Influence of Air Pollution Exposure on the microRNA Content of Neuronal Extracellular Vesicles in Subjects Affected by Major Depressive Disorder

Abstract

Major depressive disorder (MDD) has been recently linked to air pollution exposure; nevertheless, the biological mechanisms underlying this association remain underinvestigated. Air pollution might modulate the microRNA (miRNA) content of neuron-derived extracellular vesicles (NdEVs), potentially mirroring brain epigenetic alterations. In the present study, we investigated the relationship between air pollution, NdEV miRNAs, and MDD severity in a population of 200 patients with depression. After signing informed consent, participants compiled questionnaires about demographics, lifestyle and clinical history, and donated a blood sample. MDD severity was assessed by five scales. Particulate matter ≤2.5 μm (PM_2.5) and nitrogen dioxide (NO₂) exposure was assigned based on participants’ residential address. Plasma NdEVs were obtained by L1CAM immunocapture. NdEV miRNAs were queried by RT-qPCR (microarray) following a two-stage approach. Associations between air pollutants, NdEV miRNAs, and MDD severity were assessed by multivariable regression models. The regulatory function of NdEV miRNAs was investigated by gene target and pathway analysis. As a result, exposure to NO₂ was associated with decreased levels of miR-191 and miR-24, while PM_2.5 was negatively associated with miR-191, miR-223, miR-24, miR-320, miR-451, miR-572, and miR-638. Decreased miR-126, miR-19b, miR-320, miR-451, miR-572, and miR-638 were associated with higher MDD severity scores. Target genes at the interface between air pollution exposure and MDD severity were mainly involved in inflammation and cell cycle regulation. These findings suggest that air pollutants might modulate MDD severity by triggering NdEV miRNA alterations. Longitudinal studies are needed to evaluate whether NdEV miRNAs might serve as novel biomarkers for MDD prognosis.

1. Introduction

Poor air quality represents one of the major environmental threats to human health. It is estimated that more than 90% of global population is exposed to excessive concentrations of air pollutants, including particulate matter (PM) and gaseous molecules such as nitrogen and carbon oxides. Exposure to air pollutants is known to increase the risk for many diseases, mainly at the cardiovascular and respiratory levels; however, recent evidence also supports a relationship with psychiatric disorders such as depression.

Major depressive disorder (MDD) is a disabling mood disorder characterized by a wide spectrum of behavioral and cognitive symptoms. MDD affects more than 280 million people worldwide and is projected to become the leading cause of global disease burden by 2030. Despite its huge social and economic impact, there is only a partial understanding of the molecular mechanisms underlying MDD. This is partly due to the fact that this pathology is characterized by a complex pattern of biological alterations, which include unbalanced neurotransmission, hormonal dysregulation, and neuroinflammation. The main hypothesis to explain the association between air pollution and MDD revolves around the pro-inflammatory and pro-oxidative action of several airborne pollutants, exerted either with a direct action on the central nervous system or indirectly, by triggering the release of pro-inflammatory mediators that reach the brain. In either case, air pollution exposure would result in neurotoxicity and vascular brain lesions, which may exacerbate depressive symptoms.

Evidence from in vivo and in vitro studies suggests that exposure to environmental stressors can cause epigenetic changes in the brain. Among them, air pollution has been associated with alterations of circulating microRNAs (miRNAs), which are key players in gene expression regulation. Of note, MDD patients show a peculiar profile of circulating miRNAs, which correlates with MDD severity and can be modulated by antidepressants. , Nevertheless, pollution-induced miRNA alterations occurring in the human brain remain unexplored, mainly due to the lack of minimally invasive methods for their investigation.

In this framework, circulating neuron-derived extracellular vesicles (NdEVs) might provide a powerful system to assess brain miRNA alterations elicited by air pollution exposure. NdEVs, which are characterized by the presence of the L1 cell adhesion molecule (L1CAM) on their lipidic membrane, are generated by neurons and secreted into the extracellular space, where they can cross the blood–brain barrier and travel to peripheral tissues. , By shuttling their molecular cargo from parental cells to recipient ones, NdEVs might orchestrate a systemic response to environmental and pathological cues; indeed, alterations in the number, size and content of circulating NdEVs have been observed in multiple neurological and psychiatric diseases − and in response to lifestyle interventions or stress. − We have previously reported a positive association between increasing air pollution exposure levels and MDD severity in subjects with MDD. In this context, this study aims to determine whether exposure to air pollution might influence the levels of NdEVs miRNA, and if changes in such miRNAs might be associated with illness severity, in a subset (n = 200) of the same population.

2. Methods

2.1. Subject Characteristics and MDD Severity Assessment

The study was conducted on 200 patients with MDD, as summarized in Figure . Participants were randomly sampled from the larger population (n = 416) of the DeprAir study (CARIPLO Foundation −2019–3354; https://deprair.com/en/homepage/). Subjects of both sexes aged 18–65 and with a previous diagnosis of MDD were considered eligible for this study. Exclusions criteria were the following: having medical conditions potentially associated with behavioral disorders (e.g., hypothyroidism, previous stroke), drug abuse in the previous month, other psychiatric disorders (except for personality disorders other than borderline personality disorder), medical conditions or treatments that are likely to strongly alter inflammatory parameters (e.g., autoimmune diseases, steroids or interferon treatment), infectious diseases, pregnancy. By signing the informed consent, participants agreed to grant access to personal medical records, donate a 30 mL blood sample, and fill in two questionnaires to collect socio-demographic data, recent residential history, lifestyle information, and clinical history.

1.

Study workflow.

For each subject, MDD diagnosis was confirmed using the Structural Clinical Interview for DSM-5 (SCID-Italian version). Briefly, depression severity was assessed with the following rating scales: Montgomery-Asberg Depression Rating Scale (MADRS), which evaluates core symptoms of MDD (score: 0–60); Hamilton Depression Rating Scale (HAM-D), assesses anxiety and somatic symptoms of MDD (score: 0–67); clinical global impression-severity of illness (CGI), which evaluates the global severity of illness (score: 1–7); global assessment of functioning (GAF), which evaluates the overall impairment associated with MDD (score: 0–100); Sheehan Disability Scale (SDS), which assesses functional impairment in five domains: work, home management, family responsibilities, perceived stress (score: 0–10 for each domain), and social support (score: 0–100).

2.2. Exposure Assessment

Air pollution exposure estimates for PM with an aerodynamic diameter ≤2.5 μm (PM_2.5) and nitrogen dioxide (NO₂). First, each patient’s residential address was translated into spatial coordinates using the web tool GPS Visualizer (https://www.gpsvisualizer.com) and geocoded using QGIS (http://qgis.osgeo.org). Then, air pollution exposure levels were assigned to each study participant using daily mean estimates derived from the Flexible Air quality Regional (FARM) model of the Lombardy region environmental protection agency (ARPA Lombardia), which are available as daily means at a 1 × 1 km resolution. ,, Daily estimates of each pollutant’s exposure were obtained for each day starting from the day of recruitment (day 0) to 365 days before recruitment (day 365). Further methodological details, including model performance, can also be found in ref .

To analyze short- and midterm effects of air pollution, different temporal windows were considered, by averaging pollutant levels of the day of recruitment with the levels of each preceding day up to 7 (lag 0–7), 14 (lag 0–14), 21 (lag 0–21), or 30 (lag 0–30) days before recruitment. We focused on short-to midterm time windows, since all the used severity scales investigate symptoms and characteristics which roughly refer to some days to few weeks preceding the evaluation, and EVs are known to be rapidly released in response to environmental cues, contributing to acute stress responses.

Exposure to apparent temperature (AT) was also estimated. Daily means of temperature, humidity, and wind speed were retrieved from the weather monitoring stations of ARPA Lombardia and then assigned to each subject considering the closest station to his/her residential address. AT was estimated following the formula that can be found in previous publication. The same temporal exposure windows used for air pollutants were applied to AT and included in the analyses.

2.3. Collection of Blood Samples and Immunoaffinity Enrichment of Neuronal EVs

Blood drawing was performed directly by the recruiting staff. Blood samples were collected into EDTA tubes, which were centrifuged at 1200 × g for 15 min to separate plasma and buffy coat fractions. To preserve the physiochemical properties of plasma-derived EVs, plasma samples were stored at −80 °C with 0.04 mol/L sucrose and thawed only once. After centrifuging 1.6 mL plasma at 1000, 2000, and 3000 × g, the supernatant was ultracentrifuged at 110,000 × g for 75 min to obtain an EV-rich pellet. Neuron-derived EVs (NdEVs) were enriched by immunoaffinity capture, according to the protocol described in, with minor modifications. After adding bovine serum albumin (BSA, #A7030, Sigma-Aldrich) and protease and phosphatase inhibitors (PPI, #78447, Thermo Fisher Scientific), total EVs were incubated with an anti-L1CAM biotinylated antibody (1:150, eBio5G3, #13–1719–82, Thermo Fisher Scientific) for 1 h on a rotating wheel at room temperature. Following addition of prewashed Pierce Streptavidin Plus UltraLink Resin (#53116, Thermo Fisher Scientific) for 30 min on a rotating wheel, the suspension was centrifuged at 200 × g for 10 min to obtain L1CAM + EVs. Then, after removing the supernatant, each sample was added with Buffer Gly-HCl pH 3.0 0,1 mol/L (#24074, Polysciences) and incubated for 10 min on a rotating wheel, allowing detachment of L1CAM + EVs from the beads. Samples were then centrifuged at 4500 × g for 10 min, and the supernatant containing L1CAM + EVs was collected and added with PPI and Buffer Tris HCl pH 8.0 1 mol/L (#J22638.AE, Thermo Fisher Scientific). Samples were finally stored at −20 °C.

2.4. Flow Cytometry and Nanoparticle Tracking Analysis

Flow cytometry analysis was carried out using the MACSQuant Analyzer (Miltenyi Biotec) according to. To evaluate EV integrity, sample aliquots were stained with 200 nmol/L 5(6)-carboxyfluorescein diacetate N-succinimidyl ester (CFSE) at 37 °C for 30 min in the dark. After CFSE staining, EVs were incubated with anti-L1CAM primary antibody (1:500, EPR18750, #ab208155, Abcam) for 30 min at 4 °C, followed by incubation with Alexa Fluor 647-conjugated secondary antibody (1:500, #A21245, Invitrogen) for 30 min at 4 °C in the dark. Data analysis was conducted using FlowJo (v. 10.8.1, Becton Dickinson).

The concentration and size distribution of EVs were assessed by Nanoparticle Tracking Analysis (NTA) using the NanoSight NS300 system (Malvern Panalytical Ltd., Malvern, UK). Four total EV samples (derived from plasma pools of healthy subjects) and four NdEVs samples (isolated from an aliquot of total EV samples) were employed for this analysis. Five 30s recordings were made for each sample. Collected data were analyzed with the NTA software (v. 3.3, Malvern Instruments Ltd.).

2.5. Extraction of EV miRNAs

EV miRNAs were extracted using the miRNeasy Mini Kit and RNeasy MinElute Cleanup Kit (#217004 and #74204, Qiagen). Briefly, 200 μL of NdEV samples were added with 1 mL QIAzol Lysis Reagent, mixed thoroughly, and incubated for 5 min; then, samples were added with 200 μL of chloroform and shaken vigorously for 30 s. Samples were then centrifuged for 15 min at 12,000 × g, and the aqueous phase was mixed with an equal volume of Ethanol 70%. The resulting solution was loaded onto a RNeasy Mini spin column and centrifuged for 30 s at 11,000 × g to remove long RNA molecules. The eluate was then added with Ethanol 100%, mixed thoroughly, transferred onto a RNeasy MinElute spin column, and centrifuged for 30 s at 11,000 × g. The column was then washed with RPE Buffer and centrifuged for 30 s at 11,000 × g, followed by washing with ethanol 80% and centrifuging for 2 min at 11,000 × g. After repeating the centrifugation to dry the membrane, miRNAs were eluted into RNase-free water and stored at −80 °C. The quality of extracted miRNAs was determined using the 2100 Bioanalyzer (Agilent).

2.6. Quantification of EV miRNA Expression Levels

The analysis of NdEV miRNAs was carried out following a two-stage approach.

First, a preliminary screening of NdEV miRNA content was carried out by profiling 754 miRNAs (plus 16 replicates of four controls: ath-miR159a, RNU44, RNU48, and rRNA U6) on three pools of plasma samples from healthy donors and three pools from subjects with MDD. Extracted miRNAs were subjected to reverse transcription (RT) and preamplification, followed by qPCR with the QuantStudio 12 K Flex OpenArray Platform (QS12KFlex, Thermo Fisher Scientific), as described in. Gene Expression Suite Software (Life Technologies) was used to process miRNA expression data. MiRNAs with Ct < 28 and AmpScore >1.20, which were present in >1/6 NdEV samples, were selected for analysis on the study population. Regarding miRNAs in 1/6 NdEV samples, they were considered only if they were found in an MDD pool (i.e., miRNAs identified only in one pool of healthy subjects were excluded). The NormFinder algorithm was used to choose the best normalization strategy. Expression levels were calculated as RQ = 2^–ΔCt, where ΔCt = Ct (measured) – Ct (normalization). Eventually, 52 miRNAs were selected (see below for further details).

The selected miRNAs and controls were quantified in triplicate in the study population using Custom miRNA OpenArray Plates (Thermo Fisher Scientific), similarly to. Briefly, RT was carried out using the Custom Primers Pool and TaqMan Micro RNA Reverse Transcriptase Kit (Thermo Fisher Scientific). Each cDNA sample was then preamplified using a PreAmp Custom Primers Pool (Thermo Fisher Scientific). Experimental and data analysis procedures were the same as in the screening phase. However, only miRNAs with Ct < 28 and AmpScore >1.24, and that were detected in >70% samples, were considered for statistical analysis. Based on NormFinder, the mean of rRNA U6 and ath-miR159a was selected as the best normalization method.

2.7. Bioinformatic and Statistical Analyses

The miRNA Enrichment Analysis and Annotation tool (miEAA) was utilized to perform functional enrichment and over-representation analysis (ORA) on the selected miRNAs, which were mapped to their validated target genes using miRTarBase. Reactome was chosen for the enrichment analysis to map miRNA target genes to biological pathways. miEAA performed ORA by conducting hypergeometric tests to evaluate the significance of the overlap between the miRNA target genes and the genes associated with each pathway in Reactome. The null hypothesis assumed that any observed overlap was due to random chance. Only pathways with p-value < 0.01 and involving at least 10 miRNA target gene hits were considered. A multiple comparison correction method based on the Benjamini-Hochberg False Discovery Rate (FDR) was applied to all the analyses to calculate the FDR p-value.

Descriptive analyses of the study population, MDD characteristics, and related severity scales were performed. Principal exposure windows of PM_2.5 and NO₂ were also summarized. The time windows of interest were chosen following a screening of different time lags between the day of blood sampling and the preceding 365 days. In the end, we focused on lags between 0 and 30 days preceding blood sampling, for two reasons: (1) they roughly corresponded to the time windows considered when evaluating the association between air pollution and MDD severity in the DeprAir study population; (2) they yielded the highest number of significant associations with miRNA levels. Continuous variables were summarized as means with standard deviations. Categorical data were synthesized reporting absolute and relative frequencies. Correlations between miRNAs of interest were expressed as Pearson correlation coefficient.

To study the association between air pollutants exposure and miRNA expression, multivariate linear regression models were used, after applying a log2-transformation to miRNA expression to achieve normal distribution. Models were adjusted for AT (same lag as for air pollution), gender, age, occupation, education, month and year of recruitment, antidepressant treatment (yes vs no), body mass index (BMI), smoking and OpenArray plate. Results were expressed as regression coefficients or slopes (β) per 10 μg/m³ increase in air pollutants concentrations, with corresponding 95% confidence intervals (95%CI).

Multivariate linear regression models were also used to evaluate the association between miRNA expression and MDD severity for all the scales, except for CGI where a multivariate ordinal regression model was used. Models were adjusted by gender, age, occupation, education, month and year of recruitment, antidepressant treatment (yes vs no), BMI, smoking and OpenArray plate. Also in these models, miRNA expression data were log2-transformed. Finally, all miRNA values were multiplied by a −1 coefficient, thus expressing results as β per 1-unit decrease in miRNA expression, with corresponding 95%CI.

To graphically represent results from the regression models, volcano plots were created, assigning a dot to each miRNA and plotting the β vs the negative logarithm of the p-value.

miRTargetLink 2.0 was employed to investigate interactions between target genes and miRNAs associated with pollutant agents and MDD, by querying miRNAs against miRbase and miRTarBase. Only gene targets with strong validation (i.e., experimental evidence of miRNA-mRNA regulation) were selected. ORA of target genes was performed with GeneTrail 3.0 to identify key pathways emerging from the resulting interaction network. All results were visualized using the R package ggplot2.

All analyses were performed using Stata (Stata Corp. 2024; Stata Statistical Software: Release 18; College Station, Texas, USA) and R 4.4.2.

3. Results

A complete descriptive analysis of the socio-demographic characteristics of the study population is reported in Table . Briefly, the mean age of the study subjects was 50.1 years and 62.5% were females. Fifty-three percent had normal weight, 40.0% were overweight or obese, and 7.0% underweight. 74.0% had completed at least a high school diploma, and 45.5% were employed at the time of recruitment. Most subjects were nonsmokers, with 57.0% having never smoked and 9.0% being former smokers. Road traffic at residential address was moderate or heavy in 76.5% cases. About 37% of study subjects were recruited during outpatient visits, 19.5% at day-hospitals, 19.0% were hospitalized, while 25.0% were patients specifically recruited for the DeprAir study.

1. Socio-Demographic Characteristics of the Study Population.

Socio-demographic characteristics	Mean (SD)/N (%)
Age (years)	50.1 (17.3)
Sex
Females	125 (62.5%)
Males	75 (37.5%)
Body mass index (BMI; kg/m2)
Underweight (BMI <18.49)	14 (7.0%)
Normal weight (18,5 < BMI <24.99)	106 (53.0%)
Overweight (25 < BMI <29.99)	50 (25.0%)
Obesity (BMI >30)	30 (15.0%)
Education level
Primary school or less	7 (3.5%)
Secondary school	4 (22.5%)
High school	88 (44.0%)
University degree	60 (30.0%)
Occupation
Employed	91 (45.5%)
Unemployed	46 (23.0%)
Retired	43 (21.5%)
Other	20 (10.0%)
Smoking status
Never smoked	114 (57.0%)
Former smoker	18 (9.0%)
Current smoker	68 (34.0%)
Secondhand smoking exposure
Yes	68 (34.0%)
No	132 (66.0%)
Residence traffic exposure
Mild	47 (23.5%)
Moderate	76 (38.0%)
Heavy	77 (38.5%)
Source of recruitment
Outpatients	73 (36.5%)
Day-hospital	39 (19.5%)
Hospitalizations	38 (19.0%)
Patients recruited for the study	50 (25.0%)

The clinical characteristics of MDD for our study population are reported in Table . The mean age of onset was 39.7 years, with 2.9 episodes of MDD experienced throughout lifetime. Most study participants had neither a family history of psychiatric disorders (58.0%), nor of MDD (71.5%). The majority did not have psychotic symptoms (93.5%), did not try to commit suicide (80.0%), and did not suffer from seasonal depression (72.0%). The most prevalent MDD subtypes were the melancholic one (39.0%) and the one with marked symptoms of anxiety (35.0%). Most subjects (90.0%) were receiving antidepressant treatment at the time of recruitment.

2. MDD Characteristics of the 200 Study Participants.

MDD characteristics	Mean (SD)/N (%)
Age at onset of MDD (years)	39.7 (18.1)
Number of MDD episodes	2.9 (3.0)
Family history of psychiatric disorders
Yes	84 (42.0%)
No	116 (58.0%)
Family history of MDD
Yes	57 (28.5%)
No	143 (71.5%)
Total duration of untreated MDD (months)	21.1 (44.7)
Total duration of MDD (years)	9.9 (12.0)
Duration of the last MDD episode (months)	9.9 (16.2)
Hospitalization for MDD
Yes	58 (29.0%)
No	142 (71.0%)
Among those hospitalized for MDD (n = 58), number of hospitalizations	1.8 (1.5)
Psychotic symptoms
Yes	13 (6.5%)
No	187 (93.5%)
Suicide attempts
Yes	40 (20.0%)
No	160 (80.0%)
Seasonality of MDD
Yes	56 (28.0%)
No	144 (72.0%)
MDD subtype
Melancholic	78 (39.0%)
Atypical	29 (14.5%)
Psychotic	6 (3.0%)
Marked symptoms of anxiety	70 (35.0%)
No prevalent type	17 (8.5%)
Lifetime substance abuse
Single abuse	32 (16.0%)
Multiple abuse	9 (4.5%)
No	159 (79.5%)
Among abusers (n = 41), type(s) of abuse
Alcohol	21 (51.2%)
Cannabis	18 (43.9%)
Heroine	4 (9.8%)
Cocaine	5 (12.2%)
LSD	3 (7.3%)
Amphetamine	2 (4.9%)
3,4-Methylenedioxymethamphetamine (MDMA)	1 (2.4%)
Other drugs	9 (22.0%)
Currently under treatment with antidepressants
Yes	180 (90.0%)
No	20 (10.0%)
Type of treatment
Selective serotonin reuptake inhibitors (SSRI)	107 (59.4%)
Serotonin and norepinephrine reuptake inhibitors (SNRI)	26 (14.4%)
Tyricyclics	15 (8.3%)
Bupropion	2 (1.1%)
Mirtazapine	10 (5.6%)
Vortioxetine	8 (4.4%)
Trazodone	5 (2.8%)
Others	7 (4.0%)

MDD severity and related social functioning is reported in Table . The mean MADRS score was 29.4, which falls into the range of moderate depression (20–34). The mean HAM-D scale score was 25.2, which is considered as severe depression (≥24). The mean GAF score was 58.4 (moderate impairment of functioning); similarly, most subjects (56.0%) were classified as moderately or markedly ill according to the CGI scale. Finally, the subdomains of the SDS scale investigating the social, domestic and working impairment due to MDD returned average scores above 6.5; analogous results were found for the perceived stress category. The average score for the perceived social support was 56.2.

3. MDD Severity of the 200 Study Participants.

MDD severity	Mean (SD)/N (%)
Montgomery-Asberg Depression Rating Scale (MADRS; 0–60)	29.4 (12.4)
Hamilton Depression Rating Scale (HAM-D; 0–67)	25.2 (12.7)
Global Assessment of Functioning (GAF; 100–0)	58.4 (13.8)
Clinical Global Impression (CGI)
Not at all ill	4 (2.0%)
Borderline mentally ill	19 (9.5%)
Mildly ill	41 (20.5%)
Moderately ill	71 (35.5%)
Markedly ill	41 (20.5%)
Severely ill	19 (9.5%)
Extremely ill	5 (2.5%)
Sheehan Disability Scale (SDS)
Impairment at work/school (0–10)	7.2 (2.6)
Impairment in relationships/social life (0–10)	7.1 (2.4)
Impairment in family life/home responsibilities (0–10)	6.7 (2.7)
Perceived stress (0–10)	6.6 (2.8)
Perceived social support (100–0)	56.2 (26.5)

Regarding exposure to air pollutants, the mean PM_2.5 exposure was 20.6 μg/m³ (±10.6) at lag 0–7, 20.6 μg/m³ (±10.2) at lag 0–14, 20.7 μg/m³ (±9.9) at lag 0–21, and 20.7 μg/m³ (±9.8) at lag 0–30. For NO₂, the mean exposure was 37.5 μg/m³ (±13.2) at lag 0–7, 37.3 μg/m³ (±12.6) at lag 0–14, 37.4 μg/m³ (±12.3) at lag 0–21, and 37.1 μg/m³ (±12.1) at lag 0–30. Regarding exposure to meteorological factors, the mean levels for AT were: 13.32 °C (±8.42) at lag 0–7, 13.31 °C (±8.23) at lag 0–14, 13.42 °C (±8.02) at lag 0–21, and 13.55 °C (±7.85) at lag 0–30 (Supporting Information, Table S1).

The efficiency of L1CAM-based immunocapture protocol, as well as the characteristics of circulating NdEVs, were assessed on a limited number of plasma samples (from healthy and MDD subjects) before applying NdEV analysis to the study population. As shown in Figure a–c, immunocaptured EVs were highly enriched in L1CAM + EVs if compared to total EVs; accordingly, depletion of L1CAM + EVs was observed in the supernatant fraction. The mean concentration of NdEVs was (1.9 ± 1.0) × 10⁶ particles/mL, corresponding to 4.5% of total EVs ((4.3 ± 1.3) × 10⁷ particles/mL). In addition, NdEVs were on average smaller (mean size = (182 ± 9) nm; min = 173 nm; max = 192 nm) than total EVs (mean size = (218 ± 12) nm; min = 205 nm; max = 229 nm) (Figure d,e).

2.

Characteristics of L1CAM + EVs. Panel A–C: validation of L1CAM + EVs immunocapture by flow cytometry. Panel A: Pre-IC = preimmunocapture; Panel B and C: post-IC = postimmunocapture. Panel D-E: NdEV analysis by NTA. Panel D: concentration of NdEVs vs total EVs; Panel E: size distribution (30–700 nm) of NdEVs vs total EVs. Panel F: Top biological processes regulated by NdEV miRNAs.

We further investigated the miRNA content of NdEVs from plasma pool samples. Of the 754 miRNAs screened, 65 (8.6%) were found in at least one NdEV sample. All these miRNAs were considered for further analyses, except for those that were present only in one pool, if that pool was from healthy subjects. Overall, 52 miRNAs were selected through this screening.

Target prediction and pathway analysis were carried out to functionally characterize the identified 52 NdEV miRNAs. MiR-1274B and miR-720 were excluded from the analysis as they are not bona fide miRNAs; indeed, they might derive from tRNA processing and their mechanisms of action is currently unknown. Overall, the selected 50 miRNAs were found to target 128 validated genes (FDR p-value < 0.01). The genes regulated by the highest number of miRNAs were TP53 (n = 18 miRNAs), KPNA6 (n = 16 miRNAs), PLAGL2 (n = 15 miRNAs), ARPP19 (n = 14 miRNAs), SLC7A11(n = 14 miRNAs), PKM (n = 13 miRNAs), TWF1 (n = 13 miRNAs), TNFRSF10B (n = 13 miRNAs), FYCO1 (n = 12 miRNAs), and DNMT1 (n = 12 miRNAs) (for the complete list of miRNAs regulating these genes, see Supporting Information, Table S2). Among the top-biological processes regulated by these miRNAs, intracellular signal transduction, cell response to stress/external cues, and immune and senescence-related processes were found to be enriched (if compared to the targets of the 754 miRNAs) (Figure f).

Despite being screened out in the first stage of NdEV miRNA discovery, miR-124a-3p and miR-9–5p were considered for analysis on the study population as they are known to be neuron-specific miRNAs, according to both literature , and database search (miRNA Tissue Atlas database and miTED database) (Supporting Information, Figure S1).

When considering the study population, only 17 out of 54 miRNAs were included for statistical analysis, as the remaining 37 were expressed in <70% of subjects (for the complete list of included miRNAs, see Supporting Information, Table S3). Despite miR-124a-3p and miR-9–5p did not reach the inclusion threshold, they were found to strongly correlate with each other (r = 0.9337).

We observed a negative association between PM_2.5 exposure and the levels of 6 miRNAs when considering different time lags. In particular, increasing PM_2.5 levels were associated with lower miR-191 (β = −0.57; 95%CI = −1.09; −0.05; p-value = 0.033), miR-24 (β = −0.49; 95%CI = −0.88; −0.11; p-value = 0.013), miR-320 (β = −0.40; 95%CI = −0.78; −0.03; p-value = 0.035), and miR-451 (β = −0.56; 95%CI = −1.08; −0.04; p-value = 0.036) at lag 0–7; miR-191 (β = −0.80; 95%CI = −1.38; −0.21; p-value = 0.009) and miR-24 (β = −0.51; 95%CI = −0.98; −0.05; p-value = 0.030) at lag 0–14; miR-191 (β = −1.10; 95%CI = −1.87; −0.34; p-value = 0.005), miR-223 (β = −0.69; 95%CI = −1.32; −0.06; p-value = 0.032), miR-24 (β = −0.74; 95%CI = −1.33; −0.15; p-value = 0.014), miR-320 (β = −0.70; 95%CI = −1.23; −0.18; p-value = 0.009), and miR-638 (β = −0.34; 95%CI = −0.69; −0.00; p-value = 0.048) at lag 0–21; miR-320 (β = −0.83; 95%CI = −1.49; −0.18; p-value = 0.013) and miR-572 (β = −0.54; 95%CI = −1.05; −0.02; p-value = 0.040) at lag 0–30 (Figure ). A complete description of the associations between PM_2.5 and NdEV miRNAs in the time windows of interest is reported in Supporting Information, Table S4.

3.

Volcano plots depicting the associations between PM_2.5 exposure and NdEV miRNAs. Panel A: lag 0–7; Panel B: lag 0–14; Panel C: lag 0–21; Panel D: lag 0–30.

NO₂ exposure was associated with decreased levels of miR-191 (β = −0.39; 95%CI = −0.74; −0.03; p-value = 0.032) and miR-24 (β = −0.29; 95%CI = −0.58; −0.01; p-value = 0.042) at lag 0–7; miR-191 (β = −0.44; 95%CI = −0.83; −0.05; p-value = 0.028) and miR-24 (β = −0.32; 95%CI = −0.64; −0.01; p-value = 0.042) were also associated with NO₂ at lag 0–14. Only miR-191 (β = −0.48; 95%CI = −0.91; −0.06; p-value = 0.027) was negatively associated with NO₂ at lag 0–21. No significant association was found at lag 0–30. Significant associations between NO₂ exposure and NdEV miRNAs are summarized in Figure ; for complete results, see Supporting Information, Table S5.

4.

Volcano plots depicting the associations between NO₂ exposure and NdEV miRNAs. Panel A: lag 0–7; Panel B: lag 0–14; Panel C: lag 0–21; Panel D: lag 0–30.

Multiple miRNAs were associated with MDD severity scales. In particular, decreasing miR-126 (β = 1.43; 95%CI = 2.83; 0.03; p-value = 0.046), miR-19b (β = 1.42; 95%CI = 0.06; 2.79; p-value = 0.042), and miR-572 (β = 3.13; 95%CI = 1.09; 5.17; p-value = 0.003) were associated with higher MADRS score, while decrements of miR-572 (β = 3.08; 95%CI = 1.10; 5.07; p-value = 0.003) were associated with higher HAM-D scores.

Similarly, decrements of miR-126 (β = −1.67; 95%CI = −3.18; −0.16; p-value = 0.030) and miR-572 (β = −2.30; 95%CI = −4.61; 0.00; p-value = 0.050) were associated with worse functioning, as indicated by lower GAF scores.

In addition, decreasing levels of 6 miRNAs, i.e., miR-126 (β = −0.34; 95%CI = −0.11; −0.57; p-value = 0.004), miR-19b (β = 0.25; 95%CI = 0.04; 0.47; p-value = 0.021), miR-320 (β = 0.28; 95%CI = 0.02; 0.54; p-value = 0.032), miR-572 (β = 0.46; 95%CI = 0.15; 0.78; p-value = 0.004), miR-638 (β = 0.41; 95%CI = 0.04; 0.79; p-value = 0.029), and miR-451­(β = 0.24; 95%CI = 0.02; 0.46; p-value = 0.036) were associated with higher CGI scores.

Associations between NdEV miRNAs and MDD severity scales (MADRS, HAM-D, GAF and CGI) are summarized in Figure . The complete list of results is reported in Supporting Information, Table S6.

5.

Volcano plots depicting the associations between decreasing NdEV miRNA levels and MDD severity. Panel A: MADRS; Panel B: HAM-D; Panel C: GAF; Panel D: CGI.

Finally, decreasing miR-572 levels were associated with an increasing score of the home relationship domain of the SDS scale (β = 0.46; 95%CI = 0.05; 0.87; p-value = 0.028) (Supporting Information, Figure S2). Instead, no associations were found between NdEV miRNAs and the disability–impairment at work, family responsibilities, perceived stress, and perceived social support domains of the SDS scale. For the complete analysis regarding the association between miRNAs and the five domains of the SDS scale, refer to Supporting Information, Table S7.

Next, a gene target analysis was conducted to investigate the genes regulated by NdEV miRNAs associated with air pollution and MDD severity. As shown in Figure a, 35 genes are regulated by one miRNA associated with both air pollution and MDD severity (i.e., gray dots around miR-572, miR-451a, and miR-638, n = 24), or by 2+ miRNAs associated with air pollution or MDD severity (i.e., yellow dots connecting miRNAs of different colors, n = 11). For the complete list of genes regulated by miRNAs associated with air pollution and/or MDD severity, refer to Supporting Information.

6.

Panel A: Gene targets of NdEV miRNAs associated with air pollution and/or with MDD severity. Green squares: miRNAs associated with both air pollution and MDD severity; red squares: miRNAs associated with MDD severity; blue squares: miRNA associated with air pollutants; gray dots: genes targeted by only one miRNA; yellow dots: genes targeted by 2+ miRNAs. Panel B: Top biological pathways regulated by miRNA target genes at the interface between air pollution and MDD severity.

As the 35 target genes lie at the interface between air pollution and MDD severity, we conducted a pathway analysis to determine the top pathways regulated by such genes. As shown in Figure b, these genes are mainly implied in inflammatory responses and immune regulation, as well as in cell cycle progression and senescence.

4. Discussion

In the present study, we aimed at investigating the complex interplay between air pollution exposure, the miRNA content of NdEVs, and MDD severity. For this purpose, we first evaluated the characteristics of plasmatic NdEVs, which were enriched starting from total EVs by immunoaffinity capture, an isolation technique which relies on the formation of an immunocomplex between the neuron-enriched L1CAM adhesion molecule and an antibody targeting its ectodomain. , Although other surface proteins have been proposed as NdEV markers over time, we leveraged on L1CAM as it is highly enriched in NdEVs and remains the most widely employed marker for the immunocapture of this EV subset; however, since this choice was not guided by an experimental screening of candidate biomarkers, we cannot exclude that other membrane-bound NdEV biomarkers reported in literature (e.g., ATP1A3, NCAM) might be equally or more informative. The efficiency of this protocol proved to be overall high, with more than 90% of particles corresponding to L1CAM + EVs in the NdEV-enriched fraction. Based on NTA, immunocaptured EVs were on average smaller than total EVs (182 ± 9 nm vs 218 ± 12 nm) and accounted for approximately 4.5% of total particles. Although the number of EV samples analyzed was limited, and they were derived from healthy subjects whose NdEVs may not fully represent those of our study population, characterization by NTA and flow cytometry supports the robustness of our protocol in preventing major EV rupture (as broken vesicles would have been CFSE-negative and peaked around 0–100 nm). In addition, the average percentage of NdEVs is generally in accordance with what observed by Saeedi and colleagues, who similarly reported that NdEVs have a smaller diameter than total EVs; intriguingly, they also observed that NdEVs were even smaller in patients with MDD if compared to healthy controls, suggesting that NdEV size might be also related to pathophysiological states. Also, the percentage of immunocaptured EVs is in line to that measured by NTA by Mustapic and co-workers (9.6%), considering that total EVs were obtained by a precipitation method with higher yield than ultracentrifugation.

Following this preliminary characterization of plasmatic NdEVs, their miRNA content was queried to identify miRNAs enriched in this EV subpopulation. The limited number of miRNA species identified within NdEVs (52 out of 754 miRNAs screened) is consistent with what expected for a well-defined EV subpopulation, as miRNA expression patterns are cell-/tissue-specific, and most EVs contain a low copy number of miRNAs; in addition, the two neuron-enriched miRNAs (miR-9 and miR-124) were found to be highly correlated with each other, thus supporting the specificity of our methodology. We cannot exclude the possibility that the limited number of NdEV miRNAs detected reflects technical challenges; however, appropriate controls were included at all stages of miRNA extraction and quantification, and all 54 miRNAs were expressed in total EVs during the screening phase, suggesting that the assays were properly designed. Overall, in silico target prediction for the selected miRNA panel identified 128 putative gene targets. Among them, SLC7A11 encodes a cystine/glutamate antiporter with antioxidant functions, which has been implied in multiple neuroinflammation-based pathologies of the central nervous system; of note, this gene is highly expressed in astrocytes and its expression levels are strongly modulated by external stressors. Instead, TP53 and FYCO1 are involved in stress response and the regulation of senescence and apoptosis, contributing to DNA repair and autophagy, respectively. , Similarly, other highly enriched genes are involved in cell adaptation to external stimuli (e.g., ARPP19, a cell cycle regulator, and DNMT1, implied in epigenetic remodeling) and neuroinflammation (e.g., TNFRSF10B, encoding a Tumor Necrosis Factor receptor). However, further mechanistic studies are needed to elucidate the regulatory role of these NdEV-borne miRNAs in recipient cells.

Then, we analyzed the 54 miRNAs selected in the study population, whose demographic and clinical characteristics were in line with those of the larger DeprAir study population. Here, we observed that PM_2.5 exposure was negatively associated with the levels of miR-24, miR-191, miR-223, miR-320, miR-451, and miR-638 in the time lags considered, while NO₂ exposure was associated with decreased levels of miR-24 and miR-191 in the same time windows. In addition, miR-19b, miR-126, miR-320, miR-451, miR-572, and miR-638 were found to be associated with at least one MDD severity scale.

To the best of our knowledge, this is the first study analyzing the effect of air pollutants on NdEVs. NO₂ is a traffic-derived gaseous pollutant that can easily cross the epithelial barrier in the airways; once in an aqueous environment, it can spontaneously decompose to form nitric acid (HNO₃) and nitric oxide (NO). Although little is known about the effects of NO₂ inhalation on the brain, high levels of NO can increase the permeability of the blood–brain barrier, resulting in the uncontrolled, potentially deleterious passage of molecules across the blood–brain interface. Instead, PM_2.5 is thought to modulate EV trafficking and content via indirect mechanisms, as the larger size of its particulate components is likely to prevent them from reaching the brain. However, it is also possible that the mode of action of these pollutants might converge on shared pathways; for instance, miR-24 and miR-191 are negatively associated with both NO₂ and PM_2.5 in the same time lags. MiR-24 and miR-191 are widely expressed miRNA that play a key protective role in multiple organs by downregulating inflammation and promoting the recovery of injured tissues; − therefore, a reduction in these miRNAs might impair the ability to counteract air pollution-induced systemic inflammation. In this perspective, we speculate that air pollution-induced alterations in NdEV miRNA levels might also contribute to MDD pathogenesis. Notably, reduced levels of four miRNAs associated with air pollutants (i.e., miR-320, miR-451, miR-572, and miR-638) were also associated with increased MDD severity. Accordingly, previous studies have found that plasmatic miR-320 levels are strongly reduced in MDD patients if compared to healthy controls, and that increase of that miRNA can ease depressive symptoms in rats. Similarly, miR-451 is an anti-inflammatory miRNA which is involved in the pathogenesis of several neurological and psychiatric disorders. − A previous report has shown that MDD subjects have lower circulating miR-451 if compared to healthy controls, and that antidepressant treatment increased its expression levels. Moreover, circulating miR-451 was found to negatively correlate with MDD severity and has been proposed as a potential target for MDD therapy, as its overexpression in the prefrontal cortex ameliorated MDD symptoms in mice. Similar to miR-451, miR-572a pleiotropic miRNA also implied in several neuropsychiatric conditionswas previously found to be upregulated by antidepressants. In addition, miR-638 seems to exert an anti-inflammatory action; however, this miRNA has been mainly studied in the context of cancer research, while its role in the brain and in psychiatric disorders remain largely unknown.

To better understand the biological relevance of our findings, we conducted target gene and pathway enrichment analyses on miRNAs significantly associated with air pollution and/or MDD severity. Interestingly, numerous genes were found to be shared targets of NdEV miRNAs, with TP53 being regulated by the highest number of miRNAs (miR-19b, miR-24, and miR-638). Of note, the gene targets regulated by miRNAs associated with both air pollution and MDD severity were found to be mainly implied in the regulation of the inflammatory response (interleukin signaling and granulocyte activation), cell senescence (modulation of TP53-related pathways and SASP), and cell cycle (Cyclin-dependent pathways), suggesting that such processes might play a major role in mediating the noxious effect of air pollutants on MDD exacerbation. Interestingly, it has been hypothesized that air pollution fuels low-grade inflammation by inducing macrophage senescence in the lungs, with systemic consequences and therefore potentially influencing also brain physiology; however, the involvement of neuronal EVs at the nexus between air pollution exposure and depression remains largely unknown.

Despite being innovative and having several strengths, the present study has also some limitations. First, it is a cross-sectional study, thus it is not possible to establish a cause-and-effect relationship or analyze longitudinal time trends. Second, as we decided to exclude subjects with medical conditions potentially associated with behavioral disorders (e.g., hypothyroidism, previous stroke), we cannot assume that our results can be translated to these subjects as well. Third, we could not conduct a mediation analysis because at lag0–14, the exposure window for which we previously reported an association between air pollution and MDD severity, no miRNA was associated with both air pollution and MDD severity. Regarding the characterization of EVs, analysis of canonical biomarkers (such as tetraspanins) will be warranted to further validate NdEV enrichment by immunoaffinity capture. Also, this study does not investigate whether the identified differences in miRNA cargo that are associated with air pollution or with MDD severity have a functional impact on recipient cells, and which types of cells are targeted by such a “miRNA message”. Future research in vitro or on animal models is needed to elucidate the biological significance of these miRNA alterations.

5. Conclusions

In the present study, we observed an inverse association between air pollution exposure and the expression levels of miRNAs carried by NdEVs, and between such miRNAs and MDD severity. Our findings might pave the way for the identification of minimally invasive, tissue-specific biomarkers mirroring the effect of environmental exposures on mental health, with important implications for personalized medicine, disease prevention, and pollution biomonitoring. The potentiality of NdEV miRNAs as “environmental sensors” and as prognostic biomarkers for MDD should be further investigated in longitudinal studies. Future research is also needed to determine whether the observed associations are causal, thereby clarifying whether plasma NdEVs and their miRNA cargo mediate the noxious effect of air pollution on the exacerbation of MDD, possibly by modulating inflammation and cell cycle-related processes.

The data sets generated during the current study are available from the corresponding author on reasonable request.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/envhealth.5c00336.

• Additional experimental details, including gene targets of NdEV miRNAs, and extended tables reporting the estimates of the tested associations (PDF)

• Genes regulated by miRNAs associated with air pollution and/or MDD severity (XLSX)

Conceptualization: PM, LF, VB, MB, MC; methodology: PM, EB, LD, SI; investigation and formal analysis: PM, RM, ED, LD, GN, FL; visualization: PM, EB, SI, DB; data curation: EB, SI, DB; supervision: LF, VB, MB, MC; writing - original draft: PM; writing - review and editing: all authors; funding acquisition: VB, MC.

Funding for this study was provided by the Italian Ministry of University and Research (FREEDOM study, grant n.: 2022WNW97F) and by the Cariplo Foundation (DeprAir project, grant n.: 2019–3354); the funders had no further role in study design; in the collection, analysis and interpretation of data; in the writing of the report; and in the decision to submit the paper for publication.

The authors declare no competing financial interest.

The present study was performed in accordance with the Declaration of Helsinki and was approved by the local Institutional Review Board of the Fondazione IRCCS Ca’ Granda Ospedale Maggiore Policlinico. All participants provided written informed consent at recruitment.