Abstract

   Immune mechanisms contribute to the neuropathology of Alzheimer’s
   disease (AD) but the role of adaptive immune cells is unclear. Here we
   show that the brain CD8^+ T cell compartment is dysregulated in AD
   patients and in the 3xTg-AD mouse model, accumulating activated CD103^–
   tissue-resident memory T cells that produce large amounts of granzyme K
   (GrK). These CD103^–CD8^+ T cells originate from the circulation and
   migrate into the brain using LFA-1 integrin. Ablation of brain
   CD103^–CD8^+ T cells in 3xTg-AD mice ameliorates cognitive decline and
   reduces neuropathology. GrK induces neuronal dysfunction and tau
   hyperphosphorylation in human and mouse cells via protease-activated
   receptor-1 (PAR-1), which is expressed at higher levels in the AD
   brain, revealing a key immune-mediated neurotoxic axis. We conclude
   that communication between CD8^+ T cells and the nervous system is
   altered in AD, paving the way for therapies targeting T cell-dependent
   neurotoxic inflammation.

   Subject terms: Neuroimmunology, Alzheimer's disease
     __________________________________________________________________

   Neuroimmune communication in Alzheimer’s disease (AD) is poorly
   understood. Here, the authors show that CD103–CD8+ resident memory T
   cells induce neurotoxicity and neuroinflammation in AD via the granzyme
   K–PAR-1 axis.

Introduction

   The dysregulation of innate and adaptive immunity is a driving force in
   the development of Alzheimer’s disease (AD), the most common form of
   dementia, affecting more than 32 million people worldwide^[66]1.
   Classical AD neuropathology is characterized by β-amyloid (Aβ)
   deposition, tau hyperphosphorylation, and the loss of neurons and
   synapses, but more recent evidence shows that chronic inflammation
   promoted by local and peripheral immune cells is also a hallmark of
   AD^[67]2,[68]3.

   CD8^+ T cells are one of the most intriguing components of the AD
   immunological landscape. During immune responses, these cytotoxic cells
   produce classical granzymes (such as GrA and GrB), perforin, and a
   plethora of cytokines^[69]4. Most CD8^+ T cells in healthy non-lymphoid
   organs have a tissue-resident memory (Trm) phenotype, fulfill a local
   protective role, and can rapidly mount immune responses^[70]5. However,
   CD8^+ T cells also infiltrate the brains of AD patients and equivalent
   animal models^[71]6–[72]13. Aging is the main risk factor for AD and
   recent studies have shown that CD8^+ T cells also accumulate as the
   brain ages, promoting axonal degeneration by releasing GrB^[73]14. But
   although CD8^+ T cells are found in close proximity to neuronal
   structures in AD, it is unclear whether this communication between
   adaptive immune cells and neural cells promotes disease
   development^[74]9. Furthermore, recent studies suggest that the
   meningeal compartment, which is highly enriched in immune cell
   populations, may also contribute to AD, but the role of meningeal CD8^+
   T cells in disease development is unclear^[75]15.

   In most previous studies of AD brains, CD8^+ T cells have been
   considered as a single homogeneous population^[76]6–[77]8,[78]10.
   However, there is clear evidence in other tissues to support the
   presence of CD8^+ T cell subsets with district roles during homeostasis
   and immune responses^[79]4. The population of CD8^+CD45RA^+CCR7^- T
   effector memory (T[EMRA]) cells expands in the peripheral blood of
   individuals with mild cognitive impairment (MCI) and AD, and in the
   cerebrospinal fluid (CSF) following infection with Epstein–Barr virus
   (EBV), suggesting particular CD8^+ T cell phenotypes may also play a
   role in AD^[80]9. Furthermore, CD69^+CD103^+ and CD69^+CD103^–CD8^+ Trm
   cells have been observed in the human brain^[81]16, but the findings
   involved a heterogeneous group of subjects with multiple sclerosis
   (MS), various types of dementia and bipolar disorders, as well as
   controls with no brain disease. AD patients formed a minority in this
   mixed group, so it was not possible to determine how CD8^+ Trm cell
   populations are represented in the AD brain.

   The presence of CD8^+ T cell subsets has also been proposed in
   transgenic animal models of AD. However, these studies involved mice
   with separate Aβ or tau pathology and showed that CD8^+ T cells may
   have protective or deleterious effects depending on the disease stage
   and type of pathology^[82]7,[83]10–[84]13,[85]17. In APP/PS1 mice with
   late-stage Aβ pathology, the abundance of CD103^+CD8^+ T cells
   increased, and their transcriptomic profile was similar to wild-type
   (WT) mice, potentially explaining why the depletion of CD8^+ T cells
   does not affect amyloid pathology^[86]10,[87]11. The Trm marker CD103
   (αE integrin) is also more abundant in the 5xFAD mouse model, which
   develops an early and aggressive amyloid pathology, together with
   higher levels of CXCR6 and PD-1 compared to WT controls, and this CD8^+
   T cell subset has a protective anti-Aβ role suppressing the activation
   of microglia^[88]13. Although CD8^+ T cells generally seem to be
   deleterious in a model of pure tauopathy^[89]17, the role of specific
   subsets is less clear. P301S mice expressing APOE4 accumulate activated
   CD11c^+, KLRE1^+ and ISG15^+ CD8^+ T cells while the pool of
   TOX^+PDCD1^+ exhausted CD8^+ T cells is depleted. However, P301S mice
   lacking APOE show no increase in the abundance of brain CD8^+ T cells
   despite tauopathy, suggesting additional signals are needed for CD8^+ T
   cell responses in AD^[90]12.

   Aβ and tau pathologies have a synergistic effect in AD and diagnosis
   requires the presence of both hallmarks^[91]18–[92]20. Here we
   performed single-cell RNA sequencing (scRNAseq) in the 3xTg-AD mouse
   model, which develops both amyloid and tau pathologies, to investigate
   how brain and meningeal CD8^+ T cell subsets may shape the
   neuropathology of AD. We found that CD8^+ Trm cells are strongly
   dysregulated in 3xTg-AD mice, with the number of activated CD103^–
   cells increasing as the CD103 expression level declines in the CD8^+ T
   cell population. We showed that the brain is the main site of CD8^+ T
   cell dysregulation and found that CD103^–CD8^+ T cells originate from
   the circulation and invade the brain using a mechanism dependent on
   integrin LFA-1. We also observed higher levels of granzyme K (GrK) in
   the CD103^–CD8^+ T cells of 3xTg-AD mice and patients with AD compared
   to controls and discovered a role for GrK in the induction of neuronal
   dysfunction through the activation of protease-activated receptor 1
   (PAR-1). Finally, we demonstrated that GrK–PAR-1 interaction induces
   tau hyperphosphorylation, revealing a critical immune-mediated
   neurotoxic axis. Together, our data show that dysfunctional
   communication between the immune system and central nervous system
   (CNS) mediated by CD103^–CD8^+ T cells and GrK–PAR-1 signaling
   contributes to the development of AD, identifying key molecular
   mechanisms that can be targeted to prevent immune-mediated neurotoxic
   inflammation.

Results

CD103^–CD8^+ Trm cells accumulate in the brains of 3xTg-AD mice

   Previous studies suggested that CD8^+ T cells play a detrimental role
   in AD, but the subsets that drive the disease and the underlying
   mechanisms that contribute to disease development are still
   unclear^[93]6–[94]12. We therefore compared the phenotypes and
   functions of CD8^+ T cells in WT and 3xTg-AD transgenic mice, the
   latter developing amyloid and tau pathologies representing
   neuropathological characteristics of human AD patients. Single-cell RNA
   sequencing (scRNAseq) of CD45^HIGH leukocytes isolated from the
   meninges and brains of 3xTg-AD (n = 8) mice as well as sex and
   age-matched WT controls (n = 8) at the onset of cognitive
   deficit^[95]21,[96]22 (Fig. [97]1a) revealed 13 cell types, including
   CD8^+ and CD4^+ T cells, neutrophils, B cells, natural killer (NK)
   cells, innate lymphoid cells (ILCs), macrophages, and microglia
   (Fig. [98]1b; Supp. Fig. [99]1a-c). Clustering analysis applied to the
   whole CD8^+ T cell population (3,098 cells representing 19.06% of all
   CD45^HIGH leukocytes; Supp. Fig. [100]1c) revealed the presence of five
   subsets (Fig. [101]1c), each characterized by the strong expression of
   known marker genes (Fig. [102]1d-f; Supp. Data [103]1): (i) T central
   memory (Tcm) cells (4.52%) expressing Sell, Ccr7, Nsg2, Dapl1 and
   Cmah^[104]23,[105]24; (ii) T effector cells (11.43%) expressing Gzma,
   S1pr5, Cx3cr1, Zeb2, Klrg1 and Klf2^[106]24,[107]25; (iii) CD103^– Trm
   cells (64.59%) expressing Gzmk, Cxcr6, Cxcr3, Ltb, Xcl1, Tnf and
   Eomes^[108]26,[109]27; (iv) proliferating Trm (Trm^PROL) cells (4.29%)
   expressing Chek1, Stmn1, Cdk1, Mki67, Mcm2, Mcm5, Mcm7, Tfdp1 and
   Pola2^[110]28,[111]29 and (v) CD103^+ Trm cells (15.24%) expressing
   Gstp3, Foxo1, Il10, Il17a and Il2ra, as well as Itgae, encoding CD103,
   which is related to the tissue residence phenotype^[112]30–[113]32.
   Cd69, previously associated with the tissue residence phenotype but
   also with the activation of T cells^[114]5, was expressed by most of
   the CD8^+ T cells in both 3xTg-AD and WT mice (Supp. Fig. [115]1d).
   Flow cytometry experiments confirmed this phenotypic characterization
   of CD8^+ T cells showing that: (i) Tcm highly expressed CD62L, encoded
   by Sell gene, and were negative for CD69; (ii) T effector highly
   expressed CX3CR1, granzyme A (GrA) and granzyme B (GrB), and were
   negative for CD69; (iii) CD103^– Trm cells were positive for CD69^+ and
   expressed high levels of Eomes; and (iv) CD103^+ Trm cells were
   positive for CD69^+ and expressed low levels of Eomes (Fig. [116]2a, b;
   Supp. Fig. [117]5f–h). Very few Trm^PROL cells were found in the
   scRNAseq dataset and these cells were not characterized by flow
   cytometry.

Fig. 1. Characterization of CD8 T cell phenotype in 3xTg-AD and WT mice.

   [118]Fig. 1
   [119]Open in a new tab

   a Graphical overview of the scRNAseq experimental design created in
   BioRender. Terrabuio, E. (2025) [120]https://BioRender.com/f3lmkb2. b
   UMAP plot showing CD45^HIGH leukocytes detected in the brains and
   meninges of wild-type (WT; n = 8) and 3xTg-AD (n = 8) mice. CD8^+ T
   cells (n = 3,098) are shown in red. c UMAP plots showing the five
   subsets of CD8^+ T cells detected in the brains and meninges of WT
   (n = 8) and 3xTg-AD (n = 8) mice. d Bubble plot reporting the
   phenotypic marker genes for each CD8^+ T cell subset. Transcript levels
   are color-coded. e Normalized expression of known marker genes on UMAP
   plot. Transcript levels are color-coded. f Violin plots showing the
   expression of marker genes for each CD8^+ T cell subset. White dashed
   line indicates the median expression level.

Fig. 2. The CD8^+ Trm compartment is altered in the brains of 3xTg-AD mice.

   [121]Fig. 2
   [122]Open in a new tab

   a, b Violin plots (g) and bubble plot (h) showing the expression of
   phenotypic markers detected by flow cytometry in the brains and
   meninges of WT (n = 3 pools of two organs each) and 3xTg-AD (n = 3
   pools of two organs each) mice. White dashed line indicates the median
   expression level in violin plots (g). MFI expressions in the bubble
   plot (h) are scaled and color-coded. P-values are based on two-way
   ANOVA - multiple comparisons. Source data are provided as a Source Data
   file. c Donut plots indicating the distribution of CD8^+ T cell subsets
   in the brains of WT (left) and 3xTg-AD (right) mice d Heat map showing
   the expression of residency genes in the brains and meninges of WT and
   3xTg-AD mice. Transcript levels are color-coded. e Violin plots showing
   gene expression in the brains of WT (n = 8) and 3xTg-AD (n = 8) mice. f
   Heat map showing the expression in CD45^HIGH leukocytes of genes in
   brains of WT (n = 8) and 3xTg- AD (n = 8) mice. Transcript levels are
   color-coded.

   To better understand differences in the CD8^+ population between
   3xTg-AD and WT mice, we separately analyzed the phenotypic changes in
   the brain and meningeal compartments. Notably, scRNAseq analysis
   revealed an almost three-fold reduction in the abundance of
   CD103^+CD8^+ Trm cells in the brains of 3xTg-AD mice (WT = 27.44%,
   3xTg-AD = 9.8%) with a parallel increase in the abundance of
   CD103^–CD8^+ Trm cells (WT = 45.9%, 3xTg-AD = 73.69%) (Fig. [123]2c).
   Notably, the probability to observe CD103^–CD8^+ Trm cells was
   significantly higher (OR = 3.30; P-value = 0) in the brain of 3xTg-AD
   mice compared to those of WT controls (Data source Fig. [124]2). In
   contrast, the probability to observe CD103^+CD8^+ Trm cells was
   significantly lower (OR = 0.29; P-value = 0.0038) in the brain of
   3xTg-AD mice compared to those of WT controls (Data source
   Fig. [125]2). However, these changes were not evident in the meninges,
   where there was no significant accumulation of CD103^– cells
   (OR = 0.72; P-value = 0.09) and the proportion of CD103^+ cells
   (WT = 11.97%, 3xTg- AD = 19.01%; OR = 1.72; P-value = 0) increased only
   slightly (Supp. Fig. [126]1e; Data source Supp. Fig.[127]1). These data
   suggest that the major alterations in the CD8^+ Trm cell compartment
   occur in the brain, as corroborated by data showing the downregulation
   of several transcription factor (TF) genes related to the CD103^+
   tissue residence phenotype (Fabp5, Stat1, Stat3, Stat4, Stat5a, Stat5b,
   Smad4, Smad6, Smad7 and Smad9) and the suppression of genes encoding
   key cytokines involved in the differentiation and maintenance of
   CD103^+ Trm cells (Il15, Il7, Tgfb3, Tgfb2, Tgfb1, Il21, Tnf and Il33;
   Fig. [128]2d, f). Moreover, the expression level of Itgae was reduced,
   whereas the expression of Eomes and S1pr1, which inhibits the
   CD103^+CD8^+ T cell phenotype^[129]5,[130]31, was upregulated in CD8^+
   T cells from the brains of 3xTg-AD mice compared to WT controls
   (Fig. [131]2e). These data highlight the dysregulation of brain CD8^+ T
   cells and the depletion of the CD103^+ Trm population, presumably
   affecting CNS immunity and contributing to disease development.

   The results above were supported by unbiased cell fate trajectory
   analysis showing that CD8^+ T cell differentiation followed a tightly
   organized trajectory along with the pseudo-time progression
   (Fig. [132]3a), starting from a common root and ending with two
   differentiation states distinguished by Itgae (Arm A) and Eomes (Arm B)
   expression (Fig. [133]3b). Only slight differences were observed in the
   meninges whereas there were clear differences in the brain
   (Fig. [134]3c, Supp. Fig. [135]1f–l), where Arm A of the trajectory
   plot was populated by a lower proportion of CD103^+CD8^+ Trm cells in
   3xTg-AD mice compared to WT controls (WT = 26.1%, 3xTg-AD = 14.9%)
   accompanied by a higher percentage of CD103^–CD8^+ Trm cells in Arm B
   (WT = 54.5%, 3xTg-AD = 75.8%; Fig. [136]3c; Supp. Fig. [137]1h-l). This
   supports the hypothesis that the brain is a fundamental point of CD8^+
   T cell dysregulation in 3xTg-AD mice.

Fig. 3. GrK expression is upregulated in brain CD103^–CD8^+ Trm cells of
3xTg-AD mice.

   [138]Fig. 3
   [139]Open in a new tab

   a Pseudo-time ordered trajectory plot of CD8^+ T cells in the brains
   and meninges of WT and 3xTg- AD mice. b Expression of genes in the
   trajectory plots. Transcript levels are color-coded. c Trajectory plot
   indicating the distribution of brain CD103^– and CD103^+CD8^+ Trm cells
   in WT and 3xTg-AD mice. d, e Abundances of brain CD103^– (d) and
   CD103^+ (e) CD8^+ Trm cells in WT (n = 4) and 3xTg-AD (n = 4) mice.
   Data are means ± SD. P-values based on two-tailed Student’s t-test. f
   Scatter plot showing KEGG pathway enrichment analysis for CD103^– and
   CD103^+ CD8^+ Trm cells in the brains of 3xTg-AD mice (PD = “Parkinson
   disease”, ALS = “Amyotrophic lateral sclerosis”). g Bubble plot showing
   BP-GO GSEA analysis for brain CD103^–CD8^+ Trm cells in 3xTg-AD mice. h
   Normalized gene expression on UMAP plot of brain and meningeal CD8^+ T
   cells. Transcript levels are color-coded. i Violin plots showing gene
   expression in CD103^– and CD103^+CD8^+ Trm cells in the brain and
   meninges. j Violin plot showing gene expression in CD8^+ T cells in the
   brains of WT and 3xTg-AD mice. k Representative flow cytometry plot
   showing GrK expression in brain CD8^+ T cells in WT and 3xTg-AD mice. l
   Bar plots showing the percentage of GrK^–, GrK^MED, and GrK^HIGH
   CD103^– CD8^+ Trm cells in the brains of WT (n = 5) and 3xTg-AD (n = 5)
   mice. Data are means ± SD. P-values based on two-tailed Student’s
   t-test. m Scatter plot showing MFIs of CD103 and GrK in CD8^+ T cells
   in the brains of WT (n = 5) and 3xTg-AD (n = 5) mice. n, o
   Representative histograms (n) and violin plots (o) showing GrK MFI for
   brain CD103^– (red) and CD103^+ (blue) CD8^+ Trm cells in 3xTg-AD mice.
   White dashed lines indicate median expressions. P-values based on
   two-tailed Student’s t-test. p CD8^+ T cells (squares) in the brains of
   WT and 3xTg-AD mice detected by immunofluorescence microscopy. Scale
   bar = 5 µm or 2 µm for zoomed images.

   Flow cytometry showed that most brain CD8^+ T cells were CD69^+ (Supp.
   Fig. [140]1m) and clearly validated the presence of immune
   dysregulation in the brain CD8^+ Trm cell compartment. We observed a
   dramatic increase in both the proportion and absolute numbers of
   CD69^+CD103^– Trm cells in the brains of 6-month-old 3xTg-AD mice
   compared to sex/age-matched WT controls (WT proportion = 48.025%,
   3xTg-AD = 88.08%; WT number = 6,224.72, 3xTg-AD = 12,760.5), with
   CD69^+CD103^+ Trm cells showing the opposite profile (Fig. [141]3d, e).
   However, we detected no significant differences between genotypes in
   the numbers of CD103^–CD8^+ Trm cells populating the meninges (WT
   number = 9,018.66, 3xTg-AD = 8,895.18), highlighting the brain as a key
   site for pathological changes in 3xTg-AD mice (Supp. Fig. [142]1n, o).

Brain CD103^–CD8^+ T cells show an activated phenotype strongly associated
with AD

   Gene set enrichment analysis (GSEA) using the Kyoto Encyclopedia of
   Genes and Genomes (KEGG) database reported “pathway of
   neurodegeneration – multiple diseases”, “Alzheimer’s disease” (AD), and
   “Huntington disease” (HD) as the three most enriched terms in the
   brains of 3xTg-AD mice (Supp. Fig. [143]1p), whereas these pathways
   were not enriched in the meninges (Supp. Fig. [144]1q). Importantly,
   whereas CD103^+CD8^+ Trm cells were negatively associated with the AD
   pathway suggesting a protective role, CD103^–CD8^+ Trm cells
   infiltrating the brains of 3xTg-AD mice showed a strong positive
   association with this pathway (Supp. Data [145]1), further suggesting
   these cells may be specifically involved in AD (Fig. [146]3f).
   Moreover, GSEA applied to the biological process Gene Ontology (BP-GO)
   database revealed an activated phenotype for CD103^–CD8^+ Trm cells
   populating the brains of 3xTg-AD mice compared to WT controls. Indeed,
   the five most enriched biological processes included “Immune system
   process” (GO:0002376) and “Regulation of immune system process”
   (GO:0002682), indicating pathways involved in the development of immune
   responses, “Nitrogen compound transport” (GO:0071705), which is related
   to T cell activation^[147]33, and “Hydrolase activity” (GO:0016787),
   which mediates the secretion of cytotoxic granules, suggesting a
   potential pro-inflammatory and cytotoxic role for CD103^–CD8^+ T cells
   in AD^[148]34 (Fig. [149]3g). Collectively, our data show a profound
   dysregulation of the CD8^+ Trm cell compartment in the brains of
   3xTg-AD mice, with a loss of CD103^+ cells and the accumulation of
   activated CD103^– cells potentially contributing to neurotoxicity and
   AD development.

CD103^–CD8^+ Trm cells produce granzyme K in the brains of 3xTg-AD mice

   We next sought to identify the pathological mechanisms mediated by
   activated CD103^– cells under AD-like conditions and found that Gzmk,
   encoding GrK, was one of the most strongly expressed marker genes in
   the global CD103^–CD8^+ Trm cell population (brain and meninges;
   Fig. [150]3h, i; Supp. Fig. [151]2a; Supp. Data [152]1). Interestingly,
   the genes encoding other granzymes (Gzma and Gzmb) and perforin-1
   (Prf1) were not found in the list of marker genes in CD103^–CD8^+ Trm
   cells compared to other subsets (Supp. Data [153]1). Flow cytometry
   experiments validated these data confirming the low expression of GrA
   and GrB but high expression of GrK in the CD103^– CD8^+ Trm population
   (Fig. [154]2a, b), suggesting GrK may have a selective functional role
   in the CD103^– CD8^+ Trm cell subpopulation in AD.

   Importantly, Gzmk expression was upregulated in the brains but not in
   the meninges of 3xTg-AD mice compared to WT controls (Fig. [155]3j;
   Supp. Fig. [156]2b). Flow cytometry confirmed the scRNAseq data,
   showing a dramatic increase in the percentage of GrK-producing
   CD103^–CD8^+ Trm cells (both GrK^MED and GrK^HIGH) infiltrating the
   brain, but not the meninges, of 3xTg-AD mice compared to WT controls,
   whereas no differences were observed for brain-infiltrating
   CD103^–GrK^– CD8^+ T cells (Fig. [157]3k–m; Supp. Fig. [158]2c).
   Importantly, we also found that the strong increase in intracellular
   GrK expression was selective for the CD103^–CD8^+ Trm cell population
   (86% of which produced GrK) when compared to the CD103^+ population
   (Fig. [159]3n, o), suggesting the production of GrK is a key
   pathological mechanism of CD103^–CD8^+ Trm cells in the brains of
   3xTg-AD mice.

GrK^+CD103^–CD8^+ Trm cells induce neuronal dysfunction via the GrK–PAR-1
axis

   GrK is associated with pro-inflammatory and cytotoxic
   activities^[160]35, suggesting that GrK produced by CD103^–CD8^+ T
   cells may induce neuronal dysfunction in AD. Immunofluorescence
   staining revealed the presence of GrK^+CD8^+ T cells in close proximity
   to hippocampal neurons in the brains of 3xTg- AD mice, whereas CD8^+ T
   cells with lower GrK expression were predominantly detected in WT
   brains (Fig. [161]3p; Supp. Fig. [162]2j). These results indicate that
   GrK^+CD103^–CD8^+ Trm cells may contribute to neuronal alterations in
   3xTg-AD mice.

   To confirm that GrK^+CD103^–CD8^+ Trm cells directly induce neuronal
   dysfunction in AD, we used wide-field high-resolution microscopy for
   the live imaging of primary hippocampal neurons isolated from 3xTg-AD
   mice co-cultured with CD103^– or CD103^+ CD8^+ Trm cells obtained from
   the livers of 3xTg-AD mice, an abundant source of CD8^+ Trm
   cells^[163]36 (Fig. [164]4a). Immunofluorescence staining confirmed the
   high density of GrK granules inside CD103^–CD8^+ Trm cells compared to
   CD103^+ counterparts (Fig. [165]4b). Time-lapse fluorescence microscopy
   revealed that primary neurons in contact with GrK^+CD103^–CD8^+ Trm
   cells, but not those in contact with CD103^+CD8^+ Trm cells, showed
   significantly higher cytoplasmic calcium (Ca^2+) levels compared to the
   negative control, clearly indicating that GrK^+CD103^–CD8^+ Trm cells
   induce intracellular Ca^2+ dysregulation (Fig. [166]4c), which is
   associated with neuronal alterations^[167]37. Importantly, purified
   active GrK directly induced intracellular Ca^2+ release in a
   dose-dependent manner, unequivocally demonstrating the detrimental role
   of this enzyme in the context of neuronal dysfunction (Fig. [168]4d).

Fig. 4. GrK^+CD103^–CD8^+ Trm cells induce neuronal alterations by engaging
PAR-1.

   [169]Fig. 4
   [170]Open in a new tab

   a Graphical overview of the experimental design. Created in BioRender.
   Terrabuio, E. (2025) [171]https://BioRender.com/ir5h5rf. b
   Immunofluorescence staining showing GrK granules inside CD103^+ and
   CD103^– CD8^+ Trm cells. Scale bar = 3 µm. c Intracellular Ca^2+
   release in neurons co-cultured with CD103^+ or CD103^– CD8^+ Trm cells.
   The last time point is shown in the right panel. Data are means ± SD
   from four independent experiments. P-values based on two-way
   ANOVA-multiple comparisons. Ctrl^– = neurons alone. Ctrl^+ =
   ionomycin-stimulated neurons (10 µM). Source data are provided as a
   Source Data file. d Intracellular Ca^2+ release in neurons treated with
   purified active GrK or vehicle. The last time point is shown in the
   right panel. Data are means ± SD from three independent experiments.
   P-values based on two-way ANOVA-multiple comparisons. Source data are
   provided as a Source Data file. e Intracellular Ca^2+ release in
   neurons co-cultured for 5 h with CD103^–CD8^+ Trm cells in the
   presence/absence of the PAR-1 inhibitor [172]SCH79797 (100 nM). The
   last time point (300 min) is shown in the right panel. Data are
   means ± SD from four independent experiments. P-values based on two-way
   ANOVA-multiple comparisons. Ctrl^– = neurons alone. Ctrl^+ =
   ionomycin-stimulated neurons (10 µM). Source data are provided as a
   Source Data file. f, g Intracellular Ca^2+ release in neurons cultured
   for 5 h with purified active GrK alone (150 nM) or with [173]SCH79797
   (50 nM, or 100 nM). Ctrl^– = neurons alone. The last time point is
   shown in the right panel. Data are means ± SD from three independent
   experiments. P-values based on two-way ANOVA-multiple comparisons.
   Representative images are shown in (g). Red = intracellular Ca^2+
   release. Scale bar = 20 µm. Source data are provided as a Source Data
   file. h Immunofluorescence microscopy showing GrK^+CD103^–CD8^+ Trm
   cells near the soma (cell I) and dendrites (cell II) of a PAR-1^+
   neuron. Cell morphology was visualized by wide-field imaging using a
   DIC filter. Scale bar = 5 µm (or 2 µm for zoomed images I, and II).

   GrK can bind the thrombin receptor PAR-1^[174]38, which is implicated
   in synaptic plasticity and memory formation in healthy mice, but also
   plays a negative role in some brain pathologies^[175]39–[176]41.
   Importantly, immunofluorescence staining revealed a significant
   increase in both the number and area of PAR-1^+ neurons in the
   hippocampus of 3xTg-AD mice compared to WT controls, suggesting that
   signaling via PAR-1, a G-protein coupled receptor (GPCR), is
   dysregulated in AD (Supp. Fig. [177]2g, h). Importantly, wide-field
   time-lapse microscopy showed that the PAR-1 inhibitor [178]SCH79797
   strongly reduced intra-neuronal Ca^2+ release in the presence of
   GrK^+CD103^–CD8^+ Trm cells (Fig. [179]4e). Moreover, our data clearly
   showed the strong, dose-dependent inhibition of intracellular Ca^2+
   release in neurons cultured with 150 nM purified active GrK in the
   presence of [180]SCH79797 (Fig. [181]4f, g; Supplementary
   Movies [182]1-[183]3). Immunofluorescence staining confirmed the
   presence of GrK^+CD103^–CD8^+ Trm cells in close contact with PAR-1^+
   neurons in in vitro cultures (Fig. [184]4h), and of GrK^+ CD8^+ T cells
   in the hippocampus of 3xTg-AD mice (Supp. Fig. [185]2i), further
   supporting a role for GrK–PAR-1 binding in the neuronal dysfunction
   underlying AD. These data demonstrate that the GrK–PAR-1 axis is a key
   immune-mediated pathological mechanism promoting neuronal dysfunction
   in AD.

Brain CD103^–CD8^+ Trm cells originate from the circulation and are
detrimental in 3xTg-AD mice

   The intraperitoneal (ip) treatment of mice with an anti-CD8 antibody
   depletes circulating CD8^+ T cells leaving the brain Trm compartment
   unaltered^[186]42,[187]43. To determine the origin of CD103^– T cells,
   we therefore used this method to deplete circulating CD8^+ T cells in
   3xTg-AD mice and sex-matched WT controls (Supp. Fig. [188]3a-h). The
   anti-CD8 treatment significantly reduced the abundance of brain
   CD103^–CD8^+ Trm cells compared to mice treated with an isotype control
   antibody, reaching the levels of the control groups (Fig. [189]5a),
   suggesting this cell subset is replenished from the blood. However, we
   detected no differences in the abundance of CD103^+CD8^+ Trm cells
   after CD8^+ T cell depletion, in 3xTg-AD mice and WT controls
   (Fig. [190]5a), in line with previous data showing these cells are
   spared by systemic depletion^[191]43. Importantly, the ablation of
   CD103^–CD8^+ Trm cells in the brains of 3xTg-AD mice was paralleled by
   an amelioration of cognitive functions in behavioral tests, confirming
   that CD103^–CD8^+ Trm cells contribute to memory decline in 3xTg-AD
   mice. Particularly, the Morris water maze (MWM) test showed improved
   hippocampal-dependent learning after CD8^+ T cell depletion in 3xTg-AD
   mice compared to controls, as indicated by the learning curve recorded
   during the training days of the test (Fig. [192]5b). In addition, the
   probe test showed that 3xTg-AD mice devoid of circulating CD8^+ T cells
   actively searched the platform, maintaining the memory of the original
   platform position and achieving better escape performance, as also
   shown by the greater path efficiency and significant reduction in body
   rotations (Fig. [193]5c–e). The contextual fear conditioning (CFC)
   associative learning task revealed an amelioration of associative
   memory following the depletion of circulating CD8^+ T cells in 3xTg-AD
   mice, as shown by the significant increase in the percentage of
   freezing responses compared to control animals treated with an isotype
   control (Fig. [194]5f).

Fig. 5. Detrimental CD103^–CD8^+ Trm cells in the brain originate from the
circulation and their depletion ameliorates disease in 3xTg-AD mice.

   [195]Fig. 5
   [196]Open in a new tab

   a Flow cytometry showing the percentage of brain CD103^– and
   CD103^+CD8^+ Trm cells after anti-CD8 antibody treatment. Anti-CD8 (WT,
   n = 6; 3xTg-AD, n = 9), isotype-control (WT, n = 6; 3xTg-AD, n = 9).
   Data are means ± SD from two independent experiments and P-values are
   based on one-way ANOVA-multiple comparisons was used. b Time spent
   during training (MWM test). Anti-CD8 (WT, n = 10; 3xTg-AD, n = 21),
   isotype-control (WT, n = 22; 3xTg-AD, n = 18). Data are means ± SD from
   two-independent experiments (two-way ANOVA-multiple comparisons). c
   Representative tracking of three mice/group (MWM test). d, e Violin
   plots showing path efficiency (d) and number of body rotations (e) (MWM
   test). Data are from two-independent experiments one-way ANOVA-multiple
   comparisons). f Bar plot showing the percentage of freezing during the
   CFC test after anti-CD8 treatment. Anti-CD8 (WT, n = 9; 3xTg-AD,
   n = 16), isotype-control (WT, n = 21; 3xTg-AD, n = 13). Data are
   means ± SD from two-independent experiments. One-way ANOVA-multiple
   comparisons was used. g–i Immunohistochemical staining of the
   hippocampus after CD8 T cell depletion showing Aβ-load (g), the levels
   of hyperphosphorylated (h) and total (i) tau protein. Anti-CD8 (n = 3),
   isotype-control (n = 3). Scale bar = 20 µm. Data are means ± SEM.
   Two-tailed Student’s t-test was used. ROI = 624.7 µm×501.22 µm. j–l
   ELISA showing Aβ 1−40 and Aβ 1−42 (j) levels of tau
   hyperphosphorylation (k) and total tau (l) in soluble and insoluble
   fractions of brain homogenates after anti-CD8 treatment. Anti-CD8
   (n = 4), isotype control (n = 4). Data from three-independent
   experiments are means ± SEM (two-tailed Student’s t-test). m Dot blot
   showing insoluble oligomeric and fibrillar forms of Aβ in brain
   homogenates after anti-CD8 treatment. Anti-CD8 (n = 4), isotype-control
   (n = 4). Data from three-independent experiments are shown as
   means ± SEM. P-values based on two-tailed Student’s t-test. n
   Immunofluorescence staining showing hippocampal oligomeric and
   fibrillar Aβ. Scale bar = 10 µm.

   To confirm whether the amelioration of cognitive deficits by anti-CD8a
   antibody treatment correlates with the loss of neuropathological
   hallmarks of AD, we stained coronal murine brain sections with anti-Aβ
   (6E10), anti-phospho-tau (AT180), and anti-total-tau (HT7) antibodies.
   We observed a significant decrease in both the Aβ load and levels of
   tau hyperphosphorylation in the hippocampus of 3xTg-AD mice following
   the depletion of circulating CD8^+ T cells compared to animals treated
   with an isotype control (Fig. [197]5g, h), whereas the levels of total
   tau were unchanged (Fig. [198]5i). These data were confirmed by ELISA
   experiments, showing significantly less Aβ1-40 and Aβ1-42 deposition,
   and tau hyperphosphorylation (pT231, AT180) in the soluble and
   insoluble fractions of brain homogenates from 3xTg-AD mice depleted of
   CD8 + T cells compared to isotype controls (Fig. [199]5j, k), while no
   significant differences were observed in the levels of total tau
   (Fig. [200]5l). Similarly, we observed significantly lower levels of
   insoluble oligomeric (A11 antibody) and fibrillar (OC antibody) Aβ
   forms in the brain of 3xTg-AD mice depleted of CD8 + T cells compared
   to isotype controls (Fig. [201]5m, Supp. Fig. [202]3j, Supp. Data 5-8).
   Importantly, our immunofluorescence staining showed that the oligomeric
   and fibrillar forms of Aβ were located at the intraneuronal level in
   the hippocampus of 9 9-month-old 3xTg-AD mice, and both co-localized
   with the 6E10 signal (Fig. [203]5n).

   Collectively, these results demonstrate that CD103^–CD8^+ Trm cells
   originate from the circulation, infiltrate the brain, and promote
   memory decline and neuropathological changes that characterize AD,
   strongly supporting their role in AD pathogenesis.

LFA-1 integrin mediates the infiltration of CD103^–CD8^+ T cells into the
brain

   Next, we investigated the molecular mechanisms that allow circulating
   CD8^+ T cells to infiltrate the brains of 3xTg-AD mice. Our recent work
   suggests that α4-integrins are not involved in CD8^+ T cell
   extravasation in 3xTg-AD mice^[204]44. However, our flow cytometry data
   showed a significant increase in both the percentage and mean
   fluorescence intensity (MFI) of LFA-1^+ circulating CD8^+ T cells in
   the blood of 3xTg-AD mice compared to age-matched WT controls
   (Fig. [205]6a). These results, together with our earlier data showing
   the upregulation of brain endothelial intracellular adhesion molecule 1
   (ICAM-1), the counterligand of LFA-1^[206]22, suggest a role for LFA-1
   integrin in the trafficking of CD8^+ T cells into the brain. To
   understand the role of LFA-1 in more detail, we crossed 3xTg-AD mice
   with Itgal^–/– mice lacking functional LFA-1 integrin and observed a
   significant decrease in the percentage of CD103^–CD8^+ Trm cells in the
   brains of 3xTg-AD/Itgal^–/– mice compared to 3xTg-AD animals
   (Fig. [207]6b, c). Neuropathological studies showed a significantly
   lower Aβ load and significantly less tau hyperphosphorylation (AT180)
   in the brains of 3xTg-AD/Itgal^–/– mice compared to sex- and
   age-matched 3xTg-AD controls, but no significant difference in the
   levels of total tau (Fig. [208]6d–f). Overall, these data suggest a
   pivotal role for LFA-1 in the accumulation of CD103^–CD8^+ Trm cells in
   the brain, and further supports the origin of detrimental CD103^–CD8^+
   Trm cells in the circulation.

Fig. 6. LFA-1 integrin controls the accumulation of CD8^+CD103^– cells in the
brains of 3xTg- AD mice.

   [209]Fig. 6
   [210]Open in a new tab

   a Flow cytometry showing the percentage of LFA-1^+ CD8^+ T cells (left)
   and corresponding MFIs (right) in the blood of WT (n = 6) and 3xTg-AD
   (n = 6) mice. Data are means ± SD, with P-values based on two-tailed
   Student’s t-test. b Flow cytometry showing the percentage of CD45+
   leukocytes among live cells. Data are means ± SEM, with P-values based
   on two-tailed Student’s t-test. c Flow cytometry showing the percentage
   of CD103^– (left) and CD103^+ (right) CD8^+ Trm cells in the brains of
   3xTg-AD (n = 6) and 3xTg-AD/Itgal^–/– (n = 6) mice. Data are
   means ± SEM, with P-values based on two-tailed Student’s t-test. d–f
   Representative immunohistochemical staining of the hippocampus in
   3xTg-AD (n = 3) and 3xTg-AD/Itgal^–/– (n = 3) mice, showing the Aβ load
   (c), and the levels of hyperphosphorylated (d) and total (e) tau
   protein. Scale bar = 20 µm. Data are means ± SEM, with P-values based
   on two-tailed Student’s t-test. g UMAP plot showing immune cell
   populations in human blood (n = 3 AD patients and n = 2 negative
   controls, NCs) by scRNAseq. h Normalized expression of CD3G, CD3D and
   CD3E on a UMAP plot. Transcript levels are color- coded i CD3^+ T cell
   cluster subsets: UMAP plot showing CD8^+ and CD4^+ T cell
   subpopulations in human blood (n = 3 AD patients and n = 2 NCs). j
   Normalized expression of CD8A, CD8B and CD4 on a UMAP plot. Transcript
   levels are color- coded. k CD8^+ T cell cluster subsets: UMAP plot
   showing early active, Tem, and naïve-like CD8^+ T cells in human blood
   (n = 3 AD patients and n = 2 NCs). l Radar plot showing AUCell score
   using known genes. m Heat map reporting the expression values (mean
   calculated using the Hurdle model) of ITGAL in AD patients (n = 3) and
   NCs (n = 2). Transcript levels are color-coded.

Circulating CD8^+ T cells express more LFA-1 in AD patients

   To understand the translational potential of our data, we also
   investigated the role of LFA-1 in the trafficking of circulating CD8^+
   T cells in AD patients. We analyzed a published scRNAseq dataset
   (accession [211]GSE181279) of CD45^+ leukocytes isolated from the blood
   of three AD patients and two negative controls (NCs)^[212]45
   (Fig. [213]6g–m). We used known gene signatures to characterize CD8^+ T
   cells (n = 10,162) in the dataset (Fig. [214]6i, j), identifying: (i)
   naïve; (ii) early activated; and (iii) CD8^+ Tem cells^[215]46
   (Fig. [216]6k, l; Supp. Data [217]2). Interestingly, these results
   showed that ITGAL, encoding the CD11a subunit of LFA-1 integrin, was
   upregulated in the whole population of circulating CD8^+ T cells in AD
   patients 1 and 3, compared to controls (Fig. [218]6m). Particularly,
   ITGAL expression was upregulated in the Tem subset of circulating CD8^+
   T cells in AD patients compared to controls (Fig. [219]6m) but was not
   differentially expressed in the early activated or naïve CD8^+ T cell
   populations (Fig. [220]6m). These results agree with our data showing a
   higher percentage of circulating LFA- 1^+CD44^+KLRG1^+CD8^+ Tem cells
   and the stronger expression of LFA-1 on Tem cells in 3xTg-AD mice
   compared to WT controls (Supp. Fig. [221]4a), whereas no significant
   differences were observed in the KLRG1^–CD44^+ and KLRG1^–CD44^–
   subpopulations of circulating CD8^+ T cells (Supp. Fig. [222]4b, c).
   Collectively, these data confirm that circulating activated CD8^+ T
   cells, particularly the Tem subset, are more activated and have a
   greater LFA-1-dependent migration capacity in AD patients, as well as
   3xTg-AD mice.

GrK-producing CD8^+ T cells accumulate in the CSF and brains of AD patients

   We also analyzed a published scRNAseq dataset of CSF immune cells
   (accession [223]GSE134578)^[224]9 (Fig. [225]7a, b), revealing two
   subsets of CD8^+ T cells (Fig. [226]7c, d): (i) CD69^+ and PRDM1^+ Trm
   cells^[227]24,[228]47, and (ii) KLF2-expressing CD8^+ T cells (Supp.
   Data [229]3). The CD8^+ Trm population could be divided into two
   further subpopulations (Fig. [230]7e; Supp. Data [231]3): (i)
   CD103^–CD8^+ Trm cells characterized by the expression of GZMK, EOMES,
   and S1PR1, and (ii) ITGAE-expressing CD103^+ CD8^+ Trm cells
   characterized by the expression of CD7, ITGA1, KLRB1, and CCR6^[232]32.
   We observed an increase in the abundance of CD103^–CD8^+ Trm cells
   during disease progression, whereas the number of CD103^+CD8^+ Trm
   cells declined, supporting our data from 3xTg-AD mice (Fig. [233]7f).
   Importantly, we also observed a significant increase in GZMK expression
   in CD103^–CD8^+ Trm cells in AD patients compared to healthy controls
   and patients with MCI (Fig. [234]7 g, h), strongly suggesting that
   GrK-producing CD103^–CD8^+ Trm cells also play a detrimental role in
   the development of AD in humans.

Fig. 7. GrK^+CD103^–CD8^+ Trm cells accumulate in the CSF of AD patients.

   [235]Fig. 7
   [236]Open in a new tab

   a UMAP plot showing leukocyte populations detected in human CSF (n = 4
   AD, n = 5 MCI, and n = 9 healthy controls, HCs) by scRNAseq. b
   Normalized expression of CD3E, CD8B and CD4 in the UMAP plot.
   Transcript levels are color-coded. c CD8^+ T cell cluster subsets: UMAP
   plot showing KLF2^+ and Trm cell subpopulations in human CSF (n = 4 AD,
   n = 5 MCI, and n = 9 HCs). d Normalized expression of KLF2, CD69 and
   PRDM1 in a UMAP plot. Transcript levels are color-coded. e Trm cell
   cluster subsets: UMAP plot showing CD103^– and CD103^+ Trm cells in
   human CSF (n = 4 AD, n = 5 MCI, and n = 9 HCs). f Percentage of CD103^–
   and CD103^+ CD8^+ cells in human CSF (n = 4 AD, n = 5 MCI, and n = 9
   HCs). Data are means ± SD. Source data are provided as a Source Data
   file. g Violin plot reporting GZMK expression in Trm cells of n = 4 AD
   patients, n = 5 MCI subjects, and n = 9 HCs. The white dot represents
   median expression. h Normalized expression of GZMK in a UMAP plot of
   all Trm cells (panel 1) and Trm cells from HCs (panel 2), MCI subjects
   (panel 3), and AD patients (panel 4).

   We used immunofluorescence staining to determine whether GrK-producing
   CD103^–CD8^+ Trm cells also accumulate in the human brain during AD
   (Supp. Table [237]1). Importantly, whereas GrK^–CD103^+ CD8^+ T cells
   were predominantly detected in the hippocampus of controls, we observed
   the presence of both intraparenchymal and intravascular
   GrK^+CD103^–CD8^+ T cells near the hippocampal neurons of AD patients
   (Fig. [238]8a). These observations match our 3xTg-AD mice, strongly
   supporting a role for GrK-producing cells in AD. In further agreement
   with our mouse data, we observed a significant increase in the
   abundance of intraparenchymal CD103^–CD8^+ T cells in the hippocampus
   of AD patients but a lower percentage of CD103^+CD8^+ T cells
   (Fig. [239]8b). Although not statistically significant, we observed a
   slight increase in the percentage of both CD103^–CD8^+ and GrK^+CD103^–
   CD8^+ T cells at the intravascular level in the hippocampus of AD
   patients compared to controls (Fig. [240]8c). Importantly, our
   quantifications showed a significant increase in the number of
   intraparenchymal GrK^+CD103^–CD8^+ T cells in the hippocampus of AD
   patients compared to age-matched controls, suggesting a role for GrK in
   the neuronal dysfunction associated with AD in humans (Fig. [241]8b).

Fig. 8. GrK^+CD103^–CD8^+ Trm cells accumulate in the brains of AD patients.

   [242]Fig. 8
   [243]Open in a new tab

   a Immunofluorescence staining of control (CTRL) and AD human brain
   tissues. Scale bar = 40 µm, or 15 µm in the zoomed images. b Bar plots
   showing the percentage of intraparenchymal CD103^–, CD103^+, and
   GrK^+CD103^– CD8^+ T cells in the hippocampus of HCs (n = 3) and AD
   patients (n = 3). Data are means ± SD (two-tailed Student’s t-test).
   Source data are provided as a Source Data file. c Bar plots showing the
   percentage of intravascular CD103^–and GrK^+CD103^– CD8^+ T cells in
   the hippocampus of HCs (n = 3) and AD patients (n = 3). Data are
   means ± SD (two-tailed Student’s t-test). Source data are provided as a
   Source Data file.

GrK-PAR-1 signaling induces functional alterations and tau
hyperphosphorylation in human neuronal cells

   Finally, we assessed whether GrK can induce neuronal alterations via
   PAR-1 in human cells by measuring intracellular Ca^2+ released by
   differentiated SH-SY5Y human neuroblastoma cells treated with purified
   GrK in the presence or absence of the PAR-1 inhibitor [244]SCH79797
   (Fig. [245]9a–c). We observed a significant increase in the release of
   intracellular Ca^2+ by GrK alone, whereas cells cultured with GrK in
   the presence of [246]SCH79797 showed intracellular Ca^2+ levels
   comparable to the negative control (Fig. [247]9b, c). Next, we showed
   that recombinant GrK alone significantly increased the
   hyperphosphorylation of tau on serine residues (pS199 and pS396), but
   not threonine (pT231), in differentiated SH-SY5Y cells, whereas tau
   hyperphosphorylation on serine residues was prevented in the presence
   of [248]SCH79797^[249]48 (Fig. [250]10a–d; Supp. Fig. [251]4f–h;). No
   differences were observed in total tau protein levels, suggesting GrK
   has a specific effect on the signaling machinery leading to tau
   hyperphosphorylation (Supp. Fig. [252]4e, i).

Fig. 9. GrK induces functional alterations in human neuronal cells via PAR-1.

   [253]Fig. 9
   [254]Open in a new tab

   a Immunofluorescence staining for βIII-tubulin, MAP-2, pNF-H, Nestin
   and PAR-1 in undifferentiated and differentiated SH-SY5Y human
   neuroblastoma cells. b Intracellular Ca^2+ release in differentiated
   SH-SY5Y cells cultured in the presence of 150 nM GrK, or 150 nM GrK and
   100 nM [255]SCH79797. The last time point is shown in the bottom panel.
   Ctrl^– = neurons alone. Ctrl^+ = ionomycin-stimulated neurons (10 µM).
   Data are means ± SEM from three independent experiments. P-values based
   on two-way ANOVA-multiple comparisons. c Representative images of
   differentiated SH-SY5Y cells at different time points cultured without
   GrK (Ctrl^–), in the presence of 150 nM active GrK, or in the presence
   of 150 nM active GrK plus 100 nM [256]SCH79797. The red signal shows
   intracellular Ca^2+ release. Scale bar = 20 µm.

Fig. 10. GrK induces tau hyperphosphorylation and alteration of signaling
pathways in human neuronal cells via PAR-1.

   [257]Fig. 10
   [258]Open in a new tab

   a–c Bar plots and representative images showing tau phosphorylation
   levels on serine and threonine (AT8, AT100) (d, e) and only threonine
   (AT180) (f) residues in differentiated SH-SY5Y cells cultured alone
   (Ctrl^–), in the presence of 100 nM [259]SCH79797, in the presence of
   150 nM active GrK alone, and the presence of both. Data are means ± SD
   of three independent experiments (two-way ANOVA-multiple comparisons).
   Scale bar = 5 µm. d ELISA showing the levels of phosphorylation on the
   pS199, pS396, and pT231 residues of tau protein in differentiated
   SH-SY5Y cells cultured alone (Ctrl^–), in the presence of 100 nM
   [260]SCH79797, in the presence of 150 nM active GrK alone, and in the
   presence of both. Data are means ± SD of three independent experiments.
   P-values are based on two-way ANOVA multiple comparisons. e, f Protein
   enrichment analysis from three independent proteomic experiments
   showing the best enriched clusters of terms (h) and terms belonging to
   the “Alzheimer’s/neurodegeneration” and “kinase” pathways (i) in
   SH-SY5Y human neuroblastoma cells cultured in the absence of GrK and
   [261]SCH79797 (Ctrl^–), in the presence of 100 nM [262]SCH79797 alone,
   in the presence of 150 nM active GrK alone, and in the presence of
   both. Source data are provided as a Source Data file.

   To understand downstream signaling pathways induced by GrK, we analyzed
   the proteome of differentiated human neuroblastoma SH-SY5Y cells
   (treated with recombinant GrK or untreated) in the presence or absence
   of the PAR-1 inhibitor [263]SCH79797 (Fig. [264]10e). We found that GrK
   significantly increased the abundance of 84 proteins, including
   proteins involved in the development of AD (Supporting Data [265]4).
   Pathway enrichment analysis applied to upregulated proteins revealed
   that the “Alzheimer’s/Neurodegeneration” and “Kinase pathway” clusters
   were the most enriched (Fig. [266]10e). In the
   “Alzheimer’s/Neurodegeneration” cluster, the most significantly
   enriched pathways included “amyloid fiber formation”,
   “neurodegenerative diseases”, and “deregulated CDK5 triggers multiple
   neurodegeneration” (Fig. [267]10f, Supporting Data [268]4). The “Kinase
   pathway” cluster of terms was characterized by the upregulation of
   pathways related to the activation of MAP kinases^[269]49, which
   mediate tau hyperphosphorylation, and the “Post-translational protein
   phosphorylation pathway” (Fig. [270]10f). We also found that GrK
   induced the upregulation of “NF-κB signaling”, which in neurons is
   associated with the modulation of ion channel protein expression,
   supporting Ca^2+ increases at intraneuronal levels, and with a higher
   Aβ load^[271]50. Notably, the induction of all these pathways was
   prevented when GrK treatment took place in the presence of
   [272]SCH79797, leading to the significant upregulation of only 12
   proteins (Supporting Data [273]4), whereas enrichment analysis detected
   no pathways significantly related to these proteins. Together, these
   data clearly support our results in 3xTg-AD mice, suggesting that
   CD103^–CD8^+ T cells play a negative role in AD and that blocking GrK
   activity and its interaction with PAR-1 may have therapeutic value to
   reduce neuronal dysfunction and cognitive decline in AD.

Discussion

   CD8^+ Trm cells form a defensive line against infections and cancer but
   may also play a role in the development of chronic inflammatory
   conditions^[274]5. Human and mouse brains contain CD103^+CD8^+ and
   CD103^–CD8^+ Trm cells^[275]11,[276]13,[277]16,[278]42,[279]51, but
   their role in AD pathogenesis is unclear. Our scRNAseq analysis
   revealed an altered brain CD8^+ T cell compartment with more abundant
   CD103^– Trm cells and the inhibition of signals promoting CD103
   expression in the 3xTg-AD mouse model, which develops the diagnostic
   neuropathological features of human AD (amyloid and tau pathologies).
   We showed that CD103^–CD8^+ T cells originate from the circulation, use
   LFA-1 integrin to extravasate into the brain and selectively produce
   larger quantities of GrK in 3xTg-AD mice and human AD patients compared
   to controls. GrK induced neuronal dysfunction and tau
   hyperphosphorylation in mouse and human cells via PAR-1, revealing a
   key immune mechanism that promotes neurotoxic inflammation and brain
   damage in AD (Supp. Fig. [280]6).

   Previous studies in transgenic mice with separate Aβ or tau pathology
   showed the existence of CD8^+ T cell subsets with protective or
   deleterious effects depending on the disease stage and pathological
   hallmarks^[281]7,[282]10–[283]13,[284]17. Particularly, CD8^+ T cells
   in APP/PS1 mice with advanced β-amyloid pathology show a transcriptomic
   profile similar to old WT animals and higher levels of CD103 were
   detected by flow cytometry^[285]11. Brain CD8^+ T cells of 5xFAD mice
   with aggressive late-stage Aβ pathology had higher CD103 abundance
   together with higher expression of CXCR6 and PD-1 (compared to WT
   mice), inhibiting Aβ accumulation by suppressing the activation of
   microglia^[286]13. Previous studies using models of tau pathology
   suggested brain-infiltrating CD8^+ T cells promote disease development,
   although additional signals appear to be essential for CD8^+ T cell
   accumulation in the CNS of such mice^[287]12,[288]17. Particularly, old
   P301S mice lacking APOE do not accumulate more brain CD8^+ T cells
   despite tau pathology, suggesting that CD8^+ T cell infiltration into
   the CNS requires APOE4^[289]12. In this context, our data show that the
   co-occurrence of Aβ and tau pathologies significantly reduces the
   abundance of intraparenchymal CD103^+CD8^+ Trm cells in the brains of
   3xTg-AD mice compared to WT animals due to the downregulation of genes
   encoding transcription factors and cytokines involved in the
   differentiation and maintenance of CD103^+ Trm cells^[290]31.
   Particularly, our results indicate strong suppression of the
   homeostatic and anti-inflammatory cytokine IL-33, which is required for
   the generation and survival of brain CD103^+CD8^+ Trm cells^[291]52,
   and are in line with previous studies showing the loss of IL-33 in AD
   patients^[292]53.

   Our results also show that CD103^–CD8^+ Trm cells are more abundant in
   the brains of 3xTg-AD mice than WT controls. The accumulation of these
   cells (and the loss of CD103^+ cells) was also observed in the brains
   of MS patients^[293]54 and in primary Sjögren’s syndrome^[294]55
   suggesting common dysregulation of the CD8^+ T cell compartment in
   these diseases. Similarly, CD103^–CD8^+ Trm cells accumulated in the
   guts of patients with inflammatory bowel disease (IBD), and the
   rebalancing of CD103^–CD8^+ and CD103^+CD8^+ Trm cells was associated
   with disease remission^[295]56, suggesting CD103^– cells promote
   disease pathogenesis. Indeed, our data demonstrate that CD103^–CD8^+ T
   cells are detrimental in 3xTg-AD mice and their depletion in the brain
   ameliorates neuropathology and cognitive deficits. Importantly, our
   data from AD patients show an increase in the abundance of
   intraparenchymal CD103^–CD8^+ T cells in the hippocampus while the
   percentage of CD103^+CD8^+ T cells declines, suggesting CD103^– cells
   may also promote disease development in human AD patients.

   We also found that brain CD103^–CD8^+ T cells originate from the
   circulation in 3xTg-AD mice. Previous studies have shown that
   intratissutal cytotoxic CD103^–CD8^+ Trm cells originate from
   circulating CXCR3^+ “exKLRG1” effector cells^[296]26 and our
   single-cell data showed that Cxcr3 was one of the best marker genes for
   the CD103^–CD8^+ Trm cell population, providing additional evidence
   that these cells migrate from the blood into the brain. CXCR3 was
   recently shown to regulate CD8^+ T cell infiltration and neuronal
   damage in a human 3D model of neuro-immune interactions, and CXCL10
   (the chemokine ligand of CXCR3) accumulates in AD, suggesting that the
   trafficking of detrimental CD8^+ T cells in the AD brain is mediated by
   CXCR3–CXCL10 interaction^[297]57. Our data also show the massive
   depletion of CD103^–CD8^+ T cells in the brains of 3xTg-AD/Itgal^–/–
   mice lacking LFA-1, suggesting a role for this integrin in the
   migration of detrimental CD8^+ T cells into the brain during AD.
   Interestingly, an LFA-1 blockade in 3xTg-AD mice reduces neutrophil
   migration into the CNS and mitigates disease symptoms, suggesting a key
   role for this integrin in the migration of circulating leukocytes into
   the AD brain^[298]22. Our results also showed that the percentage of
   circulating LFA-1^+CD44^+KLRG1^+CD8^+ Tem cells increased (along with
   higher LFA-1 levels) in 3xTg-AD mice compared to WT controls,
   suggesting that circulating CD8^+ T cells are stickier and more prone
   to migrate into the AD brain. This is also supported by the higher
   expression of ICAM- 1, the vascular ligand of LFA-1, on the brain
   endothelial cells of AD mouse models and human AD patients, indicating
   that vascular inflammation may promote the recruitment of circulating
   activated CD8^+ T cells into the brain. Finally, our analysis of a
   published dataset showed that ITGAL (encoding the CD11a subunit of
   LFA-1) was upregulated in the whole population of circulating CD8^+ T
   cells in AD, suggesting that the LFA-1–ICAM-1 and CXCR3–CXCL10 axes may
   act in concert to promote the invasion of the brain by peripheral CD8^+
   T cells in AD.

   We found that most CD103^+CD8^+ and CD103^–CD8^+ T cells expressed
   CD69, a marker of both “residency” and T cell activation, suggesting
   CNS Trm cells may have effector activities^[299]5. However,
   CD103^+CD8^+ and CD103^–CD8^+ T cells have distinct transcriptional
   phenotypes and functions in the normal intestine or during pathological
   conditions, including a model of brain viral
   infection^[300]11,[301]26,[302]32,[303]51,[304]56. The analysis of our
   datasets from 3xTg-AD and WT mice are consistent with these data and
   unbiased cell fate trajectory analysis clearly showed that brain CD8^+
   T cell differentiation occurred on a tightly organized trajectory
   starting from a common root and ending with two differentiation states
   distinguished by the expression of Itgae and Eomes. However, only
   CD103^– (not CD103^+) CD8^+ Trm cells infiltrating the brains of
   3xTg-AD mice were strongly associated with the “Alzheimer’s disease”
   pathway, suggesting these cells have distinct pathological functions
   specifically promoting AD. Importantly, CD103^–CD8^+ Trm cells
   accumulating in the brains of AD patients and 3xTg-AD mice were
   characterized by no significant increase in Gzmb and Prf1 expression
   combined with the strong expression of Gzmk and high levels of GrK
   protein (but low levels of GrA and GrB proteins, a distinct phenotype
   for CD8^+ Trm cells in AD. Indeed, recent studies performed in other
   contexts have shown that GrK is an atypical cytotoxic molecule that can
   be produced and released independently of GrA and GrB and can act in
   the absence of perforin^[305]58,[306]59, thus supporting the phenotype
   of the CD103^–GrK^+CD8^+ T cells we found in AD. GrK^+ T cells may
   respond to cytokine stimuli alone, given that TCR stimulation can
   downregulate Gzmk expression, suggesting that GrK^+ T cells, despite
   their activated phenotype, are not antigen-stimulated T cells in
   AD^[307]59.

   Higher numbers of GrK^+CD8^+ T cells have been reported in several
   pathologies, including immune aging, cancer and autoimmune
   diseases^[308]55,[309]58–[310]61, suggesting AD pathogenesis may share
   common immune mechanisms with other inflammatory conditions. In a
   heterogeneous group of subjects, GrK was more abundant in brain
   CD103^–CD69^+CD8^+ T cells than CD103^+CD69^+CD8^+ cells populations,
   but most brain donors had no neurological disease or were patients with
   MS or Parkinson’s diseases and no data could be correlated with
   AD^[311]16. A recent study found an increased percentage of
   CD69^+CD103^+ cells in the CSF of a small group of AD patients,
   relative to the total population of memory CD45RA^–CD8^+ T
   cells^[312]62. However, these authors did not apply their CSF analysis
   to Trm cells (the majority of T cells in the brain parenchyma) and did
   not consider that CSF cells contain a significant population of KLF2^+
   cells with a circulation origin, as shown by our data^[313]62. CD8^+ T
   cells have also been found in the leptomeninges of AD patients^[314]9,
   and despite recent studies showing the abundance of meningeal immune
   cells and the role of CNS borders during neurodegenerative and
   neuroinflammatory diseases^[315]15, our results reveal that the number
   of CD103^–CD8^+ Trm cells does not change significantly and GrK
   secretion does not increase in the meninges, suggesting the brain is
   the main site of CD8^+ Trm cell dysregulation.

   How CD8^+ T cells communicate with neuronal cells and potentially
   induce deleterious effects in AD is unclear. The accumulation of CD8^+
   T cells in the CNS of aged mice was recently found to promote axonal
   degeneration and age-related cognitive and motor decline through the
   release of GrB^[316]14, which we did not find in our studies of 3xTg-AD
   mice. In addition, age-related alteration of brain functions has also
   been associated with clonally expanded CD8^+ T cells producing INF-γ,
   which infiltrate old neurogenic niches, inhibiting the proliferation of
   neural stem cells^[317]63. The ablation of circulating CD8^+ T cells in
   APP/PS1 mice does not reduce the Aβ load or cognitive decline but
   increases the expression of neuronal genes such as Arc and Npas4,
   suggesting that brain CD8^+ T cells may modulate synaptic
   plasticity^[318]10. Nevertheless, CD8^+ T cells were found close to
   neurons in AD patients, but their ability to exert direct neurotoxic
   functions has not been shown^[319]9. Our data fill this knowledge gap
   and demonstrate that CD103^–CD8^+ Trm cells induce neuronal dysfunction
   by releasing GrK. Indeed, our high-resolution live imaging studies
   showed that neurons from 3xTg-AD mice cultured in the presence of
   GrK^+CD103^–CD8^+ Trm cells undergo profound functional changes.
   Notably, purified active GrK also induced neuronal dysfunction,
   confirming the neurotoxic role of this molecule. Our proteomic data on
   human neuronal cells further supported these observations, showing that
   GrK-PAR-1 interactions activate several intracellular pathways involved
   in neurodegeneration and AD. Moreover, we show that GrK^+CD8^+ T cells
   accumulated in the parenchyma near hippocampal neurons in AD patients
   and 3xTg-AD mice, whereas GrK^–CD8^+ T cells were preferentially
   located in the healthy hippocampus, further suggesting a neurotoxic
   role for GrK. Our data thus suggest that GrK is the basis of a key
   immune mechanism that induces brain neurotoxicity with a selective role
   in the mediation of CD103^–CD8^+ Trm cell-dependent neuronal changes in
   AD.

   We found that GrK released by CD103^–CD8^+ Trm cells activates PAR-1
   (the classical thrombin receptor) on primary AD neurons, leading to
   functional alterations and potentially affecting neuronal
   networks^[320]64. GrK may also activate PAR-1-expressing glial cells,
   such as astrocytes, which release glutamate following PAR-1 engagement,
   further contributing to neuronal alterations^[321]65. Previous studies
   have shown that GrK cleavage and PAR-1 activation also induce the
   phosphorylation of ERK1/2 and MAPK, triggering the secretion of
   pro-inflammatory cytokines such as IL-1β, MCP-1, IL-6, and IL-8, as
   well as upregulating the expression of adhesion molecules on monocytes,
   fibroblasts and cells of epithelial origin^[322]35. Considering these
   pro-inflammatory activities of GrK, and our data showing that the
   ablation of brain CD103^–CD8^+ T cells (the majority expressing GrK)
   mitigates AD in mouse models, we propose that GrK–PAR-1 interactions
   promote chronic neuroinflammation and neurodegeneration in AD.

   PAR-1 has a negative role in several neuropathological conditions and
   its inhibition reduces brain damage and inflammation in animal models
   of brain ischemia^[323]66. Although not related to CD8^+ T cells, PAR-1
   inhibition also ameliorates cognitive performance in rats injected
   intracerebrally with Aβ peptide 1–42^[324]40. However, PAR-1 engagement
   can also promote learning and synaptic plasticity under physiological
   conditions^[325]39. Notably, our data show an increase in neuronal
   PAR-1 expression in 3xTg-AD mice, suggesting that chronic alterations
   in PAR-1 signaling promote immune-driven neuropathology in AD.

   Importantly, our data demonstrate that neuronal PAR-1 activation by GrK
   in mice and humans induces tau hyperphosphorylation in vitro^[326]48
   and these results are in line with previous studies showing that the
   number of CD8^+ T cells in the brains of AD patients correlates with
   tau pathology and Braak staging^[327]10,[328]17. Also, the percentage
   of T cells in the brains of mice with tau pathology was higher than
   that detected in animals with amyloidosis^[329]12. Systemic CD8^+ T
   cell ablation has no effect on behavioral changes or neuropathology in
   APP-PS1 mice^[330]10, further highlighting the pathological link
   between tau and CD8^+ T cells. Along with our work, this evidence may
   explain why an increase in the abundance of CD103^–CD8^+ Trm cells was
   not observed in mouse models of amyloidosis^[331]11,[332]13. However,
   we found that the ablation of brain CD103^–CD8^+ T cells (86% of which
   produce GrK) in 3xTg-AD mice reduced both amyloid and tau pathologies.
   The effect of CD8^+ T cell inhibition on amyloidosis is supported by
   the fact that Aβ and tau pathologies have a synergistic effect in AD
   and tau pathology itself may favor Aβ deposition^[333]19. Together,
   these data indicate that CD8^+ T cells can directly sustain tau
   pathology during AD, mechanistically linking the invasion of the brain
   by CD8^+ T cells and neuroinflammation with the development of AD
   neuropathology.

   In conclusion, we show that dysregulated CD103^–CD8^+ T cells and
   GrK–PAR-1 signaling contribute to changes in communication between the
   immune system and CNS, leading to neuroinflammation and neuronal
   dysfunction in AD. GrK inhibitors have already been developed to treat
   inflammatory diseases, including myocardial infarction^[334]35, whereas
   drugs targeting PAR-1 are already used to block platelet
   aggregation^[335]67, suggesting that the manipulation of GrK–PAR-1
   interactions may also benefit AD patients.

Methods

   All chemicals, kits, and antibodies are listed in Supp. Table [336]2.

Ethics statement

   Formalin-fixed paraffin-embedded (FFPE) hippocampal sections of
   controls and Alzheimer’s disease (AD) cases were obtained from the
   Medical Research Council (MRC) London Neurodegenerative Disease Brain
   Bank. All material was collected from donors (females aged 51-69 at
   time of death as listed in Supp. Table [337]1) from whom written
   informed consent for brain autopsy and the use of the material and
   clinical information for research purposes had been obtained by MRC
   London Neurodegenerative Disease Brain Bank, approved by the NRES
   Committee London – City & East (REC references: 08/H0704/128 + 5). Our