Anti-hamster IgG was obtained from SouthernBiotech, and anti-PD1 blocking antibody (G4) and anti-PDL1 antibody (10B5) were produced in-house. Detailed information on the antibodies used in the present study is presented in Supplementary Table S1.

For this study, we used PD1 knockout (KO) mice, which has been described previously (Yao et al., 2009), APP/PS1 mice (obtained from GemPharmatech Co., Ltd.), 5×FAD mice, which has been described previously (An et al., 2019; Zheng et al., 2021), and their age-matched C57BL/6 wild-type (WT) mice. The age of PD1 KO, APP/PS1, and 5×FAD mice ranged from 2–3, 9–12, and 9–10 months, respectively. The experiments involving the mice were approved by the Experimental Animal Ethics Committee of Fujian Medical University (FJMU IACUC 2018-034).

Synthetic β-amyloid (1–42) peptides (corresponding to the human Aβ sequence) were provided by China Peptides Co., Ltd. The Aβ1-42 powder was dissolved in sterile saline solution to a concentration of 1 μg/μl. The solution was then aged for 96 h at 37°C. Aβ aggregates or the corresponding vehicle (4 μl/5 min/mouse) was injected into the lateral ventricle (i.c.v.) using a microsyringe at the following position: −0.2 mm anteroposterior (AP), +1.0 mm mediolateral (ML), and −2.4 mm dorsoventral (DV) relative to the bregma (Amin et al., 2017; Wu et al., 2018). The control group mice received the first injection of hamster IgG (0.30 mg/mice) 3 days after the surgical procedures and the second injection 7 days after surgery. The Aβ1-42 + anti-PD1 group was injected with the PD1-blocking antibody following the same dosing program. Twelve days after surgery, all mice were sacrificed and the brain tissue was sampled for analysis (Zhang et al., 2020).

The dosing program of 5×FAD transgenic mice was based on a previous study (Rosenzweig et al., 2019). The mice were injected (i.p., 10 mg/kg) twice with anti-PD1 blocking antibody or anti-hamster IgG at a 3-day interval. The behavioral tests were conducted after 1 month, and all mice were sacrificed a month after the behavioral test. The Morris water maze (MWM) was performed to evaluate the spatial learning and memory ability of 5×FAD mice, as previously described (Qi et al., 2016; An et al., 2019). Briefly, a hidden escape platform (diameter: 100 mm, height: 230 mm) was placed at the center of one quadrant and 15 mm beneath the surface water in a pool of diameter 1,200 mm. Non-fat milk powder was added to turn the water opaque, and the temperature was maintained at 23–25°C. The behavioral test consisted of training for 5 days and a probe trial on day 6. During training, all mice were subjected to four trials per day, and the inter-trial interval was at least 10 min. For each trial, a mouse was released into warm water from the selected starting locations and allowed to locate the hidden platform within 1 min. The probe test was conducted to evaluate spatial memory ability on day 6. Mean escape latency, swimming speed, and number of platform crossing were analyzed using a computer equipped with Morris 2.8.1 software provided by Mobile Datum Co. (China).

For immunohistochemical evaluation of the mouse brain tissues, the mice were sacrificed and their brains were harvested, postfixed in 4% buffered paraformaldehyde overnight, embedded in paraffin, and cut into sections. Xylene was used to deparaffinize the paraffin-embedded tissue sections. The resulting sections were rehydrated with a gradient series of alcohol, and antigen retrieval was performed with citric buffer at 120°C for 10 min. To eliminate the influence of endogenous peroxidase and protein, 3% H₂O₂ and 10% fetal bovine serum (FBS) were successively added on the sections. Subsequently, the sections were incubated with the appropriate concentration of primary antibody overnight at 4°C, and then with HRP-conjugated secondary antibody for 30 min at room temperature. 3,3ʹ-Diaminobenzidine was added on the sections under a microscope and allowed to react for 2–5 min; the reaction was then stopped by adding H₂O. The sections were then stained with hematoxylin and dehydrated with a gradient series of alcohol. After permeabilization with xylene, the sections were covered with permanent mounting medium. Positive staining was detected and calculated using Image-Pro Plus software for Windows operating system.

Mouse hippocampal extracts were prepared by homogenizing the tissue in ice-cold RIPA lysis buffer supplemented with 1% (v/v) phenylmethanesulfonyl fluoride, as previously described (Gan et al., 2021). Boiled protein samples were subjected to SDS-PAGE, followed by semi-dry film transfer of the proteins onto polyvinylidene fluoride membranes. The membranes were blocked for 2 h with 10% (w/v) non-fat milk, and then probed with different antibodies overnight and HRP-conjugated secondary antibodies the next day. Various immunocomplexes were detected using the ChemiDoc XRS + system (BioRad).

The brains of WT C57BL/6 mice were harvested and homogenized with a lysis buffer (Beyotime Biotechnology). Nonspecific IgG (Beyotime Biotechnology) or PDL1 antibody was added to the lysates, which were then incubated for 3–4 h. Thereafter, 40 μl of protein A/G beads (Santa Cruz Biotechnology) was added to the lysates and incubated for another 1–2 h at 4°C. The precipitates were washed with washing buffer at least five times and then separated by 10% SDS-PAGE for immunoblotting.

SH-SY5Y and SH-SY5Y-APP cell lines were provided by Professor Tae Ho Lee; these cell lines have been described elsewhere (Kim et al., 2016; Chen et al., 2020). The cells were cultured in Hyclone DME-F12 supplemented with FBS (10% v/v).

The cells were harvested and incubated with different primary antibodies for 30 min at 4°C, and then washed with PBS (1% FBS). The samples were analyzed using FACSVerse, and the data were analyzed using FlowJo software. APC-conjugated PD1 antibody and APC-conjugated PDL1 antibody were purchased from Biolegend.

All data are presented as mean ± SD and analyzed using GraphPad Prism version 8.0 software for statistical analysis. A one-way or two-way analysis of variance followed by Dunnett’s post-hoc test when appropriate or paired Student’s t-test was used to calculate statistical significance.

Under normal physiological conditions, human primary neuroimmune cells express very low levels of PD1/PDL1, but neuronal PD1/PDL1 can be immediately induced by drug abuse or non-toxic doses of alcohol to contribute to neuroinflammation and neurodegeneration (Mishra et al., 2015; Mishra et al., 2020). The AD model generated by an intracerebroventricular injection of Aβ is commonly used for the following reasons: convenience and high cost-performance ratio. As shown in Figures 1A–C, the brain PD1 and PDL1 signals increased after Aβ42 insult compared with those after vehicle administration. SH-SY5Y cells stably overexpressing human APP have been widely used as an in vitro model to mimic AD pathology. We used flow cytometry and western blotting to compare the expression of PD1 and PDL1 between SH-SY5Y and SH-SY5Y-APP cells. There was an increase in the expression of PD1 and PDL1 in SH-SY5Y-APP cells, as shown in Figures 2A–E. Moreover, we performed western blotting and immunohistochemistry to detect the expression of PD1 and PDL1 in the brain of APP/PS1 and 5×FAD mice, respectively. The hippocampal levels of both PD1 and PDL1 were increased in APP/PS1 mice compared with those in the age-matched WT mice (Figures 2F–H). The PD1 and PDL1 levels were increased in a wide range of brain regions, including the cortex and hippocampus, in 5×FAD mice compared with those in the age matched WT mice (Figures 3A–C). These results suggest that PD1 and PDL1 are upregulated under AD conditions in vitro and in vivo.

The hyperphosphorylation of tau, especially Thr231, which is a phosphorylation site of tau, mainly mediated by GSK3β (Israel et al., 2012; Moszczynski et al., 2015), peptidyl-prolyl cis-trans isomerase NIMA-interacting 1 (Kimura et al., 2013; Kim et al., 2021), cyclin-dependent protein kinase 5 (CDK5) (Crespo-Biel et al., 2007), and death-associated protein kinase 1 (DAPK1) (Kim et al., 2014; Chen et al., 2020), significantly decreases post anti-PD1 antibody treatment (Rosenzweig et al., 2019). Therefore, we sought to determine whether treatment with PD1-blocking antibodies affects the activities of some key tau kinases in the AD brain. We analyzed GSK3β as a downstream target molecule of PD1/PDL1 based on the following findings: 1) PDL1 interacts with GSK3β in tumor cells and GSK3β regulates PD1 level in CTL (Li et al., 2016); 2) GSK3β phosphorylates tau mainly at Thr231, Ser262, and Ser396 in the AD brain; and 3) GSK3β activity is closely associated with learning/memory impairment in AD. As low phosphorylated GSK3β (p-GSK3β) (Ser9) levels have been widely reported in various AD mouse models, such as APP/PS1 and 5×FAD (Crouch et al., 2009; Wang et al., 2019), GSK3β is considered the most important molecule in AD pathophysiology and a pivotal marker for neurodegeneration in AD (Takashima, 2006; Lauretti et al., 2020). In our study, the p-GSK3β (Ser9) level significantly decreased after Aβ insult (Figures 1A,D) and inversely correlated with PD1 or PDL1 expression, based on Pearson’s correlation efficient (Figures 1E,F; R² = 0.6263, p = 0.0340 and R² = 0.6358, p = 0.0317, respectively). To further verify the relationship between GSK3β and PD1 and PDL1, two other transgenic mice were used in the following experiments. In APP/PS1 mice, decreased p-GSK3β (Ser9) level (Figures 2F,I,J) inversely correlated with the PD1 or PDL1 level (Figures 2K,L; R² = 0.6261, p = 0.0193 and R² = 0.6536, p = 0.0151, respectively). Moreover, activated GSK3β level in 5×FAD mice was significantly increased compared to that in WT littermates (Figures 3A,D). Furthermore, as shown in Figures 3E,F, an inverse correlation was observed between PD1 or PDL1 expression and p-GSK3β (Ser9) level (R² = 0.5383, p = 0.0383 and R² = 0.7131, p = 0.0083, respectively). Moreover, co-immunoprecipitation and double immunofluorescence assay revealed the PDL1/GSK3β immune complex in the brain and in SH-SY5Y-APP cells, respectively (Figure 3G and Supplementary Figure S1), suggesting that GSK3β might act directly downstream of the PD1/PDL1 axis. These results demonstrate that the PD1/PDL1 axis may be involved in AD pathology through GSK3β.

As PD1 is considered a regulator of GSK3β phosphorylation in different AD models, we aimed to clarify whether the upregulation of GSK3β activity in AD conditions is associated with PD1. PD1 deficiency tends to increase pS473-AKT level in normal Kupffer cells and restores AKT activation after murine polymicrobial sepsis attack, suggesting that PD1 KO protects cells from injury stimuli (Wang et al., 2016). Unexpectedly, in the present study, the phosphorylation of GSK3β at Ser9 in the brain of PD1 KO mice was significantly increased compared with that in age-matched WT mice, as shown in Figures 4A–D, suggesting that PD1 deficiency downregulates the activity of GSK3β under normal physiological conditions. We also investigated the effect of PD1 deficiency on GSK3β activity and tau hyperphosphorylation (p-tau Thr231 and Ser396) after an intracerebroventricular Aβ42 insult. Consistent with previous results, the pSer9 level in GSK3β decreased significantly after Aβ exposure. Aβ-treated PD1 KO mice showed a significant increase in the pSer9 level and decrease in the p-tau Thr231 and Ser396 levels compared with the control mice (Figures 4E–I). These results indicate that PD1 is an important regulator of GSK3β activity under normal physiological and AD conditions.

Next, we examined the effect of PD1-blocking antibody on Aβ-induced GSK3β activation and tau hyperphosphorylation. PD1-blocking antibody increased the expression of phosphorylated GSK3β at Ser9 and reduced the p-tau Thr231 and Ser396 levels after Aβ42 administration compared with those in the control, as shown in Figures 5A–E. Similarly, we used 5×FAD mice to investigate the protective effect of PD1-blocking antibody. We examined pSer9-GSK3β and total GSK3β levels in the hippocampus of 5×FAD mice by western blotting. The results showed a significant increase in pSer9-GSK3β rather than the total GSK3β level in the brain of anti-PD1-treated 5×FAD mice compared with that in the IgG-treated mice (Figures 5F–H), suggesting that GSK3β activity decreased after PD1 blockade. We further explored the effects of immunotherapy on tau hyperphosphorylation. Consistent with the results of a previous study (Rosenzweig et al., 2019), we found that both p-tau Thr231 and Ser396 levels decreased after intervention with PD1-blocking antibody (Figures 5F,I,J). These results indicate that blocking the PD1/PDL1 axis significantly reduces GSK3β activity and tau hyperphosphorylation in different AD models.

To determine the potential role of PD1 blockade in the learning and memory abilities in 5×FAD mice, hippocampus-dependent cognitive performance was evaluated using the MWM 1 month after the antibody treatment. As shown in Figure 6A, the 5×FAD mice presented an increase in the mean escape latency from the hidden platform compared with the controls (p < 0.01). Moreover, the average escape latency of the anti-PD1+5×FAD group was significantly lower than that of the IgG+5×FAD group (p < 0.05). However, there was no significant difference in the mean swimming speed during the training phase among the three groups (Figure 6B, p > 0.05). In the probe trial, the number of platform crossing was determined for 1 min on day 6 of the test. As expected, 5×FAD mice treated with control IgG had fewer platform crossings than the normal control subjects, and this was reversed by PD1-blocking antibody treatment (Figure 6C). These results suggest that the application of PD1-blocking antibody alleviates impaired cognitive performance in 5×FAD mice.