Cytotoxic Natural Killer Cells Disrupt Nerve Fibres Through Granzyme H in Atheriosclerotic Cerebral Small Vessel Disease

Background: Circulating natural killer cells (NK cells) are enriched in the central nervous system in atheriosclerotic cerebral small vessel disease (aCSVD) rat models, but their resulting effects and underlying mechanism remain to be investigated. Methods: A total of 32 patients with aCSVD and 28 healthy control patients were recruited for this study. According to white matter hyperintensity (WMH) burden, which is closely related to the severity of aCSVD, 20 participants were divided into two groups: burden of 0-1 (n=10) and burden of 2-3 (n=10). All participants participated in proteomics analysis of their cytotoxic NK cells and serum, and 3 participated in proteomics analysis of their cerebrospinal uid (CSF). In vitro BBB models and a co-culture system with primary human neurons were utilized to verify the pathogenic behaviours of cytotoxic NK cells. Results: In aCSVD patients with a high WMH burden, integrin β2 (ITGB2), cathepsin D (CTSD) and granzyme H (GZMH) were highly expressed in cytotoxic NK cells. ITGB2 interacted with intercellular adhesion molecule 1 (ICAM1) in vascular endothelial cells and promoted the adhesion of cytotoxic NK cells in vitro. Moreover, inhibition of CTSD reduced the destruction of type IV collagen (COL4A) in the extracellular matrix of the BBB and the leakiness of the BBB in vitro and in vivo, indicating that synthetic CTSD in cytotoxic NK cells participates in BBB damage. After passing through the leaky BBB, GZMH disruption on demyelinated nerve bres was reversed by cotreatment with the inhibitor 3,4-DCIC, suggesting that cytotoxic NK cell-released GZMH is crucial for the disruption of demyelinated nerve bres during WMH in aCSVD. Conclusions: Cytotoxic NK cells contribute to the CTSD-induced damage to the BBB and GZMH-induced disruption of demyelinated nerve bres during WMH in aCSVD. Our work

Lei Wei Third A liated Hospital of Sun Yat-Sen University granule toxins and granzymes or by classical caspase-dependent apoptosis [9]. In acute cerebral vessel disease, NK cells can pass through the disrupted BBB [12]. When cytotoxic NK cells are in direct contact with endothelial cells, angiogenesis is attenuated [13]. Whether cytotoxic NK cells aggravate the intercellular junctions of endothelial cells and penetrate the leaky BBB caused by chronic hypoperfusion in aCSVD remains to be clari ed.
Much less information is known about the relationship between cytotoxic NK cells and WMH. In rodents, NK cells are more abundant in the central microvascular system than in the peripheral blood [6,14]. Cytotoxic NK cells in ltrate and disintegrate intact sensory afferents through induction of NKG2D/RAE1 after injury in the peripheral nervous system (PNS) [15]. A similar phenomenon of cytotoxic NK cell migration also exists in the central nervous system (CNS) [16]. Cytotoxic NK cells can in ltrate the brains of patients after acute stroke and rodents in an induced stroke model through multiple pathways to exacerbate brain infarction [17][18][19]. Whether cytotoxic NK cells have direct contact with neurons and disrupt neurons in patients with aCSVD remains to be explored.
Based on the above premises, we con rmed that cytotoxic NK cells were more abundant in patients with aCSVD with a high WMH burden than in control patients, giving potential research value. Proteins are key components and regulators of biological processes and are related to the occurrence, development and prognosis of diseases. We aimed to explore the differential proteins between cytotoxic NK cells from patients with aCSVD with a high WMH burden and control patients by proteomics by adopting rigorous quality control standards. The objectives of this research were to use proteomics technology to determine whether mechanistic insights into the adverse effects of cytotoxic NK cells on the CNS could be discerned with bioinformatics analysis tools. Due to the di culty in obtaining brain specimens from patients with aCSVD, we co-cultured cytotoxic NK cells from patients with aCSVD with primary human neurons in a BBB model to verify the reliability of the proteomic results. Indeed, we found that cytotoxic NK cells were enriched around neural axon hillocks and disrupted nerve bres by releasing granzyme H (GZMH).

Study population
This was a prospective, single-centred trial that recruited 28 controls and 32 patients with aCSVD aged 50-90 years between January 01, 2020, and May 30, 2020 at the Department of Neurology in the Third A liated Hospital of Sun Yat-Sen University. The diagnostic criteria of aCSVD included (i) vascular risk factors for atherosclerosis; (ii) at least one common symptom, such as gait and balance disturbance or an emotional or sleeping disorder; and (iii) magnetic resonance imaging (MRI: T1 FLAIR, T2 FLAIR, susceptibility weighted imaging (SWI), arterial spin labelling (ASL, CBF), diffusion tensor imaging (DTI)) of the brain, including enlarged perivascular spaces (PVSs), deep and periventricular WMH, cerebral microbleeds (CMBs), lacunar infarction (LI), and cerebral atrophy [20]. The exclusion criteria included (i) a history of acute macrovascular disease; (ii) a history of severe liver or kidney diseases; (iii) a history of rectum or colon diseases; ( ) a history of malignant tumours; and ( ) disturbance of consciousness or death affecting the investigation. The PVSs, WMH, CMBs and LI burden combined were de ned as the aCSVD burden, and the WMH burden was determined according to the Fazekas scale and ranged from 0 to 3 [21]. All of these participants received MRI and gait impairment analyses, including pace, gait variation, gait symmetry, balance in six different directions, standing angle, standing time, sitting angle and sitting time. The participants also received laboratory blood tests, including measurements of the white blood cells (WBCs), neutrophils (NEUTs), lymphocytes (LYMPHs), monocytes (MONOs), eosinophils (EOs), basophilic granulocytes (BASOs), NK cells, high homocysteine (HCY), blood platelets (PLTs), glycosylated haemoglobin (HbA1c), total cholesterol (TC), triglycerides (TGs), low-density lipoprotein cholesterol (LDL-C) and high-density lipoprotein (HDL). After these tests, each participant donated 10 ml of blood and 1 ml of CSF for proteomics analyses.
This study protocol was approved by the ethical committee of the Third A liated Hospital of Sun Yat-Sen University. All participants signed written informed consent forms and were informed of all procedures.

Animals
Male Wistar-Kyoto rats (WKYs) and spontaneously hypertensive rats (SHRs) weighing 150-180 g were purchased from Beijing Vital River Laboratory Animal Technology Co. Ltd. (Beijing, China). They were housed under speci c pathogen-free conditions with free access to water and food. Eight months later, the WKY rats (controls) and SHRs (aCSVD model rats) were sacri ced (sodium pentobarbital, 100 mg/kg i.p.) [6].

Preparation of cytotoxic NK cells and ow cytometry
Blood samples from participants and rats were collected and centrifuged at 3000 rpm for 10 min within 1 h of collection. Serum was collected in 500 μl Eppendorf tubes and frozen at -80 °C. The lower precipitate from the blood and 1 ml of ddH 2 O were mixed with 5 ml of red blood cell lysis buffer at 4 °C for 15 min and then centrifuged at 1000 rpm for 5 min. The supernatant was discarded, and the cells were washed twice. Cells were resuspended in a tube in 100 μl of PBS containing 2 % BSA. One microlitre of antibody was added to the tube and fully blended followed by incubation at 4 °C for 30 min in the dark. Antibodies included an FITC anti-human CD3 antibody, a PE/Cyanine7 anti-human CD19 antibody, an APC antihuman CD16 antibody, and a PE anti-human CD56 antibody (BioLegend, USA). Each tube was centrifuged at 1000 rpm for 5 min, and the supernatant was discarded. The cells were then washed 2 times. Next, 500 µl of PBS containing 2 % BSA was used to resuspend the cells. The CD3 lo CD19 lo CD56 dim CD16 bright NK cells were analysed and isolated with a ow cytometer (BD, USA).
A total of 1×10 7 CD56 dim CD16 bright NK cells in a volume of 100 μl were put in 500 μl of liquid that was mixed with a methanol solution (60 % v/v) and a 0.85 % ammonium bicarbonate solution (wt/vol) at pH 7.4 and -20 °C. This tube was slightly shaken for 10 s and centrifuged at 1000 rpm for 1 min at 4 °C. The supernatant was removed, and the tube was placed in liquid nitrogen for 30 s and then stored at -80 °C.

Proteomics
Samples were mixed with 200 μl of 6 M guanidine hydrochloride solution containing 1 % protease inhibitor and ultrasonicated in an ice bath for 15 min. Next, 2 μl of 1 M dithiothreitol (DTT) was added to the tubes, followed by oscillation at 56 °C for 1.5 h. The tubes were then cooled to room temperature and 2 μl of 2 M iodoacetamide (IAA) was added for reaction in the dark for 30 min. The buffer solution was removed with 200 μl of a 50 mM ABC moisture-washed ultra ltration membrane. Next, 200 μl of 50 mM ABC solution was added to remove the guanidine hydrochloride. Buffer solution (100 μl, pH 8) and 1 μg of the protease trypsin were added for reaction at 37 °C for 12 h. Five microlitres of tri uoroacetic acid (TFA) was added followed by centrifugation to collect the peptide mixtures, which were desalinated with a C18 column. Analysis of the peptide mixtures was carried out by reversed-phase liquid chromatography (RPLC; Easy-nano 1000, Thermo, USA) and mass spectrometry (MS; Q Exactive Plus, Thermo, USA) using the data acquisition software Xcalibur (Thermo, USA). Raw les were quantitatively analysed by Protein Discovery 2.2.0 (Thermo, USA), and the database utilized was UniProtKB (2015_04, 42 121).

BBB model and study design
We used primary human brain microvascular endothelial cells (HBMECs; Zhongqiaoxinzhou, China) and primary human astrocytes (Zhongqiaoxinzhou, China) to establish the BBB model in vitro [23,24].
HBMECs were seeded in the upper layer of Transwell membranes (pore size: 12.0 μm) at a density of 1×10 6 /ml. After 24 h in a cell incubator at 37 °C, human astrocytes (HAs; Zhongqiaoxinzhou, China) were seeded on the opposite side of the Transwell membranes for another 24 h. Seven days later, both the HBMECs and HA were overgrown, producing the BBB model. The medium (90 % DMEM-H+10% FBS, endothelial cell medium, ScienCell, USA) was changed every three days.
Primary human neurons (HNs; ScienCell, USA) were cultured in neuronal medium (ScienCell, USA) in poly-L-lysine-coated 24-well plates at a density of 20000 cells/cm 2 . The medium was changed every two days.
Ten days later, the neurons had formed synaptic connections with other neurons and were used for coculture.
The BBB model (Transwell) was placed in a 24-well plate with HNs, and CD56 dim CD16 bright NK cells from the participants were seeded in the upper Transwell chamber. We used glucose-oxygen deprivation (OGD) to simulate WMH hypoperfusion [25,26]. This was the complete co-culture system required for this study.

Scanning electron microscopy (SEM)
Transwell membranes (1 cm 3 , with HBMECs and HAs) were placed in 4 % glutaraldehyde precooled at 4°C and xed at 4 °C overnight. Samples were then washed twice with PBS for 10 min each time. The membranes were xed with 1 % osmic acid precooled at 4 °C for 1 h followed by soaking twice with PBS for 10 min each time. The samples were then dehydrated with acetone/isoamyl acetate (1:1) for 10 min and then in isoamyl acetate for 30 min. After drying and vacuum spraying, the samples were observed with a scanning electron microscope (SU8100, Hitachi, Japan).

Enzyme-linked immunosorbent assay (ELISA)
Rat NK cells were rapidly frozen with liquid nitrogen, maintained at 2-8 °C after melting, and 500 μl of PBS (pH 7.4) was added for homogenization with a grinder. The cell culture supernatants, serum and CSF were collected in sterile tubes. The samples were centrifuged for 20 min at 2000-3000 rpm and the supernatants were collected carefully. Standards and samples were added according to the reagent instructions of the human ITGB2, CTSD, GZMH, FLNA, S1PR1, NEFM, B2M and NPTX2 ELISA kits (YI FEI XUE BIO TECH, China). The plates were closed with closure plate membranes and then incubated for 30 min at 37 °C. The washing solution was diluted with distilled water at a ratio of 1:1. The liquid in the plates was discarded, and then washing buffer was added for 30 s. This process was repeated ve times and then the plates were patted dry. Fifty microlitres of HRP-conjugate reagent was added to each well followed by incubation for 30 min at 37 °C and washing another ve times. Fifty microlitres of each chromogen solution A and chromogen solution B were added to the wells, which were incubated in the dark for 15 min at 37 °C. Finally, 50 μl of stop solution was added to each well. The absorbance of the blank well was set to zero, and the absorbance values of the other wells were measured at 450 nm within 15 min.
Haematoxylin-eosin (HE) staining The brain para n sections that had been baked were placed in xylene for 10 min. The sections were then transferred to anhydrous alcohol, 90 % alcohol, 80 % alcohol, 70 % alcohol and distilled water for approximately 2 min in each solution successively. Then, the cells were transferred to haematoxylin for 10 min and washed for approximately 2 min. The cells were transferred to differentiation solution (1% hydrochloric acid alcohol), and the sections were faded followed by washing in running water for 30 min, immersion in eosin solution for 3 min and washing with water. The brain slices were placed in 80 % alcohol for 2 min, 90 % alcohol for 4 min and anhydrous alcohol for 8 min successively. Finally, the slices were transferred to xylene for 3 min and xylene II for 5 min. The slices were sealed with resinene.

Luxol fast blue (LFB) staining
The slices were sealed in 0.1 % LFB solution at 60 °C for 10 h and then washed with distilled water and placed in 95 % alcohol. The slices were then transferred to 0.05 % lithium carbonate solution for 10 s and 70 % alcohol until the grey and white matter could be clearly distinguished. The slices were washed and redyed in 0.25 % tar-purple solution with a few drops of glacial acetic acid for 10 min. The slices were transferred to 70 % alcohol until the nuclei turned red and then dehydrated twice with n-butanol for 5 min each time. Finally, the slices were transferred to xylene for 3 min and xylene II for 5 min and sealed with resinene.

Statistical analysis
Experimental data are expressed as the mean ± SD using PRISM 7.0 (GraphPad Inc., USA) or SPSS 20.0 (IBM corp., USA). T tests, one-way ANOVA, correlation analysis, and fold changes from volcano plots were used for statistical analysis. The proteomics data were normalized according to z-score normalization [27]. Values of p < 0.05 were considered statistically signi cant. Heat maps and cluster maps were drawn with the online tool Heatmapper (http://www.heatmapper.ca/expression/). Volcano plots were analysed with the ImageGP tool (http://www.ehbio.com/ImageGP/index.php/Home/Index/). GO annotations and protein network analyses of the identi ed proteins were analysed using STRING (version 11.0, http://string-db.org).

Results
Increased cytotoxic NK cells was related to sparse nerve bres in patients with aCSVD with a high WMH burden.
We regrouped the 60 participants according to their WMH burden. As shown in Table S1, 38 participants were in the WMH burden 0-1 group, and the other 21 patients with aCSVD were in the WMH burden 2-3 group. The patients with a high WMH burden had hypoperfusion as observed by arterial spin labelling (ASL) imaging and sparse white matter bres (WMFs) by diffusion tensor imaging (DTI) (both p<0.05, Fig. 1, A; Table S1). We measured balance and gait impairment and found that high WMH burden patients had a prolonged sitting time (p=0.011, Fig. 1, C) and a higher number of NK cells in their blood (p=0.011, Fig. 1, D). Furthermore, a high WMH burden was moderately correlated with hypoperfusion (r=0.573, p=0.000), sparse WMFs (r=0.639, p=0.000), sitting time (r=0.476, p=0.025) and number of NK cells (r=0.509, p=0.000) (Fig. 1, E). NK cells also had a medium correlation with hypoperfusion (r=0.342, p=0.026) and sparse WMFs (r=0.511, p=0.002) (Fig. 1, E). These ndings suggested that the causal relationship between NK cells and sparse WMFs deserves further investigation. Next, we analysed the ve subtypes of NK cells by FC (Fig. 1, F). Compared with WMH burden 0-1 participants, the proportion of cytotoxic NK cells (CD56 dim CD16 bright ) increased most signi cantly in the patients with WMH burdens of 2-3 (p=0.000, Fig. 1, G). These results indicate that an increase in the number of cytotoxic NK cells was closely related to sparse nerve bres during WMH of aCSVD.
Cytotoxic NK cells in aCSVD migrated through ITGB2 and negatively regulated neural projection development by FLNA.
To explore the relationship between cytotoxic NK cells and sparse nerve bres in aCSVD, we isolated cytotoxic NK cells (CD56 dim CD16 bright ) from the WMH burden 0-1 and WMH burden 2-3 groups by FC (n=10). Based on proteomics analysis of the cytotoxic NK cells, 976 proteins were upregulated in the cytotoxic NK cells from the high WMH burden group, which indicated that they were in a highly activated state (all p<0.05, Fig. S1). We showed the cluster of the rst 300 differential proteins in the heat map (all p<0.01, Fig. 2, A). Gene ontology (GO) enrichment of cytotoxic NK cells found that the cellular component (CC) mainly included focal adhesions and extracellular exosomes (including cathepsin D; CTSD) (both p=0.000, Fig. 1, B), molecular function (MF) integrin binding (p=0.028), actin lament binding (p=0.000), and peptidase activity (CTSD, p=0.024, Fig. 2, E). The main biological processes (BPs), including NK cell activation (p=0.047), cell migration (p=0.028), regulation of cell shape (p=0.001), cell-cell adhesion (p=0.005) and cellular extravasation (p=0.008) were closely related to the key protein integrin β2 (ITGB2) (Fig. 2, C). GZMH was responsible for the BP of cytolysis (p=0.001, Fig. 2, C). Filamin-A (FLNA) was involved in cell junction assembly (p=0.000), positive regulation of actin lament bundle assembly (p=0.011), negative regulation of neuron projection development (p=0.021) and negative regulation of apoptotic processes (p=0.000, Fig. 2, C). In addition, FLNA participated in cell-extracellular matrix interactions through reactome pathway analysis (p=0.020, Fig. 2, D). The levels of ITGB2, GZMH, CTSD and FLNA were 2-fold higher in cytotoxic NK cells from patients with aCSVD with a 2-3 WMH burden than in NK cells from patients with aCSVD with a 0-1 WMH burden (all p<0.01, Fig. 2, F and G). In the interaction map, ITGB2 in cytotoxic NK cells interacted with intercellular adhesion molecule 1 (ICAM1) expressed in endothelial cells (interaction score: 0.989, Fig. 3, H) and promoted the adhesion of cytotoxic NK cells and endothelial cells. The above results suggest that the cytotoxic NK cells crossed the damaged BBB through ITGB2 and negatively regulated neural projection development through FLNA.
B2M mediated the lymphocyte immunoreaction in serum from patients with aCSVD.
To explore whether the serum components are involved in the in ammatory response of cytotoxic NK cells, we collected serum from the above mentioned patients for proteomic tests. The results showed that 35 proteins were upregulated and 16 proteins were downregulated (all p<0.05, Fig. 3, A). Among these upregulated proteins, differential β2-microglobulin (B2M) not only was related to extracellular exosomes in CC (p=0.000, Fig. 3, B), positive regulation of receptor binding (p=0.017) and negative regulation of neurogenesis (p=0.049) in BP (p=0.017, Fig. 3, E) but also participated in the immunoregulatory interactions between lymphoid and non-lymphoid cells from the reactome pathway analysis (p=0.000, Fig. 3, D). An interaction map clari ed that B2M in serum interacted with ICAM1 (interaction score: 0.944, Fig. 3, H) and then promoted the adhesion of NK cells and endothelial cells in the vasculature [29,30]. B2M in serum also interacted with CTSD (interaction score: 0.920, Fig. 3, H), which promoted the immune response of cytotoxic NK cells [13]. Correspondingly, downregulated sphingosine 1-phosphate receptor 1 (S1PR1) was involved in S1PR receptor activity in MF (p=0.017, Fig. 3, C), blood vessel maturation (p=0.019) and the S1PR signalling pathway in BP (p=0.019, Fig. S2, A). Both B2M and S1PR1 were changed by more than 2-fold (both p<0.05, Fig. 3, F and G). These results strongly suggest that upregulated B2M in serum might promote cytotoxic NK cell binding and interaction with other cells, such as endothelial cells, through ICAM1. In addition, downregulation of S1PR1 might be harmful to blood vessel maturation.
Axon, dendritic, and neuro lament fragments were observed in the CSF from patients with aCSVD.
To further identify the disruption of nerve bres in patients with aCSVD, we examined 3 CSF samples from the two groups by proteomics. There were 20 upregulated proteins and 44 downregulated proteins in the CSF of patients with aCSVD with a high WMH burden (all p<0.05, Fig. 4, A). The CC for CSF contained extracellular exosomes, axonal growth cones, focal adhesions, dendritic spine membranes and neuro laments (all p<0.05, Fig. 4, B). Neuro lament medium polypeptide (NEFM) took part in the structural constituent of the cytoskeleton in MF (p=0.039, Fig. 4, C) and neuro lament bundle assembly in BP (p=0.010, Fig. 4, D). On the other hand, the downregulation of neuronal pentraxin-2 (NPTX2) was mainly responsible for the regulation of postsynaptic neurotransmitter receptor activity in BP (p=0.001, Fig. 4, D). Both NEFM and NPTX2 changed by more than 2-fold (both p<0.01, Fig. 4, E and F), indicating that a large number of nerve bres were disrupted and the transmission of synaptic activity was reduced in patients with aCSVD with a high WMH burden. The 29 common BPs in the CSF and cytotoxic NK cells included the regulation of lymphocyte migration, negative regulation of neuron projection development and negative regulation of neuron apoptotic processes (all p<0.05, Fig. S2, B), demonstrating that cytotoxic NK cells migrated and disrupted nerve bres in the CNS of patients with aCSVD.
The volcano plot displayed a two-fold change in cadherin-5 (CDH5, p=0.006, Fig. 4, F and G) in the CSF located on the external side of the plasma membrane in CC (p=0.007, Fig. 4, B), showing functions in signalling receptor binding (p=0.039, Fig. 4, C) and participating in cell-cell adhesion via plasma membrane adhesion molecules (p=0.008, Fig. 4, D). Even though CDH5 interacted with FLNA in cytotoxic NK cells (interaction score: 0.227, Fig. 3, H), we still thought that CDH5 was not the target protein for cytotoxic NK cell-targeting neurons because CDH5 is often expressed in endothelial cells and not neurons [31]. Instead, differential CDH5 in CSF proved the leakage of the damaged BBB in patients with aCSVD with a high WMH burden.
Cytotoxic NK cells exacerbated the BBB damage by CTSD to cross the damaged BBB.
To understand whether cytotoxic NK cells pass through the BBB in the WMH area of patients with aCSVD, we used an OGD cell model to simulate hypoperfusion in WMH [23,32]. We observed the inner sides and outer sides of Transwell membranes covered with HBMECs and HA, respectively, by examining the ECM and found that the connections between these cells were very tight, similar to the normal BBB (Fig. 5, A). Intercellular connections of HBMECs consist of intercellular adhesion molecule 1 (ICAM1), collagen (COL4A), laminin-R (LR), and so on [33][34][35]. ICAM1, COL4A and LR were highly expressed by HBMECs in vitro on the fourth day (all p<0.05, Fig. 5 Fig. S4, A and B) were consistent with the results from the serum of the participants, which synergistically promoted cytotoxic NK cell migration, adhesion and damage to the BBB [38,39]. We also observed that cytotoxic NK cells from patients with aCSVD with a high WMH burden crossed the damaged BBB model in the lower layer (Fig. 5, M; Fig. S3, E). These results suggested that cytotoxic NK cells adhered to endothelial cells with ITGB2, exacerbated BBB damage through CTSD and then crossed the leaky BBB in patients with aCSVD.
Due to the decrease in nerve bres as observed by MRI and the leaky BBB, we explored the effects of cytotoxic NK cells on neurons from patients with aCSVD with a high WMH burden. In the rat model of aCSVD, the thicknesses of both the subcortical white matter (SWM) and corpus callosum (CC) decreased (both p<0.05, Fig. S5, A and B). Additionally, in this area, myelin was vacuolated (p<0.05, Fig. S5, A and C). Combined with the high WMH signal in patients with aCSVD, we speculated that most nerve bres were demyelinated [1,23]. Moreover, the intensities of the nerve bres also decreased in this area of the aCSVD model (p<0.05, Fig. S5, A and D), which was consistent with the sparse WMF observed from DTI in patients with aCSVD (Fig. 1, A). To explore the relationship between increased cytotoxic NK cells and demyelinated nerve bres in white matter, we co-cultured cytotoxic NK cells isolated from participants with HAs for 4 h in vitro (Fig. 6, A). Interestingly, cytotoxic NK cells from patients with aCSVD with a high WMH burden were enriched around neural axon hillocks and disrupted the nerve bres without myelin sheaths, which decreased the number and length of neural axons and dendrites (all p<0.05, Fig. 6, A, D and E). In the co-cultured supernatants and CSF from the rat model, the increase in NEFM and decrease in NPTX2 veri ed nerve bre injury (all p<0.05, Fig. 6, B and C; Fig. S5, E and F). Cytotoxic NK cells from patients with aCSVD adhered to neural axon hillocks by highly expressing FLNA in vitro (Fig. 6, F) and in vivo (p<0.05, Fig. S5, G and H). In vivo, cytotoxic NK cells from the aCSVD model synthesized a large number of GZMHs (p<0.05, Fig. S5, I). The damaged neurons were also rich in GZMH in the SWM of the aCSVD model (Fig. S5, K and J). To further explore the mechanism by which cytotoxic NK cells disrupt nerve bres, an inhibitor of GZMH was injected into each rat's tail vein and co-cultured with cytotoxic NK cells from patients with aCSVD, which reduced the disruption of nerve bres (all p<0.05, Fig. 6, G and H; Fig. S5, A, D and E). Additionally, phosphorylation of the death receptor (p-DR), which is involved in apoptosis, did not change in vitro or in vivo, indicating that the apoptotic pathway was not activated (Fig.   6, H and G; Fig. S5, K and J). These results demonstrated that GZMH released by cytotoxic NK cells was a key determinant of the disruption of demyelinated nerve bres during WMH in patients with aCSVD.

Discussion
Our study establishes a previously unrecognized detrimental effect of cytotoxic NK cells in aCSVD. The data in this study suggest that cytotoxic NK cells highly express ITGB2 and bind with ICAM1 in the endothelial cells of the BBB through immune synapses, releasing CTSD and exacerbating BBB damage.
After crossing the damaged BBB, cytotoxic NK cells are enriched around neural axon hillocks and disrupt nerve bres by releasing GZMH during WMH in aCSVD patients.
Our study revealed alterations in protein levels in cytotoxic NK cells, serum and CSF from patients with aCSVD with a high WMH burden. In terms of cytotoxic NK cells, a large number of proteins were highly expressed, indicating that these cytotoxic NK cells were signi cantly activated. These differential proteins accelerate cytotoxic NK cell adhesion, transformation and migration and play killing roles through immune synapses, including lysis and phagocytosis. Certain differentially upregulated proteins in serum also contributed to the migration, adhesion and immunoreactions of lymphocytes, including cytotoxic NK cells. Among them, upregulated B2M was involved in modulating the immune response, secreting extracellular exosomes and negatively regulating neuron projection development [29]. Upregulated FLNA in cytotoxic NK cells also played a role in the negative regulation of neuron projection development.
However, the expression of its downstream proteins were decreased in CSF, including moesin and serine/threonine-protein kinase 10 [40,41]. Considering that all of these results were derived from existing databases through bioinformatics analysis, new components or pathways that were not in the database could not be summarized. Therefore, we speculated that cytotoxic NK cells were enriched around demyelinated neurons through the action of FLNA or B2M, but their targeted proteins need further exploration.
Previous studies models of in ammation in the CNS and acute cerebrovascular diseases have demonstrated the in ltration of NK cells through the severely damaged BBB [42,43]. Our results revealed that the direct contact between cytotoxic NK cells and endothelial cells broke the connections between cells, such as the interactions with COL4A. Notably, we found cytotoxic NK cells expressing ITGB2 in aCSVD. ITGB2 not only is a receptor for ICAM1 in endothelial cells but also contributes to the cytotoxicity of NK cells [30,44]. Our ndings supported the pivotal role of cytotoxic NK cells in the adhesion of endothelial cells and exacerbated damage to the BBB by secreting proteases.
Until now, pathological nerve bre damage during WMH in aCSVD has remained controversial [4,21,45]. severe aCSVD, and certain autopsy results have con rmed that the density of demyelinated nerve bres decreases in severe aCSVD [2,4,48,49]. In this study, we showed that sparse nerve bres were present in patients with aCSVD with WMH, and this phenomenon coexisted with demyelination in a rat model of aCSVD. Therefore, all of the above evidence con rmed that demyelinated nerve bres were sparse during WMH in aCSVD.
Another new nding from this study was that cytotoxic NK cells disrupted demyelinated nerve bres through GZMH in patients with aCSVD with a high WMH burden. Interestingly, after disrupting demyelinated nerve bres, neurosomes remained. GO analysis showed that the neuronal apoptotic signalling pathway was negatively regulated. p-DR plays a role in neuronal apoptosis, including normal neurosome death and axonal pruning [50]. In our experiments, the phosphorylated DR-induced classical apoptotic pathway was not activated, and it was con rmed that the dissolution of nerve bres was not dominated by the apoptotic pathway.
There are some shortcomings in our research. On the one hand, the sample size in these experiments was small. On the other hand, we only inhibited proteinase activity by treatment with 3,4-DCIC in cytotoxic NK cells, and CTSD and GZMH knockout mice are needed in a follow-up study. Furthermore, the factors that induce NK cell cytotoxicity in aCSVD remain unclear. Further studies of the mechanisms are needed to solve these problems in patients with aCSVD.

Conclusion
The phenomenon of cytotoxic NK cells exacerbating BBB damage through ITGB2 and CTSD and disrupting nerve bres by releasing GZMH during WMH in aCSVD are presented here. Our results propose a new pathological mechanism of cytotoxic NK cells to disrupt nerve bres during WMH in aCSVD and provide a potential therapeutic target for aCSVD.

Availability of data and materials
All data generated and analysed during this study are included in this published article, and its supplementary information les are available from the corresponding authors upon reasonable request.

Competing interests
The authors declare that they have no competing interests.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.