Analysis of AD-associated proteome changes by quantitative proteomics
To investigate brain proteome alterations associated with sporadic AD, we analyzed brain samples from eight clinically and neuropathologically characterized AD patients and eight age-matched control subjects (Additional file 1: Table S1). Proteins were extracted from the dorsolateral prefrontal cortex tissues of these individuals by using the detergent sodium dodecyl sulfate (SDS) because it is the most effective reagent for solubilizing tissues and cells to achieve complete extraction of proteins [87]. We used a recently developed, filter-aided sample preparation (FASP) method [85, 87] for detergent removal and protein digestion to obtain high-purity peptides from the brain samples. Subsequent LC-MS/MS proteomic analysis using the high-resolution high-mass-accuracy LTQ-Orbitrap Elite mass spectrometer identified a total of 39,819 distinct peptides, corresponding to 6679 unique proteins. Due to stochastic nature of “shotgun” label-free quantitative proteomics, protein identification or abundance data are sometimes missing in certain samples [35]. The proteins with missing data in any sample were excluded in our analysis, resulting in the final quantification of 1968 proteins with complete data across all 16 brain samples from AD and control cases (Additional file 2: Table S2).
Differential expression analysis identifies proteins with altered abundance in AD
We performed differential expression analysis of quantitative proteomics data using the thresholds of ±1.3-fold change in AD over the control (P < 0.05) and identified 487 differentially expressed proteins (262 up-regulated proteins and 225 down-regulated proteins) in AD at FDR < 0.11 (Fig. 1a and Additional file 2: Table S2). Unsupervised hierarchical clustering analysis based on the protein abundances in the 16 individual brain samples showed that the identified differentially expressed proteins can serve as a proteomic signature for distinguishing AD versus control cases (Fig. 1b). The heat map illustrated an overall reproducibility as well as individual heterogeneity of protein expression profiles among different subjects within the AD or control group (Fig. 1b).
The list of the identified dysregulated proteins in AD (Additional file 2: Table S2) includes a number of proteins that have been previously shown by our group and others to be differentially expressed in AD brain, such as DJ-1, APOE, clusterin (CLU), and UCH-L1 [1, 17, 19, 55]. In addition, our proteomic analysis also identified 322 novel proteins that have not been previously reported as differentially expressed in AD, such as serine/threonine protein kinase 39 (STK39) and DIABLO/Smac (Additional file 3: Table S3). To validate our proteomic results, we performed Western blot analysis of STK39 and Smac expression in AD and control brains (Fig. 1c-g). We found that, in accordance with the proteomic data (Additional file 2: Table S2), STK39 protein level was significantly decreased in AD versus control (Fig. 1c, d). STK39 is an important kinase that has been associated with hypertension, Parkinson’s disease, and autism [50, 67, 84]. Our results indicate, for the first time, a link between STK39 and AD. In addition, our Western blot analysis also validated Smac, a key regulator of apoptosis [40], as an up-regulated protein in AD brain (Fig. 1e-f). Together, these results provide support for the robustness of our label-free quantitative proteomic analysis.
Next, we performed gene ontology (GO) enrichment analysis of the identified differentially expressed proteins to gain insights into the cellular functions and biological processes that are affected in AD brain (Fig. 2; Additional file 4: Table S4). We found that down-regulated proteins in AD were significantly enriched with GO categories linking to ion transport, mitochondrial function, synaptic transmission, myelin sheath, cell-cell adhesion, cytoskeleton organization, and endocytosis, whereas up-regulated proteins in AD were overrepresented with GO terms associated with metabolic process, immune response, cell-cell adhesion, exocytosis, vesicle-mediated transport, response to oxidative stress, translation, and regulation of apoptotic signaling (Fig. 2; Additional file 4: Table S4).
Co-expression network analysis uncovers AD-associated protein network alterations
To gain systems-level insights into the brain proteome changes in AD, we performed protein co-expression network analysis by using WGCNA, a data-driven network approach which uses pairwise correlation relationships of proteins and their topological overlap to organize the proteome into a network of biologically meaningful modules of co-expressed proteins [45, 90, 92]. We applied WGCNA to our entire proteomic data set of all proteins with no missing values (n = 1968 proteins) and constructed a protein co-expression network from protein expression profiles across all AD and control samples. Our WGCNA analysis identified 24 network modules of strongly co-expressed proteins (Fig. 3a; Additional file 5: Table S5). These modules, color coded according to the convention of WGCNA [45, 92], were labeled M1 to M24 based on the module size, ranging from the largest (M1: 223 proteins) to the smallest (M24: 30 proteins) (Fig. 3b). We found that several modules were significantly enriched for brain-specific GO categories, including mitochondria and synaptic transmission (M4), neuron part (M6), nervous system development (M7), myelin sheath and axonal organization (M12), and action potential (M24), whereas other modules were associated with GO categories linked to discrete cellular structures and functions, such as proteostasis and RNA homeostasis (M1), metabolism and lipid homeostasis (M2), cell morphogenesis (M3), mitochondria and cell adhesion (M5), hormone activity (M8), membrane assembly (M9), ion and protein transport (M10), signaling and cytoskeleton regulation (M11), hydrolase activity (M13), ribosome (M14), immune response (M15), inflammatory response (M16), and extracellular region (M17) (Fig. 3b; Additional file 6: Table S6).
To identify disease-relevant modules associated with AD phenotypic traits, we assessed the module-trait relationships by determining the biweight midcorrelations between each module eigenprotein (the module representative which summarizes protein expression profiles in the module [32]) and various disease-related traits or sample variables (Fig. 4). We identified 11 modules that were significantly correlated with AD status, amyloid plaque pathology (frontal cortex neuritic plaque frequency), and/or neurofibrillary tangle pathology (Braak stage), including 5 positive correlated modules (M1, M2, M15, M16, and M19) and 6 negatively correlated modules (M4, M5, M10, M11, M12, and M13). None of the modules showed significant correlation with age, gender, ApoE genotype, or postmortem interval (Fig. 4), confirming that the identified AD-correlated modules are not due to any of the potential confounding factors. Our analysis showed that most of the positively correlated modules (M1, M2, M15, and M16) had significantly increased module expression levels in AD (Fig. 5b), whereas most of the negatively correlated modules (M4, M5, M10, M11, and M13) had significantly decreased module expression levels in AD (Fig. 5c).
We then assessed the inter-modular relationships by performing eigenprotein network analysis as described [32, 44] to construct a higher-order meta-network based on pairwise correlation relationships of module eigenproteins. The module eigenprotein meta-network revealed the inter-modular connectivity of 24 co-expression modules in brain proteome, showing a hierarchical organization of highly interconnected modules into meta-modules, i.e., groups of highly correlated module eigenproteins (Fig. 5a). Interestingly, the eigenproteins of all modules positively correlated with AD phenotypes (M1, M2, M15, M16, and M19) were clustered in a single meta-module (Fig. 4 and Fig. 5a), suggesting close relationships among the pathways and processes associated with these positively correlated modules. In addition, we identified a meta-module containing eigenproteins from 5 out of the 6 modules negatively correlated with AD phenotypes (M4, M5, M10, M11, and M13), indicating that the corresponding pathways and processes for these negatively correlated modules may also be related.
AD-associated network modules and hub proteins reveal multiple dysregulated pathways in AD brain
Highly connected hub nodes are central to a network’s architecture and function [2, 7], and intramodular hub proteins in disease-related WGCNA modules have emerged as key targets for biomarker and therapeutic development [12, 27, 33, 46, 54, 82, 88]. Intramodular hub proteins can be identified by using module membership (kME), a measure of intramodular connectivity [32, 46]. The top 10 highly connected hub proteins for each of the identified AD-related modules are shown in the center of network plots (Figs. 6 and 7). Unsupervised hierarchical clustering analysis based on the hub protein expression profiles showed that the identified top hub proteins serve as a molecular signature to differentiate AD and control cases (Fig. 8c). We found that the top hub proteins of the modules with positive correlation to AD phenotypes were often up-regulated in AD (Fig. 8a,c), whereas the top hub proteins of the negative correlated modules were often down-regulated in AD (Fig. 8b,c), consistent with the proposed role of hub proteins as key drivers of protein co-expression modules [32, 33]. We assessed the molecular and functional characteristics of each AD-associated module based on its top hub proteins and gene ontology enrichment analysis of module proteins to gain insights into the biological roles of AD-related modules (Additional file 6: Table S6).
Our analyses revealed that M1, the largest module positively correlated with AD phenotypes (Fig. 4), was significantly enriched with GO categories and hub proteins linked to pathways that control protein homeostasis, or “proteostasis” (Fig. 6 and Additional file 6: Table S6), including 11 protein translation machinery components (EIF2S2, EIF3A, EIF4A2, EIF4B, RPLP1, RPL3, RPL10, RPS6, RPS7, RPS14, and RPS17) with 40S ribosome subunit RPS7 as a top hub protein; 19 molecular chaperones and cochaperones (AHSA1, CDC37, BAG5, CANX, DNAJA2, DNAJA3, DNAJC13, FKBP4, ERO1A, GNB4, GANAB, PDIA3, PFDN5, PFDN6, TBCD, CCT4/TCP1-delta, CCT5/TCP1-epsilon, CCT6A/TCP1-zeta-1, and CLU) with Hsp70 cochaperone DNAJA2 as a top hub protein; and 11 proteasome complex components (PSMA2, PSMC1, PSMC2, PSMC4, PSMC6, PSMD1, PSMD12, PSMD13, PSMD14, RAD23A, and RAD23B) with 26S proteasome regulatory subunit PSMD1 as a top hub protein (Fig. 6 and Additional file 5: Table S5). The overrepresentation of the proteostasis machinery components in this AD-related module supports the involvement of dysregulated proteostasis in AD pathophysiology [43, 78, 89].
In addition to proteostasis, RNA homeostasis-related proteins and pathways were also enriched in the M1 module, as demonstrated by the presence of 12 ribonucleoproteins involved in RNA processing (HNRNPC, HNRNPK, HNRNPL, ALYREF, GCN1L1, SSB, NPM1, LUC7L3, TROVE2, EFTUD2, RUVBL1, and SNRPE) with heterogeneous nuclear ribonucleoprotein K (HNRNPK) as a top hub protein (Fig. 6 and Additional file 5: Table S5). Our finding of HNRNPK, a major RNA-binding protein which functions in regulation of transcription, RNA splicing, mRNA stability, and translation [9], as an up-regulated M1 hub protein in AD (Fig. 8) reveals a previously unrecognized role of HNRNPK in AD pathophysiology. Corroborating with our results, another related M1 module member, HNRNPC, has been reported to be increased in AD and promote APP translation [10, 66]. Additionally, we identified pro-apoptotic factors HTRA2 and AIFM1 as top hub proteins up-regulated in AD (Fig. 6, Additional file 2: Table S2, and Additional file 5: Table S5), indicating enhanced apoptotic signaling is another key feature of this module.
The relevance of the M1 module to AD is further strengthened by its association with APOE and CLU (Fig. 6), two well-established, genetic risk factors for sporadic AD [22]. Our analyses showed that both APOE and CLU proteins were up-regulated in AD (Additional file 2: Table S2) and had high intramodular connectivity values (Additional file 5: Table S5), supporting their role as important determinants of M1 module functions. In addition, we found the fat mass and obesity-associated protein FTO, an AD risk factor which genetically interacts with APOE [38, 68], was the most highly connected hub protein of the M1 module (Fig. 6 and Additional file 5: Table S5). FTO, a demethylase which regulates 6-methyladenosine modifications of mRNAs, has also been linked to increased risk for obesity and type 2 diabetes [52]. Another M1 hub protein, SORBS1 (Fig. 6), which functions in insulin signaling, has also been associated with obesity and type 2 diabetes [51]. The finding of obesity and diabetes-associated FTO and SORBS1 as top hub proteins in AD-related M1 module is consistent with increasing evidence indicating the presence of shared pathways in the pathogenesis of AD, obesity, and diabetes [65].
M2, a 152-member module with positive correlation to AD phenotypes (Fig. 4), was highly enriched with GO categories, enzymes, and hub proteins linked to metabolic processes and pathways (Fig. 6 and Additional file 6: Table S6). The most prominent feature of this module is the presence of over 40 proteins that function in the carboxylic acid metabolism with serine racemase (SRR) and enolase 1 (ENO1) as top hub proteins (Fig. 6 and Additional file 5: Table S5). SRR, an enzyme for catalyzing the conversion of L-serine to D-serine (an essential co-agonist of the NMDA receptor) [15], was up-regulated by more than two folds in AD (Additional file 2: Table S2), which may lead to over-activated NMDA receptors, thereby contributing to AD pathophysiology. The M2 module was also highly enriched with proteins involved in the unsaturated fatty acid metabolic process (ACAA1, ACOX1, EPHX2, HSD17B4, LTA4H, PTGDS, PTGR1, PTGR2, GSTM2, GSTM3, GSTP1, and MIF), highlighting a link between dysregulated unsaturated fatty acid metabolism and AD pathophysiology. Furthermore, the M2 module was also significantly enriched with regulators of lipid metabolism (AGK, ACAA2, ALDH3A2, ANXA1, ANXA2, ANXA4, ANXA5, ASAH1, APPL2, DBI, ESYT1, GM2A, HADHA, INPP1, PAFAH1B3, ERLIN2, SLC44A2, PCYT2, PLCD3, and PRDX6) with annexin A5 (ANXA5) as a top hub protein (Fig. 6 and Additional file 5: Table S5). These findings provide new insights into the molecular basis of dysregulated lipid homeostasis in AD brain [26, 60].
The identified top M2 hub proteins also include all three members of the ezrin-radixin-moesin (ERM) family, ezrin (EZR), radixin (RDX), and moesin (MSN), which were up-regulated in AD (Fig. 6 and Additional file 2: Table S2), suggesting a role of ERM proteins in AD. The ERM proteins are FERM (4.1 protein, ezrin, radixin, moesin) domain-containing proteins that function as plasma membrane–cytoskeleton linkers to regulate membrane dynamics, cell adhesion, migration, signal transduction, and immune response [64]. Interestingly, another FERM domain-containing protein, FERMT2, was also identified as an up-regulated M2 hub protein with high intramodular connectivity (Fig. 6, Additional file 2: Table S2, and Additional file 5: Table S5). Our finding, together with the reports of FERMT2 as a genetic risk factor for AD [22] and a modulator of APP metabolism and tau neurotoxicity [16, 72], supports the involvement of FERMT2 in AD pathogenesis.
M15, a 57-member module positively correlated with AD phenotypes (Fig. 4), was significantly enriched with GO terms and proteins linked to immune response (ALCAM, ALAD, GAPDH, CYB5R3, DDX3X, CAPN1, PPIA, PYGB, EIF2AK2, CAB39, TTR, PDAP1, HIST1H2BK, QARS, VAPA, and PNP) with GAPDH, PPIA, CYB5R3, and PYGB as top hub proteins (Fig. 6 and Additional file 6: Table S6). Our finding of PPIA, CYB5R3, and PYGB, which are associated with neutrophil activation in immune response [31], as up-regulated M15 hub proteins (Fig. 6 and Additional file 2: Table S2) supports a role of neutrophil-dependent immune response in AD pathophysiology [91]. The enrichment of several aminoacyl-tRNA synthetases for protein translation (SARS, QARS, and NARS) in this module (Fig. 6) is in agreement with AD-associated, protein translation alteration identified in the M1 module.
M16, a 55-member, AD-positively correlated module (Fig. 4), was significantly enriched with GO categories and proteins linked to inflammatory response (C3, FHL1, A2M, CD44, FN1, HP, SERPINA1, and SERPINA3) with complement C3 and alpha-2-macroglobulin (A2M) as top hub proteins (Fig. 6 and Additional file 6: Table S6). Our finding of C3 and A2M, two key components of inflammatory response [42, 63], as up-regulated hub proteins in AD brain (Fig. 6 and Additional file 2: Table S2) supports their potential as candidate AD biomarkers [39] and the link between neuroinflammation and AD pathogenesis [14]. Another key molecular feature of M16 module is significant overrepresentation of extracellular matrix proteins (COL6A1, COL6A2, COL6A3, COL18A1, LAMA5, FLNA, and FLNB) as hub proteins (Fig. 6), providing evidence for the involvement of extracellular matrix dysfunction in AD [49].
M19, a 48-member module with positive correlation to AD phenotypes (Fig. 4), was highly enriched with GO terms and proteins linked to small GTPase-mediated trafficking and signaling (RAB1A, RAB1B, RAB3C, RAB4A, RAB4B, RAB6B, RAB8B, RAB10, RAB12, RAB14, RAB23, RAB35, ARF4, and ARF5) with Rab GTPases RAB1A, RAB1B, RAB4A, and RAB4B as top hub proteins (Fig. 6 and Additional file 6: Table S6). The enriched Rab and ARF GTPases function as key regulators of the following trafficking pathways: ER-to-Golgi transport (RAB1A and RAB1B), synaptic vesicle exocytosis and neurotransmitter release (RAB3C), endosome-to-plasma membrane recycling (RAB4A, RAB4B, RAB23, and RAB35), intra-Golgi traffic and endosome-to-Golgi transport (RAB6B), trans-Golgi network (TGN)-to-plasma membrane transport (RAB8B, RAB10, RAB12, RAB14, ARF4, and ARF5), Golgi-to-ER retrograde transport (ARF4 and ARF5), and autophagosome formation (RAB1A, RAB1B, RAB12, and RAB23) [24, 34, 75]. In addition, this module also contained endocytic trafficking regulators, NECAP1 and SORT1 (Fig. 6). The enrichment of the vesicular trafficking machinery components in the AD-correlated M19 module highlights the dysregulation of multiple trafficking pathways in AD brain.
M4, a 119-member module with negative correlation to AD phenotypes (Fig. 4), was highly enriched with GO categories and proteins linked to mitochondrial processes (Fig. 7, Additional file 5: Table S5, and Additional file 6: Table S6), including the mitochondrial contact site and cristae organizing system (MICOS) components (IMMT/MIC60 and CHCHD3/MIC19), mitochondrial import machinery components (TOMM70 and TIMM9), mitochondrial carrier system components (SLC25A3 and SLC25A12), mitochondrial inner membrane fusion GTPase OPA1, electron transport chain subunits (MT-ND5, NDUFA12, NDUFS5, NDUFS7, SDHA, COX4I1, COX5B, COX6C), and mitochondrial ATP synthase subunits (ATP5A1, ATP5B, and ATP5J). The identification of IMMT and SLC25A3 as down-regulated hub proteins in AD brain (Fig. 7 and Additional file 2: Table S2) reveals a previously unrecognized role of the MICOS system and mitochondrial carrier system in AD pathophysiology. The M4 module was also enriched with GO categories and proteins associated with synaptic structure and function (Fig. 7, Additional file 5: Table S5, and Additional file 6: Table S6), including key presynaptic proteins involved in synaptic vesicle trafficking and neurotransmitter release (SNAP25, STX1B, SYP, SV2B, VAMP1, RPH3A, and RAB3GAP2), glutamate receptor subunits (GRIA2/GluR2 and GRIA4/GluR4), GABAA receptor β1 subunit (GABRB1), and postsynaptic density proteins (SYNGAP1, DLG1/SAP97, DLG3/SAP102, and SRGAP3). The finding of these synaptic proteins in the AD-down-regulated M4 module is consistent with a loss of synaptic function in AD [28].
M5, a 106-member module negatively correlated with AD phenotypes (Fig. 4), was highly enriched with GO terms and proteins linked to oxidative phosphorylation (MT-ND3, NDUFA2, NDUFA7, NDUFAB1, NDUFB11, NDUFS1, NDUFS8, CYC1, UQCRC1, COX5A, COX6B1, and ATP5D) with MT-ND3 and UQCRC1 as top hub proteins, synaptic cell adhesion (CADM2/SynCAM2, CADM3/SynCAM3, NCAM1, LSAMP, NTM, OPCML) with CADM2/SynCAM2 and LSAMP as top hub proteins, synaptic vesicle exocytosis (STX1A, CPLX2, and SYT12), and signal transduction (PPP1CB, MARK2, PAK1, PRKCE, PRKCG, SLK, STK39, RHOA, RHOB, RHOC, RHOG, and CDC42) with PPP1CB, RHOA, and STK39 as top hub proteins (Fig. 7, Additional file 5: Table S5, and Additional file 6: Table S6). These findings further support the involvement of impaired mitochondrial and synaptic functions and dysregulated signaling in AD pathophysiology.
M10, a 62-member module with the most significant negative correlation to AD phenotypes (Fig. 4), was highly enriched with ion-transporting ATPases, such as Na+/K+ ATPase subunits (ATP1A2, ATP1B1, and ATP1B2) for establishing the electrochemical gradients of Na and K ions across the plasma membrane and the H+-transporting, vacuolar ATPase subunit ATP6V1G2 for lysosomal acidification (Fig. 7 and Additional file 6: Table S6), supporting a loss of brain cell ion homeostasis in AD pathogenesis [20, 81]. The M10 module was also significantly enriched with GO terms and proteins linked to transmembrane transport and vesicle-mediated transport (Fig. 7 and Additional file 6: Table S6), including mitochondrial protein import (MTX2), endocytosis (DNM2, DNM3), endosome-to-lysosome trafficking and synaptic vesicle biogenesis (AP3D1), ER-to-Golgi transport (BCAP31), intra-Golgi trafficking (USO1, NAPA, NAPG), TGN-to-plasma membrane transport (AP1M1), endosome-to-plasma membrane recycling (SNX27), and autophagy (GABARAPL2, RTN3, and UBQLN1). The enrichment of these transport machinery components in the AD-down-regulated M10 module indicates impairment of multiple transport pathways in AD brain.
M11, a 62-member module negatively correlated with AD phenotypes (Fig. 4), has heterotrimeric G-protein subunits (GNAI1, GNAI2, GNAI3, and GNAO1) and Src family of tyrosine kinases (FYN and YES1) as top hub proteins (Fig. 7), highlighting the involvement of altered intracellular signaling in AD pathophysiology. In addition, the M11 module was significantly enriched with GO terms and proteins linked to regulation of actin cytoskeleton (CRMP1, CRMP4/DPYSL3, ABI1, FSCN1, and WASF1) and microtubule cytoskeleton (TUBAL3, TBCB, NDRG1, NDRG2, SHTN1, and SNCG) (Fig. 7 and Additional file 6: Table S6), consistent with impaired actin and microtubule dynamics in AD brain [5, 25]. The association of the retromer complex components (SNX1 and SNX2) with the M11 module supports a link between retromer dysfunction and AD pathogenesis [73].
M12, a 61-member module with negative correlation to neurofibrillary tangle pathology but not amyloid plaque pathology (Fig. 4), is characterized by highly significant enrichment of GO terms and proteins linked to myelin sheath (CNP, MAG, MBP, OMG, PLP1, MOG, PMP2, CLDN11, and ERMN) and the organization of paranodal and juxtaparanodal regions of axon at the node of Ranvier (MAG, ERMN, CNTNAP1, CNTN2, and KCNA2) with CNP, MAG, OMG, and PLP1 as top hub proteins (Fig. 7 and Additional file 6: Table S6). These results, together with our finding of OMG and PLP1 as down-regulated hub proteins in AD (Fig. 7 and Additional file 2: Table S2), support the involvement of myelin degeneration, impaired myelin-axon interactions, and node of Ranvier dysfunction in AD pathogenesis [8]. The M12 module was also significantly enriched with neurofilament proteins (NEFL, NEFM, and INA) and microtubule-binding proteins involved in the control of microtubule polymerization or stabilization (CRYAB, MAPRE1, DST, CRMP2/DPYSL2, CLASP2, and MAP1B) and axonal transport (DCTN1 and DCTN4), indicating an association of impaired neurofilament and microtubule functions with Tau aggregation in AD. Our finding of BIN1, the second most prevalent genetic risk factor for sporadic AD [22], as a member of the M12 module with negative correlation to neurofibrillary tangle pathology (Fig. 7) is consistent with recent evidence indicating that BIN1 negatively regulates the propagation of Tau pathology [13].
M13 is a 61-member module down-regulated in AD (Fig. 5) with negative correlation to AD phenotypes (Fig. 4). More than one-third of proteins in this module are associated with hydrolase activity, represented by top hub proteins such as deubiquitinating enzyme OTUB1, small GTPase RAC1, adenosylhomocysteinase-like proteins AHCYL1 and AHCYL2, protein tyrosine phosphatase PTPRZ1, and platelet-activating factor acetylhydrolase subunit PAFAH1B1 (Fig. 7 and Additional file 5: Table S5). In addition, the M13 module was also significantly enriched with GO terms and proteins linked to organization of the actin cytoskeleton (ARPC1A, CFL1, CFL2, CTTN, RAC1, PAFAH1B1, NF1, SPTAN1, and SPTBN2) and microtubule cytoskeleton (MAPRE3, KLC2, NIT2, TBCA, TUBA8, MAP7D1) (Fig. 7 and Additional file 6: Table S6), supporting the involvement of impaired actin and microtubule dynamics in AD pathophysiology [5, 25].