The relationship between social adversity, micro-RNA expression and post-traumatic stress in a prospective, community-based cohort

Abstract

Epigenetic processes serve as both mediators of and responders to social and environmental challenges, influencing biological outcomes. Therefore, pinpointing epigenetic factors associated with social adversity and traumatic stress enables understanding of the mechanisms underlying vulnerability and resilience. We hypothesized that micro-RNA (miRNA) expression may be associated with post-traumatic stress disorder symptom severity in the context of social adversity. To test this hypothesis, we leveraged longitudinal data from the Detroit Neighborhood Health Study, a community-based, prospective cohort of predominantly African Americans. This includes blood-derived RNA samples from 389 participants in wave 2 and 243 participants in wave 4. Social adversity data were available for all included participants (n = 483). Results identified 86 miRNAs that are associated with social adversities (financial difficulties, perceived discrimination, cumulative trauma) and post-traumatic stress severity. These miRNAs are involved primarily in the immune response, brain and neural function, as well as cell cycle and differentiation, and 23 (25%) have previously been associated with conditions related to post-traumatic stress disorder, including traumatic brain injury and stress response. Our findings offer a fresh perspective on understanding the epigenetic role of miRNA in the interaction between social adversity and traumatic stress.

Wang et al. analyzed data from the Detroit Neighborhood Health Study to explore links between social adversity, post-traumatic stress and epigenetic regulation through blood-derived micro-RNA profiles.

Main

Post-traumatic stress disorder (PTSD) is unique among diagnosed mental health conditions because diagnostic criteria require one or more discrete external events (that is, trauma exposure) for its emergence. This environmentally induced etiology indicates that the emergence of PTSD can be mediated by environmental factors, promoting an interaction between genes and the environment. Epigenetic mechanisms are one way environmental exposures can be embedded into the biology of organisms¹. Epigenetic factors exert their influence by altering the activity of genes via activation and suppression of expression². Environmentally inducible epigenetic mechanisms include DNA methylation, histone modification and the actions of non-coding RNA, including micro-RNAs (miRNAs)³. A micro-RNA (miRNA) is a type of small non-coding RNA molecule that plays an important role in the regulation of gene expression^4–7. These transcripts are short, typically ~22 base pairs in length. miRNAs act post-transcriptionally to disrupt gene expression by binding to complementary sequences in mRNA (messenger RNA) transcripts. This binding primarily silences gene expression. miRNAs are thus potent epigenetic regulators and have been implicated in a wide range of diseases and developmental pathways⁸. The role of miRNAs has been examined in a variety of mental disorders, including schizophrenia, depression and PTSD⁹. Previous studies have reported that individuals with PTSD from both military and civilian backgrounds exhibit altered miRNA expression patterns^10–13. However, less is known about how variation in miRNA expressions linked to differences in social adversity prospectively influences risk for PTSD or traumatic stress.

A large body of literature has established links between exposure to social adversity and adverse mental and physical health^14–16. A key feature in this literature is the premise that individual-level differences in exposure to different situations and/or stimuli, combined with individual differences in responses to these stimuli, produce population-level differences in the toll exacted by these exposures^17,18. For example, increased risk for adverse health outcomes associated with low socioeconomic position (SEP)—commonly assessed through indicators such as education, employment status and income—are at greater risk for adverse health outcomes¹⁹. This increased risk is hypothesized to result from greater exposure to psychosocial factors such as financial and/or occupational stressors, discrimination and greater exposure to crime¹⁸. Importantly, many of these exposures have been shown to impact social and behavioral biology, including loneliness²⁰, discrimination²¹ and stressful life event (SLE) exposure²².

Importantly, there is growing evidence that traumatic stress has a building-block or cumulative effect: individuals exposed to multiple adverse events, especially in the context of chronic social adversity, show a heightened risk for developing PTSD and other mental health disorders, along with more severe symptom profiles and functional impairments^23,24. This accumulating burden is hypothesized to produce long-lasting biological changes through epigenetic regulation²⁵, including altered miRNA expression²⁶. miRNAs may thus represent critical molecular mediators linking chronic adversity to long-term mental health outcomes.

Differences in SLE exposure account for a large proportion of variation in risk for stress-related psychopathology^27,28, including PTSD^29,30. This pattern is supported by findings in the Detroit Neighborhood Health Study (DNHS) cohort—the focus of the present study—where study individuals of low SEP have been exposed to a greater number of SLEs, and this greater burden of exposure to acute and chronic SLEs, such as unemployment, difficulty accessing healthcare and other financial problems, explains a substantial portion of the relationship between SEP and PTSD in the Detroit population³¹. However, in this previous work, even after controlling for SLEs, persons with low SEP still showed greater mean post-traumatic stress levels than those with high SEP, suggesting there are additional sources of vulnerability to stress-related psychopathology that remain to be clarified.

One possibility is that chronic exposure to social adversity in low-SEP individuals induces long-term biological changes—such as miRNA dysregulation—that shape stress reactivity and influence the severity of PTSD symptoms, even after trauma exposure. We sought to observe how miRNAs, a particular epigenetic class of molecule, respond to exposure to traumatic stress—in particular, post-traumatic symptom severity (PTSS). Previous research suggested that some specific miRNAs show differential expression in individuals with PTSD when compared with both control and resilient individuals³². Indeed, these same miRNAs have been identified as components of a co-expression network closely associated with both trauma exposure and PTSD symptoms. However, PTSD is a categorical variable (that is present or absent at a given time point) while symptom severity represents a continuous measure. We reasoned that these and other miRNAs could potentially be described as modifiers in influencing the connection between exposure to social adversities and PTSS. To test this viewpoint, we examined PTSS data from the DNHS, a prospective population-based longitudinal cohort based in Detroit, Michigan³³. The DNHS includes extensive data on social adversities collected in five annual waves from 2008 to 2013 along with biospecimens collected in a subset of participants in four of the five sampling waves. Because trauma exposure and PTSD outcomes were assessed over time, this dataset is appropriate for examining the longitudinal epigenetic effects on PTSS in individuals who have experienced trauma and social adversity.

Small RNA sequencing was conducted on samples from the DNHS in waves 2 and 4

To investigate the influence of miRNA in modifying the connection between social adversity and PTSS, we conducted small RNA sequencing and focused our analysis specifically on annotated miRNAs derived from blood samples from 483 individuals. Our dataset includes 632 individual blood samples, comprising 389 samples from wave 2 and 243 from wave 4. A total of 185 participants were sampled in both waves. Following quality control, adapter trimming and merging of paired-end reads, approximately 10 billion merged reads were mapped back to annotated mature miRNAs, averaging ~14.7 million reads per sample (Supplementary Fig. 1). To assess the reproducibility of our sequencing results, each sequencing plate included Qiagen Human XpressRef Universal Total RNA as a positive control. The mean Pearson correlation among these control samples was 0.98, indicating a high level of reproducibility in our sequencing data. The miRNA with the most read counts identified was hsa-mir-16, with eight miRNAs from the hsa-let-7 family nearly as abundant. The hsa-let-7 family is known for its important roles in gene expression regulation and involvement in various physiological activities, including metabolic regulation, cell differentiation and proliferation³⁴. Other dominant miRNAs quantified by total read counts across all samples, such as hsa-miR-486-5p, have previously been demonstrated to be abundant in red blood cells, further highlighting their prevalence in the present context³⁵.

The sequencing results from two distinct waves of sampling in multiple years offer a unique opportunity to explore the longitudinal influence of miRNA on the association between social adversity and PTSS: (1) the miRNA profiles measured in wave 2 can be used to understand the prospective relationship between PTSS in wave 3 and social adversity reported in waves 2 and 3. This approach captures both prior exposures and concurrent stressors that may influence downstream psychological outcomes; (2) the miRNAs from wave 4 can be utilized to study the relationship between PTSS in wave 4 and social adversity reported in waves 3 and 4. We opted to focus on wave 4 rather than wave 5 because the latter sampled a smaller number of DNHS participants.

PTSS can be estimated by lifetime social adversities

To identify the best statistical distribution for modeling PTSS, we evaluated multiple candidate distributions to ensure the selection of an appropriate general linear regression model. To do so, we utilized maximum log-likelihood estimation to optimize the parameters and employed the Bayesian information criterion (BIC)³⁶ to evaluate four frequently used distributions: negative binomial (NB), Poisson, gamma and Weibull. The Gamma and NB distributions provide the best fit for waves 3 and 4 PTSS, as evidenced by their lowest BIC profiles (Supplementary Table 1 and Supplementary Figs. 2 and 3). The 1,000 bootstrapping replicates demonstrate similar performance of Gamma and NB on PTSS profile fitting (P = 0.605 and P = 0.628 for BIC of NB fit in waves 3 and 4, respectively; Supplementary Fig. 4). We opted for NB due to its flexibility in modeling overdispersed continuous count-like data such as PTSS variables (as shown in Supplementary Figs. 2b and 3) and its widespread use in modeling survey-based data^37–39.

We employed NB regression for the testing of key factors contributing to PTSS. These factors include lifetime and current social adversity profiles for key factors identified as important to prospective PTSS in our previous work⁴⁰ (that is, cumulative trauma, loneliness, perceived discrimination, financial difficulties and emotional mistreatment). In addition, we investigated the role of genetic background of individuals, quantified as a polygenic risk score (PRS), and cell surrogate proportion⁴¹, in contributing to PTSS in the context of all adversity factors. The PRS for PTSD was computed using PRSice-2^42,43 based on the most recent genome-wide association study (GWAS) of PTSD among individuals of African ancestry⁴⁴ (see Methods for detail).

We utilized individual-level social adversity data—including cumulative trauma, perceived discrimination, loneliness, financial difficulties and emotional mistreatment—from either wave 2 or a combination of waves 2 and 3 to predict PTSS in wave 3, while wave 4 PTSS was predicted using adversity factors from wave 4 (Supplementary Tables 2–4; see Methods for detail).

Our findings from the modeling of wave 4 PTSS suggest comparable model performance when using solely lifetime social adversity profiles in contrast to incorporating both lifetime and current profiles, or along with PRS (tenfold cross-validation Spearman correlation of 0.624 versus 0.623; Supplementary Table 4). A similar pattern emerged in the prediction of PTSS for wave 3, where using both waves 2 and 3 social adversity profiles resulted in tenfold cross-validation Spearman correlations of 0.632 for using solely lifetime profiles and 0.639 for the combination of both lifetime and current profiles (Supplementary Table 3). These results also suggest a limited contribution of emotional mistreatment to PTSS prediction when considering only lifetime social adversities. (F test P > 0.5; Supplementary Tables 3 and 4). An exception occurred when utilizing exclusively wave 2 social adversity profiles for predicting wave 3 PTSS, where there was approximately a 0.05 decrease in the Spearman correlation coefficient when using solely lifetime profiles compared with using a combination of lifetime and current profiles (Supplementary Table 2).

The incorporation of PRS or cell surrogate proportion led to a reduction of Spearman correlation, typically ranging 0.01 to 0.05 in most instances compared with utilizing lifetime social adversity profiles (Supplementary Tables 2–4). Our findings indicate that NB regression using only lifetime social adversity yields comparable model performance to including both current and lifetime profiles. Given the larger sample size accessible for lifetime social adversity (wave 3 prediction; Supplementary Tables 2 and 3) and the reduced model feature space, we opted to include only lifetime profiles for our subsequent analyses. This model omits the inclusion of PRS and cell surrogate, given their limited contribution to PTSS prediction (Supplementary Table 4).

Several earlier studies have shown that prior PTSS has the highest predictive capability for the future risk of PTSS^40,45–48. In this Article, we also sought to integrate past PTSS data into our NB regression models for forecasting PTSS in waves 3 and 4. We found strong Spearman correlation coefficients for our predictions. Wave 3 PTSS predicted from wave 2 social adversity factors determined a Spearman correlation of 0.962 in tenfold cross-validation (detailed data not shown). Similarly, the combination of waves 2 and 3 factors led to a correlation of 0.958 for wave 3 predictions, and the combination of waves 4 and 3 factors yielded a correlation of 0.94 for wave 4 predictions (detailed data not shown). These findings validate our earlier observations that prior post-traumatic psychopathology ranked highest in predicting the prospective risk of PTSD⁴⁰. However, it is important to note that these levels were largely influenced by the substantial correlation between PTSS measurements in different waves, as illustrated in Supplementary Table 5, where correlations exceeded 0.86. In addition, given our focus on exploring the link between social adversity and PTSS, incorporating PTSS data from previous waves might restrict the impact of other factors on PTSS prediction. For example, our established NB regression demonstrated the significant contribution of lifetime social adversity factors to wave 4 PTSS (as indicated by the F test, P < 0.05; Supplementary Table 4). However, when PTSS from prior waves was incorporated into the model, the contribution of these factors diminished (data not shown). Notably, we observed significant contributions only from trauma and wave 3 PTSS to wave 4 PTSS. Besides the prior wave PTSS profile, the stronger association between lifetime social adversity profiles and PTSS compared with other factors leads us to use lifetime social adversity to model PTSS.

Identifying miRNA signatures that modify the relation between social adversity and PTSS

NB regression was used to assess the impact of miRNAs on PTSS, incorporating all five previously mentioned social adversities. A total of 323 individuals from wave 3 and 243 individuals from wave 4 with social adversity profiles and miRNA expression data were included in this analysis (Table 1 and Supplementary Fig. 5). PTSS was represented by the sum of miRNA expression levels and lifetime social adversities (individual variables of cumulative trauma, loneliness, perceived discrimination and financial difficulties; see Methods for details) within the NB regression framework. Emotional mistreatment was initially evaluated alongside other adversity domains, but it demonstrated minimal predictive value for PTSS within our modeling framework. Specifically, emotional mistreatment was not significantly associated with PTSS in the NB regression analysis (P > 0.05), as shown in the F tests (Supplementary Tables 3 and 4), when considered in the context of other lifetime adversity measures. The Wald test³⁶ was employed to examine the direct contribution of miRNA expression to PTSS. We identified four miRNAs contributing to PTSS in wave 3 (Supplementary Table 6), and zero miRNAs contributed to PTSS in wave 4. We further conducted research to identify whether miRNAs might modify the relationship between social adversity and PTSS. Two variables exhibit interaction when the impact of one variable is contingent on the presence or values of the other variable⁴⁹. In our case, the effect of social adversity or the association with PTSS could depend on miRNA abundance, which is the typical case of interaction between variables. The interaction of any two variables related to PTSS profiles can be tested by NB regression (Fig. 1). We included all five lifetime social adversity measures in the NB regression, aligning with the model for exploring the main effect of miRNAs. The coefficient c of the multiplication term indicates the level of the interaction effect, and the Wald test can evaluate its statistical significance⁵⁰. We calculated a score for each lifetime social adversity profile, illustrating the influence of each miRNA interaction with a particular social adversity on PTSS in either wave 3 or wave 4 (Methods), where the z score of each miRNA’s interaction is the interaction coefficient c divided by its standard error. We identified 74 miRNAs that significantly modify the association between perceived discrimination and PTSS in wave 3, one miRNA that modifies the relationship between lifetime financial difficulties and PTSS in wave 3, and ten miRNAs that modify the association between lifetime cumulative trauma and PTSS in wave 4 (false discovery rate (FDR) < 0.1; Fig. 2a and Supplementary Tables 7–9). We list all modulation scores of the identified miRNAs modifying the association between social adversity and PTSS in Supplementary Table 10. The results include 23 miRNAs that have been previously reported in association with conditions or traits related to PTSD^{11,13,51–64}, including traumatic brain injury and stress response (Supplementary Table 11).

Table 1.

Demographic characteristics of the individuals involved in the statistical framework for studying the modulatory effects of miRNA signatures on the relationship between social adversity and PTSS

Demographic profile	Wave 2	Wave 3	Wave 4
Age, mean (s.d.)	56.5 (15.5)	57.4 (15.6)	59.8 (14.1)
Gender, female (%)	187 (57.8)		140 (57.6)
Self-reported race, Black or African American (%)	256 (79.5)		196 (80.7)
Self-reported race, American Indian or Alaska Native (%)	3 (0.9)		3 (1.2)
Self-reported race, Native Hawaiian or Other Pacific Islander (%)	1 (0.3)		1 (0.4)
Self-reported race, white (%)	46 (14.2)		35 (14.4)
Self-reported race, other (includes those who specified only Hispanic for this question and mixed races) (%)	16 (5.0)		8 (3.3)
Marital status, married (%)	87 (26.9)	85 (26.3)	67 (27.6)
Marital status, divorced (%)	74 (22.9)	71 (22)	44 (18.1)
Marital status, separated (%)	15 (4.6)	16 (5)	45 (18.5)
Marital status, widowed (%)	46 (14.2)	53 (16.4)	13 (5.3)
Marital status, never been married (%)	101 (31.3)	98 (30.3)	63 (25.9)
Marital status, living with partner (%)			11 (4.5)
Education, kindergarten to grade 8 (%)	3 (0.9)	3 (0.9)	1 (0.4)
Education, some high school (grades 9–11) (%)	39 (12.1)	34 (10.5)	24 (9.9)
Education, high school equivalency (GED) (%)	16 (5)	21 (6.5)	10 (4.1)
Education, high school graduate (grade 12) (%)	71 (22)	69 (21.4)	57 (23.5)
Education, some college or technical training (%)	117 (36.2)	112 (34.7)	88 (36.2)
Education, college graduate (4-year) (%)	42 (13)	41 (12.7)	33 (13.6)
Education, graduate work (%)	35 (10.8)	43 (13.3)	30 (12.3)
Employment status, full time (%)	85 (26.3)	66 (20.4)
Employment status, part time (%)	23 (7.1)	37 (11.5)
Employment status, not employed (%)	212 (65.6)	218 (67.5)
Lifetime financial problems, yes (%)	202 (62.5)		165 (67.9)
Lifetime emotionally mistreated, yes (%)	102 (31.6)		95 (39.1)
Loneliness scale, mean (s.d.)		1.4 (0.5)	1.4 (0.6)
Perceived discrimination, mean (s.d.)		3.4 (0.5)	3.3 (0.6)
PTSS, mean (s.d.)		36.5 (15.4)	39.6 (16.2)

Fig. 1. The interaction test within a general linear regression is employed to detect miRNAs linked to the impact of social adversity on PTSS.

The coefficient c signifies the interaction effect between social adversity and miRNA expression levels concerning PTSS outcomes. The graphs illustrate the association slopes between social adversity and PTSS, with light and dark blue denoting slopes before and after elevated miRNA expression levels. In the plot on the left, heightened miRNA levels are associated with an increase in PTSS in the context of a particular social adversity exposure, whereas the plot on the right suggests the converse effect, with increased miRNAs showing a potentially protective effect against PTSS in the context of a particular social adversity exposure.

Fig. 2. The statistical framework reveals the influence of miRNA signatures on the impact of social adversity on PTSS.

a, miRNAs play a significant moderating role in the associations between social adversity and PTSS, as determined by a two-sided Wald test with an FDR threshold of 0.1. The modulation scores of miRNAs, defined as the z score, are obtained by dividing the coefficient (c) by the standard error in NB regression. The 23 highlighted miRNAs are those associated with previously reported signatures linked to PTSD-related phenotypes. We listed all modulation scores of the identified miRNAs modifying the association between social adversity and PTSS in Supplementary Table 10. b, The expression of miR-208 is associated with the connection between financial difficulties and PTSS. This relationship is illustrated in the scatter plot. Samples are categorized according to their miR-208 expression profiles. Among participant samples lacking miR-208 expression, there is a positive correlation between financial problems and PTSS. However, for samples expressing miR-208, this correlation diminishes. c–e, The relationship between PTSS and perceived discrimination. A lower value on the horizontal axis indicates that participant perceived higher levels of discrimination. The orange and black samples specifically represent participant samples with high and low miRNA expression based on the median profile. The association between perceived discrimination and PTSS is reduced when miRNA levels are lower. f,g, As cumulative trauma increases, PTSS also increases even more among those with higher miRNA levels in the examples shown.

Our findings provide new insight into a group of miRNAs that overlap with those identified in a small pilot investigation conducted in a human clinical cohort¹³, where expression profiles differed between individuals with PTSD and both trauma-exposed individuals without PTSD (referred to as resilient) and nonexposed controls. The linear combination of these miRNA concentrations in the originally reported cluster showed significant differences between individuals with PTSD and resilient individuals, as well as between trauma-exposed individuals and nonexposed controls. However, none of these miRNAs individually showed differential expression in this study¹³. By contrast, our results suggest that several of these miRNAs may influence PTSD symptoms by modulating the effects of social adversities, providing a complementary perspective on their potential regulatory role in trauma-related outcomes. For example, the PTSS profiles from wave 2 to wave 3 in the individuals expressing miR-208 are less associated with financial problems, as evidenced by the shallower slope of the red line in Fig. 2b. Conversely, participants with higher expression levels of miR-27a, miR-34a and miR-505 may experience a protective effect against PTSS in the context of the investigated social adversities, at least at lower levels of perceived discrimination. By contrast, participants with higher expression levels of miR-411 show the opposite trend, with their PTSS being worsened in the context of the investigated social adversities (Fig. 2c–f). An additional instance of our model’s application is demonstrated with miR-431. Previous findings from a mouse model indicated that miR-431 has the potential to heighten the susceptibility of the inner ear to noise-induced trauma⁶¹. Our findings also suggest that the augmentation of miR-431 results worsens PTSS in response to cumulative trauma (Fig. 2g).

The identified miRNA signatures are involved in immune response, cell cycle and differentiation, as well as brain and neural functions

Our next goal was to identify the biological pathways targeted by the miRNAs we identified as potentially involved in the modulating relationship between social adversity and PTSS (that is, 74 miRNAs for perceived discrimination, 10 miRNAs for cumulative trauma and 1 miRNA for financial difficulties). To this end, we downloaded annotated target genes for each miRNA identified in our study from miRDB (https://mirdb.org/) with a stringent binding score cut-off of 80 (ref. ⁶⁵). We then performed Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis⁶⁶ for the miRNAs identified for each social adversity. This procedure identified 94 unique KEGG terms enriched in the target genes of miRNAs that modulate PTSS through interactions with social adversities (Supplementary Table 12). These terms (among others) are categorized into hormone metabolism, immunity, cell cycle and differentiation, brain and neural functions, and diseases (Fig. 3). We presented these categories according to whether they are involved in modulating perceived discrimination, trauma or both (the financial problem was excluded because only a single miRNA was found to influence the relationship between lifetime financial difficulties and PTSS, providing insufficient input for robust statistical interpretation).

Fig. 3. KEGG pathway enrichment analysis was conducted to explore miRNAs that could potentially modulate the association between PTSS and perceived discrimination, or between PTSS and cumulative trauma.

Each pixel reflects the adjusted P value of KEGG pathway enrichment for miRNA target genes identified for modulating the relationship between PTSS and specific social adversities. Color intensity reflects the level of statistical significance, based on two-sided Fisher’s exact tests with FDR-adjusted P values.

Discussion

This study demonstrates a significant, prospective association between blood-based measures of miRNA and traumatic stress following exposure to social adversity in a community-based, prospective cohort of predominantly African American participants. In the context of specific social adversity experiences, individuals exhibiting distinct miRNA expression patterns may be at an elevated risk of PTSS. Specifically, we found a group of miRNAs that can either intensify or diminish the correlation between social adversity and PTSS as their expression profiles increase. Conversely, individuals lacking this miRNA expression pattern may enjoy a protective effect against PTSS, even when experiencing similar levels of social adversity. Several of these identified miRNAs have been previously linked to traumatic brain injury^62–64, underscoring and strengthening earlier research that has demonstrated an association between traumatic brain injury and PTSD^67,68. Furthermore, the target genes of the identified miRNAs primarily participate in immune response, cell cycle regulation and differentiation, brain and neural functions, and other diseases related to PTSD. These findings offer additional evidence connecting the aforementioned biological pathways to PTSD. Importantly, our findings emphasize a fresh viewpoint on these pathways (immune response, cell cycle regulation and differentiation, brain and neural functions, and other diseases) implicated in PTSD, potentially affecting the interaction between social adversity and PTSS. Our study contributes to the understanding of how social adversity influences psychological outcomes, driven by underlying biological molecular mechanisms.

Our analysis indicates that lifetime social adversity has a similar predictive capacity compared to the linear combination of all current social adversity factors. Lifetime social adversity, which aggregates the average profiles across all considered waves, also ensures a higher sample size as some individuals missed specific adversity surveys in certain waves (Supplementary Tables 3 and 4). Within our statistical framework, lifetime social adversities were employed to investigate whether miRNA modified PTSS following exposure to social adversity. Perceived discrimination and loneliness data are available only for wave 3. The wave 3 and the following wave 4 PTSS survey enabled us to explore potential miRNA contributions, measured in wave 2 and wave 4, to PTSS outcomes following exposure to perceived discrimination and loneliness reported in wave 3. We identified miRNAs in wave 2 that significantly modified the association between perceived discrimination and PTSS in wave 3. To assess the temporal stability of these miRNAs, we then examined the modulation scores of these miRNAs in wave 4, again in relation to perceived discrimination. Importantly, the miRNAs identified from wave 2 showed significantly higher modulation scores in wave 4 compared with other miRNAs (Supplementary Fig. 6a; Wilcoxon test < 1 × 10⁻³). This observation partially validates our study framework, demonstrating that miRNAs identified in one wave retain or exhibit consistent effects in subsequent waves. Similarly, the identified miRNAs in wave 4, significantly modifying the association between lifetime cumulative trauma and PTSS, also demonstrated significant modulation score compared with other miRNAs in wave 2 (Supplementary Fig. 6b; Wilcoxon test < 1 × 10⁻³). These findings underscore the robustness of our study design in identifying modifying miRNAs.

On ordering all KEGG terms according to their enrichment P-adjustment values (Fig. 3 and Supplementary Table 12), we observed a cluster of functional terms that are shared by two groups of miRNAs, which modulate perceived discrimination and cumulative trauma. The Ras signaling pathway has previously been linked to PTSD⁶⁹, whereas the TGF-beta and Hippo signaling pathways have been recognized as significant connections to stress-regulated pathways^70,71. These pathways are intricately involved in cell growth, proliferation and differentiation processes⁷². Moreover, the pathways identified include long-term potentiation, glutamatergic synapses, MAPK, cAMP, neurotrophins and cGMP-PKG, which collectively form vital components of the intricate network governing synaptic plasticity, neuronal signaling and cognitive function in the brain^73–77. Our findings also highlight the involvement of two immune response pathways, endocytosis and mitophagy, which have been previously associated with PTSD^78,79. Overall, our results offer new insights into the pathways previously reported in PTSD, elucidating their regulation potentially via miRNA and their impact on the relationship between social adversities and PTSS.

Our research introduces a well-specified statistical framework for identifying a biological molecular signature that underlies associations between social adversity and PTSS. This framework can be applied to investigate any third factor that may potentially modify known one-to-one relationships. A similar framework has been employed to explore the human immune response over time to varying levels of racial discrimination⁸⁰. While this robust framework has the potential to uncover protective biological factors that counteract adverse influences, it falls short of establishing a causal relationship between the identified miRNA and the heightened risk of PTSD. Further experimentation is necessary to validate the contribution of the identified miRNA or related biological factors. This could include functional assays using miRNA mimics or inhibitors in immune or neuronal cell models, in vivo manipulation in stress-responsive animal models, and longer-term longitudinal studies to assess whether miRNA expression prospectively predicts PTSD symptoms or mediates the effects of adversity. Nonetheless, our study lays the groundwork for understanding the inherent vulnerability factors influencing psychological health disparities.

We included many lifetime social adversity factors, except emotional mistreatment. Although emotional mistreatment is widely recognized in the trauma literature as a meaningful predictor of stress-related outcomes^81,82, it was excluded from our miRNA moderation analyses due to its limited contribution (Wald test > 0.05; Supplementary Tables 3 and 4) in our normal NB regression analysis to explore the lifetime social adversity to PTSS outcomes. It is important to emphasize that the social adversity data—including cumulative trauma, perceived discrimination, loneliness, financial difficulties and emotional mistreatment—were derived from annual survey responses, reflecting a potentially prolonged period of individual exposure to social adversity. By contrast, miRNAs were extracted at a specific time point, capturing the expression profile only at that particular moment. Even though the miRNA maintains a high consistency between waves 2 and 4, as reflected by a Spearman correlation greater than 0.75 in the majority of individuals (Supplementary Fig. 7), the fluctuation in miRNA expression over time could still influence their ultimate contributions to PTSS. Although sequencing multiple time points and statistically testing the consistent modulation potential of miRNA across all waves is a logical approach, as demonstrated in our studies and consistent in waves 3 and 4 (Supplementary Fig. 5), we should aim to enhance experimental design and statistical testing to simultaneously incorporate all sequencing data comprehensively. Importantly, the variation in biospecimen availability and participant retention across waves introduces potential for selective attrition and sampling bias, which may impact the generalizability or stability of observed miRNA associations. Although we did not formally assess subsample comparability in this study, previous work using the DNHS cohort has shown that individuals retained across multiple waves tend to report lower baseline levels of trauma and psychiatric symptoms than those who were lost to follow-up⁸³, highlighting the potential for bias. We now explicitly acknowledge this limitation and emphasize the importance of future research designs that incorporate harmonized sampling protocols, formal assessments of attrition-related bias and analytic frameworks capable of integrating data across time points.

In general, it is both challenging and essential to pinpoint the biological factors that influence the outcome of traumatic stress among individuals exposed to social adversities. To address this, we have introduced a robust statistical framework to comprehend how miRNAs, an understudied epigenetic component in both community-based contexts and PTSD, regulate the connection between social adversity and PTSS. Several miRNAs identified here target genes involved in pathways previously associated with PTSD, suggesting that genomic variation in this understudied element can help improve our understanding of vulnerability to PTSD in the face of exposure to distinct social adversities. Overall, our discoveries provide new insight into how miRNA shapes the interaction between social adversities and PTSS.

Methods

Ethics statement

This study was conducted in accordance with all relevant ethical regulations. All study protocols were reviewed and approved by the institutional review boards at the University of Michigan and the University of North Carolina. Written informed consent was obtained from all participants before participation and again at the time of biospecimen collection.

DNHS

In this investigation, we utilized data from the DNHS, a prospective population-based longitudinal cohort consisting of individuals residing in Detroit, Michigan^33,84. All participants involved in this study were aged 18 years or older and self-identified predominantly as African American. The primary objective of the DNHS was to examine how biological variation, stressful and traumatic life experiences, and environmental factors contribute to the prediction of psychopathology and behavior. Participants underwent structured telephone interviews annually between 2008 and 2013 to assess various aspects, including perceptions of their neighborhoods, mental and physical health status, social support, exposure to traumatic events, symptoms of PTSD, depression and generalized anxiety, as well as alcohol and tobacco use. Informed consent was obtained at the outset of each interview and reconfirmed at the time of specimen collection. Further details regarding DNA and RNA isolation procedures are provided in refs. ^85,86, respectively. The institutional review boards at the University of Michigan and University of North Carolina reviewed and approved this study.

PRS

A PRS for PTSD was calculated for each participant using summary statistics from the most recent GWAS of individuals with African ancestry from the Psychiatrics Genomics Consortiums PTSD Working Group excluding participants from the DNHS⁴⁴. For each participant, PRSs for PTSD were calculated by taking the weighted sum of risk alleles, with each allele weighted by the z scores from the GWAS summary statistics using PRSice-2^42,43. Risk alleles for PTSD were selected using P-value thresholds ranging from 5 × 10⁻⁸ to 1 with the best-fitting PRS determined by the maximum variance in DNHS PTSD diagnosis explained on the basis of Nagelkerke’s R (ref. ⁸⁷).

Measures

In this study, exposure to traumatic events throughout life was assessed using a survey instrument administered across waves of the DNHS³³. We operationalized adversity into two distinct categories: (1) potentially traumatic events (PTEs) and (2) social adversity. While trauma exposure reflects acute or chronic PTEs (for example, assault, injury, disaster), social adversity encompasses psychosocial stressors such as financial problems, emotional mistreatment, discrimination and perceived isolation.

Trauma exposure (PTEs) was assessed using a 19-item inventory derived from validated trauma surveys, as described in ref. ³³. Participants were asked whether they had ever experienced each of 19 potentially traumatic events, including serious accidents, assaults, witnessing violence or unexpected death of a loved one. The total cumulative trauma score was calculated by summing the number of different endorsed PTEs.

Social adversity was assessed through five domains. Perceived discrimination was assessed using the Everyday Discrimination Scale, a nine-item self-report questionnaire administered in wave 3 (ref. ⁸⁸). A total score was derived by summing responses, with higher values indicating greater perceived discrimination. Loneliness was gauged using a standard three-item UCLA-based scale in wave 3 (ref. ⁸⁹), where a higher score indicates increased loneliness. Emotional mistreatment and financial problems were evaluated as binary variables based on participant responses to standardized DNHS survey items asking whether they had every experienced these stressor across waves^84,90. Cumulative trauma (described above) was also considered in interaction modeling given its established relevance to PTSD.

Our outcome measure was a continuous measure of PTSD severity, calculated by summing scores for 17 symptoms (that is, PTSD Checklist–Civilian Version (PCL-C))⁹¹ stemming from the most severe lifetime trauma experienced by each participant. Participants self-reported demographic information, including age, sex, race, education, marital status and employment status. The current social adversity was characterized by either the accumulation of adverse social events (cumulative trauma) or the experience of any social adversity events (emotional mistreatment and financial problems) since the last interview in the preceding survey wave. Perceived discrimination and loneliness were assessed as mentioned above and available only in wave 3. Lifetime trauma was determined by totaling the cumulative trauma exposures reported across all previous sampling waves. Lifetime emotional mistreatment and financial problems were assessed according to whether any emotional mistreatment or financial issues were experienced, considering all previous waves.

Small RNA sequencing

The quality and quantity of the total RNA samples are assessed using Qubit fluorometric concentration (Invitrogen) and Tapestation electrophoresis integrity (Agilent). This ensured that only high-quality RNA was used for subsequent library preparation. In total, 638 samples (389 from wave 2 and 243 from wave 4) were selected originating from 483 unique participants. Library preparation was performed following the QIAseq miRNA Prep Kit protocol (QIAGEN) with QIAseq miRNA unique dual indices (set A–H; Qiagen). Final complementary DNA libraries were assessed for quality via Qubit (Invitrogen) and TapeStation (Agilent). miRNA sequencing was performed on the NextSeq 2000 platform (Illumina) using the P3 reagents/flow cell with 66 base pairs single read and 10 base pairs dual indexing (Illumina). Each sequencing plate included Qiagen XpressRef Universal Total RNA as a positive control.

Small RNA mapping

The quality control of sequencing results was performed by FASTQC (v0.11.9)⁹² and MULTIQC⁹³. The sequencing adapter was removed by the Trim Galore (v1.18)⁹⁴. The forward and reverse reads were merged by using the PEAR (v0.9.6)⁹⁵. The mirDeep2 package^96,97 was implemented to map the merged reads back to the human genome GRCh38 (hg38). The precursor and mature miRNA information was downloaded from miRbase (Release 22)⁹⁸. Correlation across positive control samples was consistently high (mean Pearson r = 0.98), indicating minimal technical batch effects. As batch-driven variation was not evident, no batch correction or adjustment was applied in the downstream analyses.

Cell surrogate proportion

The cell surrogate proportion was calculated by CellCODE (v0.99.0)⁴¹ to estimate the relative composition of BCell, CD8T, CD4T, NKCell, Mono and Gran cells.

Statistical modeling

Exploratory analysis revealed that PTSS exhibited right-skewed, overdispersed distributions, which motivated comparison across four candidate distributions: Poisson, gamma, NB and Weibull. Maximum log-likelihood estimation was used to optimize the parameters, and we employed the BIC to rank the model-fitting performance by 1,000 bootstrap replicates (R v4.2.0, MASS v7.3.6, glmnet v4.4.1). The NB distribution provided the best fit across waves (Supplementary Table 1 and Supplementary Figs. 2–4), making it the most appropriate choice given the overdispersion present in the PTSS data. The NB framework was also selected for its flexibility in handling overdispersed count-like data and its compatibility with survey-based data structures^37–39.

We then used Spearman correlation from tenfold cross-validation in NB regression to estimate the contributions of different factors, including gender, social adversities, PRS and cell surrogate proportion, to PTSS prediction. In the tenfold cross-validation process, the dataset was partitioned into ten roughly equal-sized subsets, referred to as folds. Our linear model was trained using nine of these folds, reserving one fold for testing the model’s performance. This procedure was repeated ten times, ensuring that each of the ten folds served once as the test set. The performance of the model was averaged across all ten iterations to derive an estimation of its performance. The tenfold cross-validation in NB regression with L1 penalty was also utilized to validate the results.

An interaction test within NB regression was utilized to pinpoint miRNAs that might influence the interplay between social adversities and PTSS. In statistical terms, two variables interact when the impact of one variable is dependent on the state of the other⁴⁹. A multiplication term in a multivariate linear model can assess this interaction effect between two variables. We constructed a linear model

$$\begin{array}{c}\mathrm{P}T\mathrm{SS}=a_j\times \mathrm{social}\_{\mathrm{adversity}}_j+b\times \mathrm{miRNA}\hfill \\ +c\times \left(\mathrm{social}\_{\mathrm{adversity}}_j\times \mathrm{miRNA}\right)\hfill \\ +{\sum }_{i\ne j}{a}_i\times \mathrm{social}\_{\mathrm{adversity}}_i,\hfill \end{array}$$

where ‘miRNA’ denotes the expression level of a specific miRNA under examination. The model enables a direct investigation of the contribution of a specific social adversity $a_j\times {\mathrm{social\_adversity}}_j$ and a specific miRNA $b\times \mathrm{miRNA}$, as well as the interaction between social adversities and miRNA $c\times ({\mathrm{social\_adversity}}_j\times \mathrm{miRNA})$ to PTSS. In addition, we controlled for the contributions of other social adversities, represented as ${\sum }_{i\ne j}{a}_i\times {\mathrm{social\_adversity}}_i$. The primary contribution of each miRNA to PTSS was indicated as $b\times \mathrm{miRNA}$, where $b$ represents the linear coefficient. The relationship between the investigated social adversity and PTSS can be described as $a_j+c\times \mathrm{miRNA}$. The estimate (c) represents the modulatory effects of miRNA on PTSS. If the coefficient c is positive, it implies that higher miRNA expression levels are associated with increased PTSS in the context of a given social adversity (positive effect). Conversely, if c is negative, higher miRNA levels are associated with reduced PTSS in the context of a given social adversity, suggesting a protective effect. The modulation score for each miRNA is defined as the Wald test $z$ score, calculated as the coefficient $c$ divided by its standard error. To identify significant miRNAs in the interaction test, we employed FDRs to adjust the two-sided Wald test P values, selecting miRNAs with an FDR below 0.1 as significant candidates in the interaction test. This threshold is supported by previous studies applying similar thresholds in miRNA and epigenomic research involving modest effect sizes and high dimensionality⁹⁹.

The main effects of miRNA were modeled using the previously described linear NB framework without interaction terms. Specifically, the linear model can be expressed as $\mathrm{PTSS}={\sum }_i{a}_i\times {\mathrm{social\_adversity}}_i$$+b\times \mathrm{miRNA}$. The Wald test was then applied to evaluate whether the expression of the specific miRNA significantly contributes to PTSS.

KEGG enrichment analysis

To elucidate the biological pathways influenced by miRNAs, we retrieved all annotated target genes for each miRNA identified in our study from the miRDB database (https://mirdb.org/) using a stringent binding score threshold of 80 (ref. ⁶⁵). To assess the functional relevance of these miRNAs, we performed pathway enrichment analysis to determine whether specific KEGG pathways were significantly enriched among the target genes. Our focus was on miRNAs identified in the modulation of the relationship between social adversities and PTSS. The enrichment analysis was conducted using a hypergeometric test with P values adjusted by the FDR, and only pathways with an adjusted P value < 0.1 were considered significantly enriched.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

C.W. contributed to the conceptualization and methodology of the study, performed the analyses, interpreted the results and led the writing of the original draft as well as manuscript review and editing. M.U. contributed to funding acquisition, study supervision, conceptualization, methodology, interpretation of the results and manuscript review and editing. A.W. contributed to the methodology and data analyses and participated in manuscript review and editing. Z.G. contributed to the methodology and data analyses. A.R. contributed to the methodology and data analyses and participated in manuscript review and editing. A.E.A. contributed to funding acquisition and manuscript review and editing. K.K. contributed to manuscript review and editing. M.M. contributed to data analyses. D.E.W. contributed to funding acquisition, study supervision, conceptualization, methodology, interpretation of the results and the writing of the original draft, as well as manuscript review and editing.

Peer review

Peer review information

Nature Mental Health thanks the anonymous reviewers for their contribution to the peer review of this work.

Data availability

The data supporting the findings of this study have been deposited in the Database of Genotypes and Phenotypes (dbGaP) under accession number phs000560.v3.p1. Access to the controlled data will be granted to qualified investigators for research use consistent with the informed consent of study participants. Requests for access should be submitted through the dbGaP accession page and will be reviewed by the NIH Data Access Committee (DAC) in accordance with NIH Genomic Data Sharing Policy. Investigators may contact the corresponding author for additional information regarding data content and access procedures. Data use is subject to the conditions outlined in the dbGaP Data Use Certification Agreement, including restrictions on re-identification and redistribution.

Code availability

The R scripts used for data analysis are available via GitHub (https://github.com/CharleyWang/miRNA_SocialAdversity_PTSS_Framework) and have been archived in Zenodo (10.5281/zenodo.17619725)¹⁰⁰.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s44220-025-00581-6.