Protein Kinase Activity Decreases with Higher Braak Stages of Alzheimer’s Disease Pathology

Abstract

Alzheimer’s disease (AD) is characterized by a long pre-clinical phase (20–30 years), during which significant brain pathology manifests itself. Disease mechanisms associated with pathological hallmarks remain elusive. Most processes associated with AD pathogenesis, such as inflammation, synaptic dysfunction, and hyper-phosphorylation of tau are dependent on protein kinase activity. The objective of this study was to determine the involvement of protein kinases in AD pathogenesis. Protein kinase activity was determined in postmortem hippocampal brain tissue of 60 patients at various stages of AD and 40 non-demented controls (Braak stages 0-VI) using a peptide-based microarray platform. We observed an overall decrease of protein kinase activity that correlated with disease progression. The phosphorylation of 96.7% of the serine/threonine peptides and 37.5% of the tyrosine peptides on the microarray decreased significantly with increased Braak stage (p-value <0.01). Decreased activity was evident at pre-clinical stages of AD pathology (Braak I-II). Increased phosphorylation was not observed for any peptide. STRING analysis in combination with pathway analysis and identification of kinases responsible for peptide phosphorylation showed the interactions between well-known proteins in AD pathology, including the Ephrin-receptor A1 (EphA1), a risk gene for AD, and sarcoma tyrosine kinase (Src), which is involved in memory formation. Additionally, kinases that have not previously been associated with AD were identified, e.g., protein tyrosine kinase 6 (PTK6/BRK), feline sarcoma oncogene kinase (FES), and fyn-associated tyrosine kinase (FRK). The identified protein kinases are new biomarkers and potential drug targets for early (pre-clinical) intervention.

INTRODUCTION

Alzheimer’s disease (AD) is an irreversible, progressive neurodegenerative disorder and the most prevalent form of dementia. Memory loss and cognitive decline due to synaptic failure and neuronal death are clinical symptoms of AD. The disease is further characterized by a long pre-clinical phase (20–30 years), during which significant brain pathology manifests itself [1, 2]. The identification of causative genes and risk factors as well as the characterization of clinical and pathological features of AD during the last 25 years support the existence of a long pre-clinical phase. However, the disease mechanisms involved in the early stages of AD pathology remain elusive.

The modification of proteins by reversible phosphorylation is a key mechanism in the regulation of a large number of physiological processes. Abnormal protein kinase activity can cause diseases by altering the phosphorylation of those proteins that are critical for normal cellular and metabolic processes. The level of protein phosphorylation is controlled by the opposing activities of protein kinases and phosphatases. Accumulating evidence reveals a role of protein kinases in the brains of AD patients [3]. Protein kinases such as glycogen synthase kinase 3β (GSK3β), p25/Cyclin-dependent kinase 5 (Cdk5), dual-specific tyrosine regulated kinase 1A (Dyrk1A), and mitogen-activated protein kinases (MAPKs) increase in activity and/or expression, while protein kinases such as protein kinase C (PKC) and protein kinase A (PKA) decrease in activity [4]. An increasing number of studies support a role of aberrant protein phosphorylation in the progression of AD, by the altering of cellular processes such as subcellular localization, ligand binding, protein folding, and protein-protein interaction. For example, both hyperphosphorylation of tau and amyloid-β (Aβ) production contribute to AD pathology [3, 5, 6]. However, most studies focus on the changes in activity or phosphorylation of a single protein kinase. So far, overall changes in protein kinase activity in human brain tissue during the progression of AD have not been investigated. Changes in the phosphorylation state of proteins, which are independent of changes in total expression level, will provide insights into molecular pathways and mechanisms such as synaptic transmission and neuroplasticity [7].

The objective of this study was to determine the changes in protein kinase activity during AD pathogenesis. To that end, frozen hippocampal brain sections were selected from patients in all stages of AD and in non-demented controls (Braak stages 0-VI; N = 100) and the protein kinase activities were determined using a peptide-based microarray platform (PamChip ^®). This platform allows for the investigation of protein kinase activity of small amounts (μL) of protein lysates by monitoring changes in phosphorylation of the peptides on the arrays [8–10].

The results revealed an overall decrease in protein kinase activity; a decrease was already evident at early stages of the disease (Braak I-II), i.e., at a point in time preceding the clinical manifestation of AD. The phosphorylation of 139 Serine/Threonine (STK) peptides and 54 Tyrosine (PTK) peptides (out of 142 peptides each) was decreased significantly (p-value <0.01). Using the STRING online database, a network of kinases implicated in AD were identified, e.g., Ephrin receptor A1 (EphA1), a risk gene for AD [11, 12], and sarcoma tyrosine kinase (Src), a synaptic kinase involved in memory formation [13, 14]. Additionally, protein kinases that have not yet been linked to AD pathology were found, such as protein tyrosine kinase 6 (PTK6/BRK) and fyn-associated tyrosine kinase (FRK) with STRING protein-protein network analysis and pathway analysis.

MATERIALS AND METHODS

Case selection

Human brain tissue was obtained from the Netherlands Brain Bank (NBB, Amsterdam, The Netherlands). Prior to death, all donors gave written informed consent in accordance with the Declaration of Helsinki for the use of their brain tissue and medical records for research purposes. This study was approved by the ethics committee of the NBB. Dementia status at death was determined on the basis of clinical information available during the last year of life. Neuropathological diagnosis was performed using histochemical stainings (hematoxilin and eosin, periodic acid Schiff-Luxol fast blue). Analysis of formalin-fixed, paraffin-embedded tissue from different parts of the brain, including the frontal cortex (F2), temporal pole cortex, parietal cortex (superior and inferior lobule), occipital pole cortex and the hippocampus (essentially CA1 and entorhinal area of the parahippocampal gyrus) was performed. Hippocampus and cortical areas were stained with methenamine silver [15], and Gallyas silver staining was used for neurofilament proteins. Immunohistochemistry was performed using antibodies raised against hyperphosphorylated tau (AT8) and Aβ (1–17, Dako [16]). Staging of AD pathology was evaluated according to modified assessment of Braak and Alafuzoff [17]. Severity of dementia was measured using the Global Deterioration Scale of Reisberg (GDS) [18]. Cases with and without clinical neurological disease diagnosis were processed identically. Patients with co-morbidities like Parkinson’s disease, frontotemporal dementia (FTLD), or dementia with Lewy bodies were excluded from the study. In total, 60 patients with confirmed AD and 40 non-demented controls were included, with a median age at death of 84 years for women and 79 years for men (Table 1). Age, gender, clinical diagnosis, pH of the cerebrospinal fluid (CSF), and Braak score for neurofibrillary tangles of all cases used in this study are listed in Supplementary Table 1. Postmortem delay (PMD) of all cases was between 2.5 and 15.5 hours with a median of less than 6 hours.

Table 1.

Patient characteristics

	Total	Male	Female	Range Male	Female
N	100	36	64
AD	60	19	41
Mean Age [years]	81	76	83
Median Age [years]	83	79	84	(51–39)	(50–100)
Mean PMD [hours]	5.65	6.09	5.44
Median PMD [hours]	5.20	5.35	5.18	(3.40–12.55)	(2.55–15.40)
Braak stage (tau)	Total	Male	Female	Braak stage (Aβ)	O	A	B	C	not det.
0	4	4	0		4
1	18	7	11		6	6	5	1
2	15	5	10		7	1	6	1
3	15	7	8		1	2	6	5	1
4	14	4	10				1	12	1
5	17	3	14				1	15	1
6	17	6	11					17

AD, Alzheimer’s disease; PMD, postmortem delay; Braak stage for tau (left) and amyloid-β (right); not det., not determined.

Preparation of brain tissue lysates

Twenty 10μm-thick frozen hippocampal tissue slices were taken from the hippocampus containing the subiculum and CA1-4 region. The hippocampal region was routinely divided into 5 parts from anterior to posterior at autopsy. We used the middle region (3) at the level of the corpus geniculatum laterale. Sections contained the hippocampal formation including the entorhinal region. All samples were obtained in the same medical center following a strict standardized protocol. Sections were cut and lysed at 0 °C using M-PER (Mammalian Protein Extraction Reagent, Thermo Scientific, MA, USA) lysis buffer (0.1 g/ml) containing Protease Inhibitor Cocktail (Roche, Basel, Switzerland) and Phosphatase Inhibitor Cocktail (Roche, Basel, Switzerland). After centrifugation (10 min, 4 °C, 10.000× g), the pellet was washed and re-centrifuged. The supernatants were pooled, snap frozen in 100μl aliquots and stored at –80 °C. The protein concentration was determined using the Bradford Lowry Assay (Bio-Rad Protein Assay) with bovine serum albumin (BSA) as the standard. Lysates are very susceptible to freeze-thawing which can result in up to 30% loss of kinase activity [10, 19]. Therefore, frozen aliquots were never re-frozen, but used directly for kinase activity determination.

Protein kinase activity profiling

Kinase activity profiles were determined using the PamChip ^® 96 serine/threonine (STK) and protein tyrosine (PTK) peptide microarray system from PamGene International B.V. (‘s-Hertogenbosch, The Netherlands) according to the instructions of the manufacturer, as described previously [8, 19–21].

The PamChip ^® 96 contains 96 identical arrays grouped on 24 strips each containing 4 arrays. Each array is pre-printed with either 142 tyrosine or serine/threonine containing peptides (13 amino acids long) derived from known human phosphorylation sites. The peptides are covalently attached to a porous matrix via a spacer. Brain tissue lysates are pumped up and down through the porous array in the presence of ATP to facilitate the phosphorylation of peptides by protein kinases in the lysates. The optimal sample input was determined by testing a concentration range for three out of the 100 samples.

In order to minimize the variation between plates, all PamChip ^® 96 array plates used in this study came from the same production batch and all plates were run on the same PamStation instrument. The experimental design resulted in a total of 14 STK and 14 PTK plates with a minimum of 6 technical replicates per plate. The term technical replicate refers to identical samples from the same patient using the same assay mixture but different arrays. Early and late Braak stages were distributed evenly over the 96 array plates. Several samples were analyzed on more than one plate to allow assessment of variation between experimental runs. For technical reasons, one sample was excluded from the analysis of the STK arrays (TIS138 (Braak II) and 97 samples were analyzed on the PTK arrays (excluded: TIS138 (Braak II), TIS156 (Braak V) and TIS141 (Braak VI)).

The arrays on the PamChip ^® 96 STK plate were incubated with Odyssey blocking buffer (LI-COR Biosciences, Lincoln, Nebraska, USA) for 30 cycles (15 min) to prevent non-specific binding. Plates were washed three times with kinase assay buffer (50 mM Tris-HCl pH 7.5, 10 mM MgCl₂, 1 mM EGTA, 2 mM DTT, 0.01% Brij35). The reaction mixture contained 0.01 % BSA in kinase assay buffer supplemented with anti-phospho-Ser/Thr antibodies (PamGene International BV [19]). All reaction mixtures contained 0.5% DMSO in a total assay volume of 40μl. For each STK assay, 0.5μg of protein was used, and the enzymatic assay was started by adding ATP (final concentration 100μM; Sigma-Aldrich, St. Louis, MO, USA). The incubation mixture was pumped up and down through the porous membrane for 60 cycles (in total 30 min). After washing of the arrays, they were incubated for 60 min with a secondary antibody (polyclonal swine anti-rabbit Immunoglobulin/FITC). Table 2 provides details about the antibodies. Images at 50 ms exposure time were captured every 10 min with an integrated CCD-based optical system in combination with Evolve software (version 1.5, PamGene International BV). After removal of the secondary antibody and a wash step, post-wash images were taken at different exposure times (20, 50, 100, and 200 ms).

Table 2.

Overview of STK and PTK antibodies

Name, type	Species	Epitope (immunogen)	Source
PY20-FITC	monoclonal mouse	phospho-tyrosine	AbD Serotec, Bio-Rad, Hercules, CA, USA
#9624	monoclonal rabbit	phospho-PKA substrate (RRXY*/T*)	Cell Signaling, Danvers, Massachuseets, USA
#9614	monoclonal rabbit	phospho-AKT substrate (RXXS*/T*)	Cell Signaling, Danvers, Massachuseets, USA
#2325	monoclonal rabbit	phospho-MAPKC/CDK substrate   (PXS*P or S*PXR/K)	Cell Signaling, Danvers, Massachuseets, USA
STK detection	polyclonal swine	anti-rabbit fluorescein isothiocyanate labelled	PamGene International BV

The PTK assay mixture contained the same kinase assay buffer, 100μM ATP and 0.01% BSA, supplemented with 4μl protein kinase (PK)-additive (PamGene International BV), 10 mM Dithiothreitol (DTT, Fluka, Sigma-Aldrich, St. Louis, MO, USA) and fluorescein isothiocyanate (FITC) labeled anti-phosphotyrosine antibody ((PamGene International BV, ‘s-Hertogenbosch, The Netherlands). For each PTK assay, 7.5μg of protein was used. Since a labeled antibody is present in the PTK assay mixture, peptide phosphorylation was monitored during the incubation with assay mixture, by taking images every 5 min at 50 ms exposure time, allowing real time recording of the reaction kinetics (one-step reaction). After washing of the array, fluorescence was detected at different exposure times (20, 50, 100, and 200 ms).

Signal quantification

The fluorescent signal intensity for each peptide was analyzed using BioNavigator 6.1 software (PamGene International BV, ‘s-Hertogenbosch, The Netherlands) a statistical analysis and visualization software tool with an App-based infrastructure (https://www.pamgene.com/en/bionavigator.htm). For signal quantification, the slope of the fluorescent signal versus exposure time was calculated in order to increase the dynamic range and to filter out time differences between plates. Saturated signals were excluded. Visual quality control was performed to exclude defective arrays from the analysis.

Reproducibility of sample preparation

In order to establish the reproducibility and robustness of the sample preparation and the kinase profiling method, nineteen of the one hundred cases were cut again, lysed and both tyrosine and serine/threonine activities were determined. The phosphorylation profiles for the two N = 19 datasets were compared (Fig. 2). For each peptide, the signal intensity was expressed with respect to the mean signal for the 19 samples. This centering step was used to account for differences in sample input and experimental differences.

Fig.2.

Reproducibility of sample preparation and protein kinase activity profiling. Relative signal intensity of each peptide from N = 100 study (A) and N = 19 study (B). Each column represents one case (for details about cases, see Supplementary Table 1). Centering (expressing the signal for each peptide relative to the mean for the cohort) was performed to make differences between samples more visible.

Statistical analysis

The chip design resulted in fluorescent data with a complex correlation structure. In a prior study, the plate-to-plate, strip-to-strip (within a plate) and array-to-array (within a strip) variability had been determined. We developed a linear mixed-effects model that analyzed the signals of all peptides jointly while taking the correlation structure into account, thus enabling proper inference on the fixed effects of interest, i.e., the difference in signal strength per Braak stages for each peptide.

We modeled the change of log(signals) over log(time). The obtained STK and PTK median kinase signal intensities were analyzed without correction for the local background in order to avoid additional variability in the signal measurements. The model included common effects (for all peptides) and peptide-specific plate, strip and array random effects, with strip nested in the plate and the array nested in the strip and the plate. The measurement error was modeled using a peptide-specific variance component covariance matrix that allowed for heterogeneous variances among exposure time points.

STRING protein-protein interaction analysis

The Search Tool for Retrieval of Interacting Genes/Proteins (STRING) database (http://string-db.org) was used to construct a protein-protein interaction (PPI) network. In order to keep the PPI network readable, a sub-selection of the most significant peptides was made (p-value corrected for multiple testing <10^–10, Supplementary Table 2). Those peptides (N = 122) were traced back to their corresponding protein of origin by using Basic Local Alignment Search Tool (BLAST; http://blast.ncbi.nlm.nih.gov/Blast.cgi) software. UniProtID’s of the proteins of origin (100% identity) were included in the protein data set. After removal of duplicated UniProtIDs, 107 unique UniProtIDs remained and were entered into the STRING database. The settings “multiple names” and organism “Homo sapiens” were selected. For “interactors wanted” the option “proteins” was chosen. Known and predicted associations were scored and integrated, resulting in a comprehensive protein network (Fig. 3) [22].

Fig.3.

STRING protein-protein interaction network. Network of the interactions between 107 proteins containing the most significant peptides. Stars indicate highly connected proteins with a focus on kinases. Different colors represent different kinds of evidence of connection between proteins. Green represents neighborhood, red gene fusion, bright blue co-occurrence, black co-expression, pink experimental evidence, turquoise database evidence, light green evidence from text mining, and violet homology between the two proteins.

Pathway analysis

MetaCore^TM from Thomson Reuters was used for pathway analysis. The same dataset as for the STRING analysis (107 UniProtIDs) was used as input. UniProtIDs are mapped onto pathways that are constructed from curated literature data. The top 50 pathways were then grouped according to processes indicated in the pathway name.

Identification of upstream kinases

The upstream protein kinases able to phosphorylate residues in peptides on the PTK and STKarrays were identified in the Human Protein ReferenceDatabase (http://www.hprd.org) [23], in Phosphosite(http://www.phosphosite.org) and Reactome (http://www.reactome.org) [24]. These kinases were projected on the kinase phylogenetic tree using the Kinome Render tool (http://bcb.med.usherbrooke.ca/kinomerender.php). When databases used different names to indicate a kinase, the kinase names were converted to those used in Kinome Render via their UniProtID. For kinases linked to multiple UniProtIDs, only the ID used in the Kinome Render tool was retained.

RESULTS

A large cohort (N = 100) of AD patients in all Braak stages (n = 60) and non-demented controls (n = 40) was selected for this study (Table 1). All human hippocampal brain samples originated from the same medical center. Possible confounding factors are listed in Supplementary Table 1. No significant correlations were found between those factors. We observed a downward trend with increasing Braak stage for pH of the CSF (R² = 0.176) and for brain weight (R² = 0.314) (data not shown).

Decreased protein kinase activity was observed at early stages of AD

For 100 well-characterized cases of patients with AD and non-demented controls (CON) at all different stages (Braak 0-VI), freshly frozen hippocampal brain tissue extracts were prepared and protein kinase activity was determined using the serine/threonine kinase (STK) and protein tyrosine kinase (PTK) PamChip ^® arrays (see Supplementary Table 1 and Supplementary Figure 3). All brain tissue samples showed protein kinase activity (Supplementary Figure 3). The experimental workflow is represented in Supplementary Figure 1. The mean value per peptide for each sample was calculated, and the data was visualized in a heatmap (Supplementary Figure 3). Peptides were sorted by the signal intensity per peptide for the first sample (highest intensity on top, TIS140, Braak 0). The mean signal intensity per Braak stage was calculated (Fig. 1).

Fig.1.

Phosphorylation profiles (mean ²log signal intensity per Braak stage) for STK (A) and PTK (C). Red indicates high signal intensity, i.e., high protein kinase activity, while green and blue represent moderate and low signal intensity, i.e., low protein kinase activity. The box plots show the range of median ²log signal intensity for each sample per Braak stage for STK (B) and PTK (D). The boxes contain the values of 50 % of the samples, 25% above and 25% under the median, with the horizontal line representing the median of the data. The spacings between the different parts of the box indicate the degree of dispersion (spread) and skewness in the data. The length of the whiskers is 1.5 times the interquartile range. Whiskers reflect the variation, while outliers would be plotted as individual points.

As a result of nonspecific binding, the background of the PTK array was higher because a 15-fold greater protein concentration (for PTK 7.5μg protein per array, for STK 0.5μg/array) is required compared to the STK assay. This could indicate that brain extracts contain more STK kinase activity than PTK kinase activity. It is also known that STK phosphorylation is more abundant (98% ) than tyrosine phosphorylation, which accounts for less than 2% of the kinome [25].

Decreased phosphorylation was observed for AD (Braak VI) compared to controls (Braak I) in both the STK and the PTK assays (Fig. 1 and Supplementary Figures 2 and 3). The overall signal intensities of the STK arrays were higher (Fig. 1A, B) than of the PTK arrays (Fig. 1C, D). No phosphorylation signal was visible in the absence of ATP (Supplementary Figure 2). For each sample, the signal intensities for STK and PTK arrays did not correlate with each other (R² = 0.052), e.g., a high signal intensity on the STK chip does not imply high signal intensity on the PTK chip as well.

To investigate whether age, disease duration, gender, ApoE genotype, and brain weight correlated with the signal intensity of the samples, we performed continuous one-way ANOVA analyses with each one of these factors as a continuous factor. None of the factors correlated significantly with the observed phosphorylation signal intensity, neither for the PTK nor the STK assays (Supplementary Figure 4), except for brain weight. In addition, we investigated whether there was a correlation between PMD and the kinase activity. The samples were sorted by (increasing) PMD (Supplementary Figure 5A and B, respectively). One-way ANOVA analysis with PMD as continuous factor resulted in three peptides with a p-value <0.01 (Supplementary Figure 5). When the same analysis was done with the pH of CSF, a significant correlation between the pH of the CSF and both STK and PTK signal intensity was found (Supplementary Figure 6A and B, respectively).

Statistical analysis: Identification of peptides that show significantly altered phosphorylation profiles at different Braak stages

The mixed effects model (Materials and Methods) was fitted to the STK and PTK phosphorylation data separately. The mean peptide-specific change in signals at the Braak stages was consistent with a linear decrease. For instance, the mean change between Braak stages 0 and I was similar to the mean change between Braak stage IV and V. Hence, Braak stage was included as a continuous, peptide-specific fixed effect. This resulted in an average signal (one value) per Braak stage per peptide. In this way, the model provided one estimated effect for changes in signal intensity over Braak stage for each peptide (Supplementary Table 2). The estimates represent the estimated mean effect sizes on a binary log scale with associated p-values. Therefore, the more negative the estimate, the bigger the decrease in protein kinase activity. A p-value below 0.01 was considered significant. An estimate of 1 on the binary log scale was treated as no effect. Due to the testing of all peptides simultaneously, reported p-values are corrected for multiple testing by controlling their False Discovery Rate (FDR) at 5% .

139/142 STK peptides and 54/142 PTK peptides showed a significant decrease in phosphorylation with Braak stage (p-value <0.01 and Supplementary Table 2). These peptides are derived from proteins with a wide variety of functions, e.g., transcription factors, structural proteins, protein kinases, cell adhesion proteins, ion channels, and transport proteins. Nopeptides were observed to have a significant increase of phosphorylation with increasing Braak stage.

Reproducibility of sample preparation

The reproducibility of sample preparation was assessed by repeating the cutting, lysis, and kinase activity determination for 19 samples selected at random. Most of the repeat samples (N = 19) gave similar signals as the initial samples (N = 100, Fig. 2). Only 3 samples from Braak V and Braak VI showed small differences, which are most likely due to cutting tissue from a slightly different section of the hippocampal tissue block. The data showed that sample cutting, lysis, and the protein kinase activity profiling have a high reproducibility.

Protein-protein interaction network (STRING analysis)

Building interaction networks of proteins is beneficial for understanding their role in complex biological pathways. Since in most molecular pathways serine/threonine kinases and tyrosine kinases occur inthe same pathway and sometimes even interact, we combined the two datasets into one. In order to keep the network readable, a selection was made of the most significant peptides (p-value corrected for multiple testing <10^–10, Supplementary Table 2), resulting in a selection of 122 peptides that changed the most over Braak stages. The corresponding 107 proteins from which those peptides were derived were used for the STRING analysis. The outcome of the STRING analysis is shown in Fig. 3. Each protein is represented as a node with edged interactions. Selected proteins are marked with a star and discussed in this manuscript. They correspond to the outcome of the Metacore^TM pathwayanalysis (Fig. 4) and identification of upstream kinases (Fig. 5). Protein kinases such as Src and Janus kinase 2 (JAK2) are highly connected, whereas Myristoylated alanine-rich C-kinase substrate (MARCKS) and Vasodilator-stimulated phosphoprotein (VASP) are at the edges of the network. Different colors represent different kinds of evidence of connections between proteins.

Fig.4.

The 50 most significant MetaCore^TM pathways grouped in processes.

Fig.5.

Phylogenetic tree of the human kinome. Kinases known to differentially phosphorylate phospho-sites present on the PTK and STK PamChip arrays (122 peptides) are projected on the tree. The size of the circle indicates the number of sites phosphorylated (range 1–10). For reasons of visualization, the circle size does not increase with 10+ phosphorylation sites. Black circles indicate upstream kinases presented as highly connected nodes in the STRING analysis (Fig. 3). A) Upstream kinases identified from HPRD, Reactome, or Phosphosite database. B) Venn diagram for the overlap between these databases for the 122 most significantly changed peptides.

Proteins with a strong functional interactive pattern include Src, rapidly accelerated fibrosarcoma 1 serine kinase (RAF1), JAK2, mammalian target of Rapamycin (mTOR/FRAP), members of the Ephrin receptor family (EPHA1, EPHA2, EPHA7, EPHB1), nuclear factor kappa-light-chain-enhancer of activated B cells (NFκB1), phosphatidylinositol 3-kinase regulatory subunit alpha (PIK3R1), receptor tyrosine-protein kinase erbB-2 (ERBB2), microtubule-associated protein tau (MAPT), cAMP response element-binding protein 1 (CREB1), the estrogen receptor (ESR1), and 3-phosphoinositide-dependent protein kinase 1 (PDPK1). The protein kinase C beta (PKCβ, PRKCB), protein kinase CK2 (CSNK2A1) and Aurora A kinase were found in the center of the STRING network. In addition, protein kinases that are not known to be linked to AD were identified, e.g., PTK6/BRK, feline sarcoma tyrosine kinase (FES), and FRK.

Pathway analysis of top 50 pathways

A pathway analysis (MetaCore^TM) using the same protein input yielded 50 pathways likely to be affected in AD. These pathways are highly significant, as their p-values range from 2.38*10^–12 to 2.7*10^–6. A p-value of 10^–4 is considered significant. These pathways were grouped according to processes and are represented as a pie chart (Fig. 4). 52 of the 107 UniProt IDs were represented in these 50 pathways. Identification of kinases known to phosphorylate the original phospho-sites from the HPRD, Reactome, and Phosphosite databases gave 83 upstream kinases. For the Protein IDs not represented in the pathways, 39 upstream kinases were identified.

Identification of putative upstream kinases

Since a number of peptides were not represented in the pathway diagrams, we extended the upstream kinase analysis. For all phosphosites in this peptide set (N = 122) the kinases that are known to phosphorylate the sites were extracted from the HPRD, Reactome, and Phosphosite databases and visualized on the kinome phylogenetic tree (Fig. 5A). The number of kinases identified in each database and the overlap between the databases is visualized in a Venn diagram (Fig. 5B). Kinases that are present in the phylogenetic tree and the STRING network as well as the MetaCore™ pathways are marked with black circles (Fig. 5A).

DISCUSSION

In the present study, we used protein kinase activity profiling to investigate the changes in protein kinase activity associated with AD progression. We analyzed a cohort of 100 postmortem hippocampal brain tissue samples of all stages at AD and non-demented controls, with short postmortem delay and without any co-morbidities. To our knowledge, this is the first report of a study in which all Braak stages are studied in such a large cohort from one medical center [26]. Our main finding was that overall protein kinase activities decrease with disease progression. Interestingly, we found that as early as the pre-clinical phase of AD, protein kinase activities decrease, indicating the importance to target early events in the pathophysiology of AD, such as synaptic loss and inflammation. Hence, protein kinases are promising targets for (early) pharmacological intervention in AD.

In this study we used two types of microarrays to determine, in a multiplex fashion, the activity of tyrosine and serine/threonine protein kinases in hippocampal protein extracts by phosphorylation of specific peptides present on the array [8, 20]. We have shown previously that protein kinase activity profiling on peptide microarrays can be used to reveal novel protein kinases and new signaling pathways involved in AD pathology [27, 28]. McGuire et al. recently profiled protein kinases of the frontal cortex of patients with schizophrenia, revealing both a modest increase and decrease in kinase activity in patients with the disease compared to controls with no mental illness [29]. Investigating protein kinase activity, as opposed to determining the presence of a kinase, is an unbiased and more direct measure to monitor global changes occurring in the cell or tissue.

Detection of changes by generic antibodies is unbiased and does not require a priori hypotheses about what proteins could be involved. An advantage of the PamChip ^® kinase activity microarrays (PamGene International BV) is their robustness and the small amount of protein that is required (i.e., typically 2.5–5μg of total protein for one PTK array and 0.25–1μg per STK array). This is in contrast to much higher amounts required for other proteomics technologies (e.g., several milligrams of total protein required for mass spectrometry analysis [30]). The PamChip ^® 96 peptide microarray platform produced highly reproducible data (Fig. 2). We found that the development of a robust experimental design in combination with the use of highly detailed standard operating procedures for the handling of the material guaranteed highly reproducible data sets.

Brain alterations can only be examined after autopsy when diverse ante- and post-mortem events may well have damaged the living proteome. It has been reported that protein phosphorylation of human brain tissue is affected by the PMD, i.e., the time from death to extraction and freezing of the brain tissue at –80°C [31–34]. Most phosphoproteins are stable for 24 h or longer after autopsy of mouse brains [35] and human brains [10]. In addition, it has been shown that most postmortem changes of the phospho-proteome in tumor samples occur during the first 30 min after death [36] while changes after that time are far less drastic. In this study we selected brain tissue with a short postmortem delay (4–6 h) and showed that the observed changes in kinase activity of human hippocampal brain tissue cannot be explained by differences in PMD. We concluded that PMD is a random factor, not associated with Braak stage or kinaseactivity.

In this cohort age, gender, differences in brain weight (except for a few peptides) or ApoE genotype did not correlate to the differences in phosphorylation (Supplementary Figure 4). PMD is a random factor, not associated with Braak stage. The pH value might also reflect brain alterations. For brain tissue, pH values above 7.0 were associated with violent death and mean pH values below 6.5 with a slow death, i.e., following a prolonged agonal phase [37]. Unfortunately, neither pH measurements of the brain itself nor information about the agonal status of patients prior to death were available for this cohort, though data for the pH of the CSF was available. Due to the large sample size of our cohort (N = 100) and the random distribution of causes of death between Braak stages, we do not consider agonal state to be a significant factor in the outcome of this study. Hardy et al. found a correlation between cortical pH and pH of the CSF [38] which was also supported by others [39–41]. We consequently used the pH measurements of the CSF to act as a surrogate for the total brain pH. We discovered that the pH of the CSF correlated to a certain extent with Braak stage. The average pH of the CSF from patients in higher Braak stages was considerably lower (Supplementary Figure 6). A lower pH of the CSF promotes apoptosis [42] and could be associated with disease. However, not all samples with a low pH had low kinase activity. For that reason we conclude that changes in pH do not explain the variation in kinase activity between the samples. In accordance with others [43], we did not find any correlation between pH and PMD.

We observed changes in protein kinase activity from early Braak stages. In order to gain more insight into the potential biological mechanisms underlying the observed phenomenon, we used curated biological knowledge associated with the proteins represented by the detected phosphorylated peptides. We performed three different types of analysis: a protein-protein interaction network analysis, a pathway analysis, and an identification of direct upstream kinase analysis. The STRING protein-protein interaction tool allows the formation of an interaction network. The connectivity network revealed proteins with the strongest functional interacting pattern. Some of these have been implicated in AD (Src, EphA1, CK2), while for others little to nothing is known about their involvement in AD pathogenesis (PTK6/BRK, FES, Aurora A). The MetaCore tool facilitates the construction of differentially modulated pathways. The top 50 pathways revealed a profound involvement of developmental processes (44% ), signal transduction processes (16% ), and apoptosis and survival processes (10% ). Proteins like the lipid kinase PI3K and the protein kinases PDPK, RAF, PKA, PKB, PKC, and Src are present in many pathways, while other important nodes (CK2, PKC, Aurora A) are less frequently present in the pathway diagrams. Kinases known to phosphorylate the phosphorylation sites in the peptides were identified in databases (HPRD, Reactome, Phosphosite). A phylogenetic tree was created from curated literature data (HPRD, Reactome, and Phosphosite databases) using the Kinome Render tool. PKA, PKB, and the PKC family are well represented in many pathways. GSK3β appears as a kinase able to phosphorylate Myc, which is an important node in the protein-protein interaction network and connected to other regulators of cell division like p53, Aurora A and Rb. The protein kinases PKA, PKB and PKC are well represented among the upstream kinases. Several other kinases that were present as important nodes in the protein-protein interaction analysis also appear in this set of upstream kinases: Src, JAK2, RAF, mTor/FRAP, PKC, CK2 and FES. Since kinases of the same family share substrate preferences, many family members are likely to appear in the phylogenetic tree (Fig. 5). However, it is worth considering that only one family member might actually be active.

The interaction analysis, the pathway analysis and the upstream kinases analysis (Figs. 3–5A) are based on published and curated literature data. This reflects current knowledge and is biased towards known pathways and well-investigated kinases. In the following section, we will discuss a few selected protein kinases and pathways that were found with protein-protein interaction analysis, pathway analysis, and upstream kinase analysis and their potential role in AD pathogenesis.

An important pathway represented in the interaction analysis is the phosphatidylinositol 3-kinase (PI3K)/AKT/mTOR pathway, which is involved in neuronal proliferation, aging, and synaptic plasticity and is deregulated in several models of neurodegeneration [44, 45], including the Tg2576 transgenic mouse model for AD [46], murine cultured neurons [47] and brains of patients with Down syndrome [48]. In AD, synaptic plasticity decreases and neurons become dysfunctional, then visibly degenerate, and finally, synaptic connections are decreased [49]. The loss of synapses is an early event in the pathogenesis of AD and correlates best with loss of cognitive function of AD patients [50]. A significant loss and altered distribution of the major negative regulator of AKT, PTEN (phosphatase and tensin homologue deleted on chromosome 10), has been detected in the hippocampal CA1 region at end stages of AD and in human cultured AD neurons [51]. In the adult human brain, PI3K signaling is involved in the regulation of long-term potentiation (LTP) in the hippocampus, which is impaired in patients with AD. Decreased activity of the PI3K/AKT pathway may thus contribute to neuronal dysfunction and reduced synaptic plasticity. Memory consolidation by promoting cell survival via mTOR activation might be beneficial for future drug targets [45, 52].

When PDPK1 phosphorylation is reduced, downstream signaling pathways involved in insulin regulation, protein kinase B (AKT or PKB), protein kinase C (PKC) and serum- and glucocorticoid-induced protein kinase (SGK) become aberrant. Interestingly, AKT and PKC are downregulated in AD [4, 51, 53]. In this study we showed that reduction of PDPK1 phosphorylation is an early event in AD. One of the hypotheses about the possible role of insulin resistance is that downregulation of insulin receptor/PI3K/AKT pathway would lead to lowered inhibition of glycogen synthase kinase 3 beta (GSK3β) and subsequently hyperphosphorylation of tau [54]. Although the molecular mechanisms underlying insulin-dependent regulation of synaptic plasticity and cognitive function are not clearly understood, they seem to involve the activation of PI3K and MAPK signaling pathways [55]. Therefore, supplementing insulin at early stages of the disease may lead to reduced AD pathophysiology and neuronal repair due to the activation of PKC signaling via extracellular-signal-regulated kinases 1/2 (ERK1/2), MAPK, and Src [56].

Src is an important node found in the interaction analysis. We suggest that the observed decrease of Src activity in early stages of AD contributes to the loss of synaptic plasticity. In healthy individuals, Src catalyzes the induction of long term potentiation in the CA1 region of the hippocampus in an N-methyl-D-aspartate receptor (NMDAR) mediated manner [13, 57]. It has been reported that the inhibition of Src prevents the induction of long-term memory [13] and that the PKC-FAK2-Src cascade pathway is essential for memory formation [58]. PKC is downregulated in AD [4, 59] with the potential consequence of a down-regulation of the downstream Src. Together these findings suggest that activation of PKC or Src could be beneficial for synaptic strength and memory formation in patients with AD.

The erythropoietin-producing hepatocellular (Eph) receptor tyrosine kinases (RTK) contribute to aberrant synaptic function associated with neurodegeneration [60]. They are subdivided into two distinct classes (A and B), based on their structural similarity and affinity. Class A receptors are anchored to the cell surface via a glycosylphosphatidylinositol motif, whereas EphB receptors are type I transmembrane proteins [61]. We recently reported that the distribution of EphA4 RTK is altered in human hippocampal brain tissue of AD patients compared to controls [62]. This altered distribution pattern was observed at early stages of the disease suggesting a reduced availability of EphA4 is likely to contribute to synaptic dysfunction, an early event in AD pathogenesis [62]. A decrease in EphB2 and EphA4 expression was observed in hAβPPswe-ind mice, a mouse model for AD. The decrease coincided with the onset of memory decline [63]. Furthermore, a signaling cascade has been proposed in which activation of EphB receptors leads to the activation of Src and subsequently the enhancement of NMDAR tyrosine phosphorylation and thereby upregulation of NMDAR function and LTP [64]. This suggests that Src and EphB share common pathways. The decrease in protein kinase activity of several members of the Eph-receptor RTKs, which was observed in this study, is in accordance with previous reports, thereby validating our approach.

Another relevant kinase, which was present in the interaction network and the phylogenetic trees, was the protein kinase CK2, formerly called casein kinase 2. Interestingly, CK2 was one of the first protein kinases identified with abnormal activity in the AD brain [65]. Different roles for CK2 in AD have been suggested, including roles in synaptic plasticity [66, 67], AβPP processing [68–70], tau accumulation [71, 72], and insulin signaling [73]. Recently CK2 was found to be present in human astrocytes, and its involvement in neuroinflammatory events during AD pathogenesis was reported (Rosenberger et al., in preparation).

Janus kinase 2 (JAK2) is part of the JAK2/STAT3 pathway and phosphorylation of JAK2 in BV-2 microglia is downregulated by Aβ. Microglia and astrocytes are the immune cells of the brain and are thus directly involved in inflammatory responses that contribute to AD pathogenesis and progression [74–77]. Recently, hydroxyl-safflor yellow A (HSYA), a novel chalcone glycoside, was reported to upregulate the JAK2/STAT3/NF-κB pathway, thereby suppressing microglia activation and reducing the inflammatory response in mice that had Aβ_1-42 injected into the hippocampus to introduce Aβ pathology [78]. The authors showed that HSYA treatment of Aβ_1-42 mice ameliorated the spatial learning and memory deficits, by using a Morris water maze test. These results indicated that HSYA might be a promising drug target for the treatment of AD at early stages.

While the kinases discussed thus far play a role in AD that has been well-documented, we have identified additional protein kinases that have not been previously associated with AD, such as PTK6/BRK, FES, and FRK. PTK6/BRK can phosphorylate PDPK, PLCG, and RET. Interestingly, FRK phosphorylates several members of the EphA family, EPOR, Fer, FES, PI3K and PDPK. FES is known to phosphorylate PDPK, FES, and RET.

The FRK protein shows ~60% amino acid similarity with the tau kinase Fyn, though its function is not known. Fyn interacts with both Aβ and tau and plays an important role in regulating synaptic plasticity. A reduction in Fyn activation may lead to impairments in long-term potentiation and could affect cognitive function in humans. Interestingly, Fyn knockout mice have an age-dependent reduction in dendritic spines, the single contact points between an axon and a dendrite critical for synaptic function [79].

The cytosolic protein tyrosine kinase FES acts downstream of cell surface receptors and plays a role in the regulation of the actin cytoskeleton, microtubule assembly, cell attachment and cell spreading. FES has not been associated with AD. However, FES has been reported to play a role in the regulation of cell differentiation and promotes neuritic outgrowth in response to nerve growth factor signaling. The c-Fes tyrosine kinase cooperates with the breakpoint cluster region protein (Bcr) to induce neurite extension in a Rac- and Cdc42-dependent manner [80].

Another kinase recently implicated in age-related neurodegeneration (inflammation, neurovascular unit compromise, and exhaustion) is the protein kinase C (PKC). PKC isoforms have numerous functions important for memory, blood-brain barrier maintenance, and injury repair that change with aging [81]. Furthermore we identified PTK6/BRK, which has not been associated with AD before. PTK6/BRK is involved in the formation and regulation of neurons and was recently linked to focal adhesion kinase (FAK) and AKT [82]. PTK6/BRK augments proliferation and promotes cell survival in breast, colon, and skin cancers [83]. The precise role of FRK and PTK6/BRK in the early stages of AD remains to be determined. The identified protein kinases need to be investigated further to establish their functional role in AD pathogenesis and their potential as biomarkers and/or drug targets.

The Aurora A kinase, also known as serine/threonine-protein kinase 6 (STK6), is a member of a family of mitotic serine/threonine kinases. It is implicated in important processes during mitosis and meiosis and is integral for healthy cell proliferation. Aurora A is activated by one or more phosphorylations, with peak activity during the G2 phase to M phase transition in the cell cycle [84]. In the interaction analysis, several nodes involved in cell division were present: Myc, p53, and Rb. Aurora A dysregulation has been associated with high occurrence of cancer [85], but to date, no link to AD has been reported.

In conclusion, we analyzed protein kinase activity in a well-defined cohort of 100 human post-mortem brains, which included all Braak stages, and wediscovered aberrant activities even in the early stages of AD (Braak I and II) before clinical symptoms are apparent. Several processes are affected during AD, and in our analysis, we could confirm the involvement of several proteins already implicated in AD. Furthermore, kinases that have previously not yet been associated with AD were identified (Aurora A, FRK, FES, and PTK6/BRK).

We anticipate that the protein kinases that change early in the disease might be attractive drug targets. The currently available AD therapeutics, unfortunately, have minor effects on improving cognition or slowing down the progression of the disease. This may be because most therapeutics are directed against relatively late events in AD pathology, when irreversible damage has already occurred in the brain [86–88]. In the future, pharmacological interventions will have to be directed against pre-clinical events that can now be identified by protein kinase activityprofiling.

Appendix

The supplementary material is available in the electronic version of this article: http://dx.doi.org/10.3233/JAD-150429.