hexokinase, kinesin, and plectin) with high AD/AMC ratios and interactome influence >2, as predicted by neural-network analysis. Tau, a fifth interacting partner of 14-3-3, could not be pursued because we were unable to find an FDA-approved drug binding to the interface of these two disordered proteins. SH-SY5Y-APP Sw human neuroblastoma cells were transfected with siRNA constructs specifically targeting the genes encoding each of these 14-3-3 binding partners. The impact on amyloid aggregation of each knockdown was quantified by thioflavin T staining. Total amyloid fluorescence per cell was suppressed 20-50% by these siRNA treatments (Figure 2a,b). We also quantified levels of total sarkosyl-insoluble protein recovered from these cells after acrylamide gel electrophoresis and SYPRO-Ruby staining; protein per lane was reduced 15-30% after knockdowns (Figure 2c). Total 14-3-3 protein levels in aggregates also declined after knockdowns of interacting proteins (Figure 2d) in contrast to levels of GRP78, a key mediator of the endoplasmic reticulum unfolded protein response (UPR ER ) [19], which were not altered significantly by two of these knockdowns and only modestly (~25%) by a third (Figure 2d,e).

Identification of Top 14-3-3 Interactions as Potential Targets to Disrupt Aggregation
We assessed siRNA knockdowns targeting 14-3-3G, 14-3-3Z, and four of the abundant interacting proteins (ankyrin, hexokinase, kinesin, and plectin) wit AD/AMC ratios and interactome influence >2, as predicted by neural-network a Tau, a fifth interacting partner of 14-3-3, could not be pursued because we were to find an FDA-approved drug binding to the interface of these two disordered p SH-SY5Y-APPSw human neuroblastoma cells were transfected with siRNA con specifically targeting the genes encoding each of these 14-3-3 binding partners. T pact on amyloid aggregation of each knockdown was quantified by thioflavin T st Total amyloid fluorescence per cell was suppressed 20-50% by these siRNA trea (Figure 2a,b). We also quantified levels of total sarkosyl-insoluble protein rec from these cells after acrylamide gel electrophoresis and SYPRO-Ruby staining; per lane was reduced 15-30% after knockdowns (Figure 2c). Total 14-3-3 protein in aggregates also declined after knockdowns of interacting proteins (Figure 2d) trast to levels of GRP78, a key mediator of the endoplasmic reticulum unfolded response (UPR ER ) [19], which were not altered significantly by two of these knock and only modestly (~25%) by a third (Figure 2d,e). (c) Intensity of SYPRO-Ruby-stained g for sarkosyl-insoluble aggregates isolated from SY5Y-APPSw cells exposed to RNAi constr western blot probed with antibody to GRP78, 14-3-3, or GAPDH; gels were loaded with to tein extracted from SH-SY5Y-APPSw cells treated with RNAi constructs indicated; (e) mea of band intensity for 14-3-3 or GRP78, each normalized to GAPDH. (b,c,e) Significance b two-tailed t-tests: * p < 0.05; ** p < 0.005; *** p < 10 -7 ; **** p < 10 -17 . For each assay, 100 < N < 1

Knockdown of Key Interacting Partners of 14-3-3 Tested in Nematodes
We next assessed the impact of RNAi-mediated knockdowns targeting the closest C. elegans orthologs of the above proteins in two nematode models of neurodegenerative aggregation: strain AM141, expressing a polyglutamine-array marker (Q40::YFP) in muscle cells as a model of huntingtin-like aggregation in Huntington's disease (HD) neurons; and CL2355, which forms AD-like amyloid deposits in neurons expressing human Aβ  . In the Huntington model, total aggregate intensity per worm was reduced 40-63% by RNAimediated knockdown relative to empty-feeding-vector controls (Figure 3a). Worm images (Figure 3b) illustrate the decrease in both the number and fluorescence intensity (Q40::YFP content) of aggregates. We also assessed knockdowns in worms expressing neuronal βamyloid (CL2355), which show reduced chemotaxis, a measure of chemosensory response. The decline in chemotaxis due to neuron-specific induction of a human Aβ 42 transgene and subsequent AD-like amyloid deposition left just over a third of worms moving toward the n-butanol chemo-attractant. Knockdown of each target elicited substantial and significant rescue of chemotaxis relative to control worms, elevating their levels from 36% (FV controls) to 53-80% chemotaxis (Figure 3c). These interventions thus restored 27-70% of the deficit relative to wild-type worms (99 ± 1% chemotaxis, not shown). worms expressing neuronal β-amyloid (CL2355), which show reduced chemotaxis, a measure of chemosensory response. The decline in chemotaxis due to neuron-specific induction of a human Aβ42 transgene and subsequent AD-like amyloid deposition left just over a third of worms moving toward the n-butanol chemo-attractant. Knockdown of each target elicited substantial and significant rescue of chemotaxis relative to control worms, elevating their levels from 36% (FV controls) to 53-80% chemotaxis ( Figure 3c). These interventions thus restored 27-70% of the deficit relative to wild-type worms (99 ± 1% chemotaxis, not shown).

In Silico Screening Identifies Drugs Predicted to Disrupt 14-3-3::Hexokinase Binding
Since the 14-3-3::hexokinase interaction is enriched in AD aggregates, we posited that disruption of their complex by a small molecule may relieve protein aggregation. Considering that both 14-3-3G and hexokinase are required for normal biological functions, targeting one or both of these proteins is likely to be toxic or to have adverse side effects. However, targeting the interface between 14-3-3 and hexokinase provides a feasible alternative that may safely oppose protein aggregation and its associated functional declines. We therefore targeted a druggable pocket created at the 14-3-3γ-hexokinase interface in the bound structure observed at 70 ns, a metastable point in the simulation when the binding pocket had begun to increase in volume (Figure 4d). To identify drugs that may disrupt the interaction between 14-3-3γ and hexokinase, we initially screened drugs from an FDA-approved library comprising >2300 compounds via in silico docking simulations. To improve screening efficiency, docking was conducted in three stages, increasing the stringency for successive screens. We began with virtual docking of the entire FDA-approved drug library, using SiBiolead to run AutoDock in high-throughput mode. The top 10% of molecules (230 drugs) were then re-docked using Glide in its standard ("high-precision") mode. The top 10% (23 drugs) from Glide were pursued by assessing the free energy of each 14-3-3γ::HXK::drug complex in implicit solvent using the Schrödinger MM-GBSA module. The top five candidates from this third-stage analysis (Table 2) were pursued.

In Silico Screening Identifies Drugs Predicted to Disrupt 14-3-3::Hexokinase Binding
Since the 14-3-3::hexokinase interaction is enriched in AD aggregates, we posited that disruption of their complex by a small molecule may relieve protein aggregation. Considering that both 14-3-3G and hexokinase are required for normal biological functions, targeting one or both of these proteins is likely to be toxic or to have adverse side effects. However, targeting the interface between 14-3-3 and hexokinase provides a feasible alternative that may safely oppose protein aggregation and its associated functional declines. We therefore targeted a druggable pocket created at the 14-3-3γ-hexokinase interface in the bound structure observed at 70 ns, a metastable point in the simulation when the binding pocket had begun to increase in volume (Figure 4d). To identify drugs that may disrupt the interaction between 14-3-3γ and hexokinase, we initially screened drugs from an FDA-approved library comprising >2300 compounds via in silico docking simulations. To improve screening efficiency, docking was conducted in three stages, increasing the stringency for successive screens. We began with virtual docking of the entire FDA-approved drug library, using SiBiolead to run AutoDock in high-throughput mode. The top 10% of molecules (230 drugs) were then re-docked using Glide in its standard ("high-precision") mode. The top 10% (23 drugs) from Glide were pursued by assessing the free energy of each 14-3-3γ::HXK::drug complex in implicit solvent using the Schrödinger MM-GBSA module. The top five candidates from this third-stage analysis ( Table 2) were pursued. The above FDA-approved drugs were docked to the 14-3-3::HXK binding pocket in atomistic molecular dynamic simulations. The predicted binding sites and poses of the top-ranked two drugs (lumacaftor and conivaptan) are shown in Figure 5a,b. The ability of these drugs to disrupt the interaction between 14-3-3γ and human hexokinase (HXK) was predicted using molecular-dynamic simulations. Figure 5c,d depicts the RMSD during a 200-nsec simulation for the 14-3-3G::HXK complex, with or without binding of lumacaftor or conivaptan at the protein-protein interface. The RMSD plots illustrate progressive expansion of the 14-3-3G::HXK interaction complex in the presence of either drug relative to the 14-3-3::HXK complex alone. These predictions support pursuit of the drugs conivaptan and lumacaftor as candidates to disrupt the 14-3-3::HXK interaction and thereby relieve protein aggregation. Astemizole −10 Antihistamine used to treat allergies

Drug Binding to Predict Disruption of the 14-3-3γ::HXK Complex
The above FDA-approved drugs were docked to the 14-3-3::HXK binding pocket in atomistic molecular dynamic simulations. The predicted binding sites and poses of the top-ranked two drugs (lumacaftor and conivaptan) are shown in Figure 5a,b. The ability of these drugs to disrupt the interaction between 14-3-3γ and human hexokinase (HXK) was predicted using molecular-dynamic simulations. Figure 5c,d depicts the RMSD during a 200-nsec simulation for the 14-3-3G::HXK complex, with or without binding of lumacaftor or conivaptan at the protein-protein interface. The RMSD plots illustrate progressive expansion of the 14-3-3G::HXK interaction complex in the presence of either drug relative to the 14-3-3::HXK complex alone. These predictions support pursuit of the drugs conivaptan and lumacaftor as candidates to disrupt the 14-3-3::HXK interaction and thereby relieve protein aggregation.

Top-Ranked Drugs Rescue C. elegans and Human-Cell Aggregation Models
All five top-ranked drugs ( Table 2) were tested for rescue of protein aggregation in C. elegans strain AM141 (a model of polyglutamine aggregation in Huntington's disease). The total intensity of aggregates was assessed after exposure of nematodes to each drug at two concentrations. The top two drugs (conivaptan and lumacaftor, each at 10 μM) reduced protein aggregation by 60-70% (Figure 6a). A third drug, asfemilzole at 10 μM, reduced aggregation ~25%. We also tested these drugs in a C. elegans model of AD-like amyloidosis expressing neuronal Aβ1-42 and consequently suffering impaired chemotaxis. Chemoattraction to n-butanol was 38.5% for untreated worms (vs. 99 ± 1% for worms not expressing Aβ1-42), but restored to 82% by conivaptan, 72% by digitoxin, and 60% by lumacaftor (each at 10 μM; Figure 6b).

Hexokinase Is Recovered from Neuroblastoma Cells by 14-3-3 Immuno-Pulldown
We then asked whether human neuroblastoma (SH-SY5Y-APP Sw ) cells expressing the APP Sw double mutant observed in familial AD [20] contain aggregates in which any 14-3-3 protein interacts with hexokinase. We first quantified 14-3-3 paralogs and hexokinase in cell lysates to ensure that their levels were not severely depleted by the drug. This appeared to be the case (Figure 8a,b); although several treatment groups differed significantly from controls, the difference was <20%. We used a co-immunopulldown (co-IP) strategy to recover and quantify binding of 14-3-3 to hexokinase. Complexes or aggregates isolated by IP using magnetic beads coated with antibody to the conserved 14-3-3 core were recovered, resuspended in hot loading buffer, and electrophoresed on acrylamide gels, and their western blots were probed with antibody to hexokinase or the 14-3-3 conserved core. Co-IP of hexokinase (normalized to 14-3-3 recovery) was reduced >45% by treatment with drugs specific to the binding of the interface between 14-3-3 and hexokinase (Figure 8c,d).

Hexokinase Is Recovered from Neuroblastoma Cells by 14-3-3 Immuno-Pulldown
We then asked whether human neuroblastoma (SH-SY5Y-APPSw) cells expressing the APPSw double mutant observed in familial AD [20] contain aggregates in which any 14-3-3 protein interacts with hexokinase. We first quantified 14-3-3 paralogs and hexokinase in cell lysates to ensure that their levels were not severely depleted by the drug. This appeared to be the case (Figure 8a,b); although several treatment groups differed significantly from controls, the difference was <20%. We used a co-immunopulldown (co-IP) strategy to recover and quantify binding of 14-3-3 to hexokinase. Complexes or aggregates isolated by IP using magnetic beads coated with antibody to the conserved 14-3-3 core were recovered, resuspended in hot loading buffer, and electrophoresed on acrylamide gels, and their western blots were probed with antibody to hexokinase or the 14-3-3 conserved core. Co-IP of hexokinase (normalized to 14-3-3 recovery) was reduced >45% by treatment with drugs specific to the binding of the interface between 14-3-3 and hexokinase (Figure 8c,d). Acrylamide gels were loaded with 50 μg total protein per lane extracted from SH-SY5Y-APPSw cells after a 48 h exposure to FDA-approved drugs lumacaftor or conivaptan. Western blots were probed with antibodies to the 14-3-3 conserved core, hexokinase, or GAPDH. (b) Mean ± SD band intensities of hexokinase lanes (N = 3) normalized to the corresponding 14-3-3 band. (c) Antibody to the 14-3-3 conserved core was used for immunopulldown of 14-3-3 complexes from neuroblastoma cells either treated for 48 hr with drugs or untreated. Gel lanes were loaded with protein recovered from coated magnetic beads and electrophoresed, and western blots were probed to detect hexokinase or 14-3-3 proteins. (d) Histogram of hexokinase recovery (signals on western blots as in (c)), normalized to 14-3-3; the band signal is plotted for mean ± SD band intensity ratios of replicates. Significances according to two-tailed heteroscedastic t-tests (b,d), each N = 4: * p = 0.05; ** p < 10 -2 ; *** p < 10 -3 .

Discussion
Previous studies have documented the critical roles played by 14-3-3 proteins in diverse neurological and other age-associated disorders [1,2,5]. In eukaryotes, 14-3-3 paralogs are conserved adapter proteins involved in multiple physiological processes, such as signal transduction, translation, protein trafficking, and apoptosis [21]. Deficiencies in specific 14-3-3 paralogs in knockout mice result in neurotransmitter imbalance and altered behavior that has been likened to schizophrenia [22].
In the current study, we used computational methods to predict disordered regions of 14-3-3 paralogs that fail to attain stable conformations on their own, resulting in inde- (a) Acrylamide gels were loaded with 50 µg total protein per lane extracted from SH-SY5Y-APP Sw cells after a 48 h exposure to FDA-approved drugs lumacaftor or conivaptan. Western blots were probed with antibodies to the 14-3-3 conserved core, hexokinase, or GAPDH. (b) Mean ± SD band intensities of hexokinase lanes (N = 3) normalized to the corresponding 14-3-3 band. (c) Antibody to the 14-3-3 conserved core was used for immunopulldown of 14-3-3 complexes from neuroblastoma cells either treated for 48 hr with drugs or untreated. Gel lanes were loaded with protein recovered from coated magnetic beads and electrophoresed, and western blots were probed to detect hexokinase or 14-3-3 proteins. (d) Histogram of hexokinase recovery (signals on western blots as in (c)), normalized to 14-3-3; the band signal is plotted for mean ± SD band intensity ratios of replicates. Significances according to two-tailed heteroscedastic t-tests (b,d), each N = 4: * p = 0.05; ** p < 10 −2 ; *** p < 10 −3 .

Discussion
Previous studies have documented the critical roles played by 14-3-3 proteins in diverse neurological and other age-associated disorders [1,2,5]. In eukaryotes, 14-3-3 paralogs are conserved adapter proteins involved in multiple physiological processes, such as signal transduction, translation, protein trafficking, and apoptosis [21]. Deficiencies in specific 14-3-3 paralogs in knockout mice result in neurotransmitter imbalance and altered behavior that has been likened to schizophrenia [22].
In the current study, we used computational methods to predict disordered regions of 14-3-3 paralogs that fail to attain stable conformations on their own, resulting in indeterminate (or partner-determined) structures. These paralogs are rich in basic amino acid residues, aromatic amino acids, and amphipathic amino acids relative to acidic amino acids. We observed an increased sequestration of 14-3-3 proteins into sarkosyl-insoluble aggregates during aging and also in disease states both in heart and brain [15,23,24]. In order to target protein-protein interaction interfaces for the treatment of neurodegenerative diseases, including Alzheimer's disease, we proposed to use small molecules as protein-protein interaction inhibitors to counteract aggregate progression by breaking critical interactions needed for aggregate growth.
We first identified 14-3-3 interacting partners by performing crosslinking analysis of hippocampal aggregate proteins isolated from Alzheimer's disease vs. age-matched controls. A total of 85 interacting proteins were associated with 14-3-3 only in tissue from AD, whereas only 26 protein partners were unique to controls. Previous studies showed that 14-3-3θ (theta) acts as a chaperone to assist in refolding of disordered proteins, such as α-synuclein, thus reducing its toxicity; 14-3-3θ-overexpressing mice are protected from toxic effects of α-synuclein fibrils [25].
Since disruption of 14-3-3 protein results in the development of diabetic cardiomyopathy [28] and neurodegeneration [22,25,[29][30][31], we used an approach that targeted protein-protein interactions unique to the disease state. Pharmacological disruption of protein-protein interactions should not affect the normal biological functions mediated by either of the interacting proteins, thus holding the promise of very limited side effects. The validity of our approach was supported by the screening of an FDA-approved drug library, leading to the discovery of existing drugs that could be repurposed to prevent specific pro-aggregative interactions of 14-3-3 paralogs (See Supplementary Figure S1). This provides an express route to novel therapeutics available immediately but, at the same time, offers the promise that screening of large structural libraries of small molecules is likely to lead to even better drug candidates with highly specific (and non-essential) targets. Such screens will be the subject of a forthcoming paper.
In conclusion, we demonstrated that drugs targeting the interfaces of 14-3-3 paralogs with their interacting partners show promise for the reduction of aggregation and the improvement of associated physiological functions. Such disease-specific protein-protein interaction inhibitors have the potential to prevent, slow, or reverse aggregation associated with neurodegenerative diseases and other age-progressive disorders.

siRNA Knockdowns of Human Cells and Thioflavin T Staining to Quantify Amyloids
Exponentially growing cultures of SY5Y-APP Sw (human neuroblastoma) cells were trypsinized and replated at 8000-10,000 cells/well in 96-well plates and grown for 16 h at 37 • C in DMEM + F12 (Life Technologies; Carlsbad, CA, USA) medium supplemented with 10% (v/v) fetal bovine serum. When cells reached~40% confluence, siRNAs were transfected by lipofection (RNAiMax, Life Technologies; Carlsbad, CA, USA) to target genes encoding ANK3 (SAS1_Hs_00065571), YWHAZ (SASI_Hs01_00210839), YWHAG (SASI_Hs01_00201711), KIF5C (SAS1_Hs_ 00065571), PLEC (SAS1_Hs_00039321), or NUCL (SAS1_Hs_00217223), all obtained from Millipore-Sigma (St. Louis, MO, USA) and used as directed by the manufacturer. Transfected cells were assayed for protein aggregation 48 h later by fixation in 4% formaldehyde and staining with 0.1% w/v Thioflavin T. After four washes in PBS, cells were covered with Antifade + DAPI (Life Technologies) and fluorescence was captured in both blue and green channels (Keyence fluorescence microscope with motorized stage) for automated well-by-well imaging, with nine fields per well. Thioflavin T fluorescence intensity was divided by the number of DAPI + nuclei to obtain normalized values (amyloid fluorescence per cell), summarized as means ± SD.

Molecular Dynamic (MD) Simulations of 14-3-3 Paralogs
The X-ray crystallographic structures of 14-3-3 paralogs (14-3-3S/σ, 14-3-3G/γ, 14-3-3Z/ζ) were obtained from the PDB database (www.rcsb.org; accessed on 25 May 2021) or modeled using I-TASSER (https://zhanggroup.org/I-TASSER/ accessed on 2 June 2021), an online server that uses fold-recognition and ab initio modeling [35,36]. To understand the dynamic behavior of protein structure, target proteins interacting with 14-3-3 paralogs (e.g., hexokinase for 14-3-3G) were simulated using Schrödinger Maestro (version 11.9.011). Each protein went through a preparation stage of preprocessing with default parameters, including pH, using Maestro protein preparation wizard, after which an orthorhombic box was created around the protein with a minimized volume that varied with the protein. The system was first neutralized using the appropriate salt concentration (Na + , Cl − ), and a further 0.15M NaCl was added to the system to mimic the physiological conditions. The temperature and pressure were held constant (300 • K; 1.1023 bar). A random seed number was entered prior to starting each simulation, which was maintained for 200 ns (or as indicated) and replicated ≥3 times with new seeds.

MD Simulation Identifies a Druggable Target for 14-3-3G Interaction with Hexokinase
Docking studies were conducted using Hex, an interactive protein-docking and molecular-superposition program. Hex identifies the most stable interaction of two proteins (i.e., with the most negative ∆G). The complex of 14-3-3G with hexokinase was exported and simulated using Schrödinger Maestro. After MD simulation, the complex was captured, usually at an RMSD "plateau". The druggable pocket or interface was determined using the Discovery Studio Receptor Cavities plug-in to predict likely druggable sites.

High-Throughput Computational Screen to Identify Novel PPII Molecules to Disrupt the 14-3-3G-Hexokinase Interaction
The interface at which 14-3-3G and hexokinase interact was screened against two structural drug libraries, each at three successive stages of increasing stringency. The FDAapproved drug library (https://www.fda.gov/drugs accessed on 8 June 2021) was prepared using Biovia Discovery Studio and was virtually screened as follows. The first screen was conducted with SiBiolead (www.sibiolead.com accessed on 12 June 2021) running highthroughput screening with AutoDock. This program accepts proteins in PDB format as inputs, along with amino acid numbers that define a grid box around the targeted binding region. The user specifies the library to screen and initiates the search. The time to completion depends on the sizes of the library and the protein.
We retrieved the top 10% of drugs from the first-stage high-throughput docking screen as inputs for second-stage docking at higher stringency. This stage used high-precision Glide docking implemented within Schrödinger Maestro. Inputs for Glide docking included a protein specified in PDB format and ligands in *.mae format. To convert ligands in the library from *.mol2 to *.mae format, we used the Ligprep plug-in under Maestro, which works in "batch" mode, returning ligands as a single file in *.mae format. Glide docking predicts binding poses for a drug with high precision. Grid boxes were formed around the receptor region by specifying amino acid residue numbers; default values were used for all other parameters.
The top 10% of drugs from the second-stage screen were advanced to stage three, in which we used the MM-GBSA plug-in (molecular mechanics with generalized Born surface area) from Schrödinger Maestro to predict binding free energies for all ligands based on inputs from Glide docking, employing core and ligand settings. The output lists top candidates for specific binding to a target, which can be either a protein or a protein-protein interface (Supplementary Figure S1).

Statistical Analyses
For assays in which N was <10, differences between control and experimental groups were assessed for significance by the Fisher-Behrens heteroscedastic t-test (appropriate to samples of unequal or unknown variance) [37]. In some cases, the significance of experimental reproducibility was also evaluated in this way, treating each experiment as a single point. For N > 10, two-tailed t-tests were utilized unless 1-tailed tests were justified by previous results. Within experiments, differences in proportions (fractional paralysis or chemotaxis) were assessed by chi-squared or Fisher exact tests, as appropriate based on sample counts. For assays with multiple end-points, the threshold for significance was adjusted to p < 0.01 to reduce the frequency of type I errors.

Patents
A patent application is in preparation for the drugs described herein.
Supplementary Materials: The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/ijms232415457/s1, Figure S1: FDA-approved drug screening.  Data Availability Statement: Full data will be made available to the research community upon request.