Antigen-specific age-related memory CD8 T cells induce and track Alzheimer’s-like neurodegeneration

PBS, or CD8 T cells from mouse wild-type donors were injected into young (6 to 8 wk) B6.Foxn1 hosts (PBS and wt-CD8 groups, respectively). Limited analyses of young B6.Foxn1 hosts injected with wild-type CD4 T cells or CD4 plus CD8 T cells together (wt-CD4 and wt-CD8+CD4 groups, respectively), were also performed (SI Appendix, Fig. S1A). Females were exclusively used to avoid male-specific autoimmune disease dynamics, some of which are associated with CD8 T cell homeostatic expansion, in related mouse strains (19, 20). The donor T cells acquire age-related resident memory-like (T_RM) phenotype, as well as age-related genotype and function as they expand. CD8 T cells reactive to a non-Aβ epitope on APP then selectively accumulate in the brain of these ^hiT (for “homeostatically induced T cell”) mice (18). Circulating APP-reactive CD8 T_RM cells were rapidly homeostatically expanded in ^hiT mice (SI Appendix, Fig. S1B), similar to the reported gradual expansion of age-related CD8 T_RM cells in aging humans (21).

We examined Triton-soluble brain extracts as we previously described (22), to assess potentially toxic Aβ in ^hiT mice by western blot (WB) using 4G8 (human- and rodent-specific APP/Aβ) and ab14220 (rodent-specific APP/Aβ) antibodies. This revealed relative upregulation of high molecular weight species presumed to be APP and its cleavage products (APP^Cl) in brains of wt-CD8 group ^hiT mice 3 and 10 wk after injection (Fig. 1A and SI Appendix, Fig. S2 A and B). Additional isoform- and species-specific antibodies in ELISA revealed that endogenous Aβ1-40 but not Aβ1-42 was relatively increased in wt-CD8 group mice 10 wk postinjection (Fig. 1B), with WBs and tissue staining using 4G8 antibody (recognizing rodent and human Aβ/APP) confirming prominent amyloidosis in the cortex and hippocampus (SI Appendix, Fig. S2B and Fig. 1 C–F, respectively). By 6 mo postinjection, wt-CD8 and wt-CD8+CD4 groups exhibited increased Aβ deposition in brain vasculature, whereas wt-CD4 group did not (Fig. 1 D and E and SI Appendix, Fig. S2 C and D), further suggesting CD8 T cells are necessary and sufficient to promote amyloidosis. Aβ plaques in wt-CD8 group brains 15 mo postinjection exhibited typical morphology, but exhibited small compact cores at best, were chiefly detergent-soluble (SI Appendix, Fig. S3), and were minimally stained by curcumin or ThioS (Fig. 1 C and F).

Detergent-soluble phospho-tau (pTau) was slightly (30%) but significantly increased at 10 wk postinjection in the wt-CD8 group forebrain, while pTau paired helical filaments (PHFs, which mature to form NFTs in AD) were increased nearly fivefold on WBs (Fig. 1 G and H). Starting at 6 mo postinjection, fibril-staining reagents (Gallyas silver, curcumin, and ThioS), each stained cellular inclusions within the wt-CD8 group hippocampus (Fig. 1 I–K and SI Appendix, Fig. S4A), with similar structures seen in simultaneously stained human AD cortex (SI Appendix, Fig. S4B), but not in brains of either AD-transgenic Tg2576 (AD-Tg) mice or AD-Tg rats (Fig. 1 I–K) (22). Tissue immunofluorescence (IF) for pTau, Aβ, and DAPI prior to Gallyas staining revealed these structures were derived from pTau⁺ neurons with intact nuclei, and that silver staining was superimposable with that of pTau but not Aβ (Fig. 1L). These data suggest that the predominant T cells in ^hiT mice (which resemble resident-memory phenotype CD8 T cells or CD8 T_RM) promote coordinated and robust deposition of Aβ and fibrillar NFT-like inclusions in the mouse brain.

Additional images support our previous findings that CD8 T cell infiltration, microgliosis, and astrogliosis (SI Appendix, Fig. S5 A–C, D and E, and F–I, respectively), are exclusively seen in wt-CD8 group mice (18). Cerebral plaques in these animals were closely associated with gliosis, as is common in human AD (SI Appendix, Fig. S5 H and I). In brain regions containing both T and glial cells, Aβ plaque burden correlated more strongly with hippocampal CD8 T cell levels than astrocytic or microglial levels (SI Appendix, Fig. S5 J–L), suggesting a unique relationship between CD8 T cells and amyloid pathology.

To examine mechanisms of ^hiT cell mediated neuropathology, we injected PBS, or CD8 T cells from wild-type, Perforin 1-deficient, or IFNγ-deficient donors into young (6 to 8 wk) B6.Foxn1 hosts (PBS, wt-CD8, PrfKO-CD8, and IfnγKO-CD8 groups, respectively), and analyzed neuropathology by ELISA or WB, as well as by tissue staining. CD8 T cells from all donor strains expanded in circulation of B6.Foxn1 recipients (18). In marked contrast to wt-CD8 or IfnγKO-CD8, however, PrfKO-CD8 brains exhibited no significant increase in Aβ, plaques, pTau, PHFs, or NFT-like inclusions in any region. At 15 mo postinjection, Aβ1-40 was significantly elevated only in the wt-CD8 group brain by ELISA (Fig. 2A). Nevertheless, Aβ plaques and silver-stained neurons were evident within the entorhinal cortex and hippocampus by tissue IF in both wt-CD8 and IfnγKO-CD8 groups (Fig. 2B). Tau PHFs were also only elevated in wt-CD8 brains at 15 mo by WB (Fig. 2C), but Gallyas tissue staining of IfnγKO-CD8 brains similarly revealed NFT-like inclusions in the entorhinal cortex and hippocampus, but not in the cingulate cortex as in wt-CD8 group brains (Fig. 2D). Silver-stained neurons in IfnγKO-CD8 brain were derived from pTau⁺ neurons by sequential staining as shown for wt-CD8 brain (Fig. 2E). Aβ plaques in wt-CD8 and IfnγKO-CD8 group brains were of similar morphology (Fig. 2 F and G), but those in the IfnγKO-CD8 group were significantly smaller (Fig. 2 F and H) and appeared only in the entorhinal cortex and hippocampus but not in the cingulate cortex as in the wt-CD8 group (Fig. 2B). These data suggested that blocking Perforin production completely prevented all overt AD-like pathology, while blocking IFNγ production led to smaller Aβ plaques and to limited distribution of NFT-like structures reminiscent of early-stage AD (23).

To determine whether neurodegeneration was evident in ^hiT mice, we stained and counted neurons positive for the neuron specific nuclear protein, NeuN, in CA1, CA2, and CA3 of the hippocampus. Brain mass was also assessed, and NeuN and synaptic protein signals quantified on WBs. Loss of NeuN⁺ cells in wt-CD8 group mice was visually apparent in hippocampal immunostains (Fig. 3 A and B), and was verified by NeuN⁺ cell counts at 15 mo postinjection (Fig. 3C). Loss of brain mass in the wt-CD8 group progressed from 5% at 6 mo, to 10% 15 mo postinjection (Fig. 3D), comparable to terminal brain atrophy in human AD (24). WBs of brains 15 mo postinjection confirmed proportional decreases in NeuN, the postsynaptic submembrane protein, drebrin, and the presynaptic vesicle protein, Synaptophysin (Fig. 3 E and F). Loss in brain mass correlated with decreased NeuN within PBS and CD8 injection groups, establishing a direct relationship between brain atrophy and neuronal loss (Fig. 3G). Significant brain loss was not observed in wt-CD4 group but was evident in wt-CD8+CD4 group, suggesting that CD4 T cells fail to either mediate or mitigate neuronal loss, or modulate its induction by CD8 T cells (SI Appendix, Fig. S6A).

We assessed cognitive performance in ^hiT mice using three independent tests for learning and memory: contextual and cued FC, SAB in the Y-maze, and BM performance. Open-field testing (OFT) was performed prior to each of these three behavioral tests to rule out influences due to injection-associated mobility variations (Fig. 3H and SI Appendix, Fig. S7A), as well as to monitor motor dysfunction evident in other T cell-associated neurological conditions such as multiple sclerosis.

Significant motor deficits occurred with age but did not distinguish any groups including the PBS group from others (SI Appendix, Fig. S7A). In contrast, contextual FC was reduced in the wt-CD8 group relative to PBS controls 6 mo after T cell injection, with both contextual and cued learning impaired at 11 mo (Fig. 3I). These results suggest that CD8 T cells in ^hiT mice alone mediate early damage to the hippocampus (required for contextual FC), with additional later damage to the amygdala (involved in cued FC). This progressive behavioral deficit pattern mimics one typical of cognitive decline in human AD (25). Intriguingly, only cued learning was impaired in wt-CD8+CD4 but not wt-CD4 group mice at the earlier time point (SI Appendix, Fig. S6B), suggesting that CD4 T cells modulate whether the hippocampus or amygdala is damaged first in ^hiT mice, but neither induce cognitive decline themselves nor alter the ability of CD8 T cells to do so.

SAB in the Y-maze is a well-established assay for hippocampal integrity (26, 27), based on the preference of mice to alternately explore two alleys. The lowest possible score (50%) indicates random alley choice, due to either no working memory of the previous alley entered, or complete lack of preference. SAB testing 12 mo postinjection revealed a score of 55 to 56% in PBS controls, comparable to wild-type mice (28), but was reduced to 50% in the wt-CD8 group (Fig. 3J). This is consistent with an absence of working memory in the wt-CD8 group.

BM testing was performed at 14 mo postinjection and represents a more focused measure of hippocampus-dependent learning and memory. This test is not confounded by decreased ability to swim, or stress-induced behaviors caused by swim tests (29). Relative to PBS controls, wt-CD8 mice showed no ability to learn the maze over the initial 4-d training period (Fig. 3K). Given this initial learning deficit, it was not surprising that wt-CD8 mice were also profoundly impaired in subsequent memory retention and reversal phases of the maze (Fig. 3 L–N).

Contextual FC at 6 and 11 mo correlated with brain mass (SI Appendix, Fig. S7B), as did latency to solve the BM (SI Appendix, Fig. S7C), further underscoring the relationship of cognitive decline to physical neurodegeneration. Poor performance of ^hiT mice on BM (total latency below median = BM^lo) also exhibited significant association with increased pTau PHFs exclusively (SI Appendix, Fig. S7D), but not with any form of Aβ or detergent-soluble pTau (SI Appendix, Fig. S7 E and F).

Taken together, these three independent tests suggest that fully functional CD8 T cells in ^hiT mice mediate severe, progressive impairment of hippocampus-dependent learning and memory independent of locomotor activity, and that both the progressive pattern of cognitive loss and its association with cerebral pathology in the ^hiT model mirrors features typical of clinical AD (30).

Neuronal and cognitive loss differences were also evident between wt-CD8 and KO-CD8 groups: Neither IfnγKO-CD8 nor PrfKO-CD8 brains showed significant evidence of neuronal loss, and both were indistinguishable from PBS controls in level of spontaneous alternation (Fig. 3J). IfnγKO-CD8 mice, however, appeared initially similar to PBS controls in the training sessions of the BM, but became more similar to the impaired wt-CD8 group mice at many of the later stages (Fig. 3 K–N). This suggests that both IFNγ and PRF1 deficiency eliminates robust neuronal loss and either subdues cognitive decline (in IfnγKO-CD8 group) or eliminates it altogether (in PrfKO-CD8 group).

The dramatic impact of PRF1 deficiency on all aspects of AD-like neurodegeneration is most consistent with a direct role for lytic elimination of neural cells by CD8 T cells in the ^hiT mouse model. CD8 T cells normally reside in the brain, including those of C57BL/6 mice (31). These cells require expression of MHC I on their targets to eliminate them in an antigen-specific manner. Most neuronal lineage cells are MHC I-negative unless induced by cytokines such as IFNγ, but catecholaminergic neurons within the locus coeruleus (32) and neural progenitors constitutively express MHC I (33). To determine whether these cells were eliminated early and independent of IFNγ, we examined Doublecortin(DCX)-positive neural progenitors within the subventricular zone (SVZ) of mature (6-mo-old) wild-type C57BL/6 mice, as well as in wt-CD8 and IfnγKO-CD8 group ^hiT mice 15 mo postinjection. CD8 T cells were observed in close proximity to DCX⁺ neural progenitors in the SVZ rostral migratory stream of wild-type C57BL/6 mice, consistent with the possibility that CD8 T cells might directly eliminate these cells under disease conditions (SI Appendix, Fig. S8A). Accordingly, DCX⁺ neurons were dramatically decreased in SVZ and throughout the brain in wt-CD8 and IfnγKO-CD8 group ^hiT mice (SI Appendix, Fig. S8 B and C). Thus, even in the absence of more widespread neuronal loss (as in the IfnγKO-CD8 group), CD8 T cells in ^hiT mice promoted DCX⁺ neural progenitor elimination. This suggests that neuronal lineage cells that constitutively express MHC I are eliminated first by ^hiT cells.

To gain further mechanistic insights into how CD8 T cells in ^hiT mice promote AD-like pathology, we performed RNAseq analysis on mouse forebrains from experimental and control groups. Focusing on changes ≥10% relative to PBS, each of wt-CD8, IfnγKO-CD8, and PrfKO-CD8 groups exhibited significant upregulation of the T cell-specific marker, CD3ε, while they also altered 1,947, 309, and 1,901 additional transcripts, respectively (SI Appendix, Fig. S9 A–J). Since upregulation of CD3ε was at odds with our previous observation that CD8 T cells were undetectable in PrfKO-CD8 group brain by tissue staining (18), we reexamined T cells in this group using potentially more sensitive flow cytometric analysis (SI Appendix, Fig. S10). We confirmed significant accumulation of CD8 T cells in PrfKO-CD8 group brain relative to control B6.Foxn1, albeit lower than in wild-type C57BL/6 controls previously reported as comparable to the wt-CD8 group (SI Appendix, Fig. S9 A and B) (18). Thus, all groups injected with CD8 T cells harbored elevated levels of those cells in the brain, consistent with their CD8ε mRNA upregulation. This provides relevant context for further gene expression analysis.

Two gene expression pathways were prominently altered in the wt-CD8 group exclusively: BioPlanet AD and a group of largely overlapping pathways related to cytoplasmic ribosomal proteins and translation (SI Appendix, Figs. S9H and S11 A–C). In addition, a group of largely overlapping pathways related to electron transport chain and oxidative phosphorylation was up-regulated in both wt-CD8 and IfnγKO-CD8 but not PrfKO-CD8 brain (SI Appendix, Figs. S9H and S11 A–C), and as such was uniquely associated with both delayed/arrested and later AD-like neuropathology. Electron transport chain- and ribosomal protein-related pathways are known to be involved in CD8 T cell-mediated effector activity as well as AD pathophysiology, and IFNγ is also known to modulate ribosomal proteins (34). Within cell type-specific genes, only neuronal genes exhibited net upregulation, whereas most nonneuronal genes were down-regulated (Fig. 4A). Moreover, multiple pathways implicated in AD were affected in the wt-CD8 group (SI Appendix, Figs. S9 H–J and S12). To further determine relevance of these modest changes to AD, we focused on 82 genes identified in prior genome-wide epidemiological studies (“GWAS genes”) (35) whose AD-associated variants often manifest as subtle expression changes. Among this GWAS subset, 14, 2, and 11 genes were significantly altered in wt-CD8, IfnγKO-CD8, and PrfKO-CD8 groups, respectively, with 8 and 1 of these also altered in wt-CD8 and IfnγKO-CD8 groups respectively, relative to PrfKO-CD8 (P = 0.034; two-sided Fisher’s exact test; Fig. 4B). Known functions of GWAS genes altered in wt-CD8 and PrfKO-CD8 brains included multiple biological pathways implicated in AD, including Aβ/Tau pathology gene expression and cell signaling, endocytosis, inflammation/immunity, and lipid/cholesterol metabolism (SI Appendix, Fig. S9 H and J), whereas those altered in IfnγKO-CD8 brains covered a more restricted subset involved in Aβ/Tau pathology and inflammation/immunity. This supports a model in which Perforin controls a critical disease-initiating event reflected by electron transport chain gene modulation, whereas IFNγ modulates most other gene expression risk factors that may worsen disease pace, assure disease progression, and/or increase disease severity (Fig. 4C).

Given the varied impact of CD8 T cells on different rodent models of hereditary neurodegeneration, it was imperative that the potential impact of ^hiT analogues was examined in human patients. We first examined levels of the larger “parental” memory/aging (KLRG1⁺) CD8 population in blood, and then the APP-specific KLRG1⁺ CD8 subpopulation derived from it, by flow cytometry in HLA-A2⁺ individuals from two separate cohorts (Fig. 5 A–F, with gating parameters in SI Appendix, Figs. S13 and S15A). As expected, the parental KLRG1⁺ CD8 population increased with age while KLRG1^- CD8 cells did not (SI Appendix, Fig. S14). The KLRG1⁺ subpopulation also exhibited slight expansion in age-related cognitive decline or MCI-AD (Fig. 5 B and D; and SI Appendix, Fig. S15 B and C, with representative gating shown in SI Appendix, Fig. S15A). By contrast, APP-specific KLRG1⁺ CD8 T cells were markedly decreased in MCI and/or AD (Fig. 5 C and D), and this trend was maintained in both females and males (SI Appendix, Fig. S16 A and B). The decrease in APP-specific KLRG1⁺ CD8 T cells correlated with poor cognitive performance in both cohorts independent of age (Fig. 5 E and F), exhibiting a linear correlation in the cohort subjected to the more sensitive MoCA test (Fig. 5E).

Cytolytic and antigen-specific T cell markers in AD brains were next examined (Fig. 5 A and G–J). Upregulation of Perforin1 in the AD brain was evident in two independent patient cohorts, corresponded with CD8 upregulation on WBs, and exhibited the expected punctate (endocytic) cellular pattern by tissue IF (Fig. 5 G and H). Similarly, T cells stained by CD8 and APP-specific pHLA multimers were more prevalent by tissue IF in the AD brain from the UCD cohort, despite no prescreening of specimens for HLA-A2 to which the APP-specific multimer binds (Fig. 5 I and J). Taken together, these data suggest that parental KLRG1⁺ CD8 T cells accumulate with aging in blood. By contrast, APP-specific KLRG1⁺ CD8 T cells in blood decrease in proportion to cognitive decline in AD and prodromal conditions, just as they accumulate in the AD brain.

Decreased APP-specific KLRG1⁺ CD8 T cells in blood correlated significantly with reduced Aβ1-42 and increased total Tau (but only marginally with p-tau181) in CSF within the largest patient cohort (UA; Fig. 6A), suggesting correspondence with existing AD biomarkers. Levels of parental KLRG1⁺ CD8 T cells also correlated uniquely with CSF Aβ1-42 in AD patients independent of age (Fig. 6B). While neither parental nor APP-specific CD8 cells correlated linearly with MMSE (as with MoCA) (Fig. 6 A and B), the latter did exhibit a correlation with hi/lo MMSE score (Fig. 5F). The lower sensitivity of MMSE vs. MoCA in detecting cognitive impairment may account for this difference (36). Overall, these trends suggested association of age-related CD8 T cell dynamics with some of the most specific established AD biomarkers, hinting at potential AD biomarker utility for both parental CD8 T_RM and APP-specific subpopulations.

To test this, we first examined publicly accessible data for multiple CD8 T_RM markers in blood from independent patient cohorts (CD3D/E, TCRZ, CD8A/CD8B, CD103, CD122, CD127, CD44, as available in database). Using GFAP as an up-regulated standard (SI Appendix, Fig. S17 A and B), or simply compared to normal aging controls (SI Appendix, Fig. S17 C and D), we verified significant upregulation of multiple CD8 T_RM markers at various stages of AD in two independent cohorts, with CD8 and TCRZ genes up-regulated at the earliest stage of AD in a severity-graded cohort (SI Appendix, Fig. S17B). We separated the largest of these cohorts into T cell high and low groups based on median CD3 expression to exclude patients with age-related T cell lymphopenia, and excluded those under 65 to eliminate potentially confounding early-onset AD patients and young controls (SI Appendix, Fig. S6D). Comodulation of CD103, CD44, and CD8A as a group, as well as levels of CD103 alone, exhibited intriguing biomarker potential in ROC plots (Fig. 6C). In addition, APP-specific KLRG1⁺ CD8 T cell levels quantified by flow cytometry exhibited even better biomarker potential than CD8 T_RM genes (Fig. 6D), corresponding better to MCI-AD than the highly touted plasma biomarker, pTau-217 (SI Appendix, Fig. S17 A–C), corresponded to AD itself (37). This demonstrates that the APP-specific CD8 T cell subpopulation and its parental KLRG1⁺ population are uniquely modulated in AD, and suggests that their levels may be useful and potentially superior disease biomarkers.

Female C57BL/6, B6.Foxn1 mice, and congenic and/or syngeneic knockout strains were housed in a pathogen–free vivarium under standard conditions on a 12-h light/12-h dark cycle. Recipient animals were 8- to 10-wk-old female B6.Foxn1 mice; donors were 5- to 8-wk-old female C57BL/6 or B6.CD45.1 congenic mice. Numbers of animals used per test are specified in SI Appendix, Table S1.

Splenic CD8⁺ T cells from C57BL/6J female donors (5 to 8 wk old) were purified using anti-CD8 or anti-CD4 immunobeads (Miltenyi Biotech, Sunnyvale, CA). 3 × 10⁶ purified CD8 or CD4 T cells, or 3 × 10⁶ native splenocytes for CD8+CD4 T cell hosts, were intravenously injected in 50 µL of PBS into female B6.Foxn1 nude hosts. Cell numbers were based on prior publications demonstrating homeostatic expansion at similar doses (18, 53, 54). Transfer efficiency into B6.Foxn1 hosts was validated by persistence of ≥5% CD8⁺, CD4⁺, or both, T cells within splenic lymphocytes 3 wk after injection (18).

The brain and spleen were harvested from mice and perfused with saline under deep anesthesia using a ketamine and xylazine (40 to 50 mg/kg i.p.) cocktail. Whole brains were weighed after removal of the cerebellum, brainstem, and olfactory bulb. Right hemispheres were flash frozen at −80 °C for protein studies, and homogenized. Cell lysates were separated into detergent-soluble and detergent-insoluble fractions. Left hemispheres were immersion fixed in 4% paraformaldehyde for 24 to 48 h and used for immunohistochemical staining.

Brain sections were mounted on slides and incubated at 4 °C overnight with primary antibody in blocking buffer, rinsed, and incubated 90 min in fluorochrome- or biotin-conjugated secondary antibody, with or without curcumin, or with ThioS alone. Sections were mounted with DAPI (Invitrogen) unless otherwise indicated, and images analyzed with ImageJ (NIH). Antibodies used: anti-Aβ/APP antibody (ab14220; clone 4G8\); Anti-pTau pS199/202 antibody (Invitrogen) and AT8; β-actin (clone AC-74, Sigma) due to cross-reactivity of GAPDH with IgG H chain (55); GAPDH used as control for all other markers; anti-GFAP (Dako); anti-NeuN antibody (Chemicon); anti-Iba1 (Wako, Ltd.); anti-CD8 (clone 53-6.72, BD Pharmingen); anti-doublecortin (DCX; polyclonal sc-8066, Santa Cruz Biotechnology); secondary antibodies (HRP, Alexa Flour-488, -594, -647; Invitrogen). control HLA-A2 and custom APP epitope dextramers (predicted affinity < 100 nM per NetMHC version 3.4 [APP_(471–479)/HLA-A2], were manufactured by Immudex.

Triton-soluble cell lysates were electrophoresed and blotted onto 0.2 µm nitrocellulose. Membranes were blocked and incubated in primary and secondary antibodies 1 h at room temperature with ≥3 washes between each, developed with ECL (GE Healthcare Biosciences; Pittsburgh, PA), and exposed onto Hyperfilm. WBs were performed a minimum of three times with comparable results per antibody. Bands were quantified using ImageJ software and expressed as (% area of test band)/(average % area of control bands).

Supernatant from homogenized brain was used for Triton-soluble Aβ. Insoluble pellets from Triton-homogenized brain were resuspended in 10 volumes 5M guanidine HCl 4 h to generate guanidine-soluble Aβ. Triton- and guanidine-soluble samples were subjected to analysis using species- and isoform-specific antibodies in Soluble and Insoluble Aβ ELISA kits (Invitrogen, Life Technologies). Absorbance was read on a SPECTRAmax Plus384 microplate reader (Molecular Devices, Sunnyvale, CA) and data analyzed in Graphpad PRISM. ELISAs were carried out a minimum of three times with comparable results per antibody/kit, with samples tested in triplicate and standard curves included on each plate.

Lymphocytes stained with respective antibodies were analyzed by three-color flow cytometry. Flow cytometry was performed with CD8-APC, HLA-A2-PerCP-Cy5.5, KLRG1-FITC antibodies (all Biolegend), and MHC A*0201-PE (Immudex) dextramers on a FACSCanto flow cytometer (BD Biosciences, San Jose, CA). Antibodies and multimers were incubated with whole-blood single-cell suspension in PBS with 2% FBS, on ice for 30 min, washed, and analyzed. FSC vs SSC was used to distinguish between lymphocytes and dead cells/debris, with exclusion of cell multiplets. 100,000 to 300,000 flow events were acquired and analyzed. FlowJo software was used for data analysis using identical gating (shown in SI Appendix, Figs. S13 and S15A).

Free floating brain sections were placed in 5% periodic acid 3 min, washed 2×, and placed in silver iodide solution 1 min, followed by incubation in 0.5% acetic acid 5 min (2×), and rinsed with dH₂0. Sections were incubated in developer for ~10 min until sections were pale brown/gray, stopped in 0.5% acetic acid 5 min, rinsed in dH₂O, and mounted. Stained sections were examined by microscopy. Stained neurons were counted from CA2, and proportions within total neurons visually quantified in triplicate from entorhinal and cingulate cortex.

Whole-number neuronal estimates were performed using the optical fractionator method (56) with stereological software (Stereo Investigator; MBF Bioscience). Paramedian sagittal serial sections spaced 50 µm apart were stained with NeuN. CA1, CA2, CA3, and other regions of interest were defined according to the Paxinos and Watson mouse brain atlas. A grid was placed randomly over the ROI, and cells were counted within three-dimensional optical dissectors (50 µm, 50 µm, 10 µm) using a 100× objective. Estimated totals weighted by section thickness were obtained with Stereo Investigator software, yielding a coefficient of error 0.10.

Testing order was randomized by alternating control and treatment group animal runs. Testing started at the same time (±1.5 h) for tests run over multiple days, with early and late times alternated for intergroup randomization. Testing was performed during the light phase exclusively. OFT was performed preceding all other behavioral tests to rule out motility effects.

rRNA was depleted with NEBNext® rRNA Depletion Kit v2 and libraries prepared. Data were analyzed by ROSALIND® (https://rosalind.bio/), with their HyperScale architecture (San Diego, CA). Database sources referenced for enrichment analysis included Interpro, NCBI, MSigDB, REACTOME, and WikiPathways. Enrichment was calculated relative to a set of background genes relevant for the experiment.

Cohort 1: 40 control individuals (CTRL); 52 MCI patients with an AD-characteristic CSF biomarker profile (MCI-AD); 36 MCI patients not displaying an AD-characteristic CSF biomarker profile (MCI); 50 sporadic AD patients with an AD-characteristic CSF biomarker profile (AD). CSF samples were collected at Middelheim General Hospital (Antwerp, Belgium) according to described protocols (58). Inclusion criteria for controls: 1) no neurological or psychiatric antecedents and 2) no organic disease involving the central nervous system upon extensive clinical examination. MCI patients were diagnosed applying Petersen’s diagnostic criteria (59) without dementia (60). AD dementia was clinically diagnosed according to NINCDS/ADRDA and IWG-2 criteria (61).

Cohort 2: Cognitive Testing. 29 self-referred memory clinic patients from Cedars-Sinai Department of Neurosurgery clinically diagnosed as normal (n = 6), MCI (n = 18), dementia (n = 3), or uncertain (n = 2; on basis of cognitive testing) and were administered Montreal Cognitive Assessment (MoCA) (n = 22). HLA-A2-negative patients determined by flow cytometry were excluded from analysis, as were patients with ≤2.5% CD8⁺ cells in lymphocyte gates, or ≥20% variation in any staining parameter between duplicate samples.

Cohort 3: Brain western and IHC. Hippocampal lysates from 13 autopsy-confirmed sporadic AD patients, and 5 age-matched normal controls were run on WBs using anti-CD8 or anti-PRF1. Hippocampal sections from 10 autopsy-confirmed sporadic AD patients, and 10 age-matched normal controls were stained with anti-CD8-fluorescein plus APP_(471–479)/HLA-A2-PE. HLA-A2-negative samples were not excluded from western and IHC/IF analysis.

Power analyses for sample sizes and statistical methods are detailed in SI Appendix, Supplementary Test: Detailed Materials & Methods. Briefly, GraphPad Prism (version 5.0b; San Diego, CA) was used to analyze the data using appropriate statistical tests when data was normally or nonnormally distributed. Subject numbers and methods of reagent validation are shown in SI Appendix, Table S1, and all histograms depict average ± SEM.

All animal procedures were approved prior to performance by the Cedars-Sinai Institutional Animal Care and Use Committee. The Cedars-Sinai Institutional Review Board designated the analysis of deidentified human brain specimens from UC Davis exempt from committee review. Brain specimens were collected, stored, and disseminated with prior approval by the UC Davis Medical Center Institutional Review Board. Sampling for cohort 1 was approved by the Medical Ethics Committee of the Hospital Network Antwerp (ZNA), Antwerp, Belgium.

R.C., R.N.P., J.A.R., D.K.I., B.A.W., and C.J.W. designed research; A.P., A.R., M.J., R.M.C., R.C., N.G., R.N.P., G.D., A.M., D.G., L.-W.J., D.V.D., Y.V., H.D.R., P.P.D.D., D.K.I., B.A.W., and C.J.W. performed research; A.R., R.C., R.N.P., L.-W.J., Y.V., H.D.R., P.P.D.D., K.L.B., D.K.I., B.A.W., and C.J.W. contributed new reagents/analytic tools; A.P., A.R., M.J., R.M.C., R.C., N.G., R.N.P., G.D., A.M., H.S., D.V.D., Y.V., H.D.R., P.P.D.D., D.K.I., B.A.W., and C.J.W. analyzed data; R.N.P., L.-W.J., and P.P.D.D. advised on paper framing & design; J.A.R. facilitated collaborations, advised on paper framing & design; K.L.B. and C.J.W. facilitated collaborations, supervised facilities and personnel; and A.P., A.R., R.C., and C.J.W. wrote the paper.

C.J.W. is the author of patents PCT/US2016/049598, WO2017/040594, and PCT/US2019/017879. R.C. and K.L.B. are co-authors on patent PCT/US2019/017879. PCT/US2016/049598 and WO 2017/040594 are licensed by Cedars-Sinai Medical Center to T-Neuro Pharma, Inc. C.J.W. has received salary and ownership interest in T-Neuro Pharma, Inc.