Depletion of NK Cells Improves Cognitive Function in the Alzheimer Disease Mouse Model

Alzheimer disease (AD) is a devastating disease with unmet therapeutic needs. A hallmark of AD is chronic neuroinflammation mediated by dysregulated microglia (1–4). Microglia, the predominant brain-resident immune cells, appear to play dual-wedged roles in the pathogenesis of AD (3, 5–7). Although microglial phagocytosis is important in restraining the accumulation of amyloid β (Aβ) and other toxic substances, deregulated microglia activation may lead to chronic neuroinflammation, synaptic loss, and impaired neurogenesis (3, 5–7). Chronic microglial activation in AD is demonstrated by multiple dynamic changes including altered morphology, abnormal proliferation, and increased secretion of neurotoxic proinflammatory mediators (8, 9). The cellular and molecular pathways that regulate microglia activity and brain inflammation in AD, however, remain largely unclear.

Increasing evidence indicates that no-glia immune cells might also play important roles in AD pathogenesis. The discovery that Aβ possesses antimicrobial activities has led to interesting infectious-disease hypotheses of AD (10–13). Peripheral innate immune cells, such as neutrophils, have been found to play important roles in promoting cognitive decline in mouse models of AD (14). Recent evidence also suggests implication of lymphocytes in AD progression. Genetic deletion of lymphocytes in the Alzheimer Disease mouse model resulted in accelerated amyloid pathology and exacerbated neuroinflammation (15). In addition, amplification of regulatory T cells delayed progression of AD-like pathology in APPPS1 transgenic mice (16–18). Our recent work also indicates that activation of group-2 innate lymphoid cells can alleviate aging-associated cognitive decline (19). The precise roles of other adaptive and innate lymphocytes in AD development and exacerbation, however, remain poorly understood.

NK cells are innate cytotoxic lymphocytes that can be rapidly induced to secrete cytotoxic molecules and inflammatory cytokines such as granzymes, cathepsins, perforins, IFN-γ, and TNF-α (20). NK cells are known for their capability to kill infected and malignant cells and to mediate Ab-dependent cellular cytotoxicity. NK cells might also be involved in inflammatory disorders. Certain subsets of NK cells may produce multiple inflammatory cytokines and chemokines (21). The cytotoxic molecules secreted by NK cells, such as granzymes and cathepsins, may also possess proinflammatory functions when released into the extracellular microenvironment (22–28). However, the precise roles for NK cells in tissue homeostasis and inflammation remain largely unknown.

In this study, we used a triple-transgenic AD mouse model (3xTg-AD) and anti-NK1.1–depleting Abs to interrogate the role of NK cells in AD-associated cognitive decline. We found that NK cells exhibited an enhanced proinflammatory profile in 3xTg-AD. Depletion of NK cells by anti-NK1.1 treatment significantly improved cognitive function of 3xTg-AD mice. Depletion of NK cells did not affect amyloid pathology but repressed neuroinflammation and stimulated neurogenesis. Notably, microglia in NK cell–depleted 3xTg-AD mice exhibited a homeostatic-like morphology, reduced proliferative responses, and decreased expression of multiple proinflammatory cytokines. Together, our results reveal a striking role for NK cells in promoting neuroinflammation and AD-associated cognitive decline and indicate that targeting NK cells might unlock novel strategies to combat neurodegenerative diseases.

Both 3xTg-AD and control B6129SF2/J were obtained from Mutant Mouse Resource and Research Center at The Jackson Laboratory or The Jackson Laboratory and bred in the animal facility of Albany Medical College. Seven- to eight-month-old female mice were used in this study. For anti-NK1.1 treatment, mice were treated with 25 μg anti-NK1.1 (clone PK136; Bio X Cell) Abs or isotype control (clone C1.18.4; Bio X Cell) every 4 d for 4 wk. For anti-CD1d treatment, mice were treated with 500 μg anti-CD1d (clone 19G11; Bio X Cell) Abs or isotype control every other day for 4 wk. Water Maze tests were performed on the day after the last treatment. Specifically, for anti-Nk1.1 treatment, mice were treated with anti-NK1.1 Abs or isotype controls on days 1, 5, 9, 13, 17, 21, 25, and 29, and Water Maze test was performed on day 30. For anti-CD1d treatment, mice were treated with anti-CD1d Abs or isotype control on days 1, 3, 7, 9, 11, 13, 15, 17, 19, 21, 23, 25, 27, and 29, and Water Maze test was performed on day 30. All animal experiments were performed according to protocols approved by the Institutional Animal Care and Use Committee at Albany Medical College.

For isolation of hematopoietic cells in the whole-brain tissue, mice were perfused with 50 ml of PBS. The whole-brain tissue, including brain parenchyma, leptomeninges, choroid plexus, and perivascular space tissue, was harvested. The tissue was minced with scissors and digested with 0.2 mg/ml of Liberase (Roche) and 0.1 mg/ml DNase I (Roche) for 30 min at 37°C. Cells were filtered through a 70-μM strainer, followed by gradient centrifugation with 40% Percoll (GE Healthcare).

Abs were purchased from BioLegend or Thermo Fisher Scientific. Abs for flow cytometry analysis included anti-B220 (RA3-6B2), anti-NK1.1 (PK136), anti-CD11b (M1/70), anti-CD3 (2C11), anti-CD45.2 (104), and anti-Ki67 (16A8). Mouse CD1d tetramers PBS57 were obtained from the National Institutes of Health tetramer core (29). Intracellular staining of Ki67 was performed using the Foxp3 Fix/Perm Kit (Thermo Fisher Scientific) according to the manufacturer’s instructions. 5-Ethynyl-2′-deoxyuridine (EdU) was detected using the Click-iT Plus EdU Flow Cytometry Assay Kit (Thermo Fisher Scientific) following the manufacturer’s instructions. Flow cytometric analysis was performed using an FACSCanto analyzer (BD Biosciences). Cell sorting was performed using an FACSAria II sorter (BD Biosciences).

The concentrations of soluble and insoluble β-amyloid were measured using the LEGEND MAX β-Amyloid x-42 ELISA Kit (BioLegend) according to the manufacturer’s instructions. Specifically, brain samples were homogenized using a glass homogenizer with six passes on ice in TBS containing protease inhibitors. Samples were centrifuged at 350,000 × g for 20 min. The supernatant containing soluble β-amyloid was collected. The pellet was resuspended in TBS containing 1% Triton X-100 and centrifuged again for collection of insoluble β-amyloid. ELISA assays were performed immediately.

For measurement of in vivo β-amyloid uptake, mice were injected i.p. with methoxy-X04 (4920; Tocris Bioscience) at the dosage of 10 mg/kg. Mice were euthanized at 3 h after methoxy-X04 injection, and flow cytometry analysis was performed to examine microglia uptake of methoxy-X04.

For quantitative PCR (QPCR), RNeasy Plus Mini Kit (QIAGEN) was used to extract mRNA from samples. cDNA was synthesized using the Superscript II kit (Thermo Fisher Scientific). QPCR was performed with TaqMan probes (Thermo Fisher Scientific).

RNA sequencing (RNA-seq) was performed at the Center for Functional Genomics at the University at Albany. For bulk RNA-seq, cells were sorted by FACS. cDNA was generated with the SMART-Seq v4 Ultra Low Input RNA kit (Takara) according to the manufacturer’s instructions. libraries were prepared using Nextera XT DNA Library Prep Kit (Illumina). Libraries were sequenced with a using NextSeq 500 (Illumina). Single-end, 75-bp high-throughput sequencing was performed. Data were aligned and normalized using STAR aligner. Differentially expressed genes were identified by DEseq2. Gene pathway analysis was performed with the Database for Annotation, Visualization and Integrated Discovery (30, 31).

For single-cell RNA-seq (scRNA-seq), cells were sorted by flow FACS. Libraries were generated by 5′ gene expression kit (10X Genomics) using the Chromium Single-Cell Controller (10X Genomics). Double-end, 75-bp high-throughput RNA-seq was performed using NextSeq 500 (Illumina). Initial data analysis was performed by Cellranger 3.1.1. Data were normalized and scaled, and mitochondria regression was performed using Seurat 3.1.1. Uniform Manifold Approximation and Projection (UMAP) was used for cell clustering. A Wilcoxon rank-sum test was used to determine significance of differentially expressed genes.

scRNA-seq data for NK cells and microglia and bulk RNA-seq data for microglia were deposited in the Gene Expression Omnibus under the accession numbers GSE142853 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE142853), GSE142858 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE142858), and GSE142875 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE142875), respectively.

Morris Water Maze was performed at 4 wk after anti-NK1.1 treatment as we previously described (32). Specifically, mice for Water Maze test were housed with five mice per Allentown cage. Mice were habituated to the testing facility for 1 h before the behavior tests each day. A circular pool of opaque water with a diameter of 125 cm was used. Water temperature was kept at 21–22°C. The maze was conceptually divided into four quadrants, with visual cues on each side of the pool. On day 1 (visual trial), mice were trained to escape the maze by swimming to a clear plastic platform in the target quadrant. The platform was submerged by 1 cm and made visible by black tape around the platform and a 10-cm black cylinder on top of it. Mice were trained for five trials of up to 3 min each until the mouse found that platform and stayed on it for 10 s. If a mouse failed to escape the maze within 3 min, it was guided slowly to the platform by dragging its tail and stayed on the platform for 10 s. On day 2 (hidden trial), mice were trained to escape the maze with the same platform, but the visual cues on the platform were removed. Mice were again trained for five trials for up to 3 min each. On day 3 (probe trial), the platform was removed, and the mice were assessed by one test of 3 min. The behavior tests were recorded and analyzed by using the ANY-maze software (Stoelting).

For immunofluorescence, mice were perfused with 50 ml of PBS, followed by 50 ml of 4% paraformaldehyde. The brains were harvested and fixed for 24 h in 4% paraformaldehyde. The samples were transferred to 30% sucrose in PBS and frozen at OCT in −80°C until sectioning. At 40 μM, sections were prepared using a Leica CM1950 cryostat. For staining of microglia, tissues were stained with goat anti–Iba-1 (Thermo Fisher Scientific) and Cy5-conjugated anti-goat secondary Ab (Jackson ImmunoResearch Laboratories). Slides were imaged using a Zeiss Axio Observer fluorescence microscope with a 20× objective. The Zeiss Zen Blue software (ZEISS) was used to process the images.

For in vivo EdU labeling, mice were injected i.p. with six dosages of EdU (100 mg/kg) over 2 d. EdU was detected by the Click-iT Plus EdU Imaging Kit (Invitrogen) according to the manufacturer’s instructions. After EdU detection, the sections were stained with anti-NeuN rabbit polyclonal Ab (A60) (MilliporeSigma) and anti-rabbit Rhodamine secondary Ab (The Jackson Laboratory). Slides were imaged using a Zeiss Axio Observer fluorescence microscope with a 10× objective. Images were processed by the Zeiss Zen Blue software (ZEISS).

Mann–Whitney U test was used to compare the differences in Water Maze tests. Wilcoxon rank-sum test was used to compare gene expression difference between two groups for scRNA-seq data. Student t tests were used to calculate statistical significance for all other data. A p value < 0.05 was considered significant.

Previous studies indicate that NK cell activity is altered in AD patients and mouse models of AD (33–41). However, the precise transcriptomal changes of NK cells in the inflamed microenvironment of AD remains unclear. In this study, we examined NK cells of middle-aged (7–8 mo) triple-transgenic 3xTg-AD and control wild-type mice. At this age, 3xTg-AD mice exhibit declined cognitive function, immunoreactivity to amyloid β, and mild plaques (42–44). NK cells were identified as CD45⁺CD3⁻B220⁻NK1.1⁺DX5⁺ lymphocytes (Fig. 1A). The number of NK cells was moderately decreased in the whole-brain tissue of 3xTg-AD mice compared with wild-type control mice (Fig. 1B). Of note, in this study, we examined NK cells in the whole-brain tissue, including brain parenchyma, choroid plexus, leptomeninges, and perivascular space tissue. We noted that NK cells were enriched in the barrier tissues (choroid plexus, leptomeninges, and perivascular space tissues), whereas brain parenchyma was devoid of NK cells (Supplemental Fig. 1A). We thus refer to these NK cells as “brain-associated NK cells.” To obtain sufficient numbers of brain-associated NK cells for in-depth transcriptomal analysis, we isolate NK cells from the whole-brain tissue for scRNA-seq. We sorted NK cells from the whole-brain tissue by FACS and performed scRNA-seq. Recent work indicates that spleen and blood NK cells can be broadly divided into two subsets: Thy1⁻CD7^−/low NK₁ cells that express high amounts of cytotoxic molecules and Thy1⁺CD7^hi NK₂ cells that express Xcl1 and other effector molecules (45). The heterogeneity and single-cell transcriptomes of brain-associated NK cells remain unknown. Using UMAP analysis (L. McInnes, J. Healy, and J. Melville, manuscript posted on arXiv), our scRNA-seq results indicated that the majority of brain-associated NK cells were Thy1⁻CD7^−/low NK₁ cells that express relatively high amounts of cytotoxic molecules (Fig. 1C, 1D). Expression of Ifng was minimal among all brain-associated NK cell subsets (Fig. 1D), and expression of Tnf and Il17a was not detectable (data not shown). Interestingly, UMAP also identified a subset of NK cells with a very distinct transcriptomal profile in 3xTg-AD mice (NK_1AD) (Fig. 1C). NK_1AD cells were marked by high expression of cytotoxic molecules Cstb and Ctsc (Fig. 1D). This subset was hardly distinguishable in control wild-type mice (Fig. 1C). Gene enrichment pathway analysis demonstrated that cytotoxic molecules and lymphocyte receptor signaling molecules were overrepresented in genes highly expressed in NK_1AD cells (Fig. 1E). In particular, among all NK cell subsets, NK_1AD cells expressed the highest amounts of the cytotoxic molecules Ctsc and Ctsd, proinflammatory chemokines Ccl3 and Ccl4, and the adhesion molecule Icam1. The lymphocyte receptor signaling molecules highly expressed on NK_1AD were predominantly NK cell activation molecules, including Nfatc1, Tbx21, Nfkbia, Il12rb, and Klra9, suggesting that they are a hyperactivate proinflammatory subset (Fig. 1E–G). Other than NK_1AD cells, the other NK₁ cells in the brains of wild-type and 3xTg-AD mice can be further divided into two subsets, named NK_1a and NK_1b in this study (Fig. 1C). NK_1a cells expressed lower amounts of cytotoxic molecules and activating genes than NK_1b and NK_1AD subset, suggesting that they might be at a relatively inactive state (Fig. 1D, 1G). Interestingly, compared with than those in control wild-type mice, NK_1a and/or NK_1b subsets in 3xTg-AD mice expressed higher amounts of cytotoxic and other effector molecules such Ctsc, Ctsd, Ccl3, and Ccl4, indicating enhanced activity (Fig. 1H). NK cells in 3xTg-AD mice also expressed higher amounts of the adhesion molecule ICAM1, which was verified by flow cytometric analysis (Fig. 1H, Supplemental Fig. 1B, 1C). We were not able to obtain reliable Abs to examine Ctsc and Ctsd protein levels by flow cytometric analysis; thus, future efforts to verify changes in Ctsc and Ctsd protein expression would be worthwhile. Together, the brains in 3xTg-AD mice harbor a unique subset of NK_1AD subset with a hyperactive proinflammatory profile, and their other NK subsets also exhibit an enhanced proinflammatory profile than those in control wild-type mice.

Notably, all three brain-associated NK₁ subsets, including NK_1AD, express high amounts of the trafficking molecule S1pr5 (Supplemental Fig. 1D). The expression of genes characteristic of tissue-resident lymphocytes, such as Itgae and CD69, was barely detectable in brain-associated NK cells (Supplemental Fig. 1D). Thus, NK_1AD might be circulating NK cells that were recruited to brain barrier tissues.

We next sought to understand whether the presence of NK cells might influence the cognitive function of 3xTg-AD mice. The 3xTg-AD mice exhibit progressive loss of cognitive function as early as 3–6 mo, and impairment in cognitive function is very obvious in 7–8-mo-olds (44). We depleted NK cells in middle-aged 3xTg-AD mice using anti-NK1.1–neutralizing Ab. Mice were treated with Abs for 1 mo to allow sufficient time for potential effects of immune cells to take place. Flow cytometry analysis verified efficient depletion of NK cells (Fig. 2A). We used a Morris Water Maze test (46) to examine the cognitive function of 3xTg-AD mice that were treated with anti-NK1.1 Ab or isotype control (Fig. 2B). Mice were trained with a visible platform on day 1, followed by training with an invisible platform at the same place on day 2. Mice were then allowed to freely swim for 3 min in the probe trial (no platform) on day 3. The activity and behavior of mice in the probe trial were recorded. We compared the mobility and cognitive function indexes between mice treated with isotype control and those treated with anti-Nk1.1 Abs. NK cell depletion did not affect the general mobility of 3xTg-AD mice, demonstrated by comparable mean swimming speed between mice treated with anti-NK1.1 Abs and those treated with isotype controls (Fig. 2C). Notably, the percentage of time spent in the target quadrant by isotype-treated 3xTg-AD mice was comparable to the predicted chance level (25%) in the probe trial, indicating a loss of hippocampal-dependent spatial recognition (Fig. 2D). Mice treated with anti-NK1.1 Abs, however, spent significantly higher percentage of time in the target quadrant, suggesting improved spatial memory (Fig. 2D). Compared with control mice, mice treated with anti-NK1.1 Abs also exhibited increased numbers of entries into target quadrant, reduced latency, and increased path efficiency, verifying improved cognitive function (Fig. 2E–H). Together, multiplex indicators suggest that NK cell depletion by anti-NK1.1 treatment alleviated the cognitive decline of 3xTg-AD mice.

Of note, anti-NK1.1 depletion Abs may also deplete NKT cells. We thus examined NKT cells in 3xTg-AD mice using CD1d tetramers. Notably, NKT cells were barely detectable in the whole-brain tissue of 3xTg-AD mice (Supplemental Fig. 2A). In addition, treatment with CD1d-neutralizing Abs did not significantly affect the cognitive function of 3xTg-AD mice (Supplemental Fig. 2B). Thus, NK cells, but not NKT cells, might play a more predominant role in promoting cognitive decline in 3xTg-AD mice.

Declined neurogenesis is another notable feature of AD, which may contribute to declined cognitive function in AD (47). We thus examined neurogenesis in 3xTg-AD mice after 1 mo of treatment with anti-NK1.1–depleting Ab or isotype control. We injected mice with six doses of EdU in 2 d to label proliferating cells and then colabeled cells with NeuN to identify new neurons that had been born and survived to maturity. Compared with control mice, NK cell–depleted mice exhibited decreased number of EdU⁺ NeuN⁺ cells in the subventricular zone (SVZ) and in the hippocampal dentate gyrus region, indicating enhanced neurogenesis (Fig. 2I–K). Thus, depletion of NK cells stimulates neurogenesis in 3xTg-AD mice.

We next examined the effects of NK cell on β-amyloid pathologies. Results with ELISA assays indicated that the concentrations of soluble and insoluble β-amyloid in the brain of 3xTg-AD mice were not significantly altered by NK cell depletion (Fig. 3A, 3B). To further examine whether anti-Nk1.1 depletion might influence microglia uptake of β-amyloid, we injected mice with methoxy-X04, a fluorescence probe for β-amyloid that can cross blood-brain barrier. Percentages of methoxy-X04⁺ microglia were comparable in 3xTg-AD mice treated with anti-NK1.1–depleting Abs and those treated with isotype controls, indicating that NK cell depletion did not affect β-amyloid uptake by microglia (Fig. 3C, 3D). Together, NK cell depletion might improve cognitive function through mechanisms other than regulation of β-amyloid pathologies.

A hallmark of AD is neuroinflammation mediated by dysregulated microglia (4–9). We thus sought to determine whether NK cell depletion might affect neuroinflammation in 3xTg-AD mice. We first used flow cytometry analysis to examine microglia in 3xTg-AD mice treated with anti-NK1.1 Ab or isotype controls. The number of microglia was moderately reduced after NK cell depletion (Fig. 4A, 4B). Interestingly, microglia in NK cell–depleted 3xTg-AD mice demonstrated remarkedly reduced forward scatter values than microglia in control 3xTg-AD mice, suggesting that NK cell depletion might alter the morphology of NK cells (Fig. 4C, 4D). Indeed, immunofluorescence assays indicated that microglia in NK cell–depleted 3xTg-AD mice exhibited a homeostatic-like morphology in contrast to the more amoeboid-like morphology in control 3xTg-AD mice (Fig. 4E). We next performed bulk RNA-seq to examine the transcriptomal changes in microglia. Notably, RNA-seq suggested that anti-NK1.1 treatment reduced the expression of many proliferative genes such as Mik67 and Cdc42 as well as proinflammatory genes such as Tnf and Il1 (Fig. 4F–H). Consistent with decreased expression of proliferative genes, microglial proliferative responses in 3xTg-AD mice were significantly reduced by anti-NK1.1 treatment, demonstrated by increased proportion of microglia in the quiescent G0 phase and reduced percentage of cells in the activating cycling S/G2 phases (Fig. 4I, 4J). QPCR verified that depletion of NK cells led to greatly reduced expression of Mki67 as well as proinflammatory cytokines Tnf, Il1a, Il1b, and Il18 by microglia (Fig. 4K). Together, these results indicate that depletion of NK cells lead to reduced microglial inflammation in 3xTg-AD mice.

We next performed scRNA-seq analysis to better understand the effects of NK cell depletion on the transcriptomal changes of microglia in 3xTg-AD mice. Interestingly, UMAP analysis indicated that the expression of proinflammatory cytokines such as Il-18 were broadly spread among all microglia clusters in 3xTg-AD mice, indicating that the capability to produce proinflammatory cytokines were not restricted to a specific microglia subset in 3xTg-AD mice (Fig. 4L). Depletion of NK cells by anti-NK1.1 treatment led to reduced frequency of microglia expressing proinflammatory cytokines Il18 and Il1b (Fig. 4L–N). More notably, the average expression of many proinflammatory cytokines, including Il-18, Tnf, and Il1b, by cytokine-expressing microglia was markedly reduced in mice treated with anti-NK1.1 Abs (Fig. 4L–N). Thus, depletion of NK cells may both reduce the frequency of microglia that express proinflammatory cytokines and decrease the cytokine production by microglia at per cell level.

Of note, anti-NK1.1 treatment did not affect the expression of Trem2 on microglia, indicating that depletion of NK cells might not affect the general fitness of microglia (Supplemental Fig. 3A). Other innate immune cells, such as neutrophils, may also promote microglial inflammation in mouse models of AD (14). However, we did not detect significant changes in neutrophil numbers in the whole-brain tissue of 3xTg-AD mice following anti-NK1.1 treatment (Supplemental Fig. 3B, 3C). In addition, depletion of neutrophils reduces Aβ burdens in mouse models of AD (14), whereas depletion of NK cells improves cognitive function without affecting Aβ burdens (Fig. 3A, 3B). Thus, NK cells and neutrophils might promote neuroinflammation through distinctive mechanisms.

Our work indicates a striking role for NK cells in promoting neuroinflammation and exacerbating cognitive decline in 3xTg-AD mice. Depletion of NK cells repressed neuroinflammation, stimulated neurogenesis, and improved cognitive function in 3xTg-AD mice. Amyloid β concentrations, however, were unaffected by NK cell depletion. This study, to our knowledge, thus establishes a novel role for NK cells in exacerbating AD-associated cognitive decline and suggests that targeting neuroinflammation may unlock new therapies to combat AD.

Several previous studies have indicated abnormal NK cell function in human patients of ADs (33–41). The results from these studies, however, were not entirely coherent (33–41). Some studies indicate increased functional capability of NK cells in AD, whereas others suggest decreased NK cell function (33–41). This might be partly because different in vitro assays have been employed to investigate NK cell function in different studies (33–41). In addition, the various stimulators used in in vitro assays could mask the homeostatic function of NK cells in physiologic conditions. Using scRNA-seq, our study provides an unbiased understanding of the physiologic biology of brain-associated NK cells in 3xTg-AD mice and control wild-type mice. Similar experiments in human patient samples would be highly worthwhile in future efforts.

Our results indicate that brain-associated NK cells are predominantly cytotoxic NK₁ subsets with minimal cytokine expression. Ifng expression was barely detectable in brain-associated NK cells, and Tnf expression was not detected. Interestingly, such transcriptomal profile of brain-associated NK cells differs from previously published NK cell profile in the mouse spleen where equivalent percentages of NK₁ and NK₂ cells have been observed (45). Of note, the brain is a delicate organ where excessive expression of stimulating cytokines might lead to significant detrimental effects. Thus, the near absence of cytokine-expressing NK cells might represent a protective mechanism that may help maintain brain homeostasis. Our results also indicate a hyperactive profile of NK cells in 3xTg-AD mice, which might be explained by several mechanisms. Aβ oligomers may directly bind to circulating lymphocytes through receptors such as RAGE and TLRs; the ligation of which may activate NK cells. The proinflammatory microenvironment in AD mice and humans, such as increased expression of IL-18, may also contribute to NK cell hyperactivation. Thus, a positive feedback loop might be formed between NK cell hyperactivation and microglial inflammation in AD, which might contribute to exacerbated neuroinflammation and cognitive decline.

Our data have revealed striking effects of NK cell depletion on improving cognitive function in 3xTg-AD mice. Of note, NK cells possess important beneficial function in healthy individuals, such as preventing tumor development and progression. Thus, long-term depletion of NK cells, particularly in humans, might incur unwanted side effects. As such, future exploration of more targeted therapies is warranted, which necessitates a more in-depth understanding of the specific molecular pathways by which NK cells promote neuroinflammation. Our results indicate that brain-associated NK cells predominantly express cytotoxic molecules such as granzymes and cathepsins. The roles of cytotoxic molecules and NK cells in immune and inflammatory disorders, beyond their cytotoxic function, remain poorly understood. Yet, increasing evidence indicates that when secreted into the extracellular microenvironment, cytotoxic molecules such as cathepsins and granzymes can have important function in triggering myeloid cell activation and in processing proinflammatory cytokines (22, 23, 25–28). Granzymes have been found to promote the production and maturation of proinflammatory cytokines in myeloid cells and other innate cells (24, 25, 48–54). Cathepsins may also play important roles in processing proinflammatory cytokines and in modifying tissue microenvironment (22, 23, 55–59). NK cells are enriched in brain barrier tissues (choroid plexus, perivascular space, and leptomeninges). In addition to their barrier function, these tissues also play essential roles in generating cerebrospinal fluid that nitrifies the brain parenchyma. We thus speculate that when released into the cerebrospinal fluid, the cytotoxic molecules produced by NK cells might influence the behaviors of brain parenchymal cells such as microglia, and dysregulation of this process may underlie the proinflammatory property of brain-associated NK cells in AD.

Another characteristic feature of AD is sharply declined neurogenesis (47). Neurogenesis remains active in the SVZ and, to a less degree, in the hippocampus regions of healthy adult humans and mice (47, 60). Adult neurogenesis was diminished in the inflamed brains of AD patients and transgenic AD mouse models, which might contribute to declined cognitive function in AD (47, 61). We have found that NK cell depletion significantly enhanced neurogenesis in both the SVZ regions and in the hippocampus. The enhanced neurogenesis might be due to reduced neuroinflammation. Improved neurogenesis, as well as other beneficial effects conferred by reduced neuroinflammation, likely together contribute to the alleviated cognitive decline in NK cell–depleted 3xTg-AD mice.

AD remains a devastating disease with unmet therapeutic needs. The development of successfully treatments relies on a comprehensive understanding of the cellular and molecular mechanisms involved. Although amyloid β pathologies are the defining features of AD, therapeutic efforts that target amyloid β have been futile, indicating the importance of other mechanistic pathways (62). In this work, we show that NK cell depletion significantly improves cognitive function of 3xTg-AD mice without affecting amyloid β concentrations. Strikingly, in NK cell–depleted 3xTg-AD mice, microglia demonstrated homeostatic-like morphology, reduced proliferative response, and decreased expression of proinflammatory cytokines. Interestingly, these microglia retained the capability to uptake amyloid-β, indicating that NK cell depletion does not appear to affect the phagocyte capability of microglia. These microglia also exhibited intact expression of Trem2, a surface receptor that plays an essential role in promoting microglial metabolic fitness (63). Thus, depletion of NK cells might specifically reduce the proinflammatory property of microglia without affecting the general fitness of microglia. Together, these results suggest that the presence of NK cells is pivotal in promoting neuroinflammation in 3xTg-AD mice. Our work thus indicates that targeting NK cells and neuroinflammation might provide new avenues to combat AD.

Q.Y. reported a patent (U.S. Patent Application No.: 62/822,159). The other authors have no financial conflicts of interest.