The study population consists of a cross-sectional retrospective convenience sample of combat-exposed US military servicemembers and veterans, termed “warfighters.” Lifetime mTBI history, blood draws, and neuropsychological assessments were collected from 182 participants during baseline enrollment in the ongoing Chronic Effects of Neurotrauma Consortium (CENC) multi-site observational study (now LIMBIC-CENC under continuing funding). Participants were enrolled between 2015 and 2016 according to CENC protocol and approved by appropriate institutional review boards.

For each participant, mTBI diagnosis was determined based on CENC protocols detailed previously (Werner et al., 2021). Briefly, all underwent detailed structured and unstructured interviews as well as medical history reviews. Structured interviews included an identification of all potential concussive events across the participant’s lifetime using the Ohio State University TBI Identification (OSU TBI-ID) instrument. Each potential concussive event was assessed for preliminary mTBI diagnosis using the Virginia Commonwealth University retrospective concussion diagnostic interview and DOD/VA common mTBI definition. Preliminary mTBI diagnoses were vetted against unstructured interviews and medical documentation pertaining to the event and final diagnoses administered by the site principal investigator (PI). Any diagnostic uncertainty was adjudicated by a central diagnosis committee of national TBI experts. Participants were excluded if TBI was determined to be moderate, severe, or penetrating.

Sleep quality was assessed with the Pittsburgh Sleep Quality Index (PSQI) self-report questionnaire (Buysse et al., 1989; Carpenter and Andrykowski, 1998). The PSQI is a widely used measure comprised of 19 items, each item scored on a 4-point Likert scale. Items are focused on sleep-related problems and sleep quality, including sleep duration and delay in falling asleep. PSQI scores range from 0 to 21; higher scores indicate increasingly poor sleep quality.

For all cytokine biomarkers, outliers were removed at a threshold of 5x standard deviation. A conservative threshold of 5x standard deviation was chosen to maximally preserve true data variability while excluding samples that fell outside of the probable physiological range. Data were log transformed to achieve approximate normality. A large percentage of EV protein samples fell below respective lower limits of quantification for all three cytokines; frequencies of these samples were similar in mTBI and control cohorts (Supplementary Table S2). Analyses were performed with and without samples below the lower limit of quantification to evaluate the extent to which assay sensitivity may influence results. Because directionality was unchanged with inclusion or exclusion of below-limit samples, reported analyses include all evaluated samples.

Demographic characteristics were summarized by mean ± SD for continuous variables and counts for categorical variables. Cytokine concentrations were described by median and inter-quartile range. Student’s T-tests were used to compare log-transformed cytokine concentrations between mTBI and control cohorts. Separate analyses were performed for plasma-derived and EV-derived cytokine samples.

Linear relationships between sleep quality (PSQI score) and inflammatory cytokine levels were evaluated separately for mTBI and control cohorts using Spearman’s correlations. Data distribution and correlations were visualized using univariate kernel density estimates and best-fit linear regression lines. Multivariable linear regression models were used to estimate independent associations of sleep quality (PSQI score) with cytokine levels (standardized to mean of zero and standard deviation of one) with adjustment for age, sex, and body mass index (BMI) (i.e., potential confounders expected to influence inflammatory biomarker levels). Multivariable linear regression models were also constructed to evaluate potential interactions between sleep quality and additional predictors, adjusting for age, sex, and BMI. Interactions of cohort with PSQI were tested for each cytokine to evaluate the hypothesized stronger relationship of sleep quality with inflammatory biomarkers in the mTBI versus control cohort. In the mTBI cohort, we additionally evaluated whether there was an interaction effect of number of TBI events and PSQI score on cytokine levels.

Additional multivariable linear regression models were used to evaluate the potential predictive value of categorical sleep quality (“good” vs. “poor”) on cytokine levels, adjusted for age, sex, and BMI. Patients with PSQI <10 were classified as “good” sleepers and those with PSQI ≥10 as “poor” sleepers, according to established guidance on clinically significant sleep complaints in a military population (Matsangas and Mysliwiec, 2018).

Of 182 participants included in the study, 138 (76%) were in the mTBI cohort and 44 (24%) were healthy controls. The sample was predominantly male (86%) with an average age of 40 years (SD = 10.7) and BMI of 30.3 (SD = 5.4). 70% of participants identified as white, and most (87%) completed at least 1 year of college. In the mTBI cohort, the average number of TBIs was 2.5 (SD = 1.8) with 9 years (SD = 4.3) since the most severe mTBI. Demographic characteristics likely to influence inflammatory biomarker levels (i.e., age, sex, and BMI) were similar between mTBI and control cohorts, regardless of the presence or absence of clinically significant sleep complaints (Table 1).

IL-6, IL-10, and TNFα protein concentrations were quantified from each participant using EV and plasma sources (see Supplementary Table S2 for a detailed breakdown of values below the lower limit of quantification for each protein detection assay. Supplementary Table S3 provides a descriptive summary of cytokine levels with and without these values). Overall, EV IL-6 concentration was higher in the mTBI cohort [mean (SD) = 0.70 (0.90) pg/ml] versus controls[(mean (SD) = 0.32 (0.43) pg/ml], p = 0.040. No between-group differences were found for plasma IL-6, EV and plasma IL-10 or EV and plasma TNFα concentrations.

Spearman’s correlations were calculated to determine whether there may be a relationship between sleep quality (PSQI score) and cytokine levels following mTBI; correlation coefficients for each cohort are reported in Table 2. In the mTBI cohort, PSQI was positively correlated with each EV-derived cytokine (Figure 1) and with plasma-derived IL-6 (Supplementary Figure S1). Statistical significance was not reached for any PSQI-cytokine combination in the control cohort. However, a negative association between PSQI and plasma IL-10 approached significance in controls. (Table 2).

To determine whether correlations observed in the mTBI cohort might be partially explained by additional variables expected to influence inflammatory processes, multivariable linear regression models were constructed to estimate the association of PSQI with inflammatory cytokine levels after adjusting for age, sex, and BMI. In these models, PSQI remained associated with higher EV IL-10 and EV TNFα levels in mTBI patients (Table 3). Estimated associations in the mTBI cohort did not substantially vary by patient history of multiple TBI or number of TBI events (Supplementary Table S4).

For completeness, an additional set of models was calculated to estimate the hypothesized moderating effect of mTBI cohort on the relationship between PSQI and cytokine levels. Overall, PSQI was associated with higher EV IL-10 [ß (SE) = 0.08 (0.04), p = 0.04] and TNFα [ß (SE) = 0.06 (0.02), p = 0.01] independent of age, sex and BMI (Supplementary Table S5). An association with EV IL-6 approached significance. Interaction terms for PSQI with mTBI were not significant. Beta coefficients for interaction terms ranged from −0.03 to 0.09.

To evaluate clinical relevance of the observed associations between PSQI and cytokine levels in mTBI patients, the mTBI cohort was divided into clinically “good” and “poor” sleeper types according to PSQI score <10 and ≥10, respectively, as has been previously reported in the military population (Matsangas and Mysliwiec, 2018). In linear regression models of the potential predictive value of sleeper type on cytokine level in this cohort, the subset of mTBI patients who reported clinically significant sleep complaints (considered “poor” sleepers) had elevated levels of EV IL-10 compared to mTBI “good” sleepers, after controlling for age, sex and BMI: ß (SE) = 0.12 (0.04), p < 0.01 (Figure 2A). Elevations in mTBI poor sleepers approached significance for EV TNFα [ß (SE) = 0.05 (0.027), p = 0.08, Figure 2A] and plasma IL-6 [ß (SE) = 0.10 (0.05), p = 0.07] (Figure 2B). In these analyses, interactions of number of TBI events with sleeper type were not significant (Supplementary Table S6).

Initial exploratory analyses revealed significant associations between sleep quality (PSQI score) and each examined EV-derived cytokine (IL-6, IL-10, TNFα) in the mTBI cohort (Figure 1; Table 2). No associations were found for non-mTBI controls. To determine whether these findings represented a PSQI-cytokine relationship that was truly unique to the mTBI cohort, subsequent analyses were conducted to evaluate whether cohort was a moderator of PSQI-cytokine associations across participants. These analyses did not find cohort to be a statistically significant moderator (Supplementary Table S5). However, we note that the distribution of our dataset was particularly skewed toward the mTBI cohort, with mTBI individuals comprising 76% of the total population (n = 138 mTBI and n = 44 controls). The standard deviations of ß coefficients within moderator models were also notably large (Supplementary Table S5), suggesting potential statistical underpowering. Larger sample sizes may be needed to definitively support or refute a differential effect of sleep quality according to mTBI history in this population. Given our primary objective of exploring a link between inflammatory cytokines and sleep quality in mTBI, subsequent analyses were narrowed to the mTBI cohort only.

In the mTBI cohort, subsequent regression analyses controlling for age, sex, and BMI demonstrated significant relationships between PSQI and EV-derived pro-inflammatory TNFα, as well as anti-inflammatory IL-10 (Table 3). These findings are consistent with prior studies conducted in otherwise healthy individuals with chronic insomnia. Acutely, sleep deprivation has been shown to result in an increase in pro-inflammatory cytokines and concurrent decrease in anti-inflammatory cytokines (Hirotsu et al., 2012; Taraz et al., 2013). At longer timescales, this trend shifts into an upregulation of both pro-and anti-inflammatory markers (Wright et al., 2015; Ren et al., 2021). In a recent study of 110 individuals with and without chronic insomnia, Ren et al. (2021) demonstrated that both TNFα and IL-10 levels were elevated in insomnia patients, with TNFα positively correlating with PSQI score. Conversely, adenosine levels were found to be reduced and negatively correlated with PSQI. Given the established role of adenosine in both sleep and immune regulation, the authors hypothesized that reduced adenosine signaling may underlie the observed cytokine-PSQI relationships.

Regarding immune regulation in the mTBI brain, only a small number of studies have previously examined inflammatory cytokines in mTBI patients (Gill et al., 2018; Sun et al., 2019; Vedantam et al., 2021). Some studies have reported acute (days to weeks) initial elevations of IL-6 that dissipate over the subsequent 3–6 months (Sun et al., 2019; Vedantam et al., 2021). Here we show that IL-6 levels were significantly elevated at an average of 9 years since mTBI (p = 0.40). Since IL-6 production is induced by sleep restriction (Burgos et al., 2006), this finding may be indicative of a delayed secondary IL-6 upregulation associated with chronic sleep deprivation following mTBI. In support of this interpretation, sleep complaints were more common in the mTBI cohort; 59% of mTBI patients reported a PSQI ≥10 compared to 40% of controls (Table 1). Whether IL-6 remains high in some individuals or falls and rises again over the following years is unclear. Prospective studies tracking sleep quality and cytokine levels longitudinally after mTBI will be needed to characterize these interactions.

Previous studies have also found prolonged elevations of IL-10 following mTBI, which have been proposed to provide anti-inflammatory signaling in an attempt to counteract pro-inflammatory responses to injury (Gill et al., 2018; Vedantam et al., 2021). While absolute levels of IL-10 were not upregulated in our mTBI cohort compared to controls, they were significantly associated with poor sleep quality in mTBI (Figure 1; Tables 2, 3). Potentially underlying this association, IL-10 has been shown to inhibit the release of sleep-promoting regulatory molecules and to suppress non-rapid eye movement sleep in animal models (Kanaan et al., 1998; Dugas et al., 1998; Kushikata et al., 1999). One such regulatory molecule, adenosine, has potential for mechanistic parsimony, explaining the increase of both IL-10 and TNFα with worsening sleep quality as demonstrated previously in chronic insomnia (Ren et al., 2021) and now also in mTBI (Figure 1; Table 3). This potential mechanism is further supported by an extensive body of literature surrounding adenosine modulation in response to TBI (Lusardi, 2009; McGuire et al., 2019) and adenosine 2A receptors as critical components of sleep initiation (Lazarus et al., 2019; Ren et al., 2021). Future studies will be needed to explore this hypothesis further.

We additionally sought to contextualize the demonstrated associations between sleep quality and cytokine levels according to the clinical presentation of sleep complaints in mTBI participants. The mTBI cohort was divided into “good” and “poor” sleepers according to whether reported PSQI score was below or above the diagnostic threshold for identification of clinical sleep difficulties (Matsangas and Mysliwiec, 2018). According to this cut-off, the “poor” sleepers were expected to represent a group that would be naturally stratified within the current standard of care. Using this stratification, we found that mTBI patients who report poor sleep demonstrated significantly elevated EV IL-10, with trending significance for TNFα elevation (Figure 2A). These findings suggest that inflammatory alterations associated with sleep difficulties are not confined to subclinical gradations but rather can be inferred at the clinical level.

The finding that cytokine levels differ according to the presentation of sleep complaints is especially relevant to clinical outcomes as increasing evidence demonstrates that immune dysregulation has the potential to cause neurodegenerative disease [see Verboon et al. (2021) for a brief review]. We have previously shown that poor sleep is linked to elevated biomarkers of neurodegeneration in this same cohort of mTBI patients (Werner et al., 2021). Importantly, causality cannot yet be inferred, but the association of poor sleep with both neurodegenerative and inflammatory biomarkers in the same mTBI population should be explored further. This line of exploration could yield novel therapeutic targets, potentially implicating sleep-focused interventions as an accessible therapeutic entry point. Similarly, the development of relevant biomarkers may be useful for patient stratification within clinical trials aimed at novel therapies for mTBI-associated sleep disorders.

EV-based cell-to-cell communication has been linked to pathological mechanisms in a large number of conditions, including TBI (Mustapic et al., 2017; Gurunathan et al., 2019; Paolicelli et al., 2019; Mondello et al., 2020). Whether cytokines are released into peripheral circulation in soluble form or associated with EVs may depend on regulated biological processes and vary with the functional state of biological systems (Fitzgerald et al., 2018). However, the processes that regulate cytokine release after TBI are not well understood. Because the physiologic roles of soluble cytokines derived from the plasma compartment may be different from those of EV-encapsulated/bound cytokines, both were analyzed in the present study. In support of this approach, recent work demonstrates that EV-derived samples may hold additional value compared to plasma cytokine sources. Similar to the present findings, another study of military personnel with a history of chronic mTBI found associations between levels of EV IL-10 and PTSD symptom severity (Devoto et al., 2016). These EV-specific associations may be partially the result of the EV lipid bilayer membrane that shields protein contents from degradation (Winston et al., 2019), thereby increasing signal-to-noise ratios by decreasing the amount of non-specific and non-functional proteins detected in biomarker assays (Chung et al., 2020). In addition to favorable signal strength, increasing evidence suggests that the proteins derived from EVs may serve key functional roles in cell signaling pathways and cell-to-cell transport (Mustapic et al., 2017), perhaps more so than plasma-derived proteins. Our findings support this theory that, although lower in concentration (Supplementary Table S3), EV-derived cytokines may be more clinically relevant for sleep-immune signaling following mTBI. In our population, only EV-derived cytokines were associated with sleep quality (Figures 1, 2; Tables 2, 3). The exception was an unadjusted correlation between PSQI and plasma IL-6 in the mTBI cohort (Table 2), which disappeared once models were adjusted for age, sex, and BMI (Table 3). Future biomarker studies would likely benefit from the inclusion of both plasma and EV-derived samples for the measurement of functionally relevant protein levels.

This study presents a preliminary analysis of the association between sleep quality and inflammatory cytokines in warfighters with and without a history of mTBI. Although significant associations were demonstrated within the mTBI cohort, the included population lacked the statistical power for comparison between mTBI and control cohorts. Despite this limitation, we considered the inclusion of a control cohort to be an important aspect for initial exploratory analyses and one that did not diminish the goal of exploring mTBI-relevant associations. Future large-scale studies are needed to determine whether the demonstrated PSQI-cytokine relationships are exclusive to warfighters with mTBI. The current population also consisted exclusively of warfighters; it is unclear whether these results generalize to other TBI populations.

Within this population, we are unable to report on premorbid sleep difficulties or the continuous trajectory of sleep after injury for mTBI participants. The PSQI reports only on sleep complaints experienced during the previous 30 days. Accordingly, reported PSQI-cytokine associations may not be a true representation of long-term sleep difficulties and inflammatory processes occurring continuously in the time since mTBI. However, there is evidence to indicate that sleep symptoms following mTBI may worsen over time (Ma et al., 2019; Lequerica et al., 2020), suggesting that the demonstrated associations are likely gradually building processes.

A significant strength of the current study is the use of both plasma and EV cytokines, providing a novel and important contrast of the relevance of these two protein sources as clinical biomarkers. Additionally, the utilization of a consistent and standardized measure of mTBI through the CENC protocol conferred key diagnostic confidence and relative homogeneity of our mTBI cohort. Furthermore, through this standardized protocol we were able to analyze a cohort of mTBI patients with post-injury timescales averaging nearly a decade. This timescale is significantly longer than relevant prior studies, affording a uniquely valuable preliminary window into prolonged sequelae of mTBI and the associated sleep-immune crosstalk.