Trauma Exposure, PTSD, and Methylation of the Blood Brain Barrier Claudin-5 Gene

Abstract

Posttraumatic stress disorder (PTSD) is associated with early onset of neurological conditions, but the mechanism by which PTSD relates to diseases of the central nervous system is unclear. One possibility is that PTSD perpetuates breakdown of the blood brain barrier (BBB), allowing for bidirectional passage of molecules across the periphery and central nervous system that promote neuropathology. Preclinical studies have implicated claudin-5 (CLDN5), a protein integral to the integrity of the BBB tight junctions, in the pathogenesis of depression. Based on this, we evaluated if trauma exposure and PTSD related to CLDN5 epigenetics in blood among 1,311 trauma-exposed individuals (primarily Veterans) and in the brain tissue from 100 decedents. Three (out of 19) CLDN5 DNA methylation (DNAm) probes, cg00804504, cg17411190, and cg21872764, were significantly associated with trauma exposure or PTSD severity after multiple testing correction in blood. The latter two probes also showed association with PTSD diagnosis in ventromedial prefrontal cortex. The most strongly associated DNAm probe, cg21872764, also evidenced associations with the neuropathology biomarker neurofilament light in plasma. CLDN5 expression was strongly associated with estimated proportion of brain endothelial cells. The cross-sectional associations observed in this study highlight the importance of studying the link between traumatic stress and early onset of neuropathology. Future research is needed to test the mechanistic hypothesis that trauma exposure and chronic PTSD alter CLDN5 DNAm, lead to increased BBB permeability and allow for bidirectional passage of neuroinflammatory molecules across the BBB.

1. Introduction

Posttraumatic stress disorder (PTSD) is associated with risk for early onset of neurodegeneration (Miller & Sadeh, 2014), declining cognitive function (Roberts et al., 2022), and dementia (Yaffe et al., 2010). Among older Veterans who receive VA healthcare, PTSD is associated with approximately double the risk for a dementia diagnosis compared to individuals with no lifetime history of PTSD (Yaffe et al. 2010, Logue et al., 2022). These neurological alterations among individuals with PTSD may partially reflect an accelerated cellular aging process (Sugden et al., 2022) whereby the stress of chronic psychiatric symptoms, such as emotional and physiological reactivity, sleep disturbance, and associated poor health behaviors, lay the groundwork for advanced biological aging (Miller, et al., 2018; Bourassa & Sbarra, 2024). It is unclear how psychiatric symptoms impact central nervous system (CNS) structures and processes or how psychopathology-related aberrations in peripheral biomarkers (e.g., inflammation, glucocorticoid reactivity) relate to alterations in the CNS (e.g., neuroinflammation, neurodegeneration) and vice versa. One possibility is that psychiatric stress and associated physiological reactivity perpetuate breakdown of the blood brain barrier (BBB), allowing for bidirectional passage of molecules that promote neuroinflammation and neuropathology (Medina-Rodriguez & Beurel, 2022; Wu et al., 2022; Kealy et al., 2020).

1.1. Claudin-5, the Blood Brain Barrier, and Neurological Disease

The BBB serves to limit molecules from passing between the CNS and the periphery so that only select molecules can cross, which protects the CNS and maintains homeostasis. This is accomplished through a layer of tightly connected endothelial cells aligned along the capillaries of the brain that separate blood vessels from the CNS. The BBB is part of a larger structure referred to as the neurovascular unit, comprised of astrocytes, microglia, mural and endothelial cells, and neurons, which coordinates cellular signaling and allows blood, nutrients, neuroimmune molecules, and CNS waste to cross the BBB (McConnell & Mishra, 2022). The endothelial cells of the neurovascular unit are connected to each other through tight junction proteins which form a physical barrier (akin to a seal) and regulate passive transfer of molecules across it. Several types of claudin and occludin proteins control the membranes at the tight junctions. Chief among these is claudin-5 (CLDN5), which is expressed primarily in brain endothelial cells and is critical to tight junction integrity and resistance to small molecule transfer across the BBB (Greene et al., 2019; Hashimoto et al., 2023).

Numerous factors alter CLDN5 expression, including inflammatory proteins such as tumor necrosis factor alpha (TNFα) and associated nuclear factor kappa B signaling, glucocorticoids, and circadian-related proteins (Greene et al., 2019). Downregulation of CLDN5 is associated with increased BBB leakiness and passage of peripheral immune and inflammatory cells into the CNS, which can be neurotoxic (Greene et al., 2019). CLDN5 downregulation is also associated with movement of central neuropathology biomarkers into the periphery. For example, downregulation of CLDN5 in the brain allows for transfer of amyloid beta (Aβ), a primary indicator of Alzheimer’s disease (Jack et al., 2024), from the CNS into the periphery via the BBB; Aβ itself may autoregulate CLDN5 expression to allow for its clearance from the brain (Keaney et al., 2015). Neurofilament light (NFL), a cytoskeletal protein produced in neuronal cytoplasm, is a marker of axonal injury that is correlated with neuropathological disease (Jack et al., 2024; Yuan & Nixon, 2021). It can be measured in the CNS and is also detectable in plasma and serum (from cerebral spinal fluid; Yuan & Nixon, 2021). NFL is also associated with BBB permeability (Friis et al., 2022) and CLDN5 levels (Li et al., 2021), so its presence in the periphery may be a sign of increased BBB permeability. Another neurology protein with relevance to CLDN5 is glial fibrillary acidic protein (GFAP), which forms part of the astrocyte cytoskeleton structure. GFAP-immunoreactive astrocytes are reduced in concert with alterations in CLDN5 protein levels and organization (Camire et al., 2015; Sántha et al., 2015) and GFAP levels are relevant to dementia risk (Kim et al., 2023; Gonzales et al., 2022). Alterations in Aβs, NFL, and GFAP may reflect, in part, the underlying role of CLDN5 in BBB disruption and neurological disease. Consistent with this, reduced CLDN5 expression and associated disruption to the BBB at the tight junctions is associated with numerous neurological conditions including traumatic brain injury sequela, Alzheimer’s disease, cognitive decline, multiple sclerosis, epilepsy, and stroke (Greene et al., 2019; Hashimoto et al., 2023).

1.2. Claudin-5 and Stress-Related Psychopathology

In addition to associations with neurological conditions, CLDN5 is also relevant to stress-related psychopathology. Much of the evidence for this comes from rodent models of depression. Chronic social stress is associated with reduced cldn5 expression in the nucleus accumbens of stress-susceptible male mice and in the prefrontal cortex of stress-susceptible female mice, and this co-occurs with depression-like phenotypes and increased BBB permeability (Menard et al., 2017; Dion-Albert et al., 2022). Both the reduced cldn5 expression in the nucleus accumbens and the depression-like phenotype can be reversed with lengthy administration of the antidepressant imipramine (Menard et al., 2017). Rodent studies also suggest that acute and chronic stress are related to reduced cldn5 expression in the hippocampus, loss of regional BBB integrity, and behavioral and cognitive impairments (Menard et al., 2017; Sun et al., 2024; Ni et al., 2022). Experimentally manipulated hippocampal cldn5 downregulation was also associated with increased depression-like behaviors in rodents and reversed with long term administration of fluoxetine (Sun et al., 2024). Stress-related cldn5 downregulation may allow for inflammatory molecules to pass into the CNS: experimentally stressed mice with selectively downregulated cldn5 expression showed depression-like phenotypes and infiltration of the inflammatory molecules interleukin (IL)-6 and TNFα from the periphery into the nucleus accumbens (Menard et al., 2017) and hippocampus (Sun et al., 2024), respectively. Secretion of TNFα and IL-6 by microglia may also be sufficient to damage the BBB via reduced expression of CLDN5 per in vitro studies (Camire et al., 2015). Together, this raises the possibility of both central and peripheral inflammation leading to tight junction damage.

Postmortem human studies also suggest that CLDN5 expression is substantially reduced in the nucleus accumbens from untreated depressed men and women (Menard et al., 2017), and in the ventromedial prefrontal cortex (vmPFC) of women with depression (Dion-Albert et al., 2022). Similarly, postmortem studies suggest that CLDN5 protein is reduced in the hippocampi of men and women with depression and schizophrenia, with greater reductions in CLDN5 expression as a function of more years of chronic psychiatric symptoms (Greene et al., 2020).

1.3. CLDN5 Genotypes and DNA Methylation

A handful of studies have examined CLDN5 candidate genotypes in association with psychiatric outcomes. The minor allele of the rs10314 CLDN5 variant may increase risk for schizophrenia (Greene et al., 2018; Sun et al., 2004). A UK Biobank candidate SNP study with over 275k individuals found that CLDN5 variant rs885985 interacted with an IL-6 variant and self-reported life stress to predict depression (i.e., a 3-way interaction effect), and this was replicated in a second cohort (Gal et al., 2023). Associations between these genotypes and PTSD risk have not been evaluated in candidate gene studies, though the most recent genome wide association study of PTSD featuring over 1.2 million participants did not report a significant association with either variant (ps ≥ .20; Nievergelt et al., 2024).

Given the evidence for CLDN5 genotypes and expression in BBB integrity and stress-associated psycho- and neuropathology, it is reasonable to wonder about the role of DNA methylation (DNAm) in this process, particularly among aging populations. This is because DNAm is responsive to the environment and physiological processes (Tobi et al., 2009), known to change with age (Christensen et al., 2009), and is a primary driver of gene expression. It may serve as the environmentally sensitive intermediate process by which substantial stressors like chronic PTSD symptoms exert effects on BBB.

To our knowledge, the only CLDN5 DNAm study relevant to understanding neuropathology stems from a longitudinal cognitive trajectory and brain donation study with a cohort of over 600 individuals (Hüls et al., 2022). That study found that cognitive trajectory was strongly (at the genome-wide significance level) associated with postmortem DNAm at two CLDN5 loci in the dorsolateral prefrontal cortex (dlPFC). The probes, cs16773741 and cg05460329, were associated with greater cognitive decline over time among individuals with and without obvious neuropathology. Though the results supported a role for CLDN5 DNAm in cognitive outcomes, associations between DNAm and psychopathology were not examined.

1.4. Study Aims & Hypotheses

Given evidence for a role for CLDN5 in susceptibility to depression following stress (Menard et al., 2017), and phenotypic and genetic similarities between PTSD and depression (Dalvie et al., 2021; Ringwald et al., 2023), our aim was to examine, for the first time, the association between PTSD and CLDN5 epigenetics. Our primary hypothesis was that chronic PTSD symptoms would be associated with alterations in CLDN5 DNAm. We first examined this in blood DNAm in a cohort of over 1300 trauma-exposed individuals who were assessed for PTSD using structured diagnostic interviews. We also evaluated if CLDN5 candidate genotypes accounted for associations between PTSD and CLDN5 DNAm. We then tested our primary hypothesis in postmortem brain tissue samples from 100 donors by examining associations between PTSD diagnosis and CLDN5 DNAm in vmPFC, dlPFC, and motor cortex. The postmortem tissue afforded the opportunity to examine associations between CLDN5 DNAm and expression and between CLDN5 expression and estimated brain endothelial cell proportions.

Our second hypothesis was that CLDN5 DNAm would be associated with peripheral measures of neuropathology and inflammation. We examined cross-sectional associations between CLDN5 DNAm and the neuropathology biomarkers Aβ−42/Aβ−40 (i.e., their ratio), NFL, GFAP, and the inflammatory biomarkers IL-6 and TNFα, all measured in blood in our trauma-exposed cohort. As reviewed above, these were selected based on their relevance to CLDN5 (Keaney et al., 2015), aging and neurodegeneration (Jack et al., 2024), and PTSD (Wolf et al., 2024; Miller et al., 2024; Peruzzolo et al., 2022). Evidence for these associations could be a reflection of a leaky BBB allowing for neuropathology biomarkers to pass from the brain to blood.

2. Methods

2.1. Participants and Procedures

Participants (N = 1311) were drawn from two cohorts and the data were combined for analyses. The National Center for PTSD (NCPTSD) cohort (n = 752) included trauma-exposed male (60.6%) and female Veterans from a range of war eras (i.e., WWII through post-9/11) and a subset of their live-in partners with trauma exposure (n = 186) who underwent identical procedures (Wolf et al., 2024, Logue et al., 2013). On average, the NCPTSD cohort was in their early 50s (range: 19 to 75 years). The Translational Research Center for TBI and Stress Disorders (TRACTS) cohort (n = 559) included male (90.3%) and female post-9/11 combat-exposed Veterans with a mean age in their early 30s (McGlinchey et al., 2017). Both cohorts included overlapping assessment and blood draw procedures. The demographic characteristics of the combined cohort are listed in Table 1.

Table 1.

Demographic Characteristics of the Cohorts

	Human Subjects (N = 1311)		Brain Bank (N = 100)
Variable	% (n)	M (SD)	% (n)	M (SD)
Sex (male)	76.1 (998)		60.0 (60)
Age		44.24 (13.68)		43.70 (12.93)
Veteran	85.8 (1125)		22.0 (22)
Ethnicity: Hispanic or Latino/a	16.2 (213)		NA
Race
White	76.8 (1007)		76.0 (76)
Black	13.0 (171)		23.0 (23)
American Indian	5.0 (66)		NA
Asian	1.8 (23)		NA
Other	8.6 (113)		1.0 (1)
Current PTSD Diagnosis	48.2 (632)		40.0 (40)
Current PTSD Severity		.33 (.22)		NA
Lifetime Trauma Count		3.14 (2.40)		NA

Note. Ethnicity and race were self-reported and options were not mutually exclusive; percentages were combined into “other” for reporting purposes due to small cell sizes for some response options such as Hawaiian/Pacific Islander, other, and unknown. PTSD severity reflects the harmonized PTSD severity score that captures symptom severity ratings a percentage of the maximum possible severity score. Some characteristics were not available in the brain bank cohort and are labeled as NA (not applicable).

Across the two cohorts, participants were recruited from VA PTSD clinics, research recruitment databases and flyers, and military and Veteran community events. Exclusion criteria included acute substance intoxication and acute psychotic or safety (suicidal and homicidal) concerns. Additionally, the TRACTS protocol excluded Veterans with a neurological or cognitive disorder diagnosis other than those related to traumatic brain injury (TBI). Participants in both cohorts completed psychological interviews, self-report surveys, blood draw, and physiological measurements. The studies were approved by the VA Boston Healthcare System IRB and all participants provided written informed consent.

2.2. Measures

In the NCPTSD cohort, current PTSD was assessed with the Clinician Administered PTSD Scale for DSM-IV (CAPS; Blake et al., 1995) or DSM-5 (CAPS-5; Weathers et al., 2018), based on the prevailing version of the DSM at the time of study participation. These interviews were videotaped and approximately 25% were reviewed by an independent rater. Intraclass correlation coefficients for PTSD symptom severity totals across two timepoints were r ≥ .84 and diagnostic agreement for current PTSD diagnoses were K ≥ .87. For the TRACTS cohort, the CAPS for DSM-IV was administered and all interviews were audio recorded; 23 were reviewed by an independent rater which yielded intraclass correlation r = .92 for PTSD symptom severity and K = .68 for PTSD diagnostic status. To harmonize current PTSD severity scores (sum of all CAPS PTSD items) across the two versions of the measure (which have different possible ranges), we calculated a standardized score which reflected the total symptom severity score as a fraction of the maximum possible total score on each measure. These scores have a potential range of 0 to 1. Trauma exposure was assessed with the self-report Traumatic Life Events Questionnaire (TLEQ; Kubany et al., 2000) which assesses exposure to 23 types of traumatic experiences. We tabulated the number of different types of traumatic events endorsed for use in our analyses. Comorbid major depressive disorder (MDD) diagnoses were assessed with the Structured Clinical Interview for DSM-IV (First et al., 1994) or DSM-5 (First et al., 2015) Disorders.

2.3. Biomarkers

2.3.1. DNA and DNAm.

Fasting morning peripheral blood samples were drawn into 10 mL EDTA tubes which were centrifuged the same day. Plasma, serum, and buffy coat samples were stored at −80° C until ready for analysis. DNA was extracted from buffy coat for genotype and DNAm ascertainment using Qiagen reagents on a Qiagen Gentra AutoPure machine. Genotypes were interrogated on the Illumina HumanOmni2.5–8 Beadchip. DNAm was obtained on the Illumina Infinium EPIC Beadchip. DNA and DNAm plates were balanced for sex and PTSD diagnostic status. Details of the genotyping data quality control (QC) and imputation procedures can be found in the Supplementary Materials. We computed ancestry (for use in the DNAm analyses) and ancestry substructure (within the EUR ancestry subset, for use in the genotype analyses) principal components (PCs) from 100,000 randomly chosen genotypes with minor allele frequency (MAF) of ≥ 5% (Nievergelt et al., 2019).

We followed a pipeline developed by the PGC PTSD Epigenetics Working Group (Ratanatharathorn et al., 2017) and its updated version at https://github.com/PGC-PTSD-EWAS/EPIC_QC to process DNAm data (Supplementary Materials). Proportional estimates of six DNAm-based cell types (B cells, CD4+T cells, CD8+T cells, natural killer cells, monocytes, and neutrophils) were computed using the IDOL algorithm (Koestler et al., 2016) in the Bioconductor package EpiDISH (Teschendorff et al., 2017) in R. A DNAm-based smoking score was calculated by taking the product of the methylation M values at 39 smoking-associated CpG sites and the effect size estimates of their association with smoking pack years (Li et al., 2018).

2.3.2. Simoa Neuropathology & Inflammatory Biomarkers.

The following Simoa analytes were analyzed from peripheral plasma samples in the NCPTSD and TRACTS cohorts: IL-6, TNFα, GFAP, NFL, Aβ−42, and Aβ−40 (the latter two were combined into a single variable reflecting the ratio of Aβ−42 to Aβ−40; Lewczuk et al., 2004). As previously described (Wolf et al., 2024; Miller et al., 2024), Simoa assays for the NCPTSD cohort were obtained directly from the Quanterix Accelerator Lab (Quanterix Corporation, Billerica, MA; see Supplementary Materials for QC and other details). In the TRACTS cohort, the plasma samples were analyzed (in duplicate) on an in-house HD-1 analyzer (Quanterix, Billerica, MA; see Supplementary Materials).

2.4. Postmortem Brain Bank Cohort

Data from the VA National PTSD Brain Bank cohort were from n = 117 donors originally collected by the Lieber Institute for Brain Development at Johns Hopkins University (Friedman et al., 2017; Mighdoll et al., 2018). Left hemisphere postmortem brains were acquired and three brain regions relevant to PTSD and aging were available to our group: dlPFC, vmPFC, and motor cortex. We have previously described the methods for obtaining these regions and ascertaining genotypes, DNAm, and RNA sequence data (Wolf et al., 2021; Zhao et al., 2022). This cohort excluded individuals with neurodegenerative disease, neuritic pathology, or severe TBI as determined by board-certified neuropathologists. Determination of psychiatric diagnoses was based on medical record review and next-of-kin interviews performed by board-certified neuropathologists. The interviews included the PTSD Checklist for DSM-5 adapted for postmortem use, the MINI International Neuropsychiatric Interview 6.0, and the Liber Psychological Autopsy Interview (Mighdoll et al., 2018). Two independent board-certified psychiatrists reviewed the records and psychiatric diagnostic determinations and made confidence ratings for the assigned diagnoses using a 1–5 scale. PTSD cases had confidence ratings of 3 or greater.

DNAm was obtained on the Illumina Infinium MethylationEPIC beadchip with processing and QC procedures as described for plasma cells and in previous publications (Logue et al., 2020). The proportion of neurons was estimated from the methylation data using the CETS package (Guintivano et al., 2013). RNA was obtained from 25 mg of tissue via Qiagen RNeasy Fibrous Tissue Minikit. Illumina TruSeq Stranded total RNA kit with globin depletion and a Hiseq 2500 for paired-end 75bp reads for library sequencing. RNA integrity values were available for a subset of samples and found to be acceptable. Quality surrogate variables (qSVs) were generated to assess RNA degradation using the Bioconductor package sva (Leek et al., 2012). Samples were excluded if < 50% uniquely mapped reads or if they were outliers per evaluation of log-transformed counts from the regularized log transformations in DESeq2 (Love et al., 2014). Estimated proportional cell types (astrocytes, endothelial cells, microglia, mural cells, neurons, oligodendrocytes, and red blood cells) were generated for each brain region per BrainInABlender (Hagenauer et al., 1995).

We used the largest possible dataset for each set of brain bank analyses. For analyses examining associations between PTSD and CLDN5 DNAm across brain regions, decedents were excluded if they had a bipolar disorder diagnosis (n = 5). Among the remaining 112 donors, n = 42 had PTSD (n = 35 of this group had comorbid MDD), n = 41 had MDD without PTSD and n = 29 were controls. Given our efforts to replicate associations with PTSD from our living cohort, we excluded the cases with MDD without PTSD. We further excluded two individuals missing antidepressant use data and one motor cortex sample that failed DNAm QC, yielding a final sample size of n = 69 with dlPFC and vmPFC data, and n = 68 for motor cortex data for the DNAm and PTSD analyses.

For the expression analyses, we focused on associations between biological variables (e.g., CLDN5 expression with estimated cell types, CLDN5 DNAm with expression) and did not include psychological diagnoses in the analyses. Therefore, we included all subjects regardless of MDD status (bipolar cases were still excluded). RNA was available for a subset of the non-bipolar donors (n = 92–96 across brain regions). Individuals whose RNA data failed QC procedures (n = 2 – 5 across brain regions) or had missing genotypes for determining ancestry (n = 1) were removed from expression analyses. This yielded a sample size of n = 93 for dlPFC, 86 for vmPFC, and 89 for motor cortex. Collectively, data from 100 donors were included in either DNAm and PTSD or expression-related brain bank analyses, and their demographic characteristics are listed in Table 1.

2.5. Data Analyses

All analyses involved multiple linear regression models conducted in SPSS v 29.0.1.0 (IBM Corporation) or R version 4.0.5. We first evaluated the 19 CLDN5 DNAm probes (the outcome variable) on the EPIC chip that surpassed QC in association with current PTSD symptom severity in our largest possible cohort (N = 1311). These regressions covaried for age, sex, five WBCs (CD8 and CD4-T cells, monocytes, b cells, and natural killer cells), and the top 3 ancestry PCs. We employed a false discovery rate (FDR; Benjamini & Hochberg, 1995) adjusted p-value (p_-adj) threshold of p_-adj < .05. Sensitivity analyses which further covaried for smoking (using the DNAm smoking score), lifetime trauma exposure, and MDD were conducted for models with significant PTSD associations with a given probe. Next, we examined the candidate CLDN5 variants rs10314 and rs885985, for their associations with the PTSD-associated CLDN5 loci to determine if they accounted for associations between PTSD and CLDN5 DNAm (in regressions covarying for age, sex, WBCs, and 3 ancestry substructure PCs). This was conducted in the EUR subset (n = 873) due to concerns of population stratification. We also examined the SNP associations with trauma exposure and PTSD severity in analogous regression models. To test our hypothesis that CLDN5 DNAm alterations would relate to increased levels of neurology and inflammation biomarkers in blood, we examined CLDN5 DNAm loci (that were significantly related to PTSD) in association with peripheral IL-6, TNFα, GFAP, Aβ−42/Aβ−40, and NFL while covarying for sex, age, and WBCs and employing FDR p-value adjustments for the five Simoa biomarkers (n = 1097 of mixed ancestry due to missing Simoa data).

Finally, we examined data derived from postmortem brain tissue to test for replication of results. In the three brain regions, we tested associations between PTSD diagnosis (versus controls) and the CLDN5 loci that were identified as being significantly associated with PTSD in blood, covarying for age at death, sex, postmortem interval (PMI), 3 PCs, proportion of neurons, and antidepressant use (Menard et al., 2017). For completeness, we also tested for association between PTSD and all 19 CLDN5 DNAm probes on the chip in each brain region. We extended the analyses we could conduct in blood by examining the PTSD-associated blood CLDN5 DNAm loci in association with CLDN5 expression in the three brain regions, covarying for age at death, sex, PMI, 3 PCs, 3 qSVs, sequencing run ID, and RNA-derived brain cell type estimates. We examined CLDN5 expression in each brain region in association with the 7 types of estimated cell types to test hypothesized associations between CLDN5 and endothelial cells, covarying for age at death, sex, PMI, 3 PCs, 3 qSVs, and sequencing run ID, and correcting across the cell types and brain regions using FDR.

3. Results

3.1. Peripheral CLDN5 DNAm and Current PTSD Severity

Analyses examining associations between current PTSD symptom severity and each of the 19 CLDN5 probes derived from blood samples revealed 2 nominally significant and an additional 3 FDR-corrected significant associations (Table 2). The associations that surpassed the multiple testing correction were between PTSD severity and cg21872764 (β = .086, p = .002, p-_adj = .032), cg00804505 (β = −.080, p = .004, p-_adj = .032), and cg17411190 (β = .068, p = .005, p-_adj = .032). These three CpG sites were not highly correlated with each other (strongest r = .262; Table S1), suggesting that their associations with PTSD severity represented independent effects.

Table 2.

Associations between Current PTSD Severity and CLDN5 DNA Methylation in Blood (N = 1311)

Probe	B (unstd)	β	p	p -adj
cg00189989	.038	.041	.144	.249
cg00804504	−.094	−.080	.004	.032
cg00811132	.074	.050	.061	.193
cg04463638	−.029	−.027	.334	.428
cg05460329	−.045	−.058	.038	.144
cg05498726	−.028	−.025	.360	.428
cg06315607	−.017	−.012	.660	.660
cg06340942	−.055	−.044	.110	.232
cg09092054	−.036	−.035	.187	.232
cg09446908	−.074	−.046	.103	.232
cg11450827	.038	.046	.110	.232
cg13114849	.025	.043	.123	.234
cg14553765	−.039	−.032	.254	.371
cg16773741	−.066	−.059	.032	.144
cg17411190	.098	.068	.005	.032
cg17577122	−.028	−.018	.523	.585
cg17583256	.018	.015	.582	.614
cg20486569	.036	.022	.341	.428
cg21872764	.128	.086	.002	.032

Note. Covariates included in the model were sex, age, proportional white blood cell types, and the first 3 ancestry principal components. Significant associations are shown in bold font. Unstd = unstandardized; adj = adjusted.

To test the three associations between PTSD severity and the CpG loci for potential confounds, we added lifetime trauma exposure, DNAm smoking score, and MDD diagnosis to each model (each in separate regressions). For the follow-up models involving cg00804505 and cg17411190, none of these covariates were significantly associated with the probe, while PTSD severity remained significantly associated. When adding lifetime trauma exposure to the model involving cg21872764, the association between the probe and PTSD severity was no longer significant (p = .137) while trauma exposure was strongly related to DNAm at this locus (β = .109, p = .000185). Additional analysis revealed that the association between trauma exposure and cg21872764 (and only this probe) would withstand an FDR correction across all 19 probes (p_-adj = .000589). This association between trauma exposure and the probe also remained significant with the DNAm smoking score and MDD in the model. We examined CLDN5 candidate SNPs (rs10314 and rs885985) for their associations with the three trauma/PTSD-associated DNAm probes in the EUR cohort (n = 873). These variants are 158 to 3254 bp away from the 3 DNAm loci. The variant rs10314 was associated with two of the PTSD-related probes: cg00804504 (p_-adj = 5.08E-34) and cg17411190 (p_-adj = 0.000125; Table S2). Similarly, rs885985 was associated with all 3 PTSD-associated probes: cg00804504 (p_-adj = 5.02E-18), cg17411190 (p_-adj = 7.69E-18), and cg21872764 (p_-adj = 0.0000908). Neither SNP was associated with lifetime trauma exposure (ps = .094 to .249) nor current PTSD severity (ps = .081 to .129), covarying for age, sex, and 3 ancestry substructure PCs. Though the SNPs were associated with the 3 PTSD-associated DNAm probes, PTSD severity and trauma exposure were still significantly associated with their respective DNAm probes with the SNPs included in the models (Table S3).

3.2. CLDN5 DNAm and Biomarkers of Inflammation and Neuropathology

We examined cross-sectional associations between the 3 FDR-significant trauma/PTSD-associated loci and peripheral IL-6, TNFα, GFAP, Aβ−42/Aβ−40 ratio, and NFL (covarying for sex, age, and WBCs). There was an association between the peak probe cg21872764 and NFL (β = .066, p = .011, p_-adj = .033; Table 3). A follow-up cross-sectional mediation path model revealed an indirect association between trauma exposure and NFL via cg21872764 (indirect β = .009, p = .015; Figure 1; Supplementary Materials).

Table 3.

Cross-sectional Associations between Peripheral CLDN5 DNA Methylation and Inflammation and Neuropathology Biomarkers (n = 1097)

	IL-6		TNFα		GFAP		Aβ−42/Aβ−40		NFL
Probe	β	p	β	p	β	p	β	p	β	p
cg00804504	−.041	.160	−.006	.843	−.043	.128	.043	.100	−.043	.092
cg17411190	−.021	.538	−.003	.921	.058	.085	−.040	.192	−.012	.680
cg21872764	−.022	.462	.031	.296	−.011	.721	.025	.352	.066	.011 a

Note. Covariates included sex, age, and proportional white blood cells. Significant associations are shown in bold font. IL = interleukin; TNFα = tumor necrosis factor alpha; GFAP = glial fibrillary acidic protein; Aβ = amyloid beta; NFL = neurofilament light.

^aThe FDR corrected p-value for this probe (corrected across the 5 biomarkers) was significant at .033.

Figure 1.

shows the results of the mediation path model in which trauma exposure was indirectly associated with NFL levels via DNAm at cg21872764, covarying for potential demographic and methodological confounds.

3.3. Follow-up in Postmortem Brain Tissue

Analyses in the brain bank cohort revealed that two of the three CLDN5 loci that were associated with trauma/PTSD in blood were also associated in brain tissue. PTSD was associated with cg17411190 (B = 0.305, β = 0.369, p = 0.005) and cg21872764 (B = 0.308, β = 0.246, p = 0.034), both in vmPFC (Table 4). The probes showed the same direction of association in both tissues. For completeness, we examined the remaining 16 CLDN5 probes on the chip in association with PTSD, but none were significant after FDR correction across the probes and brain regions (Table S4).

Table 4.

Associations between PTSD Diagnosis and Three CLDN5 DNAm Probes in Postmortem Brain Tissue

	dlPFC (n = 69)			vmPFC (n = 69)			Motor Cortex (n = 68)
CPG	B	β	p	B	β	p	B	β	p
cg00804504	−0.338	−0.285	0.055	−0.109	−0.104	0.494	0.035	0.030	0.842
cg17411190	0.114	0.129	0.381	0.305	0.369	0.005	0.196	0.238	0.148
cg21872764	0.243	0.196	0.137	0.308	0.246	0.034	0.144	0.129	0.383

Note. Significant associations are shown in bold font. dlPFC = dorsolateral prefrontal cortex; vmPFC = ventromedial prefrontal cortex.

None of the three PTSD-associated DNAm loci were significantly associated with CLDN5 expression in any brain region. For completeness, we examined all CLDN5 probes on the chip and found two nominally significant associations with CLDN5 expression that did not withstand correction for multiple testing: cg09446908 (B = −0.343, β = −0.220, p = 0.005) and cg17577122 (B = −0.118, β = −0.174, p = 0.028), both in vmPFC (Table S5). Comparison of CLDN5 expression with estimated cell types confirmed expected associations between CLDN5 RNA and endothelial cells in all three brain regions (βs = 0.731 – 0.743 and p_-adj = 7.336×10⁻¹⁷ – 4.907×10⁻¹²; Table 5). Multiple testing adjusted associations also emerged between CLDN5 expression and estimated mural cells, astrocytes, and microglia in all three regions, neurons in dlPFC and motor cortex, and red blood cells in dlPFC and vmPFC (Table 5). The direction of these associations was positive except for neurons.

Table 5.

Associations between the CLDN5 Expression and Estimated Cell Type Proportions in Postmortem Brain Tissues

	dlPFC (n = 93)				vmPFC (n = 86)				Motor cortex (n = 89)
Cell Type	B	β	p	p -adj	B	β	p	p -adj	B	β	p	p -adj
Astrocyte	0.570	0.566	4.424E-08	1.548E-07	0.218	0.247	0.025	0.033	0.368	0.326	3.539E-04	0.001
Endothelial	0.680	0.739	3.493E-18	7.336E-17	0.609	0.731	7.010E-13	4.907E-12	0.621	0.743	1.747E-13	1.835E-12
Microglia	0.702	0.584	1.085E-08	4.557E-08	0.500	0.410	0.001	0.001	0.451	0.363	0.002	0.004
Mural	0.551	0.625	2.730E-11	1.433E-10	0.397	0.505	9.667E-08	2.900E-07	0.349	0.495	3.592E-07	9.430E-07
Neuron	−0.358	−0.413	2.988E-06	6.973E-06	−0.154	−0.152	0.174	0.202	−0.276	−0.264	0.003	0.005
Oligodendrocyte	0.067	0.054	0.561	0.589	0.037	0.027	0.829	0.829	0.121	0.086	0.289	0.319
RBC	0.267	0.296	0.005	0.008	0.285	0.291	0.012	0.017	0.134	0.161	0.126	0.156

Note. p_-adj = adjusted p-value via an FDR correction across 21 tests (3 regions * 7 cell types). Significant associations are shown in bold font. dlPFC = dorsolateral prefrontal cortex; vmPFC = ventromedial prefrontal cortex; RBC = red blood cells.

4. Discussion

4.1. Overview

In this study, we expanded on the burgeoning literature suggesting a role for CLDN5 in BBB permeability, depression, and neuroinflammation in stressed organisms (Menard et al., 2017). We found preliminary evidence that the number of different types of traumatic events experienced across the lifetime was associated with CLDN5 DNAm at cg21872764. Two additional CLDN5 DNAm loci, cg00804505 and cg17411190, were associated with PTSD severity. Two of these trauma and PTSD-associated probes (the peak probe, cg21872764 and cg17411190) also showed associations with PTSD when methylation was measured in vmPFC, a region previously implicated in postmortem CLDN5 expression studies of depression (Dion-Albert et al., 2022). This is consistent with evidence for correlations between these loci in blood and brain when measured in frontal and temporal regions in reference databases (e.g., Blood-Brain Epigenetic Concordance or BECon; Edgar et al., 2017). We did not observe associations between PTSD diagnosis and CLDN5 DNAm in dlPFC or motor cortex. This may suggest the vmPFC is more vulnerable to stress-related alterations in CLDN5 DNAm. Meta-analyses suggest that PTSD is associated with vmPFC hypoactivity in the context of amygdala hyperactivity (the two regions evidence bidirectional projections between them), consistent with the notion of insufficient inhibition of emotional arousal in PTSD (Hayes et al., 2012). Whether BBB integrity and CLDN5 play a role in this is unknown and requires additional research.

Associations between trauma and PTSD and the CLDN5 loci were not better accounted for by rs10314 and rs885985, despite these genotypes showing associations with CLDN5 DNAm. Results are broadly consistent with the preclinical literature suggesting a role for cldn5 in stress-related depression-like phenotypes (Menard et al., 2017). Our results raise the possibility that CLDN5 is associated with both the stressor itself (i.e., trauma exposure) and the chronic psychiatric stress response (i.e., PTSD) and suggest the need for future mechanistic research evaluating the role of CLDN5 epigenetics in linking trauma and PTSD to BBB degradation and related neuropathology.

We found that the peak trauma and PTSD-associated locus (in blood and brain, respectively), cg21872764, evidenced a positive cross-sectional association with peripheral NFL. Additionally, our preliminary cross-sectional mediation model suggested that the association between trauma exposure and plasma NFL was mediated by blood DNAm at cg21872764. Prior preclinical research suggests that BBB permeability is strongly correlated with serum NFL levels at multiple timepoints following experimental head injury (Arena et al., 2002). Thus, to the extent that cg21872764 DNAm in blood is a marker for the same locus in brain (Edgar et al., 2017), DNAm at this locus may signal stress-associated BBB disruption. Analysis of data from brain tissue also revealed that CLDN5 expression was associated with cell type estimates in each brain region, with the most significant and consistent effects evident for endothelial cells, which were positively associated with CLDN5 expression. We observed nominally significant associations between CLDN5 probes cg09446908 and cg17577122 (that were not associated with trauma or PTSD in blood or brain) and reduced CLND5 expression in vmPFC. One possible explanation for this pattern of results is that trauma and PTSD-associated alterations in CLDN5 DNAm give rise to CLDN5 downregulation in endothelial cells, allowing for central biomarkers of neuropathology, like NFL, to cross the BBB to the periphery. Consistent with this possibility, DNAm at two CLDN5 loci that were nominally associated with PTSD severity in blood in this study, cg05460329 and cg16773741, were previously found to be associated with worse cognitive performance over time (at the epigenome-wide level of statistical significance) when measured in dlPFC (Hüls et al., 2022). As these data are cross-sectional, we can make no causal or directional claims, however, the evidence for relationships between PTSD, CLDN5 DNAm, and biomarkers of neuropathology highlight the need for future studies to test this.

4.2. Future Research

Additional research is needed to test the replicability of these associations, determine their overlap with other stress-related psychiatric conditions like depression, and evaluate temporal associations between PTSD, CLDN5 DNAm, and neuropathology biomarkers. CLDN5 epigenetics may be a useful target for future intervention research aimed at improving BBB integrity and reducing psychological symptoms. The antidepressants imipramine and fluoxetine (Menard et al., 2017; Sun et al., 2024) and the mood stabilizer lithium (Taler et al., 2020) appear to modulate cldn5 expression, reduce stress-associated depressive phenotypes, and protect the brain from inflammatory insults in rodents. Similarly, experimental agents, such as a glycogen synthase kinase-3 inhibitor (Cheng et al., 2018) have been shown to reduce depression-like symptoms and alter cldn5 expression in rodent brains. Nonpharmacological interventions may also influence CLDN5 expression. A recent study found that positive environmental conditions (an enriched environment) mitigated the effects of early life stress (maternal separation) on cldn5 expression, BBB permeability, and depressive-like behaviors in mice (Ansari et al., 2025). Collectively, these studies suggest that CLDN5 epigenetics may be altered through environmental or pharmacological approaches, with concomitant improvement in BBB integrity and symptom reduction. Future research could evaluate if any of these interventions operating on CLDN5 also reduce future neurological risk among those with PTSD.

4.3. Study Limitations

The results of this study should be considered preliminary in light of several limitations. The study was comprised of primarily male Veterans with PTSD and may not generalize to non-Veterans or women. Further, the relatively young mean age of the TRACTS subset (in their early 30s) and the exclusion of Veterans with cognitive or neurological disease (other than that related to TBI) from that cohort may have limited the range of NFL in plasma and made it more difficult to observe associations with CLDN5 DNAm. This concern is offset to an extent by the inclusion of the older NCPTSD cohort (57% of the sample) without these exclusions. Still, follow-up studies in older Veterans with greater neurological risk are needed to more comprehensively evaluate CLDN5 DNAm in association with PTSD and neurological impairment. The brain bank cohort was small and power was limited in these analyses. We only had access to data from three brain regions and it is possible that associations between CLDN5 DNAm and expression may differ in other regions. The differential associations observed across brain regions require replication in larger samples. We did not have a measure of trauma exposure in the brain bank cohort that would allow us to disentangle its associations with CLDN5 DNAm apart from those of PTSD. We also could not cleanly separate effects related to PTSD from those associated with MDD in the brain bank because the majority of PTSD cases had comorbid MDD. Brain bank results were broadly consistent with those from our living cohort (in which we were able to covary for MDD), helping to mitigate this concern. We did not have data concerning CLDN5 protein levels in blood or brain, BBB staining metrics, or BBB permeability neuroimaging (e.g., via positron emission tomography) that would have allowed for more direct examination of BBB permeability. Most importantly, we cannot infer causality or temporality from these cross-sectional associations and additional research is needed to address such questions.

4.4. Conclusions

In this first-ever study of associations between traumatic stress and a gene critically responsible for tight junction integrity of the BBB, we found evidence for cross-sectional associations between trauma exposure, PTSD, and CLDN5 DNAm and between CLDN5 DNAm and markers of neuropathology. We suspect that chronic PTSD symptoms function as a stressor that may alter CLDN5 DNAm, leading to decreased CLDN5 expression in brain endothelial cells and increased BBB permeability, allowing for bidirectional passage of neuroinflammatory molecules across the BBB. These preliminary cross-sectional results may help to explain the link between traumatic stress and neuropathology and highlight the role of DNAm in sensitivity to environmental insult. If these results are replicated and future studies support a mechanistic role for CLDN5, this could inform the development of novel therapeutic approaches for reducing risk for neurological disease among those with PTSD.