2.1 Characteristics of the study population and detection of DNMT3A-driven CHIP

A total of 15,166 patients were recruited from August 2015 to March 2018 in the CNSR-III, which is a nationwide and comprehensive clinical evaluation registry of ischemic stroke or TIA in China based on etiological classification, imaging, and biological markers.²⁰ The CNSR-III comprised 93.3% AIS patients (n = 14,146) and 6.7% TIA patients (n = 1020). To explore the association of DNMT3A-driven CHIP with neurological functional disability of ischemic stroke, we included 10,241 AIS/TIA patients with qualified WGS data²¹ in this study. The patients with loss of follow-up (n = 114), history of malignancies (n = 94), TIA diagnosis (n = 679), recurrent stroke at 3 months (n = 612), and CHIP driven by other genes (n = 218) were excluded (Figure 1). Finally, 8524 AIS patients were enrolled in this study.

The baseline characteristics of the 8,524 patients are summarized in Table 1. The median age of the patients was 62 years (interquartile range [IQR], 54.0−70.0), and 90 of 8524 patients (1.1%) were identified as DNMT3A-driven CHIP carriers. DNMT3A-driven CHIP carriers were 5 years older than patients without the DNMT3A mutation. DNMT3A-driven CHIP carriers were more likely to be female than non-DNMT3A carriers (42.4% vs. 30.2%, p = 0.01), which might have led to a significantly lower rate of smoking habits in DNMT3A-driven CHIP carriers (23.3% vs. 33.3%, p = 0.045). Except for coronary heart disease, traditional cardiovascular risk factors, such as a history of AIS/TIA, hypertension, diabetes mellitus, and hypercholesterolemia, were not significantly different between DNMT3A-driven CHIP carriers and non-DNMT3A carriers. The levels of low-density lipoprotein, cholesterol, and interleukin-1 receptor antagonist were significantly elevated in DNMT3A-driven CHIP carriers.

2.2 DNMT3A-CHIP is associated with neurological functional disability in acute ischemic stroke

During a follow-up period of 3 months, we observed a higher rate of neurological functional disability in DNMT3A-driven CHIP carriers than that in non-DNMT3A carriers (Figure 2A). Multivariable analysis showed that the odds ratio (OR) with a 95% confidence interval (CI) of DNMT3A-driven CHIP for risk of neurological functional disability was 1.39 (1.27−1.52; Figure 2B). The associations between DNMT3A-driven CHIP and neurological functional disability in the prespecified subgroups are shown in Figure 3. The effect of DNMT3A-driven CHIP on functional disability was adjusted by sex, baseline high-sensitivity C-reactive protein (hsCRP) levels, baseline interleukin-6 (IL-6) levels, and the modified Rankin scale (mRS) score at discharge. DNMT3A-driven CHIP increased the risk of neurological functional disability only in females (OR, 2.45; 95% CI, 2.14−2.80; p_interaction < 0.001), patients with baseline hsCRP > 3.0 mg/L (OR, 1.58; 95% CI, 1.41−1.78; p_interaction = 0.002), baseline IL-6 > 3.4 mg/L (OR, 1.87; 95% CI, 1.66−2.11; p < 0.001), and patients with mRS ≥ 2 at discharge (OR, 1.37; 95% CI, 1.23−1.53; p_interaction = 0.03).

2.3 The DNMT3A inhibitor RG108 significantly increased infarct volume and worsened neurobehavioral deficits in mice with transient middle cerebral artery occlusion

To explore the potential mechanisms underlying the effect of DNMT3A mutation on neurological functional disability in AIS, the function of DNMT3A was inhibited using the small-molecule compound SGI1027²² or RG108 (10 mg/kg)²³ in mice with transient middle cerebral artery occlusion (tMCAO; Figure 4A). The Preliminary experiment results show that RG108 has superiority over SGI-1027 in DNMT3A functional inhibition (Figures S1 and S2A). RG108 is recognized as a specific inhibitor of both DNMT3A and DNMT1.^{13, 17} At 24 h post-tMCAO, we assessed infarct volume using T2-weighted imaging (T2WI), apparent diffusion coefficient (ADC) parameters of magnetic resonance imaging (MRI; Figure 4B), and 2,3,5-triphenyltetrazolium chloride (TTC) staining (Figure 4D). Our results showed a significant approximately 20% increase in infarct volume in the RG108 group as compared with the vehicle group (T2WI: 118.6 mm³ vs. 97.96 mm³; ADC: 120.2 mm³ vs. 101.0 mm³; TTC: 37.39% vs. 28.99%; Figure 4C, E). During 18–24 h post-tMCAO, three different behavioral tests were used to assess neurological deficits after stroke. First, we scored neurological function using the modified Garcia test,^{24, 25} which showed that the RG108 group had a lower mean neurological score than the vehicle group (5.62 vs. 6.85, p < 0.05, Figure 4F). Second, the foot misplacement apparatus (FMA) approach^{26, 27} showed that mice treated with RG108 tended to have more average missed steps (14.8 vs. 5.3, p < 0.05, Figure 4G). Third, an open-field test was used to assess the motor activity of rodents. The RG108 and vehicle groups displayed a significantly decreased total distance compared with the sham group (Figure 4H). Besides, the RG108 group has a decreased mean total distance compared with the vehicle group (777.2 vs. 2561.6, p < 0.05; Figure 4H).

RG108 treatment for DNMT3A inhibition resulted in an expansion of infarct volume and deterioration of neurological deficits after cerebral ischemia in mice. Brain tissue and peripheral blood were collected to verify the molecular effect of RG108 on DNMT3A. We quantified DNMT3A using western blotting in tissues of the ischemic penumbra (the area adjacent to the infarct in the 11 o'clock direction of the brain slice) and the control area (the contralateral side of the penumbra, the 13 o'clock direction of the brain slice; Figure 5A). Compared with the sham group, the DNMT3A level was significantly increased in the control area and penumbra in mice with tMCAO, regardless of whether they received RG108. However, RG108 further elevated the DNMT3A level in the penumbra area compared with the vehicle group (Figure 5A, B; Figure S2B). ELISA of peripheral blood cells showed that 5-methylcytosine (5-mC) in the RG108 group was significantly lower than that in the vehicle group, which could be due to DNMT3A function loss (Figure 5C). Taken together, cerebral ischemia increased DNMT3A levels in the brain, particularly in the penumbral area. RG108 inhibits DNMT3A function, which leads to a compensatory increase in DNMT3A protein expression.

2.4 DNMT3A inhibition boosts neuroinflammation after ischemic stroke

To explore the underlying mechanism of DNMT3A inhibition, we performed a transcriptomic analysis of the ischemic penumbra cortex of mice treated with RG108 or vehicle (Figure 6A). Bulk RNA-sequencing was used on ischemic penumbra tissues harvested from mice 24 h post-tMCAO. We processed samples from six mice, with three serving as a vehicle group and three treated with RG108. Principal component analysis showed that the RG108 group differed dramatically from the vehicle group, and there was a relatively small within-group variability in the transcriptome. Meanwhile, the correlation analysis of gene expression between samples showed similar results to those of the principal component analysis (Figure 6B). We found 616 differentially expressed genes (DEGs) upregulated (fold change > 2 and adjusted p-value < 0.05) and 91 DEGs downregulated (fold change ← 2 and adjusted p-value < 0.05) in the RG108 group compared with the vehicle group (Figure 6C, D; Tables S3 and S4). Upregulated DEGs were used for biological pathway analysis based on gene ontology (GO) enrichment, GO networks, and the Kyoto Encyclopedia of Genes and Genomes (KEGG). RG108 is involved in proinflammatory biological processes, including leukocyte migration and cell–cell adhesion, cytokine production and interaction, extracellular structure and matrix organization, granulocyte/neutrophil migration, and cell–cell adhesion (Figure S3A–D, Tables S5 and S6).

2.5 Neutrophil migration and chemotaxis are enhanced by DNMT3A inhibition in ischemic stroke

To avoid the omission of some important but less obviously altered genes in GO and KEGG enrichment analyses, gene set enrichment analysis (GSEA) was performed with all the genes in the experiment.²⁸ GSEA showed that the top pathways included neutrophil/granulocyte/leukocyte migration and chemotaxis, and positive regulation of tumor necrosis factor superfamily cytokine production, which were similar to the results of GO terms and KEGG pathway enrichment analysis (Figure 6E–H, Tables S7). According to pathway analysis, three pathways were mainly enriched: neutrophil migration and chemotaxis, leukocyte migration and cell–cell adhesion, and cytokine production and interaction. The expression of the top 20 DEGs is shown in the corresponding heatmaps (Figure 6I–K). To further verify the results of these enrichment analyses, we performed quantitative real-time PCR and ELISA for a subset of the top DEGs in the indicated pathway. Compared with the sham group, nearly all genes related to the three pathways were significantly upregulated in the poststroke brain. On the other hand, four genes (Il11, Il6, Cxcl1, and Cxcl2) of the cytokine production and interaction pathway, six genes (Il6, Cxcl1, Cxcl2, S100a8, Sell, and Thbs1) of leukocyte migration and cell−cell adhesion, and four genes (Cxcl1, Cxcl2, S100a8, and Sell) of neutrophil migration and chemotaxis were significantly upregulated in the RG108 group compared with the vehicle group (Figure 6L). As expected, the level of the proinflammatory cytokine IL-6 in the serum of the RG108 group was also higher than that in the vehicle group (Figure 6M).

2.6 Increased neutrophils in the peripheral blood and ischemic hemisphere with DNMT3A inhibition

To further validate the results of RNA sequencing, we tested the immune response in the peripheral blood and ischemic hemisphere. First, we performed peripheral hematological parameter analysis at three different time points: baseline, 3 days after continuous administration (tMCAO 0 h), and 24 h after tMCAO. White blood cell (WBC, Figure 7A) and neutrophil (Figure 7B) counts continuously decreased in all three groups (sham, vehicle, and RG108). At 24 h after tMCAO, compared with the vehicle group, the RG108 group had many more neutrophils (Figure 7B). In addition, the percentage of neutrophils comprising the WBC increased from 27.03% to 33.16% (p < 0.01, Figure 7C). Therefore, DNMT3A inhibition promotes peripheral blood neutrophil proliferation after ischemic reperfusion.

Based on these results, we tested myeloid cells in the ischemic hemisphere 24 h after tMCAO by flow cytometry,^{29, 30} and CD45⁺CD11b⁺ myeloid cells were concatenated after manual gating to perform subclustering with high-dimensional analysis (Figure 7D, E; Figure S4). CD11b⁺CD45^low microglia comprised the largest population of immune cells in the brain (Figure 7D). Compared with the sham group, the vehicle and RG108 groups had more infiltrated CD45⁺CD11b⁺Ly6G⁺ neutrophils and CD45⁺CD11b⁺Ly6G⁻F4/80⁺ macrophages in the ischemic hemisphere, including increased CXCR2⁺ activated cells^31-33 (Figure 7D–G). The vehicle and RG108 groups had fewer microglia and more CXCR2⁺-activated microglia (Figure 7H). In addition, compared with the vehicle group, the RG108 group had much more infiltrated neutrophils (Figure 7F) than macrophages (Figure 7G). However, CXCR2⁺-activated neutrophils and macrophage cells did not show differences between the vehicle and RG108 groups (Figure 7F, G). Taken together, DNMT3A inhibition promotes the infiltration of neutrophils into the ischemic hemisphere after ischemic reperfusion. To explore proliferation gene expression in neutrophils, we have conducted additional flow cytometry analysis of Ki67 expression in neutrophils from peripheral blood and ischemic brain tissue in tMCAO model mice. The results show that there are more Ki67-positive neutrophils both circulating and infiltrating into brain tissue with RG108 treated (Figure S5A–D).

4.1 Study population

Patients enrolled in this study were selected from the Third China National Stroke Registry (CNSR-III),²⁰ a nationwide prospective registry for patients presenting to hospitals with acute ischemic cerebrovascular events, including AIS and TIA. Patients from 201 hospitals across China were recruited for this study from August 2015 to March 2018. The diagnosis of AIS was based on the WHO criteria and confirmed by MRI or computed tomography of the brain. A total of 8524 AIS patients without a history of malignancy and recurrent stroke at 3 months were included in this study. More details on the patient enrollment process are shown in Figure S1. Signed informed consent was obtained from all the patients before they were enrolled in the study. The study protocol was approved by the ethics committee of Beijing Tiantan Hospital. Patients at 3-month follow-up visits were interviewed face-to-face by trained research coordinators. The mRS was used to evaluate the functional dependence of each patient (range 0 [no symptoms] to 6 [death]).⁵⁸ The primary outcome in the current study was neurological functional disability, defined as an mRS score of 3−6.

4.2 Whole-genome sequencing and somatic mutation detection

Samples preservation and WGS analysis protocols were published as previous.²¹ Total DNA was isolated from peripheral WBCs using a magnetic blood genomic DNA kit (DP329; TIANGEN Biotech Co Ltd). WGS was performed using the BGISEQ-500 platform (BGI Genomics) with an average depth greater than 30× for each subject.²¹ GATK MuTect2 software (https://software.broadinstitute.org/gatk) was used to detect the putative somatic variants of DNMT3A. All variants with a VAF of at least 0.05 (5%) were considered. Variants that had been previously reported in the literature and/or the Catalog of Somatic Mutations in Cancer (http://cancer.sanger.ac.uk/cancergenome/projects/cosmic/) were used in the subsequent analysis.

4.3 Animals

Experiments were performed on C57Bl/6J male mice (age 8−10 weeks, weight 25 g). Animals were fed on a normal chow diet, had ad libitum access to food and water, and maintained at 24°C under a 12 h light/dark cycle.

Apart from G*Power, no other statistical methods were used to predetermine the sample size.^{59, 60} The number of experimental mice for each group is specified in the corresponding figure captions. The mice were randomized into each group and the investigators were blinded to allocation during experiments and outcome assessment.

4.4 Drug administration

RG108 (N-phthalyl-L-tryptophan, DNA methylation inhibitor), 10 mg/kg (Selleck Chemicals; catalog number: S2821) or SGI-1027, 10 mg/kg (Selleck Chemicals; catalog number: S7276) was dissolved separately in dimethyl sulfoxide and then diluted in normal saline. RG108/ SGI-1027 and vehicle were injected via the tail vein once a day 3 days before surgery and once again at the onset of reperfusion, four times in total.

4.5 Cerebral ischemia model: transient middle cerebral artery occlusion

To induce ischemia/reperfusion brain injury, mice from both drug- and vehicle-injected groups were subjected to tMCAO surgery, as previously described.^{12, 61} The tMCAO experiments were performed in a blinded manner. The detailed procedure of the tMCAO surgery can be found in the Supporting Information.

4.6 MR imaging of cerebral infarcts with mice

MRI scans were performed at the Beijing Neurosurgical Institute Imaging Facility using an echo planar capable, 7.0 T small animal MRI system (Bruker; BioSpec70/20USR). Scans were performed in anesthetized (2.5% ethobrom) mice 24 h poststroke, as previously described.⁶²

4.7 Infarct analysis by TTC staining

At 24 h after cerebral ischemia, the mice were deeply anesthetized with isoflurane. The brains were rapidly removed and sliced into seven 1-mm thick slices. The infarct volume was measured by TTC staining for approximately 30 min, as previously described.¹² Using ImageJ software, we scanned and calculated the relative area of the corresponding region in each brain slice. The detailed calculating procedure of the infarct volume can be found in the Supporting Information.

4.8 Neurological function assessment

Mouse neurological function was assessed by experimenters without knowing any information about the mice. Briefly, we conducted a modified Garcia Test 18−24 h after tMCAO to assess neurological deficits, and five tests were performed, including body proprioception, vibrissae touch, limb symmetry, lateral turning, and forelimb walking. The scores on each test ranged from 0−3, as previously described.^{24, 25}

4.9 Behavioral testing

All behavioral tests were carried out 24 h after tMCAO. The investigators were blinded to the allocation during the experiments and the outcome assessment.

Foot misplacement test: Mice were placed in a testing room for 60 min before testing. An automatic foot misplacement apparatus (Bioseb In Vivo Research Instruments; # BIO-FMA) with a mouse corridor was used.

The open field test: Mice were placed in the open field from the same corner (50 cm × 50 cm × 40 cm, a total of 25 squares), after which they were adapted for 1 min. Their activities in the open field within 30 min were photographed and recorded.

The detailed procedure of the behavioral testing can be found in the Supplementary Information (SI).

4.10 RNA-sequencing and data analysis

Mice were anesthetized 24 h after tMCAO and perfused with cooled phosphate-buffered saline. The ischemic penumbra tissue was quickly dissected and separated on ice (the penumbra area was defined as adjacent to the infarct area, which was milky white after stripping the skull). Six mice were sampled (three in the vehicle group and three in the RG108 group). DEseq2⁶³ was used for principal component analysis and identification of differentially expressed genes. Genes with an expression |log₂fold change| > 1 and q-value (adjusted p-value) < 0.05 were defined as statistically significantly differentially expressed.

All heatmaps, volcanic maps, and violin diagrams were visualized using ggplot2 in R, version 4.1.0. The detailed procedure of the sample treatment and data analysis can be found in the Supporting Information.

4.11 ELISA assay

The levels of IL-6 and 5-mC in the peripheral blood were detected using IL-6 (Abcam, ab222503) and 5-mC (Epigentek; P1030) ELISA kits according to the manufacturer's instructions. Absorption at 450 nm was determined using a microplate reader (Tecan Trading AG; # SPARK). The concentration of IL-6 and 5-mC were determined according to the standard curve generated simultaneously.

4.12 Flow cytometry and hematological cell counts

Mouse peripheral blood and brain samples were collected and processed for single-cell suspensions. Blood was drawn under deep anesthesia, and hematologic parameters were assessed using an automated cell counter machine (BC-5000 Vet, Mindray Animal Medical). After blood collection, the brain was perfused and dissected, with the ischemic hemisphere treated with collagenase and digested at 37°C. The resulting mixture was centrifuged, and the pellet was resuspended and purified using Percoll gradient centrifugation to isolate viable cells. These were then resuspended in the FACS buffer for further analysis. A panel of fluorophore-conjugated antibodies was used to label various immune cell types, including myeloid, microglia, neutrophil, and macrophage populations. The stained cells were analyzed using a Beckman flow cytometer (CytoFLEX S).

Flow cytometry data were subsequently analyzed using the Flow Jo software (BD, V10.8). Manual gating and subclustering of each sample was performed as indicated in the supplementary material (Figure S4), followed by unsupervised high-dimensional subclustering analysis. The cell counts and the activated percentage of the indicated myeloid cell subgroups were calculated. The antibodies and reagents used in this study are listed in Table S2. The detailed procedure and additional data can be found in the Supporting Information.

4.13 Data analysis and statistics