Single-cell transcriptomics of neuroinflammation and cerebrovascular endothelial cells in the aged rat hippocampus

Summary

Perioperative neurocognitive disorders (PNDs) are common in the elderly, with hippocampal neuroinflammation playing a key role in their pathogenesis. To characterize the cellular landscape of neuroinflammation, we performed single-cell RNA sequencing (scRNA-seq) of hippocampal cells from aged rats on day 3 after systemic lipopolysaccharide (LPS) injection. We identified distinct transcriptomic alterations in microglia, infiltrating immune cells, and notably cerebrovascular endothelial cells (ECs). ECs exhibited prominent oxidative stress–related changes, including the upregulation of DNA-damage-inducible transcript 4 (REDD1). In vitro REDD1 knockdown alleviated oxidative stress and promoted ECs survival. Our findings offer a high-resolution map of the neuroinflammatory microenvironment in the aged hippocampus and underscore the role of ECs dysfunction, blood-brain barrier (BBB) disruption, and immune infiltration in PNDs-related pathology.

Graphical abstract

Highlights

• •LPS activates microglia, recruits immune cells, and alters EC phenotype in the aged brain
• •Apoptosis pathways were significantly activated in ECs
• •Oxidative stress-related DEGs were enriched in ECs, including Ddit4/Redd1
• •Ddit4/Redd1 knockdown reduced oxidative stress and improved ECs survival in vitro

Geriatrics; Neuroscience; Immunology

Introduction

Perioperative neurocognitive disorders (PNDs) in the elderly are associated with slower recovery, reduced postoperative life quality, and even shorter long-term survival.¹ Accumulating evidence suggests that neuroinflammation induced by aseptic trauma from surgical stress or perioperative infection contributes to PNDs progression.² The hippocampal region is critical for cognitive dysfunction, which is especially vulnerable to neuroinflammation. Therefore, understanding the mechanisms of neuroinflammation triggered by surgical stress in the aging hippocampus is essential for reducing the incidence of PNDs.

Our previous studies have shown that patients with preoperative cognitive impairment had higher baseline plasma LPS levels compared to those with normal cognitive function, and plasma LPS levels increased postoperatively in all patients with high inflammatory cytokines.³ The administration of the bacterial toxin LPS is a widely used model to study the mechanisms of neuroinflammation⁴ as well as PNDs.⁵ Peripheral innate immune cells such as monocytes are activated by LPS and release inflammatory factors, including tumor necrosis factor α (TNF-α), interleukin (IL)-1β, and IL-6, which contribute to the loss of blood–brain barrier (BBB) integrity,⁶ activation of microglia⁷ and infiltration of peripheral immune cells into parenchyma.⁶ In addition, peripheral pro-inflammatory cytokines upregulate the expression of intercellular cell adhesion molecule-1 (ICAM-1) and vascular cell adhesion molecule-1 (VCAM-1) by cerebrovascular endothelial cells (ECs), which drives further recruitment of immune cells.^8,9

It was reported that peripheral monocyte infiltration occurs only in the presence of substantial BBB dysfunction,¹⁰ and ECs are the main constituent cells of the BBB. Thus, ECs damage may be critical for neuroinflammation. Prior scRNA-seq studies related to neuroinflammation have primarily focused on immune cells, such as microglia,¹¹ macrophages¹² or T cell.¹³ However, few studies have investigated the alterations in ECs under inflammatory conditions, particularly in aging populations with heightened susceptibility to PNDs. The mechanisms through which transcriptomic alterations in ECs influence BBB function and neuroinflammatory microenvironment remain elusive. We hypothesized that ECs are a major target of peripheral inflammatory signals, leading to BBB damage, microglial activation, immune cells infiltration, and ultimately leading to neuroinflammation aggravation, neural degeneration, and cognitive dysfunction.

In the present study, we applied scRNA-seq to explore the inflammatory gene expression profiles and intercellular signaling pathways active in the inflammatory microenvironment of aging rat hippocampus. Furthermore, we explored the role of specific differentially expressed genes in the ECs, providing a potential therapeutic target for the prevention and treatment of BBB damage and neuroinflammation underlying PNDs.

Results

Cognitive deficits and single-cell levels in the hippocampus of aging rats following lipopolysaccharide-induced neuroinflammation

To assess the effects of LPS-induced neuroinflammation on higher cognitive function, spatial learning and memory were evaluated using the Morris water maze test.¹⁴ The LPS group exhibited a more aimless swimming trajectory, reflecting deficits in spatial learning and memory (Figure 1B). During the five consecutive training (spatial learning) days, aging rats from two groups achieved significantly shorter escape latencies. However, latencies were significantly prolonged in the LPS group on days 14 compared to the NC group (p < 0.05) (Figure 1C). In contrast, there was no significant difference in swimming speeds between groups except for training day 1 (Figure S1A). During the probe trial on day 7, LPS treated aging rats spent significantly less time in the former platform quadrant (p = 0.019) and made fewer platform location crossovers (p = 0.015) than the NC group (Figures 1D and 1E).

Figure 1.

The establishment and landscape of neuroinflammatory microenvironment in aging rats hippocampus

(A) Morris water maze (MWM) training protocol. Spatial acquisition trials and probe trials were performed from day 1 to day 7.

(B) Representative swimming trajectory on the probe test day 7 between two groups.

(C) Statistical plot of escape latency during acquisition training on days 1–5 post-intervention between LPS and NC groups of aging rats.

(D and E) Rats treated with LPS showed reduced the number of platform crossovers and time spent in the target quadrant compared with the NC group on the day 7 probe test.

(F and G) RT-PCR shows differences in mRNA expression for Tnf-α and Il-1β in the aging rats hippocampal tissues. Data are presented as mean ± SEM, independent t-test, ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001. Behavior results n = 8 rats/group, qPCR results n = 4–5 rats/group.

(H) Graphical diagram of single cell isolation and RNA-seq experimental setup.

(I) UMAP plot visualizes the clustering of single cells colored by cell types and the distributions among two groups.

(J) Dot plot represents the top 3 most highly expressed genes within each cell type relative to the rest of clusters.

(K) Band plots show the distribution of DEGs (dots) in the 8 major cell types that induced by LPS. Red dots represent DEGs showing significant (p Value _FDR < 0.01 and |avg_log2FC| > 0.25) up or down-regulation of expression response to LPS stimulation, and black dots represent DEGs that are outside of the threshold.

To confirm that this cognitive dysfunction induced by LPS was associated with neuroinflammation, we measured the expression levels of the proinflammatory genes Tnf-α and Il-1β. Intraperitoneal injection of LPS markedly upregulated the hippocampal expression of Tnf-α mRNA compared to the hippocampus of the NC group on days 1, 3, and 7 post-injection (P_D1 = 0.002, P_D3 = 0.0001, P_D7 = 0.001), with peak expression on day 3 (Figure 1F). Similarly, hippocampal Il-1β mRNA expression was also significantly higher than that in the NC group on days 1, 3, and 7 (P_{D1, 3} < 0.0001, P_D7 = 0.001) (Figure 1G). Thus, LPS induced neuroinflammatory signaling within the hippocampus, with the difference from controls most pronounced on day 3 after injection.

Based on these results, we conducted single-cell transcriptome analysis (scRNA-seq) on aging rat hippocampus on day 3 after LPS or vehicle injection using the 10X Genomics platform to examine changes in the neuroinflammatory microenvironment (Figure 1H). After filtering out cells with low quality reads (Figure S1B), we obtained transcriptome datasets from 19,325 cells with a mean of ∼2000 genes detected per cell and a median of ∼6,000 unique molecular identifiers (UMIs) per cell from both LPS and NC groups. The gene expression profiles of hippocampal cell types were identified using scRNA-seq, including microglia, choroid plexus cells, ependymal cells, ECs, T cells, B cells, oligodendrocytes, pericytes, monocytes, fibroblasts, macrophages, neutrophils, astrocytes, neuronal restricted progenitors, and erythroid cells (Figure 1I). Each cell type was successfully annotated by the detection of known marker genes such as C1qa for microglia,^15,16 Cldn5 for ECs,¹⁷ Clec10a for macrophages,¹⁸ and S100a6 for monocytes (Figure 1J).¹⁹ Compared to NCs, LPS significantly increased the proportions of microglia (76.6% vs. 73.9%), macrophages (1.8% vs. 0.4%), and monocytes (1.2% vs. 0.6%) but decreased the proportion of ECs (5.6% vs. 6.4%). Further, group comparisons revealed large numbers of genes differentially expressed by microglia and ECs between groups (Figure 1K).

Unsupervised clustering revealed a diversity of microglial subtypes

To assess the degree of phenotypic heterogeneity among the major resident immune cells in the aging rat hippocampus, we first analyzed group differences in microglial expression profiles using UMAP dimensionality reduction. Microglial re-clustering yielded eight subpopulations (MG0–MG7) (Figure 2A) with significant differences in relative distribution between LPS and NC groups (Figure 2B). Specifically, the proportion of subpopulation MG0 was 41.7% greater, while those of MG4, MG5, and MG7 were 32.8%, 39.9%, and 30.4% greater, respectively, in the LPS group. In contrast, the proportions of MG1, MG2, and MG6 were reduced in the LPS group. The proportion of MG3 showed no significant change between groups.

Figure 2.

Single-cell transcriptome analysis of the microglia

(A) UMAP shows the distribution of each subtype of microglia.

(B) The sector graph shows the composition of cells in subclusters by groups.

(C) Violin plot depicts the expression levels of known core signature genes for each microglia subcluster.

(D) Representative immunofluorescence double staining images of NFKBIA (red), IBA1 (green), and nuclei were labeled with DAPI located in hippocampus in the NC and LPS groups. Scale bar = 75 μm or 25 μm. Quantitative analysis of the proportion of NFKBIA⁺ cells in microglia (IBA1+) in hippocampal DG subregion. Data are shown as mean ± SEM, independent samples t-test, n = 4, ∗∗p < 0.01.

(E) Marker genes enriched KEGG pathway analyses in various microglia subpopulations.

(F) GO analysis shows the top five signaling pathways across the four subpopulations, MG0, MG4, MG5 and MG7.

Marker gene profiles were then examined to identify each cluster (Figures 2C and S2A). Subpopulation MG0 highly expressed genes encoding the small ribosomal subunit (Rps) and large ribosomal subunit (Rpl) proteins (Rpl3, Rpl17, Rpl5, Rpl37), as well as the complement gene C3, which is released by microglia and involved in synaptic clearance and redevelopment.^16,20 The MG4 subpopulation was enriched in Ccl3, Ccl4, and Il1b, genes that promote the migration of peripheral monocytes, lymphocytes, and neutrophils into the brain parenchyma, thereby creating a neuroinflammatory microenvironment.^21,22 In addition, MG4 highly expressed Egr1, which is involved in neuronal synaptic pruning and neuronal cell maturation. A specific proinflammatory microglial type has been identified in both human and rodent tissues that overexpresses immunoregulatory genes (Il1a, Il1b, Ccl3, Ccl4) and immediate-early genes such as C-fos and Egr1.²³ Like MG4, MG5 also highly expressed pro-inflammatory genes, as well as the Rano class I histocompatibility antigen genes RT1- Ba, RT1-Db1, RT1-Da, RT1-B, and CD74 (MHC class II invariant chain, Ii). Microglia with high MHC class II expression are implicated in neuroinflammation-related disorders.²⁴ In addition, MG4 and MG5 subpopulations also exhibited strong expression of the stress signaling-associated target gene Nfkbia (Figure S2A). Nfkbia transcription is induced by NF-κB signaling, making its mRNA a marker of pathway activation.²⁵ Further, immunofluorescence staining of the MG4/5 marker IκBα confirmed that the proportion of this subpopulation was markedly increased in the dentate gyrus of LPS-treated rats (Mean_LPS = 33.86%, Mean_NC = 7.28%, p = 0.03) (Figure 2D). Finally, MG7 was defined as a type I interferon-associated subpopulation as evidenced by high expression of the interferon-induced genes Ifit3, Ifit2, and Ifi27l2b. This interferon-associated microglia subtype has also been found in AD²⁶ and aged rodents.²⁷ In contrast to these proinflammatory phenotypes, MG1 and MG2 represent resting state subpopulations strongly expressing homeostatic genes such as Gpr34, P2ry12, Hexb, Tmem119, Ctss, and Cx3cr1.^16,28

According to the KEGG database, the expression profiles of inflammation-associated microglia subpopulations MG0, MG4, MG5, and MG7 were all highly enriched in genes of the IL-17 and TNF signaling pathways (Figure 2E). Gene Ontogeny (GO) function analysis further revealed that the MG0 profile was highly enriched in genes related to ribosomal biogenesis, MG4 in genes associated with IL-1β and LPS responses, and MG5 in genes associated with antigen presentation and immunomodulation (Figure 2F). We also applied “AddModuleScore” and “AUCell” Seurat functions to identify the inflammatory signatures of MG4 and MG5 subtypes (Figure S2B).

Differential expression gene analysis of microglial subtypes

The proportions of microglial subtypes MG0, MG4, MG5, and MG7 were significantly modulated following LPS exposure, so we further cataloged the LPS-dependent DEGs in each cluster. These subtypes exhibited 65 common DEGs following LPS stimulation compared to the control condition, including ribosomal genes (Rpl/Rps), the complement gene C3, and the inflammatory and proliferation-related genes Il1b, Adgre1,²⁹ Cxcl13,³⁰ and Top1 (Figures 3A and S2C).³¹ Immunofluorescence staining of hippocampal slices from aging rats verified the stronger expression of complement C3 and IL-1β in microglia from LPS group rats compared to NC group rats (P_C3 = 0.01, P_IL-1β = 0.04) (Figures 3B and 3C).

Figure 3.

Differential gene expression patterns in microglia subclusters

(A) Violin plot shows the expression of the representative DEGs overlapped in MG0, MG4, MG5 and MG7 subpopulations, including Rpl8, C3, Il1b, Ftl1, C1qb, Cxcl13, and Top1.

(B and C) Representative images of immunostaining for C3, IL-1β and Iba1 in the hippocampus of aging rats on day 3 after LPS injection. Quantification of C3 and IL-1β immunostaining signal in microglia in LPS and NC groups in DG region. Scale bar, 75 μm or 25 μm. Data are shown as mean ± SEM, independent samples t-test, n = 4, ∗p < 0.05.

(D) Violin plots demonstrate the distribution of subpopulation-specific DEGs in four microglia subpopulations between LPS and NC groups, including the gene Ccr5 in MG0, Cd68 in MG4, Ccl2 in MG5 and Ifi27l2b in MG7.

Furthermore, we found unique DEGs in four microglia subtypes following LPS stimulation (Figures 3D and S2B). Compared to NCs, Serpinb1a, Il18, and Ccr5, genes associated with IL-1β production and signaling, were differentially expressed only in MG0 (Figure S2C). The Ccr5 gene encodes a receptor, Ccr5, that binding multiple chemokine ligands, including Ccl3, Ccl4, and Ccl5.³² Further, CCR5-dependent migration of microglia to the vasculature contributes to the diffusion of peripheral inflammatory mediators.³² In MG4 cells, LPS induced the differential expression of Ccrl2, Cxcl2, and Cd68, all of which are associated with chemokine-mediated signaling pathways. The chemokine CXCL2 released from microglia in ischemic animals contributes to the recruitment of neutrophils,³³ while CD68 is a lysosomal protein and specific marker of activated phagocytizing microglia.³⁴ Finally, MG5 differentially expressed Tlr2 and Ccl2, genes associated with the macrophage colony stimulating factor response.

Lipopolysaccharide-induced proliferation of macrophages and expression of proinflammatory genes

CNS-associated macrophages (CAMs) are a small population of resident immune cells situated at the CNS border^19,35 that isolate the CNS from the periphery but also mediate communication between the CNS parenchyma and periphery.³⁶ The major subtype of CAMs in hippocampus is the perivascular macrophages (PVMs).^37,38 In pathological or inflammatory states, the number of macrophages in brain parenchyma increases due to the proliferation of CAMs and the recruitment of peripheral monocytes from blood, which form monocyte-derived macrophages (MDMs).³⁹

These PVMs were successfully distinguished from other hippocampal cells by the high expression of known marker genes Mrc1,¹⁹ Pf4,⁴⁰ Ms4a7³⁹ and low expression of Ptprc (Cd45) compared to peripheral immune cells (Figure 4A).⁴¹ To investigate changes in PVM subtypes during inflammation, we performed re-clustering analysis, yielding 3 subclusters (PVM0–PVM2) (Figure 4B). The proportion of all PVM subtypes was much higher in the LPS group than the NC group (Figure S3A), suggesting that neuroinflammation is associated with the proliferation of resident macrophages. The PVM0 subcluster highly expressed Lyve1 and Ccl3, genes associated with myeloid differentiation, migration, and chemokine responses (Figure S3B), while PVM1 overexpressed Cd74, RT1-Bb, and Lilrb4, genes associated with MHC class II antigen presentation and immune regulation (Figure S3C).

Figure 4.

Differential gene expression patterns in macrophage subclusters

(A) Violin plot shows the marker genes that distinguish perivascular macrophages (PVMs) and monocyte-derived macrophages (MDMs) from other immune cell types, including gene Pf4, Mrc1, Ms4a7, Ptprc, S100a6, Crip1, Lyz2.

(B) UMAP plot reveals that PVMs in the aging rat hippocampus in two groups are present in 3 distinct clusters, PVM0-PVM2.

(C) UMAP plot shows the distribution of 2 subpopulations of MDMs in the LPS and NC groups.

(D) Bar graph shows the percentage of the MDM subgroups in the LPS and NC groups respectively.

Monocyte-derived macrophages, the main leukocytes infiltrating from the periphery into THE brain parenchyma under LPS exposure, highly expressed S100a6, Crip1, Lyz2, and Cd45 (Figure 4A).^19,40 These MDMs accounted for a low percentage of the total cell population in the aging hippocampus of NC rats but were more abundant in the LPS group (Figures 4C and 4D). Re-clustering analysis further divided MDMs into Mdm0 and Mdm1. The Mdm0 subpopulation highly expressed RT1 and other genes associated with T cell-mediated immune responses and antigen presentation (Figure S3D), while Mdm1 highly expressed S100a8, Lgals3, Prdx6, and other genes related to cellular detoxification. The injection of LPS increased the proportion of Mdm1 cells and upregulated expression of the phagocytic vesicle/endocytic vesicle-related genes Vim and Capg (online Table S3).

Unsupervised clustering revealed a diversity of endothelial cell subtypes under lipopolysaccharide stimulation

The disruption of the BBB is strongly implicated in the invasion of peripheral monocytes into the brain parenchyma, and the structural and functional integrity of ECs is essential for maintaining BBB function. Unsupervised clustering reveals six EC subtypes (EC0–EC5) in both LPS and NC groups (Figures 5A and 5B), with EC0 constituting a significantly greater subpopulation in the LPS group compared to the NC group (59.45% vs. 40.71%) (Figure 5C). In accord with previous studies,^17,42,43 the EC4 subtype highly expressed the arterial-EC markers Fbln5, Bmx, Efnb2, and Vegfc (Figure 5D), while we found that EC3 overexpressed the venous-EC marker gene Nr2f2. Finally, EC0 and EC5 highly expressed Rgcc, Cxcl12, and Slc16a1, markers of capillary-ECs, while EC0 also highly expressed Mfsd2a (Capillary-EC) (Figure 5E), which encodes the BBB-associated lipid transporter.⁴² Other subclusters, such as EC1 and EC2, exhibited intermediate phenotypes and were therefore difficult to define precisely.

Figure 5.

Single-cell transcriptome analysis of the cerebral vascular endothelial cells

(A–C) UMAP plot and bar plot showing the distribution of 6 subpopulations of cerebral vascular endothelial cells in the LPS and NC groups.

(D) Violin plot shows the gene expression related to vascular origin (arterial, venous, and capillary), including arterial endothelial cell marker genes Fbln5, Bmx, Efnb2, Vegfc. The venous endothelial cells highly expressed gene Nr2f and capillary endothelial cells highly expressed gene Rgcc and Slc16a1.

(E) Expression profiles of EC0 Marker genes including Mfge8, Lrg1, Lgals9, Cldn5, Ocln, Tjp1, Ddit4/Redd1, Mfsd2a are shown using the UMAP visualization approach.

(F) Marker genes in the EC0 subpopulation are enriched with GO functional analysis.

(G) Volcano plot depicts the DEGs at overall level of cerebral vascular endothelial cells between LPS and NC groups. DEGs (|log2(fold change)| > 1, p Value FDR <0.05, Difference = |pct.1- pct.2 | > 0.2) were colored (red for upregulated DEGs and blue for downregulated DEGs.

(H and I) GO analysis shows the upregulated signaling pathway at overall level of cerebral vascular endothelial cells and EC0 subpopulation respectively.

Treatment with LPS increased the number of EC0 cells (Capillary-EC) in the hippocampus of aging rats. To further characterize the Capillary-EC, we performed GO term enrichment analysis, which revealed that EC0 marker genes are enriched in genes associated with blood vessel morphogenesis, the regulation of epithelial cell proliferation, and cell-cell adhesion (Figure 5F). In addition, EC0 highly expressed the tight junction–related genes such as Tjp1 (ZO-1), Cldn5, and Ocln (Figure 5E).

LPS injection also upregulated the expression levels of transcriptomic ribosomes (Rpl/Rps), complement C3, Vcam1, Lrg1, Vim, Ddit4/Redd1, and Sult1a1 (Figure 5G). Corresponding functional and pathway analyses of DEGs compared to the NC group indicated that LPS upregulated the vascular cortex apoptotic signaling pathway genes Ddit4, Rps7, Ppial4d, Tpt1, Rps3, Rpl26, Clu, Hint1, and Rpl11 (Figure 5H). In addition, LPS induced EC0 cell overexpression of Ddit4, Txnip, Angptl4, Eif4, and Rpl11, Angptl4, Eif4ebp1, Ucp2, Ndrg1, Slc1a1, Prdx6, and Gpx1, genes associated with decreased oxygen and ROS responses (Figure 5I). DDIT4, also known as REDD1, is a key mediator of cellular stress responses, primarily through the repression of the central metabolic regulator mTORC1, but also independently by promoting oxidative stress via increased ROS production and suppression of antioxidant defenses.⁴⁴ In particular, thioredoxin binding protein (TXNIP), which has been identified as a downstream target of DDIT4 and is known to promote intracellular oxidative stress⁴⁵ was found in our study to be associated with responses to reactive oxygen species and hydrogen peroxide.

Cell-to-cell interaction analysis between inflammation-activated immune cells and endothelial cells

We then investigated the relationships and functional interactions among the major constituent cells of the neuroinflammatory microenvironment induced by LPS using the CellChat R package. Some forms of intercellular communication among ECs and immune cells were detected in both groups, while others were dependent on inflammatory status (Figure 6A). Treatment with LPS enhanced the complement, CCL chemokine, vascular endothelial growth factor (VEGF), semaphorin 3 (SEMA3), and insulin-like growth factor (IGF) signaling pathways. We further examined the sources of the Igf1–Igf1r ligand–receptor pair (Figure 6B) and found that microglia and macrophages (PVMs) are the most prominent source of the Igf1 ligand, while the receptor Igf1r is expressed mainly by ECs. Further, this IGF pathway between microglia and ECs was more active in the LPS group, as LPS exposure induced the overexpression of Igf1 in microglia and Igf1r in ECs (Figures 6C and 6D).

Figure 6.

Cell–cell interactions between immune cells and cerebral vascular endothelial cells

(A) The comparison of the signaling pathway based on the communication information flow between LPS and NC group.

(B) Bar plots show the relative contribution of each ligand–receptor pair in IGF signaling pathway (up). Hierarchical plots show the inferred IGF signaling networks in NC and LPS group (down).

(C and D) Dot plots and violin plots show the expression of genes related to IGF signaling pathway between LPS and NC groups.

(E) Bubble plots show the exchange of signaling pathway between endothelial cells (ligand source) and immune cells (receptor source) in two groups.

(F) Circle plots show the inferred SEMA3 signaling networks in NC and LPS groups.

(G) Violin plots show the expression of genes including genes Sema3d, Sema3a, Sema3c, Cx3cl1, Hbegf, and Lgals9 between LPS and NC groups.

Figure 6E presents the expression levels of other gene pairs by ECs and immune cells under LPS and control conditions. The SEMA3 signaling pathway was activated following LPS treatment (Figure 6F), with ECs being the primary ligand source and acting in both autocrine and paracrine pathways. Notably, we found that LPS elevated the expression of multiple genes related to the SEMA3 signaling pathway, mainly in ECs and microglia. In addition, Cx3cl1–Cx3cr1, Hbegf–Egfr, and Lgals9–Cd45 signaling pathways from ECs to microglia, macrophages (PVMs), and MDMs were significantly enhanced in the LPS group. The CX3CL1 protein synthesized in ECs acts as an adhesion factor to capture CX3CR1-expressing immune cells, leading to leukocyte infiltration.⁴⁶ ECs also release the tandem repeat cysteine protein Galectin-9 into the inflammatory environment to promote the recruitment of leukocytes and thereby expand the inflammatory activity of macrophages.⁴⁷ Notably, LPS induced the overexpression of Sema3d, Sema3a, Sema3c, and Cx3cl1, which are the corresponding ligand genes of the aforementioned pathways in ECs (Figure 7G), indicating that ECs contribute to enhanced peripheral immune cell infiltration into brain parenchyma following exposure to LPS.

Figure 7.

Cerebral vascular endothelial cells REDD1 expression induced by LPS exposure

(A) Representative immunofluorescence double staining images of REDD1 (red) and CD31 (green) in the aging rat hippocampus on day 3 after a single intraperitoneal injection of LPS. Scale bar = 25 μm. Quantitative analysis of REDD1 fluorescence in endothelial cells (CD31⁺).

(B) Representative confocal microscopic images of NC and LPS rats hippocampus stained for IgG. Nuclei are labeled with DAPI. Scale bars: 75 μm or 25 μm. Quantitative analysis of IgG extravascular deposits in two groups at hippocampus DG region.

(C) Representative Western blot bands and quantitative analysis of REDD1 protein in hCMEC/D3 cells stimulated by LPS at an intervening concentration of 100 ng/mL for 24 h (p < 0.0001).

(D) Changes in REDD1 mRNA expression were measured by qPCR in hCMEC/D3 cells stimulated by LPS at an intervening concentration of 100 ng/mL for 24 h. Data are expressed as mean ± SEM, independent samples t-test, n ≥ 3, ∗∗p < 0.01, ∗∗∗∗p < 0.0001.

REDD1 expression increases after lipopolysaccharide exposure

To validate the key DEGs involved in LPS-induced vascular endothelial cell dysfunction, we focused on the gene REDD1, which is related to EC apoptosis and oxidative stress response in our study, and previously reported to contribute to endothelial senescence by repressing oxidative stress.⁴⁸ The hippocampi of aging rats were harvested on day 3 after the injection of LPS and analyzed for REDD1 expression by immunofluorescence co-staining with the ECs marker protein CD31. Indeed, REDD1 was highly co-localized with CD31, and this co-expression was increased by LPS (p < 0.0001) (Figure 7A).

Single-cell sequencing results also revealed an increased number of peripheral-derived monocytes in the hippocampus after LPS exposure. In addition, differential expression analysis of Ddit4/Redd1, Ppial4d, Tpt1, Bcl2l1, Bad, Pmaip1, and Ybx3 indicated that these ECs were functional stratified into apoptosis. Taken together, these results suggest that LPS markedly impairs BBB function by altering EC abundance and phenotype. To directly test BBB integrity, we measured the rate of IgG extravasation (Figure 7B) and found a significantly greater accumulation in the hippocampus of the LPS group (p = 0.0003), confirming that LPS stimulation leads to BBB hyperpermeability.

To investigate the molecular mechanisms in isolation from other physiological variables, we conducted in vitro experiments using the human cerebral microvascular EC line hCMEC/D3. The REDD1 mRNA expression in hCMEC/D3 cells was markedly and dose-dependently enhanced by LPS (Figure S4), reaching a peak at 24 h after 100 ng/mL LPS administration. Further, the stimulation of hCMEC/D3 cells with 100 ng/mL LPS for 24 h substantially increased both REDD1 protein levels (p < 0.0001) (Figure 7C) and REDD1 mRNA expression (p = 0.002) (Figure 7D) compared to the NC group.

REDD1 knockdown in hCMEC/D3 cells alleviated lipopolysaccharide-induced oxidative stress and apoptosis

Cultures of hCMEC/D3 cells were randomly divided into four treatment groups according to LPS intervention and REDD1 knockdown using sh-REDD1#3: PBS + shNC, PBS + shREDD1, LPS + shNC and LPS + shREDD1. Treatment with 100 ng/mL LPS and shNC transfection led to a significant accumulation of ROS in ECs, as detected by the dihydroethidium (DHE) probe 24 h after LPS exposure, compared to PBS + shNC cells (p < 0.0001) (Figures 8A and 8B). While REDD1 knockdown prior to LPS stimulation significantly reduced ROS levels (LPS + shNC vs. LPS + shREDD1, p = 0.004).

Figure 8.

REDD1 gene knock-down decreased oxidative stress related cell damage and apoptosis

(A and B) ROS production assessed by DHE (red) staining of hCMEC/D3 cells after interventions. Nuclei are labeled with DAPI. Relative intracellular DHE fluorescence among four groups: PBS+shNC, PBS+shREDD1, LPS+shNC, and LPS+shREDD1 group. Scale bar = 50 μm.

(C) Relative fluorescence density of intracellular TMRM (red).

(D) Relative ATP production in hCMEC/D3 cells after LPS stimulation and gene transfection.

(E and F) Representative images of the fluorescence assay (left panel) and quantification (right panel) of double-staining of TUNEL (green) and DAPI (blue) on hCMEC/D3 cells among four groups. Scale bars = 50 μm.

(G) The hCMEC/D3 cells supernatants were collected for LDH release assays. Data are expressed as mean ± SEM, one-way ANOVA with Tukey’s test, n ≥ 3, ∗p < 0.05, ∗∗p < 0.01, ∗∗∗∗p < 0.0001. Data represent three experiments for in vitro assays.

Oxidative stress responses are associated with mitochondrial dysfunction, and increased oxidative stress can lead to a loss of mitochondrial membrane potential (Δψm) and ATP synthesis capacity. Therefore, changes in Δψm were examined using TMRM after LPS stimulation in cells transfected with shNC or shREDD1. Treatment with LPS significantly reduced (depolarized) Δψm in shNC-transfected cells compared PBS (vehicle) treatment, as evidenced by a marked reduction in the relative fluorescence intensity of TMRM (PBS+shNC vs. LPS+shNC, p=0.025) (Figure 8C). However, REDD1 gene knockdown significantly reversed this effect of LPS on Δψm compared to the LPS+shNC group (p=0.003). In addition, cellular ATP was slightly reduced by the LPS stimulation of hCMEC/D3 cells (Mean_PBS+shNC=15.50 nmol/g, Mean_LPS+shNC =12.66 nmol/g) (Figure 8D), and this effect was also alleviated by REDD1 knockdown (Mean_LPS+shNC = 12.66nmol/g, Mean_LPS+shREDD1 =18.21 nmol/g,p= 0.015). Therefore, knockdown of REDD1 in hCMEC/D3 cells can sustain mitochondrial biogenesis and reduce oxidative stress under LPS stimulation. These effects of REDD1 downregulation may, in turn, protect neurovascular ECs against damage under inflammatory stress, thus preventing BBB damage and induction of neuroinflammation and associated neurological impairments. Indeed, the oxidative stress response induced by 100 ng/mL LPS treatment for 24 h significantly increased the proportion of TUNEL-positive hCMEC/D3 cells (Mean_{PBS + shNC} = 4.83% vs. Mean_{LPS + shNC} = 18.23%, p = 0.003) (Figures 8E and 8F), and this proapoptotic effect was partially reversed by REDD1 knockdown (Mean_{LPS + shNC} = 18.23% vs. Mean_{LPS + shREDD1} = 6.38%, p = 0.002). In addition, the release of lactate dehydrogenase (LDH) from hCMEC/D3 cells induced by LPS (1.34-fold compared to the PBS + shNC treatment group, p = 0.001) was reduced to 0.85-fold by REDD1 knockdown (LPS + shNC vs. LPS + shREDD1, p = 0.017).

Discussion

We examined the gene expression changes in aging rat hippocampus during the acute phase of LPS-induced neuroinflammation by scRNA-seq analysis,¹⁶ which revealed significant impacts on cell-type composition, subset-specific gene expression, enriched pathways, and cell–cell communication. The major findings of this study are that (i) LPS induced phenotypic stratification of microglia into various proinflammatory subtypes expressing distinct sets of cytokines, complement, and chemokines to recruit peripheral immune cells; (ii) peripheral-derived monocytes infiltrate the brain parenchyma under LPS exposure and further shift the brain microenvironment toward a neuroinflammatory state; (iii) among genes upregulated by LPS, DDIT4/REDD1 may be a particularly important mediator of EC oxidative stress and mitochondrial dysfunction, leading to loss of BBB integrity. We speculate that this loss of BBB integrity facilitates peripheral immune cell infiltration and further exacerbates neuroinflammation, ultimately leading to neuronal damage and cognitive dysfunction as observed in neuroinflammatory diseases such as PNDs.

Microglia are the major resident immune cells in the brain parenchyma, and single-cell transcriptome sequencing in different neurological disorders has demonstrated marked phenotypic diversity.^16,40,49 In the present study, we found that the proportions of microglia that function in immune homeostasis (MG1 and MG2) were reduced in the LPS group, in accord with previous studies.⁵⁰ Homeostatic microglia maintain the microenvironment, but under pathological conditions, gradually transform to phenotypes with reduced expression of homeostatic molecules.⁵¹ In this study, these disease-associated microglial subtypes expressed high levels of complement C3 and ribosomal genes (MG0), proinflammatory cytokines (MG4/5), or interferon-associated markers (G7) by day 3 after LPS injection. The ribosomal genes upregulated in MG0 cells by LPS have also been detected in aging,⁵² Alzheimer’s disease⁴⁹ and spinal cord injury.¹⁶ This increase in ribosomal genes enhances the protein synthesis capacity of microglia required for phenotypic transformation.¹⁶ Microglia subtypes with high C3 expression were also found in human spinal cord pain models associated with synapse elimination¹⁶ and dendrite loss.^53,54 In addition, inflammatory microglia (MG4/5) highly expressed Ccl3 and Ccl4 following LPS injection. Upregulated microglia Ccl3 and Ccl4 expression levels have also been reported in senescence,⁵⁵ neurodegenerative diseases,⁵⁶ and experimental autoimmune encephalitis.³⁹ We provide evidence that microglia subpopulations are similarly diverse during LPS-induced inflammation, driving the recruitment of peripheral immune cells into the central nervous system and creating a proinflammatory microenvironment.

Analysis of ligand and receptor expression levels revealed that LPS induced microglia to overexpress ligands Ccl3, Ccl4, and receptor Ccr5 to recruit more activated microglia and amplify the inflammatory response through an autocrine pathway. In addition, LPS induced the activation of interferon-associated microglial subsets in the hippocampus of aging rats. Although the proportion of interferon-associated microglia was relatively small, this subtype has been detected in traumatic brain injury and is strongly associated with macrophage infiltration and cognitive recovery.⁵⁷

Treatment with LPS also induced the differential expression of many distinct genes in specific microglia subpopulations, creating further functional heterogeneity. For instance, only MG0 cells demonstrated the substantial upregulation of Ccr5, a chemokine receptor secreted by microglia and PVMs. Haruwaka and colleagues demonstrated that systemic inflammation induces the CCR5-dependent migration of brain-resident microglia to the perivascular space, where they impair BBB integrity by phagocytosing astrocyte terminal pseudopods. Alternatively, MG5 cells exhibited upregulated expression of Ccl2, which can expand the inflammatory response by recruiting peripheral and central immune cells into brain lesions.⁵⁸ CD68 is a transmembrane glycoprotein expressed on lysosomes, endosomes, and plasma membranes of microglia and macrophages, and is an important marker of activated phagocytic microglia. LPS induced an increase in microglia inflammation-related phenotypes and enhanced activation. In turn, these activated cells produced a variety of inflammatory factors, chemokines, and complement components that act on central and peripheral immune cells and cause the spread of neuroinflammation.⁵⁹

PVMs are a small population of CNS resident immune cells located in the perivascular space that respond differently under homeostatic and disease conditions.⁵⁹ These cells highly expressed the chemokine genes Ccl6, Ccl24, and Ccl8, which are presumably involved in the recruitment of peripheral leukocytes to the CNS.⁶⁰ A subpopulation of PVMs (PVM0) with the high expression of Lyve1 and Ccl3 was also identified in the current study. In addition, functional analysis of PVM0 cells revealed high expression of genes related to immunomodulatory functions such as leukocyte migration and myeloid leukocyte differentiation.

In previous studies, most monocytes highly expressing the marker genes Ly6c and Ccr2 were defined as monocytes of peripheral origin. Ochocka and coworkers found that monocyte-differentiated macrophages strongly expressed Ifitm2, S100a6, and S100a11, and overexpressed Cd274, which encodes the immune checkpoint inhibitor PD-L1, but did not express detectable Ly6c or Ccr2.¹⁹ In the present study, we found a phenotype of MDMs strongly expressing S100a6 and S100a11 as well as peripheral immune cell marker genes. The number of MDMs was increased in the hippocampus of the LPS group, and functional analysis indicated that these cells contribute to T-cell-mediated immune responses, cellular detoxification, and oxidative stress. Zheng and colleagues reported that proinflammatory monocytes/macrophages in the cerebral hemispheres of ischemic animals expressed high levels of genes associated with ROS detoxification, such as Prdx5 and Prdx6.⁴⁰ In the LPS group, Ccl3–Ccr1 ligand–receptor signaling was enhanced, indicating that chemokines released by microglia promote MDM invasion into the brain. It was reported that the elimination of microglia delayed recruitment of MDMs, while the amplification of chemokine signaling between microglia and MDMs exacerbated pathological changes in traumatic spinal cord injury.⁵⁰ Thus, microglia have an important role in the recruitment of MDMs.

The BBB regulates the exchange of nutrients and signaling molecules between the peripheral blood and brain, thereby maintaining the stability of the CNS microenvironment essential for normal neurological function.⁶¹ Inflammatory-activated microglia promote the invasion of circulating immune cells through the release of cytokines, while the upregulation of adhesion molecules on ECs promotes BBB damage and exacerbates neuroinflammation.^62,63 In addition, peripheral stimuli can influence BBB function directly by modulating ECs gene expression. In disease states, capillary ECs undergo more pronounced transcriptome alterations than other ECs, including the upregulation of genes related to intrinsic immunity and oxidative stress pathways.¹⁷ In the present study, we found that LPS treatment increased the proportion of capillary ECs (EC0 cells), a subpopulation that strongly expressed genes involved in the positive regulation of cell migration and cell-cell adhesion. These results suggest that LPS stimulation enhances the adhesion of peripheral immune cells to the brain capillary endothelium and thereby facilitates the invasion of peripheral immune cells. Many studies have demonstrated that loss of BBB integrity is associated with the downregulation of tight junction proteins such as CLDN5 and TJP1 in ECs.⁶⁴ In the current study, LPS increased the proportion of EC0 cells highly expressing tight junction protein in the hippocampus, which may be a compensatory response to repair BBB damage and prevent the excessive entry of leukocytes into the brain under inflammatory conditions. A similar alteration was found in the brain of a schizophrenia model.⁶⁵ Functional enrichment of EC0 cell marker genes also suggests involvement in vascular endothelial growth factor receptor signaling and blood vessel morphogenesis. During vascular inflammation or injury, ECs proliferate and migrate to form new vessels (angiogenesis).⁶⁶ We hypothesize that under stimulation by peripheral inflammatory mediators, capillary ECs proliferate and highly express cell adhesion and connexin-related genes to regulate the migration of immune cells into the central nervous system. We plan to further validate these findings in future animal experiments by investigating the temporal expression patterns of tight junction proteins in ECs following LPS treatment.

To further explore the functional implications of these LPS-induced changes in ECs gene expression, we focused on the NC versus LPS group DEGs Vcam1, Ddit4, Lrg1, and Vim. Among them, VCAM1 is a well described adhesion factor recruiting leukocytes into the CNS.^67,68 In addition, Ddit4 upregulation was detected in a mouse model of LPS-induced neuroinflammation and found to regulate vascular inflammation through the NF-κB pathway,⁶⁹ while another study found that Ddit4 is associated with apoptosis induced by oxidative stress in ECs.⁷⁰ Functional enrichment results suggested that these DEGs contribute to ribosome-related functions and activation of apoptotic pathways involving p53. Subpopulation-specific DEG analysis indicated that LPS specifically upregulated genes such as Ddit4 and its downstream effector Txnip in EC0 cells, implicating pathways related to hypoxia, mitochondrial dysfunction, and apoptosis. It has been demonstrated that apoptosis of human brain ECs can be suppressed either by directly silencing REDD1⁷¹ or by limiting mitochondrial ROS accumulation.⁷² Further, enhanced apoptosis of ECs directly affects BBB permeability and brain function.⁷³ The present study suggests that LPS stimulation can induce the proliferation of EC0 cells but also induces oxidative stress-related and apoptotic pathways that impair BBB function.

LPS enhanced Igf1–Igf1r pathway signaling between microglia and ECs. Previous studies have shown that IGF-1 protects CNS function by inducing angiogenesis and vascular remodeling in the hippocampus,⁷⁴ while therapies that increase IGF-1 levels during the recovery period were found to reduce infarct volume and apoptosis rates, and to improve cognitive deficits in animal models of stroke.⁷⁴ A subpopulation of microglia in the hippocampus of senescent AD model mice highly expressed Igf1 mRNA, and microglia were found to be a major source of IGF-1.⁷⁵ The expression of IGF-1R by ECs also enhanced regenerative repair of the endothelium. Microglia and macrophages were found to promote endogenous angiogenesis in animal studies of spinal cord injury by interacting and communicating with ECs through the IGF signaling pathway.⁷⁶ In this study, Igf1–Igf1r pathway signaling between microglia and ECs in the LPS group may induce angiogenesis in the hippocampus. Semaphorin3 (SEMA3) signaling was significantly stronger in ECs of the LPS group than in those of the NC group. SEMA3A may contribute to secondary neurological deficits, BBB damage, and associated leakage after traumatic brain injury by increasing VE-cadherin serine phosphorylation and weakening endothelial cell-cell adhesion.⁷⁷ The results of the present study suggest that LPS induces endothelial cell autocrine and microglial cell paracrine SEMA3 signaling, which acts on ECs to cause BBB damage. At the same time, ECs release Galectin-9 (encoded by Lagals9), which acts as a recruiting agent for further immune cell infiltration.

The disruption of BBB integrity by LPS is often accompanied by the secretion of inflammatory mediators such as IL-1β and TNF-α from ECs and activation of inflammatory vesicles, so neuroinflammation is closely associated with loss of BBB function.^78,79 It has also been reported that LPS stimulation damages the BBB by inducing oxidative stress,⁸⁰ an imbalance between the generation of prooxidants and cellular antioxidant capacity that contributes to cellular senescence, endothelial damage, and multiple diseases, including most neurodegenerative disorders.^81,82 Excessive ROS generation can damage DNA, disrupt mitochondrial metabolic function, and inhibit protein synthesis, leading to the activation of the apoptotic cascade.⁸³ In the current study, the stimulation of hCMEC/D3 cells with LPS increased ROS generation, depolarized the mitochondrial membrane potential necessary for oxidative phosphorylation, and reduced cellular energy charge (ATP concentration), whereas DDIT4/REDD1 knockdown reduced all of these deleterious effects. The depletion of REDD1 also delayed the production of ROS in human umbilical vein endothelial cells after exposure to LPS, accompanied by an increase in the expression of antioxidant enzymes.⁷¹ Previous studies have also found that REDD1 can decrease mitochondria-associated endoplasmic reticulum membrane integrity and the mitochondrial metabolic rate, which in turn reduces oxygen consumption and ATP synthesis, at least in part by impairing the mitochondrial electron transport chain.⁸⁴ In contrast, the drug thujaplicin can protect the barrier function of human umbilical vein ECs by ameliorating mitochondrial dysfunction.⁸⁵

LPS stimulation also promoted the release of intracellular LDH and the proportion of TUNEL-positive hCMEC/D3 cells, suggesting the promotion of both apoptosis and necrosis, while REDD1 knockdown reversed both effects. Previous studies have found that the overexpression of REDD1 promotes apoptosis in differentiated neuronal cells and increases sensitivity to hypoxia and oxidative stress.⁸⁶ Further, the inhibition of REDD1 was reported to attenuate cell damage, inflammatory responses, and apoptosis by decreasing autophagic pathways.⁸⁷ Animal studies have also demonstrated that autophagy-associated apoptosis induced by REDD1 is regulated by forming a complex with TXNIP.⁸⁸ Our scRNA-seq results indicate that Txnip, as well as Redd1, is differentially expressed on ECs following LPS treatment, strongly suggesting that the REDD1/TXNIP complex is critical for maintaining BBB function. Further studies will extend our investigation to the downstream target Txnip, aiming to validate the role of the REDD1/TXNIP complex in regulating BBB function in vascular endothelial cells.

In this study, scRNA-seq revealed complex changes in cellular gene expression profiles, cellular phenotypes, intracellular signaling pathways, and intercellular signaling pathways during the induction of neuroinflammation by the bacterial toxin LPS in aging rat hippocampus. LPS treatment increased the proportions of proinflammatory microglia and PVMs, reduced the proportions of macrophages involved in immune homeostasis, increased the abundance of peripheral monocyte-derived macrophages, and enhanced the expression of proapoptotic signaling genes such as Ddit4/Redd1 in cerebrovascular endothelial cells. Ligand–receptor pair analyses also suggested that signaling interactions among immune cells and vascular endothelial cells contribute to BBB damage and further recruitment, retention, and infiltration of immune cells, thereby spreading peripheral inflammation to the CNS. Further, REDD1 knockdown suppressed LPS-induced EC oxidative stress and apoptosis, suggesting that interventions that reduce REDD1 expression or function may preserve vascular endothelial barrier function, thereby preventing neuroinflammation and associated neurocognitive dysfunction.

Limitations of the study

Although scRNA-seq measurements of the cellular transcriptome can reveal detailed molecular changes associated with disease processes as well as complex changes in cell phenotype and intercellular signaling, it is limited for the detection of dynamic changes. Thus, further study is needed to reveal the temporal changes during the development of LPS-induced neuroinflammation. In addition, the scope of scRNA-seq analysis was limited to the hippocampal region and male subjects. To achieve a more comprehensive understanding of neuroinflammatory mechanisms, future research should extend the investigation to other brain regions, include both sexes, cover a broader age range, and explore various disease models. Another limitation of this study is the absence of in vivo validation using animal models. As this work represents an early-stage basic investigation, we focused primarily on exploring the underlying mechanisms. We acknowledge the importance of animal-based validation and plan to incorporate in vivo studies, including the use of aged animal models and perioperative human specimens, in future research to further substantiate our findings and clarify the role of REDD1. Furthermore, although astrocytes play a critical role in neuroinflammation, their representation in our single-cell dataset was limited, which may be attributed to current technical challenges in isolating astrocytes. Further studies will aim to optimize cell sorting methods and consider alternative animal models to enable more detailed investigation of astrocyte involvement.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Prof. Tianlong Wang (w_tl5595@hotmail.com).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •Data: The datasets generated during this study have been deposited in ArrayExpress and are publicly available under the accession number E-MTAB-15343.
• •Code: All computational code used in this study is based on publicly available R packages, which are listed in the key resources table.
• •Other information: Additional data or materials are available from the lead contact upon reasonable request.

Acknowledgments

This work was supported by Beijing Municipal Health Commission of China (No. Jing2019-2) and Post-subsidy funds for National Clinical Research Center, Ministry of Science and Technology (China) of China (No. 303-01-001-0272-03).

Author contributions

FYL and ZJL jointly completed the experiment of this study. FYL wrote the original draft. FYL, HQF, and ZJL processed the data. TLW, HQF and FYL designed the study. TLW and YW contributed to the project administration. All authors read and approved the final article.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
REDD1 specific antibody	Proteintech	Cat# 10638-1-AP; RRID: AB_2245711
CD31 antibody	Abcam	Cat# ab64543; RRID: AB_1141558
Donkey anti-Rabbit IgG (H+L) Highly Cross-Adsorbed Secondary Antibody, Alexa Fluor™ 594	Invitrogen	Cat# A-21207; RRID: AB_141637
Donkey anti-Mouse IgG (H+L) Highly Cross-Adsorbed Secondary Antibody, Alexa Fluor™ 488	Invitrogen	Cat# A-21202; RRID: AB_141607
Goat anti-Rat IgG (H+L) Cross-Adsorbed Secondary Antibody, Alexa Fluor™ 488	Invitrogen	Cat# A-11006, RRID:AB_2534074
Goat anti-Rabbit IgG-HRP Antibody	Absin Bioscience	Cat# abs20002A, RRID:AB_2716554
Biological samples
Rat hippocampus	This paper	N/A
Chemicals, peptides, and recombinant proteins
Bovine Serum Albumin	Sigma-Aldrich	Cat# A7030
Lipopolysaccharide	Sigma-Aldrich	Cat# 055:B5
TRIzol reagent	Ambion	Cat# 15596018
DHE	Thermo Scientific	Cat# D11347
DAPI	ZSGB-BIO	Cat# ZLI-9600
Critical commercial assays
Adult Brain Dissociation Kit	Miltenyi Biotec	Cat# 130-107-677
Lipofectamine™ 3000 Transfection Reagent	Invitrogen	Cat# L3000008
Transcriptor First Strand cDNA synthesis kit	Roche Diagnostics	Cat# 04897030001
Advanced™ Universal SYBR Green Supermix	Bio-Rad	Cat# 1725271
BCA Protein Assay Kit	Thermo Scientific	Cat# 23227
Enhanced ATP Assay Kit	Beyotime Biotechnology	Cat# S0027
TUNEL staining kit	Beyotime Biotechnology	Cat# C1090
LDH Detection Kit	Beyotime Biotechnology	Cat# C0016
Deposited data
Single-cell RNA-seq data	This paper	E-MTAB-15343
Experimental models: Cell lines
hCMEC/D3	Fuheng Biotechnology	Cat# FH1110
Oligonucleotides
Primer: Il1β forward (5′ to 3′): ATGAGAGCATCCAGCT TCAAATC	This paper	N/A
Primer: Il1β reverse (5′ to 3′): CACACTAGCAGGTCGT CATCATC	This paper	N/A
Primer: Tnf forward (5′ to 3′): CAAGAGCCCTTGCC CTAA	This paper	N/A
Primer: Tnf reverse (5′ to 3′): CAGAGCAATGACTCC AAAGTA	This paper	N/A
Primer: Gapdh forward (5′ to 3′): ACAGCAACAGGG TGGTGGAC	This paper	N/A
Primer: Gapdh reverse (5′ to 3′): TTTGAGGGTGCAG CGAACTT	This paper	N/A
Primer: REDD1 forward (5′ to 3′): GAGCCTGGAGA GCTCGGACT	This paper	N/A
Primer: REDD1 reverse (5′ to 3′): CTGCATCAGGTTG GCACACA	This paper	N/A
Primer: GAPDH forward (5′ to 3′): CGCCACAGTTTC CCGGAGGG	This paper	N/A
Primer: GAPDH reverse (5′ to 3′): CCCTCCAAAATCA AGTGGGG	This paper	N/A
Software and algorithms
R V4.3.	Cran, The R Foundation	https://cran.r-project.org/
Seurat (v4.1.0, R package)	Satija Lab	https://satijalab.org/seurat/; RRID:SCR_016341
Kyoto Encyclopedia of Genes and Genomes (KEGG)	KEGG/Kanehisa	https://www.genome.jp/kegg/; RRID:SCR_012773
CellRanger V3	10x Genomics	https://support.10xgenomics.com/single-cell-gene-expression/software/pipelines/latest/what-is-cell-ranger
CellChat (R package)	Github/Jin Lab	https://github.com/sqjin/CellChat
Metascape	Zhou Lab, Johns Hopkins University	https://metascape.org/gp/index.html#/main/step1; RRID:SCR_016620
GraphPad Prism 9	GraphPad Software, San Diego, CA, USA	http://www.graphpad.com/; RRID:SCR_002798

Experimental model and study participant details

Animals and neuroinflammatory model

Aged male Wistar rats (20 months old) were obtained from Dossy Laboratory Animal Company (Chengdu, China) and housed in a temperature and humidity controlled environment under a 12-h/12-h light/dark cycle. All animal experiments were approved by the Institutional Ethics Committee of Capital Medical University (EEI-02020-203) and complied with the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. Rats received intraperitoneal injection of LPS (2 mg/kg, Sigma-Aldrich, 055:B5)⁸⁹ and were sacrificed on day1, 3, and 7 after injection respectively, while negative control (NC) group rats received equal-volume saline injection.

Cell culture

The human cerebral microvascular endothelial cell line hCMEC/D3 (FH1110) was obtained from Shanghai Fuheng Biotechnology (Shanghai, China). Cells were maintained at 37°C in Endothelial Cell Media (#1,001) supplemented with 1% Endothelial Cell Growth Supplement (ECGS, #1,052), 5% fetal bovine serum (FBS, #0025), and 1% penicillin/streptomycin solution (P/S, #0503) under a humidified atmosphere of 5% CO₂ and 95% air. Based on preliminary concentration gradient tests, hCMEC/D3 cells were stimulated with 100 ng/mL LPS for 24 hours in complete medium to establish the experimental group. PBS-treated cells under identical conditions were used as the control group.

Method details

Morris water maze test

One day after treatment, rats in the LPS and NC groups (n = 8 per group) were examined for spatial learning and memory capacity in the Morris water maze. A detailed description of the training protocols can be found in our previous study.⁸⁹ Briefly, the test was conducted in a plastic circular pool measuring 150 cm in diameter and 60 cm in depth filled with opaque room temperature water. The tank walls displayed four distinct visual clues, dividing into four quadrants. Swimming speed, latency to find the platform, and the number of target quadrant/platform crossover were recorded by a computer tracking system (Ethovision, Noldus Information Technologies, Wageningen, The Netherlands). Rats were first trained over five consecutive days to find the escape platform located 1.5 cm below the water surface in one quadrant. Rats were released individually into the water from every quadrant and allowed to swim freely for 60 s to find the platform. Once found, the rat was allowed to remain for 5 s. Otherwise, it would be placed on it for 20 s. On day 7, a probe trial was conducted without platform, and both the time spent in the platform location quadrant and the number of platform crossings were recorded over 30 s following release from the most distal quadrant. The timeline for the MWM testing is shown in Figure 1A.

Tissue dissociation for 10x Genomics

Pair of hippocampi from LPS and NC group rats (n = 3 per group) were harvested on day 3 after LPS injection and pooled by group for scRNA-seq. Briefly, rats were anesthetized and perfused through the heart with ice-cold PBS. The hippocampus was immediately isolated and enzymatically dissociated using the Adult Brain Dissociation Kit according to the manufacturer’s instructions (Miltenyi Biotec, 130-107-677). After removal of debris and red blood cells, the suspended single cells were collected in Dulbecco’s phosphate buffered saline (DPBS) supplemented with 0.04% bovine serum albumin. Cells were stained with trypan blue and the proportion of viable cells (unstained) was determined using a Luna-FL™ automated cell counter (Logos Biosystems, Anyang, Kyonggi-do, South Korea). Samples with viable cell counts above 85% were loaded onto a 10x Genomics Chromium chip according to the manufacturer's instructions.

Single-cell RNA library preparation and sequencing

Libraries were prepared using the Chromium Single Cell 3' GEM, Library & Gel Bead Kit (10X Genomics) according to the manufacturer’s instructions. Briefly, a gel bead emulsion (GEM) was prepared by mixing the single cell suspension, gel beads, and oils via the 10x Genomics Chromium controller. In suspension, cells were fragmented and lysed, thereby releasing mRNA and initiating reverse transcription into cDNA in the presence of Poly-dT primer and reverse transcriptase. The cDNA was generated, amplified and assessed for quality using an Agilent Bioanalyzer 2,100, and sequenced using an Illumina NovaSeq 6,000 PE150 system.

Single-cell RNA sequencing data analysis

ScRNA-seqdata were further analyzed using 10x Cell Ranger (version2.2.0, https://support.10xgenomics.com/single-cell-gene-expression/software/pipelines/latest/what-is-cellranger), processed with unique molecular identifiers (UMI) tools, and aligned to the rat reference genome using STAR (Spliced Transcripts Alignment to a Reference). Data normalization, detailed analysis, and visualization were performed using the Seurat package. For initial quality control of the extracted gene-cell matrices, we filtered cells using Seurat parameters as below: i) nFeature_RNA >200 & nFeature_RNA <8,000 for number of genes per cell; ii)percentage of mitochondrial genes less than 20%; iii) minimum expressing cells = 3. Results were then normalized to total expression and log-transformed. Principal component analysis was conducted using the Seurat R Package to identify the first 20 principal components in the feature-barcode matrices for dimensionality reduction. Next, t-distributed stochastic neighbor embedding (t-SNE) and the Uniform Manifold Approximation and Projection (UMAP) algorithm were used for visualization of the reduced data in two-dimensional space and expression similarity analysis, respectively. The parameters and thresholds used in Seurat are detailed in online Table S1.

Differential expression analysis and cell type identification

The cellular identity of each cluster yielded by similarity analysis was determined by finding cluster-specific marker genes and differentially expressed genes (DEGs) using the Seurat FindAllMarkers function, with the minimum fraction of cells expressing the gene set to over 25%. A log2-fold change (Log2FC) in expression was used as an estimate of the log2 ratio of mean gene expression in a cluster to that in all other clusters and cells, based on a negative binomial test. The p-values of the output data reported here were adjusted for multiple comparisons using the Benjamini-Hochberg procedure. Cell type annotations were derived mainly from previous studies on marker genes and the Cell Marker database.

Gene enrichment and pathway analysis

Absolute log2FC > 0.25 and P < 0.05 with false discovery rate correction (P_FDR) were set as thresholds for significant DEGs. In this study, we conducted the following differential expression comparisons: i) DEGs between one cell type and all other cell types for cell-type specific marker genes; ii) DEGs between LPS and NC samples for identifying inflammation-associated changes at the cell-type level; iii) DEGs between one subcluster of a given cell type and the remaining subclusters of the same cell type for determining subcluster marker genes; iv) DEGs between LPS and NC samples in a given cell subcluster for determining subcluster-specific DEGs. All DEGs derived from these comparisons served as inputs to Metascape (https://metascape.org) or the cluster profiler package for gene enrichment analysis (Gene Ontology, KEGG Pathway), and to calculate accumulative hypergeometric P values and enrichment factors for filtering.

Prediction and quantification of cell–cell communications using CellChat

Cell–cell communication pathways were predicted and visualized using the R package CellChat. Specifically, we isolated microglial, CNS-associated macrophage (CAM), monocyte-derived macrophage (MDM), and ECs clusters from scRNA-seq datasets for each experimental condition. Each gene expression matrix and associated cell metadata set was used to create a CellChat object for both experimental conditions using the createCellChat function. We set the ligand–receptor interaction database as CellChat.DB and retained only signaling genes in each CellChat object to reduce computational load. Overexpressed genes and overexpressed ligand–receptor interactions were identified using the identifyOverExpressedGenes and identifyOverExpressedInteractions functions, respectively. Then, we calculated the cell–cell communication probability using the computeCommunProb function. Finally, we filtered out cell–cell communications if present in only a few cells per type (5 by default in CellChat). The parameters and thresholds used in CellChat are detailed in online Table S1.

Immunofluorescence and microscopy

Aging rats were anesthetized and perfused through the heart with PBS containing 4% paraformaldehyde. Brains were rapidly removed, postfixed in 4% paraformaldehyde, dehydrated in 30% sucrose, then rapidly frozen and stored at −80°C. Brain slices (20 μm) were prepared using a cryostat, washed, permeabilized, blocked, and incubated with the indicated primary antibodies at room temperature for 2 h and then at 4°C overnight. The next day, slices were washed and incubated with secondary antibody at room temperature for 2 h. Nuclei were then counterstained with DAPI. Images were obtained with a confocal laser scanning microscope (Leica TCS2, Leica Microsystems, Mannheim, Germany) and analyzed using Optimas 6.5 (CyberMetrics, Scottsdale, AZ, USA) by an independent observer. Fluorescence signals were quantified in ImageJ (NIH, USA) using the Color Threshold tool and analyzed via ROI Manager or Analyze Particles. All analyses were performed with consistent parameters and in a blinded manner.

Real-time quantitative PCR analysis

Total RNA was extracted from hippocampal tissues (n = 4/5 per group) and cells using TRIzol reagent (Ambion, Austin, TX, USA) according to the manufacturer’s instructions, assessed for purity using a NanoVue Plus Spectrophotometer (GE Healthcare, Little Chalfont, Buckinghamshire, UK), and reverse transcribed to cDNA using a Transcriptor First Strand cDNA synthesis kit (Roche Diagnostics, Meylan, France). Quantitative PCR was performed on a Bio-Rad CFX96 thermal cycler using Advanced™ Universal SYBR Green Supermix (Bio-Rad, Hercules, CA, USA) according to the manufacturer’ s protocol. The reaction conditions were 95°C for 30 s followed by 40 cycles of 10 s at 95°C and 30 s at 60°C. Target gene expression was normalized to that of Gapdh. The primer sequences used were as follows: 5′-ATGAGAGCATCCAGCTTCAAATC-3′ (sense) and 5′-CACACTAGCAGGTCGTCATCATC-3′ (antisense) for rat Il1β; 5′-CAAGAGCCCTTGCCCTAA-3′ (sense) and 5′-CAGAGCAATGACTCCAAAGTA-3′ (antisense) for rat Tnf; 5′-ACAGCAACAGGGTGGTGGAC-3′ (sense) and 5′-TTTGAGGGTGCAGCGAACTT-3′ (antisense) for rat Gapdh 5′- GAGCCTGGAGAGCTCGGACT-3′ (sense) and 5′- CTGCATCAGGTTGGCACACA-3′ (antisense) for human REDD1; 5′- CGCCACAGTTTCCCGGAGGG-3′ (sense) and 5′- CCCTCCAAAATCAAGTGGGG-3′ (antisense) for human GAPDH.

Western blotting

Cells were harvested, lysed in RIPA buffer supplemented with 1% protease inhibitors (Solarbio, Beijing, China), and centrifuged at 12,000 rpm for 15–30 min at 4°C. Supernatants were retained and protein concentrations determined using a BCA Protein Assay Kit (Thermo Fisher Scientific, Waltham, MA) according to the manufacturer's instructions. Equal amounts of protein were separated by SDS-polyacrylamide gel electrophoresis and transferred to polyvinylidene difluoride (PVDF) membranes. Membranes were blocked with 5% skim milk for at least 2 h at room temperature and incubated with the indicated primary antibodies overnight at 4°C. The next day, the membranes were incubated with horseradish peroxidase (HRP)-linked secondary antibody for 1 h at room temperature. Immunolabeling was visualized with an enhanced chemiluminescence detection reagent, and signals were quantified using ImageJ. Band density values were normalized to β-actin as the gel loading control.

Transfection with small interfering RNA

Cells were seeded in 24-well or 6-well plates, allowed to reach 60%–70% confluence, and then transfected with targeted short hairpin RNA (h-REDD1#1, sh-REDD1#2, sh-REDD1#3) or negative control (sh-NC) in OptiMEM (#31,985,062, Gibco) using Lipofectamine 3000 (#L3000008, Invitrogen) according to the manufacturer's protocol. After transfection, the cells were cultured with nonpenicillin or streptomycin in low FBS culture medium for 8 h. Related genes or proteins were detected by PCR, immunostaining, or Western blotting 48 h after transfection. The sequences of the shRNA and negative control are shown in online Table S2. These sequences were synthesized by Genechem (Shanghai, China).

Detection of intracellular reactive oxygen species (ROS)

Intracellular reactive oxygen species (ROS) accumulation in hCMEC/D3 cells was measured using the fluorescent probe dihydroethidium (DHE, D11347, Thermo Scientific). Briefly, cells were incubated with 5 μM DHE at 37°C for 30 min under darkness, then counterstained with DAPI. Specifically, fluorescence was observed under a fluorescence microscope (excitation at 518 nm, emission at 606 nm), and quantitative analysis was performed using ImageJ software. Co-localization in cultured cells was assessed by intensity correlation analysis with PDM (product of the differences from the mean) image generation.

Detection of mitochondrial membrane potential

To measure mitochondrial membrane potential, hCMEC/D3 cells were incubated with 200 nM tetramethylrhodamine, methyl ester (TMRM) (#T668, Invitrogen) at 37°C for 30 min. Fluorescence was measured using excitation at 550 nm and emission at 576 nm. Co-localization analysis was conducted using intensity correlation analysis with PDM image generation.

Measurements of cellular ATP levels

Cellular ATP levels were measured using the Enhanced ATP Assay Kit (Beyotime Biotechnology, S0027) according to the manufacturer’s protocol. Briefly, cells were lysed in ATP lysis buffer. Total lysate protein concentration was measured by BCA Protein Assay Kit (Thermo Fisher Scientific, Waltham, MA), and ATP content determined using a luminometer (SoftMax Pro6, Molecular Devices, San Jose, USA). Cellular ATP levels were normalized to total protein content (nmol/g protein).

Terminal deoxynucleotidyl transferase dUTP nick-end labeling (TUNEL) assay for apoptosis

After the indicated treatment, hCMEC/D3 cells were washed, fixed with 4% paraformaldehyde in PBS for 30 min at 37°C, permeabilized with 0.3% Triton X-100 in PBS for 5 min, and stained using a TUNEL staining kit according to the manufacturer’s instructions (C1090, Beyotime Biotechnology, China). TUNEL-positive cells were counted using ImageJ by setting a fluorescence intensity threshold. The percentage of TUNEL-positive cells was calculated as the number of TUNEL-positive nuclei divided by the total number of DAPI-stained nuclei in the same field.

Lactate dehydrogenase (LDH) release assay

Cytotoxicity was assessed by measuring LDH release into the culture medium using a commercial LDH Detection Kit (C0016, Beyotime Biotechnology, China) according to the manufacturer's instructions. In brief, hCMEC/D3 cells were seeded in 24-well plates and cultured in low FBS medium overnight. After transfection and treatment, the plates were centrifuged at 400g for 5 min. Then, 60 μL cell-free supernatant was incubated with 30 μL LDH substrate solution for 30 min, and absorbance read at 490 nm on a microplate spectrophotometer. The LDH release activity is expressed as fold-increase over the control group.

Quantification and statistical analysis

All data are presented as the mean ± standard error of the mean. Cell experiments were performed at least three times. Data analysis were performed blind to the conditions of the experiments. Group means were compared by independent sample t-test or one-way analysis of variance followed by Tukey’s or Dunnett’s post hoc test for pair-wise comparisons as indicated. All statistical calculations were performed using R Studio (2022.02.1 Build 461) and GraphPad Prism version 9 (GraphPad Software, San Diego, CA, USA). A P < 0.05 was considered statistically significant for all tests.

Published: August 11, 2025