Aβ₂₅₋₃₅ was purchased from Sigma (Sigma, A4559). N2 supplement was purchased from Gibco (Gibco, 17502048). As an example of 2-pyrrolidone derivative compounds GFKP-19 was kindly provided by a collaboration laboratory, which wishes to not release the detailed structure of this compound along with the present paper. Aβ₂₅₋₃₅ was dissolved in DMSO at a concentration of 3 mM. GFKP-19 was dissolved in DMSO at a concentration of 10 mM. 3-(4, 5-dimethylthiazolyl-2)-2, 5-diphenyl tetrazolium bromide (MTT) was purchased from MP Biomedicals (MP Biomedicals, 02102227). The dialyzed fetal bovine serum (FBS) was from Hyclone (Hyclone, SH30079.03). The Amicon Ultra 15 mL centrifugal filter with a membrane NMWL of 30 kDa was purchased from Merck Millipore (Merck Millipore, UFC903024).

Neuro-2a cells purchased from American Type Culture Collection (ATCC, CCL-131) were cultured in DMEM media supplemented with 10% FBS, 1% penicillin/streptomycin, growing in humidified incubator at 37°C and 5% CO₂. N2 supplement (Gibco, 17502048) was used as a serum-free supplement during drug treatment. Neuro-2a cells were evenly seeded into 96-well plates with a density of 5 × 10³ cells per well. At the second day, cells were pretreated with or without GFKP-19 (0.1 or 1 μM) for 1 h, and Aβ₂₅₋₃₅ (1 μM) was added in the presence of GFKP-19 and all cells were incubated for another 24 h. Control cells were treated with corresponding concentration of DMSO.

Cell viability was measured by MTT assays (Mosmann, 1983; Ulukaya et al., 2004). After treatments above, cells were incubated with 0.5 mg/mL MTT for 4 h. Then, the supernatant was removed carefully and 150 μL DMSO was added to dissolve the formazan. The plate was shaken for 10 min, and absorbance at 490 nm was obtained with a microplate reader (BIO-RAD Model 3550).

DMEM media for SILAC (Thermo, 89985) was supplemented with 10% dialyzed FBS, 1% penicillin/streptomycin and before drug treatment, and the dialyzed FBS was replaced by N2 supplement during drug treatment. The light SILAC amino acids L-lysine (Sigma, L8662) and L-arginine (Sigma, A8094), the medium SILAC amino acids L-lysine-²H₄ (Sigma, 616192) and L-arginine-¹³C₆ (Sigma, 643440), or the heavy SILAC amino acids L-lysine-¹³C615N₂ (Sigma, 608041) and L-arginine-¹³C615N₄ (Sigma, 608033) were added with a final concentration of 28 mg/L for the arginine and 50 mg/L for the lysine amino acids to generate light, medium, and heavy SILAC-labeling media. Neuro-2a cells were labeled as previously published with light, medium and heavy SILAC-labeling media (Ong and Mann, 2006). After eight passages, full incorporation of SILAC amino acid was confirmed.

Light SILAC-labeling cells were taken as controls, whereas medium SILAC-labeling cells were treated with Aβ₂₅₋₃₅ (1 μM) for 4 or 24 h, and heavy SILAC-labeling cells were pretreated with GFKP-19 for 1 h followed exposure to Aβ₂₅₋₃₅ (1 μM) and GFKP-19 for another 4 or 24 h. All cells were, respectively, lysed in buffer containing 4% SDS, 0.1M DTT, 100 mM tris (pH 7.5). The protein concentrations were evaluated by tryptophan fluorescence emission assay (Kulak et al., 2014) and evaluated by SDS-polyacrylamide gel electrophresis (SDS-PAGE). 100 ng/μL tryptophan was used to build a standard calibration curve (0, 0.125, 0.25, 0.5, 0.75, 100, and 150 μL), and 1 μL of sample was diluted in 200 μL of 8 M urea for reading protein amount. An excitation wavelength of 295 nm was adopted to determine protein concentrations whereby tryptophan fluorescence emission at 350 nm. The assaying instrument used was SpectraMax M5 (Molecular Devices, Sunnyvale, CA). Protein extracts derived from different treatments were then mixed at the ratio of 1:1:1 at the protein level. A total of 15 mg protein mixture was alkylated with iodoacetamide and digested by trypsin following the filiter-aided sample preparation method (FASP) reported (Wisniewski et al., 2009). The enzymatic hydrolysis efficiency was around 40~50%. The peptides were subjected to TiO₂ enrichment to get phosphopeptides as described below.

Phosphopeptides were enriched by using Titansphere-chromatography (TiO₂) as described (Thingholm et al., 2006; Selenica et al., 2007) with minor modifications. Briefly, 1 mg of slurry of 5 μm Titansphere material (GL Sciences) in 80% ACN/1% TFA containing saturated glutamic acid was added directly to the 1 mg tryptic peptides. The slurry was incubated for 15–30 min at room temperature with end-over-end rotation, spunbriefly on a bench-top centrifuge, and the supernatant was collected. Titansphere material with phosphopeptides bounded was washed two times with 400 μL 30% ACN/0.5% TFA and 2 times with 400 μL 70% ACN/0.1% TFA, then transferred on top of a C8 disc (Empore) placed in 200 μL pipette-tip. Bound phosphopeptides were eluted by passing 1 times 50 μL 0.6% NH₄OH/20% ACN and 2 times 50 μL 1% NH₄OH/40% ACN through the beads. Elutes were then dried almost to completeness. Normally the yield is 1% of the original peptides input. All phosphopeptides got at this step were subjected to fractionation.

Samples were separated by high-pH reversed-phase high-performance liquid chromatography (HPLC). The fractionation procedure was modified from previously described protocol (Batth et al., 2014). A Waters Xbridge BEH C18 2.5 μm 3.0 × 100 mm XP column and an Agilent 1290 HPLC were used. Mobile phase A: 10 mM ammonium formate, 100% H₂O (adjusted to pH 10 with ammonium hydroxide); B: 10 mM ammonium formate, 10% H₂O, 90% ACN (adjusted to pH 10 with ammonium hydroxide). Dried phosphopeptides were dissolved in mobile phase A. The temperature of the column was operated at 45°C. After loaded onto the column, samples were separated using a 66 min gradient at a flow rate of 200 μL/min, as follows: 0 min 5% B; 50 min 25% B; 54 min 60% B; 56 min 70% B; 61 min 70% B; 63 min 5% B; 66 min 5% B. All fractions were collected at 2 min intervals. 28 fractions were collected between 0 and 56 min and were pooled into 14 fractions by combining fractions (1 and 15 were combined as fraction 1; 2 and 16 were combined as fraction 2; 3 and 17 were combined as fraction 3, and so on, until 14 and 28 as fraction 14).

All LC-MS/MS analysis were performed on an EASY n-LC 1000 system directly connected to a Q Exactive (Thermo Fisher Scientific, San Jose, CA) or a Q Exactive HF (Thermo Fisher Scientific, San Jose, CA) mass spectrometer which equipped with a nanospray ion source. Phosphopeptide fractions were dissolved in solvent A (0.1% formic acid, 100% H₂O) and loaded onto a reversed-phase column, which was packed in-house with 3 μm C18 ReproSil particles (Dr. Maisch GmbH). The inner diameter of the column was 75 μm. The solvent B was 0.1% formic acid in acetonitrile. Samples were separated with a 2 h gradient at a flow rate of 300 nL/min. The gradient was set up as follows: 0–4 min, 3–4% B; 106 min, 20% B; 114 min, 35% B; 117–120 min, 90% B. Each sample was analyzed twice by Q Exactive or Q Exactive HF, respectively.

For Q Exactive, the instrument parameters were set as: The scan range was between 300 and 1500 m/z and full scan resolution was set at 70,000; fragmentation was performed with the high-energy collisional dissociation (HCD) with a normalized collision energy (NCE) of 28%; one full scan was followed by 15 data-dependent MS/MS scans with a resolution of 17,500 and isolation window of 2.0 m/z.

For Q Exactive HF, the instrument parameters were as follows: The full scan used a mass window of 350–1800 m/z at a target resolution of 60,000; precursor ions were fragmented by HCD with a NCE of 27%; one full scan was followed by 20 data-dependent MS/MS scans with a resolution of 30,000 and isolation window of 1.6 m/z.

The MS parameters are firstly set up according to the vender's instructions and then optimized in house to maximize the analytical power of each machine.

The raw data files were processed using software MaxQuant (Cox and Mann, 2008) version 1.5.1.0 with an false-discovery rater (FDR) <0.01 at all the levels of proteins, peptides and modifications, to generate peak lists. The MS/MS spectra were searched against the Mouse UniProt FASTA database (July 2016) containing 51418 entries. Enzyme specificity was set to trypsin, allowing for cleavage N-terminal to proline and between aspartic acid and proline. The search included cysteine carbamidomethylation as a fixed modification and N-acetylation of protein, oxidation of methionine and/or phosphorylation of serine, threonine tyrosine residue (STY) as variable modifications. The missed cleavage number allowed was set as no more than 3. The SILAC based quantification of peptides and proteins was also performed by MaxQuant. SILAC ratios were normalized so that the median ratio was 1 to correct unequal sample mixing. Bioinformatics analyses were carried out with Microsoft Excel, GraphPad, Perseus (Tyanova et al., 2016), and R (https://www.r-project.org/) statistical computing software. Functional annotations were made through UniProtKB (http://www.uniprot.org/uniprot/), the Database for Annotation, Visualization and Integrated Discovery (DAVID) (https://david.ncifcrf.gov/), motif-x (Schwartz and Gygi, 2005), and Ingenuity Pathway Analysis (IPA, version 8.5, Ingenuity Systems, Redwood City, CA). Interaction network was made with software Cytoscape (http://www.cytoscape.org) and the physical interaction was provided by the SIGnaling Network Open Resource (SIGNOR) (Perfetto et al., 2016). Categorical annotation was performed in the Perseus software environment (Tyanova et al., 2016), which is a following up analysis tool in MaxQuant workflow. Categorical annotation was supplied in the form of GO biological process, molecular function, and cellular component. Enrichment for these categories was evaluated by Fisher exact test. Multiple hypothesis testing was controlled by using a Benjamini-Hochberg FDR threshold of 0.02. The annotation matrix algorithm was used to compute the difference of any significant protein annotation term from the overall intensity distribution as described (Cox et al., 2012). Motif-X was used to test for significantly overrepresented motifs, with the following parameters: P-value threshold ≤ 1.0E-6, with the entire mouse database as background database. The fold increase of motif was visualized with the Pheatmap package in R.

To determine whether GFKP-19 can protect neuro-2a cells (a brain neuroblastoma cell line) against the cytotoxicity induced by Aβ₂₅₋₃₅, neuro-2a cells were treated with 1 μM Aβ₂₅₋₃₅ in the absence or presence of GFKP-19 at different doses. When treated with GFKP-19 and Aβ₂₅₋₃₅, cells were pretreated with GFKP-19 for 1 h before Aβ₂₅₋₃₅ was added in. According to cell viability assayed by MTT method, the applicable concentration of Aβ₂₅₋₃₅ was decided according to the results of prevent experiments (Figure S1A) and previous report (Shih et al., 2011). Furthermore, we found that the treatment with Aβ₂₅₋₃₅ significantly decreased cell viability as assessed, and GFKP-19 was able to reverse the change (Figure 1A and Figure S1B). Thus, the above cell viability measurements qualified the samples for further proteomic analysis.

To gain global insight into the protective mechanism of GFKP-19, we combined the quantification method SILAC, high-pH reversed phase HPLC for fractionation, TiO₂ chromatography for phosphopeptide enrichment and high accuracy mass spectrometry characterization (Figure 1B and Methods). The study was carried out on neuro-2a cells, which were cultured in light, medium and heavy SILAC-labeling media paralleling three different treatment conditions selected based on the cytotoxicity experiments (DMSO, 1 μM Aβ₂₅₋₃₅, 1 μM Aβ₂₅₋₃₅ + 1 μM GFKP-19). After treated with drugs for 4 or 24 h, cells were lysed and the proteins were mixed at the ratio of 1:1:1 and digested by trypsin with the method FASP. Subsequently, peptides were incubated with TiO₂ and phosphopeptides were enriched. Next, phosphopeptides of each sample were separated into 14 fractions to decrease the complexity of the samples. The light, medium and heavy stable isotope labeled phosphopeptides were detected with highly accurate LC-MS/MS which incorporates a nanoLC system and a Q Exactive (QE) or a Q Exactive HF (HF). QE is a high-performance quadrupole Orbitrap mass spectrometer (Michalski et al., 2011) and HF denotes the improved version of QE which feature several innovations (Scheltema et al., 2014). Mass spectrometric data generated from QE and HF of three experiment groups was searched in MaxQuant using rigorous criteria. Totally, 25715 phosphosites were identified, corresponding to 5221 phosphoproteins. Among them, 18505 Class I phosphosites (of which localization probability is at least 0.75 and the score difference is 5 or higher, Table S1) (Pan et al., 2009) were identified and 14761 of them (that is, 80%) were quantified (Figure 1C).

Next, we analyzed the coverage of cellular functions of the deep phosphoproteome measured. To do so, we compared the Gene Ontology (GO) annotation for cellular compartment between murine genome and identified phosphoproteins in neuro-2a cells. GO cellular component analysis, using DAVID, revealed organelle, cytoplasm, nucleus, cytoskeleton, chromosome, chromatin, endosome and vesicle to be overrepresented, indicating that the functions of these parts of cells are based on significant phosphorylation. As expected, extracellular proteins in membrane, focal adhesion and extracellular region were underrepresented (Figure 2A). This is in agreement with previous reports proving low phosphorylation in the extracellular matrix (Weintz et al., 2010). The phosphoproteins identified were classified with Ingenuity IPA for protein class analysis (Figure 2B). These results suggest the phosphoproteome reflected essential biological functions, such as kinase (N = 344), phosphatase (N = 99), transcription regulators (N = 649) and other processes and their regulations between conditions, demonstrating the direct relevance of studying phosphoproteome in Aβ cytotoxicity.

We then focused on the quantified phosphoproteomic regulation. Benefiting from the fact that the experimental replicates were performed at two independent mass spectrometers and analyzed separately, we first correlated the ratios Aβ₂₅₋₃₅/Control (Ab vs. Con), (Aβ₂₅₋₃₅+GFKP-19)/Control (GF vs. Con) and (Aβ₂₅₋₃₅+GFKP-19)/Aβ₂₅₋₃₅ (GF vs. Ab) obtained from two mass spectrometer (QE and HF). We concluded three important points in our biological and experimental system. Firstly, the global regulation of phosphoproteome is minor between conditions, because the fold-changes (FCs) of majority (>90%) of the phosphosites remained below 1.5 FCs. Also, GF vs. Con at 4 h only have minor change suggesting that GFKP-19 has small effect on normal neuro-2a cells. This validates our experimental approach where specific, rather than broad, phosphoprotein regulations (e.g., those triggered by cell death) were captured. Secondly, we found the correlations of Aβ₂₅₋₃₅/Control (Ab vs. Con) and (Aβ₂₅₋₃₅+GFKP-19)/Control (GF vs. Con), as well as the numbers of significantly regulated phosphorylation events were both higher at 4 h compared to those of 24 h (Figure 2C). For the 4 h data of QE and HF, 5.5 and 6.3% of the phosphosites have the FC of Ab vs. Con at least 1.5; 1.3 and 2.0% of the phosphosites have the FC of GF vs. Con at least 1.5; 4.0 and 4.7% of the phosphosites have the FC of GF vs. Ab at least 1.5. For the 24 h data of QE and HF, 2.1 and 2.7% of the phosphosites have the FC of Ab vs. Con at least 1.5; 1.2 and 1.3% of the phosphosites have the FC of GF vs. Con at least 1.5; 1.0 and 1.5% of the phosphosites have the FC of GF vs. Ab at least 1.5. This means that 4 h is a good time point to inspect phosphorylation regulation, while the status at 24 h only have minor phosphoproteomic changes. This is consistent to the rapid responsive role of phosphorylation in cellular systems (Roux and Thibault, 2013). Last but not the least, we managed to obtain overall decent correlations between the data sets generated from the two mass spectrometers, despite of the global phosphoproteomic changes were minor. The correlation of the repeat experiments between QE and HF can be seen in Figure S2A. As expected, HF, which has the higher instrumental configuration, generated a larger data set compared to QE (Figure S2B). The biological compositions of phosphoproteome reflected by the two instruments are also very consistent. Just as an example, for the 4 h data of QE, identified phosphosites of class I sites consist of 88.3% phosphor-Ser sites, 11.2% phosphor-Thr sites, and 0.5% phosphor-Tyr sites; for the 4 h data of HF, most phosphosites of class I sites were on Ser (88.1%) and Thr (11.0%), whereas 0.8% phosphosites were at Tyr (Figure S2C). The high inter-instrument correlation essentially provides a benchmark for our quantification data, suggesting the phosphoproteomic regulations, although they were mostly of <1.5 FCs, are not due to experimental noise. To summarize, our integrative method captured the phosphoproteomic snapshots at two time points between three conditions (Ab vs. Con, GF vs. Con, and GF vs. Ab), revealing phosphoproteome regulations are globally of a small extent but biologically specific.

Given the relatively slow and mild effects on nerve cells of neuron protective drugs, we set a filter criteria to better understand the proteins whose phosphorylation were regulated by Aβ₂₅₋₃₅ and GFKP-19. This is because we hypothesize the more significantly regulated phosphoproteome events are more closely related to Aβ cytotoxicity and the protective mechanisms of GFKP-19. The substantially changed phosphosites were screened as follows: first, for both Con vs. Ab and GF vs. Ab, at least one of twice the FC up than 1.5 was quantified by QE or HF; second, both Con vs. Ab and GF vs. Ab undergo FC at least 1.5 or one of them experience more than 1.2 FCs and another at least 1.5 FCs (Figure 3A, Table S2). These highlighted phosphosites could be critical to GFKP-19 function in response to the cell toxicity of Aβ₂₅₋₃₅, and analysis with heat map show very small change in GF vs. Con and opposite change between Ab vs. Con and GF vs. Ab (Figure S3). To interrogate which cell function the filtered phosphosites participate in, GO annotation enrichment for cellular function of the phosphoproteins was done. Notably, cellular function like cytoskeleton protein binding, tubulin binding, microtubule binding, cell projection, cell junction have been enriched (Benjamini-Hochberg FDR threshold of 0.02 for all processes, Figure 3B). Tau plays a vital role in the microtubule assembly and stability. Main post-translational modification of tau is phosphorylation which results in tau aggregation in pathological situations (Avila et al., 2004). Neuron morphology is formed by cytoskeletal scaffolding. Hyperphosphorylation of tau is a premise for its aberrant aggregation and the formation of NFTs, which is a main neuropathological characteristic of AD. All of this suggests that Aβ₂₅₋₃₅ and GFKP-19 treatment related to the cell function that involved in AD progression.

To acquire a broad understand of the phosphorylation sites and their sequence pattern significantly changed in experiment group Ab vs. Con, GF vs. Con and GF vs. Ab, we carried out motif enrichment with motif-x (Schwartz and Gygi, 2005) with a P-value of ≤ 1.0E-6. The analysis was done separately to the phosphopeptides that were up regulated or down regulated in Ab vs. Con, GF vs. Con and GF vs. Ab (Figure S4). More motifs were enriched for down regulated phosphosites of GF vs. Ab. Taking focus on the results of GF vs. Ab (Figure 4A), in the phosphosites up regulated we found an enrichment of proline-directed (phosphor-Ser-X-X-X-phospho-Ser-Pro), basophilic (Arg-phospho-Ser-X-phospho-Ser and Arg-X-X-phospho-Ser-X-X-X-phospho-Ser), and basophilic at left and basophilic at right (Arg-X-X-phospho-Ser-X-Asp) motifs, whereas the phosphosites down regulated were enriched for basophilic and proline-directed (phosphor-Ser-Pro-X-X-X-Arg) motif (Figure 4B). The phosphorylation of proteins with these motifs can possibly be regulated by Aβ₂₅₋₃₅ and GFKP-19. Further experiments are needed to verify the functional relevance of AD for these phosphopeptide motif patterns.

How does GFKP-19 protect neuro-2a cells from the cytotoxicity of Aβ₂₅₋₃₅? β-Amyloid developing SPs in AD brain can lead to the generation of superoxide radicals which react with nitric oxide (NO) producing peroxynitrite and induces cell injury subsequently. Meanwhile, NO has been suggested to be associated to the pathogenesis of AD (Law et al., 2001). NO is a free radical and can bind to superoxide anions to form peroxynitrite (Eliasson et al., 1999), causing significant oxidative stress. NO is produced by nitric oxide synthase (NOS) which has three isoforms NOS1, NOS2, and NOS3. Overexpression of all three isoforms of NOS is significantly involved in AD and the initiation of the early progression of disease (de la Monte and Wands, 2006). Phosphorylation of NOS1 at S852 activates NOS1 then increases NO production and protein phosphatase PP2B can dephosphorylate NOS1 at S852 (Foster et al., 2009). We found the phosphorylation of NOS1 at S847, which is corresponding to S852 at human NOS1, is decreased by Aβ₂₅₋₃₅ treatment and rescued with GFKP-19 (Table S2). This suggests that GFKP-19 may bring down the oxidative damage and inflammation induced by NO through suppressing its synthase NOS1.

Apoptosis-stimulating protein of p53 (ASPP2; encoded by the Tp53bp2 gene) has been identified as an activator of the p53 family. ASPP2 stimulates tumor suppressor protein p53-mediated apoptotic response (Samuels-Lev et al., 2001; Bergamaschi et al., 2004). ASPP2 can be phosphorylated at S827 by the RAS/MAPK pathway which increases its binding to p53 and enhances the transcription function of p53 (Godin-Heymann et al., 2013). In our study, the phosphorylation of ASPP2 at S479 was detected to increase when treated with Aβ₂₅₋₃₅ and was reversed with GFKP-19 treatment (Table S2). The site of S479 has been identified in previous report (Mertins et al., 2014), but the extract relationship between the ASPP2 protein function and this phosphosite is still unknown.

As we know, NFTs induced by hyperphosphorylation of tau protein is one of the main neuropathological characteristics of AD and has been a direction of AD treatment. Pre-clinical studies on small molecule inhibitors of primary kinases involved in tau phosphorylation to prevent hyperphosphorylation of tau have shown promise (Khlistunova et al., 2007; Pickhardt et al., 2007). According to previous study, there are six isoforms of tau in human central nervous system (CNS) ringing in size from 352 to 441 amino acids, which generated by alternative splicing (Hanger et al., 2009). Many kinases have been proved to be tau kinases in earlier investigations (Vázquez-Higuera et al., 2011; Martin et al., 2013). To identify the targets of these tau kinases, large-scale studies aimed at determining phosphosites were done, and produced much information as reference. Proline-directed protein kinases of tau target Ser and Thr preceding a Pro residue (SP/TP motif). Mitogen-activated protein kinases (MAPK), such as p38 is contained in proline-directed protein kinases (Martin et al., 2013).

In our results, 20 tau phosphosites were quantified and six of them showed up regulation after Aβ₂₅₋₃₅ treated in line with previous reports that Aβ fibrils can increase tau phosphorylation (Busciglio et al., 1995), and down regulated with GFKP-19 treatment. The six phosphosites are S491, T523, S648, S688, S692, and S696 (Figure 5A, Figure S5, and Table S3), consistent to the phosphosites S199, T231, S356, S396, S400, and S404 in human tau isoform which had been reported before (Evans et al., 2000; Thornton et al., 2011; Luo et al., 2014). The phosphorylation of at S199 residue of tau is enriched in section of neurons vulnerable to neurodegeneration and is an early marker of tau pathology (Maurage et al., 2003). Phosphorylation at T231 has been assumed an initiating event for the formation of NFTs and NFTs induce neuronal cell death (Lasagna-Reeves et al., 2012). The serine 356 site located in KXGS motif promote tau disconnection from microtubules which was probably a key site refer to AD (Abraha et al., 2000; Haase et al., 2004). The phosphorylation of tau at sites Ser (396–404) was reported to represent an early event in AD (Mondragón-Rodríguez et al., 2014). These suggest that GFKP-19 can reverse Aβ₂₅₋₃₅ treatment induced up regulation of tau phosphorylation.

To identify if GFKP-19 regulates the phosphorylation of tau through its kinase or phosphatases, interaction network was produced with physical interaction prediction through SIGNOR (Figure 5B). Protein phosphatase PP1 and PP2B are phosphatases of tau dephosphorylating tau at S199, S202, T205, T212, S214, S235, S262, S396, S404, and S409 and total phosphatase activity and the activity of PP2B toward tau were increased in AD brain (Liu et al., 2005). Expression product of Ppp3ca is the catalytic subunit alpha isoform of PP2B. We found the phosphorylation of Ppp3ca at S462 was up regulated after treated with Aβ₂₅₋₃₅ and was down regulated by GFKP-19. The phosphorylation level of PP1 detected in our study was unchanged. Tyrosine-protein kinase FYN has been reported to phosphorylate tau at T18, but the site was not detected in our study, and the phosphorylation of FYN at Y185 was slightly up regulated with GFKP-19 treatment (Table S1), however, microtubule binding behavior of tau was not significantly altered by tyrosine phosphorylation of tau induced by FYN (Lee et al., 2004). Microtubule-associated protein/microtubule affinity-regulating kinases (MARKs) have been proved to phosphorylate the KXGS motifs of tau's repeat domain which is crucial for microtubules interactions and aggregation (Drewes et al., 1997; Biernat et al., 2002). MARK2 can phosphorylates tau at S262 and the interaction between MARK2 and tau was reported to be enhanced in AD brain section (Gu et al., 2013). MARK2 was also demonstrated to selectively phosphorylate eight serine residues in the repeat domain and C-terminal tail of tau and impact the structure and microtubule binding of tau (Schwalbe et al., 2013) and significant increase tau phosphorylation level at S396/404 and decrease total tau levels (Cavallini et al., 2013). We found that the phosphorylation of MARK2 at S499 and S533 were down regulated after treated with Aβ₂₅₋₃₅ and reversed by GFKP-19, while these two sites were not previously discovered for MARK2. Mitogen-activated protein kinases (MAPK) are involved in tau hyperphosphorylation, and p38 MAPK has been reported to phosphorylate tau at S202, S205, S231, S356, and S404 (Hanger et al., 2009). MAPK1 (ERK2) can reduce phosphorylation of tau at S396 (Cavallini et al., 2013). Dual phosphorylation of MAPK1 at T183 and Y185 could be necessary for the activation of MAPK1 (Robbins et al., 1993) and we found slightly increase in the phosphorylation level of MAPK1 at these two sites induced by GFKP-19 treatment (Table S1). MAPK14 (p38α) had been reported to be a potential kinase of tau, and was related to a later onset of AD (Vázquez-Higuera et al., 2011). Phosphorylation of MAPK14 at T180 is necessary for its activation, while phosphorylation at both T180 and Y182 was more active than phosphorylation at T180 only (Zhang et al., 2008). Although we did not quantify the phosphorylation of MAPK14 at Y182, we indeed detected the change of the phosphorylation of MAPK14 at T180, a very close phosphosite (Figure S6 and Table S4). The phosphorylation level of MAPK14 at T180 was up regulated after Aβ₂₅₋₃₅ treated, and down regulated with GFKP-19 treatment, suggesting that GFKP-19 may decrease tau phosphorylation through down regulating the phosphorylation level of MAPK14 at T180.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.