Traumatic events and mental health: The amplifying effects of pre-trauma systemic inflammation

Abstract

Background.

Traumatic experiences are strongly predictive of adverse mental health outcomes. Experimental studies have demonstrated that systemic inflammation can increase reactivity to threatening stimuli. It is not known whether naturally occurring inflammation amplifies the impact of traumatic experiences on mental health. Here we test whether incident traumatic events are more predictive of adverse mental health outcomes for individuals with greater pre-trauma systemic inflammation in a racially and ethnically diverse cohort study of youth assigned male at birth who identify as sexual or gender minorities (ages 16–29, n=518), a group at high risk for trauma exposure.

Methods.

Measures of inflammation, depression symptom severity, and perceived stress were measured at baseline. One year later, depression symptom severity and perceived stress were measured again, and participants reported the traumatic events they had experienced in the intervening year.

Results.

In a model adjusted for baseline depression symptom severity and other key covariates, we found that higher baseline levels of interleukin-1β amplified the effect of incident trauma exposure on depression symptom severity at follow-up (β=0.234, SE=0.080, P=0.004). In a model adjusted for baseline perceived stress and other key covariates, we found that higher baseline scores on a multi-marker inflammatory index amplified the effect of incident trauma exposure on perceived stress at follow-up (β=0.243, SE=0.083, P=0.003).

Conclusions.

These findings suggest that greater pre-trauma inflammation may predict poorer mental health following trauma exposure. Understanding how inflammation interacts with trauma to shape mental health may generate novel insights for preventing and treating the debilitating psychological consequences of trauma.

1. Introduction

Traumatic experiences are strongly predictive of adverse mental health outcomes. One study found that depression severity scores were 71% higher among individuals who had experienced at least one traumatic event in the previous year, after adjusting for pre-trauma depression severity (1). Another study reported that having experienced a major stressful life event in the previous year was associated with higher risks for major depressive disorder, posttraumatic stress disorder (PTSD), anxiety disorders, and greater perceived stress (2). Yet, there is considerable variation in how trauma impacts subsequent mental health. For example, a sizeable minority of individuals exposed to traumatic stressors later develop PTSD (~10–20%), but a majority of individuals exposed to such stressors do not (3). Several factors have been identified that may predict an individual’s risk of developing PTSD in the aftermath of a traumatic experience (e.g., sex/gender, prior adversity, perceived life threat during trauma) (4, 5), but the neurobiological mechanisms that account for differences in trauma-related mental health are still poorly understood. Previous experimental research suggests that elevated systemic inflammation can increase the aversiveness of stressful or threatening stimuli (6–13). It is not known whether naturally occurring inflammation amplifies the aversiveness of traumatic experiences. Here we test whether traumatic events are more deleterious to subsequent mental health among individuals with higher levels of pre-trauma systemic inflammation.

Inflammation is a core component of the immune system’s response to a wide range of pathologies, including infections, toxin exposures, and tissue damage (14–16). Acute inflammation is critical for mounting an effective immune response to infection and injury, but chronic inflammation predicts greater morbidity and mortality from a variety of non-communicable conditions (e.g., cancer, cerebrovascular disease, cardiovascular disease) (17–19). Inflammation may also be a risk factor for adverse mental health outcomes, such as depression and anxiety. Some studies have reported that elevated systemic inflammation is associated with greater risk of depression or anxiety (20–23), but others fail to detect such associations or report mixed results, even with relatively large sample sizes (24–27). Further work is needed to clarify the potential relationship between inflammation and mental health.

Multiple dimensions of adversity (e.g., bereavement, social isolation, poverty, trauma exposure) have been linked to pro-inflammatory patterns of immune regulation in humans (28–31). A set of similar regulatory responses to adversity, including pro-inflammatory shifts in gene expression, have also been described in non-human primates (32–34), in mouse models (35, 36), and even in fish (37), which has prompted some researchers to refer to these coordinated changes the Conserved Transcriptional Response to Adversity (CTRA). Acute inflammation is critical for mitigating mortality risk from the tissue damage and bacterial infections that often accompany injuries (14–16). Adverse life experiences may often serve as cues of increased mortality risk from violence and injury (38, 39). This may have resulted in selection for the CTRA, if it improved chances of survival in the event of a physical injury following an adverse experience (32).

Systemic inflammation can induce a variety of changes in behavior, motivation, and perception. These changes, often called “sickness behavior”, include increased lethargy, increased anhedonia, altered patterns of food consumption, and altered social behavior (10, 40–42). Sickness behavior may help sick hosts mount an effective immune response by prioritizing the allocation of metabolic resources to the immune system (40, 43, 44), but chronic activation of sickness behavior may play a role in the development of depressive disorders (45, 46).

Sickness behavior is initiated through pro-inflammatory cytokine signaling pathways from the peripheral immune system to the central nervous system (44, 47). There are multiple mechanisms by which peripheral inflammatory signals reach the brain, including a pathway mediated by the afferent vagus nerve (48), passive diffusion of pro-inflammatory cytokines across the blood-brain barrier (49), and selective transport of pro-inflammatory cytokines across the blood-brain barrier (50). There is also in vitro evidence that the pro-inflammatory cytokine interleukin-1β (IL-1β) upregulates neurotoxic activity in the indolamine-2,3-dioxygenase-tryptophan-kynurenine pathway, leading to decreased neurogenesis in hippocampal progenitor cells, which may play a role in the pathogenesis of depressive disorders (51).

Studies with animal models demonstrate that elevated inflammation can increase reactivity to stressful or threatening stimuli (6–13). For example, a study of rhesus monkeys found that experimentally induced inflammation led to an increase in agitated reactions to threatening behavior from a human experimenter (12). Pro-inflammatory immune profiles were associated with greater depression-like behavior in studies of rodents exposed to stressors, and these depression-like responses to stressful stimuli were attenuated by blocking pro-inflammatory pathways (7–9).

Several human studies have produced similar findings. For example, a neuroimaging study in humans found that experimentally induced inflammation led to greater amygdala activity in response to images of faces depicting fear (6). A study of older adults who had experienced spousal bereavement found that higher levels of systemic inflammation, as measured by a combined inflammatory index of stimulated cytokines measured shortly after bereavement, predicted greater odds of future depression (52). Another study found that greater systemic inflammation (as measured by a combined multi-marker inflammatory index) amplified the statistical effect of poverty (a chronic stressor) on neural reactivity to both threatening and rewarding stimuli (53). A study of adolescent girls found that greater salivary cytokine reactivity to a laboratory stressor predicted stronger effects of subsequent interpersonal life stressors on depression symptoms (54). However, the relationship between salivary cytokine production and systemic inflammation is unclear (55), and the observed effects could reflect general physiological reactivity to stress exposure, rather than systemic inflammation per se. Studies of combat-exposed military personnel have reported that greater inflammatory activity (measured in various ways) before deployment predicts greater risk of PTSD after deployment (56–58). The studies reviewed above provide important data on inflammation, stress exposure, and mental health, but they also feature certain limitations. For the experimental studies (6–8, 12), it is not clear whether the findings translate to real-world mental health outcomes. Some prior human studies focused on the interaction between stress exposure and inflammation have measured inflammation after the onset of the stressor (52, 53), which makes it impossible to determine the temporal sequence of inflammation, stress exposure, and the outcome variable. The studies of military personnel (56–58), like most prior studies of inflammation and mental health (20–27), did not explicitly model the interaction of stress exposure and inflammation in predicting mental health, so it is not clear whether inflammation amplifies the adverse impact of trauma exposure or increases PTSD risk through some other mechanism. In this paper, we combine individual strengths of prior studies by: (1) measuring systemic inflammation prior to the onset of trauma exposure; (2) explicitly modeling the interaction between systemic inflammation and trauma exposure in predicting subsequent mental health; and (3) measuring tangible stressors (traumatic events) and mental health outcomes (depression symptom severity and perceived stress).

Drawing on the research reviewed above, we hypothesized that traumatic events are more deleterious to mental health among individuals with higher levels of pre-trauma systemic inflammation. We tested this hypothesis using data from a Chicago-based cohort study of sexual and gender minority youth assigned male at birth (ages 16–29, n=518). Measures of systemic inflammation, depression symptom severity, and perceived stress were collected at baseline. One year later, depression symptom severity and perceived stress were measured again, and participants reported the traumatic events they had experienced in the intervening year. Our research question is of particular importance for members of this population, who are at high risk for experiencing traumatic events (59), exhibit high levels of systemic inflammation (60, 61), and experience high burdens of depression, anxiety, suicidality, and substance use (62–70).

2. Materials and methods

2.1. Recruitment and study design

The data for this study were collected as part of RADAR, a Chicago-based cohort study of sexual and gender minority youth assigned male at birth. The RADAR study aims to understand multilevel influences on a syndemic of health concerns related to HIV (e.g., substance use, mental health, intimate partner violence) (71). Baseline data collection for RADAR began in 2015. Initial enrollees included members of two previous cohort studies, Project Q2 and Crew 450, as well as a third cohort of newly recruited participants who were 16–20 years of age, were assigned male at birth, reported a sexual encounter with a man in the previous year or identified as gay, bisexual or transgender, and spoke English. Recruitment was expanded through enrollee referrals to friends and romantic partners who met the selection criteria. Upon recruitment into the study, participants were invited to a centrally located community-based site for data collection. After providing informed consent, participants completed an interview administered by a trained study staff member. At the same visit, a trained phlebotomist drew a blood sample into a 4 mL BD Vacutainer K₂ ethylenediaminetetraacetic acid-treated tube (BD 368021, Fisher Scientific 02–683-153) via antecubital venipuncture and centrifuged to separate plasma. Plasma was then divided into 0.5 mL aliquots and frozen for storage until the samples were analyzed to measure inflammatory markers. As the recruitment window approached for a participant’s one-year follow-up visit, study staff reached out to the participant via the participant’s preferred contact method to schedule a follow-up visit.

Soluble markers of systemic inflammation were initially measured for all HIV seropositive participants and a sample of HIV seronegative controls matched on age, gender, and race/ethnicity (60). Inflammatory markers were subsequently measured for additional HIV seronegative participants at baseline in order to increase statistical power. All study protocols were approved by Northwestern University’s IRB.

2.2. Traumatic events

Traumatic events were assessed using an index of traumatic experiences adapted from the PTSD module of the Computerized Diagnostic Interview for DSM-IV (72). These items were modified to assess exposure to traumatic events unrelated to military combat. Similar adaptations of the Computerized Diagnostic Interview have been used to measure trauma exposure in studies of sexual and gender minorities (73) and in other populations (1). Participants reported whether they had experienced each of the following in the past year: (a) being shot or stabbed; (b) being mugged or threatened with a weapon, or experiencing a break-in or robbery; (c) being raped or sexually assaulted; (d) being in a disaster like a fire, flood, earthquake, tornado, hurricane, bombing or plane crash; (e) experiencing an unexpected sudden death of a close friend or relative; (f) being diagnosed with a life-threatening illness; (g) being in a serious accident; (h) seeing someone being seriously injured or killed; (h) unexpectedly discovering a dead body; or (i) being kicked out of a caregiver’s house. The latter item was added because of its particular importance for sexual and gender minority youth, who experience high rates of homelessness.

2.3. Mental health

Depressive symptoms were measured using the Patient Reported Outcomes Measurement Information System (PROMIS) Depression –Short Form 8a instrument, which measures the presence and severity of non-somatic depressive symptoms (sadness; guilt; self-criticism; worthlessness; loneliness; interpersonal alienation; loss of interest, meaning, and purpose) over the past seven days (74). Participants were asked to report how often in the past week they had experienced each of the feelings mentioned above on a five-point scale ranging from “Never” to “Always”. The PROMIS Depression score was calculated by adding the numeric value of each response and dividing by the number of items that were answered by the participant. The PROMIS Depression –Short Form 8a exhibits high levels of reliability, high levels of information provided, and low levels of differential item functioning across racial/ethnic groups, age, gender, and socioeconomic groups (75). Cronbach’s α for the PROMIS Depression –Short Form 8a in our analytic sample was 0.94 at baseline and 0.95 at one-year follow-up. Statistical analyses in this paper were conducted using the PROMIS Depression score as a continuous variable.

Perceived stress was measured using Cohen’s Perceived Stress Scale, 10-item version (PSS-10), which measures perceived stress in the past month (76). The PSS-10 is designed to assess how unpredictable, uncontrollable, and overloaded participants find their lives. The PSS-10 asks participants how often in the past month they’ve experienced a series of scenarios that indicate perceived stress, or lack thereof. Participants report the frequency of each scenario on five-point scale, ranging from “Never” to “Very Often”. The PSS-10 is a widely used measure of perceived stress and exhibits strong reliability and validity in a diverse range of populations (76–78). Cronbach’s α for the PSS-10 in our analytic sample was 0.79 at baseline and 0.75 at one-year follow-up. Statistical analyses in this paper were conducted using the PSS-10 score as a continuous variable.

2.4. Inflammation measures

Soluble markers of inflammation were measured in plasma samples using the MESO QuickPlex SQ 120 electrochemiluminescence Meso Scale Discovery (MSD) immunoassay platform (Rockville, MD, USA). An MSD V-PLEX Custom Proinflammatory Panel 1 kit (human) was used to measure IL-1β (0.05–375 pg/mL), IL-6 (0.06–488 pg/mL), and TNF-α, (0.04–248 pg/mL). CRP was measured using the MSD V-PLEX Plus Human CRP kit (0.001–49.6 mg/L). Because CRP values were higher than expected, CRP was also measured in a subset of 10 plasma samples using a Roche/Hitachi cobas c 311 platform (0.15–20.0 mg/L) (Basel, Switzerland). Values of CRP for the matched samples were highly correlated between the two instruments (r = .99). The average intra-assay coefficient of variation (CV) was 8.29% for IL-1β, 9.04% for IL-6, 8.71% for TNF-α, and 6.28% for CRP. The average inter-assay coefficient of variation (CV) was 12.93% for IL-1β, 10.36% for IL-6, 13.40% for TNF-α, and 13.59% for CRP.

In addition to these individual inflammatory markers, we also calculated the standardized sum of all four inflammatory measures to assess broad inflammatory activation across multiple pathways. This approach has been used frequently in recent studies of systemic inflammation (52, 53, 79).

2.5. Covariates

Data on age, gender, race/ethnicity, marijuana use, and tobacco use were obtained from a survey administered at baseline. Marijuana use was assessed using numeric scores on the Cannabis Use Disorder Identification Test-Revised (CUDIT-R) (80). Tobacco use was assessed using single-item measures of cigarette use (never, once or twice, occasionally but not regularly, regularly in the past, regularly now) and e-cigarette/vaporizer use (never, once or twice, occasionally but not regularly, regularly in the past, regularly now). HIV tests were conducted at baseline using the Alere Determine™ HIV1/2 Ab/Ag Combo 4th generation point-of-care test on a fingerstick blood sample (Waltham, MA). If a participant tested positive on the point-of-care test, follow-up HIV antibody and antigen testing was conducted to confirm the positive result. Height and weight were measured using standard anthropometric techniques. Body mass index (BMI) was calculated from height and weight measured at baseline when available. For participants lacking baseline anthropometric data, BMI was calculated using height and weight from a later visit.

2.6. Statistical analyses

We included all participants who had complete data for the variables and time points described above (n=518). One participant had data available for the pro-inflammatory cytokines (IL-1β, TNF-α, IL-6) but not for CRP. Thus, models with CRP as an independent variable were fit using data from 517 participants.

Values for IL-1β, TNF-α, IL-6, and CRP were natural log-transformed for statistical analysis as these biomarkers exhibited a skewed distribution that is approximately log-normal. To aid interpretability and facilitate the estimation of interaction terms, these natural log-transformed values were rescaled and centered to have a mean of 0 and standard deviation of 1. For the sake of brevity, these rescaled natural log-transformed values for the inflammatory measures are termed simply IL-1β, TNFα, IL-6, and CRP hereafter. To assess broad activation of the inflammatory response across multiple pathways, we also created a combined inflammatory index comprised of the standardized sum of IL-1β, TNF-α, IL-6, and CRP.

To test whether inflammation amplified the effect of traumatic events on mental health, we fit ordinary least squares multiple regression models that estimated interaction terms between traumatic event count and baseline inflammation in predicting mental health at follow-up. Separate models were specified for each of the inflammation measures. For the models predicting depression severity score at follow-up, the models controlled for baseline depression severity score. For the models predicting perceived stress score at follow-up, the models controlled for baseline perceived stress score. All models also included HIV status, BMI, age, gender, race/ethnicity, CUDIT-R score, cigarette use, and e-cigarette use as covariates. Models were estimated using the “lm” function in base R 3.5.0 (R Core Team, 2018). To produce standardized coefficient estimates, both outcome variables and all numeric predictors were standardized (mean=0, SD=1) prior to fitting regression models. To facilitate interpretation of interaction effects, scatterplots overlaid with marginal effect lines were created using the R package “ggplot2” (v3.1.0).

The traumatic events count variable yielded relatively small numbers of participants who had experienced more than one traumatic event. The small sample size at the upper end of the count distribution may inflate the standard errors of model parameter estimates. To evaluate this possibility, we repeated our analyses with traumatic events operationalized as a dichotomous variable (1=one or more traumatic event, 0=no traumatic events).

3. Results

Of the 518 participants in this study, 27% reported experiencing at least one traumatic event in the past year. For context, a previous epidemiological study of urban adults using a similar measure of trauma exposure found that 21% of participants reported experiencing at least one traumatic event in the past year. In our sample, ninety-six participants (18.5%) reported experiencing one traumatic event in the one-year period leading up to the follow-up visit, twenty-seven (5.2%) reported experiencing two traumatic events, and eighteen (3.5%) reported experiencing three or more traumatic events. The most frequently reported categories of trauma were the unexpected sudden death of a close friend or relative (n=76); being mugged or threatened with a weapon or experiencing a break-in or robbery (n=34); and being kicked out of a caregiver’s house (n=29). Selected descriptive statistics for the study sample, both stratified by HIV status and for the total analytic sample, are presented in Table 1.

Table 1.

Selected descriptive statistics for a sample of sexual and gender minority youth assigned male at birth from the RADAR cohort study (n=518)

	HIV positive (n=150, 29%)	HIV negative (n=368, 71%)	Total	P-value1
Age, mean (SD)	23.8 (2.5)	21.3 (3.0)	22.0 (3.1)	< 0.001
Gender, n (%)				0.087
Man	133 (28.1)	340 (71.9)	473 (91.3)
Woman	16 (43.2)	21 (56.8)	37 (7.1)
Non-binary	1 (12.5)	7 (87.5)	8 (1.5)
Race/ethnicity, n (%)				< 0.001
Black or African
American	93 (41.2)	133 (58.9)	226 (43.6)
Hispanic or Latinx	37 (22.6)	127 (77.4)	164 (31.7)
White	5 (4.9)	98 (95.1)	103 (19.9)
Other	15 (60.0)	10 (40.0)	25 (4.8)
Body mass index (kg/m2), mean (SD)	24.4 (4.8)	25.6 (6.5)	25.2 (6.0)	0.032
Past year trauma exposure
Yes	55 (36.7)	86 (23.4)	141 (27.2)	0.003
No	95 (63.3)	282 (76.6)	377 (72.8)
Traumatic event count, mean (SD)	0.6 (0.9)	0.4 (0.8)	0.4 (0.8)	0.022
Depression T-score, mean (SD)	51.0 (10.4)	51.7 (9.6)	51.5 (9.9)	0.519
Perceived stress score, mean (SD)	16.9 (6.2)	17.3 (6.7)	17.2 (6.5)	0.499
CUDIT-R score, mean (SD)	7.2 (6.5)	6.1 (6.3)	6.4 (6.3)	0.088
Cigarette use, n (%)				< 0.001
Never	35 (23.3)	128 (34.8)	163 (31.5)
Once or twice	21 (14.0)	84 (22.8)	105 (20.2)
Occasionally	18 (12.0)	69 (18.8)	87 (16.8)
Regularly in the past	15 (10.0)	29 (7.9)	44 (8.5)
Regularly now	61 (40.7)	58 (15.8)	119 (23)
E-cigarette use, n (%)				0.280
Never	88 (58.7)	211 (57.3)	299 (57.7)
Once or twice	44 (29.3)	105 (28.5)	149 (28.8)
Occasionally	13 (8.7)	38 (10.3)	51 (9.8)
Regularly in the past	0 (0.0)	8 (2.1)	8 (1.5)
Regularly now	5 (3.3)	6 (1.6)	11 (2.1)
C-reactive protein (mg/L), mean (SD)	6.3 (10.8)	3.1 (6.8)	4.0 (8.3)	< 0.001
C-reactive protein (mg/L), median (MAD)	2.7 (3.1)	1.1 (1.3)	1.4 (1.7)
Interleukin-1β (pg/mL), mean (SD)	1.97 (3.30)	0.09 (0.16)	0.63 (1.97)	< 0.001
Interleukin-1β (pg/mL), median (MAD)	0.76 (1.05)	0.06 (0.04)	0.08 (0.07)
Tumor necrosis factor-α (pg/mL), mean (SD)	3.15 (2.12)	1.89 (0.82)	2.26 (1.45)	< 0.001
Tumor necrosis factor-α (pg/mL), median (MAD)	2.57 (1.37)	1.86 (0.80)	2.00 (0.89)
Interleukin-6 (pg/mL), mean (SD)	0.57 (0.50)	0.59 (0.80)	0.58 (0.72)	0.690
Interleukin-6 (pg/mL), median (MAD)	0.44 (0.28)	0.38 (0.25)	0.39 (0.26)

HIV = human Immunodeficiency virus

CUDIT-R = Cannabis Use Disorders Identification Test-Revised

SD = standard deviation

MAD = median absolute deviation

¹P-values for differences by HIV status were generated using independent samples t-tests for numeric variables and chi-squared tests for categorical variables

In the sections below, statistical models are organized by outcome variable, first depression and then perceived stress. For each of the two outcome variables, we first describe models with trauma exposure modeled as a count variable and then describe models with trauma exposure modeled as a dichotomous variable. Next, we describe which interaction terms remain statistically significant after adjustment for multiple testing. Finally, we describe the results of additional analyses requested by reviewers.

3.1. Models predicting depressive symptoms: Trauma count

We fit a set of models predicting depression symptom severity in which trauma exposure was operationalized as a count variable (range: 0–5). In these models, the following inflammation*trauma interaction terms were in the hypothesized direction: IL-1β*traumatic event count (β=0.080, SE=0.036, P=0.026), TNF-α*traumatic event count (β=0.058, SE=0.047, P=0.225), CRP*traumatic event count (β=0.019, SE=0.028, P=0.500), and inflammation index*traumatic event count (β=0.049, SE=0.036, P=0.179). The IL-6*traumatic event count term was not in the hypothesized direction (β=−0.023, SE=0.039, P=0.560).

3.2. Models predicting depressive symptoms: Dichotomous trauma exposure

We fit another set of models that operationalized incident traumatic event exposure as a dichotomous variable (1=exposed, 0=not exposed). In these models, the following inflammation*trauma interaction terms were in the hypothesized direction: IL-1β*trauma exposure (β=0.234, SE=0.080, P=0.004), TNF-α*trauma exposure (β=0.075, SE=0.096, P=0.434), CRP*trauma exposure (β=0.033, SE=0.085, P=0.699), and inflammation index*trauma exposure (β=0.122, SE=0.085, P=0.154). The IL-6*trauma exposure term was not in the hypothesized direction (β=−0.029, SE=0.093, P=0.759). Complete coefficient estimates, standard errors, and p-values for ordinary least squares multiple regression models predicting depressive symptom severity scores at follow-up, with trauma operationalized as a dichotomous variable, are presented in Table 2.

Table 2.

Ordinary least squares multiple regression models predicting depression severity scores at follow-up in a sample of sexual and gender minority youth in the RADAR cohort study

	Inflammation Index Model	IL-1β Model	TNFα Model	IL-6 Model	CRP Model
	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)
Inflammation	0.033
	(0.049)
IL-1β		−0.038
		(0.053)
TNFα			−0.003
			(0.047)
IL-6				0.043
				(0.045)
CRP					0.059
					(0.045)
Incident Trauma	0.325***	0.318***	0.323***	0.331***	0.337***
	(0.086)	(0.086)	(0.087)	(0.086)	(0.086)
Baseline Depression Score	0.429***	0.429***	0.427***	0.426***	0.427***
	(0.040)	(0.040)	(0.040)	(0.040)	(0.040)
HIV Positive	−0.209**	−0.196*	−0.147	−0.129	−0.169*
	(0.104)	(0.109)	(0.100)	(0.095)	(0.097)
Body Mass Index	−0.024	−0.013	−0.005	−0.017	−0.026
	(0.040)	(0.038)	(0.038)	(0.041)	(0.040)
Age	0.045	0.045	0.043	0.041	0.040
	(0.042)	(0.042)	(0.042)	(0.042)	(0.042)
Gender (ref=Man)
Woman	0.009	−0.028	0.002	−0.004	−0.005
	(0.149)	(0.148)	(0.149)	(0.149)	(0.149)
Non-binary	0.713**	0.699**	0.724**	0.675**	0.708**
	(0.307)	(0.305)	(0.311)	(0.307)	(0.307)
Race/ethnicity (ref=Black/African American)
Hispanic/Latinx	0.110	0.103	0.108	0.105	0.096
	(0.095)	(0.094)	(0.095)	(0.095)	(0.095)
White	0.178	0.165	0.173	0.179	0.171
	(0.113)	(0.113)	(0.113)	(0.114)	(0.113)
Other	0.114	0.089	0.107	0.105	0.108
	(0.179)	(0.179)	(0.180)	(0.180)	(0.180)
CUDIT-R Score	0.014	0.010	0.013	0.015	0.017
	(0.042)	(0.042)	(0.042)	(0.042)	(0.042)
Cigarette Use (ref=Never)
Once or twice	−0.106	−0.105	−0.098	−0.099	−0.096
	(0.112)	(0.111)	(0.112)	(0.112)	(0.112)
Occasionally	0.060	0.075	0.069	0.077	0.053
	(0.123)	(0.122)	(0.124)	(0.123)	(0.124)
Regularly in the past	−0.269*	−0.257*	−0.265*	−0.268*	−0.281*
	(0.155)	(0.154)	(0.155)	(0.155)	(0.155)
Regularly now	−0.055	−0.056	−0.038	−0.043	−0.048
	(0.121)	(0.120)	(0.121)	(0.121)	(0.121)
E-cigarette use (ref=never)
Once or twice	0.211**	0.212**	0.208**	0.214**	0.216**
	(0.092)	(0.091)	(0.092)	(0.092)	(0.092)
Occasionally	0.190	0.169	0.206	0.211	0.202
	(0.139)	(0.139)	(0.139)	(0.139)	(0.139)
Regularly in the past	−0.330	−0.367	−0.382	−0.370	−0.278
	(0.308)	(0.306)	(0.308)	(0.308)	(0.314)
Regularly now	0.566**	0.587**	0.575**	0.584**	0.589**
	(0.265)	(0.263)	(0.265)	(0.265)	(0.265)
Inflammation*Incident Trauma	0.122
	(0.085)
IL-1β*Incident Trauma		0.234***
		(0.080)
TNFα*Incident Trauma			0.075
			(0.096)
IL-6*Incident Trauma				−0.028
				(0.093)
CRP*Incident Trauma					0.033
					(0.085)
Intercept	−0.179**	−0.180**	−0.202**	−0.207**	−0.190**
	(0.089)	(0.090)	(0.088)	(0.088)	(0.088)
Observations	517	518	518	518	517
R2	0.281	0.289	0.277	0.277	0.279

Note:

^*p<0.1

^**p<0.05

^***p<0.01

Inflammation = Combined inflammatory index

TNFα = natural log-transformed plasma tumor necrosis factor-α values

IL-6 = natural log-transformed plasma interleukin-6 values

CRP = natural log-transformed plasma C-reactive protein values

Numeric variables are standardized to produce β coefficients

3.3. Models predicting perceived stress: Trauma count

We fit a set of models predicting perceived stress score in which incident trauma exposure was operationalized as a count variable (range: 0–5). In these models, all inflammation*trauma interaction terms were in the hypothesized direction: IL-1β*traumatic event count (β=0.056, SE=0.035, P=0.111), TNF-α*traumatic event count (β=0.089, SE=0.046, P=0.057), CRP*traumatic event count (β=0.034, SE=0.028, P=0.216), inflammation index*traumatic event count (β=0.075, SE=0.036, P=0.037), and IL-6*traumatic event count (β=0.041, SE=0.038, P=0.271).

3.4. Models predicting perceived stress: Dichotomous trauma exposure

We fit another set of models that operationalized incident trauma exposure as a dichotomous variable (1=exposed, 0=not exposed). In these models, all inflammation*trauma interaction terms were in the hypothesized direction: IL-1β*trauma exposure (β=0.147, SE=0.079, P=0.063), TNF-α*traumatic event count (β=0.197, SE=0.093, P=0.035), CRP*trauma exposure (β=0.161, SE=0.082, P=0.050), inflammation index*trauma exposure (β=0.243, SE=0.083, P=0.003), and IL-6*trauma exposure (β=0.186, SE=0.090, P=0.039). Complete coefficient estimates, standard errors, and p-values for ordinary least squares multiple regression models predicting perceived stress scores at follow-up, with trauma operationalized as a dichotomous variable, are presented in Table 3.

Table 3.

Ordinary least squares multiple regression models predicting perceived stress scores at follow-up in a sample of sexual and gender minority youth in the RADAR cohort study

	Inflammation Index Model	IL-1β Model	TNFα Model	IL-6 Model	CRP Model
	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)
Inflammation	−0.047
	(0.048)
IL-1β		0.002
		(0.052)
TNFα			−0.001
			(0.045)
IL-6				−0.038
				(0.044)
CRP					−0.080*
					(0.044)
Incident Trauma	0.213**	0.209**	0.196**	0.224***	0.218***
	(0.083)	(0.084)	(0.084)	(0.084)	(0.084)
Baseline Depression Score	0.515***	0.515***	0.513***	0.518***	0.523***
	(0.036)	(0.036)	(0.036)	(0.037)	(0.037)
HIV Positive	−0.107	−0.120	−0.100	−0.064	−0.051
	(0.101)	(0.107)	(0.097)	(0.092)	(0.095)
Body Mass Index	0.002	−0.001	0.004	0.005	0.017
	(0.039)	(0.037)	(0.037)	(0.039)	(0.039)
Age	0.051	0.043	0.043	0.050	0.050
	(0.041)	(0.041)	(0.041)	(0.041)	(0.041)
Gender (ref=Man)
Woman	0.050	0.027	0.047	0.059	0.062
	(0.145)	(0.146)	(0.145)	(0.145)	(0.145)
Non-binary	−0.099	−0.139	−0.039	−0.157	−0.149
	(0.298)	(0.299)	(0.302)	(0.299)	(0.299)
Race/ethnicity (ref=Black/African American)
Hispanic/Latinx	0.080	0.066	0.078	0.073	0.083
	(0.091)	(0.092)	(0.092)	(0.092)	(0.092)
White	0.283***	0.275**	0.282**	0.281**	0.281**
	(0.109)	(0.109)	(0.109)	(0.110)	(0.109)
Other	−0.039	−0.083	−0.050	−0.058	−0.033
	(0.174)	(0.175)	(0.175)	(0.175)	(0.176)
CUDIT-R Score	0.009	0.006	0.002	0.008	0.009
	(0.040)	(0.041)	(0.041)	(0.040)	(0.041)
Cigarette Use (ref=Never)
Once or twice	−0.031	−0.019	−0.015	−0.024	−0.036
	(0.109)	(0.109)	(0.109)	(0.109)	(0.109)
Occasionally	−0.021	−0.010	−0.023	−0.010	0.005
	(0.120)	(0.120)	(0.120)	(0.120)	(0.121)
Regularly in the past	−0.223	−0.228	−0.221	−0.224	−0.225
	(0.150)	(0.151)	(0.151)	(0.151)	(0.151)
Regularly now	−0.055	−0.050	−0.035	−0.036	−0.042
	(0.118)	(0.118)	(0.117)	(0.118)	(0.118)
E-cigarette use (ref=never)
Once or twice	0.045	0.041	0.030	0.052	0.042
	(0.089)	(0.089)	(0.090)	(0.090)	(0.090)
Occasionally	0.057	0.059	0.076	0.081	0.068
	(0.135)	(0.136)	(0.135)	(0.135)	(0.136)
Regularly in the past	0.031	−0.018	−0.046	−0.005	0.008
	(0.298)	(0.298)	(0.298)	(0.298)	(0.304)
Regularly now	0.174	0.224	0.177	0.203	0.178
	(0.257)	(0.258)	(0.258)	(0.258)	(0.258)
Inflammation*Incident Trauma	0.243***
	(0.083)
IL-1β*Incident Trauma		0.147*
		(0.079)
TNFα*Incident Trauma			0.197**
			(0.093)
IL-6*Incident Trauma				0.186**
				(0.090)
CRP*Incident Trauma					0.161*
					(0.082)
Intercept	−0.115	−0.101	−0.116	−0.129	−0.131
	(0.085)	(0.087)	(0.085)	(0.085)	(0.085)
Observations	517	518	518	518	517
R2	0.336	0.330	0.331	0.330	0.331

Note:

^*p<0.1

^**p<0.05

^***p<0.01

Inflammation = Combined inflammatory index

IL-1β = natural log-transformed interleukin-1β

TNFα = natural log-transformed plasma tumor necrosis factor-α values

IL-6 = natural log-transformed plasma interleukin-6 values

CRP = natural log-transformed plasma C-reactive protein values

Numeric variables are standardized to produce β coefficients

3.5. Adjustment for multiple testing

In the models reported above, we included two mental health outcome variables, two methods of operationalizing trauma exposure, and five inflammation variables, for a total of twenty hypothesis tests. Eighteen of the twenty interaction terms we tested were in the hypothesized direction. To account for multiple testing, we applied a Benjamini-Hochberg false discovery rate (FDR) adjustment to the 20 p-values reported above (FDR < 0.05) (81). After FDR adjustment, the interaction terms that remained statistically significant were: (a) the IL-1β*dichotomous trauma term predicting depression symptom severity (Figure 1) and (b) the inflammation index*dichotomous trauma term predicting perceived stress (Figure 2). We therefore focus our discussion on these two findings.

Figure 1. The interaction between incident traumatic event exposure and pre-trauma IL-1β (natural log-transformed) levels in predicting depressive symptoms follow-up.

The scatterplot depicts depressive symptom severity by IL-1β level and trauma exposure. The plotted marginal effect lines illustrate the synergistic interaction between trauma exposure and IL-1βin predicting depressive symptom severity.

Figure 2. The interaction between incident traumatic event exposure and pre-trauma inflammation index scores in predicting perceived stress at follow-up.

The inflammation index score is the standardized sum of log-transformed IL-1β, TNF-α, IL-6, and CRP levels. The scatterplot depicts perceived stress scores by inflammation level and trauma exposure. The plotted marginal effect lines illustrate the synergistic interaction between trauma exposure and inflammation in predicting perceived stress.

3.6. Additional analyses requested by reviewers

Several additional analyses were requested by reviewers. For the sake of brevity, we restrict the scope of these additional analyses to the models whose interaction terms remained statistically significant after adjustment for multiple testing in our original analyses: (a) the IL-1β*dichotomous trauma model predicting depression symptom severity and (b) the inflammation index*dichotomous trauma model predicting perceived stress.

In addition to measuring past-year trauma exposure at one-year follow-up, we also measured it at baseline. We re-ran our models adding past-year trauma exposure measured at baseline as a covariate. These models produced similar effect estimates as the original model for both the IL-1β*dichotomous trauma interaction term predicting depression symptoms (β=0.229, SE=0.080, P=0.004) and the inflammation index*dichotomous trauma interaction term predicting perceived stress (β=0.233, SE=0.082, P=0.005).

At one-year follow-up we also collected a measure of the frequency of LGBT victimization (e.g., being threatened, punched, followed, etc. due to LGBT status/identity) in the past six months. Higher values on this measure indicate more frequent experiences of LGBT victimization. We re-ran our models including LGBT victimization and its interaction with the inflammatory measure as covariates. These models produced similar effect estimates as the original model for both the IL-1β*dichotomous trauma interaction term predicting depression symptoms (β=0.262, SE=0.087, P=0.003) and the inflammation index*dichotomous trauma interaction term predicting perceived stress (β=0.222, SE=0.083, P=0.007). In the model predicting depression symptoms, neither the main effect of LGBT victimization (β=0.065, SE=0.043, P=0.130) nor the IL-1β*LGBT victimization interaction term (β=−0.045, SE=0.036, P=0.207) exhibited a statistically significant association with depressive symptoms above and beyond the other terms in the model. In the model predicting perceived stress, greater LGBT victimization predicted greater perceived stress (β=0.120, SE=0.046, P=0.008) above and beyond the other terms in the model, but the inflammation index*LGBT victimization term was not statistically significant at P<0.05 (β=−0.001, SE=0.040, P=0.979).

We re-ran our models stratified by race/ethnicity. The IL-1β*dichotomous trauma interaction term predicting depression symptoms was estimated separately for Black (β=0.259, SE=0.123, P=0.035), Latinx/Hispanic (β=0.198, SE=0.166, P=0.234), white (β=0.496, SE=0.187, P=0.010), and other (β=1.232, SE=0.357, P=0.007) participants. The inflammation index*dichotomous trauma interaction term predicting perceived stress was also estimated separately for Black (β=0.308, SE=0.125, P=0.014), Latinx/Hispanic (β=0.051, SE=0.171, P=0.767), white (β=0.245, SE=0.189, P=0.199), and other (β=0.259, SE=0.123, P=0.035) participants.

Finally, we re-ran our models excluding participants with CRP levels more than three standard deviations above the mean (n=10). These models yielded estimates similar to our original analyses for both the IL-1β*dichotomous trauma interaction term predicting depression symptoms (β=0.228, SE=0.081, P=0.005) and the inflammation index*dichotomous trauma interaction term predicting perceived stress (β=0.186, SE=0.083, P=0.025).

4. Discussion

In this study, we used a longitudinal cohort design to test whether pre-trauma systemic inflammation amplifies the effects of subsequent traumatic events on mental health in a sample of sexual and gender minority youth. Controlling for baseline perceived stress and other key covariates, we found that higher pre-trauma scores on a multi-marker inflammatory index amplified the effect of incident trauma exposure on perceived stress at follow-up. This finding suggests that broad activation of the inflammatory response prior to experiencing a traumatic event may contribute to a more intense subjective experience of stress in the aftermath of a traumatic event. This finding is consistent with previous studies reporting that military personnel who exhibited pro-inflammatory immune profiles prior to deployment were more likely to develop subsequent PTSD (56–58). Figure 2 illustrates the moderating effect of systemic inflammation. At average levels of systemic inflammation, the model estimates a very small effect of trauma exposure on perceived stress. At the highest levels of systemic inflammation, the model estimates that perceived stress is up to nearly a full standard deviation higher among those who were exposed to trauma compared to those who were not.

Controlling for baseline depression symptom severity and other key covariates, we found that higher pre-trauma levels of IL-1β amplified the effect of incident trauma exposure on depression symptom severity at follow-up. Figure 1 illustrates the moderating effect of IL-1β. At average levels of IL-1β, the model estimates a very small effect of trauma exposure on depression symptom severity. At the highest levels of IL-1β, the model estimates that depression symptom severity is up to nearly a full standard deviation higher among those who were exposed to trauma compared to those who were not.Previous work suggests that IL-1β can induce neurotoxic activity in the indolamine-2,3-dioxygenase-tryptophan-kynurenine pathway, leading to decreased neurogenesis in hippocampal progenitor cells (51). Disrupted neurogenesis in the hippocampus may impair recovery following activation of the stress response (82). Impaired post-stress recovery, in turn, may exacerbate depressive symptoms in the aftermath of a traumatic experience (83).

Our results suggest that the trauma-amplifying effects of inflammation are stronger and more consistent when traumatic event exposure is operationalized as a dichotomous variable rather than a cumulative count variable. The small number of participants at the higher end of the count distribution may have led to inflated standard error estimates when using a cumulative count measure of trauma exposure. Further work is needed to investigate different approaches to modeling trauma exposure and how these modeling decisions affect conclusions.

Previous research on the association between inflammation and mental health has reported inconsistent results, with some studies identifying elevated inflammation as a risk factor for adverse mental health outcomes (20–23) and other studies failing to detect these associations or reporting mixed findings (24–27). Most prior studies of inflammation and mental health have not modeled the interaction between stress exposure and inflammation in predicting mental health (20–27, 56–58). Several human and animal experimental studies have modeled the interaction between stress exposure and systemic inflammation (6–8, 12), but these studies focus on behavioral and neural reactivity as outcomes, rather than mental health. Some previous human studies have measured inflammation after the onset of the focal stressor (52, 53), which makes it impossible to determine the temporal sequence of inflammation, stress exposure, and mental health. One prior study found that adolescent girls who exhibited greater salivary pro-inflammatory activity in response to a laboratory stressor also experienced greater depression symptom severity in the aftermath of an interpersonal life stressor (54). However, the relationship between salivary cytokines and systemic inflammation is unclear (55), which limits the conclusions that can be drawn from that study about the amplifying effects of systemic inflammation. In this paper, we combined individual strengths of previous studies by: (1) measuring systemic inflammation prior to the onset of trauma exposure; (2) explicitly modeling the interaction between systemic inflammation and trauma exposure in predicting subsequent mental health; and (3) measuring tangible stressors (traumatic events) and mental health outcomes (depression symptom severity and perceived stress). Our findings suggest that at least some associations between inflammation and mental health may be conditional on the occurrence of traumatic events. Inconsistencies in previous findings may therefore be explained in part by unmeasured interaction effects. Future studies of systemic inflammation and mental health should incorporate measures of trauma exposure and explicitly model the interaction between inflammation and trauma exposure in predicting mental health.

Sexual and gender minority youth assigned male at birth experience disproportionately high rates of depression, anxiety, suicidality, and substance use compared to their heterosexual and cisgender counterparts (62–70).This study illustrates how high rates of trauma exposure (59) and elevated systemic inflammation (60, 61) in this population may interact synergistically to increase risks for adverse mental health outcomes. Initiatives aimed at improving mental health should consider testing integrative interventions designed to reduce both trauma exposure and systemic inflammation. Given the high burden of adverse mental health outcomes among sexual and gender minority youth, future studies should investigate whether systemic inflammation interacts with LGBTQ-specific forms of adversity (e.g., anti-gay stigma, microaggressions), as well as other forms of adversity, to predict mental health.

While this study provides important new findings on the underpinnings of mental health risk in sexual and gender minority youth, there are limitations that should be noted. Our findings may not generalize to other populations. Further work is needed to assess whether similar patterns are present in other groups at high risk for exposure to traumatic events (e.g., combat-exposed military personnel, SGM individuals assigned female at birth), in other age groups (e.g., children, older adults), in groups inhabiting different inflammation ecologies (e.g., high-parasite risk environments), and in representative samples drawn from the general population.

In our study, inflammation was only measured at one point in time, making it impossible to determine whether a high concentration of an inflammatory marker represents chronically high levels or an acute increase. Future studies should measure inflammation at multiple timepoints and model how trauma, inflammation, and mental health relate to one another over time. When measuring basal inflammation, interleukin-1 receptor antagonist is often used as a measure of interleukin-1 activity in lieu of IL-1β because circulating concentrations of IL-1β are often low. Future studies should test whether our results replicate when interleukin-1 receptor antagonist is used instead of IL-1β.

It is possible that some of the apparent neuropsychological effects of inflammation may stem from the underlying causes of inflammation (e.g., current infection, injury, illness, complications from HIV) rather than directly from inflammation itself. Along the same lines, certain medications (e.g., immunosuppressants, anti-depressants, exogenous hormones) may influence both inflammation and mental health. While we included HIV status, tobacco use, and BMI in the model, we did not have access to comprehensive data on our participants’ current health status or medication use. Future studies should test whether our findings replicate when comprehensive measures of current health status and medication use are included in the model.

In this study, we asked participants about trauma exposure in the past year, but we did not collect finer-grained data on how recently each traumatic event occurred. Traumatic events that occurred more recently may have stronger effects on depressive symptoms and perceived stress. Future studies should collect high-resolution data on the timing of each traumatic event to account for the recency of trauma exposure. While cohort studies are important, there is also a need for human experimental studies that manipulate both inflammation (e.g., via lipopolysaccharide administration or typhoid vaccination) and stress exposure (e.g., social evaluative or social exclusion stress paradigms) in order to assess the moderating effects of inflammation on stress reactivity and recovery.

The results of our study suggest that pre-trauma systemic inflammation may predict vulnerability to adverse mental health outcomes in the aftermath of traumatic events. Our findings dovetail with previous experimental research demonstrating that systemic inflammation can increase neural and behavioral reactivity to stressful or threatening stimuli. Understanding how systemic inflammation interacts with traumatic events to shape mental health may lead to novel insights for preventing and treating the debilitating psychological consequences of traumatic events.

Highlights.

• Experiments have shown that inflammation can amplify the aversiveness of threats

• We hypothesized that inflammation amplifies the impact of trauma on mental health

• We tested this hypothesis in a longitudinal study of sexual and gender minority youth

• Pre-trauma inflammation amplified the effects of trauma on subsequent mental health

• Studies on the interaction of inflammation and trauma may lead to new interventions

List of abbreviations

PTSD

Posttraumatic stress disorder

CTRA

Conserved transcriptional response to adversity

IL-1β

interleukin-1β

IL-6:

interleukin-6

TNF-α

tumor necrosis factor-α

CRP

C-reactive protein

HIV

human immunodeficiency virus

PROMIS

Patient Reported Outcomes Measurement Information System

CUDIT-R

Cannabis Use Disorder Identification Test-Revised

BMI

body mass index

SGM

Sexual and gender minorities

AMAB

Assigned male at birth