Authors: Dario Figueroa Velez (aDepartment of Pathology, Boston Children’s Hospital and Harvard Medical School, Boston, MA 02115, USA), Reza Rahimian (bDouglas Mental Health University Institute, Dept. of Psychiatry, McGill University, Canada), Christine Hehnly (aDepartment of Pathology, Boston Children’s Hospital and Harvard Medical School, Boston, MA 02115, USA), Jordan C. Benson (aDepartment of Pathology, Boston Children’s Hospital and Harvard Medical School, Boston, MA 02115, USA), Isabella Sacharczyk (aDepartment of Pathology, Boston Children’s Hospital and Harvard Medical School, Boston, MA 02115, USA), Gustavo Turecki (bDouglas Mental Health University Institute, Dept. of Psychiatry, McGill University, Canada), Naguib Mechawar (bDouglas Mental Health University Institute, Dept. of Psychiatry, McGill University, Canada), Maria K. Lehtinen (aDepartment of Pathology, Boston Children’s Hospital and Harvard Medical School, Boston, MA 02115, USA)
Categories: Article, Bipolar disorder, Choroid plexus, Inflammation, Blood-CSF barrier, Cytokine
Source: Brain, behavior, and immunity
Authors: Dario Figueroa Velez, Reza Rahimian, Christine Hehnly, Jordan C. Benson, Isabella Sacharczyk, Gustavo Turecki, Naguib Mechawar, Maria K. Lehtinen
Inflammation has emerged as a prominent feature of bipolar disorder (BD) pathophysiology, drawing attention to brain barriers known to regulate immune-brain interactions. While perturbation of the blood–brain barrier has been reported in BD, the blood-cerebrospinal fluid (CSF) barrier formed largely by the choroid plexus (ChP) remains underexamined. To address this gap in knowledge, we used a multiplex array to measure cytokine protein abundance in postmortem ChP tissue from individuals with BD and unaffected controls, revealing elevated levels of CCL2 and SPP1, factors associated with monocyte and macrophage recruitment and activation. In contrast, expression of cytokines involved in tissue homeostasis, trophic support, and immune signaling, including OSM, IGF-1, CX3CL1, TGFB3, GDNF, LIF, BDNF, SCF, and FGFs, was reduced. Several cytokines, including CCL2 and PLGF, exhibited condition-specific divergent age trajectories. Bulk RNA sequencing of the same cohort revealed a modest set of differentially expressed genes, including transcripts associated with oxidative stress, mitochondrial function, and immune regulation that were upregulated in BD. Notably, the BD CSF biomarker NELL2 was downregulated in the ChP. Gene set enrichment analysis highlighted activation of inflammatory and cellular stress pathways, as well as reduced expression of junction-related gene programs. These findings suggest a shift in ChP function in BD characterized by increased pro-inflammatory signaling and reduced trophic and barrier-supportive activity. Together, these data identify the ChP as an active site of immune dysregulation in BD and support the broader notion of brain barrier dysfunction in mood disorder pathology.
Bipolar disorder (BD) is a life-long psychiatric condition afflicting ~ 1% of the global population (Merikangas et al., 2011; Grande et al., 2016) and represents a leading contributor to disability (Alonso et al., 2011; GBD 2019 Mental Disorders Collaborators, 2022; Grande et al., 2013). Although the etiology of BD is unknown, a growing body of research suggests persistent immune activation is central to its pathophysiology (Munkholm et al., 2013; O’Brien et al., 2006; Brietzke et al., 2009; Rowland et al., 2018; Modabbernia et al., 2013; Ghoryani et al., 2019; Curzytek and Leskiewicz, 2021; Berk et al., 2011; Goldsmith et al., 2016; Sneeboer et al., 2019; Giridharan et al., 2020; Watkins et al., 2014). Sustained, low-grade inflammation in BD patients is evidenced by elevated circulating pro-inflammatory cytokines including IL–6, TNF–α, IL–8 (CXCL8), and CCL2 (Brietzke et al., 2009; Barbosa et al., 2013; Poletti et al., 2024; Misiak et al., 2020; Solmi et al., 2021; Isgren et al., 2017; Isgren et al., 2015) and by inflammation-related transcriptional signatures in postmortem brain tissue (Pacifico and Davis, 2017; Logotheti et al., 2013). It is unclear whether this activation originates peripherally or centrally, or if it is a primary disease process or a secondary response (Barichello et al., 2021; Benedetti et al., 2020).
The choroid plexus (ChP) is the principal source of cerebrospinal fluid (CSF) and accounts for most of the surface area of the blood-CSF interface (Damkier et al., 2013; Cserr, 1971; Wolburg and Paulus, 2010; Lun et al., 2015). This anatomy not only enables surveillance of both the blood and CSF but also allows the ChP to quickly alter CSF composition through active solute transport, secretion of signaling molecules, and immune cell trafficking (Damkier et al., 2013; Solar et al., 2020; Xu et al., 2021; Lehtinen et al., 2011; Xu et al., 2024; Saunders et al., 2012). Growing evidence implicates the ChP as an immunologically active structure capable of sensing inflammation and modulating cytokine and leukocyte entry into the CNS (Xu et al., 2024; Marques et al., 2009; Xu et al., 2025; Engelhardt et al., 2001; Cui et al., 2020; Baruch et al., 2015; Thompson et al., 2022). In rodent models, ChP transcriptomic profiles shift markedly in response to immune activation, and similar molecular changes have been observed in infection and neurodegeneration (Xu et al., 2024; Marques et al., 2009; Engelhardt et al., 2001). Multiple human imaging studies have reported ChP enlargement in psychiatric illnesses, particularly in mood disorders, where ChP volume correlates with peripheral inflammatory markers and ventricular dilation (Bravi et al., 2024; Cao et al., 2023; Althubaity et al., 2022; Lizano et al., 2019; Gong et al., 2025). However, it remains unknown whether the ChP in BD exhibits localized inflammatory signatures or contributes to disease pathology through changes in barrier or signaling function.
To address these questions, we compared cytokine, transcriptomic, and histological characteristics of ChP tissue from individuals with BD to controls. In BD we found elevated levels of monocyte- and macrophage-attracting chemokines (CCL2 and SPP1) alongside reduced expression of cytokines implicated in tissue repair and homeostasis (IGF-1, CX3CL1, TGFB3, GDNF, and LIF). Transcriptomic analysis identified markers of oxidative stress (e.g., MT2A, HMOX1, IER3), mitochondrial activity (e.g., MT-ND5, MTHFD2), and immune signaling (e.g., LIF, IER3), and revealed the BD biomarker NELL2 was downregulated ChP BD (Goteson et al., 2025). Gene set enrichment analysis (GSEA) reinforced findings of immune activity and were consistent with a loss of barrier integrity. Together, these findings provide the first cell- and tissue-level evidence of localized inflammation at the ChP in BD, suggesting that changes in blood-CSF barrier function and immune signaling may contribute to disease pathophysiology.
Postmortem human brain samples were provided by the Douglas-Bell Canada Brain Bank (DBCBB) (Montreal, Canada) with ethical approval from the Douglas Research Ethics Board and an IRB-approved protocol at Boston Children’s Hospital. Causes of death were ascertained by the Quebec Coroner’s Office, and psychiatric diagnoses determined by psychological autopsy, as previously described (Dumais et al., 2005). BD samples were from diagnosed individuals who died by suicide during a depressive episode. Controls were from psychiatrically and neurologically healthy individuals who died suddenly. Exclusion criteria included subjects with ongoing or prior immune or inflammatory illness. Cases with evidence of agonal death and likely terminal hypoxia were also excluded. Through a collaboration between the McGill Group for Suicide Studies and the Quebec Coroner’s Office, information for each sample included clinical vignettes created by a panel of psychiatrists from medical records, coroner’s reports, toxicological analyses, and other sources of information (Dumais et al., 2005). Demographic, postmortem, and clinical metadata for all subjects are provided in Supplementary Table 1.
Tissue samples from the lateral ventricle ChP were fresh-frozen for cytokine and gene expression analyses or formalin-fixed paraffin-embedded (FFPE) for histological analysis. Subject groups were matched for sex, age, postmortem interval (PMI), brain pH, and body mass index (BMI; Supplementary Table 2).
Protein lysate was obtained by homogenization of 50 mg of frozen ChP in 1X Cell Lysis Buffer (included in the Ray Biotech kit) supplemented with protease inhibitor cocktail (Roche). Protein concentration was determined by TECAN microplate reader for each sample using the BCA Protein Assay (Thermo Fisher Scientific).
Human antibody array (RayBiotech Human Cytokine Array C5) was used to measure pro-inflammatory cytokines, anti-inflammatory cytokines, chemokines and growth factors in frozen archived lateral ventricle ChP samples. This array allows the semi-quantitative detection of 80 human proteins in tissue lysate at the same time. Antibody arrays were removed carefully from the plastic packaging, and each membrane was placed into a well of the incubation tray. 2 ml of blocking buffer was pipetted into each well and membranes were incubated for 2 h at room temperature. After blocking, 1 ml of diluted sample (150 μg protein in 1 ml of blocking each membrane) was added to each well. The wells were incubated overnight at 4°C on a shaker. Samples were washed with 2 ml of two different wash buffers (5 times in total) for 5 min at room temperature. 1 ml of the prepared biotinylated antibody cocktail was pipetted into each well and membranes were incubated overnight at 4°C. All membranes were washed again as described. 2 ml of 1X HRP-streptavidin was added into each well and membranes were incubated for 2 h at room temperature. Following the last series of washes, chemiluminescence detection was completed using detection buffers provided by the kit. Background correction was performed by subtracting the mean signal of the negative control spots from each target spot. To normalize for inter-array variability, spot intensities on each array were adjusted relative to the mean signal of the Positive Control spots on that array, using the following X(Ny)=X(y)*P1/P(y) where X(y) is the background-subtracted intensity of spot X on array y, P1 is the mean intensity of the positive control spots on the reference array, and P(y) is the mean intensity of all positive control spots. After normalization, values that were negative and outliers identified using a ROUT outlier test (GraphPad Prism, Q set to 1%) were removed.
The sampling distribution of the mean and error for each cytokine was estimated using 10,000 bootstrap iterations with replacement. Group differences were assessed using a null-shifted bootstrap approach, in which group-specific distributions were symmetrically shifted to test a zero-difference null hypothesis. The empirical p-value was calculated as the proportion of bootstrap iterations under the null-shifted distributions that were as extreme or more extreme than the observed group differences, yielding a two-sided p-value. Multiple comparisons were corrected using a False Discovery Rate (FDR) procedure (Benjamini-Hochberg method), and significance was set at a q value of 0.1, consistent with the bulk sequencing analysis. For cytokines showing significant group differences between BD and Ctrl, associations between cytokine abundance and age were evaluated using Pearson correlation analyses (GraphPad Prism). These analyses were used to assess whether age-related effects plausibly accounted for observed group differences, with summary statistics reported in Supplementary Table 3.
RNA was extracted from ChP tissue using RNeasy Kits (Qiagen) according to the manufacturer’s instructions. Tissue was homogenized using a pellet pestle in buffer RLT (Qiagen) supplemented with 1% 2-mercaptoethanol (MilliporeSigma). RNA was purified using spin columns and eluted in RNase-free water. RNA quantity and purity were initially assessed by spectrophotometry (Nanodrop, A260/280 > 2.0).
Purified RNA (500 ng) was reverse transcribed using the NEB First Strand cDNA Synthesis Kit (New England Biolabs) according to the manufacturer’s protocol. For each duplex RT-qPCR reaction, 100 ng of cDNA equivalent was used with TaqMan probes. Each reaction included a probe for the internal reference gene HPRT1 and one of the following target MT2A, HMOX1, IER3, RGS8BP, ARRDC3, ERRFI1, NELL2, or SLC39A11. Amplification and fluorescence quantification were performed using the StepOne Real-Time PCR System. Relative expression was calculated using the 2^−ΔΔCt^ method (Livak and Schmittgen, 2001), normalizing each target to HPRT1.
For transcriptomic profiling, total RNA extracted from ChP tissue was submitted to Novogene for library preparation, alignment, and quantification. Novogene aligned to the human GRCh39 reference genome using STAR (Dobin et al., 2013), and gene-level count matrices were generated with featureCounts (Liao et al., 2014). These counts were used as input for differential expression and gene set enrichment analyses with DESeq2 (Love et al., 2014).
Differential expression analysis was performed using the DESeq2 package (v1.46.0) in R (R4.1). Raw gene counts obtained from featureCounts were filtered to remove lowly expressed genes, retaining only genes with an average count > 10 across samples. Data were normalized using DESeq2′s default size factor estimation, and dispersion estimates were calculated for each gene. To identify potential technical and biological confounders, we performed principal component analysis (PCA) on variance-stabilized data. The first 5 principal components were evaluated against metadata (RIN, sex, age, PMI, BMI, pH and Condition) using linear models, with covariate significance assessed by ANOVA F-test and variance contributions calculated as the percentage of each PC’s total sum of squares attributable to each term. In the unfiltered dataset, RIN was identified as a dominant driver of variance, explaining 69.8% of PC1 variance and significantly associated with PC1 (F-test p = 0.0003). Samples with RIN < 5 were subsequently excluded to minimize this bias (Supplementary Table 4). The final differential expression model used a design formula of ~ Sex + Age + Condition. Differentially expressed genes (DEGs) were defined as those with an adjusted p-value < 0.1 (Benjamini-Hochberg; Supplementary Table 5). Volcano plots and heatmaps were generated using ggplot2 (https://ggplot2.tidyverse.org) and pheatmap (Kolde, 2025), respectively.
Gene set enrichment analysis was performed using the clusterProfiler package (v4.14.6) with Hallmark gene sets from the Molecular Signatures Database (MSigDB v2025.1.Hs). DESeq2-derived log2 fold-changes were mapped to ENTREZ IDs via org.Hs.eg.db. GSEA was run using the GSEA function with the following minGSSize = 15, maxGSSize = 500, pvalueCutoff = 1, and pAdjustMethod = “BH”. Gene sets with false discovery rate (FDR) q-values < 0.1 were considered significantly enriched. Ridgeplots were generated to visualize enrichment using the enrichplot package.
To identify genes recurrently contributing to multiple enriched pathways, leading-edge subsets were extracted from Hallmark gene sets with FDR q-values < 0.1 using the core_enrichment field from the GSEAResult object in clusterProfiler (Wu et al., 2021). Gene symbols were mapped to ENSEMBL IDs using the org.Hs.eg.db database for integration with the normalized expression matrix. A binary presence-absence matrix was constructed to indicate which leading-edge genes appeared in each enriched pathway. Genes were ranked by frequency of occurrence across gene sets, and those appearing in multiple pathways were considered shared drivers of the transcriptional programs. Overlap patterns were visualized using binary heatmaps generated via the pheatmap package.
To provide cellular context for inflammation-associated transcriptional changes, DEGs identified in human BD ChP were compared with a published mouse ChP single-cell RNA-seq inflammatory model (Xu et al., 2024). Human genes were mapped to mouse orthologs by gene symbol, and expression patterns were examined across annotated ChP cell populations. A subset of representative DEGs was selected for visualization based on clear association with defined inflammatory or activated cell states in the mouse dataset.
FFPE tissue blocks were sectioned at ~ 15 μm using a Leica microtome. For immunohistochemistry, a subset of sections were sent to the BCH histology laboratory core for CD45 staining. For immunofluorescence staining, slides were deparaffinized, rehydrated through graded alcohol, and subjected to antigen retrieval in citric buffer (10 mM, pH 6.0) using a steam bath (100°C) for 30 min and then removed from the bath until the liquid reached room temperature (30–60 min). Sections were permeabilized for 30 min (0.3% Triton X-100 in 1XPBS) and background autofluorescence minimized using TrueBlack lipofuscin quencher (Biotum) followed by a 30 min incubation period in animal-free blocker (VectorLabs). Sections were incubated with chicken anti-Iba1 (SynapticSystems, 234 009, 200) in animal-free blocker overnight at 4°C. The next day, sections were washed in 1XPBS (0, 0, 5, 10, 20 min) and incubated with Alexa Fluor 488 goat anti-chicken (Invitrogen, 1,000) for 1 h at room temperature and then washed with 1XPBS (0, 0, 5, 10 min). Sections were then counterstained with Hoechst 33,342 (Invitrogen, H3570, 10,000) and mounted using ProLong Gold Antifade Mountant (Invitrogen). Sections were imaged using a 40x water-immersion objective and Airyscan mode (Zeiss, LSM980). Image analysis was performed in ImageJ and cell counts were quantified within user-defined ROIs outlining a cross section of the epithelium and performed in a blinded manner.
To characterize the inflammatory landscape of the ChP in BD, we first profiled 80 cytokines using a membrane-based antibody array to determine their relative abundance (Fig. 1A). Outliers for each cytokine identified and excluded using a ROUT test (Q = 1%). Among the differentially expressed cytokines, CCL2 and SPP1 were elevated in BD samples (Fig. 1B). These factors are classically associated with immune cell recruitment, including monocytes, neutrophils, and macrophages. In contrast, OSM, SCF, IGF-1, FGF-4, FGF-9, CXCL6, CX3CL1, GDNF, TNFSF14, LIF, TGF-β3, and BDNF were reduced in BD (Fig. 1C). Many of these factors are linked to tissue homeostasis, trophic support, and immune signaling. Together, these patterns suggest a shift in the local cytokine environment characterized by selective immune activation alongside reduced trophic and regulatory factors. No apparent sex-related differences in cytokine levels were observed (Fig. 1B,C; triangle = female, circle = male). Similarly, most differentially expressed cytokines did not show age-dependent correlations. However, CCL2 and PLGF exhibited condition-specific divergent age CCL2 declined with age in controls but increased in BD, whereas PLGF increased with age in controls but decreased in BD (Fig. 1D). Leptin showed a consistent negative correlation with age in both groups. Statistical information for cytokine-age correlations, including group-specific significance values, is provided in Supplementary Table 3.
To determine whether immune cell abundance at the ChP epithelium was altered in BD, we selected representative subsets of BD and control samples based on their cytokine expression profiles, prioritizing those near the mean expression of upregulated immune recruitment signals. Cells exhibiting chromogenic immunostaining for CD45 as well as nuclear hematoxylin staining and positioned adjacent to the epithelium and not within vessels, were taken to be leukocytes (Fig. 2A). Leukocyte counts were normalized to the epithelial perimeter, and no differences were observed between BD and control samples (Fig. 2B). We performed immunofluorescence staining for IBA1 and quantified positive cells with clearly identifiable macrophage-like morphology, defined as ramified or amoeboid cells with visible processes. Given prior reports describing variability in IBA1^+^ density and morphology across brain regions in postmortem tissue from individuals with BD (Sneeboer et al., 2019; Giridharan et al., 2020; Petrasch-Parwez et al., 2020; Hercher et al., 2014) quantification was limited to cells with clearly identifiable macrophage-like morphology, defined as ramified or amoeboid cells with visible processes (Fig. 2C). Normalized IBA1^+^ cell counts did not differ between BD and control samples (Fig. 2D).
We used Novogene for bulk sequencing of RNA extracted from ChP samples in-house. Principal component analysis (PCA) of variance-stabilized expression data revealed that RNA integrity (RIN) was the dominant driver of variance along PC1, explaining 69.8% of PC1 variance (p = 0.0003) in the full dataset. Three samples with RIN < 5 (samples 2, 6, and 19) were identified as outliers and excluded. Following their removal, RIN no longer significantly associated with PC1 (p = 0.087, 24.7% of PC1 variance), while Condition emerged as a contributor to PC2 (Fig. 3A, Supplementary Table 4). To assess tissue specificity, we examined the expression of canonical marker genes associated with ChP, meninges, and neurons. Heatmap of rlog-transformed counts revealed strong expression of choroid plexus-enriched genes (e.g., TTR) and low expression of meningeal (e.g., FOXC1) and neuronal (e.g., RBFOX3, a.k.a. NeuN) markers, suggesting minimal contamination from adjacent tissue. Differential expression analysis identified a limited set of genes differing between BD and control groups (Fig. 3C). Notably, HMOX1 and MT2A, both associated with oxidative stress and metal ion regulation, were upregulated in BD. Additional upregulated transcripts included RGS9BP, ERRFI1, ARRDC3, and IER3, genes involved in signal regulation and cellular stress pathways. In contrast, NELL2, a secreted glycoprotein implicated in neuronal differentiation and maintenance, was decreased in BD. Six of the 43 differentially expressed genes were mitochondrially encoded, including subunits of the NADH dehydrogenase complex (ND2, ND5, ND6) and mitochondrial tRNAs (MT-TC, MT-TL1, MT-TP), suggesting possible disruption of respiratory function. Other affected genes included HILPDA, involved in hypoxia-induced lipid storage, THBS1, a matricellular protein linked to inflammation and extracellular matrix remodeling, and FOXO3, a transcription factor responsive to oxidative stress. RT-qPCR validation of selected targets confirmed the direction of change observed by bulk sequencing (Fig. 3D). These findings suggest that ChP gene expression differences in BD reflected stress response, mitochondrial activity, and neurotrophic signaling.
To better understand the cellular origin of inflammation-related differentially expressed genes (DEGs), we compared them to a ChP single-cell RNA sequencing dataset from a mouse model of CNS inflammation induced by intracerebroventricular LPS administration (Xu et al., 2024). Several BD DEGs showed cell-specific expression in the inflammatory model (Fig. 3E). Specifically, HMOX1 was expressed by inflammatory macrophages, fibroblasts, and monocytes; IER3 was upregulated in inflammatory epithelial and fibroblast cells; and MT2A was broadly expressed but most prominent in inflammatory fibroblasts, with additional expression in epithelial cells and resident macrophages. These observations suggest that some of the DEGs identified in BD ChP reflect a cell-specific inflammatory signature involving macrophages, fibroblasts, and epithelial cells.
Gene set enrichment (GSE) analysis using the Hallmark collection revealed transcriptional changes in BD ChP consistent with immune activation, cellular stress, metabolic reprogramming, and loss of cell junction components (Fig. 4A). Immune-related gene sets were among the most strongly positively enriched, including TNFα signaling via NFκB, IL6-JAK-STAT3 signaling, IL2-STAT5 signaling, TGF signaling, and inflammatory response, indicating coordinated activation of cytokine signaling pathways and transcriptional programs linked to both innate and adaptive immune responses. In contrast, interferon alpha response and allograft rejection sets were significantly negatively enriched, suggesting suppression of antiviral and antigen-presentation-related transcriptional programs. Positive enrichment of gene sets such as hypoxia, unfolded protein response, p53 pathway, apoptosis, and DNA damage response programs reflect cellular stress. Positive enrichment of glycolysis, heme metabolism, and coagulation gene sets suggests a shift in metabolism and vascular signaling. Gene sets linked to epithelial organization and remodeling, including epithelial-mesenchymal transition, TGFβ signaling, myogenesis, and KRAS signaling down, were also positively enriched, whereas the apical junction gene set was strongly negatively enriched, consistent with inflammation related loss of barrier integrity. Together, these results demonstrate that BD choroid plexus exhibits simultaneous immune activation and stress responses, metabolic and structural remodeling, and compromised junctional integrity.
To identify the core transcriptional drivers underpinning the enriched pathways, we performed a leading-edge gene overlap analysis across significantly enriched Hallmark sets (Fig. 4B). This analysis highlighted key genes including MYC, IL6, JUN, CDKN1A, GADD45A, IRF1, DDIT4, and STC2. The frequent appearance of these genes across leading-edge subsets of multiple hallmark sets suggests they represent a robust transcriptional signature associated with immune activation, inflammation, and cellular stress in BD ChP. Notably, although the overall interferon alpha response was negatively enriched, the interferon-associated transcription factor IRF1 and the interferon-stimulated gene ISG20 appeared within the leading-edge subsets of several positively enriched pathways. This observation suggests selective or partial activation of interferon-related pathways despite broader suppression. Together, these results provide a refined, gene-level resolution of the BD ChP transcriptome, highlighting specific molecular components potentially underpinning the complex immune dysfunction observed in BD patients.
To assess whether ChP cytokine protein changes may reflect transcript level, we compared the cytokine array protein log2 fold-changes with corresponding bulk RNA seq mRNA fold-changes. Of the cytokines with altered protein levels in BD, four showed the same directional change in mRNA levels (CCL2, SPP1, OSM, TGF-β3.), whereas four others showed opposite directionality (SCF, CX3CL1, BDNF, LIF) (Fig. 5A). Per-sample concordance heatmaps showed that these concordant and discordant relationships were consistent across BD samples. Whereas all cytokines upregulated at the protein level showed consistent concordance across most BD samples (Fig. 5B), only two downregulated cytokines (OSM and TGF-β3) exhibited a similar pattern, with the remainder more frequently showing discordance or lacking detectable mRNA expression (Fig. 5C).
This study provides the first direct evidence of localized immune dysregulation at the ChP in individuals with BD. We identified a distinct cytokine profile marked by elevated CCL2 and SPP1, chemokines involved in myeloid cell recruitment, alongside reduced levels of trophic and regulatory cytokines such as OSM, IGF-1, CX3CL1, TGFB3, GDNF, LIF, BDNF, SCF, and FGFs. Complementary RNA-seq analysis revealed a modest but coherent set of differentially expressed genes related to oxidative stress, mitochondrial function, and inflammatory signaling. These transcriptional signatures were validated using RT-qPCR and pathway enrichment analyses, supporting the conclusion that the ChP is a site of immune and epithelial disruption in BD.
Immune activity in ChP tissue from BP patients is supported by findings of increased levels of the cytokines CCL2 and SPP1 (Osteopontin), which are associated with monocyte and macrophage recruitment. These data are in line with several previous reports of elevated levels of CCL2 in blood (Modabbernia et al., 2013; Berk et al., 2011) and CSF (Isgren et al., 2017; Isgren et al., 2015). While CCL2 has been implicated in BD, SPP1 remains largely uncharacterized in this context. SPP1 is an extracellular matrix-associated factor secreted by immune cells that is upregulated in activated myeloid states across multiple neuroinflammatory contexts and has been linked to myeloid cell survival, tissue remodeling, and inflammation (Rabenstein et al., 2016; De Schepper et al., 2023; Rosmus et al., 2022; Zawadzki et al., 2025). Recent evidence suggests SPP1^+^ microglia play a fundamental role in maintaining structural integrity at tissue boundaries (Lawrence et al., 2024), raising the possibility that elevated SPP1 in BD reflects a reparative immune response at the ChP. Despite this reparative signal, the pro-inflammatory ChP profile was paired with a marked suppression of trophic and regulatory cytokines, including IGF-1, CX3CL1, LIF, and TGF-β3, factors implicated in neuroprotective processes, homeostatic immune signaling, and tissue maintenance, suggesting impaired inflammation resolution mechanisms at the blood-CSF barrier in BD.
Consistent with the presence of pathological immune activation, our bulk RNA-seq analysis of ChP tissue revealed differential expression of genes linked to oxidative stress, inflammation, and mitochondrial function. The presence of similar molecular changes reported across other brain compartments supports the hypothesis that dysregulated cellular stress responses are a general feature of BD brains (Steckert et al., 2010; Pfaffenseller et al., 2014; Viswanath et al., 2015). Notably, the MT2A metal-binding protein involved in cellular stress regulation was upregulated in the ChP, as it has been shown to be in the brain parenchyma (Logotheti et al., 2013; Ling et al., 2016). Two additional differentially expressed genes (DEGs), HMOX1 and IER3, were elevated in BD ChP. HMOX1 encodes heme oxygenase 1, a key stress-response enzyme that regulates oxidative damage and modulates immune activity. Although it has not been directly linked to BD, HMOX1 upregulation has been reported in major depression (Simon et al., 2021), implicating it in the depressive phase of BD. IER3 is rapidly induced by cellular stress and inflammatory stimuli; while not previously associated with BD, it is known to mediate stress-resistance pathways and cytokine responses in other inflammatory contexts (Arlt and Schafer, 2011). The concurrent upregulation of these inflammation-responsive genes indicates that local stress and immune signaling are active in the BD ChP, potentially contributing to transcriptomic changes possibly independent of marked immune cell infiltration.
Pathway-level analysis using the Hallmark gene set collection also pointed toward broad immune, metabolic, and, perhaps, structural remodeling in BD ChP. Enrichment of TNFα/NF-κB, IL-6/JAK-STAT3, IL-2/STAT5, interferon, and inflammatory response pathways indicates coordinated innate and adaptive immune activation, as previously reported in BD and depression (Misiak et al., 2020; Pfaffenseller et al., 2014; Viswanath et al., 2015; Brietzke and Kapczinski, 2008). Enrichment of stress-related pathways, including hypoxia, unfolded protein response, apoptosis, and p53 signaling, reflects oxidative and inflammatory pressure (Viswanath et al., 2015; Simon et al., 2021). Glycolysis, coagulation, and heme metabolism pathways further support a shift in metabolic and vascular states (Disler et al., 2019).
Our detection of concordant RNA and protein upregulation of CCL2 and SPP1 within the ChP is consistent with local increases in production within the ChP and epithelial cells are the most likely source of elevated cytokines. Since monocytes and macrophages secrete CCL2 and SPP1 in chronic inflammatory conditions (Galione and Davis, 2018) we cannot exclude a contribution from circulating immune cells. Nonetheless, we did not detect increased CD45^+^ or Iba1^+^ cells, leaving intrinsic ChP activation as the most likely driver of the observed inflammatory profile.
Assignment of inflammatory markers and stress-response genes to specific cell types is difficult without spatial transcriptomic or proteomic resolution. There are currently no high-quality human single-cell reference atlases available. Therefore, we compared the DEGs observed here with a mouse ChP single-cell RNA-seq inflammatory model (Xu et al., 2024) to assess whether these changes show a conceptual overlap with specific activated ChP cell states. We anticipated relevance to humans because human and mouse ChP share many transcriptional programs. Although there are species-enriched differences at baseline (Janssen et al., 2013), most DEGs identified here have not been shown to be strongly species-specific, with the notable exception of NELL2, a secreted glycoprotein enriched in human but not mouse ChP (Janssen et al., 2013; Shaker et al., 2022), suggesting a primate-specific marker of ChP dysfunction in BD. Future studies leveraging human single-cell profiling of the ChP in mood disorders will help resolve specific cell-type inflammatory states.
Interpretation of these immune alterations must be considered in light of clinical context, as our BD cohort died by suicide during acute depressive episodes. Therefore, we cannot discern the relative contribution of the presence of the severe depressive state at the time of death to our findings (Deep-Soboslay et al., 2011). Inflammatory alterations have been reported in major depressive disorder (Bravi et al., 2024). Moreover, a previous meta-analysis of BD studies found that CCL2 was elevated during the depressive phase but not during other mood states (Misiak et al., 2020) and it has been suggested that this association underlies general depression (Curzytek and Leskiewicz, 2021; Eyre et al., 2016; Leighton et al., 2018). Further, suicidality in individuals with mood disorders has been associated with inflammatory alterations across peripheral blood, CSF, and brain tissue, even after adjustment for depression and medication exposure (Zhao et al., 2019; Janelidze et al., 2011; Black and Miller, 2015; Garcia-Gutierrez et al., 2025).
The presence of ChP pathology in BD raises the question of whether barrier dysfunction contributes to disease mechanisms. Prior work in depression has identified cytoskeletal and extracellular matrix-related transcriptional alterations in the choroid plexus and proposed that such structural remodeling may compromise blood-CSF barrier properties (Turner et al., 2014). We were unable to visualize tight junction proteins by immunostaining, likely due to protein degradation inherent in the nature of the samples. However, pathway-level analysis of bulk RNA-seq data revealed positive enrichment of epithelial–mesenchymal transition and TGF-β signaling programs, together with less apical junction gene sets, consistent with active remodeling and reduced barrier cohesion. In our cohort, CCL2 decreased with age in controls but increased in BD, a divergence that may reflect progressive remodeling of the ChP immune landscape over time in BD.
Limitations of our study include constraints on sample size by tissue availability and biological covariate matching. Small cohorts in high-dimensional settings risk both overfitting of limited data as well as false negatives due to wide confidence intervals. BD subjects in this study were on psychotropic medications at the time of and in the months preceding death (Supplementary Table 1). Antipsychotic treatment has been associated with suppression of inflammatory signaling, although evidence for these effects is derived largely from non-BD mood disorder cohorts (McNamara and Lotrich, 2012; Miller and Raison, 2016). Importantly, antipsychotics were frequently administered in combination with antidepressants, anticonvulsants, and benzodiazepines, precluding isolation of medication-specific immune effects. Though applicable to few subjects in this cohort, antidepressants such as selective serotonin reuptake inhibitors (SSRIs) have been shown to modulate cytokine and chemokine expression (Su et al., 2015; Tynan et al., 2012; Rahimian et al., 2021). Lastly, while individuals with active inflammatory or immune disorders were largely excluded from this study, it remains possible that undiagnosed medical comorbidities contributed to inter-individual variability in our observations. Collectively, these factors limit attribution of specific immune alterations to medication exposure and should be considered when interpreting effect sizes.
Together, our results position the ChP as a neuroimmune interface affected in BD and highlight barrier-associated signaling as a potential contributor to peripheral-central communication in psychiatric disease. Future studies assessing ChP barrier function such as secretory activity and structural integrity using experimental models, and a direct comparison of ChP molecular signatures across mood disorders and clinical conditions are needed to clarify the functional and disease-specific relevance of the alterations identified in this study.