We searched for ‘‘AD’’ in GEO database,¹ and microarray datasets GSE1297 from the GPL96 platform, GSE63063 chip from the GPL10558 platform and the GSE110226 chip from the GPL10379 platform were downloaded. In GSE1297, according to the MMSE score, hippocampal samples were divided into control group (MMSE > 26, n = 9), incipient AD group (MMSE: 20–26, n = 7), moderated AD group (MMSE: 14–19, n = 8), and severe AD group (MMSE < 14, n = 7). GSE63063 contained 135 control samples and 139 AD patients’ blood samples, while GSE110226 included six samples of normal choroid plexus and seven samples of choroid plexus from AD patients. See the Supplementary Material form for detailed patient information.

The GEO2R interactive online tool² was applied to convert the raw data into a recognizable format. The differential expression genes (DEGs) were identified with P < 0.05 and | logFC| >1 as threshold values. The expression of genes obtained by VENN intersection was corrected by Bonferroni method.

The online enrichment platform David³ was used to conduct enrichment analysis of kyoto encyclopedia of genes and genomes (KEGG) pathway and gene ontology (GO) function on DEGs and screen the top-10 significant biological pathways with P < 0.05. Then, we used the bisoGenet plug-in in Cytoscape3.6.1 software to analyze the protein interaction of DEGs. Simultaneously, the protein interactions with the target gene PTAFR were evaluated through the String online tool.⁴

C57BL/6 mice were obtained from the Experimental Animal Center of China Medical University (Shenyang, China), and APP/PS1 transgenic mice were obtained from Jackson laboratory (Maine, USA). All animal care and experimental procedures were in compliance with the “Ethical Standards for Animal Laboratory Animals” of China Medical University. We collected the hippocampus of 12-month-old APP/PS1 mice (n = 5) and 12-month-old C57BL/6 mice (n = 5). We chose to study only female mice because incidence of AD is biased toward female (Jiao et al., 2016). The right hemisphere was used for Western Blotting, while the left hemisphere was immersed in paraformaldehyde for immunofluorescence staining. Blood was obtained from the orbit. Then, the blood samples were centrifuged at 3,000 rpm for 10 min at room temperature. The bottom layer was collected and used for RNA extraction.

The cells used in our experiment are BV2, SH-SY5Y, and SVGp12. Mouse-derived microglial BV2 cells and human-derived neuroblastoma SH-SY5Y cells were purchased from the Institute of Basic Medicine Chinese Academy of Medical Sciences (Beijing, China) while human-derived astrocytes SVGp12 were purchased from Bei Na Chuang Lian company (Beijing, China). Both BV2 and SVGp12 cells were cultured in DMEM medium (Invitrogen, Chicago, IL, United States), and SH-SY5Y cells were cultured in DMEM/F12 medium (Invitrogen, Chicago, IL, United States). All cells were supplemented with 10% fetal bovine serum (Thermo Fisher Scientific, Waltham, MA, United States) and 100 U/ml penicillin-streptomycin (Solarbio, Beijing, China), and placed them in a 37^°C, 5% CO₂ incubator for culture. The specific cell experiments are as follows. Both SVGp12 and SH-SY5Y were treated with 20 μM Aβ for 48 h to establish AD models. BV2 cells were treated with 1 μg/ml of LPS (Sigma-Aldrich, St. Louis, MO, United States) and 10 μM of Aβ (Sigma-Aldrich, St. Louis, MO, United States) for 48 h to establish AD inflammation model. The supernatant of BV2 treated in each group was added to Aβ-treated SH-SY5Y cells for 48 h for SH-SY5Y conditioned culture experiments. Cells of passage 5–20 were used for experiments.

We transfected the si-PTAFR plasmid (Sangon Biotech, Shanghai, China) into BV2 cells as required, and the transfection reagent was Lipofectamine 3000 (Thermo Fisher Scientific, Waltham, MA, United States). In the nucleotide sequence of si-PTAFR, the sense strand is GCUAUGGGUCUUUGCUAACUUTT; the anti-sense strand is AAGUUAGCAAAGACCCAUAGCTT. When the cell density was about 60–70%, cells were transfected by using Lipofectamine 3000 according to the manufacturer’s instructions for 24 h. 1 μg/ml of LPS and 10 μM of Aβ (Aβ25-35 were placed in a 37^°C incubator for 7 days before use) were incubated for another 24 h, then we collected the corresponding proteins and mRNAs and stored them at –80^°C. The specific operation flow chart is shown in Supplementary Figure 1.

We used the reverse transcription kit (Eric Biotechnology Company, Shanghai, China) to reverse transcribe the extracted total RNA into complementary cDNA, and then conducted the qPCR kit (Eric Biotechnology Company, Shanghai, China) according to the instructions required by the system (5 μl SYBR, 0.2 μl upstream and downstream primers, 0.2 μl ROX, 1 μl cDNA, 3.4 μl DEPC water) for real-time quantitative polymerase chain reaction (PCR) detection. The primers listed in this article are PTAFR, IL10, STAT3, and IL6. All primers were obtained from Sangon Biotech (Shanghai, China). The primer sequences are shown in Table 1. The results were processed by 2^–ΔΔCT method to compare the relative expression of RNA.

The BV2 cells and brain tissue were homogenized with protein lysate containing protease inhibitors (Beyotime Biotechnology, Shanghai, China) and the proteins were extracted, then quantified with a BCA kit (Beyotime Biotechnology, Shanghai, China). Equal amounts of protein were separated by SDS-polyacrylamide gels and transferred to polyvinylidene fluoride membranes (Millipore, Bedford, MA, United States). Then, the membrane was incubated in a blocking solution (a mixture of TBST containing 0.1% Tween-20 and 5% BSA) for 1 h at room temperature and placed in a primary antibody containing PTAFR (Abcam104162, 1:200, Abcam, Cambridge, MA, United States), IL10 (Wanlei03088, 1:1,000, Wanlei, Shenyang, China), STAT3 (CST12640s, 1:1,000, CST, Boston, MA, United States), IL6 (Wanlei02841, 1:1,000, Wanlei, Shenyang, China), MAP2 (CST4542s, 1:1,000, CST, Boston, MA, United States), and Syn (CST4329s, 1:1,000, CST, Boston, MA, United States) at 4^°C overnight. The next day, after washing with TBST, the membrane was incubated with the corresponding HRP secondary antibody (Proteintech, Chicago, IL, United States). The immune response band was observed by enhanced chemiluminescence (ECL) with luminescence and quantified by measuring the density of each band using Image-J software.

The cell survival viability was assessed by CCK8 assay (DOJINDO, Japan). First, SH-SY5Y cells were spread in a 96-well plate (n = 5,000 cells/well) and cultured for 24 h. Then, the medium was changed to the transfected and modeled BV2 cell supernatant for 1 h. One microgram/milliliter LPS and 10 μM Aβ were given, and the culture was continued for 24 h. After that, we changed the medium to serum-free DMEM medium containing 10% CCK8 reagent, 100 μl per well, and incubate at 37^°C for 2 h. Finally, we used a microplate reader to detect the absorbance at 450 nm wavelength and calculated the cell viability.

For apoptosis assays, SH-SY5Y cells were seeded in a 6-well plate (n = 2 * 10⁵ cells/well) and cultured for 24 h. Then, the medium was changed to the transfected and modeled BV2 cell supernatant for 1 h. One microgram/milliliter LPS and 10 μM Aβ were given, and the culture was continued for 24 h. Cells were washed twice with pre-cooled PBS, incubated with Annexin V-FITC and PI (BD Biosciences, United States) in the dark. Ultimately, cell apoptosis was analyzed by flow cytometer.

BV2 cells and SH-SY5Y cells were inoculated on a sterile cover glass into 12-well plates. After transfection and conditioned culture, the cells were fixed with 4% paraformaldehyde at room temperature. After washing with PBS, the cells were permeabilized with 0.5% TritonX-100 (Sigma–Aldrich, United States) for 20 min. The cells and brain tissue were blocked with goat serum (Boster Biological Technology, Wuhan, China) and stained with PTAFR or MAP2 in a humid box at 4^°C overnight. Afterward, they were incubated with TRITC-conjugated rabbit anti-goat IgG (Thermo Fisher Scientific, Waltham, MA, United States) for 1 h at 37^°C in a dark and humid box, followed by counterstaining with DAPI (Beyotime Biotechnology, Shanghai, China). Finally, the immunofluorescence image was obtained by laser scanning of a confocal microscope.

The protein secondary structure of PTAFR were downloaded from the PDB website⁵ while the three-dimensional structure of EGCG, donepezil, curcumin, memantine, and Huperzine A were acquired from the PubChem website.⁶ Then, we imported the protein and small molecule drug structure into the molecular operating environment (MOE) (2018 version) software and convert the protein secondary structure into the tertiary structure. Finally, we performed molecular docking between protein and drugs after the small molecule drug is optimized.

All data were statistically analyzed using GraphPad Prism 8.0.1 version. The data were expressed as mean ± standard deviation. Differences between groups were evaluated by one-way analysis of variance. All determinations were repeated three times. P < 0.05 was considered statistically significant.

To obtain the DEGs related to AD progression, we extracted the GSE 1297 chip from the GEO database. The GSE 1297 included the expression profiling of the hippocampus from 22 postmortem brain samples of AD patients at different stages of severity (incipient, moderate, and severe AD). The DEGs were identified with the set criteria of | logFC| >1 and P < 0.05, as shown in Figure 1A. A total of 174 DEGs were up-regulated, while 54 were down-regulated in incipient AD samples; 270 were up-regulated, while 157 were down-regulated in the moderate AD samples, and 688 were up-regulated, while 367 were down-regulated in severe AD samples (Figure 1B). Next, using the VENN graph network tool,⁷ the intersecting DEGs that were significant in all the stages of AD were obtained; a total of 32 DEGs were overlapping (Figure 1C and Table 2). They were further analyzed by plotting a heat map using the Sangerbox program (Figure 1D). A total of 25 genes were significantly up-regulated spanning across the incipient, moderate, and severe stages of AD, including, CR1, OGFOD3, AKAP13, MRPS12, ZDHHC17, ERF, UMOD, GRP107, PTGER4P2-CDK2AP2P2, HRP, RAD51B, NPAT, FGF20, RPL21P28, PTAFR, IL9R, AVPR2, LTB4R, PMS2P9, MYRF, SLC16A5, ATP11A, ITGB3, BGN, and LOC389906. A total of six genes, including TNFRSF25, E2F5, BICD2, RFXAP, TAC1, and B4GALT6 were significantly down-regulated consistently in each stage of AD. Only ITGB1 was up-regulated in the incipient and moderate AD stages, whereas down-regulated in the severe AD stage (Table 2). Therefore, based on the characteristics of biomarkers, we focused on the 25 up-regulated DEGs consistently in each stage of AD.

To identify the dominant pathways related to the DEGs and their biological roles, we performed enrichment analysis using the KEGG pathway enrichment and GO annotation using the David network enrichment tool (see text footnote 3). The top-10 statistically significant biological pathways were obtained (Figure 2A). The biological processes, included hematopoietic cell lineage, viral entry into host cells, mesodermal cell differentiation, inflammatory response, cell adhesion mediated by integrin, cell-substrate adhesion, heterotypic cell-cell adhesion, positive regulation of cell proliferation, extracellular matrix organization, and leukocyte cell-cell adhesion. Next, we also evaluated the interactions between the 32 DEGs using the bisoGenet plug-in in the Cytoscape 3.6.1 software. A total of 22 genes showed possible interactions with the core pathogenic gene, APP, including PTAFR, BGN, E2F5, BICD2, SLC16A5, AVPR2, TNFRSF25, IL9R, RFXAP, AKAP13, CR1, NPAT, RAD51B, GPR107, ITGB1, ITGB3, ERF, MRPS12, TAC1, OGFOD3, LTB4R, and ZDHHC17 (Figure 2B).

Given the 32 DEGs in the hippocampus of AD patients in different stages, these were reasonably considered as potential biomarkers for AD diagnosis. The expressions of potential biomarkers significantly change with the occurrence and progression of AD, thus, the changes in DEGs in the hippocampus needed to synchronize with the changes in blood and cerebrospinal fluid (CSF). We used the CSF chip (GSE110226) and the blood chip (GSE63063) for further investigation. The VENN graph network tool was used to identify potential target DEGs simultaneously upregulated in blood, CSF, and hippocampus of AD patients. Finally, two DEGs, PTAFR and AKAP13, were obtained (Figure 3A). Next, the possible correlations between the mRNA expressions of PTAFR and AKAP13 with MMSE scores were examined. We found that the mRNA expression of PTAFR was significantly correlated with the MMSE score (P = 0.0006), while that of AKAP13 was not statistically significant (P = 0.1308), shown in Figure 3B. Additionally, the Braak staging was more strongly correlated with PTAFR than AKAP13 (PTAFR, P < 0.0001; AKAP13, P = 0.0274; Figure 3C). We also found that the expression of PTAFR positively correlated with the neurofibrillary tangle scores, unlike AKAP13 (PTAFR, P = 0.0015; AKAP13, P = 0.1078; Figure 3D). The mRNA expression of PTAFR gradually increased with AD progression from incipient to severe (P < 0.05, Figure 3E), while AKAP13 did not exhibit this trend (Supplementary Figure 2A). Moreover, both PTAFR and AKAP3 were upregulated in the choroid plexus and peripheral blood (Figure 3F and Supplementary Figure 2A). PTAFR was also highly expressed in the entorhinal cortex, hippocampus, and temporal cortex of AD brains, while AKAP13 showed high expression in the hippocampus only (Figure 3H and Supplementary Figure 2C). Furthermore, we investigated the possible mechanism underlying PTAFR involvement in AD progression. Using the AlzData online network platform⁸ (Xu et al., 2018), we found that only PTAFR was highly expressed in microglia, which contributes to the inflammatory responses in the pathogenesis of AD, while AKAP13 was expressed in many cell lines (Figure 3G and Supplementary Figure 2B). Taken together, our findings suggested that PTAFR may be a more important target biomarker having higher efficacy for AD diagnosis.

Our previous results using the GEO database indicated that PTAFR was the most relevant potential biomarker, which was highly expressed in AD patients. Thus, we validated its expression both in vivo and in vitro. We performed the assays using a 12-month-old APP/PS1 double transgenic AD mouse model. Indeed, PTAFR expression in APP/PS1 mice was significantly upregulated both at the mRNA and protein levels (P < 0.01, Figures 4A–C), as compared to the age-matched C57 BL/6J control mice. The peripheral blood levels of PTAFR in APP/PS1 mice also showed a significant increase (P < 0.01, Figure 4A).

Next, we validated these results in vitro. As mentioned previously, the predictive results based on the AlzData platform showed that PTAFR was specifically and highly expressed in microglia as compared to other neural cells in the central nervous system. A homologous sequence alignment of murine and human PTAFR genes showed high homology between the two (Supplementary Figure 3C). Therefore, we examined the PTAFR mRNA and protein levels in different cells, including the BV2 (microglia-like), SH-SY5Y (neuron-like), and SVGp12 (astrocyte-like) cell lines. We found that LPS+Aβ induction significantly enhanced the mRNA and protein level expressions of PTAFR in the BV2 cells (P < 0.01, Figures 4A,B,D), while there were no significant changes in the SVGp12 or SH-SY5Y cells as compared to the corresponding control groups (P > 0.05, Supplementary Figures 3A,B), which indicated that PTAFR was highly expressed in the microglia.

PTAFR is closely associated with the secretion of inflammatory factors upon kidney injury and in retinal neovascularization (Latchoumycandane et al., 2015a,b; Bhosle et al., 2016), indicating that neuroinflammation mediated by microglia may be the possible mechanism underlying PTAFR involvement in AD progression. The specific experimental flow chart is shown in Supplementary Figure 4. First, we found that PTAFR was closely associated with the inflammatory factors IL-10 and STAT3 using the STRING online tool (see text footnote 4) (Figure 5A). Extensive studies show that the IL10-STAT3 pathway is involved in the occurrence of many diseases (Zhang et al., 2017; Campana et al., 2018; Shirakawa et al., 2018; Cevey et al., 2019; Degboe et al., 2019; Wang et al., 2019). Thus, we reasonably hypothesized that PTAFR kindled the microglia-mediated neuroinflammation through the IL10-STAT3 signaling pathway and exaggerated the microenvironment of neurons in the progression of AD. After LPS and Aβ treatments, the mRNA and protein levels of IL-10 reduced significantly, while those of STAT3 and IL-6 were elevated, following PTAFR up-regulation (P < 0.05, Figures 5B–D). Further, we silenced the PTAFR gene in BV2 cells and found that the mRNA and protein levels of IL-10 increased, while those of STAT3 and IL-6 decreased substantially (P < 0.05, Figures 5B–D and Supplementary Figures 5A–F).

Next, the conditional medium (CM) of the treated BV2 cells was used to culture SH-SY5Y cells treated with LPS+Aβ to investigate the subsequent inflammatory efficacy of microglia on neurons. The CM of LPS+Aβ-induced BV2 cells significantly reduced the cell viability of SH-SY5Y cells (P < 0.0001, Figures 5E,F), while the CM of LPS+Aβ-induced BV2 cells after PTAFR silencing, could significantly improve the cell viability in SH-SY5Y cells; these findings were consistent in both CCK8 and Annexin V/PI flow cytometry assays (P < 0.001, Figures 5E,F). Moreover, the CM also remarkedly enhanced the expression of neuroplasticity indices, MAP2 and Syn (Figures 5G,H). Taken together, PTAFR, a potential biomarker, exaggerated the microglia-mediated microenvironment by upregulating inflammatory factors through IL10-STAT3 signaling.

As described previously, PTAFR played an important role in exaggerating the microglia-mediated neuronal microenvironment through the IL10-STAT3 signaling pathway, which was closely correlated with AD progression. Therefore, it could be a potential biomarker or an essential target for the R&D of new anti-AD drugs. Thus, we performed targeted molecular docking of PTAFR with several drugs commonly used in clinical and scientific research for AD treatment, including donepezil, memantine, EGCG, curcumin, and Huperzine A. The molecular docking was performed using the MOE software. The S-value obtained after docking is used for evaluation of the possible binding; S < –7 was considered as having a significant probability of binding. EGCG is a potent compound that can exert anti-AD effects; the docking results showed that it could theoretically bind with PTAFR (S = –7.7826, Figure 6A). Donepezil is a commonly used medication used to treat mild-to-moderate AD. We also identified the possible binding sites (S = –7.5199, Figure 6B). Curcumin has neuroprotective efficacy and could bind to PTAFR, S = –7.5698 (Figure 6C). However, the S-values for memantine and Huperzine A were -5.3495 and S = –5.3781, respectively (Figures 6D,E). The compounds with planar structures and multiple benzene rings, such as EGCG, curcumin, and donepezil had a greater probability of binding to PTAFR, while those compounds showing stereo conformation, such as memantine and huperzine A, showed a significantly lesser probability of binding with PTAFR. These results indicated that PTAFR could bind to some anti-AD drugs and may have implications as a potential target for the treatment of AD in the future.