Postmortem brain tissues derived from the hippocampus, CA1 region of the hippocampal cortex (21), entorhinal (Brodmann area 34) and perirhinal cortices (Brodmann area 35) (22) were dissected out by the micropunch technique as previously described (23). The histopathological analysis and disease staging were based on the presence of the NFTs, neuropil threads, and neuritic plaques, following standardized criteria as previously described (24). A total of 56 tissue samples from eleven AD (Braak stage III-IV) and twelve control brains were analyzed ( Table S1 ). All brain samples were obtained from the Human Brain Tissue Bank (Semmelweis University, Budapest, Hungary). Human brain microdissection procedures were approved by the Regional and Institutional Committee of Science and Research Ethics of Scientific Council of Health (34/2002/TUKEB-13716/2013/EHR, Budapest, Hungary) and the Code of Ethics of the World Medical Association (Declaration of Helsinki).

The frozen tissue samples were sectioned using a cryostat (LEICA CM1950), and ten to twenty 10 μm slices were collected in Eppendorf tubes for proteomic analysis. Samples were resuspended and vortexed in 100-300 μL of protein extraction buffer (50 mM dithiothreitol (DTT), 2% sodium dodecyl sulfate (SDS), 100 mM tris(hydroxymethyl)aminomethane hydrochloride (Tris-HCl) pH 8.6). The lysates were incubated at 95°C for 5 min with constant shaking at 500 rpm (Thermomixer Comfort, Eppendorf AG), and then sonicated in the Bioruptor Plus UCD-300 (Diagenode) (40 cycles, 15s on/off) followed by centrifugation at 20,000 g for 20 min at 18°C (Thermo Scientific Sorvall ST 8R Centrifuge). The supernatants were collected and transferred to new tubes. The proteins were alkylated with 100 mM iodoacetamide for 30 min in the dark at room temperature (RT), and then precipitated using 9 volumes of cold ethanol overnight at -20°C. The pellet was washed three times with 1 mL of 90% ethanol, then dried at room temperature and preserved at -20°C until further use.

The protein pellet was dissolved in a solution of 1% SDS, 0.5% sodium deoxycholate (SDC), 100 mM tetraethylammonium bromide (TEAB) (pH 8.0). For acetylation stoichiometry analysis, the reagent N-acetoxy-succinimide-d3 (NAS-d3) was synthesized as described previously (25–27). The unmodified lysine residues were chemically acetylated by incubation with a 100-fold molar excess of NAS-d3 in dimethyl sulfoxide at RT for 1h. This labeling step was repeated once more. The samples were finally incubated with 5% hydroxylamine for 20 min to revert unspecific reactions such as the O-acetylation of hydroxylated amino acids. The samples were precipitated using cold ethanol as described before to remove the excess of reagents and then resuspended in 0.5% SDC, 50 mM ammonium bicarbonate (pH 8.0). Protein amounts were estimated using the BCA protein determination assay (BCA™ Protein Assay Kit).

The proteins were digested using trypsin (enzyme: substrate, 1:50) at 37°C overnight (Sequencing Grade Modified Trypsin, Promega, Madison, WI, USA). After trypsinization the peptides were cleaned up using ethyl-acetate extraction (1:1, samples/solvent) under acidic conditions (final concentration of trifluoroacetic acid (TFA) was 0.5%). The organic phase was discarded, and an additional cleaning step was conducted using ethyl acetate (1:1). The organic phase was discarded, and peptides were dried in a speed-vac for 5 min to evaporate the remaining solvent. The peptides were resuspended in 2% acetonitrile (ACN) with 0.1% TFA and peptide determination was performed using the Pierce™ Quantitative Colorimetric Peptide Assay. The samples were used for global proteome and acetylome analyses.

For phosphoproteome analysis, the protein pellet was dissolved in a solution of 0.5% SDC, 100 mM ammonium bicarbonate (pH 8.0), and the proteins were digested using trypsin (enzyme: substrate, 1:50) at 37°C overnight. After digestion, the peptides were cleaned up using ethyl-acetate extraction as described before. The peptides were resuspended in 40 μL of 0.1% TFA and peptide concentration was determined using the Pierce™ Quantitative Colorimetric Peptide Assay. For phosphopeptide enrichment, 55 µg of peptides were submitted to an automatic workflow on the AssayMAP Bravo system (Agilent Technologies) using the Fe(III)-IMAC assay as previously described (Murillo et al, 2018). The phosphopeptides were resuspended in 2% ACN with 0.1% TFA before nLC-MS/MS analysis.

One microgram of the peptides was injected for each global proteome analysis and the total amount of enriched peptides was used for phosphoproteome analysis. The samples were analyzed on a Q-Exactive HF-X mass spectrometer coupled to an UltiMate 3000 RSLCnano system (Thermo Scientific). Peptides were concentrated on an Acclaim PepMap™ 100 (75 μm × 2 cm, nanoViper) trap-column and separated on a PepMap RSLC C18 (2 μm, 100 Å, 75 μm x 25 cm) analytical column. The nanoLC system was operated at a flow rate of 300 nL/min and the column temperature was set to 45°C. The solvents used for the nonlinear gradient were A (0.1% formic acid) and B (0.08% formic acid in 80% ACN).

For total proteome and acetylome analysis, the gradient started with 2% solvent B and increased to 27% during 112 min, in the next 10 min solvent B increased to 35%, then to 50% in 7 min and finally, it augmented to 90% in 8 min and it was kept at 90% for 5 min. Samples were analyzed using a top 20 data-dependent acquisition (DDA) method, the spray voltage was set to 1.85 kV. MS1 full scans were acquired with 120,000 (@ 200 m/z) resolution, target AGC value of 3e6, and maximum injection time of 100 ms. For MS2 analysis the 20 most intense ions were selected for higher-energy collisional dissociation (HCD) with an NCE of 28. MS2 spectra were acquired with 15,000 (@ 200 m/z) resolution, target AGC value of 1e5, and maximum injection time of 50 ms. The ion selection threshold was set to 1.6e5 and the dynamic exclusion to 40 s.

The phosphoproteome analysis was performed using a gradient starting with 4% solvent B and increased to 27% during 120 min, in the next 15 min solvent B increased to 45% and finally, it augmented to 98% in 1 min and it was kept at 98% for 5 min. Samples were analyzed using a top 15 DDA method, the spray voltage was set to 1.85 kV. MS1 full scans were acquired with 120,000 resolution, target AGC value of 3e6, and maximum injection time of 50 ms. For MS2 analysis the 15 most intense ions were selected for fragmentation with an NCE of 25. MS2 spectra were acquired with 60,000 resolution, target AGC value of 1e5, and maximum injection time of 120 ms. The ion selection threshold was set to 7.2e3 and the dynamic exclusion to 30 s.

Raw files from global proteomics were analyzed in Proteome Discoverer 2.2 software environment (Thermo Scientific) using SEQUEST HT as search engine against the UniProtKB human database (version from 01/15/2019, canonical and isoforms, 42,356 sequences). The following search parameters were employed: Arg-C as the cleavage enzyme, carbamidomethylation at cysteine residues as a static modification, methionine oxidation, acetylation [normal (d0) and heavy (d3)] at lysine residues, and at protein N-terminus as dynamic modifications, 10 ppm tolerance for precursor ions and 0.02 Da for fragment ions. A maximum of 2 missed cleavages were allowed and the minimum peptide length was set to six amino acids. The filter High Confidence (FDR q value = 0.01) was applied on both peptide and protein levels.

For the phosphoproteome analysis, the same software, protein database, and search engine were used with the following search parameters: Trypsin as the cleavage enzyme, carbamidomethylation at cysteine residues as a static modification, phosphorylation of serine, threonine and tyrosine, oxidation of methionine and acetylation at protein N-terminus as dynamic modifications, 10 ppm tolerance for precursor ions and 0.02 Da for fragment ions. A maximum of 3 missed cleavages were allowed and the minimum peptide length was set to six amino acids. The ptmRS algorithm was used for scoring phosphorylation sites considering site probability threshold >75. The filter High Confidence (FDR q value = 0.01) was applied on both peptide and protein levels.

Lysine acetylation identification and stoichiometry calculations were performed with the Pview software (28). For peptide identification, the maximum tolerance was set to 10 ppm for precursor and 0.02 Da for fragment ions, and up to 1% FDR was allowed. For the stoichiometry calculations, the isotopic tolerance was set to 3.5 ppm and at least four peaks in the XIC, allowing only one missing point in between. The MS proteomic data have been deposited to the ProteomeXchange Consortium via the PRIDE (29) partner repository with the data set identifier PXD027173.

Pre-processing was performed for each brain region individually within the Perseus software (version 1.6.2.3) (30) both for global proteomics and phosphoproteomics. Technical replicates from phosphoproteomic analysis were averaged by taking their median. For qualitative comparison of the disease groups, we considered proteins, phosphopeptides and acetylated peptides that have at least 70% valid values in one and less than 10% valid values in the other condition (AD and control). For quantitative analysis, proteins and phosphopeptides with minimum 66% valid values in each disease group (AD and control), plus “on-off-like” proteins (less than 33% valid values in one group and at least 66% valid values in the other) were kept for further processing. Intensities were log2-transformed and median normalized, followed by replacing the missing values from a normal distribution (width = 0.3, downshift = 1.8). The phosphorylation state of proteins was calculated based on phosphopeptide intensities corrected for total protein abundance.

Further statistical analyses were conducted in R (version 4.0). Sex, age and postmortem intervals (PMI) were compared between the groups (AD and control) in each region using Fisher’s exact test (for binary data) and Wilcoxon rank sum test (for continuous numerical values) to determine whether these variables differ between groups. The tests showed that the disease groups were unbalanced with respect to age, a tendency for higher age in AD group was observable in HP and CA1, and the Wilcoxon rank sum test was significant in EC and PRC ( Figure S1 ). Linear regression analysis was performed to address the proteomic, phosphoproteomic and phosphorylation status-related changes associated with disease category. Proteins and phosphopeptides with p-value < 0.05 and log2 fold change ≥ |0.7| were considered statistically significant. To filter out the effect of sex, age and PMI in each brain region, proteins and phosphopeptides that were significantly affected by these covariates (linear regression analysis, p-value < 0.05) were excluded from the biological interpretation of the results. Data derived from the global proteome and phosphoproteome analyses were combined to detect the kinases present in our dataset.

To follow AD progression through the comparison of different brain regions, log2-intensities of each protein and phosphopeptide in the AD samples were subtracted by the median log2-intensity of the same protein and phosphopeptide in the control samples. These log2 fold change values were then compared across regions using an ANOVA test followed by a Tukey HSD post-hoc test. Proteins and phosphopeptides with significant ANOVA p-value (p-value < 0.05) were then subjected to hierarchical clustering (complete linkage, Euclidean distance). The tree cutting was performed via the cutreeDynamic function from the dynamicTreeCut R package (version 1.63-1) with the following settings: minClusterSize = 100, method = hybrid, deepSplit = 1. Furthermore, pre-ranked Gene Set Enrichment Analysis via the clusterProfiler R package (vs. 3.18.1) (31) was performed on the log2 fold change differences across the regions, utilizing the Hallmark (32) and the ImmuneSigDB gene sets (33).

Data pre-processing for the acetyl stoichiometry analysis was performed in Excel and in R. The stoichiometry calculations were performed as reported previously (34). Acetylation ratios were calculated by taking the sum of endogenously acetylated peptides’ intensity divided by the sum of all peptides’ intensity (sum of endogenously and chemically acetylated). For comparison of the peptides’ acetylation ratio between disease groups, a Wilcoxon test was applied on peptides with at least 40% valid values in each disease group. Spatial progression of the brain areas was not addressed at the acetylation level due to the high percentage of missing values.

Additional bioinformatics analyses were performed using Reactome (35), and visualizations were done utilizing Cytoscape (36), Coral (37), and KEGG mapper (38).

To dissect the proteomic profiles of AD vulnerable brain areas in the limbic stage, we developed an analytical workflow ( Figure 1 ) that integrates global and phosphoproteome analysis, including analysis of both phosphopeptide abundance and phosphorylation status of peptides (see Methods section), and acetylation stoichiometry of peptides. Table 1 summarizes the total number of identified and quantified protein groups, phosphopeptides and acetylated peptides derived from global-, phosphoproteome and acetylome analyses, respectively in each AD-brain area.

When comparing disease groups, we first focused on qualitative analysis of the data (see Methods section) considering the proteins predominantly represented in one condition (i.e., on-off proteins), and then performed quantitative analysis of each region and dissection of spatial progression of AD across brain vulnerable areas.

The global proteome and PTM analysis conducted so far assessed the individual characteristics of each brain area vulnerable to AD, comparing the imbalances in protein expression and PTM status between AD and control cases. Our data revealed the predominant dysregulation of biological pathways related to the immune response, APP processing, Tau phosphorylation, cytoskeleton remodeling, and mRNA processing. These protein changes have a unique biological signature in each AD brain region. However, to provide a deep understanding of the brain region vulnerability in AD, we focused on comparing the fold changes/ratios (AD vs. control) in protein and phosphopeptide abundances, as well as in phosphorylation and acetylation status of the proteins/peptides identified across all brain areas. In this analysis, changes from HP to CA1 followed by EC and PRC were revealed. This approach allows us to find molecular trajectories shaped by alterations in protein abundances and variations in their PTM status that are potentially associated with the spatial progression of the disease.

Cluster analysis of the global proteome data ( Figure 6A and Supplementary Table S7 ) showed 121 proteins with an increased expression profile, mainly localized in clusters 1 and 4 ( Figure 6A ). Cluster 1 revealed a gradual increase in protein expression primarily related to acetyl-CoA and amino acid metabolism (e.g., protein JTV-1, glycine amidinotransferase, and the mitochondrial 2-oxoisovalerate dehydrogenase subunit beta), especially serine metabolism (e.g., 3-phosphoglycerate dehydrogenase, cytoplasmic serine-tRNA ligase). In the same cluster, we observed proteins, such as hnRNPD0, involved in mRNA processing. The progressive increase of critical regulators of the immune response, such as E3 ubiquitin-protein ligase Itchy homolog ( Figure 6B ), a key regulator of Th2, as well as proteins involved in T-cell receptor activation, NF-κB pathway (e.g., core-binding factor subunit beta, pirin), “eat-me” signals [e.g., multiple epidermal growth factor-like domains protein 10 (MEGF10)], and microglia/macrophages markers (e.g., monocyte differentiation antigen CD14) were also identified. Cluster 4 was characterized by increasing levels of proteins linked to PI3K/AKT signaling (e.g., GRB2-associated-binding protein 2, fibroblast growth factor receptor 3), metabolism (e.g., glutamate dehydrogenase, nicotinamide phosphoribosyltransferase, fatty-acid amide hydrolase), and APP processing such as cathepsin D ( Figure 6B ). In addition, proteins related to the innate and adaptive immune response (e.g., butyrophilin subfamily 3 member A3, multiple EGF-like domains protein 10) and antigen presentation (MHC class II antigen DRA) were also found in the same protein cluster. Lastly, our analysis uncovered 77 proteins with decreasing expression profile, most of them grouped in cluster 3. These proteins are mainly involved in synaptic function (e.g., glutamate ionotropic receptor, gamma-aminobutyric acid receptor, synaptotagmin-7) ( Figure 6B ), and in the co-activation of the transcriptional activities of several nuclear receptors such as PPARG (e.g., nuclear receptor coactivator 7), an essential promoter of the sustained activation of the M2 anti-inflammatory profile (44).

Phosphoproteome analysis revealed a progressive increase in the phosphopeptide abundances and phosphorylation status of 82 and 60 proteins, and a decrease of 167 and 119 proteins, respectively. Phosphorylation status analysis displayed 100% overlap with the phosphopeptide abundance data ( Supplementary Table S7 ). Protein clusters with increased phosphorylation profiles are mostly distributed in three clusters, i.e., clusters 4, 5 and 7 ( Figure 6C ). Cluster 4 includes proteins that are associated with pre-mRNA processing events (e.g., SRm160, CD2 tail-binding protein 2), positive regulation of NF-κB signaling (SAM and SH3 domain-containing protein 1) and maintaining immune self-tolerance (e.g., SAPS domain family member 3). Cluster 5 is characterized by oxidative deamination (e.g., amine oxidase), cytoskeleton regulation (e.g., neurofilament medium polypeptide, filamin-A, amphiphysin II), T-cell activation (e.g., leukocyte common antigen) and the increment of CD45, a hematopoietic cell marker (45). In cluster 7, we found proteins involved in the Rho GTPases pathway (e.g., PH-interacting protein, Rho GTPase-activating protein 44) and the TGF-beta signaling pathway modulation (e.g., protein PML). We also detected a continuous increase in the phosphorylation status of well-established AD markers such as Tau and glial fibrillary acidic protein ( Figure 6D ) across the four brain regions. Proteins from clusters 2, 3 and 6 showed decreasing phosphorylation profiles ( Figure 6C ). Cluster 2 is composed of proteins from the CaMK family ( Figure 6D ) that mediates the phosphorylation of CREB (e.g., CAMK2A, CAMK2D, CAMK2B), mTOR signaling (e.g., S6K-beta-1), and negative regulation of NF-κB (e.g., hKSR2). Vesicle transport (e.g., clathrin heavy chain, DNAJC6), ubiquitination (e.g., E3 ubiquitin-protein ligase HUWE1, E3 ubiquitin-protein ligase RNF31), modulation of Wnt signaling (e.g., cyclin-dependent kinase 14), alternative splicing of mRNAs by proteins such as RNA-binding protein 15 ( Figure 6D ), are the main biological pathways related to cluster 3. Finally, cluster 6 is represented by proteins that modulate the actin cytoskeleton (e.g., Abl interactor 2), APP processing (e.g., amyloid-beta A4 precursor protein-binding family A member 2), and mRNA metabolism (e.g., Pumilio homolog 2, putative RNA-binding protein 15B).

The proteomic profiles of vulnerable brain areas during the AD limbic stage revealed an imbalance of proteins related to the immune response. Currently, the hypothesis that links microbial insults to immune system activation in AD, which ultimately leads to the production of antimicrobial peptides (AMPs), has been strengthened (46). To explore the distribution of proteins with antimicrobial activities across the four brain regions and their changes in expression and/or post-translational modification status, we compared our complete proteomic data set with the recently published UDAMP database (47). A total of 39 AMPs were detected, 26 of which were found in all brain regions ( Figure 7A ). The HP and CA1 regions display the same AMP profiles while EC and PRC exhibited exclusive proteins with AMP functions (e.g., proline-rich protein 4 in EC, and neutrophil defensin 1, C-X-C motif chemokine 14, bone marrow stromal antigen 2 and neutrophil gelatinase-associated lipocalin in PRC) ( Figure 7B ). Despite this overlap, the data shows differences in changes in their relative abundance when comparing AD with controls across different brain areas.

Figure 7C highlights the interaction networks and dysregulation pattern of proteins with AMP function in each vulnerable region of AD. In particular, the most prominent alterations occur in brain areas with pronounced pro-inflammatory proteomic profiles, such as HP. Three AMP proteins showed increased abundances in HP, such as lactotransferrin, cathepsin G, and zinc-alpha-2-glycoprotein. Also, in HP the phosphorylation status of the pro-neuropeptide Y stands out in AD samples compared to controls. The phosphorylated form of ribonuclease K6 was found exclusively in AD samples from the CA1 region, but not in controls. Conversely, no significant changes in phosphorylation level or global protein expression were observed between AD and controls in EC. A decrease in the abundance of protein S100-A9 was observed in AD samples from the PRC. Taken together, the data reveals that proteins with reported AMP functions are similarly distributed in different vulnerable AD areas in the limbic stage of the disease, but the changes in their abundances are more evident in the hippocampus.