Coordinated response of vascular and glial cells in a pig model of sepsis

Abstract

Sepsis is a deadly syndrome arising from overactivation of the immune system in response to a pathogen. Severe sepsis can result in sepsis-associated encephalopathy (SAE) that significantly contributes to morbidity and long-term cognitive impairment. Studying the molecular underpinnings of SAE in the human brain is challenging, and rodent models have limited translational applicability due to significant neuroanatomical and immune differences. To bridge this gap, we employed a Yorkshire pig model to investigate changes in the brain cortex during the early stages of SAE using a uropathogenic strain of E. coli to model bacterial sepsis. Bulk RNAseq of cortical brain demonstrated profound global shifts in immune and inflammatory pathways, cytokine signaling, vascular dysfunction, and homeostatic dysregulation. To gain a more granular understanding of the changes in this model, we performed single-nucleus RNA sequencing (snRNAseq) on the same cortical brain tissue. This snRNAseq analysis revealed dramatic alterations in vascular cell populations, accompanied by concomitant glial cell activation. Surprisingly minimal changes were noted in neurons, likely due to the very early nature of this insult. The neurovascular unit, including vascular leptomeningeal cells (VLMCs), mural cells, endothelial cells, astrocytes, and microglia, showed strong upregulation of genes involved in TNF signaling, cellular responses to cytokine stimuli, and cytokine signaling in the immune system. Conversely, downregulated genes were enriched in pathways related to the disruption of metabolic and structural functions, including genes involved in maintaining the blood-brain barrier (BBB). Cell-cell communication analysis also identified significant rewiring of intercellular signaling in the septic brain, with an overall increase in the number and strength of cell to cell interactions in the E. coli pigs compared to saline controls. In saline controls, cell-cell communication was dominated by homeostatic signaling from microglia and astrocytes. In contrast, the septic pigs showed a significant shift toward inflammatory and immune-related signaling pathways, mainly involving microglia and astrocytes, along with the emergence of signaling pathways from mural cells and VLMCs. Our findings reveal that systemic sepsis triggers a rapid shift from homeostatic to inflammatory signaling within the neurovascular unit, driven by the emergence of pro-inflammatory crosstalk between vascular cells, astrocytes, and microglia. This porcine model effectively captures the acute transcriptional landscape of the septic brain, offering a critical window for therapeutic strategies aimed at stabilizing the BBB and preventing long-term cognitive decline in sepsis survivors.

Introduction

Sepsis is a life-threatening, dysfunctional response to infection, most often triggered by bacteria but can also arise from viral or fungal sources [1, 2]. Sepsis is characterized by a dysregulated and complex host inflammatory response that encompasses both pro-inflammatory and compensatory anti-inflammatory processes, ultimately leading to tissue damage and systemic organ dysfunction [3–6]. Sepsis affects an estimated 49 million people globally each year and is a leading cause of critical illness, hospitalization, and death [7, 8]. A common and severe complication of sepsis is sepsis-associated encephalopathy (SAE), manifesting as cognitive decline, inattention, mood alterations, and delirium – often resulting in lasting impairments [9]. It’s essential to note that SAE may occur without direct infection of the central nervous system (CNS) and is an early feature of infection that may appear before other systemic features of sepsis are obvious [9]. It is also important to note that sepsis survivors, even those who do not experience SAE, commonly experience lasting “brain fog” and are at high risk for cognitive impairments, mood changes, and symptoms that can resemble post-traumatic stress disorder [10, 11]. Despite extensive research into the acute inflammatory cascade of sepsis, the precise mechanisms associated with SAE remain underinvestigated. Recent single-nucleus transcriptomic analysis of the human hippocampus by Liu et al. (2025) has begun to bridge this gap, revealing that sepsis induces a coordinated neurovascular and glial response characterized by astrocyte-microglial crosstalk and blood-brain barrier (BBB) dysfunction [12]. However, given that the brain is an early target of septic injury [9, 13], the foundational molecular events leading to chronic cognitive and behavioral dysfunction may precede clinical sepsis diagnosis. Understanding these early molecular changes in the brain is therefore critical for developing effective interventions.

Currently, there are no definitive diagnostic markers for early sepsis or for predicting susceptibility to post-sepsis syndromes such as cognitive decline. Rodent models have provided fundamental insights into the pathophysiology and transcriptomic landscape of sepsis in a controlled experimental framework that human studies lack [14–16]. However, rodent models have significant neuroanatomical and physiological differences from humans, which limit the translation of findings, particularly regarding complex transcriptional changes within a larger brain containing substantially more white matter. Thus, we utilized a Yorkshire pig model, which allows precise control of infection, antibiotic-free observation, and quantification of the hyper-acute molecular landscape that immediately follows the septic insult. Additionally, our prior work comparing the septic porcine and murine brain transcriptomes demonstrated that while shared pathways of dysregulation exist, there are notable differences in the magnitude and pathways of the molecular response between these species (Olney et al. 2024; PMC10765757) [17]. Pigs are a more physiologically relevant model for studying human sepsis than rodents [17–19]; they offer a comparable immune system, similar renal physiology, and a large gyrencephalic brain that closely resembles the human central nervous system (CNS) [20, 21]. While some studies have examined endotoxemia in large animal models [22–24], none, to our knowledge, have specifically investigated the molecular effects of true bacterial sepsis on the brain in a large animal. Our previous work [17], using a porcine model of lipopolysaccharide (LPS)-induced endotoxemia, showed significant gene expression changes in the brain; however, LPS only partially mimics the clinical sepsis caused by pathogenic organisms. To address this critical translational gap, we used a Yorkshire pig model treated with Escherichia coli (E. coli).

In this study, the pig brain cortex was interrogated using a multimodal approach, combining bulk tissue RNA sequencing (RNAseq) and single-nucleus RNA sequencing (snRNAseq) to map the transcriptomic response during hyper-acute stages of septic response. This multimodal strategy enables cross-referencing global gene expression shifts with high-resolution cell-type-specific alterations. To further strengthen the translatability of this model, our analysis utilized Ingenuity Pathway Analysis (IPA) and CellChat2 to identify disrupted signaling axes and key ligand-receptor (L-R) pairs involved in neurovascular communication. Notably, our findings reveal molecular signatures, specifically regarding glial activation and vascular dysfunction, that converge with recent single-nucleus observations in the human hippocampus septic brain reported by Liu et al. 2025; PMC12477015 [12]. Our work bridges the gap between clinical human brain sepsis data and potential therapeutic strategies to prevent long-term cognitive decline in sepsis survivors by identifying early molecular drivers of SAE in a controlled porcine model.

Results

Clinical and pathological signs of systemic inflammation

Our model of sepsis utilized outbred Yorkshire female swine, continuously infused intravenously with either Escherichia coli (E. coli, n = 4) or saline (control, n = 4). The E. coli infusion targeted a dose of 1.2 ×10^9 colony forming units (CFU)/Kg over a maximum of 6 h, while control pigs received an equivalent volume of 0.9% normal saline (1.00 mL/h). Time to euthanasia varied somewhat among E. coli pigs (199 – 362 min) due to individual progression of septic shock (Supplemental Table 1). In contrast, all saline controls were euthanized at 300 min, showing no signs of organ dysfunction. Clinical monitoring was performed via surgically placed cannulas in the jugular vein and carotid artery, and supportive care was administered as needed. Analysis of physiological parameters revealed differences between E. coli and saline pigs (Supplemental Fig. 1). At euthanasia, E. coli pigs exhibited elevated heart rates compared to saline controls (p = 0.01), with values exceeding 150 beats per minute (Supplemental Fig. 1A). There was no significant difference in the starting body temperature, though the final body temperature was marginally elevated by 6.3°F (p = 0.09) between groups (Supplemental Fig. 1B). Peripheral blood oxygen saturation (SpO2) values at the start of the experiment were comparable between the treatment groups, but the final SpO2 measurements were lower in E. coli pigs (p = 0.04), typically below 98%, compared to saline controls, which maintained SpO2 above 98% (Supplemental Fig. 1C). Mean arterial pressure (MAP) showed no significant difference between treatment groups at the start of the experiment or final time points (Supplemental Fig. 1D). Collectively, these changes are indicative of an early septic response, and importantly, animals received continuous infusions of phenylephrine and norepinephrine at increasing doses to maintain MAP at 60 mmHg and prevent hypotension. To better understand the underlying molecular mechanisms driving these physiological changes, we next performed bulk RNA sequencing to characterize the global transcriptional response to septic shock within the brain.

Global increased inflammatory dysregulation in the brain following E. coli infusion

To explore the global transcriptional changes in the brain during sepsis, we performed bulk tissue RNAseq on cortical samples from pigs treated with E. coli (n = 4) versus saline controls (n = 4). Principal Component Analysis (PCA) of the bulk RNAseq data revealed that the primary source of transcriptional variation was driven by the experimental condition (Supplemental Fig. 2A).To identify potential confounding variables, we assessed the correlation between principal components and clinical parameters. PC1 was highly correlated with the clinical state, specifically the final end heart rate and end temperature (Supplemental Fig. 2B). Variance partition analysis confirmed that these clinical markers explain a significant portion of the total variance, and their effects are collinear with the treatment condition (Supplemental Fig. 2B). Given individual differences in the progression of septic shock, the duration of infection measured in minutes from infusion to euthanasia in the E. coli group varied (199–362 min). We used a parsimonious model to compare the E. coli and saline groups, with each coefficient corresponding to a group mean, and included minutes from injection to euthanasia (duration_min) as a covariate to account for temporal differences in disease progression. Differential gene expression analysis between the E. coli and saline groups revealed robust transcriptional dysregulation (q < 0.05 & |log2FC | > 1), with 9 downregulated and 264 upregulated genes (Fig. 1A and Supplemental Table 2). Notably, highly upregulated genes included CCL2, CXCL10, and RSAD2, which are known to be involved in inflammatory response. Conversely, OCLN, SLC52 A3, and SLCA38A5 were among the most significantly downregulated genes. The OCLN gene encodes the protein occludin, an integral component of tight junctions in the endothelial cells that form the blood brain barrier (BBB). During sepsis, a dysregulated and exaggerated inflammatory response may cause systemic BBB damage, allowing inflammatory mediators, pathogens, and immune cells from the periphery to enter the brain [25, 26]. This infiltration may trigger neuroinflammation, which can result in brain dysfunction, delirium, and cognitive impairment [25, 26]. The downregulation of SLC52A3 and SLC38A5 also indicates localized immunometabolic reprogramming. SLC52A3 is crucial for transporting riboflavin, a precursor of Flavin Adenine Dinucleotide (FAD), a vital cofactor for oxidative metabolism and the mitochondrial electron transport chain [27]. At the same time, decreased expression of SLC38A5, a transporter that transports glutamine, suggests inhibition of the TCA cycle, reduced oxidative phosphorylation, and possibly lower glutathione production [28]. These findings imply that, within the septic porcine brain, there is a gene-signaling pattern associated with decreased energy use. This response may be a neuroprotective mechanism aimed at maintaining brain integrity during systemic inflammation, even as it reflects broader dysregulation of the inflammatory response. Gene enrichment analysis of the downregulated genes were associated with pathways such as vitamin transport and organic anion transport (Supplmental Table 3), likely due to impairments in the vascular cells that are known to express these transporters. In contrast, gene enrichment analysis among all the upregulated genes showed significant enrichment of several inflammatory response pathways, including cytokine signaling within the immune system and the TNF signaling pathway, as well as pathways involved in positive regulation of programmed cell death, cell migration, vasculature development, and blood vessel formation, indicating a widespread cellular response (Fig. 1B and Supplemental Table 2 & 3). Collectively, the bulk RNAseq data provide evidence that early sepsis induces an inflammatory response in the brain, characterized by the upregulation of key cytokine and chemokine genes and a concurrent disruption of barrier integrity, suggesting a link between systemic inflammation and cerebral dysfunction. While the bulk tissue analyses provide a global response, it does not resolve which specific cell types are transcriptionally altered due to the septic insult; thus, we next explored the cell type specific transcriptional response in the septic porcine brain.

Fig. 1. Global transcriptional alterations in the septic pig brain reveal robust upregulation of proinflammatory response and downregulation of blood brain barrier maintenance.

A Volcano plot of differentially expressed genes for E. coli (n = 4) versus saline control (n = 4) pigs. Genes that are differentially expressed, with a q < 0.05, are indicated in blue for downregulated (log₂ fold change < -1) and red for upregulated (log₂ fold change > 1). Genes that are not differentially expressed, q > 0.05, are shown in gray. B Gene ontology (GO) analysis shows enriched upregulated and downregulated pathways. The x-axis is the gene count contributing to the enrichment pathways listed on the y-axis. The color of the bar indicates the − log₁₀ p-value.

Global transcriptional alterations are driven by vascular and glial responses in the acute septic porcine brain

To evaluate the cell-specific molecular landscape of the porcine brain during acute sepsis, we performed single-nucleus RNA sequencing (snRNAseq) on cortical samples from E. coli and saline control pigs. Unsupervised clustering of the snRNAseq data identified eight distinct cellular populations, which were annotated using canonical marker genes (Fig. 2A, B). Differential expression analysis (q < 0.05 and |log2FC | > 1) revealed that the septic insult triggers a robust, cell-type-specific transcriptional response. The endothelial cluster exhibited the highest degree of dysregulation with 2369 differentially expressed genes (1003 downregulated and 1366 upregulated), followed by microglia (1478 DEGs), mural cells (966 DEGs), and vascular leptomeningeal cells (398 DEGs) (Fig. 2C). In contrast, astrocytes (227 DEGs), oligodendrocytes (250 DEGs), oligodendrocyte progenitor cells (41 DEGs), and neurons (0 DEGs) showed comparatively fewer or no transcriptional alterations at this early time point (Fig. 2C and Supplemental Table 4).To determine the drivers of the global septic signature, we assessed the overlap between cell-type-specific DEGs and the bulk tissue RNAseq signal (Fig. 2D and Supplemental Table 5). Our analysis demonstrated that the transcriptomic profile of the bulk tissue is predominantly shaped by the vascular and innate immune compartments. Specifically, the endothelial response captured most of the bulk DEG signal (120 shared DEGs), with contributions also observed from mural cells (73 shared DEGs) and microglia (23 shared DEGs) (Fig. 2D, E). These findings suggest that the vascular unit and microglia serve as the primary early responders to systemic sepsis within the brain parenchyma, driving the initial stages of neuroinflammatory and cerebrovascular dysfunction.

Fig. 2. Single-nucleus RNA sequencing identifies distinct cell populations in the porcine brain.

A Uniform Manifold Approximation and Projection (UMAP) plot showing the major cell clusters in porcine brain tissue within the saline and E. coli pigs. Each point represents a single nucleus, colored by its assigned cell type. Light gray represents nuclei that are within one treatment group but not the other. B Dot plot illustrating the expression of canonical marker genes used to annotate the identified cell type clusters. The color intensity of each dot reflects the average expression level of the gene within that cluster, while the size indicates the percentage of cells within the cluster expressing that gene. C Number of differentially expressed genes (q < 0.05 & absolute log₂ fold change > 1) in each cell type. Blue indicates downregulation (log₂ fold change < -1), and red indicates upregulation (log₂ fold change > 1). D Venn diagram showing the overlap of differentially expressed genes (DEGs) identified through single-nucleus RNA sequencing (snRNA, purple) and bulk tissue RNA sequencing (Bulk, blue). Many of the genes identified in the bulk analysis (n = 261; 84%) are also differentially expressed within the snRNAseq dataset. E Contribution of cell type specific DEGs to the overall bulk DEG signal. The bar graph displays the total number of DEGs from each identified cell cluster that overlap with the bulk tissue DEG list. The bulk DEG signal is largely captured by the endothelial and mural DEGs, indicating these cell types are the main contributors to the global transcriptomic shift observed during the hyper-acute stage of the septic response. The not recovered are the DEGs that are differentially expressed in the bulk signal but are not differentially expressed in the snRNA data or show the opposite expression pattern. One gene, CMAH, is upregulated in the bulk but is downregulated in snRNAseq data.

Immune and vascular permeability pathways are dysregulated in endothelial cells following E. coli challenge

We first focused on endothelial cells as these are the “first responders” to the damage and inflammatory signals present in peripheral blood [29]. We identified substantial differential gene expression in endothelial cells in response to a septic insult (Fig. 3A). Specifically, 1003 genes were downregulated, while 1366 genes were upregulated in the E. coli compared to saline controls (q < 0.05 and |log2FC | > 1). Metascape gene set enrichment analysis revealed a prominent enrichment of upregulated genes broadly linked to immune and inflammatory responses (Fig. 3B and Supplemental Tables 4 & 6). Notably, pathways related to cytokine signaling in the immune system, positive regulation of programmed cell death, response to virus, cellular response to cytokine stimulus, positive regulation of cytokine production, and the TNF signaling pathway were among the most significantly upregulated pathways from this enrichment analysis. Conversely, processes such as tRNA metabolic process, lipid modification, DNA metabolic process, and lipid metabolism showed a general trend toward downregulation. The downregulation of lipid-processing transcripts may indicate a shift away from fatty acid oxidation toward glycolysis [30]. Ingenuity Pathway Analysis (IPA) was used to further identify the biological pathways most significantly impacted by these transcriptional changes in endothelial cells (Fig. 3C). The IPA network for endothelial differentially expressed genes (DEGs) highlighted several key molecules and their relationships. Specifically, predicted activation was observed for pathways related to leukocyte viability, antiviral response, and inflammatory signaling, involving IFNG, IL1B, and TNF. VEGFA also emerged as a central node, consistent with its role in vascular permeability and angiogenesis, which can be disrupted during sepsis [31, 32]. The network further suggested predicted activation of the I-kappaB kinase/NF-kappaB cascade, a crucial regulator of immune responses. Conversely, predicted inhibition was seen for processes such as vascular endothelial cell movement and blood cell viability. These findings underscore a significant alteration in endothelial cell function during early sepsis, shifting toward pro-inflammatory and immune-modulatory responses while potentially disrupting normal vascular processes.

Fig. 3. Endothelial transcriptional dysregulation and pathway analysis.

A Differential gene expression within the endothelial cell type cluster. Volcano plot showing significantly altered genes in endothelial cells from E. coli versus saline control pigs. Downregulated genes (log₂ fold change(FC) < -1) in blue (1003 genes) and upregulated genes (log₂(FC) > 1) are shown in red (1366 genes), all with q-value < 0.05. B Metascape gene set enrichment bar graph showing pathways associated with the enriched gene sets. Upregulated pathways include cytokine signaling and VEGFA VEGFR2 signaling, while downregulated pathways include lipid and DNA metabolic processes. The bar length indicates the gene count, and the color intensity reflects the statistical significance. C Ingenuity Pathway Analysis (IPA) of the endothelial differentially expressed genes reveals canonical pathways and upstream regulators. Orange nodes indicate predicted activation, and blue nodes indicate predicted inhibition. Lines depict relationships: solid for direct, dashed for indirect, and arrows for activation (orange) or inhibition (blue). Notable elements include IL6, IRF7, TNF, and pathways related to antiviral response and inflammation.

Dysregulated genes within mural cells are enriched in the interferon alpha/beta signaling pathway

Differential expression analysis between E. coli and saline control pigs within the mural cell cluster revealed significant transcriptional changes. As depicted in Fig. 4A, a total of 966 genes were differentially expressed (q < 0.05 and |log2FC | > 1) in mural cells from septic animals compared to controls (Supplemental Table 4). Specifically, 258 genes were downregulated, and 708 genes were upregulated, indicating a robust transcriptional response in mural cells during early sepsis. Among the most significantly upregulated pathways were those related to immune system activation and inflammation, including cytokine signaling in the immune system, innate immune response, VEGFA-VEGFR2 signaling, cellular response to cytokine stimulus, and TNF signaling pathway (Fig. 4B and Supplemental Table 6). Pathways associated with cellular movement and migration, such as positive regulation of cell migration and integrin cell surface interactions, were also significantly enriched. Conversely, pathways related to metabolic processes, such as aldosterone-regulated sodium reabsorption and intracellular monoatomic cation homeostasis, showed downregulation, again highlighting the transporter functions of these vascular cells. The IPA network analysis of DEGs in mural cells highlights interactions among CD40, DOCK8, and IL6, indicating their roles in enhancing immune responses, including leukocyte activation and the maturation of antigen-presenting cells (Fig. 4C). Key cytokines such as IFNG, IL1A, and IL1B are central to immune regulation within this network. Furthermore, the hepatic fibrosis signaling pathway is significantly influenced by CD40, FGF2, and RAC1, suggesting a strong theme of tissue remodeling and fibrotic processes within mural cells, with the involvement of TGFB3 and VEGFA further supporting this role. Genes like CD40, DOCK8, FGF2, and IL1A are linked to increased cell movement and migration, which is central for processes such as immune cell trafficking and tissue repair. The network also demonstrates a robust influence of cytokines, including IFNG, IL1A, and TNF, on pathways such as IL6 signaling and TLR4 activation, which are central to the inflammatory responses and host defense mechanisms observed in sepsis. Finally, IRF7 and RIGI emerge as key players in enhancing antimicrobial and antiviral responses, indicated by their connections to IL6 and TLR4, underscoring the network’s involvement in host defense against infections. These findings suggest that brain mural cells undergo significant transcriptional reprogramming during the early stages of E. coli-induced sepsis, contributing to the neuroinflammatory response and potentially influencing brain pathology through altered immune activation, tissue remodeling, and cellular migration.

Fig. 4. Transcriptional alterations and pathway analysis in mural cells from a porcine sepsis model reveal inflammatory activation and disrupted vascular pathways.

A Volcano plot showing significantly altered genes in endothelial cells from E. coli versus saline control pigs. Downregulated genes (log₂(FC) < -1) in blue (258 genes) and upregulated genes (log₂(FC) > 1) are shown in red (708 genes), all with q-value < 0.05. B Metascape gene set enrichment bar graph showing pathways associated with the enriched gene sets. Upregulated pathways include cytokine signaling in the immune system and the TNF signaling pathway, while downregulated pathways include vascular transport and metal ion transport. The length of each bar corresponds to the number of genes within the pathway, and the color intensity indicates the statistical significance (-log10(p-value)). C A subset of the IPA network of the mural differentially expressed genes shows predicted activation (orange) and inhibition (blue) of key molecules and pathways. Notable activated components include TNF, IFNG, IL1A, and IL6, with a predicted activation of Interferon alpha/beta signaling, antimicrobial response, and immune response of leukocytes. Solid lines represent direct interactions, while dashed lines represent indirect interactions.

Microglia in septic pig brains reveal pro-inflammatory enrichment and metabolic dysregulation

Microglia act as the resident immune cells of the brain and are pivotal in the neuroinflammatory response [33]. Differential expression analysis comparing microglia from septic pigs to controls identified a robust transcriptional shift, with 577 genes downregulated and 901 genes upregulated (q < 0.05 and |log2(FC)∣ > 1) (Fig. 5A and Supplemental Table 4). Enrichment analysis revealed significant upregulation of pathways related to cellular responses to stimuli, cytokine signaling in the immune system, and regulation of chemotaxis. These findings underscore a strong pro-inflammatory and immune-activated state in microglia during early sepsis. Conversely, pathways related to homeostatic functions such as protein processing in the endoplasmic reticulum, nervous system development, and fatty acid metabolic process were significantly downregulated (Fig. 5B and Supplemental Table 6). We further leveraged IPA to construct a network of interacting genes and their predicted effects on cellular functions (Fig. 5C). This analysis highlighted a central role for key inflammatory mediators, including TNF, IL6, and IL1B, which are connected to a broad range of predicted activated functions. Specifically, the network predicts activation of pathways involved in cell proliferation of T lymphocytes, proliferation of immune cells, and hematopoiesis of mononuclear leukocytes. The network also illustrates a link between these inflammatory signals and metabolic dysregulation, with a predicted activation of lipid synthesis and a connection to the autophagy pathway. The transcription factors EGR1 and BHLHE40 emerge as central nodes, suggesting their critical role in mediating the transcriptional response. These results demonstrate a complex and coordinated transcriptional program in microglia during early sepsis, which not only drives a pro-inflammatory state but also disrupts key metabolic and developmental processes.

Fig. 5. Disrupted nervous system pathways are observed in microglia from E. coli pigs compared with saline.

A Volcano plot of differentially expressed genes (DEGs) in microglia from E. coli versus saline control pigs. Downregulated genes are shown in blue (577 genes) and upregulated genes in red (901 genes), with significance thresholds set at a false discovery rate (q) < 0.05 and |log2(FC)| > 1. B Top canonical pathways and upstream regulators identified by Metascape of the microglial DEGs. The bar graph displays pathways that are either upregulated (top half) or downregulated (bottom half) in the septic brain. Upregulated pathways include Cytokine Signaling in the Immune system and Nervous system development, while downregulated pathways include fatty acid metabolic process and Neutrophil degranulation. The length of each bar corresponds to the number of genes within the pathway, and the color intensity indicates the statistical significance (-log10(p-value)). C A subset of the IPA network derived from the differentially expressed genes in microglia, showing predicted activation (orange) and inhibition (blue) of key molecular pathways. Notable activated components include TNF, IFNG, IL1A, and IL6, with a predicted activation of cell proliferation of T lymphocytes, hematopoiesis of mononuclear leukocytes, and proliferation of immune cells. Solid lines represent direct interactions, while dashed lines represent indirect interactions.

Early septic response is primarily confined to glial and vascular cells within the brain

Our snRNAseq has shown that the acute septic response to E. coli causes specific transcriptional changes in different brain cell populations. We observed significant alterations in gene expression in endothelial cells (Fig. 3), mural cells (Fig. 4), and microglia (Fig. 5). Astrocytes, vascular leptomeningeal cells (VLMCs), oligodendrocytes, and oligodendrocyte precursor cells (OPCs) also reveal significant transcriptional alterations in E. coli pigs compared to saline controls (Fig. 2 and Supplemental Fig. 3 & 4). In astrocytes, 49 genes were significantly downregulated, and 178 were upregulated following E. coli exposure (Supplemental Fig. 4A), with a strong enrichment of upregulated pathways related to intercellular communication and fundamental cellular functions (Supplemental Fig. 4B). Key upregulated pathways included interferon type I signaling, cytoplasmic ribosomal proteins, the TRBP-containing complex, and lipid transport. Downregulated pathways included monoatomic anion transport, transport of small molecules, and lipid transport, suggesting a significant disruption of astrocytic homeostatic and supportive functions. Collectively, these findings point to a shift in astrocytic function from a supportive, homeostatic role to a reactive, inflammatory state.

Our analysis of the transcriptional profile of vascular leptomeningeal cells (VLMCs) highlights their central role in mediating vascular disruption during the early stages of sepsis. VLMCs exhibited a strong inflammatory transcriptional signature, with 163 downregulated and 235 upregulated genes (Supplemental Fig. 3C). This inflammatory signature, characterized by the upregulation of pathways such as cytokine signaling in the immune system and the TNF signaling pathway, suggests that VLMCs actively respond to the systemic inflammatory state (Supplemental Fig. 4D). Additionally, the downregulation of pathways involved in cellular homeostasis and ECM organization provides further evidence of a breakdown in these fundamental functions. Disruption of the ECM can weaken the integrity of the blood-brain barrier (BBB), which is vital for protecting the brain from systemic inflammation and invading pathogens. The combined downregulation of structural and homeostatic pathways with heightened inflammatory activation indicates that VLMCs may play a role in increasing vascular permeability and facilitating immune cell migration into the brain parenchyma—hallmarks of septic encephalopathy [34]. While VLMCs show a strong inflammatory response that likely contributes to vascular disruption, our analysis reveals that this neuroinflammatory cascade is not isolated to VLMCs.

Our findings demonstrate that both oligodendrocytes and oligodendrocyte precursor cells (OPCs) undergo transcriptional changes in response to systemic sepsis, indicating that these glial cells are integral components of the central nervous system’s early inflammatory and stress response. These changes, particularly the shift away from normal maintenance and regenerative functions, likely contribute to the neurological dysfunction observed in SAE. Oligodendrocytes showed a significant transcriptional response to sepsis, with 23 genes downregulated and 227 upregulated (Supplemental Fig. 5A). Upregulated pathways reflected a generalized cellular stress and metabolic response, including cellular responses to stimuli, autophagosome organization, and photodynamic unfolded protein response (Supplemental Fig. 4B). OPCs also exhibited a transcriptional response, with 36 genes upregulated and 5 downregulated (Supplemental Fig. 4C). Upregulated pathways in OPCs included protein processing in the endoplasmic reticulum, cellular responses to stress, and regulation of T cell activation (Supplemental Fig. 4D). An interesting finding in OPCs is the downregulation of the cell-cell adhesion pathway. This pathway, which is essential for communication between OPCs and neurons and for their successful integration into neural circuits, is necessary before OPCs differentiate into mature oligodendrocytes. Together, these findings demonstrate that sepsis diverts these glial cells from their typical homeostatic and regenerative roles toward a maladaptive stress and inflammatory response.

Despite the extensive neuroinflammatory response observed in glial and vascular cells, a striking finding from our snRNAseq analysis was the lack of significant transcriptional changes in the neuronal populations (Supplemental Fig. 5E). This suggests a selective vulnerability and a hierarchical response within the brain during the early stages of sepsis. The observed neuronal dysfunction in septic encephalopathy is likely to be a secondary consequence of the widespread vascular- and glial-mediated inflammation and disruption of the neurovascular unit [35]. These observations are consistent with a recent human hippocampus sepsis single-nucleus study by Liu et al. 2025, which reported a disrupted neurovascular unit driven by dysfunctional astrocyte-microglia crosstalk and pro-inflammatory signaling [12].

Multi-cell type integration of inflammation and impaired regeneration

To better understand the dysregulated pathways that are either shared among different cell types or specific to a cell type, we performed a multi-gene Metascape analysis. Our analysis, shown in Supplemental Fig. 6, revealed distinct yet coordinated responses across nonneuronal cell types. The upregulated pathways indicate a strong neuroinflammatory response, primarily driven by cells of the neurovascular unit, including VLMCs, mural cells, endothelial cells, astrocytes, and microglia (Supplemental Fig. 6A). This multi-cell activation plays a key role in propagating systemic inflammatory signals to the brain [35]. Key pathways that were significantly upregulated include the TNF signaling pathway, cellular response to cytokine stimuli, and cytokine signaling in the immune system. Additionally, we observed activation of pathways related to the innate immune response, positive regulation of cytokine production, and cell activation across several cell types. This suggests a comprehensive and coordinated inflammatory reaction within the brain in response to septic conditions. Conversely, the downregulated pathways highlighted widespread disruption of metabolic and structural functions (Supplemental Fig. 6B). VLMCs and mural cells exhibited suppressed functions related to extracellular matrix (ECM) organization, which may weaken the structural integrity of the BBB. Endothelial and mural cells also showed significant downregulation of pathways associated with vascular transport, further indicating a potential breakdown of the BBB and overall disruption of vascular health. Additionally, metabolic processes, including carbohydrate and fatty acid metabolism, were downregulated in endothelial cells and microglia. Metabolic imbalance has been linked to energy deficiency, oxidative stress, and altered brain signaling, which may collectively lead to neuronal dysfunction [36]. Overall, the strong inflammatory response and metabolic and structural dysregulation likely contribute to the development of sepsis-associated encephalopathy.

Cell-cell communication analysis identifies significant rewiring of intercellular signaling in the septic brain

To further explore the cell-type-specific effects of sepsis, we employed CellChat2 to investigate the communication networks among different cell types [37]. Our analysis revealed a profound rewiring of cell-cell interactions in the brains of pigs exposed to E. coli compared to saline controls (Fig. 6). The number of interactions, reflecting the overall complexity of the communication network, and the interaction strength, representing the probability of interaction between all identified cell types, were largely increased in E. coli pigs than in saline controls (Fig. 6A). The number of interactions was significantly altered in the septic brain, with several cell types showing increased signaling capabilities. Specifically, VLMCs, mural, and endothelial cells emerged as key sender and receiver cells, exhibiting a dramatic increase in both incoming (top bar) and outgoing (side bar) signaling interactions (Fig. 6A, left plot). The VLMCs, in particular, represent a significant source of autocrine and paracrine signaling within the vascular compartment. Interestingly, the interaction strength is noticeably increased in astrocytes (Fig. 6A, right plot) but is relatively weak in the VLMCs. This suggests that the VLMCs are attempting more connections, but the functional strength of those connections is only slightly increased in the E. coli pigs compared to the saline controls (Fig. 6A, right plot). Interestingly, the interaction strength between the sender VLMC and receiver oligodendrocytes was dramatically reduced in the septic state (Fig. 6A). VLMCs likely contribute to blood-brain barrier and perivascular homeostasis through extracellular matrix deposition and signaling, including collagen and laminin [38]. During sepsis, disrupted communication within the neurovascular unit may activate glial cells, including astrocytes, leading to the release of neurotoxins that ultimately impair oligodendrocyte support, potentially contributing to demyelination [39]. Also noteworthy is a marked reduction in both the number and strength of neuron-derived signaling interactions, particularly toward microglia and astrocytes, indicating a loss of neuronal communication with key glial populations during sepsis. This reduction in neuron-to-glia signaling is consistent with prior evidence that neuronal homeostatic cues regulating microglial and astrocyte states are diminished during neuroinflammation, potentially contributing to glial activation and dysregulation [40, 41]. Specifically, the neuron signaling was reduced in sepsis, with loss of ligand–receptor interactions involved in neurotrophic support, synaptic plasticity, and axon guidance, while structural adhesion and synaptic scaffolding interactions were relatively preserved (Supplemental Fig. 7). In saline controls, cell–cell communication was dominated by homeostatic and neuroprotective signaling pathways, including FGF1, NEGR1, and FGF2, with prominent contributions from microglia and astrocytes (Fig. 6B). In contrast, sepsis induced a marked reorganization of intercellular signaling networks, characterized by the emergence of inflammatory and immune-related pathways (Fig. 6C). While some homeostatic signaling was retained, septic pigs exhibited increased signaling through pathways such as TNFRSF21, CD74, and TGFB1 receptor signaling, consistent with activation of immune and stress-response programs. The identification of altered NEGR1, SORL1, and APP signaling (Fig. 6B,C) is also particularly notable, as these pathways are classically associated with neuronal adhesion and neurodegenerative processes. Their dysregulation in microglia and astrocytes during acute sepsis suggests early disruption of the homeostatic glial–neuronal interface. Collectively, these findings support a model in which sepsis induces a network-level shift toward a glial- and vascular-dominated signaling environment.

Fig. 6. Sepsis-induced changes in intercellular communication within the brain.

A Heatmap showing the differential number of interactions (left) and differential interaction strength (right) between all cell types in the septic brain of E. coli compared to the saline control pigs. The color scale represents the relative change, with red indicating an increase and blue indicating a decrease in the E. coli group relative to the saline control. The marginal bars on the top and right summarize the total incoming and outgoing signal changes, respectively. Specifically, the top colored bar plot shows the sum of each column of the absolute values in the heatmap (incoming signaling). The right colored bar plot displays the sum of each row of the absolute values (outgoing signaling). The height of the bars reflects the extent of change in terms of the number of interactions or the strength of interactions between E. coli compared to the saline control pigs. B Circle plot showing the major cell-cell communication pathways in the saline control group. Ribbons represent communication from a source cell type (colored arc) to a target cell type, with the thickness of the ribbon corresponding to the interaction strength. Key homeostatic pathways such as NEGR1, FGF1, and FGFR2 signaling are primarily involved in microglia and astrocytes. C Circle plot illustrating the rewired communication network in the E. coli pig brains. The plot illustrates a shift from homeostatic to inflammatory pathways, marked by the emergence of robust, new signaling cascades, including TNFRSF21, CD74, and APP. These signals are primarily sent by microglia and mural cells, indicating their central role in the neuroinflammatory response.

To further resolve these changes in cell-to-cell communication, we analyzed cell-type-specific shifts in incoming and outgoing signaling (Fig. 7). Specifically, the analysis shows the varying strengths of outgoing (x-axis) and incoming (y-axis) interactions within signaling pathways for each cell type in E. coli pigs compared to saline controls (Fig. 7). The presence of “E. coli-specific” pathways indicates the development of sepsis-induced communication networks, meaning these pathways were significantly increased during sepsis. For example, microglia showed a significant decrease in both outgoing and incoming NEGR signaling, indicating a loss of homeostatic communication, while also displaying a sepsis-specific increase in incoming contactin (CNTN) signaling. The contactin family proteins are cell adhesion molecules essential for nervous system development and function, as well as for mediating interactions between neurons and other cells, including glial cells [42]. These findings support a transition away from homeostatic neuron–glia interactions toward alternative adhesion-mediated communication.

Fig. 7. Differential signaling changes within individual cell types in the septic brain.

Scatter plots comparing the differential outgoing (x-axis) and incoming (y-axis) interaction strength of signaling pathways for each cell type in the E. coli versus the saline control pigs. Each point represents a signaling pathway, and its position reflects the magnitude of change in its sending and receiving capabilities. The color and shape of the points indicate if the pathway is shared between the two groups or if it is “Ecoli specific,” meaning it is significantly active in the septic condition. Notably, endothelial cells exhibit a strong, E. coli-specific signal for incoming COLLAGEN and LAMININ, and decreased signaling of OCLN, supporting the notion of a compromised blood-brain barrier.

Astrocytes showed a broad increase in both incoming and outgoing signaling, including enhanced laminin pathway activity, indicating increased engagement with ECM-associated signaling (Fig. 7). Similarly, endothelial, VLMC, and mural cells demonstrated increased signaling in response to incoming collagen and laminin, with these pathways being significantly active in endothelial and VLMCs only in the E. coli condition (Fig. 7). These findings point to a coordinated shift toward ECM remodeling and vascular-associated communication networks during sepsis, with increased collagen signaling as a critical component [43]. Consistent with these vascular changes, endothelial cells exhibited decreased signaling through the tight junction pathway occludin (OCLN), which was labeled as “Saline specific” in this analysis. This indicates that the signaling pathway associated with OCLN is statistically significant and active during normal conditions (Saline), but it is lost or significantly downregulated during an E. coli infection (Fig. 7). This observation, combined with decreased OCLN expression in bulk RNAseq data (Fig. 1), strongly supports a sepsis-induced compromise of the BBB and other vascular cells. In parallel, multiple cell types, including oligodendrocytes, OPCs, astrocytes, and neurons, showed increased NCAM signaling, a pathway involved in cell adhesion and synaptic plasticity, suggesting a stress-induced remodeling of cell–cell contacts. Overall, the CellChat2 analysis provides an overview of how sepsis disrupts normal intercellular communication networks, replacing them with a robust pro-inflammatory signaling cascade and vascular disruption, which likely drive sepsis-associated encephalopathy.

Discussion

The early altered mechanisms underlying sepsis-associated encephalopathy (SAE), specifically the link between systemic inflammation and brain dysfunction, remain incompletely understood. In this study, we leveraged a clinically relevant porcine model, combined with bulk and single-nucleus transcriptomics, to provide a comprehensive view of the early brain transcriptomic response to sepsis. Our findings demonstrate that the acute septic brain undergoes a rapid and coordinated reprogramming characterized by (i) activation of conserved systemic inflammatory pathways, (ii) pronounced neurovascular and glial dysfunction, and (iii) disruption of intercellular communication networks. Consistent with our prior multi-organ analysis across the brain, blood, and kidney in an LPS pig model (Olney et al. 2024; PMC10765757) [17], we again observe strong transcriptional upregulation, particularly in NF-κB and cytokine signaling pathways. In our prior multi-organ analysis, there was a conserved inflammatory signature across tissues, suggesting that the brain is not insulated from systemic immune activation but instead engages in a coordinated host response to bacterial insult. However, this response is not merely a reflection of peripheral changes. The identification of brain-specific alterations, including the downregulation of metabolic transporters such as SLC52A3 and SLC38A5 (Fig. 1), indicates that the brain simultaneously initiates distinct, tissue-specific programs. These findings support a model in which SAE arises from the intersection of systemic inflammation and brain metabolic and cellular adaptations.

A central finding of this study is the prominent involvement of the neurovascular unit. Bulk RNAseq revealed downregulation of the tight junction gene OCLN, while single-nucleus and CellChat2 analyses demonstrated reduced OCLN-mediated signaling within endothelial cells, providing convergent evidence of early blood–brain barrier (BBB) disruption. BBB dysfunction is a well-established feature of sepsis and is thought to facilitate the entry of inflammatory mediators, thereby amplifying neuroinflammation and contributing to cognitive dysfunction [26]. Importantly, our multi-modal approach links transcriptional changes to functional communication deficits, suggesting that disruption of endothelial signaling is not only structural but also reflects a breakdown in coordinated cellular interactions required for barrier integrity (Fig. 7). In parallel with vascular dysfunction, we observed extensive transcriptional reprogramming in glial populations. Microglia exhibited a robust pro-inflammatory phenotype, consistent with their role as the primary immune effector cells in the brain, while endothelial, mural, and VLMC populations showed strong activation of cytokine and ECM-related pathways. The coordinated upregulation of collagen and laminin signaling across vascular-associated cell types indicates active remodeling of the ECM. These changes may reflect an attempt to stabilize the neurovascular interface under inflammatory stress but may simultaneously contribute to barrier dysfunction and altered cellular signaling. Collagen signaling has been previously proposed as a biomarker of sepsis [44, 45], further supporting the relevance of these pathways. Notably, the endothelial and microglial cell clusters showed marked downregulation of genes involved in metabolism and lipid processing (Figs. 3 and 5). In the context of neuroinflammation, this metabolic shift might enable endothelial and microglial cells to quickly produce energy and biosynthetic precursors needed for cytokine production and barrier remodeling. These findings suggest that metabolic reprogramming is not merely a secondary consequence of systemic collapse, but a primary regulatory pivot essential for orchestrating the brain’s acute inflammatory response [30].

A particularly striking observation is the relative lack of transcriptional changes in neurons during this early sepsis response time window. Despite widespread inflammation and vascular disruption, neurons showed no significant differential gene expression, suggesting that early neuronal dysfunction in SAE is not driven by intrinsic transcriptional responses. Instead, our ligand–receptor analysis indicates that neurons undergo a functional decoupling from the surrounding cellular network. Specifically, we observe a selective loss of neuron-derived signaling pathways associated with neurotrophic support, synaptic plasticity, and axon guidance, while preserving structural adhesion and synaptic scaffolding interactions (Supplemental Fig. 7). This pattern suggests that neurons remain structurally connected but lose their capacity to actively regulate glial and vascular cell states. This finding has important implications for understanding the pathogenesis of SAE. Neurons maintain microglial and astrocyte homeostasis through ligand–receptor signaling, and disruption of these signals can promote glial activation and dysregulation of inflammation [33, 46, 47]. Our data provide systems-level evidence supporting this model, indicating that reduced neuronal signaling may contribute to the shift toward a glial- and vascular-dominated inflammatory environment. In this context, neuronal dysfunction may arise not from immediate cell death, but from impaired communication and metabolic stress induced by the surrounding inflammatory response. Consistent with this interpretation, we did not observe enrichment of canonical apoptotic pathways in neurons or oligodendrocytes at this early time point. Instead, alterations in APP, NEGR1, and SORL1 signaling suggest the initiation of sub-lethal stress responses that may precede neurodegeneration. In contrast to neurons, oligodendrocytes exhibited a strong transcriptional stress response, including activation of unfolded protein response and autophagy-related pathways, indicating early vulnerability within the white matter compartment (Supplemental Fig. 5). Together, these findings support a model in which early SAE begins with signaling dysfunction and metabolic stress which may be later followed by structural damage and neuronal death.

Finally, our findings support a growing body of human and experimental evidence positioning glial and vascular dysfunction at the center of SAE pathogenesis. Recent single-nucleus analyses of human septic brain hippocampus tissue have similarly identified microglial activation and neurovascular alterations (Liu et al., 2025; PMC12477015), reinforcing the biological relevance of our observations. The concordance between our data and human studies highlights the translational strength of the porcine model, which more closely recapitulates human brain architecture and physiology than traditional rodent systems [17–19]. Nonetheless, we acknowledge several limitations of our study. Although the porcine model offers important advantages, species-specific differences may still constrain direct extrapolation to human disease. Additionally, our analysis focuses on an early time point (within 6 h of infection), and pigs were given phenylephrine and norepinephrine to maintain hemodynamic stability (MAP at 60 mmHg), limiting insight into later-stage immune cell infiltration and overt neuronal death. While our dual bulk and single-nucleus transcriptomic approach provides high-resolution insights into cellular states and communication networks, they do not directly capture protein-level changes. Future studies incorporating immunohistochemistry and longitudinal sampling will help extend these findings. Despite these limitations, our study provides a comprehensive and integrative framework for understanding early brain responses to sepsis, supporting a model in which systemic inflammation rapidly engages the neurovascular unit, driving BBB disruption, glial activation, and large-scale remodeling of intercellular communication networks. Within this altered environment, neurons become functionally disengaged from regulatory signaling while remaining structurally intact, leading to impaired neuro-glial communication, which could contribute to early cognitive dysfunction. Collectively, these findings shift the conceptual framework of SAE away from neuron-centric mechanisms toward a model dominated by vascular and glial dysfunction, identifying the neurovascular unit as a critical and potentially actionable therapeutic target.

Materials and methods

Experimental design

Animal housing and experimental procedures were conducted within the Mayo Clinic Arizona animal facility under an approved IACUC protocol (#A0005412-20). Outbred Yorkshire female swine (Sus scrofa domesticus) were randomly selected to receive continuous intravenous (i.v.) infusion of Escherichia coli (E. coli, n = 4) or saline (control, n = 4). Animals were aged 4.5 months and weighed between 74.4 – 86 Kg (Supplemental Table 1). Animals were group-housed with ad libitum access to water and Purina Lab Porcine Grower Diet 5084. Adult males were not included in these experiments as they are too large and aggressive for safe housing and handling. At experiment onset, anesthesia was induced with intramuscular glycopyrrolate (0.02 mg/Kg), telazol (5 mg/Kg), and xylazine (2 mg/Kg), followed by maintenance anesthesia with continuous isoflurane (1.5-2.5%). Once anesthetized, cannulas were surgically placed in the jugular vein and carotid artery for clinical monitoring and infusion procedures. A continuous i.v. infusion of E. coli (clinical isolate 41949 obtained from Brigham and Women’s Hospital, Crimson Biorepository) was initiated for a target dose of 1.2×10^9 colony forming units (CFU)/Kg over a maximum of 6 h, while saline controls received an equivalent volume of 0.9% normal saline (1.00 mL/h). Time to euthanasia varied among pigs exposed to E. coli (199 – 362 min) due to individual differences in the progression of septic shock. In contrast, all saline control pigs were euthanized at 300 min, having exhibited no signs of organ dysfunction. To replicate the treatment provided in an intensive care unit (ICU), vital signs were continuously monitored, and supportive care was provided as needed throughout the experiment. Measurement of clinical parameters included heart rate, oxygen saturation (SpO₂), blood pressure, mean arterial pressure (MAP), and body temperature (Supplemental Table 1). Supportive measures included maintenance i.v. PlasmaLyte fluid replacement (1.5 mL/Kg/hr), supplemental oxygen (100%) for Sp02 < 90, i.v. phenylephrine (1 mcg/Kg/min) as needed for MAP < 60, i.v. norepinephrine (0.01-1 ug/Kg/min) as needed for MAP < 60, and i.v. 5% dextrose in water (D5W) for glucose <70 mg/dl. Humane endpoints were established by the attending veterinarian prior to the onset of experiments. These included uncontrolled bleeding or uncorrectable clinical deterioration, as indicated by severe changes in vital signs or cardiac rhythms, multi-organ dysfunction unresponsive to supportive measures, as designated by the veterinary technicians, veterinarian, or research investigators. Brain (prefrontal cortex) samples were collected post-mortem and flash-frozen immediately at −80 °C or fixed in formalin for subsequent analysis. No blinding was done in this study.

Bulk RNAseq analysis

A total of 0.5 grams of forebrain tissue was used to prepare total brain RNA from n = 4 E. coli pigs and n = 4 saline-treated pigs with the Qiagen RNeasy Mini Kit (cat # 74104). Samples were QC’d on the Agilent Tapestation, and bulk RNAseq libraries prepared with TruSeq stranded mRNA (cat # 20020594) and sequenced to a depth of ~50 M paired-end reads on an Illumina NovaSeq S4 instrument at the Mayo Clinic Genome Sequencing Core. FASTQ files were trimmed for adapter content and polyA/polyG artifacts using BBDuk, part of the BBMap package [48].FastQC and MultiQC were used to assess read quality before and after trimming [49, 50]. Trimmed reads were aligned to the Ensembl reference genome Sscrofa11.1 v107 using STAR 2.7.11b [51]. Quantification estimates for each sample were obtained using STAR following the twopassMode alignment and quantMode to get the read counts for each geneID. A count matrix of gene rows by sample columns was created and read into R for analysis. Mitochondrially-encoded genes (MT-genes) were excluded during the normalization pipeline to prevent highly abundant mitochondrial transcripts from masking more subtle nuclear-encoded differential expressions, and only protein-coding genes were retained. Lowly expressed genes were removed from the analysis by keeping only genes with a count greater than or equal to one count per million (CPM) in all samples within at least one group, 13,692 protein coding genes were retained for downstream differential expression analysis. The edgeR package converted library sizes into effective library sizes using the Trimmed Mean of M-values (TMM) [52]. Differential expression was performed using the limma/voom pipeline [53, 54]. A model was created to compare E. coli and saline where each coefficient corresponds to a group mean and minutes from injection till the killing was added as a covariate. P-values were adjusted for multiple tests using the Benjamin-Hochberg correction, denoted as q-value. Specifically, the q-value is derived by adjusting the distribution of p-values across all tested genes to control the expected proportion of false positives among the set of significant findings. There is no fixed raw p-value cutoff that corresponds to q < 0.05. Instead, the effective p-value threshold depends on the overall p-value distribution. We used a logarithm base 2 of the fold change (log2FC) to denote the ratio of change between experimental and control animals. Genes were identified as differentially expressed when the absolute log2FC exceeded 1.0 and the q-value < 0.05. This strict criterion guarantees that the detected changes are not only statistically reliable after adjusting for multiple testing, but also biologically robust in magnitude.

Single-nucleus RNAseq analysis

Frozen brain samples from the same pigs used for bulk RNAseq analysis (n = 4 pigs per group) were cut on dry ice into approximately 500 mg pieces. All buffers and washes contained RNase inhibitor (Protector RNase Inhibitor, MilliporeSigma, Cat. No. 3335402001). For nuclei lysis, 2 mL of ice-cold Nuclei Extraction Buffer (Miltenyi Biotec, Cat. No. 130-128-024) containing RNase inhibitor and BSA was added to pre-cooled gentleMACS C Tubes. Tissue pieces were immediately transferred to the C Tube to prevent RNA degradation. Dissociation was performed on a gentleMACS Dissociator using the program “4C_nuclei_1”. After dissociation, the C Tube was immediately placed on ice. The nuclei suspension was then passed through a 100 µm MACS Smart Strainer (Miltenyi Biotec). To maximize nuclei recovery, the strainer was washed with 2 mL of ice-cold lysis buffer, and the used C Tube was rinsed with ice-cold lysis buffer, with all washes added to the strainer. The flow-through was then passed through a 30 µm MACS Smart Strainer, which was subsequently discarded. The filtered nuclei suspension was centrifuged at 300 × g for 5 min at 4 °C. Low-binding tubes were used for all steps following the initial dissociation.The supernatant was entirely aspirated, and the cell pellet was carefully resuspended in cold buffer following the Miltenyi Biotec protocol. Nuclei were counted using propidium iodide on an automated cell counter. cDNA libraries were prepared using the 10X Genomics Chromium Next GEM Single Cell 3’ Kit, v4. The Kapa mRNA Hyper Prep Cat# KR1352 was used for library construction. 1.2 nM cDNA will be loaded onto Illumina NovaSeq 6000 S4 for sequencing.

Alignment of the single nucleus 10x Genomics data to the Ensembl reference genome Sscrofa11.1 v107 was carried out using the Cellranger v9.0.0 pipeline [55]. Cellbender v0.3.2 was used to remove technical noise from the single nucleus data that may result from enzymatic processes and produce library fragments leading to contamination [56]. The counts adjusted single nucleus count data from Cellbender was then imported into R v4.3.0 using the Seurat v5.0.2 package for quality filtering, dimensionality reduction, and clustering [57]. Doublets were identified and removed with DoubletFinder [58]. Data integration was performed using Harmony [59]. For cell clustering of the snRNAseq datasets, a two-dimensional Uniform Manifold Approximation and Projection (UMAP) [60] was generated. This dimensionality reduction allowed us to visualize which cell populations group together based on gene expression levels. Expression of cell-specific markers (Fig. 2) was used to annotate the cell clusters. Cell expression data were aggregated for pseudobulk analysis. Specifically, the Seurat command AggregateExpression was used to sum gene counts of all cells from the same sample for each cell type. The aggregation was limited to protein-coding genes. After aggregation, cell-type-specific gene filtering focused only on genes with a mean expression above the first quartile within each cell type. Following pseudobulk creation and filtering out lowly expressed genes, differential expression analysis between E. coli and saline within each cell type was performed using the DESeq2 method, which relies on a model based on the negative binomial distribution [61]. Genes were identified as differentially expressed when the absolute log2FC exceeded 1.0 and the q-value was less than 0.05. This strict criterion ensures that the observed changes are not only statistically reliable after adjusting for multiple testing but also biologically meaningful in magnitude. The q-values are calculated using the Benjamin-Hochberg correction. The q-value is obtained by adjusting the distribution of p-values across all tested genes to control the expected proportion of false positives among significant findings. There is no fixed raw p-value cutoff that corresponds to q < 0.05; instead, the effective p-value threshold varies according to the overall p-value distribution within each tested cell type.

Gene set enrichment and network analysis

To understand the biological significance of differentially expressed genes (DEGs) identified from both bulk tissue and snRNA sequencing data, we performed several bioinformatics analyses. We first utilized the Metascape web tool (https://metascape.org), which integrates data from multiple functional annotation databases to provide comprehensive insights into gene sets [62]. For both bulk RNAseq and single-cell type DEGs, we performed a functional enrichment analysis to identify enriched gene ontology (GO) terms, pathways, and other biological annotations. Metascape uses a hypergeometric distribution to assess enrichment, and we considered terms with a corrected p-value (adjusted Fisher’s exact p-value) of less than 0.05 as significantly enriched. This approach allowed for the identification of pathways that were either up- or downregulated in the E. coli pigs compared to saline-treated pigs, facilitating a broader understanding of the cellular response to sepsis.

To further explore the biological context and identify key regulatory networks, we employed Qiagen’s Ingenuity Pathway Analysis (IPA) [63]. IPA leverages a curated knowledge base of molecular interactions to analyze gene expression data. We used IPA to identify significantly altered canonical pathways and predict upstream regulators—molecules like transcription factors or cytokines—that could explain the observed changes in gene expression. This analysis provided a mechanistic framework for interpreting the transcriptional alterations observed in the brain tissue of septic pigs.

Cell-cell communication analysis

To systematically investigate intercellular communication changes in the brain in response to sepsis, we applied CellChat v2 (v.2.1.0) [37] to our snRNAseq data. CellChat is a computational tool that infers and analyzes cell-cell communication networks by quantifying the signaling probability between different cell types. It uses a database of ligand-receptor interactions, including those with multi-subunit structures, to model communication based on gene expression levels. Signaling crosstalk was inferred using the CellChatDB.human database, focusing on protein-protein interactions. Given that porcine gene symbols in our dataset align with human nomenclature and the high evolutionary conservation of neurovascular signaling components between these species, the human database provided a robust framework for inference. Prior to communication analysis, we applied a minimum expression threshold of 10 cells per cell group using the filterCommunication function to ensure that signaling was not driven by rare, low-frequency outliers. We further refined the data by considering only over-expressed ligands and receptors within each condition (identifyOverExpressedGenes). Communication probabilities were quantified using the ‘triMean’ statistical method, which calculates the average gene expression of multi-subunit signaling complexes while minimizing the influence of extreme values. Significant interactions were defined using CellChat’s internal permutation-based testing with a significance threshold of p < 0.05. The comparative analysis between the E. coli and saline groups used the mergeCellChat function. Interactions were quantitatively categorized as ‘increased’ or ‘decreased’ based on the differential communication probability (strength) and the total information flow (the sum of communication probabilities) between conditions. Differential expression of signaling genes was further validated using a Wilcoxon rank-sum test with a log-fold change threshold of 0.05 and a p-value threshold of 0.05 to map condition-specific signaling enrichment.

Analytical concordance

To strengthen the biological interpretations of the septic porcine brain, we captured molecular changes across three distinct levels of resolution. Our approach identifies key transcriptomic signatures by assessing the concordance between (i) global gene expression shifts in bulk tissue RNAseq, (ii) cell-type-specific transcriptional alterations in snRNAseq, and (iii) intercellular signaling probability changes inferred through CellChat v2. To resolve the cellular origins of the bulk tissue signal (Fig. 2D), we implemented a dominance-based assignment logic for differentially expressed genes (DEGs). Specifically, for each DEG identified in the bulk RNAseq (q < 0.05, |log₂FC | > 1), those genes were cross-referenced with the DEGs identified in each cell type. In instances where a bulk DEG was significantly regulated across multiple cell types in the snRNAseq data, the gene was assigned to the “dominant” cell type—defined as the cluster exhibiting the greatest absolute fold change ( | log₂FC | ) in the same regulatory direction as the bulk signal. Genes that failed to meet significance thresholds or displayed discordant directional changes between bulk and the snRNAseq data were categorized as “Not recovered.” This multi-layered strategy allows us to prioritize molecular pathways and signaling axes that show consistent regulation at both the tissue and single-nucleus levels, providing a high-fidelity map of the acute neurovascular and glial response to sepsis while ensuring findings are supported by multiple independent lines of transcriptomic evidence.

Statistical analysis

All statistical analyses were conducted using R v4.3.0. To assess data distribution, the Shapiro–Wilk test was performed for each experimental group. Comparison of baseline (start) and terminal (end) measurements between Yorkshire pigs receiving continuous intravenous infusion of E. coli (n = 4) or saline (n = 4). For each parameter, the mean and standard deviation are reported (Supplemental Table 7). Data distribution was assessed using the Shapiro–Wilk test, and homogeneity of variance was evaluated via Levene’s test. Significant differences between groups were determined using a two-sample t-test, with the Wilcoxon rank-sum test as a nonparametric alternative for non-normally distributed variables. Statistical significance was defined as p < 0.05 (Supplemental Table 7).

Author contributions

LT and JF designed the experiments. NG, KG, PH, ASP, PCD, MS, RW, ML, and EF provided maintenance for the pigs and additional technical support. NG, a board-certified veterinarian, oversaw the experiment. CdÁ performed tissue collection. KO, CdÁ, KT, LT, KG, and JB conducted the research. KO performed the bulk RNAseq, snRNAseq, and statistical analyses and created the visualizations. KO, JB, and JF wrote the manuscript with input from all contributing authors. All authors read and approved the final manuscript.

Funding

JF discloses support for this research from the Mayo Foundation and the Translational Genomics Research Institute. LT discloses support for this research from the Mayo Clinic and ASU Team Science Grant.

Data availability

The raw data are deposited in the Sequence Read Archive (SRA), accessible via PRJNA1172687 for the bulk RNAseq and PRJNA1175698 for the snRNAseq. The processed counts data is available on GEO under GSE335778. The code used for analysis is available at https://github.com/olneykimberly/Ecoli_pigs. Interactive Shiny app to explore the single-nucleus data: https://fryerlab.shinyapps.io/ecoli_pigs_snRNA/ and bulk RNAseq: https://rpubs.com/olneykimberly/Ecoli_vs_Saline.

Code availability

The raw data are deposited in the Sequence Read Archive (SRA), accessible via PRJNA1172687 for the bulk RNAseq and PRJNA1254309 for the snRNAseq. The processed counts data is available on GEO under GSE335778. The code used for analysis is available at https://github.com/olneykimberly/Ecoli_pigs. Interactive Shiny app to explore the single-nucleus data: https://fryerlab.shinyapps.io/ecoli_pigs_snRNA/ and bulk RNAseq: https://rpubs.com/olneykimberly/Ecoli_vs_Saline.

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

Approval for the study was obtained from the Institutional Animal Care and Use Committee (IACUC) (protocol number A00005412-00) and the Animal Care and Use Review Office of the US Army Medical Research and Materiel Command, Office of the Department of Defense. The work was conducted at an institution accredited by the American Association for Accreditation of Laboratory Animal Care. AAALAC accredited and USDA registered facility, and in accordance with NIH guidelines.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41398-026-04301-1.